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