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 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size); 470 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size, 471 struct ubuf_info *uarg); 472 473 static inline void sock_zerocopy_get(struct ubuf_info *uarg) 474 { 475 refcount_inc(&uarg->refcnt); 476 } 477 478 void sock_zerocopy_put(struct ubuf_info *uarg); 479 void sock_zerocopy_put_abort(struct ubuf_info *uarg); 480 481 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success); 482 483 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 484 struct msghdr *msg, int len, 485 struct ubuf_info *uarg); 486 487 /* This data is invariant across clones and lives at 488 * the end of the header data, ie. at skb->end. 489 */ 490 struct skb_shared_info { 491 __u8 __unused; 492 __u8 meta_len; 493 __u8 nr_frags; 494 __u8 tx_flags; 495 unsigned short gso_size; 496 /* Warning: this field is not always filled in (UFO)! */ 497 unsigned short gso_segs; 498 struct sk_buff *frag_list; 499 struct skb_shared_hwtstamps hwtstamps; 500 unsigned int gso_type; 501 u32 tskey; 502 503 /* 504 * Warning : all fields before dataref are cleared in __alloc_skb() 505 */ 506 atomic_t dataref; 507 508 /* Intermediate layers must ensure that destructor_arg 509 * remains valid until skb destructor */ 510 void * destructor_arg; 511 512 /* must be last field, see pskb_expand_head() */ 513 skb_frag_t frags[MAX_SKB_FRAGS]; 514 }; 515 516 /* We divide dataref into two halves. The higher 16 bits hold references 517 * to the payload part of skb->data. The lower 16 bits hold references to 518 * the entire skb->data. A clone of a headerless skb holds the length of 519 * the header in skb->hdr_len. 520 * 521 * All users must obey the rule that the skb->data reference count must be 522 * greater than or equal to the payload reference count. 523 * 524 * Holding a reference to the payload part means that the user does not 525 * care about modifications to the header part of skb->data. 526 */ 527 #define SKB_DATAREF_SHIFT 16 528 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 529 530 531 enum { 532 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 533 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 534 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 535 }; 536 537 enum { 538 SKB_GSO_TCPV4 = 1 << 0, 539 540 /* This indicates the skb is from an untrusted source. */ 541 SKB_GSO_DODGY = 1 << 1, 542 543 /* This indicates the tcp segment has CWR set. */ 544 SKB_GSO_TCP_ECN = 1 << 2, 545 546 SKB_GSO_TCP_FIXEDID = 1 << 3, 547 548 SKB_GSO_TCPV6 = 1 << 4, 549 550 SKB_GSO_FCOE = 1 << 5, 551 552 SKB_GSO_GRE = 1 << 6, 553 554 SKB_GSO_GRE_CSUM = 1 << 7, 555 556 SKB_GSO_IPXIP4 = 1 << 8, 557 558 SKB_GSO_IPXIP6 = 1 << 9, 559 560 SKB_GSO_UDP_TUNNEL = 1 << 10, 561 562 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 563 564 SKB_GSO_PARTIAL = 1 << 12, 565 566 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 567 568 SKB_GSO_SCTP = 1 << 14, 569 570 SKB_GSO_ESP = 1 << 15, 571 572 SKB_GSO_UDP = 1 << 16, 573 }; 574 575 #if BITS_PER_LONG > 32 576 #define NET_SKBUFF_DATA_USES_OFFSET 1 577 #endif 578 579 #ifdef NET_SKBUFF_DATA_USES_OFFSET 580 typedef unsigned int sk_buff_data_t; 581 #else 582 typedef unsigned char *sk_buff_data_t; 583 #endif 584 585 /** 586 * struct sk_buff - socket buffer 587 * @next: Next buffer in list 588 * @prev: Previous buffer in list 589 * @tstamp: Time we arrived/left 590 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 591 * @sk: Socket we are owned by 592 * @dev: Device we arrived on/are leaving by 593 * @cb: Control buffer. Free for use by every layer. Put private vars here 594 * @_skb_refdst: destination entry (with norefcount bit) 595 * @sp: the security path, used for xfrm 596 * @len: Length of actual data 597 * @data_len: Data length 598 * @mac_len: Length of link layer header 599 * @hdr_len: writable header length of cloned skb 600 * @csum: Checksum (must include start/offset pair) 601 * @csum_start: Offset from skb->head where checksumming should start 602 * @csum_offset: Offset from csum_start where checksum should be stored 603 * @priority: Packet queueing priority 604 * @ignore_df: allow local fragmentation 605 * @cloned: Head may be cloned (check refcnt to be sure) 606 * @ip_summed: Driver fed us an IP checksum 607 * @nohdr: Payload reference only, must not modify header 608 * @pkt_type: Packet class 609 * @fclone: skbuff clone status 610 * @ipvs_property: skbuff is owned by ipvs 611 * @tc_skip_classify: do not classify packet. set by IFB device 612 * @tc_at_ingress: used within tc_classify to distinguish in/egress 613 * @tc_redirected: packet was redirected by a tc action 614 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect 615 * @peeked: this packet has been seen already, so stats have been 616 * done for it, don't do them again 617 * @nf_trace: netfilter packet trace flag 618 * @protocol: Packet protocol from driver 619 * @destructor: Destruct function 620 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 621 * @_nfct: Associated connection, if any (with nfctinfo bits) 622 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 623 * @skb_iif: ifindex of device we arrived on 624 * @tc_index: Traffic control index 625 * @hash: the packet hash 626 * @queue_mapping: Queue mapping for multiqueue devices 627 * @xmit_more: More SKBs are pending for this queue 628 * @ndisc_nodetype: router type (from link layer) 629 * @ooo_okay: allow the mapping of a socket to a queue to be changed 630 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 631 * ports. 632 * @sw_hash: indicates hash was computed in software stack 633 * @wifi_acked_valid: wifi_acked was set 634 * @wifi_acked: whether frame was acked on wifi or not 635 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 636 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 637 * @dst_pending_confirm: need to confirm neighbour 638 * @napi_id: id of the NAPI struct this skb came from 639 * @secmark: security marking 640 * @mark: Generic packet mark 641 * @vlan_proto: vlan encapsulation protocol 642 * @vlan_tci: vlan tag control information 643 * @inner_protocol: Protocol (encapsulation) 644 * @inner_transport_header: Inner transport layer header (encapsulation) 645 * @inner_network_header: Network layer header (encapsulation) 646 * @inner_mac_header: Link layer header (encapsulation) 647 * @transport_header: Transport layer header 648 * @network_header: Network layer header 649 * @mac_header: Link layer header 650 * @tail: Tail pointer 651 * @end: End pointer 652 * @head: Head of buffer 653 * @data: Data head pointer 654 * @truesize: Buffer size 655 * @users: User count - see {datagram,tcp}.c 656 */ 657 658 struct sk_buff { 659 union { 660 struct { 661 /* These two members must be first. */ 662 struct sk_buff *next; 663 struct sk_buff *prev; 664 665 union { 666 struct net_device *dev; 667 /* Some protocols might use this space to store information, 668 * while device pointer would be NULL. 669 * UDP receive path is one user. 670 */ 671 unsigned long dev_scratch; 672 }; 673 }; 674 struct rb_node rbnode; /* used in netem & tcp stack */ 675 }; 676 struct sock *sk; 677 678 union { 679 ktime_t tstamp; 680 u64 skb_mstamp; 681 }; 682 /* 683 * This is the control buffer. It is free to use for every 684 * layer. Please put your private variables there. If you 685 * want to keep them across layers you have to do a skb_clone() 686 * first. This is owned by whoever has the skb queued ATM. 687 */ 688 char cb[48] __aligned(8); 689 690 union { 691 struct { 692 unsigned long _skb_refdst; 693 void (*destructor)(struct sk_buff *skb); 694 }; 695 struct list_head tcp_tsorted_anchor; 696 }; 697 698 #ifdef CONFIG_XFRM 699 struct sec_path *sp; 700 #endif 701 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 702 unsigned long _nfct; 703 #endif 704 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 705 struct nf_bridge_info *nf_bridge; 706 #endif 707 unsigned int len, 708 data_len; 709 __u16 mac_len, 710 hdr_len; 711 712 /* Following fields are _not_ copied in __copy_skb_header() 713 * Note that queue_mapping is here mostly to fill a hole. 714 */ 715 __u16 queue_mapping; 716 717 /* if you move cloned around you also must adapt those constants */ 718 #ifdef __BIG_ENDIAN_BITFIELD 719 #define CLONED_MASK (1 << 7) 720 #else 721 #define CLONED_MASK 1 722 #endif 723 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset) 724 725 __u8 __cloned_offset[0]; 726 __u8 cloned:1, 727 nohdr:1, 728 fclone:2, 729 peeked:1, 730 head_frag:1, 731 xmit_more:1, 732 __unused:1; /* one bit hole */ 733 734 /* fields enclosed in headers_start/headers_end are copied 735 * using a single memcpy() in __copy_skb_header() 736 */ 737 /* private: */ 738 __u32 headers_start[0]; 739 /* public: */ 740 741 /* if you move pkt_type around you also must adapt those constants */ 742 #ifdef __BIG_ENDIAN_BITFIELD 743 #define PKT_TYPE_MAX (7 << 5) 744 #else 745 #define PKT_TYPE_MAX 7 746 #endif 747 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 748 749 __u8 __pkt_type_offset[0]; 750 __u8 pkt_type:3; 751 __u8 pfmemalloc:1; 752 __u8 ignore_df:1; 753 754 __u8 nf_trace:1; 755 __u8 ip_summed:2; 756 __u8 ooo_okay:1; 757 __u8 l4_hash:1; 758 __u8 sw_hash:1; 759 __u8 wifi_acked_valid:1; 760 __u8 wifi_acked:1; 761 762 __u8 no_fcs:1; 763 /* Indicates the inner headers are valid in the skbuff. */ 764 __u8 encapsulation:1; 765 __u8 encap_hdr_csum:1; 766 __u8 csum_valid:1; 767 __u8 csum_complete_sw:1; 768 __u8 csum_level:2; 769 __u8 csum_not_inet:1; 770 771 __u8 dst_pending_confirm:1; 772 #ifdef CONFIG_IPV6_NDISC_NODETYPE 773 __u8 ndisc_nodetype:2; 774 #endif 775 __u8 ipvs_property:1; 776 __u8 inner_protocol_type:1; 777 __u8 remcsum_offload:1; 778 #ifdef CONFIG_NET_SWITCHDEV 779 __u8 offload_fwd_mark:1; 780 __u8 offload_mr_fwd_mark:1; 781 #endif 782 #ifdef CONFIG_NET_CLS_ACT 783 __u8 tc_skip_classify:1; 784 __u8 tc_at_ingress:1; 785 __u8 tc_redirected:1; 786 __u8 tc_from_ingress:1; 787 #endif 788 789 #ifdef CONFIG_NET_SCHED 790 __u16 tc_index; /* traffic control index */ 791 #endif 792 793 union { 794 __wsum csum; 795 struct { 796 __u16 csum_start; 797 __u16 csum_offset; 798 }; 799 }; 800 __u32 priority; 801 int skb_iif; 802 __u32 hash; 803 __be16 vlan_proto; 804 __u16 vlan_tci; 805 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 806 union { 807 unsigned int napi_id; 808 unsigned int sender_cpu; 809 }; 810 #endif 811 #ifdef CONFIG_NETWORK_SECMARK 812 __u32 secmark; 813 #endif 814 815 union { 816 __u32 mark; 817 __u32 reserved_tailroom; 818 }; 819 820 union { 821 __be16 inner_protocol; 822 __u8 inner_ipproto; 823 }; 824 825 __u16 inner_transport_header; 826 __u16 inner_network_header; 827 __u16 inner_mac_header; 828 829 __be16 protocol; 830 __u16 transport_header; 831 __u16 network_header; 832 __u16 mac_header; 833 834 /* private: */ 835 __u32 headers_end[0]; 836 /* public: */ 837 838 /* These elements must be at the end, see alloc_skb() for details. */ 839 sk_buff_data_t tail; 840 sk_buff_data_t end; 841 unsigned char *head, 842 *data; 843 unsigned int truesize; 844 refcount_t users; 845 }; 846 847 #ifdef __KERNEL__ 848 /* 849 * Handling routines are only of interest to the kernel 850 */ 851 #include <linux/slab.h> 852 853 854 #define SKB_ALLOC_FCLONE 0x01 855 #define SKB_ALLOC_RX 0x02 856 #define SKB_ALLOC_NAPI 0x04 857 858 /* Returns true if the skb was allocated from PFMEMALLOC reserves */ 859 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 860 { 861 return unlikely(skb->pfmemalloc); 862 } 863 864 /* 865 * skb might have a dst pointer attached, refcounted or not. 866 * _skb_refdst low order bit is set if refcount was _not_ taken 867 */ 868 #define SKB_DST_NOREF 1UL 869 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 870 871 #define SKB_NFCT_PTRMASK ~(7UL) 872 /** 873 * skb_dst - returns skb dst_entry 874 * @skb: buffer 875 * 876 * Returns skb dst_entry, regardless of reference taken or not. 877 */ 878 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 879 { 880 /* If refdst was not refcounted, check we still are in a 881 * rcu_read_lock section 882 */ 883 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 884 !rcu_read_lock_held() && 885 !rcu_read_lock_bh_held()); 886 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 887 } 888 889 /** 890 * skb_dst_set - sets skb dst 891 * @skb: buffer 892 * @dst: dst entry 893 * 894 * Sets skb dst, assuming a reference was taken on dst and should 895 * be released by skb_dst_drop() 896 */ 897 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 898 { 899 skb->_skb_refdst = (unsigned long)dst; 900 } 901 902 /** 903 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 904 * @skb: buffer 905 * @dst: dst entry 906 * 907 * Sets skb dst, assuming a reference was not taken on dst. 908 * If dst entry is cached, we do not take reference and dst_release 909 * will be avoided by refdst_drop. If dst entry is not cached, we take 910 * reference, so that last dst_release can destroy the dst immediately. 