1 /* 2 * Definitions for the 'struct sk_buff' memory handlers. 3 * 4 * Authors: 5 * Alan Cox, <[email protected]> 6 * Florian La Roche, <[email protected]> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 */ 13 14 #ifndef _LINUX_SKBUFF_H 15 #define _LINUX_SKBUFF_H 16 17 #include <linux/kernel.h> 18 #include <linux/kmemcheck.h> 19 #include <linux/compiler.h> 20 #include <linux/time.h> 21 #include <linux/bug.h> 22 #include <linux/cache.h> 23 #include <linux/rbtree.h> 24 #include <linux/socket.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 atomic_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 atomic_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 unsigned short _unused; 492 unsigned char nr_frags; 493 __u8 tx_flags; 494 unsigned short gso_size; 495 /* Warning: this field is not always filled in (UFO)! */ 496 unsigned short gso_segs; 497 struct sk_buff *frag_list; 498 struct skb_shared_hwtstamps hwtstamps; 499 unsigned int gso_type; 500 u32 tskey; 501 __be32 ip6_frag_id; 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 }; 572 573 #if BITS_PER_LONG > 32 574 #define NET_SKBUFF_DATA_USES_OFFSET 1 575 #endif 576 577 #ifdef NET_SKBUFF_DATA_USES_OFFSET 578 typedef unsigned int sk_buff_data_t; 579 #else 580 typedef unsigned char *sk_buff_data_t; 581 #endif 582 583 /** 584 * struct sk_buff - socket buffer 585 * @next: Next buffer in list 586 * @prev: Previous buffer in list 587 * @tstamp: Time we arrived/left 588 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 589 * @sk: Socket we are owned by 590 * @dev: Device we arrived on/are leaving by 591 * @cb: Control buffer. Free for use by every layer. Put private vars here 592 * @_skb_refdst: destination entry (with norefcount bit) 593 * @sp: the security path, used for xfrm 594 * @len: Length of actual data 595 * @data_len: Data length 596 * @mac_len: Length of link layer header 597 * @hdr_len: writable header length of cloned skb 598 * @csum: Checksum (must include start/offset pair) 599 * @csum_start: Offset from skb->head where checksumming should start 600 * @csum_offset: Offset from csum_start where checksum should be stored 601 * @priority: Packet queueing priority 602 * @ignore_df: allow local fragmentation 603 * @cloned: Head may be cloned (check refcnt to be sure) 604 * @ip_summed: Driver fed us an IP checksum 605 * @nohdr: Payload reference only, must not modify header 606 * @pkt_type: Packet class 607 * @fclone: skbuff clone status 608 * @ipvs_property: skbuff is owned by ipvs 609 * @tc_skip_classify: do not classify packet. set by IFB device 610 * @tc_at_ingress: used within tc_classify to distinguish in/egress 611 * @tc_redirected: packet was redirected by a tc action 612 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect 613 * @peeked: this packet has been seen already, so stats have been 614 * done for it, don't do them again 615 * @nf_trace: netfilter packet trace flag 616 * @protocol: Packet protocol from driver 617 * @destructor: Destruct function 618 * @_nfct: Associated connection, if any (with nfctinfo bits) 619 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 620 * @skb_iif: ifindex of device we arrived on 621 * @tc_index: Traffic control index 622 * @hash: the packet hash 623 * @queue_mapping: Queue mapping for multiqueue devices 624 * @xmit_more: More SKBs are pending for this queue 625 * @ndisc_nodetype: router type (from link layer) 626 * @ooo_okay: allow the mapping of a socket to a queue to be changed 627 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 628 * ports. 629 * @sw_hash: indicates hash was computed in software stack 630 * @wifi_acked_valid: wifi_acked was set 631 * @wifi_acked: whether frame was acked on wifi or not 632 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 633 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 634 * @dst_pending_confirm: need to confirm neighbour 635 * @napi_id: id of the NAPI struct this skb came from 636 * @secmark: security marking 637 * @mark: Generic packet mark 638 * @vlan_proto: vlan encapsulation protocol 639 * @vlan_tci: vlan tag control information 640 * @inner_protocol: Protocol (encapsulation) 641 * @inner_transport_header: Inner transport layer header (encapsulation) 642 * @inner_network_header: Network layer header (encapsulation) 643 * @inner_mac_header: Link layer header (encapsulation) 644 * @transport_header: Transport layer header 645 * @network_header: Network layer header 646 * @mac_header: Link layer header 647 * @tail: Tail pointer 648 * @end: End pointer 649 * @head: Head of buffer 650 * @data: Data head pointer 651 * @truesize: Buffer size 652 * @users: User count - see {datagram,tcp}.c 653 */ 654 655 struct sk_buff { 656 union { 657 struct { 658 /* These two members must be first. */ 659 struct sk_buff *next; 660 struct sk_buff *prev; 661 662 union { 663 ktime_t tstamp; 664 u64 skb_mstamp; 665 }; 666 }; 667 struct rb_node rbnode; /* used in netem & tcp stack */ 668 }; 669 struct sock *sk; 670 671 union { 672 struct net_device *dev; 673 /* Some protocols might use this space to store information, 674 * while device pointer would be NULL. 675 * UDP receive path is one user. 676 */ 677 unsigned long dev_scratch; 678 }; 679 /* 680 * This is the control buffer. It is free to use for every 681 * layer. Please put your private variables there. If you 682 * want to keep them across layers you have to do a skb_clone() 683 * first. This is owned by whoever has the skb queued ATM. 684 */ 685 char cb[48] __aligned(8); 686 687 unsigned long _skb_refdst; 688 void (*destructor)(struct sk_buff *skb); 689 #ifdef CONFIG_XFRM 690 struct sec_path *sp; 691 #endif 692 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 693 unsigned long _nfct; 694 #endif 695 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 696 struct nf_bridge_info *nf_bridge; 697 #endif 698 unsigned int len, 699 data_len; 700 __u16 mac_len, 701 hdr_len; 702 703 /* Following fields are _not_ copied in __copy_skb_header() 704 * Note that queue_mapping is here mostly to fill a hole. 705 */ 706 kmemcheck_bitfield_begin(flags1); 707 __u16 queue_mapping; 708 709 /* if you move cloned around you also must adapt those constants */ 710 #ifdef __BIG_ENDIAN_BITFIELD 711 #define CLONED_MASK (1 << 7) 712 #else 713 #define CLONED_MASK 1 714 #endif 715 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset) 716 717 __u8 __cloned_offset[0]; 718 __u8 cloned:1, 719 nohdr:1, 720 fclone:2, 721 peeked:1, 722 head_frag:1, 723 xmit_more:1, 724 __unused:1; /* one bit hole */ 725 kmemcheck_bitfield_end(flags1); 726 727 /* fields enclosed in headers_start/headers_end are copied 728 * using a single memcpy() in __copy_skb_header() 729 */ 730 /* private: */ 731 __u32 headers_start[0]; 732 /* public: */ 733 734 /* if you move pkt_type around you also must adapt those constants */ 735 #ifdef __BIG_ENDIAN_BITFIELD 736 #define PKT_TYPE_MAX (7 << 5) 737 #else 738 #define PKT_TYPE_MAX 7 739 #endif 740 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 741 742 __u8 __pkt_type_offset[0]; 743 __u8 pkt_type:3; 744 __u8 pfmemalloc:1; 745 __u8 ignore_df:1; 746 747 __u8 nf_trace:1; 748 __u8 ip_summed:2; 749 __u8 ooo_okay:1; 750 __u8 l4_hash:1; 751 __u8 sw_hash:1; 752 __u8 wifi_acked_valid:1; 753 __u8 wifi_acked:1; 754 755 __u8 no_fcs:1; 756 /* Indicates the inner headers are valid in the skbuff. */ 757 __u8 encapsulation:1; 758 __u8 encap_hdr_csum:1; 759 __u8 csum_valid:1; 760 __u8 csum_complete_sw:1; 761 __u8 csum_level:2; 762 __u8 csum_not_inet:1; 763 764 __u8 dst_pending_confirm:1; 765 #ifdef CONFIG_IPV6_NDISC_NODETYPE 766 __u8 ndisc_nodetype:2; 767 #endif 768 __u8 ipvs_property:1; 769 __u8 inner_protocol_type:1; 770 __u8 remcsum_offload:1; 771 #ifdef CONFIG_NET_SWITCHDEV 772 __u8 offload_fwd_mark:1; 773 #endif 774 #ifdef CONFIG_NET_CLS_ACT 775 __u8 tc_skip_classify:1; 776 __u8 tc_at_ingress:1; 777 __u8 tc_redirected:1; 778 __u8 tc_from_ingress:1; 779 #endif 780 781 #ifdef CONFIG_NET_SCHED 782 __u16 tc_index; /* traffic control index */ 783 #endif 784 785 union { 786 __wsum csum; 787 struct { 788 __u16 csum_start; 789 __u16 csum_offset; 790 }; 791 }; 792 __u32 priority; 793 int skb_iif; 794 __u32 hash; 795 __be16 vlan_proto; 796 __u16 vlan_tci; 797 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 798 union { 799 unsigned int napi_id; 800 unsigned int sender_cpu; 801 }; 802 #endif 803 #ifdef CONFIG_NETWORK_SECMARK 804 __u32 secmark; 805 #endif 806 807 union { 808 __u32 mark; 809 __u32 reserved_tailroom; 810 }; 811 812 union { 813 __be16 inner_protocol; 814 __u8 inner_ipproto; 815 }; 816 817 __u16 inner_transport_header; 818 __u16 inner_network_header; 819 __u16 inner_mac_header; 820 821 __be16 protocol; 822 __u16 transport_header; 823 __u16 network_header; 824 __u16 mac_header; 825 826 /* private: */ 827 __u32 headers_end[0]; 828 /* public: */ 829 830 /* These elements must be at the end, see alloc_skb() for details. */ 831 sk_buff_data_t tail; 832 sk_buff_data_t end; 833 unsigned char *head, 834 *data; 835 unsigned int truesize; 836 refcount_t users; 837 }; 838 839 #ifdef __KERNEL__ 840 /* 841 * Handling routines are only of interest to the kernel 842 */ 843 #include <linux/slab.h> 844 845 846 #define SKB_ALLOC_FCLONE 0x01 847 #define SKB_ALLOC_RX 0x02 848 #define SKB_ALLOC_NAPI 0x04 849 850 /* Returns true if the skb was allocated from PFMEMALLOC reserves */ 851 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 852 { 853 return unlikely(skb->pfmemalloc); 854 } 855 856 /* 857 * skb might have a dst pointer attached, refcounted or not. 858 * _skb_refdst low order bit is set if refcount was _not_ taken 859 */ 860 #define SKB_DST_NOREF 1UL 861 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 862 863 #define SKB_NFCT_PTRMASK ~(7UL) 864 /** 865 * skb_dst - returns skb dst_entry 866 * @skb: buffer 867 * 868 * Returns skb dst_entry, regardless of reference taken or not. 869 */ 870 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 871 { 872 /* If refdst was not refcounted, check we still are in a 873 * rcu_read_lock section 874 */ 875 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 876 !rcu_read_lock_held() && 877 !rcu_read_lock_bh_held()); 878 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 879 } 880 881 /** 882 * skb_dst_set - sets skb dst 883 * @skb: buffer 884 * @dst: dst entry 885 * 886 * Sets skb dst, assuming a reference was taken on dst and should 887 * be released by skb_dst_drop() 888 */ 889 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 890 { 891 skb->_skb_refdst = (unsigned long)dst; 892 } 893 894 /** 895 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 896 * @skb: buffer 897 * @dst: dst entry 898 * 899 * Sets skb dst, assuming a reference was not taken on dst. 900 * If dst entry is cached, we do not take reference and dst_release 901 * will be avoided by refdst_drop. If dst entry is not cached, we take 902 * reference, so that last dst_release can destroy the dst immediately. 903 */ 904 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 905 { 906 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 907 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 908 } 909 910 /** 911 * skb_dst_is_noref - Test if skb dst isn't refcounted 912 * @skb: buffer 913 */ 914 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 915 { 916 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 917 } 918 919 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 920 { 921 return (struct rtable *)skb_dst(skb); 922 } 923 924 /* For mangling skb->pkt_type from user space side from applications 925 * such as nft, tc, etc, we only allow a conservative subset of 926 * possible pkt_types to be set. 927 */ 928 static inline bool skb_pkt_type_ok(u32 ptype) 929 { 930 return ptype <= PACKET_OTHERHOST; 931 } 932 933 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 934 { 935 #ifdef CONFIG_NET_RX_BUSY_POLL 936 return skb->napi_id; 937 #else 938 return 0; 939 #endif 940 } 941 942 /* decrement the reference count and return true if we can free the skb */ 943 static inline bool skb_unref(struct sk_buff *skb) 944 { 945 if (unlikely(!skb)) 946 return false; 947 if (likely(refcount_read(&skb->users) == 1)) 948 smp_rmb(); 949 else if (likely(!refcount_dec_and_test(&skb->users))) 950 return false; 951 952 return true; 953 } 954 955 void skb_release_head_state(struct sk_buff *skb); 956 void kfree_skb(struct sk_buff *skb); 957 void kfree_skb_list(struct sk_buff *segs); 958 void skb_tx_error(struct sk_buff *skb); 959 void consume_skb(struct sk_buff *skb); 960 void consume_stateless_skb(struct sk_buff *skb); 961 void __kfree_skb(struct sk_buff *skb); 962 extern struct kmem_cache *skbuff_head_cache; 963 964 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 965 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 966 bool *fragstolen, int *delta_truesize); 967 968 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 969 int node); 970 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 971 struct sk_buff *build_skb(void *data, unsigned int frag_size); 972 static inline struct sk_buff *alloc_skb(unsigned int size, 973 gfp_t priority) 974 { 975 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 976 } 977 978 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 979 unsigned long data_len, 980 int max_page_order, 981 int *errcode, 982 gfp_t gfp_mask); 983 984 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 985 struct sk_buff_fclones { 986 struct sk_buff skb1; 987 988 struct sk_buff skb2; 989 990 refcount_t fclone_ref; 991 }; 992 993 /** 994 * skb_fclone_busy - check if fclone is busy 995 * @sk: socket 996 * @skb: buffer 997 * 998 * Returns true if skb is a fast clone, and its clone is not freed. 