1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * Definitions for the 'struct sk_buff' memory handlers. 4 * 5 * Authors: 6 * Alan Cox, <[email protected]> 7 * Florian La Roche, <[email protected]> 8 */ 9 10 #ifndef _LINUX_SKBUFF_H 11 #define _LINUX_SKBUFF_H 12 13 #include <linux/kernel.h> 14 #include <linux/compiler.h> 15 #include <linux/time.h> 16 #include <linux/bug.h> 17 #include <linux/bvec.h> 18 #include <linux/cache.h> 19 #include <linux/rbtree.h> 20 #include <linux/socket.h> 21 #include <linux/refcount.h> 22 23 #include <linux/atomic.h> 24 #include <asm/types.h> 25 #include <linux/spinlock.h> 26 #include <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 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 36 #include <linux/netfilter/nf_conntrack_common.h> 37 #endif 38 #include <net/net_debug.h> 39 #include <net/dropreason-core.h> 40 #include <net/netmem.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))) 275 276 struct ahash_request; 277 struct net_device; 278 struct scatterlist; 279 struct pipe_inode_info; 280 struct iov_iter; 281 struct napi_struct; 282 struct bpf_prog; 283 union bpf_attr; 284 struct skb_ext; 285 struct ts_config; 286 287 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 288 struct 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 int physinif; 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 */ 322 struct 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 336 337 struct 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 }; 347 348 struct 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 355 356 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 357 * segment using its current segmentation instead. 358 */ 359 #define GSO_BY_FRAGS 0xFFFF 360 361 typedef struct skb_frag { 362 netmem_ref netmem; 363 unsigned int len; 364 unsigned int offset; 365 } skb_frag_t; 366 367 /** 368 * skb_frag_size() - Returns the size of a skb fragment 369 * @frag: skb fragment 370 */ 371 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 372 { 373 return frag->len; 374 } 375 376 /** 377 * skb_frag_size_set() - Sets the size of a skb fragment 378 * @frag: skb fragment 379 * @size: size of fragment 380 */ 381 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 382 { 383 frag->len = size; 384 } 385 386 /** 387 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 388 * @frag: skb fragment 389 * @delta: value to add 390 */ 391 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 392 { 393 frag->len += delta; 394 } 395 396 /** 397 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 398 * @frag: skb fragment 399 * @delta: value to subtract 400 */ 401 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 402 { 403 frag->len -= delta; 404 } 405 406 /** 407 * skb_frag_must_loop - Test if %p is a high memory page 408 * @p: fragment's page 409 */ 410 static inline bool skb_frag_must_loop(struct page *p) 411 { 412 #if defined(CONFIG_HIGHMEM) 413 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 414 return true; 415 #endif 416 return false; 417 } 418 419 /** 420 * skb_frag_foreach_page - loop over pages in a fragment 421 * 422 * @f: skb frag to operate on 423 * @f_off: offset from start of f->netmem 424 * @f_len: length from f_off to loop over 425 * @p: (temp var) current page 426 * @p_off: (temp var) offset from start of current page, 427 * non-zero only on first page. 428 * @p_len: (temp var) length in current page, 429 * < PAGE_SIZE only on first and last page. 430 * @copied: (temp var) length so far, excluding current p_len. 431 * 432 * A fragment can hold a compound page, in which case per-page 433 * operations, notably kmap_atomic, must be called for each 434 * regular page. 435 */ 436 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 437 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 438 p_off = (f_off) & (PAGE_SIZE - 1), \ 439 p_len = skb_frag_must_loop(p) ? \ 440 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 441 copied = 0; \ 442 copied < f_len; \ 443 copied += p_len, p++, p_off = 0, \ 444 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 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 */ 462 struct 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 */ 470 enum { 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 */ 500 enum { 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 struct ubuf_info_ops { 529 void (*complete)(struct sk_buff *, struct ubuf_info *, 530 bool zerocopy_success); 531 /* has to be compatible with skb_zcopy_set() */ 532 int (*link_skb)(struct sk_buff *skb, struct ubuf_info *uarg); 533 }; 534 535 /* 536 * The callback notifies userspace to release buffers when skb DMA is done in 537 * lower device, the skb last reference should be 0 when calling this. 538 * The zerocopy_success argument is true if zero copy transmit occurred, 539 * false on data copy or out of memory error caused by data copy attempt. 540 * The ctx field is used to track device context. 541 * The desc field is used to track userspace buffer index. 542 */ 543 struct ubuf_info { 544 const struct ubuf_info_ops *ops; 545 refcount_t refcnt; 546 u8 flags; 547 }; 548 549 struct ubuf_info_msgzc { 550 struct ubuf_info ubuf; 551 552 union { 553 struct { 554 unsigned long desc; 555 void *ctx; 556 }; 557 struct { 558 u32 id; 559 u16 len; 560 u16 zerocopy:1; 561 u32 bytelen; 562 }; 563 }; 564 565 struct mmpin { 566 struct user_struct *user; 567 unsigned int num_pg; 568 } mmp; 569 }; 570 571 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 572 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \ 573 ubuf) 574 575 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 576 void mm_unaccount_pinned_pages(struct mmpin *mmp); 577 578 /* Preserve some data across TX submission and completion. 579 * 580 * Note, this state is stored in the driver. Extending the layout 581 * might need some special care. 582 */ 583 struct xsk_tx_metadata_compl { 584 __u64 *tx_timestamp; 585 }; 586 587 /* This data is invariant across clones and lives at 588 * the end of the header data, ie. at skb->end. 589 */ 590 struct skb_shared_info { 591 __u8 flags; 592 __u8 meta_len; 593 __u8 nr_frags; 594 __u8 tx_flags; 595 unsigned short gso_size; 596 /* Warning: this field is not always filled in (UFO)! */ 597 unsigned short gso_segs; 598 struct sk_buff *frag_list; 599 union { 600 struct skb_shared_hwtstamps hwtstamps; 601 struct xsk_tx_metadata_compl xsk_meta; 602 }; 603 unsigned int gso_type; 604 u32 tskey; 605 606 /* 607 * Warning : all fields before dataref are cleared in __alloc_skb() 608 */ 609 atomic_t dataref; 610 unsigned int xdp_frags_size; 611 612 /* Intermediate layers must ensure that destructor_arg 613 * remains valid until skb destructor */ 614 void * destructor_arg; 615 616 /* must be last field, see pskb_expand_head() */ 617 skb_frag_t frags[MAX_SKB_FRAGS]; 618 }; 619 620 /** 621 * DOC: dataref and headerless skbs 622 * 623 * Transport layers send out clones of payload skbs they hold for 624 * retransmissions. To allow lower layers of the stack to prepend their headers 625 * we split &skb_shared_info.dataref into two halves. 626 * The lower 16 bits count the overall number of references. 627 * The higher 16 bits indicate how many of the references are payload-only. 628 * skb_header_cloned() checks if skb is allowed to add / write the headers. 629 * 630 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr 631 * (via __skb_header_release()). Any clone created from marked skb will get 632 * &sk_buff.hdr_len populated with the available headroom. 633 * If there's the only clone in existence it's able to modify the headroom 634 * at will. The sequence of calls inside the transport layer is:: 635 * 636 * <alloc skb> 637 * skb_reserve() 638 * __skb_header_release() 639 * skb_clone() 640 * // send the clone down the stack 641 * 642 * This is not a very generic construct and it depends on the transport layers 643 * doing the right thing. In practice there's usually only one payload-only skb. 644 * Having multiple payload-only skbs with different lengths of hdr_len is not 645 * possible. The payload-only skbs should never leave their owner. 646 */ 647 #define SKB_DATAREF_SHIFT 16 648 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 649 650 651 enum { 652 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 653 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 654 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 655 }; 656 657 enum { 658 SKB_GSO_TCPV4 = 1 << 0, 659 660 /* This indicates the skb is from an untrusted source. */ 661 SKB_GSO_DODGY = 1 << 1, 662 663 /* This indicates the tcp segment has CWR set. */ 664 SKB_GSO_TCP_ECN = 1 << 2, 665 666 SKB_GSO_TCP_FIXEDID = 1 << 3, 667 668 SKB_GSO_TCPV6 = 1 << 4, 669 670 SKB_GSO_FCOE = 1 << 5, 671 672 SKB_GSO_GRE = 1 << 6, 673 674 SKB_GSO_GRE_CSUM = 1 << 7, 675 676 SKB_GSO_IPXIP4 = 1 << 8, 677 678 SKB_GSO_IPXIP6 = 1 << 9, 679 680 SKB_GSO_UDP_TUNNEL = 1 << 10, 681 682 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 683 684 SKB_GSO_PARTIAL = 1 << 12, 685 686 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 687 688 SKB_GSO_SCTP = 1 << 14, 689 690 SKB_GSO_ESP = 1 << 15, 691 692 SKB_GSO_UDP = 1 << 16, 693 694 SKB_GSO_UDP_L4 = 1 << 17, 695 696 SKB_GSO_FRAGLIST = 1 << 18, 697 }; 698 699 #if BITS_PER_LONG > 32 700 #define NET_SKBUFF_DATA_USES_OFFSET 1 701 #endif 702 703 #ifdef NET_SKBUFF_DATA_USES_OFFSET 704 typedef unsigned int sk_buff_data_t; 705 #else 706 typedef unsigned char *sk_buff_data_t; 707 #endif 708 709 enum skb_tstamp_type { 710 SKB_CLOCK_REALTIME, 711 SKB_CLOCK_MONOTONIC, 712 SKB_CLOCK_TAI, 713 __SKB_CLOCK_MAX = SKB_CLOCK_TAI, 714 }; 715 716 /** 717 * DOC: Basic sk_buff geometry 718 * 719 * struct sk_buff itself is a metadata structure and does not hold any packet 720 * data. All the data is held in associated buffers. 721 * 722 * &sk_buff.head points to the main "head" buffer. The head buffer is divided 723 * into two parts: 724 * 725 * - data buffer, containing headers and sometimes payload; 726 * this is the part of the skb operated on by the common helpers 727 * such as skb_put() or skb_pull(); 728 * - shared info (struct skb_shared_info) which holds an array of pointers 729 * to read-only data in the (page, offset, length) format. 730 * 731 * Optionally &skb_shared_info.frag_list may point to another skb. 732 * 733 * Basic diagram may look like this:: 734 * 735 * --------------- 736 * | sk_buff | 737 * --------------- 738 * ,--------------------------- + head 739 * / ,----------------- + data 740 * / / ,----------- + tail 741 * | | | , + end 742 * | | | | 743 * v v v v 744 * ----------------------------------------------- 745 * | headroom | data | tailroom | skb_shared_info | 746 * ----------------------------------------------- 747 * + [page frag] 748 * + [page frag] 749 * + [page frag] 750 * + [page frag] --------- 751 * + frag_list --> | sk_buff | 752 * --------- 753 * 754 */ 755 756 /** 757 * struct sk_buff - socket buffer 758 * @next: Next buffer in list 759 * @prev: Previous buffer in list 760 * @tstamp: Time we arrived/left 761 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 762 * for retransmit timer 763 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 764 * @list: queue head 765 * @ll_node: anchor in an llist (eg socket defer_list) 766 * @sk: Socket we are owned by 767 * @dev: Device we arrived on/are leaving by 768 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 769 * @cb: Control buffer. Free for use by every layer. Put private vars here 770 * @_skb_refdst: destination entry (with norefcount bit) 771 * @len: Length of actual data 772 * @data_len: Data length 773 * @mac_len: Length of link layer header 774 * @hdr_len: writable header length of cloned skb 775 * @csum: Checksum (must include start/offset pair) 776 * @csum_start: Offset from skb->head where checksumming should start 777 * @csum_offset: Offset from csum_start where checksum should be stored 778 * @priority: Packet queueing priority 779 * @ignore_df: allow local fragmentation 780 * @cloned: Head may be cloned (check refcnt to be sure) 781 * @ip_summed: Driver fed us an IP checksum 782 * @nohdr: Payload reference only, must not modify header 783 * @pkt_type: Packet class 784 * @fclone: skbuff clone status 785 * @ipvs_property: skbuff is owned by ipvs 786 * @inner_protocol_type: whether the inner protocol is 787 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 788 * @remcsum_offload: remote checksum offload is enabled 789 * @offload_fwd_mark: Packet was L2-forwarded in hardware 790 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 791 * @tc_skip_classify: do not classify packet. set by IFB device 792 * @tc_at_ingress: used within tc_classify to distinguish in/egress 793 * @redirected: packet was redirected by packet classifier 794 * @from_ingress: packet was redirected from the ingress path 795 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 796 * @peeked: this packet has been seen already, so stats have been 797 * done for it, don't do them again 798 * @nf_trace: netfilter packet trace flag 799 * @protocol: Packet protocol from driver 800 * @destructor: Destruct function 801 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 802 * @_sk_redir: socket redirection information for skmsg 803 * @_nfct: Associated connection, if any (with nfctinfo bits) 804 * @skb_iif: ifindex of device we arrived on 805 * @tc_index: Traffic control index 806 * @hash: the packet hash 807 * @queue_mapping: Queue mapping for multiqueue devices 808 * @head_frag: skb was allocated from page fragments, 809 * not allocated by kmalloc() or vmalloc(). 810 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 811 * @pp_recycle: mark the packet for recycling instead of freeing (implies 812 * page_pool support on driver) 813 * @active_extensions: active extensions (skb_ext_id types) 814 * @ndisc_nodetype: router type (from link layer) 815 * @ooo_okay: allow the mapping of a socket to a queue to be changed 816 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 817 * ports. 818 * @sw_hash: indicates hash was computed in software stack 819 * @wifi_acked_valid: wifi_acked was set 820 * @wifi_acked: whether frame was acked on wifi or not 821 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 822 * @encapsulation: indicates the inner headers in the skbuff are valid 823 * @encap_hdr_csum: software checksum is needed 824 * @csum_valid: checksum is already valid 825 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 826 * @csum_complete_sw: checksum was completed by software 827 * @csum_level: indicates the number of consecutive checksums found in 828 * the packet minus one that have been verified as 829 * CHECKSUM_UNNECESSARY (max 3) 830 * @dst_pending_confirm: need to confirm neighbour 831 * @decrypted: Decrypted SKB 832 * @slow_gro: state present at GRO time, slower prepare step required 833 * @tstamp_type: When set, skb->tstamp has the 834 * delivery_time clock base of skb->tstamp. 835 * @napi_id: id of the NAPI struct this skb came from 836 * @sender_cpu: (aka @napi_id) source CPU in XPS 837 * @alloc_cpu: CPU which did the skb allocation. 838 * @secmark: security marking 839 * @mark: Generic packet mark 840 * @reserved_tailroom: (aka @mark) number of bytes of free space available 841 * at the tail of an sk_buff 842 * @vlan_all: vlan fields (proto & tci) 843 * @vlan_proto: vlan encapsulation protocol 844 * @vlan_tci: vlan tag control information 845 * @inner_protocol: Protocol (encapsulation) 846 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 847 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 848 * @inner_transport_header: Inner transport layer header (encapsulation) 849 * @inner_network_header: Network layer header (encapsulation) 850 * @inner_mac_header: Link layer header (encapsulation) 851 * @transport_header: Transport layer header 852 * @network_header: Network layer header 853 * @mac_header: Link layer header 854 * @kcov_handle: KCOV remote handle for remote coverage collection 855 * @tail: Tail pointer 856 * @end: End pointer 857 * @head: Head of buffer 858 * @data: Data head pointer 859 * @truesize: Buffer size 860 * @users: User count - see {datagram,tcp}.c 861 * @extensions: allocated extensions, valid if active_extensions is nonzero 862 */ 863 864 struct sk_buff { 865 union { 866 struct { 867 /* These two members must be first to match sk_buff_head. */ 868 struct sk_buff *next; 869 struct sk_buff *prev; 870 871 union { 872 struct net_device *dev; 873 /* Some protocols might use this space to store information, 874 * while device pointer would be NULL. 875 * UDP receive path is one user. 876 */ 877 unsigned long dev_scratch; 878 }; 879 }; 880 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 881 struct list_head list; 882 struct llist_node ll_node; 883 }; 884 885 struct sock *sk; 886 887 union { 888 ktime_t tstamp; 889 u64 skb_mstamp_ns; /* earliest departure time */ 890 }; 891 /* 892 * This is the control buffer. It is free to use for every 893 * layer. Please put your private variables there. If you 894 * want to keep them across layers you have to do a skb_clone() 895 * first. This is owned by whoever has the skb queued ATM. 896 */ 897 char cb[48] __aligned(8); 898 899 union { 900 struct { 901 unsigned long _skb_refdst; 902 void (*destructor)(struct sk_buff *skb); 903 }; 904 struct list_head tcp_tsorted_anchor; 905 #ifdef CONFIG_NET_SOCK_MSG 906 unsigned long _sk_redir; 907 #endif 908 }; 909 910 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 911 unsigned long _nfct; 912 #endif 913 unsigned int len, 914 data_len; 915 __u16 mac_len, 916 hdr_len; 917 918 /* Following fields are _not_ copied in __copy_skb_header() 919 * Note that queue_mapping is here mostly to fill a hole. 920 */ 921 __u16 queue_mapping; 922 923 /* if you move cloned around you also must adapt those constants */ 924 #ifdef __BIG_ENDIAN_BITFIELD 925 #define CLONED_MASK (1 << 7) 926 #else 927 #define CLONED_MASK 1 928 #endif 929 