1 /* 2 * Definitions for the 'struct sk_buff' memory handlers. 3 * 4 * Authors: 5 * Alan Cox, <[email protected]> 6 * Florian La Roche, <[email protected]> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 */ 13 14 #ifndef _LINUX_SKBUFF_H 15 #define _LINUX_SKBUFF_H 16 17 #include <linux/kernel.h> 18 #include <linux/compiler.h> 19 #include <linux/time.h> 20 #include <linux/bug.h> 21 #include <linux/cache.h> 22 #include <linux/rbtree.h> 23 #include <linux/socket.h> 24 #include <linux/refcount.h> 25 26 #include <linux/atomic.h> 27 #include <asm/types.h> 28 #include <linux/spinlock.h> 29 #include <linux/net.h> 30 #include <linux/textsearch.h> 31 #include <net/checksum.h> 32 #include <linux/rcupdate.h> 33 #include <linux/hrtimer.h> 34 #include <linux/dma-mapping.h> 35 #include <linux/netdev_features.h> 36 #include <linux/sched.h> 37 #include <linux/sched/clock.h> 38 #include <net/flow_dissector.h> 39 #include <linux/splice.h> 40 #include <linux/in6.h> 41 #include <linux/if_packet.h> 42 #include <net/flow.h> 43 44 /* The interface for checksum offload between the stack and networking drivers 45 * is as follows... 46 * 47 * A. IP checksum related features 48 * 49 * Drivers advertise checksum offload capabilities in the features of a device. 50 * From the stack's point of view these are capabilities offered by the driver, 51 * a driver typically only advertises features that it is capable of offloading 52 * to its device. 53 * 54 * The checksum related features are: 55 * 56 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one 57 * IP (one's complement) checksum for any combination 58 * of protocols or protocol layering. The checksum is 59 * computed and set in a packet per the CHECKSUM_PARTIAL 60 * interface (see below). 61 * 62 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain 63 * TCP or UDP packets over IPv4. These are specifically 64 * unencapsulated packets of the form IPv4|TCP or 65 * IPv4|UDP where the Protocol field in the IPv4 header 66 * is TCP or UDP. The IPv4 header may contain IP options 67 * This feature cannot be set in features for a device 68 * with NETIF_F_HW_CSUM also set. This feature is being 69 * DEPRECATED (see below). 70 * 71 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain 72 * TCP or UDP packets over IPv6. These are specifically 73 * unencapsulated packets of the form IPv6|TCP or 74 * IPv4|UDP where the Next Header field in the IPv6 75 * header is either TCP or UDP. IPv6 extension headers 76 * are not supported with this feature. This feature 77 * cannot be set in features for a device with 78 * NETIF_F_HW_CSUM also set. This feature is being 79 * DEPRECATED (see below). 80 * 81 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload. 82 * This flag is used only used to disable the RX checksum 83 * feature for a device. The stack will accept receive 84 * checksum indication in packets received on a device 85 * regardless of whether NETIF_F_RXCSUM is set. 86 * 87 * B. Checksumming of received packets by device. Indication of checksum 88 * verification is in set skb->ip_summed. Possible values are: 89 * 90 * CHECKSUM_NONE: 91 * 92 * Device did not checksum this packet e.g. due to lack of capabilities. 93 * The packet contains full (though not verified) checksum in packet but 94 * not in skb->csum. Thus, skb->csum is undefined in this case. 95 * 96 * CHECKSUM_UNNECESSARY: 97 * 98 * The hardware you're dealing with doesn't calculate the full checksum 99 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 100 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY 101 * if their checksums are okay. skb->csum is still undefined in this case 102 * though. A driver or device must never modify the checksum field in the 103 * packet even if checksum is verified. 104 * 105 * CHECKSUM_UNNECESSARY is applicable to following protocols: 106 * TCP: IPv6 and IPv4. 107 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 108 * zero UDP checksum for either IPv4 or IPv6, the networking stack 109 * may perform further validation in this case. 110 * GRE: only if the checksum is present in the header. 111 * SCTP: indicates the CRC in SCTP header has been validated. 112 * FCOE: indicates the CRC in FC frame has been validated. 113 * 114 * skb->csum_level indicates the number of consecutive checksums found in 115 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY. 116 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 117 * and a device is able to verify the checksums for UDP (possibly zero), 118 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to 119 * two. If the device were only able to verify the UDP checksum and not 120 * GRE, either because it doesn't support GRE checksum of because GRE 121 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 122 * not considered in this case). 123 * 124 * CHECKSUM_COMPLETE: 125 * 126 * This is the most generic way. The device supplied checksum of the _whole_ 127 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the 128 * hardware doesn't need to parse L3/L4 headers to implement this. 129 * 130 * Notes: 131 * - Even if device supports only some protocols, but is able to produce 132 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 133 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 134 * 135 * CHECKSUM_PARTIAL: 136 * 137 * A checksum is set up to be offloaded to a device as described in the 138 * output description for CHECKSUM_PARTIAL. This may occur on a packet 139 * received directly from another Linux OS, e.g., a virtualized Linux kernel 140 * on the same host, or it may be set in the input path in GRO or remote 141 * checksum offload. For the purposes of checksum verification, the checksum 142 * referred to by skb->csum_start + skb->csum_offset and any preceding 143 * checksums in the packet are considered verified. Any checksums in the 144 * packet that are after the checksum being offloaded are not considered to 145 * be verified. 146 * 147 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload 148 * in the skb->ip_summed for a packet. Values are: 149 * 150 * CHECKSUM_PARTIAL: 151 * 152 * The driver is required to checksum the packet as seen by hard_start_xmit() 153 * from skb->csum_start up to the end, and to record/write the checksum at 154 * offset skb->csum_start + skb->csum_offset. A driver may verify that the 155 * csum_start and csum_offset values are valid values given the length and 156 * offset of the packet, however they should not attempt to validate that the 157 * checksum refers to a legitimate transport layer checksum-- it is the 158 * purview of the stack to validate that csum_start and csum_offset are set 159 * correctly. 160 * 161 * When the stack requests checksum offload for a packet, the driver MUST 162 * ensure that the checksum is set correctly. A driver can either offload the 163 * checksum calculation to the device, or call skb_checksum_help (in the case 164 * that the device does not support offload for a particular checksum). 165 * 166 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of 167 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate 168 * checksum offload capability. 169 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based 170 * on network device checksumming capabilities: if a packet does not match 171 * them, skb_checksum_help or skb_crc32c_help (depending on the value of 172 * csum_not_inet, see item D.) is called to resolve the checksum. 173 * 174 * CHECKSUM_NONE: 175 * 176 * The skb was already checksummed by the protocol, or a checksum is not 177 * required. 178 * 179 * CHECKSUM_UNNECESSARY: 180 * 181 * This has the same meaning on as CHECKSUM_NONE for checksum offload on 182 * output. 183 * 184 * CHECKSUM_COMPLETE: 185 * Not used in checksum output. If a driver observes a packet with this value 186 * set in skbuff, if should treat as CHECKSUM_NONE being set. 187 * 188 * D. Non-IP checksum (CRC) offloads 189 * 190 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of 191 * offloading the SCTP CRC in a packet. To perform this offload the stack 192 * will set set csum_start and csum_offset accordingly, set ip_summed to 193 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in 194 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c. 195 * A driver that supports both IP checksum offload and SCTP CRC32c offload 196 * must verify which offload is configured for a packet by testing the 197 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve 198 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 199 * 200 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of 201 * offloading the FCOE CRC in a packet. To perform this offload the stack 202 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 203 * accordingly. Note the there is no indication in the skbuff that the 204 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports 205 * both IP checksum offload and FCOE CRC offload must verify which offload 206 * is configured for a packet presumably by inspecting packet headers. 207 * 208 * E. Checksumming on output with GSO. 209 * 210 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload 211 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 212 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as 213 * part of the GSO operation is implied. If a checksum is being offloaded 214 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset 215 * are set to refer to the outermost checksum being offload (two offloaded 216 * checksums are possible with UDP encapsulation). 217 */ 218 219 /* Don't change this without changing skb_csum_unnecessary! */ 220 #define CHECKSUM_NONE 0 221 #define CHECKSUM_UNNECESSARY 1 222 #define CHECKSUM_COMPLETE 2 223 #define CHECKSUM_PARTIAL 3 224 225 /* Maximum value in skb->csum_level */ 226 #define SKB_MAX_CSUM_LEVEL 3 227 228 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 229 #define SKB_WITH_OVERHEAD(X) \ 230 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 231 #define SKB_MAX_ORDER(X, ORDER) \ 232 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 233 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 234 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 235 236 /* return minimum truesize of one skb containing X bytes of data */ 237 #define SKB_TRUESIZE(X) ((X) + \ 238 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 239 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 240 241 struct net_device; 242 struct scatterlist; 243 struct pipe_inode_info; 244 struct iov_iter; 245 struct napi_struct; 246 struct bpf_prog; 247 union bpf_attr; 248 struct skb_ext; 249 250 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 251 struct nf_conntrack { 252 atomic_t use; 253 }; 254 #endif 255 256 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 257 struct nf_bridge_info { 258 enum { 259 BRNF_PROTO_UNCHANGED, 260 BRNF_PROTO_8021Q, 261 BRNF_PROTO_PPPOE 262 } orig_proto:8; 263 u8 pkt_otherhost:1; 264 u8 in_prerouting:1; 265 u8 bridged_dnat:1; 266 __u16 frag_max_size; 267 struct net_device *physindev; 268 269 /* always valid & non-NULL from FORWARD on, for physdev match */ 270 struct net_device *physoutdev; 271 union { 272 /* prerouting: detect dnat in orig/reply direction */ 273 __be32 ipv4_daddr; 274 struct in6_addr ipv6_daddr; 275 276 /* after prerouting + nat detected: store original source 277 * mac since neigh resolution overwrites it, only used while 278 * skb is out in neigh layer. 279 */ 280 char neigh_header[8]; 281 }; 282 }; 283 #endif 284 285 struct sk_buff_head { 286 /* These two members must be first. */ 287 struct sk_buff *next; 288 struct sk_buff *prev; 289 290 __u32 qlen; 291 spinlock_t lock; 292 }; 293 294 struct sk_buff; 295 296 /* To allow 64K frame to be packed as single skb without frag_list we 297 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 298 * buffers which do not start on a page boundary. 299 * 300 * Since GRO uses frags we allocate at least 16 regardless of page 301 * size. 302 */ 303 #if (65536/PAGE_SIZE + 1) < 16 304 #define MAX_SKB_FRAGS 16UL 305 #else 306 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 307 #endif 308 extern int sysctl_max_skb_frags; 309 310 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 311 * segment using its current segmentation instead. 312 */ 313 #define GSO_BY_FRAGS 0xFFFF 314 315 typedef struct skb_frag_struct skb_frag_t; 316 317 struct skb_frag_struct { 318 struct { 319 struct page *p; 320 } page; 321 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 322 __u32 page_offset; 323 __u32 size; 324 #else 325 __u16 page_offset; 326 __u16 size; 327 #endif 328 }; 329 330 /** 331 * skb_frag_size - Returns the size of a skb fragment 332 * @frag: skb fragment 333 */ 334 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 335 { 336 return frag->size; 337 } 338 339 /** 340 * skb_frag_size_set - Sets the size of a skb fragment 341 * @frag: skb fragment 342 * @size: size of fragment 343 */ 344 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 345 { 346 frag->size = size; 347 } 348 349 /** 350 * skb_frag_size_add - Incrementes the size of a skb fragment by %delta 351 * @frag: skb fragment 352 * @delta: value to add 353 */ 354 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 355 { 356 frag->size += delta; 357 } 358 359 /** 360 * skb_frag_size_sub - Decrements the size of a skb fragment by %delta 361 * @frag: skb fragment 362 * @delta: value to subtract 363 */ 364 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 365 { 366 frag->size -= delta; 367 } 368 369 /** 370 * skb_frag_must_loop - Test if %p is a high memory page 371 * @p: fragment's page 372 */ 373 static inline bool skb_frag_must_loop(struct page *p) 374 { 375 #if defined(CONFIG_HIGHMEM) 376 if (PageHighMem(p)) 377 return true; 378 #endif 379 return false; 380 } 381 382 /** 383 * skb_frag_foreach_page - loop over pages in a fragment 384 * 385 * @f: skb frag to operate on 386 * @f_off: offset from start of f->page.p 387 * @f_len: length from f_off to loop over 388 * @p: (temp var) current page 389 * @p_off: (temp var) offset from start of current page, 390 * non-zero only on first page. 391 * @p_len: (temp var) length in current page, 392 * < PAGE_SIZE only on first and last page. 393 * @copied: (temp var) length so far, excluding current p_len. 394 * 395 * A fragment can hold a compound page, in which case per-page 396 * operations, notably kmap_atomic, must be called for each 397 * regular page. 398 */ 399 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 400 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 401 p_off = (f_off) & (PAGE_SIZE - 1), \ 402 p_len = skb_frag_must_loop(p) ? \ 403 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 404 copied = 0; \ 405 copied < f_len; \ 406 copied += p_len, p++, p_off = 0, \ 407 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 408 409 #define HAVE_HW_TIME_STAMP 410 411 /** 412 * struct skb_shared_hwtstamps - hardware time stamps 413 * @hwtstamp: hardware time stamp transformed into duration 414 * since arbitrary point in time 415 * 416 * Software time stamps generated by ktime_get_real() are stored in 417 * skb->tstamp. 418 * 419 * hwtstamps can only be compared against other hwtstamps from 420 * the same device. 421 * 422 * This structure is attached to packets as part of the 423 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 424 */ 425 struct skb_shared_hwtstamps { 426 ktime_t hwtstamp; 427 }; 428 429 /* Definitions for tx_flags in struct skb_shared_info */ 430 enum { 431 /* generate hardware time stamp */ 432 SKBTX_HW_TSTAMP = 1 << 0, 433 434 /* generate software time stamp when queueing packet to NIC */ 435 SKBTX_SW_TSTAMP = 1 << 1, 436 437 /* device driver is going to provide hardware time stamp */ 438 SKBTX_IN_PROGRESS = 1 << 2, 439 440 /* device driver supports TX zero-copy buffers */ 441 SKBTX_DEV_ZEROCOPY = 1 << 3, 442 443 /* generate wifi status information (where possible) */ 444 SKBTX_WIFI_STATUS = 1 << 4, 445 446 /* This indicates at least one fragment might be overwritten 447 * (as in vmsplice(), sendfile() ...) 448 * If we need to compute a TX checksum, we'll need to copy 449 * all frags to avoid possible bad checksum 450 */ 451 SKBTX_SHARED_FRAG = 1 << 5, 452 453 /* generate software time stamp when entering packet scheduling */ 454 SKBTX_SCHED_TSTAMP = 1 << 6, 455 }; 456 457 #define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG) 458 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 459 SKBTX_SCHED_TSTAMP) 460 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 461 462 /* 463 * The callback notifies userspace to release buffers when skb DMA is done in 464 * lower device, the skb last reference should be 0 when calling this. 465 * The zerocopy_success argument is true if zero copy transmit occurred, 466 * false on data copy or out of memory error caused by data copy attempt. 467 * The ctx field is used to track device context. 468 * The desc field is used to track userspace buffer index. 