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/bvec.h> 18 #include <linux/cache.h> 19 #include <linux/rbtree.h> 20 #include <linux/socket.h> 21 #include <linux/refcount.h> 22 23 #include <linux/atomic.h> 24 #include <asm/types.h> 25 #include <linux/spinlock.h> 26 #include <net/checksum.h> 27 #include <linux/rcupdate.h> 28 #include <linux/dma-mapping.h> 29 #include <linux/netdev_features.h> 30 #include <net/flow_dissector.h> 31 #include <linux/in6.h> 32 #include <linux/if_packet.h> 33 #include <linux/llist.h> 34 #include <linux/page_frag_cache.h> 35 #include <net/flow.h> 36 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 37 #include <linux/netfilter/nf_conntrack_common.h> 38 #endif 39 #include <net/net_debug.h> 40 #include <net/dropreason-core.h> 41 #include <net/netmem.h> 42 43 /** 44 * DOC: skb checksums 45 * 46 * The interface for checksum offload between the stack and networking drivers 47 * is as follows... 48 * 49 * IP checksum related features 50 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 51 * 52 * Drivers advertise checksum offload capabilities in the features of a device. 53 * From the stack's point of view these are capabilities offered by the driver. 54 * A driver typically only advertises features that it is capable of offloading 55 * to its device. 56 * 57 * .. flat-table:: Checksum related device features 58 * :widths: 1 10 59 * 60 * * - %NETIF_F_HW_CSUM 61 * - The driver (or its device) is able to compute one 62 * IP (one's complement) checksum for any combination 63 * of protocols or protocol layering. The checksum is 64 * computed and set in a packet per the CHECKSUM_PARTIAL 65 * interface (see below). 66 * 67 * * - %NETIF_F_IP_CSUM 68 * - Driver (device) is only able to checksum plain 69 * TCP or UDP packets over IPv4. These are specifically 70 * unencapsulated packets of the form IPv4|TCP or 71 * IPv4|UDP where the Protocol field in the IPv4 header 72 * is TCP or UDP. The IPv4 header may contain IP options. 73 * This feature cannot be set in features for a device 74 * with NETIF_F_HW_CSUM also set. This feature is being 75 * DEPRECATED (see below). 76 * 77 * * - %NETIF_F_IPV6_CSUM 78 * - Driver (device) is only able to checksum plain 79 * TCP or UDP packets over IPv6. These are specifically 80 * unencapsulated packets of the form IPv6|TCP or 81 * IPv6|UDP where the Next Header field in the IPv6 82 * header is either TCP or UDP. IPv6 extension headers 83 * are not supported with this feature. This feature 84 * cannot be set in features for a device with 85 * NETIF_F_HW_CSUM also set. This feature is being 86 * DEPRECATED (see below). 87 * 88 * * - %NETIF_F_RXCSUM 89 * - Driver (device) performs receive checksum offload. 90 * This flag is only used to disable the RX checksum 91 * feature for a device. The stack will accept receive 92 * checksum indication in packets received on a device 93 * regardless of whether NETIF_F_RXCSUM is set. 94 * 95 * Checksumming of received packets by device 96 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 97 * 98 * Indication of checksum verification is set in &sk_buff.ip_summed. 99 * Possible values are: 100 * 101 * - %CHECKSUM_NONE 102 * 103 * Device did not checksum this packet e.g. due to lack of capabilities. 104 * The packet contains full (though not verified) checksum in packet but 105 * not in skb->csum. Thus, skb->csum is undefined in this case. 106 * 107 * - %CHECKSUM_UNNECESSARY 108 * 109 * The hardware you're dealing with doesn't calculate the full checksum 110 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums 111 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY 112 * if their checksums are okay. &sk_buff.csum is still undefined in this case 113 * though. A driver or device must never modify the checksum field in the 114 * packet even if checksum is verified. 115 * 116 * %CHECKSUM_UNNECESSARY is applicable to following protocols: 117 * 118 * - TCP: IPv6 and IPv4. 119 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 120 * zero UDP checksum for either IPv4 or IPv6, the networking stack 121 * may perform further validation in this case. 122 * - GRE: only if the checksum is present in the header. 123 * - SCTP: indicates the CRC in SCTP header has been validated. 124 * - FCOE: indicates the CRC in FC frame has been validated. 125 * 126 * &sk_buff.csum_level indicates the number of consecutive checksums found in 127 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY. 128 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 129 * and a device is able to verify the checksums for UDP (possibly zero), 130 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to 131 * two. If the device were only able to verify the UDP checksum and not 132 * GRE, either because it doesn't support GRE checksum or because GRE 133 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 134 * not considered in this case). 135 * 136 * - %CHECKSUM_COMPLETE 137 * 138 * This is the most generic way. The device supplied checksum of the _whole_ 139 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the 140 * hardware doesn't need to parse L3/L4 headers to implement this. 141 * 142 * Notes: 143 * 144 * - Even if device supports only some protocols, but is able to produce 145 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 146 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 147 * 148 * - %CHECKSUM_PARTIAL 149 * 150 * A checksum is set up to be offloaded to a device as described in the 151 * output description for CHECKSUM_PARTIAL. This may occur on a packet 152 * received directly from another Linux OS, e.g., a virtualized Linux kernel 153 * on the same host, or it may be set in the input path in GRO or remote 154 * checksum offload. For the purposes of checksum verification, the checksum 155 * referred to by skb->csum_start + skb->csum_offset and any preceding 156 * checksums in the packet are considered verified. Any checksums in the 157 * packet that are after the checksum being offloaded are not considered to 158 * be verified. 159 * 160 * Checksumming on transmit for non-GSO 161 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 162 * 163 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet. 164 * Values are: 165 * 166 * - %CHECKSUM_PARTIAL 167 * 168 * The driver is required to checksum the packet as seen by hard_start_xmit() 169 * from &sk_buff.csum_start up to the end, and to record/write the checksum at 170 * offset &sk_buff.csum_start + &sk_buff.csum_offset. 171 * A driver may verify that the 172 * csum_start and csum_offset values are valid values given the length and 173 * offset of the packet, but it should not attempt to validate that the 174 * checksum refers to a legitimate transport layer checksum -- it is the 175 * purview of the stack to validate that csum_start and csum_offset are set 176 * correctly. 177 * 178 * When the stack requests checksum offload for a packet, the driver MUST 179 * ensure that the checksum is set correctly. A driver can either offload the 180 * checksum calculation to the device, or call skb_checksum_help (in the case 181 * that the device does not support offload for a particular checksum). 182 * 183 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of 184 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate 185 * checksum offload capability. 186 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based 187 * on network device checksumming capabilities: if a packet does not match 188 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of 189 * &sk_buff.csum_not_inet, see :ref:`crc`) 190 * is called to resolve the checksum. 191 * 192 * - %CHECKSUM_NONE 193 * 194 * The skb was already checksummed by the protocol, or a checksum is not 195 * required. 196 * 197 * - %CHECKSUM_UNNECESSARY 198 * 199 * This has the same meaning as CHECKSUM_NONE for checksum offload on 200 * output. 