911 */ 912 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 913 { 914 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 915 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 916 } 917 918 /** 919 * skb_dst_is_noref - Test if skb dst isn't refcounted 920 * @skb: buffer 921 */ 922 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 923 { 924 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 925 } 926 927 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 928 { 929 return (struct rtable *)skb_dst(skb); 930 } 931 932 /* For mangling skb->pkt_type from user space side from applications 933 * such as nft, tc, etc, we only allow a conservative subset of 934 * possible pkt_types to be set. 935 */ 936 static inline bool skb_pkt_type_ok(u32 ptype) 937 { 938 return ptype <= PACKET_OTHERHOST; 939 } 940 941 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 942 { 943 #ifdef CONFIG_NET_RX_BUSY_POLL 944 return skb->napi_id; 945 #else 946 return 0; 947 #endif 948 } 949 950 /* decrement the reference count and return true if we can free the skb */ 951 static inline bool skb_unref(struct sk_buff *skb) 952 { 953 if (unlikely(!skb)) 954 return false; 955 if (likely(refcount_read(&skb->users) == 1)) 956 smp_rmb(); 957 else if (likely(!refcount_dec_and_test(&skb->users))) 958 return false; 959 960 return true; 961 } 962 963 void skb_release_head_state(struct sk_buff *skb); 964 void kfree_skb(struct sk_buff *skb); 965 void kfree_skb_list(struct sk_buff *segs); 966 void skb_tx_error(struct sk_buff *skb); 967 void consume_skb(struct sk_buff *skb); 968 void __consume_stateless_skb(struct sk_buff *skb); 969 void __kfree_skb(struct sk_buff *skb); 970 extern struct kmem_cache *skbuff_head_cache; 971 972 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 973 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 974 bool *fragstolen, int *delta_truesize); 975 976 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 977 int node); 978 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 979 struct sk_buff *build_skb(void *data, unsigned int frag_size); 980 static inline struct sk_buff *alloc_skb(unsigned int size, 981 gfp_t priority) 982 { 983 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 984 } 985 986 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 987 unsigned long data_len, 988 int max_page_order, 989 int *errcode, 990 gfp_t gfp_mask); 991 992 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 993 struct sk_buff_fclones { 994 struct sk_buff skb1; 995 996 struct sk_buff skb2; 997 998 refcount_t fclone_ref; 999 }; 1000 1001 /** 1002 * skb_fclone_busy - check if fclone is busy 1003 * @sk: socket 1004 * @skb: buffer 1005 * 1006 * Returns true if skb is a fast clone, and its clone is not freed. 1007 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1008 * so we also check that this didnt happen. 1009 */ 1010 static inline bool skb_fclone_busy(const struct sock *sk, 1011 const struct sk_buff *skb) 1012 { 1013 const struct sk_buff_fclones *fclones; 1014 1015 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1016 1017 return skb->fclone == SKB_FCLONE_ORIG && 1018 refcount_read(&fclones->fclone_ref) > 1 && 1019 fclones->skb2.sk == sk; 1020 } 1021 1022 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1023 gfp_t priority) 1024 { 1025 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1026 } 1027 1028 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1029 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1030 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1031 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1032 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1033 gfp_t gfp_mask, bool fclone); 1034 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1035 gfp_t gfp_mask) 1036 { 1037 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1038 } 1039 1040 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1041 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1042 unsigned int headroom); 1043 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1044 int newtailroom, gfp_t priority); 1045 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1046 int offset, int len); 1047 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1048 int offset, int len); 1049 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1050 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1051 1052 /** 1053 * skb_pad - zero pad the tail of an skb 1054 * @skb: buffer to pad 1055 * @pad: space to pad 1056 * 1057 * Ensure that a buffer is followed by a padding area that is zero 1058 * filled. Used by network drivers which may DMA or transfer data 1059 * beyond the buffer end onto the wire. 1060 * 1061 * May return error in out of memory cases. The skb is freed on error. 1062 */ 1063 static inline int skb_pad(struct sk_buff *skb, int pad) 1064 { 1065 return __skb_pad(skb, pad, true); 1066 } 1067 #define dev_kfree_skb(a) consume_skb(a) 1068 1069 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 1070 int getfrag(void *from, char *to, int offset, 1071 int len, int odd, struct sk_buff *skb), 1072 void *from, int length); 1073 1074 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1075 int offset, size_t size); 1076 1077 struct skb_seq_state { 1078 __u32 lower_offset; 1079 __u32 upper_offset; 1080 __u32 frag_idx; 1081 __u32 stepped_offset; 1082 struct sk_buff *root_skb; 1083 struct sk_buff *cur_skb; 1084 __u8 *frag_data; 1085 }; 1086 1087 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1088 unsigned int to, struct skb_seq_state *st); 1089 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1090 struct skb_seq_state *st); 1091 void skb_abort_seq_read(struct skb_seq_state *st); 1092 1093 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1094 unsigned int to, struct ts_config *config); 1095 1096 /* 1097 * Packet hash types specify the type of hash in skb_set_hash. 1098 * 1099 * Hash types refer to the protocol layer addresses which are used to 1100 * construct a packet's hash. The hashes are used to differentiate or identify 1101 * flows of the protocol layer for the hash type. Hash types are either 1102 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1103 * 1104 * Properties of hashes: 1105 * 1106 * 1) Two packets in different flows have different hash values 1107 * 2) Two packets in the same flow should have the same hash value 1108 * 1109 * A hash at a higher layer is considered to be more specific. A driver should 1110 * set the most specific hash possible. 1111 * 1112 * A driver cannot indicate a more specific hash than the layer at which a hash 1113 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1114 * 1115 * A driver may indicate a hash level which is less specific than the 1116 * actual layer the hash was computed on. For instance, a hash computed 1117 * at L4 may be considered an L3 hash. This should only be done if the 1118 * driver can't unambiguously determine that the HW computed the hash at 1119 * the higher layer. Note that the "should" in the second property above 1120 * permits this. 1121 */ 1122 enum pkt_hash_types { 1123 PKT_HASH_TYPE_NONE, /* Undefined type */ 1124 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1125 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1126 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1127 }; 1128 1129 static inline void skb_clear_hash(struct sk_buff *skb) 1130 { 1131 skb->hash = 0; 1132 skb->sw_hash = 0; 1133 skb->l4_hash = 0; 1134 } 1135 1136 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1137 { 1138 if (!skb->l4_hash) 1139 skb_clear_hash(skb); 1140 } 1141 1142 static inline void 1143 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1144 { 1145 skb->l4_hash = is_l4; 1146 skb->sw_hash = is_sw; 1147 skb->hash = hash; 1148 } 1149 1150 static inline void 1151 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1152 { 1153 /* Used by drivers to set hash from HW */ 1154 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1155 } 1156 1157 static inline void 1158 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1159 { 1160 __skb_set_hash(skb, hash, true, is_l4); 1161 } 1162 1163 void __skb_get_hash(struct sk_buff *skb); 1164 u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1165 u32 skb_get_poff(const struct sk_buff *skb); 1166 u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1167 const struct flow_keys *keys, int hlen); 1168 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1169 void *data, int hlen_proto); 1170 1171 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1172 int thoff, u8 ip_proto) 1173 { 1174 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1175 } 1176 1177 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1178 const struct flow_dissector_key *key, 1179 unsigned int key_count); 1180 1181 bool __skb_flow_dissect(const struct sk_buff *skb, 1182 struct flow_dissector *flow_dissector, 1183 void *target_container, 1184 void *data, __be16 proto, int nhoff, int hlen, 1185 unsigned int flags); 1186 1187 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1188 struct flow_dissector *flow_dissector, 1189 void *target_container, unsigned int flags) 1190 { 1191 return __skb_flow_dissect(skb, flow_dissector, target_container, 1192 NULL, 0, 0, 0, flags); 1193 } 1194 1195 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1196 struct flow_keys *flow, 1197 unsigned int flags) 1198 { 1199 memset(flow, 0, sizeof(*flow)); 1200 return __skb_flow_dissect(skb, &flow_keys_dissector, flow, 1201 NULL, 0, 0, 0, flags); 1202 } 1203 1204 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow, 1205 void *data, __be16 proto, 1206 int nhoff, int hlen, 1207 unsigned int flags) 1208 { 1209 memset(flow, 0, sizeof(*flow)); 1210 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow, 1211 data, proto, nhoff, hlen, flags); 1212 } 1213 1214 static inline __u32 skb_get_hash(struct sk_buff *skb) 1215 { 1216 if (!skb->l4_hash && !skb->sw_hash) 1217 __skb_get_hash(skb); 1218 1219 return skb->hash; 1220 } 1221 1222 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1223 { 1224 if (!skb->l4_hash && !skb->sw_hash) { 1225 struct flow_keys keys; 1226 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1227 1228 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1229 } 1230 1231 return skb->hash; 1232 } 1233 1234 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1235 1236 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1237 { 1238 return skb->hash; 1239 } 1240 1241 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1242 { 1243 to->hash = from->hash; 1244 to->sw_hash = from->sw_hash; 1245 to->l4_hash = from->l4_hash; 1246 }; 1247 1248 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1249 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1250 { 1251 return skb->head + skb->end; 1252 } 1253 1254 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1255 { 1256 return skb->end; 1257 } 1258 #else 1259 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1260 { 1261 return skb->end; 1262 } 1263 1264 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1265 { 1266 return skb->end - skb->head; 1267 } 1268 #endif 1269 1270 /* Internal */ 1271 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1272 1273 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1274 { 1275 return &skb_shinfo(skb)->hwtstamps; 1276 } 1277 1278 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1279 { 1280 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY; 1281 1282 return is_zcopy ? skb_uarg(skb) : NULL; 1283 } 1284 1285 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg) 1286 { 1287 if (skb && uarg && !skb_zcopy(skb)) { 1288 sock_zerocopy_get(uarg); 1289 skb_shinfo(skb)->destructor_arg = uarg; 1290 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG; 1291 } 1292 } 1293 1294 /* Release a reference on a zerocopy structure */ 1295 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy) 1296 { 1297 struct ubuf_info *uarg = skb_zcopy(skb); 1298 1299 if (uarg) { 1300 if (uarg->callback == sock_zerocopy_callback) { 1301 uarg->zerocopy = uarg->zerocopy && zerocopy; 1302 sock_zerocopy_put(uarg); 1303 } else { 1304 uarg->callback(uarg, zerocopy); 1305 } 1306 1307 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1308 } 1309 } 1310 1311 /* Abort a zerocopy operation and revert zckey on error in send syscall */ 1312 static inline void skb_zcopy_abort(struct sk_buff *skb) 1313 { 1314 struct ubuf_info *uarg = skb_zcopy(skb); 1315 1316 if (uarg) { 1317 sock_zerocopy_put_abort(uarg); 1318 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1319 } 1320 } 1321 1322 /** 1323 * skb_queue_empty - check if a queue is empty 1324 * @list: queue head 1325 * 1326 * Returns true if the queue is empty, false otherwise. 1327 */ 1328 static inline int skb_queue_empty(const struct sk_buff_head *list) 1329 { 1330 return list->next == (const struct sk_buff *) list; 1331 } 1332 1333 /** 1334 * skb_queue_is_last - check if skb is the last entry in the queue 1335 * @list: queue head 1336 * @skb: buffer 1337 * 1338 * Returns true if @skb is the last buffer on the list. 1339 */ 1340 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1341 const struct sk_buff *skb) 1342 { 1343 return skb->next == (const struct sk_buff *) list; 1344 } 1345 1346 /** 1347 * skb_queue_is_first - check if skb is the first entry in the queue 1348 * @list: queue head 1349 * @skb: buffer 1350 * 1351 * Returns true if @skb is the first buffer on the list. 1352 */ 1353 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1354 const struct sk_buff *skb) 1355 { 1356 return skb->prev == (const struct sk_buff *) list; 1357 } 1358 1359 /** 1360 * skb_queue_next - return the next packet in the queue 1361 * @list: queue head 1362 * @skb: current buffer 1363 * 1364 * Return the next packet in @list after @skb. It is only valid to 1365 * call this if skb_queue_is_last() evaluates to false. 1366 */ 1367 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1368 const struct sk_buff *skb) 1369 { 1370 /* This BUG_ON may seem severe, but if we just return then we 1371 * are going to dereference garbage. 1372 */ 1373 BUG_ON(skb_queue_is_last(list, skb)); 1374 return skb->next; 1375 } 1376 1377 /** 1378 * skb_queue_prev - return the prev packet in the queue 1379 * @list: queue head 1380 * @skb: current buffer 1381 * 1382 * Return the prev packet in @list before @skb. It is only valid to 1383 * call this if skb_queue_is_first() evaluates to false. 1384 */ 1385 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1386 const struct sk_buff *skb) 1387 { 1388 /* This BUG_ON may seem severe, but if we just return then we 1389 * are going to dereference garbage. 1390 */ 1391 BUG_ON(skb_queue_is_first(list, skb)); 1392 return skb->prev; 1393 } 1394 1395 /** 1396 * skb_get - reference buffer 1397 * @skb: buffer to reference 1398 * 1399 * Makes another reference to a socket buffer and returns a pointer 1400 * to the buffer. 