999 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1000 * so we also check that this didnt happen. 1001 */ 1002 static inline bool skb_fclone_busy(const struct sock *sk, 1003 const struct sk_buff *skb) 1004 { 1005 const struct sk_buff_fclones *fclones; 1006 1007 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1008 1009 return skb->fclone == SKB_FCLONE_ORIG && 1010 refcount_read(&fclones->fclone_ref) > 1 && 1011 fclones->skb2.sk == sk; 1012 } 1013 1014 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1015 gfp_t priority) 1016 { 1017 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1018 } 1019 1020 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1021 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1022 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1023 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1024 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1025 gfp_t gfp_mask, bool fclone); 1026 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1027 gfp_t gfp_mask) 1028 { 1029 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1030 } 1031 1032 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1033 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1034 unsigned int headroom); 1035 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1036 int newtailroom, gfp_t priority); 1037 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1038 int offset, int len); 1039 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1040 int offset, int len); 1041 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1042 int skb_pad(struct sk_buff *skb, int pad); 1043 #define dev_kfree_skb(a) consume_skb(a) 1044 1045 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 1046 int getfrag(void *from, char *to, int offset, 1047 int len, int odd, struct sk_buff *skb), 1048 void *from, int length); 1049 1050 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1051 int offset, size_t size); 1052 1053 struct skb_seq_state { 1054 __u32 lower_offset; 1055 __u32 upper_offset; 1056 __u32 frag_idx; 1057 __u32 stepped_offset; 1058 struct sk_buff *root_skb; 1059 struct sk_buff *cur_skb; 1060 __u8 *frag_data; 1061 }; 1062 1063 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1064 unsigned int to, struct skb_seq_state *st); 1065 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1066 struct skb_seq_state *st); 1067 void skb_abort_seq_read(struct skb_seq_state *st); 1068 1069 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1070 unsigned int to, struct ts_config *config); 1071 1072 /* 1073 * Packet hash types specify the type of hash in skb_set_hash. 1074 * 1075 * Hash types refer to the protocol layer addresses which are used to 1076 * construct a packet's hash. The hashes are used to differentiate or identify 1077 * flows of the protocol layer for the hash type. Hash types are either 1078 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1079 * 1080 * Properties of hashes: 1081 * 1082 * 1) Two packets in different flows have different hash values 1083 * 2) Two packets in the same flow should have the same hash value 1084 * 1085 * A hash at a higher layer is considered to be more specific. A driver should 1086 * set the most specific hash possible. 1087 * 1088 * A driver cannot indicate a more specific hash than the layer at which a hash 1089 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1090 * 1091 * A driver may indicate a hash level which is less specific than the 1092 * actual layer the hash was computed on. For instance, a hash computed 1093 * at L4 may be considered an L3 hash. This should only be done if the 1094 * driver can't unambiguously determine that the HW computed the hash at 1095 * the higher layer. Note that the "should" in the second property above 1096 * permits this. 1097 */ 1098 enum pkt_hash_types { 1099 PKT_HASH_TYPE_NONE, /* Undefined type */ 1100 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1101 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1102 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1103 }; 1104 1105 static inline void skb_clear_hash(struct sk_buff *skb) 1106 { 1107 skb->hash = 0; 1108 skb->sw_hash = 0; 1109 skb->l4_hash = 0; 1110 } 1111 1112 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1113 { 1114 if (!skb->l4_hash) 1115 skb_clear_hash(skb); 1116 } 1117 1118 static inline void 1119 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1120 { 1121 skb->l4_hash = is_l4; 1122 skb->sw_hash = is_sw; 1123 skb->hash = hash; 1124 } 1125 1126 static inline void 1127 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1128 { 1129 /* Used by drivers to set hash from HW */ 1130 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1131 } 1132 1133 static inline void 1134 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1135 { 1136 __skb_set_hash(skb, hash, true, is_l4); 1137 } 1138 1139 void __skb_get_hash(struct sk_buff *skb); 1140 u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1141 u32 skb_get_poff(const struct sk_buff *skb); 1142 u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1143 const struct flow_keys *keys, int hlen); 1144 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1145 void *data, int hlen_proto); 1146 1147 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1148 int thoff, u8 ip_proto) 1149 { 1150 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1151 } 1152 1153 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1154 const struct flow_dissector_key *key, 1155 unsigned int key_count); 1156 1157 bool __skb_flow_dissect(const struct sk_buff *skb, 1158 struct flow_dissector *flow_dissector, 1159 void *target_container, 1160 void *data, __be16 proto, int nhoff, int hlen, 1161 unsigned int flags); 1162 1163 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1164 struct flow_dissector *flow_dissector, 1165 void *target_container, unsigned int flags) 1166 { 1167 return __skb_flow_dissect(skb, flow_dissector, target_container, 1168 NULL, 0, 0, 0, flags); 1169 } 1170 1171 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1172 struct flow_keys *flow, 1173 unsigned int flags) 1174 { 1175 memset(flow, 0, sizeof(*flow)); 1176 return __skb_flow_dissect(skb, &flow_keys_dissector, flow, 1177 NULL, 0, 0, 0, flags); 1178 } 1179 1180 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow, 1181 void *data, __be16 proto, 1182 int nhoff, int hlen, 1183 unsigned int flags) 1184 { 1185 memset(flow, 0, sizeof(*flow)); 1186 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow, 1187 data, proto, nhoff, hlen, flags); 1188 } 1189 1190 static inline __u32 skb_get_hash(struct sk_buff *skb) 1191 { 1192 if (!skb->l4_hash && !skb->sw_hash) 1193 __skb_get_hash(skb); 1194 1195 return skb->hash; 1196 } 1197 1198 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1199 { 1200 if (!skb->l4_hash && !skb->sw_hash) { 1201 struct flow_keys keys; 1202 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1203 1204 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1205 } 1206 1207 return skb->hash; 1208 } 1209 1210 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1211 1212 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1213 { 1214 return skb->hash; 1215 } 1216 1217 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1218 { 1219 to->hash = from->hash; 1220 to->sw_hash = from->sw_hash; 1221 to->l4_hash = from->l4_hash; 1222 }; 1223 1224 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1225 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1226 { 1227 return skb->head + skb->end; 1228 } 1229 1230 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1231 { 1232 return skb->end; 1233 } 1234 #else 1235 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1236 { 1237 return skb->end; 1238 } 1239 1240 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1241 { 1242 return skb->end - skb->head; 1243 } 1244 #endif 1245 1246 /* Internal */ 1247 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1248 1249 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1250 { 1251 return &skb_shinfo(skb)->hwtstamps; 1252 } 1253 1254 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1255 { 1256 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY; 1257 1258 return is_zcopy ? skb_uarg(skb) : NULL; 1259 } 1260 1261 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg) 1262 { 1263 if (skb && uarg && !skb_zcopy(skb)) { 1264 sock_zerocopy_get(uarg); 1265 skb_shinfo(skb)->destructor_arg = uarg; 1266 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG; 1267 } 1268 } 1269 1270 /* Release a reference on a zerocopy structure */ 1271 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy) 1272 { 1273 struct ubuf_info *uarg = skb_zcopy(skb); 1274 1275 if (uarg) { 1276 if (uarg->callback == sock_zerocopy_callback) { 1277 uarg->zerocopy = uarg->zerocopy && zerocopy; 1278 sock_zerocopy_put(uarg); 1279 } else { 1280 uarg->callback(uarg, zerocopy); 1281 } 1282 1283 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1284 } 1285 } 1286 1287 /* Abort a zerocopy operation and revert zckey on error in send syscall */ 1288 static inline void skb_zcopy_abort(struct sk_buff *skb) 1289 { 1290 struct ubuf_info *uarg = skb_zcopy(skb); 1291 1292 if (uarg) { 1293 sock_zerocopy_put_abort(uarg); 1294 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1295 } 1296 } 1297 1298 /** 1299 * skb_queue_empty - check if a queue is empty 1300 * @list: queue head 1301 * 1302 * Returns true if the queue is empty, false otherwise. 1303 */ 1304 static inline int skb_queue_empty(const struct sk_buff_head *list) 1305 { 1306 return list->next == (const struct sk_buff *) list; 1307 } 1308 1309 /** 1310 * skb_queue_is_last - check if skb is the last entry in the queue 1311 * @list: queue head 1312 * @skb: buffer 1313 * 1314 * Returns true if @skb is the last buffer on the list. 1315 */ 1316 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1317 const struct sk_buff *skb) 1318 { 1319 return skb->next == (const struct sk_buff *) list; 1320 } 1321 1322 /** 1323 * skb_queue_is_first - check if skb is the first entry in the queue 1324 * @list: queue head 1325 * @skb: buffer 1326 * 1327 * Returns true if @skb is the first buffer on the list. 1328 */ 1329 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1330 const struct sk_buff *skb) 1331 { 1332 return skb->prev == (const struct sk_buff *) list; 1333 } 1334 1335 /** 1336 * skb_queue_next - return the next packet in the queue 1337 * @list: queue head 1338 * @skb: current buffer 1339 * 1340 * Return the next packet in @list after @skb. It is only valid to 1341 * call this if skb_queue_is_last() evaluates to false. 1342 */ 1343 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1344 const struct sk_buff *skb) 1345 { 1346 /* This BUG_ON may seem severe, but if we just return then we 1347 * are going to dereference garbage. 1348 */ 1349 BUG_ON(skb_queue_is_last(list, skb)); 1350 return skb->next; 1351 } 1352 1353 /** 1354 * skb_queue_prev - return the prev packet in the queue 1355 * @list: queue head 1356 * @skb: current buffer 1357 * 1358 * Return the prev packet in @list before @skb. It is only valid to 1359 * call this if skb_queue_is_first() evaluates to false. 1360 */ 1361 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1362 const struct sk_buff *skb) 1363 { 1364 /* This BUG_ON may seem severe, but if we just return then we 1365 * are going to dereference garbage. 1366 */ 1367 BUG_ON(skb_queue_is_first(list, skb)); 1368 return skb->prev; 1369 } 1370 1371 /** 1372 * skb_get - reference buffer 1373 * @skb: buffer to reference 1374 * 1375 * Makes another reference to a socket buffer and returns a pointer 1376 * to the buffer. 1377 */ 1378 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1379 { 1380 refcount_inc(&skb->users); 1381 return skb; 1382 } 1383 1384 /* 1385 * If users == 1, we are the only owner and are can avoid redundant 1386 * atomic change. 1387 */ 1388 1389 /** 1390 * skb_cloned - is the buffer a clone 1391 * @skb: buffer to check 1392 * 1393 * Returns true if the buffer was generated with skb_clone() and is 1394 * one of multiple shared copies of the buffer. Cloned buffers are 1395 * shared data so must not be written to under normal circumstances. 1396 */ 1397 static inline int skb_cloned(const struct sk_buff *skb) 1398 { 1399 return skb->cloned && 1400 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1401 } 1402 1403 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1404 { 1405 might_sleep_if(gfpflags_allow_blocking(pri)); 1406 1407 if (skb_cloned(skb)) 1408 return pskb_expand_head(skb, 0, 0, pri); 1409 1410 return 0; 1411 } 1412 1413 /** 1414 * skb_header_cloned - is the header a clone 1415 * @skb: buffer to check 1416 * 1417 * Returns true if modifying the header part of the buffer requires 1418 * the data to be copied. 