469 */ 470 struct ubuf_info { 471 void (*callback)(struct ubuf_info *, bool zerocopy_success); 472 union { 473 struct { 474 unsigned long desc; 475 void *ctx; 476 }; 477 struct { 478 u32 id; 479 u16 len; 480 u16 zerocopy:1; 481 u32 bytelen; 482 }; 483 }; 484 refcount_t refcnt; 485 486 struct mmpin { 487 struct user_struct *user; 488 unsigned int num_pg; 489 } mmp; 490 }; 491 492 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 493 494 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 495 void mm_unaccount_pinned_pages(struct mmpin *mmp); 496 497 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size); 498 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size, 499 struct ubuf_info *uarg); 500 501 static inline void sock_zerocopy_get(struct ubuf_info *uarg) 502 { 503 refcount_inc(&uarg->refcnt); 504 } 505 506 void sock_zerocopy_put(struct ubuf_info *uarg); 507 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 508 509 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success); 510 511 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len); 512 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 513 struct msghdr *msg, int len, 514 struct ubuf_info *uarg); 515 516 /* This data is invariant across clones and lives at 517 * the end of the header data, ie. at skb->end. 518 */ 519 struct skb_shared_info { 520 __u8 __unused; 521 __u8 meta_len; 522 __u8 nr_frags; 523 __u8 tx_flags; 524 unsigned short gso_size; 525 /* Warning: this field is not always filled in (UFO)! */ 526 unsigned short gso_segs; 527 struct sk_buff *frag_list; 528 struct skb_shared_hwtstamps hwtstamps; 529 unsigned int gso_type; 530 u32 tskey; 531 532 /* 533 * Warning : all fields before dataref are cleared in __alloc_skb() 534 */ 535 atomic_t dataref; 536 537 /* Intermediate layers must ensure that destructor_arg 538 * remains valid until skb destructor */ 539 void * destructor_arg; 540 541 /* must be last field, see pskb_expand_head() */ 542 skb_frag_t frags[MAX_SKB_FRAGS]; 543 }; 544 545 /* We divide dataref into two halves. The higher 16 bits hold references 546 * to the payload part of skb->data. The lower 16 bits hold references to 547 * the entire skb->data. A clone of a headerless skb holds the length of 548 * the header in skb->hdr_len. 549 * 550 * All users must obey the rule that the skb->data reference count must be 551 * greater than or equal to the payload reference count. 552 * 553 * Holding a reference to the payload part means that the user does not 554 * care about modifications to the header part of skb->data. 555 */ 556 #define SKB_DATAREF_SHIFT 16 557 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 558 559 560 enum { 561 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 562 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 563 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 564 }; 565 566 enum { 567 SKB_GSO_TCPV4 = 1 << 0, 568 569 /* This indicates the skb is from an untrusted source. */ 570 SKB_GSO_DODGY = 1 << 1, 571 572 /* This indicates the tcp segment has CWR set. */ 573 SKB_GSO_TCP_ECN = 1 << 2, 574 575 SKB_GSO_TCP_FIXEDID = 1 << 3, 576 577 SKB_GSO_TCPV6 = 1 << 4, 578 579 SKB_GSO_FCOE = 1 << 5, 580 581 SKB_GSO_GRE = 1 << 6, 582 583 SKB_GSO_GRE_CSUM = 1 << 7, 584 585 SKB_GSO_IPXIP4 = 1 << 8, 586 587 SKB_GSO_IPXIP6 = 1 << 9, 588 589 SKB_GSO_UDP_TUNNEL = 1 << 10, 590 591 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 592 593 SKB_GSO_PARTIAL = 1 << 12, 594 595 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 596 597 SKB_GSO_SCTP = 1 << 14, 598 599 SKB_GSO_ESP = 1 << 15, 600 601 SKB_GSO_UDP = 1 << 16, 602 603 SKB_GSO_UDP_L4 = 1 << 17, 604 }; 605 606 #if BITS_PER_LONG > 32 607 #define NET_SKBUFF_DATA_USES_OFFSET 1 608 #endif 609 610 #ifdef NET_SKBUFF_DATA_USES_OFFSET 611 typedef unsigned int sk_buff_data_t; 612 #else 613 typedef unsigned char *sk_buff_data_t; 614 #endif 615 616 /** 617 * struct sk_buff - socket buffer 618 * @next: Next buffer in list 619 * @prev: Previous buffer in list 620 * @tstamp: Time we arrived/left 621 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 622 * @sk: Socket we are owned by 623 * @dev: Device we arrived on/are leaving by 624 * @cb: Control buffer. Free for use by every layer. Put private vars here 625 * @_skb_refdst: destination entry (with norefcount bit) 626 * @sp: the security path, used for xfrm 627 * @len: Length of actual data 628 * @data_len: Data length 629 * @mac_len: Length of link layer header 630 * @hdr_len: writable header length of cloned skb 631 * @csum: Checksum (must include start/offset pair) 632 * @csum_start: Offset from skb->head where checksumming should start 633 * @csum_offset: Offset from csum_start where checksum should be stored 634 * @priority: Packet queueing priority 635 * @ignore_df: allow local fragmentation 636 * @cloned: Head may be cloned (check refcnt to be sure) 637 * @ip_summed: Driver fed us an IP checksum 638 * @nohdr: Payload reference only, must not modify header 639 * @pkt_type: Packet class 640 * @fclone: skbuff clone status 641 * @ipvs_property: skbuff is owned by ipvs 642 * @offload_fwd_mark: Packet was L2-forwarded in hardware 643 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 644 * @tc_skip_classify: do not classify packet. set by IFB device 645 * @tc_at_ingress: used within tc_classify to distinguish in/egress 646 * @tc_redirected: packet was redirected by a tc action 647 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect 648 * @peeked: this packet has been seen already, so stats have been 649 * done for it, don't do them again 650 * @nf_trace: netfilter packet trace flag 651 * @protocol: Packet protocol from driver 652 * @destructor: Destruct function 653 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 654 * @_nfct: Associated connection, if any (with nfctinfo bits) 655 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 656 * @skb_iif: ifindex of device we arrived on 657 * @tc_index: Traffic control index 658 * @hash: the packet hash 659 * @queue_mapping: Queue mapping for multiqueue devices 660 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 661 * @active_extensions: active extensions (skb_ext_id types) 662 * @ndisc_nodetype: router type (from link layer) 663 * @ooo_okay: allow the mapping of a socket to a queue to be changed 664 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 665 * ports. 666 * @sw_hash: indicates hash was computed in software stack 667 * @wifi_acked_valid: wifi_acked was set 668 * @wifi_acked: whether frame was acked on wifi or not 669 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 670 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 671 * @dst_pending_confirm: need to confirm neighbour 672 * @decrypted: Decrypted SKB 673 * @napi_id: id of the NAPI struct this skb came from 674 * @secmark: security marking 675 * @mark: Generic packet mark 676 * @vlan_proto: vlan encapsulation protocol 677 * @vlan_tci: vlan tag control information 678 * @inner_protocol: Protocol (encapsulation) 679 * @inner_transport_header: Inner transport layer header (encapsulation) 680 * @inner_network_header: Network layer header (encapsulation) 681 * @inner_mac_header: Link layer header (encapsulation) 682 * @transport_header: Transport layer header 683 * @network_header: Network layer header 684 * @mac_header: Link layer header 685 * @tail: Tail pointer 686 * @end: End pointer 687 * @head: Head of buffer 688 * @data: Data head pointer 689 * @truesize: Buffer size 690 * @users: User count - see {datagram,tcp}.c 691 * @extensions: allocated extensions, valid if active_extensions is nonzero 692 */ 693 694 struct sk_buff { 695 union { 696 struct { 697 /* These two members must be first. */ 698 struct sk_buff *next; 699 struct sk_buff *prev; 700 701 union { 702 struct net_device *dev; 703 /* Some protocols might use this space to store information, 704 * while device pointer would be NULL. 705 * UDP receive path is one user. 706 */ 707 unsigned long dev_scratch; 708 }; 709 }; 710 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 711 struct list_head list; 712 }; 713 714 union { 715 struct sock *sk; 716 int ip_defrag_offset; 717 }; 718 719 union { 720 ktime_t tstamp; 721 u64 skb_mstamp_ns; /* earliest departure time */ 722 }; 723 /* 724 * This is the control buffer. It is free to use for every 725 * layer. Please put your private variables there. If you 726 * want to keep them across layers you have to do a skb_clone() 727 * first. This is owned by whoever has the skb queued ATM. 728 */ 729 char cb[48] __aligned(8); 730 731 union { 732 struct { 733 unsigned long _skb_refdst; 734 void (*destructor)(struct sk_buff *skb); 735 }; 736 struct list_head tcp_tsorted_anchor; 737 }; 738 739 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 740 unsigned long _nfct; 741 #endif 742 unsigned int len, 743 data_len; 744 __u16 mac_len, 745 hdr_len; 746 747 /* Following fields are _not_ copied in __copy_skb_header() 748 * Note that queue_mapping is here mostly to fill a hole. 749 */ 750 __u16 queue_mapping; 751 752 /* if you move cloned around you also must adapt those constants */ 753 #ifdef __BIG_ENDIAN_BITFIELD 754 #define CLONED_MASK (1 << 7) 755 #else 756 #define CLONED_MASK 1 757 #endif 758 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset) 759 760 __u8 __cloned_offset[0]; 761 __u8 cloned:1, 762 nohdr:1, 763 fclone:2, 764 peeked:1, 765 head_frag:1, 766 pfmemalloc:1; 767 #ifdef CONFIG_SKB_EXTENSIONS 768 __u8 active_extensions; 769 #endif 770 /* fields enclosed in headers_start/headers_end are copied 771 * using a single memcpy() in __copy_skb_header() 772 */ 773 /* private: */ 774 __u32 headers_start[0]; 775 /* public: */ 776 777 /* if you move pkt_type around you also must adapt those constants */ 778 #ifdef __BIG_ENDIAN_BITFIELD 779 #define PKT_TYPE_MAX (7 << 5) 780 #else 781 #define PKT_TYPE_MAX 7 782 #endif 783 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 784 785 __u8 __pkt_type_offset[0]; 786 __u8 pkt_type:3; 787 __u8 ignore_df:1; 788 __u8 nf_trace:1; 789 __u8 ip_summed:2; 790 __u8 ooo_okay:1; 791 792 __u8 l4_hash:1; 793 __u8 sw_hash:1; 794 __u8 wifi_acked_valid:1; 795 __u8 wifi_acked:1; 796 __u8 no_fcs:1; 797 /* Indicates the inner headers are valid in the skbuff. */ 798 __u8 encapsulation:1; 799 __u8 encap_hdr_csum:1; 800 __u8 csum_valid:1; 801 802 #ifdef __BIG_ENDIAN_BITFIELD 803 #define PKT_VLAN_PRESENT_BIT 7 804 #else 805 #define PKT_VLAN_PRESENT_BIT 0 806 #endif 807 #define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset) 808 __u8 __pkt_vlan_present_offset[0]; 809 __u8 vlan_present:1; 810 __u8 csum_complete_sw:1; 811 __u8 csum_level:2; 812 __u8 csum_not_inet:1; 813 __u8 dst_pending_confirm:1; 814 #ifdef CONFIG_IPV6_NDISC_NODETYPE 815 __u8 ndisc_nodetype:2; 816 #endif 817 818 __u8 ipvs_property:1; 819 __u8 inner_protocol_type:1; 820 __u8 remcsum_offload:1; 821 #ifdef CONFIG_NET_SWITCHDEV 822 __u8 offload_fwd_mark:1; 823 __u8 offload_l3_fwd_mark:1; 824 #endif 825 #ifdef CONFIG_NET_CLS_ACT 826 __u8 tc_skip_classify:1; 827 __u8 tc_at_ingress:1; 828 __u8 tc_redirected:1; 829 __u8 tc_from_ingress:1; 830 #endif 831 #ifdef CONFIG_TLS_DEVICE 832 __u8 decrypted:1; 833 #endif 834 835 #ifdef CONFIG_NET_SCHED 836 __u16 tc_index; /* traffic control index */ 837 #endif 838 839 union { 840 __wsum csum; 841 struct { 842 __u16 csum_start; 843 __u16 csum_offset; 844 }; 845 }; 846 __u32 priority; 847 int skb_iif; 848 __u32 hash; 849 __be16 vlan_proto; 850 __u16 vlan_tci; 851 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 852 union { 853 unsigned int napi_id; 854 unsigned int sender_cpu; 855 }; 856 #endif 857 #ifdef CONFIG_NETWORK_SECMARK 858 __u32 secmark; 859 #endif 860 861 union { 862 __u32 mark; 863 __u32 reserved_tailroom; 864 }; 865 866 union { 867 __be16 inner_protocol; 868 __u8 inner_ipproto; 869 }; 870 871 __u16 inner_transport_header; 872 __u16 inner_network_header; 873 __u16 inner_mac_header; 874 875 __be16 protocol; 876 __u16 transport_header; 877 __u16 network_header; 878 __u16 mac_header; 879 880 /* private: */ 881 __u32 headers_end[0]; 882 /* public: */ 883 884 /* These elements must be at the end, see alloc_skb() for details. */ 885 sk_buff_data_t tail; 886 sk_buff_data_t end; 887 unsigned char *head, 888 *data; 889 unsigned int truesize; 890 refcount_t users; 891 892 #ifdef CONFIG_SKB_EXTENSIONS 893 /* only useable after checking ->active_extensions != 0 */ 894 struct skb_ext *extensions; 895 #endif 896 }; 897 898 #ifdef __KERNEL__ 899 /* 900 * Handling routines are only of interest to the kernel 901 */ 902 903 #define SKB_ALLOC_FCLONE 0x01 904 #define SKB_ALLOC_RX 0x02 905 #define SKB_ALLOC_NAPI 0x04 906 907 /** 908 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 909 * @skb: buffer 910 */ 911 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 912 { 913 return unlikely(skb->pfmemalloc); 914 } 915 916 /* 917 * skb might have a dst pointer attached, refcounted or not. 918 * _skb_refdst low order bit is set if refcount was _not_ taken 919 */ 920 #define SKB_DST_NOREF 1UL 921 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 922 923 #define SKB_NFCT_PTRMASK ~(7UL) 924 /** 925 * skb_dst - returns skb dst_entry 926 * @skb: buffer 927 * 928 * Returns skb dst_entry, regardless of reference taken or not. 929 */ 930 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 931 { 932 /* If refdst was not refcounted, check we still are in a 933 * rcu_read_lock section 934 */ 935 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 936 !rcu_read_lock_held() && 937 !rcu_read_lock_bh_held()); 938 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 939 } 940 941 /** 942 * skb_dst_set - sets skb dst 943 * @skb: buffer 944 * @dst: dst entry 945 * 946 * Sets skb dst, assuming a reference was taken on dst and should 947 * be released by skb_dst_drop() 948 */ 949 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 950 { 951 skb->_skb_refdst = (unsigned long)dst; 952 } 953 954 /** 955 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 956 * @skb: buffer 957 * @dst: dst entry 958 * 959 * Sets skb dst, assuming a reference was not taken on dst. 960 * If dst entry is cached, we do not take reference and dst_release 961 * will be avoided by refdst_drop. If dst entry is not cached, we take 962 * reference, so that last dst_release can destroy the dst immediately. 963 */ 964 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 965 { 966 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 967 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 968 } 969 970 /** 971 * skb_dst_is_noref - Test if skb dst isn't refcounted 972 * @skb: buffer 973 */ 974 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 975 { 976 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 977 } 978 979 /** 980 * skb_rtable - Returns the skb &rtable 981 * @skb: buffer 982 */ 983 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 984 { 985 return (struct rtable *)skb_dst(skb); 986 } 987 988 /* For mangling skb->pkt_type from user space side from applications 989 * such as nft, tc, etc, we only allow a conservative subset of 990 * possible pkt_types to be set. 991 */ 992 static inline bool skb_pkt_type_ok(u32 ptype) 993 { 994 return ptype <= PACKET_OTHERHOST; 995 } 996 997 /** 998 * skb_napi_id - Returns the skb's NAPI id 999 * @skb: buffer 1000 */ 1001 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1002 { 1003 #ifdef CONFIG_NET_RX_BUSY_POLL 1004 return skb->napi_id; 1005 #else 1006 return 0; 1007 #endif 1008 } 1009 1010 /** 1011 * skb_unref - decrement the skb's reference count 1012 * @skb: buffer 1013 * 1014 * Returns true if we can free the skb. 1015 */ 1016 static inline bool skb_unref(struct sk_buff *skb) 1017 { 1018 if (unlikely(!skb)) 1019 return false; 1020 if (likely(refcount_read(&skb->users) == 1)) 1021 smp_rmb(); 1022 else if (likely(!refcount_dec_and_test(&skb->users))) 1023 return false; 1024 1025 return true; 1026 } 1027 1028 void skb_release_head_state(struct sk_buff *skb); 1029 void kfree_skb(struct sk_buff *skb); 1030 void kfree_skb_list(struct sk_buff *segs); 1031 void skb_tx_error(struct sk_buff *skb); 1032 void consume_skb(struct sk_buff *skb); 1033 void __consume_stateless_skb(struct sk_buff *skb); 1034 void __kfree_skb(struct sk_buff *skb); 1035 extern struct kmem_cache *skbuff_head_cache; 1036 1037 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1038 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1039 bool *fragstolen, int *delta_truesize); 1040 1041 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1042 int node); 1043 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1044 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1045 struct sk_buff *build_skb_around(struct sk_buff *skb, 1046 void *data, unsigned int frag_size); 1047 1048 /** 1049 * alloc_skb - allocate a network buffer 1050 * @size: size to allocate 1051 * @priority: allocation mask 1052 * 1053 * This function is a convenient wrapper around __alloc_skb(). 