201 * 202 * - %CHECKSUM_COMPLETE 203 * 204 * Not used in checksum output. If a driver observes a packet with this value 205 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set. 206 * 207 * .. _crc: 208 * 209 * Non-IP checksum (CRC) offloads 210 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 211 * 212 * .. flat-table:: 213 * :widths: 1 10 214 * 215 * * - %NETIF_F_SCTP_CRC 216 * - This feature indicates that a device is capable of 217 * offloading the SCTP CRC in a packet. To perform this offload the stack 218 * will set csum_start and csum_offset accordingly, set ip_summed to 219 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication 220 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c. 221 * A driver that supports both IP checksum offload and SCTP CRC32c offload 222 * must verify which offload is configured for a packet by testing the 223 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to 224 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 225 * 226 * * - %NETIF_F_FCOE_CRC 227 * - This feature indicates that a device is capable of offloading the FCOE 228 * CRC in a packet. To perform this offload the stack will set ip_summed 229 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset 230 * accordingly. Note that there is no indication in the skbuff that the 231 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports 232 * both IP checksum offload and FCOE CRC offload must verify which offload 233 * is configured for a packet, presumably by inspecting packet headers. 234 * 235 * Checksumming on output with GSO 236 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 237 * 238 * In the case of a GSO packet (skb_is_gso() is true), checksum offload 239 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 240 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as 241 * part of the GSO operation is implied. If a checksum is being offloaded 242 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and 243 * csum_offset are set to refer to the outermost checksum being offloaded 244 * (two offloaded checksums are possible with UDP encapsulation). 245 */ 246 247 /* Don't change this without changing skb_csum_unnecessary! */ 248 #define CHECKSUM_NONE 0 249 #define CHECKSUM_UNNECESSARY 1 250 #define CHECKSUM_COMPLETE 2 251 #define CHECKSUM_PARTIAL 3 252 253 /* Maximum value in skb->csum_level */ 254 #define SKB_MAX_CSUM_LEVEL 3 255 256 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 257 #define SKB_WITH_OVERHEAD(X) \ 258 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 259 260 /* For X bytes available in skb->head, what is the minimal 261 * allocation needed, knowing struct skb_shared_info needs 262 * to be aligned. 263 */ 264 #define SKB_HEAD_ALIGN(X) (SKB_DATA_ALIGN(X) + \ 265 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 266 267 #define SKB_MAX_ORDER(X, ORDER) \ 268 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 269 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 270 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 271 272 /* return minimum truesize of one skb containing X bytes of data */ 273 #define SKB_TRUESIZE(X) ((X) + \ 274 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 275 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 276 277 struct ahash_request; 278 struct net_device; 279 struct scatterlist; 280 struct pipe_inode_info; 281 struct iov_iter; 282 struct napi_struct; 283 struct bpf_prog; 284 union bpf_attr; 285 struct skb_ext; 286 struct ts_config; 287 288 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 289 struct nf_bridge_info { 290 enum { 291 BRNF_PROTO_UNCHANGED, 292 BRNF_PROTO_8021Q, 293 BRNF_PROTO_PPPOE 294 } orig_proto:8; 295 u8 pkt_otherhost:1; 296 u8 in_prerouting:1; 297 u8 bridged_dnat:1; 298 u8 sabotage_in_done:1; 299 __u16 frag_max_size; 300 int physinif; 301 302 /* always valid & non-NULL from FORWARD on, for physdev match */ 303 struct net_device *physoutdev; 304 union { 305 /* prerouting: detect dnat in orig/reply direction */ 306 __be32 ipv4_daddr; 307 struct in6_addr ipv6_daddr; 308 309 /* after prerouting + nat detected: store original source 310 * mac since neigh resolution overwrites it, only used while 311 * skb is out in neigh layer. 312 */ 313 char neigh_header[8]; 314 }; 315 }; 316 #endif 317 318 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 319 /* Chain in tc_skb_ext will be used to share the tc chain with 320 * ovs recirc_id. It will be set to the current chain by tc 321 * and read by ovs to recirc_id. 322 */ 323 struct tc_skb_ext { 324 union { 325 u64 act_miss_cookie; 326 __u32 chain; 327 }; 328 __u16 mru; 329 __u16 zone; 330 u8 post_ct:1; 331 u8 post_ct_snat:1; 332 u8 post_ct_dnat:1; 333 u8 act_miss:1; /* Set if act_miss_cookie is used */ 334 u8 l2_miss:1; /* Set by bridge upon FDB or MDB miss */ 335 }; 336 #endif 337 338 struct sk_buff_head { 339 /* These two members must be first to match sk_buff. */ 340 struct_group_tagged(sk_buff_list, list, 341 struct sk_buff *next; 342 struct sk_buff *prev; 343 ); 344 345 __u32 qlen; 346 spinlock_t lock; 347 }; 348 349 struct sk_buff; 350 351 #ifndef CONFIG_MAX_SKB_FRAGS 352 # define CONFIG_MAX_SKB_FRAGS 17 353 #endif 354 355 #define MAX_SKB_FRAGS CONFIG_MAX_SKB_FRAGS 356 357 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 358 * segment using its current segmentation instead. 359 */ 360 #define GSO_BY_FRAGS 0xFFFF 361 362 typedef struct skb_frag { 363 netmem_ref netmem; 364 unsigned int len; 365 unsigned int offset; 366 } skb_frag_t; 367 368 /** 369 * skb_frag_size() - Returns the size of a skb fragment 370 * @frag: skb fragment 371 */ 372 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 373 { 374 return frag->len; 375 } 376 377 /** 378 * skb_frag_size_set() - Sets the size of a skb fragment 379 * @frag: skb fragment 380 * @size: size of fragment 381 */ 382 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 383 { 384 frag->len = size; 385 } 386 387 /** 388 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 389 * @frag: skb fragment 390 * @delta: value to add 391 */ 392 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 393 { 394 frag->len += delta; 395 } 396 397 /** 398 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 399 * @frag: skb fragment 400 * @delta: value to subtract 401 */ 402 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 403 { 404 frag->len -= delta; 405 } 406 407 /** 408 * skb_frag_must_loop - Test if %p is a high memory page 409 * @p: fragment's page 410 */ 411 static inline bool skb_frag_must_loop(struct page *p) 412 { 413 #if defined(CONFIG_HIGHMEM) 414 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 415 return true; 416 #endif 417 return false; 418 } 419 420 /** 421 * skb_frag_foreach_page - loop over pages in a fragment 422 * 423 * @f: skb frag to operate on 424 * @f_off: offset from start of f->netmem 425 * @f_len: length from f_off to loop over 426 * @p: (temp var) current page 427 * @p_off: (temp var) offset from start of current page, 428 * non-zero only on first page. 429 * @p_len: (temp var) length in current page, 430 * < PAGE_SIZE only on first and last page. 431 * @copied: (temp var) length so far, excluding current p_len. 432 * 433 * A fragment can hold a compound page, in which case per-page 434 * operations, notably kmap_atomic, must be called for each 435 * regular page. 436 */ 437 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 438 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 439 p_off = (f_off) & (PAGE_SIZE - 1), \ 440 p_len = skb_frag_must_loop(p) ? \ 441 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 442 copied = 0; \ 443 copied < f_len; \ 444 copied += p_len, p++, p_off = 0, \ 445 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 446 447 /** 448 * struct skb_shared_hwtstamps - hardware time stamps 449 * @hwtstamp: hardware time stamp transformed into duration 450 * since arbitrary point in time 451 * @netdev_data: address/cookie of network device driver used as 452 * reference to actual hardware time stamp 453 * 454 * Software time stamps generated by ktime_get_real() are stored in 455 * skb->tstamp. 456 * 457 * hwtstamps can only be compared against other hwtstamps from 458 * the same device. 