1401 */ 1402 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1403 { 1404 refcount_inc(&skb->users); 1405 return skb; 1406 } 1407 1408 /* 1409 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1410 */ 1411 1412 /** 1413 * skb_cloned - is the buffer a clone 1414 * @skb: buffer to check 1415 * 1416 * Returns true if the buffer was generated with skb_clone() and is 1417 * one of multiple shared copies of the buffer. Cloned buffers are 1418 * shared data so must not be written to under normal circumstances. 1419 */ 1420 static inline int skb_cloned(const struct sk_buff *skb) 1421 { 1422 return skb->cloned && 1423 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1424 } 1425 1426 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1427 { 1428 might_sleep_if(gfpflags_allow_blocking(pri)); 1429 1430 if (skb_cloned(skb)) 1431 return pskb_expand_head(skb, 0, 0, pri); 1432 1433 return 0; 1434 } 1435 1436 /** 1437 * skb_header_cloned - is the header a clone 1438 * @skb: buffer to check 1439 * 1440 * Returns true if modifying the header part of the buffer requires 1441 * the data to be copied. 1442 */ 1443 static inline int skb_header_cloned(const struct sk_buff *skb) 1444 { 1445 int dataref; 1446 1447 if (!skb->cloned) 1448 return 0; 1449 1450 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1451 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1452 return dataref != 1; 1453 } 1454 1455 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1456 { 1457 might_sleep_if(gfpflags_allow_blocking(pri)); 1458 1459 if (skb_header_cloned(skb)) 1460 return pskb_expand_head(skb, 0, 0, pri); 1461 1462 return 0; 1463 } 1464 1465 /** 1466 * __skb_header_release - release reference to header 1467 * @skb: buffer to operate on 1468 */ 1469 static inline void __skb_header_release(struct sk_buff *skb) 1470 { 1471 skb->nohdr = 1; 1472 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1473 } 1474 1475 1476 /** 1477 * skb_shared - is the buffer shared 1478 * @skb: buffer to check 1479 * 1480 * Returns true if more than one person has a reference to this 1481 * buffer. 1482 */ 1483 static inline int skb_shared(const struct sk_buff *skb) 1484 { 1485 return refcount_read(&skb->users) != 1; 1486 } 1487 1488 /** 1489 * skb_share_check - check if buffer is shared and if so clone it 1490 * @skb: buffer to check 1491 * @pri: priority for memory allocation 1492 * 1493 * If the buffer is shared the buffer is cloned and the old copy 1494 * drops a reference. A new clone with a single reference is returned. 1495 * If the buffer is not shared the original buffer is returned. When 1496 * being called from interrupt status or with spinlocks held pri must 1497 * be GFP_ATOMIC. 1498 * 1499 * NULL is returned on a memory allocation failure. 1500 */ 1501 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1502 { 1503 might_sleep_if(gfpflags_allow_blocking(pri)); 1504 if (skb_shared(skb)) { 1505 struct sk_buff *nskb = skb_clone(skb, pri); 1506 1507 if (likely(nskb)) 1508 consume_skb(skb); 1509 else 1510 kfree_skb(skb); 1511 skb = nskb; 1512 } 1513 return skb; 1514 } 1515 1516 /* 1517 * Copy shared buffers into a new sk_buff. We effectively do COW on 1518 * packets to handle cases where we have a local reader and forward 1519 * and a couple of other messy ones. The normal one is tcpdumping 1520 * a packet thats being forwarded. 1521 */ 1522 1523 /** 1524 * skb_unshare - make a copy of a shared buffer 1525 * @skb: buffer to check 1526 * @pri: priority for memory allocation 1527 * 1528 * If the socket buffer is a clone then this function creates a new 1529 * copy of the data, drops a reference count on the old copy and returns 1530 * the new copy with the reference count at 1. If the buffer is not a clone 1531 * the original buffer is returned. When called with a spinlock held or 1532 * from interrupt state @pri must be %GFP_ATOMIC 1533 * 1534 * %NULL is returned on a memory allocation failure. 1535 */ 1536 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1537 gfp_t pri) 1538 { 1539 might_sleep_if(gfpflags_allow_blocking(pri)); 1540 if (skb_cloned(skb)) { 1541 struct sk_buff *nskb = skb_copy(skb, pri); 1542 1543 /* Free our shared copy */ 1544 if (likely(nskb)) 1545 consume_skb(skb); 1546 else 1547 kfree_skb(skb); 1548 skb = nskb; 1549 } 1550 return skb; 1551 } 1552 1553 /** 1554 * skb_peek - peek at the head of an &sk_buff_head 1555 * @list_: list to peek at 1556 * 1557 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1558 * be careful with this one. A peek leaves the buffer on the 1559 * list and someone else may run off with it. You must hold 1560 * the appropriate locks or have a private queue to do this. 1561 * 1562 * Returns %NULL for an empty list or a pointer to the head element. 1563 * The reference count is not incremented and the reference is therefore 1564 * volatile. Use with caution. 1565 */ 1566 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1567 { 1568 struct sk_buff *skb = list_->next; 1569 1570 if (skb == (struct sk_buff *)list_) 1571 skb = NULL; 1572 return skb; 1573 } 1574 1575 /** 1576 * skb_peek_next - peek skb following the given one from a queue 1577 * @skb: skb to start from 1578 * @list_: list to peek at 1579 * 1580 * Returns %NULL when the end of the list is met or a pointer to the 1581 * next element. The reference count is not incremented and the 1582 * reference is therefore volatile. Use with caution. 1583 */ 1584 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1585 const struct sk_buff_head *list_) 1586 { 1587 struct sk_buff *next = skb->next; 1588 1589 if (next == (struct sk_buff *)list_) 1590 next = NULL; 1591 return next; 1592 } 1593 1594 /** 1595 * skb_peek_tail - peek at the tail of an &sk_buff_head 1596 * @list_: list to peek at 1597 * 1598 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1599 * be careful with this one. A peek leaves the buffer on the 1600 * list and someone else may run off with it. You must hold 1601 * the appropriate locks or have a private queue to do this. 1602 * 1603 * Returns %NULL for an empty list or a pointer to the tail element. 1604 * The reference count is not incremented and the reference is therefore 1605 * volatile. Use with caution. 1606 */ 1607 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1608 { 1609 struct sk_buff *skb = list_->prev; 1610 1611 if (skb == (struct sk_buff *)list_) 1612 skb = NULL; 1613 return skb; 1614 1615 } 1616 1617 /** 1618 * skb_queue_len - get queue length 1619 * @list_: list to measure 1620 * 1621 * Return the length of an &sk_buff queue. 1622 */ 1623 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1624 { 1625 return list_->qlen; 1626 } 1627 1628 /** 1629 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1630 * @list: queue to initialize 1631 * 1632 * This initializes only the list and queue length aspects of 1633 * an sk_buff_head object. This allows to initialize the list 1634 * aspects of an sk_buff_head without reinitializing things like 1635 * the spinlock. It can also be used for on-stack sk_buff_head 1636 * objects where the spinlock is known to not be used. 1637 */ 1638 static inline void __skb_queue_head_init(struct sk_buff_head *list) 1639 { 1640 list->prev = list->next = (struct sk_buff *)list; 1641 list->qlen = 0; 1642 } 1643 1644 /* 1645 * This function creates a split out lock class for each invocation; 1646 * this is needed for now since a whole lot of users of the skb-queue 1647 * infrastructure in drivers have different locking usage (in hardirq) 1648 * than the networking core (in softirq only). In the long run either the 1649 * network layer or drivers should need annotation to consolidate the 1650 * main types of usage into 3 classes. 1651 */ 1652 static inline void skb_queue_head_init(struct sk_buff_head *list) 1653 { 1654 spin_lock_init(&list->lock); 1655 __skb_queue_head_init(list); 1656 } 1657 1658 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1659 struct lock_class_key *class) 1660 { 1661 skb_queue_head_init(list); 1662 lockdep_set_class(&list->lock, class); 1663 } 1664 1665 /* 1666 * Insert an sk_buff on a list. 1667 * 1668 * The "__skb_xxxx()" functions are the non-atomic ones that 1669 * can only be called with interrupts disabled. 1670 */ 1671 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1672 struct sk_buff_head *list); 1673 static inline void __skb_insert(struct sk_buff *newsk, 1674 struct sk_buff *prev, struct sk_buff *next, 1675 struct sk_buff_head *list) 1676 { 1677 newsk->next = next; 1678 newsk->prev = prev; 1679 next->prev = prev->next = newsk; 1680 list->qlen++; 1681 } 1682 1683 static inline void __skb_queue_splice(const struct sk_buff_head *list, 1684 struct sk_buff *prev, 1685 struct sk_buff *next) 1686 { 1687 struct sk_buff *first = list->next; 1688 struct sk_buff *last = list->prev; 1689 1690 first->prev = prev; 1691 prev->next = first; 1692 1693 last->next = next; 1694 next->prev = last; 1695 } 1696 1697 /** 1698 * skb_queue_splice - join two skb lists, this is designed for stacks 1699 * @list: the new list to add 1700 * @head: the place to add it in the first list 1701 */ 1702 static inline void skb_queue_splice(const struct sk_buff_head *list, 1703 struct sk_buff_head *head) 1704 { 1705 if (!skb_queue_empty(list)) { 1706 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1707 head->qlen += list->qlen; 1708 } 1709 } 1710 1711 /** 1712 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1713 * @list: the new list to add 1714 * @head: the place to add it in the first list 1715 * 1716 * The list at @list is reinitialised 1717 */ 1718 static inline void skb_queue_splice_init(struct sk_buff_head *list, 1719 struct sk_buff_head *head) 1720 { 1721 if (!skb_queue_empty(list)) { 1722 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1723 head->qlen += list->qlen; 1724 __skb_queue_head_init(list); 1725 } 1726 } 1727 1728 /** 1729 * skb_queue_splice_tail - join two skb lists, each list being a queue 1730 * @list: the new list to add 1731 * @head: the place to add it in the first list 1732 */ 1733 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1734 struct sk_buff_head *head) 1735 { 1736 if (!skb_queue_empty(list)) { 1737 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1738 head->qlen += list->qlen; 1739 } 1740 } 1741 1742 /** 1743 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1744 * @list: the new list to add 1745 * @head: the place to add it in the first list 1746 * 1747 * Each of the lists is a queue. 1748 * The list at @list is reinitialised 1749 */ 1750 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1751 struct sk_buff_head *head) 1752 { 1753 if (!skb_queue_empty(list)) { 1754 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1755 head->qlen += list->qlen; 1756 __skb_queue_head_init(list); 1757 } 1758 } 1759 1760 /** 1761 * __skb_queue_after - queue a buffer at the list head 1762 * @list: list to use 1763 * @prev: place after this buffer 1764 * @newsk: buffer to queue 1765 * 1766 * Queue a buffer int the middle of a list. This function takes no locks 1767 * and you must therefore hold required locks before calling it. 1768 * 1769 * A buffer cannot be placed on two lists at the same time. 1770 */ 1771 static inline void __skb_queue_after(struct sk_buff_head *list, 1772 struct sk_buff *prev, 1773 struct sk_buff *newsk) 1774 { 1775 __skb_insert(newsk, prev, prev->next, list); 1776 } 1777 1778 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1779 struct sk_buff_head *list); 1780 1781 static inline void __skb_queue_before(struct sk_buff_head *list, 1782 struct sk_buff *next, 1783 struct sk_buff *newsk) 1784 { 1785 __skb_insert(newsk, next->prev, next, list); 1786 } 1787 1788 /** 1789 * __skb_queue_head - queue a buffer at the list head 1790 * @list: list to use 1791 * @newsk: buffer to queue 1792 * 1793 * Queue a buffer at the start of a list. This function takes no locks 1794 * and you must therefore hold required locks before calling it. 1795 * 1796 * A buffer cannot be placed on two lists at the same time. 1797 */ 1798 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1799 static inline void __skb_queue_head(struct sk_buff_head *list, 1800 struct sk_buff *newsk) 1801 { 1802 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1803 } 1804 1805 /** 1806 * __skb_queue_tail - queue a buffer at the list tail 1807 * @list: list to use 1808 * @newsk: buffer to queue 1809 * 1810 * Queue a buffer at the end of a list. This function takes no locks 1811 * and you must therefore hold required locks before calling it. 1812 * 1813 * A buffer cannot be placed on two lists at the same time. 1814 */ 1815 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1816 static inline void __skb_queue_tail(struct sk_buff_head *list, 1817 struct sk_buff *newsk) 1818 { 1819 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1820 } 1821 1822 /* 1823 * remove sk_buff from list. _Must_ be called atomically, and with 1824 * the list known.. 1825 */ 1826 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1827 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1828 { 1829 struct sk_buff *next, *prev; 1830 1831 list->qlen--; 1832 next = skb->next; 1833 prev = skb->prev; 1834 skb->next = skb->prev = NULL; 1835 next->prev = prev; 1836 prev->next = next; 1837 } 1838 1839 /** 1840 * __skb_dequeue - remove from the head of the queue 1841 * @list: list to dequeue from 1842 * 1843 * Remove the head of the list. This function does not take any locks 1844 * so must be used with appropriate locks held only. The head item is 1845 * returned or %NULL if the list is empty. 