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 }; 682 struct sock *sk; 683 684 union { 685 ktime_t tstamp; 686 u64 skb_mstamp; 687 }; 688 /* 689 * This is the control buffer. It is free to use for every 690 * layer. Please put your private variables there. If you 691 * want to keep them across layers you have to do a skb_clone() 692 * first. This is owned by whoever has the skb queued ATM. 693 */ 694 char cb[48] __aligned(8); 695 696 union { 697 struct { 698 unsigned long _skb_refdst; 699 void (*destructor)(struct sk_buff *skb); 700 }; 701 struct list_head tcp_tsorted_anchor; 702 }; 703 704 #ifdef CONFIG_XFRM 705 struct sec_path *sp; 706 #endif 707 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 708 unsigned long _nfct; 709 #endif 710 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 711 struct nf_bridge_info *nf_bridge; 712 #endif 713 unsigned int len, 714 data_len; 715 __u16 mac_len, 716 hdr_len; 717 718 /* Following fields are _not_ copied in __copy_skb_header() 719 * Note that queue_mapping is here mostly to fill a hole. 720 */ 721 __u16 queue_mapping; 722 723 /* if you move cloned around you also must adapt those constants */ 724 #ifdef __BIG_ENDIAN_BITFIELD 725 #define CLONED_MASK (1 << 7) 726 #else 727 #define CLONED_MASK 1 728 #endif 729 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset) 730 731 __u8 __cloned_offset[0]; 732 __u8 cloned:1, 733 nohdr:1, 734 fclone:2, 735 peeked:1, 736 head_frag:1, 737 xmit_more:1, 738 __unused:1; /* one bit hole */ 739 740 /* fields enclosed in headers_start/headers_end are copied 741 * using a single memcpy() in __copy_skb_header() 742 */ 743 /* private: */ 744 __u32 headers_start[0]; 745 /* public: */ 746 747 /* if you move pkt_type around you also must adapt those constants */ 748 #ifdef __BIG_ENDIAN_BITFIELD 749 #define PKT_TYPE_MAX (7 << 5) 750 #else 751 #define PKT_TYPE_MAX 7 752 #endif 753 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 754 755 __u8 __pkt_type_offset[0]; 756 __u8 pkt_type:3; 757 __u8 pfmemalloc:1; 758 __u8 ignore_df:1; 759 760 __u8 nf_trace:1; 761 __u8 ip_summed:2; 762 __u8 ooo_okay:1; 763 __u8 l4_hash:1; 764 __u8 sw_hash:1; 765 __u8 wifi_acked_valid:1; 766 __u8 wifi_acked:1; 767 768 __u8 no_fcs:1; 769 /* Indicates the inner headers are valid in the skbuff. */ 770 __u8 encapsulation:1; 771 __u8 encap_hdr_csum:1; 772 __u8 csum_valid:1; 773 __u8 csum_complete_sw:1; 774 __u8 csum_level:2; 775 __u8 csum_not_inet:1; 776 777 __u8 dst_pending_confirm:1; 778 #ifdef CONFIG_IPV6_NDISC_NODETYPE 779 __u8 ndisc_nodetype:2; 780 #endif 781 __u8 ipvs_property:1; 782 __u8 inner_protocol_type:1; 783 __u8 remcsum_offload:1; 784 #ifdef CONFIG_NET_SWITCHDEV 785 __u8 offload_fwd_mark:1; 786 __u8 offload_mr_fwd_mark:1; 787 #endif 788 #ifdef CONFIG_NET_CLS_ACT 789 __u8 tc_skip_classify:1; 790 __u8 tc_at_ingress:1; 791 __u8 tc_redirected:1; 792 __u8 tc_from_ingress:1; 793 #endif 794 795 #ifdef CONFIG_NET_SCHED 796 __u16 tc_index; /* traffic control index */ 797 #endif 798 799 union { 800 __wsum csum; 801 struct { 802 __u16 csum_start; 803 __u16 csum_offset; 804 }; 805 }; 806 __u32 priority; 807 int skb_iif; 808 __u32 hash; 809 __be16 vlan_proto; 810 __u16 vlan_tci; 811 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 812 union { 813 unsigned int napi_id; 814 unsigned int sender_cpu; 815 }; 816 #endif 817 #ifdef CONFIG_NETWORK_SECMARK 818 __u32 secmark; 819 #endif 820 821 union { 822 __u32 mark; 823 __u32 reserved_tailroom; 824 }; 825 826 union { 827 __be16 inner_protocol; 828 __u8 inner_ipproto; 829 }; 830 831 __u16 inner_transport_header; 832 __u16 inner_network_header; 833 __u16 inner_mac_header; 834 835 __be16 protocol; 836 __u16 transport_header; 837 __u16 network_header; 838 __u16 mac_header; 839 840 /* private: */ 841 __u32 headers_end[0]; 842 /* public: */ 843 844 /* These elements must be at the end, see alloc_skb() for details. */ 845 sk_buff_data_t tail; 846 sk_buff_data_t end; 847 unsigned char *head, 848 *data; 849 unsigned int truesize; 850 refcount_t users; 851 }; 852 853 #ifdef __KERNEL__ 854 /* 855 * Handling routines are only of interest to the kernel 856 */ 857 #include <linux/slab.h> 858 859 860 #define SKB_ALLOC_FCLONE 0x01 861 #define SKB_ALLOC_RX 0x02 862 #define SKB_ALLOC_NAPI 0x04 863 864 /* Returns true if the skb was allocated from PFMEMALLOC reserves */ 865 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 866 { 867 return unlikely(skb->pfmemalloc); 868 } 869 870 /* 871 * skb might have a dst pointer attached, refcounted or not. 872 * _skb_refdst low order bit is set if refcount was _not_ taken 873 */ 874 #define SKB_DST_NOREF 1UL 875 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 876 877 #define SKB_NFCT_PTRMASK ~(7UL) 878 /** 879 * skb_dst - returns skb dst_entry 880 * @skb: buffer 881 * 882 * Returns skb dst_entry, regardless of reference taken or not. 883 */ 884 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 885 { 886 /* If refdst was not refcounted, check we still are in a 887 * rcu_read_lock section 888 */ 889 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 890 !rcu_read_lock_held() && 891 !rcu_read_lock_bh_held()); 892 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 893 } 894 895 /** 896 * skb_dst_set - sets skb dst 897 * @skb: buffer 898 * @dst: dst entry 899 * 900 * Sets skb dst, assuming a reference was taken on dst and should 901 * be released by skb_dst_drop() 902 */ 903 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 904 { 905 skb->_skb_refdst = (unsigned long)dst; 906 } 907 908 /** 909 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 910 * @skb: buffer 911 * @dst: dst entry 912 * 913 * Sets skb dst, assuming a reference was not taken on dst. 914 * If dst entry is cached, we do not take reference and dst_release 915 * will be avoided by refdst_drop. If dst entry is not cached, we take 916 * reference, so that last dst_release can destroy the dst immediately. 917 */ 918 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 919 { 920 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 921 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 922 } 923 924 /** 925 * skb_dst_is_noref - Test if skb dst isn't refcounted 926 * @skb: buffer 927 */ 928 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 929 { 930 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 931 } 932 933 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 934 { 935 return (struct rtable *)skb_dst(skb); 936 } 937 938 /* For mangling skb->pkt_type from user space side from applications 939 * such as nft, tc, etc, we only allow a conservative subset of 940 * possible pkt_types to be set. 941 */ 942 static inline bool skb_pkt_type_ok(u32 ptype) 943 { 944 return ptype <= PACKET_OTHERHOST; 945 } 946 947 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 948 { 949 #ifdef CONFIG_NET_RX_BUSY_POLL 950 return skb->napi_id; 951 #else 952 return 0; 953 #endif 954 } 955 956 /* decrement the reference count and return true if we can free the skb */ 957 static inline bool skb_unref(struct sk_buff *skb) 958 { 959 if (unlikely(!skb)) 960 return false; 961 if (likely(refcount_read(&skb->users) == 1)) 962 smp_rmb(); 963 else if (likely(!refcount_dec_and_test(&skb->users))) 964 return false; 965 966 return true; 967 } 968 969 void skb_release_head_state(struct sk_buff *skb); 970 void kfree_skb(struct sk_buff *skb); 971 void kfree_skb_list(struct sk_buff *segs); 972 void skb_tx_error(struct sk_buff *skb); 973 void consume_skb(struct sk_buff *skb); 974 void __consume_stateless_skb(struct sk_buff *skb); 975 void __kfree_skb(struct sk_buff *skb); 976 extern struct kmem_cache *skbuff_head_cache; 977 978 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 979 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 980 bool *fragstolen, int *delta_truesize); 981 982 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 983 int node); 984 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 985 struct sk_buff *build_skb(void *data, unsigned int frag_size); 986 static inline struct sk_buff *alloc_skb(unsigned int size, 987 gfp_t priority) 988 { 989 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 990 } 991 992 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 993 unsigned long data_len, 994 int max_page_order, 995 int *errcode, 996 gfp_t gfp_mask); 997 998 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 999 struct sk_buff_fclones { 1000 struct sk_buff skb1; 1001 1002 struct sk_buff skb2; 1003 1004 refcount_t fclone_ref; 1005 }; 1006 1007 /** 1008 * skb_fclone_busy - check if fclone is busy 1009 * @sk: socket 1010 * @skb: buffer 1011 * 1012 * Returns true if skb is a fast clone, and its clone is not freed. 1013 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1014 * so we also check that this didnt happen. 1015 */ 1016 static inline bool skb_fclone_busy(const struct sock *sk, 1017 const struct sk_buff *skb) 1018 { 1019 const struct sk_buff_fclones *fclones; 1020 1021 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1022 1023 return skb->fclone == SKB_FCLONE_ORIG && 1024 refcount_read(&fclones->fclone_ref) > 1 && 1025 fclones->skb2.sk == sk; 1026 } 1027 1028 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1029 gfp_t priority) 1030 { 1031 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1032 } 1033 1034 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1035 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1036 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1037 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1038 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1039 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1040 gfp_t gfp_mask, bool fclone); 1041 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1042 gfp_t gfp_mask) 1043 { 1044 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1045 } 1046 1047 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1048 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1049 unsigned int headroom); 1050 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1051 int newtailroom, gfp_t priority); 1052 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1053 int offset, int len); 1054 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1055 int offset, int len); 1056 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1057 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1058 1059 /** 1060 * skb_pad - zero pad the tail of an skb 1061 * @skb: buffer to pad 1062 * @pad: space to pad 1063 * 1064 * Ensure that a buffer is followed by a padding area that is zero 1065 * filled. Used by network drivers which may DMA or transfer data 1066 * beyond the buffer end onto the wire. 1067 * 1068 * May return error in out of memory cases. The skb is freed on error. 1069 */ 1070 static inline int skb_pad(struct sk_buff *skb, int pad) 1071 { 1072 return __skb_pad(skb, pad, true); 1073 } 1074 #define dev_kfree_skb(a) consume_skb(a) 1075 1076 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 1077 int getfrag(void *from, char *to, int offset, 1078 int len, int odd, struct sk_buff *skb), 1079 void *from, int length); 1080 1081 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1082 int offset, size_t size); 1083 1084 struct skb_seq_state { 1085 __u32 lower_offset; 1086 __u32 upper_offset; 1087 __u32 frag_idx; 1088 __u32 stepped_offset; 1089 struct sk_buff *root_skb; 1090 struct sk_buff *cur_skb; 1091 __u8 *frag_data; 1092 }; 1093 1094 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1095 unsigned int to, struct skb_seq_state *st); 1096 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1097 struct skb_seq_state *st); 1098 void skb_abort_seq_read(struct skb_seq_state *st); 1099 1100 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1101 unsigned int to, struct ts_config *config); 1102 1103 /* 1104 * Packet hash types specify the type of hash in skb_set_hash. 1105 * 1106 * Hash types refer to the protocol layer addresses which are used to 1107 * construct a packet's hash. The hashes are used to differentiate or identify 1108 * flows of the protocol layer for the hash type. Hash types are either 1109 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1110 * 1111 * Properties of hashes: 1112 * 1113 * 1) Two packets in different flows have different hash values 1114 * 2) Two packets in the same flow should have the same hash value 1115 * 1116 * A hash at a higher layer is considered to be more specific. A driver should 1117 * set the most specific hash possible. 