459 * 460 * This structure is attached to packets as part of the 461 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 462 */ 463 struct skb_shared_hwtstamps { 464 union { 465 ktime_t hwtstamp; 466 void *netdev_data; 467 }; 468 }; 469 470 /* Definitions for tx_flags in struct skb_shared_info */ 471 enum { 472 /* generate hardware time stamp */ 473 SKBTX_HW_TSTAMP_NOBPF = 1 << 0, 474 475 /* generate software time stamp when queueing packet to NIC */ 476 SKBTX_SW_TSTAMP = 1 << 1, 477 478 /* device driver is going to provide hardware time stamp */ 479 SKBTX_IN_PROGRESS = 1 << 2, 480 481 /* generate hardware time stamp based on cycles if supported */ 482 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3, 483 484 /* generate wifi status information (where possible) */ 485 SKBTX_WIFI_STATUS = 1 << 4, 486 487 /* determine hardware time stamp based on time or cycles */ 488 SKBTX_HW_TSTAMP_NETDEV = 1 << 5, 489 490 /* generate software time stamp when entering packet scheduling */ 491 SKBTX_SCHED_TSTAMP = 1 << 6, 492 493 /* used for bpf extension when a bpf program is loaded */ 494 SKBTX_BPF = 1 << 7, 495 }; 496 497 #define SKBTX_HW_TSTAMP (SKBTX_HW_TSTAMP_NOBPF | SKBTX_BPF) 498 499 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 500 SKBTX_SCHED_TSTAMP | \ 501 SKBTX_BPF) 502 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \ 503 SKBTX_HW_TSTAMP_USE_CYCLES | \ 504 SKBTX_ANY_SW_TSTAMP) 505 506 /* Definitions for flags in struct skb_shared_info */ 507 enum { 508 /* use zcopy routines */ 509 SKBFL_ZEROCOPY_ENABLE = BIT(0), 510 511 /* This indicates at least one fragment might be overwritten 512 * (as in vmsplice(), sendfile() ...) 513 * If we need to compute a TX checksum, we'll need to copy 514 * all frags to avoid possible bad checksum 515 */ 516 SKBFL_SHARED_FRAG = BIT(1), 517 518 /* segment contains only zerocopy data and should not be 519 * charged to the kernel memory. 520 */ 521 SKBFL_PURE_ZEROCOPY = BIT(2), 522 523 SKBFL_DONT_ORPHAN = BIT(3), 524 525 /* page references are managed by the ubuf_info, so it's safe to 526 * use frags only up until ubuf_info is released 527 */ 528 SKBFL_MANAGED_FRAG_REFS = BIT(4), 529 }; 530 531 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG) 532 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \ 533 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS) 534 535 struct ubuf_info_ops { 536 void (*complete)(struct sk_buff *, struct ubuf_info *, 537 bool zerocopy_success); 538 /* has to be compatible with skb_zcopy_set() */ 539 int (*link_skb)(struct sk_buff *skb, struct ubuf_info *uarg); 540 }; 541 542 /* 543 * The callback notifies userspace to release buffers when skb DMA is done in 544 * lower device, the skb last reference should be 0 when calling this. 545 * The zerocopy_success argument is true if zero copy transmit occurred, 546 * false on data copy or out of memory error caused by data copy attempt. 547 * The ctx field is used to track device context. 548 * The desc field is used to track userspace buffer index. 549 */ 550 struct ubuf_info { 551 const struct ubuf_info_ops *ops; 552 refcount_t refcnt; 553 u8 flags; 554 }; 555 556 struct ubuf_info_msgzc { 557 struct ubuf_info ubuf; 558 559 union { 560 struct { 561 unsigned long desc; 562 void *ctx; 563 }; 564 struct { 565 u32 id; 566 u16 len; 567 u16 zerocopy:1; 568 u32 bytelen; 569 }; 570 }; 571 572 struct mmpin { 573 struct user_struct *user; 574 unsigned int num_pg; 575 } mmp; 576 }; 577 578 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 579 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \ 580 ubuf) 581 582 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 583 void mm_unaccount_pinned_pages(struct mmpin *mmp); 584 585 /* Preserve some data across TX submission and completion. 586 * 587 * Note, this state is stored in the driver. Extending the layout 588 * might need some special care. 589 */ 590 struct xsk_tx_metadata_compl { 591 __u64 *tx_timestamp; 592 }; 593 594 /* This data is invariant across clones and lives at 595 * the end of the header data, ie. at skb->end. 596 */ 597 struct skb_shared_info { 598 __u8 flags; 599 __u8 meta_len; 600 __u8 nr_frags; 601 __u8 tx_flags; 602 unsigned short gso_size; 603 /* Warning: this field is not always filled in (UFO)! */ 604 unsigned short gso_segs; 605 struct sk_buff *frag_list; 606 union { 607 struct skb_shared_hwtstamps hwtstamps; 608 struct xsk_tx_metadata_compl xsk_meta; 609 }; 610 unsigned int gso_type; 611 u32 tskey; 612 613 /* 614 * Warning : all fields before dataref are cleared in __alloc_skb() 615 */ 616 atomic_t dataref; 617 618 union { 619 struct { 620 u32 xdp_frags_size; 621 u32 xdp_frags_truesize; 622 }; 623 624 /* 625 * Intermediate layers must ensure that destructor_arg 626 * remains valid until skb destructor. 627 */ 628 void *destructor_arg; 629 }; 630 631 /* must be last field, see pskb_expand_head() */ 632 skb_frag_t frags[MAX_SKB_FRAGS]; 633 }; 634 635 /** 636 * DOC: dataref and headerless skbs 637 * 638 * Transport layers send out clones of payload skbs they hold for 639 * retransmissions. To allow lower layers of the stack to prepend their headers 640 * we split &skb_shared_info.dataref into two halves. 641 * The lower 16 bits count the overall number of references. 642 * The higher 16 bits indicate how many of the references are payload-only. 643 * skb_header_cloned() checks if skb is allowed to add / write the headers. 644 * 645 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr 646 * (via __skb_header_release()). Any clone created from marked skb will get 647 * &sk_buff.hdr_len populated with the available headroom. 648 * If there's the only clone in existence it's able to modify the headroom 649 * at will. The sequence of calls inside the transport layer is:: 650 * 651 * <alloc skb> 652 * skb_reserve() 653 * __skb_header_release() 654 * skb_clone() 655 * // send the clone down the stack 656 * 657 * This is not a very generic construct and it depends on the transport layers 658 * doing the right thing. In practice there's usually only one payload-only skb. 659 * Having multiple payload-only skbs with different lengths of hdr_len is not 660 * possible. The payload-only skbs should never leave their owner. 661 */ 662 #define SKB_DATAREF_SHIFT 16 663 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 664 665 666 enum { 667 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 668 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 669 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 670 }; 671 672 enum { 673 SKB_GSO_TCPV4 = 1 << 0, 674 675 /* This indicates the skb is from an untrusted source. */ 676 SKB_GSO_DODGY = 1 << 1, 677 678 /* This indicates the tcp segment has CWR set. */ 679 SKB_GSO_TCP_ECN = 1 << 2, 680 681 SKB_GSO_TCP_FIXEDID = 1 << 3, 682 683 SKB_GSO_TCPV6 = 1 << 4, 684 685 SKB_GSO_FCOE = 1 << 5, 686 687 SKB_GSO_GRE = 1 << 6, 688 689 SKB_GSO_GRE_CSUM = 1 << 7, 690 691 SKB_GSO_IPXIP4 = 1 << 8, 692 693 SKB_GSO_IPXIP6 = 1 << 9, 694 695 SKB_GSO_UDP_TUNNEL = 1 << 10, 696 697 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 698 699 SKB_GSO_PARTIAL = 1 << 12, 700 701 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 702 703 SKB_GSO_SCTP = 1 << 14, 704 705 SKB_GSO_ESP = 1 << 15, 706 707 SKB_GSO_UDP = 1 << 16, 708 709 SKB_GSO_UDP_L4 = 1 << 17, 710 711 SKB_GSO_FRAGLIST = 1 << 18, 712 }; 713 714 #if BITS_PER_LONG > 32 715 #define NET_SKBUFF_DATA_USES_OFFSET 1 716 #endif 717 718 #ifdef NET_SKBUFF_DATA_USES_OFFSET 719 typedef unsigned int sk_buff_data_t; 720 #else 721 typedef unsigned char *sk_buff_data_t; 722 #endif 723 724 enum skb_tstamp_type { 725 SKB_CLOCK_REALTIME, 726 SKB_CLOCK_MONOTONIC, 727 SKB_CLOCK_TAI, 728 __SKB_CLOCK_MAX = SKB_CLOCK_TAI, 729 }; 730 731 /** 732 * DOC: Basic sk_buff geometry 733 * 734 * struct sk_buff itself is a metadata structure and does not hold any packet 735 * data. All the data is held in associated buffers. 