1419 */ 1420 static inline int skb_header_cloned(const struct sk_buff *skb) 1421 { 1422 int dataref; 1423 1424 if (!skb->cloned) 1425 return 0; 1426 1427 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1428 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1429 return dataref != 1; 1430 } 1431 1432 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1433 { 1434 might_sleep_if(gfpflags_allow_blocking(pri)); 1435 1436 if (skb_header_cloned(skb)) 1437 return pskb_expand_head(skb, 0, 0, pri); 1438 1439 return 0; 1440 } 1441 1442 /** 1443 * skb_header_release - release reference to header 1444 * @skb: buffer to operate on 1445 * 1446 * Drop a reference to the header part of the buffer. This is done 1447 * by acquiring a payload reference. You must not read from the header 1448 * part of skb->data after this. 1449 * Note : Check if you can use __skb_header_release() instead. 1450 */ 1451 static inline void skb_header_release(struct sk_buff *skb) 1452 { 1453 BUG_ON(skb->nohdr); 1454 skb->nohdr = 1; 1455 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1456 } 1457 1458 /** 1459 * __skb_header_release - release reference to header 1460 * @skb: buffer to operate on 1461 * 1462 * Variant of skb_header_release() assuming skb is private to caller. 1463 * We can avoid one atomic operation. 1464 */ 1465 static inline void __skb_header_release(struct sk_buff *skb) 1466 { 1467 skb->nohdr = 1; 1468 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1469 } 1470 1471 1472 /** 1473 * skb_shared - is the buffer shared 1474 * @skb: buffer to check 1475 * 1476 * Returns true if more than one person has a reference to this 1477 * buffer. 1478 */ 1479 static inline int skb_shared(const struct sk_buff *skb) 1480 { 1481 return refcount_read(&skb->users) != 1; 1482 } 1483 1484 /** 1485 * skb_share_check - check if buffer is shared and if so clone it 1486 * @skb: buffer to check 1487 * @pri: priority for memory allocation 1488 * 1489 * If the buffer is shared the buffer is cloned and the old copy 1490 * drops a reference. A new clone with a single reference is returned. 1491 * If the buffer is not shared the original buffer is returned. When 1492 * being called from interrupt status or with spinlocks held pri must 1493 * be GFP_ATOMIC. 1494 * 1495 * NULL is returned on a memory allocation failure. 1496 */ 1497 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1498 { 1499 might_sleep_if(gfpflags_allow_blocking(pri)); 1500 if (skb_shared(skb)) { 1501 struct sk_buff *nskb = skb_clone(skb, pri); 1502 1503 if (likely(nskb)) 1504 consume_skb(skb); 1505 else 1506 kfree_skb(skb); 1507 skb = nskb; 1508 } 1509 return skb; 1510 } 1511 1512 /* 1513 * Copy shared buffers into a new sk_buff. We effectively do COW on 1514 * packets to handle cases where we have a local reader and forward 1515 * and a couple of other messy ones. The normal one is tcpdumping 1516 * a packet thats being forwarded. 1517 */ 1518 1519 /** 1520 * skb_unshare - make a copy of a shared buffer 1521 * @skb: buffer to check 1522 * @pri: priority for memory allocation 1523 * 1524 * If the socket buffer is a clone then this function creates a new 1525 * copy of the data, drops a reference count on the old copy and returns 1526 * the new copy with the reference count at 1. If the buffer is not a clone 1527 * the original buffer is returned. When called with a spinlock held or 1528 * from interrupt state @pri must be %GFP_ATOMIC 1529 * 1530 * %NULL is returned on a memory allocation failure. 1531 */ 1532 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1533 gfp_t pri) 1534 { 1535 might_sleep_if(gfpflags_allow_blocking(pri)); 1536 if (skb_cloned(skb)) { 1537 struct sk_buff *nskb = skb_copy(skb, pri); 1538 1539 /* Free our shared copy */ 1540 if (likely(nskb)) 1541 consume_skb(skb); 1542 else 1543 kfree_skb(skb); 1544 skb = nskb; 1545 } 1546 return skb; 1547 } 1548 1549 /** 1550 * skb_peek - peek at the head of an &sk_buff_head 1551 * @list_: list to peek at 1552 * 1553 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1554 * be careful with this one. A peek leaves the buffer on the 1555 * list and someone else may run off with it. You must hold 1556 * the appropriate locks or have a private queue to do this. 1557 * 1558 * Returns %NULL for an empty list or a pointer to the head element. 1559 * The reference count is not incremented and the reference is therefore 1560 * volatile. Use with caution. 1561 */ 1562 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1563 { 1564 struct sk_buff *skb = list_->next; 1565 1566 if (skb == (struct sk_buff *)list_) 1567 skb = NULL; 1568 return skb; 1569 } 1570 1571 /** 1572 * skb_peek_next - peek skb following the given one from a queue 1573 * @skb: skb to start from 1574 * @list_: list to peek at 1575 * 1576 * Returns %NULL when the end of the list is met or a pointer to the 1577 * next element. The reference count is not incremented and the 1578 * reference is therefore volatile. Use with caution. 1579 */ 1580 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1581 const struct sk_buff_head *list_) 1582 { 1583 struct sk_buff *next = skb->next; 1584 1585 if (next == (struct sk_buff *)list_) 1586 next = NULL; 1587 return next; 1588 } 1589 1590 /** 1591 * skb_peek_tail - peek at the tail of an &sk_buff_head 1592 * @list_: list to peek at 1593 * 1594 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1595 * be careful with this one. A peek leaves the buffer on the 1596 * list and someone else may run off with it. You must hold 1597 * the appropriate locks or have a private queue to do this. 1598 * 1599 * Returns %NULL for an empty list or a pointer to the tail element. 1600 * The reference count is not incremented and the reference is therefore 1601 * volatile. Use with caution. 1602 */ 1603 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1604 { 1605 struct sk_buff *skb = list_->prev; 1606 1607 if (skb == (struct sk_buff *)list_) 1608 skb = NULL; 1609 return skb; 1610 1611 } 1612 1613 /** 1614 * skb_queue_len - get queue length 1615 * @list_: list to measure 1616 * 1617 * Return the length of an &sk_buff queue. 1618 */ 1619 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1620 { 1621 return list_->qlen; 1622 } 1623 1624 /** 1625 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1626 * @list: queue to initialize 1627 * 1628 * This initializes only the list and queue length aspects of 1629 * an sk_buff_head object. This allows to initialize the list 1630 * aspects of an sk_buff_head without reinitializing things like 1631 * the spinlock. It can also be used for on-stack sk_buff_head 1632 * objects where the spinlock is known to not be used. 1633 */ 1634 static inline void __skb_queue_head_init(struct sk_buff_head *list) 1635 { 1636 list->prev = list->next = (struct sk_buff *)list; 1637 list->qlen = 0; 1638 } 1639 1640 /* 1641 * This function creates a split out lock class for each invocation; 1642 * this is needed for now since a whole lot of users of the skb-queue 1643 * infrastructure in drivers have different locking usage (in hardirq) 1644 * than the networking core (in softirq only). In the long run either the 1645 * network layer or drivers should need annotation to consolidate the 1646 * main types of usage into 3 classes. 1647 */ 1648 static inline void skb_queue_head_init(struct sk_buff_head *list) 1649 { 1650 spin_lock_init(&list->lock); 1651 __skb_queue_head_init(list); 1652 } 1653 1654 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1655 struct lock_class_key *class) 1656 { 1657 skb_queue_head_init(list); 1658 lockdep_set_class(&list->lock, class); 1659 } 1660 1661 /* 1662 * Insert an sk_buff on a list. 1663 * 1664 * The "__skb_xxxx()" functions are the non-atomic ones that 1665 * can only be called with interrupts disabled. 1666 */ 1667 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1668 struct sk_buff_head *list); 1669 static inline void __skb_insert(struct sk_buff *newsk, 1670 struct sk_buff *prev, struct sk_buff *next, 1671 struct sk_buff_head *list) 1672 { 1673 newsk->next = next; 1674 newsk->prev = prev; 1675 next->prev = prev->next = newsk; 1676 list->qlen++; 1677 } 1678 1679 static inline void __skb_queue_splice(const struct sk_buff_head *list, 1680 struct sk_buff *prev, 1681 struct sk_buff *next) 1682 { 1683 struct sk_buff *first = list->next; 1684 struct sk_buff *last = list->prev; 1685 1686 first->prev = prev; 1687 prev->next = first; 1688 1689 last->next = next; 1690 next->prev = last; 1691 } 1692 1693 /** 1694 * skb_queue_splice - join two skb lists, this is designed for stacks 1695 * @list: the new list to add 1696 * @head: the place to add it in the first list 1697 */ 1698 static inline void skb_queue_splice(const struct sk_buff_head *list, 1699 struct sk_buff_head *head) 1700 { 1701 if (!skb_queue_empty(list)) { 1702 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1703 head->qlen += list->qlen; 1704 } 1705 } 1706 1707 /** 1708 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1709 * @list: the new list to add 1710 * @head: the place to add it in the first list 1711 * 1712 * The list at @list is reinitialised 1713 */ 1714 static inline void skb_queue_splice_init(struct sk_buff_head *list, 1715 struct sk_buff_head *head) 1716 { 1717 if (!skb_queue_empty(list)) { 1718 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1719 head->qlen += list->qlen; 1720 __skb_queue_head_init(list); 1721 } 1722 } 1723 1724 /** 1725 * skb_queue_splice_tail - join two skb lists, each list being a queue 1726 * @list: the new list to add 1727 * @head: the place to add it in the first list 1728 */ 1729 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1730 struct sk_buff_head *head) 1731 { 1732 if (!skb_queue_empty(list)) { 1733 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1734 head->qlen += list->qlen; 1735 } 1736 } 1737 1738 /** 1739 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1740 * @list: the new list to add 1741 * @head: the place to add it in the first list 1742 * 1743 * Each of the lists is a queue. 1744 * The list at @list is reinitialised 1745 */ 1746 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1747 struct sk_buff_head *head) 1748 { 1749 if (!skb_queue_empty(list)) { 1750 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1751 head->qlen += list->qlen; 1752 __skb_queue_head_init(list); 1753 } 1754 } 1755 1756 /** 1757 * __skb_queue_after - queue a buffer at the list head 1758 * @list: list to use 1759 * @prev: place after this buffer 1760 * @newsk: buffer to queue 1761 * 1762 * Queue a buffer int the middle of a list. This function takes no locks 1763 * and you must therefore hold required locks before calling it. 1764 * 1765 * A buffer cannot be placed on two lists at the same time. 1766 */ 1767 static inline void __skb_queue_after(struct sk_buff_head *list, 1768 struct sk_buff *prev, 1769 struct sk_buff *newsk) 1770 { 1771 __skb_insert(newsk, prev, prev->next, list); 1772 } 1773 1774 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1775 struct sk_buff_head *list); 1776 1777 static inline void __skb_queue_before(struct sk_buff_head *list, 1778 struct sk_buff *next, 1779 struct sk_buff *newsk) 1780 { 1781 __skb_insert(newsk, next->prev, next, list); 1782 } 1783 1784 /** 1785 * __skb_queue_head - queue a buffer at the list head 1786 * @list: list to use 1787 * @newsk: buffer to queue 1788 * 1789 * Queue a buffer at the start of a list. This function takes no locks 1790 * and you must therefore hold required locks before calling it. 1791 * 1792 * A buffer cannot be placed on two lists at the same time. 1793 */ 1794 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1795 static inline void __skb_queue_head(struct sk_buff_head *list, 1796 struct sk_buff *newsk) 1797 { 1798 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1799 } 1800 1801 /** 1802 * __skb_queue_tail - queue a buffer at the list tail 1803 * @list: list to use 1804 * @newsk: buffer to queue 1805 * 1806 * Queue a buffer at the end of a list. This function takes no locks 1807 * and you must therefore hold required locks before calling it. 1808 * 1809 * A buffer cannot be placed on two lists at the same time. 1810 */ 1811 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1812 static inline void __skb_queue_tail(struct sk_buff_head *list, 1813 struct sk_buff *newsk) 1814 { 1815 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1816 } 1817 1818 /* 1819 * remove sk_buff from list. _Must_ be called atomically, and with 1820 * the list known.. 1821 */ 1822 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1823 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1824 { 1825 struct sk_buff *next, *prev; 1826 1827 list->qlen--; 1828 next = skb->next; 1829 prev = skb->prev; 1830 skb->next = skb->prev = NULL; 1831 next->prev = prev; 1832 prev->next = next; 1833 } 1834 1835 /** 1836 * __skb_dequeue - remove from the head of the queue 1837 * @list: list to dequeue from 1838 * 1839 * Remove the head of the list. This function does not take any locks 1840 * so must be used with appropriate locks held only. The head item is 1841 * returned or %NULL if the list is empty. 