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 930 931 /* private: */ 932 __u8 __cloned_offset[0]; 933 /* public: */ 934 __u8 cloned:1, 935 nohdr:1, 936 fclone:2, 937 peeked:1, 938 head_frag:1, 939 pfmemalloc:1, 940 pp_recycle:1; /* page_pool recycle indicator */ 941 #ifdef CONFIG_SKB_EXTENSIONS 942 __u8 active_extensions; 943 #endif 944 945 /* Fields enclosed in headers group are copied 946 * using a single memcpy() in __copy_skb_header() 947 */ 948 struct_group(headers, 949 950 /* private: */ 951 __u8 __pkt_type_offset[0]; 952 /* public: */ 953 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 954 __u8 ignore_df:1; 955 __u8 dst_pending_confirm:1; 956 __u8 ip_summed:2; 957 __u8 ooo_okay:1; 958 959 /* private: */ 960 __u8 __mono_tc_offset[0]; 961 /* public: */ 962 __u8 tstamp_type:2; /* See skb_tstamp_type */ 963 #ifdef CONFIG_NET_XGRESS 964 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 965 __u8 tc_skip_classify:1; 966 #endif 967 __u8 remcsum_offload:1; 968 __u8 csum_complete_sw:1; 969 __u8 csum_level:2; 970 __u8 inner_protocol_type:1; 971 972 __u8 l4_hash:1; 973 __u8 sw_hash:1; 974 #ifdef CONFIG_WIRELESS 975 __u8 wifi_acked_valid:1; 976 __u8 wifi_acked:1; 977 #endif 978 __u8 no_fcs:1; 979 /* Indicates the inner headers are valid in the skbuff. */ 980 __u8 encapsulation:1; 981 __u8 encap_hdr_csum:1; 982 __u8 csum_valid:1; 983 #ifdef CONFIG_IPV6_NDISC_NODETYPE 984 __u8 ndisc_nodetype:2; 985 #endif 986 987 #if IS_ENABLED(CONFIG_IP_VS) 988 __u8 ipvs_property:1; 989 #endif 990 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 991 __u8 nf_trace:1; 992 #endif 993 #ifdef CONFIG_NET_SWITCHDEV 994 __u8 offload_fwd_mark:1; 995 __u8 offload_l3_fwd_mark:1; 996 #endif 997 __u8 redirected:1; 998 #ifdef CONFIG_NET_REDIRECT 999 __u8 from_ingress:1; 1000 #endif 1001 #ifdef CONFIG_NETFILTER_SKIP_EGRESS 1002 __u8 nf_skip_egress:1; 1003 #endif 1004 #ifdef CONFIG_SKB_DECRYPTED 1005 __u8 decrypted:1; 1006 #endif 1007 __u8 slow_gro:1; 1008 #if IS_ENABLED(CONFIG_IP_SCTP) 1009 __u8 csum_not_inet:1; 1010 #endif 1011 1012 #if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS) 1013 __u16 tc_index; /* traffic control index */ 1014 #endif 1015 1016 u16 alloc_cpu; 1017 1018 union { 1019 __wsum csum; 1020 struct { 1021 __u16 csum_start; 1022 __u16 csum_offset; 1023 }; 1024 }; 1025 __u32 priority; 1026 int skb_iif; 1027 __u32 hash; 1028 union { 1029 u32 vlan_all; 1030 struct { 1031 __be16 vlan_proto; 1032 __u16 vlan_tci; 1033 }; 1034 }; 1035 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1036 union { 1037 unsigned int napi_id; 1038 unsigned int sender_cpu; 1039 }; 1040 #endif 1041 #ifdef CONFIG_NETWORK_SECMARK 1042 __u32 secmark; 1043 #endif 1044 1045 union { 1046 __u32 mark; 1047 __u32 reserved_tailroom; 1048 }; 1049 1050 union { 1051 __be16 inner_protocol; 1052 __u8 inner_ipproto; 1053 }; 1054 1055 __u16 inner_transport_header; 1056 __u16 inner_network_header; 1057 __u16 inner_mac_header; 1058 1059 __be16 protocol; 1060 __u16 transport_header; 1061 __u16 network_header; 1062 __u16 mac_header; 1063 1064 #ifdef CONFIG_KCOV 1065 u64 kcov_handle; 1066 #endif 1067 1068 ); /* end headers group */ 1069 1070 /* These elements must be at the end, see alloc_skb() for details. */ 1071 sk_buff_data_t tail; 1072 sk_buff_data_t end; 1073 unsigned char *head, 1074 *data; 1075 unsigned int truesize; 1076 refcount_t users; 1077 1078 #ifdef CONFIG_SKB_EXTENSIONS 1079 /* only usable after checking ->active_extensions != 0 */ 1080 struct skb_ext *extensions; 1081 #endif 1082 }; 1083 1084 /* if you move pkt_type around you also must adapt those constants */ 1085 #ifdef __BIG_ENDIAN_BITFIELD 1086 #define PKT_TYPE_MAX (7 << 5) 1087 #else 1088 #define PKT_TYPE_MAX 7 1089 #endif 1090 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1091 1092 /* if you move tc_at_ingress or tstamp_type 1093 * around, you also must adapt these constants. 1094 */ 1095 #ifdef __BIG_ENDIAN_BITFIELD 1096 #define SKB_TSTAMP_TYPE_MASK (3 << 6) 1097 #define SKB_TSTAMP_TYPE_RSHIFT (6) 1098 #define TC_AT_INGRESS_MASK (1 << 5) 1099 #else 1100 #define SKB_TSTAMP_TYPE_MASK (3) 1101 #define TC_AT_INGRESS_MASK (1 << 2) 1102 #endif 1103 #define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset) 1104 1105 #ifdef __KERNEL__ 1106 /* 1107 * Handling routines are only of interest to the kernel 1108 */ 1109 1110 #define SKB_ALLOC_FCLONE 0x01 1111 #define SKB_ALLOC_RX 0x02 1112 #define SKB_ALLOC_NAPI 0x04 1113 1114 /** 1115 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1116 * @skb: buffer 1117 */ 1118 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1119 { 1120 return unlikely(skb->pfmemalloc); 1121 } 1122 1123 /* 1124 * skb might have a dst pointer attached, refcounted or not. 1125 * _skb_refdst low order bit is set if refcount was _not_ taken 1126 */ 1127 #define SKB_DST_NOREF 1UL 1128 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1129 1130 /** 1131 * skb_dst - returns skb dst_entry 1132 * @skb: buffer 1133 * 1134 * Returns skb dst_entry, regardless of reference taken or not. 1135 */ 1136 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1137 { 1138 /* If refdst was not refcounted, check we still are in a 1139 * rcu_read_lock section 1140 */ 1141 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1142 !rcu_read_lock_held() && 1143 !rcu_read_lock_bh_held()); 1144 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1145 } 1146 1147 /** 1148 * skb_dst_set - sets skb dst 1149 * @skb: buffer 1150 * @dst: dst entry 1151 * 1152 * Sets skb dst, assuming a reference was taken on dst and should 1153 * be released by skb_dst_drop() 1154 */ 1155 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1156 { 1157 skb->slow_gro |= !!dst; 1158 skb->_skb_refdst = (unsigned long)dst; 1159 } 1160 1161 /** 1162 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1163 * @skb: buffer 1164 * @dst: dst entry 1165 * 1166 * Sets skb dst, assuming a reference was not taken on dst. 1167 * If dst entry is cached, we do not take reference and dst_release 1168 * will be avoided by refdst_drop. If dst entry is not cached, we take 1169 * reference, so that last dst_release can destroy the dst immediately. 1170 */ 1171 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1172 { 1173 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1174 skb->slow_gro |= !!dst; 1175 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1176 } 1177 1178 /** 1179 * skb_dst_is_noref - Test if skb dst isn't refcounted 1180 * @skb: buffer 1181 */ 1182 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1183 { 1184 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1185 } 1186 1187 /* For mangling skb->pkt_type from user space side from applications 1188 * such as nft, tc, etc, we only allow a conservative subset of 1189 * possible pkt_types to be set. 1190 */ 1191 static inline bool skb_pkt_type_ok(u32 ptype) 1192 { 1193 return ptype <= PACKET_OTHERHOST; 1194 } 1195 1196 /** 1197 * skb_napi_id - Returns the skb's NAPI id 1198 * @skb: buffer 1199 */ 1200 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1201 { 1202 #ifdef CONFIG_NET_RX_BUSY_POLL 1203 return skb->napi_id; 1204 #else 1205 return 0; 1206 #endif 1207 } 1208 1209 static inline bool skb_wifi_acked_valid(const struct sk_buff *skb) 1210 { 1211 #ifdef CONFIG_WIRELESS 1212 return skb->wifi_acked_valid; 1213 #else 1214 return 0; 1215 #endif 1216 } 1217 1218 /** 1219 * skb_unref - decrement the skb's reference count 1220 * @skb: buffer 1221 * 1222 * Returns true if we can free the skb. 1223 */ 1224 static inline bool skb_unref(struct sk_buff *skb) 1225 { 1226 if (unlikely(!skb)) 1227 return false; 1228 if (!IS_ENABLED(CONFIG_DEBUG_NET) && likely(refcount_read(&skb->users) == 1)) 1229 smp_rmb(); 1230 else if (likely(!refcount_dec_and_test(&skb->users))) 1231 return false; 1232 1233 return true; 1234 } 1235 1236 static inline bool skb_data_unref(const struct sk_buff *skb, 1237 struct skb_shared_info *shinfo) 1238 { 1239 int bias; 1240 1241 if (!skb->cloned) 1242 return true; 1243 1244 bias = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1; 1245 1246 if (atomic_read(&shinfo->dataref) == bias) 1247 smp_rmb(); 1248 else if (atomic_sub_return(bias, &shinfo->dataref)) 1249 return false; 1250 1251 return true; 1252 } 1253 1254 void __fix_address sk_skb_reason_drop(struct sock *sk, struct sk_buff *skb, 1255 enum skb_drop_reason reason); 1256 1257 static inline void 1258 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason) 1259 { 1260 sk_skb_reason_drop(NULL, skb, reason); 1261 } 1262 1263 /** 1264 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1265 * @skb: buffer to free 1266 */ 1267 static inline void kfree_skb(struct sk_buff *skb) 1268 { 1269 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1270 } 1271 1272 void skb_release_head_state(struct sk_buff *skb); 1273 void kfree_skb_list_reason(struct sk_buff *segs, 1274 enum skb_drop_reason reason); 1275 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1276 void skb_tx_error(struct sk_buff *skb); 1277 1278 static inline void kfree_skb_list(struct sk_buff *segs) 1279 { 1280 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1281 } 1282 1283 #ifdef CONFIG_TRACEPOINTS 1284 void consume_skb(struct sk_buff *skb); 1285 #else 1286 static inline void consume_skb(struct sk_buff *skb) 1287 { 1288 return kfree_skb(skb); 1289 } 1290 #endif 1291 1292 void __consume_stateless_skb(struct sk_buff *skb); 1293 void __kfree_skb(struct sk_buff *skb); 1294 1295 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1296 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1297 bool *fragstolen, int *delta_truesize); 1298 1299 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1300 int node); 1301 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1302 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1303 struct sk_buff *build_skb_around(struct sk_buff *skb, 1304 void *data, unsigned int frag_size); 1305 void skb_attempt_defer_free(struct sk_buff *skb); 1306 1307 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1308 struct sk_buff *slab_build_skb(void *data); 1309 1310 /** 1311 * alloc_skb - allocate a network buffer 1312 * @size: size to allocate 1313 * @priority: allocation mask 1314 * 1315 * This function is a convenient wrapper around __alloc_skb(). 1316 */ 1317 static inline struct sk_buff *alloc_skb(unsigned int size, 1318 gfp_t priority) 1319 { 1320 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1321 } 1322 1323 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1324 unsigned long data_len, 1325 int max_page_order, 1326 int *errcode, 1327 gfp_t gfp_mask); 1328 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1329 1330 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1331 struct sk_buff_fclones { 1332 struct sk_buff skb1; 1333 1334 struct sk_buff skb2; 1335 1336 refcount_t fclone_ref; 1337 }; 1338 1339 /** 1340 * skb_fclone_busy - check if fclone is busy 1341 * @sk: socket 1342 * @skb: buffer 1343 * 1344 * Returns true if skb is a fast clone, and its clone is not freed. 1345 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1346 * so we also check that didn't happen. 1347 */ 1348 static inline bool skb_fclone_busy(const struct sock *sk, 1349 const struct sk_buff *skb) 1350 { 1351 const struct sk_buff_fclones *fclones; 1352 1353 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1354 1355 return skb->fclone == SKB_FCLONE_ORIG && 1356 refcount_read(&fclones->fclone_ref) > 1 && 1357 READ_ONCE(fclones->skb2.sk) == sk; 1358 } 1359 1360 /** 1361 * alloc_skb_fclone - allocate a network buffer from fclone cache 1362 * @size: size to allocate 1363 * @priority: allocation mask 1364 * 1365 * This function is a convenient wrapper around __alloc_skb(). 1366 */ 1367 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1368 gfp_t priority) 1369 { 1370 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1371 } 1372 1373 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1374 void skb_headers_offset_update(struct sk_buff *skb, int off); 1375 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1376 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1377 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1378 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1379 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1380 gfp_t gfp_mask, bool fclone); 1381 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1382 gfp_t gfp_mask) 1383 { 1384 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1385 } 1386 1387 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1388 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1389 unsigned int headroom); 1390 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1391 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1392 int newtailroom, gfp_t priority); 1393 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1394 int offset, int len); 1395 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1396 int offset, int len); 1397 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1398 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1399 1400 /** 1401 * skb_pad - zero pad the tail of an skb 1402 * @skb: buffer to pad 1403 * @pad: space to pad 1404 * 1405 * Ensure that a buffer is followed by a padding area that is zero 1406 * filled. Used by network drivers which may DMA or transfer data 1407 * beyond the buffer end onto the wire. 1408 * 1409 * May return error in out of memory cases. The skb is freed on error. 1410 */ 1411 static inline int skb_pad(struct sk_buff *skb, int pad) 1412 { 1413 return __skb_pad(skb, pad, true); 1414 } 1415 #define dev_kfree_skb(a) consume_skb(a) 1416 1417 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1418 int offset, size_t size, size_t max_frags); 1419 1420 struct skb_seq_state { 1421 __u32 lower_offset; 1422 __u32 upper_offset; 1423 __u32 frag_idx; 1424 __u32 stepped_offset; 1425 struct sk_buff *root_skb; 1426 struct sk_buff *cur_skb; 1427 __u8 *frag_data; 1428 __u32 frag_off; 1429 }; 1430 1431 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1432 unsigned int to, struct skb_seq_state *st); 1433 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1434 struct skb_seq_state *st); 1435 void skb_abort_seq_read(struct skb_seq_state *st); 1436 int skb_copy_seq_read(struct skb_seq_state *st, int offset, void *to, int len); 1437 1438 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1439 unsigned int to, struct ts_config *config); 1440 1441 /* 1442 * Packet hash types specify the type of hash in skb_set_hash. 1443 * 1444 * Hash types refer to the protocol layer addresses which are used to 1445 * construct a packet's hash. The hashes are used to differentiate or identify 1446 * flows of the protocol layer for the hash type. Hash types are either 1447 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1448 * 1449 * Properties of hashes: 1450 * 1451 * 1) Two packets in different flows have different hash values 1452 * 2) Two packets in the same flow should have the same hash value 1453 * 1454 * A hash at a higher layer is considered to be more specific. A driver should 1455 * set the most specific hash possible. 1456 * 1457 * A driver cannot indicate a more specific hash than the layer at which a hash 1458 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1459 * 1460 * A driver may indicate a hash level which is less specific than the 1461 * actual layer the hash was computed on. For instance, a hash computed 1462 * at L4 may be considered an L3 hash. This should only be done if the 1463 * driver can't unambiguously determine that the HW computed the hash at 1464 * the higher layer. Note that the "should" in the second property above 1465 * permits this. 1466 */ 1467 enum pkt_hash_types { 1468 PKT_HASH_TYPE_NONE, /* Undefined type */ 1469 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1470 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1471 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1472 }; 1473 1474 static inline void skb_clear_hash(struct sk_buff *skb) 1475 { 1476 skb->hash = 0; 1477 skb->sw_hash = 0; 1478 skb->l4_hash = 0; 1479 } 1480 1481 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1482 { 1483 if (!skb->l4_hash) 1484 skb_clear_hash(skb); 1485 } 1486 1487 static inline void 1488 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1489 { 1490 skb->l4_hash = is_l4; 1491 skb->sw_hash = is_sw; 1492 skb->hash = hash; 1493 } 1494 1495 static inline void 1496 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1497 { 1498 /* Used by drivers to set hash from HW */ 1499 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1500 } 1501 1502 static inline void 1503 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1504 { 1505 __skb_set_hash(skb, hash, true, is_l4); 1506 } 1507 1508 u32 __skb_get_hash_symmetric_net(const struct net *net, const struct sk_buff *skb); 1509 1510 static inline u32 __skb_get_hash_symmetric(const struct sk_buff *skb) 1511 { 1512 return __skb_get_hash_symmetric_net(NULL, skb); 1513 } 1514 1515 void __skb_get_hash_net(const struct net *net, struct sk_buff *skb); 1516 u32 skb_get_poff(const struct sk_buff *skb); 1517 u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1518 const struct flow_keys_basic *keys, int hlen); 1519 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1520 const void *data, int hlen_proto); 1521 1522 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1523 int thoff, u8 ip_proto) 1524 { 1525 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1526 } 1527 1528 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1529 const struct flow_dissector_key *key, 1530 unsigned int key_count); 1531 1532 struct bpf_flow_dissector; 1533 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1534 __be16 proto, int nhoff, int hlen, unsigned int flags); 1535 1536 bool __skb_flow_dissect(const struct net *net, 1537 const struct sk_buff *skb, 1538 struct flow_dissector *flow_dissector, 1539 void *target_container, const void *data, 1540 __be16 proto, int nhoff, int hlen, unsigned int flags); 1541 1542 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1543 struct flow_dissector *flow_dissector, 1544 void *target_container, unsigned int flags) 1545 { 1546 return __skb_flow_dissect(NULL, skb, flow_dissector, 1547 target_container, NULL, 0, 0, 0, flags); 1548 } 1549 1550 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1551 struct flow_keys *flow, 1552 unsigned int flags) 1553 { 1554 memset(flow, 0, sizeof(*flow)); 1555 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1556 flow, NULL, 0, 0, 0, flags); 1557 } 1558 1559 static inline bool 1560 skb_flow_dissect_flow_keys_basic(const struct net *net, 1561 const struct sk_buff *skb, 1562 struct flow_keys_basic *flow, 1563 const void *data, __be16 proto, 1564 int nhoff, int hlen, unsigned int flags) 1565 { 1566 memset(flow, 0, sizeof(*flow)); 1567 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1568 data, proto, nhoff, hlen, flags); 1569 } 1570 1571 void skb_flow_dissect_meta(const struct sk_buff *skb, 1572 struct flow_dissector *flow_dissector, 1573 void *target_container); 1574 1575 /* Gets a skb connection tracking info, ctinfo map should be a 1576 * map of mapsize to translate enum ip_conntrack_info states 1577 * to user states. 