1118 * 1119 * A driver cannot indicate a more specific hash than the layer at which a hash 1120 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1121 * 1122 * A driver may indicate a hash level which is less specific than the 1123 * actual layer the hash was computed on. For instance, a hash computed 1124 * at L4 may be considered an L3 hash. This should only be done if the 1125 * driver can't unambiguously determine that the HW computed the hash at 1126 * the higher layer. Note that the "should" in the second property above 1127 * permits this. 1128 */ 1129 enum pkt_hash_types { 1130 PKT_HASH_TYPE_NONE, /* Undefined type */ 1131 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1132 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1133 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1134 }; 1135 1136 static inline void skb_clear_hash(struct sk_buff *skb) 1137 { 1138 skb->hash = 0; 1139 skb->sw_hash = 0; 1140 skb->l4_hash = 0; 1141 } 1142 1143 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1144 { 1145 if (!skb->l4_hash) 1146 skb_clear_hash(skb); 1147 } 1148 1149 static inline void 1150 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1151 { 1152 skb->l4_hash = is_l4; 1153 skb->sw_hash = is_sw; 1154 skb->hash = hash; 1155 } 1156 1157 static inline void 1158 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1159 { 1160 /* Used by drivers to set hash from HW */ 1161 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1162 } 1163 1164 static inline void 1165 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1166 { 1167 __skb_set_hash(skb, hash, true, is_l4); 1168 } 1169 1170 void __skb_get_hash(struct sk_buff *skb); 1171 u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1172 u32 skb_get_poff(const struct sk_buff *skb); 1173 u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1174 const struct flow_keys_basic *keys, int hlen); 1175 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1176 void *data, int hlen_proto); 1177 1178 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1179 int thoff, u8 ip_proto) 1180 { 1181 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1182 } 1183 1184 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1185 const struct flow_dissector_key *key, 1186 unsigned int key_count); 1187 1188 bool __skb_flow_dissect(const struct sk_buff *skb, 1189 struct flow_dissector *flow_dissector, 1190 void *target_container, 1191 void *data, __be16 proto, int nhoff, int hlen, 1192 unsigned int flags); 1193 1194 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1195 struct flow_dissector *flow_dissector, 1196 void *target_container, unsigned int flags) 1197 { 1198 return __skb_flow_dissect(skb, flow_dissector, target_container, 1199 NULL, 0, 0, 0, flags); 1200 } 1201 1202 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1203 struct flow_keys *flow, 1204 unsigned int flags) 1205 { 1206 memset(flow, 0, sizeof(*flow)); 1207 return __skb_flow_dissect(skb, &flow_keys_dissector, flow, 1208 NULL, 0, 0, 0, flags); 1209 } 1210 1211 static inline bool 1212 skb_flow_dissect_flow_keys_basic(const struct sk_buff *skb, 1213 struct flow_keys_basic *flow, void *data, 1214 __be16 proto, int nhoff, int hlen, 1215 unsigned int flags) 1216 { 1217 memset(flow, 0, sizeof(*flow)); 1218 return __skb_flow_dissect(skb, &flow_keys_basic_dissector, flow, 1219 data, proto, nhoff, hlen, flags); 1220 } 1221 1222 void 1223 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1224 struct flow_dissector *flow_dissector, 1225 void *target_container); 1226 1227 static inline __u32 skb_get_hash(struct sk_buff *skb) 1228 { 1229 if (!skb->l4_hash && !skb->sw_hash) 1230 __skb_get_hash(skb); 1231 1232 return skb->hash; 1233 } 1234 1235 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1236 { 1237 if (!skb->l4_hash && !skb->sw_hash) { 1238 struct flow_keys keys; 1239 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1240 1241 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1242 } 1243 1244 return skb->hash; 1245 } 1246 1247 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1248 1249 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1250 { 1251 return skb->hash; 1252 } 1253 1254 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1255 { 1256 to->hash = from->hash; 1257 to->sw_hash = from->sw_hash; 1258 to->l4_hash = from->l4_hash; 1259 }; 1260 1261 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1262 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1263 { 1264 return skb->head + skb->end; 1265 } 1266 1267 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1268 { 1269 return skb->end; 1270 } 1271 #else 1272 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1273 { 1274 return skb->end; 1275 } 1276 1277 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1278 { 1279 return skb->end - skb->head; 1280 } 1281 #endif 1282 1283 /* Internal */ 1284 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1285 1286 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1287 { 1288 return &skb_shinfo(skb)->hwtstamps; 1289 } 1290 1291 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1292 { 1293 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY; 1294 1295 return is_zcopy ? skb_uarg(skb) : NULL; 1296 } 1297 1298 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg) 1299 { 1300 if (skb && uarg && !skb_zcopy(skb)) { 1301 sock_zerocopy_get(uarg); 1302 skb_shinfo(skb)->destructor_arg = uarg; 1303 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG; 1304 } 1305 } 1306 1307 /* Release a reference on a zerocopy structure */ 1308 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy) 1309 { 1310 struct ubuf_info *uarg = skb_zcopy(skb); 1311 1312 if (uarg) { 1313 if (uarg->callback == sock_zerocopy_callback) { 1314 uarg->zerocopy = uarg->zerocopy && zerocopy; 1315 sock_zerocopy_put(uarg); 1316 } else { 1317 uarg->callback(uarg, zerocopy); 1318 } 1319 1320 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1321 } 1322 } 1323 1324 /* Abort a zerocopy operation and revert zckey on error in send syscall */ 1325 static inline void skb_zcopy_abort(struct sk_buff *skb) 1326 { 1327 struct ubuf_info *uarg = skb_zcopy(skb); 1328 1329 if (uarg) { 1330 sock_zerocopy_put_abort(uarg); 1331 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1332 } 1333 } 1334 1335 /** 1336 * skb_queue_empty - check if a queue is empty 1337 * @list: queue head 1338 * 1339 * Returns true if the queue is empty, false otherwise. 1340 */ 1341 static inline int skb_queue_empty(const struct sk_buff_head *list) 1342 { 1343 return list->next == (const struct sk_buff *) list; 1344 } 1345 1346 /** 1347 * skb_queue_is_last - check if skb is the last entry in the queue 1348 * @list: queue head 1349 * @skb: buffer 1350 * 1351 * Returns true if @skb is the last buffer on the list. 1352 */ 1353 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1354 const struct sk_buff *skb) 1355 { 1356 return skb->next == (const struct sk_buff *) list; 1357 } 1358 1359 /** 1360 * skb_queue_is_first - check if skb is the first entry in the queue 1361 * @list: queue head 1362 * @skb: buffer 1363 * 1364 * Returns true if @skb is the first buffer on the list. 1365 */ 1366 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1367 const struct sk_buff *skb) 1368 { 1369 return skb->prev == (const struct sk_buff *) list; 1370 } 1371 1372 /** 1373 * skb_queue_next - return the next packet in the queue 1374 * @list: queue head 1375 * @skb: current buffer 1376 * 1377 * Return the next packet in @list after @skb. It is only valid to 1378 * call this if skb_queue_is_last() evaluates to false. 1379 */ 1380 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1381 const struct sk_buff *skb) 1382 { 1383 /* This BUG_ON may seem severe, but if we just return then we 1384 * are going to dereference garbage. 1385 */ 1386 BUG_ON(skb_queue_is_last(list, skb)); 1387 return skb->next; 1388 } 1389 1390 /** 1391 * skb_queue_prev - return the prev packet in the queue 1392 * @list: queue head 1393 * @skb: current buffer 1394 * 1395 * Return the prev packet in @list before @skb. It is only valid to 1396 * call this if skb_queue_is_first() evaluates to false. 1397 */ 1398 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1399 const struct sk_buff *skb) 1400 { 1401 /* This BUG_ON may seem severe, but if we just return then we 1402 * are going to dereference garbage. 1403 */ 1404 BUG_ON(skb_queue_is_first(list, skb)); 1405 return skb->prev; 1406 } 1407 1408 /** 1409 * skb_get - reference buffer 1410 * @skb: buffer to reference 1411 * 1412 * Makes another reference to a socket buffer and returns a pointer 1413 * to the buffer. 1414 */ 1415 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1416 { 1417 refcount_inc(&skb->users); 1418 return skb; 1419 } 1420 1421 /* 1422 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1423 */ 1424 1425 /** 1426 * skb_cloned - is the buffer a clone 1427 * @skb: buffer to check 1428 * 1429 * Returns true if the buffer was generated with skb_clone() and is 1430 * one of multiple shared copies of the buffer. Cloned buffers are 1431 * shared data so must not be written to under normal circumstances. 1432 */ 1433 static inline int skb_cloned(const struct sk_buff *skb) 1434 { 1435 return skb->cloned && 1436 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1437 } 1438 1439 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1440 { 1441 might_sleep_if(gfpflags_allow_blocking(pri)); 1442 1443 if (skb_cloned(skb)) 1444 return pskb_expand_head(skb, 0, 0, pri); 1445 1446 return 0; 1447 } 1448 1449 /** 1450 * skb_header_cloned - is the header a clone 1451 * @skb: buffer to check 1452 * 1453 * Returns true if modifying the header part of the buffer requires 1454 * the data to be copied. 1455 */ 1456 static inline int skb_header_cloned(const struct sk_buff *skb) 1457 { 1458 int dataref; 1459 1460 if (!skb->cloned) 1461 return 0; 1462 1463 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1464 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1465 return dataref != 1; 1466 } 1467 1468 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1469 { 1470 might_sleep_if(gfpflags_allow_blocking(pri)); 1471 1472 if (skb_header_cloned(skb)) 1473 return pskb_expand_head(skb, 0, 0, pri); 1474 1475 return 0; 1476 } 1477 1478 /** 1479 * __skb_header_release - release reference to header 1480 * @skb: buffer to operate on 1481 */ 1482 static inline void __skb_header_release(struct sk_buff *skb) 1483 { 1484 skb->nohdr = 1; 1485 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1486 } 1487 1488 1489 /** 1490 * skb_shared - is the buffer shared 1491 * @skb: buffer to check 1492 * 1493 * Returns true if more than one person has a reference to this 1494 * buffer. 1495 */ 1496 static inline int skb_shared(const struct sk_buff *skb) 1497 { 1498 return refcount_read(&skb->users) != 1; 1499 } 1500 1501 /** 1502 * skb_share_check - check if buffer is shared and if so clone it 1503 * @skb: buffer to check 1504 * @pri: priority for memory allocation 1505 * 1506 * If the buffer is shared the buffer is cloned and the old copy 1507 * drops a reference. A new clone with a single reference is returned. 1508 * If the buffer is not shared the original buffer is returned. When 1509 * being called from interrupt status or with spinlocks held pri must 1510 * be GFP_ATOMIC. 1511 * 1512 * NULL is returned on a memory allocation failure. 1513 */ 1514 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1515 { 1516 might_sleep_if(gfpflags_allow_blocking(pri)); 1517 if (skb_shared(skb)) { 1518 struct sk_buff *nskb = skb_clone(skb, pri); 1519 1520 if (likely(nskb)) 1521 consume_skb(skb); 1522 else 1523 kfree_skb(skb); 1524 skb = nskb; 1525 } 1526 return skb; 1527 } 1528 1529 /* 1530 * Copy shared buffers into a new sk_buff. We effectively do COW on 1531 * packets to handle cases where we have a local reader and forward 1532 * and a couple of other messy ones. The normal one is tcpdumping 1533 * a packet thats being forwarded. 1534 */ 1535 1536 /** 1537 * skb_unshare - make a copy of a shared buffer 1538 * @skb: buffer to check 1539 * @pri: priority for memory allocation 1540 * 1541 * If the socket buffer is a clone then this function creates a new 1542 * copy of the data, drops a reference count on the old copy and returns 1543 * the new copy with the reference count at 1. If the buffer is not a clone 1544 * the original buffer is returned. When called with a spinlock held or 1545 * from interrupt state @pri must be %GFP_ATOMIC 1546 * 1547 * %NULL is returned on a memory allocation failure. 1548 */ 1549 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1550 gfp_t pri) 1551 { 1552 might_sleep_if(gfpflags_allow_blocking(pri)); 1553 if (skb_cloned(skb)) { 1554 struct sk_buff *nskb = skb_copy(skb, pri); 1555 1556 /* Free our shared copy */ 1557 if (likely(nskb)) 1558 consume_skb(skb); 1559 else 1560 kfree_skb(skb); 1561 skb = nskb; 1562 } 1563 return skb; 1564 } 1565 1566 /** 1567 * skb_peek - peek at the head of an &sk_buff_head 1568 * @list_: list to peek at 1569 * 1570 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1571 * be careful with this one. A peek leaves the buffer on the 1572 * list and someone else may run off with it. You must hold 1573 * the appropriate locks or have a private queue to do this. 1574 * 1575 * Returns %NULL for an empty list or a pointer to the head element. 1576 * The reference count is not incremented and the reference is therefore 1577 * volatile. Use with caution. 