1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * Definitions for the 'struct sk_buff' memory handlers. 4 * 5 * Authors: 6 * Alan Cox, <[email protected]> 7 * Florian La Roche, <[email protected]> 8 */ 9 10 #ifndef _LINUX_SKBUFF_H 11 #define _LINUX_SKBUFF_H 12 13 #include <linux/kernel.h> 14 #include <linux/compiler.h> 15 #include <linux/time.h> 16 #include <linux/bug.h> 17 #include <linux/cache.h> 18 #include <linux/rbtree.h> 19 #include <linux/socket.h> 20 #include <linux/refcount.h> 21 22 #include <linux/atomic.h> 23 #include <asm/types.h> 24 #include <linux/spinlock.h> 25 #include <linux/net.h> 26 #include <linux/textsearch.h> 27 #include <net/checksum.h> 28 #include <linux/rcupdate.h> 29 #include <linux/hrtimer.h> 30 #include <linux/dma-mapping.h> 31 #include <linux/netdev_features.h> 32 #include <linux/sched.h> 33 #include <linux/sched/clock.h> 34 #include <net/flow_dissector.h> 35 #include <linux/splice.h> 36 #include <linux/in6.h> 37 #include <linux/if_packet.h> 38 #include <net/flow.h> 39 40 /* The interface for checksum offload between the stack and networking drivers 41 * is as follows... 42 * 43 * A. IP checksum related features 44 * 45 * Drivers advertise checksum offload capabilities in the features of a device. 46 * From the stack's point of view these are capabilities offered by the driver, 47 * a driver typically only advertises features that it is capable of offloading 48 * to its device. 49 * 50 * The checksum related features are: 51 * 52 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one 53 * IP (one's complement) checksum for any combination 54 * of protocols or protocol layering. The checksum is 55 * computed and set in a packet per the CHECKSUM_PARTIAL 56 * interface (see below). 57 * 58 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain 59 * TCP or UDP packets over IPv4. These are specifically 60 * unencapsulated packets of the form IPv4|TCP or 61 * IPv4|UDP where the Protocol field in the IPv4 header 62 * is TCP or UDP. The IPv4 header may contain IP options 63 * This feature cannot be set in features for a device 64 * with NETIF_F_HW_CSUM also set. This feature is being 65 * DEPRECATED (see below). 66 * 67 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain 68 * TCP or UDP packets over IPv6. These are specifically 69 * unencapsulated packets of the form IPv6|TCP or 70 * IPv4|UDP where the Next Header field in the IPv6 71 * header is either TCP or UDP. IPv6 extension headers 72 * are not supported with this feature. This feature 73 * cannot be set in features for a device with 74 * NETIF_F_HW_CSUM also set. This feature is being 75 * DEPRECATED (see below). 76 * 77 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload. 78 * This flag is used only used to disable the RX checksum 79 * feature for a device. The stack will accept receive 80 * checksum indication in packets received on a device 81 * regardless of whether NETIF_F_RXCSUM is set. 82 * 83 * B. Checksumming of received packets by device. Indication of checksum 84 * verification is in set skb->ip_summed. Possible values are: 85 * 86 * CHECKSUM_NONE: 87 * 88 * Device did not checksum this packet e.g. due to lack of capabilities. 89 * The packet contains full (though not verified) checksum in packet but 90 * not in skb->csum. Thus, skb->csum is undefined in this case. 91 * 92 * CHECKSUM_UNNECESSARY: 93 * 94 * The hardware you're dealing with doesn't calculate the full checksum 95 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 96 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY 97 * if their checksums are okay. skb->csum is still undefined in this case 98 * though. A driver or device must never modify the checksum field in the 99 * packet even if checksum is verified. 100 * 101 * CHECKSUM_UNNECESSARY is applicable to following protocols: 102 * TCP: IPv6 and IPv4. 103 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 104 * zero UDP checksum for either IPv4 or IPv6, the networking stack 105 * may perform further validation in this case. 106 * GRE: only if the checksum is present in the header. 107 * SCTP: indicates the CRC in SCTP header has been validated. 108 * FCOE: indicates the CRC in FC frame has been validated. 109 * 110 * skb->csum_level indicates the number of consecutive checksums found in 111 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY. 112 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 113 * and a device is able to verify the checksums for UDP (possibly zero), 114 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to 115 * two. If the device were only able to verify the UDP checksum and not 116 * GRE, either because it doesn't support GRE checksum of because GRE 117 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 118 * not considered in this case). 119 * 120 * CHECKSUM_COMPLETE: 121 * 122 * This is the most generic way. The device supplied checksum of the _whole_ 123 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the 124 * hardware doesn't need to parse L3/L4 headers to implement this. 125 * 126 * Notes: 127 * - Even if device supports only some protocols, but is able to produce 128 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 129 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 130 * 131 * CHECKSUM_PARTIAL: 132 * 133 * A checksum is set up to be offloaded to a device as described in the 134 * output description for CHECKSUM_PARTIAL. This may occur on a packet 135 * received directly from another Linux OS, e.g., a virtualized Linux kernel 136 * on the same host, or it may be set in the input path in GRO or remote 137 * checksum offload. For the purposes of checksum verification, the checksum 138 * referred to by skb->csum_start + skb->csum_offset and any preceding 139 * checksums in the packet are considered verified. Any checksums in the 140 * packet that are after the checksum being offloaded are not considered to 141 * be verified. 142 * 143 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload 144 * in the skb->ip_summed for a packet. Values are: 145 * 146 * CHECKSUM_PARTIAL: 147 * 148 * The driver is required to checksum the packet as seen by hard_start_xmit() 149 * from skb->csum_start up to the end, and to record/write the checksum at 150 * offset skb->csum_start + skb->csum_offset. A driver may verify that the 151 * csum_start and csum_offset values are valid values given the length and 152 * offset of the packet, however they should not attempt to validate that the 153 * checksum refers to a legitimate transport layer checksum-- it is the 154 * purview of the stack to validate that csum_start and csum_offset are set 155 * correctly. 156 * 157 * When the stack requests checksum offload for a packet, the driver MUST 158 * ensure that the checksum is set correctly. A driver can either offload the 159 * checksum calculation to the device, or call skb_checksum_help (in the case 160 * that the device does not support offload for a particular checksum). 161 * 162 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of 163 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate 164 * checksum offload capability. 165 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based 166 * on network device checksumming capabilities: if a packet does not match 167 * them, skb_checksum_help or skb_crc32c_help (depending on the value of 168 * csum_not_inet, see item D.) is called to resolve the checksum. 169 * 170 * CHECKSUM_NONE: 171 * 172 * The skb was already checksummed by the protocol, or a checksum is not 173 * required. 174 * 175 * CHECKSUM_UNNECESSARY: 176 * 177 * This has the same meaning on as CHECKSUM_NONE for checksum offload on 178 * output. 179 * 180 * CHECKSUM_COMPLETE: 181 * Not used in checksum output. If a driver observes a packet with this value 182 * set in skbuff, if should treat as CHECKSUM_NONE being set. 183 * 184 * D. Non-IP checksum (CRC) offloads 185 * 186 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of 187 * offloading the SCTP CRC in a packet. To perform this offload the stack 188 * will set set csum_start and csum_offset accordingly, set ip_summed to 189 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in 190 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c. 191 * A driver that supports both IP checksum offload and SCTP CRC32c offload 192 * must verify which offload is configured for a packet by testing the 193 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve 194 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 195 * 196 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of 197 * offloading the FCOE CRC in a packet. To perform this offload the stack 198 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 199 * accordingly. Note the there is no indication in the skbuff that the 200 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports 201 * both IP checksum offload and FCOE CRC offload must verify which offload 202 * is configured for a packet presumably by inspecting packet headers. 203 * 204 * E. Checksumming on output with GSO. 205 * 206 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload 207 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 208 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as 209 * part of the GSO operation is implied. If a checksum is being offloaded 210 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset 211 * are set to refer to the outermost checksum being offload (two offloaded 212 * checksums are possible with UDP encapsulation). 