1054 */ 1055 static inline struct sk_buff *alloc_skb(unsigned int size, 1056 gfp_t priority) 1057 { 1058 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1059 } 1060 1061 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1062 unsigned long data_len, 1063 int max_page_order, 1064 int *errcode, 1065 gfp_t gfp_mask); 1066 1067 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1068 struct sk_buff_fclones { 1069 struct sk_buff skb1; 1070 1071 struct sk_buff skb2; 1072 1073 refcount_t fclone_ref; 1074 }; 1075 1076 /** 1077 * skb_fclone_busy - check if fclone is busy 1078 * @sk: socket 1079 * @skb: buffer 1080 * 1081 * Returns true if skb is a fast clone, and its clone is not freed. 1082 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1083 * so we also check that this didnt happen. 1084 */ 1085 static inline bool skb_fclone_busy(const struct sock *sk, 1086 const struct sk_buff *skb) 1087 { 1088 const struct sk_buff_fclones *fclones; 1089 1090 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1091 1092 return skb->fclone == SKB_FCLONE_ORIG && 1093 refcount_read(&fclones->fclone_ref) > 1 && 1094 fclones->skb2.sk == sk; 1095 } 1096 1097 /** 1098 * alloc_skb_fclone - allocate a network buffer from fclone cache 1099 * @size: size to allocate 1100 * @priority: allocation mask 1101 * 1102 * This function is a convenient wrapper around __alloc_skb(). 1103 */ 1104 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1105 gfp_t priority) 1106 { 1107 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1108 } 1109 1110 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1111 void skb_headers_offset_update(struct sk_buff *skb, int off); 1112 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1113 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1114 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1115 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1116 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1117 gfp_t gfp_mask, bool fclone); 1118 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1119 gfp_t gfp_mask) 1120 { 1121 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1122 } 1123 1124 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1125 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1126 unsigned int headroom); 1127 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1128 int newtailroom, gfp_t priority); 1129 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1130 int offset, int len); 1131 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1132 int offset, int len); 1133 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1134 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1135 1136 /** 1137 * skb_pad - zero pad the tail of an skb 1138 * @skb: buffer to pad 1139 * @pad: space to pad 1140 * 1141 * Ensure that a buffer is followed by a padding area that is zero 1142 * filled. Used by network drivers which may DMA or transfer data 1143 * beyond the buffer end onto the wire. 1144 * 1145 * May return error in out of memory cases. The skb is freed on error. 1146 */ 1147 static inline int skb_pad(struct sk_buff *skb, int pad) 1148 { 1149 return __skb_pad(skb, pad, true); 1150 } 1151 #define dev_kfree_skb(a) consume_skb(a) 1152 1153 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1154 int offset, size_t size); 1155 1156 struct skb_seq_state { 1157 __u32 lower_offset; 1158 __u32 upper_offset; 1159 __u32 frag_idx; 1160 __u32 stepped_offset; 1161 struct sk_buff *root_skb; 1162 struct sk_buff *cur_skb; 1163 __u8 *frag_data; 1164 }; 1165 1166 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1167 unsigned int to, struct skb_seq_state *st); 1168 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1169 struct skb_seq_state *st); 1170 void skb_abort_seq_read(struct skb_seq_state *st); 1171 1172 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1173 unsigned int to, struct ts_config *config); 1174 1175 /* 1176 * Packet hash types specify the type of hash in skb_set_hash. 1177 * 1178 * Hash types refer to the protocol layer addresses which are used to 1179 * construct a packet's hash. The hashes are used to differentiate or identify 1180 * flows of the protocol layer for the hash type. Hash types are either 1181 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1182 * 1183 * Properties of hashes: 1184 * 1185 * 1) Two packets in different flows have different hash values 1186 * 2) Two packets in the same flow should have the same hash value 1187 * 1188 * A hash at a higher layer is considered to be more specific. A driver should 1189 * set the most specific hash possible. 1190 * 1191 * A driver cannot indicate a more specific hash than the layer at which a hash 1192 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1193 * 1194 * A driver may indicate a hash level which is less specific than the 1195 * actual layer the hash was computed on. For instance, a hash computed 1196 * at L4 may be considered an L3 hash. This should only be done if the 1197 * driver can't unambiguously determine that the HW computed the hash at 1198 * the higher layer. Note that the "should" in the second property above 1199 * permits this. 1200 */ 1201 enum pkt_hash_types { 1202 PKT_HASH_TYPE_NONE, /* Undefined type */ 1203 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1204 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1205 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1206 }; 1207 1208 static inline void skb_clear_hash(struct sk_buff *skb) 1209 { 1210 skb->hash = 0; 1211 skb->sw_hash = 0; 1212 skb->l4_hash = 0; 1213 } 1214 1215 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1216 { 1217 if (!skb->l4_hash) 1218 skb_clear_hash(skb); 1219 } 1220 1221 static inline void 1222 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1223 { 1224 skb->l4_hash = is_l4; 1225 skb->sw_hash = is_sw; 1226 skb->hash = hash; 1227 } 1228 1229 static inline void 1230 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1231 { 1232 /* Used by drivers to set hash from HW */ 1233 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1234 } 1235 1236 static inline void 1237 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1238 { 1239 __skb_set_hash(skb, hash, true, is_l4); 1240 } 1241 1242 void __skb_get_hash(struct sk_buff *skb); 1243 u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1244 u32 skb_get_poff(const struct sk_buff *skb); 1245 u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1246 const struct flow_keys_basic *keys, int hlen); 1247 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1248 void *data, int hlen_proto); 1249 1250 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1251 int thoff, u8 ip_proto) 1252 { 1253 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1254 } 1255 1256 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1257 const struct flow_dissector_key *key, 1258 unsigned int key_count); 1259 1260 #ifdef CONFIG_NET 1261 int skb_flow_dissector_prog_query(const union bpf_attr *attr, 1262 union bpf_attr __user *uattr); 1263 int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr, 1264 struct bpf_prog *prog); 1265 1266 int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr); 1267 #else 1268 static inline int skb_flow_dissector_prog_query(const union bpf_attr *attr, 1269 union bpf_attr __user *uattr) 1270 { 1271 return -EOPNOTSUPP; 1272 } 1273 1274 static inline int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr, 1275 struct bpf_prog *prog) 1276 { 1277 return -EOPNOTSUPP; 1278 } 1279 1280 static inline int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr) 1281 { 1282 return -EOPNOTSUPP; 1283 } 1284 #endif 1285 1286 struct bpf_flow_dissector; 1287 bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1288 __be16 proto, int nhoff, int hlen); 1289 1290 bool __skb_flow_dissect(const struct net *net, 1291 const struct sk_buff *skb, 1292 struct flow_dissector *flow_dissector, 1293 void *target_container, 1294 void *data, __be16 proto, int nhoff, int hlen, 1295 unsigned int flags); 1296 1297 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1298 struct flow_dissector *flow_dissector, 1299 void *target_container, unsigned int flags) 1300 { 1301 return __skb_flow_dissect(NULL, skb, flow_dissector, 1302 target_container, NULL, 0, 0, 0, flags); 1303 } 1304 1305 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1306 struct flow_keys *flow, 1307 unsigned int flags) 1308 { 1309 memset(flow, 0, sizeof(*flow)); 1310 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1311 flow, NULL, 0, 0, 0, flags); 1312 } 1313 1314 static inline bool 1315 skb_flow_dissect_flow_keys_basic(const struct net *net, 1316 const struct sk_buff *skb, 1317 struct flow_keys_basic *flow, void *data, 1318 __be16 proto, int nhoff, int hlen, 1319 unsigned int flags) 1320 { 1321 memset(flow, 0, sizeof(*flow)); 1322 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1323 data, proto, nhoff, hlen, flags); 1324 } 1325 1326 void 1327 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1328 struct flow_dissector *flow_dissector, 1329 void *target_container); 1330 1331 static inline __u32 skb_get_hash(struct sk_buff *skb) 1332 { 1333 if (!skb->l4_hash && !skb->sw_hash) 1334 __skb_get_hash(skb); 1335 1336 return skb->hash; 1337 } 1338 1339 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1340 { 1341 if (!skb->l4_hash && !skb->sw_hash) { 1342 struct flow_keys keys; 1343 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1344 1345 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1346 } 1347 1348 return skb->hash; 1349 } 1350 1351 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1352 1353 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1354 { 1355 return skb->hash; 1356 } 1357 1358 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1359 { 1360 to->hash = from->hash; 1361 to->sw_hash = from->sw_hash; 1362 to->l4_hash = from->l4_hash; 1363 }; 1364 1365 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1366 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1367 { 1368 return skb->head + skb->end; 1369 } 1370 1371 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1372 { 1373 return skb->end; 1374 } 1375 #else 1376 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1377 { 1378 return skb->end; 1379 } 1380 1381 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1382 { 1383 return skb->end - skb->head; 1384 } 1385 #endif 1386 1387 /* Internal */ 1388 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1389 1390 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1391 { 1392 return &skb_shinfo(skb)->hwtstamps; 1393 } 1394 1395 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1396 { 1397 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY; 1398 1399 return is_zcopy ? skb_uarg(skb) : NULL; 1400 } 1401 1402 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1403 bool *have_ref) 1404 { 1405 if (skb && uarg && !skb_zcopy(skb)) { 1406 if (unlikely(have_ref && *have_ref)) 1407 *have_ref = false; 1408 else 1409 sock_zerocopy_get(uarg); 1410 skb_shinfo(skb)->destructor_arg = uarg; 1411 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG; 1412 } 1413 } 1414 1415 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1416 { 1417 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1418 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG; 1419 } 1420 1421 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1422 { 1423 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1424 } 1425 1426 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1427 { 1428 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1429 } 1430 1431 /* Release a reference on a zerocopy structure */ 1432 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy) 1433 { 1434 struct ubuf_info *uarg = skb_zcopy(skb); 1435 1436 if (uarg) { 1437 if (uarg->callback == sock_zerocopy_callback) { 1438 uarg->zerocopy = uarg->zerocopy && zerocopy; 1439 sock_zerocopy_put(uarg); 1440 } else if (!skb_zcopy_is_nouarg(skb)) { 1441 uarg->callback(uarg, zerocopy); 1442 } 1443 1444 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1445 } 1446 } 1447 1448 /* Abort a zerocopy operation and revert zckey on error in send syscall */ 1449 static inline void skb_zcopy_abort(struct sk_buff *skb) 1450 { 1451 struct ubuf_info *uarg = skb_zcopy(skb); 1452 1453 if (uarg) { 1454 sock_zerocopy_put_abort(uarg, false); 1455 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1456 } 1457 } 1458 1459 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1460 { 1461 skb->next = NULL; 1462 } 1463 1464 static inline void skb_list_del_init(struct sk_buff *skb) 1465 { 1466 __list_del_entry(&skb->list); 1467 skb_mark_not_on_list(skb); 1468 } 1469 1470 /** 1471 * skb_queue_empty - check if a queue is empty 1472 * @list: queue head 1473 * 1474 * Returns true if the queue is empty, false otherwise. 1475 */ 1476 static inline int skb_queue_empty(const struct sk_buff_head *list) 1477 { 1478 return list->next == (const struct sk_buff *) list; 1479 } 1480 1481 /** 1482 * skb_queue_is_last - check if skb is the last entry in the queue 1483 * @list: queue head 1484 * @skb: buffer 1485 * 1486 * Returns true if @skb is the last buffer on the list. 1487 */ 1488 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1489 const struct sk_buff *skb) 1490 { 1491 return skb->next == (const struct sk_buff *) list; 1492 } 1493 1494 /** 1495 * skb_queue_is_first - check if skb is the first entry in the queue 1496 * @list: queue head 1497 * @skb: buffer 1498 * 1499 * Returns true if @skb is the first buffer on the list. 1500 */ 1501 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1502 const struct sk_buff *skb) 1503 { 1504 return skb->prev == (const struct sk_buff *) list; 1505 } 1506 1507 /** 1508 * skb_queue_next - return the next packet in the queue 1509 * @list: queue head 1510 * @skb: current buffer 1511 * 1512 * Return the next packet in @list after @skb. It is only valid to 1513 * call this if skb_queue_is_last() evaluates to false. 1514 */ 1515 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1516 const struct sk_buff *skb) 1517 { 1518 /* This BUG_ON may seem severe, but if we just return then we 1519 * are going to dereference garbage. 1520 */ 1521 BUG_ON(skb_queue_is_last(list, skb)); 1522 return skb->next; 1523 } 1524 1525 /** 1526 * skb_queue_prev - return the prev packet in the queue 1527 * @list: queue head 1528 * @skb: current buffer 1529 * 1530 * Return the prev packet in @list before @skb. It is only valid to 1531 * call this if skb_queue_is_first() evaluates to false. 1532 */ 1533 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1534 const struct sk_buff *skb) 1535 { 1536 /* This BUG_ON may seem severe, but if we just return then we 1537 * are going to dereference garbage. 1538 */ 1539 BUG_ON(skb_queue_is_first(list, skb)); 1540 return skb->prev; 1541 } 1542 1543 /** 1544 * skb_get - reference buffer 1545 * @skb: buffer to reference 1546 * 1547 * Makes another reference to a socket buffer and returns a pointer 1548 * to the buffer. 1549 */ 1550 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1551 { 1552 refcount_inc(&skb->users); 1553 return skb; 1554 } 1555 1556 /* 1557 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1558 */ 1559 1560 /** 1561 * skb_cloned - is the buffer a clone 1562 * @skb: buffer to check 1563 * 1564 * Returns true if the buffer was generated with skb_clone() and is 1565 * one of multiple shared copies of the buffer. Cloned buffers are 1566 * shared data so must not be written to under normal circumstances. 1567 */ 1568 static inline int skb_cloned(const struct sk_buff *skb) 1569 { 1570 return skb->cloned && 1571 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1572 } 1573 1574 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1575 { 1576 might_sleep_if(gfpflags_allow_blocking(pri)); 1577 1578 if (skb_cloned(skb)) 1579 return pskb_expand_head(skb, 0, 0, pri); 1580 1581 return 0; 1582 } 1583 1584 /** 1585 * skb_header_cloned - is the header a clone 1586 * @skb: buffer to check 1587 * 1588 * Returns true if modifying the header part of the buffer requires 1589 * the data to be copied. 1590 */ 1591 static inline int skb_header_cloned(const struct sk_buff *skb) 1592 { 1593 int dataref; 1594 1595 if (!skb->cloned) 1596 return 0; 1597 1598 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1599 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1600 return dataref != 1; 1601 } 1602 1603 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1604 { 1605 might_sleep_if(gfpflags_allow_blocking(pri)); 1606 1607 if (skb_header_cloned(skb)) 1608 return pskb_expand_head(skb, 0, 0, pri); 1609 1610 return 0; 1611 } 1612 1613 /** 1614 * __skb_header_release - release reference to header 1615 * @skb: buffer to operate on 1616 */ 1617 static inline void __skb_header_release(struct sk_buff *skb) 1618 { 1619 skb->nohdr = 1; 1620 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1621 } 1622 1623 1624 /** 1625 * skb_shared - is the buffer shared 1626 * @skb: buffer to check 1627 * 1628 * Returns true if more than one person has a reference to this 1629 * buffer. 