1 /* 2 * Definitions for the 'struct sk_buff' memory handlers. 3 * 4 * Authors: 5 * Alan Cox, <gw4pts@gw4pts.ampr.org> 6 * Florian La Roche, <rzsfl@rz.uni-sb.de> 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 */ 857 858 #define SKB_ALLOC_FCLONE 0x01 859 #define SKB_ALLOC_RX 0x02 860 #define SKB_ALLOC_NAPI 0x04 861 862 /* Returns true if the skb was allocated from PFMEMALLOC reserves */ 863 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 864 { 865 return unlikely(skb->pfmemalloc); 866 } 867 868 /* 869 * skb might have a dst pointer attached, refcounted or not. 870 * _skb_refdst low order bit is set if refcount was _not_ taken 871 */ 872 #define SKB_DST_NOREF 1UL 873 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 874 875 #define SKB_NFCT_PTRMASK ~(7UL) 876 /** 877 * skb_dst - returns skb dst_entry 878 * @skb: buffer 879 * 880 * Returns skb dst_entry, regardless of reference taken or not. 881 */ 882 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 883 { 884 /* If refdst was not refcounted, check we still are in a 885 * rcu_read_lock section 886 */ 887 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 888 !rcu_read_lock_held() && 889 !rcu_read_lock_bh_held()); 890 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 891 } 892 893 /** 894 * skb_dst_set - sets skb dst 895 * @skb: buffer 896 * @dst: dst entry 897 * 898 * Sets skb dst, assuming a reference was taken on dst and should 899 * be released by skb_dst_drop() 900 */ 901 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 902 { 903 skb->_skb_refdst = (unsigned long)dst; 904 } 905 906 /** 907 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 908 * @skb: buffer 909 * @dst: dst entry 910 * 911 * Sets skb dst, assuming a reference was not taken on dst. 912 * If dst entry is cached, we do not take reference and dst_release 913 * will be avoided by refdst_drop. If dst entry is not cached, we take 914 * reference, so that last dst_release can destroy the dst immediately. 915 */ 916 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 917 { 918 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 919 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 920 } 921 922 /** 923 * skb_dst_is_noref - Test if skb dst isn't refcounted 924 * @skb: buffer 925 */ 926 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 927 { 928 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 929 } 930 931 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 932 { 933 return (struct rtable *)skb_dst(skb); 934 } 935 936 /* For mangling skb->pkt_type from user space side from applications 937 * such as nft, tc, etc, we only allow a conservative subset of 938 * possible pkt_types to be set. 939 */ 940 static inline bool skb_pkt_type_ok(u32 ptype) 941 { 942 return ptype <= PACKET_OTHERHOST; 943 } 944 945 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 946 { 947 #ifdef CONFIG_NET_RX_BUSY_POLL 948 return skb->napi_id; 949 #else 950 return 0; 951 #endif 952 } 953 954 /* decrement the reference count and return true if we can free the skb */ 955 static inline bool skb_unref(struct sk_buff *skb) 956 { 957 if (unlikely(!skb)) 958 return false; 959 if (likely(refcount_read(&skb->users) == 1)) 960 smp_rmb(); 961 else if (likely(!refcount_dec_and_test(&skb->users))) 962 return false; 963 964 return true; 965 } 966 967 void skb_release_head_state(struct sk_buff *skb); 968 void kfree_skb(struct sk_buff *skb); 969 void kfree_skb_list(struct sk_buff *segs); 970 void skb_tx_error(struct sk_buff *skb); 971 void consume_skb(struct sk_buff *skb); 972 void __consume_stateless_skb(struct sk_buff *skb); 973 void __kfree_skb(struct sk_buff *skb); 974 extern struct kmem_cache *skbuff_head_cache; 975 976 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 977 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 978 bool *fragstolen, int *delta_truesize); 979 980 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 981 int node); 982 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 983 struct sk_buff *build_skb(void *data, unsigned int frag_size); 984 static inline struct sk_buff *alloc_skb(unsigned int size, 985 gfp_t priority) 986 { 987 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 988 } 989 990 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 991 unsigned long data_len, 992 int max_page_order, 993 int *errcode, 994 gfp_t gfp_mask); 995 996 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 997 struct sk_buff_fclones { 998 struct sk_buff skb1; 999 1000 struct sk_buff skb2; 1001 1002 refcount_t fclone_ref; 1003 }; 1004 1005 /** 1006 * skb_fclone_busy - check if fclone is busy 1007 * @sk: socket 1008 * @skb: buffer 1009 * 1010 * Returns true if skb is a fast clone, and its clone is not freed. 1011 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1012 * so we also check that this didnt happen. 1013 */ 1014 static inline bool skb_fclone_busy(const struct sock *sk, 1015 const struct sk_buff *skb) 1016 { 1017 const struct sk_buff_fclones *fclones; 1018 1019 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1020 1021 return skb->fclone == SKB_FCLONE_ORIG && 1022 refcount_read(&fclones->fclone_ref) > 1 && 1023 fclones->skb2.sk == sk; 1024 } 1025 1026 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1027 gfp_t priority) 1028 { 1029 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1030 } 1031 1032 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1033 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1034 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1035 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1036 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1037 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1038 gfp_t gfp_mask, bool fclone); 1039 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1040 gfp_t gfp_mask) 1041 { 1042 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1043 } 1044 1045 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1046 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1047 unsigned int headroom); 1048 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1049 int newtailroom, gfp_t priority); 1050 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1051 int offset, int len); 1052 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1053 int offset, int len); 1054 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1055 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1056 1057 /** 1058 * skb_pad - zero pad the tail of an skb 1059 * @skb: buffer to pad 1060 * @pad: space to pad 1061 * 1062 * Ensure that a buffer is followed by a padding area that is zero 1063 * filled. Used by network drivers which may DMA or transfer data 1064 * beyond the buffer end onto the wire. 1065 * 1066 * May return error in out of memory cases. The skb is freed on error. 1067 */ 1068 static inline int skb_pad(struct sk_buff *skb, int pad) 1069 { 1070 return __skb_pad(skb, pad, true); 1071 } 1072 #define dev_kfree_skb(a) consume_skb(a) 1073 1074 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 1075 int getfrag(void *from, char *to, int offset, 1076 int len, int odd, struct sk_buff *skb), 1077 void *from, int length); 1078 1079 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1080 int offset, size_t size); 1081 1082 struct skb_seq_state { 1083 __u32 lower_offset; 1084 __u32 upper_offset; 1085 __u32 frag_idx; 1086 __u32 stepped_offset; 1087 struct sk_buff *root_skb; 1088 struct sk_buff *cur_skb; 1089 __u8 *frag_data; 1090 }; 1091 1092 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1093 unsigned int to, struct skb_seq_state *st); 1094 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1095 struct skb_seq_state *st); 1096 void skb_abort_seq_read(struct skb_seq_state *st); 1097 1098 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1099 unsigned int to, struct ts_config *config); 1100 1101 /* 1102 * Packet hash types specify the type of hash in skb_set_hash. 1103 * 1104 * Hash types refer to the protocol layer addresses which are used to 1105 * construct a packet's hash. The hashes are used to differentiate or identify 1106 * flows of the protocol layer for the hash type. Hash types are either 1107 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1108 * 1109 * Properties of hashes: 1110 * 1111 * 1) Two packets in different flows have different hash values 1112 * 2) Two packets in the same flow should have the same hash value 1113 * 1114 * A hash at a higher layer is considered to be more specific. A driver should 1115 * set the most specific hash possible. 1116 * 1117 * A driver cannot indicate a more specific hash than the layer at which a hash 1118 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1119 * 1120 * A driver may indicate a hash level which is less specific than the 1121 * actual layer the hash was computed on. For instance, a hash computed 1122 * at L4 may be considered an L3 hash. This should only be done if the 1123 * driver can't unambiguously determine that the HW computed the hash at 1124 * the higher layer. Note that the "should" in the second property above 1125 * permits this. 1126 */ 1127 enum pkt_hash_types { 1128 PKT_HASH_TYPE_NONE, /* Undefined type */ 1129 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1130 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1131 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1132 }; 1133 1134 static inline void skb_clear_hash(struct sk_buff *skb) 1135 { 1136 skb->hash = 0; 1137 skb->sw_hash = 0; 1138 skb->l4_hash = 0; 1139 } 1140 1141 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1142 { 1143 if (!skb->l4_hash) 1144 skb_clear_hash(skb); 1145 } 1146 1147 static inline void 1148 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1149 { 1150 skb->l4_hash = is_l4; 1151 skb->sw_hash = is_sw; 1152 skb->hash = hash; 1153 } 1154 1155 static inline void 1156 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1157 { 1158 /* Used by drivers to set hash from HW */ 1159 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1160 } 1161 1162 static inline void 1163 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1164 { 1165 __skb_set_hash(skb, hash, true, is_l4); 1166 } 1167 1168 void __skb_get_hash(struct sk_buff *skb); 1169 u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1170 u32 skb_get_poff(const struct sk_buff *skb); 1171 u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1172 const struct flow_keys_basic *keys, int hlen); 1173 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1174 void *data, int hlen_proto); 1175 1176 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1177 int thoff, u8 ip_proto) 1178 { 1179 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1180 } 1181 1182 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1183 const struct flow_dissector_key *key, 1184 unsigned int key_count); 1185 1186 bool __skb_flow_dissect(const struct sk_buff *skb, 1187 struct flow_dissector *flow_dissector, 1188 void *target_container, 1189 void *data, __be16 proto, int nhoff, int hlen, 1190 unsigned int flags); 1191 1192 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1193 struct flow_dissector *flow_dissector, 1194 void *target_container, unsigned int flags) 1195 { 1196 return __skb_flow_dissect(skb, flow_dissector, target_container, 1197 NULL, 0, 0, 0, flags); 1198 } 1199 1200 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1201 struct flow_keys *flow, 1202 unsigned int flags) 1203 { 1204 memset(flow, 0, sizeof(*flow)); 1205 return __skb_flow_dissect(skb, &flow_keys_dissector, flow, 1206 NULL, 0, 0, 0, flags); 1207 } 1208 1209 static inline bool 1210 skb_flow_dissect_flow_keys_basic(const struct sk_buff *skb, 1211 struct flow_keys_basic *flow, void *data, 1212 __be16 proto, int nhoff, int hlen, 1213 unsigned int flags) 1214 { 1215 memset(flow, 0, sizeof(*flow)); 1216 return __skb_flow_dissect(skb, &flow_keys_basic_dissector, flow, 1217 data, proto, nhoff, hlen, flags); 1218 } 1219 1220 void 1221 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1222 struct flow_dissector *flow_dissector, 1223 void *target_container); 1224 1225 static inline __u32 skb_get_hash(struct sk_buff *skb) 1226 { 1227 if (!skb->l4_hash && !skb->sw_hash) 1228 __skb_get_hash(skb); 1229 1230 return skb->hash; 1231 } 1232 1233 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1234 { 1235 if (!skb->l4_hash && !skb->sw_hash) { 1236 struct flow_keys keys; 1237 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1238 1239 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1240 } 1241 1242 return skb->hash; 1243 } 1244 1245 