1578 */ 1579 void 1580 skb_flow_dissect_ct(const struct sk_buff *skb, 1581 struct flow_dissector *flow_dissector, 1582 void *target_container, 1583 u16 *ctinfo_map, size_t mapsize, 1584 bool post_ct, u16 zone); 1585 void 1586 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1587 struct flow_dissector *flow_dissector, 1588 void *target_container); 1589 1590 void skb_flow_dissect_hash(const struct sk_buff *skb, 1591 struct flow_dissector *flow_dissector, 1592 void *target_container); 1593 1594 static inline __u32 skb_get_hash_net(const struct net *net, struct sk_buff *skb) 1595 { 1596 if (!skb->l4_hash && !skb->sw_hash) 1597 __skb_get_hash_net(net, skb); 1598 1599 return skb->hash; 1600 } 1601 1602 static inline __u32 skb_get_hash(struct sk_buff *skb) 1603 { 1604 if (!skb->l4_hash && !skb->sw_hash) 1605 __skb_get_hash_net(NULL, skb); 1606 1607 return skb->hash; 1608 } 1609 1610 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1611 { 1612 if (!skb->l4_hash && !skb->sw_hash) { 1613 struct flow_keys keys; 1614 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1615 1616 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1617 } 1618 1619 return skb->hash; 1620 } 1621 1622 __u32 skb_get_hash_perturb(const struct sk_buff *skb, 1623 const siphash_key_t *perturb); 1624 1625 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1626 { 1627 return skb->hash; 1628 } 1629 1630 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1631 { 1632 to->hash = from->hash; 1633 to->sw_hash = from->sw_hash; 1634 to->l4_hash = from->l4_hash; 1635 }; 1636 1637 static inline int skb_cmp_decrypted(const struct sk_buff *skb1, 1638 const struct sk_buff *skb2) 1639 { 1640 #ifdef CONFIG_SKB_DECRYPTED 1641 return skb2->decrypted - skb1->decrypted; 1642 #else 1643 return 0; 1644 #endif 1645 } 1646 1647 static inline bool skb_is_decrypted(const struct sk_buff *skb) 1648 { 1649 #ifdef CONFIG_SKB_DECRYPTED 1650 return skb->decrypted; 1651 #else 1652 return false; 1653 #endif 1654 } 1655 1656 static inline void skb_copy_decrypted(struct sk_buff *to, 1657 const struct sk_buff *from) 1658 { 1659 #ifdef CONFIG_SKB_DECRYPTED 1660 to->decrypted = from->decrypted; 1661 #endif 1662 } 1663 1664 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1665 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1666 { 1667 return skb->head + skb->end; 1668 } 1669 1670 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1671 { 1672 return skb->end; 1673 } 1674 1675 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1676 { 1677 skb->end = offset; 1678 } 1679 #else 1680 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1681 { 1682 return skb->end; 1683 } 1684 1685 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1686 { 1687 return skb->end - skb->head; 1688 } 1689 1690 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1691 { 1692 skb->end = skb->head + offset; 1693 } 1694 #endif 1695 1696 extern const struct ubuf_info_ops msg_zerocopy_ubuf_ops; 1697 1698 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1699 struct ubuf_info *uarg); 1700 1701 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1702 1703 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, 1704 struct sk_buff *skb, struct iov_iter *from, 1705 size_t length); 1706 1707 int zerocopy_fill_skb_from_iter(struct sk_buff *skb, 1708 struct iov_iter *from, size_t length); 1709 1710 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1711 struct msghdr *msg, int len) 1712 { 1713 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len); 1714 } 1715 1716 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1717 struct msghdr *msg, int len, 1718 struct ubuf_info *uarg); 1719 1720 /* Internal */ 1721 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1722 1723 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1724 { 1725 return &skb_shinfo(skb)->hwtstamps; 1726 } 1727 1728 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1729 { 1730 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1731 1732 return is_zcopy ? skb_uarg(skb) : NULL; 1733 } 1734 1735 static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1736 { 1737 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1738 } 1739 1740 static inline bool skb_zcopy_managed(const struct sk_buff *skb) 1741 { 1742 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS; 1743 } 1744 1745 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1746 const struct sk_buff *skb2) 1747 { 1748 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1749 } 1750 1751 static inline void net_zcopy_get(struct ubuf_info *uarg) 1752 { 1753 refcount_inc(&uarg->refcnt); 1754 } 1755 1756 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1757 { 1758 skb_shinfo(skb)->destructor_arg = uarg; 1759 skb_shinfo(skb)->flags |= uarg->flags; 1760 } 1761 1762 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1763 bool *have_ref) 1764 { 1765 if (skb && uarg && !skb_zcopy(skb)) { 1766 if (unlikely(have_ref && *have_ref)) 1767 *have_ref = false; 1768 else 1769 net_zcopy_get(uarg); 1770 skb_zcopy_init(skb, uarg); 1771 } 1772 } 1773 1774 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1775 { 1776 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1777 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1778 } 1779 1780 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1781 { 1782 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1783 } 1784 1785 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1786 { 1787 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1788 } 1789 1790 static inline void net_zcopy_put(struct ubuf_info *uarg) 1791 { 1792 if (uarg) 1793 uarg->ops->complete(NULL, uarg, true); 1794 } 1795 1796 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1797 { 1798 if (uarg) { 1799 if (uarg->ops == &msg_zerocopy_ubuf_ops) 1800 msg_zerocopy_put_abort(uarg, have_uref); 1801 else if (have_uref) 1802 net_zcopy_put(uarg); 1803 } 1804 } 1805 1806 /* Release a reference on a zerocopy structure */ 1807 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1808 { 1809 struct ubuf_info *uarg = skb_zcopy(skb); 1810 1811 if (uarg) { 1812 if (!skb_zcopy_is_nouarg(skb)) 1813 uarg->ops->complete(skb, uarg, zerocopy_success); 1814 1815 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1816 } 1817 } 1818 1819 void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1820 1821 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1822 { 1823 if (unlikely(skb_zcopy_managed(skb))) 1824 __skb_zcopy_downgrade_managed(skb); 1825 } 1826 1827 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1828 { 1829 skb->next = NULL; 1830 } 1831 1832 static inline void skb_poison_list(struct sk_buff *skb) 1833 { 1834 #ifdef CONFIG_DEBUG_NET 1835 skb->next = SKB_LIST_POISON_NEXT; 1836 #endif 1837 } 1838 1839 /* Iterate through singly-linked GSO fragments of an skb. */ 1840 #define skb_list_walk_safe(first, skb, next_skb) \ 1841 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1842 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1843 1844 static inline void skb_list_del_init(struct sk_buff *skb) 1845 { 1846 __list_del_entry(&skb->list); 1847 skb_mark_not_on_list(skb); 1848 } 1849 1850 /** 1851 * skb_queue_empty - check if a queue is empty 1852 * @list: queue head 1853 * 1854 * Returns true if the queue is empty, false otherwise. 1855 */ 1856 static inline int skb_queue_empty(const struct sk_buff_head *list) 1857 { 1858 return list->next == (const struct sk_buff *) list; 1859 } 1860 1861 /** 1862 * skb_queue_empty_lockless - check if a queue is empty 1863 * @list: queue head 1864 * 1865 * Returns true if the queue is empty, false otherwise. 1866 * This variant can be used in lockless contexts. 1867 */ 1868 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1869 { 1870 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1871 } 1872 1873 1874 /** 1875 * skb_queue_is_last - check if skb is the last entry in the queue 1876 * @list: queue head 1877 * @skb: buffer 1878 * 1879 * Returns true if @skb is the last buffer on the list. 1880 */ 1881 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1882 const struct sk_buff *skb) 1883 { 1884 return skb->next == (const struct sk_buff *) list; 1885 } 1886 1887 /** 1888 * skb_queue_is_first - check if skb is the first entry in the queue 1889 * @list: queue head 1890 * @skb: buffer 1891 * 1892 * Returns true if @skb is the first buffer on the list. 1893 */ 1894 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1895 const struct sk_buff *skb) 1896 { 1897 return skb->prev == (const struct sk_buff *) list; 1898 } 1899 1900 /** 1901 * skb_queue_next - return the next packet in the queue 1902 * @list: queue head 1903 * @skb: current buffer 1904 * 1905 * Return the next packet in @list after @skb. It is only valid to 1906 * call this if skb_queue_is_last() evaluates to false. 1907 */ 1908 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1909 const struct sk_buff *skb) 1910 { 1911 /* This BUG_ON may seem severe, but if we just return then we 1912 * are going to dereference garbage. 1913 */ 1914 BUG_ON(skb_queue_is_last(list, skb)); 1915 return skb->next; 1916 } 1917 1918 /** 1919 * skb_queue_prev - return the prev packet in the queue 1920 * @list: queue head 1921 * @skb: current buffer 1922 * 1923 * Return the prev packet in @list before @skb. It is only valid to 1924 * call this if skb_queue_is_first() evaluates to false. 1925 */ 1926 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1927 const struct sk_buff *skb) 1928 { 1929 /* This BUG_ON may seem severe, but if we just return then we 1930 * are going to dereference garbage. 1931 */ 1932 BUG_ON(skb_queue_is_first(list, skb)); 1933 return skb->prev; 1934 } 1935 1936 /** 1937 * skb_get - reference buffer 1938 * @skb: buffer to reference 1939 * 1940 * Makes another reference to a socket buffer and returns a pointer 1941 * to the buffer. 1942 */ 1943 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1944 { 1945 refcount_inc(&skb->users); 1946 return skb; 1947 } 1948 1949 /* 1950 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1951 */ 1952 1953 /** 1954 * skb_cloned - is the buffer a clone 1955 * @skb: buffer to check 1956 * 1957 * Returns true if the buffer was generated with skb_clone() and is 1958 * one of multiple shared copies of the buffer. Cloned buffers are 1959 * shared data so must not be written to under normal circumstances. 1960 */ 1961 static inline int skb_cloned(const struct sk_buff *skb) 1962 { 1963 return skb->cloned && 1964 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1965 } 1966 1967 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1968 { 1969 might_sleep_if(gfpflags_allow_blocking(pri)); 1970 1971 if (skb_cloned(skb)) 1972 return pskb_expand_head(skb, 0, 0, pri); 1973 1974 return 0; 1975 } 1976 1977 /* This variant of skb_unclone() makes sure skb->truesize 1978 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 1979 * 1980 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 1981 * when various debugging features are in place. 1982 */ 1983 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 1984 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 1985 { 1986 might_sleep_if(gfpflags_allow_blocking(pri)); 1987 1988 if (skb_cloned(skb)) 1989 return __skb_unclone_keeptruesize(skb, pri); 1990 return 0; 1991 } 1992 1993 /** 1994 * skb_header_cloned - is the header a clone 1995 * @skb: buffer to check 1996 * 1997 * Returns true if modifying the header part of the buffer requires 1998 * the data to be copied. 1999 */ 2000 static inline int skb_header_cloned(const struct sk_buff *skb) 2001 { 2002 int dataref; 2003 2004 if (!skb->cloned) 2005 return 0; 2006 2007 dataref = atomic_read(&skb_shinfo(skb)->dataref); 2008 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 2009 return dataref != 1; 2010 } 2011 2012 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 2013 { 2014 might_sleep_if(gfpflags_allow_blocking(pri)); 2015 2016 if (skb_header_cloned(skb)) 2017 return pskb_expand_head(skb, 0, 0, pri); 2018 2019 return 0; 2020 } 2021 2022 /** 2023 * __skb_header_release() - allow clones to use the headroom 2024 * @skb: buffer to operate on 2025 * 2026 * See "DOC: dataref and headerless skbs". 2027 */ 2028 static inline void __skb_header_release(struct sk_buff *skb) 2029 { 2030 skb->nohdr = 1; 2031 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 2032 } 2033 2034 2035 /** 2036 * skb_shared - is the buffer shared 2037 * @skb: buffer to check 2038 * 2039 * Returns true if more than one person has a reference to this 2040 * buffer. 2041 */ 2042 static inline int skb_shared(const struct sk_buff *skb) 2043 { 2044 return refcount_read(&skb->users) != 1; 2045 } 2046 2047 /** 2048 * skb_share_check - check if buffer is shared and if so clone it 2049 * @skb: buffer to check 2050 * @pri: priority for memory allocation 2051 * 2052 * If the buffer is shared the buffer is cloned and the old copy 2053 * drops a reference. A new clone with a single reference is returned. 2054 * If the buffer is not shared the original buffer is returned. When 2055 * being called from interrupt status or with spinlocks held pri must 2056 * be GFP_ATOMIC. 2057 * 2058 * NULL is returned on a memory allocation failure. 2059 */ 2060 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 2061 { 2062 might_sleep_if(gfpflags_allow_blocking(pri)); 2063 if (skb_shared(skb)) { 2064 struct sk_buff *nskb = skb_clone(skb, pri); 2065 2066 if (likely(nskb)) 2067 consume_skb(skb); 2068 else 2069 kfree_skb(skb); 2070 skb = nskb; 2071 } 2072 return skb; 2073 } 2074 2075 /* 2076 * Copy shared buffers into a new sk_buff. We effectively do COW on 2077 * packets to handle cases where we have a local reader and forward 2078 * and a couple of other messy ones. The normal one is tcpdumping 2079 * a packet that's being forwarded. 2080 */ 2081 2082 /** 2083 * skb_unshare - make a copy of a shared buffer 2084 * @skb: buffer to check 2085 * @pri: priority for memory allocation 2086 * 2087 * If the socket buffer is a clone then this function creates a new 2088 * copy of the data, drops a reference count on the old copy and returns 2089 * the new copy with the reference count at 1. If the buffer is not a clone 2090 * the original buffer is returned. When called with a spinlock held or 2091 * from interrupt state @pri must be %GFP_ATOMIC 2092 * 2093 * %NULL is returned on a memory allocation failure. 2094 */ 2095 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2096 gfp_t pri) 2097 { 2098 might_sleep_if(gfpflags_allow_blocking(pri)); 2099 if (skb_cloned(skb)) { 2100 struct sk_buff *nskb = skb_copy(skb, pri); 2101 2102 /* Free our shared copy */ 2103 if (likely(nskb)) 2104 consume_skb(skb); 2105 else 2106 kfree_skb(skb); 2107 skb = nskb; 2108 } 2109 return skb; 2110 } 2111 2112 /** 2113 * skb_peek - peek at the head of an &sk_buff_head 2114 * @list_: list to peek at 2115 * 2116 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2117 * be careful with this one. A peek leaves the buffer on the 2118 * list and someone else may run off with it. You must hold 2119 * the appropriate locks or have a private queue to do this. 2120 * 2121 * Returns %NULL for an empty list or a pointer to the head element. 2122 * The reference count is not incremented and the reference is therefore 2123 * volatile. Use with caution. 2124 */ 2125 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2126 { 2127 struct sk_buff *skb = list_->next; 2128 2129 if (skb == (struct sk_buff *)list_) 2130 skb = NULL; 2131 return skb; 2132 } 2133 2134 /** 2135 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2136 * @list_: list to peek at 2137 * 2138 * Like skb_peek(), but the caller knows that the list is not empty. 2139 */ 2140 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2141 { 2142 return list_->next; 2143 } 2144 2145 /** 2146 * skb_peek_next - peek skb following the given one from a queue 2147 * @skb: skb to start from 2148 * @list_: list to peek at 2149 * 2150 * Returns %NULL when the end of the list is met or a pointer to the 2151 * next element. The reference count is not incremented and the 2152 * reference is therefore volatile. Use with caution. 2153 */ 2154 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2155 const struct sk_buff_head *list_) 2156 { 2157 struct sk_buff *next = skb->next; 2158 2159 if (next == (struct sk_buff *)list_) 2160 next = NULL; 2161 return next; 2162 } 2163 2164 /** 2165 * skb_peek_tail - peek at the tail of an &sk_buff_head 2166 * @list_: list to peek at 2167 * 2168 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2169 * be careful with this one. A peek leaves the buffer on the 2170 * list and someone else may run off with it. You must hold 2171 * the appropriate locks or have a private queue to do this. 2172 * 2173 * Returns %NULL for an empty list or a pointer to the tail element. 