1630 */ 1631 static inline int skb_shared(const struct sk_buff *skb) 1632 { 1633 return refcount_read(&skb->users) != 1; 1634 } 1635 1636 /** 1637 * skb_share_check - check if buffer is shared and if so clone it 1638 * @skb: buffer to check 1639 * @pri: priority for memory allocation 1640 * 1641 * If the buffer is shared the buffer is cloned and the old copy 1642 * drops a reference. A new clone with a single reference is returned. 1643 * If the buffer is not shared the original buffer is returned. When 1644 * being called from interrupt status or with spinlocks held pri must 1645 * be GFP_ATOMIC. 1646 * 1647 * NULL is returned on a memory allocation failure. 1648 */ 1649 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1650 { 1651 might_sleep_if(gfpflags_allow_blocking(pri)); 1652 if (skb_shared(skb)) { 1653 struct sk_buff *nskb = skb_clone(skb, pri); 1654 1655 if (likely(nskb)) 1656 consume_skb(skb); 1657 else 1658 kfree_skb(skb); 1659 skb = nskb; 1660 } 1661 return skb; 1662 } 1663 1664 /* 1665 * Copy shared buffers into a new sk_buff. We effectively do COW on 1666 * packets to handle cases where we have a local reader and forward 1667 * and a couple of other messy ones. The normal one is tcpdumping 1668 * a packet thats being forwarded. 1669 */ 1670 1671 /** 1672 * skb_unshare - make a copy of a shared buffer 1673 * @skb: buffer to check 1674 * @pri: priority for memory allocation 1675 * 1676 * If the socket buffer is a clone then this function creates a new 1677 * copy of the data, drops a reference count on the old copy and returns 1678 * the new copy with the reference count at 1. If the buffer is not a clone 1679 * the original buffer is returned. When called with a spinlock held or 1680 * from interrupt state @pri must be %GFP_ATOMIC 1681 * 1682 * %NULL is returned on a memory allocation failure. 1683 */ 1684 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1685 gfp_t pri) 1686 { 1687 might_sleep_if(gfpflags_allow_blocking(pri)); 1688 if (skb_cloned(skb)) { 1689 struct sk_buff *nskb = skb_copy(skb, pri); 1690 1691 /* Free our shared copy */ 1692 if (likely(nskb)) 1693 consume_skb(skb); 1694 else 1695 kfree_skb(skb); 1696 skb = nskb; 1697 } 1698 return skb; 1699 } 1700 1701 /** 1702 * skb_peek - peek at the head of an &sk_buff_head 1703 * @list_: list to peek at 1704 * 1705 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1706 * be careful with this one. A peek leaves the buffer on the 1707 * list and someone else may run off with it. You must hold 1708 * the appropriate locks or have a private queue to do this. 1709 * 1710 * Returns %NULL for an empty list or a pointer to the head element. 1711 * The reference count is not incremented and the reference is therefore 1712 * volatile. Use with caution. 1713 */ 1714 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1715 { 1716 struct sk_buff *skb = list_->next; 1717 1718 if (skb == (struct sk_buff *)list_) 1719 skb = NULL; 1720 return skb; 1721 } 1722 1723 /** 1724 * __skb_peek - peek at the head of a non-empty &sk_buff_head 1725 * @list_: list to peek at 1726 * 1727 * Like skb_peek(), but the caller knows that the list is not empty. 1728 */ 1729 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 1730 { 1731 return list_->next; 1732 } 1733 1734 /** 1735 * skb_peek_next - peek skb following the given one from a queue 1736 * @skb: skb to start from 1737 * @list_: list to peek at 1738 * 1739 * Returns %NULL when the end of the list is met or a pointer to the 1740 * next element. The reference count is not incremented and the 1741 * reference is therefore volatile. Use with caution. 1742 */ 1743 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1744 const struct sk_buff_head *list_) 1745 { 1746 struct sk_buff *next = skb->next; 1747 1748 if (next == (struct sk_buff *)list_) 1749 next = NULL; 1750 return next; 1751 } 1752 1753 /** 1754 * skb_peek_tail - peek at the tail of an &sk_buff_head 1755 * @list_: list to peek at 1756 * 1757 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1758 * be careful with this one. A peek leaves the buffer on the 1759 * list and someone else may run off with it. You must hold 1760 * the appropriate locks or have a private queue to do this. 1761 * 1762 * Returns %NULL for an empty list or a pointer to the tail element. 1763 * The reference count is not incremented and the reference is therefore 1764 * volatile. Use with caution. 1765 */ 1766 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1767 { 1768 struct sk_buff *skb = list_->prev; 1769 1770 if (skb == (struct sk_buff *)list_) 1771 skb = NULL; 1772 return skb; 1773 1774 } 1775 1776 /** 1777 * skb_queue_len - get queue length 1778 * @list_: list to measure 1779 * 1780 * Return the length of an &sk_buff queue. 1781 */ 1782 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1783 { 1784 return list_->qlen; 1785 } 1786 1787 /** 1788 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1789 * @list: queue to initialize 1790 * 1791 * This initializes only the list and queue length aspects of 1792 * an sk_buff_head object. This allows to initialize the list 1793 * aspects of an sk_buff_head without reinitializing things like 1794 * the spinlock. It can also be used for on-stack sk_buff_head 1795 * objects where the spinlock is known to not be used. 1796 */ 1797 static inline void __skb_queue_head_init(struct sk_buff_head *list) 1798 { 1799 list->prev = list->next = (struct sk_buff *)list; 1800 list->qlen = 0; 1801 } 1802 1803 /* 1804 * This function creates a split out lock class for each invocation; 1805 * this is needed for now since a whole lot of users of the skb-queue 1806 * infrastructure in drivers have different locking usage (in hardirq) 1807 * than the networking core (in softirq only). In the long run either the 1808 * network layer or drivers should need annotation to consolidate the 1809 * main types of usage into 3 classes. 1810 */ 1811 static inline void skb_queue_head_init(struct sk_buff_head *list) 1812 { 1813 spin_lock_init(&list->lock); 1814 __skb_queue_head_init(list); 1815 } 1816 1817 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1818 struct lock_class_key *class) 1819 { 1820 skb_queue_head_init(list); 1821 lockdep_set_class(&list->lock, class); 1822 } 1823 1824 /* 1825 * Insert an sk_buff on a list. 1826 * 1827 * The "__skb_xxxx()" functions are the non-atomic ones that 1828 * can only be called with interrupts disabled. 1829 */ 1830 static inline void __skb_insert(struct sk_buff *newsk, 1831 struct sk_buff *prev, struct sk_buff *next, 1832 struct sk_buff_head *list) 1833 { 1834 newsk->next = next; 1835 newsk->prev = prev; 1836 next->prev = prev->next = newsk; 1837 list->qlen++; 1838 } 1839 1840 static inline void __skb_queue_splice(const struct sk_buff_head *list, 1841 struct sk_buff *prev, 1842 struct sk_buff *next) 1843 { 1844 struct sk_buff *first = list->next; 1845 struct sk_buff *last = list->prev; 1846 1847 first->prev = prev; 1848 prev->next = first; 1849 1850 last->next = next; 1851 next->prev = last; 1852 } 1853 1854 /** 1855 * skb_queue_splice - join two skb lists, this is designed for stacks 1856 * @list: the new list to add 1857 * @head: the place to add it in the first list 1858 */ 1859 static inline void skb_queue_splice(const struct sk_buff_head *list, 1860 struct sk_buff_head *head) 1861 { 1862 if (!skb_queue_empty(list)) { 1863 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1864 head->qlen += list->qlen; 1865 } 1866 } 1867 1868 /** 1869 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1870 * @list: the new list to add 1871 * @head: the place to add it in the first list 1872 * 1873 * The list at @list is reinitialised 1874 */ 1875 static inline void skb_queue_splice_init(struct sk_buff_head *list, 1876 struct sk_buff_head *head) 1877 { 1878 if (!skb_queue_empty(list)) { 1879 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1880 head->qlen += list->qlen; 1881 __skb_queue_head_init(list); 1882 } 1883 } 1884 1885 /** 1886 * skb_queue_splice_tail - join two skb lists, each list being a queue 1887 * @list: the new list to add 1888 * @head: the place to add it in the first list 1889 */ 1890 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1891 struct sk_buff_head *head) 1892 { 1893 if (!skb_queue_empty(list)) { 1894 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1895 head->qlen += list->qlen; 1896 } 1897 } 1898 1899 /** 1900 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1901 * @list: the new list to add 1902 * @head: the place to add it in the first list 1903 * 1904 * Each of the lists is a queue. 1905 * The list at @list is reinitialised 1906 */ 1907 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1908 struct sk_buff_head *head) 1909 { 1910 if (!skb_queue_empty(list)) { 1911 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1912 head->qlen += list->qlen; 1913 __skb_queue_head_init(list); 1914 } 1915 } 1916 1917 /** 1918 * __skb_queue_after - queue a buffer at the list head 1919 * @list: list to use 1920 * @prev: place after this buffer 1921 * @newsk: buffer to queue 1922 * 1923 * Queue a buffer int the middle of a list. This function takes no locks 1924 * and you must therefore hold required locks before calling it. 1925 * 1926 * A buffer cannot be placed on two lists at the same time. 1927 */ 1928 static inline void __skb_queue_after(struct sk_buff_head *list, 1929 struct sk_buff *prev, 1930 struct sk_buff *newsk) 1931 { 1932 __skb_insert(newsk, prev, prev->next, list); 1933 } 1934 1935 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1936 struct sk_buff_head *list); 1937 1938 static inline void __skb_queue_before(struct sk_buff_head *list, 1939 struct sk_buff *next, 1940 struct sk_buff *newsk) 1941 { 1942 __skb_insert(newsk, next->prev, next, list); 1943 } 1944 1945 /** 1946 * __skb_queue_head - queue a buffer at the list head 1947 * @list: list to use 1948 * @newsk: buffer to queue 1949 * 1950 * Queue a buffer at the start of a list. This function takes no locks 1951 * and you must therefore hold required locks before calling it. 1952 * 1953 * A buffer cannot be placed on two lists at the same time. 1954 */ 1955 static inline void __skb_queue_head(struct sk_buff_head *list, 1956 struct sk_buff *newsk) 1957 { 1958 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1959 } 1960 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1961 1962 /** 1963 * __skb_queue_tail - queue a buffer at the list tail 1964 * @list: list to use 1965 * @newsk: buffer to queue 1966 * 1967 * Queue a buffer at the end of a list. This function takes no locks 1968 * and you must therefore hold required locks before calling it. 1969 * 1970 * A buffer cannot be placed on two lists at the same time. 1971 */ 1972 static inline void __skb_queue_tail(struct sk_buff_head *list, 1973 struct sk_buff *newsk) 1974 { 1975 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1976 } 1977 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1978 1979 /* 1980 * remove sk_buff from list. _Must_ be called atomically, and with 1981 * the list known.. 1982 */ 1983 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1984 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1985 { 1986 struct sk_buff *next, *prev; 1987 1988 list->qlen--; 1989 next = skb->next; 1990 prev = skb->prev; 1991 skb->next = skb->prev = NULL; 1992 next->prev = prev; 1993 prev->next = next; 1994 } 1995 1996 /** 1997 * __skb_dequeue - remove from the head of the queue 1998 * @list: list to dequeue from 1999 * 2000 * Remove the head of the list. This function does not take any locks 2001 * so must be used with appropriate locks held only. The head item is 2002 * returned or %NULL if the list is empty. 2003 */ 2004 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2005 { 2006 struct sk_buff *skb = skb_peek(list); 2007 if (skb) 2008 __skb_unlink(skb, list); 2009 return skb; 2010 } 2011 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2012 2013 /** 2014 * __skb_dequeue_tail - remove from the tail of the queue 2015 * @list: list to dequeue from 2016 * 2017 * Remove the tail of the list. This function does not take any locks 2018 * so must be used with appropriate locks held only. The tail item is 2019 * returned or %NULL if the list is empty. 2020 */ 2021 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2022 { 2023 struct sk_buff *skb = skb_peek_tail(list); 2024 if (skb) 2025 __skb_unlink(skb, list); 2026 return skb; 2027 } 2028 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2029 2030 2031 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2032 { 2033 return skb->data_len; 2034 } 2035 2036 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2037 { 2038 return skb->len - skb->data_len; 2039 } 2040 2041 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2042 { 2043 unsigned int i, len = 0; 2044 2045 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2046 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2047 return len; 2048 } 2049 2050 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2051 { 2052 return skb_headlen(skb) + __skb_pagelen(skb); 2053 } 2054 2055 /** 2056 * __skb_fill_page_desc - initialise a paged fragment in an skb 2057 * @skb: buffer containing fragment to be initialised 2058 * @i: paged fragment index to initialise 2059 * @page: the page to use for this fragment 2060 * @off: the offset to the data with @page 2061 * @size: the length of the data 2062 * 2063 * Initialises the @i'th fragment of @skb to point to &size bytes at 2064 * offset @off within @page. 2065 * 2066 * Does not take any additional reference on the fragment. 2067 */ 2068 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2069 struct page *page, int off, int size) 2070 { 2071 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2072 2073 /* 2074 * Propagate page pfmemalloc to the skb if we can. The problem is 2075 * that not all callers have unique ownership of the page but rely 2076 * on page_is_pfmemalloc doing the right thing(tm). 2077 */ 2078 frag->page.p = page; 2079 frag->page_offset = off; 2080 skb_frag_size_set(frag, size); 2081 2082 page = compound_head(page); 2083 if (page_is_pfmemalloc(page)) 2084 skb->pfmemalloc = true; 2085 } 2086 2087 /** 2088 * skb_fill_page_desc - initialise a paged fragment in an skb 2089 * @skb: buffer containing fragment to be initialised 2090 * @i: paged fragment index to initialise 2091 * @page: the page to use for this fragment 2092 * @off: the offset to the data with @page 2093 * @size: the length of the data 2094 * 2095 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2096 * @skb to point to @size bytes at offset @off within @page. In 2097 * addition updates @skb such that @i is the last fragment. 2098 * 2099 * Does not take any additional reference on the fragment. 