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1246 1247 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1248 { 1249 return skb->hash; 1250 } 1251 1252 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1253 { 1254 to->hash = from->hash; 1255 to->sw_hash = from->sw_hash; 1256 to->l4_hash = from->l4_hash; 1257 }; 1258 1259 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1260 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1261 { 1262 return skb->head + skb->end; 1263 } 1264 1265 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1266 { 1267 return skb->end; 1268 } 1269 #else 1270 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1271 { 1272 return skb->end; 1273 } 1274 1275 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1276 { 1277 return skb->end - skb->head; 1278 } 1279 #endif 1280 1281 /* Internal */ 1282 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1283 1284 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1285 { 1286 return &skb_shinfo(skb)->hwtstamps; 1287 } 1288 1289 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1290 { 1291 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY; 1292 1293 return is_zcopy ? skb_uarg(skb) : NULL; 1294 } 1295 1296 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg) 1297 { 1298 if (skb && uarg && !skb_zcopy(skb)) { 1299 sock_zerocopy_get(uarg); 1300 skb_shinfo(skb)->destructor_arg = uarg; 1301 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG; 1302 } 1303 } 1304 1305 /* Release a reference on a zerocopy structure */ 1306 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy) 1307 { 1308 struct ubuf_info *uarg = skb_zcopy(skb); 1309 1310 if (uarg) { 1311 if (uarg->callback == sock_zerocopy_callback) { 1312 uarg->zerocopy = uarg->zerocopy && zerocopy; 1313 sock_zerocopy_put(uarg); 1314 } else { 1315 uarg->callback(uarg, zerocopy); 1316 } 1317 1318 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1319 } 1320 } 1321 1322 /* Abort a zerocopy operation and revert zckey on error in send syscall */ 1323 static inline void skb_zcopy_abort(struct sk_buff *skb) 1324 { 1325 struct ubuf_info *uarg = skb_zcopy(skb); 1326 1327 if (uarg) { 1328 sock_zerocopy_put_abort(uarg); 1329 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1330 } 1331 } 1332 1333 /** 1334 * skb_queue_empty - check if a queue is empty 1335 * @list: queue head 1336 * 1337 * Returns true if the queue is empty, false otherwise. 1338 */ 1339 static inline int skb_queue_empty(const struct sk_buff_head *list) 1340 { 1341 return list->next == (const struct sk_buff *) list; 1342 } 1343 1344 /** 1345 * skb_queue_is_last - check if skb is the last entry in the queue 1346 * @list: queue head 1347 * @skb: buffer 1348 * 1349 * Returns true if @skb is the last buffer on the list. 1350 */ 1351 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1352 const struct sk_buff *skb) 1353 { 1354 return skb->next == (const struct sk_buff *) list; 1355 } 1356 1357 /** 1358 * skb_queue_is_first - check if skb is the first entry in the queue 1359 * @list: queue head 1360 * @skb: buffer 1361 * 1362 * Returns true if @skb is the first buffer on the list. 1363 */ 1364 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1365 const struct sk_buff *skb) 1366 { 1367 return skb->prev == (const struct sk_buff *) list; 1368 } 1369 1370 /** 1371 * skb_queue_next - return the next packet in the queue 1372 * @list: queue head 1373 * @skb: current buffer 1374 * 1375 * Return the next packet in @list after @skb. It is only valid to 1376 * call this if skb_queue_is_last() evaluates to false. 1377 */ 1378 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1379 const struct sk_buff *skb) 1380 { 1381 /* This BUG_ON may seem severe, but if we just return then we 1382 * are going to dereference garbage. 1383 */ 1384 BUG_ON(skb_queue_is_last(list, skb)); 1385 return skb->next; 1386 } 1387 1388 /** 1389 * skb_queue_prev - return the prev packet in the queue 1390 * @list: queue head 1391 * @skb: current buffer 1392 * 1393 * Return the prev packet in @list before @skb. It is only valid to 1394 * call this if skb_queue_is_first() evaluates to false. 1395 */ 1396 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1397 const struct sk_buff *skb) 1398 { 1399 /* This BUG_ON may seem severe, but if we just return then we 1400 * are going to dereference garbage. 1401 */ 1402 BUG_ON(skb_queue_is_first(list, skb)); 1403 return skb->prev; 1404 } 1405 1406 /** 1407 * skb_get - reference buffer 1408 * @skb: buffer to reference 1409 * 1410 * Makes another reference to a socket buffer and returns a pointer 1411 * to the buffer. 1412 */ 1413 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1414 { 1415 refcount_inc(&skb->users); 1416 return skb; 1417 } 1418 1419 /* 1420 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1421 */ 1422 1423 /** 1424 * skb_cloned - is the buffer a clone 1425 * @skb: buffer to check 1426 * 1427 * Returns true if the buffer was generated with skb_clone() and is 1428 * one of multiple shared copies of the buffer. Cloned buffers are 1429 * shared data so must not be written to under normal circumstances. 1430 */ 1431 static inline int skb_cloned(const struct sk_buff *skb) 1432 { 1433 return skb->cloned && 1434 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1435 } 1436 1437 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1438 { 1439 might_sleep_if(gfpflags_allow_blocking(pri)); 1440 1441 if (skb_cloned(skb)) 1442 return pskb_expand_head(skb, 0, 0, pri); 1443 1444 return 0; 1445 } 1446 1447 /** 1448 * skb_header_cloned - is the header a clone 1449 * @skb: buffer to check 1450 * 1451 * Returns true if modifying the header part of the buffer requires 1452 * the data to be copied. 1453 */ 1454 static inline int skb_header_cloned(const struct sk_buff *skb) 1455 { 1456 int dataref; 1457 1458 if (!skb->cloned) 1459 return 0; 1460 1461 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1462 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1463 return dataref != 1; 1464 } 1465 1466 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1467 { 1468 might_sleep_if(gfpflags_allow_blocking(pri)); 1469 1470 if (skb_header_cloned(skb)) 1471 return pskb_expand_head(skb, 0, 0, pri); 1472 1473 return 0; 1474 } 1475 1476 /** 1477 * __skb_header_release - release reference to header 1478 * @skb: buffer to operate on 1479 */ 1480 static inline void __skb_header_release(struct sk_buff *skb) 1481 { 1482 skb->nohdr = 1; 1483 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1484 } 1485 1486 1487 /** 1488 * skb_shared - is the buffer shared 1489 * @skb: buffer to check 1490 * 1491 * Returns true if more than one person has a reference to this 1492 * buffer. 1493 */ 1494 static inline int skb_shared(const struct sk_buff *skb) 1495 { 1496 return refcount_read(&skb->users) != 1; 1497 } 1498 1499 /** 1500 * skb_share_check - check if buffer is shared and if so clone it 1501 * @skb: buffer to check 1502 * @pri: priority for memory allocation 1503 * 1504 * If the buffer is shared the buffer is cloned and the old copy 1505 * drops a reference. A new clone with a single reference is returned. 1506 * If the buffer is not shared the original buffer is returned. When 1507 * being called from interrupt status or with spinlocks held pri must 1508 * be GFP_ATOMIC. 1509 * 1510 * NULL is returned on a memory allocation failure. 1511 */ 1512 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1513 { 1514 might_sleep_if(gfpflags_allow_blocking(pri)); 1515 if (skb_shared(skb)) { 1516 struct sk_buff *nskb = skb_clone(skb, pri); 1517 1518 if (likely(nskb)) 1519 consume_skb(skb); 1520 else 1521 kfree_skb(skb); 1522 skb = nskb; 1523 } 1524 return skb; 1525 } 1526 1527 /* 1528 * Copy shared buffers into a new sk_buff. We effectively do COW on 1529 * packets to handle cases where we have a local reader and forward 1530 * and a couple of other messy ones. The normal one is tcpdumping 1531 * a packet thats being forwarded. 1532 */ 1533 1534 /** 1535 * skb_unshare - make a copy of a shared buffer 1536 * @skb: buffer to check 1537 * @pri: priority for memory allocation 1538 * 1539 * If the socket buffer is a clone then this function creates a new 1540 * copy of the data, drops a reference count on the old copy and returns 1541 * the new copy with the reference count at 1. If the buffer is not a clone 1542 * the original buffer is returned. When called with a spinlock held or 1543 * from interrupt state @pri must be %GFP_ATOMIC 1544 * 1545 * %NULL is returned on a memory allocation failure. 1546 */ 1547 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1548 gfp_t pri) 1549 { 1550 might_sleep_if(gfpflags_allow_blocking(pri)); 1551 if (skb_cloned(skb)) { 1552 struct sk_buff *nskb = skb_copy(skb, pri); 1553 1554 /* Free our shared copy */ 1555 if (likely(nskb)) 1556 consume_skb(skb); 1557 else 1558 kfree_skb(skb); 1559 skb = nskb; 1560 } 1561 return skb; 1562 } 1563 1564 /** 1565 * skb_peek - peek at the head of an &sk_buff_head 1566 * @list_: list to peek at 1567 * 1568 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1569 * be careful with this one. A peek leaves the buffer on the 1570 * list and someone else may run off with it. You must hold 1571 * the appropriate locks or have a private queue to do this. 1572 * 1573 * Returns %NULL for an empty list or a pointer to the head element. 1574 * The reference count is not incremented and the reference is therefore 1575 * volatile. Use with caution. 1576 */ 1577 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1578 { 1579 struct sk_buff *skb = list_->next; 1580 1581 if (skb == (struct sk_buff *)list_) 1582 skb = NULL; 1583 return skb; 1584 } 1585 1586 /** 1587 * skb_peek_next - peek skb following the given one from a queue 1588 * @skb: skb to start from 1589 * @list_: list to peek at 1590 * 1591 * Returns %NULL when the end of the list is met or a pointer to the 1592 * next element. The reference count is not incremented and the 1593 * reference is therefore volatile. Use with caution. 1594 */ 1595 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1596 const struct sk_buff_head *list_) 1597 { 1598 struct sk_buff *next = skb->next; 1599 1600 if (next == (struct sk_buff *)list_) 1601 next = NULL; 1602 return next; 1603 } 1604 1605 /** 1606 * skb_peek_tail - peek at the tail of an &sk_buff_head 1607 * @list_: list to peek at 1608 * 1609 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1610 * be careful with this one. A peek leaves the buffer on the 1611 * list and someone else may run off with it. You must hold 1612 * the appropriate locks or have a private queue to do this. 1613 * 1614 * Returns %NULL for an empty list or a pointer to the tail element. 1615 * The reference count is not incremented and the reference is therefore 1616 * volatile. Use with caution. 1617 */ 1618 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1619 { 1620 struct sk_buff *skb = list_->prev; 1621 1622 if (skb == (struct sk_buff *)list_) 1623 skb = NULL; 1624 return skb; 1625 1626 } 1627 1628 /** 1629 * skb_queue_len - get queue length 1630 * @list_: list to measure 1631 * 1632 * Return the length of an &sk_buff queue. 1633 */ 1634 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1635 { 1636 return list_->qlen; 1637 } 1638 1639 /** 1640 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1641 * @list: queue to initialize 1642 * 1643 * This initializes only the list and queue length aspects of 1644 * an sk_buff_head object. This allows to initialize the list 1645 * aspects of an sk_buff_head without reinitializing things like 1646 * the spinlock. It can also be used for on-stack sk_buff_head 1647 * objects where the spinlock is known to not be used. 1648 */ 1649 static inline void __skb_queue_head_init(struct sk_buff_head *list) 1650 { 1651 list->prev = list->next = (struct sk_buff *)list; 1652 list->qlen = 0; 1653 } 1654 1655 /* 1656 * This function creates a split out lock class for each invocation; 1657 * this is needed for now since a whole lot of users of the skb-queue 1658 * infrastructure in drivers have different locking usage (in hardirq) 1659 * than the networking core (in softirq only). In the long run either the 1660 * network layer or drivers should need annotation to consolidate the 1661 * main types of usage into 3 classes. 1662 */ 1663 static inline void skb_queue_head_init(struct sk_buff_head *list) 1664 { 1665 spin_lock_init(&list->lock); 1666 __skb_queue_head_init(list); 1667 } 1668 1669 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1670 struct lock_class_key *class) 1671 { 1672 skb_queue_head_init(list); 1673 lockdep_set_class(&list->lock, class); 1674 } 1675 1676 /* 1677 * Insert an sk_buff on a list. 1678 * 1679 * The "__skb_xxxx()" functions are the non-atomic ones that 1680 * can only be called with interrupts disabled. 