2174 * The reference count is not incremented and the reference is therefore 2175 * volatile. Use with caution. 2176 */ 2177 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2178 { 2179 struct sk_buff *skb = READ_ONCE(list_->prev); 2180 2181 if (skb == (struct sk_buff *)list_) 2182 skb = NULL; 2183 return skb; 2184 2185 } 2186 2187 /** 2188 * skb_queue_len - get queue length 2189 * @list_: list to measure 2190 * 2191 * Return the length of an &sk_buff queue. 2192 */ 2193 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2194 { 2195 return list_->qlen; 2196 } 2197 2198 /** 2199 * skb_queue_len_lockless - get queue length 2200 * @list_: list to measure 2201 * 2202 * Return the length of an &sk_buff queue. 2203 * This variant can be used in lockless contexts. 2204 */ 2205 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2206 { 2207 return READ_ONCE(list_->qlen); 2208 } 2209 2210 /** 2211 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2212 * @list: queue to initialize 2213 * 2214 * This initializes only the list and queue length aspects of 2215 * an sk_buff_head object. This allows to initialize the list 2216 * aspects of an sk_buff_head without reinitializing things like 2217 * the spinlock. It can also be used for on-stack sk_buff_head 2218 * objects where the spinlock is known to not be used. 2219 */ 2220 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2221 { 2222 list->prev = list->next = (struct sk_buff *)list; 2223 list->qlen = 0; 2224 } 2225 2226 /* 2227 * This function creates a split out lock class for each invocation; 2228 * this is needed for now since a whole lot of users of the skb-queue 2229 * infrastructure in drivers have different locking usage (in hardirq) 2230 * than the networking core (in softirq only). In the long run either the 2231 * network layer or drivers should need annotation to consolidate the 2232 * main types of usage into 3 classes. 2233 */ 2234 static inline void skb_queue_head_init(struct sk_buff_head *list) 2235 { 2236 spin_lock_init(&list->lock); 2237 __skb_queue_head_init(list); 2238 } 2239 2240 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2241 struct lock_class_key *class) 2242 { 2243 skb_queue_head_init(list); 2244 lockdep_set_class(&list->lock, class); 2245 } 2246 2247 /* 2248 * Insert an sk_buff on a list. 2249 * 2250 * The "__skb_xxxx()" functions are the non-atomic ones that 2251 * can only be called with interrupts disabled. 2252 */ 2253 static inline void __skb_insert(struct sk_buff *newsk, 2254 struct sk_buff *prev, struct sk_buff *next, 2255 struct sk_buff_head *list) 2256 { 2257 /* See skb_queue_empty_lockless() and skb_peek_tail() 2258 * for the opposite READ_ONCE() 2259 */ 2260 WRITE_ONCE(newsk->next, next); 2261 WRITE_ONCE(newsk->prev, prev); 2262 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2263 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2264 WRITE_ONCE(list->qlen, list->qlen + 1); 2265 } 2266 2267 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2268 struct sk_buff *prev, 2269 struct sk_buff *next) 2270 { 2271 struct sk_buff *first = list->next; 2272 struct sk_buff *last = list->prev; 2273 2274 WRITE_ONCE(first->prev, prev); 2275 WRITE_ONCE(prev->next, first); 2276 2277 WRITE_ONCE(last->next, next); 2278 WRITE_ONCE(next->prev, last); 2279 } 2280 2281 /** 2282 * skb_queue_splice - join two skb lists, this is designed for stacks 2283 * @list: the new list to add 2284 * @head: the place to add it in the first list 2285 */ 2286 static inline void skb_queue_splice(const struct sk_buff_head *list, 2287 struct sk_buff_head *head) 2288 { 2289 if (!skb_queue_empty(list)) { 2290 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2291 head->qlen += list->qlen; 2292 } 2293 } 2294 2295 /** 2296 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2297 * @list: the new list to add 2298 * @head: the place to add it in the first list 2299 * 2300 * The list at @list is reinitialised 2301 */ 2302 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2303 struct sk_buff_head *head) 2304 { 2305 if (!skb_queue_empty(list)) { 2306 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2307 head->qlen += list->qlen; 2308 __skb_queue_head_init(list); 2309 } 2310 } 2311 2312 /** 2313 * skb_queue_splice_tail - join two skb lists, each list being a queue 2314 * @list: the new list to add 2315 * @head: the place to add it in the first list 2316 */ 2317 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2318 struct sk_buff_head *head) 2319 { 2320 if (!skb_queue_empty(list)) { 2321 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2322 head->qlen += list->qlen; 2323 } 2324 } 2325 2326 /** 2327 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2328 * @list: the new list to add 2329 * @head: the place to add it in the first list 2330 * 2331 * Each of the lists is a queue. 2332 * The list at @list is reinitialised 2333 */ 2334 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2335 struct sk_buff_head *head) 2336 { 2337 if (!skb_queue_empty(list)) { 2338 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2339 head->qlen += list->qlen; 2340 __skb_queue_head_init(list); 2341 } 2342 } 2343 2344 /** 2345 * __skb_queue_after - queue a buffer at the list head 2346 * @list: list to use 2347 * @prev: place after this buffer 2348 * @newsk: buffer to queue 2349 * 2350 * Queue a buffer int the middle of a list. This function takes no locks 2351 * and you must therefore hold required locks before calling it. 2352 * 2353 * A buffer cannot be placed on two lists at the same time. 2354 */ 2355 static inline void __skb_queue_after(struct sk_buff_head *list, 2356 struct sk_buff *prev, 2357 struct sk_buff *newsk) 2358 { 2359 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2360 } 2361 2362 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2363 struct sk_buff_head *list); 2364 2365 static inline void __skb_queue_before(struct sk_buff_head *list, 2366 struct sk_buff *next, 2367 struct sk_buff *newsk) 2368 { 2369 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2370 } 2371 2372 /** 2373 * __skb_queue_head - queue a buffer at the list head 2374 * @list: list to use 2375 * @newsk: buffer to queue 2376 * 2377 * Queue a buffer at the start of a list. This function takes no locks 2378 * and you must therefore hold required locks before calling it. 2379 * 2380 * A buffer cannot be placed on two lists at the same time. 2381 */ 2382 static inline void __skb_queue_head(struct sk_buff_head *list, 2383 struct sk_buff *newsk) 2384 { 2385 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2386 } 2387 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2388 2389 /** 2390 * __skb_queue_tail - queue a buffer at the list tail 2391 * @list: list to use 2392 * @newsk: buffer to queue 2393 * 2394 * Queue a buffer at the end of a list. This function takes no locks 2395 * and you must therefore hold required locks before calling it. 2396 * 2397 * A buffer cannot be placed on two lists at the same time. 2398 */ 2399 static inline void __skb_queue_tail(struct sk_buff_head *list, 2400 struct sk_buff *newsk) 2401 { 2402 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2403 } 2404 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2405 2406 /* 2407 * remove sk_buff from list. _Must_ be called atomically, and with 2408 * the list known.. 2409 */ 2410 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2411 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2412 { 2413 struct sk_buff *next, *prev; 2414 2415 WRITE_ONCE(list->qlen, list->qlen - 1); 2416 next = skb->next; 2417 prev = skb->prev; 2418 skb->next = skb->prev = NULL; 2419 WRITE_ONCE(next->prev, prev); 2420 WRITE_ONCE(prev->next, next); 2421 } 2422 2423 /** 2424 * __skb_dequeue - remove from the head of the queue 2425 * @list: list to dequeue from 2426 * 2427 * Remove the head of the list. This function does not take any locks 2428 * so must be used with appropriate locks held only. The head item is 2429 * returned or %NULL if the list is empty. 2430 */ 2431 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2432 { 2433 struct sk_buff *skb = skb_peek(list); 2434 if (skb) 2435 __skb_unlink(skb, list); 2436 return skb; 2437 } 2438 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2439 2440 /** 2441 * __skb_dequeue_tail - remove from the tail of the queue 2442 * @list: list to dequeue from 2443 * 2444 * Remove the tail of the list. This function does not take any locks 2445 * so must be used with appropriate locks held only. The tail item is 2446 * returned or %NULL if the list is empty. 2447 */ 2448 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2449 { 2450 struct sk_buff *skb = skb_peek_tail(list); 2451 if (skb) 2452 __skb_unlink(skb, list); 2453 return skb; 2454 } 2455 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2456 2457 2458 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2459 { 2460 return skb->data_len; 2461 } 2462 2463 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2464 { 2465 return skb->len - skb->data_len; 2466 } 2467 2468 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2469 { 2470 unsigned int i, len = 0; 2471 2472 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2473 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2474 return len; 2475 } 2476 2477 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2478 { 2479 return skb_headlen(skb) + __skb_pagelen(skb); 2480 } 2481 2482 static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag, 2483 netmem_ref netmem, int off, 2484 int size) 2485 { 2486 frag->netmem = netmem; 2487 frag->offset = off; 2488 skb_frag_size_set(frag, size); 2489 } 2490 2491 static inline void skb_frag_fill_page_desc(skb_frag_t *frag, 2492 struct page *page, 2493 int off, int size) 2494 { 2495 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size); 2496 } 2497 2498 static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo, 2499 int i, netmem_ref netmem, 2500 int off, int size) 2501 { 2502 skb_frag_t *frag = &shinfo->frags[i]; 2503 2504 skb_frag_fill_netmem_desc(frag, netmem, off, size); 2505 } 2506 2507 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2508 int i, struct page *page, 2509 int off, int size) 2510 { 2511 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off, 2512 size); 2513 } 2514 2515 /** 2516 * skb_len_add - adds a number to len fields of skb 2517 * @skb: buffer to add len to 2518 * @delta: number of bytes to add 2519 */ 2520 static inline void skb_len_add(struct sk_buff *skb, int delta) 2521 { 2522 skb->len += delta; 2523 skb->data_len += delta; 2524 skb->truesize += delta; 2525 } 2526 2527 /** 2528 * __skb_fill_netmem_desc - initialise a fragment in an skb 2529 * @skb: buffer containing fragment to be initialised 2530 * @i: fragment index to initialise 2531 * @netmem: the netmem to use for this fragment 2532 * @off: the offset to the data with @page 2533 * @size: the length of the data 2534 * 2535 * Initialises the @i'th fragment of @skb to point to &size bytes at 2536 * offset @off within @page. 2537 * 2538 * Does not take any additional reference on the fragment. 2539 */ 2540 static inline void __skb_fill_netmem_desc(struct sk_buff *skb, int i, 2541 netmem_ref netmem, int off, int size) 2542 { 2543 struct page *page = netmem_to_page(netmem); 2544 2545 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size); 2546 2547 /* Propagate page pfmemalloc to the skb if we can. The problem is 2548 * that not all callers have unique ownership of the page but rely 2549 * on page_is_pfmemalloc doing the right thing(tm). 2550 */ 2551 page = compound_head(page); 2552 if (page_is_pfmemalloc(page)) 2553 skb->pfmemalloc = true; 2554 } 2555 2556 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2557 struct page *page, int off, int size) 2558 { 2559 __skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2560 } 2561 2562 static inline void skb_fill_netmem_desc(struct sk_buff *skb, int i, 2563 netmem_ref netmem, int off, int size) 2564 { 2565 __skb_fill_netmem_desc(skb, i, netmem, off, size); 2566 skb_shinfo(skb)->nr_frags = i + 1; 2567 } 2568 2569 /** 2570 * skb_fill_page_desc - initialise a paged fragment in an skb 2571 * @skb: buffer containing fragment to be initialised 2572 * @i: paged fragment index to initialise 2573 * @page: the page to use for this fragment 2574 * @off: the offset to the data with @page 2575 * @size: the length of the data 2576 * 2577 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2578 * @skb to point to @size bytes at offset @off within @page. In 2579 * addition updates @skb such that @i is the last fragment. 2580 * 2581 * Does not take any additional reference on the fragment. 2582 */ 2583 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2584 struct page *page, int off, int size) 2585 { 2586 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2587 } 2588 2589 /** 2590 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2591 * @skb: buffer containing fragment to be initialised 2592 * @i: paged fragment index to initialise 2593 * @page: the page to use for this fragment 2594 * @off: the offset to the data with @page 2595 * @size: the length of the data 2596 * 2597 * Variant of skb_fill_page_desc() which does not deal with 2598 * pfmemalloc, if page is not owned by us. 2599 */ 2600 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2601 struct page *page, int off, 2602 int size) 2603 { 2604 struct skb_shared_info *shinfo = skb_shinfo(skb); 2605 2606 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2607 shinfo->nr_frags = i + 1; 2608 } 2609 2610 void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem, 2611 int off, int size, unsigned int truesize); 2612 2613 static inline void skb_add_rx_frag(struct sk_buff *skb, int i, 2614 struct page *page, int off, int size, 2615 unsigned int truesize) 2616 { 2617 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size, 2618 truesize); 2619 } 2620 2621 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2622 unsigned int truesize); 2623 2624 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2625 2626 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2627 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2628 { 2629 return skb->head + skb->tail; 2630 } 2631 2632 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2633 { 2634 skb->tail = skb->data - skb->head; 2635 } 2636 2637 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2638 { 2639 skb_reset_tail_pointer(skb); 2640 skb->tail += offset; 2641 } 2642 2643 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2644 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2645 { 2646 return skb->tail; 2647 } 2648 2649 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2650 { 2651 skb->tail = skb->data; 2652 } 2653 2654 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2655 { 2656 skb->tail = skb->data + offset; 2657 } 2658 2659 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2660 2661 static inline void skb_assert_len(struct sk_buff *skb) 2662 { 2663 #ifdef CONFIG_DEBUG_NET 2664 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2665 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2666 #endif /* CONFIG_DEBUG_NET */ 2667 } 2668 2669 /* 2670 * Add data to an sk_buff 2671 */ 2672 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2673 void *skb_put(struct sk_buff *skb, unsigned int len); 2674 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2675 { 2676 void *tmp = skb_tail_pointer(skb); 2677 SKB_LINEAR_ASSERT(skb); 2678 skb->tail += len; 2679 skb->len += len; 2680 return tmp; 2681 } 2682 2683 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2684 { 2685 void *tmp = __skb_put(skb, len); 2686 2687 memset(tmp, 0, len); 2688 return tmp; 2689 } 2690 2691 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2692 unsigned int len) 2693 { 2694 void *tmp = __skb_put(skb, len); 2695 2696 memcpy(tmp, data, len); 2697 return tmp; 2698 } 2699 2700 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2701 { 2702 *(u8 *)__skb_put(skb, 1) = val; 2703 } 2704 2705 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2706 { 2707 void *tmp = skb_put(skb, len); 2708 2709 memset(tmp, 0, len); 2710 2711 return tmp; 2712 } 2713 2714 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2715 unsigned int len) 2716 { 2717 void *tmp = skb_put(skb, len); 2718 2719 memcpy(tmp, data, len); 2720 2721 return tmp; 2722 } 2723 2724 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2725 { 2726 *(u8 *)skb_put(skb, 1) = val; 2727 } 2728 2729 void *skb_push(struct sk_buff *skb, unsigned int len); 2730 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2731 { 2732 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2733 2734 skb->data -= len; 2735 skb->len += len; 2736 return skb->data; 2737 } 2738 2739 void *skb_pull(struct sk_buff *skb, unsigned int len); 2740 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2741 { 2742 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2743 2744 skb->len -= len; 2745 if (unlikely(skb->len < skb->data_len)) { 2746 #if defined(CONFIG_DEBUG_NET) 2747 skb->len += len; 2748 pr_err("__skb_pull(len=%u)\n", len); 2749 skb_dump(KERN_ERR, skb, false); 2750 #endif 2751 BUG(); 2752 } 2753 return skb->data += len; 2754 } 2755 2756 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2757 { 2758 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2759 } 2760 2761 void *skb_pull_data(struct sk_buff *skb, size_t len); 2762 2763 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2764 2765 static inline enum skb_drop_reason 2766 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len) 2767 { 2768 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2769 2770 if (likely(len <= skb_headlen(skb))) 2771 return SKB_NOT_DROPPED_YET; 2772 2773 if (unlikely(len > skb->len)) 2774 return SKB_DROP_REASON_PKT_TOO_SMALL; 2775 2776 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb)))) 2777 return SKB_DROP_REASON_NOMEM; 2778 2779 return SKB_NOT_DROPPED_YET; 2780 } 2781 2782 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2783 { 2784 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 2785 } 2786 2787 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2788 { 2789 if (!pskb_may_pull(skb, len)) 2790 return NULL; 2791 2792 skb->len -= len; 2793 return skb->data += len; 2794 } 2795 2796 void skb_condense(struct sk_buff *skb); 2797 2798 /** 2799 * skb_headroom - bytes at buffer head 2800 * @skb: buffer to check 2801 * 2802 * Return the number of bytes of free space at the head of an &sk_buff. 2803 */ 2804 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2805 { 2806 return skb->data - skb->head; 2807 } 2808 2809 /** 2810 * skb_tailroom - bytes at buffer end 2811 * @skb: buffer to check 2812 * 2813 * Return the number of bytes of free space at the tail of an sk_buff 2814 */ 2815 static inline int skb_tailroom(const struct sk_buff *skb) 2816 { 2817 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2818 } 2819 2820 /** 2821 * skb_availroom - bytes at buffer end 2822 * @skb: buffer to check 2823 * 2824 * Return the number of bytes of free space at the tail of an sk_buff 2825 * allocated by sk_stream_alloc() 2826 */ 2827 static inline int skb_availroom(const struct sk_buff *skb) 2828 { 2829 if (skb_is_nonlinear(skb)) 2830 return 0; 2831 2832 return skb->end - skb->tail - skb->reserved_tailroom; 2833 } 2834 2835 /** 2836 * skb_reserve - adjust headroom 2837 * @skb: buffer to alter 2838 * @len: bytes to move 2839 * 2840 * Increase the headroom of an empty &sk_buff by reducing the tail 2841 * room. This is only allowed for an empty buffer. 