1846 */ 1847 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1848 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1849 { 1850 struct sk_buff *skb = skb_peek(list); 1851 if (skb) 1852 __skb_unlink(skb, list); 1853 return skb; 1854 } 1855 1856 /** 1857 * __skb_dequeue_tail - remove from the tail of the queue 1858 * @list: list to dequeue from 1859 * 1860 * Remove the tail of the list. This function does not take any locks 1861 * so must be used with appropriate locks held only. The tail item is 1862 * returned or %NULL if the list is empty. 1863 */ 1864 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1865 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1866 { 1867 struct sk_buff *skb = skb_peek_tail(list); 1868 if (skb) 1869 __skb_unlink(skb, list); 1870 return skb; 1871 } 1872 1873 1874 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1875 { 1876 return skb->data_len; 1877 } 1878 1879 static inline unsigned int skb_headlen(const struct sk_buff *skb) 1880 { 1881 return skb->len - skb->data_len; 1882 } 1883 1884 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 1885 { 1886 unsigned int i, len = 0; 1887 1888 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 1889 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1890 return len; 1891 } 1892 1893 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 1894 { 1895 return skb_headlen(skb) + __skb_pagelen(skb); 1896 } 1897 1898 /** 1899 * __skb_fill_page_desc - initialise a paged fragment in an skb 1900 * @skb: buffer containing fragment to be initialised 1901 * @i: paged fragment index to initialise 1902 * @page: the page to use for this fragment 1903 * @off: the offset to the data with @page 1904 * @size: the length of the data 1905 * 1906 * Initialises the @i'th fragment of @skb to point to &size bytes at 1907 * offset @off within @page. 1908 * 1909 * Does not take any additional reference on the fragment. 1910 */ 1911 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1912 struct page *page, int off, int size) 1913 { 1914 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1915 1916 /* 1917 * Propagate page pfmemalloc to the skb if we can. The problem is 1918 * that not all callers have unique ownership of the page but rely 1919 * on page_is_pfmemalloc doing the right thing(tm). 1920 */ 1921 frag->page.p = page; 1922 frag->page_offset = off; 1923 skb_frag_size_set(frag, size); 1924 1925 page = compound_head(page); 1926 if (page_is_pfmemalloc(page)) 1927 skb->pfmemalloc = true; 1928 } 1929 1930 /** 1931 * skb_fill_page_desc - initialise a paged fragment in an skb 1932 * @skb: buffer containing fragment to be initialised 1933 * @i: paged fragment index to initialise 1934 * @page: the page to use for this fragment 1935 * @off: the offset to the data with @page 1936 * @size: the length of the data 1937 * 1938 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1939 * @skb to point to @size bytes at offset @off within @page. In 1940 * addition updates @skb such that @i is the last fragment. 1941 * 1942 * Does not take any additional reference on the fragment. 1943 */ 1944 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1945 struct page *page, int off, int size) 1946 { 1947 __skb_fill_page_desc(skb, i, page, off, size); 1948 skb_shinfo(skb)->nr_frags = i + 1; 1949 } 1950 1951 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1952 int size, unsigned int truesize); 1953 1954 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1955 unsigned int truesize); 1956 1957 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1958 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1959 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1960 1961 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1962 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1963 { 1964 return skb->head + skb->tail; 1965 } 1966 1967 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1968 { 1969 skb->tail = skb->data - skb->head; 1970 } 1971 1972 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1973 { 1974 skb_reset_tail_pointer(skb); 1975 skb->tail += offset; 1976 } 1977 1978 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 1979 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1980 { 1981 return skb->tail; 1982 } 1983 1984 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1985 { 1986 skb->tail = skb->data; 1987 } 1988 1989 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1990 { 1991 skb->tail = skb->data + offset; 1992 } 1993 1994 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1995 1996 /* 1997 * Add data to an sk_buff 1998 */ 1999 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2000 void *skb_put(struct sk_buff *skb, unsigned int len); 2001 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2002 { 2003 void *tmp = skb_tail_pointer(skb); 2004 SKB_LINEAR_ASSERT(skb); 2005 skb->tail += len; 2006 skb->len += len; 2007 return tmp; 2008 } 2009 2010 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2011 { 2012 void *tmp = __skb_put(skb, len); 2013 2014 memset(tmp, 0, len); 2015 return tmp; 2016 } 2017 2018 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2019 unsigned int len) 2020 { 2021 void *tmp = __skb_put(skb, len); 2022 2023 memcpy(tmp, data, len); 2024 return tmp; 2025 } 2026 2027 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2028 { 2029 *(u8 *)__skb_put(skb, 1) = val; 2030 } 2031 2032 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2033 { 2034 void *tmp = skb_put(skb, len); 2035 2036 memset(tmp, 0, len); 2037 2038 return tmp; 2039 } 2040 2041 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2042 unsigned int len) 2043 { 2044 void *tmp = skb_put(skb, len); 2045 2046 memcpy(tmp, data, len); 2047 2048 return tmp; 2049 } 2050 2051 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2052 { 2053 *(u8 *)skb_put(skb, 1) = val; 2054 } 2055 2056 void *skb_push(struct sk_buff *skb, unsigned int len); 2057 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2058 { 2059 skb->data -= len; 2060 skb->len += len; 2061 return skb->data; 2062 } 2063 2064 void *skb_pull(struct sk_buff *skb, unsigned int len); 2065 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2066 { 2067 skb->len -= len; 2068 BUG_ON(skb->len < skb->data_len); 2069 return skb->data += len; 2070 } 2071 2072 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2073 { 2074 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2075 } 2076 2077 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2078 2079 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len) 2080 { 2081 if (len > skb_headlen(skb) && 2082 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 2083 return NULL; 2084 skb->len -= len; 2085 return skb->data += len; 2086 } 2087 2088 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2089 { 2090 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 2091 } 2092 2093 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 2094 { 2095 if (likely(len <= skb_headlen(skb))) 2096 return 1; 2097 if (unlikely(len > skb->len)) 2098 return 0; 2099 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 2100 } 2101 2102 void skb_condense(struct sk_buff *skb); 2103 2104 /** 2105 * skb_headroom - bytes at buffer head 2106 * @skb: buffer to check 2107 * 2108 * Return the number of bytes of free space at the head of an &sk_buff. 2109 */ 2110 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2111 { 2112 return skb->data - skb->head; 2113 } 2114 2115 /** 2116 * skb_tailroom - bytes at buffer end 2117 * @skb: buffer to check 2118 * 2119 * Return the number of bytes of free space at the tail of an sk_buff 2120 */ 2121 static inline int skb_tailroom(const struct sk_buff *skb) 2122 { 2123 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2124 } 2125 2126 /** 2127 * skb_availroom - 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 * allocated by sk_stream_alloc() 2132 */ 2133 static inline int skb_availroom(const struct sk_buff *skb) 2134 { 2135 if (skb_is_nonlinear(skb)) 2136 return 0; 2137 2138 return skb->end - skb->tail - skb->reserved_tailroom; 2139 } 2140 2141 /** 2142 * skb_reserve - adjust headroom 2143 * @skb: buffer to alter 2144 * @len: bytes to move 2145 * 2146 * Increase the headroom of an empty &sk_buff by reducing the tail 2147 * room. This is only allowed for an empty buffer. 2148 */ 2149 static inline void skb_reserve(struct sk_buff *skb, int len) 2150 { 2151 skb->data += len; 2152 skb->tail += len; 2153 } 2154 2155 /** 2156 * skb_tailroom_reserve - adjust reserved_tailroom 2157 * @skb: buffer to alter 2158 * @mtu: maximum amount of headlen permitted 2159 * @needed_tailroom: minimum amount of reserved_tailroom 2160 * 2161 * Set reserved_tailroom so that headlen can be as large as possible but 2162 * not larger than mtu and tailroom cannot be smaller than 2163 * needed_tailroom. 2164 * The required headroom should already have been reserved before using 2165 * this function. 2166 */ 2167 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2168 unsigned int needed_tailroom) 2169 { 2170 SKB_LINEAR_ASSERT(skb); 2171 if (mtu < skb_tailroom(skb) - needed_tailroom) 2172 /* use at most mtu */ 2173 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2174 else 2175 /* use up to all available space */ 2176 skb->reserved_tailroom = needed_tailroom; 2177 } 2178 2179 #define ENCAP_TYPE_ETHER 0 2180 #define ENCAP_TYPE_IPPROTO 1 2181 2182 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2183 __be16 protocol) 2184 { 2185 skb->inner_protocol = protocol; 2186 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2187 } 2188 2189 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2190 __u8 ipproto) 2191 { 2192 skb->inner_ipproto = ipproto; 2193 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2194 } 2195 2196 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2197 { 2198 skb->inner_mac_header = skb->mac_header; 2199 skb->inner_network_header = skb->network_header; 2200 skb->inner_transport_header = skb->transport_header; 2201 } 2202 2203 static inline void skb_reset_mac_len(struct sk_buff *skb) 2204 { 2205 skb->mac_len = skb->network_header - skb->mac_header; 2206 } 2207 2208 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2209 *skb) 2210 { 2211 return skb->head + skb->inner_transport_header; 2212 } 2213 2214 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2215 { 2216 return skb_inner_transport_header(skb) - skb->data; 2217 } 2218 2219 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2220 { 2221 skb->inner_transport_header = skb->data - skb->head; 2222 } 2223 2224 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2225 const int offset) 2226 { 2227 skb_reset_inner_transport_header(skb); 2228 skb->inner_transport_header += offset; 2229 } 2230 2231 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2232 { 2233 return skb->head + skb->inner_network_header; 2234 } 2235 2236 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2237 { 2238 skb->inner_network_header = skb->data - skb->head; 2239 } 2240 2241 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2242 const int offset) 2243 { 2244 skb_reset_inner_network_header(skb); 2245 skb->inner_network_header += offset; 2246 } 2247 2248 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2249 { 2250 return skb->head + skb->inner_mac_header; 2251 } 2252 2253 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2254 { 2255 skb->inner_mac_header = skb->data - skb->head; 2256 } 2257 2258 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2259 const int offset) 2260 { 2261 skb_reset_inner_mac_header(skb); 2262 skb->inner_mac_header += offset; 2263 } 2264 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2265 { 2266 return skb->transport_header != (typeof(skb->transport_header))~0U; 2267 } 2268 2269 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2270 { 2271 return skb->head + skb->transport_header; 2272 } 2273 2274 static inline void skb_reset_transport_header(struct sk_buff *skb) 2275 { 2276 skb->transport_header = skb->data - skb->head; 2277 } 2278 2279 static inline void skb_set_transport_header(struct sk_buff *skb, 2280 const int offset) 2281 { 2282 skb_reset_transport_header(skb); 2283 skb->transport_header += offset; 2284 } 2285 2286 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2287 { 2288 return skb->head + skb->network_header; 2289 } 2290 2291 static inline void skb_reset_network_header(struct sk_buff *skb) 2292 { 2293 skb->network_header = skb->data - skb->head; 2294 } 2295 2296 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2297 { 2298 skb_reset_network_header(skb); 2299 skb->network_header += offset; 2300 } 2301 2302 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2303 { 2304 return skb->head + skb->mac_header; 2305 } 2306 2307 static inline int skb_mac_offset(const struct sk_buff *skb) 2308 { 2309 return skb_mac_header(skb) - skb->data; 2310 } 2311 2312 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2313 { 2314 return skb->network_header - skb->mac_header; 2315 } 2316 2317 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2318 { 2319 return skb->mac_header != (typeof(skb->mac_header))~0U; 2320 } 2321 2322 static inline void skb_reset_mac_header(struct sk_buff *skb) 2323 { 2324 skb->mac_header = skb->data - skb->head; 2325 } 2326 2327 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2328 { 2329 skb_reset_mac_header(skb); 2330 skb->mac_header += offset; 2331 } 2332 2333 static inline void skb_pop_mac_header(struct sk_buff *skb) 2334 { 2335 skb->mac_header = skb->network_header; 2336 } 2337 2338 static inline void skb_probe_transport_header(struct sk_buff *skb, 2339 const int offset_hint) 2340 { 2341 struct flow_keys keys; 2342 2343 if (skb_transport_header_was_set(skb)) 2344 return; 2345 else if (skb_flow_dissect_flow_keys(skb, &keys, 0)) 2346 skb_set_transport_header(skb, keys.control.thoff); 2347 else 2348 skb_set_transport_header(skb, offset_hint); 2349 } 2350 2351 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2352 { 2353 if (skb_mac_header_was_set(skb)) { 2354 const unsigned char *old_mac = skb_mac_header(skb); 2355 2356 skb_set_mac_header(skb, -skb->mac_len); 2357 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2358 } 2359 } 2360 2361 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2362 { 2363 return skb->csum_start - skb_headroom(skb); 2364 } 2365 2366 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2367 { 2368 return skb->head + skb->csum_start; 2369 } 2370 2371 static inline int skb_transport_offset(const struct sk_buff *skb) 2372 { 2373 return skb_transport_header(skb) - skb->data; 2374 } 2375 2376 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2377 { 2378 return skb->transport_header - skb->network_header; 2379 } 2380 2381 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2382 { 2383 return skb->inner_transport_header - skb->inner_network_header; 2384 } 2385 2386 static inline int skb_network_offset(const struct sk_buff *skb) 2387 { 2388 return skb_network_header(skb) - skb->data; 2389 } 2390 2391 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2392 { 2393 return skb_inner_network_header(skb) - skb->data; 2394 } 2395 2396 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2397 { 2398 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2399 } 2400 2401 /* 2402 * CPUs often take a performance hit when accessing unaligned memory 2403 * locations. The actual performance hit varies, it can be small if the 2404 * hardware handles it or large if we have to take an exception and fix it 2405 * in software. 