213 */ 214 215 /* Don't change this without changing skb_csum_unnecessary! */ 216 #define CHECKSUM_NONE 0 217 #define CHECKSUM_UNNECESSARY 1 218 #define CHECKSUM_COMPLETE 2 219 #define CHECKSUM_PARTIAL 3 220 221 /* Maximum value in skb->csum_level */ 222 #define SKB_MAX_CSUM_LEVEL 3 223 224 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 225 #define SKB_WITH_OVERHEAD(X) \ 226 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 227 #define SKB_MAX_ORDER(X, ORDER) \ 228 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 229 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 230 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 231 232 /* return minimum truesize of one skb containing X bytes of data */ 233 #define SKB_TRUESIZE(X) ((X) + \ 234 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 235 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 236 237 struct net_device; 238 struct scatterlist; 239 struct pipe_inode_info; 240 struct iov_iter; 241 struct napi_struct; 242 struct bpf_prog; 243 union bpf_attr; 244 struct skb_ext; 245 246 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 247 struct nf_conntrack { 248 atomic_t use; 249 }; 250 #endif 251 252 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 253 struct nf_bridge_info { 254 enum { 255 BRNF_PROTO_UNCHANGED, 256 BRNF_PROTO_8021Q, 257 BRNF_PROTO_PPPOE 258 } orig_proto:8; 259 u8 pkt_otherhost:1; 260 u8 in_prerouting:1; 261 u8 bridged_dnat:1; 262 __u16 frag_max_size; 263 struct net_device *physindev; 264 265 /* always valid & non-NULL from FORWARD on, for physdev match */ 266 struct net_device *physoutdev; 267 union { 268 /* prerouting: detect dnat in orig/reply direction */ 269 __be32 ipv4_daddr; 270 struct in6_addr ipv6_daddr; 271 272 /* after prerouting + nat detected: store original source 273 * mac since neigh resolution overwrites it, only used while 274 * skb is out in neigh layer. 275 */ 276 char neigh_header[8]; 277 }; 278 }; 279 #endif 280 281 struct sk_buff_head { 282 /* These two members must be first. */ 283 struct sk_buff *next; 284 struct sk_buff *prev; 285 286 __u32 qlen; 287 spinlock_t lock; 288 }; 289 290 struct sk_buff; 291 292 /* To allow 64K frame to be packed as single skb without frag_list we 293 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 294 * buffers which do not start on a page boundary. 295 * 296 * Since GRO uses frags we allocate at least 16 regardless of page 297 * size. 298 */ 299 #if (65536/PAGE_SIZE + 1) < 16 300 #define MAX_SKB_FRAGS 16UL 301 #else 302 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 303 #endif 304 extern int sysctl_max_skb_frags; 305 306 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 307 * segment using its current segmentation instead. 308 */ 309 #define GSO_BY_FRAGS 0xFFFF 310 311 typedef struct skb_frag_struct skb_frag_t; 312 313 struct skb_frag_struct { 314 struct { 315 struct page *p; 316 } page; 317 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 318 __u32 page_offset; 319 __u32 size; 320 #else 321 __u16 page_offset; 322 __u16 size; 323 #endif 324 }; 325 326 /** 327 * skb_frag_size - Returns the size of a skb fragment 328 * @frag: skb fragment 329 */ 330 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 331 { 332 return frag->size; 333 } 334 335 /** 336 * skb_frag_size_set - Sets the size of a skb fragment 337 * @frag: skb fragment 338 * @size: size of fragment 339 */ 340 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 341 { 342 frag->size = size; 343 } 344 345 /** 346 * skb_frag_size_add - Incrementes the size of a skb fragment by %delta 347 * @frag: skb fragment 348 * @delta: value to add 349 */ 350 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 351 { 352 frag->size += delta; 353 } 354 355 /** 356 * skb_frag_size_sub - Decrements the size of a skb fragment by %delta 357 * @frag: skb fragment 358 * @delta: value to subtract 359 */ 360 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 361 { 362 frag->size -= delta; 363 } 364 365 /** 366 * skb_frag_must_loop - Test if %p is a high memory page 367 * @p: fragment's page 368 */ 369 static inline bool skb_frag_must_loop(struct page *p) 370 { 371 #if defined(CONFIG_HIGHMEM) 372 if (PageHighMem(p)) 373 return true; 374 #endif 375 return false; 376 } 377 378 /** 379 * skb_frag_foreach_page - loop over pages in a fragment 380 * 381 * @f: skb frag to operate on 382 * @f_off: offset from start of f->page.p 383 * @f_len: length from f_off to loop over 384 * @p: (temp var) current page 385 * @p_off: (temp var) offset from start of current page, 386 * non-zero only on first page. 387 * @p_len: (temp var) length in current page, 388 * < PAGE_SIZE only on first and last page. 389 * @copied: (temp var) length so far, excluding current p_len. 390 * 391 * A fragment can hold a compound page, in which case per-page 392 * operations, notably kmap_atomic, must be called for each 393 * regular page. 394 */ 395 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 396 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 397 p_off = (f_off) & (PAGE_SIZE - 1), \ 398 p_len = skb_frag_must_loop(p) ? \ 399 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 400 copied = 0; \ 401 copied < f_len; \ 402 copied += p_len, p++, p_off = 0, \ 403 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 404 405 #define HAVE_HW_TIME_STAMP 406 407 /** 408 * struct skb_shared_hwtstamps - hardware time stamps 409 * @hwtstamp: hardware time stamp transformed into duration 410 * since arbitrary point in time 411 * 412 * Software time stamps generated by ktime_get_real() are stored in 413 * skb->tstamp. 414 * 415 * hwtstamps can only be compared against other hwtstamps from 416 * the same device. 417 * 418 * This structure is attached to packets as part of the 419 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 420 */ 421 struct skb_shared_hwtstamps { 422 ktime_t hwtstamp; 423 }; 424 425 /* Definitions for tx_flags in struct skb_shared_info */ 426 enum { 427 /* generate hardware time stamp */ 428 SKBTX_HW_TSTAMP = 1 << 0, 429 430 /* generate software time stamp when queueing packet to NIC */ 431 SKBTX_SW_TSTAMP = 1 << 1, 432 433 /* device driver is going to provide hardware time stamp */ 434 SKBTX_IN_PROGRESS = 1 << 2, 435 436 /* device driver supports TX zero-copy buffers */ 437 SKBTX_DEV_ZEROCOPY = 1 << 3, 438 439 /* generate wifi status information (where possible) */ 440 SKBTX_WIFI_STATUS = 1 << 4, 441 442 /* This indicates at least one fragment might be overwritten 443 * (as in vmsplice(), sendfile() ...) 444 * If we need to compute a TX checksum, we'll need to copy 445 * all frags to avoid possible bad checksum 446 */ 447 SKBTX_SHARED_FRAG = 1 << 5, 448 449 /* generate software time stamp when entering packet scheduling */ 450 SKBTX_SCHED_TSTAMP = 1 << 6, 451 }; 452 453 #define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG) 454 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 455 SKBTX_SCHED_TSTAMP) 456 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 457 458 /* 459 * The callback notifies userspace to release buffers when skb DMA is done in 460 * lower device, the skb last reference should be 0 when calling this. 461 * The zerocopy_success argument is true if zero copy transmit occurred, 462 * false on data copy or out of memory error caused by data copy attempt. 463 * The ctx field is used to track device context. 464 * The desc field is used to track userspace buffer index. 465 */ 466 struct ubuf_info { 467 void (*callback)(struct ubuf_info *, bool zerocopy_success); 468 union { 469 struct { 470 unsigned long desc; 471 void *ctx; 472 }; 473 struct { 474 u32 id; 475 u16 len; 476 u16 zerocopy:1; 477 u32 bytelen; 478 }; 479 }; 480 refcount_t refcnt; 481 482 struct mmpin { 483 struct user_struct *user; 484 unsigned int num_pg; 485 } mmp; 486 }; 487 488 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 489 490 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 491 void mm_unaccount_pinned_pages(struct mmpin *mmp); 492 493 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size); 494 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size, 495 struct ubuf_info *uarg); 496 497 static inline void sock_zerocopy_get(struct ubuf_info *uarg) 498 { 499 refcount_inc(&uarg->refcnt); 500 } 501 502 void sock_zerocopy_put(struct ubuf_info *uarg); 503 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 504 505 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success); 506 507 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len); 508 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 509 struct msghdr *msg, int len, 510 struct ubuf_info *uarg); 511 512 /* This data is invariant across clones and lives at 513 * the end of the header data, ie. at skb->end. 514 */ 515 struct skb_shared_info { 516 __u8 __unused; 517 __u8 meta_len; 518 __u8 nr_frags; 519 __u8 tx_flags; 520 unsigned short gso_size; 521 /* Warning: this field is not always filled in (UFO)! */ 522 unsigned short gso_segs; 523 struct sk_buff *frag_list; 524 struct skb_shared_hwtstamps hwtstamps; 525 unsigned int gso_type; 526 u32 tskey; 527 528 /* 529 * Warning : all fields before dataref are cleared in __alloc_skb() 530 */ 531 atomic_t dataref; 532 533 /* Intermediate layers must ensure that destructor_arg 534 * remains valid until skb destructor */ 535 void * destructor_arg; 536 537 /* must be last field, see pskb_expand_head() */ 538 skb_frag_t frags[MAX_SKB_FRAGS]; 539 }; 540 541 /* We divide dataref into two halves. The higher 16 bits hold references 542 * to the payload part of skb->data. The lower 16 bits hold references to 543 * the entire skb->data. A clone of a headerless skb holds the length of 544 * the header in skb->hdr_len. 545 * 546 * All users must obey the rule that the skb->data reference count must be 547 * greater than or equal to the payload reference count. 