736 * 737 * &sk_buff.head points to the main "head" buffer. The head buffer is divided 738 * into two parts: 739 * 740 * - data buffer, containing headers and sometimes payload; 741 * this is the part of the skb operated on by the common helpers 742 * such as skb_put() or skb_pull(); 743 * - shared info (struct skb_shared_info) which holds an array of pointers 744 * to read-only data in the (page, offset, length) format. 745 * 746 * Optionally &skb_shared_info.frag_list may point to another skb. 747 * 748 * Basic diagram may look like this:: 749 * 750 * --------------- 751 * | sk_buff | 752 * --------------- 753 * ,--------------------------- + head 754 * / ,----------------- + data 755 * / / ,----------- + tail 756 * | | | , + end 757 * | | | | 758 * v v v v 759 * ----------------------------------------------- 760 * | headroom | data | tailroom | skb_shared_info | 761 * ----------------------------------------------- 762 * + [page frag] 763 * + [page frag] 764 * + [page frag] 765 * + [page frag] --------- 766 * + frag_list --> | sk_buff | 767 * --------- 768 * 769 */ 770 771 /** 772 * struct sk_buff - socket buffer 773 * @next: Next buffer in list 774 * @prev: Previous buffer in list 775 * @tstamp: Time we arrived/left 776 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 777 * for retransmit timer 778 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 779 * @list: queue head 780 * @ll_node: anchor in an llist (eg socket defer_list) 781 * @sk: Socket we are owned by 782 * @dev: Device we arrived on/are leaving by 783 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 784 * @cb: Control buffer. Free for use by every layer. Put private vars here 785 * @_skb_refdst: destination entry (with norefcount bit) 786 * @len: Length of actual data 787 * @data_len: Data length 788 * @mac_len: Length of link layer header 789 * @hdr_len: writable header length of cloned skb 790 * @csum: Checksum (must include start/offset pair) 791 * @csum_start: Offset from skb->head where checksumming should start 792 * @csum_offset: Offset from csum_start where checksum should be stored 793 * @priority: Packet queueing priority 794 * @ignore_df: allow local fragmentation 795 * @cloned: Head may be cloned (check refcnt to be sure) 796 * @ip_summed: Driver fed us an IP checksum 797 * @nohdr: Payload reference only, must not modify header 798 * @pkt_type: Packet class 799 * @fclone: skbuff clone status 800 * @ipvs_property: skbuff is owned by ipvs 801 * @inner_protocol_type: whether the inner protocol is 802 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 803 * @remcsum_offload: remote checksum offload is enabled 804 * @offload_fwd_mark: Packet was L2-forwarded in hardware 805 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 806 * @tc_skip_classify: do not classify packet. set by IFB device 807 * @tc_at_ingress: used within tc_classify to distinguish in/egress 808 * @redirected: packet was redirected by packet classifier 809 * @from_ingress: packet was redirected from the ingress path 810 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 811 * @peeked: this packet has been seen already, so stats have been 812 * done for it, don't do them again 813 * @nf_trace: netfilter packet trace flag 814 * @protocol: Packet protocol from driver 815 * @destructor: Destruct function 816 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 817 * @_sk_redir: socket redirection information for skmsg 818 * @_nfct: Associated connection, if any (with nfctinfo bits) 819 * @skb_iif: ifindex of device we arrived on 820 * @tc_index: Traffic control index 821 * @hash: the packet hash 822 * @queue_mapping: Queue mapping for multiqueue devices 823 * @head_frag: skb was allocated from page fragments, 824 * not allocated by kmalloc() or vmalloc(). 825 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 826 * @pp_recycle: mark the packet for recycling instead of freeing (implies 827 * page_pool support on driver) 828 * @active_extensions: active extensions (skb_ext_id types) 829 * @ndisc_nodetype: router type (from link layer) 830 * @ooo_okay: allow the mapping of a socket to a queue to be changed 831 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 832 * ports. 833 * @sw_hash: indicates hash was computed in software stack 834 * @wifi_acked_valid: wifi_acked was set 835 * @wifi_acked: whether frame was acked on wifi or not 836 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 837 * @encapsulation: indicates the inner headers in the skbuff are valid 838 * @encap_hdr_csum: software checksum is needed 839 * @csum_valid: checksum is already valid 840 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 841 * @csum_complete_sw: checksum was completed by software 842 * @csum_level: indicates the number of consecutive checksums found in 843 * the packet minus one that have been verified as 844 * CHECKSUM_UNNECESSARY (max 3) 845 * @unreadable: indicates that at least 1 of the fragments in this skb is 846 * unreadable. 847 * @dst_pending_confirm: need to confirm neighbour 848 * @decrypted: Decrypted SKB 849 * @slow_gro: state present at GRO time, slower prepare step required 850 * @tstamp_type: When set, skb->tstamp has the 851 * delivery_time clock base of skb->tstamp. 852 * @napi_id: id of the NAPI struct this skb came from 853 * @sender_cpu: (aka @napi_id) source CPU in XPS 854 * @alloc_cpu: CPU which did the skb allocation. 855 * @secmark: security marking 856 * @mark: Generic packet mark 857 * @reserved_tailroom: (aka @mark) number of bytes of free space available 858 * at the tail of an sk_buff 859 * @vlan_all: vlan fields (proto & tci) 860 * @vlan_proto: vlan encapsulation protocol 861 * @vlan_tci: vlan tag control information 862 * @inner_protocol: Protocol (encapsulation) 863 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 864 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 865 * @inner_transport_header: Inner transport layer header (encapsulation) 866 * @inner_network_header: Network layer header (encapsulation) 867 * @inner_mac_header: Link layer header (encapsulation) 868 * @transport_header: Transport layer header 869 * @network_header: Network layer header 870 * @mac_header: Link layer header 871 * @kcov_handle: KCOV remote handle for remote coverage collection 872 * @tail: Tail pointer 873 * @end: End pointer 874 * @head: Head of buffer 875 * @data: Data head pointer 876 * @truesize: Buffer size 877 * @users: User count - see {datagram,tcp}.c 878 * @extensions: allocated extensions, valid if active_extensions is nonzero 879 */ 880 881 struct sk_buff { 882 union { 883 struct { 884 /* These two members must be first to match sk_buff_head. */ 885 struct sk_buff *next; 886 struct sk_buff *prev; 887 888 union { 889 struct net_device *dev; 890 /* Some protocols might use this space to store information, 891 * while device pointer would be NULL. 892 * UDP receive path is one user. 893 */ 894 unsigned long dev_scratch; 895 }; 896 }; 897 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 898 struct list_head list; 899 struct llist_node ll_node; 900 }; 901 902 struct sock *sk; 903 904 union { 905 ktime_t tstamp; 906 u64 skb_mstamp_ns; /* earliest departure time */ 907 }; 908 /* 909 * This is the control buffer. It is free to use for every 910 * layer. Please put your private variables there. If you 911 * want to keep them across layers you have to do a skb_clone() 912 * first. This is owned by whoever has the skb queued ATM. 913 */ 914 char cb[48] __aligned(8); 915 916 union { 917 struct { 918 unsigned long _skb_refdst; 919 void (*destructor)(struct sk_buff *skb); 920 }; 921 struct list_head tcp_tsorted_anchor; 922 #ifdef CONFIG_NET_SOCK_MSG 923 unsigned long _sk_redir; 924 #endif 925 }; 926 927 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 928 unsigned long _nfct; 929 #endif 930 unsigned int len, 931 data_len; 932 __u16 mac_len, 933 hdr_len; 934 935 /* Following fields are _not_ copied in __copy_skb_header() 936 * Note that queue_mapping is here mostly to fill a hole. 937 */ 938 __u16 queue_mapping; 939 940 /* if you move cloned around you also must adapt those constants */ 941 #ifdef __BIG_ENDIAN_BITFIELD 942 #define CLONED_MASK (1 << 7) 943 #else 944 #define CLONED_MASK 1 945 #endif 946 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 947 948 /* private: */ 949 __u8 __cloned_offset[0]; 950 /* public: */ 951 __u8 cloned:1, 952 nohdr:1, 953 fclone:2, 954 peeked:1, 955 head_frag:1, 956 pfmemalloc:1, 957 pp_recycle:1; /* page_pool recycle indicator */ 958 #ifdef CONFIG_SKB_EXTENSIONS 959 __u8 active_extensions; 960 #endif 961 962 /* Fields enclosed in headers group are copied 963 * using a single memcpy() in __copy_skb_header() 964 */ 965 struct_group(headers, 966 967 /* private: */ 968 __u8 __pkt_type_offset[0]; 969 /* public: */ 970 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 971 __u8 ignore_df:1; 972 __u8 dst_pending_confirm:1; 973 __u8 ip_summed:2; 974 __u8 ooo_okay:1; 975 976 /* private: */ 977 __u8 __mono_tc_offset[0]; 978 /* public: */ 979 __u8 tstamp_type:2; /* See skb_tstamp_type */ 980 #ifdef CONFIG_NET_XGRESS 981 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 982 __u8 tc_skip_classify:1; 983 #endif 984 __u8 remcsum_offload:1; 985 __u8 csum_complete_sw:1; 986 __u8 csum_level:2; 987 __u8 inner_protocol_type:1; 988 989 __u8 l4_hash:1; 990 __u8 sw_hash:1; 991 #ifdef CONFIG_WIRELESS 992 __u8 wifi_acked_valid:1; 993 __u8 wifi_acked:1; 994 #endif 995 __u8 no_fcs:1; 996 /* Indicates the inner headers are valid in the skbuff. */ 997 __u8 encapsulation:1; 998 __u8 encap_hdr_csum:1; 999 __u8 csum_valid:1; 1000 #ifdef CONFIG_IPV6_NDISC_NODETYPE 1001 __u8 ndisc_nodetype:2; 1002 #endif 1003 1004 #if IS_ENABLED(CONFIG_IP_VS) 1005 __u8 ipvs_property:1; 1006 #endif 1007 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 1008 __u8 nf_trace:1; 1009 #endif 1010 #ifdef CONFIG_NET_SWITCHDEV 1011 __u8 offload_fwd_mark:1; 1012 __u8 offload_l3_fwd_mark:1; 1013 #endif 1014 __u8 redirected:1; 1015 #ifdef CONFIG_NET_REDIRECT 1016 __u8 from_ingress:1; 1017 #endif 1018 #ifdef CONFIG_NETFILTER_SKIP_EGRESS 1019 __u8 nf_skip_egress:1; 1020 #endif 1021 #ifdef CONFIG_SKB_DECRYPTED 1022 __u8 decrypted:1; 1023 #endif 1024 __u8 slow_gro:1; 1025 #if IS_ENABLED(CONFIG_IP_SCTP) 1026 __u8 csum_not_inet:1; 1027 #endif 1028 __u8 unreadable:1; 1029 #if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS) 1030 __u16 tc_index; /* traffic control index */ 1031 #endif 1032 1033 u16 alloc_cpu; 1034 1035 union { 1036 __wsum csum; 1037 struct { 1038 __u16 csum_start; 1039 __u16 csum_offset; 1040 }; 1041 }; 1042 __u32 priority; 1043 int skb_iif; 1044 __u32 hash; 1045 union { 1046 u32 vlan_all; 1047 struct { 1048 __be16 vlan_proto; 1049 __u16 vlan_tci; 1050 }; 1051 }; 1052 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1053 union { 1054 unsigned int napi_id; 1055 unsigned int sender_cpu; 1056 }; 1057 #endif 1058 #ifdef CONFIG_NETWORK_SECMARK 1059 __u32 secmark; 1060 #endif 1061 1062 union { 1063 __u32 mark; 1064 __u32 reserved_tailroom; 1065 }; 1066 1067 union { 1068 __be16 inner_protocol; 1069 __u8 inner_ipproto; 1070 }; 1071 1072 __u16 inner_transport_header; 1073 __u16 inner_network_header; 1074 __u16 inner_mac_header; 1075 1076 __be16 protocol; 1077 __u16 transport_header; 1078 __u16 network_header; 1079 __u16 mac_header; 1080 1081 #ifdef CONFIG_KCOV 1082 u64 kcov_handle; 1083 #endif 1084 1085 ); /* end headers group */ 1086 1087 /* These elements must be at the end, see alloc_skb() for details. */ 1088 sk_buff_data_t tail; 1089 sk_buff_data_t end; 1090 unsigned char *head, 1091 *data; 1092 unsigned int truesize; 1093 refcount_t users; 1094 1095 #ifdef CONFIG_SKB_EXTENSIONS 1096 /* only usable after checking ->active_extensions != 0 */ 1097 struct skb_ext *extensions; 1098 #endif 1099 }; 1100 1101 /* if you move pkt_type around you also must adapt those constants */ 1102 #ifdef __BIG_ENDIAN_BITFIELD 1103 #define PKT_TYPE_MAX (7 << 5) 1104 #else 1105 #define PKT_TYPE_MAX 7 1106 #endif 1107 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1108 1109 /* if you move tc_at_ingress or tstamp_type 1110 * around, you also must adapt these constants. 1111 */ 1112 #ifdef __BIG_ENDIAN_BITFIELD 1113 #define SKB_TSTAMP_TYPE_MASK (3 << 6) 1114 #define SKB_TSTAMP_TYPE_RSHIFT (6) 1115 #define TC_AT_INGRESS_MASK (1 << 5) 1116 #else 1117 #define SKB_TSTAMP_TYPE_MASK (3) 1118 #define TC_AT_INGRESS_MASK (1 << 2) 1119 #endif 1120 #define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset) 1121 1122 #ifdef __KERNEL__ 1123 /* 1124 * Handling routines are only of interest to the kernel 1125 */ 1126 1127 #define SKB_ALLOC_FCLONE 0x01 1128 #define SKB_ALLOC_RX 0x02 1129 #define SKB_ALLOC_NAPI 0x04 1130 1131 /** 1132 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1133 * @skb: buffer 1134 */ 1135 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1136 { 1137 return unlikely(skb->pfmemalloc); 1138 } 1139 1140 /* 1141 * skb might have a dst pointer attached, refcounted or not. 1142 * _skb_refdst low order bit is set if refcount was _not_ taken 1143 */ 1144 #define SKB_DST_NOREF 1UL 1145 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1146 1147 /** 1148 * skb_dst - returns skb dst_entry 1149 * @skb: buffer 1150 * 1151 * Returns: skb dst_entry, regardless of reference taken or not. 1152 */ 1153 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1154 { 1155 /* If refdst was not refcounted, check we still are in a 1156 * rcu_read_lock section 1157 */ 1158 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1159 !rcu_read_lock_held() && 1160 !rcu_read_lock_bh_held()); 1161 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1162 } 1163 1164 /** 1165 * skb_dst_set - sets skb dst 1166 * @skb: buffer 1167 * @dst: dst entry 1168 * 1169 * Sets skb dst, assuming a reference was taken on dst and should 1170 * be released by skb_dst_drop() 1171 */ 1172 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1173 { 1174 skb->slow_gro |= !!dst; 1175 skb->_skb_refdst = (unsigned long)dst; 1176 } 1177 1178 /** 1179 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1180 * @skb: buffer 1181 * @dst: dst entry 1182 * 1183 * Sets skb dst, assuming a reference was not taken on dst. 1184 * If dst entry is cached, we do not take reference and dst_release 1185 * will be avoided by refdst_drop. If dst entry is not cached, we take 1186 * reference, so that last dst_release can destroy the dst immediately. 1187 */ 1188 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1189 { 1190 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1191 skb->slow_gro |= !!dst; 1192 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1193 } 1194 1195 /** 1196 * skb_dst_is_noref - Test if skb dst isn't refcounted 1197 * @skb: buffer 1198 */ 1199 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1200 { 1201 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1202 } 1203 1204 /* For mangling skb->pkt_type from user space side from applications 1205 * such as nft, tc, etc, we only allow a conservative subset of 1206 * possible pkt_types to be set. 1207 */ 1208 static inline bool skb_pkt_type_ok(u32 ptype) 1209 { 1210 return ptype <= PACKET_OTHERHOST; 1211 } 1212 1213 /** 1214 * skb_napi_id - Returns the skb's NAPI id 1215 * @skb: buffer 1216 */ 1217 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1218 { 1219 #ifdef CONFIG_NET_RX_BUSY_POLL 1220 return skb->napi_id; 1221 #else 1222 return 0; 1223 #endif 1224 } 1225 1226 static inline bool skb_wifi_acked_valid(const struct sk_buff *skb) 1227 { 1228 #ifdef CONFIG_WIRELESS 1229 return skb->wifi_acked_valid; 1230 #else 1231 return 0; 1232 #endif 1233 } 1234 1235 /** 1236 * skb_unref - decrement the skb's reference count 1237 * @skb: buffer 1238 * 1239 * Returns: true if we can free the skb. 1240 */ 1241 static inline bool skb_unref(struct sk_buff *skb) 1242 { 1243 if (unlikely(!skb)) 1244 return false; 1245 if (!IS_ENABLED(CONFIG_DEBUG_NET) && likely(refcount_read(&skb->users) == 1)) 1246 smp_rmb(); 1247 else if (likely(!refcount_dec_and_test(&skb->users))) 1248 return false; 1249 1250 return true; 1251 } 1252 1253 static inline bool skb_data_unref(const struct sk_buff *skb, 1254 struct skb_shared_info *shinfo) 1255 { 1256 int bias; 1257 1258 if (!skb->cloned) 1259 return true; 1260 