1578 */ 1579 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1580 { 1581 struct sk_buff *skb = list_->next; 1582 1583 if (skb == (struct sk_buff *)list_) 1584 skb = NULL; 1585 return skb; 1586 } 1587 1588 /** 1589 * skb_peek_next - peek skb following the given one from a queue 1590 * @skb: skb to start from 1591 * @list_: list to peek at 1592 * 1593 * Returns %NULL when the end of the list is met or a pointer to the 1594 * next element. The reference count is not incremented and the 1595 * reference is therefore volatile. Use with caution. 1596 */ 1597 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1598 const struct sk_buff_head *list_) 1599 { 1600 struct sk_buff *next = skb->next; 1601 1602 if (next == (struct sk_buff *)list_) 1603 next = NULL; 1604 return next; 1605 } 1606 1607 /** 1608 * skb_peek_tail - peek at the tail of an &sk_buff_head 1609 * @list_: list to peek at 1610 * 1611 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1612 * be careful with this one. A peek leaves the buffer on the 1613 * list and someone else may run off with it. You must hold 1614 * the appropriate locks or have a private queue to do this. 1615 * 1616 * Returns %NULL for an empty list or a pointer to the tail element. 1617 * The reference count is not incremented and the reference is therefore 1618 * volatile. Use with caution. 1619 */ 1620 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1621 { 1622 struct sk_buff *skb = list_->prev; 1623 1624 if (skb == (struct sk_buff *)list_) 1625 skb = NULL; 1626 return skb; 1627 1628 } 1629 1630 /** 1631 * skb_queue_len - get queue length 1632 * @list_: list to measure 1633 * 1634 * Return the length of an &sk_buff queue. 1635 */ 1636 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1637 { 1638 return list_->qlen; 1639 } 1640 1641 /** 1642 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1643 * @list: queue to initialize 1644 * 1645 * This initializes only the list and queue length aspects of 1646 * an sk_buff_head object. This allows to initialize the list 1647 * aspects of an sk_buff_head without reinitializing things like 1648 * the spinlock. It can also be used for on-stack sk_buff_head 1649 * objects where the spinlock is known to not be used. 1650 */ 1651 static inline void __skb_queue_head_init(struct sk_buff_head *list) 1652 { 1653 list->prev = list->next = (struct sk_buff *)list; 1654 list->qlen = 0; 1655 } 1656 1657 /* 1658 * This function creates a split out lock class for each invocation; 1659 * this is needed for now since a whole lot of users of the skb-queue 1660 * infrastructure in drivers have different locking usage (in hardirq) 1661 * than the networking core (in softirq only). In the long run either the 1662 * network layer or drivers should need annotation to consolidate the 1663 * main types of usage into 3 classes. 1664 */ 1665 static inline void skb_queue_head_init(struct sk_buff_head *list) 1666 { 1667 spin_lock_init(&list->lock); 1668 __skb_queue_head_init(list); 1669 } 1670 1671 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1672 struct lock_class_key *class) 1673 { 1674 skb_queue_head_init(list); 1675 lockdep_set_class(&list->lock, class); 1676 } 1677 1678 /* 1679 * Insert an sk_buff on a list. 1680 * 1681 * The "__skb_xxxx()" functions are the non-atomic ones that 1682 * can only be called with interrupts disabled. 1683 */ 1684 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1685 struct sk_buff_head *list); 1686 static inline void __skb_insert(struct sk_buff *newsk, 1687 struct sk_buff *prev, struct sk_buff *next, 1688 struct sk_buff_head *list) 1689 { 1690 newsk->next = next; 1691 newsk->prev = prev; 1692 next->prev = prev->next = newsk; 1693 list->qlen++; 1694 } 1695 1696 static inline void __skb_queue_splice(const struct sk_buff_head *list, 1697 struct sk_buff *prev, 1698 struct sk_buff *next) 1699 { 1700 struct sk_buff *first = list->next; 1701 struct sk_buff *last = list->prev; 1702 1703 first->prev = prev; 1704 prev->next = first; 1705 1706 last->next = next; 1707 next->prev = last; 1708 } 1709 1710 /** 1711 * skb_queue_splice - join two skb lists, this is designed for stacks 1712 * @list: the new list to add 1713 * @head: the place to add it in the first list 1714 */ 1715 static inline void skb_queue_splice(const struct sk_buff_head *list, 1716 struct sk_buff_head *head) 1717 { 1718 if (!skb_queue_empty(list)) { 1719 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1720 head->qlen += list->qlen; 1721 } 1722 } 1723 1724 /** 1725 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1726 * @list: the new list to add 1727 * @head: the place to add it in the first list 1728 * 1729 * The list at @list is reinitialised 1730 */ 1731 static inline void skb_queue_splice_init(struct sk_buff_head *list, 1732 struct sk_buff_head *head) 1733 { 1734 if (!skb_queue_empty(list)) { 1735 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1736 head->qlen += list->qlen; 1737 __skb_queue_head_init(list); 1738 } 1739 } 1740 1741 /** 1742 * skb_queue_splice_tail - join two skb lists, each list being a queue 1743 * @list: the new list to add 1744 * @head: the place to add it in the first list 1745 */ 1746 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1747 struct sk_buff_head *head) 1748 { 1749 if (!skb_queue_empty(list)) { 1750 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1751 head->qlen += list->qlen; 1752 } 1753 } 1754 1755 /** 1756 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1757 * @list: the new list to add 1758 * @head: the place to add it in the first list 1759 * 1760 * Each of the lists is a queue. 1761 * The list at @list is reinitialised 1762 */ 1763 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1764 struct sk_buff_head *head) 1765 { 1766 if (!skb_queue_empty(list)) { 1767 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1768 head->qlen += list->qlen; 1769 __skb_queue_head_init(list); 1770 } 1771 } 1772 1773 /** 1774 * __skb_queue_after - queue a buffer at the list head 1775 * @list: list to use 1776 * @prev: place after this buffer 1777 * @newsk: buffer to queue 1778 * 1779 * Queue a buffer int the middle of a list. This function takes no locks 1780 * and you must therefore hold required locks before calling it. 1781 * 1782 * A buffer cannot be placed on two lists at the same time. 1783 */ 1784 static inline void __skb_queue_after(struct sk_buff_head *list, 1785 struct sk_buff *prev, 1786 struct sk_buff *newsk) 1787 { 1788 __skb_insert(newsk, prev, prev->next, list); 1789 } 1790 1791 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1792 struct sk_buff_head *list); 1793 1794 static inline void __skb_queue_before(struct sk_buff_head *list, 1795 struct sk_buff *next, 1796 struct sk_buff *newsk) 1797 { 1798 __skb_insert(newsk, next->prev, next, list); 1799 } 1800 1801 /** 1802 * __skb_queue_head - queue a buffer at the list head 1803 * @list: list to use 1804 * @newsk: buffer to queue 1805 * 1806 * Queue a buffer at the start of a list. This function takes no locks 1807 * and you must therefore hold required locks before calling it. 1808 * 1809 * A buffer cannot be placed on two lists at the same time. 1810 */ 1811 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1812 static inline void __skb_queue_head(struct sk_buff_head *list, 1813 struct sk_buff *newsk) 1814 { 1815 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1816 } 1817 1818 /** 1819 * __skb_queue_tail - queue a buffer at the list tail 1820 * @list: list to use 1821 * @newsk: buffer to queue 1822 * 1823 * Queue a buffer at the end of a list. This function takes no locks 1824 * and you must therefore hold required locks before calling it. 1825 * 1826 * A buffer cannot be placed on two lists at the same time. 1827 */ 1828 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1829 static inline void __skb_queue_tail(struct sk_buff_head *list, 1830 struct sk_buff *newsk) 1831 { 1832 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1833 } 1834 1835 /* 1836 * remove sk_buff from list. _Must_ be called atomically, and with 1837 * the list known.. 1838 */ 1839 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1840 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1841 { 1842 struct sk_buff *next, *prev; 1843 1844 list->qlen--; 1845 next = skb->next; 1846 prev = skb->prev; 1847 skb->next = skb->prev = NULL; 1848 next->prev = prev; 1849 prev->next = next; 1850 } 1851 1852 /** 1853 * __skb_dequeue - remove from the head of the queue 1854 * @list: list to dequeue from 1855 * 1856 * Remove the head of the list. This function does not take any locks 1857 * so must be used with appropriate locks held only. The head item is 1858 * returned or %NULL if the list is empty. 1859 */ 1860 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1861 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1862 { 1863 struct sk_buff *skb = skb_peek(list); 1864 if (skb) 1865 __skb_unlink(skb, list); 1866 return skb; 1867 } 1868 1869 /** 1870 * __skb_dequeue_tail - remove from the tail of the queue 1871 * @list: list to dequeue from 1872 * 1873 * Remove the tail of the list. This function does not take any locks 1874 * so must be used with appropriate locks held only. The tail item is 1875 * returned or %NULL if the list is empty. 1876 */ 1877 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1878 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1879 { 1880 struct sk_buff *skb = skb_peek_tail(list); 1881 if (skb) 1882 __skb_unlink(skb, list); 1883 return skb; 1884 } 1885 1886 1887 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1888 { 1889 return skb->data_len; 1890 } 1891 1892 static inline unsigned int skb_headlen(const struct sk_buff *skb) 1893 { 1894 return skb->len - skb->data_len; 1895 } 1896 1897 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 1898 { 1899 unsigned int i, len = 0; 1900 1901 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 1902 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1903 return len; 1904 } 1905 1906 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 1907 { 1908 return skb_headlen(skb) + __skb_pagelen(skb); 1909 } 1910 1911 /** 1912 * __skb_fill_page_desc - initialise a paged fragment in an skb 1913 * @skb: buffer containing fragment to be initialised 1914 * @i: paged fragment index to initialise 1915 * @page: the page to use for this fragment 1916 * @off: the offset to the data with @page 1917 * @size: the length of the data 1918 * 1919 * Initialises the @i'th fragment of @skb to point to &size bytes at 1920 * offset @off within @page. 1921 * 1922 * Does not take any additional reference on the fragment. 1923 */ 1924 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1925 struct page *page, int off, int size) 1926 { 1927 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1928 1929 /* 1930 * Propagate page pfmemalloc to the skb if we can. The problem is 1931 * that not all callers have unique ownership of the page but rely 1932 * on page_is_pfmemalloc doing the right thing(tm). 1933 */ 1934 frag->page.p = page; 1935 frag->page_offset = off; 1936 skb_frag_size_set(frag, size); 1937 1938 page = compound_head(page); 1939 if (page_is_pfmemalloc(page)) 1940 skb->pfmemalloc = true; 1941 } 1942 1943 /** 1944 * skb_fill_page_desc - initialise a paged fragment in an skb 1945 * @skb: buffer containing fragment to be initialised 1946 * @i: paged fragment index to initialise 1947 * @page: the page to use for this fragment 1948 * @off: the offset to the data with @page 1949 * @size: the length of the data 1950 * 1951 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1952 * @skb to point to @size bytes at offset @off within @page. In 1953 * addition updates @skb such that @i is the last fragment. 1954 * 1955 * Does not take any additional reference on the fragment. 