2842 */ 2843 static inline void skb_reserve(struct sk_buff *skb, int len) 2844 { 2845 skb->data += len; 2846 skb->tail += len; 2847 } 2848 2849 /** 2850 * skb_tailroom_reserve - adjust reserved_tailroom 2851 * @skb: buffer to alter 2852 * @mtu: maximum amount of headlen permitted 2853 * @needed_tailroom: minimum amount of reserved_tailroom 2854 * 2855 * Set reserved_tailroom so that headlen can be as large as possible but 2856 * not larger than mtu and tailroom cannot be smaller than 2857 * needed_tailroom. 2858 * The required headroom should already have been reserved before using 2859 * this function. 2860 */ 2861 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2862 unsigned int needed_tailroom) 2863 { 2864 SKB_LINEAR_ASSERT(skb); 2865 if (mtu < skb_tailroom(skb) - needed_tailroom) 2866 /* use at most mtu */ 2867 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2868 else 2869 /* use up to all available space */ 2870 skb->reserved_tailroom = needed_tailroom; 2871 } 2872 2873 #define ENCAP_TYPE_ETHER 0 2874 #define ENCAP_TYPE_IPPROTO 1 2875 2876 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2877 __be16 protocol) 2878 { 2879 skb->inner_protocol = protocol; 2880 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2881 } 2882 2883 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2884 __u8 ipproto) 2885 { 2886 skb->inner_ipproto = ipproto; 2887 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2888 } 2889 2890 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2891 { 2892 skb->inner_mac_header = skb->mac_header; 2893 skb->inner_network_header = skb->network_header; 2894 skb->inner_transport_header = skb->transport_header; 2895 } 2896 2897 static inline void skb_reset_mac_len(struct sk_buff *skb) 2898 { 2899 skb->mac_len = skb->network_header - skb->mac_header; 2900 } 2901 2902 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2903 *skb) 2904 { 2905 return skb->head + skb->inner_transport_header; 2906 } 2907 2908 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2909 { 2910 return skb_inner_transport_header(skb) - skb->data; 2911 } 2912 2913 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2914 { 2915 skb->inner_transport_header = skb->data - skb->head; 2916 } 2917 2918 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2919 const int offset) 2920 { 2921 skb_reset_inner_transport_header(skb); 2922 skb->inner_transport_header += offset; 2923 } 2924 2925 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2926 { 2927 return skb->head + skb->inner_network_header; 2928 } 2929 2930 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2931 { 2932 skb->inner_network_header = skb->data - skb->head; 2933 } 2934 2935 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2936 const int offset) 2937 { 2938 skb_reset_inner_network_header(skb); 2939 skb->inner_network_header += offset; 2940 } 2941 2942 static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb) 2943 { 2944 return skb->inner_network_header > 0; 2945 } 2946 2947 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2948 { 2949 return skb->head + skb->inner_mac_header; 2950 } 2951 2952 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2953 { 2954 skb->inner_mac_header = skb->data - skb->head; 2955 } 2956 2957 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2958 const int offset) 2959 { 2960 skb_reset_inner_mac_header(skb); 2961 skb->inner_mac_header += offset; 2962 } 2963 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2964 { 2965 return skb->transport_header != (typeof(skb->transport_header))~0U; 2966 } 2967 2968 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2969 { 2970 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 2971 return skb->head + skb->transport_header; 2972 } 2973 2974 static inline void skb_reset_transport_header(struct sk_buff *skb) 2975 { 2976 skb->transport_header = skb->data - skb->head; 2977 } 2978 2979 static inline void skb_set_transport_header(struct sk_buff *skb, 2980 const int offset) 2981 { 2982 skb_reset_transport_header(skb); 2983 skb->transport_header += offset; 2984 } 2985 2986 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2987 { 2988 return skb->head + skb->network_header; 2989 } 2990 2991 static inline void skb_reset_network_header(struct sk_buff *skb) 2992 { 2993 skb->network_header = skb->data - skb->head; 2994 } 2995 2996 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2997 { 2998 skb_reset_network_header(skb); 2999 skb->network_header += offset; 3000 } 3001 3002 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 3003 { 3004 return skb->mac_header != (typeof(skb->mac_header))~0U; 3005 } 3006 3007 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 3008 { 3009 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3010 return skb->head + skb->mac_header; 3011 } 3012 3013 static inline int skb_mac_offset(const struct sk_buff *skb) 3014 { 3015 return skb_mac_header(skb) - skb->data; 3016 } 3017 3018 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 3019 { 3020 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3021 return skb->network_header - skb->mac_header; 3022 } 3023 3024 static inline void skb_unset_mac_header(struct sk_buff *skb) 3025 { 3026 skb->mac_header = (typeof(skb->mac_header))~0U; 3027 } 3028 3029 static inline void skb_reset_mac_header(struct sk_buff *skb) 3030 { 3031 skb->mac_header = skb->data - skb->head; 3032 } 3033 3034 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 3035 { 3036 skb_reset_mac_header(skb); 3037 skb->mac_header += offset; 3038 } 3039 3040 static inline void skb_pop_mac_header(struct sk_buff *skb) 3041 { 3042 skb->mac_header = skb->network_header; 3043 } 3044 3045 static inline void skb_probe_transport_header(struct sk_buff *skb) 3046 { 3047 struct flow_keys_basic keys; 3048 3049 if (skb_transport_header_was_set(skb)) 3050 return; 3051 3052 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 3053 NULL, 0, 0, 0, 0)) 3054 skb_set_transport_header(skb, keys.control.thoff); 3055 } 3056 3057 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 3058 { 3059 if (skb_mac_header_was_set(skb)) { 3060 const unsigned char *old_mac = skb_mac_header(skb); 3061 3062 skb_set_mac_header(skb, -skb->mac_len); 3063 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 3064 } 3065 } 3066 3067 /* Move the full mac header up to current network_header. 3068 * Leaves skb->data pointing at offset skb->mac_len into the mac_header. 3069 * Must be provided the complete mac header length. 3070 */ 3071 static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len) 3072 { 3073 if (skb_mac_header_was_set(skb)) { 3074 const unsigned char *old_mac = skb_mac_header(skb); 3075 3076 skb_set_mac_header(skb, -full_mac_len); 3077 memmove(skb_mac_header(skb), old_mac, full_mac_len); 3078 __skb_push(skb, full_mac_len - skb->mac_len); 3079 } 3080 } 3081 3082 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 3083 { 3084 return skb->csum_start - skb_headroom(skb); 3085 } 3086 3087 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 3088 { 3089 return skb->head + skb->csum_start; 3090 } 3091 3092 static inline int skb_transport_offset(const struct sk_buff *skb) 3093 { 3094 return skb_transport_header(skb) - skb->data; 3095 } 3096 3097 static inline u32 skb_network_header_len(const struct sk_buff *skb) 3098 { 3099 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3100 return skb->transport_header - skb->network_header; 3101 } 3102 3103 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 3104 { 3105 return skb->inner_transport_header - skb->inner_network_header; 3106 } 3107 3108 static inline int skb_network_offset(const struct sk_buff *skb) 3109 { 3110 return skb_network_header(skb) - skb->data; 3111 } 3112 3113 static inline int skb_inner_network_offset(const struct sk_buff *skb) 3114 { 3115 return skb_inner_network_header(skb) - skb->data; 3116 } 3117 3118 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 3119 { 3120 return pskb_may_pull(skb, skb_network_offset(skb) + len); 3121 } 3122 3123 /* 3124 * CPUs often take a performance hit when accessing unaligned memory 3125 * locations. The actual performance hit varies, it can be small if the 3126 * hardware handles it or large if we have to take an exception and fix it 3127 * in software. 3128 * 3129 * Since an ethernet header is 14 bytes network drivers often end up with 3130 * the IP header at an unaligned offset. The IP header can be aligned by 3131 * shifting the start of the packet by 2 bytes. Drivers should do this 3132 * with: 3133 * 3134 * skb_reserve(skb, NET_IP_ALIGN); 3135 * 3136 * The downside to this alignment of the IP header is that the DMA is now 3137 * unaligned. On some architectures the cost of an unaligned DMA is high 3138 * and this cost outweighs the gains made by aligning the IP header. 3139 * 3140 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 3141 * to be overridden. 3142 */ 3143 #ifndef NET_IP_ALIGN 3144 #define NET_IP_ALIGN 2 3145 #endif 3146 3147 /* 3148 * The networking layer reserves some headroom in skb data (via 3149 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 3150 * the header has to grow. In the default case, if the header has to grow 3151 * 32 bytes or less we avoid the reallocation. 3152 * 3153 * Unfortunately this headroom changes the DMA alignment of the resulting 3154 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 3155 * on some architectures. An architecture can override this value, 3156 * perhaps setting it to a cacheline in size (since that will maintain 3157 * cacheline alignment of the DMA). It must be a power of 2. 3158 * 3159 * Various parts of the networking layer expect at least 32 bytes of 3160 * headroom, you should not reduce this. 3161 * 3162 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 3163 * to reduce average number of cache lines per packet. 3164 * get_rps_cpu() for example only access one 64 bytes aligned block : 3165 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 3166 */ 3167 #ifndef NET_SKB_PAD 3168 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 3169 #endif 3170 3171 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 3172 3173 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3174 { 3175 if (WARN_ON(skb_is_nonlinear(skb))) 3176 return; 3177 skb->len = len; 3178 skb_set_tail_pointer(skb, len); 3179 } 3180 3181 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3182 { 3183 __skb_set_length(skb, len); 3184 } 3185 3186 void skb_trim(struct sk_buff *skb, unsigned int len); 3187 3188 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3189 { 3190 if (skb->data_len) 3191 return ___pskb_trim(skb, len); 3192 __skb_trim(skb, len); 3193 return 0; 3194 } 3195 3196 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3197 { 3198 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3199 } 3200 3201 /** 3202 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3203 * @skb: buffer to alter 3204 * @len: new length 3205 * 3206 * This is identical to pskb_trim except that the caller knows that 3207 * the skb is not cloned so we should never get an error due to out- 3208 * of-memory. 3209 */ 3210 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3211 { 3212 int err = pskb_trim(skb, len); 3213 BUG_ON(err); 3214 } 3215 3216 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3217 { 3218 unsigned int diff = len - skb->len; 3219 3220 if (skb_tailroom(skb) < diff) { 3221 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3222 GFP_ATOMIC); 3223 if (ret) 3224 return ret; 3225 } 3226 __skb_set_length(skb, len); 3227 return 0; 3228 } 3229 3230 /** 3231 * skb_orphan - orphan a buffer 3232 * @skb: buffer to orphan 3233 * 3234 * If a buffer currently has an owner then we call the owner's 3235 * destructor function and make the @skb unowned. The buffer continues 3236 * to exist but is no longer charged to its former owner. 3237 */ 3238 static inline void skb_orphan(struct sk_buff *skb) 3239 { 3240 if (skb->destructor) { 3241 skb->destructor(skb); 3242 skb->destructor = NULL; 3243 skb->sk = NULL; 3244 } else { 3245 BUG_ON(skb->sk); 3246 } 3247 } 3248 3249 /** 3250 * skb_orphan_frags - orphan the frags contained in a buffer 3251 * @skb: buffer to orphan frags from 3252 * @gfp_mask: allocation mask for replacement pages 3253 * 3254 * For each frag in the SKB which needs a destructor (i.e. has an 3255 * owner) create a copy of that frag and release the original 3256 * page by calling the destructor. 3257 */ 3258 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3259 { 3260 if (likely(!skb_zcopy(skb))) 3261 return 0; 3262 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3263 return 0; 3264 return skb_copy_ubufs(skb, gfp_mask); 3265 } 3266 3267 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3268 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3269 { 3270 if (likely(!skb_zcopy(skb))) 3271 return 0; 3272 return skb_copy_ubufs(skb, gfp_mask); 3273 } 3274 3275 /** 3276 * __skb_queue_purge_reason - empty a list 3277 * @list: list to empty 3278 * @reason: drop reason 3279 * 3280 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3281 * the list and one reference dropped. This function does not take the 3282 * list lock and the caller must hold the relevant locks to use it. 3283 */ 3284 static inline void __skb_queue_purge_reason(struct sk_buff_head *list, 3285 enum skb_drop_reason reason) 3286 { 3287 struct sk_buff *skb; 3288 3289 while ((skb = __skb_dequeue(list)) != NULL) 3290 kfree_skb_reason(skb, reason); 3291 } 3292 3293 static inline void __skb_queue_purge(struct sk_buff_head *list) 3294 { 3295 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3296 } 3297 3298 void skb_queue_purge_reason(struct sk_buff_head *list, 3299 enum skb_drop_reason reason); 3300 3301 static inline void skb_queue_purge(struct sk_buff_head *list) 3302 { 3303 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3304 } 3305 3306 unsigned int skb_rbtree_purge(struct rb_root *root); 3307 void skb_errqueue_purge(struct sk_buff_head *list); 3308 3309 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3310 3311 /** 3312 * netdev_alloc_frag - allocate a page fragment 3313 * @fragsz: fragment size 3314 * 3315 * Allocates a frag from a page for receive buffer. 3316 * Uses GFP_ATOMIC allocations. 3317 */ 3318 static inline void *netdev_alloc_frag(unsigned int fragsz) 3319 { 3320 return __netdev_alloc_frag_align(fragsz, ~0u); 3321 } 3322 3323 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3324 unsigned int align) 3325 { 3326 WARN_ON_ONCE(!is_power_of_2(align)); 3327 return __netdev_alloc_frag_align(fragsz, -align); 3328 } 3329 3330 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3331 gfp_t gfp_mask); 3332 3333 /** 3334 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3335 * @dev: network device to receive on 3336 * @length: length to allocate 3337 * 3338 * Allocate a new &sk_buff and assign it a usage count of one. The 3339 * buffer has unspecified headroom built in. Users should allocate 3340 * the headroom they think they need without accounting for the 3341 * built in space. The built in space is used for optimisations. 3342 * 3343 * %NULL is returned if there is no free memory. Although this function 3344 * allocates memory it can be called from an interrupt. 