1842 */ 1843 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1844 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1845 { 1846 struct sk_buff *skb = skb_peek(list); 1847 if (skb) 1848 __skb_unlink(skb, list); 1849 return skb; 1850 } 1851 1852 /** 1853 * __skb_dequeue_tail - remove from the tail of the queue 1854 * @list: list to dequeue from 1855 * 1856 * Remove the tail of the list. This function does not take any locks 1857 * so must be used with appropriate locks held only. The tail item is 1858 * returned or %NULL if the list is empty. 1859 */ 1860 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1861 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1862 { 1863 struct sk_buff *skb = skb_peek_tail(list); 1864 if (skb) 1865 __skb_unlink(skb, list); 1866 return skb; 1867 } 1868 1869 1870 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1871 { 1872 return skb->data_len; 1873 } 1874 1875 static inline unsigned int skb_headlen(const struct sk_buff *skb) 1876 { 1877 return skb->len - skb->data_len; 1878 } 1879 1880 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 1881 { 1882 unsigned int i, len = 0; 1883 1884 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 1885 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1886 return len; 1887 } 1888 1889 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 1890 { 1891 return skb_headlen(skb) + __skb_pagelen(skb); 1892 } 1893 1894 /** 1895 * __skb_fill_page_desc - initialise a paged fragment in an skb 1896 * @skb: buffer containing fragment to be initialised 1897 * @i: paged fragment index to initialise 1898 * @page: the page to use for this fragment 1899 * @off: the offset to the data with @page 1900 * @size: the length of the data 1901 * 1902 * Initialises the @i'th fragment of @skb to point to &size bytes at 1903 * offset @off within @page. 1904 * 1905 * Does not take any additional reference on the fragment. 1906 */ 1907 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1908 struct page *page, int off, int size) 1909 { 1910 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1911 1912 /* 1913 * Propagate page pfmemalloc to the skb if we can. The problem is 1914 * that not all callers have unique ownership of the page but rely 1915 * on page_is_pfmemalloc doing the right thing(tm). 1916 */ 1917 frag->page.p = page; 1918 frag->page_offset = off; 1919 skb_frag_size_set(frag, size); 1920 1921 page = compound_head(page); 1922 if (page_is_pfmemalloc(page)) 1923 skb->pfmemalloc = true; 1924 } 1925 1926 /** 1927 * skb_fill_page_desc - initialise a paged fragment in an skb 1928 * @skb: buffer containing fragment to be initialised 1929 * @i: paged fragment index to initialise 1930 * @page: the page to use for this fragment 1931 * @off: the offset to the data with @page 1932 * @size: the length of the data 1933 * 1934 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1935 * @skb to point to @size bytes at offset @off within @page. In 1936 * addition updates @skb such that @i is the last fragment. 1937 * 1938 * Does not take any additional reference on the fragment. 1939 */ 1940 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1941 struct page *page, int off, int size) 1942 { 1943 __skb_fill_page_desc(skb, i, page, off, size); 1944 skb_shinfo(skb)->nr_frags = i + 1; 1945 } 1946 1947 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1948 int size, unsigned int truesize); 1949 1950 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1951 unsigned int truesize); 1952 1953 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1954 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1955 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1956 1957 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1958 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1959 { 1960 return skb->head + skb->tail; 1961 } 1962 1963 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1964 { 1965 skb->tail = skb->data - skb->head; 1966 } 1967 1968 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1969 { 1970 skb_reset_tail_pointer(skb); 1971 skb->tail += offset; 1972 } 1973 1974 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 1975 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1976 { 1977 return skb->tail; 1978 } 1979 1980 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1981 { 1982 skb->tail = skb->data; 1983 } 1984 1985 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1986 { 1987 skb->tail = skb->data + offset; 1988 } 1989 1990 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1991 1992 /* 1993 * Add data to an sk_buff 1994 */ 1995 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1996 void *skb_put(struct sk_buff *skb, unsigned int len); 1997 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 1998 { 1999 void *tmp = skb_tail_pointer(skb); 2000 SKB_LINEAR_ASSERT(skb); 2001 skb->tail += len; 2002 skb->len += len; 2003 return tmp; 2004 } 2005 2006 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2007 { 2008 void *tmp = __skb_put(skb, len); 2009 2010 memset(tmp, 0, len); 2011 return tmp; 2012 } 2013 2014 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2015 unsigned int len) 2016 { 2017 void *tmp = __skb_put(skb, len); 2018 2019 memcpy(tmp, data, len); 2020 return tmp; 2021 } 2022 2023 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2024 { 2025 *(u8 *)__skb_put(skb, 1) = val; 2026 } 2027 2028 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2029 { 2030 void *tmp = skb_put(skb, len); 2031 2032 memset(tmp, 0, len); 2033 2034 return tmp; 2035 } 2036 2037 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2038 unsigned int len) 2039 { 2040 void *tmp = skb_put(skb, len); 2041 2042 memcpy(tmp, data, len); 2043 2044 return tmp; 2045 } 2046 2047 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2048 { 2049 *(u8 *)skb_put(skb, 1) = val; 2050 } 2051 2052 void *skb_push(struct sk_buff *skb, unsigned int len); 2053 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2054 { 2055 skb->data -= len; 2056 skb->len += len; 2057 return skb->data; 2058 } 2059 2060 void *skb_pull(struct sk_buff *skb, unsigned int len); 2061 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2062 { 2063 skb->len -= len; 2064 BUG_ON(skb->len < skb->data_len); 2065 return skb->data += len; 2066 } 2067 2068 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2069 { 2070 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2071 } 2072 2073 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2074 2075 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len) 2076 { 2077 if (len > skb_headlen(skb) && 2078 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 2079 return NULL; 2080 skb->len -= len; 2081 return skb->data += len; 2082 } 2083 2084 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2085 { 2086 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 2087 } 2088 2089 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 2090 { 2091 if (likely(len <= skb_headlen(skb))) 2092 return 1; 2093 if (unlikely(len > skb->len)) 2094 return 0; 2095 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 2096 } 2097 2098 void skb_condense(struct sk_buff *skb); 2099 2100 /** 2101 * skb_headroom - bytes at buffer head 2102 * @skb: buffer to check 2103 * 2104 * Return the number of bytes of free space at the head of an &sk_buff. 2105 */ 2106 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2107 { 2108 return skb->data - skb->head; 2109 } 2110 2111 /** 2112 * skb_tailroom - bytes at buffer end 2113 * @skb: buffer to check 2114 * 2115 * Return the number of bytes of free space at the tail of an sk_buff 2116 */ 2117 static inline int skb_tailroom(const struct sk_buff *skb) 2118 { 2119 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2120 } 2121 2122 /** 2123 * skb_availroom - bytes at buffer end 2124 * @skb: buffer to check 2125 * 2126 * Return the number of bytes of free space at the tail of an sk_buff 2127 * allocated by sk_stream_alloc() 2128 */ 2129 static inline int skb_availroom(const struct sk_buff *skb) 2130 { 2131 if (skb_is_nonlinear(skb)) 2132 return 0; 2133 2134 return skb->end - skb->tail - skb->reserved_tailroom; 2135 } 2136 2137 /** 2138 * skb_reserve - adjust headroom 2139 * @skb: buffer to alter 2140 * @len: bytes to move 2141 * 2142 * Increase the headroom of an empty &sk_buff by reducing the tail 2143 * room. This is only allowed for an empty buffer. 2144 */ 2145 static inline void skb_reserve(struct sk_buff *skb, int len) 2146 { 2147 skb->data += len; 2148 skb->tail += len; 2149 } 2150 2151 /** 2152 * skb_tailroom_reserve - adjust reserved_tailroom 2153 * @skb: buffer to alter 2154 * @mtu: maximum amount of headlen permitted 2155 * @needed_tailroom: minimum amount of reserved_tailroom 2156 * 2157 * Set reserved_tailroom so that headlen can be as large as possible but 2158 * not larger than mtu and tailroom cannot be smaller than 2159 * needed_tailroom. 2160 * The required headroom should already have been reserved before using 2161 * this function. 2162 */ 2163 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2164 unsigned int needed_tailroom) 2165 { 2166 SKB_LINEAR_ASSERT(skb); 2167 if (mtu < skb_tailroom(skb) - needed_tailroom) 2168 /* use at most mtu */ 2169 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2170 else 2171 /* use up to all available space */ 2172 skb->reserved_tailroom = needed_tailroom; 2173 } 2174 2175 #define ENCAP_TYPE_ETHER 0 2176 #define ENCAP_TYPE_IPPROTO 1 2177 2178 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2179 __be16 protocol) 2180 { 2181 skb->inner_protocol = protocol; 2182 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2183 } 2184 2185 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2186 __u8 ipproto) 2187 { 2188 skb->inner_ipproto = ipproto; 2189 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2190 } 2191 2192 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2193 { 2194 skb->inner_mac_header = skb->mac_header; 2195 skb->inner_network_header = skb->network_header; 2196 skb->inner_transport_header = skb->transport_header; 2197 } 2198 2199 static inline void skb_reset_mac_len(struct sk_buff *skb) 2200 { 2201 skb->mac_len = skb->network_header - skb->mac_header; 2202 } 2203 2204 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2205 *skb) 2206 { 2207 return skb->head + skb->inner_transport_header; 2208 } 2209 2210 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2211 { 2212 return skb_inner_transport_header(skb) - skb->data; 2213 } 2214 2215 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2216 { 2217 skb->inner_transport_header = skb->data - skb->head; 2218 } 2219 2220 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2221 const int offset) 2222 { 2223 skb_reset_inner_transport_header(skb); 2224 skb->inner_transport_header += offset; 2225 } 2226 2227 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2228 { 2229 return skb->head + skb->inner_network_header; 2230 } 2231 2232 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2233 { 2234 skb->inner_network_header = skb->data - skb->head; 2235 } 2236 2237 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2238 const int offset) 2239 { 2240 skb_reset_inner_network_header(skb); 2241 skb->inner_network_header += offset; 2242 } 2243 2244 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2245 { 2246 return skb->head + skb->inner_mac_header; 2247 } 2248 2249 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2250 { 2251 skb->inner_mac_header = skb->data - skb->head; 2252 } 2253 2254 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2255 const int offset) 2256 { 2257 skb_reset_inner_mac_header(skb); 2258 skb->inner_mac_header += offset; 2259 } 2260 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2261 { 2262 return skb->transport_header != (typeof(skb->transport_header))~0U; 2263 } 2264 2265 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2266 { 2267 return skb->head + skb->transport_header; 2268 } 2269 2270 static inline void skb_reset_transport_header(struct sk_buff *skb) 2271 { 2272 skb->transport_header = skb->data - skb->head; 2273 } 2274 2275 static inline void skb_set_transport_header(struct sk_buff *skb, 2276 const int offset) 2277 { 2278 skb_reset_transport_header(skb); 2279 skb->transport_header += offset; 2280 } 2281 2282 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2283 { 2284 return skb->head + skb->network_header; 2285 } 2286 2287 static inline void skb_reset_network_header(struct sk_buff *skb) 2288 { 2289 skb->network_header = skb->data - skb->head; 2290 } 2291 2292 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2293 { 2294 skb_reset_network_header(skb); 2295 skb->network_header += offset; 2296 } 2297 2298 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2299 { 2300 return skb->head + skb->mac_header; 2301 } 2302 2303 static inline int skb_mac_offset(const struct sk_buff *skb) 2304 { 2305 return skb_mac_header(skb) - skb->data; 2306 } 2307 2308 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2309 { 2310 return skb->network_header - skb->mac_header; 2311 } 2312 2313 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2314 { 2315 return skb->mac_header != (typeof(skb->mac_header))~0U; 2316 } 2317 2318 static inline void skb_reset_mac_header(struct sk_buff *skb) 2319 { 2320 skb->mac_header = skb->data - skb->head; 2321 } 2322 2323 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2324 { 2325 skb_reset_mac_header(skb); 2326 skb->mac_header += offset; 2327 } 2328 2329 