1681 */ 1682 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1683 struct sk_buff_head *list); 1684 static inline void __skb_insert(struct sk_buff *newsk, 1685 struct sk_buff *prev, struct sk_buff *next, 1686 struct sk_buff_head *list) 1687 { 1688 newsk->next = next; 1689 newsk->prev = prev; 1690 next->prev = prev->next = newsk; 1691 list->qlen++; 1692 } 1693 1694 static inline void __skb_queue_splice(const struct sk_buff_head *list, 1695 struct sk_buff *prev, 1696 struct sk_buff *next) 1697 { 1698 struct sk_buff *first = list->next; 1699 struct sk_buff *last = list->prev; 1700 1701 first->prev = prev; 1702 prev->next = first; 1703 1704 last->next = next; 1705 next->prev = last; 1706 } 1707 1708 /** 1709 * skb_queue_splice - join two skb lists, this is designed for stacks 1710 * @list: the new list to add 1711 * @head: the place to add it in the first list 1712 */ 1713 static inline void skb_queue_splice(const struct sk_buff_head *list, 1714 struct sk_buff_head *head) 1715 { 1716 if (!skb_queue_empty(list)) { 1717 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1718 head->qlen += list->qlen; 1719 } 1720 } 1721 1722 /** 1723 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1724 * @list: the new list to add 1725 * @head: the place to add it in the first list 1726 * 1727 * The list at @list is reinitialised 1728 */ 1729 static inline void skb_queue_splice_init(struct sk_buff_head *list, 1730 struct sk_buff_head *head) 1731 { 1732 if (!skb_queue_empty(list)) { 1733 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1734 head->qlen += list->qlen; 1735 __skb_queue_head_init(list); 1736 } 1737 } 1738 1739 /** 1740 * skb_queue_splice_tail - join two skb lists, each list being a queue 1741 * @list: the new list to add 1742 * @head: the place to add it in the first list 1743 */ 1744 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1745 struct sk_buff_head *head) 1746 { 1747 if (!skb_queue_empty(list)) { 1748 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1749 head->qlen += list->qlen; 1750 } 1751 } 1752 1753 /** 1754 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1755 * @list: the new list to add 1756 * @head: the place to add it in the first list 1757 * 1758 * Each of the lists is a queue. 1759 * The list at @list is reinitialised 1760 */ 1761 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1762 struct sk_buff_head *head) 1763 { 1764 if (!skb_queue_empty(list)) { 1765 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1766 head->qlen += list->qlen; 1767 __skb_queue_head_init(list); 1768 } 1769 } 1770 1771 /** 1772 * __skb_queue_after - queue a buffer at the list head 1773 * @list: list to use 1774 * @prev: place after this buffer 1775 * @newsk: buffer to queue 1776 * 1777 * Queue a buffer int the middle of a list. This function takes no locks 1778 * and you must therefore hold required locks before calling it. 1779 * 1780 * A buffer cannot be placed on two lists at the same time. 1781 */ 1782 static inline void __skb_queue_after(struct sk_buff_head *list, 1783 struct sk_buff *prev, 1784 struct sk_buff *newsk) 1785 { 1786 __skb_insert(newsk, prev, prev->next, list); 1787 } 1788 1789 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1790 struct sk_buff_head *list); 1791 1792 static inline void __skb_queue_before(struct sk_buff_head *list, 1793 struct sk_buff *next, 1794 struct sk_buff *newsk) 1795 { 1796 __skb_insert(newsk, next->prev, next, list); 1797 } 1798 1799 /** 1800 * __skb_queue_head - queue a buffer at the list head 1801 * @list: list to use 1802 * @newsk: buffer to queue 1803 * 1804 * Queue a buffer at the start of a list. This function takes no locks 1805 * and you must therefore hold required locks before calling it. 1806 * 1807 * A buffer cannot be placed on two lists at the same time. 1808 */ 1809 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1810 static inline void __skb_queue_head(struct sk_buff_head *list, 1811 struct sk_buff *newsk) 1812 { 1813 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1814 } 1815 1816 /** 1817 * __skb_queue_tail - queue a buffer at the list tail 1818 * @list: list to use 1819 * @newsk: buffer to queue 1820 * 1821 * Queue a buffer at the end of a list. This function takes no locks 1822 * and you must therefore hold required locks before calling it. 1823 * 1824 * A buffer cannot be placed on two lists at the same time. 1825 */ 1826 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1827 static inline void __skb_queue_tail(struct sk_buff_head *list, 1828 struct sk_buff *newsk) 1829 { 1830 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1831 } 1832 1833 /* 1834 * remove sk_buff from list. _Must_ be called atomically, and with 1835 * the list known.. 1836 */ 1837 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1838 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1839 { 1840 struct sk_buff *next, *prev; 1841 1842 list->qlen--; 1843 next = skb->next; 1844 prev = skb->prev; 1845 skb->next = skb->prev = NULL; 1846 next->prev = prev; 1847 prev->next = next; 1848 } 1849 1850 /** 1851 * __skb_dequeue - remove from the head of the queue 1852 * @list: list to dequeue from 1853 * 1854 * Remove the head of the list. This function does not take any locks 1855 * so must be used with appropriate locks held only. The head item is 1856 * returned or %NULL if the list is empty. 1857 */ 1858 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1859 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1860 { 1861 struct sk_buff *skb = skb_peek(list); 1862 if (skb) 1863 __skb_unlink(skb, list); 1864 return skb; 1865 } 1866 1867 /** 1868 * __skb_dequeue_tail - remove from the tail of the queue 1869 * @list: list to dequeue from 1870 * 1871 * Remove the tail of the list. This function does not take any locks 1872 * so must be used with appropriate locks held only. The tail item is 1873 * returned or %NULL if the list is empty. 1874 */ 1875 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1876 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1877 { 1878 struct sk_buff *skb = skb_peek_tail(list); 1879 if (skb) 1880 __skb_unlink(skb, list); 1881 return skb; 1882 } 1883 1884 1885 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1886 { 1887 return skb->data_len; 1888 } 1889 1890 static inline unsigned int skb_headlen(const struct sk_buff *skb) 1891 { 1892 return skb->len - skb->data_len; 1893 } 1894 1895 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 1896 { 1897 unsigned int i, len = 0; 1898 1899 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 1900 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1901 return len; 1902 } 1903 1904 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 1905 { 1906 return skb_headlen(skb) + __skb_pagelen(skb); 1907 } 1908 1909 /** 1910 * __skb_fill_page_desc - initialise a paged fragment in an skb 1911 * @skb: buffer containing fragment to be initialised 1912 * @i: paged fragment index to initialise 1913 * @page: the page to use for this fragment 1914 * @off: the offset to the data with @page 1915 * @size: the length of the data 1916 * 1917 * Initialises the @i'th fragment of @skb to point to &size bytes at 1918 * offset @off within @page. 1919 * 1920 * Does not take any additional reference on the fragment. 1921 */ 1922 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1923 struct page *page, int off, int size) 1924 { 1925 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1926 1927 /* 1928 * Propagate page pfmemalloc to the skb if we can. The problem is 1929 * that not all callers have unique ownership of the page but rely 1930 * on page_is_pfmemalloc doing the right thing(tm). 1931 */ 1932 frag->page.p = page; 1933 frag->page_offset = off; 1934 skb_frag_size_set(frag, size); 1935 1936 page = compound_head(page); 1937 if (page_is_pfmemalloc(page)) 1938 skb->pfmemalloc = true; 1939 } 1940 1941 /** 1942 * skb_fill_page_desc - initialise a paged fragment in an skb 1943 * @skb: buffer containing fragment to be initialised 1944 * @i: paged fragment index to initialise 1945 * @page: the page to use for this fragment 1946 * @off: the offset to the data with @page 1947 * @size: the length of the data 1948 * 1949 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1950 * @skb to point to @size bytes at offset @off within @page. In 1951 * addition updates @skb such that @i is the last fragment. 1952 * 1953 * Does not take any additional reference on the fragment. 1954 */ 1955 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1956 struct page *page, int off, int size) 1957 { 1958 __skb_fill_page_desc(skb, i, page, off, size); 1959 skb_shinfo(skb)->nr_frags = i + 1; 1960 } 1961 1962 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1963 int size, unsigned int truesize); 1964 1965 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1966 unsigned int truesize); 1967 1968 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1969 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1970 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1971 1972 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1973 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1974 { 1975 return skb->head + skb->tail; 1976 } 1977 1978 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1979 { 1980 skb->tail = skb->data - skb->head; 1981 } 1982 1983 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1984 { 1985 skb_reset_tail_pointer(skb); 1986 skb->tail += offset; 1987 } 1988 1989 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 1990 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1991 { 1992 return skb->tail; 1993 } 1994 1995 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1996 { 1997 skb->tail = skb->data; 1998 } 1999 2000 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2001 { 2002 skb->tail = skb->data + offset; 2003 } 2004 2005 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2006 2007 /* 2008 * Add data to an sk_buff 2009 */ 2010 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2011 void *skb_put(struct sk_buff *skb, unsigned int len); 2012 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2013 { 2014 void *tmp = skb_tail_pointer(skb); 2015 SKB_LINEAR_ASSERT(skb); 2016 skb->tail += len; 2017 skb->len += len; 2018 return tmp; 2019 } 2020 2021 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2022 { 2023 void *tmp = __skb_put(skb, len); 2024 2025 memset(tmp, 0, len); 2026 return tmp; 2027 } 2028 2029 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2030 unsigned int len) 2031 { 2032 void *tmp = __skb_put(skb, len); 2033 2034 memcpy(tmp, data, len); 2035 return tmp; 2036 } 2037 2038 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2039 { 2040 *(u8 *)__skb_put(skb, 1) = val; 2041 } 2042 2043 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2044 { 2045 void *tmp = skb_put(skb, len); 2046 2047 memset(tmp, 0, len); 2048 2049 return tmp; 2050 } 2051 2052 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2053 unsigned int len) 2054 { 2055 void *tmp = skb_put(skb, len); 2056 2057 memcpy(tmp, data, len); 2058 2059 return tmp; 2060 } 2061 2062 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2063 { 2064 *(u8 *)skb_put(skb, 1) = val; 2065 } 2066 2067 void *skb_push(struct sk_buff *skb, unsigned int len); 2068 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2069 { 2070 skb->data -= len; 2071 skb->len += len; 2072 return skb->data; 2073 } 2074 2075 void *skb_pull(struct sk_buff *skb, unsigned int len); 2076 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2077 { 2078 skb->len -= len; 2079 BUG_ON(skb->len < skb->data_len); 2080 return skb->data += len; 2081 } 2082 2083 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2084 { 2085 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2086 } 2087 2088 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2089 2090 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len) 2091 { 2092 if (len > skb_headlen(skb) && 2093 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 2094 return NULL; 2095 skb->len -= len; 2096 return skb->data += len; 2097 } 2098 2099 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2100 { 2101 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 2102 } 2103 2104 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 2105 { 2106 if (likely(len <= skb_headlen(skb))) 2107 return 1; 2108 if (unlikely(len > skb->len)) 2109 return 0; 2110 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 2111 } 2112 2113 void skb_condense(struct sk_buff *skb); 2114 2115 /** 2116 * skb_headroom - bytes at buffer head 2117 * @skb: buffer to check 2118 * 2119 * Return the number of bytes of free space at the head of an &sk_buff. 2120 */ 2121 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2122 { 2123 return skb->data - skb->head; 2124 } 2125 2126 /** 2127 * skb_tailroom - bytes at buffer end 2128 * @skb: buffer to check 2129 * 2130 * Return the number of bytes of free space at the tail of an sk_buff 2131 */ 2132 static inline int skb_tailroom(const struct sk_buff *skb) 2133 { 2134 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2135 } 2136 2137 /** 2138 * skb_availroom - bytes at buffer end 2139 * @skb: buffer to check 2140 * 2141 * Return the number of bytes of free space at the tail of an sk_buff 2142 * allocated by sk_stream_alloc() 2143 */ 2144 static inline int skb_availroom(const struct sk_buff *skb) 2145 { 2146 if (skb_is_nonlinear(skb)) 2147 return 0; 2148 2149 return skb->end - skb->tail - skb->reserved_tailroom; 2150 } 2151 2152 /** 2153 * skb_reserve - adjust headroom 2154 * @skb: buffer to alter 2155 * @len: bytes to move 2156 * 2157 * Increase the headroom of an empty &sk_buff by reducing the tail 2158 * room. This is only allowed for an empty buffer. 