2406 * 2407 * Since an ethernet header is 14 bytes network drivers often end up with 2408 * the IP header at an unaligned offset. The IP header can be aligned by 2409 * shifting the start of the packet by 2 bytes. Drivers should do this 2410 * with: 2411 * 2412 * skb_reserve(skb, NET_IP_ALIGN); 2413 * 2414 * The downside to this alignment of the IP header is that the DMA is now 2415 * unaligned. On some architectures the cost of an unaligned DMA is high 2416 * and this cost outweighs the gains made by aligning the IP header. 2417 * 2418 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2419 * to be overridden. 2420 */ 2421 #ifndef NET_IP_ALIGN 2422 #define NET_IP_ALIGN 2 2423 #endif 2424 2425 /* 2426 * The networking layer reserves some headroom in skb data (via 2427 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2428 * the header has to grow. In the default case, if the header has to grow 2429 * 32 bytes or less we avoid the reallocation. 2430 * 2431 * Unfortunately this headroom changes the DMA alignment of the resulting 2432 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2433 * on some architectures. An architecture can override this value, 2434 * perhaps setting it to a cacheline in size (since that will maintain 2435 * cacheline alignment of the DMA). It must be a power of 2. 2436 * 2437 * Various parts of the networking layer expect at least 32 bytes of 2438 * headroom, you should not reduce this. 2439 * 2440 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2441 * to reduce average number of cache lines per packet. 2442 * get_rps_cpus() for example only access one 64 bytes aligned block : 2443 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2444 */ 2445 #ifndef NET_SKB_PAD 2446 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2447 #endif 2448 2449 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2450 2451 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 2452 { 2453 if (unlikely(skb_is_nonlinear(skb))) { 2454 WARN_ON(1); 2455 return; 2456 } 2457 skb->len = len; 2458 skb_set_tail_pointer(skb, len); 2459 } 2460 2461 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2462 { 2463 __skb_set_length(skb, len); 2464 } 2465 2466 void skb_trim(struct sk_buff *skb, unsigned int len); 2467 2468 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2469 { 2470 if (skb->data_len) 2471 return ___pskb_trim(skb, len); 2472 __skb_trim(skb, len); 2473 return 0; 2474 } 2475 2476 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2477 { 2478 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2479 } 2480 2481 /** 2482 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2483 * @skb: buffer to alter 2484 * @len: new length 2485 * 2486 * This is identical to pskb_trim except that the caller knows that 2487 * the skb is not cloned so we should never get an error due to out- 2488 * of-memory. 2489 */ 2490 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2491 { 2492 int err = pskb_trim(skb, len); 2493 BUG_ON(err); 2494 } 2495 2496 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 2497 { 2498 unsigned int diff = len - skb->len; 2499 2500 if (skb_tailroom(skb) < diff) { 2501 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 2502 GFP_ATOMIC); 2503 if (ret) 2504 return ret; 2505 } 2506 __skb_set_length(skb, len); 2507 return 0; 2508 } 2509 2510 /** 2511 * skb_orphan - orphan a buffer 2512 * @skb: buffer to orphan 2513 * 2514 * If a buffer currently has an owner then we call the owner's 2515 * destructor function and make the @skb unowned. The buffer continues 2516 * to exist but is no longer charged to its former owner. 2517 */ 2518 static inline void skb_orphan(struct sk_buff *skb) 2519 { 2520 if (skb->destructor) { 2521 skb->destructor(skb); 2522 skb->destructor = NULL; 2523 skb->sk = NULL; 2524 } else { 2525 BUG_ON(skb->sk); 2526 } 2527 } 2528 2529 /** 2530 * skb_orphan_frags - orphan the frags contained in a buffer 2531 * @skb: buffer to orphan frags from 2532 * @gfp_mask: allocation mask for replacement pages 2533 * 2534 * For each frag in the SKB which needs a destructor (i.e. has an 2535 * owner) create a copy of that frag and release the original 2536 * page by calling the destructor. 2537 */ 2538 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2539 { 2540 if (likely(!skb_zcopy(skb))) 2541 return 0; 2542 if (skb_uarg(skb)->callback == sock_zerocopy_callback) 2543 return 0; 2544 return skb_copy_ubufs(skb, gfp_mask); 2545 } 2546 2547 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 2548 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 2549 { 2550 if (likely(!skb_zcopy(skb))) 2551 return 0; 2552 return skb_copy_ubufs(skb, gfp_mask); 2553 } 2554 2555 /** 2556 * __skb_queue_purge - empty a list 2557 * @list: list to empty 2558 * 2559 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2560 * the list and one reference dropped. This function does not take the 2561 * list lock and the caller must hold the relevant locks to use it. 2562 */ 2563 void skb_queue_purge(struct sk_buff_head *list); 2564 static inline void __skb_queue_purge(struct sk_buff_head *list) 2565 { 2566 struct sk_buff *skb; 2567 while ((skb = __skb_dequeue(list)) != NULL) 2568 kfree_skb(skb); 2569 } 2570 2571 void skb_rbtree_purge(struct rb_root *root); 2572 2573 void *netdev_alloc_frag(unsigned int fragsz); 2574 2575 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2576 gfp_t gfp_mask); 2577 2578 /** 2579 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2580 * @dev: network device to receive on 2581 * @length: length to allocate 2582 * 2583 * Allocate a new &sk_buff and assign it a usage count of one. The 2584 * buffer has unspecified headroom built in. Users should allocate 2585 * the headroom they think they need without accounting for the 2586 * built in space. The built in space is used for optimisations. 2587 * 2588 * %NULL is returned if there is no free memory. Although this function 2589 * allocates memory it can be called from an interrupt. 2590 */ 2591 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2592 unsigned int length) 2593 { 2594 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2595 } 2596 2597 /* legacy helper around __netdev_alloc_skb() */ 2598 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2599 gfp_t gfp_mask) 2600 { 2601 return __netdev_alloc_skb(NULL, length, gfp_mask); 2602 } 2603 2604 /* legacy helper around netdev_alloc_skb() */ 2605 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2606 { 2607 return netdev_alloc_skb(NULL, length); 2608 } 2609 2610 2611 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2612 unsigned int length, gfp_t gfp) 2613 { 2614 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2615 2616 if (NET_IP_ALIGN && skb) 2617 skb_reserve(skb, NET_IP_ALIGN); 2618 return skb; 2619 } 2620 2621 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2622 unsigned int length) 2623 { 2624 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2625 } 2626 2627 static inline void skb_free_frag(void *addr) 2628 { 2629 page_frag_free(addr); 2630 } 2631 2632 void *napi_alloc_frag(unsigned int fragsz); 2633 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2634 unsigned int length, gfp_t gfp_mask); 2635 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2636 unsigned int length) 2637 { 2638 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2639 } 2640 void napi_consume_skb(struct sk_buff *skb, int budget); 2641 2642 void __kfree_skb_flush(void); 2643 void __kfree_skb_defer(struct sk_buff *skb); 2644 2645 /** 2646 * __dev_alloc_pages - allocate page for network Rx 2647 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2648 * @order: size of the allocation 2649 * 2650 * Allocate a new page. 2651 * 2652 * %NULL is returned if there is no free memory. 2653 */ 2654 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2655 unsigned int order) 2656 { 2657 /* This piece of code contains several assumptions. 2658 * 1. This is for device Rx, therefor a cold page is preferred. 2659 * 2. The expectation is the user wants a compound page. 2660 * 3. If requesting a order 0 page it will not be compound 2661 * due to the check to see if order has a value in prep_new_page 2662 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2663 * code in gfp_to_alloc_flags that should be enforcing this. 2664 */ 2665 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 2666 2667 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2668 } 2669 2670 static inline struct page *dev_alloc_pages(unsigned int order) 2671 { 2672 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 2673 } 2674 2675 /** 2676 * __dev_alloc_page - allocate a page for network Rx 2677 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2678 * 2679 * Allocate a new page. 2680 * 2681 * %NULL is returned if there is no free memory. 2682 */ 2683 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2684 { 2685 return __dev_alloc_pages(gfp_mask, 0); 2686 } 2687 2688 static inline struct page *dev_alloc_page(void) 2689 { 2690 return dev_alloc_pages(0); 2691 } 2692 2693 /** 2694 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2695 * @page: The page that was allocated from skb_alloc_page 2696 * @skb: The skb that may need pfmemalloc set 2697 */ 2698 static inline void skb_propagate_pfmemalloc(struct page *page, 2699 struct sk_buff *skb) 2700 { 2701 if (page_is_pfmemalloc(page)) 2702 skb->pfmemalloc = true; 2703 } 2704 2705 /** 2706 * skb_frag_page - retrieve the page referred to by a paged fragment 2707 * @frag: the paged fragment 2708 * 2709 * Returns the &struct page associated with @frag. 2710 */ 2711 static inline struct page *skb_frag_page(const skb_frag_t *frag) 2712 { 2713 return frag->page.p; 2714 } 2715 2716 /** 2717 * __skb_frag_ref - take an addition reference on a paged fragment. 2718 * @frag: the paged fragment 2719 * 2720 * Takes an additional reference on the paged fragment @frag. 2721 */ 2722 static inline void __skb_frag_ref(skb_frag_t *frag) 2723 { 2724 get_page(skb_frag_page(frag)); 2725 } 2726 2727 /** 2728 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2729 * @skb: the buffer 2730 * @f: the fragment offset. 2731 * 2732 * Takes an additional reference on the @f'th paged fragment of @skb. 2733 */ 2734 static inline void skb_frag_ref(struct sk_buff *skb, int f) 2735 { 2736 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2737 } 2738 2739 /** 2740 * __skb_frag_unref - release a reference on a paged fragment. 2741 * @frag: the paged fragment 2742 * 2743 * Releases a reference on the paged fragment @frag. 2744 */ 2745 static inline void __skb_frag_unref(skb_frag_t *frag) 2746 { 2747 put_page(skb_frag_page(frag)); 2748 } 2749 2750 /** 2751 * skb_frag_unref - release a reference on a paged fragment of an skb. 2752 * @skb: the buffer 2753 * @f: the fragment offset 2754 * 2755 * Releases a reference on the @f'th paged fragment of @skb. 2756 */ 2757 static inline void skb_frag_unref(struct sk_buff *skb, int f) 2758 { 2759 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2760 } 2761 2762 /** 2763 * skb_frag_address - gets the address of the data contained in a paged fragment 2764 * @frag: the paged fragment buffer 2765 * 2766 * Returns the address of the data within @frag. The page must already 2767 * be mapped. 2768 */ 2769 static inline void *skb_frag_address(const skb_frag_t *frag) 2770 { 2771 return page_address(skb_frag_page(frag)) + frag->page_offset; 2772 } 2773 2774 /** 2775 * skb_frag_address_safe - 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. Checks that the page 2779 * is mapped and returns %NULL otherwise. 2780 */ 2781 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2782 { 2783 void *ptr = page_address(skb_frag_page(frag)); 2784 if (unlikely(!ptr)) 2785 return NULL; 2786 2787 return ptr + frag->page_offset; 2788 } 2789 2790 /** 2791 * __skb_frag_set_page - sets the page contained in a paged fragment 2792 * @frag: the paged fragment 2793 * @page: the page to set 2794 * 2795 * Sets the fragment @frag to contain @page. 2796 */ 2797 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2798 { 2799 frag->page.p = page; 2800 } 2801 2802 /** 2803 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2804 * @skb: the buffer 2805 * @f: the fragment offset 2806 * @page: the page to set 2807 * 2808 * Sets the @f'th fragment of @skb to contain @page. 2809 */ 2810 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2811 struct page *page) 2812 { 2813 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2814 } 2815 2816 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2817 2818 /** 2819 * skb_frag_dma_map - maps a paged fragment via the DMA API 2820 * @dev: the device to map the fragment to 2821 * @frag: the paged fragment to map 2822 * @offset: the offset within the fragment (starting at the 2823 * fragment's own offset) 2824 * @size: the number of bytes to map 2825 * @dir: the direction of the mapping (``PCI_DMA_*``) 2826 * 2827 * Maps the page associated with @frag to @device. 