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