1956 */ 1957 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1958 struct page *page, int off, int size) 1959 { 1960 __skb_fill_page_desc(skb, i, page, off, size); 1961 skb_shinfo(skb)->nr_frags = i + 1; 1962 } 1963 1964 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1965 int size, unsigned int truesize); 1966 1967 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1968 unsigned int truesize); 1969 1970 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1971 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1972 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1973 1974 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1975 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1976 { 1977 return skb->head + skb->tail; 1978 } 1979 1980 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1981 { 1982 skb->tail = skb->data - skb->head; 1983 } 1984 1985 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1986 { 1987 skb_reset_tail_pointer(skb); 1988 skb->tail += offset; 1989 } 1990 1991 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 1992 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1993 { 1994 return skb->tail; 1995 } 1996 1997 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1998 { 1999 skb->tail = skb->data; 2000 } 2001 2002 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2003 { 2004 skb->tail = skb->data + offset; 2005 } 2006 2007 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2008 2009 /* 2010 * Add data to an sk_buff 2011 */ 2012 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2013 void *skb_put(struct sk_buff *skb, unsigned int len); 2014 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2015 { 2016 void *tmp = skb_tail_pointer(skb); 2017 SKB_LINEAR_ASSERT(skb); 2018 skb->tail += len; 2019 skb->len += len; 2020 return tmp; 2021 } 2022 2023 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2024 { 2025 void *tmp = __skb_put(skb, len); 2026 2027 memset(tmp, 0, len); 2028 return tmp; 2029 } 2030 2031 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2032 unsigned int len) 2033 { 2034 void *tmp = __skb_put(skb, len); 2035 2036 memcpy(tmp, data, len); 2037 return tmp; 2038 } 2039 2040 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2041 { 2042 *(u8 *)__skb_put(skb, 1) = val; 2043 } 2044 2045 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2046 { 2047 void *tmp = skb_put(skb, len); 2048 2049 memset(tmp, 0, len); 2050 2051 return tmp; 2052 } 2053 2054 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2055 unsigned int len) 2056 { 2057 void *tmp = skb_put(skb, len); 2058 2059 memcpy(tmp, data, len); 2060 2061 return tmp; 2062 } 2063 2064 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2065 { 2066 *(u8 *)skb_put(skb, 1) = val; 2067 } 2068 2069 void *skb_push(struct sk_buff *skb, unsigned int len); 2070 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2071 { 2072 skb->data -= len; 2073 skb->len += len; 2074 return skb->data; 2075 } 2076 2077 void *skb_pull(struct sk_buff *skb, unsigned int len); 2078 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2079 { 2080 skb->len -= len; 2081 BUG_ON(skb->len < skb->data_len); 2082 return skb->data += len; 2083 } 2084 2085 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2086 { 2087 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2088 } 2089 2090 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2091 2092 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len) 2093 { 2094 if (len > skb_headlen(skb) && 2095 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 2096 return NULL; 2097 skb->len -= len; 2098 return skb->data += len; 2099 } 2100 2101 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2102 { 2103 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 2104 } 2105 2106 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 2107 { 2108 if (likely(len <= skb_headlen(skb))) 2109 return 1; 2110 if (unlikely(len > skb->len)) 2111 return 0; 2112 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 2113 } 2114 2115 void skb_condense(struct sk_buff *skb); 2116 2117 /** 2118 * skb_headroom - bytes at buffer head 2119 * @skb: buffer to check 2120 * 2121 * Return the number of bytes of free space at the head of an &sk_buff. 2122 */ 2123 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2124 { 2125 return skb->data - skb->head; 2126 } 2127 2128 /** 2129 * skb_tailroom - bytes at buffer end 2130 * @skb: buffer to check 2131 * 2132 * Return the number of bytes of free space at the tail of an sk_buff 2133 */ 2134 static inline int skb_tailroom(const struct sk_buff *skb) 2135 { 2136 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2137 } 2138 2139 /** 2140 * skb_availroom - bytes at buffer end 2141 * @skb: buffer to check 2142 * 2143 * Return the number of bytes of free space at the tail of an sk_buff 2144 * allocated by sk_stream_alloc() 2145 */ 2146 static inline int skb_availroom(const struct sk_buff *skb) 2147 { 2148 if (skb_is_nonlinear(skb)) 2149 return 0; 2150 2151 return skb->end - skb->tail - skb->reserved_tailroom; 2152 } 2153 2154 /** 2155 * skb_reserve - adjust headroom 2156 * @skb: buffer to alter 2157 * @len: bytes to move 2158 * 2159 * Increase the headroom of an empty &sk_buff by reducing the tail 2160 * room. This is only allowed for an empty buffer. 2161 */ 2162 static inline void skb_reserve(struct sk_buff *skb, int len) 2163 { 2164 skb->data += len; 2165 skb->tail += len; 2166 } 2167 2168 /** 2169 * skb_tailroom_reserve - adjust reserved_tailroom 2170 * @skb: buffer to alter 2171 * @mtu: maximum amount of headlen permitted 2172 * @needed_tailroom: minimum amount of reserved_tailroom 2173 * 2174 * Set reserved_tailroom so that headlen can be as large as possible but 2175 * not larger than mtu and tailroom cannot be smaller than 2176 * needed_tailroom. 2177 * The required headroom should already have been reserved before using 2178 * this function. 2179 */ 2180 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2181 unsigned int needed_tailroom) 2182 { 2183 SKB_LINEAR_ASSERT(skb); 2184 if (mtu < skb_tailroom(skb) - needed_tailroom) 2185 /* use at most mtu */ 2186 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2187 else 2188 /* use up to all available space */ 2189 skb->reserved_tailroom = needed_tailroom; 2190 } 2191 2192 #define ENCAP_TYPE_ETHER 0 2193 #define ENCAP_TYPE_IPPROTO 1 2194 2195 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2196 __be16 protocol) 2197 { 2198 skb->inner_protocol = protocol; 2199 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2200 } 2201 2202 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2203 __u8 ipproto) 2204 { 2205 skb->inner_ipproto = ipproto; 2206 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2207 } 2208 2209 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2210 { 2211 skb->inner_mac_header = skb->mac_header; 2212 skb->inner_network_header = skb->network_header; 2213 skb->inner_transport_header = skb->transport_header; 2214 } 2215 2216 static inline void skb_reset_mac_len(struct sk_buff *skb) 2217 { 2218 skb->mac_len = skb->network_header - skb->mac_header; 2219 } 2220 2221 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2222 *skb) 2223 { 2224 return skb->head + skb->inner_transport_header; 2225 } 2226 2227 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2228 { 2229 return skb_inner_transport_header(skb) - skb->data; 2230 } 2231 2232 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2233 { 2234 skb->inner_transport_header = skb->data - skb->head; 2235 } 2236 2237 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2238 const int offset) 2239 { 2240 skb_reset_inner_transport_header(skb); 2241 skb->inner_transport_header += offset; 2242 } 2243 2244 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2245 { 2246 return skb->head + skb->inner_network_header; 2247 } 2248 2249 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2250 { 2251 skb->inner_network_header = skb->data - skb->head; 2252 } 2253 2254 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2255 const int offset) 2256 { 2257 skb_reset_inner_network_header(skb); 2258 skb->inner_network_header += offset; 2259 } 2260 2261 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2262 { 2263 return skb->head + skb->inner_mac_header; 2264 } 2265 2266 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2267 { 2268 skb->inner_mac_header = skb->data - skb->head; 2269 } 2270 2271 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2272 const int offset) 2273 { 2274 skb_reset_inner_mac_header(skb); 2275 skb->inner_mac_header += offset; 2276 } 2277 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2278 { 2279 return skb->transport_header != (typeof(skb->transport_header))~0U; 2280 } 2281 2282 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2283 { 2284 return skb->head + skb->transport_header; 2285 } 2286 2287 static inline void skb_reset_transport_header(struct sk_buff *skb) 2288 { 2289 skb->transport_header = skb->data - skb->head; 2290 } 2291 2292 static inline void skb_set_transport_header(struct sk_buff *skb, 2293 const int offset) 2294 { 2295 skb_reset_transport_header(skb); 2296 skb->transport_header += offset; 2297 } 2298 2299 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2300 { 2301 return skb->head + skb->network_header; 2302 } 2303 2304 static inline void skb_reset_network_header(struct sk_buff *skb) 2305 { 2306 skb->network_header = skb->data - skb->head; 2307 } 2308 2309 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2310 { 2311 skb_reset_network_header(skb); 2312 skb->network_header += offset; 2313 } 2314 2315 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2316 { 2317 return skb->head + skb->mac_header; 2318 } 2319 2320 static inline int skb_mac_offset(const struct sk_buff *skb) 2321 { 2322 return skb_mac_header(skb) - skb->data; 2323 } 2324 2325 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2326 { 2327 return skb->network_header - skb->mac_header; 2328 } 2329 2330 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2331 { 2332 return skb->mac_header != (typeof(skb->mac_header))~0U; 2333 } 2334 2335 static inline void skb_reset_mac_header(struct sk_buff *skb) 2336 { 2337 skb->mac_header = skb->data - skb->head; 2338 } 2339 2340 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2341 { 2342 skb_reset_mac_header(skb); 2343 skb->mac_header += offset; 2344 } 2345 2346 static inline void skb_pop_mac_header(struct sk_buff *skb) 2347 { 2348 skb->mac_header = skb->network_header; 2349 } 2350 2351 static inline void skb_probe_transport_header(struct sk_buff *skb, 2352 const int offset_hint) 2353 { 2354 struct flow_keys_basic keys; 2355 2356 if (skb_transport_header_was_set(skb)) 2357 return; 2358 2359 if (skb_flow_dissect_flow_keys_basic(skb, &keys, 0, 0, 0, 0, 0)) 2360 skb_set_transport_header(skb, keys.control.thoff); 2361 else 2362 skb_set_transport_header(skb, offset_hint); 2363 } 2364 2365 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2366 { 2367 if (skb_mac_header_was_set(skb)) { 2368 const unsigned char *old_mac = skb_mac_header(skb); 2369 2370 skb_set_mac_header(skb, -skb->mac_len); 2371 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2372 } 2373 } 2374 2375 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2376 { 2377 return skb->csum_start - skb_headroom(skb); 2378 } 2379 2380 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2381 { 2382 return skb->head + skb->csum_start; 2383 } 2384 2385 static inline int skb_transport_offset(const struct sk_buff *skb) 2386 { 2387 return skb_transport_header(skb) - skb->data; 2388 } 2389 2390 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2391 { 2392 return skb->transport_header - skb->network_header; 2393 } 2394 2395 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2396 { 2397 return skb->inner_transport_header - skb->inner_network_header; 2398 } 2399 2400 static inline int skb_network_offset(const struct sk_buff *skb) 2401 { 2402 return skb_network_header(skb) - skb->data; 2403 } 2404 2405 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2406 { 2407 return skb_inner_network_header(skb) - skb->data; 2408 } 2409 2410 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2411 { 2412 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2413 } 2414 2415 /* 2416 * CPUs often take a performance hit when accessing unaligned memory 2417 * locations. The actual performance hit varies, it can be small if the 2418 * hardware handles it or large if we have to take an exception and fix it 2419 * in software. 2420 * 2421 * Since an ethernet header is 14 bytes network drivers often end up with 2422 * the IP header at an unaligned offset. The IP header can be aligned by 2423 * shifting the start of the packet by 2 bytes. Drivers should do this 2424 * with: 2425 * 2426 * skb_reserve(skb, NET_IP_ALIGN); 2427 * 2428 * The downside to this alignment of the IP header is that the DMA is now 2429 * unaligned. On some architectures the cost of an unaligned DMA is high 2430 * and this cost outweighs the gains made by aligning the IP header. 2431 * 2432 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2433 * to be overridden. 2434 */ 2435 #ifndef NET_IP_ALIGN 2436 #define NET_IP_ALIGN 2 2437 #endif 2438 2439 /* 2440 * The networking layer reserves some headroom in skb data (via 2441 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2442 * the header has to grow. In the default case, if the header has to grow 2443 * 32 bytes or less we avoid the reallocation. 2444 * 2445 * Unfortunately this headroom changes the DMA alignment of the resulting 2446 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2447 * on some architectures. An architecture can override this value, 2448 * perhaps setting it to a cacheline in size (since that will maintain 2449 * cacheline alignment of the DMA). It must be a power of 2. 2450 * 2451 * Various parts of the networking layer expect at least 32 bytes of 2452 * headroom, you should not reduce this. 2453 * 2454 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2455 * to reduce average number of cache lines per packet. 