2828 */ 2829 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2830 const skb_frag_t *frag, 2831 size_t offset, size_t size, 2832 enum dma_data_direction dir) 2833 { 2834 return dma_map_page(dev, skb_frag_page(frag), 2835 frag->page_offset + offset, size, dir); 2836 } 2837 2838 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2839 gfp_t gfp_mask) 2840 { 2841 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2842 } 2843 2844 2845 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2846 gfp_t gfp_mask) 2847 { 2848 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2849 } 2850 2851 2852 /** 2853 * skb_clone_writable - is the header of a clone writable 2854 * @skb: buffer to check 2855 * @len: length up to which to write 2856 * 2857 * Returns true if modifying the header part of the cloned buffer 2858 * does not requires the data to be copied. 2859 */ 2860 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2861 { 2862 return !skb_header_cloned(skb) && 2863 skb_headroom(skb) + len <= skb->hdr_len; 2864 } 2865 2866 static inline int skb_try_make_writable(struct sk_buff *skb, 2867 unsigned int write_len) 2868 { 2869 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 2870 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2871 } 2872 2873 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2874 int cloned) 2875 { 2876 int delta = 0; 2877 2878 if (headroom > skb_headroom(skb)) 2879 delta = headroom - skb_headroom(skb); 2880 2881 if (delta || cloned) 2882 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2883 GFP_ATOMIC); 2884 return 0; 2885 } 2886 2887 /** 2888 * skb_cow - copy header of skb when it is required 2889 * @skb: buffer to cow 2890 * @headroom: needed headroom 2891 * 2892 * If the skb passed lacks sufficient headroom or its data part 2893 * is shared, data is reallocated. If reallocation fails, an error 2894 * is returned and original skb is not changed. 2895 * 2896 * The result is skb with writable area skb->head...skb->tail 2897 * and at least @headroom of space at head. 2898 */ 2899 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2900 { 2901 return __skb_cow(skb, headroom, skb_cloned(skb)); 2902 } 2903 2904 /** 2905 * skb_cow_head - skb_cow but only making the head writable 2906 * @skb: buffer to cow 2907 * @headroom: needed headroom 2908 * 2909 * This function is identical to skb_cow except that we replace the 2910 * skb_cloned check by skb_header_cloned. It should be used when 2911 * you only need to push on some header and do not need to modify 2912 * the data. 2913 */ 2914 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2915 { 2916 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2917 } 2918 2919 /** 2920 * skb_padto - pad an skbuff up to a minimal size 2921 * @skb: buffer to pad 2922 * @len: minimal length 2923 * 2924 * Pads up a buffer to ensure the trailing bytes exist and are 2925 * blanked. If the buffer already contains sufficient data it 2926 * is untouched. Otherwise it is extended. Returns zero on 2927 * success. The skb is freed on error. 2928 */ 2929 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2930 { 2931 unsigned int size = skb->len; 2932 if (likely(size >= len)) 2933 return 0; 2934 return skb_pad(skb, len - size); 2935 } 2936 2937 /** 2938 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2939 * @skb: buffer to pad 2940 * @len: minimal length 2941 * @free_on_error: free buffer on error 2942 * 2943 * Pads up a buffer to ensure the trailing bytes exist and are 2944 * blanked. If the buffer already contains sufficient data it 2945 * is untouched. Otherwise it is extended. Returns zero on 2946 * success. The skb is freed on error if @free_on_error is true. 2947 */ 2948 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len, 2949 bool free_on_error) 2950 { 2951 unsigned int size = skb->len; 2952 2953 if (unlikely(size < len)) { 2954 len -= size; 2955 if (__skb_pad(skb, len, free_on_error)) 2956 return -ENOMEM; 2957 __skb_put(skb, len); 2958 } 2959 return 0; 2960 } 2961 2962 /** 2963 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2964 * @skb: buffer to pad 2965 * @len: minimal length 2966 * 2967 * Pads up a buffer to ensure the trailing bytes exist and are 2968 * blanked. If the buffer already contains sufficient data it 2969 * is untouched. Otherwise it is extended. Returns zero on 2970 * success. The skb is freed on error. 2971 */ 2972 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2973 { 2974 return __skb_put_padto(skb, len, true); 2975 } 2976 2977 static inline int skb_add_data(struct sk_buff *skb, 2978 struct iov_iter *from, int copy) 2979 { 2980 const int off = skb->len; 2981 2982 if (skb->ip_summed == CHECKSUM_NONE) { 2983 __wsum csum = 0; 2984 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 2985 &csum, from)) { 2986 skb->csum = csum_block_add(skb->csum, csum, off); 2987 return 0; 2988 } 2989 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 2990 return 0; 2991 2992 __skb_trim(skb, off); 2993 return -EFAULT; 2994 } 2995 2996 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2997 const struct page *page, int off) 2998 { 2999 if (skb_zcopy(skb)) 3000 return false; 3001 if (i) { 3002 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 3003 3004 return page == skb_frag_page(frag) && 3005 off == frag->page_offset + skb_frag_size(frag); 3006 } 3007 return false; 3008 } 3009 3010 static inline int __skb_linearize(struct sk_buff *skb) 3011 { 3012 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3013 } 3014 3015 /** 3016 * skb_linearize - convert paged skb to linear one 3017 * @skb: buffer to linarize 3018 * 3019 * If there is no free memory -ENOMEM is returned, otherwise zero 3020 * is returned and the old skb data released. 3021 */ 3022 static inline int skb_linearize(struct sk_buff *skb) 3023 { 3024 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3025 } 3026 3027 /** 3028 * skb_has_shared_frag - can any frag be overwritten 3029 * @skb: buffer to test 3030 * 3031 * Return true if the skb has at least one frag that might be modified 3032 * by an external entity (as in vmsplice()/sendfile()) 3033 */ 3034 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3035 { 3036 return skb_is_nonlinear(skb) && 3037 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 3038 } 3039 3040 /** 3041 * skb_linearize_cow - make sure skb is linear and writable 3042 * @skb: buffer to process 3043 * 3044 * If there is no free memory -ENOMEM is returned, otherwise zero 3045 * is returned and the old skb data released. 3046 */ 3047 static inline int skb_linearize_cow(struct sk_buff *skb) 3048 { 3049 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3050 __skb_linearize(skb) : 0; 3051 } 3052 3053 static __always_inline void 3054 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3055 unsigned int off) 3056 { 3057 if (skb->ip_summed == CHECKSUM_COMPLETE) 3058 skb->csum = csum_block_sub(skb->csum, 3059 csum_partial(start, len, 0), off); 3060 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3061 skb_checksum_start_offset(skb) < 0) 3062 skb->ip_summed = CHECKSUM_NONE; 3063 } 3064 3065 /** 3066 * skb_postpull_rcsum - update checksum for received skb after pull 3067 * @skb: buffer to update 3068 * @start: start of data before pull 3069 * @len: length of data pulled 3070 * 3071 * After doing a pull on a received packet, you need to call this to 3072 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3073 * CHECKSUM_NONE so that it can be recomputed from scratch. 3074 */ 3075 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3076 const void *start, unsigned int len) 3077 { 3078 __skb_postpull_rcsum(skb, start, len, 0); 3079 } 3080 3081 static __always_inline void 3082 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3083 unsigned int off) 3084 { 3085 if (skb->ip_summed == CHECKSUM_COMPLETE) 3086 skb->csum = csum_block_add(skb->csum, 3087 csum_partial(start, len, 0), off); 3088 } 3089 3090 /** 3091 * skb_postpush_rcsum - update checksum for received skb after push 3092 * @skb: buffer to update 3093 * @start: start of data after push 3094 * @len: length of data pushed 3095 * 3096 * After doing a push on a received packet, you need to call this to 3097 * update the CHECKSUM_COMPLETE checksum. 3098 */ 3099 static inline void skb_postpush_rcsum(struct sk_buff *skb, 3100 const void *start, unsigned int len) 3101 { 3102 __skb_postpush_rcsum(skb, start, len, 0); 3103 } 3104 3105 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3106 3107 /** 3108 * skb_push_rcsum - push skb and update receive checksum 3109 * @skb: buffer to update 3110 * @len: length of data pulled 3111 * 3112 * This function performs an skb_push on the packet and updates 3113 * the CHECKSUM_COMPLETE checksum. It should be used on 3114 * receive path processing instead of skb_push unless you know 3115 * that the checksum difference is zero (e.g., a valid IP header) 3116 * or you are setting ip_summed to CHECKSUM_NONE. 3117 */ 3118 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3119 { 3120 skb_push(skb, len); 3121 skb_postpush_rcsum(skb, skb->data, len); 3122 return skb->data; 3123 } 3124 3125 /** 3126 * pskb_trim_rcsum - trim received skb and update checksum 3127 * @skb: buffer to trim 3128 * @len: new length 3129 * 3130 * This is exactly the same as pskb_trim except that it ensures the 3131 * checksum of received packets are still valid after the operation. 3132 */ 3133 3134 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3135 { 3136 if (likely(len >= skb->len)) 3137 return 0; 3138 if (skb->ip_summed == CHECKSUM_COMPLETE) 3139 skb->ip_summed = CHECKSUM_NONE; 3140 return __pskb_trim(skb, len); 3141 } 3142 3143 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3144 { 3145 if (skb->ip_summed == CHECKSUM_COMPLETE) 3146 skb->ip_summed = CHECKSUM_NONE; 3147 __skb_trim(skb, len); 3148 return 0; 3149 } 3150 3151 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3152 { 3153 if (skb->ip_summed == CHECKSUM_COMPLETE) 3154 skb->ip_summed = CHECKSUM_NONE; 3155 return __skb_grow(skb, len); 3156 } 3157 3158 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3159 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 3160 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 3161 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3162 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3163 3164 #define skb_queue_walk(queue, skb) \ 3165 for (skb = (queue)->next; \ 3166 skb != (struct sk_buff *)(queue); \ 3167 skb = skb->next) 3168 3169 #define skb_queue_walk_safe(queue, skb, tmp) \ 3170 for (skb = (queue)->next, tmp = skb->next; \ 3171 skb != (struct sk_buff *)(queue); \ 3172 skb = tmp, tmp = skb->next) 3173 3174 #define skb_queue_walk_from(queue, skb) \ 3175 for (; skb != (struct sk_buff *)(queue); \ 3176 skb = skb->next) 3177 3178 #define skb_rbtree_walk(skb, root) \ 3179 for (skb = skb_rb_first(root); skb != NULL; \ 3180 skb = skb_rb_next(skb)) 3181 3182 #define skb_rbtree_walk_from(skb) \ 3183 for (; skb != NULL; \ 3184 skb = skb_rb_next(skb)) 3185 3186 #define skb_rbtree_walk_from_safe(skb, tmp) \ 3187 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3188 skb = tmp) 3189 3190 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 3191 for (tmp = skb->next; \ 3192 skb != (struct sk_buff *)(queue); \ 3193 skb = tmp, tmp = skb->next) 3194 3195 #define skb_queue_reverse_walk(queue, skb) \ 3196 for (skb = (queue)->prev; \ 3197 skb != (struct sk_buff *)(queue); \ 3198 skb = skb->prev) 3199 3200 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3201 for (skb = (queue)->prev, tmp = skb->prev; \ 3202 skb != (struct sk_buff *)(queue); \ 3203 skb = tmp, tmp = skb->prev) 3204 3205 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3206 for (tmp = skb->prev; \ 3207 skb != (struct sk_buff *)(queue); \ 3208 skb = tmp, tmp = skb->prev) 3209 3210 static inline bool skb_has_frag_list(const struct sk_buff *skb) 3211 { 3212 return skb_shinfo(skb)->frag_list != NULL; 3213 } 3214 3215 static inline void skb_frag_list_init(struct sk_buff *skb) 3216 { 3217 skb_shinfo(skb)->frag_list = NULL; 3218 } 3219 3220 #define skb_walk_frags(skb, iter) \ 3221 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3222 3223 3224 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, 3225 const struct sk_buff *skb); 3226 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3227 struct sk_buff_head *queue, 3228 unsigned int flags, 3229 void (*destructor)(struct sock *sk, 3230 struct sk_buff *skb), 3231 int *peeked, int *off, int *err, 3232 struct sk_buff **last); 3233 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags, 3234 void (*destructor)(struct sock *sk, 3235 struct sk_buff *skb), 3236 int *peeked, int *off, int *err, 3237 struct sk_buff **last); 3238 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 3239 void (*destructor)(struct sock *sk, 3240 struct sk_buff *skb), 3241 int *peeked, int *off, int *err); 3242 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 3243 int *err); 3244 unsigned int datagram_poll(struct file *file, struct socket *sock, 3245 struct poll_table_struct *wait); 3246 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3247 struct iov_iter *to, int size); 3248 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3249 struct msghdr *msg, int size) 3250 { 3251 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3252 } 3253 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3254 struct msghdr *msg); 3255 