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