2159 */ 2160 static inline void skb_reserve(struct sk_buff *skb, int len) 2161 { 2162 skb->data += len; 2163 skb->tail += len; 2164 } 2165 2166 /** 2167 * skb_tailroom_reserve - adjust reserved_tailroom 2168 * @skb: buffer to alter 2169 * @mtu: maximum amount of headlen permitted 2170 * @needed_tailroom: minimum amount of reserved_tailroom 2171 * 2172 * Set reserved_tailroom so that headlen can be as large as possible but 2173 * not larger than mtu and tailroom cannot be smaller than 2174 * needed_tailroom. 2175 * The required headroom should already have been reserved before using 2176 * this function. 2177 */ 2178 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2179 unsigned int needed_tailroom) 2180 { 2181 SKB_LINEAR_ASSERT(skb); 2182 if (mtu < skb_tailroom(skb) - needed_tailroom) 2183 /* use at most mtu */ 2184 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2185 else 2186 /* use up to all available space */ 2187 skb->reserved_tailroom = needed_tailroom; 2188 } 2189 2190 #define ENCAP_TYPE_ETHER 0 2191 #define ENCAP_TYPE_IPPROTO 1 2192 2193 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2194 __be16 protocol) 2195 { 2196 skb->inner_protocol = protocol; 2197 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2198 } 2199 2200 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2201 __u8 ipproto) 2202 { 2203 skb->inner_ipproto = ipproto; 2204 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2205 } 2206 2207 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2208 { 2209 skb->inner_mac_header = skb->mac_header; 2210 skb->inner_network_header = skb->network_header; 2211 skb->inner_transport_header = skb->transport_header; 2212 } 2213 2214 static inline void skb_reset_mac_len(struct sk_buff *skb) 2215 { 2216 skb->mac_len = skb->network_header - skb->mac_header; 2217 } 2218 2219 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2220 *skb) 2221 { 2222 return skb->head + skb->inner_transport_header; 2223 } 2224 2225 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2226 { 2227 return skb_inner_transport_header(skb) - skb->data; 2228 } 2229 2230 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2231 { 2232 skb->inner_transport_header = skb->data - skb->head; 2233 } 2234 2235 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2236 const int offset) 2237 { 2238 skb_reset_inner_transport_header(skb); 2239 skb->inner_transport_header += offset; 2240 } 2241 2242 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2243 { 2244 return skb->head + skb->inner_network_header; 2245 } 2246 2247 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2248 { 2249 skb->inner_network_header = skb->data - skb->head; 2250 } 2251 2252 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2253 const int offset) 2254 { 2255 skb_reset_inner_network_header(skb); 2256 skb->inner_network_header += offset; 2257 } 2258 2259 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2260 { 2261 return skb->head + skb->inner_mac_header; 2262 } 2263 2264 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2265 { 2266 skb->inner_mac_header = skb->data - skb->head; 2267 } 2268 2269 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2270 const int offset) 2271 { 2272 skb_reset_inner_mac_header(skb); 2273 skb->inner_mac_header += offset; 2274 } 2275 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2276 { 2277 return skb->transport_header != (typeof(skb->transport_header))~0U; 2278 } 2279 2280 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2281 { 2282 return skb->head + skb->transport_header; 2283 } 2284 2285 static inline void skb_reset_transport_header(struct sk_buff *skb) 2286 { 2287 skb->transport_header = skb->data - skb->head; 2288 } 2289 2290 static inline void skb_set_transport_header(struct sk_buff *skb, 2291 const int offset) 2292 { 2293 skb_reset_transport_header(skb); 2294 skb->transport_header += offset; 2295 } 2296 2297 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2298 { 2299 return skb->head + skb->network_header; 2300 } 2301 2302 static inline void skb_reset_network_header(struct sk_buff *skb) 2303 { 2304 skb->network_header = skb->data - skb->head; 2305 } 2306 2307 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2308 { 2309 skb_reset_network_header(skb); 2310 skb->network_header += offset; 2311 } 2312 2313 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2314 { 2315 return skb->head + skb->mac_header; 2316 } 2317 2318 static inline int skb_mac_offset(const struct sk_buff *skb) 2319 { 2320 return skb_mac_header(skb) - skb->data; 2321 } 2322 2323 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2324 { 2325 return skb->network_header - skb->mac_header; 2326 } 2327 2328 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2329 { 2330 return skb->mac_header != (typeof(skb->mac_header))~0U; 2331 } 2332 2333 static inline void skb_reset_mac_header(struct sk_buff *skb) 2334 { 2335 skb->mac_header = skb->data - skb->head; 2336 } 2337 2338 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2339 { 2340 skb_reset_mac_header(skb); 2341 skb->mac_header += offset; 2342 } 2343 2344 static inline void skb_pop_mac_header(struct sk_buff *skb) 2345 { 2346 skb->mac_header = skb->network_header; 2347 } 2348 2349 static inline void skb_probe_transport_header(struct sk_buff *skb, 2350 const int offset_hint) 2351 { 2352 struct flow_keys_basic keys; 2353 2354 if (skb_transport_header_was_set(skb)) 2355 return; 2356 2357 if (skb_flow_dissect_flow_keys_basic(skb, &keys, 0, 0, 0, 0, 0)) 2358 skb_set_transport_header(skb, keys.control.thoff); 2359 else 2360 skb_set_transport_header(skb, offset_hint); 2361 } 2362 2363 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2364 { 2365 if (skb_mac_header_was_set(skb)) { 2366 const unsigned char *old_mac = skb_mac_header(skb); 2367 2368 skb_set_mac_header(skb, -skb->mac_len); 2369 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2370 } 2371 } 2372 2373 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2374 { 2375 return skb->csum_start - skb_headroom(skb); 2376 } 2377 2378 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2379 { 2380 return skb->head + skb->csum_start; 2381 } 2382 2383 static inline int skb_transport_offset(const struct sk_buff *skb) 2384 { 2385 return skb_transport_header(skb) - skb->data; 2386 } 2387 2388 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2389 { 2390 return skb->transport_header - skb->network_header; 2391 } 2392 2393 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2394 { 2395 return skb->inner_transport_header - skb->inner_network_header; 2396 } 2397 2398 static inline int skb_network_offset(const struct sk_buff *skb) 2399 { 2400 return skb_network_header(skb) - skb->data; 2401 } 2402 2403 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2404 { 2405 return skb_inner_network_header(skb) - skb->data; 2406 } 2407 2408 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2409 { 2410 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2411 } 2412 2413 /* 2414 * CPUs often take a performance hit when accessing unaligned memory 2415 * locations. The actual performance hit varies, it can be small if the 2416 * hardware handles it or large if we have to take an exception and fix it 2417 * in software. 2418 * 2419 * Since an ethernet header is 14 bytes network drivers often end up with 2420 * the IP header at an unaligned offset. The IP header can be aligned by 2421 * shifting the start of the packet by 2 bytes. Drivers should do this 2422 * with: 2423 * 2424 * skb_reserve(skb, NET_IP_ALIGN); 2425 * 2426 * The downside to this alignment of the IP header is that the DMA is now 2427 * unaligned. On some architectures the cost of an unaligned DMA is high 2428 * and this cost outweighs the gains made by aligning the IP header. 2429 * 2430 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2431 * to be overridden. 2432 */ 2433 #ifndef NET_IP_ALIGN 2434 #define NET_IP_ALIGN 2 2435 #endif 2436 2437 /* 2438 * The networking layer reserves some headroom in skb data (via 2439 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2440 * the header has to grow. In the default case, if the header has to grow 2441 * 32 bytes or less we avoid the reallocation. 2442 * 2443 * Unfortunately this headroom changes the DMA alignment of the resulting 2444 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2445 * on some architectures. An architecture can override this value, 2446 * perhaps setting it to a cacheline in size (since that will maintain 2447 * cacheline alignment of the DMA). It must be a power of 2. 2448 * 2449 * Various parts of the networking layer expect at least 32 bytes of 2450 * headroom, you should not reduce this. 2451 * 2452 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2453 * to reduce average number of cache lines per packet. 2454 * get_rps_cpus() for example only access one 64 bytes aligned block : 2455 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2456 */ 2457 #ifndef NET_SKB_PAD 2458 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2459 #endif 2460 2461 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2462 2463 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 2464 { 2465 if (unlikely(skb_is_nonlinear(skb))) { 2466 WARN_ON(1); 2467 return; 2468 } 2469 skb->len = len; 2470 skb_set_tail_pointer(skb, len); 2471 } 2472 2473 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2474 { 2475 __skb_set_length(skb, len); 2476 } 2477 2478 void skb_trim(struct sk_buff *skb, unsigned int len); 2479 2480 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2481 { 2482 if (skb->data_len) 2483 return ___pskb_trim(skb, len); 2484 __skb_trim(skb, len); 2485 return 0; 2486 } 2487 2488 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2489 { 2490 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2491 } 2492 2493 /** 2494 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2495 * @skb: buffer to alter 2496 * @len: new length 2497 * 2498 * This is identical to pskb_trim except that the caller knows that 2499 * the skb is not cloned so we should never get an error due to out- 2500 * of-memory. 2501 */ 2502 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2503 { 2504 int err = pskb_trim(skb, len); 2505 BUG_ON(err); 2506 } 2507 2508 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 2509 { 2510 unsigned int diff = len - skb->len; 2511 2512 if (skb_tailroom(skb) < diff) { 2513 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 2514 GFP_ATOMIC); 2515 if (ret) 2516 return ret; 2517 } 2518 __skb_set_length(skb, len); 2519 return 0; 2520 } 2521 2522 /** 2523 * skb_orphan - orphan a buffer 2524 * @skb: buffer to orphan 2525 * 2526 * If a buffer currently has an owner then we call the owner's 2527 * destructor function and make the @skb unowned. The buffer continues 2528 * to exist but is no longer charged to its former owner. 2529 */ 2530 static inline void skb_orphan(struct sk_buff *skb) 2531 { 2532 if (skb->destructor) { 2533 skb->destructor(skb); 2534 skb->destructor = NULL; 2535 skb->sk = NULL; 2536 } else { 2537 BUG_ON(skb->sk); 2538 } 2539 } 2540 2541 /** 2542 * skb_orphan_frags - orphan the frags contained in a buffer 2543 * @skb: buffer to orphan frags from 2544 * @gfp_mask: allocation mask for replacement pages 2545 * 2546 * For each frag in the SKB which needs a destructor (i.e. has an 2547 * owner) create a copy of that frag and release the original 2548 * page by calling the destructor. 