3345 */ 3346 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3347 unsigned int length) 3348 { 3349 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3350 } 3351 3352 /* legacy helper around __netdev_alloc_skb() */ 3353 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3354 gfp_t gfp_mask) 3355 { 3356 return __netdev_alloc_skb(NULL, length, gfp_mask); 3357 } 3358 3359 /* legacy helper around netdev_alloc_skb() */ 3360 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3361 { 3362 return netdev_alloc_skb(NULL, length); 3363 } 3364 3365 3366 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3367 unsigned int length, gfp_t gfp) 3368 { 3369 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3370 3371 if (NET_IP_ALIGN && skb) 3372 skb_reserve(skb, NET_IP_ALIGN); 3373 return skb; 3374 } 3375 3376 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3377 unsigned int length) 3378 { 3379 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3380 } 3381 3382 static inline void skb_free_frag(void *addr) 3383 { 3384 page_frag_free(addr); 3385 } 3386 3387 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3388 3389 static inline void *napi_alloc_frag(unsigned int fragsz) 3390 { 3391 return __napi_alloc_frag_align(fragsz, ~0u); 3392 } 3393 3394 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3395 unsigned int align) 3396 { 3397 WARN_ON_ONCE(!is_power_of_2(align)); 3398 return __napi_alloc_frag_align(fragsz, -align); 3399 } 3400 3401 struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int length); 3402 void napi_consume_skb(struct sk_buff *skb, int budget); 3403 3404 void napi_skb_free_stolen_head(struct sk_buff *skb); 3405 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason); 3406 3407 /** 3408 * __dev_alloc_pages - allocate page for network Rx 3409 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3410 * @order: size of the allocation 3411 * 3412 * Allocate a new page. 3413 * 3414 * %NULL is returned if there is no free memory. 3415 */ 3416 static inline struct page *__dev_alloc_pages_noprof(gfp_t gfp_mask, 3417 unsigned int order) 3418 { 3419 /* This piece of code contains several assumptions. 3420 * 1. This is for device Rx, therefore a cold page is preferred. 3421 * 2. The expectation is the user wants a compound page. 3422 * 3. If requesting a order 0 page it will not be compound 3423 * due to the check to see if order has a value in prep_new_page 3424 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3425 * code in gfp_to_alloc_flags that should be enforcing this. 3426 */ 3427 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3428 3429 return alloc_pages_node_noprof(NUMA_NO_NODE, gfp_mask, order); 3430 } 3431 #define __dev_alloc_pages(...) alloc_hooks(__dev_alloc_pages_noprof(__VA_ARGS__)) 3432 3433 /* 3434 * This specialized allocator has to be a macro for its allocations to be 3435 * accounted separately (to have a separate alloc_tag). 3436 */ 3437 #define dev_alloc_pages(_order) __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, _order) 3438 3439 /** 3440 * __dev_alloc_page - allocate a page for network Rx 3441 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3442 * 3443 * Allocate a new page. 3444 * 3445 * %NULL is returned if there is no free memory. 3446 */ 3447 static inline struct page *__dev_alloc_page_noprof(gfp_t gfp_mask) 3448 { 3449 return __dev_alloc_pages_noprof(gfp_mask, 0); 3450 } 3451 #define __dev_alloc_page(...) alloc_hooks(__dev_alloc_page_noprof(__VA_ARGS__)) 3452 3453 /* 3454 * This specialized allocator has to be a macro for its allocations to be 3455 * accounted separately (to have a separate alloc_tag). 3456 */ 3457 #define dev_alloc_page() dev_alloc_pages(0) 3458 3459 /** 3460 * dev_page_is_reusable - check whether a page can be reused for network Rx 3461 * @page: the page to test 3462 * 3463 * A page shouldn't be considered for reusing/recycling if it was allocated 3464 * under memory pressure or at a distant memory node. 3465 * 3466 * Returns false if this page should be returned to page allocator, true 3467 * otherwise. 3468 */ 3469 static inline bool dev_page_is_reusable(const struct page *page) 3470 { 3471 return likely(page_to_nid(page) == numa_mem_id() && 3472 !page_is_pfmemalloc(page)); 3473 } 3474 3475 /** 3476 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3477 * @page: The page that was allocated from skb_alloc_page 3478 * @skb: The skb that may need pfmemalloc set 3479 */ 3480 static inline void skb_propagate_pfmemalloc(const struct page *page, 3481 struct sk_buff *skb) 3482 { 3483 if (page_is_pfmemalloc(page)) 3484 skb->pfmemalloc = true; 3485 } 3486 3487 /** 3488 * skb_frag_off() - Returns the offset of a skb fragment 3489 * @frag: the paged fragment 3490 */ 3491 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3492 { 3493 return frag->offset; 3494 } 3495 3496 /** 3497 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3498 * @frag: skb fragment 3499 * @delta: value to add 3500 */ 3501 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3502 { 3503 frag->offset += delta; 3504 } 3505 3506 /** 3507 * skb_frag_off_set() - Sets the offset of a skb fragment 3508 * @frag: skb fragment 3509 * @offset: offset of fragment 3510 */ 3511 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3512 { 3513 frag->offset = offset; 3514 } 3515 3516 /** 3517 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3518 * @fragto: skb fragment where offset is set 3519 * @fragfrom: skb fragment offset is copied from 3520 */ 3521 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3522 const skb_frag_t *fragfrom) 3523 { 3524 fragto->offset = fragfrom->offset; 3525 } 3526 3527 /** 3528 * skb_frag_page - retrieve the page referred to by a paged fragment 3529 * @frag: the paged fragment 3530 * 3531 * Returns the &struct page associated with @frag. 3532 */ 3533 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3534 { 3535 return netmem_to_page(frag->netmem); 3536 } 3537 3538 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb, 3539 unsigned int headroom); 3540 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb, 3541 struct bpf_prog *prog); 3542 /** 3543 * skb_frag_address - gets the address of the data contained in a paged fragment 3544 * @frag: the paged fragment buffer 3545 * 3546 * Returns the address of the data within @frag. The page must already 3547 * be mapped. 3548 */ 3549 static inline void *skb_frag_address(const skb_frag_t *frag) 3550 { 3551 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3552 } 3553 3554 /** 3555 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3556 * @frag: the paged fragment buffer 3557 * 3558 * Returns the address of the data within @frag. Checks that the page 3559 * is mapped and returns %NULL otherwise. 3560 */ 3561 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3562 { 3563 void *ptr = page_address(skb_frag_page(frag)); 3564 if (unlikely(!ptr)) 3565 return NULL; 3566 3567 return ptr + skb_frag_off(frag); 3568 } 3569 3570 /** 3571 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3572 * @fragto: skb fragment where page is set 3573 * @fragfrom: skb fragment page is copied from 3574 */ 3575 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3576 const skb_frag_t *fragfrom) 3577 { 3578 fragto->netmem = fragfrom->netmem; 3579 } 3580 3581 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3582 3583 /** 3584 * skb_frag_dma_map - maps a paged fragment via the DMA API 3585 * @dev: the device to map the fragment to 3586 * @frag: the paged fragment to map 3587 * @offset: the offset within the fragment (starting at the 3588 * fragment's own offset) 3589 * @size: the number of bytes to map 3590 * @dir: the direction of the mapping (``PCI_DMA_*``) 3591 * 3592 * Maps the page associated with @frag to @device. 3593 */ 3594 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 3595 const skb_frag_t *frag, 3596 size_t offset, size_t size, 3597 enum dma_data_direction dir) 3598 { 3599 return dma_map_page(dev, skb_frag_page(frag), 3600 skb_frag_off(frag) + offset, size, dir); 3601 } 3602 3603 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3604 gfp_t gfp_mask) 3605 { 3606 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3607 } 3608 3609 3610 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3611 gfp_t gfp_mask) 3612 { 3613 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3614 } 3615 3616 3617 /** 3618 * skb_clone_writable - is the header of a clone writable 3619 * @skb: buffer to check 3620 * @len: length up to which to write 3621 * 3622 * Returns true if modifying the header part of the cloned buffer 3623 * does not requires the data to be copied. 3624 */ 3625 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3626 { 3627 return !skb_header_cloned(skb) && 3628 skb_headroom(skb) + len <= skb->hdr_len; 3629 } 3630 3631 static inline int skb_try_make_writable(struct sk_buff *skb, 3632 unsigned int write_len) 3633 { 3634 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3635 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3636 } 3637 3638 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3639 int cloned) 3640 { 3641 int delta = 0; 3642 3643 if (headroom > skb_headroom(skb)) 3644 delta = headroom - skb_headroom(skb); 3645 3646 if (delta || cloned) 3647 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3648 GFP_ATOMIC); 3649 return 0; 3650 } 3651 3652 /** 3653 * skb_cow - copy header of skb when it is required 3654 * @skb: buffer to cow 3655 * @headroom: needed headroom 3656 * 3657 * If the skb passed lacks sufficient headroom or its data part 3658 * is shared, data is reallocated. If reallocation fails, an error 3659 * is returned and original skb is not changed. 3660 * 3661 * The result is skb with writable area skb->head...skb->tail 3662 * and at least @headroom of space at head. 3663 */ 3664 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3665 { 3666 return __skb_cow(skb, headroom, skb_cloned(skb)); 3667 } 3668 3669 /** 3670 * skb_cow_head - skb_cow but only making the head writable 3671 * @skb: buffer to cow 3672 * @headroom: needed headroom 3673 * 3674 * This function is identical to skb_cow except that we replace the 3675 * skb_cloned check by skb_header_cloned. It should be used when 3676 * you only need to push on some header and do not need to modify 3677 * the data. 3678 */ 3679 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3680 { 3681 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3682 } 3683 3684 /** 3685 * skb_padto - pad an skbuff up to a minimal size 3686 * @skb: buffer to pad 3687 * @len: minimal length 3688 * 3689 * Pads up a buffer to ensure the trailing bytes exist and are 3690 * blanked. If the buffer already contains sufficient data it 3691 * is untouched. Otherwise it is extended. Returns zero on 3692 * success. The skb is freed on error. 3693 */ 3694 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3695 { 3696 unsigned int size = skb->len; 3697 if (likely(size >= len)) 3698 return 0; 3699 return skb_pad(skb, len - size); 3700 } 3701 3702 /** 3703 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3704 * @skb: buffer to pad 3705 * @len: minimal length 3706 * @free_on_error: free buffer on error 3707 * 3708 * Pads up a buffer to ensure the trailing bytes exist and are 3709 * blanked. If the buffer already contains sufficient data it 3710 * is untouched. Otherwise it is extended. Returns zero on 3711 * success. The skb is freed on error if @free_on_error is true. 3712 */ 3713 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3714 unsigned int len, 3715 bool free_on_error) 3716 { 3717 unsigned int size = skb->len; 3718 3719 if (unlikely(size < len)) { 3720 len -= size; 3721 if (__skb_pad(skb, len, free_on_error)) 3722 return -ENOMEM; 3723 __skb_put(skb, len); 3724 } 3725 return 0; 3726 } 3727 3728 /** 3729 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3730 * @skb: buffer to pad 3731 * @len: minimal length 3732 * 3733 * Pads up a buffer to ensure the trailing bytes exist and are 3734 * blanked. If the buffer already contains sufficient data it 3735 * is untouched. Otherwise it is extended. Returns zero on 3736 * success. The skb is freed on error. 3737 */ 3738 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3739 { 3740 return __skb_put_padto(skb, len, true); 3741 } 3742 3743 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i) 3744 __must_check; 3745 3746 static inline int skb_add_data(struct sk_buff *skb, 3747 struct iov_iter *from, int copy) 3748 { 3749 const int off = skb->len; 3750 3751 if (skb->ip_summed == CHECKSUM_NONE) { 3752 __wsum csum = 0; 3753 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 3754 &csum, from)) { 3755 skb->csum = csum_block_add(skb->csum, csum, off); 3756 return 0; 3757 } 3758 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3759 return 0; 3760 3761 __skb_trim(skb, off); 3762 return -EFAULT; 3763 } 3764 3765 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3766 const struct page *page, int off) 3767 { 3768 if (skb_zcopy(skb)) 3769 return false; 3770 if (i) { 3771 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3772 3773 return page == skb_frag_page(frag) && 3774 off == skb_frag_off(frag) + skb_frag_size(frag); 3775 } 3776 return false; 3777 } 3778 3779 static inline int __skb_linearize(struct sk_buff *skb) 3780 { 3781 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3782 } 3783 3784 /** 3785 * skb_linearize - convert paged skb to linear one 3786 * @skb: buffer to linarize 3787 * 3788 * If there is no free memory -ENOMEM is returned, otherwise zero 3789 * is returned and the old skb data released. 3790 */ 3791 static inline int skb_linearize(struct sk_buff *skb) 3792 { 3793 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3794 } 3795 3796 /** 3797 * skb_has_shared_frag - can any frag be overwritten 3798 * @skb: buffer to test 3799 * 3800 * Return true if the skb has at least one frag that might be modified 3801 * by an external entity (as in vmsplice()/sendfile()) 3802 */ 3803 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3804 { 3805 return skb_is_nonlinear(skb) && 3806 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 3807 } 3808 3809 /** 3810 * skb_linearize_cow - make sure skb is linear and writable 3811 * @skb: buffer to process 3812 * 3813 * If there is no free memory -ENOMEM is returned, otherwise zero 3814 * is returned and the old skb data released. 3815 */ 3816 static inline int skb_linearize_cow(struct sk_buff *skb) 3817 { 3818 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3819 __skb_linearize(skb) : 0; 3820 } 3821 3822 static __always_inline void 3823 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3824 unsigned int off) 3825 { 3826 if (skb->ip_summed == CHECKSUM_COMPLETE) 3827 skb->csum = csum_block_sub(skb->csum, 3828 csum_partial(start, len, 0), off); 3829 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3830 skb_checksum_start_offset(skb) < 0) 3831 skb->ip_summed = CHECKSUM_NONE; 3832 } 3833 3834 /** 3835 * skb_postpull_rcsum - update checksum for received skb after pull 3836 * @skb: buffer to update 3837 * @start: start of data before pull 3838 * @len: length of data pulled 3839 * 3840 * After doing a pull on a received packet, you need to call this to 3841 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3842 * CHECKSUM_NONE so that it can be recomputed from scratch. 3843 */ 3844 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3845 const void *start, unsigned int len) 3846 { 3847 if (skb->ip_summed == CHECKSUM_COMPLETE) 3848 skb->csum = wsum_negate(csum_partial(start, len, 3849 wsum_negate(skb->csum))); 3850 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3851 skb_checksum_start_offset(skb) < 0) 3852 skb->ip_summed = CHECKSUM_NONE; 3853 } 3854 3855 static __always_inline void 3856 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3857 unsigned int off) 3858 { 3859 if (skb->ip_summed == CHECKSUM_COMPLETE) 3860 skb->csum = csum_block_add(skb->csum, 3861 csum_partial(start, len, 0), off); 3862 } 3863 3864 /** 3865 * skb_postpush_rcsum - update checksum for received skb after push 3866 * @skb: buffer to update 3867 * @start: start of data after push 3868 * @len: length of data pushed 3869 * 3870 * After doing a push on a received packet, you need to call this to 3871 * update the CHECKSUM_COMPLETE checksum. 3872 */ 3873 static inline void skb_postpush_rcsum(struct sk_buff *skb, 3874 const void *start, unsigned int len) 3875 { 3876 __skb_postpush_rcsum(skb, start, len, 0); 3877 } 3878 3879 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3880 3881 /** 3882 * skb_push_rcsum - push skb and update receive checksum 3883 * @skb: buffer to update 3884 * @len: length of data pulled 3885 * 3886 * This function performs an skb_push on the packet and updates 3887 * the CHECKSUM_COMPLETE checksum. It should be used on 3888 * receive path processing instead of skb_push unless you know 3889 * that the checksum difference is zero (e.g., a valid IP header) 3890 * or you are setting ip_summed to CHECKSUM_NONE. 3891 */ 3892 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3893 { 3894 skb_push(skb, len); 3895 skb_postpush_rcsum(skb, skb->data, len); 3896 return skb->data; 3897 } 3898 3899 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 3900 /** 3901 * pskb_trim_rcsum - trim received skb and update checksum 3902 * @skb: buffer to trim 3903 * @len: new length 3904 * 3905 * This is exactly the same as pskb_trim except that it ensures the 3906 * checksum of received packets are still valid after the operation. 3907 * It can change skb pointers. 