2456 * get_rps_cpus() for example only access one 64 bytes aligned block : 2457 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2458 */ 2459 #ifndef NET_SKB_PAD 2460 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2461 #endif 2462 2463 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2464 2465 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 2466 { 2467 if (unlikely(skb_is_nonlinear(skb))) { 2468 WARN_ON(1); 2469 return; 2470 } 2471 skb->len = len; 2472 skb_set_tail_pointer(skb, len); 2473 } 2474 2475 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2476 { 2477 __skb_set_length(skb, len); 2478 } 2479 2480 void skb_trim(struct sk_buff *skb, unsigned int len); 2481 2482 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2483 { 2484 if (skb->data_len) 2485 return ___pskb_trim(skb, len); 2486 __skb_trim(skb, len); 2487 return 0; 2488 } 2489 2490 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2491 { 2492 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2493 } 2494 2495 /** 2496 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2497 * @skb: buffer to alter 2498 * @len: new length 2499 * 2500 * This is identical to pskb_trim except that the caller knows that 2501 * the skb is not cloned so we should never get an error due to out- 2502 * of-memory. 2503 */ 2504 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2505 { 2506 int err = pskb_trim(skb, len); 2507 BUG_ON(err); 2508 } 2509 2510 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 2511 { 2512 unsigned int diff = len - skb->len; 2513 2514 if (skb_tailroom(skb) < diff) { 2515 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 2516 GFP_ATOMIC); 2517 if (ret) 2518 return ret; 2519 } 2520 __skb_set_length(skb, len); 2521 return 0; 2522 } 2523 2524 /** 2525 * skb_orphan - orphan a buffer 2526 * @skb: buffer to orphan 2527 * 2528 * If a buffer currently has an owner then we call the owner's 2529 * destructor function and make the @skb unowned. The buffer continues 2530 * to exist but is no longer charged to its former owner. 2531 */ 2532 static inline void skb_orphan(struct sk_buff *skb) 2533 { 2534 if (skb->destructor) { 2535 skb->destructor(skb); 2536 skb->destructor = NULL; 2537 skb->sk = NULL; 2538 } else { 2539 BUG_ON(skb->sk); 2540 } 2541 } 2542 2543 /** 2544 * skb_orphan_frags - orphan the frags contained in a buffer 2545 * @skb: buffer to orphan frags from 2546 * @gfp_mask: allocation mask for replacement pages 2547 * 2548 * For each frag in the SKB which needs a destructor (i.e. has an 2549 * owner) create a copy of that frag and release the original 2550 * page by calling the destructor. 2551 */ 2552 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2553 { 2554 if (likely(!skb_zcopy(skb))) 2555 return 0; 2556 if (skb_uarg(skb)->callback == sock_zerocopy_callback) 2557 return 0; 2558 return skb_copy_ubufs(skb, gfp_mask); 2559 } 2560 2561 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 2562 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 2563 { 2564 if (likely(!skb_zcopy(skb))) 2565 return 0; 2566 return skb_copy_ubufs(skb, gfp_mask); 2567 } 2568 2569 /** 2570 * __skb_queue_purge - empty a list 2571 * @list: list to empty 2572 * 2573 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2574 * the list and one reference dropped. This function does not take the 2575 * list lock and the caller must hold the relevant locks to use it. 2576 */ 2577 void skb_queue_purge(struct sk_buff_head *list); 2578 static inline void __skb_queue_purge(struct sk_buff_head *list) 2579 { 2580 struct sk_buff *skb; 2581 while ((skb = __skb_dequeue(list)) != NULL) 2582 kfree_skb(skb); 2583 } 2584 2585 void skb_rbtree_purge(struct rb_root *root); 2586 2587 void *netdev_alloc_frag(unsigned int fragsz); 2588 2589 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2590 gfp_t gfp_mask); 2591 2592 /** 2593 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2594 * @dev: network device to receive on 2595 * @length: length to allocate 2596 * 2597 * Allocate a new &sk_buff and assign it a usage count of one. The 2598 * buffer has unspecified headroom built in. Users should allocate 2599 * the headroom they think they need without accounting for the 2600 * built in space. The built in space is used for optimisations. 2601 * 2602 * %NULL is returned if there is no free memory. Although this function 2603 * allocates memory it can be called from an interrupt. 2604 */ 2605 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2606 unsigned int length) 2607 { 2608 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2609 } 2610 2611 /* legacy helper around __netdev_alloc_skb() */ 2612 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2613 gfp_t gfp_mask) 2614 { 2615 return __netdev_alloc_skb(NULL, length, gfp_mask); 2616 } 2617 2618 /* legacy helper around netdev_alloc_skb() */ 2619 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2620 { 2621 return netdev_alloc_skb(NULL, length); 2622 } 2623 2624 2625 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2626 unsigned int length, gfp_t gfp) 2627 { 2628 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2629 2630 if (NET_IP_ALIGN && skb) 2631 skb_reserve(skb, NET_IP_ALIGN); 2632 return skb; 2633 } 2634 2635 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2636 unsigned int length) 2637 { 2638 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2639 } 2640 2641 static inline void skb_free_frag(void *addr) 2642 { 2643 page_frag_free(addr); 2644 } 2645 2646 void *napi_alloc_frag(unsigned int fragsz); 2647 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2648 unsigned int length, gfp_t gfp_mask); 2649 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2650 unsigned int length) 2651 { 2652 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2653 } 2654 void napi_consume_skb(struct sk_buff *skb, int budget); 2655 2656 void __kfree_skb_flush(void); 2657 void __kfree_skb_defer(struct sk_buff *skb); 2658 2659 /** 2660 * __dev_alloc_pages - allocate page for network Rx 2661 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2662 * @order: size of the allocation 2663 * 2664 * Allocate a new page. 2665 * 2666 * %NULL is returned if there is no free memory. 2667 */ 2668 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2669 unsigned int order) 2670 { 2671 /* This piece of code contains several assumptions. 2672 * 1. This is for device Rx, therefor a cold page is preferred. 2673 * 2. The expectation is the user wants a compound page. 2674 * 3. If requesting a order 0 page it will not be compound 2675 * due to the check to see if order has a value in prep_new_page 2676 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2677 * code in gfp_to_alloc_flags that should be enforcing this. 2678 */ 2679 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 2680 2681 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2682 } 2683 2684 static inline struct page *dev_alloc_pages(unsigned int order) 2685 { 2686 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 2687 } 2688 2689 /** 2690 * __dev_alloc_page - allocate a page for network Rx 2691 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2692 * 2693 * Allocate a new page. 2694 * 2695 * %NULL is returned if there is no free memory. 2696 */ 2697 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2698 { 2699 return __dev_alloc_pages(gfp_mask, 0); 2700 } 2701 2702 static inline struct page *dev_alloc_page(void) 2703 { 2704 return dev_alloc_pages(0); 2705 } 2706 2707 /** 2708 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2709 * @page: The page that was allocated from skb_alloc_page 2710 * @skb: The skb that may need pfmemalloc set 2711 */ 2712 static inline void skb_propagate_pfmemalloc(struct page *page, 2713 struct sk_buff *skb) 2714 { 2715 if (page_is_pfmemalloc(page)) 2716 skb->pfmemalloc = true; 2717 } 2718 2719 /** 2720 * skb_frag_page - retrieve the page referred to by a paged fragment 2721 * @frag: the paged fragment 2722 * 2723 * Returns the &struct page associated with @frag. 2724 */ 2725 static inline struct page *skb_frag_page(const skb_frag_t *frag) 2726 { 2727 return frag->page.p; 2728 } 2729 2730 /** 2731 * __skb_frag_ref - take an addition reference on a paged fragment. 2732 * @frag: the paged fragment 2733 * 2734 * Takes an additional reference on the paged fragment @frag. 2735 */ 2736 static inline void __skb_frag_ref(skb_frag_t *frag) 2737 { 2738 get_page(skb_frag_page(frag)); 2739 } 2740 2741 /** 2742 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2743 * @skb: the buffer 2744 * @f: the fragment offset. 2745 * 2746 * Takes an additional reference on the @f'th paged fragment of @skb. 2747 */ 2748 static inline void skb_frag_ref(struct sk_buff *skb, int f) 2749 { 2750 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2751 } 2752 2753 /** 2754 * __skb_frag_unref - release a reference on a paged fragment. 2755 * @frag: the paged fragment 2756 * 2757 * Releases a reference on the paged fragment @frag. 2758 */ 2759 static inline void __skb_frag_unref(skb_frag_t *frag) 2760 { 2761 put_page(skb_frag_page(frag)); 2762 } 2763 2764 /** 2765 * skb_frag_unref - release a reference on a paged fragment of an skb. 2766 * @skb: the buffer 2767 * @f: the fragment offset 2768 * 2769 * Releases a reference on the @f'th paged fragment of @skb. 2770 */ 2771 static inline void skb_frag_unref(struct sk_buff *skb, int f) 2772 { 2773 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2774 } 2775 2776 /** 2777 * skb_frag_address - gets the address of the data contained in a paged fragment 2778 * @frag: the paged fragment buffer 2779 * 2780 * Returns the address of the data within @frag. The page must already 2781 * be mapped. 2782 */ 2783 static inline void *skb_frag_address(const skb_frag_t *frag) 2784 { 2785 return page_address(skb_frag_page(frag)) + frag->page_offset; 2786 } 2787 2788 /** 2789 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2790 * @frag: the paged fragment buffer 2791 * 2792 * Returns the address of the data within @frag. Checks that the page 2793 * is mapped and returns %NULL otherwise. 2794 */ 2795 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2796 { 2797 void *ptr = page_address(skb_frag_page(frag)); 2798 if (unlikely(!ptr)) 2799 return NULL; 2800 2801 return ptr + frag->page_offset; 2802 } 2803 2804 /** 2805 * __skb_frag_set_page - sets the page contained in a paged fragment 2806 * @frag: the paged fragment 2807 * @page: the page to set 2808 * 2809 * Sets the fragment @frag to contain @page. 2810 */ 2811 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2812 { 2813 frag->page.p = page; 2814 } 2815 2816 /** 2817 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2818 * @skb: the buffer 2819 * @f: the fragment offset 2820 * @page: the page to set 2821 * 2822 * Sets the @f'th fragment of @skb to contain @page. 2823 */ 2824 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2825 struct page *page) 2826 { 2827 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2828 } 2829 2830 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2831 2832 /** 2833 * skb_frag_dma_map - maps a paged fragment via the DMA API 2834 * @dev: the device to map the fragment to 2835 * @frag: the paged fragment to map 2836 * @offset: the offset within the fragment (starting at the 2837 * fragment's own offset) 2838 * @size: the number of bytes to map 2839 * @dir: the direction of the mapping (``PCI_DMA_*``) 2840 * 2841 * Maps the page associated with @frag to @device. 2842 */ 2843 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2844 const skb_frag_t *frag, 2845 size_t offset, size_t size, 2846 enum dma_data_direction dir) 2847 { 2848 return dma_map_page(dev, skb_frag_page(frag), 2849 frag->page_offset + offset, size, dir); 2850 } 2851 2852 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2853 gfp_t gfp_mask) 2854 { 2855 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2856 } 2857 2858 2859 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2860 gfp_t gfp_mask) 2861 { 2862 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2863 } 2864 2865 2866 /** 2867 * skb_clone_writable - is the header of a clone writable 2868 * @skb: buffer to check 2869 * @len: length up to which to write 2870 * 2871 * Returns true if modifying the header part of the cloned buffer 2872 * does not requires the data to be copied. 2873 */ 2874 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2875 { 2876 return !skb_header_cloned(skb) && 2877 skb_headroom(skb) + len <= skb->hdr_len; 2878 } 2879 2880 static inline int skb_try_make_writable(struct sk_buff *skb, 2881 unsigned int write_len) 2882 { 2883 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 2884 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2885 } 2886 2887 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2888 int cloned) 2889 { 2890 int delta = 0; 2891 2892 if (headroom > skb_headroom(skb)) 2893 delta = headroom - skb_headroom(skb); 2894 2895 if (delta || cloned) 2896 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2897 GFP_ATOMIC); 2898 return 0; 2899 } 2900 2901 /** 2902 * skb_cow - copy header of skb when it is required 2903 * @skb: buffer to cow 2904 * @headroom: needed headroom 2905 * 2906 * If the skb passed lacks sufficient headroom or its data part 2907 * is shared, data is reallocated. If reallocation fails, an error 2908 * is returned and original skb is not changed. 