548 * 549 * Holding a reference to the payload part means that the user does not 550 * care about modifications to the header part of skb->data. 551 */ 552 #define SKB_DATAREF_SHIFT 16 553 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 554 555 556 enum { 557 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 558 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 559 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 560 }; 561 562 enum { 563 SKB_GSO_TCPV4 = 1 << 0, 564 565 /* This indicates the skb is from an untrusted source. */ 566 SKB_GSO_DODGY = 1 << 1, 567 568 /* This indicates the tcp segment has CWR set. */ 569 SKB_GSO_TCP_ECN = 1 << 2, 570 571 SKB_GSO_TCP_FIXEDID = 1 << 3, 572 573 SKB_GSO_TCPV6 = 1 << 4, 574 575 SKB_GSO_FCOE = 1 << 5, 576 577 SKB_GSO_GRE = 1 << 6, 578 579 SKB_GSO_GRE_CSUM = 1 << 7, 580 581 SKB_GSO_IPXIP4 = 1 << 8, 582 583 SKB_GSO_IPXIP6 = 1 << 9, 584 585 SKB_GSO_UDP_TUNNEL = 1 << 10, 586 587 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 588 589 SKB_GSO_PARTIAL = 1 << 12, 590 591 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 592 593 SKB_GSO_SCTP = 1 << 14, 594 595 SKB_GSO_ESP = 1 << 15, 596 597 SKB_GSO_UDP = 1 << 16, 598 599 SKB_GSO_UDP_L4 = 1 << 17, 600 }; 601 602 #if BITS_PER_LONG > 32 603 #define NET_SKBUFF_DATA_USES_OFFSET 1 604 #endif 605 606 #ifdef NET_SKBUFF_DATA_USES_OFFSET 607 typedef unsigned int sk_buff_data_t; 608 #else 609 typedef unsigned char *sk_buff_data_t; 610 #endif 611 612 /** 613 * struct sk_buff - socket buffer 614 * @next: Next buffer in list 615 * @prev: Previous buffer in list 616 * @tstamp: Time we arrived/left 617 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 618 * @sk: Socket we are owned by 619 * @dev: Device we arrived on/are leaving by 620 * @cb: Control buffer. Free for use by every layer. Put private vars here 621 * @_skb_refdst: destination entry (with norefcount bit) 622 * @sp: the security path, used for xfrm 623 * @len: Length of actual data 624 * @data_len: Data length 625 * @mac_len: Length of link layer header 626 * @hdr_len: writable header length of cloned skb 627 * @csum: Checksum (must include start/offset pair) 628 * @csum_start: Offset from skb->head where checksumming should start 629 * @csum_offset: Offset from csum_start where checksum should be stored 630 * @priority: Packet queueing priority 631 * @ignore_df: allow local fragmentation 632 * @cloned: Head may be cloned (check refcnt to be sure) 633 * @ip_summed: Driver fed us an IP checksum 634 * @nohdr: Payload reference only, must not modify header 635 * @pkt_type: Packet class 636 * @fclone: skbuff clone status 637 * @ipvs_property: skbuff is owned by ipvs 638 * @offload_fwd_mark: Packet was L2-forwarded in hardware 639 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 640 * @tc_skip_classify: do not classify packet. set by IFB device 641 * @tc_at_ingress: used within tc_classify to distinguish in/egress 642 * @tc_redirected: packet was redirected by a tc action 643 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect 644 * @peeked: this packet has been seen already, so stats have been 645 * done for it, don't do them again 646 * @nf_trace: netfilter packet trace flag 647 * @protocol: Packet protocol from driver 648 * @destructor: Destruct function 649 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 650 * @_nfct: Associated connection, if any (with nfctinfo bits) 651 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 652 * @skb_iif: ifindex of device we arrived on 653 * @tc_index: Traffic control index 654 * @hash: the packet hash 655 * @queue_mapping: Queue mapping for multiqueue devices 656 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 657 * @active_extensions: active extensions (skb_ext_id types) 658 * @ndisc_nodetype: router type (from link layer) 659 * @ooo_okay: allow the mapping of a socket to a queue to be changed 660 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 661 * ports. 662 * @sw_hash: indicates hash was computed in software stack 663 * @wifi_acked_valid: wifi_acked was set 664 * @wifi_acked: whether frame was acked on wifi or not 665 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 666 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 667 * @dst_pending_confirm: need to confirm neighbour 668 * @decrypted: Decrypted SKB 669 * @napi_id: id of the NAPI struct this skb came from 670 * @secmark: security marking 671 * @mark: Generic packet mark 672 * @vlan_proto: vlan encapsulation protocol 673 * @vlan_tci: vlan tag control information 674 * @inner_protocol: Protocol (encapsulation) 675 * @inner_transport_header: Inner transport layer header (encapsulation) 676 * @inner_network_header: Network layer header (encapsulation) 677 * @inner_mac_header: Link layer header (encapsulation) 678 * @transport_header: Transport layer header 679 * @network_header: Network layer header 680 * @mac_header: Link layer header 681 * @tail: Tail pointer 682 * @end: End pointer 683 * @head: Head of buffer 684 * @data: Data head pointer 685 * @truesize: Buffer size 686 * @users: User count - see {datagram,tcp}.c 687 * @extensions: allocated extensions, valid if active_extensions is nonzero 688 */ 689 690 struct sk_buff { 691 union { 692 struct { 693 /* These two members must be first. */ 694 struct sk_buff *next; 695 struct sk_buff *prev; 696 697 union { 698 struct net_device *dev; 699 /* Some protocols might use this space to store information, 700 * while device pointer would be NULL. 701 * UDP receive path is one user. 702 */ 703 unsigned long dev_scratch; 704 }; 705 }; 706 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 707 struct list_head list; 708 }; 709 710 union { 711 struct sock *sk; 712 int ip_defrag_offset; 713 }; 714 715 union { 716 ktime_t tstamp; 717 u64 skb_mstamp_ns; /* earliest departure time */ 718 }; 719 /* 720 * This is the control buffer. It is free to use for every 721 * layer. Please put your private variables there. If you 722 * want to keep them across layers you have to do a skb_clone() 723 * first. This is owned by whoever has the skb queued ATM. 724 */ 725 char cb[48] __aligned(8); 726 727 union { 728 struct { 729 unsigned long _skb_refdst; 730 void (*destructor)(struct sk_buff *skb); 731 }; 732 struct list_head tcp_tsorted_anchor; 733 }; 734 735 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 736 unsigned long _nfct; 737 #endif 738 unsigned int len, 739 data_len; 740 __u16 mac_len, 741 hdr_len; 742 743 /* Following fields are _not_ copied in __copy_skb_header() 744 * Note that queue_mapping is here mostly to fill a hole. 745 */ 746 __u16 queue_mapping; 747 748 /* if you move cloned around you also must adapt those constants */ 749 #ifdef __BIG_ENDIAN_BITFIELD 750 #define CLONED_MASK (1 << 7) 751 #else 752 #define CLONED_MASK 1 753 #endif 754 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset) 755 756 __u8 __cloned_offset[0]; 757 __u8 cloned:1, 758 nohdr:1, 759 fclone:2, 760 peeked:1, 761 head_frag:1, 762 pfmemalloc:1; 763 #ifdef CONFIG_SKB_EXTENSIONS 764 __u8 active_extensions; 765 #endif 766 /* fields enclosed in headers_start/headers_end are copied 767 * using a single memcpy() in __copy_skb_header() 768 */ 769 /* private: */ 770 __u32 headers_start[0]; 771 /* public: */ 772 773 /* if you move pkt_type around you also must adapt those constants */ 774 #ifdef __BIG_ENDIAN_BITFIELD 775 #define PKT_TYPE_MAX (7 << 5) 776 #else 777 #define PKT_TYPE_MAX 7 778 #endif 779 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 780 781 __u8 __pkt_type_offset[0]; 782 __u8 pkt_type:3; 783 __u8 ignore_df:1; 784 __u8 nf_trace:1; 785 __u8 ip_summed:2; 786 __u8 ooo_okay:1; 787 788 __u8 l4_hash:1; 789 __u8 sw_hash:1; 790 __u8 wifi_acked_valid:1; 791 __u8 wifi_acked:1; 792 __u8 no_fcs:1; 793 /* Indicates the inner headers are valid in the skbuff. */ 794 __u8 encapsulation:1; 795 __u8 encap_hdr_csum:1; 796 __u8 csum_valid:1; 797 798 #ifdef __BIG_ENDIAN_BITFIELD 799 #define PKT_VLAN_PRESENT_BIT 7 800 #else 801 #define PKT_VLAN_PRESENT_BIT 0 802 #endif 803 #define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset) 804 __u8 __pkt_vlan_present_offset[0]; 805 __u8 vlan_present:1; 806 __u8 csum_complete_sw:1; 807 __u8 csum_level:2; 808 __u8 csum_not_inet:1; 809 __u8 dst_pending_confirm:1; 810 #ifdef CONFIG_IPV6_NDISC_NODETYPE 811 __u8 ndisc_nodetype:2; 812 #endif 813 814 __u8 ipvs_property:1; 815 __u8 inner_protocol_type:1; 816 __u8 remcsum_offload:1; 817 #ifdef CONFIG_NET_SWITCHDEV 818 __u8 offload_fwd_mark:1; 819 __u8 offload_l3_fwd_mark:1; 820 #endif 821 #ifdef CONFIG_NET_CLS_ACT 822 __u8 tc_skip_classify:1; 823 __u8 tc_at_ingress:1; 824 __u8 tc_redirected:1; 825 __u8 tc_from_ingress:1; 826 #endif 827 #ifdef CONFIG_TLS_DEVICE 828 __u8 decrypted:1; 829 #endif 830 831 #ifdef CONFIG_NET_SCHED 832 __u16 tc_index; /* traffic control index */ 833 #endif 834 835 union { 836 __wsum csum; 837 struct { 838 __u16 csum_start; 839 __u16 csum_offset; 840 }; 841 }; 842 __u32 priority; 843 int skb_iif; 844 __u32 hash; 845 __be16 vlan_proto; 846 __u16 vlan_tci; 847 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 848 union { 849 unsigned int napi_id; 850 unsigned int sender_cpu; 851 }; 852 #endif 853 #ifdef CONFIG_NETWORK_SECMARK 854 __u32 secmark; 855 #endif 856 857 union { 858 __u32 mark; 859 __u32 reserved_tailroom; 860 }; 861 862 union { 863 __be16 inner_protocol; 864 __u8 inner_ipproto; 865 }; 866 867 __u16 inner_transport_header; 868 __u16 inner_network_header; 869 __u16 inner_mac_header; 870 871 __be16 protocol; 872 __u16 transport_header; 873 __u16 network_header; 874 __u16 mac_header; 875 876 /* private: */ 877 __u32 headers_end[0]; 878 /* public: */ 879 880 /* These elements must be at the end, see alloc_skb() for details. */ 881 sk_buff_data_t tail; 882 sk_buff_data_t end; 883 unsigned char *head, 884 *data; 885 unsigned int truesize; 886 refcount_t users; 887 888 #ifdef CONFIG_SKB_EXTENSIONS 889 /* only useable after checking ->active_extensions != 0 */ 890 struct skb_ext *extensions; 891 #endif 892 }; 893 894 #ifdef __KERNEL__ 895 /* 896 * Handling routines are only of interest to the kernel 897 */ 898 899 #define SKB_ALLOC_FCLONE 0x01 900 #define SKB_ALLOC_RX 0x02 901 #define SKB_ALLOC_NAPI 0x04 902 903 /** 904 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 905 * @skb: buffer 906 */ 907 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 908 { 909 return unlikely(skb->pfmemalloc); 910 } 911 912 /* 913 * skb might have a dst pointer attached, refcounted or not. 914 * _skb_refdst low order bit is set if refcount was _not_ taken 915 */ 916 #define SKB_DST_NOREF 1UL 917 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 918 919 #define SKB_NFCT_PTRMASK ~(7UL) 920 /** 921 * skb_dst - returns skb dst_entry 922 * @skb: buffer 923 * 924 * Returns skb dst_entry, regardless of reference taken or not. 925 */ 926 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 927 { 928 /* If refdst was not refcounted, check we still are in a 929 * rcu_read_lock section 930 */ 931 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 932 !rcu_read_lock_held() && 933 !rcu_read_lock_bh_held()); 934 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 935 } 936 937 /** 938 * skb_dst_set - sets skb dst 939 * @skb: buffer 940 * @dst: dst entry 941 * 942 * Sets skb dst, assuming a reference was taken on dst and should 943 * be released by skb_dst_drop() 944 */ 945 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 946 { 947 skb->_skb_refdst = (unsigned long)dst; 948 } 949 950 /** 951 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 952 * @skb: buffer 953 * @dst: dst entry 954 * 955 * Sets skb dst, assuming a reference was not taken on dst. 956 * If dst entry is cached, we do not take reference and dst_release 957 * will be avoided by refdst_drop. If dst entry is not cached, we take 958 * reference, so that last dst_release can destroy the dst immediately. 