static inline void skb_pop_mac_header(struct sk_buff *skb) 2330 { 2331 skb->mac_header = skb->network_header; 2332 } 2333 2334 static inline void skb_probe_transport_header(struct sk_buff *skb, 2335 const int offset_hint) 2336 { 2337 struct flow_keys keys; 2338 2339 if (skb_transport_header_was_set(skb)) 2340 return; 2341 else if (skb_flow_dissect_flow_keys(skb, &keys, 0)) 2342 skb_set_transport_header(skb, keys.control.thoff); 2343 else 2344 skb_set_transport_header(skb, offset_hint); 2345 } 2346 2347 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2348 { 2349 if (skb_mac_header_was_set(skb)) { 2350 const unsigned char *old_mac = skb_mac_header(skb); 2351 2352 skb_set_mac_header(skb, -skb->mac_len); 2353 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2354 } 2355 } 2356 2357 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2358 { 2359 return skb->csum_start - skb_headroom(skb); 2360 } 2361 2362 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2363 { 2364 return skb->head + skb->csum_start; 2365 } 2366 2367 static inline int skb_transport_offset(const struct sk_buff *skb) 2368 { 2369 return skb_transport_header(skb) - skb->data; 2370 } 2371 2372 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2373 { 2374 return skb->transport_header - skb->network_header; 2375 } 2376 2377 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2378 { 2379 return skb->inner_transport_header - skb->inner_network_header; 2380 } 2381 2382 static inline int skb_network_offset(const struct sk_buff *skb) 2383 { 2384 return skb_network_header(skb) - skb->data; 2385 } 2386 2387 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2388 { 2389 return skb_inner_network_header(skb) - skb->data; 2390 } 2391 2392 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2393 { 2394 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2395 } 2396 2397 /* 2398 * CPUs often take a performance hit when accessing unaligned memory 2399 * locations. The actual performance hit varies, it can be small if the 2400 * hardware handles it or large if we have to take an exception and fix it 2401 * in software. 2402 * 2403 * Since an ethernet header is 14 bytes network drivers often end up with 2404 * the IP header at an unaligned offset. The IP header can be aligned by 2405 * shifting the start of the packet by 2 bytes. Drivers should do this 2406 * with: 2407 * 2408 * skb_reserve(skb, NET_IP_ALIGN); 2409 * 2410 * The downside to this alignment of the IP header is that the DMA is now 2411 * unaligned. On some architectures the cost of an unaligned DMA is high 2412 * and this cost outweighs the gains made by aligning the IP header. 2413 * 2414 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2415 * to be overridden. 2416 */ 2417 #ifndef NET_IP_ALIGN 2418 #define NET_IP_ALIGN 2 2419 #endif 2420 2421 /* 2422 * The networking layer reserves some headroom in skb data (via 2423 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2424 * the header has to grow. In the default case, if the header has to grow 2425 * 32 bytes or less we avoid the reallocation. 2426 * 2427 * Unfortunately this headroom changes the DMA alignment of the resulting 2428 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2429 * on some architectures. An architecture can override this value, 2430 * perhaps setting it to a cacheline in size (since that will maintain 2431 * cacheline alignment of the DMA). It must be a power of 2. 2432 * 2433 * Various parts of the networking layer expect at least 32 bytes of 2434 * headroom, you should not reduce this. 2435 * 2436 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2437 * to reduce average number of cache lines per packet. 2438 * get_rps_cpus() for example only access one 64 bytes aligned block : 2439 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2440 */ 2441 #ifndef NET_SKB_PAD 2442 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2443 #endif 2444 2445 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2446 2447 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 2448 { 2449 if (unlikely(skb_is_nonlinear(skb))) { 2450 WARN_ON(1); 2451 return; 2452 } 2453 skb->len = len; 2454 skb_set_tail_pointer(skb, len); 2455 } 2456 2457 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2458 { 2459 __skb_set_length(skb, len); 2460 } 2461 2462 void skb_trim(struct sk_buff *skb, unsigned int len); 2463 2464 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2465 { 2466 if (skb->data_len) 2467 return ___pskb_trim(skb, len); 2468 __skb_trim(skb, len); 2469 return 0; 2470 } 2471 2472 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2473 { 2474 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2475 } 2476 2477 /** 2478 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2479 * @skb: buffer to alter 2480 * @len: new length 2481 * 2482 * This is identical to pskb_trim except that the caller knows that 2483 * the skb is not cloned so we should never get an error due to out- 2484 * of-memory. 2485 */ 2486 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2487 { 2488 int err = pskb_trim(skb, len); 2489 BUG_ON(err); 2490 } 2491 2492 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 2493 { 2494 unsigned int diff = len - skb->len; 2495 2496 if (skb_tailroom(skb) < diff) { 2497 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 2498 GFP_ATOMIC); 2499 if (ret) 2500 return ret; 2501 } 2502 __skb_set_length(skb, len); 2503 return 0; 2504 } 2505 2506 /** 2507 * skb_orphan - orphan a buffer 2508 * @skb: buffer to orphan 2509 * 2510 * If a buffer currently has an owner then we call the owner's 2511 * destructor function and make the @skb unowned. The buffer continues 2512 * to exist but is no longer charged to its former owner. 2513 */ 2514 static inline void skb_orphan(struct sk_buff *skb) 2515 { 2516 if (skb->destructor) { 2517 skb->destructor(skb); 2518 skb->destructor = NULL; 2519 skb->sk = NULL; 2520 } else { 2521 BUG_ON(skb->sk); 2522 } 2523 } 2524 2525 /** 2526 * skb_orphan_frags - orphan the frags contained in a buffer 2527 * @skb: buffer to orphan frags from 2528 * @gfp_mask: allocation mask for replacement pages 2529 * 2530 * For each frag in the SKB which needs a destructor (i.e. has an 2531 * owner) create a copy of that frag and release the original 2532 * page by calling the destructor. 2533 */ 2534 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2535 { 2536 if (likely(!skb_zcopy(skb))) 2537 return 0; 2538 if (skb_uarg(skb)->callback == sock_zerocopy_callback) 2539 return 0; 2540 return skb_copy_ubufs(skb, gfp_mask); 2541 } 2542 2543 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 2544 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 2545 { 2546 if (likely(!skb_zcopy(skb))) 2547 return 0; 2548 return skb_copy_ubufs(skb, gfp_mask); 2549 } 2550 2551 /** 2552 * __skb_queue_purge - empty a list 2553 * @list: list to empty 2554 * 2555 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2556 * the list and one reference dropped. This function does not take the 2557 * list lock and the caller must hold the relevant locks to use it. 2558 */ 2559 void skb_queue_purge(struct sk_buff_head *list); 2560 static inline void __skb_queue_purge(struct sk_buff_head *list) 2561 { 2562 struct sk_buff *skb; 2563 while ((skb = __skb_dequeue(list)) != NULL) 2564 kfree_skb(skb); 2565 } 2566 2567 void skb_rbtree_purge(struct rb_root *root); 2568 2569 void *netdev_alloc_frag(unsigned int fragsz); 2570 2571 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2572 gfp_t gfp_mask); 2573 2574 /** 2575 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2576 * @dev: network device to receive on 2577 * @length: length to allocate 2578 * 2579 * Allocate a new &sk_buff and assign it a usage count of one. The 2580 * buffer has unspecified headroom built in. Users should allocate 2581 * the headroom they think they need without accounting for the 2582 * built in space. The built in space is used for optimisations. 2583 * 2584 * %NULL is returned if there is no free memory. Although this function 2585 * allocates memory it can be called from an interrupt. 2586 */ 2587 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2588 unsigned int length) 2589 { 2590 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2591 } 2592 2593 /* legacy helper around __netdev_alloc_skb() */ 2594 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2595 gfp_t gfp_mask) 2596 { 2597 return __netdev_alloc_skb(NULL, length, gfp_mask); 2598 } 2599 2600 /* legacy helper around netdev_alloc_skb() */ 2601 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2602 { 2603 return netdev_alloc_skb(NULL, length); 2604 } 2605 2606 2607 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2608 unsigned int length, gfp_t gfp) 2609 { 2610 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2611 2612 if (NET_IP_ALIGN && skb) 2613 skb_reserve(skb, NET_IP_ALIGN); 2614 return skb; 2615 } 2616 2617 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2618 unsigned int length) 2619 { 2620 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2621 } 2622 2623 static inline void skb_free_frag(void *addr) 2624 { 2625 page_frag_free(addr); 2626 } 2627 2628 void *napi_alloc_frag(unsigned int fragsz); 2629 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2630 unsigned int length, gfp_t gfp_mask); 2631 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2632 unsigned int length) 2633 { 2634 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2635 } 2636 void napi_consume_skb(struct sk_buff *skb, int budget); 2637 2638 void __kfree_skb_flush(void); 2639 void __kfree_skb_defer(struct sk_buff *skb); 2640 2641 /** 2642 * __dev_alloc_pages - allocate page for network Rx 2643 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2644 * @order: size of the allocation 2645 * 2646 * Allocate a new page. 2647 * 2648 * %NULL is returned if there is no free memory. 2649 */ 2650 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2651 unsigned int order) 2652 { 2653 /* This piece of code contains several assumptions. 2654 * 1. This is for device Rx, therefor a cold page is preferred. 2655 * 2. The expectation is the user wants a compound page. 2656 * 3. If requesting a order 0 page it will not be compound 2657 * due to the check to see if order has a value in prep_new_page 2658 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2659 * code in gfp_to_alloc_flags that should be enforcing this. 2660 */ 2661 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2662 2663 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2664 } 2665 2666 static inline struct page *dev_alloc_pages(unsigned int order) 2667 { 2668 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 2669 } 2670 2671 /** 2672 * __dev_alloc_page - allocate a page for network Rx 2673 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2674 * 2675 * Allocate a new page. 2676 * 2677 * %NULL is returned if there is no free memory. 2678 */ 2679 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2680 { 2681 return __dev_alloc_pages(gfp_mask, 0); 2682 } 2683 2684 static inline struct page *dev_alloc_page(void) 2685 { 2686 return dev_alloc_pages(0); 2687 } 2688 2689 /** 2690 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2691 * @page: The page that was allocated from skb_alloc_page 2692 * @skb: The skb that may need pfmemalloc set 2693 */ 2694 static inline void skb_propagate_pfmemalloc(struct page *page, 2695 struct sk_buff *skb) 2696 { 2697 if (page_is_pfmemalloc(page)) 2698 skb->pfmemalloc = true; 2699 } 2700 2701 /** 2702 * skb_frag_page - retrieve the page referred to by a paged fragment 2703 * @frag: the paged fragment 2704 * 2705 * Returns the &struct page associated with @frag. 2706 */ 2707 static inline struct page *skb_frag_page(const skb_frag_t *frag) 2708 { 2709 return frag->page.p; 2710 } 2711 2712 /** 2713 * __skb_frag_ref - take an addition reference on a paged fragment. 2714 * @frag: the paged fragment 2715 * 2716 * Takes an additional reference on the paged fragment @frag. 2717 */ 2718 static inline void __skb_frag_ref(skb_frag_t *frag) 2719 { 2720 get_page(skb_frag_page(frag)); 2721 } 2722 2723 /** 2724 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2725 * @skb: the buffer 2726 * @f: the fragment offset. 2727 * 2728 * Takes an additional reference on the @f'th paged fragment of @skb. 2729 */ 2730 static inline void skb_frag_ref(struct sk_buff *skb, int f) 2731 { 2732 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2733 } 2734 2735 /** 2736 * __skb_frag_unref - release a reference on a paged fragment. 2737 * @frag: the paged fragment 2738 * 2739 * Releases a reference on the paged fragment @frag. 2740 */ 2741 static inline void __skb_frag_unref(skb_frag_t *frag) 2742 { 2743 put_page(skb_frag_page(frag)); 2744 } 2745 2746 /** 2747 * skb_frag_unref - release a reference on a paged fragment of an skb. 2748 * @skb: the buffer 2749 * @f: the fragment offset 2750 * 2751 * Releases a reference on the @f'th paged fragment of @skb. 2752 */ 2753 static inline void skb_frag_unref(struct sk_buff *skb, int f) 2754 { 2755 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2756 } 2757 2758 /** 2759 * skb_frag_address - gets the address of the data contained in a paged fragment 2760 * @frag: the paged fragment buffer 2761 * 2762 * Returns the address of the data within @frag. The page must already 2763 * be mapped. 