2100 */ 2101 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2102 struct page *page, int off, int size) 2103 { 2104 __skb_fill_page_desc(skb, i, page, off, size); 2105 skb_shinfo(skb)->nr_frags = i + 1; 2106 } 2107 2108 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 2109 int size, unsigned int truesize); 2110 2111 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2112 unsigned int truesize); 2113 2114 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2115 2116 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2117 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2118 { 2119 return skb->head + skb->tail; 2120 } 2121 2122 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2123 { 2124 skb->tail = skb->data - skb->head; 2125 } 2126 2127 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2128 { 2129 skb_reset_tail_pointer(skb); 2130 skb->tail += offset; 2131 } 2132 2133 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2134 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2135 { 2136 return skb->tail; 2137 } 2138 2139 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2140 { 2141 skb->tail = skb->data; 2142 } 2143 2144 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2145 { 2146 skb->tail = skb->data + offset; 2147 } 2148 2149 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2150 2151 /* 2152 * Add data to an sk_buff 2153 */ 2154 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2155 void *skb_put(struct sk_buff *skb, unsigned int len); 2156 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2157 { 2158 void *tmp = skb_tail_pointer(skb); 2159 SKB_LINEAR_ASSERT(skb); 2160 skb->tail += len; 2161 skb->len += len; 2162 return tmp; 2163 } 2164 2165 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2166 { 2167 void *tmp = __skb_put(skb, len); 2168 2169 memset(tmp, 0, len); 2170 return tmp; 2171 } 2172 2173 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2174 unsigned int len) 2175 { 2176 void *tmp = __skb_put(skb, len); 2177 2178 memcpy(tmp, data, len); 2179 return tmp; 2180 } 2181 2182 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2183 { 2184 *(u8 *)__skb_put(skb, 1) = val; 2185 } 2186 2187 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2188 { 2189 void *tmp = skb_put(skb, len); 2190 2191 memset(tmp, 0, len); 2192 2193 return tmp; 2194 } 2195 2196 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2197 unsigned int len) 2198 { 2199 void *tmp = skb_put(skb, len); 2200 2201 memcpy(tmp, data, len); 2202 2203 return tmp; 2204 } 2205 2206 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2207 { 2208 *(u8 *)skb_put(skb, 1) = val; 2209 } 2210 2211 void *skb_push(struct sk_buff *skb, unsigned int len); 2212 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2213 { 2214 skb->data -= len; 2215 skb->len += len; 2216 return skb->data; 2217 } 2218 2219 void *skb_pull(struct sk_buff *skb, unsigned int len); 2220 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2221 { 2222 skb->len -= len; 2223 BUG_ON(skb->len < skb->data_len); 2224 return skb->data += len; 2225 } 2226 2227 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2228 { 2229 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2230 } 2231 2232 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2233 2234 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len) 2235 { 2236 if (len > skb_headlen(skb) && 2237 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 2238 return NULL; 2239 skb->len -= len; 2240 return skb->data += len; 2241 } 2242 2243 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2244 { 2245 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 2246 } 2247 2248 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 2249 { 2250 if (likely(len <= skb_headlen(skb))) 2251 return 1; 2252 if (unlikely(len > skb->len)) 2253 return 0; 2254 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 2255 } 2256 2257 void skb_condense(struct sk_buff *skb); 2258 2259 /** 2260 * skb_headroom - bytes at buffer head 2261 * @skb: buffer to check 2262 * 2263 * Return the number of bytes of free space at the head of an &sk_buff. 2264 */ 2265 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2266 { 2267 return skb->data - skb->head; 2268 } 2269 2270 /** 2271 * skb_tailroom - bytes at buffer end 2272 * @skb: buffer to check 2273 * 2274 * Return the number of bytes of free space at the tail of an sk_buff 2275 */ 2276 static inline int skb_tailroom(const struct sk_buff *skb) 2277 { 2278 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2279 } 2280 2281 /** 2282 * skb_availroom - bytes at buffer end 2283 * @skb: buffer to check 2284 * 2285 * Return the number of bytes of free space at the tail of an sk_buff 2286 * allocated by sk_stream_alloc() 2287 */ 2288 static inline int skb_availroom(const struct sk_buff *skb) 2289 { 2290 if (skb_is_nonlinear(skb)) 2291 return 0; 2292 2293 return skb->end - skb->tail - skb->reserved_tailroom; 2294 } 2295 2296 /** 2297 * skb_reserve - adjust headroom 2298 * @skb: buffer to alter 2299 * @len: bytes to move 2300 * 2301 * Increase the headroom of an empty &sk_buff by reducing the tail 2302 * room. This is only allowed for an empty buffer. 2303 */ 2304 static inline void skb_reserve(struct sk_buff *skb, int len) 2305 { 2306 skb->data += len; 2307 skb->tail += len; 2308 } 2309 2310 /** 2311 * skb_tailroom_reserve - adjust reserved_tailroom 2312 * @skb: buffer to alter 2313 * @mtu: maximum amount of headlen permitted 2314 * @needed_tailroom: minimum amount of reserved_tailroom 2315 * 2316 * Set reserved_tailroom so that headlen can be as large as possible but 2317 * not larger than mtu and tailroom cannot be smaller than 2318 * needed_tailroom. 2319 * The required headroom should already have been reserved before using 2320 * this function. 2321 */ 2322 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2323 unsigned int needed_tailroom) 2324 { 2325 SKB_LINEAR_ASSERT(skb); 2326 if (mtu < skb_tailroom(skb) - needed_tailroom) 2327 /* use at most mtu */ 2328 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2329 else 2330 /* use up to all available space */ 2331 skb->reserved_tailroom = needed_tailroom; 2332 } 2333 2334 #define ENCAP_TYPE_ETHER 0 2335 #define ENCAP_TYPE_IPPROTO 1 2336 2337 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2338 __be16 protocol) 2339 { 2340 skb->inner_protocol = protocol; 2341 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2342 } 2343 2344 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2345 __u8 ipproto) 2346 { 2347 skb->inner_ipproto = ipproto; 2348 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2349 } 2350 2351 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2352 { 2353 skb->inner_mac_header = skb->mac_header; 2354 skb->inner_network_header = skb->network_header; 2355 skb->inner_transport_header = skb->transport_header; 2356 } 2357 2358 static inline void skb_reset_mac_len(struct sk_buff *skb) 2359 { 2360 skb->mac_len = skb->network_header - skb->mac_header; 2361 } 2362 2363 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2364 *skb) 2365 { 2366 return skb->head + skb->inner_transport_header; 2367 } 2368 2369 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2370 { 2371 return skb_inner_transport_header(skb) - skb->data; 2372 } 2373 2374 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2375 { 2376 skb->inner_transport_header = skb->data - skb->head; 2377 } 2378 2379 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2380 const int offset) 2381 { 2382 skb_reset_inner_transport_header(skb); 2383 skb->inner_transport_header += offset; 2384 } 2385 2386 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2387 { 2388 return skb->head + skb->inner_network_header; 2389 } 2390 2391 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2392 { 2393 skb->inner_network_header = skb->data - skb->head; 2394 } 2395 2396 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2397 const int offset) 2398 { 2399 skb_reset_inner_network_header(skb); 2400 skb->inner_network_header += offset; 2401 } 2402 2403 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2404 { 2405 return skb->head + skb->inner_mac_header; 2406 } 2407 2408 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2409 { 2410 skb->inner_mac_header = skb->data - skb->head; 2411 } 2412 2413 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2414 const int offset) 2415 { 2416 skb_reset_inner_mac_header(skb); 2417 skb->inner_mac_header += offset; 2418 } 2419 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2420 { 2421 return skb->transport_header != (typeof(skb->transport_header))~0U; 2422 } 2423 2424 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2425 { 2426 return skb->head + skb->transport_header; 2427 } 2428 2429 static inline void skb_reset_transport_header(struct sk_buff *skb) 2430 { 2431 skb->transport_header = skb->data - skb->head; 2432 } 2433 2434 static inline void skb_set_transport_header(struct sk_buff *skb, 2435 const int offset) 2436 { 2437 skb_reset_transport_header(skb); 2438 skb->transport_header += offset; 2439 } 2440 2441 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2442 { 2443 return skb->head + skb->network_header; 2444 } 2445 2446 static inline void skb_reset_network_header(struct sk_buff *skb) 2447 { 2448 skb->network_header = skb->data - skb->head; 2449 } 2450 2451 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2452 { 2453 skb_reset_network_header(skb); 2454 skb->network_header += offset; 2455 } 2456 2457 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2458 { 2459 return skb->head + skb->mac_header; 2460 } 2461 2462 static inline int skb_mac_offset(const struct sk_buff *skb) 2463 { 2464 return skb_mac_header(skb) - skb->data; 2465 } 2466 2467 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2468 { 2469 return skb->network_header - skb->mac_header; 2470 } 2471 2472 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2473 { 2474 return skb->mac_header != (typeof(skb->mac_header))~0U; 2475 } 2476 2477 static inline void skb_reset_mac_header(struct sk_buff *skb) 2478 { 2479 skb->mac_header = skb->data - skb->head; 2480 } 2481 2482 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2483 { 2484 skb_reset_mac_header(skb); 2485 skb->mac_header += offset; 2486 } 2487 2488 static inline void skb_pop_mac_header(struct sk_buff *skb) 2489 { 2490 skb->mac_header = skb->network_header; 2491 } 2492 2493 static inline void skb_probe_transport_header(struct sk_buff *skb) 2494 { 2495 struct flow_keys_basic keys; 2496 2497 if (skb_transport_header_was_set(skb)) 2498 return; 2499 2500 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 2501 NULL, 0, 0, 0, 0)) 2502 skb_set_transport_header(skb, keys.control.thoff); 2503 } 2504 2505 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2506 { 2507 if (skb_mac_header_was_set(skb)) { 2508 const unsigned char *old_mac = skb_mac_header(skb); 2509 2510 skb_set_mac_header(skb, -skb->mac_len); 2511 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2512 } 2513 } 2514 2515 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2516 { 2517 return skb->csum_start - skb_headroom(skb); 2518 } 2519 2520 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2521 { 2522 return skb->head + skb->csum_start; 2523 } 2524 2525 static inline int skb_transport_offset(const struct sk_buff *skb) 2526 { 2527 return skb_transport_header(skb) - skb->data; 2528 } 2529 2530 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2531 { 2532 return skb->transport_header - skb->network_header; 2533 } 2534 2535 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2536 { 2537 return skb->inner_transport_header - skb->inner_network_header; 2538 } 2539 2540 static inline int skb_network_offset(const struct sk_buff *skb) 2541 { 2542 return skb_network_header(skb) - skb->data; 2543 } 2544 2545 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2546 { 2547 return skb_inner_network_header(skb) - skb->data; 2548 } 2549 2550 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2551 { 2552 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2553 } 2554 2555 /* 2556 * CPUs often take a performance hit when accessing unaligned memory 2557 * locations. The actual performance hit varies, it can be small if the 2558 * hardware handles it or large if we have to take an exception and fix it 2559 * in software. 2560 * 2561 * Since an ethernet header is 14 bytes network drivers often end up with 2562 * the IP header at an unaligned offset. The IP header can be aligned by 2563 * shifting the start of the packet by 2 bytes. Drivers should do this 2564 * with: 2565 * 2566 * skb_reserve(skb, NET_IP_ALIGN); 2567 * 2568 * The downside to this alignment of the IP header is that the DMA is now 2569 * unaligned. On some architectures the cost of an unaligned DMA is high 2570 * and this cost outweighs the gains made by aligning the IP header. 2571 * 2572 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2573 * to be overridden. 2574 */ 2575 #ifndef NET_IP_ALIGN 2576 #define NET_IP_ALIGN 2 2577 #endif 2578 2579 /* 2580 * The networking layer reserves some headroom in skb data (via 2581 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2582 * the header has to grow. In the default case, if the header has to grow 2583 * 32 bytes or less we avoid the reallocation. 2584 * 2585 * Unfortunately this headroom changes the DMA alignment of the resulting 2586 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2587 * on some architectures. An architecture can override this value, 2588 * perhaps setting it to a cacheline in size (since that will maintain 2589 * cacheline alignment of the DMA). It must be a power of 2. 2590 * 2591 * Various parts of the networking layer expect at least 32 bytes of 2592 * headroom, you should not reduce this. 2593 * 2594 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2595 * to reduce average number of cache lines per packet. 2596 * get_rps_cpus() for example only access one 64 bytes aligned block : 2597 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2598 */ 2599 #ifndef NET_SKB_PAD 2600 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2601 #endif 2602 2603 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2604 2605 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 2606 { 2607 if (WARN_ON(skb_is_nonlinear(skb))) 2608 return; 2609 skb->len = len; 2610 skb_set_tail_pointer(skb, len); 2611 } 2612 2613 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2614 { 2615 __skb_set_length(skb, len); 2616 } 2617 2618 void skb_trim(struct sk_buff *skb, unsigned int len); 2619 2620 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2621 { 2622 if (skb->data_len) 2623 return ___pskb_trim(skb, len); 2624 __skb_trim(skb, len); 2625 return 0; 2626 } 2627 2628 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2629 { 2630 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2631 } 2632 2633 /** 2634 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2635 * @skb: buffer to alter 2636 * @len: new length 2637 * 2638 * This is identical to pskb_trim except that the caller knows that 2639 * the skb is not cloned so we should never get an error due to out- 2640 * of-memory. 2641 */ 2642 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2643 { 2644 int err = pskb_trim(skb, len); 2645 BUG_ON(err); 2646 } 2647 2648 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 2649 { 2650 unsigned int diff = len - skb->len; 2651 2652 if (skb_tailroom(skb) < diff) { 2653 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 2654 GFP_ATOMIC); 2655 if (ret) 2656 return ret; 2657 } 2658 __skb_set_length(skb, len); 2659 return 0; 2660 } 2661 2662 /** 2663 * skb_orphan - orphan a buffer 2664 * @skb: buffer to orphan 2665 * 2666 * If a buffer currently has an owner then we call the owner's 2667 * destructor function and make the @skb unowned. The buffer continues 2668 * to exist but is no longer charged to its former owner. 2669 */ 2670 static inline void skb_orphan(struct sk_buff *skb) 2671 { 2672 if (skb->destructor) { 2673 skb->destructor(skb); 2674 skb->destructor = NULL; 2675 skb->sk = NULL; 2676 } else { 2677 BUG_ON(skb->sk); 2678 } 2679 } 2680 2681 /** 2682 * skb_orphan_frags - orphan the frags contained in a buffer 2683 * @skb: buffer to orphan frags from 2684 * @gfp_mask: allocation mask for replacement pages 2685 * 2686 * For each frag in the SKB which needs a destructor (i.e. has an 2687 * owner) create a copy of that frag and release the original 2688 * page by calling the destructor. 2689 */ 2690 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2691 { 2692 if (likely(!skb_zcopy(skb))) 2693 return 0; 2694 if (skb_uarg(skb)->callback == sock_zerocopy_callback) 2695 return 0; 2696 return skb_copy_ubufs(skb, gfp_mask); 2697 } 2698 2699 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 2700 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 2701 { 2702 if (likely(!skb_zcopy(skb))) 2703 return 0; 2704 return skb_copy_ubufs(skb, gfp_mask); 2705 } 2706 2707 /** 2708 * __skb_queue_purge - empty a list 2709 * @list: list to empty 2710 * 2711 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2712 * the list and one reference dropped. This function does not take the 2713 * list lock and the caller must hold the relevant locks to use it. 2714 */ 2715 static inline void __skb_queue_purge(struct sk_buff_head *list) 2716 { 2717 struct sk_buff *skb; 2718 while ((skb = __skb_dequeue(list)) != NULL) 2719 kfree_skb(skb); 2720 } 2721 void skb_queue_purge(struct sk_buff_head *list); 2722 2723 unsigned int skb_rbtree_purge(struct rb_root *root); 2724 2725 void *netdev_alloc_frag(unsigned int fragsz); 2726 2727 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2728 gfp_t gfp_mask); 2729 2730 /** 2731 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2732 * @dev: network device to receive on 2733 * @length: length to allocate 2734 * 2735 * Allocate a new &sk_buff and assign it a usage count of one. The 2736 * buffer has unspecified headroom built in. Users should allocate 2737 * the headroom they think they need without accounting for the 2738 * built in space. The built in space is used for optimisations. 2739 * 2740 * %NULL is returned if there is no free memory. Although this function 2741 * allocates memory it can be called from an interrupt. 