959 */ 960 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 961 { 962 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 963 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 964 } 965 966 /** 967 * skb_dst_is_noref - Test if skb dst isn't refcounted 968 * @skb: buffer 969 */ 970 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 971 { 972 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 973 } 974 975 /** 976 * skb_rtable - Returns the skb &rtable 977 * @skb: buffer 978 */ 979 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 980 { 981 return (struct rtable *)skb_dst(skb); 982 } 983 984 /* For mangling skb->pkt_type from user space side from applications 985 * such as nft, tc, etc, we only allow a conservative subset of 986 * possible pkt_types to be set. 987 */ 988 static inline bool skb_pkt_type_ok(u32 ptype) 989 { 990 return ptype <= PACKET_OTHERHOST; 991 } 992 993 /** 994 * skb_napi_id - Returns the skb's NAPI id 995 * @skb: buffer 996 */ 997 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 998 { 999 #ifdef CONFIG_NET_RX_BUSY_POLL 1000 return skb->napi_id; 1001 #else 1002 return 0; 1003 #endif 1004 } 1005 1006 /** 1007 * skb_unref - decrement the skb's reference count 1008 * @skb: buffer 1009 * 1010 * Returns true if we can free the skb. 1011 */ 1012 static inline bool skb_unref(struct sk_buff *skb) 1013 { 1014 if (unlikely(!skb)) 1015 return false; 1016 if (likely(refcount_read(&skb->users) == 1)) 1017 smp_rmb(); 1018 else if (likely(!refcount_dec_and_test(&skb->users))) 1019 return false; 1020 1021 return true; 1022 } 1023 1024 void skb_release_head_state(struct sk_buff *skb); 1025 void kfree_skb(struct sk_buff *skb); 1026 void kfree_skb_list(struct sk_buff *segs); 1027 void skb_tx_error(struct sk_buff *skb); 1028 void consume_skb(struct sk_buff *skb); 1029 void __consume_stateless_skb(struct sk_buff *skb); 1030 void __kfree_skb(struct sk_buff *skb); 1031 extern struct kmem_cache *skbuff_head_cache; 1032 1033 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1034 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1035 bool *fragstolen, int *delta_truesize); 1036 1037 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1038 int node); 1039 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1040 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1041 struct sk_buff *build_skb_around(struct sk_buff *skb, 1042 void *data, unsigned int frag_size); 1043 1044 /** 1045 * alloc_skb - allocate a network buffer 1046 * @size: size to allocate 1047 * @priority: allocation mask 1048 * 1049 * This function is a convenient wrapper around __alloc_skb(). 1050 */ 1051 static inline struct sk_buff *alloc_skb(unsigned int size, 1052 gfp_t priority) 1053 { 1054 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1055 } 1056 1057 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1058 unsigned long data_len, 1059 int max_page_order, 1060 int *errcode, 1061 gfp_t gfp_mask); 1062 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1063 1064 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1065 struct sk_buff_fclones { 1066 struct sk_buff skb1; 1067 1068 struct sk_buff skb2; 1069 1070 refcount_t fclone_ref; 1071 }; 1072 1073 /** 1074 * skb_fclone_busy - check if fclone is busy 1075 * @sk: socket 1076 * @skb: buffer 1077 * 1078 * Returns true if skb is a fast clone, and its clone is not freed. 1079 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1080 * so we also check that this didnt happen. 1081 */ 1082 static inline bool skb_fclone_busy(const struct sock *sk, 1083 const struct sk_buff *skb) 1084 { 1085 const struct sk_buff_fclones *fclones; 1086 1087 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1088 1089 return skb->fclone == SKB_FCLONE_ORIG && 1090 refcount_read(&fclones->fclone_ref) > 1 && 1091 fclones->skb2.sk == sk; 1092 } 1093 1094 /** 1095 * alloc_skb_fclone - allocate a network buffer from fclone cache 1096 * @size: size to allocate 1097 * @priority: allocation mask 1098 * 1099 * This function is a convenient wrapper around __alloc_skb(). 1100 */ 1101 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1102 gfp_t priority) 1103 { 1104 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1105 } 1106 1107 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1108 void skb_headers_offset_update(struct sk_buff *skb, int off); 1109 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1110 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1111 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1112 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1113 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1114 gfp_t gfp_mask, bool fclone); 1115 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1116 gfp_t gfp_mask) 1117 { 1118 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1119 } 1120 1121 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1122 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1123 unsigned int headroom); 1124 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1125 int newtailroom, gfp_t priority); 1126 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1127 int offset, int len); 1128 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1129 int offset, int len); 1130 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1131 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1132 1133 /** 1134 * skb_pad - zero pad the tail of an skb 1135 * @skb: buffer to pad 1136 * @pad: space to pad 1137 * 1138 * Ensure that a buffer is followed by a padding area that is zero 1139 * filled. Used by network drivers which may DMA or transfer data 1140 * beyond the buffer end onto the wire. 1141 * 1142 * May return error in out of memory cases. The skb is freed on error. 1143 */ 1144 static inline int skb_pad(struct sk_buff *skb, int pad) 1145 { 1146 return __skb_pad(skb, pad, true); 1147 } 1148 #define dev_kfree_skb(a) consume_skb(a) 1149 1150 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1151 int offset, size_t size); 1152 1153 struct skb_seq_state { 1154 __u32 lower_offset; 1155 __u32 upper_offset; 1156 __u32 frag_idx; 1157 __u32 stepped_offset; 1158 struct sk_buff *root_skb; 1159 struct sk_buff *cur_skb; 1160 __u8 *frag_data; 1161 }; 1162 1163 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1164 unsigned int to, struct skb_seq_state *st); 1165 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1166 struct skb_seq_state *st); 1167 void skb_abort_seq_read(struct skb_seq_state *st); 1168 1169 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1170 unsigned int to, struct ts_config *config); 1171 1172 /* 1173 * Packet hash types specify the type of hash in skb_set_hash. 1174 * 1175 * Hash types refer to the protocol layer addresses which are used to 1176 * construct a packet's hash. The hashes are used to differentiate or identify 1177 * flows of the protocol layer for the hash type. Hash types are either 1178 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1179 * 1180 * Properties of hashes: 1181 * 1182 * 1) Two packets in different flows have different hash values 1183 * 2) Two packets in the same flow should have the same hash value 1184 * 1185 * A hash at a higher layer is considered to be more specific. A driver should 1186 * set the most specific hash possible. 1187 * 1188 * A driver cannot indicate a more specific hash than the layer at which a hash 1189 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1190 * 1191 * A driver may indicate a hash level which is less specific than the 1192 * actual layer the hash was computed on. For instance, a hash computed 1193 * at L4 may be considered an L3 hash. This should only be done if the 1194 * driver can't unambiguously determine that the HW computed the hash at 1195 * the higher layer. Note that the "should" in the second property above 1196 * permits this. 1197 */ 1198 enum pkt_hash_types { 1199 PKT_HASH_TYPE_NONE, /* Undefined type */ 1200 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1201 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1202 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1203 }; 1204 1205 static inline void skb_clear_hash(struct sk_buff *skb) 1206 { 1207 skb->hash = 0; 1208 skb->sw_hash = 0; 1209 skb->l4_hash = 0; 1210 } 1211 1212 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1213 { 1214 if (!skb->l4_hash) 1215 skb_clear_hash(skb); 1216 } 1217 1218 static inline void 1219 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1220 { 1221 skb->l4_hash = is_l4; 1222 skb->sw_hash = is_sw; 1223 skb->hash = hash; 1224 } 1225 1226 static inline void 1227 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1228 { 1229 /* Used by drivers to set hash from HW */ 1230 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1231 } 1232 1233 static inline void 1234 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1235 { 1236 __skb_set_hash(skb, hash, true, is_l4); 1237 } 1238 1239 void __skb_get_hash(struct sk_buff *skb); 1240 u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1241 u32 skb_get_poff(const struct sk_buff *skb); 1242 u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1243 const struct flow_keys_basic *keys, int hlen); 1244 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1245 void *data, int hlen_proto); 1246 1247 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1248 int thoff, u8 ip_proto) 1249 { 1250 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1251 } 1252 1253 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1254 const struct flow_dissector_key *key, 1255 unsigned int key_count); 1256 1257 #ifdef CONFIG_NET 1258 int skb_flow_dissector_prog_query(const union bpf_attr *attr, 1259 union bpf_attr __user *uattr); 1260 int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr, 1261 struct bpf_prog *prog); 1262 1263 int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr); 1264 #else 1265 static inline int skb_flow_dissector_prog_query(const union bpf_attr *attr, 1266 union bpf_attr __user *uattr) 1267 { 1268 return -EOPNOTSUPP; 1269 } 1270 1271 