2549 */ 2550 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2551 { 2552 if (likely(!skb_zcopy(skb))) 2553 return 0; 2554 if (skb_uarg(skb)->callback == sock_zerocopy_callback) 2555 return 0; 2556 return skb_copy_ubufs(skb, gfp_mask); 2557 } 2558 2559 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 2560 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 2561 { 2562 if (likely(!skb_zcopy(skb))) 2563 return 0; 2564 return skb_copy_ubufs(skb, gfp_mask); 2565 } 2566 2567 /** 2568 * __skb_queue_purge - empty a list 2569 * @list: list to empty 2570 * 2571 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2572 * the list and one reference dropped. This function does not take the 2573 * list lock and the caller must hold the relevant locks to use it. 2574 */ 2575 void skb_queue_purge(struct sk_buff_head *list); 2576 static inline void __skb_queue_purge(struct sk_buff_head *list) 2577 { 2578 struct sk_buff *skb; 2579 while ((skb = __skb_dequeue(list)) != NULL) 2580 kfree_skb(skb); 2581 } 2582 2583 void skb_rbtree_purge(struct rb_root *root); 2584 2585 void *netdev_alloc_frag(unsigned int fragsz); 2586 2587 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2588 gfp_t gfp_mask); 2589 2590 /** 2591 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2592 * @dev: network device to receive on 2593 * @length: length to allocate 2594 * 2595 * Allocate a new &sk_buff and assign it a usage count of one. The 2596 * buffer has unspecified headroom built in. Users should allocate 2597 * the headroom they think they need without accounting for the 2598 * built in space. The built in space is used for optimisations. 2599 * 2600 * %NULL is returned if there is no free memory. Although this function 2601 * allocates memory it can be called from an interrupt. 2602 */ 2603 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2604 unsigned int length) 2605 { 2606 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2607 } 2608 2609 /* legacy helper around __netdev_alloc_skb() */ 2610 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2611 gfp_t gfp_mask) 2612 { 2613 return __netdev_alloc_skb(NULL, length, gfp_mask); 2614 } 2615 2616 /* legacy helper around netdev_alloc_skb() */ 2617 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2618 { 2619 return netdev_alloc_skb(NULL, length); 2620 } 2621 2622 2623 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2624 unsigned int length, gfp_t gfp) 2625 { 2626 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2627 2628 if (NET_IP_ALIGN && skb) 2629 skb_reserve(skb, NET_IP_ALIGN); 2630 return skb; 2631 } 2632 2633 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2634 unsigned int length) 2635 { 2636 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2637 } 2638 2639 static inline void skb_free_frag(void *addr) 2640 { 2641 page_frag_free(addr); 2642 } 2643 2644 void *napi_alloc_frag(unsigned int fragsz); 2645 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2646 unsigned int length, gfp_t gfp_mask); 2647 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2648 unsigned int length) 2649 { 2650 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2651 } 2652 void napi_consume_skb(struct sk_buff *skb, int budget); 2653 2654 void __kfree_skb_flush(void); 2655 void __kfree_skb_defer(struct sk_buff *skb); 2656 2657 /** 2658 * __dev_alloc_pages - allocate page for network Rx 2659 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2660 * @order: size of the allocation 2661 * 2662 * Allocate a new page. 2663 * 2664 * %NULL is returned if there is no free memory. 2665 */ 2666 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2667 unsigned int order) 2668 { 2669 /* This piece of code contains several assumptions. 2670 * 1. This is for device Rx, therefor a cold page is preferred. 2671 * 2. The expectation is the user wants a compound page. 2672 * 3. If requesting a order 0 page it will not be compound 2673 * due to the check to see if order has a value in prep_new_page 2674 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2675 * code in gfp_to_alloc_flags that should be enforcing this. 2676 */ 2677 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 2678 2679 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2680 } 2681 2682 static inline struct page *dev_alloc_pages(unsigned int order) 2683 { 2684 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 2685 } 2686 2687 /** 2688 * __dev_alloc_page - allocate a page for network Rx 2689 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2690 * 2691 * Allocate a new page. 2692 * 2693 * %NULL is returned if there is no free memory. 2694 */ 2695 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2696 { 2697 return __dev_alloc_pages(gfp_mask, 0); 2698 } 2699 2700 static inline struct page *dev_alloc_page(void) 2701 { 2702 return dev_alloc_pages(0); 2703 } 2704 2705 /** 2706 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2707 * @page: The page that was allocated from skb_alloc_page 2708 * @skb: The skb that may need pfmemalloc set 2709 */ 2710 static inline void skb_propagate_pfmemalloc(struct page *page, 2711 struct sk_buff *skb) 2712 { 2713 if (page_is_pfmemalloc(page)) 2714 skb->pfmemalloc = true; 2715 } 2716 2717 /** 2718 * skb_frag_page - retrieve the page referred to by a paged fragment 2719 * @frag: the paged fragment 2720 * 2721 * Returns the &struct page associated with @frag. 2722 */ 2723 static inline struct page *skb_frag_page(const skb_frag_t *frag) 2724 { 2725 return frag->page.p; 2726 } 2727 2728 /** 2729 * __skb_frag_ref - take an addition reference on a paged fragment. 2730 * @frag: the paged fragment 2731 * 2732 * Takes an additional reference on the paged fragment @frag. 2733 */ 2734 static inline void __skb_frag_ref(skb_frag_t *frag) 2735 { 2736 get_page(skb_frag_page(frag)); 2737 } 2738 2739 /** 2740 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2741 * @skb: the buffer 2742 * @f: the fragment offset. 2743 * 2744 * Takes an additional reference on the @f'th paged fragment of @skb. 2745 */ 2746 static inline void skb_frag_ref(struct sk_buff *skb, int f) 2747 { 2748 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2749 } 2750 2751 /** 2752 * __skb_frag_unref - release a reference on a paged fragment. 2753 * @frag: the paged fragment 2754 * 2755 * Releases a reference on the paged fragment @frag. 2756 */ 2757 static inline void __skb_frag_unref(skb_frag_t *frag) 2758 { 2759 put_page(skb_frag_page(frag)); 2760 } 2761 2762 /** 2763 * skb_frag_unref - release a reference on a paged fragment of an skb. 2764 * @skb: the buffer 2765 * @f: the fragment offset 2766 * 2767 * Releases a reference on the @f'th paged fragment of @skb. 2768 */ 2769 static inline void skb_frag_unref(struct sk_buff *skb, int f) 2770 { 2771 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2772 } 2773 2774 /** 2775 * skb_frag_address - gets the address of the data contained in a paged fragment 2776 * @frag: the paged fragment buffer 2777 * 2778 * Returns the address of the data within @frag. The page must already 2779 * be mapped. 2780 */ 2781 static inline void *skb_frag_address(const skb_frag_t *frag) 2782 { 2783 return page_address(skb_frag_page(frag)) + frag->page_offset; 2784 } 2785 2786 /** 2787 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2788 * @frag: the paged fragment buffer 2789 * 2790 * Returns the address of the data within @frag. Checks that the page 2791 * is mapped and returns %NULL otherwise. 2792 */ 2793 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2794 { 2795 void *ptr = page_address(skb_frag_page(frag)); 2796 if (unlikely(!ptr)) 2797 return NULL; 2798 2799 return ptr + frag->page_offset; 2800 } 2801 2802 /** 2803 * __skb_frag_set_page - sets the page contained in a paged fragment 2804 * @frag: the paged fragment 2805 * @page: the page to set 2806 * 2807 * Sets the fragment @frag to contain @page. 2808 */ 2809 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2810 { 2811 frag->page.p = page; 2812 } 2813 2814 /** 2815 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2816 * @skb: the buffer 2817 * @f: the fragment offset 2818 * @page: the page to set 2819 * 2820 * Sets the @f'th fragment of @skb to contain @page. 2821 */ 2822 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2823 struct page *page) 2824 { 2825 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2826 } 2827 2828 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2829 2830 /** 2831 * skb_frag_dma_map - maps a paged fragment via the DMA API 2832 * @dev: the device to map the fragment to 2833 * @frag: the paged fragment to map 2834 * @offset: the offset within the fragment (starting at the 2835 * fragment's own offset) 2836 * @size: the number of bytes to map 2837 * @dir: the direction of the mapping (``PCI_DMA_*``) 2838 * 2839 * Maps the page associated with @frag to @device. 2840 */ 2841 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2842 const skb_frag_t *frag, 2843 size_t offset, size_t size, 2844 enum dma_data_direction dir) 2845 { 2846 return dma_map_page(dev, skb_frag_page(frag), 2847 frag->page_offset + offset, size, dir); 2848 } 2849 2850 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2851 gfp_t gfp_mask) 2852 { 2853 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2854 } 2855 2856 2857 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2858 gfp_t gfp_mask) 2859 { 2860 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2861 } 2862 2863 2864 /** 2865 * skb_clone_writable - is the header of a clone writable 2866 * @skb: buffer to check 2867 * @len: length up to which to write 2868 * 2869 * Returns true if modifying the header part of the cloned buffer 2870 * does not requires the data to be copied. 2871 */ 2872 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2873 { 2874 return !skb_header_cloned(skb) && 2875 skb_headroom(skb) + len <= skb->hdr_len; 2876 } 2877 2878 static inline int skb_try_make_writable(struct sk_buff *skb, 2879 unsigned int write_len) 2880 { 2881 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 2882 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2883 } 2884 2885 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2886 int cloned) 2887 { 2888 int delta = 0; 2889 2890 if (headroom > skb_headroom(skb)) 2891 delta = headroom - skb_headroom(skb); 2892 2893 if (delta || cloned) 2894 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2895 GFP_ATOMIC); 2896 return 0; 2897 } 2898 2899 /** 2900 * skb_cow - copy header of skb when it is required 2901 * @skb: buffer to cow 2902 * @headroom: needed headroom 2903 * 2904 * If the skb passed lacks sufficient headroom or its data part 2905 * is shared, data is reallocated. If reallocation fails, an error 2906 * is returned and original skb is not changed. 2907 * 2908 * The result is skb with writable area skb->head...skb->tail 2909 * and at least @headroom of space at head. 2910 */ 2911 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2912 { 2913 return __skb_cow(skb, headroom, skb_cloned(skb)); 2914 } 2915 2916 /** 2917 * skb_cow_head - skb_cow but only making the head writable 2918 * @skb: buffer to cow 2919 * @headroom: needed headroom 2920 * 2921 * This function is identical to skb_cow except that we replace the 2922 * skb_cloned check by skb_header_cloned. It should be used when 2923 * you only need to push on some header and do not need to modify 2924 * the data. 2925 */ 2926 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2927 { 2928 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2929 } 2930 2931 /** 2932 * skb_padto - pad an skbuff up to a minimal size 2933 * @skb: buffer to pad 2934 * @len: minimal length 2935 * 2936 * Pads up a buffer to ensure the trailing bytes exist and are 2937 * blanked. If the buffer already contains sufficient data it 2938 * is untouched. Otherwise it is extended. Returns zero on 2939 * success. The skb is freed on error. 