2909 * 2910 * The result is skb with writable area skb->head...skb->tail 2911 * and at least @headroom of space at head. 2912 */ 2913 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2914 { 2915 return __skb_cow(skb, headroom, skb_cloned(skb)); 2916 } 2917 2918 /** 2919 * skb_cow_head - skb_cow but only making the head writable 2920 * @skb: buffer to cow 2921 * @headroom: needed headroom 2922 * 2923 * This function is identical to skb_cow except that we replace the 2924 * skb_cloned check by skb_header_cloned. It should be used when 2925 * you only need to push on some header and do not need to modify 2926 * the data. 2927 */ 2928 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2929 { 2930 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2931 } 2932 2933 /** 2934 * skb_padto - pad an skbuff up to a minimal size 2935 * @skb: buffer to pad 2936 * @len: minimal length 2937 * 2938 * Pads up a buffer to ensure the trailing bytes exist and are 2939 * blanked. If the buffer already contains sufficient data it 2940 * is untouched. Otherwise it is extended. Returns zero on 2941 * success. The skb is freed on error. 2942 */ 2943 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2944 { 2945 unsigned int size = skb->len; 2946 if (likely(size >= len)) 2947 return 0; 2948 return skb_pad(skb, len - size); 2949 } 2950 2951 /** 2952 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2953 * @skb: buffer to pad 2954 * @len: minimal length 2955 * @free_on_error: free buffer on error 2956 * 2957 * Pads up a buffer to ensure the trailing bytes exist and are 2958 * blanked. If the buffer already contains sufficient data it 2959 * is untouched. Otherwise it is extended. Returns zero on 2960 * success. The skb is freed on error if @free_on_error is true. 2961 */ 2962 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len, 2963 bool free_on_error) 2964 { 2965 unsigned int size = skb->len; 2966 2967 if (unlikely(size < len)) { 2968 len -= size; 2969 if (__skb_pad(skb, len, free_on_error)) 2970 return -ENOMEM; 2971 __skb_put(skb, len); 2972 } 2973 return 0; 2974 } 2975 2976 /** 2977 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2978 * @skb: buffer to pad 2979 * @len: minimal length 2980 * 2981 * Pads up a buffer to ensure the trailing bytes exist and are 2982 * blanked. If the buffer already contains sufficient data it 2983 * is untouched. Otherwise it is extended. Returns zero on 2984 * success. The skb is freed on error. 2985 */ 2986 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2987 { 2988 return __skb_put_padto(skb, len, true); 2989 } 2990 2991 static inline int skb_add_data(struct sk_buff *skb, 2992 struct iov_iter *from, int copy) 2993 { 2994 const int off = skb->len; 2995 2996 if (skb->ip_summed == CHECKSUM_NONE) { 2997 __wsum csum = 0; 2998 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 2999 &csum, from)) { 3000 skb->csum = csum_block_add(skb->csum, csum, off); 3001 return 0; 3002 } 3003 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3004 return 0; 3005 3006 __skb_trim(skb, off); 3007 return -EFAULT; 3008 } 3009 3010 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3011 const struct page *page, int off) 3012 { 3013 if (skb_zcopy(skb)) 3014 return false; 3015 if (i) { 3016 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 3017 3018 return page == skb_frag_page(frag) && 3019 off == frag->page_offset + skb_frag_size(frag); 3020 } 3021 return false; 3022 } 3023 3024 static inline int __skb_linearize(struct sk_buff *skb) 3025 { 3026 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3027 } 3028 3029 /** 3030 * skb_linearize - convert paged skb to linear one 3031 * @skb: buffer to linarize 3032 * 3033 * If there is no free memory -ENOMEM is returned, otherwise zero 3034 * is returned and the old skb data released. 3035 */ 3036 static inline int skb_linearize(struct sk_buff *skb) 3037 { 3038 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3039 } 3040 3041 /** 3042 * skb_has_shared_frag - can any frag be overwritten 3043 * @skb: buffer to test 3044 * 3045 * Return true if the skb has at least one frag that might be modified 3046 * by an external entity (as in vmsplice()/sendfile()) 3047 */ 3048 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3049 { 3050 return skb_is_nonlinear(skb) && 3051 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 3052 } 3053 3054 /** 3055 * skb_linearize_cow - make sure skb is linear and writable 3056 * @skb: buffer to process 3057 * 3058 * If there is no free memory -ENOMEM is returned, otherwise zero 3059 * is returned and the old skb data released. 3060 */ 3061 static inline int skb_linearize_cow(struct sk_buff *skb) 3062 { 3063 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3064 __skb_linearize(skb) : 0; 3065 } 3066 3067 static __always_inline void 3068 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3069 unsigned int off) 3070 { 3071 if (skb->ip_summed == CHECKSUM_COMPLETE) 3072 skb->csum = csum_block_sub(skb->csum, 3073 csum_partial(start, len, 0), off); 3074 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3075 skb_checksum_start_offset(skb) < 0) 3076 skb->ip_summed = CHECKSUM_NONE; 3077 } 3078 3079 /** 3080 * skb_postpull_rcsum - update checksum for received skb after pull 3081 * @skb: buffer to update 3082 * @start: start of data before pull 3083 * @len: length of data pulled 3084 * 3085 * After doing a pull on a received packet, you need to call this to 3086 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3087 * CHECKSUM_NONE so that it can be recomputed from scratch. 3088 */ 3089 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3090 const void *start, unsigned int len) 3091 { 3092 __skb_postpull_rcsum(skb, start, len, 0); 3093 } 3094 3095 static __always_inline void 3096 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3097 unsigned int off) 3098 { 3099 if (skb->ip_summed == CHECKSUM_COMPLETE) 3100 skb->csum = csum_block_add(skb->csum, 3101 csum_partial(start, len, 0), off); 3102 } 3103 3104 /** 3105 * skb_postpush_rcsum - update checksum for received skb after push 3106 * @skb: buffer to update 3107 * @start: start of data after push 3108 * @len: length of data pushed 3109 * 3110 * After doing a push on a received packet, you need to call this to 3111 * update the CHECKSUM_COMPLETE checksum. 3112 */ 3113 static inline void skb_postpush_rcsum(struct sk_buff *skb, 3114 const void *start, unsigned int len) 3115 { 3116 __skb_postpush_rcsum(skb, start, len, 0); 3117 } 3118 3119 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3120 3121 /** 3122 * skb_push_rcsum - push skb and update receive checksum 3123 * @skb: buffer to update 3124 * @len: length of data pulled 3125 * 3126 * This function performs an skb_push on the packet and updates 3127 * the CHECKSUM_COMPLETE checksum. It should be used on 3128 * receive path processing instead of skb_push unless you know 3129 * that the checksum difference is zero (e.g., a valid IP header) 3130 * or you are setting ip_summed to CHECKSUM_NONE. 3131 */ 3132 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3133 { 3134 skb_push(skb, len); 3135 skb_postpush_rcsum(skb, skb->data, len); 3136 return skb->data; 3137 } 3138 3139 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 3140 /** 3141 * pskb_trim_rcsum - trim received skb and update checksum 3142 * @skb: buffer to trim 3143 * @len: new length 3144 * 3145 * This is exactly the same as pskb_trim except that it ensures the 3146 * checksum of received packets are still valid after the operation. 3147 */ 3148 3149 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3150 { 3151 if (likely(len >= skb->len)) 3152 return 0; 3153 return pskb_trim_rcsum_slow(skb, len); 3154 } 3155 3156 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3157 { 3158 if (skb->ip_summed == CHECKSUM_COMPLETE) 3159 skb->ip_summed = CHECKSUM_NONE; 3160 __skb_trim(skb, len); 3161 return 0; 3162 } 3163 3164 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3165 { 3166 if (skb->ip_summed == CHECKSUM_COMPLETE) 3167 skb->ip_summed = CHECKSUM_NONE; 3168 return __skb_grow(skb, len); 3169 } 3170 3171 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3172 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 3173 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 3174 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3175 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3176 3177 #define skb_queue_walk(queue, skb) \ 3178 for (skb = (queue)->next; \ 3179 skb != (struct sk_buff *)(queue); \ 3180 skb = skb->next) 3181 3182 #define skb_queue_walk_safe(queue, skb, tmp) \ 3183 for (skb = (queue)->next, tmp = skb->next; \ 3184 skb != (struct sk_buff *)(queue); \ 3185 skb = tmp, tmp = skb->next) 3186 3187 #define skb_queue_walk_from(queue, skb) \ 3188 for (; skb != (struct sk_buff *)(queue); \ 3189 skb = skb->next) 3190 3191 #define skb_rbtree_walk(skb, root) \ 3192 for (skb = skb_rb_first(root); skb != NULL; \ 3193 skb = skb_rb_next(skb)) 3194 3195 #define skb_rbtree_walk_from(skb) \ 3196 for (; skb != NULL; \ 3197 skb = skb_rb_next(skb)) 3198 3199 #define skb_rbtree_walk_from_safe(skb, tmp) \ 3200 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3201 skb = tmp) 3202 3203 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 3204 for (tmp = skb->next; \ 3205 skb != (struct sk_buff *)(queue); \ 3206 skb = tmp, tmp = skb->next) 3207 3208 #define skb_queue_reverse_walk(queue, skb) \ 3209 for (skb = (queue)->prev; \ 3210 skb != (struct sk_buff *)(queue); \ 3211 skb = skb->prev) 3212 3213 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3214 for (skb = (queue)->prev, tmp = skb->prev; \ 3215 skb != (struct sk_buff *)(queue); \ 3216 skb = tmp, tmp = skb->prev) 3217 3218 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3219 for (tmp = skb->prev; \ 3220 skb != (struct sk_buff *)(queue); \ 3221 skb = tmp, tmp = skb->prev) 3222 3223 static inline bool skb_has_frag_list(const struct sk_buff *skb) 3224 { 3225 return skb_shinfo(skb)->frag_list != NULL; 3226 } 3227 3228 static inline void skb_frag_list_init(struct sk_buff *skb) 3229 { 3230 skb_shinfo(skb)->frag_list = NULL; 3231 } 3232 3233 #define skb_walk_frags(skb, iter) \ 3234 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3235 3236 3237 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, 3238 const struct sk_buff *skb); 3239 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3240 struct sk_buff_head *queue, 3241 unsigned int flags, 3242 void (*destructor)(struct sock *sk, 3243 struct sk_buff *skb), 3244 int *peeked, int *off, int *err, 3245 struct sk_buff **last); 3246 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags, 3247 void (*destructor)(struct sock *sk, 3248 struct sk_buff *skb), 3249 int *peeked, int *off, int *err, 3250 struct sk_buff **last); 3251 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 3252 void (*destructor)(struct sock *sk, 3253 struct sk_buff *skb), 3254 int *peeked, int *off, int *err); 3255 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 3256 int *err); 3257 __poll_t datagram_poll(struct file *file, struct socket *sock, 3258 struct poll_table_struct *wait); 3259 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3260 struct iov_iter *to, int size); 3261 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3262 struct msghdr *msg, int size) 3263 { 3264 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3265 } 3266 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3267 struct msghdr *msg); 3268 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3269 struct iov_iter *from, int len); 3270 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3271 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3272 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3273 static inline void skb_free_datagram_locked(struct sock *sk, 3274 struct sk_buff *skb) 3275 { 3276 __skb_free_datagram_locked(sk, skb, 0); 3277 } 3278 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3279 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3280 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3281 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3282 int len, __wsum csum); 3283 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3284 struct pipe_inode_info *pipe, unsigned int len, 3285 unsigned int flags); 3286 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 3287 int len); 3288 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 3289 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3290 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3291 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3292 int len, int hlen); 3293 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3294 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3295 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3296 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu); 3297 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len); 3298 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3299 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3300 int skb_ensure_writable(struct sk_buff *skb, int write_len); 3301 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3302 int skb_vlan_pop(struct sk_buff *skb); 3303 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3304 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3305 gfp_t gfp); 3306 3307 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3308 { 3309 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 3310 } 3311 3312 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3313 { 3314 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3315 } 3316 3317 struct skb_checksum_ops { 3318 __wsum (*update)(const void *mem, int len, __wsum wsum); 3319 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3320 }; 3321 3322 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 3323 3324 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3325 __wsum csum, const struct skb_checksum_ops *ops); 3326 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3327 __wsum csum); 3328 3329 static inline void * __must_check 3330 __skb_header_pointer(const struct sk_buff *skb, int offset, 3331 int len, void *data, int hlen, void *buffer) 3332 { 3333 if (hlen - offset >= len) 3334 return data + offset; 3335 3336 if (!skb || 3337 skb_copy_bits(skb, offset, buffer, len) < 0) 3338 return NULL; 3339 3340 return buffer; 3341 } 3342 3343 static inline void * __must_check 3344 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 3345 { 3346 return __skb_header_pointer(skb, offset, len, skb->data, 3347 skb_headlen(skb), buffer); 3348 } 3349 3350 /** 3351 * skb_needs_linearize - check if we need to linearize a given skb 3352 * depending on the given device features. 