static inline int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr, 1272 struct bpf_prog *prog) 1273 { 1274 return -EOPNOTSUPP; 1275 } 1276 1277 static inline int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr) 1278 { 1279 return -EOPNOTSUPP; 1280 } 1281 #endif 1282 1283 struct bpf_flow_dissector; 1284 bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1285 __be16 proto, int nhoff, int hlen); 1286 1287 bool __skb_flow_dissect(const struct net *net, 1288 const struct sk_buff *skb, 1289 struct flow_dissector *flow_dissector, 1290 void *target_container, 1291 void *data, __be16 proto, int nhoff, int hlen, 1292 unsigned int flags); 1293 1294 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1295 struct flow_dissector *flow_dissector, 1296 void *target_container, unsigned int flags) 1297 { 1298 return __skb_flow_dissect(NULL, skb, flow_dissector, 1299 target_container, NULL, 0, 0, 0, flags); 1300 } 1301 1302 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1303 struct flow_keys *flow, 1304 unsigned int flags) 1305 { 1306 memset(flow, 0, sizeof(*flow)); 1307 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1308 flow, NULL, 0, 0, 0, flags); 1309 } 1310 1311 static inline bool 1312 skb_flow_dissect_flow_keys_basic(const struct net *net, 1313 const struct sk_buff *skb, 1314 struct flow_keys_basic *flow, void *data, 1315 __be16 proto, int nhoff, int hlen, 1316 unsigned int flags) 1317 { 1318 memset(flow, 0, sizeof(*flow)); 1319 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1320 data, proto, nhoff, hlen, flags); 1321 } 1322 1323 void 1324 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1325 struct flow_dissector *flow_dissector, 1326 void *target_container); 1327 1328 static inline __u32 skb_get_hash(struct sk_buff *skb) 1329 { 1330 if (!skb->l4_hash && !skb->sw_hash) 1331 __skb_get_hash(skb); 1332 1333 return skb->hash; 1334 } 1335 1336 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1337 { 1338 if (!skb->l4_hash && !skb->sw_hash) { 1339 struct flow_keys keys; 1340 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1341 1342 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1343 } 1344 1345 return skb->hash; 1346 } 1347 1348 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1349 1350 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1351 { 1352 return skb->hash; 1353 } 1354 1355 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1356 { 1357 to->hash = from->hash; 1358 to->sw_hash = from->sw_hash; 1359 to->l4_hash = from->l4_hash; 1360 }; 1361 1362 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1363 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1364 { 1365 return skb->head + skb->end; 1366 } 1367 1368 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1369 { 1370 return skb->end; 1371 } 1372 #else 1373 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1374 { 1375 return skb->end; 1376 } 1377 1378 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1379 { 1380 return skb->end - skb->head; 1381 } 1382 #endif 1383 1384 /* Internal */ 1385 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1386 1387 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1388 { 1389 return &skb_shinfo(skb)->hwtstamps; 1390 } 1391 1392 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1393 { 1394 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY; 1395 1396 return is_zcopy ? skb_uarg(skb) : NULL; 1397 } 1398 1399 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1400 bool *have_ref) 1401 { 1402 if (skb && uarg && !skb_zcopy(skb)) { 1403 if (unlikely(have_ref && *have_ref)) 1404 *have_ref = false; 1405 else 1406 sock_zerocopy_get(uarg); 1407 skb_shinfo(skb)->destructor_arg = uarg; 1408 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG; 1409 } 1410 } 1411 1412 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1413 { 1414 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1415 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG; 1416 } 1417 1418 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1419 { 1420 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1421 } 1422 1423 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1424 { 1425 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1426 } 1427 1428 /* Release a reference on a zerocopy structure */ 1429 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy) 1430 { 1431 struct ubuf_info *uarg = skb_zcopy(skb); 1432 1433 if (uarg) { 1434 if (skb_zcopy_is_nouarg(skb)) { 1435 /* no notification callback */ 1436 } else if (uarg->callback == sock_zerocopy_callback) { 1437 uarg->zerocopy = uarg->zerocopy && zerocopy; 1438 sock_zerocopy_put(uarg); 1439 } else { 1440 uarg->callback(uarg, zerocopy); 1441 } 1442 1443 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1444 } 1445 } 1446 1447 /* Abort a zerocopy operation and revert zckey on error in send syscall */ 1448 static inline void skb_zcopy_abort(struct sk_buff *skb) 1449 { 1450 struct ubuf_info *uarg = skb_zcopy(skb); 1451 1452 if (uarg) { 1453 sock_zerocopy_put_abort(uarg, false); 1454 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1455 } 1456 } 1457 1458 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1459 { 1460 skb->next = NULL; 1461 } 1462 1463 static inline void skb_list_del_init(struct sk_buff *skb) 1464 { 1465 __list_del_entry(&skb->list); 1466 skb_mark_not_on_list(skb); 1467 } 1468 1469 /** 1470 * skb_queue_empty - check if a queue is empty 1471 * @list: queue head 1472 * 1473 * Returns true if the queue is empty, false otherwise. 1474 */ 1475 static inline int skb_queue_empty(const struct sk_buff_head *list) 1476 { 1477 return list->next == (const struct sk_buff *) list; 1478 } 1479 1480 /** 1481 * skb_queue_is_last - check if skb is the last entry in the queue 1482 * @list: queue head 1483 * @skb: buffer 1484 * 1485 * Returns true if @skb is the last buffer on the list. 1486 */ 1487 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1488 const struct sk_buff *skb) 1489 { 1490 return skb->next == (const struct sk_buff *) list; 1491 } 1492 1493 /** 1494 * skb_queue_is_first - check if skb is the first entry in the queue 1495 * @list: queue head 1496 * @skb: buffer 1497 * 1498 * Returns true if @skb is the first buffer on the list. 1499 */ 1500 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1501 const struct sk_buff *skb) 1502 { 1503 return skb->prev == (const struct sk_buff *) list; 1504 } 1505 1506 /** 1507 * skb_queue_next - return the next packet in the queue 1508 * @list: queue head 1509 * @skb: current buffer 1510 * 1511 * Return the next packet in @list after @skb. It is only valid to 1512 * call this if skb_queue_is_last() evaluates to false. 1513 */ 1514 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1515 const struct sk_buff *skb) 1516 { 1517 /* This BUG_ON may seem severe, but if we just return then we 1518 * are going to dereference garbage. 1519 */ 1520 BUG_ON(skb_queue_is_last(list, skb)); 1521 return skb->next; 1522 } 1523 1524 /** 1525 * skb_queue_prev - return the prev packet in the queue 1526 * @list: queue head 1527 * @skb: current buffer 1528 * 1529 * Return the prev packet in @list before @skb. It is only valid to 1530 * call this if skb_queue_is_first() evaluates to false. 1531 */ 1532 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1533 const struct sk_buff *skb) 1534 { 1535 /* This BUG_ON may seem severe, but if we just return then we 1536 * are going to dereference garbage. 1537 */ 1538 BUG_ON(skb_queue_is_first(list, skb)); 1539 return skb->prev; 1540 } 1541 1542 /** 1543 * skb_get - reference buffer 1544 * @skb: buffer to reference 1545 * 1546 * Makes another reference to a socket buffer and returns a pointer 1547 * to the buffer. 1548 */ 1549 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1550 { 1551 refcount_inc(&skb->users); 1552 return skb; 1553 } 1554 1555 /* 1556 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1557 */ 1558 1559 /** 1560 * skb_cloned - is the buffer a clone 1561 * @skb: buffer to check 1562 * 1563 * Returns true if the buffer was generated with skb_clone() and is 1564 * one of multiple shared copies of the buffer. Cloned buffers are 1565 * shared data so must not be written to under normal circumstances. 1566 */ 1567 static inline int skb_cloned(const struct sk_buff *skb) 1568 { 1569 return skb->cloned && 1570 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1571 } 1572 1573 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1574 { 1575 might_sleep_if(gfpflags_allow_blocking(pri)); 1576 1577 if (skb_cloned(skb)) 1578 return pskb_expand_head(skb, 0, 0, pri); 1579 1580 return 0; 1581 } 1582 1583 /** 1584 * skb_header_cloned - is the header a clone 1585 * @skb: buffer to check 1586 * 1587 * Returns true if modifying the header part of the buffer requires 1588 * the data to be copied. 1589 */ 1590 static inline int skb_header_cloned(const struct sk_buff *skb) 1591 { 1592 int dataref; 1593 1594 if (!skb->cloned) 1595 return 0; 1596 1597 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1598 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1599 return dataref != 1; 1600 } 1601 1602 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1603 { 1604 might_sleep_if(gfpflags_allow_blocking(pri)); 1605 1606 if (skb_header_cloned(skb)) 1607 return pskb_expand_head(skb, 0, 0, pri); 1608 1609 return 0; 1610 } 1611 1612 /** 1613 * __skb_header_release - release reference to header 1614 * @skb: buffer to operate on 1615 */ 1616 static inline void __skb_header_release(struct sk_buff *skb) 1617 { 1618 skb->nohdr = 1; 1619 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1620 } 1621 1622 1623 /** 1624 * skb_shared - is the buffer shared 1625 * @skb: buffer to check 1626 * 1627 * Returns true if more than one person has a reference to this 1628 * buffer. 1629 */ 1630 static inline int skb_shared(const struct sk_buff *skb) 1631 { 1632 return refcount_read(&skb->users) != 1; 1633 } 1634 1635 /** 1636 * skb_share_check - check if buffer is shared and if so clone it 1637 * @skb: buffer to check 1638 * @pri: priority for memory allocation 1639 * 1640 * If the buffer is shared the buffer is cloned and the old copy 1641 * drops a reference. A new clone with a single reference is returned. 1642 * If the buffer is not shared the original buffer is returned. When 1643 * being called from interrupt status or with spinlocks held pri must 1644 * be GFP_ATOMIC. 