2764 */ 2765 static inline void *skb_frag_address(const skb_frag_t *frag) 2766 { 2767 return page_address(skb_frag_page(frag)) + frag->page_offset; 2768 } 2769 2770 /** 2771 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2772 * @frag: the paged fragment buffer 2773 * 2774 * Returns the address of the data within @frag. Checks that the page 2775 * is mapped and returns %NULL otherwise. 2776 */ 2777 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2778 { 2779 void *ptr = page_address(skb_frag_page(frag)); 2780 if (unlikely(!ptr)) 2781 return NULL; 2782 2783 return ptr + frag->page_offset; 2784 } 2785 2786 /** 2787 * __skb_frag_set_page - sets the page contained in a paged fragment 2788 * @frag: the paged fragment 2789 * @page: the page to set 2790 * 2791 * Sets the fragment @frag to contain @page. 2792 */ 2793 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2794 { 2795 frag->page.p = page; 2796 } 2797 2798 /** 2799 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2800 * @skb: the buffer 2801 * @f: the fragment offset 2802 * @page: the page to set 2803 * 2804 * Sets the @f'th fragment of @skb to contain @page. 2805 */ 2806 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2807 struct page *page) 2808 { 2809 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2810 } 2811 2812 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2813 2814 /** 2815 * skb_frag_dma_map - maps a paged fragment via the DMA API 2816 * @dev: the device to map the fragment to 2817 * @frag: the paged fragment to map 2818 * @offset: the offset within the fragment (starting at the 2819 * fragment's own offset) 2820 * @size: the number of bytes to map 2821 * @dir: the direction of the mapping (``PCI_DMA_*``) 2822 * 2823 * Maps the page associated with @frag to @device. 2824 */ 2825 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2826 const skb_frag_t *frag, 2827 size_t offset, size_t size, 2828 enum dma_data_direction dir) 2829 { 2830 return dma_map_page(dev, skb_frag_page(frag), 2831 frag->page_offset + offset, size, dir); 2832 } 2833 2834 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2835 gfp_t gfp_mask) 2836 { 2837 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2838 } 2839 2840 2841 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2842 gfp_t gfp_mask) 2843 { 2844 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2845 } 2846 2847 2848 /** 2849 * skb_clone_writable - is the header of a clone writable 2850 * @skb: buffer to check 2851 * @len: length up to which to write 2852 * 2853 * Returns true if modifying the header part of the cloned buffer 2854 * does not requires the data to be copied. 2855 */ 2856 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2857 { 2858 return !skb_header_cloned(skb) && 2859 skb_headroom(skb) + len <= skb->hdr_len; 2860 } 2861 2862 static inline int skb_try_make_writable(struct sk_buff *skb, 2863 unsigned int write_len) 2864 { 2865 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 2866 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2867 } 2868 2869 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2870 int cloned) 2871 { 2872 int delta = 0; 2873 2874 if (headroom > skb_headroom(skb)) 2875 delta = headroom - skb_headroom(skb); 2876 2877 if (delta || cloned) 2878 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2879 GFP_ATOMIC); 2880 return 0; 2881 } 2882 2883 /** 2884 * skb_cow - copy header of skb when it is required 2885 * @skb: buffer to cow 2886 * @headroom: needed headroom 2887 * 2888 * If the skb passed lacks sufficient headroom or its data part 2889 * is shared, data is reallocated. If reallocation fails, an error 2890 * is returned and original skb is not changed. 2891 * 2892 * The result is skb with writable area skb->head...skb->tail 2893 * and at least @headroom of space at head. 2894 */ 2895 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2896 { 2897 return __skb_cow(skb, headroom, skb_cloned(skb)); 2898 } 2899 2900 /** 2901 * skb_cow_head - skb_cow but only making the head writable 2902 * @skb: buffer to cow 2903 * @headroom: needed headroom 2904 * 2905 * This function is identical to skb_cow except that we replace the 2906 * skb_cloned check by skb_header_cloned. It should be used when 2907 * you only need to push on some header and do not need to modify 2908 * the data. 2909 */ 2910 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2911 { 2912 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2913 } 2914 2915 /** 2916 * skb_padto - pad an skbuff up to a minimal size 2917 * @skb: buffer to pad 2918 * @len: minimal length 2919 * 2920 * Pads up a buffer to ensure the trailing bytes exist and are 2921 * blanked. If the buffer already contains sufficient data it 2922 * is untouched. Otherwise it is extended. Returns zero on 2923 * success. The skb is freed on error. 2924 */ 2925 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2926 { 2927 unsigned int size = skb->len; 2928 if (likely(size >= len)) 2929 return 0; 2930 return skb_pad(skb, len - size); 2931 } 2932 2933 /** 2934 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2935 * @skb: buffer to pad 2936 * @len: minimal length 2937 * 2938 * Pads up a buffer to ensure the trailing bytes exist and are 2939 * blanked. If the buffer already contains sufficient data it 2940 * is untouched. Otherwise it is extended. Returns zero on 2941 * success. The skb is freed on error. 2942 */ 2943 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2944 { 2945 unsigned int size = skb->len; 2946 2947 if (unlikely(size < len)) { 2948 len -= size; 2949 if (skb_pad(skb, len)) 2950 return -ENOMEM; 2951 __skb_put(skb, len); 2952 } 2953 return 0; 2954 } 2955 2956 static inline int skb_add_data(struct sk_buff *skb, 2957 struct iov_iter *from, int copy) 2958 { 2959 const int off = skb->len; 2960 2961 if (skb->ip_summed == CHECKSUM_NONE) { 2962 __wsum csum = 0; 2963 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 2964 &csum, from)) { 2965 skb->csum = csum_block_add(skb->csum, csum, off); 2966 return 0; 2967 } 2968 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 2969 return 0; 2970 2971 __skb_trim(skb, off); 2972 return -EFAULT; 2973 } 2974 2975 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2976 const struct page *page, int off) 2977 { 2978 if (skb_zcopy(skb)) 2979 return false; 2980 if (i) { 2981 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2982 2983 return page == skb_frag_page(frag) && 2984 off == frag->page_offset + skb_frag_size(frag); 2985 } 2986 return false; 2987 } 2988 2989 static inline int __skb_linearize(struct sk_buff *skb) 2990 { 2991 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2992 } 2993 2994 /** 2995 * skb_linearize - convert paged skb to linear one 2996 * @skb: buffer to linarize 2997 * 2998 * If there is no free memory -ENOMEM is returned, otherwise zero 2999 * is returned and the old skb data released. 3000 */ 3001 static inline int skb_linearize(struct sk_buff *skb) 3002 { 3003 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3004 } 3005 3006 /** 3007 * skb_has_shared_frag - can any frag be overwritten 3008 * @skb: buffer to test 3009 * 3010 * Return true if the skb has at least one frag that might be modified 3011 * by an external entity (as in vmsplice()/sendfile()) 3012 */ 3013 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3014 { 3015 return skb_is_nonlinear(skb) && 3016 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 3017 } 3018 3019 /** 3020 * skb_linearize_cow - make sure skb is linear and writable 3021 * @skb: buffer to process 3022 * 3023 * If there is no free memory -ENOMEM is returned, otherwise zero 3024 * is returned and the old skb data released. 3025 */ 3026 static inline int skb_linearize_cow(struct sk_buff *skb) 3027 { 3028 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3029 __skb_linearize(skb) : 0; 3030 } 3031 3032 static __always_inline void 3033 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3034 unsigned int off) 3035 { 3036 if (skb->ip_summed == CHECKSUM_COMPLETE) 3037 skb->csum = csum_block_sub(skb->csum, 3038 csum_partial(start, len, 0), off); 3039 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3040 skb_checksum_start_offset(skb) < 0) 3041 skb->ip_summed = CHECKSUM_NONE; 3042 } 3043 3044 /** 3045 * skb_postpull_rcsum - update checksum for received skb after pull 3046 * @skb: buffer to update 3047 * @start: start of data before pull 3048 * @len: length of data pulled 3049 * 3050 * After doing a pull on a received packet, you need to call this to 3051 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3052 * CHECKSUM_NONE so that it can be recomputed from scratch. 3053 */ 3054 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3055 const void *start, unsigned int len) 3056 { 3057 __skb_postpull_rcsum(skb, start, len, 0); 3058 } 3059 3060 static __always_inline void 3061 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3062 unsigned int off) 3063 { 3064 if (skb->ip_summed == CHECKSUM_COMPLETE) 3065 skb->csum = csum_block_add(skb->csum, 3066 csum_partial(start, len, 0), off); 3067 } 3068 3069 /** 3070 * skb_postpush_rcsum - update checksum for received skb after push 3071 * @skb: buffer to update 3072 * @start: start of data after push 3073 * @len: length of data pushed 3074 * 3075 * After doing a push on a received packet, you need to call this to 3076 * update the CHECKSUM_COMPLETE checksum. 3077 */ 3078 static inline void skb_postpush_rcsum(struct sk_buff *skb, 3079 const void *start, unsigned int len) 3080 { 3081 __skb_postpush_rcsum(skb, start, len, 0); 3082 } 3083 3084 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3085 3086 /** 3087 * skb_push_rcsum - push skb and update receive checksum 3088 * @skb: buffer to update 3089 * @len: length of data pulled 3090 * 3091 * This function performs an skb_push on the packet and updates 3092 * the CHECKSUM_COMPLETE checksum. It should be used on 3093 * receive path processing instead of skb_push unless you know 3094 * that the checksum difference is zero (e.g., a valid IP header) 3095 * or you are setting ip_summed to CHECKSUM_NONE. 3096 */ 3097 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3098 { 3099 skb_push(skb, len); 3100 skb_postpush_rcsum(skb, skb->data, len); 3101 return skb->data; 3102 } 3103 3104 /** 3105 * pskb_trim_rcsum - trim received skb and update checksum 3106 * @skb: buffer to trim 3107 * @len: new length 3108 * 3109 * This is exactly the same as pskb_trim except that it ensures the 3110 * checksum of received packets are still valid after the operation. 3111 */ 3112 3113 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3114 { 3115 if (likely(len >= skb->len)) 3116 return 0; 3117 if (skb->ip_summed == CHECKSUM_COMPLETE) 3118 skb->ip_summed = CHECKSUM_NONE; 3119 return __pskb_trim(skb, len); 3120 } 3121 3122 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3123 { 3124 if (skb->ip_summed == CHECKSUM_COMPLETE) 3125 skb->ip_summed = CHECKSUM_NONE; 3126 __skb_trim(skb, len); 3127 return 0; 3128 } 3129 3130 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3131 { 3132 if (skb->ip_summed == CHECKSUM_COMPLETE) 3133 skb->ip_summed = CHECKSUM_NONE; 3134 return __skb_grow(skb, len); 3135 } 3136 3137 #define skb_queue_walk(queue, skb) \ 3138 for (skb = (queue)->next; \ 3139 skb != (struct sk_buff *)(queue); \ 3140 skb = skb->next) 3141 3142 #define skb_queue_walk_safe(queue, skb, tmp) \ 3143 for (skb = (queue)->next, tmp = skb->next; \ 3144 skb != (struct sk_buff *)(queue); \ 3145 skb = tmp, tmp = skb->next) 3146 3147 #define skb_queue_walk_from(queue, skb) \ 3148 for (; skb != (struct sk_buff *)(queue); \ 3149 skb = skb->next) 3150 3151 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 3152 for (tmp = skb->next; \ 3153 skb != (struct sk_buff *)(queue); \ 3154 skb = tmp, tmp = skb->next) 3155 3156 #define skb_queue_reverse_walk(queue, skb) \ 3157 for (skb = (queue)->prev; \ 3158 skb != (struct sk_buff *)(queue); \ 3159 skb = skb->prev) 3160 3161 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3162 for (skb = (queue)->prev, tmp = skb->prev; \ 3163 skb != (struct sk_buff *)(queue); \ 3164 skb = tmp, tmp = skb->prev) 3165 3166 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3167 for (tmp = skb->prev; \ 3168 skb != (struct sk_buff *)(queue); \ 3169 skb = tmp, tmp = skb->prev) 3170 3171 static inline bool skb_has_frag_list(const struct sk_buff *skb) 3172 { 3173 return skb_shinfo(skb)->frag_list != NULL; 3174 } 3175 3176 static inline void skb_frag_list_init(struct sk_buff *skb) 3177 { 3178 skb_shinfo(skb)->frag_list = NULL; 3179 } 3180 3181 #define skb_walk_frags(skb, iter) \ 3182 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3183 3184 3185 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, 3186 const struct sk_buff *skb); 3187 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3188 struct sk_buff_head *queue, 3189 unsigned int flags, 3190 void (*destructor)(struct sock *sk, 3191 struct sk_buff *skb), 3192 int *peeked, int *off, int *err, 3193 struct sk_buff **last); 3194 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags, 3195 void (*destructor)(struct sock *sk, 3196 struct sk_buff *skb), 