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3256 struct iov_iter *from, int len); 3257 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3258 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3259 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3260 static inline void skb_free_datagram_locked(struct sock *sk, 3261 struct sk_buff *skb) 3262 { 3263 __skb_free_datagram_locked(sk, skb, 0); 3264 } 3265 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3266 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3267 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3268 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3269 int len, __wsum csum); 3270 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3271 struct pipe_inode_info *pipe, unsigned int len, 3272 unsigned int flags); 3273 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 3274 int len); 3275 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 3276 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3277 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3278 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3279 int len, int hlen); 3280 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3281 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3282 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3283 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 3284 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu); 3285 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3286 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3287 int skb_ensure_writable(struct sk_buff *skb, int write_len); 3288 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3289 int skb_vlan_pop(struct sk_buff *skb); 3290 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3291 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3292 gfp_t gfp); 3293 3294 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3295 { 3296 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 3297 } 3298 3299 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3300 { 3301 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3302 } 3303 3304 struct skb_checksum_ops { 3305 __wsum (*update)(const void *mem, int len, __wsum wsum); 3306 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3307 }; 3308 3309 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 3310 3311 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3312 __wsum csum, const struct skb_checksum_ops *ops); 3313 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3314 __wsum csum); 3315 3316 static inline void * __must_check 3317 __skb_header_pointer(const struct sk_buff *skb, int offset, 3318 int len, void *data, int hlen, void *buffer) 3319 { 3320 if (hlen - offset >= len) 3321 return data + offset; 3322 3323 if (!skb || 3324 skb_copy_bits(skb, offset, buffer, len) < 0) 3325 return NULL; 3326 3327 return buffer; 3328 } 3329 3330 static inline void * __must_check 3331 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 3332 { 3333 return __skb_header_pointer(skb, offset, len, skb->data, 3334 skb_headlen(skb), buffer); 3335 } 3336 3337 /** 3338 * skb_needs_linearize - check if we need to linearize a given skb 3339 * depending on the given device features. 3340 * @skb: socket buffer to check 3341 * @features: net device features 3342 * 3343 * Returns true if either: 3344 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3345 * 2. skb is fragmented and the device does not support SG. 3346 */ 3347 static inline bool skb_needs_linearize(struct sk_buff *skb, 3348 netdev_features_t features) 3349 { 3350 return skb_is_nonlinear(skb) && 3351 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3352 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3353 } 3354 3355 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3356 void *to, 3357 const unsigned int len) 3358 { 3359 memcpy(to, skb->data, len); 3360 } 3361 3362 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3363 const int offset, void *to, 3364 const unsigned int len) 3365 { 3366 memcpy(to, skb->data + offset, len); 3367 } 3368 3369 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3370 const void *from, 3371 const unsigned int len) 3372 { 3373 memcpy(skb->data, from, len); 3374 } 3375 3376 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3377 const int offset, 3378 const void *from, 3379 const unsigned int len) 3380 { 3381 memcpy(skb->data + offset, from, len); 3382 } 3383 3384 void skb_init(void); 3385 3386 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3387 { 3388 return skb->tstamp; 3389 } 3390 3391 /** 3392 * skb_get_timestamp - get timestamp from a skb 3393 * @skb: skb to get stamp from 3394 * @stamp: pointer to struct timeval to store stamp in 3395 * 3396 * Timestamps are stored in the skb as offsets to a base timestamp. 3397 * This function converts the offset back to a struct timeval and stores 3398 * it in stamp. 3399 */ 3400 static inline void skb_get_timestamp(const struct sk_buff *skb, 3401 struct timeval *stamp) 3402 { 3403 *stamp = ktime_to_timeval(skb->tstamp); 3404 } 3405 3406 static inline void skb_get_timestampns(const struct sk_buff *skb, 3407 struct timespec *stamp) 3408 { 3409 *stamp = ktime_to_timespec(skb->tstamp); 3410 } 3411 3412 static inline void __net_timestamp(struct sk_buff *skb) 3413 { 3414 skb->tstamp = ktime_get_real(); 3415 } 3416 3417 static inline ktime_t net_timedelta(ktime_t t) 3418 { 3419 return ktime_sub(ktime_get_real(), t); 3420 } 3421 3422 static inline ktime_t net_invalid_timestamp(void) 3423 { 3424 return 0; 3425 } 3426 3427 static inline u8 skb_metadata_len(const struct sk_buff *skb) 3428 { 3429 return skb_shinfo(skb)->meta_len; 3430 } 3431 3432 static inline void *skb_metadata_end(const struct sk_buff *skb) 3433 { 3434 return skb_mac_header(skb); 3435 } 3436 3437 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 3438 const struct sk_buff *skb_b, 3439 u8 meta_len) 3440 { 3441 const void *a = skb_metadata_end(skb_a); 3442 const void *b = skb_metadata_end(skb_b); 3443 /* Using more efficient varaiant than plain call to memcmp(). */ 3444 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64 3445 u64 diffs = 0; 3446 3447 switch (meta_len) { 3448 #define __it(x, op) (x -= sizeof(u##op)) 3449 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 3450 case 32: diffs |= __it_diff(a, b, 64); 3451 case 24: diffs |= __it_diff(a, b, 64); 3452 case 16: diffs |= __it_diff(a, b, 64); 3453 case 8: diffs |= __it_diff(a, b, 64); 3454 break; 3455 case 28: diffs |= __it_diff(a, b, 64); 3456 case 20: diffs |= __it_diff(a, b, 64); 3457 case 12: diffs |= __it_diff(a, b, 64); 3458 case 4: diffs |= __it_diff(a, b, 32); 3459 break; 3460 } 3461 return diffs; 3462 #else 3463 return memcmp(a - meta_len, b - meta_len, meta_len); 3464 #endif 3465 } 3466 3467 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 3468 const struct sk_buff *skb_b) 3469 { 3470 u8 len_a = skb_metadata_len(skb_a); 3471 u8 len_b = skb_metadata_len(skb_b); 3472 3473 if (!(len_a | len_b)) 3474 return false; 3475 3476 return len_a != len_b ? 3477 true : __skb_metadata_differs(skb_a, skb_b, len_a); 3478 } 3479 3480 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 3481 { 3482 skb_shinfo(skb)->meta_len = meta_len; 3483 } 3484 3485 static inline void skb_metadata_clear(struct sk_buff *skb) 3486 { 3487 skb_metadata_set(skb, 0); 3488 } 3489 3490 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3491 3492 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3493 3494 void skb_clone_tx_timestamp(struct sk_buff *skb); 3495 bool skb_defer_rx_timestamp(struct sk_buff *skb); 3496 3497 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3498 3499 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3500 { 3501 } 3502 3503 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3504 { 3505 return false; 3506 } 3507 3508 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3509 3510 /** 3511 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3512 * 3513 * PHY drivers may accept clones of transmitted packets for 3514 * timestamping via their phy_driver.txtstamp method. These drivers 3515 * must call this function to return the skb back to the stack with a 3516 * timestamp. 3517 * 3518 * @skb: clone of the the original outgoing packet 3519 * @hwtstamps: hardware time stamps 3520 * 3521 */ 3522 void skb_complete_tx_timestamp(struct sk_buff *skb, 3523 struct skb_shared_hwtstamps *hwtstamps); 3524 3525 void __skb_tstamp_tx(struct sk_buff *orig_skb, 3526 struct skb_shared_hwtstamps *hwtstamps, 3527 struct sock *sk, int tstype); 3528 3529 /** 3530 * skb_tstamp_tx - queue clone of skb with send time stamps 3531 * @orig_skb: the original outgoing packet 3532 * @hwtstamps: hardware time stamps, may be NULL if not available 3533 * 3534 * If the skb has a socket associated, then this function clones the 3535 * skb (thus sharing the actual data and optional structures), stores 3536 * the optional hardware time stamping information (if non NULL) or 3537 * generates a software time stamp (otherwise), then queues the clone 3538 * to the error queue of the socket. Errors are silently ignored. 3539 */ 3540 void skb_tstamp_tx(struct sk_buff *orig_skb, 3541 struct skb_shared_hwtstamps *hwtstamps); 3542 3543 /** 3544 * skb_tx_timestamp() - Driver hook for transmit timestamping 3545 * 3546 * Ethernet MAC Drivers should call this function in their hard_xmit() 3547 * function immediately before giving the sk_buff to the MAC hardware. 3548 * 3549 * Specifically, one should make absolutely sure that this function is 3550 * called before TX completion of this packet can trigger. Otherwise 3551 * the packet could potentially already be freed. 3552 * 3553 * @skb: A socket buffer. 3554 */ 3555 static inline void skb_tx_timestamp(struct sk_buff *skb) 3556 { 3557 skb_clone_tx_timestamp(skb); 3558 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 3559 skb_tstamp_tx(skb, NULL); 3560 } 3561 3562 /** 3563 * skb_complete_wifi_ack - deliver skb with wifi status 3564 * 3565 * @skb: the original outgoing packet 3566 * @acked: ack status 3567 * 3568 */ 3569 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3570 3571 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3572 __sum16 __skb_checksum_complete(struct sk_buff *skb); 3573 3574 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3575 { 3576 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3577 skb->csum_valid || 3578 (skb->ip_summed == CHECKSUM_PARTIAL && 3579 skb_checksum_start_offset(skb) >= 0)); 3580 } 3581 3582 /** 3583 * skb_checksum_complete - Calculate checksum of an entire packet 3584 * @skb: packet to process 3585 * 3586 * This function calculates the checksum over the entire packet plus 3587 * the value of skb->csum. The latter can be used to supply the 3588 * checksum of a pseudo header as used by TCP/UDP. It returns the 3589 * checksum. 3590 * 3591 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3592 * this function can be used to verify that checksum on received 3593 * packets. In that case the function should return zero if the 3594 * checksum is correct. In particular, this function will return zero 3595 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3596 * hardware has already verified the correctness of the checksum. 3597 */ 3598 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3599 { 3600 return skb_csum_unnecessary(skb) ? 3601 0 : __skb_checksum_complete(skb); 3602 } 3603 3604 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3605 { 3606 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3607 if (skb->csum_level == 0) 3608 skb->ip_summed = CHECKSUM_NONE; 3609 else 3610 skb->csum_level--; 3611 } 3612 } 3613 3614 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3615 { 3616 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3617 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3618 skb->csum_level++; 3619 } else if (skb->ip_summed == CHECKSUM_NONE) { 3620 skb->ip_summed = CHECKSUM_UNNECESSARY; 3621 skb->csum_level = 0; 3622 } 3623 } 3624 3625 /* Check if we need to perform checksum complete validation. 3626 * 3627 * Returns true if checksum complete is needed, false otherwise 3628 * (either checksum is unnecessary or zero checksum is allowed). 3629 */ 3630 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3631 bool zero_okay, 3632 __sum16 check) 3633 { 3634 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3635 skb->csum_valid = 1; 3636 __skb_decr_checksum_unnecessary(skb); 3637 return false; 3638 } 3639 3640 return true; 3641 } 3642 3643 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3644 * in checksum_init. 3645 */ 3646 #define CHECKSUM_BREAK 76 3647 3648 /* Unset checksum-complete 3649 * 3650 * Unset checksum complete can be done when packet is being modified 3651 * (uncompressed for instance) and checksum-complete value is 3652 * invalidated. 3653 */ 3654 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3655 { 3656 if (skb->ip_summed == CHECKSUM_COMPLETE) 3657 skb->ip_summed = CHECKSUM_NONE; 3658 } 3659 3660 /* Validate (init) checksum based on checksum complete. 3661 * 3662 * Return values: 3663 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3664 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3665 * checksum is stored in skb->csum for use in __skb_checksum_complete 3666 * non-zero: value of invalid checksum 3667 * 3668 */ 3669 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3670 bool complete, 3671 __wsum psum) 3672 { 3673 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3674 if (!csum_fold(csum_add(psum, skb->csum))) { 3675 skb->csum_valid = 1; 3676 return 0; 3677 } 3678 } 3679 3680 skb->csum = psum; 3681 3682 if (complete || skb->len <= CHECKSUM_BREAK) { 3683 __sum16 csum; 3684 3685 csum = __skb_checksum_complete(skb); 3686 skb->csum_valid = !csum; 3687 return csum; 3688 } 3689 3690 return 0; 3691 } 3692 3693 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3694 { 3695 return 0; 3696 } 3697 3698 /* Perform checksum validate (init). Note that this is a macro since we only 3699 * want to calculate the pseudo header which is an input function if necessary. 