1645 * 1646 * NULL is returned on a memory allocation failure. 1647 */ 1648 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1649 { 1650 might_sleep_if(gfpflags_allow_blocking(pri)); 1651 if (skb_shared(skb)) { 1652 struct sk_buff *nskb = skb_clone(skb, pri); 1653 1654 if (likely(nskb)) 1655 consume_skb(skb); 1656 else 1657 kfree_skb(skb); 1658 skb = nskb; 1659 } 1660 return skb; 1661 } 1662 1663 /* 1664 * Copy shared buffers into a new sk_buff. We effectively do COW on 1665 * packets to handle cases where we have a local reader and forward 1666 * and a couple of other messy ones. The normal one is tcpdumping 1667 * a packet thats being forwarded. 1668 */ 1669 1670 /** 1671 * skb_unshare - make a copy of a shared buffer 1672 * @skb: buffer to check 1673 * @pri: priority for memory allocation 1674 * 1675 * If the socket buffer is a clone then this function creates a new 1676 * copy of the data, drops a reference count on the old copy and returns 1677 * the new copy with the reference count at 1. If the buffer is not a clone 1678 * the original buffer is returned. When called with a spinlock held or 1679 * from interrupt state @pri must be %GFP_ATOMIC 1680 * 1681 * %NULL is returned on a memory allocation failure. 1682 */ 1683 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1684 gfp_t pri) 1685 { 1686 might_sleep_if(gfpflags_allow_blocking(pri)); 1687 if (skb_cloned(skb)) { 1688 struct sk_buff *nskb = skb_copy(skb, pri); 1689 1690 /* Free our shared copy */ 1691 if (likely(nskb)) 1692 consume_skb(skb); 1693 else 1694 kfree_skb(skb); 1695 skb = nskb; 1696 } 1697 return skb; 1698 } 1699 1700 /** 1701 * skb_peek - peek at the head of an &sk_buff_head 1702 * @list_: list to peek at 1703 * 1704 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1705 * be careful with this one. A peek leaves the buffer on the 1706 * list and someone else may run off with it. You must hold 1707 * the appropriate locks or have a private queue to do this. 1708 * 1709 * Returns %NULL for an empty list or a pointer to the head element. 1710 * The reference count is not incremented and the reference is therefore 1711 * volatile. Use with caution. 1712 */ 1713 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1714 { 1715 struct sk_buff *skb = list_->next; 1716 1717 if (skb == (struct sk_buff *)list_) 1718 skb = NULL; 1719 return skb; 1720 } 1721 1722 /** 1723 * __skb_peek - peek at the head of a non-empty &sk_buff_head 1724 * @list_: list to peek at 1725 * 1726 * Like skb_peek(), but the caller knows that the list is not empty. 1727 */ 1728 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 1729 { 1730 return list_->next; 1731 } 1732 1733 /** 1734 * skb_peek_next - peek skb following the given one from a queue 1735 * @skb: skb to start from 1736 * @list_: list to peek at 1737 * 1738 * Returns %NULL when the end of the list is met or a pointer to the 1739 * next element. The reference count is not incremented and the 1740 * reference is therefore volatile. Use with caution. 1741 */ 1742 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1743 const struct sk_buff_head *list_) 1744 { 1745 struct sk_buff *next = skb->next; 1746 1747 if (next == (struct sk_buff *)list_) 1748 next = NULL; 1749 return next; 1750 } 1751 1752 /** 1753 * skb_peek_tail - peek at the tail of an &sk_buff_head 1754 * @list_: list to peek at 1755 * 1756 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1757 * be careful with this one. A peek leaves the buffer on the 1758 * list and someone else may run off with it. You must hold 1759 * the appropriate locks or have a private queue to do this. 1760 * 1761 * Returns %NULL for an empty list or a pointer to the tail element. 1762 * The reference count is not incremented and the reference is therefore 1763 * volatile. Use with caution. 1764 */ 1765 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1766 { 1767 struct sk_buff *skb = list_->prev; 1768 1769 if (skb == (struct sk_buff *)list_) 1770 skb = NULL; 1771 return skb; 1772 1773 } 1774 1775 /** 1776 * skb_queue_len - get queue length 1777 * @list_: list to measure 1778 * 1779 * Return the length of an &sk_buff queue. 1780 */ 1781 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1782 { 1783 return list_->qlen; 1784 } 1785 1786 /** 1787 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1788 * @list: queue to initialize 1789 * 1790 * This initializes only the list and queue length aspects of 1791 * an sk_buff_head object. This allows to initialize the list 1792 * aspects of an sk_buff_head without reinitializing things like 1793 * the spinlock. It can also be used for on-stack sk_buff_head 1794 * objects where the spinlock is known to not be used. 1795 */ 1796 static inline void __skb_queue_head_init(struct sk_buff_head *list) 1797 { 1798 list->prev = list->next = (struct sk_buff *)list; 1799 list->qlen = 0; 1800 } 1801 1802 /* 1803 * This function creates a split out lock class for each invocation; 1804 * this is needed for now since a whole lot of users of the skb-queue 1805 * infrastructure in drivers have different locking usage (in hardirq) 1806 * than the networking core (in softirq only). In the long run either the 1807 * network layer or drivers should need annotation to consolidate the 1808 * main types of usage into 3 classes. 1809 */ 1810 static inline void skb_queue_head_init(struct sk_buff_head *list) 1811 { 1812 spin_lock_init(&list->lock); 1813 __skb_queue_head_init(list); 1814 } 1815 1816 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1817 struct lock_class_key *class) 1818 { 1819 skb_queue_head_init(list); 1820 lockdep_set_class(&list->lock, class); 1821 } 1822 1823 /* 1824 * Insert an sk_buff on a list. 1825 * 1826 * The "__skb_xxxx()" functions are the non-atomic ones that 1827 * can only be called with interrupts disabled. 1828 */ 1829 static inline void __skb_insert(struct sk_buff *newsk, 1830 struct sk_buff *prev, struct sk_buff *next, 1831 struct sk_buff_head *list) 1832 { 1833 newsk->next = next; 1834 newsk->prev = prev; 1835 next->prev = prev->next = newsk; 1836 list->qlen++; 1837 } 1838 1839 static inline void __skb_queue_splice(const struct sk_buff_head *list, 1840 struct sk_buff *prev, 1841 struct sk_buff *next) 1842 { 1843 struct sk_buff *first = list->next; 1844 struct sk_buff *last = list->prev; 1845 1846 first->prev = prev; 1847 prev->next = first; 1848 1849 last->next = next; 1850 next->prev = last; 1851 } 1852 1853 /** 1854 * skb_queue_splice - join two skb lists, this is designed for stacks 1855 * @list: the new list to add 1856 * @head: the place to add it in the first list 1857 */ 1858 static inline void skb_queue_splice(const struct sk_buff_head *list, 1859 struct sk_buff_head *head) 1860 { 1861 if (!skb_queue_empty(list)) { 1862 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1863 head->qlen += list->qlen; 1864 } 1865 } 1866 1867 /** 1868 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1869 * @list: the new list to add 1870 * @head: the place to add it in the first list 1871 * 1872 * The list at @list is reinitialised 1873 */ 1874 static inline void skb_queue_splice_init(struct sk_buff_head *list, 1875 struct sk_buff_head *head) 1876 { 1877 if (!skb_queue_empty(list)) { 1878 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1879 head->qlen += list->qlen; 1880 __skb_queue_head_init(list); 1881 } 1882 } 1883 1884 /** 1885 * skb_queue_splice_tail - join two skb lists, each list being a queue 1886 * @list: the new list to add 1887 * @head: the place to add it in the first list 1888 */ 1889 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1890 struct sk_buff_head *head) 1891 { 1892 if (!skb_queue_empty(list)) { 1893 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1894 head->qlen += list->qlen; 1895 } 1896 } 1897 1898 /** 1899 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1900 * @list: the new list to add 1901 * @head: the place to add it in the first list 1902 * 1903 * Each of the lists is a queue. 1904 * The list at @list is reinitialised 1905 */ 1906 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1907 struct sk_buff_head *head) 1908 { 1909 if (!skb_queue_empty(list)) { 1910 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1911 head->qlen += list->qlen; 1912 __skb_queue_head_init(list); 1913 } 1914 } 1915 1916 /** 1917 * __skb_queue_after - queue a buffer at the list head 1918 * @list: list to use 1919 * @prev: place after this buffer 1920 * @newsk: buffer to queue 1921 * 1922 * Queue a buffer int the middle of a list. This function takes no locks 1923 * and you must therefore hold required locks before calling it. 1924 * 1925 * A buffer cannot be placed on two lists at the same time. 1926 */ 1927 static inline void __skb_queue_after(struct sk_buff_head *list, 1928 struct sk_buff *prev, 1929 struct sk_buff *newsk) 1930 { 1931 __skb_insert(newsk, prev, prev->next, list); 1932 } 1933 1934 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1935 struct sk_buff_head *list); 1936 1937 static inline void __skb_queue_before(struct sk_buff_head *list, 1938 struct sk_buff *next, 1939 struct sk_buff *newsk) 1940 { 1941 __skb_insert(newsk, next->prev, next, list); 1942 } 1943 1944 /** 1945 * __skb_queue_head - queue a buffer at the list head 1946 * @list: list to use 1947 * @newsk: buffer to queue 1948 * 1949 * Queue a buffer at the start of a list. This function takes no locks 1950 * and you must therefore hold required locks before calling it. 1951 * 1952 * A buffer cannot be placed on two lists at the same time. 1953 */ 1954 static inline void __skb_queue_head(struct sk_buff_head *list, 1955 struct sk_buff *newsk) 1956 { 1957 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1958 } 1959 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1960 1961 /** 1962 * __skb_queue_tail - queue a buffer at the list tail 1963 * @list: list to use 1964 * @newsk: buffer to queue 1965 * 1966 * Queue a buffer at the end of a list. This function takes no locks 1967 * and you must therefore hold required locks before calling it. 1968 * 1969 * A buffer cannot be placed on two lists at the same time. 1970 */ 1971 static inline void __skb_queue_tail(struct sk_buff_head *list, 1972 struct sk_buff *newsk) 1973 { 1974 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1975 } 1976 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1977 1978 /* 1979 * remove sk_buff from list. _Must_ be called atomically, and with 1980 * the list known.. 