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