2940 */ 2941 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2942 { 2943 unsigned int size = skb->len; 2944 if (likely(size >= len)) 2945 return 0; 2946 return skb_pad(skb, len - size); 2947 } 2948 2949 /** 2950 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2951 * @skb: buffer to pad 2952 * @len: minimal length 2953 * @free_on_error: free buffer on error 2954 * 2955 * Pads up a buffer to ensure the trailing bytes exist and are 2956 * blanked. If the buffer already contains sufficient data it 2957 * is untouched. Otherwise it is extended. Returns zero on 2958 * success. The skb is freed on error if @free_on_error is true. 2959 */ 2960 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len, 2961 bool free_on_error) 2962 { 2963 unsigned int size = skb->len; 2964 2965 if (unlikely(size < len)) { 2966 len -= size; 2967 if (__skb_pad(skb, len, free_on_error)) 2968 return -ENOMEM; 2969 __skb_put(skb, len); 2970 } 2971 return 0; 2972 } 2973 2974 /** 2975 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2976 * @skb: buffer to pad 2977 * @len: minimal length 2978 * 2979 * Pads up a buffer to ensure the trailing bytes exist and are 2980 * blanked. If the buffer already contains sufficient data it 2981 * is untouched. Otherwise it is extended. Returns zero on 2982 * success. The skb is freed on error. 2983 */ 2984 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2985 { 2986 return __skb_put_padto(skb, len, true); 2987 } 2988 2989 static inline int skb_add_data(struct sk_buff *skb, 2990 struct iov_iter *from, int copy) 2991 { 2992 const int off = skb->len; 2993 2994 if (skb->ip_summed == CHECKSUM_NONE) { 2995 __wsum csum = 0; 2996 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 2997 &csum, from)) { 2998 skb->csum = csum_block_add(skb->csum, csum, off); 2999 return 0; 3000 } 3001 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3002 return 0; 3003 3004 __skb_trim(skb, off); 3005 return -EFAULT; 3006 } 3007 3008 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3009 const struct page *page, int off) 3010 { 3011 if (skb_zcopy(skb)) 3012 return false; 3013 if (i) { 3014 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 3015 3016 return page == skb_frag_page(frag) && 3017 off == frag->page_offset + skb_frag_size(frag); 3018 } 3019 return false; 3020 } 3021 3022 static inline int __skb_linearize(struct sk_buff *skb) 3023 { 3024 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3025 } 3026 3027 /** 3028 * skb_linearize - convert paged skb to linear one 3029 * @skb: buffer to linarize 3030 * 3031 * If there is no free memory -ENOMEM is returned, otherwise zero 3032 * is returned and the old skb data released. 3033 */ 3034 static inline int skb_linearize(struct sk_buff *skb) 3035 { 3036 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3037 } 3038 3039 /** 3040 * skb_has_shared_frag - can any frag be overwritten 3041 * @skb: buffer to test 3042 * 3043 * Return true if the skb has at least one frag that might be modified 3044 * by an external entity (as in vmsplice()/sendfile()) 3045 */ 3046 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3047 { 3048 return skb_is_nonlinear(skb) && 3049 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 3050 } 3051 3052 /** 3053 * skb_linearize_cow - make sure skb is linear and writable 3054 * @skb: buffer to process 3055 * 3056 * If there is no free memory -ENOMEM is returned, otherwise zero 3057 * is returned and the old skb data released. 3058 */ 3059 static inline int skb_linearize_cow(struct sk_buff *skb) 3060 { 3061 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3062 __skb_linearize(skb) : 0; 3063 } 3064 3065 static __always_inline void 3066 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3067 unsigned int off) 3068 { 3069 if (skb->ip_summed == CHECKSUM_COMPLETE) 3070 skb->csum = csum_block_sub(skb->csum, 3071 csum_partial(start, len, 0), off); 3072 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3073 skb_checksum_start_offset(skb) < 0) 3074 skb->ip_summed = CHECKSUM_NONE; 3075 } 3076 3077 /** 3078 * skb_postpull_rcsum - update checksum for received skb after pull 3079 * @skb: buffer to update 3080 * @start: start of data before pull 3081 * @len: length of data pulled 3082 * 3083 * After doing a pull on a received packet, you need to call this to 3084 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3085 * CHECKSUM_NONE so that it can be recomputed from scratch. 3086 */ 3087 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3088 const void *start, unsigned int len) 3089 { 3090 __skb_postpull_rcsum(skb, start, len, 0); 3091 } 3092 3093 static __always_inline void 3094 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3095 unsigned int off) 3096 { 3097 if (skb->ip_summed == CHECKSUM_COMPLETE) 3098 skb->csum = csum_block_add(skb->csum, 3099 csum_partial(start, len, 0), off); 3100 } 3101 3102 /** 3103 * skb_postpush_rcsum - update checksum for received skb after push 3104 * @skb: buffer to update 3105 * @start: start of data after push 3106 * @len: length of data pushed 3107 * 3108 * After doing a push on a received packet, you need to call this to 3109 * update the CHECKSUM_COMPLETE checksum. 3110 */ 3111 static inline void skb_postpush_rcsum(struct sk_buff *skb, 3112 const void *start, unsigned int len) 3113 { 3114 __skb_postpush_rcsum(skb, start, len, 0); 3115 } 3116 3117 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3118 3119 /** 3120 * skb_push_rcsum - push skb and update receive checksum 3121 * @skb: buffer to update 3122 * @len: length of data pulled 3123 * 3124 * This function performs an skb_push on the packet and updates 3125 * the CHECKSUM_COMPLETE checksum. It should be used on 3126 * receive path processing instead of skb_push unless you know 3127 * that the checksum difference is zero (e.g., a valid IP header) 3128 * or you are setting ip_summed to CHECKSUM_NONE. 3129 */ 3130 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3131 { 3132 skb_push(skb, len); 3133 skb_postpush_rcsum(skb, skb->data, len); 3134 return skb->data; 3135 } 3136 3137 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 3138 /** 3139 * pskb_trim_rcsum - trim received skb and update checksum 3140 * @skb: buffer to trim 3141 * @len: new length 3142 * 3143 * This is exactly the same as pskb_trim except that it ensures the 3144 * checksum of received packets are still valid after the operation. 3145 */ 3146 3147 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3148 { 3149 if (likely(len >= skb->len)) 3150 return 0; 3151 return pskb_trim_rcsum_slow(skb, len); 3152 } 3153 3154 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3155 { 3156 if (skb->ip_summed == CHECKSUM_COMPLETE) 3157 skb->ip_summed = CHECKSUM_NONE; 3158 __skb_trim(skb, len); 3159 return 0; 3160 } 3161 3162 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3163 { 3164 if (skb->ip_summed == CHECKSUM_COMPLETE) 3165 skb->ip_summed = CHECKSUM_NONE; 3166 return __skb_grow(skb, len); 3167 } 3168 3169 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3170 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 3171 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 3172 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3173 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3174 3175 #define skb_queue_walk(queue, skb) \ 3176 for (skb = (queue)->next; \ 3177 skb != (struct sk_buff *)(queue); \ 3178 skb = skb->next) 3179 3180 #define skb_queue_walk_safe(queue, skb, tmp) \ 3181 for (skb = (queue)->next, tmp = skb->next; \ 3182 skb != (struct sk_buff *)(queue); \ 3183 skb = tmp, tmp = skb->next) 3184 3185 #define skb_queue_walk_from(queue, skb) \ 3186 for (; skb != (struct sk_buff *)(queue); \ 3187 skb = skb->next) 3188 3189 #define skb_rbtree_walk(skb, root) \ 3190 for (skb = skb_rb_first(root); skb != NULL; \ 3191 skb = skb_rb_next(skb)) 3192 3193 #define skb_rbtree_walk_from(skb) \ 3194 for (; skb != NULL; \ 3195 skb = skb_rb_next(skb)) 3196 3197 #define skb_rbtree_walk_from_safe(skb, tmp) \ 3198 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3199 skb = tmp) 3200 3201 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 3202 for (tmp = skb->next; \ 3203 skb != (struct sk_buff *)(queue); \ 3204 skb = tmp, tmp = skb->next) 3205 3206 #define skb_queue_reverse_walk(queue, skb) \ 3207 for (skb = (queue)->prev; \ 3208 skb != (struct sk_buff *)(queue); \ 3209 skb = skb->prev) 3210 3211 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3212 for (skb = (queue)->prev, tmp = skb->prev; \ 3213 skb != (struct sk_buff *)(queue); \ 3214 skb = tmp, tmp = skb->prev) 3215 3216 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3217 for (tmp = skb->prev; \ 3218 skb != (struct sk_buff *)(queue); \ 3219 skb = tmp, tmp = skb->prev) 3220 3221 static inline bool skb_has_frag_list(const struct sk_buff *skb) 3222 { 3223 return skb_shinfo(skb)->frag_list != NULL; 3224 } 3225 3226 static inline void skb_frag_list_init(struct sk_buff *skb) 3227 { 3228 skb_shinfo(skb)->frag_list = NULL; 3229 } 3230 3231 #define skb_walk_frags(skb, iter) \ 3232 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3233 3234 3235 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, 3236 const struct sk_buff *skb); 3237 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3238 struct sk_buff_head *queue, 3239 unsigned int flags, 3240 void (*destructor)(struct sock *sk, 3241 struct sk_buff *skb), 3242 int *peeked, int *off, int *err, 3243 struct sk_buff **last); 3244 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags, 3245 void (*destructor)(struct sock *sk, 3246 struct sk_buff *skb), 3247 int *peeked, int *off, int *err, 3248 struct sk_buff **last); 3249 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 3250 void (*destructor)(struct sock *sk, 3251 struct sk_buff *skb), 3252 int *peeked, int *off, int *err); 3253 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 3254 int *err); 3255 __poll_t datagram_poll(struct file *file, struct socket *sock, 3256 struct poll_table_struct *wait); 3257 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3258 struct iov_iter *to, int size); 3259 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3260 struct msghdr *msg, int size) 3261 { 3262 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3263 } 3264 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3265 struct msghdr *msg); 3266 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3267 struct iov_iter *from, int len); 3268 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3269 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3270 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3271 static inline void skb_free_datagram_locked(struct sock *sk, 3272 struct sk_buff *skb) 3273 { 3274 __skb_free_datagram_locked(sk, skb, 0); 3275 } 3276 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3277 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3278 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3279 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3280 int len, __wsum csum); 3281 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3282 struct pipe_inode_info *pipe, unsigned int len, 3283 unsigned int flags); 3284 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 3285 int len); 3286 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 3287 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3288 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3289 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3290 int len, int hlen); 3291 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3292 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3293 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3294 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu); 3295 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len); 3296 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3297 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3298 int skb_ensure_writable(struct sk_buff *skb, int write_len); 3299 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3300 int skb_vlan_pop(struct sk_buff *skb); 3301 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3302 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3303 gfp_t gfp); 3304 3305 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3306 { 3307 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 3308 } 3309 3310 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3311 { 3312 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3313 } 3314 3315 struct skb_checksum_ops { 3316 __wsum (*update)(const void *mem, int len, __wsum wsum); 3317 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3318 }; 3319 3320 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 3321 3322 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3323 __wsum csum, const struct skb_checksum_ops *ops); 3324 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3325 __wsum csum); 3326 3327 static inline void * __must_check 3328 __skb_header_pointer(const struct sk_buff *skb, int offset, 3329 int len, void *data, int hlen, void *buffer) 3330 { 3331 if (hlen - offset >= len) 3332 return data + offset; 3333 3334 if (!skb || 3335 skb_copy_bits(skb, offset, buffer, len) < 0) 3336 return NULL; 3337 3338 return buffer; 3339 } 3340 3341 static inline void * __must_check 3342 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 3343 { 3344 return __skb_header_pointer(skb, offset, len, skb->data, 3345 skb_headlen(skb), buffer); 3346 } 3347 3348 /** 3349 * skb_needs_linearize - check if we need to linearize a given skb 3350 * depending on the given device features. 