2742 */ 2743 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2744 unsigned int length) 2745 { 2746 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2747 } 2748 2749 /* legacy helper around __netdev_alloc_skb() */ 2750 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2751 gfp_t gfp_mask) 2752 { 2753 return __netdev_alloc_skb(NULL, length, gfp_mask); 2754 } 2755 2756 /* legacy helper around netdev_alloc_skb() */ 2757 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2758 { 2759 return netdev_alloc_skb(NULL, length); 2760 } 2761 2762 2763 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2764 unsigned int length, gfp_t gfp) 2765 { 2766 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2767 2768 if (NET_IP_ALIGN && skb) 2769 skb_reserve(skb, NET_IP_ALIGN); 2770 return skb; 2771 } 2772 2773 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2774 unsigned int length) 2775 { 2776 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2777 } 2778 2779 static inline void skb_free_frag(void *addr) 2780 { 2781 page_frag_free(addr); 2782 } 2783 2784 void *napi_alloc_frag(unsigned int fragsz); 2785 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2786 unsigned int length, gfp_t gfp_mask); 2787 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2788 unsigned int length) 2789 { 2790 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2791 } 2792 void napi_consume_skb(struct sk_buff *skb, int budget); 2793 2794 void __kfree_skb_flush(void); 2795 void __kfree_skb_defer(struct sk_buff *skb); 2796 2797 /** 2798 * __dev_alloc_pages - allocate page for network Rx 2799 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2800 * @order: size of the allocation 2801 * 2802 * Allocate a new page. 2803 * 2804 * %NULL is returned if there is no free memory. 2805 */ 2806 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2807 unsigned int order) 2808 { 2809 /* This piece of code contains several assumptions. 2810 * 1. This is for device Rx, therefor a cold page is preferred. 2811 * 2. The expectation is the user wants a compound page. 2812 * 3. If requesting a order 0 page it will not be compound 2813 * due to the check to see if order has a value in prep_new_page 2814 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2815 * code in gfp_to_alloc_flags that should be enforcing this. 2816 */ 2817 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 2818 2819 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2820 } 2821 2822 static inline struct page *dev_alloc_pages(unsigned int order) 2823 { 2824 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 2825 } 2826 2827 /** 2828 * __dev_alloc_page - allocate a page for network Rx 2829 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2830 * 2831 * Allocate a new page. 2832 * 2833 * %NULL is returned if there is no free memory. 2834 */ 2835 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2836 { 2837 return __dev_alloc_pages(gfp_mask, 0); 2838 } 2839 2840 static inline struct page *dev_alloc_page(void) 2841 { 2842 return dev_alloc_pages(0); 2843 } 2844 2845 /** 2846 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2847 * @page: The page that was allocated from skb_alloc_page 2848 * @skb: The skb that may need pfmemalloc set 2849 */ 2850 static inline void skb_propagate_pfmemalloc(struct page *page, 2851 struct sk_buff *skb) 2852 { 2853 if (page_is_pfmemalloc(page)) 2854 skb->pfmemalloc = true; 2855 } 2856 2857 /** 2858 * skb_frag_page - retrieve the page referred to by a paged fragment 2859 * @frag: the paged fragment 2860 * 2861 * Returns the &struct page associated with @frag. 2862 */ 2863 static inline struct page *skb_frag_page(const skb_frag_t *frag) 2864 { 2865 return frag->page.p; 2866 } 2867 2868 /** 2869 * __skb_frag_ref - take an addition reference on a paged fragment. 2870 * @frag: the paged fragment 2871 * 2872 * Takes an additional reference on the paged fragment @frag. 2873 */ 2874 static inline void __skb_frag_ref(skb_frag_t *frag) 2875 { 2876 get_page(skb_frag_page(frag)); 2877 } 2878 2879 /** 2880 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2881 * @skb: the buffer 2882 * @f: the fragment offset. 2883 * 2884 * Takes an additional reference on the @f'th paged fragment of @skb. 2885 */ 2886 static inline void skb_frag_ref(struct sk_buff *skb, int f) 2887 { 2888 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2889 } 2890 2891 /** 2892 * __skb_frag_unref - release a reference on a paged fragment. 2893 * @frag: the paged fragment 2894 * 2895 * Releases a reference on the paged fragment @frag. 2896 */ 2897 static inline void __skb_frag_unref(skb_frag_t *frag) 2898 { 2899 put_page(skb_frag_page(frag)); 2900 } 2901 2902 /** 2903 * skb_frag_unref - release a reference on a paged fragment of an skb. 2904 * @skb: the buffer 2905 * @f: the fragment offset 2906 * 2907 * Releases a reference on the @f'th paged fragment of @skb. 2908 */ 2909 static inline void skb_frag_unref(struct sk_buff *skb, int f) 2910 { 2911 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2912 } 2913 2914 /** 2915 * skb_frag_address - gets the address of the data contained in a paged fragment 2916 * @frag: the paged fragment buffer 2917 * 2918 * Returns the address of the data within @frag. The page must already 2919 * be mapped. 2920 */ 2921 static inline void *skb_frag_address(const skb_frag_t *frag) 2922 { 2923 return page_address(skb_frag_page(frag)) + frag->page_offset; 2924 } 2925 2926 /** 2927 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2928 * @frag: the paged fragment buffer 2929 * 2930 * Returns the address of the data within @frag. Checks that the page 2931 * is mapped and returns %NULL otherwise. 2932 */ 2933 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2934 { 2935 void *ptr = page_address(skb_frag_page(frag)); 2936 if (unlikely(!ptr)) 2937 return NULL; 2938 2939 return ptr + frag->page_offset; 2940 } 2941 2942 /** 2943 * __skb_frag_set_page - sets the page contained in a paged fragment 2944 * @frag: the paged fragment 2945 * @page: the page to set 2946 * 2947 * Sets the fragment @frag to contain @page. 2948 */ 2949 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2950 { 2951 frag->page.p = page; 2952 } 2953 2954 /** 2955 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2956 * @skb: the buffer 2957 * @f: the fragment offset 2958 * @page: the page to set 2959 * 2960 * Sets the @f'th fragment of @skb to contain @page. 2961 */ 2962 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2963 struct page *page) 2964 { 2965 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2966 } 2967 2968 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2969 2970 /** 2971 * skb_frag_dma_map - maps a paged fragment via the DMA API 2972 * @dev: the device to map the fragment to 2973 * @frag: the paged fragment to map 2974 * @offset: the offset within the fragment (starting at the 2975 * fragment's own offset) 2976 * @size: the number of bytes to map 2977 * @dir: the direction of the mapping (``PCI_DMA_*``) 2978 * 2979 * Maps the page associated with @frag to @device. 2980 */ 2981 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2982 const skb_frag_t *frag, 2983 size_t offset, size_t size, 2984 enum dma_data_direction dir) 2985 { 2986 return dma_map_page(dev, skb_frag_page(frag), 2987 frag->page_offset + offset, size, dir); 2988 } 2989 2990 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2991 gfp_t gfp_mask) 2992 { 2993 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2994 } 2995 2996 2997 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2998 gfp_t gfp_mask) 2999 { 3000 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3001 } 3002 3003 3004 /** 3005 * skb_clone_writable - is the header of a clone writable 3006 * @skb: buffer to check 3007 * @len: length up to which to write 3008 * 3009 * Returns true if modifying the header part of the cloned buffer 3010 * does not requires the data to be copied. 3011 */ 3012 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3013 { 3014 return !skb_header_cloned(skb) && 3015 skb_headroom(skb) + len <= skb->hdr_len; 3016 } 3017 3018 static inline int skb_try_make_writable(struct sk_buff *skb, 3019 unsigned int write_len) 3020 { 3021 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3022 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3023 } 3024 3025 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3026 int cloned) 3027 { 3028 int delta = 0; 3029 3030 if (headroom > skb_headroom(skb)) 3031 delta = headroom - skb_headroom(skb); 3032 3033 if (delta || cloned) 3034 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3035 GFP_ATOMIC); 3036 return 0; 3037 } 3038 3039 /** 3040 * skb_cow - copy header of skb when it is required 3041 * @skb: buffer to cow 3042 * @headroom: needed headroom 3043 * 3044 * If the skb passed lacks sufficient headroom or its data part 3045 * is shared, data is reallocated. If reallocation fails, an error 3046 * is returned and original skb is not changed. 3047 * 3048 * The result is skb with writable area skb->head...skb->tail 3049 * and at least @headroom of space at head. 3050 */ 3051 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3052 { 3053 return __skb_cow(skb, headroom, skb_cloned(skb)); 3054 } 3055 3056 /** 3057 * skb_cow_head - skb_cow but only making the head writable 3058 * @skb: buffer to cow 3059 * @headroom: needed headroom 3060 * 3061 * This function is identical to skb_cow except that we replace the 3062 * skb_cloned check by skb_header_cloned. It should be used when 3063 * you only need to push on some header and do not need to modify 3064 * the data. 3065 */ 3066 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3067 { 3068 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3069 } 3070 3071 /** 3072 * skb_padto - pad an skbuff up to a minimal size 3073 * @skb: buffer to pad 3074 * @len: minimal length 3075 * 3076 * Pads up a buffer to ensure the trailing bytes exist and are 3077 * blanked. If the buffer already contains sufficient data it 3078 * is untouched. Otherwise it is extended. Returns zero on 3079 * success. The skb is freed on error. 3080 */ 3081 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3082 { 3083 unsigned int size = skb->len; 3084 if (likely(size >= len)) 3085 return 0; 3086 return skb_pad(skb, len - size); 3087 } 3088 3089 /** 3090 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3091 * @skb: buffer to pad 3092 * @len: minimal length 3093 * @free_on_error: free buffer on error 3094 * 3095 * Pads up a buffer to ensure the trailing bytes exist and are 3096 * blanked. If the buffer already contains sufficient data it 3097 * is untouched. Otherwise it is extended. Returns zero on 3098 * success. The skb is freed on error if @free_on_error is true. 3099 */ 3100 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len, 3101 bool free_on_error) 3102 { 3103 unsigned int size = skb->len; 3104 3105 if (unlikely(size < len)) { 3106 len -= size; 3107 if (__skb_pad(skb, len, free_on_error)) 3108 return -ENOMEM; 3109 __skb_put(skb, len); 3110 } 3111 return 0; 3112 } 3113 3114 /** 3115 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3116 * @skb: buffer to pad 3117 * @len: minimal length 3118 * 3119 * Pads up a buffer to ensure the trailing bytes exist and are 3120 * blanked. If the buffer already contains sufficient data it 3121 * is untouched. Otherwise it is extended. Returns zero on 3122 * success. The skb is freed on error. 3123 */ 3124 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 3125 { 3126 return __skb_put_padto(skb, len, true); 3127 } 3128 3129 static inline int skb_add_data(struct sk_buff *skb, 3130 struct iov_iter *from, int copy) 3131 { 3132 const int off = skb->len; 3133 3134 if (skb->ip_summed == CHECKSUM_NONE) { 3135 __wsum csum = 0; 3136 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 3137 &csum, from)) { 3138 skb->csum = csum_block_add(skb->csum, csum, off); 3139 return 0; 3140 } 3141 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3142 return 0; 3143 3144 __skb_trim(skb, off); 3145 return -EFAULT; 3146 } 3147 3148 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3149 const struct page *page, int off) 3150 { 3151 if (skb_zcopy(skb)) 3152 return false; 3153 if (i) { 3154 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 3155 3156 return page == skb_frag_page(frag) && 3157 off == frag->page_offset + skb_frag_size(frag); 3158 } 3159 return false; 3160 } 3161 3162 static inline int __skb_linearize(struct sk_buff *skb) 3163 { 3164 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3165 } 3166 3167 /** 3168 * skb_linearize - convert paged skb to linear one 3169 * @skb: buffer to linarize 3170 * 3171 * If there is no free memory -ENOMEM is returned, otherwise zero 3172 * is returned and the old skb data released. 3173 */ 3174 static inline int skb_linearize(struct sk_buff *skb) 3175 { 3176 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3177 } 3178 3179 /** 3180 * skb_has_shared_frag - can any frag be overwritten 3181 * @skb: buffer to test 3182 * 3183 * Return true if the skb has at least one frag that might be modified 3184 * by an external entity (as in vmsplice()/sendfile()) 3185 */ 3186 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3187 { 3188 return skb_is_nonlinear(skb) && 3189 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 3190 } 3191 3192 /** 3193 * skb_linearize_cow - make sure skb is linear and writable 3194 * @skb: buffer to process 3195 * 3196 * If there is no free memory -ENOMEM is returned, otherwise zero 3197 * is returned and the old skb data released. 3198 */ 3199 static inline int skb_linearize_cow(struct sk_buff *skb) 3200 { 3201 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3202 __skb_linearize(skb) : 0; 3203 } 3204 3205 static __always_inline void 3206 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3207 unsigned int off) 3208 { 3209 if (skb->ip_summed == CHECKSUM_COMPLETE) 3210 skb->csum = csum_block_sub(skb->csum, 3211 csum_partial(start, len, 0), off); 3212 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3213 skb_checksum_start_offset(skb) < 0) 3214 skb->ip_summed = CHECKSUM_NONE; 3215 } 3216 3217 /** 3218 * skb_postpull_rcsum - update checksum for received skb after pull 3219 * @skb: buffer to update 3220 * @start: start of data before pull 3221 * @len: length of data pulled 3222 * 3223 * After doing a pull on a received packet, you need to call this to 3224 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3225 * CHECKSUM_NONE so that it can be recomputed from scratch. 3226 */ 3227 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3228 const void *start, unsigned int len) 3229 { 3230 __skb_postpull_rcsum(skb, start, len, 0); 3231 } 3232 3233 static __always_inline void 3234 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3235 unsigned int off) 3236 { 3237 if (skb->ip_summed == CHECKSUM_COMPLETE) 3238 skb->csum = csum_block_add(skb->csum, 3239 csum_partial(start, len, 0), off); 3240 } 3241 3242 /** 3243 * skb_postpush_rcsum - update checksum for received skb after push 3244 * @skb: buffer to update 3245 * @start: start of data after push 3246 * @len: length of data pushed 3247 * 3248 * After doing a push on a received packet, you need to call this to 3249 * update the CHECKSUM_COMPLETE checksum. 3250 */ 3251 static inline void skb_postpush_rcsum(struct sk_buff *skb, 3252 const void *start, unsigned int len) 3253 { 3254 __skb_postpush_rcsum(skb, start, len, 0); 3255 } 3256 3257 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3258 3259 /** 3260 * skb_push_rcsum - push skb and update receive checksum 3261 * @skb: buffer to update 3262 * @len: length of data pulled 3263 * 3264 * This function performs an skb_push on the packet and updates 3265 * the CHECKSUM_COMPLETE checksum. It should be used on 3266 * receive path processing instead of skb_push unless you know 3267 * that the checksum difference is zero (e.g., a valid IP header) 3268 * or you are setting ip_summed to CHECKSUM_NONE. 3269 */ 3270 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3271 { 3272 skb_push(skb, len); 3273 skb_postpush_rcsum(skb, skb->data, len); 3274 return skb->data; 3275 } 3276 3277 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 3278 /** 3279 * pskb_trim_rcsum - trim received skb and update checksum 3280 * @skb: buffer to trim 3281 * @len: new length 3282 * 3283 * This is exactly the same as pskb_trim except that it ensures the 3284 * checksum of received packets are still valid after the operation. 3285 * It can change skb pointers. 