3700 * First we try to validate without any computation (checksum unnecessary) and 3701 * then calculate based on checksum complete calling the function to compute 3702 * pseudo header. 3703 * 3704 * Return values: 3705 * 0: checksum is validated or try to in skb_checksum_complete 3706 * non-zero: value of invalid checksum 3707 */ 3708 #define __skb_checksum_validate(skb, proto, complete, \ 3709 zero_okay, check, compute_pseudo) \ 3710 ({ \ 3711 __sum16 __ret = 0; \ 3712 skb->csum_valid = 0; \ 3713 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3714 __ret = __skb_checksum_validate_complete(skb, \ 3715 complete, compute_pseudo(skb, proto)); \ 3716 __ret; \ 3717 }) 3718 3719 #define skb_checksum_init(skb, proto, compute_pseudo) \ 3720 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3721 3722 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3723 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3724 3725 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 3726 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3727 3728 #define skb_checksum_validate_zero_check(skb, proto, check, \ 3729 compute_pseudo) \ 3730 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3731 3732 #define skb_checksum_simple_validate(skb) \ 3733 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3734 3735 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3736 { 3737 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 3738 } 3739 3740 static inline void __skb_checksum_convert(struct sk_buff *skb, 3741 __sum16 check, __wsum pseudo) 3742 { 3743 skb->csum = ~pseudo; 3744 skb->ip_summed = CHECKSUM_COMPLETE; 3745 } 3746 3747 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3748 do { \ 3749 if (__skb_checksum_convert_check(skb)) \ 3750 __skb_checksum_convert(skb, check, \ 3751 compute_pseudo(skb, proto)); \ 3752 } while (0) 3753 3754 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3755 u16 start, u16 offset) 3756 { 3757 skb->ip_summed = CHECKSUM_PARTIAL; 3758 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3759 skb->csum_offset = offset - start; 3760 } 3761 3762 /* Update skbuf and packet to reflect the remote checksum offload operation. 3763 * When called, ptr indicates the starting point for skb->csum when 3764 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3765 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3766 */ 3767 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3768 int start, int offset, bool nopartial) 3769 { 3770 __wsum delta; 3771 3772 if (!nopartial) { 3773 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3774 return; 3775 } 3776 3777 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3778 __skb_checksum_complete(skb); 3779 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3780 } 3781 3782 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3783 3784 /* Adjust skb->csum since we changed the packet */ 3785 skb->csum = csum_add(skb->csum, delta); 3786 } 3787 3788 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 3789 { 3790 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 3791 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK); 3792 #else 3793 return NULL; 3794 #endif 3795 } 3796 3797 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3798 void nf_conntrack_destroy(struct nf_conntrack *nfct); 3799 static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3800 { 3801 if (nfct && atomic_dec_and_test(&nfct->use)) 3802 nf_conntrack_destroy(nfct); 3803 } 3804 static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3805 { 3806 if (nfct) 3807 atomic_inc(&nfct->use); 3808 } 3809 #endif 3810 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3811 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3812 { 3813 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use)) 3814 kfree(nf_bridge); 3815 } 3816 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3817 { 3818 if (nf_bridge) 3819 refcount_inc(&nf_bridge->use); 3820 } 3821 #endif /* CONFIG_BRIDGE_NETFILTER */ 3822 static inline void nf_reset(struct sk_buff *skb) 3823 { 3824 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3825 nf_conntrack_put(skb_nfct(skb)); 3826 skb->_nfct = 0; 3827 #endif 3828 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3829 nf_bridge_put(skb->nf_bridge); 3830 skb->nf_bridge = NULL; 3831 #endif 3832 } 3833 3834 static inline void nf_reset_trace(struct sk_buff *skb) 3835 { 3836 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3837 skb->nf_trace = 0; 3838 #endif 3839 } 3840 3841 static inline void ipvs_reset(struct sk_buff *skb) 3842 { 3843 #if IS_ENABLED(CONFIG_IP_VS) 3844 skb->ipvs_property = 0; 3845 #endif 3846 } 3847 3848 /* Note: This doesn't put any conntrack and bridge info in dst. */ 3849 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3850 bool copy) 3851 { 3852 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3853 dst->_nfct = src->_nfct; 3854 nf_conntrack_get(skb_nfct(src)); 3855 #endif 3856 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3857 dst->nf_bridge = src->nf_bridge; 3858 nf_bridge_get(src->nf_bridge); 3859 #endif 3860 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3861 if (copy) 3862 dst->nf_trace = src->nf_trace; 3863 #endif 3864 } 3865 3866 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3867 { 3868 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3869 nf_conntrack_put(skb_nfct(dst)); 3870 #endif 3871 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3872 nf_bridge_put(dst->nf_bridge); 3873 #endif 3874 __nf_copy(dst, src, true); 3875 } 3876 3877 #ifdef CONFIG_NETWORK_SECMARK 3878 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3879 { 3880 to->secmark = from->secmark; 3881 } 3882 3883 static inline void skb_init_secmark(struct sk_buff *skb) 3884 { 3885 skb->secmark = 0; 3886 } 3887 #else 3888 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3889 { } 3890 3891 static inline void skb_init_secmark(struct sk_buff *skb) 3892 { } 3893 #endif 3894 3895 static inline bool skb_irq_freeable(const struct sk_buff *skb) 3896 { 3897 return !skb->destructor && 3898 #if IS_ENABLED(CONFIG_XFRM) 3899 !skb->sp && 3900 #endif 3901 !skb_nfct(skb) && 3902 !skb->_skb_refdst && 3903 !skb_has_frag_list(skb); 3904 } 3905 3906 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3907 { 3908 skb->queue_mapping = queue_mapping; 3909 } 3910 3911 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3912 { 3913 return skb->queue_mapping; 3914 } 3915 3916 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3917 { 3918 to->queue_mapping = from->queue_mapping; 3919 } 3920 3921 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3922 { 3923 skb->queue_mapping = rx_queue + 1; 3924 } 3925 3926 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3927 { 3928 return skb->queue_mapping - 1; 3929 } 3930 3931 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3932 { 3933 return skb->queue_mapping != 0; 3934 } 3935 3936 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 3937 { 3938 skb->dst_pending_confirm = val; 3939 } 3940 3941 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 3942 { 3943 return skb->dst_pending_confirm != 0; 3944 } 3945 3946 static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3947 { 3948 #ifdef CONFIG_XFRM 3949 return skb->sp; 3950 #else 3951 return NULL; 3952 #endif 3953 } 3954 3955 /* Keeps track of mac header offset relative to skb->head. 3956 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3957 * For non-tunnel skb it points to skb_mac_header() and for 3958 * tunnel skb it points to outer mac header. 3959 * Keeps track of level of encapsulation of network headers. 3960 */ 3961 struct skb_gso_cb { 3962 union { 3963 int mac_offset; 3964 int data_offset; 3965 }; 3966 int encap_level; 3967 __wsum csum; 3968 __u16 csum_start; 3969 }; 3970 #define SKB_SGO_CB_OFFSET 32 3971 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 3972 3973 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3974 { 3975 return (skb_mac_header(inner_skb) - inner_skb->head) - 3976 SKB_GSO_CB(inner_skb)->mac_offset; 3977 } 3978 3979 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3980 { 3981 int new_headroom, headroom; 3982 int ret; 3983 3984 headroom = skb_headroom(skb); 3985 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3986 if (ret) 3987 return ret; 3988 3989 new_headroom = skb_headroom(skb); 3990 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3991 return 0; 3992 } 3993 3994 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 3995 { 3996 /* Do not update partial checksums if remote checksum is enabled. */ 3997 if (skb->remcsum_offload) 3998 return; 3999 4000 SKB_GSO_CB(skb)->csum = res; 4001 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 4002 } 4003 4004 /* Compute the checksum for a gso segment. First compute the checksum value 4005 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 4006 * then add in skb->csum (checksum from csum_start to end of packet). 4007 * skb->csum and csum_start are then updated to reflect the checksum of the 4008 * resultant packet starting from the transport header-- the resultant checksum 4009 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 4010 * header. 4011 */ 4012 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 4013 { 4014 unsigned char *csum_start = skb_transport_header(skb); 4015 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 4016 __wsum partial = SKB_GSO_CB(skb)->csum; 4017 4018 SKB_GSO_CB(skb)->csum = res; 4019 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 4020 4021 return csum_fold(csum_partial(csum_start, plen, partial)); 4022 } 4023 4024 static inline bool skb_is_gso(const struct sk_buff *skb) 4025 { 4026 return skb_shinfo(skb)->gso_size; 4027 } 4028 4029 /* Note: Should be called only if skb_is_gso(skb) is true */ 4030 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4031 { 4032 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4033 } 4034 4035 static inline void skb_gso_reset(struct sk_buff *skb) 4036 { 4037 skb_shinfo(skb)->gso_size = 0; 4038 skb_shinfo(skb)->gso_segs = 0; 4039 skb_shinfo(skb)->gso_type = 0; 4040 } 4041 4042 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 4043 4044 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 4045 { 4046 /* LRO sets gso_size but not gso_type, whereas if GSO is really 4047 * wanted then gso_type will be set. */ 4048 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4049 4050 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 4051 unlikely(shinfo->gso_type == 0)) { 4052 __skb_warn_lro_forwarding(skb); 4053 return true; 4054 } 4055 return false; 4056 } 4057 4058 static inline void skb_forward_csum(struct sk_buff *skb) 4059 { 4060 /* Unfortunately we don't support this one. Any brave souls? */ 4061 if (skb->ip_summed == CHECKSUM_COMPLETE) 4062 skb->ip_summed = CHECKSUM_NONE; 4063 } 4064 4065 /** 4066 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 4067 * @skb: skb to check 4068 * 4069 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 4070 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 4071 * use this helper, to document places where we make this assertion. 4072 */ 4073 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 4074 { 4075 #ifdef DEBUG 4076 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 4077 #endif 4078 } 4079 4080 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 4081 4082 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 4083 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4084 unsigned int transport_len, 4085 __sum16(*skb_chkf)(struct sk_buff *skb)); 4086 4087 /** 4088 * skb_head_is_locked - Determine if the skb->head is locked down 4089 * @skb: skb to check 4090 * 4091 * The head on skbs build around a head frag can be removed if they are 4092 * not cloned. This function returns true if the skb head is locked down 4093 * due to either being allocated via kmalloc, or by being a clone with 4094 * multiple references to the head. 4095 */ 4096 static inline bool skb_head_is_locked(const struct sk_buff *skb) 4097 { 4098 return !skb->head_frag || skb_cloned(skb); 4099 } 4100 4101 /** 4102 * skb_gso_network_seglen - Return length of individual segments of a gso packet 4103 * 4104 * @skb: GSO skb 4105 * 4106 * skb_gso_network_seglen is used to determine the real size of the 4107 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 4108 * 4109 * The MAC/L2 header is not accounted for. 4110 */ 4111 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 4112 { 4113 unsigned int hdr_len = skb_transport_header(skb) - 4114 skb_network_header(skb); 4115 return hdr_len + skb_gso_transport_seglen(skb); 4116 } 4117 4118 /* Local Checksum Offload. 4119 * Compute outer checksum based on the assumption that the 4120 * inner checksum will be offloaded later. 4121 * See Documentation/networking/checksum-offloads.txt for 4122 * explanation of how this works. 4123 * Fill in outer checksum adjustment (e.g. with sum of outer 4124 * pseudo-header) before calling. 4125 * Also ensure that inner checksum is in linear data area. 4126 */ 4127 static inline __wsum lco_csum(struct sk_buff *skb) 4128 { 4129 unsigned char *csum_start = skb_checksum_start(skb); 4130 unsigned char *l4_hdr = skb_transport_header(skb); 4131 __wsum partial; 4132 4133 /* Start with complement of inner checksum adjustment */ 4134 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 4135 skb->csum_offset)); 4136 4137 /* Add in checksum of our headers (incl. outer checksum 4138 * adjustment filled in by caller) and return result. 4139 */ 4140 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 4141 } 4142 4143 #endif /* __KERNEL__ */ 4144 #endif /* _LINUX_SKBUFF_H */ 4145
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.