3351 * @skb: socket buffer to check 3352 * @features: net device features 3353 * 3354 * Returns true if either: 3355 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3356 * 2. skb is fragmented and the device does not support SG. 3357 */ 3358 static inline bool skb_needs_linearize(struct sk_buff *skb, 3359 netdev_features_t features) 3360 { 3361 return skb_is_nonlinear(skb) && 3362 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3363 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3364 } 3365 3366 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3367 void *to, 3368 const unsigned int len) 3369 { 3370 memcpy(to, skb->data, len); 3371 } 3372 3373 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3374 const int offset, void *to, 3375 const unsigned int len) 3376 { 3377 memcpy(to, skb->data + offset, len); 3378 } 3379 3380 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3381 const void *from, 3382 const unsigned int len) 3383 { 3384 memcpy(skb->data, from, len); 3385 } 3386 3387 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3388 const int offset, 3389 const void *from, 3390 const unsigned int len) 3391 { 3392 memcpy(skb->data + offset, from, len); 3393 } 3394 3395 void skb_init(void); 3396 3397 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3398 { 3399 return skb->tstamp; 3400 } 3401 3402 /** 3403 * skb_get_timestamp - get timestamp from a skb 3404 * @skb: skb to get stamp from 3405 * @stamp: pointer to struct timeval to store stamp in 3406 * 3407 * Timestamps are stored in the skb as offsets to a base timestamp. 3408 * This function converts the offset back to a struct timeval and stores 3409 * it in stamp. 3410 */ 3411 static inline void skb_get_timestamp(const struct sk_buff *skb, 3412 struct timeval *stamp) 3413 { 3414 *stamp = ktime_to_timeval(skb->tstamp); 3415 } 3416 3417 static inline void skb_get_timestampns(const struct sk_buff *skb, 3418 struct timespec *stamp) 3419 { 3420 *stamp = ktime_to_timespec(skb->tstamp); 3421 } 3422 3423 static inline void __net_timestamp(struct sk_buff *skb) 3424 { 3425 skb->tstamp = ktime_get_real(); 3426 } 3427 3428 static inline ktime_t net_timedelta(ktime_t t) 3429 { 3430 return ktime_sub(ktime_get_real(), t); 3431 } 3432 3433 static inline ktime_t net_invalid_timestamp(void) 3434 { 3435 return 0; 3436 } 3437 3438 static inline u8 skb_metadata_len(const struct sk_buff *skb) 3439 { 3440 return skb_shinfo(skb)->meta_len; 3441 } 3442 3443 static inline void *skb_metadata_end(const struct sk_buff *skb) 3444 { 3445 return skb_mac_header(skb); 3446 } 3447 3448 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 3449 const struct sk_buff *skb_b, 3450 u8 meta_len) 3451 { 3452 const void *a = skb_metadata_end(skb_a); 3453 const void *b = skb_metadata_end(skb_b); 3454 /* Using more efficient varaiant than plain call to memcmp(). */ 3455 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64 3456 u64 diffs = 0; 3457 3458 switch (meta_len) { 3459 #define __it(x, op) (x -= sizeof(u##op)) 3460 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 3461 case 32: diffs |= __it_diff(a, b, 64); 3462 case 24: diffs |= __it_diff(a, b, 64); 3463 case 16: diffs |= __it_diff(a, b, 64); 3464 case 8: diffs |= __it_diff(a, b, 64); 3465 break; 3466 case 28: diffs |= __it_diff(a, b, 64); 3467 case 20: diffs |= __it_diff(a, b, 64); 3468 case 12: diffs |= __it_diff(a, b, 64); 3469 case 4: diffs |= __it_diff(a, b, 32); 3470 break; 3471 } 3472 return diffs; 3473 #else 3474 return memcmp(a - meta_len, b - meta_len, meta_len); 3475 #endif 3476 } 3477 3478 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 3479 const struct sk_buff *skb_b) 3480 { 3481 u8 len_a = skb_metadata_len(skb_a); 3482 u8 len_b = skb_metadata_len(skb_b); 3483 3484 if (!(len_a | len_b)) 3485 return false; 3486 3487 return len_a != len_b ? 3488 true : __skb_metadata_differs(skb_a, skb_b, len_a); 3489 } 3490 3491 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 3492 { 3493 skb_shinfo(skb)->meta_len = meta_len; 3494 } 3495 3496 static inline void skb_metadata_clear(struct sk_buff *skb) 3497 { 3498 skb_metadata_set(skb, 0); 3499 } 3500 3501 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3502 3503 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3504 3505 void skb_clone_tx_timestamp(struct sk_buff *skb); 3506 bool skb_defer_rx_timestamp(struct sk_buff *skb); 3507 3508 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3509 3510 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3511 { 3512 } 3513 3514 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3515 { 3516 return false; 3517 } 3518 3519 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3520 3521 /** 3522 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3523 * 3524 * PHY drivers may accept clones of transmitted packets for 3525 * timestamping via their phy_driver.txtstamp method. These drivers 3526 * must call this function to return the skb back to the stack with a 3527 * timestamp. 3528 * 3529 * @skb: clone of the the original outgoing packet 3530 * @hwtstamps: hardware time stamps 3531 * 3532 */ 3533 void skb_complete_tx_timestamp(struct sk_buff *skb, 3534 struct skb_shared_hwtstamps *hwtstamps); 3535 3536 void __skb_tstamp_tx(struct sk_buff *orig_skb, 3537 struct skb_shared_hwtstamps *hwtstamps, 3538 struct sock *sk, int tstype); 3539 3540 /** 3541 * skb_tstamp_tx - queue clone of skb with send time stamps 3542 * @orig_skb: the original outgoing packet 3543 * @hwtstamps: hardware time stamps, may be NULL if not available 3544 * 3545 * If the skb has a socket associated, then this function clones the 3546 * skb (thus sharing the actual data and optional structures), stores 3547 * the optional hardware time stamping information (if non NULL) or 3548 * generates a software time stamp (otherwise), then queues the clone 3549 * to the error queue of the socket. Errors are silently ignored. 3550 */ 3551 void skb_tstamp_tx(struct sk_buff *orig_skb, 3552 struct skb_shared_hwtstamps *hwtstamps); 3553 3554 /** 3555 * skb_tx_timestamp() - Driver hook for transmit timestamping 3556 * 3557 * Ethernet MAC Drivers should call this function in their hard_xmit() 3558 * function immediately before giving the sk_buff to the MAC hardware. 3559 * 3560 * Specifically, one should make absolutely sure that this function is 3561 * called before TX completion of this packet can trigger. Otherwise 3562 * the packet could potentially already be freed. 3563 * 3564 * @skb: A socket buffer. 3565 */ 3566 static inline void skb_tx_timestamp(struct sk_buff *skb) 3567 { 3568 skb_clone_tx_timestamp(skb); 3569 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 3570 skb_tstamp_tx(skb, NULL); 3571 } 3572 3573 /** 3574 * skb_complete_wifi_ack - deliver skb with wifi status 3575 * 3576 * @skb: the original outgoing packet 3577 * @acked: ack status 3578 * 3579 */ 3580 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3581 3582 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3583 __sum16 __skb_checksum_complete(struct sk_buff *skb); 3584 3585 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3586 { 3587 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3588 skb->csum_valid || 3589 (skb->ip_summed == CHECKSUM_PARTIAL && 3590 skb_checksum_start_offset(skb) >= 0)); 3591 } 3592 3593 /** 3594 * skb_checksum_complete - Calculate checksum of an entire packet 3595 * @skb: packet to process 3596 * 3597 * This function calculates the checksum over the entire packet plus 3598 * the value of skb->csum. The latter can be used to supply the 3599 * checksum of a pseudo header as used by TCP/UDP. It returns the 3600 * checksum. 3601 * 3602 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3603 * this function can be used to verify that checksum on received 3604 * packets. In that case the function should return zero if the 3605 * checksum is correct. In particular, this function will return zero 3606 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3607 * hardware has already verified the correctness of the checksum. 3608 */ 3609 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3610 { 3611 return skb_csum_unnecessary(skb) ? 3612 0 : __skb_checksum_complete(skb); 3613 } 3614 3615 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3616 { 3617 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3618 if (skb->csum_level == 0) 3619 skb->ip_summed = CHECKSUM_NONE; 3620 else 3621 skb->csum_level--; 3622 } 3623 } 3624 3625 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3626 { 3627 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3628 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3629 skb->csum_level++; 3630 } else if (skb->ip_summed == CHECKSUM_NONE) { 3631 skb->ip_summed = CHECKSUM_UNNECESSARY; 3632 skb->csum_level = 0; 3633 } 3634 } 3635 3636 /* Check if we need to perform checksum complete validation. 3637 * 3638 * Returns true if checksum complete is needed, false otherwise 3639 * (either checksum is unnecessary or zero checksum is allowed). 3640 */ 3641 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3642 bool zero_okay, 3643 __sum16 check) 3644 { 3645 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3646 skb->csum_valid = 1; 3647 __skb_decr_checksum_unnecessary(skb); 3648 return false; 3649 } 3650 3651 return true; 3652 } 3653 3654 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 3655 * in checksum_init. 3656 */ 3657 #define CHECKSUM_BREAK 76 3658 3659 /* Unset checksum-complete 3660 * 3661 * Unset checksum complete can be done when packet is being modified 3662 * (uncompressed for instance) and checksum-complete value is 3663 * invalidated. 3664 */ 3665 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3666 { 3667 if (skb->ip_summed == CHECKSUM_COMPLETE) 3668 skb->ip_summed = CHECKSUM_NONE; 3669 } 3670 3671 /* Validate (init) checksum based on checksum complete. 3672 * 3673 * Return values: 3674 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3675 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3676 * checksum is stored in skb->csum for use in __skb_checksum_complete 3677 * non-zero: value of invalid checksum 3678 * 3679 */ 3680 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3681 bool complete, 3682 __wsum psum) 3683 { 3684 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3685 if (!csum_fold(csum_add(psum, skb->csum))) { 3686 skb->csum_valid = 1; 3687 return 0; 3688 } 3689 } 3690 3691 skb->csum = psum; 3692 3693 if (complete || skb->len <= CHECKSUM_BREAK) { 3694 __sum16 csum; 3695 3696 csum = __skb_checksum_complete(skb); 3697 skb->csum_valid = !csum; 3698 return csum; 3699 } 3700 3701 return 0; 3702 } 3703 3704 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3705 { 3706 return 0; 3707 } 3708 3709 /* Perform checksum validate (init). Note that this is a macro since we only 3710 * want to calculate the pseudo header which is an input function if necessary. 3711 * First we try to validate without any computation (checksum unnecessary) and 3712 * then calculate based on checksum complete calling the function to compute 3713 * pseudo header. 