3286 */ 3287 3288 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3289 { 3290 if (likely(len >= skb->len)) 3291 return 0; 3292 return pskb_trim_rcsum_slow(skb, len); 3293 } 3294 3295 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3296 { 3297 if (skb->ip_summed == CHECKSUM_COMPLETE) 3298 skb->ip_summed = CHECKSUM_NONE; 3299 __skb_trim(skb, len); 3300 return 0; 3301 } 3302 3303 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3304 { 3305 if (skb->ip_summed == CHECKSUM_COMPLETE) 3306 skb->ip_summed = CHECKSUM_NONE; 3307 return __skb_grow(skb, len); 3308 } 3309 3310 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3311 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 3312 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 3313 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3314 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3315 3316 #define skb_queue_walk(queue, skb) \ 3317 for (skb = (queue)->next; \ 3318 skb != (struct sk_buff *)(queue); \ 3319 skb = skb->next) 3320 3321 #define skb_queue_walk_safe(queue, skb, tmp) \ 3322 for (skb = (queue)->next, tmp = skb->next; \ 3323 skb != (struct sk_buff *)(queue); \ 3324 skb = tmp, tmp = skb->next) 3325 3326 #define skb_queue_walk_from(queue, skb) \ 3327 for (; skb != (struct sk_buff *)(queue); \ 3328 skb = skb->next) 3329 3330 #define skb_rbtree_walk(skb, root) \ 3331 for (skb = skb_rb_first(root); skb != NULL; \ 3332 skb = skb_rb_next(skb)) 3333 3334 #define skb_rbtree_walk_from(skb) \ 3335 for (; skb != NULL; \ 3336 skb = skb_rb_next(skb)) 3337 3338 #define skb_rbtree_walk_from_safe(skb, tmp) \ 3339 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3340 skb = tmp) 3341 3342 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 3343 for (tmp = skb->next; \ 3344 skb != (struct sk_buff *)(queue); \ 3345 skb = tmp, tmp = skb->next) 3346 3347 #define skb_queue_reverse_walk(queue, skb) \ 3348 for (skb = (queue)->prev; \ 3349 skb != (struct sk_buff *)(queue); \ 3350 skb = skb->prev) 3351 3352 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3353 for (skb = (queue)->prev, tmp = skb->prev; \ 3354 skb != (struct sk_buff *)(queue); \ 3355 skb = tmp, tmp = skb->prev) 3356 3357 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3358 for (tmp = skb->prev; \ 3359 skb != (struct sk_buff *)(queue); \ 3360 skb = tmp, tmp = skb->prev) 3361 3362 static inline bool skb_has_frag_list(const struct sk_buff *skb) 3363 { 3364 return skb_shinfo(skb)->frag_list != NULL; 3365 } 3366 3367 static inline void skb_frag_list_init(struct sk_buff *skb) 3368 { 3369 skb_shinfo(skb)->frag_list = NULL; 3370 } 3371 3372 #define skb_walk_frags(skb, iter) \ 3373 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3374 3375 3376 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, 3377 const struct sk_buff *skb); 3378 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3379 struct sk_buff_head *queue, 3380 unsigned int flags, 3381 void (*destructor)(struct sock *sk, 3382 struct sk_buff *skb), 3383 int *off, int *err, 3384 struct sk_buff **last); 3385 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags, 3386 void (*destructor)(struct sock *sk, 3387 struct sk_buff *skb), 3388 int *off, int *err, 3389 struct sk_buff **last); 3390 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 3391 void (*destructor)(struct sock *sk, 3392 struct sk_buff *skb), 3393 int *off, int *err); 3394 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 3395 int *err); 3396 __poll_t datagram_poll(struct file *file, struct socket *sock, 3397 struct poll_table_struct *wait); 3398 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3399 struct iov_iter *to, int size); 3400 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3401 struct msghdr *msg, int size) 3402 { 3403 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3404 } 3405 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3406 struct msghdr *msg); 3407 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 3408 struct iov_iter *to, int len, 3409 struct ahash_request *hash); 3410 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3411 struct iov_iter *from, int len); 3412 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3413 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3414 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3415 static inline void skb_free_datagram_locked(struct sock *sk, 3416 struct sk_buff *skb) 3417 { 3418 __skb_free_datagram_locked(sk, skb, 0); 3419 } 3420 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3421 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3422 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3423 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3424 int len, __wsum csum); 3425 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3426 struct pipe_inode_info *pipe, unsigned int len, 3427 unsigned int flags); 3428 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 3429 int len); 3430 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3431 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3432 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3433 int len, int hlen); 3434 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3435 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3436 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3437 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu); 3438 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len); 3439 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3440 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3441 int skb_ensure_writable(struct sk_buff *skb, int write_len); 3442 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3443 int skb_vlan_pop(struct sk_buff *skb); 3444 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3445 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3446 gfp_t gfp); 3447 3448 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3449 { 3450 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 3451 } 3452 3453 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3454 { 3455 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3456 } 3457 3458 struct skb_checksum_ops { 3459 __wsum (*update)(const void *mem, int len, __wsum wsum); 3460 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3461 }; 3462 3463 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 3464 3465 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3466 __wsum csum, const struct skb_checksum_ops *ops); 3467 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3468 __wsum csum); 3469 3470 static inline void * __must_check 3471 __skb_header_pointer(const struct sk_buff *skb, int offset, 3472 int len, void *data, int hlen, void *buffer) 3473 { 3474 if (hlen - offset >= len) 3475 return data + offset; 3476 3477 if (!skb || 3478 skb_copy_bits(skb, offset, buffer, len) < 0) 3479 return NULL; 3480 3481 return buffer; 3482 } 3483 3484 static inline void * __must_check 3485 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 3486 { 3487 return __skb_header_pointer(skb, offset, len, skb->data, 3488 skb_headlen(skb), buffer); 3489 } 3490 3491 /** 3492 * skb_needs_linearize - check if we need to linearize a given skb 3493 * depending on the given device features. 3494 * @skb: socket buffer to check 3495 * @features: net device features 3496 * 3497 * Returns true if either: 3498 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3499 * 2. skb is fragmented and the device does not support SG. 3500 */ 3501 static inline bool skb_needs_linearize(struct sk_buff *skb, 3502 netdev_features_t features) 3503 { 3504 return skb_is_nonlinear(skb) && 3505 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3506 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3507 } 3508 3509 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3510 void *to, 3511 const unsigned int len) 3512 { 3513 memcpy(to, skb->data, len); 3514 } 3515 3516 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3517 const int offset, void *to, 3518 const unsigned int len) 3519 { 3520 memcpy(to, skb->data + offset, len); 3521 } 3522 3523 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3524 const void *from, 3525 const unsigned int len) 3526 { 3527 memcpy(skb->data, from, len); 3528 } 3529 3530 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3531 const int offset, 3532 const void *from, 3533 const unsigned int len) 3534 { 3535 memcpy(skb->data + offset, from, len); 3536 } 3537 3538 void skb_init(void); 3539 3540 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3541 { 3542 return skb->tstamp; 3543 } 3544 3545 /** 3546 * skb_get_timestamp - get timestamp from a skb 3547 * @skb: skb to get stamp from 3548 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 3549 * 3550 * Timestamps are stored in the skb as offsets to a base timestamp. 3551 * This function converts the offset back to a struct timeval and stores 3552 * it in stamp. 3553 */ 3554 static inline void skb_get_timestamp(const struct sk_buff *skb, 3555 struct __kernel_old_timeval *stamp) 3556 { 3557 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 3558 } 3559 3560 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 3561 struct __kernel_sock_timeval *stamp) 3562 { 3563 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 3564 3565 stamp->tv_sec = ts.tv_sec; 3566 stamp->tv_usec = ts.tv_nsec / 1000; 3567 } 3568 3569 static inline void skb_get_timestampns(const struct sk_buff *skb, 3570 struct timespec *stamp) 3571 { 3572 *stamp = ktime_to_timespec(skb->tstamp); 3573 } 3574 3575 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 3576 struct __kernel_timespec *stamp) 3577 { 3578 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 3579 3580 stamp->tv_sec = ts.tv_sec; 3581 stamp->tv_nsec = ts.tv_nsec; 3582 } 3583 3584 static inline void __net_timestamp(struct sk_buff *skb) 3585 { 3586 skb->tstamp = ktime_get_real(); 3587 } 3588 3589 static inline ktime_t net_timedelta(ktime_t t) 3590 { 3591 return ktime_sub(ktime_get_real(), t); 3592 } 3593 3594 static inline ktime_t net_invalid_timestamp(void) 3595 { 3596 return 0; 3597 } 3598 3599 static inline u8 skb_metadata_len(const struct sk_buff *skb) 3600 { 3601 return skb_shinfo(skb)->meta_len; 3602 } 3603 3604 static inline void *skb_metadata_end(const struct sk_buff *skb) 3605 { 3606 return skb_mac_header(skb); 3607 } 3608 3609 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 3610 const struct sk_buff *skb_b, 3611 u8 meta_len) 3612 { 3613 const void *a = skb_metadata_end(skb_a); 3614 const void *b = skb_metadata_end(skb_b); 3615 /* Using more efficient varaiant than plain call to memcmp(). */ 3616 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64 3617 u64 diffs = 0; 3618 3619 switch (meta_len) { 3620 #define __it(x, op) (x -= sizeof(u##op)) 3621 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 3622 case 32: diffs |= __it_diff(a, b, 64); 3623 /* fall through */ 3624 case 24: diffs |= __it_diff(a, b, 64); 3625 /* fall through */ 3626 case 16: diffs |= __it_diff(a, b, 64); 3627 /* fall through */ 3628 case 8: diffs |= __it_diff(a, b, 64); 3629 break; 3630 case 28: diffs |= __it_diff(a, b, 64); 3631 /* fall through */ 3632 case 20: diffs |= __it_diff(a, b, 64); 3633 /* fall through */ 3634 case 12: diffs |= __it_diff(a, b, 64); 3635 /* fall through */ 3636 case 4: diffs |= __it_diff(a, b, 32); 3637 break; 3638 } 3639 return diffs; 3640 #else 3641 return memcmp(a - meta_len, b - meta_len, meta_len); 3642 #endif 3643 } 3644 3645 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 3646 const struct sk_buff *skb_b) 3647 { 3648 u8 len_a = skb_metadata_len(skb_a); 3649 u8 len_b = skb_metadata_len(skb_b); 3650 3651 if (!(len_a | len_b)) 3652 return false; 3653 3654 return len_a != len_b ? 3655 true : __skb_metadata_differs(skb_a, skb_b, len_a); 3656 } 3657 3658 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 3659 { 3660 skb_shinfo(skb)->meta_len = meta_len; 3661 } 3662 3663 static inline void skb_metadata_clear(struct sk_buff *skb) 3664 { 3665 skb_metadata_set(skb, 0); 3666 } 3667 3668 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3669 3670 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3671 3672 void skb_clone_tx_timestamp(struct sk_buff *skb); 3673 bool skb_defer_rx_timestamp(struct sk_buff *skb); 3674 3675 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3676 3677 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3678 { 3679 } 3680 3681 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3682 { 3683 return false; 3684 } 3685 3686 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3687 3688 /** 3689 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3690 * 3691 * PHY drivers may accept clones of transmitted packets for 3692 * timestamping via their phy_driver.txtstamp method. These drivers 3693 * must call this function to return the skb back to the stack with a 3694 * timestamp. 3695 * 3696 * @skb: clone of the the original outgoing packet 3697 * @hwtstamps: hardware time stamps 3698 * 3699 */ 3700 void skb_complete_tx_timestamp(struct sk_buff *skb, 3701 struct skb_shared_hwtstamps *hwtstamps); 3702 3703 void __skb_tstamp_tx(struct sk_buff *orig_skb, 3704 struct skb_shared_hwtstamps *hwtstamps, 3705 struct sock *sk, int tstype); 3706 3707 /** 3708 * skb_tstamp_tx - queue clone of skb with send time stamps 3709 * @orig_skb: the original outgoing packet 3710 * @hwtstamps: hardware time stamps, may be NULL if not available 3711 * 3712 * If the skb has a socket associated, then this function clones the 3713 * skb (thus sharing the actual data and optional structures), stores 3714 * the optional hardware time stamping information (if non NULL) or 3715 * generates a software time stamp (otherwise), then queues the clone 3716 * to the error queue of the socket. Errors are silently ignored. 3717 */ 3718 void skb_tstamp_tx(struct sk_buff *orig_skb, 3719 struct skb_shared_hwtstamps *hwtstamps); 3720 3721 /** 3722 * skb_tx_timestamp() - Driver hook for transmit timestamping 3723 * 3724 * Ethernet MAC Drivers should call this function in their hard_xmit() 3725 * function immediately before giving the sk_buff to the MAC hardware. 3726 * 3727 * Specifically, one should make absolutely sure that this function is 3728 * called before TX completion of this packet can trigger. Otherwise 3729 * the packet could potentially already be freed. 3730 * 3731 * @skb: A socket buffer. 3732 */ 3733 static inline void skb_tx_timestamp(struct sk_buff *skb) 3734 { 3735 skb_clone_tx_timestamp(skb); 3736 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 3737 skb_tstamp_tx(skb, NULL); 3738 } 3739 3740 /** 3741 * skb_complete_wifi_ack - deliver skb with wifi status 3742 * 3743 * @skb: the original outgoing packet 3744 * @acked: ack status 3745 * 3746 */ 3747 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3748 3749 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3750 __sum16 __skb_checksum_complete(struct sk_buff *skb); 3751 3752 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3753 { 3754 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3755 skb->csum_valid || 3756 (skb->ip_summed == CHECKSUM_PARTIAL && 3757 skb_checksum_start_offset(skb) >= 0)); 3758 } 3759 3760 /** 3761 * skb_checksum_complete - Calculate checksum of an entire packet 3762 * @skb: packet to process 3763 * 3764 * This function calculates the checksum over the entire packet plus 3765 * the value of skb->csum. The latter can be used to supply the 3766 * checksum of a pseudo header as used by TCP/UDP. It returns the 3767 * checksum. 3768 * 3769 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3770 * this function can be used to verify that checksum on received 3771 * packets. In that case the function should return zero if the 3772 * checksum is correct. In particular, this function will return zero 3773 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3774 * hardware has already verified the correctness of the checksum. 3775 */ 3776 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3777 { 3778 return skb_csum_unnecessary(skb) ? 3779 0 : __skb_checksum_complete(skb); 3780 } 3781 3782 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3783 { 3784 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3785 if (skb->csum_level == 0) 3786 skb->ip_summed = CHECKSUM_NONE; 3787 else 3788 skb->csum_level--; 3789 } 3790 } 3791 3792 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3793 { 3794 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3795 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3796 skb->csum_level++; 3797 } else if (skb->ip_summed == CHECKSUM_NONE) { 3798 skb->ip_summed = CHECKSUM_UNNECESSARY; 3799 skb->csum_level = 0; 3800 } 3801 } 3802 3803 /* Check if we need to perform checksum complete validation. 3804 * 3805 * Returns true if checksum complete is needed, false otherwise 3806 * (either checksum is unnecessary or zero checksum is allowed). 3807 */ 3808 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3809 bool zero_okay, 3810 __sum16 check) 3811 { 3812 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3813 skb->csum_valid = 1; 3814 __skb_decr_checksum_unnecessary(skb); 3815 return false; 3816 } 3817 3818 return true; 3819 } 3820 3821 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 3822 * in checksum_init. 3823 */ 3824 #define CHECKSUM_BREAK 76 3825 3826 /* Unset checksum-complete 3827 * 3828 * Unset checksum complete can be done when packet is being modified 3829 * (uncompressed for instance) and checksum-complete value is 3830 * invalidated. 3831 */ 3832 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3833 { 3834 if (skb->ip_summed == CHECKSUM_COMPLETE) 3835 skb->ip_summed = CHECKSUM_NONE; 3836 } 3837 3838 /* Validate (init) checksum based on checksum complete. 