1261 bias = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1; 1262 1263 if (atomic_read(&shinfo->dataref) == bias) 1264 smp_rmb(); 1265 else if (atomic_sub_return(bias, &shinfo->dataref)) 1266 return false; 1267 1268 return true; 1269 } 1270 1271 void __fix_address sk_skb_reason_drop(struct sock *sk, struct sk_buff *skb, 1272 enum skb_drop_reason reason); 1273 1274 static inline void 1275 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason) 1276 { 1277 sk_skb_reason_drop(NULL, skb, reason); 1278 } 1279 1280 /** 1281 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1282 * @skb: buffer to free 1283 */ 1284 static inline void kfree_skb(struct sk_buff *skb) 1285 { 1286 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1287 } 1288 1289 void skb_release_head_state(struct sk_buff *skb); 1290 void kfree_skb_list_reason(struct sk_buff *segs, 1291 enum skb_drop_reason reason); 1292 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1293 void skb_tx_error(struct sk_buff *skb); 1294 1295 static inline void kfree_skb_list(struct sk_buff *segs) 1296 { 1297 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1298 } 1299 1300 #ifdef CONFIG_TRACEPOINTS 1301 void consume_skb(struct sk_buff *skb); 1302 #else 1303 static inline void consume_skb(struct sk_buff *skb) 1304 { 1305 return kfree_skb(skb); 1306 } 1307 #endif 1308 1309 void __consume_stateless_skb(struct sk_buff *skb); 1310 void __kfree_skb(struct sk_buff *skb); 1311 1312 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1313 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1314 bool *fragstolen, int *delta_truesize); 1315 1316 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1317 int node); 1318 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1319 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1320 struct sk_buff *build_skb_around(struct sk_buff *skb, 1321 void *data, unsigned int frag_size); 1322 void skb_attempt_defer_free(struct sk_buff *skb); 1323 1324 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1325 struct sk_buff *slab_build_skb(void *data); 1326 1327 /** 1328 * alloc_skb - allocate a network buffer 1329 * @size: size to allocate 1330 * @priority: allocation mask 1331 * 1332 * This function is a convenient wrapper around __alloc_skb(). 1333 */ 1334 static inline struct sk_buff *alloc_skb(unsigned int size, 1335 gfp_t priority) 1336 { 1337 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1338 } 1339 1340 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1341 unsigned long data_len, 1342 int max_page_order, 1343 int *errcode, 1344 gfp_t gfp_mask); 1345 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1346 1347 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1348 struct sk_buff_fclones { 1349 struct sk_buff skb1; 1350 1351 struct sk_buff skb2; 1352 1353 refcount_t fclone_ref; 1354 }; 1355 1356 /** 1357 * skb_fclone_busy - check if fclone is busy 1358 * @sk: socket 1359 * @skb: buffer 1360 * 1361 * Returns: true if skb is a fast clone, and its clone is not freed. 1362 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1363 * so we also check that didn't happen. 1364 */ 1365 static inline bool skb_fclone_busy(const struct sock *sk, 1366 const struct sk_buff *skb) 1367 { 1368 const struct sk_buff_fclones *fclones; 1369 1370 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1371 1372 return skb->fclone == SKB_FCLONE_ORIG && 1373 refcount_read(&fclones->fclone_ref) > 1 && 1374 READ_ONCE(fclones->skb2.sk) == sk; 1375 } 1376 1377 /** 1378 * alloc_skb_fclone - allocate a network buffer from fclone cache 1379 * @size: size to allocate 1380 * @priority: allocation mask 1381 * 1382 * This function is a convenient wrapper around __alloc_skb(). 1383 */ 1384 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1385 gfp_t priority) 1386 { 1387 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1388 } 1389 1390 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1391 void skb_headers_offset_update(struct sk_buff *skb, int off); 1392 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1393 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1394 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1395 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1396 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1397 gfp_t gfp_mask, bool fclone); 1398 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1399 gfp_t gfp_mask) 1400 { 1401 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1402 } 1403 1404 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1405 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1406 unsigned int headroom); 1407 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1408 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1409 int newtailroom, gfp_t priority); 1410 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1411 int offset, int len); 1412 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1413 int offset, int len); 1414 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1415 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1416 1417 /** 1418 * skb_pad - zero pad the tail of an skb 1419 * @skb: buffer to pad 1420 * @pad: space to pad 1421 * 1422 * Ensure that a buffer is followed by a padding area that is zero 1423 * filled. Used by network drivers which may DMA or transfer data 1424 * beyond the buffer end onto the wire. 1425 * 1426 * May return error in out of memory cases. The skb is freed on error. 1427 */ 1428 static inline int skb_pad(struct sk_buff *skb, int pad) 1429 { 1430 return __skb_pad(skb, pad, true); 1431 } 1432 #define dev_kfree_skb(a) consume_skb(a) 1433 1434 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1435 int offset, size_t size, size_t max_frags); 1436 1437 struct skb_seq_state { 1438 __u32 lower_offset; 1439 __u32 upper_offset; 1440 __u32 frag_idx; 1441 __u32 stepped_offset; 1442 struct sk_buff *root_skb; 1443 struct sk_buff *cur_skb; 1444 __u8 *frag_data; 1445 __u32 frag_off; 1446 }; 1447 1448 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1449 unsigned int to, struct skb_seq_state *st); 1450 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1451 struct skb_seq_state *st); 1452 void skb_abort_seq_read(struct skb_seq_state *st); 1453 int skb_copy_seq_read(struct skb_seq_state *st, int offset, void *to, int len); 1454 1455 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1456 unsigned int to, struct ts_config *config); 1457 1458 /* 1459 * Packet hash types specify the type of hash in skb_set_hash. 1460 * 1461 * Hash types refer to the protocol layer addresses which are used to 1462 * construct a packet's hash. The hashes are used to differentiate or identify 1463 * flows of the protocol layer for the hash type. Hash types are either 1464 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1465 * 1466 * Properties of hashes: 1467 * 1468 * 1) Two packets in different flows have different hash values 1469 * 2) Two packets in the same flow should have the same hash value 1470 * 1471 * A hash at a higher layer is considered to be more specific. A driver should 1472 * set the most specific hash possible. 