3714 * 3715 * Return values: 3716 * 0: checksum is validated or try to in skb_checksum_complete 3717 * non-zero: value of invalid checksum 3718 */ 3719 #define __skb_checksum_validate(skb, proto, complete, \ 3720 zero_okay, check, compute_pseudo) \ 3721 ({ \ 3722 __sum16 __ret = 0; \ 3723 skb->csum_valid = 0; \ 3724 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3725 __ret = __skb_checksum_validate_complete(skb, \ 3726 complete, compute_pseudo(skb, proto)); \ 3727 __ret; \ 3728 }) 3729 3730 #define skb_checksum_init(skb, proto, compute_pseudo) \ 3731 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3732 3733 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3734 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3735 3736 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 3737 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3738 3739 #define skb_checksum_validate_zero_check(skb, proto, check, \ 3740 compute_pseudo) \ 3741 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3742 3743 #define skb_checksum_simple_validate(skb) \ 3744 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3745 3746 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3747 { 3748 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 3749 } 3750 3751 static inline void __skb_checksum_convert(struct sk_buff *skb, 3752 __sum16 check, __wsum pseudo) 3753 { 3754 skb->csum = ~pseudo; 3755 skb->ip_summed = CHECKSUM_COMPLETE; 3756 } 3757 3758 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3759 do { \ 3760 if (__skb_checksum_convert_check(skb)) \ 3761 __skb_checksum_convert(skb, check, \ 3762 compute_pseudo(skb, proto)); \ 3763 } while (0) 3764 3765 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3766 u16 start, u16 offset) 3767 { 3768 skb->ip_summed = CHECKSUM_PARTIAL; 3769 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3770 skb->csum_offset = offset - start; 3771 } 3772 3773 /* Update skbuf and packet to reflect the remote checksum offload operation. 3774 * When called, ptr indicates the starting point for skb->csum when 3775 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3776 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3777 */ 3778 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3779 int start, int offset, bool nopartial) 3780 { 3781 __wsum delta; 3782 3783 if (!nopartial) { 3784 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3785 return; 3786 } 3787 3788 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3789 __skb_checksum_complete(skb); 3790 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3791 } 3792 3793 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3794 3795 /* Adjust skb->csum since we changed the packet */ 3796 skb->csum = csum_add(skb->csum, delta); 3797 } 3798 3799 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 3800 { 3801 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 3802 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK); 3803 #else 3804 return NULL; 3805 #endif 3806 } 3807 3808 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3809 void nf_conntrack_destroy(struct nf_conntrack *nfct); 3810 static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3811 { 3812 if (nfct && atomic_dec_and_test(&nfct->use)) 3813 nf_conntrack_destroy(nfct); 3814 } 3815 static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3816 { 3817 if (nfct) 3818 atomic_inc(&nfct->use); 3819 } 3820 #endif 3821 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3822 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3823 { 3824 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use)) 3825 kfree(nf_bridge); 3826 } 3827 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3828 { 3829 if (nf_bridge) 3830 refcount_inc(&nf_bridge->use); 3831 } 3832 #endif /* CONFIG_BRIDGE_NETFILTER */ 3833 static inline void nf_reset(struct sk_buff *skb) 3834 { 3835 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3836 nf_conntrack_put(skb_nfct(skb)); 3837 skb->_nfct = 0; 3838 #endif 3839 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3840 nf_bridge_put(skb->nf_bridge); 3841 skb->nf_bridge = NULL; 3842 #endif 3843 } 3844 3845 static inline void nf_reset_trace(struct sk_buff *skb) 3846 { 3847 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3848 skb->nf_trace = 0; 3849 #endif 3850 } 3851 3852 static inline void ipvs_reset(struct sk_buff *skb) 3853 { 3854 #if IS_ENABLED(CONFIG_IP_VS) 3855 skb->ipvs_property = 0; 3856 #endif 3857 } 3858 3859 /* Note: This doesn't put any conntrack and bridge info in dst. */ 3860 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3861 bool copy) 3862 { 3863 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3864 dst->_nfct = src->_nfct; 3865 nf_conntrack_get(skb_nfct(src)); 3866 #endif 3867 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3868 dst->nf_bridge = src->nf_bridge; 3869 nf_bridge_get(src->nf_bridge); 3870 #endif 3871 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3872 if (copy) 3873 dst->nf_trace = src->nf_trace; 3874 #endif 3875 } 3876 3877 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3878 { 3879 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3880 nf_conntrack_put(skb_nfct(dst)); 3881 #endif 3882 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3883 nf_bridge_put(dst->nf_bridge); 3884 #endif 3885 __nf_copy(dst, src, true); 3886 } 3887 3888 #ifdef CONFIG_NETWORK_SECMARK 3889 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3890 { 3891 to->secmark = from->secmark; 3892 } 3893 3894 static inline void skb_init_secmark(struct sk_buff *skb) 3895 { 3896 skb->secmark = 0; 3897 } 3898 #else 3899 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3900 { } 3901 3902 static inline void skb_init_secmark(struct sk_buff *skb) 3903 { } 3904 #endif 3905 3906 static inline bool skb_irq_freeable(const struct sk_buff *skb) 3907 { 3908 return !skb->destructor && 3909 #if IS_ENABLED(CONFIG_XFRM) 3910 !skb->sp && 3911 #endif 3912 !skb_nfct(skb) && 3913 !skb->_skb_refdst && 3914 !skb_has_frag_list(skb); 3915 } 3916 3917 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3918 { 3919 skb->queue_mapping = queue_mapping; 3920 } 3921 3922 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3923 { 3924 return skb->queue_mapping; 3925 } 3926 3927 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3928 { 3929 to->queue_mapping = from->queue_mapping; 3930 } 3931 3932 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3933 { 3934 skb->queue_mapping = rx_queue + 1; 3935 } 3936 3937 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3938 { 3939 return skb->queue_mapping - 1; 3940 } 3941 3942 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3943 { 3944 return skb->queue_mapping != 0; 3945 } 3946 3947 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 3948 { 3949 skb->dst_pending_confirm = val; 3950 } 3951 3952 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 3953 { 3954 return skb->dst_pending_confirm != 0; 3955 } 3956 3957 static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3958 { 3959 #ifdef CONFIG_XFRM 3960 return skb->sp; 3961 #else 3962 return NULL; 3963 #endif 3964 } 3965 3966 /* Keeps track of mac header offset relative to skb->head. 3967 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3968 * For non-tunnel skb it points to skb_mac_header() and for 3969 * tunnel skb it points to outer mac header. 3970 * Keeps track of level of encapsulation of network headers. 3971 */ 3972 struct skb_gso_cb { 3973 union { 3974 int mac_offset; 3975 int data_offset; 3976 }; 3977 int encap_level; 3978 __wsum csum; 3979 __u16 csum_start; 3980 }; 3981 #define SKB_SGO_CB_OFFSET 32 3982 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 3983 3984 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3985 { 3986 return (skb_mac_header(inner_skb) - inner_skb->head) - 3987 SKB_GSO_CB(inner_skb)->mac_offset; 3988 } 3989 3990 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3991 { 3992 int new_headroom, headroom; 3993 int ret; 3994 3995 headroom = skb_headroom(skb); 3996 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3997 if (ret) 3998 return ret; 3999 4000 new_headroom = skb_headroom(skb); 4001 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 4002 return 0; 4003 } 4004 4005 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 4006 { 4007 /* Do not update partial checksums if remote checksum is enabled. */ 4008 if (skb->remcsum_offload) 4009 return; 4010 4011 SKB_GSO_CB(skb)->csum = res; 4012 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 4013 } 4014 4015 /* Compute the checksum for a gso segment. First compute the checksum value 4016 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 4017 * then add in skb->csum (checksum from csum_start to end of packet). 4018 * skb->csum and csum_start are then updated to reflect the checksum of the 4019 * resultant packet starting from the transport header-- the resultant checksum 4020 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 4021 * header. 4022 */ 4023 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 4024 { 4025 unsigned char *csum_start = skb_transport_header(skb); 4026 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 4027 __wsum partial = SKB_GSO_CB(skb)->csum; 4028 4029 SKB_GSO_CB(skb)->csum = res; 4030 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 4031 4032 return csum_fold(csum_partial(csum_start, plen, partial)); 4033 } 4034 4035 static inline bool skb_is_gso(const struct sk_buff *skb) 4036 { 4037 return skb_shinfo(skb)->gso_size; 4038 } 4039 4040 /* Note: Should be called only if skb_is_gso(skb) is true */ 4041 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4042 { 4043 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4044 } 4045 4046 /* Note: Should be called only if skb_is_gso(skb) is true */ 4047 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 4048 { 4049 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 4050 } 4051 4052 static inline void skb_gso_reset(struct sk_buff *skb) 4053 { 4054 skb_shinfo(skb)->gso_size = 0; 4055 skb_shinfo(skb)->gso_segs = 0; 4056 skb_shinfo(skb)->gso_type = 0; 4057 } 4058 4059 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 4060 u16 increment) 4061 { 4062 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4063 return; 4064 shinfo->gso_size += increment; 4065 } 4066 4067 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 4068 u16 decrement) 4069 { 4070 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4071 return; 4072 shinfo->gso_size -= decrement; 4073 } 4074 4075 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 4076 4077 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 4078 { 4079 /* LRO sets gso_size but not gso_type, whereas if GSO is really 4080 * wanted then gso_type will be set. */ 4081 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4082 4083 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 4084 unlikely(shinfo->gso_type == 0)) { 4085 __skb_warn_lro_forwarding(skb); 4086 return true; 4087 } 4088 return false; 4089 } 4090 4091 static inline void skb_forward_csum(struct sk_buff *skb) 4092 { 4093 /* Unfortunately we don't support this one. Any brave souls? */ 4094 if (skb->ip_summed == CHECKSUM_COMPLETE) 4095 skb->ip_summed = CHECKSUM_NONE; 4096 } 4097 4098 /** 4099 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 4100 * @skb: skb to check 4101 * 4102 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 4103 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 4104 * use this helper, to document places where we make this assertion. 4105 */ 4106 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 4107 { 4108 #ifdef DEBUG 4109 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 4110 #endif 4111 } 4112 4113 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 4114 4115 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 4116 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4117 unsigned int transport_len, 4118 __sum16(*skb_chkf)(struct sk_buff *skb)); 4119 4120 /** 4121 * skb_head_is_locked - Determine if the skb->head is locked down 4122 * @skb: skb to check 4123 * 4124 * The head on skbs build around a head frag can be removed if they are 4125 * not cloned. This function returns true if the skb head is locked down 4126 * due to either being allocated via kmalloc, or by being a clone with 4127 * multiple references to the head. 4128 */ 4129 static inline bool skb_head_is_locked(const struct sk_buff *skb) 4130 { 4131 return !skb->head_frag || skb_cloned(skb); 4132 } 4133 4134 /* Local Checksum Offload. 4135 * Compute outer checksum based on the assumption that the 4136 * inner checksum will be offloaded later. 4137 * See Documentation/networking/checksum-offloads.txt for 4138 * explanation of how this works. 4139 * Fill in outer checksum adjustment (e.g. with sum of outer 4140 * pseudo-header) before calling. 4141 * Also ensure that inner checksum is in linear data area. 4142 */ 4143 static inline __wsum lco_csum(struct sk_buff *skb) 4144 { 4145 unsigned char *csum_start = skb_checksum_start(skb); 4146 unsigned char *l4_hdr = skb_transport_header(skb); 4147 __wsum partial; 4148 4149 /* Start with complement of inner checksum adjustment */ 4150 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 4151 skb->csum_offset)); 4152 4153 /* Add in checksum of our headers (incl. outer checksum 4154 * adjustment filled in by caller) and return result. 4155 */ 4156 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 4157 } 4158 4159 #endif /* __KERNEL__ */ 4160 #endif /* _LINUX_SKBUFF_H */ 4161