1981 */ 1982 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1983 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1984 { 1985 struct sk_buff *next, *prev; 1986 1987 list->qlen--; 1988 next = skb->next; 1989 prev = skb->prev; 1990 skb->next = skb->prev = NULL; 1991 next->prev = prev; 1992 prev->next = next; 1993 } 1994 1995 /** 1996 * __skb_dequeue - remove from the head of the queue 1997 * @list: list to dequeue from 1998 * 1999 * Remove the head of the list. This function does not take any locks 2000 * so must be used with appropriate locks held only. The head item is 2001 * returned or %NULL if the list is empty. 2002 */ 2003 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2004 { 2005 struct sk_buff *skb = skb_peek(list); 2006 if (skb) 2007 __skb_unlink(skb, list); 2008 return skb; 2009 } 2010 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2011 2012 /** 2013 * __skb_dequeue_tail - remove from the tail of the queue 2014 * @list: list to dequeue from 2015 * 2016 * Remove the tail of the list. This function does not take any locks 2017 * so must be used with appropriate locks held only. The tail item is 2018 * returned or %NULL if the list is empty. 2019 */ 2020 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2021 { 2022 struct sk_buff *skb = skb_peek_tail(list); 2023 if (skb) 2024 __skb_unlink(skb, list); 2025 return skb; 2026 } 2027 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2028 2029 2030 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2031 { 2032 return skb->data_len; 2033 } 2034 2035 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2036 { 2037 return skb->len - skb->data_len; 2038 } 2039 2040 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2041 { 2042 unsigned int i, len = 0; 2043 2044 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2045 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2046 return len; 2047 } 2048 2049 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2050 { 2051 return skb_headlen(skb) + __skb_pagelen(skb); 2052 } 2053 2054 /** 2055 * __skb_fill_page_desc - initialise a paged fragment in an skb 2056 * @skb: buffer containing fragment to be initialised 2057 * @i: paged fragment index to initialise 2058 * @page: the page to use for this fragment 2059 * @off: the offset to the data with @page 2060 * @size: the length of the data 2061 * 2062 * Initialises the @i'th fragment of @skb to point to &size bytes at 2063 * offset @off within @page. 2064 * 2065 * Does not take any additional reference on the fragment. 2066 */ 2067 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2068 struct page *page, int off, int size) 2069 { 2070 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2071 2072 /* 2073 * Propagate page pfmemalloc to the skb if we can. The problem is 2074 * that not all callers have unique ownership of the page but rely 2075 * on page_is_pfmemalloc doing the right thing(tm). 2076 */ 2077 frag->page.p = page; 2078 frag->page_offset = off; 2079 skb_frag_size_set(frag, size); 2080 2081 page = compound_head(page); 2082 if (page_is_pfmemalloc(page)) 2083 skb->pfmemalloc = true; 2084 } 2085 2086 /** 2087 * skb_fill_page_desc - initialise a paged fragment in an skb 2088 * @skb: buffer containing fragment to be initialised 2089 * @i: paged fragment index to initialise 2090 * @page: the page to use for this fragment 2091 * @off: the offset to the data with @page 2092 * @size: the length of the data 2093 * 2094 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2095 * @skb to point to @size bytes at offset @off within @page. In 2096 * addition updates @skb such that @i is the last fragment. 2097 * 2098 * Does not take any additional reference on the fragment. 2099 */ 2100 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2101 struct page *page, int off, int size) 2102 { 2103 __skb_fill_page_desc(skb, i, page, off, size); 2104 skb_shinfo(skb)->nr_frags = i + 1; 2105 } 2106 2107 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 2108 int size, unsigned int truesize); 2109 2110 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2111 unsigned int truesize); 2112 2113 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2114 2115 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2116 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2117 { 2118 return skb->head + skb->tail; 2119 } 2120 2121 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2122 { 2123 skb->tail = skb->data - skb->head; 2124 } 2125 2126 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2127 { 2128 skb_reset_tail_pointer(skb); 2129 skb->tail += offset; 2130 } 2131 2132 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2133 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2134 { 2135 return skb->tail; 2136 } 2137 2138 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2139 { 2140 skb->tail = skb->data; 2141 } 2142 2143 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2144 { 2145 skb->tail = skb->data + offset; 2146 } 2147 2148 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2149 2150 /* 2151 * Add data to an sk_buff 2152 */ 2153 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2154 void *skb_put(struct sk_buff *skb, unsigned int len); 2155 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2156 { 2157 void *tmp = skb_tail_pointer(skb); 2158 SKB_LINEAR_ASSERT(skb); 2159 skb->tail += len; 2160 skb->len += len; 2161 return tmp; 2162 } 2163 2164 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2165 { 2166 void *tmp = __skb_put(skb, len); 2167 2168 memset(tmp, 0, len); 2169 return tmp; 2170 } 2171 2172 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2173 unsigned int len) 2174 { 2175 void *tmp = __skb_put(skb, len); 2176 2177 memcpy(tmp, data, len); 2178 return tmp; 2179 } 2180 2181 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2182 { 2183 *(u8 *)__skb_put(skb, 1) = val; 2184 } 2185 2186 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2187 { 2188 void *tmp = skb_put(skb, len); 2189 2190 memset(tmp, 0, len); 2191 2192 return tmp; 2193 } 2194 2195 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2196 unsigned int len) 2197 { 2198 void *tmp = skb_put(skb, len); 2199 2200 memcpy(tmp, data, len); 2201 2202 return tmp; 2203 } 2204 2205 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2206 { 2207 *(u8 *)skb_put(skb, 1) = val; 2208 } 2209 2210 void *skb_push(struct sk_buff *skb, unsigned int len); 2211 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2212 { 2213 skb->data -= len; 2214 skb->len += len; 2215 return skb->data; 2216 } 2217 2218 void *skb_pull(struct sk_buff *skb, unsigned int len); 2219 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2220 { 2221 skb->len -= len; 2222 BUG_ON(skb->len < skb->data_len); 2223 return skb->data += len; 2224 } 2225 2226 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2227 { 2228 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2229 } 2230 2231 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2232 2233 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len) 2234 { 2235 if (len > skb_headlen(skb) && 2236 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 2237 return NULL; 2238 skb->len -= len; 2239 return skb->data += len; 2240 } 2241 2242 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2243 { 2244 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 2245 } 2246 2247 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 2248 { 2249 if (likely(len <= skb_headlen(skb))) 2250 return 1; 2251 if (unlikely(len > skb->len)) 2252 return 0; 2253 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 2254 } 2255 2256 void skb_condense(struct sk_buff *skb); 2257 2258 /** 2259 * skb_headroom - bytes at buffer head 2260 * @skb: buffer to check 2261 * 2262 * Return the number of bytes of free space at the head of an &sk_buff. 2263 */ 2264 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2265 { 2266 return skb->data - skb->head; 2267 } 2268 2269 /** 2270 * skb_tailroom - bytes at buffer end 2271 * @skb: buffer to check 2272 * 2273 * Return the number of bytes of free space at the tail of an sk_buff 2274 */ 2275 static inline int skb_tailroom(const struct sk_buff *skb) 2276 { 2277 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2278 } 2279 2280 /** 2281 * skb_availroom - bytes at buffer end 2282 * @skb: buffer to check 2283 * 2284 * Return the number of bytes of free space at the tail of an sk_buff 2285 * allocated by sk_stream_alloc() 2286 */ 2287 static inline int skb_availroom(const struct sk_buff *skb) 2288 { 2289 if (skb_is_nonlinear(skb)) 2290 return 0; 2291 2292 return skb->end - skb->tail - skb->reserved_tailroom; 2293 } 2294 2295 /** 2296 * skb_reserve - adjust headroom 2297 * @skb: buffer to alter 2298 * @len: bytes to move 2299 * 2300 * Increase the headroom of an empty &sk_buff by reducing the tail 2301 * room. This is only allowed for an empty buffer. 2302 */ 2303 static inline void skb_reserve(struct sk_buff *skb, int len) 2304 { 2305 skb->data += len; 2306 skb->tail += len; 2307 } 2308 2309 /** 2310 * skb_tailroom_reserve - adjust reserved_tailroom 2311 * @skb: buffer to alter 2312 * @mtu: maximum amount of headlen permitted 2313 * @needed_tailroom: minimum amount of reserved_tailroom 2314 * 2315 * Set reserved_tailroom so that headlen can be as large as possible but 2316 * not larger than mtu and tailroom cannot be smaller than 2317 * needed_tailroom. 2318 * The required headroom should already have been reserved before using 2319 * this function. 