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