3908 */ 3909 3910 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3911 { 3912 if (likely(len >= skb->len)) 3913 return 0; 3914 return pskb_trim_rcsum_slow(skb, len); 3915 } 3916 3917 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3918 { 3919 if (skb->ip_summed == CHECKSUM_COMPLETE) 3920 skb->ip_summed = CHECKSUM_NONE; 3921 __skb_trim(skb, len); 3922 return 0; 3923 } 3924 3925 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3926 { 3927 if (skb->ip_summed == CHECKSUM_COMPLETE) 3928 skb->ip_summed = CHECKSUM_NONE; 3929 return __skb_grow(skb, len); 3930 } 3931 3932 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3933 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 3934 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 3935 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3936 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3937 3938 #define skb_queue_walk(queue, skb) \ 3939 for (skb = (queue)->next; \ 3940 skb != (struct sk_buff *)(queue); \ 3941 skb = skb->next) 3942 3943 #define skb_queue_walk_safe(queue, skb, tmp) \ 3944 for (skb = (queue)->next, tmp = skb->next; \ 3945 skb != (struct sk_buff *)(queue); \ 3946 skb = tmp, tmp = skb->next) 3947 3948 #define skb_queue_walk_from(queue, skb) \ 3949 for (; skb != (struct sk_buff *)(queue); \ 3950 skb = skb->next) 3951 3952 #define skb_rbtree_walk(skb, root) \ 3953 for (skb = skb_rb_first(root); skb != NULL; \ 3954 skb = skb_rb_next(skb)) 3955 3956 #define skb_rbtree_walk_from(skb) \ 3957 for (; skb != NULL; \ 3958 skb = skb_rb_next(skb)) 3959 3960 #define skb_rbtree_walk_from_safe(skb, tmp) \ 3961 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3962 skb = tmp) 3963 3964 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 3965 for (tmp = skb->next; \ 3966 skb != (struct sk_buff *)(queue); \ 3967 skb = tmp, tmp = skb->next) 3968 3969 #define skb_queue_reverse_walk(queue, skb) \ 3970 for (skb = (queue)->prev; \ 3971 skb != (struct sk_buff *)(queue); \ 3972 skb = skb->prev) 3973 3974 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3975 for (skb = (queue)->prev, tmp = skb->prev; \ 3976 skb != (struct sk_buff *)(queue); \ 3977 skb = tmp, tmp = skb->prev) 3978 3979 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3980 for (tmp = skb->prev; \ 3981 skb != (struct sk_buff *)(queue); \ 3982 skb = tmp, tmp = skb->prev) 3983 3984 static inline bool skb_has_frag_list(const struct sk_buff *skb) 3985 { 3986 return skb_shinfo(skb)->frag_list != NULL; 3987 } 3988 3989 static inline void skb_frag_list_init(struct sk_buff *skb) 3990 { 3991 skb_shinfo(skb)->frag_list = NULL; 3992 } 3993 3994 #define skb_walk_frags(skb, iter) \ 3995 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3996 3997 3998 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 3999 int *err, long *timeo_p, 4000 const struct sk_buff *skb); 4001 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 4002 struct sk_buff_head *queue, 4003 unsigned int flags, 4004 int *off, int *err, 4005 struct sk_buff **last); 4006 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 4007 struct sk_buff_head *queue, 4008 unsigned int flags, int *off, int *err, 4009 struct sk_buff **last); 4010 struct sk_buff *__skb_recv_datagram(struct sock *sk, 4011 struct sk_buff_head *sk_queue, 4012 unsigned int flags, int *off, int *err); 4013 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 4014 __poll_t datagram_poll(struct file *file, struct socket *sock, 4015 struct poll_table_struct *wait); 4016 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 4017 struct iov_iter *to, int size); 4018 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 4019 struct msghdr *msg, int size) 4020 { 4021 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 4022 } 4023 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 4024 struct msghdr *msg); 4025 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 4026 struct iov_iter *to, int len, 4027 struct ahash_request *hash); 4028 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 4029 struct iov_iter *from, int len); 4030 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 4031 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 4032 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 4033 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 4034 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 4035 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 4036 int len); 4037 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 4038 struct pipe_inode_info *pipe, unsigned int len, 4039 unsigned int flags); 4040 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 4041 int len); 4042 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 4043 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 4044 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 4045 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 4046 int len, int hlen); 4047 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 4048 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 4049 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 4050 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 4051 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 4052 unsigned int offset); 4053 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 4054 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 4055 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev); 4056 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 4057 int skb_vlan_pop(struct sk_buff *skb); 4058 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 4059 int skb_eth_pop(struct sk_buff *skb); 4060 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 4061 const unsigned char *src); 4062 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 4063 int mac_len, bool ethernet); 4064 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 4065 bool ethernet); 4066 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 4067 int skb_mpls_dec_ttl(struct sk_buff *skb); 4068 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 4069 gfp_t gfp); 4070 4071 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 4072 { 4073 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 4074 } 4075 4076 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 4077 { 4078 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 4079 } 4080 4081 struct skb_checksum_ops { 4082 __wsum (*update)(const void *mem, int len, __wsum wsum); 4083 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 4084 }; 4085 4086 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 4087 4088 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 4089 __wsum csum, const struct skb_checksum_ops *ops); 4090 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 4091 __wsum csum); 4092 4093 static inline void * __must_check 4094 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 4095 const void *data, int hlen, void *buffer) 4096 { 4097 if (likely(hlen - offset >= len)) 4098 return (void *)data + offset; 4099 4100 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 4101 return NULL; 4102 4103 return buffer; 4104 } 4105 4106 static inline void * __must_check 4107 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4108 { 4109 return __skb_header_pointer(skb, offset, len, skb->data, 4110 skb_headlen(skb), buffer); 4111 } 4112 4113 static inline void * __must_check 4114 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len) 4115 { 4116 if (likely(skb_headlen(skb) - offset >= len)) 4117 return skb->data + offset; 4118 return NULL; 4119 } 4120 4121 /** 4122 * skb_needs_linearize - check if we need to linearize a given skb 4123 * depending on the given device features. 4124 * @skb: socket buffer to check 4125 * @features: net device features 4126 * 4127 * Returns true if either: 4128 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4129 * 2. skb is fragmented and the device does not support SG. 4130 */ 4131 static inline bool skb_needs_linearize(struct sk_buff *skb, 4132 netdev_features_t features) 4133 { 4134 return skb_is_nonlinear(skb) && 4135 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4136 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4137 } 4138 4139 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4140 void *to, 4141 const unsigned int len) 4142 { 4143 memcpy(to, skb->data, len); 4144 } 4145 4146 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4147 const int offset, void *to, 4148 const unsigned int len) 4149 { 4150 memcpy(to, skb->data + offset, len); 4151 } 4152 4153 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4154 const void *from, 4155 const unsigned int len) 4156 { 4157 memcpy(skb->data, from, len); 4158 } 4159 4160 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4161 const int offset, 4162 const void *from, 4163 const unsigned int len) 4164 { 4165 memcpy(skb->data + offset, from, len); 4166 } 4167 4168 void skb_init(void); 4169 4170 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4171 { 4172 return skb->tstamp; 4173 } 4174 4175 /** 4176 * skb_get_timestamp - get timestamp from a skb 4177 * @skb: skb to get stamp from 4178 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4179 * 4180 * Timestamps are stored in the skb as offsets to a base timestamp. 4181 * This function converts the offset back to a struct timeval and stores 4182 * it in stamp. 4183 */ 4184 static inline void skb_get_timestamp(const struct sk_buff *skb, 4185 struct __kernel_old_timeval *stamp) 4186 { 4187 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4188 } 4189 4190 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4191 struct __kernel_sock_timeval *stamp) 4192 { 4193 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4194 4195 stamp->tv_sec = ts.tv_sec; 4196 stamp->tv_usec = ts.tv_nsec / 1000; 4197 } 4198 4199 static inline void skb_get_timestampns(const struct sk_buff *skb, 4200 struct __kernel_old_timespec *stamp) 4201 { 4202 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4203 4204 stamp->tv_sec = ts.tv_sec; 4205 stamp->tv_nsec = ts.tv_nsec; 4206 } 4207 4208 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4209 struct __kernel_timespec *stamp) 4210 { 4211 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4212 4213 stamp->tv_sec = ts.tv_sec; 4214 stamp->tv_nsec = ts.tv_nsec; 4215 } 4216 4217 static inline void __net_timestamp(struct sk_buff *skb) 4218 { 4219 skb->tstamp = ktime_get_real(); 4220 skb->tstamp_type = SKB_CLOCK_REALTIME; 4221 } 4222 4223 static inline ktime_t net_timedelta(ktime_t t) 4224 { 4225 return ktime_sub(ktime_get_real(), t); 4226 } 4227 4228 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4229 u8 tstamp_type) 4230 { 4231 skb->tstamp = kt; 4232 4233 if (kt) 4234 skb->tstamp_type = tstamp_type; 4235 else 4236 skb->tstamp_type = SKB_CLOCK_REALTIME; 4237 } 4238 4239 static inline void skb_set_delivery_type_by_clockid(struct sk_buff *skb, 4240 ktime_t kt, clockid_t clockid) 4241 { 4242 u8 tstamp_type = SKB_CLOCK_REALTIME; 4243 4244 switch (clockid) { 4245 case CLOCK_REALTIME: 4246 break; 4247 case CLOCK_MONOTONIC: 4248 tstamp_type = SKB_CLOCK_MONOTONIC; 4249 break; 4250 case CLOCK_TAI: 4251 tstamp_type = SKB_CLOCK_TAI; 4252 break; 4253 default: 4254 WARN_ON_ONCE(1); 4255 kt = 0; 4256 } 4257 4258 skb_set_delivery_time(skb, kt, tstamp_type); 4259 } 4260 4261 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4262 4263 /* It is used in the ingress path to clear the delivery_time. 4264 * If needed, set the skb->tstamp to the (rcv) timestamp. 4265 */ 4266 static inline void skb_clear_delivery_time(struct sk_buff *skb) 4267 { 4268 if (skb->tstamp_type) { 4269 skb->tstamp_type = SKB_CLOCK_REALTIME; 4270 if (static_branch_unlikely(&netstamp_needed_key)) 4271 skb->tstamp = ktime_get_real(); 4272 else 4273 skb->tstamp = 0; 4274 } 4275 } 4276 4277 static inline void skb_clear_tstamp(struct sk_buff *skb) 4278 { 4279 if (skb->tstamp_type) 4280 return; 4281 4282 skb->tstamp = 0; 4283 } 4284 4285 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4286 { 4287 if (skb->tstamp_type) 4288 return 0; 4289 4290 return skb->tstamp; 4291 } 4292 4293 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4294 { 4295 if (skb->tstamp_type != SKB_CLOCK_MONOTONIC && skb->tstamp) 4296 return skb->tstamp; 4297 4298 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4299 return ktime_get_real(); 4300 4301 return 0; 4302 } 4303 4304 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4305 { 4306 return skb_shinfo(skb)->meta_len; 4307 } 4308 4309 static inline void *skb_metadata_end(const struct sk_buff *skb) 4310 { 4311 return skb_mac_header(skb); 4312 } 4313 4314 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4315 const struct sk_buff *skb_b, 4316 u8 meta_len) 4317 { 4318 const void *a = skb_metadata_end(skb_a); 4319 const void *b = skb_metadata_end(skb_b); 4320 u64 diffs = 0; 4321 4322 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) || 4323 BITS_PER_LONG != 64) 4324 goto slow; 4325 4326 /* Using more efficient variant than plain call to memcmp(). */ 4327 switch (meta_len) { 4328 #define __it(x, op) (x -= sizeof(u##op)) 4329 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4330 case 32: diffs |= __it_diff(a, b, 64); 4331 fallthrough; 4332 case 24: diffs |= __it_diff(a, b, 64); 4333 fallthrough; 4334 case 16: diffs |= __it_diff(a, b, 64); 4335 fallthrough; 4336 case 8: diffs |= __it_diff(a, b, 64); 4337 break; 4338 case 28: diffs |= __it_diff(a, b, 64); 4339 fallthrough; 4340 case 20: diffs |= __it_diff(a, b, 64); 4341 fallthrough; 4342 case 12: diffs |= __it_diff(a, b, 64); 4343 fallthrough; 4344 case 4: diffs |= __it_diff(a, b, 32); 4345 break; 4346 default: 4347 slow: 4348 return memcmp(a - meta_len, b - meta_len, meta_len); 4349 } 4350 return diffs; 4351 } 4352 4353 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4354 const struct sk_buff *skb_b) 4355 { 4356 u8 len_a = skb_metadata_len(skb_a); 4357 u8 len_b = skb_metadata_len(skb_b); 4358 4359 if (!(len_a | len_b)) 4360 return false; 4361 4362 return len_a != len_b ? 4363 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4364 } 4365 4366 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4367 { 4368 skb_shinfo(skb)->meta_len = meta_len; 4369 } 4370 4371 static inline void skb_metadata_clear(struct sk_buff *skb) 4372 { 4373 skb_metadata_set(skb, 0); 4374 } 4375 4376 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4377 4378 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4379 4380 void skb_clone_tx_timestamp(struct sk_buff *skb); 4381 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4382 4383 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4384 4385 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4386 { 4387 } 4388 4389 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4390 { 4391 return false; 4392 } 4393 4394 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4395 4396 /** 4397 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4398 * 4399 * PHY drivers may accept clones of transmitted packets for 4400 * timestamping via their phy_driver.txtstamp method. These drivers 4401 * must call this function to return the skb back to the stack with a 4402 * timestamp. 4403 * 4404 * @skb: clone of the original outgoing packet 4405 * @hwtstamps: hardware time stamps 4406 * 4407 */ 4408 void skb_complete_tx_timestamp(struct sk_buff *skb, 4409 struct skb_shared_hwtstamps *hwtstamps); 4410 4411 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4412 struct skb_shared_hwtstamps *hwtstamps, 4413 struct sock *sk, int tstype); 4414 4415 /** 4416 * skb_tstamp_tx - queue clone of skb with send time stamps 4417 * @orig_skb: the original outgoing packet 4418 * @hwtstamps: hardware time stamps, may be NULL if not available 4419 * 4420 * If the skb has a socket associated, then this function clones the 4421 * skb (thus sharing the actual data and optional structures), stores 4422 * the optional hardware time stamping information (if non NULL) or 4423 * generates a software time stamp (otherwise), then queues the clone 4424 * to the error queue of the socket. Errors are silently ignored. 4425 */ 4426 void skb_tstamp_tx(struct sk_buff *orig_skb, 4427 struct skb_shared_hwtstamps *hwtstamps); 4428 4429 /** 4430 * skb_tx_timestamp() - Driver hook for transmit timestamping 4431 * 4432 * Ethernet MAC Drivers should call this function in their hard_xmit() 4433 * function immediately before giving the sk_buff to the MAC hardware. 4434 * 4435 * Specifically, one should make absolutely sure that this function is 4436 * called before TX completion of this packet can trigger. Otherwise 4437 * the packet could potentially already be freed. 4438 * 4439 * @skb: A socket buffer. 4440 */ 4441 static inline void skb_tx_timestamp(struct sk_buff *skb) 4442 { 4443 skb_clone_tx_timestamp(skb); 4444 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 4445 skb_tstamp_tx(skb, NULL); 4446 } 4447 4448 /** 4449 * skb_complete_wifi_ack - deliver skb with wifi status 4450 * 4451 * @skb: the original outgoing packet 4452 * @acked: ack status 4453 * 4454 */ 4455 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4456 4457 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4458 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4459 4460 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4461 { 4462 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4463 skb->csum_valid || 4464 (skb->ip_summed == CHECKSUM_PARTIAL && 4465 skb_checksum_start_offset(skb) >= 0)); 4466 } 4467 4468 /** 4469 * skb_checksum_complete - Calculate checksum of an entire packet 4470 * @skb: packet to process 4471 * 4472 * This function calculates the checksum over the entire packet plus 4473 * the value of skb->csum. The latter can be used to supply the 4474 * checksum of a pseudo header as used by TCP/UDP. It returns the 4475 * checksum. 4476 * 4477 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4478 * this function can be used to verify that checksum on received 4479 * packets. In that case the function should return zero if the 4480 * checksum is correct. In particular, this function will return zero 4481 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4482 * hardware has already verified the correctness of the checksum. 4483 */ 4484 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4485 { 4486 return skb_csum_unnecessary(skb) ? 4487 0 : __skb_checksum_complete(skb); 4488 } 4489 4490 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4491 { 4492 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4493 if (skb->csum_level == 0) 4494 skb->ip_summed = CHECKSUM_NONE; 4495 else 4496 skb->csum_level--; 4497 } 4498 } 4499 4500 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4501 { 4502 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4503 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4504 skb->csum_level++; 4505 } else if (skb->ip_summed == CHECKSUM_NONE) { 4506 skb->ip_summed = CHECKSUM_UNNECESSARY; 4507 skb->csum_level = 0; 4508 } 4509 } 4510 4511 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4512 { 4513 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4514 skb->ip_summed = CHECKSUM_NONE; 4515 skb->csum_level = 0; 4516 } 4517 } 4518 4519 /* Check if we need to perform checksum complete validation. 4520 * 4521 * Returns true if checksum complete is needed, false otherwise 4522 * (either checksum is unnecessary or zero checksum is allowed). 