2320 */ 2321 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2322 unsigned int needed_tailroom) 2323 { 2324 SKB_LINEAR_ASSERT(skb); 2325 if (mtu < skb_tailroom(skb) - needed_tailroom) 2326 /* use at most mtu */ 2327 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2328 else 2329 /* use up to all available space */ 2330 skb->reserved_tailroom = needed_tailroom; 2331 } 2332 2333 #define ENCAP_TYPE_ETHER 0 2334 #define ENCAP_TYPE_IPPROTO 1 2335 2336 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2337 __be16 protocol) 2338 { 2339 skb->inner_protocol = protocol; 2340 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2341 } 2342 2343 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2344 __u8 ipproto) 2345 { 2346 skb->inner_ipproto = ipproto; 2347 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2348 } 2349 2350 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2351 { 2352 skb->inner_mac_header = skb->mac_header; 2353 skb->inner_network_header = skb->network_header; 2354 skb->inner_transport_header = skb->transport_header; 2355 } 2356 2357 static inline void skb_reset_mac_len(struct sk_buff *skb) 2358 { 2359 skb->mac_len = skb->network_header - skb->mac_header; 2360 } 2361 2362 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2363 *skb) 2364 { 2365 return skb->head + skb->inner_transport_header; 2366 } 2367 2368 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2369 { 2370 return skb_inner_transport_header(skb) - skb->data; 2371 } 2372 2373 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2374 { 2375 skb->inner_transport_header = skb->data - skb->head; 2376 } 2377 2378 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2379 const int offset) 2380 { 2381 skb_reset_inner_transport_header(skb); 2382 skb->inner_transport_header += offset; 2383 } 2384 2385 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2386 { 2387 return skb->head + skb->inner_network_header; 2388 } 2389 2390 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2391 { 2392 skb->inner_network_header = skb->data - skb->head; 2393 } 2394 2395 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2396 const int offset) 2397 { 2398 skb_reset_inner_network_header(skb); 2399 skb->inner_network_header += offset; 2400 } 2401 2402 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2403 { 2404 return skb->head + skb->inner_mac_header; 2405 } 2406 2407 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2408 { 2409 skb->inner_mac_header = skb->data - skb->head; 2410 } 2411 2412 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2413 const int offset) 2414 { 2415 skb_reset_inner_mac_header(skb); 2416 skb->inner_mac_header += offset; 2417 } 2418 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2419 { 2420 return skb->transport_header != (typeof(skb->transport_header))~0U; 2421 } 2422 2423 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2424 { 2425 return skb->head + skb->transport_header; 2426 } 2427 2428 static inline void skb_reset_transport_header(struct sk_buff *skb) 2429 { 2430 skb->transport_header = skb->data - skb->head; 2431 } 2432 2433 static inline void skb_set_transport_header(struct sk_buff *skb, 2434 const int offset) 2435 { 2436 skb_reset_transport_header(skb); 2437 skb->transport_header += offset; 2438 } 2439 2440 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2441 { 2442 return skb->head + skb->network_header; 2443 } 2444 2445 static inline void skb_reset_network_header(struct sk_buff *skb) 2446 { 2447 skb->network_header = skb->data - skb->head; 2448 } 2449 2450 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2451 { 2452 skb_reset_network_header(skb); 2453 skb->network_header += offset; 2454 } 2455 2456 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2457 { 2458 return skb->head + skb->mac_header; 2459 } 2460 2461 static inline int skb_mac_offset(const struct sk_buff *skb) 2462 { 2463 return skb_mac_header(skb) - skb->data; 2464 } 2465 2466 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2467 { 2468 return skb->network_header - skb->mac_header; 2469 } 2470 2471 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2472 { 2473 return skb->mac_header != (typeof(skb->mac_header))~0U; 2474 } 2475 2476 static inline void skb_reset_mac_header(struct sk_buff *skb) 2477 { 2478 skb->mac_header = skb->data - skb->head; 2479 } 2480 2481 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2482 { 2483 skb_reset_mac_header(skb); 2484 skb->mac_header += offset; 2485 } 2486 2487 static inline void skb_pop_mac_header(struct sk_buff *skb) 2488 { 2489 skb->mac_header = skb->network_header; 2490 } 2491 2492 static inline void skb_probe_transport_header(struct sk_buff *skb) 2493 { 2494 struct flow_keys_basic keys; 2495 2496 if (skb_transport_header_was_set(skb)) 2497 return; 2498 2499 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 2500 NULL, 0, 0, 0, 0)) 2501 skb_set_transport_header(skb, keys.control.thoff); 2502 } 2503 2504 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2505 { 2506 if (skb_mac_header_was_set(skb)) { 2507 const unsigned char *old_mac = skb_mac_header(skb); 2508 2509 skb_set_mac_header(skb, -skb->mac_len); 2510 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2511 } 2512 } 2513 2514 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2515 { 2516 return skb->csum_start - skb_headroom(skb); 2517 } 2518 2519 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2520 { 2521 return skb->head + skb->csum_start; 2522 } 2523 2524 static inline int skb_transport_offset(const struct sk_buff *skb) 2525 { 2526 return skb_transport_header(skb) - skb->data; 2527 } 2528 2529 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2530 { 2531 return skb->transport_header - skb->network_header; 2532 } 2533 2534 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2535 { 2536 return skb->inner_transport_header - skb->inner_network_header; 2537 } 2538 2539 static inline int skb_network_offset(const struct sk_buff *skb) 2540 { 2541 return skb_network_header(skb) - skb->data; 2542 } 2543 2544 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2545 { 2546 return skb_inner_network_header(skb) - skb->data; 2547 } 2548 2549 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2550 { 2551 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2552 } 2553 2554 /* 2555 * CPUs often take a performance hit when accessing unaligned memory 2556 * locations. The actual performance hit varies, it can be small if the 2557 * hardware handles it or large if we have to take an exception and fix it 2558 * in software. 2559 * 2560 * Since an ethernet header is 14 bytes network drivers often end up with 2561 * the IP header at an unaligned offset. The IP header can be aligned by 2562 * shifting the start of the packet by 2 bytes. Drivers should do this 2563 * with: 2564 * 2565 * skb_reserve(skb, NET_IP_ALIGN); 2566 * 2567 * The downside to this alignment of the IP header is that the DMA is now 2568 * unaligned. On some architectures the cost of an unaligned DMA is high 2569 * and this cost outweighs the gains made by aligning the IP header. 2570 * 2571 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2572 * to be overridden. 2573 */ 2574 #ifndef NET_IP_ALIGN 2575 #define NET_IP_ALIGN 2 2576 #endif 2577 2578 /* 2579 * The networking layer reserves some headroom in skb data (via 2580 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2581 * the header has to grow. In the default case, if the header has to grow 2582 * 32 bytes or less we avoid the reallocation. 2583 * 2584 * Unfortunately this headroom changes the DMA alignment of the resulting 2585 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2586 * on some architectures. An architecture can override this value, 2587 * perhaps setting it to a cacheline in size (since that will maintain 2588 * cacheline alignment of the DMA). It must be a power of 2. 2589 * 2590 * Various parts of the networking layer expect at least 32 bytes of 2591 * headroom, you should not reduce this. 2592 * 2593 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2594 * to reduce average number of cache lines per packet. 2595 * get_rps_cpus() for example only access one 64 bytes aligned block : 2596 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2597 */ 2598 #ifndef NET_SKB_PAD 2599 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2600 #endif 2601 2602 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2603 2604 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 2605 { 2606 if (WARN_ON(skb_is_nonlinear(skb))) 2607 return; 2608 skb->len = len; 2609 skb_set_tail_pointer(skb, len); 2610 } 2611 2612 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2613 { 2614 __skb_set_length(skb, len); 2615 } 2616 2617 void skb_trim(struct sk_buff *skb, unsigned int len); 2618 2619 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2620 { 2621 if (skb->data_len) 2622 return ___pskb_trim(skb, len); 2623 __skb_trim(skb, len); 2624 return 0; 2625 } 2626 2627 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2628 { 2629 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2630 } 2631 2632 /** 2633 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2634 * @skb: buffer to alter 2635 * @len: new length 2636 * 2637 * This is identical to pskb_trim except that the caller knows that 2638 * the skb is not cloned so we should never get an error due to out- 2639 * of-memory. 2640 */ 2641 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2642 { 2643 int err = pskb_trim(skb, len); 2644 BUG_ON(err); 2645 } 2646 2647 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 2648 { 2649 unsigned int diff = len - skb->len; 2650 2651 if (skb_tailroom(skb) < diff) { 2652 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 2653 GFP_ATOMIC); 2654 if (ret) 2655 return ret; 2656 } 2657 __skb_set_length(skb, len); 2658 return 0; 2659 } 2660 2661 /** 2662 * skb_orphan - orphan a buffer 2663 * @skb: buffer to orphan 2664 * 2665 * If a buffer currently has an owner then we call the owner's 2666 * destructor function and make the @skb unowned. The buffer continues 2667 * to exist but is no longer charged to its former owner. 2668 */ 2669 static inline void skb_orphan(struct sk_buff *skb) 2670 { 2671 if (skb->destructor) { 2672 skb->destructor(skb); 2673 skb->destructor = NULL; 2674 skb->sk = NULL; 2675 } else { 2676 BUG_ON(skb->sk); 2677 } 2678 } 2679 2680 /** 2681 * skb_orphan_frags - orphan the frags contained in a buffer 2682 * @skb: buffer to orphan frags from 2683 * @gfp_mask: allocation mask for replacement pages 2684 * 2685 * For each frag in the SKB which needs a destructor (i.e. has an 2686 * owner) create a copy of that frag and release the original 2687 * page by calling the destructor. 2688 */ 2689 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2690 { 2691 if (likely(!skb_zcopy(skb))) 2692 return 0; 2693 if (!skb_zcopy_is_nouarg(skb) && 2694 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