3197 int *peeked, int *off, int *err, 3198 struct sk_buff **last); 3199 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 3200 void (*destructor)(struct sock *sk, 3201 struct sk_buff *skb), 3202 int *peeked, int *off, int *err); 3203 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 3204 int *err); 3205 unsigned int datagram_poll(struct file *file, struct socket *sock, 3206 struct poll_table_struct *wait); 3207 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3208 struct iov_iter *to, int size); 3209 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3210 struct msghdr *msg, int size) 3211 { 3212 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3213 } 3214 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3215 struct msghdr *msg); 3216 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3217 struct iov_iter *from, int len); 3218 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3219 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3220 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3221 static inline void skb_free_datagram_locked(struct sock *sk, 3222 struct sk_buff *skb) 3223 { 3224 __skb_free_datagram_locked(sk, skb, 0); 3225 } 3226 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3227 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3228 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3229 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3230 int len, __wsum csum); 3231 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3232 struct pipe_inode_info *pipe, unsigned int len, 3233 unsigned int flags); 3234 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 3235 int len); 3236 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 3237 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3238 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3239 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3240 int len, int hlen); 3241 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3242 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3243 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3244 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 3245 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu); 3246 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3247 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3248 int skb_ensure_writable(struct sk_buff *skb, int write_len); 3249 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3250 int skb_vlan_pop(struct sk_buff *skb); 3251 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3252 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3253 gfp_t gfp); 3254 3255 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3256 { 3257 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 3258 } 3259 3260 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3261 { 3262 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3263 } 3264 3265 struct skb_checksum_ops { 3266 __wsum (*update)(const void *mem, int len, __wsum wsum); 3267 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3268 }; 3269 3270 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 3271 3272 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3273 __wsum csum, const struct skb_checksum_ops *ops); 3274 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3275 __wsum csum); 3276 3277 static inline void * __must_check 3278 __skb_header_pointer(const struct sk_buff *skb, int offset, 3279 int len, void *data, int hlen, void *buffer) 3280 { 3281 if (hlen - offset >= len) 3282 return data + offset; 3283 3284 if (!skb || 3285 skb_copy_bits(skb, offset, buffer, len) < 0) 3286 return NULL; 3287 3288 return buffer; 3289 } 3290 3291 static inline void * __must_check 3292 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 3293 { 3294 return __skb_header_pointer(skb, offset, len, skb->data, 3295 skb_headlen(skb), buffer); 3296 } 3297 3298 /** 3299 * skb_needs_linearize - check if we need to linearize a given skb 3300 * depending on the given device features. 3301 * @skb: socket buffer to check 3302 * @features: net device features 3303 * 3304 * Returns true if either: 3305 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3306 * 2. skb is fragmented and the device does not support SG. 3307 */ 3308 static inline bool skb_needs_linearize(struct sk_buff *skb, 3309 netdev_features_t features) 3310 { 3311 return skb_is_nonlinear(skb) && 3312 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3313 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3314 } 3315 3316 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3317 void *to, 3318 const unsigned int len) 3319 { 3320 memcpy(to, skb->data, len); 3321 } 3322 3323 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3324 const int offset, void *to, 3325 const unsigned int len) 3326 { 3327 memcpy(to, skb->data + offset, len); 3328 } 3329 3330 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3331 const void *from, 3332 const unsigned int len) 3333 { 3334 memcpy(skb->data, from, len); 3335 } 3336 3337 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3338 const int offset, 3339 const void *from, 3340 const unsigned int len) 3341 { 3342 memcpy(skb->data + offset, from, len); 3343 } 3344 3345 void skb_init(void); 3346 3347 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3348 { 3349 return skb->tstamp; 3350 } 3351 3352 /** 3353 * skb_get_timestamp - get timestamp from a skb 3354 * @skb: skb to get stamp from 3355 * @stamp: pointer to struct timeval to store stamp in 3356 * 3357 * Timestamps are stored in the skb as offsets to a base timestamp. 3358 * This function converts the offset back to a struct timeval and stores 3359 * it in stamp. 3360 */ 3361 static inline void skb_get_timestamp(const struct sk_buff *skb, 3362 struct timeval *stamp) 3363 { 3364 *stamp = ktime_to_timeval(skb->tstamp); 3365 } 3366 3367 static inline void skb_get_timestampns(const struct sk_buff *skb, 3368 struct timespec *stamp) 3369 { 3370 *stamp = ktime_to_timespec(skb->tstamp); 3371 } 3372 3373 static inline void __net_timestamp(struct sk_buff *skb) 3374 { 3375 skb->tstamp = ktime_get_real(); 3376 } 3377 3378 static inline ktime_t net_timedelta(ktime_t t) 3379 { 3380 return ktime_sub(ktime_get_real(), t); 3381 } 3382 3383 static inline ktime_t net_invalid_timestamp(void) 3384 { 3385 return 0; 3386 } 3387 3388 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3389 3390 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3391 3392 void skb_clone_tx_timestamp(struct sk_buff *skb); 3393 bool skb_defer_rx_timestamp(struct sk_buff *skb); 3394 3395 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3396 3397 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3398 { 3399 } 3400 3401 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3402 { 3403 return false; 3404 } 3405 3406 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3407 3408 /** 3409 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3410 * 3411 * PHY drivers may accept clones of transmitted packets for 3412 * timestamping via their phy_driver.txtstamp method. These drivers 3413 * must call this function to return the skb back to the stack with a 3414 * timestamp. 3415 * 3416 * @skb: clone of the the original outgoing packet 3417 * @hwtstamps: hardware time stamps 3418 * 3419 */ 3420 void skb_complete_tx_timestamp(struct sk_buff *skb, 3421 struct skb_shared_hwtstamps *hwtstamps); 3422 3423 void __skb_tstamp_tx(struct sk_buff *orig_skb, 3424 struct skb_shared_hwtstamps *hwtstamps, 3425 struct sock *sk, int tstype); 3426 3427 /** 3428 * skb_tstamp_tx - queue clone of skb with send time stamps 3429 * @orig_skb: the original outgoing packet 3430 * @hwtstamps: hardware time stamps, may be NULL if not available 3431 * 3432 * If the skb has a socket associated, then this function clones the 3433 * skb (thus sharing the actual data and optional structures), stores 3434 * the optional hardware time stamping information (if non NULL) or 3435 * generates a software time stamp (otherwise), then queues the clone 3436 * to the error queue of the socket. Errors are silently ignored. 3437 */ 3438 void skb_tstamp_tx(struct sk_buff *orig_skb, 3439 struct skb_shared_hwtstamps *hwtstamps); 3440 3441 /** 3442 * skb_tx_timestamp() - Driver hook for transmit timestamping 3443 * 3444 * Ethernet MAC Drivers should call this function in their hard_xmit() 3445 * function immediately before giving the sk_buff to the MAC hardware. 3446 * 3447 * Specifically, one should make absolutely sure that this function is 3448 * called before TX completion of this packet can trigger. Otherwise 3449 * the packet could potentially already be freed. 3450 * 3451 * @skb: A socket buffer. 3452 */ 3453 static inline void skb_tx_timestamp(struct sk_buff *skb) 3454 { 3455 skb_clone_tx_timestamp(skb); 3456 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 3457 skb_tstamp_tx(skb, NULL); 3458 } 3459 3460 /** 3461 * skb_complete_wifi_ack - deliver skb with wifi status 3462 * 3463 * @skb: the original outgoing packet 3464 * @acked: ack status 3465 * 3466 */ 3467 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3468 3469 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3470 __sum16 __skb_checksum_complete(struct sk_buff *skb); 3471 3472 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3473 { 3474 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3475 skb->csum_valid || 3476 (skb->ip_summed == CHECKSUM_PARTIAL && 3477 skb_checksum_start_offset(skb) >= 0)); 3478 } 3479 3480 /** 3481 * skb_checksum_complete - Calculate checksum of an entire packet 3482 * @skb: packet to process 3483 * 3484 * This function calculates the checksum over the entire packet plus 3485 * the value of skb->csum. The latter can be used to supply the 3486 * checksum of a pseudo header as used by TCP/UDP. It returns the 3487 * checksum. 3488 * 3489 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3490 * this function can be used to verify that checksum on received 3491 * packets. In that case the function should return zero if the 3492 * checksum is correct. In particular, this function will return zero 3493 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3494 * hardware has already verified the correctness of the checksum. 3495 */ 3496 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3497 { 3498 return skb_csum_unnecessary(skb) ? 3499 0 : __skb_checksum_complete(skb); 3500 } 3501 3502 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3503 { 3504 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3505 if (skb->csum_level == 0) 3506 skb->ip_summed = CHECKSUM_NONE; 3507 else 3508 skb->csum_level--; 3509 } 3510 } 3511 3512 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3513 { 3514 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3515 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3516 skb->csum_level++; 3517 } else if (skb->ip_summed == CHECKSUM_NONE) { 3518 skb->ip_summed = CHECKSUM_UNNECESSARY; 3519 skb->csum_level = 0; 3520 } 3521 } 3522 3523 /* Check if we need to perform checksum complete validation. 3524 * 3525 * Returns true if checksum complete is needed, false otherwise 3526 * (either checksum is unnecessary or zero checksum is allowed). 3527 */ 3528 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3529 bool zero_okay, 3530 __sum16 check) 3531 { 3532 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3533 skb->csum_valid = 1; 3534 __skb_decr_checksum_unnecessary(skb); 3535 return false; 3536 } 3537 3538 return true; 3539 } 3540 3541 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3542 * in checksum_init. 3543 */ 3544 #define CHECKSUM_BREAK 76 3545 3546 /* Unset checksum-complete 3547 * 3548 * Unset checksum complete can be done when packet is being modified 3549 * (uncompressed for instance) and checksum-complete value is 3550 * invalidated. 3551 */ 3552 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3553 { 3554 if (skb->ip_summed == CHECKSUM_COMPLETE) 3555 skb->ip_summed = CHECKSUM_NONE; 3556 } 3557 3558 /* Validate (init) checksum based on checksum complete. 3559 * 3560 * Return values: 3561 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3562 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3563 * checksum is stored in skb->csum for use in __skb_checksum_complete 3564 * non-zero: value of invalid checksum 3565 * 3566 */ 3567 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3568 bool complete, 3569 __wsum psum) 3570 { 3571 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3572 if (!csum_fold(csum_add(psum, skb->csum))) { 3573 skb->csum_valid = 1; 3574 return 0; 3575 } 3576 } 3577 3578 skb->csum = psum; 3579 3580 if (complete || skb->len <= CHECKSUM_BREAK) { 3581 __sum16 csum; 3582 3583 csum = __skb_checksum_complete(skb); 3584 skb->csum_valid = !csum; 3585 return csum; 3586 } 3587 3588 return 0; 3589 } 3590 3591 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3592 { 3593 return 0; 3594 } 3595 3596 /* Perform checksum validate (init). Note that this is a macro since we only 3597 * want to calculate the pseudo header which is an input function if necessary. 3598 * First we try to validate without any computation (checksum unnecessary) and 3599 * then calculate based on checksum complete calling the function to compute 3600 * pseudo header. 