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