3353 * @skb: socket buffer to check 3354 * @features: net device features 3355 * 3356 * Returns true if either: 3357 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3358 * 2. skb is fragmented and the device does not support SG. 3359 */ 3360 static inline bool skb_needs_linearize(struct sk_buff *skb, 3361 netdev_features_t features) 3362 { 3363 return skb_is_nonlinear(skb) && 3364 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3365 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3366 } 3367 3368 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3369 void *to, 3370 const unsigned int len) 3371 { 3372 memcpy(to, skb->data, len); 3373 } 3374 3375 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3376 const int offset, void *to, 3377 const unsigned int len) 3378 { 3379 memcpy(to, skb->data + offset, len); 3380 } 3381 3382 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3383 const void *from, 3384 const unsigned int len) 3385 { 3386 memcpy(skb->data, from, len); 3387 } 3388 3389 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3390 const int offset, 3391 const void *from, 3392 const unsigned int len) 3393 { 3394 memcpy(skb->data + offset, from, len); 3395 } 3396 3397 void skb_init(void); 3398 3399 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3400 { 3401 return skb->tstamp; 3402 } 3403 3404 /** 3405 * skb_get_timestamp - get timestamp from a skb 3406 * @skb: skb to get stamp from 3407 * @stamp: pointer to struct timeval to store stamp in 3408 * 3409 * Timestamps are stored in the skb as offsets to a base timestamp. 3410 * This function converts the offset back to a struct timeval and stores 3411 * it in stamp. 3412 */ 3413 static inline void skb_get_timestamp(const struct sk_buff *skb, 3414 struct timeval *stamp) 3415 { 3416 *stamp = ktime_to_timeval(skb->tstamp); 3417 } 3418 3419 static inline void skb_get_timestampns(const struct sk_buff *skb, 3420 struct timespec *stamp) 3421 { 3422 *stamp = ktime_to_timespec(skb->tstamp); 3423 } 3424 3425 static inline void __net_timestamp(struct sk_buff *skb) 3426 { 3427 skb->tstamp = ktime_get_real(); 3428 } 3429 3430 static inline ktime_t net_timedelta(ktime_t t) 3431 { 3432 return ktime_sub(ktime_get_real(), t); 3433 } 3434 3435 static inline ktime_t net_invalid_timestamp(void) 3436 { 3437 return 0; 3438 } 3439 3440 static inline u8 skb_metadata_len(const struct sk_buff *skb) 3441 { 3442 return skb_shinfo(skb)->meta_len; 3443 } 3444 3445 static inline void *skb_metadata_end(const struct sk_buff *skb) 3446 { 3447 return skb_mac_header(skb); 3448 } 3449 3450 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 3451 const struct sk_buff *skb_b, 3452 u8 meta_len) 3453 { 3454 const void *a = skb_metadata_end(skb_a); 3455 const void *b = skb_metadata_end(skb_b); 3456 /* Using more efficient varaiant than plain call to memcmp(). */ 3457 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64 3458 u64 diffs = 0; 3459 3460 switch (meta_len) { 3461 #define __it(x, op) (x -= sizeof(u##op)) 3462 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 3463 case 32: diffs |= __it_diff(a, b, 64); 3464 case 24: diffs |= __it_diff(a, b, 64); 3465 case 16: diffs |= __it_diff(a, b, 64); 3466 case 8: diffs |= __it_diff(a, b, 64); 3467 break; 3468 case 28: diffs |= __it_diff(a, b, 64); 3469 case 20: diffs |= __it_diff(a, b, 64); 3470 case 12: diffs |= __it_diff(a, b, 64); 3471 case 4: diffs |= __it_diff(a, b, 32); 3472 break; 3473 } 3474 return diffs; 3475 #else 3476 return memcmp(a - meta_len, b - meta_len, meta_len); 3477 #endif 3478 } 3479 3480 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 3481 const struct sk_buff *skb_b) 3482 { 3483 u8 len_a = skb_metadata_len(skb_a); 3484 u8 len_b = skb_metadata_len(skb_b); 3485 3486 if (!(len_a | len_b)) 3487 return false; 3488 3489 return len_a != len_b ? 3490 true : __skb_metadata_differs(skb_a, skb_b, len_a); 3491 } 3492 3493 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 3494 { 3495 skb_shinfo(skb)->meta_len = meta_len; 3496 } 3497 3498 static inline void skb_metadata_clear(struct sk_buff *skb) 3499 { 3500 skb_metadata_set(skb, 0); 3501 } 3502 3503 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3504 3505 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3506 3507 void skb_clone_tx_timestamp(struct sk_buff *skb); 3508 bool skb_defer_rx_timestamp(struct sk_buff *skb); 3509 3510 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3511 3512 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3513 { 3514 } 3515 3516 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3517 { 3518 return false; 3519 } 3520 3521 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3522 3523 /** 3524 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3525 * 3526 * PHY drivers may accept clones of transmitted packets for 3527 * timestamping via their phy_driver.txtstamp method. These drivers 3528 * must call this function to return the skb back to the stack with a 3529 * timestamp. 3530 * 3531 * @skb: clone of the the original outgoing packet 3532 * @hwtstamps: hardware time stamps 3533 * 3534 */ 3535 void skb_complete_tx_timestamp(struct sk_buff *skb, 3536 struct skb_shared_hwtstamps *hwtstamps); 3537 3538 void __skb_tstamp_tx(struct sk_buff *orig_skb, 3539 struct skb_shared_hwtstamps *hwtstamps, 3540 struct sock *sk, int tstype); 3541 3542 /** 3543 * skb_tstamp_tx - queue clone of skb with send time stamps 3544 * @orig_skb: the original outgoing packet 3545 * @hwtstamps: hardware time stamps, may be NULL if not available 3546 * 3547 * If the skb has a socket associated, then this function clones the 3548 * skb (thus sharing the actual data and optional structures), stores 3549 * the optional hardware time stamping information (if non NULL) or 3550 * generates a software time stamp (otherwise), then queues the clone 3551 * to the error queue of the socket. Errors are silently ignored. 3552 */ 3553 void skb_tstamp_tx(struct sk_buff *orig_skb, 3554 struct skb_shared_hwtstamps *hwtstamps); 3555 3556 /** 3557 * skb_tx_timestamp() - Driver hook for transmit timestamping 3558 * 3559 * Ethernet MAC Drivers should call this function in their hard_xmit() 3560 * function immediately before giving the sk_buff to the MAC hardware. 3561 * 3562 * Specifically, one should make absolutely sure that this function is 3563 * called before TX completion of this packet can trigger. Otherwise 3564 * the packet could potentially already be freed. 3565 * 3566 * @skb: A socket buffer. 3567 */ 3568 static inline void skb_tx_timestamp(struct sk_buff *skb) 3569 { 3570 skb_clone_tx_timestamp(skb); 3571 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 3572 skb_tstamp_tx(skb, NULL); 3573 } 3574 3575 /** 3576 * skb_complete_wifi_ack - deliver skb with wifi status 3577 * 3578 * @skb: the original outgoing packet 3579 * @acked: ack status 3580 * 3581 */ 3582 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3583 3584 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3585 __sum16 __skb_checksum_complete(struct sk_buff *skb); 3586 3587 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3588 { 3589 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3590 skb->csum_valid || 3591 (skb->ip_summed == CHECKSUM_PARTIAL && 3592 skb_checksum_start_offset(skb) >= 0)); 3593 } 3594 3595 /** 3596 * skb_checksum_complete - Calculate checksum of an entire packet 3597 * @skb: packet to process 3598 * 3599 * This function calculates the checksum over the entire packet plus 3600 * the value of skb->csum. The latter can be used to supply the 3601 * checksum of a pseudo header as used by TCP/UDP. It returns the 3602 * checksum. 3603 * 3604 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3605 * this function can be used to verify that checksum on received 3606 * packets. In that case the function should return zero if the 3607 * checksum is correct. In particular, this function will return zero 3608 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3609 * hardware has already verified the correctness of the checksum. 3610 */ 3611 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3612 { 3613 return skb_csum_unnecessary(skb) ? 3614 0 : __skb_checksum_complete(skb); 3615 } 3616 3617 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3618 { 3619 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3620 if (skb->csum_level == 0) 3621 skb->ip_summed = CHECKSUM_NONE; 3622 else 3623 skb->csum_level--; 3624 } 3625 } 3626 3627 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3628 { 3629 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3630 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3631 skb->csum_level++; 3632 } else if (skb->ip_summed == CHECKSUM_NONE) { 3633 skb->ip_summed = CHECKSUM_UNNECESSARY; 3634 skb->csum_level = 0; 3635 } 3636 } 3637 3638 /* Check if we need to perform checksum complete validation. 3639 * 3640 * Returns true if checksum complete is needed, false otherwise 3641 * (either checksum is unnecessary or zero checksum is allowed). 3642 */ 3643 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3644 bool zero_okay, 3645 __sum16 check) 3646 { 3647 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3648 skb->csum_valid = 1; 3649 __skb_decr_checksum_unnecessary(skb); 3650 return false; 3651 } 3652 3653 return true; 3654 } 3655 3656 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 3657 * in checksum_init. 3658 */ 3659 #define CHECKSUM_BREAK 76 3660 3661 /* Unset checksum-complete 3662 * 3663 * Unset checksum complete can be done when packet is being modified 3664 * (uncompressed for instance) and checksum-complete value is 3665 * invalidated. 3666 */ 3667 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3668 { 3669 if (skb->ip_summed == CHECKSUM_COMPLETE) 3670 skb->ip_summed = CHECKSUM_NONE; 3671 } 3672 3673 /* Validate (init) checksum based on checksum complete. 3674 * 3675 * Return values: 3676 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3677 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3678 * checksum is stored in skb->csum for use in __skb_checksum_complete 3679 * non-zero: value of invalid checksum 3680 * 3681 */ 3682 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3683 bool complete, 3684 __wsum psum) 3685 { 3686 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3687 if (!csum_fold(csum_add(psum, skb->csum))) { 3688 skb->csum_valid = 1; 3689 return 0; 3690 } 3691 } 3692 3693 skb->csum = psum; 3694 3695 if (complete || skb->len <= CHECKSUM_BREAK) { 3696 __sum16 csum; 3697 3698 csum = __skb_checksum_complete(skb); 3699 skb->csum_valid = !csum; 3700 return csum; 3701 } 3702 3703 return 0; 3704 } 3705 3706 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3707 { 3708 return 0; 3709 } 3710 3711 /* Perform checksum validate (init). Note that this is a macro since we only 3712 * want to calculate the pseudo header which is an input function if necessary. 3713 * First we try to validate without any computation (checksum unnecessary) and 3714 * then calculate based on checksum complete calling the function to compute 3715 * pseudo header. 