3839 * 3840 * Return values: 3841 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3842 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3843 * checksum is stored in skb->csum for use in __skb_checksum_complete 3844 * non-zero: value of invalid checksum 3845 * 3846 */ 3847 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3848 bool complete, 3849 __wsum psum) 3850 { 3851 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3852 if (!csum_fold(csum_add(psum, skb->csum))) { 3853 skb->csum_valid = 1; 3854 return 0; 3855 } 3856 } 3857 3858 skb->csum = psum; 3859 3860 if (complete || skb->len <= CHECKSUM_BREAK) { 3861 __sum16 csum; 3862 3863 csum = __skb_checksum_complete(skb); 3864 skb->csum_valid = !csum; 3865 return csum; 3866 } 3867 3868 return 0; 3869 } 3870 3871 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3872 { 3873 return 0; 3874 } 3875 3876 /* Perform checksum validate (init). Note that this is a macro since we only 3877 * want to calculate the pseudo header which is an input function if necessary. 3878 * First we try to validate without any computation (checksum unnecessary) and 3879 * then calculate based on checksum complete calling the function to compute 3880 * pseudo header. 3881 * 3882 * Return values: 3883 * 0: checksum is validated or try to in skb_checksum_complete 3884 * non-zero: value of invalid checksum 3885 */ 3886 #define __skb_checksum_validate(skb, proto, complete, \ 3887 zero_okay, check, compute_pseudo) \ 3888 ({ \ 3889 __sum16 __ret = 0; \ 3890 skb->csum_valid = 0; \ 3891 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3892 __ret = __skb_checksum_validate_complete(skb, \ 3893 complete, compute_pseudo(skb, proto)); \ 3894 __ret; \ 3895 }) 3896 3897 #define skb_checksum_init(skb, proto, compute_pseudo) \ 3898 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3899 3900 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3901 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3902 3903 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 3904 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3905 3906 #define skb_checksum_validate_zero_check(skb, proto, check, \ 3907 compute_pseudo) \ 3908 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3909 3910 #define skb_checksum_simple_validate(skb) \ 3911 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3912 3913 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3914 { 3915 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 3916 } 3917 3918 static inline void __skb_checksum_convert(struct sk_buff *skb, 3919 __sum16 check, __wsum pseudo) 3920 { 3921 skb->csum = ~pseudo; 3922 skb->ip_summed = CHECKSUM_COMPLETE; 3923 } 3924 3925 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3926 do { \ 3927 if (__skb_checksum_convert_check(skb)) \ 3928 __skb_checksum_convert(skb, check, \ 3929 compute_pseudo(skb, proto)); \ 3930 } while (0) 3931 3932 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3933 u16 start, u16 offset) 3934 { 3935 skb->ip_summed = CHECKSUM_PARTIAL; 3936 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3937 skb->csum_offset = offset - start; 3938 } 3939 3940 /* Update skbuf and packet to reflect the remote checksum offload operation. 3941 * When called, ptr indicates the starting point for skb->csum when 3942 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3943 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3944 */ 3945 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3946 int start, int offset, bool nopartial) 3947 { 3948 __wsum delta; 3949 3950 if (!nopartial) { 3951 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3952 return; 3953 } 3954 3955 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3956 __skb_checksum_complete(skb); 3957 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3958 } 3959 3960 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3961 3962 /* Adjust skb->csum since we changed the packet */ 3963 skb->csum = csum_add(skb->csum, delta); 3964 } 3965 3966 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 3967 { 3968 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 3969 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK); 3970 #else 3971 return NULL; 3972 #endif 3973 } 3974 3975 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3976 void nf_conntrack_destroy(struct nf_conntrack *nfct); 3977 static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3978 { 3979 if (nfct && atomic_dec_and_test(&nfct->use)) 3980 nf_conntrack_destroy(nfct); 3981 } 3982 static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3983 { 3984 if (nfct) 3985 atomic_inc(&nfct->use); 3986 } 3987 #endif 3988 3989 #ifdef CONFIG_SKB_EXTENSIONS 3990 enum skb_ext_id { 3991 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3992 SKB_EXT_BRIDGE_NF, 3993 #endif 3994 #ifdef CONFIG_XFRM 3995 SKB_EXT_SEC_PATH, 3996 #endif 3997 SKB_EXT_NUM, /* must be last */ 3998 }; 3999 4000 /** 4001 * struct skb_ext - sk_buff extensions 4002 * @refcnt: 1 on allocation, deallocated on 0 4003 * @offset: offset to add to @data to obtain extension address 4004 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4005 * @data: start of extension data, variable sized 4006 * 4007 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4008 * to use 'u8' types while allowing up to 2kb worth of extension data. 4009 */ 4010 struct skb_ext { 4011 refcount_t refcnt; 4012 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4013 u8 chunks; /* same */ 4014 char data[0] __aligned(8); 4015 }; 4016 4017 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4018 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4019 void __skb_ext_put(struct skb_ext *ext); 4020 4021 static inline void skb_ext_put(struct sk_buff *skb) 4022 { 4023 if (skb->active_extensions) 4024 __skb_ext_put(skb->extensions); 4025 } 4026 4027 static inline void __skb_ext_copy(struct sk_buff *dst, 4028 const struct sk_buff *src) 4029 { 4030 dst->active_extensions = src->active_extensions; 4031 4032 if (src->active_extensions) { 4033 struct skb_ext *ext = src->extensions; 4034 4035 refcount_inc(&ext->refcnt); 4036 dst->extensions = ext; 4037 } 4038 } 4039 4040 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4041 { 4042 skb_ext_put(dst); 4043 __skb_ext_copy(dst, src); 4044 } 4045 4046 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4047 { 4048 return !!ext->offset[i]; 4049 } 4050 4051 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4052 { 4053 return skb->active_extensions & (1 << id); 4054 } 4055 4056 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4057 { 4058 if (skb_ext_exist(skb, id)) 4059 __skb_ext_del(skb, id); 4060 } 4061 4062 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4063 { 4064 if (skb_ext_exist(skb, id)) { 4065 struct skb_ext *ext = skb->extensions; 4066 4067 return (void *)ext + (ext->offset[id] << 3); 4068 } 4069 4070 return NULL; 4071 } 4072 #else 4073 static inline void skb_ext_put(struct sk_buff *skb) {} 4074 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4075 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4076 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4077 #endif /* CONFIG_SKB_EXTENSIONS */ 4078 4079 static inline void nf_reset(struct sk_buff *skb) 4080 { 4081 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4082 nf_conntrack_put(skb_nfct(skb)); 4083 skb->_nfct = 0; 4084 #endif 4085 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4086 skb_ext_del(skb, SKB_EXT_BRIDGE_NF); 4087 #endif 4088 } 4089 4090 static inline void nf_reset_trace(struct sk_buff *skb) 4091 { 4092 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4093 skb->nf_trace = 0; 4094 #endif 4095 } 4096 4097 static inline void ipvs_reset(struct sk_buff *skb) 4098 { 4099 #if IS_ENABLED(CONFIG_IP_VS) 4100 skb->ipvs_property = 0; 4101 #endif 4102 } 4103 4104 /* Note: This doesn't put any conntrack info in dst. */ 4105 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4106 bool copy) 4107 { 4108 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4109 dst->_nfct = src->_nfct; 4110 nf_conntrack_get(skb_nfct(src)); 4111 #endif 4112 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4113 if (copy) 4114 dst->nf_trace = src->nf_trace; 4115 #endif 4116 } 4117 4118 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 4119 { 4120 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4121 nf_conntrack_put(skb_nfct(dst)); 4122 #endif 4123 __nf_copy(dst, src, true); 4124 } 4125 4126 #ifdef CONFIG_NETWORK_SECMARK 4127 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4128 { 4129 to->secmark = from->secmark; 4130 } 4131 4132 static inline void skb_init_secmark(struct sk_buff *skb) 4133 { 4134 skb->secmark = 0; 4135 } 4136 #else 4137 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4138 { } 4139 4140 static inline void skb_init_secmark(struct sk_buff *skb) 4141 { } 4142 #endif 4143 4144 static inline int secpath_exists(const struct sk_buff *skb) 4145 { 4146 #ifdef CONFIG_XFRM 4147 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 4148 #else 4149 return 0; 4150 #endif 4151 } 4152 4153 static inline bool skb_irq_freeable(const struct sk_buff *skb) 4154 { 4155 return !skb->destructor && 4156 !secpath_exists(skb) && 4157 !skb_nfct(skb) && 4158 !skb->_skb_refdst && 4159 !skb_has_frag_list(skb); 4160 } 4161 4162 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 4163 { 4164 skb->queue_mapping = queue_mapping; 4165 } 4166 4167 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 4168 { 4169 return skb->queue_mapping; 4170 } 4171 4172 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 4173 { 4174 to->queue_mapping = from->queue_mapping; 4175 } 4176 4177 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 4178 { 4179 skb->queue_mapping = rx_queue + 1; 4180 } 4181 4182 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 4183 { 4184 return skb->queue_mapping - 1; 4185 } 4186 4187 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 4188 { 4189 return skb->queue_mapping != 0; 4190 } 4191 4192 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 4193 { 4194 skb->dst_pending_confirm = val; 4195 } 4196 4197 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 4198 { 4199 return skb->dst_pending_confirm != 0; 4200 } 4201 4202 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 4203 { 4204 #ifdef CONFIG_XFRM 4205 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 4206 #else 4207 return NULL; 4208 #endif 4209 } 4210 4211 /* Keeps track of mac header offset relative to skb->head. 4212 * It is useful for TSO of Tunneling protocol. e.g. GRE. 4213 * For non-tunnel skb it points to skb_mac_header() and for 4214 * tunnel skb it points to outer mac header. 4215 * Keeps track of level of encapsulation of network headers. 4216 */ 4217 struct skb_gso_cb { 4218 union { 4219 int mac_offset; 4220 int data_offset; 4221 }; 4222 int encap_level; 4223 __wsum csum; 4224 __u16 csum_start; 4225 }; 4226 #define SKB_SGO_CB_OFFSET 32 4227 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 4228 4229 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 4230 { 4231 return (skb_mac_header(inner_skb) - inner_skb->head) - 4232 SKB_GSO_CB(inner_skb)->mac_offset; 4233 } 4234 4235 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 4236 { 4237 int new_headroom, headroom; 4238 int ret; 4239 4240 headroom = skb_headroom(skb); 4241 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 4242 if (ret) 4243 return ret; 4244 4245 new_headroom = skb_headroom(skb); 4246 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 4247 return 0; 4248 } 4249 4250 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 4251 { 4252 /* Do not update partial checksums if remote checksum is enabled. */ 4253 if (skb->remcsum_offload) 4254 return; 4255 4256 SKB_GSO_CB(skb)->csum = res; 4257 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 4258 } 4259 4260 /* Compute the checksum for a gso segment. First compute the checksum value 4261 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 4262 * then add in skb->csum (checksum from csum_start to end of packet). 4263 * skb->csum and csum_start are then updated to reflect the checksum of the 4264 * resultant packet starting from the transport header-- the resultant checksum 4265 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 4266 * header. 4267 */ 4268 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 4269 { 4270 unsigned char *csum_start = skb_transport_header(skb); 4271 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 4272 __wsum partial = SKB_GSO_CB(skb)->csum; 4273 4274 SKB_GSO_CB(skb)->csum = res; 4275 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 4276 4277 return csum_fold(csum_partial(csum_start, plen, partial)); 4278 } 4279 4280 static inline bool skb_is_gso(const struct sk_buff *skb) 4281 { 4282 return skb_shinfo(skb)->gso_size; 4283 } 4284 4285 /* Note: Should be called only if skb_is_gso(skb) is true */ 4286 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4287 { 4288 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4289 } 4290 4291 /* Note: Should be called only if skb_is_gso(skb) is true */ 4292 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 4293 { 4294 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 4295 } 4296 4297 /* Note: Should be called only if skb_is_gso(skb) is true */ 4298 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 4299 { 4300 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 4301 } 4302 4303 static inline void skb_gso_reset(struct sk_buff *skb) 4304 { 4305 skb_shinfo(skb)->gso_size = 0; 4306 skb_shinfo(skb)->gso_segs = 0; 4307 skb_shinfo(skb)->gso_type = 0; 4308 } 4309 4310 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 4311 u16 increment) 4312 { 4313 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4314 return; 4315 shinfo->gso_size += increment; 4316 } 4317 4318 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 4319 u16 decrement) 4320 { 4321 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4322 return; 4323 shinfo->gso_size -= decrement; 4324 } 4325 4326 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 4327 4328 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 4329 { 4330 /* LRO sets gso_size but not gso_type, whereas if GSO is really 4331 * wanted then gso_type will be set. */ 4332 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4333 4334 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 4335 unlikely(shinfo->gso_type == 0)) { 4336 __skb_warn_lro_forwarding(skb); 4337 return true; 4338 } 4339 return false; 4340 } 4341 4342 static inline void skb_forward_csum(struct sk_buff *skb) 4343 { 4344 /* Unfortunately we don't support this one. Any brave souls? */ 4345 if (skb->ip_summed == CHECKSUM_COMPLETE) 4346 skb->ip_summed = CHECKSUM_NONE; 4347 } 4348 4349 /** 4350 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 4351 * @skb: skb to check 4352 * 4353 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 4354 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 4355 * use this helper, to document places where we make this assertion. 4356 */ 4357 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 4358 { 4359 #ifdef DEBUG 4360 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 4361 #endif 4362 } 4363 4364 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 4365 4366 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 4367 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4368 unsigned int transport_len, 4369 __sum16(*skb_chkf)(struct sk_buff *skb)); 4370 4371 /** 4372 * skb_head_is_locked - Determine if the skb->head is locked down 4373 * @skb: skb to check 4374 * 4375 * The head on skbs build around a head frag can be removed if they are 4376 * not cloned. This function returns true if the skb head is locked down 4377 * due to either being allocated via kmalloc, or by being a clone with 4378 * multiple references to the head. 4379 */ 4380 static inline bool skb_head_is_locked(const struct sk_buff *skb) 4381 { 4382 return !skb->head_frag || skb_cloned(skb); 4383 } 4384 4385 /* Local Checksum Offload. 4386 * Compute outer checksum based on the assumption that the 4387 * inner checksum will be offloaded later. 4388 * See Documentation/networking/checksum-offloads.rst for 4389 * explanation of how this works. 4390 * Fill in outer checksum adjustment (e.g. with sum of outer 4391 * pseudo-header) before calling. 4392 * Also ensure that inner checksum is in linear data area. 4393 */ 4394 static inline __wsum lco_csum(struct sk_buff *skb) 4395 { 4396 unsigned char *csum_start = skb_checksum_start(skb); 4397 unsigned char *l4_hdr = skb_transport_header(skb); 4398 __wsum partial; 4399 4400 /* Start with complement of inner checksum adjustment */ 4401 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 4402 skb->csum_offset)); 4403 4404 /* Add in checksum of our headers (incl. outer checksum 4405 * adjustment filled in by caller) and return result. 4406 */ 4407 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 4408 } 4409 4410 #endif /* __KERNEL__ */ 4411 #endif /* _LINUX_SKBUFF_H */ 4412