3601 * 3602 * Return values: 3603 * 0: checksum is validated or try to in skb_checksum_complete 3604 * non-zero: value of invalid checksum 3605 */ 3606 #define __skb_checksum_validate(skb, proto, complete, \ 3607 zero_okay, check, compute_pseudo) \ 3608 ({ \ 3609 __sum16 __ret = 0; \ 3610 skb->csum_valid = 0; \ 3611 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3612 __ret = __skb_checksum_validate_complete(skb, \ 3613 complete, compute_pseudo(skb, proto)); \ 3614 __ret; \ 3615 }) 3616 3617 #define skb_checksum_init(skb, proto, compute_pseudo) \ 3618 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3619 3620 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3621 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3622 3623 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 3624 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3625 3626 #define skb_checksum_validate_zero_check(skb, proto, check, \ 3627 compute_pseudo) \ 3628 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3629 3630 #define skb_checksum_simple_validate(skb) \ 3631 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3632 3633 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3634 { 3635 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 3636 } 3637 3638 static inline void __skb_checksum_convert(struct sk_buff *skb, 3639 __sum16 check, __wsum pseudo) 3640 { 3641 skb->csum = ~pseudo; 3642 skb->ip_summed = CHECKSUM_COMPLETE; 3643 } 3644 3645 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3646 do { \ 3647 if (__skb_checksum_convert_check(skb)) \ 3648 __skb_checksum_convert(skb, check, \ 3649 compute_pseudo(skb, proto)); \ 3650 } while (0) 3651 3652 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3653 u16 start, u16 offset) 3654 { 3655 skb->ip_summed = CHECKSUM_PARTIAL; 3656 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3657 skb->csum_offset = offset - start; 3658 } 3659 3660 /* Update skbuf and packet to reflect the remote checksum offload operation. 3661 * When called, ptr indicates the starting point for skb->csum when 3662 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3663 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3664 */ 3665 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3666 int start, int offset, bool nopartial) 3667 { 3668 __wsum delta; 3669 3670 if (!nopartial) { 3671 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3672 return; 3673 } 3674 3675 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3676 __skb_checksum_complete(skb); 3677 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3678 } 3679 3680 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3681 3682 /* Adjust skb->csum since we changed the packet */ 3683 skb->csum = csum_add(skb->csum, delta); 3684 } 3685 3686 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 3687 { 3688 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 3689 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK); 3690 #else 3691 return NULL; 3692 #endif 3693 } 3694 3695 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3696 void nf_conntrack_destroy(struct nf_conntrack *nfct); 3697 static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3698 { 3699 if (nfct && atomic_dec_and_test(&nfct->use)) 3700 nf_conntrack_destroy(nfct); 3701 } 3702 static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3703 { 3704 if (nfct) 3705 atomic_inc(&nfct->use); 3706 } 3707 #endif 3708 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3709 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3710 { 3711 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use)) 3712 kfree(nf_bridge); 3713 } 3714 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3715 { 3716 if (nf_bridge) 3717 refcount_inc(&nf_bridge->use); 3718 } 3719 #endif /* CONFIG_BRIDGE_NETFILTER */ 3720 static inline void nf_reset(struct sk_buff *skb) 3721 { 3722 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3723 nf_conntrack_put(skb_nfct(skb)); 3724 skb->_nfct = 0; 3725 #endif 3726 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3727 nf_bridge_put(skb->nf_bridge); 3728 skb->nf_bridge = NULL; 3729 #endif 3730 } 3731 3732 static inline void nf_reset_trace(struct sk_buff *skb) 3733 { 3734 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3735 skb->nf_trace = 0; 3736 #endif 3737 } 3738 3739 /* Note: This doesn't put any conntrack and bridge info in dst. */ 3740 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3741 bool copy) 3742 { 3743 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3744 dst->_nfct = src->_nfct; 3745 nf_conntrack_get(skb_nfct(src)); 3746 #endif 3747 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3748 dst->nf_bridge = src->nf_bridge; 3749 nf_bridge_get(src->nf_bridge); 3750 #endif 3751 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3752 if (copy) 3753 dst->nf_trace = src->nf_trace; 3754 #endif 3755 } 3756 3757 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3758 { 3759 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3760 nf_conntrack_put(skb_nfct(dst)); 3761 #endif 3762 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3763 nf_bridge_put(dst->nf_bridge); 3764 #endif 3765 __nf_copy(dst, src, true); 3766 } 3767 3768 #ifdef CONFIG_NETWORK_SECMARK 3769 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3770 { 3771 to->secmark = from->secmark; 3772 } 3773 3774 static inline void skb_init_secmark(struct sk_buff *skb) 3775 { 3776 skb->secmark = 0; 3777 } 3778 #else 3779 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3780 { } 3781 3782 static inline void skb_init_secmark(struct sk_buff *skb) 3783 { } 3784 #endif 3785 3786 static inline bool skb_irq_freeable(const struct sk_buff *skb) 3787 { 3788 return !skb->destructor && 3789 #if IS_ENABLED(CONFIG_XFRM) 3790 !skb->sp && 3791 #endif 3792 !skb_nfct(skb) && 3793 !skb->_skb_refdst && 3794 !skb_has_frag_list(skb); 3795 } 3796 3797 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3798 { 3799 skb->queue_mapping = queue_mapping; 3800 } 3801 3802 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3803 { 3804 return skb->queue_mapping; 3805 } 3806 3807 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3808 { 3809 to->queue_mapping = from->queue_mapping; 3810 } 3811 3812 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3813 { 3814 skb->queue_mapping = rx_queue + 1; 3815 } 3816 3817 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3818 { 3819 return skb->queue_mapping - 1; 3820 } 3821 3822 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3823 { 3824 return skb->queue_mapping != 0; 3825 } 3826 3827 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 3828 { 3829 skb->dst_pending_confirm = val; 3830 } 3831 3832 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 3833 { 3834 return skb->dst_pending_confirm != 0; 3835 } 3836 3837 static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3838 { 3839 #ifdef CONFIG_XFRM 3840 return skb->sp; 3841 #else 3842 return NULL; 3843 #endif 3844 } 3845 3846 /* Keeps track of mac header offset relative to skb->head. 3847 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3848 * For non-tunnel skb it points to skb_mac_header() and for 3849 * tunnel skb it points to outer mac header. 3850 * Keeps track of level of encapsulation of network headers. 3851 */ 3852 struct skb_gso_cb { 3853 union { 3854 int mac_offset; 3855 int data_offset; 3856 }; 3857 int encap_level; 3858 __wsum csum; 3859 __u16 csum_start; 3860 }; 3861 #define SKB_SGO_CB_OFFSET 32 3862 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 3863 3864 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3865 { 3866 return (skb_mac_header(inner_skb) - inner_skb->head) - 3867 SKB_GSO_CB(inner_skb)->mac_offset; 3868 } 3869 3870 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3871 { 3872 int new_headroom, headroom; 3873 int ret; 3874 3875 headroom = skb_headroom(skb); 3876 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3877 if (ret) 3878 return ret; 3879 3880 new_headroom = skb_headroom(skb); 3881 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3882 return 0; 3883 } 3884 3885 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 3886 { 3887 /* Do not update partial checksums if remote checksum is enabled. */ 3888 if (skb->remcsum_offload) 3889 return; 3890 3891 SKB_GSO_CB(skb)->csum = res; 3892 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 3893 } 3894 3895 /* Compute the checksum for a gso segment. First compute the checksum value 3896 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3897 * then add in skb->csum (checksum from csum_start to end of packet). 3898 * skb->csum and csum_start are then updated to reflect the checksum of the 3899 * resultant packet starting from the transport header-- the resultant checksum 3900 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3901 * header. 3902 */ 3903 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3904 { 3905 unsigned char *csum_start = skb_transport_header(skb); 3906 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 3907 __wsum partial = SKB_GSO_CB(skb)->csum; 3908 3909 SKB_GSO_CB(skb)->csum = res; 3910 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 3911 3912 return csum_fold(csum_partial(csum_start, plen, partial)); 3913 } 3914 3915 static inline bool skb_is_gso(const struct sk_buff *skb) 3916 { 3917 return skb_shinfo(skb)->gso_size; 3918 } 3919 3920 /* Note: Should be called only if skb_is_gso(skb) is true */ 3921 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3922 { 3923 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3924 } 3925 3926 static inline void skb_gso_reset(struct sk_buff *skb) 3927 { 3928 skb_shinfo(skb)->gso_size = 0; 3929 skb_shinfo(skb)->gso_segs = 0; 3930 skb_shinfo(skb)->gso_type = 0; 3931 } 3932 3933 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3934 3935 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3936 { 3937 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3938 * wanted then gso_type will be set. */ 3939 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3940 3941 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3942 unlikely(shinfo->gso_type == 0)) { 3943 __skb_warn_lro_forwarding(skb); 3944 return true; 3945 } 3946 return false; 3947 } 3948 3949 static inline void skb_forward_csum(struct sk_buff *skb) 3950 { 3951 /* Unfortunately we don't support this one. Any brave souls? */ 3952 if (skb->ip_summed == CHECKSUM_COMPLETE) 3953 skb->ip_summed = CHECKSUM_NONE; 3954 } 3955 3956 /** 3957 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3958 * @skb: skb to check 3959 * 3960 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3961 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3962 * use this helper, to document places where we make this assertion. 3963 */ 3964 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3965 { 3966 #ifdef DEBUG 3967 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3968 #endif 3969 } 3970 3971 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3972 3973 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3974 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 3975 unsigned int transport_len, 3976 __sum16(*skb_chkf)(struct sk_buff *skb)); 3977 3978 /** 3979 * skb_head_is_locked - Determine if the skb->head is locked down 3980 * @skb: skb to check 3981 * 3982 * The head on skbs build around a head frag can be removed if they are 3983 * not cloned. This function returns true if the skb head is locked down 3984 * due to either being allocated via kmalloc, or by being a clone with 3985 * multiple references to the head. 3986 */ 3987 static inline bool skb_head_is_locked(const struct sk_buff *skb) 3988 { 3989 return !skb->head_frag || skb_cloned(skb); 3990 } 3991 3992 /** 3993 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3994 * 3995 * @skb: GSO skb 3996 * 3997 * skb_gso_network_seglen is used to determine the real size of the 3998 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3999 * 4000 * The MAC/L2 header is not accounted for. 4001 */ 4002 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 4003 { 4004 unsigned int hdr_len = skb_transport_header(skb) - 4005 skb_network_header(skb); 4006 return hdr_len + skb_gso_transport_seglen(skb); 4007 } 4008 4009 /* Local Checksum Offload. 4010 * Compute outer checksum based on the assumption that the 4011 * inner checksum will be offloaded later. 4012 * See Documentation/networking/checksum-offloads.txt for 4013 * explanation of how this works. 4014 * Fill in outer checksum adjustment (e.g. with sum of outer 4015 * pseudo-header) before calling. 4016 * Also ensure that inner checksum is in linear data area. 4017 */ 4018 static inline __wsum lco_csum(struct sk_buff *skb) 4019 { 4020 unsigned char *csum_start = skb_checksum_start(skb); 4021 unsigned char *l4_hdr = skb_transport_header(skb); 4022 __wsum partial; 4023 4024 /* Start with complement of inner checksum adjustment */ 4025 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 4026 skb->csum_offset)); 4027 4028 /* Add in checksum of our headers (incl. outer checksum 4029 * adjustment filled in by caller) and return result. 4030 */ 4031 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 4032 } 4033 4034 #endif /* __KERNEL__ */ 4035 #endif /* _LINUX_SKBUFF_H */ 4036