3716 * 3717 * Return values: 3718 * 0: checksum is validated or try to in skb_checksum_complete 3719 * non-zero: value of invalid checksum 3720 */ 3721 #define __skb_checksum_validate(skb, proto, complete, \ 3722 zero_okay, check, compute_pseudo) \ 3723 ({ \ 3724 __sum16 __ret = 0; \ 3725 skb->csum_valid = 0; \ 3726 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3727 __ret = __skb_checksum_validate_complete(skb, \ 3728 complete, compute_pseudo(skb, proto)); \ 3729 __ret; \ 3730 }) 3731 3732 #define skb_checksum_init(skb, proto, compute_pseudo) \ 3733 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3734 3735 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3736 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3737 3738 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 3739 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3740 3741 #define skb_checksum_validate_zero_check(skb, proto, check, \ 3742 compute_pseudo) \ 3743 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3744 3745 #define skb_checksum_simple_validate(skb) \ 3746 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3747 3748 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3749 { 3750 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 3751 } 3752 3753 static inline void __skb_checksum_convert(struct sk_buff *skb, 3754 __sum16 check, __wsum pseudo) 3755 { 3756 skb->csum = ~pseudo; 3757 skb->ip_summed = CHECKSUM_COMPLETE; 3758 } 3759 3760 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3761 do { \ 3762 if (__skb_checksum_convert_check(skb)) \ 3763 __skb_checksum_convert(skb, check, \ 3764 compute_pseudo(skb, proto)); \ 3765 } while (0) 3766 3767 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3768 u16 start, u16 offset) 3769 { 3770 skb->ip_summed = CHECKSUM_PARTIAL; 3771 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3772 skb->csum_offset = offset - start; 3773 } 3774 3775 /* Update skbuf and packet to reflect the remote checksum offload operation. 3776 * When called, ptr indicates the starting point for skb->csum when 3777 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3778 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3779 */ 3780 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3781 int start, int offset, bool nopartial) 3782 { 3783 __wsum delta; 3784 3785 if (!nopartial) { 3786 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3787 return; 3788 } 3789 3790 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3791 __skb_checksum_complete(skb); 3792 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3793 } 3794 3795 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3796 3797 /* Adjust skb->csum since we changed the packet */ 3798 skb->csum = csum_add(skb->csum, delta); 3799 } 3800 3801 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 3802 { 3803 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 3804 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK); 3805 #else 3806 return NULL; 3807 #endif 3808 } 3809 3810 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3811 void nf_conntrack_destroy(struct nf_conntrack *nfct); 3812 static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3813 { 3814 if (nfct && atomic_dec_and_test(&nfct->use)) 3815 nf_conntrack_destroy(nfct); 3816 } 3817 static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3818 { 3819 if (nfct) 3820 atomic_inc(&nfct->use); 3821 } 3822 #endif 3823 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3824 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3825 { 3826 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use)) 3827 kfree(nf_bridge); 3828 } 3829 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3830 { 3831 if (nf_bridge) 3832 refcount_inc(&nf_bridge->use); 3833 } 3834 #endif /* CONFIG_BRIDGE_NETFILTER */ 3835 static inline void nf_reset(struct sk_buff *skb) 3836 { 3837 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3838 nf_conntrack_put(skb_nfct(skb)); 3839 skb->_nfct = 0; 3840 #endif 3841 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3842 nf_bridge_put(skb->nf_bridge); 3843 skb->nf_bridge = NULL; 3844 #endif 3845 } 3846 3847 static inline void nf_reset_trace(struct sk_buff *skb) 3848 { 3849 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3850 skb->nf_trace = 0; 3851 #endif 3852 } 3853 3854 static inline void ipvs_reset(struct sk_buff *skb) 3855 { 3856 #if IS_ENABLED(CONFIG_IP_VS) 3857 skb->ipvs_property = 0; 3858 #endif 3859 } 3860 3861 /* Note: This doesn't put any conntrack and bridge info in dst. */ 3862 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3863 bool copy) 3864 { 3865 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3866 dst->_nfct = src->_nfct; 3867 nf_conntrack_get(skb_nfct(src)); 3868 #endif 3869 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3870 dst->nf_bridge = src->nf_bridge; 3871 nf_bridge_get(src->nf_bridge); 3872 #endif 3873 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3874 if (copy) 3875 dst->nf_trace = src->nf_trace; 3876 #endif 3877 } 3878 3879 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3880 { 3881 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3882 nf_conntrack_put(skb_nfct(dst)); 3883 #endif 3884 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3885 nf_bridge_put(dst->nf_bridge); 3886 #endif 3887 __nf_copy(dst, src, true); 3888 } 3889 3890 #ifdef CONFIG_NETWORK_SECMARK 3891 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3892 { 3893 to->secmark = from->secmark; 3894 } 3895 3896 static inline void skb_init_secmark(struct sk_buff *skb) 3897 { 3898 skb->secmark = 0; 3899 } 3900 #else 3901 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3902 { } 3903 3904 static inline void skb_init_secmark(struct sk_buff *skb) 3905 { } 3906 #endif 3907 3908 static inline bool skb_irq_freeable(const struct sk_buff *skb) 3909 { 3910 return !skb->destructor && 3911 #if IS_ENABLED(CONFIG_XFRM) 3912 !skb->sp && 3913 #endif 3914 !skb_nfct(skb) && 3915 !skb->_skb_refdst && 3916 !skb_has_frag_list(skb); 3917 } 3918 3919 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3920 { 3921 skb->queue_mapping = queue_mapping; 3922 } 3923 3924 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3925 { 3926 return skb->queue_mapping; 3927 } 3928 3929 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3930 { 3931 to->queue_mapping = from->queue_mapping; 3932 } 3933 3934 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3935 { 3936 skb->queue_mapping = rx_queue + 1; 3937 } 3938 3939 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3940 { 3941 return skb->queue_mapping - 1; 3942 } 3943 3944 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3945 { 3946 return skb->queue_mapping != 0; 3947 } 3948 3949 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 3950 { 3951 skb->dst_pending_confirm = val; 3952 } 3953 3954 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 3955 { 3956 return skb->dst_pending_confirm != 0; 3957 } 3958 3959 static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3960 { 3961 #ifdef CONFIG_XFRM 3962 return skb->sp; 3963 #else 3964 return NULL; 3965 #endif 3966 } 3967 3968 /* Keeps track of mac header offset relative to skb->head. 3969 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3970 * For non-tunnel skb it points to skb_mac_header() and for 3971 * tunnel skb it points to outer mac header. 3972 * Keeps track of level of encapsulation of network headers. 3973 */ 3974 struct skb_gso_cb { 3975 union { 3976 int mac_offset; 3977 int data_offset; 3978 }; 3979 int encap_level; 3980 __wsum csum; 3981 __u16 csum_start; 3982 }; 3983 #define SKB_SGO_CB_OFFSET 32 3984 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 3985 3986 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3987 { 3988 return (skb_mac_header(inner_skb) - inner_skb->head) - 3989 SKB_GSO_CB(inner_skb)->mac_offset; 3990 } 3991 3992 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3993 { 3994 int new_headroom, headroom; 3995 int ret; 3996 3997 headroom = skb_headroom(skb); 3998 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3999 if (ret) 4000 return ret; 4001 4002 new_headroom = skb_headroom(skb); 4003 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 4004 return 0; 4005 } 4006 4007 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 4008 { 4009 /* Do not update partial checksums if remote checksum is enabled. */ 4010 if (skb->remcsum_offload) 4011 return; 4012 4013 SKB_GSO_CB(skb)->csum = res; 4014 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 4015 } 4016 4017 /* Compute the checksum for a gso segment. First compute the checksum value 4018 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 4019 * then add in skb->csum (checksum from csum_start to end of packet). 4020 * skb->csum and csum_start are then updated to reflect the checksum of the 4021 * resultant packet starting from the transport header-- the resultant checksum 4022 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 4023 * header. 4024 */ 4025 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 4026 { 4027 unsigned char *csum_start = skb_transport_header(skb); 4028 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 4029 __wsum partial = SKB_GSO_CB(skb)->csum; 4030 4031 SKB_GSO_CB(skb)->csum = res; 4032 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 4033 4034 return csum_fold(csum_partial(csum_start, plen, partial)); 4035 } 4036 4037 static inline bool skb_is_gso(const struct sk_buff *skb) 4038 { 4039 return skb_shinfo(skb)->gso_size; 4040 } 4041 4042 /* Note: Should be called only if skb_is_gso(skb) is true */ 4043 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4044 { 4045 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4046 } 4047 4048 /* Note: Should be called only if skb_is_gso(skb) is true */ 4049 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 4050 { 4051 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 4052 } 4053 4054 static inline void skb_gso_reset(struct sk_buff *skb) 4055 { 4056 skb_shinfo(skb)->gso_size = 0; 4057 skb_shinfo(skb)->gso_segs = 0; 4058 skb_shinfo(skb)->gso_type = 0; 4059 } 4060 4061 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 4062 u16 increment) 4063 { 4064 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4065 return; 4066 shinfo->gso_size += increment; 4067 } 4068 4069 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 4070 u16 decrement) 4071 { 4072 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4073 return; 4074 shinfo->gso_size -= decrement; 4075 } 4076 4077 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 4078 4079 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 4080 { 4081 /* LRO sets gso_size but not gso_type, whereas if GSO is really 4082 * wanted then gso_type will be set. */ 4083 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4084 4085 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 4086 unlikely(shinfo->gso_type == 0)) { 4087 __skb_warn_lro_forwarding(skb); 4088 return true; 4089 } 4090 return false; 4091 } 4092 4093 static inline void skb_forward_csum(struct sk_buff *skb) 4094 { 4095 /* Unfortunately we don't support this one. Any brave souls? */ 4096 if (skb->ip_summed == CHECKSUM_COMPLETE) 4097 skb->ip_summed = CHECKSUM_NONE; 4098 } 4099 4100 /** 4101 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 4102 * @skb: skb to check 4103 * 4104 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 4105 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 4106 * use this helper, to document places where we make this assertion. 4107 */ 4108 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 4109 { 4110 #ifdef DEBUG 4111 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 4112 #endif 4113 } 4114 4115 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 4116 4117 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 4118 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4119 unsigned int transport_len, 4120 __sum16(*skb_chkf)(struct sk_buff *skb)); 4121 4122 /** 4123 * skb_head_is_locked - Determine if the skb->head is locked down 4124 * @skb: skb to check 4125 * 4126 * The head on skbs build around a head frag can be removed if they are 4127 * not cloned. This function returns true if the skb head is locked down 4128 * due to either being allocated via kmalloc, or by being a clone with 4129 * multiple references to the head. 4130 */ 4131 static inline bool skb_head_is_locked(const struct sk_buff *skb) 4132 { 4133 return !skb->head_frag || skb_cloned(skb); 4134 } 4135 4136 /* Local Checksum Offload. 4137 * Compute outer checksum based on the assumption that the 4138 * inner checksum will be offloaded later. 4139 * See Documentation/networking/checksum-offloads.txt for 4140 * explanation of how this works. 4141 * Fill in outer checksum adjustment (e.g. with sum of outer 4142 * pseudo-header) before calling. 4143 * Also ensure that inner checksum is in linear data area. 4144 */ 4145 static inline __wsum lco_csum(struct sk_buff *skb) 4146 { 4147 unsigned char *csum_start = skb_checksum_start(skb); 4148 unsigned char *l4_hdr = skb_transport_header(skb); 4149 __wsum partial; 4150 4151 /* Start with complement of inner checksum adjustment */ 4152 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 4153 skb->csum_offset)); 4154 4155 /* Add in checksum of our headers (incl. outer checksum 4156 * adjustment filled in by caller) and return result. 4157 */ 4158 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 4159 } 4160 4161 #endif /* __KERNEL__ */ 4162 #endif /* _LINUX_SKBUFF_H */ 4163