4523 */ 4524 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4525 bool zero_okay, 4526 __sum16 check) 4527 { 4528 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4529 skb->csum_valid = 1; 4530 __skb_decr_checksum_unnecessary(skb); 4531 return false; 4532 } 4533 4534 return true; 4535 } 4536 4537 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4538 * in checksum_init. 4539 */ 4540 #define CHECKSUM_BREAK 76 4541 4542 /* Unset checksum-complete 4543 * 4544 * Unset checksum complete can be done when packet is being modified 4545 * (uncompressed for instance) and checksum-complete value is 4546 * invalidated. 4547 */ 4548 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4549 { 4550 if (skb->ip_summed == CHECKSUM_COMPLETE) 4551 skb->ip_summed = CHECKSUM_NONE; 4552 } 4553 4554 /* Validate (init) checksum based on checksum complete. 4555 * 4556 * Return values: 4557 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4558 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4559 * checksum is stored in skb->csum for use in __skb_checksum_complete 4560 * non-zero: value of invalid checksum 4561 * 4562 */ 4563 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4564 bool complete, 4565 __wsum psum) 4566 { 4567 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4568 if (!csum_fold(csum_add(psum, skb->csum))) { 4569 skb->csum_valid = 1; 4570 return 0; 4571 } 4572 } 4573 4574 skb->csum = psum; 4575 4576 if (complete || skb->len <= CHECKSUM_BREAK) { 4577 __sum16 csum; 4578 4579 csum = __skb_checksum_complete(skb); 4580 skb->csum_valid = !csum; 4581 return csum; 4582 } 4583 4584 return 0; 4585 } 4586 4587 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4588 { 4589 return 0; 4590 } 4591 4592 /* Perform checksum validate (init). Note that this is a macro since we only 4593 * want to calculate the pseudo header which is an input function if necessary. 4594 * First we try to validate without any computation (checksum unnecessary) and 4595 * then calculate based on checksum complete calling the function to compute 4596 * pseudo header. 4597 * 4598 * Return values: 4599 * 0: checksum is validated or try to in skb_checksum_complete 4600 * non-zero: value of invalid checksum 4601 */ 4602 #define __skb_checksum_validate(skb, proto, complete, \ 4603 zero_okay, check, compute_pseudo) \ 4604 ({ \ 4605 __sum16 __ret = 0; \ 4606 skb->csum_valid = 0; \ 4607 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4608 __ret = __skb_checksum_validate_complete(skb, \ 4609 complete, compute_pseudo(skb, proto)); \ 4610 __ret; \ 4611 }) 4612 4613 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4614 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4615 4616 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4617 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4618 4619 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4620 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4621 4622 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4623 compute_pseudo) \ 4624 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4625 4626 #define skb_checksum_simple_validate(skb) \ 4627 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4628 4629 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4630 { 4631 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4632 } 4633 4634 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4635 { 4636 skb->csum = ~pseudo; 4637 skb->ip_summed = CHECKSUM_COMPLETE; 4638 } 4639 4640 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4641 do { \ 4642 if (__skb_checksum_convert_check(skb)) \ 4643 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4644 } while (0) 4645 4646 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4647 u16 start, u16 offset) 4648 { 4649 skb->ip_summed = CHECKSUM_PARTIAL; 4650 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4651 skb->csum_offset = offset - start; 4652 } 4653 4654 /* Update skbuf and packet to reflect the remote checksum offload operation. 4655 * When called, ptr indicates the starting point for skb->csum when 4656 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4657 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4658 */ 4659 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4660 int start, int offset, bool nopartial) 4661 { 4662 __wsum delta; 4663 4664 if (!nopartial) { 4665 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4666 return; 4667 } 4668 4669 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4670 __skb_checksum_complete(skb); 4671 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4672 } 4673 4674 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4675 4676 /* Adjust skb->csum since we changed the packet */ 4677 skb->csum = csum_add(skb->csum, delta); 4678 } 4679 4680 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4681 { 4682 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4683 return (void *)(skb->_nfct & NFCT_PTRMASK); 4684 #else 4685 return NULL; 4686 #endif 4687 } 4688 4689 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4690 { 4691 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4692 return skb->_nfct; 4693 #else 4694 return 0UL; 4695 #endif 4696 } 4697 4698 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4699 { 4700 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4701 skb->slow_gro |= !!nfct; 4702 skb->_nfct = nfct; 4703 #endif 4704 } 4705 4706 #ifdef CONFIG_SKB_EXTENSIONS 4707 enum skb_ext_id { 4708 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4709 SKB_EXT_BRIDGE_NF, 4710 #endif 4711 #ifdef CONFIG_XFRM 4712 SKB_EXT_SEC_PATH, 4713 #endif 4714 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4715 TC_SKB_EXT, 4716 #endif 4717 #if IS_ENABLED(CONFIG_MPTCP) 4718 SKB_EXT_MPTCP, 4719 #endif 4720 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 4721 SKB_EXT_MCTP, 4722 #endif 4723 SKB_EXT_NUM, /* must be last */ 4724 }; 4725 4726 /** 4727 * struct skb_ext - sk_buff extensions 4728 * @refcnt: 1 on allocation, deallocated on 0 4729 * @offset: offset to add to @data to obtain extension address 4730 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4731 * @data: start of extension data, variable sized 4732 * 4733 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4734 * to use 'u8' types while allowing up to 2kb worth of extension data. 4735 */ 4736 struct skb_ext { 4737 refcount_t refcnt; 4738 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4739 u8 chunks; /* same */ 4740 char data[] __aligned(8); 4741 }; 4742 4743 struct skb_ext *__skb_ext_alloc(gfp_t flags); 4744 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 4745 struct skb_ext *ext); 4746 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4747 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4748 void __skb_ext_put(struct skb_ext *ext); 4749 4750 static inline void skb_ext_put(struct sk_buff *skb) 4751 { 4752 if (skb->active_extensions) 4753 __skb_ext_put(skb->extensions); 4754 } 4755 4756 static inline void __skb_ext_copy(struct sk_buff *dst, 4757 const struct sk_buff *src) 4758 { 4759 dst->active_extensions = src->active_extensions; 4760 4761 if (src->active_extensions) { 4762 struct skb_ext *ext = src->extensions; 4763 4764 refcount_inc(&ext->refcnt); 4765 dst->extensions = ext; 4766 } 4767 } 4768 4769 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4770 { 4771 skb_ext_put(dst); 4772 __skb_ext_copy(dst, src); 4773 } 4774 4775 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4776 { 4777 return !!ext->offset[i]; 4778 } 4779 4780 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4781 { 4782 return skb->active_extensions & (1 << id); 4783 } 4784 4785 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4786 { 4787 if (skb_ext_exist(skb, id)) 4788 __skb_ext_del(skb, id); 4789 } 4790 4791 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4792 { 4793 if (skb_ext_exist(skb, id)) { 4794 struct skb_ext *ext = skb->extensions; 4795 4796 return (void *)ext + (ext->offset[id] << 3); 4797 } 4798 4799 return NULL; 4800 } 4801 4802 static inline void skb_ext_reset(struct sk_buff *skb) 4803 { 4804 if (unlikely(skb->active_extensions)) { 4805 __skb_ext_put(skb->extensions); 4806 skb->active_extensions = 0; 4807 } 4808 } 4809 4810 static inline bool skb_has_extensions(struct sk_buff *skb) 4811 { 4812 return unlikely(skb->active_extensions); 4813 } 4814 #else 4815 static inline void skb_ext_put(struct sk_buff *skb) {} 4816 static inline void skb_ext_reset(struct sk_buff *skb) {} 4817 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4818 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4819 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4820 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 4821 #endif /* CONFIG_SKB_EXTENSIONS */ 4822 4823 static inline void nf_reset_ct(struct sk_buff *skb) 4824 { 4825 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4826 nf_conntrack_put(skb_nfct(skb)); 4827 skb->_nfct = 0; 4828 #endif 4829 } 4830 4831 static inline void nf_reset_trace(struct sk_buff *skb) 4832 { 4833 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4834 skb->nf_trace = 0; 4835 #endif 4836 } 4837 4838 static inline void ipvs_reset(struct sk_buff *skb) 4839 { 4840 #if IS_ENABLED(CONFIG_IP_VS) 4841 skb->ipvs_property = 0; 4842 #endif 4843 } 4844 4845 /* Note: This doesn't put any conntrack info in dst. */ 4846 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4847 bool copy) 4848 { 4849 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4850 dst->_nfct = src->_nfct; 4851 nf_conntrack_get(skb_nfct(src)); 4852 #endif 4853 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4854 if (copy) 4855 dst->nf_trace = src->nf_trace; 4856 #endif 4857 } 4858 4859 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 4860 { 4861 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4862 nf_conntrack_put(skb_nfct(dst)); 4863 #endif 4864 dst->slow_gro = src->slow_gro; 4865 __nf_copy(dst, src, true); 4866 } 4867 4868 #ifdef CONFIG_NETWORK_SECMARK 4869 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4870 { 4871 to->secmark = from->secmark; 4872 } 4873 4874 static inline void skb_init_secmark(struct sk_buff *skb) 4875 { 4876 skb->secmark = 0; 4877 } 4878 #else 4879 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4880 { } 4881 4882 static inline void skb_init_secmark(struct sk_buff *skb) 4883 { } 4884 #endif 4885 4886 static inline int secpath_exists(const struct sk_buff *skb) 4887 { 4888 #ifdef CONFIG_XFRM 4889 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 4890 #else 4891 return 0; 4892 #endif 4893 } 4894 4895 static inline bool skb_irq_freeable(const struct sk_buff *skb) 4896 { 4897 return !skb->destructor && 4898 !secpath_exists(skb) && 4899 !skb_nfct(skb) && 4900 !skb->_skb_refdst && 4901 !skb_has_frag_list(skb); 4902 } 4903 4904 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 4905 { 4906 skb->queue_mapping = queue_mapping; 4907 } 4908 4909 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 4910 { 4911 return skb->queue_mapping; 4912 } 4913 4914 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 4915 { 4916 to->queue_mapping = from->queue_mapping; 4917 } 4918 4919 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 4920 { 4921 skb->queue_mapping = rx_queue + 1; 4922 } 4923 4924 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 4925 { 4926 return skb->queue_mapping - 1; 4927 } 4928 4929 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 4930 { 4931 return skb->queue_mapping != 0; 4932 } 4933 4934 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 4935 { 4936 skb->dst_pending_confirm = val; 4937 } 4938 4939 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 4940 { 4941 return skb->dst_pending_confirm != 0; 4942 } 4943 4944 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 4945 { 4946 #ifdef CONFIG_XFRM 4947 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 4948 #else 4949 return NULL; 4950 #endif 4951 } 4952 4953 static inline bool skb_is_gso(const struct sk_buff *skb) 4954 { 4955 return skb_shinfo(skb)->gso_size; 4956 } 4957 4958 /* Note: Should be called only if skb_is_gso(skb) is true */ 4959 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4960 { 4961 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4962 } 4963 4964 /* Note: Should be called only if skb_is_gso(skb) is true */ 4965 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 4966 { 4967 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 4968 } 4969 4970 /* Note: Should be called only if skb_is_gso(skb) is true */ 4971 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 4972 { 4973 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 4974 } 4975 4976 static inline void skb_gso_reset(struct sk_buff *skb) 4977 { 4978 skb_shinfo(skb)->gso_size = 0; 4979 skb_shinfo(skb)->gso_segs = 0; 4980 skb_shinfo(skb)->gso_type = 0; 4981 } 4982 4983 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 4984 u16 increment) 4985 { 4986 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4987 return; 4988 shinfo->gso_size += increment; 4989 } 4990 4991 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 4992 u16 decrement) 4993 { 4994 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4995 return; 4996 shinfo->gso_size -= decrement; 4997 } 4998 4999 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 5000 5001 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 5002 { 5003 /* LRO sets gso_size but not gso_type, whereas if GSO is really 5004 * wanted then gso_type will be set. */ 5005 const struct skb_shared_info *shinfo = skb_shinfo(skb); 5006 5007 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 5008 unlikely(shinfo->gso_type == 0)) { 5009 __skb_warn_lro_forwarding(skb); 5010 return true; 5011 } 5012 return false; 5013 } 5014 5015 static inline void skb_forward_csum(struct sk_buff *skb) 5016 { 5017 /* Unfortunately we don't support this one. Any brave souls? */ 5018 if (skb->ip_summed == CHECKSUM_COMPLETE) 5019 skb->ip_summed = CHECKSUM_NONE; 5020 } 5021 5022 /** 5023 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 5024 * @skb: skb to check 5025 * 5026 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 5027 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 5028 * use this helper, to document places where we make this assertion. 5029 */ 5030 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 5031 { 5032 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 5033 } 5034 5035 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 5036 5037 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 5038 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 5039 unsigned int transport_len, 5040 __sum16(*skb_chkf)(struct sk_buff *skb)); 5041 5042 /** 5043 * skb_head_is_locked - Determine if the skb->head is locked down 5044 * @skb: skb to check 5045 * 5046 * The head on skbs build around a head frag can be removed if they are 5047 * not cloned. This function returns true if the skb head is locked down 5048 * due to either being allocated via kmalloc, or by being a clone with 5049 * multiple references to the head. 5050 */ 5051 static inline bool skb_head_is_locked(const struct sk_buff *skb) 5052 { 5053 return !skb->head_frag || skb_cloned(skb); 5054 } 5055 5056 /* Local Checksum Offload. 5057 * Compute outer checksum based on the assumption that the 5058 * inner checksum will be offloaded later. 5059 * See Documentation/networking/checksum-offloads.rst for 5060 * explanation of how this works. 5061 * Fill in outer checksum adjustment (e.g. with sum of outer 5062 * pseudo-header) before calling. 5063 * Also ensure that inner checksum is in linear data area. 5064 */ 5065 static inline __wsum lco_csum(struct sk_buff *skb) 5066 { 5067 unsigned char *csum_start = skb_checksum_start(skb); 5068 unsigned char *l4_hdr = skb_transport_header(skb); 5069 __wsum partial; 5070 5071 /* Start with complement of inner checksum adjustment */ 5072 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 5073 skb->csum_offset)); 5074 5075 /* Add in checksum of our headers (incl. outer checksum 5076 * adjustment filled in by caller) and return result. 5077 */ 5078 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5079 } 5080 5081 static inline bool skb_is_redirected(const struct sk_buff *skb) 5082 { 5083 return skb->redirected; 5084 } 5085 5086 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5087 { 5088 skb->redirected = 1; 5089 #ifdef CONFIG_NET_REDIRECT 5090 skb->from_ingress = from_ingress; 5091 if (skb->from_ingress) 5092 skb_clear_tstamp(skb); 5093 #endif 5094 } 5095 5096 static inline void skb_reset_redirect(struct sk_buff *skb) 5097 { 5098 skb->redirected = 0; 5099 } 5100 5101 static inline void skb_set_redirected_noclear(struct sk_buff *skb, 5102 bool from_ingress) 5103 { 5104 skb->redirected = 1; 5105 #ifdef CONFIG_NET_REDIRECT 5106 skb->from_ingress = from_ingress; 5107 #endif 5108 } 5109 5110 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5111 { 5112 #if IS_ENABLED(CONFIG_IP_SCTP) 5113 return skb->csum_not_inet; 5114 #else 5115 return 0; 5116 #endif 5117 } 5118 5119 static inline void skb_reset_csum_not_inet(struct sk_buff *skb) 5120 { 5121 skb->ip_summed = CHECKSUM_NONE; 5122 #if IS_ENABLED(CONFIG_IP_SCTP) 5123 skb->csum_not_inet = 0; 5124 #endif 5125 } 5126 5127 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5128 const u64 kcov_handle) 5129 { 5130 #ifdef CONFIG_KCOV 5131 skb->kcov_handle = kcov_handle; 5132 #endif 5133 } 5134 5135 static inline u64 skb_get_kcov_handle(struct sk_buff *skb) 5136 { 5137 #ifdef CONFIG_KCOV 5138 return skb->kcov_handle; 5139 #else 5140 return 0; 5141 #endif 5142 } 5143 5144 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5145 { 5146 #ifdef CONFIG_PAGE_POOL 5147 skb->pp_recycle = 1; 5148 #endif 5149 } 5150 5151 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter, 5152 ssize_t maxsize, gfp_t gfp); 5153 5154 #endif /* __KERNEL__ */ 5155 #endif /* _LINUX_SKBUFF_H */ 5156