1473 * 1474 * A driver cannot indicate a more specific hash than the layer at which a hash 1475 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1476 * 1477 * A driver may indicate a hash level which is less specific than the 1478 * actual layer the hash was computed on. For instance, a hash computed 1479 * at L4 may be considered an L3 hash. This should only be done if the 1480 * driver can't unambiguously determine that the HW computed the hash at 1481 * the higher layer. Note that the "should" in the second property above 1482 * permits this. 1483 */ 1484 enum pkt_hash_types { 1485 PKT_HASH_TYPE_NONE, /* Undefined type */ 1486 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1487 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1488 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1489 }; 1490 1491 static inline void skb_clear_hash(struct sk_buff *skb) 1492 { 1493 skb->hash = 0; 1494 skb->sw_hash = 0; 1495 skb->l4_hash = 0; 1496 } 1497 1498 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1499 { 1500 if (!skb->l4_hash) 1501 skb_clear_hash(skb); 1502 } 1503 1504 static inline void 1505 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1506 { 1507 skb->l4_hash = is_l4; 1508 skb->sw_hash = is_sw; 1509 skb->hash = hash; 1510 } 1511 1512 static inline void 1513 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1514 { 1515 /* Used by drivers to set hash from HW */ 1516 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1517 } 1518 1519 static inline void 1520 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1521 { 1522 __skb_set_hash(skb, hash, true, is_l4); 1523 } 1524 1525 u32 __skb_get_hash_symmetric_net(const struct net *net, const struct sk_buff *skb); 1526 1527 static inline u32 __skb_get_hash_symmetric(const struct sk_buff *skb) 1528 { 1529 return __skb_get_hash_symmetric_net(NULL, skb); 1530 } 1531 1532 void __skb_get_hash_net(const struct net *net, struct sk_buff *skb); 1533 u32 skb_get_poff(const struct sk_buff *skb); 1534 u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1535 const struct flow_keys_basic *keys, int hlen); 1536 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1537 const void *data, int hlen_proto); 1538 1539 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1540 int thoff, u8 ip_proto) 1541 { 1542 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1543 } 1544 1545 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1546 const struct flow_dissector_key *key, 1547 unsigned int key_count); 1548 1549 struct bpf_flow_dissector; 1550 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1551 __be16 proto, int nhoff, int hlen, unsigned int flags); 1552 1553 bool __skb_flow_dissect(const struct net *net, 1554 const struct sk_buff *skb, 1555 struct flow_dissector *flow_dissector, 1556 void *target_container, const void *data, 1557 __be16 proto, int nhoff, int hlen, unsigned int flags); 1558 1559 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1560 struct flow_dissector *flow_dissector, 1561 void *target_container, unsigned int flags) 1562 { 1563 return __skb_flow_dissect(NULL, skb, flow_dissector, 1564 target_container, NULL, 0, 0, 0, flags); 1565 } 1566 1567 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1568 struct flow_keys *flow, 1569 unsigned int flags) 1570 { 1571 memset(flow, 0, sizeof(*flow)); 1572 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1573 flow, NULL, 0, 0, 0, flags); 1574 } 1575 1576 static inline bool 1577 skb_flow_dissect_flow_keys_basic(const struct net *net, 1578 const struct sk_buff *skb, 1579 struct flow_keys_basic *flow, 1580 const void *data, __be16 proto, 1581 int nhoff, int hlen, unsigned int flags) 1582 { 1583 memset(flow, 0, sizeof(*flow)); 1584 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1585 data, proto, nhoff, hlen, flags); 1586 } 1587 1588 void skb_flow_dissect_meta(const struct sk_buff *skb, 1589 struct flow_dissector *flow_dissector, 1590 void *target_container); 1591 1592 /* Gets a skb connection tracking info, ctinfo map should be a 1593 * map of mapsize to translate enum ip_conntrack_info states 1594 * to user states. 1595 */ 1596 void 1597 skb_flow_dissect_ct(const struct sk_buff *skb, 1598 struct flow_dissector *flow_dissector, 1599 void *target_container, 1600 u16 *ctinfo_map, size_t mapsize, 1601 bool post_ct, u16 zone); 1602 void 1603 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1604 struct flow_dissector *flow_dissector, 1605 void *target_container); 1606 1607 void skb_flow_dissect_hash(const struct sk_buff *skb, 1608 struct flow_dissector *flow_dissector, 1609 void *target_container); 1610 1611 static inline __u32 skb_get_hash_net(const struct net *net, struct sk_buff *skb) 1612 { 1613 if (!skb->l4_hash && !skb->sw_hash) 1614 __skb_get_hash_net(net, skb); 1615 1616 return skb->hash; 1617 } 1618 1619 static inline __u32 skb_get_hash(struct sk_buff *skb) 1620 { 1621 if (!skb->l4_hash && !skb->sw_hash) 1622 __skb_get_hash_net(NULL, skb); 1623 1624 return skb->hash; 1625 } 1626 1627 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1628 { 1629 if (!skb->l4_hash && !skb->sw_hash) { 1630 struct flow_keys keys; 1631 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1632 1633 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1634 } 1635 1636 return skb->hash; 1637 } 1638 1639 __u32 skb_get_hash_perturb(const struct sk_buff *skb, 1640 const siphash_key_t *perturb); 1641 1642 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1643 { 1644 return skb->hash; 1645 } 1646 1647 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1648 { 1649 to->hash = from->hash; 1650 to->sw_hash = from->sw_hash; 1651 to->l4_hash = from->l4_hash; 1652 }; 1653 1654 static inline int skb_cmp_decrypted(const struct sk_buff *skb1, 1655 const struct sk_buff *skb2) 1656 { 1657 #ifdef CONFIG_SKB_DECRYPTED 1658 return skb2->decrypted - skb1->decrypted; 1659 #else 1660 return 0; 1661 #endif 1662 } 1663 1664 static inline bool skb_is_decrypted(const struct sk_buff *skb) 1665 { 1666 #ifdef CONFIG_SKB_DECRYPTED 1667 return skb->decrypted; 1668 #else 1669 return false; 1670 #endif 1671 } 1672 1673 static inline void skb_copy_decrypted(struct sk_buff *to, 1674 const struct sk_buff *from) 1675 { 1676 #ifdef CONFIG_SKB_DECRYPTED 1677 to->decrypted = from->decrypted; 1678 #endif 1679 } 1680 1681 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1682 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1683 { 1684 return skb->head + skb->end; 1685 } 1686 1687 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1688 { 1689 return skb->end; 1690 } 1691 1692 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1693 { 1694 skb->end = offset; 1695 } 1696 #else 1697 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1698 { 1699 return skb->end; 1700 } 1701 1702 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1703 { 1704 return skb->end - skb->head; 1705 } 1706 1707 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1708 { 1709 skb->end = skb->head + offset; 1710 } 1711 #endif 1712 1713 extern const struct ubuf_info_ops msg_zerocopy_ubuf_ops; 1714 1715 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1716 struct ubuf_info *uarg); 1717 1718 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1719 1720 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, 1721 struct sk_buff *skb, struct iov_iter *from, 1722 size_t length); 1723 1724 int zerocopy_fill_skb_from_iter(struct sk_buff *skb, 1725 struct iov_iter *from, size_t length); 1726 1727 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1728 struct msghdr *msg, int len) 1729 { 1730 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len); 1731 } 1732 1733 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1734 struct msghdr *msg, int len, 1735 struct ubuf_info *uarg); 1736 1737 /* Internal */ 1738 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1739 1740 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1741 { 1742 return &skb_shinfo(skb)->hwtstamps; 1743 } 1744 1745 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1746 { 1747 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1748 1749 return is_zcopy ? skb_uarg(skb) : NULL; 1750 } 1751 1752 static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1753 { 1754 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1755 } 1756 1757 static inline bool skb_zcopy_managed(const struct sk_buff *skb) 1758 { 1759 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS; 1760 } 1761 1762 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1763 const struct sk_buff *skb2) 1764 { 1765 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1766 } 1767 1768 static inline void net_zcopy_get(struct ubuf_info *uarg) 1769 { 1770 refcount_inc(&uarg->refcnt); 1771 } 1772 1773 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1774 { 1775 skb_shinfo(skb)->destructor_arg = uarg; 1776 skb_shinfo(skb)->flags |= uarg->flags; 1777 } 1778 1779 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1780 bool *have_ref) 1781 { 1782 if (skb && uarg && !skb_zcopy(skb)) { 1783 if (unlikely(have_ref && *have_ref)) 1784 *have_ref = false; 1785 else 1786 net_zcopy_get(uarg); 1787 skb_zcopy_init(skb, uarg); 1788 } 1789 } 1790 1791 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1792 { 1793 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1794 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1795 } 1796 1797 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1798 { 1799 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1800 } 1801 1802 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1803 { 1804 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1805 } 1806 1807 static inline void net_zcopy_put(struct ubuf_info *uarg) 1808 { 1809 if (uarg) 1810 uarg->ops->complete(NULL, uarg, true); 1811 } 1812 1813 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1814 { 1815 if (uarg) { 1816 if (uarg->ops == &msg_zerocopy_ubuf_ops) 1817 msg_zerocopy_put_abort(uarg, have_uref); 1818 else if (have_uref) 1819 net_zcopy_put(uarg); 1820 } 1821 } 1822 1823 /* Release a reference on a zerocopy structure */ 1824 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1825 { 1826 struct ubuf_info *uarg = skb_zcopy(skb); 1827 1828 if (uarg) { 1829 if (!skb_zcopy_is_nouarg(skb)) 1830 uarg->ops->complete(skb, uarg, zerocopy_success); 1831 1832 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1833 } 1834 } 1835 1836 void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1837 1838 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1839 { 1840 if (unlikely(skb_zcopy_managed(skb))) 1841 __skb_zcopy_downgrade_managed(skb); 1842 } 1843 1844 /* Return true if frags in this skb are readable by the host. */ 1845 static inline bool skb_frags_readable(const struct sk_buff *skb) 1846 { 1847 return !skb->unreadable; 1848 } 1849 1850 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1851 { 1852 skb->next = NULL; 1853 } 1854 1855 static inline void skb_poison_list(struct sk_buff *skb) 1856 { 1857 #ifdef CONFIG_DEBUG_NET 1858 skb->next = SKB_LIST_POISON_NEXT; 1859 #endif 1860 } 1861 1862 /* Iterate through singly-linked GSO fragments of an skb. */ 1863 #define skb_list_walk_safe(first, skb, next_skb) \ 1864 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1865 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1866 1867 static inline void skb_list_del_init(struct sk_buff *skb) 1868 { 1869 __list_del_entry(&skb->list); 1870 skb_mark_not_on_list(skb); 1871 } 1872 1873 /** 1874 * skb_queue_empty - check if a queue is empty 1875 * @list: queue head 1876 * 1877 * Returns true if the queue is empty, false otherwise. 1878 */ 1879 static inline int skb_queue_empty(const struct sk_buff_head *list) 1880 { 1881 return list->next == (const struct sk_buff *) list; 1882 } 1883 1884 /** 1885 * skb_queue_empty_lockless - check if a queue is empty 1886 * @list: queue head 1887 * 1888 * Returns true if the queue is empty, false otherwise. 1889 * This variant can be used in lockless contexts. 1890 */ 1891 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1892 { 1893 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1894 } 1895 1896 1897 /** 1898 * skb_queue_is_last - check if skb is the last entry in the queue 1899 * @list: queue head 1900 * @skb: buffer 1901 * 1902 * Returns true if @skb is the last buffer on the list. 1903 */ 1904 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1905 const struct sk_buff *skb) 1906 { 1907 return skb->next == (const struct sk_buff *) list; 1908 } 1909 1910 /** 1911 * skb_queue_is_first - check if skb is the first entry in the queue 1912 * @list: queue head 1913 * @skb: buffer 1914 * 1915 * Returns true if @skb is the first buffer on the list. 1916 */ 1917 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1918 const struct sk_buff *skb) 1919 { 1920 return skb->prev == (const struct sk_buff *) list; 1921 } 1922 1923 /** 1924 * skb_queue_next - return the next packet in the queue 1925 * @list: queue head 1926 * @skb: current buffer 1927 * 1928 * Return the next packet in @list after @skb. It is only valid to 1929 * call this if skb_queue_is_last() evaluates to false. 1930 */ 1931 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1932 const struct sk_buff *skb) 1933 { 1934 /* This BUG_ON may seem severe, but if we just return then we 1935 * are going to dereference garbage. 1936 */ 1937 BUG_ON(skb_queue_is_last(list, skb)); 1938 return skb->next; 1939 } 1940 1941 /** 1942 * skb_queue_prev - return the prev packet in the queue 1943 * @list: queue head 1944 * @skb: current buffer 1945 * 1946 * Return the prev packet in @list before @skb. It is only valid to 1947 * call this if skb_queue_is_first() evaluates to false. 1948 */ 1949 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1950 const struct sk_buff *skb) 1951 { 1952 /* This BUG_ON may seem severe, but if we just return then we 1953 * are going to dereference garbage. 1954 */ 1955 BUG_ON(skb_queue_is_first(list, skb)); 1956 return skb->prev; 1957 } 1958 1959 /** 1960 * skb_get - reference buffer 1961 * @skb: buffer to reference 1962 * 1963 * Makes another reference to a socket buffer and returns a pointer 1964 * to the buffer. 1965 */ 1966 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1967 { 1968 refcount_inc(&skb->users); 1969 return skb; 1970 } 1971 1972 /* 1973 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1974 */ 1975 1976 /** 1977 * skb_cloned - is the buffer a clone 1978 * @skb: buffer to check 1979 * 1980 * Returns true if the buffer was generated with skb_clone() and is 1981 * one of multiple shared copies of the buffer. Cloned buffers are 1982 * shared data so must not be written to under normal circumstances. 1983 */ 1984 static inline int skb_cloned(const struct sk_buff *skb) 1985 { 1986 return skb->cloned && 1987 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1988 } 1989 1990 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1991 { 1992 might_sleep_if(gfpflags_allow_blocking(pri)); 1993 1994 if (skb_cloned(skb)) 1995 return pskb_expand_head(skb, 0, 0, pri); 1996 1997 return 0; 1998 } 1999 2000 /* This variant of skb_unclone() makes sure skb->truesize 2001 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 2002 * 2003 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 2004 * when various debugging features are in place. 2005 */ 2006 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 2007 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 2008 { 2009 might_sleep_if(gfpflags_allow_blocking(pri)); 2010 2011 if (skb_cloned(skb)) 2012 return __skb_unclone_keeptruesize(skb, pri); 2013 return 0; 2014 } 2015 2016 /** 2017 * skb_header_cloned - is the header a clone 2018 * @skb: buffer to check 2019 * 2020 * Returns true if modifying the header part of the buffer requires 2021 * the data to be copied. 2022 */ 2023 static inline int skb_header_cloned(const struct sk_buff *skb) 2024 { 2025 int dataref; 2026 2027 if (!skb->cloned) 2028 return 0; 2029 2030 dataref = atomic_read(&skb_shinfo(skb)->dataref); 2031 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 2032 return dataref != 1; 2033 } 2034 2035 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 2036 { 2037 might_sleep_if(gfpflags_allow_blocking(pri)); 2038 2039 if (skb_header_cloned(skb)) 2040 return pskb_expand_head(skb, 0, 0, pri); 2041 2042 return 0; 2043 } 2044 2045 /** 2046 * __skb_header_release() - allow clones to use the headroom 2047 * @skb: buffer to operate on 2048 * 2049 * See "DOC: dataref and headerless skbs". 2050 */ 2051 static inline void __skb_header_release(struct sk_buff *skb) 2052 { 2053 skb->nohdr = 1; 2054 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 2055 } 2056 2057 2058 /** 2059 * skb_shared - is the buffer shared 2060 * @skb: buffer to check 2061 * 2062 * Returns true if more than one person has a reference to this 2063 * buffer. 2064 */ 2065 static inline int skb_shared(const struct sk_buff *skb) 2066 { 2067 return refcount_read(&skb->users) != 1; 2068 } 2069 2070 /** 2071 * skb_share_check - check if buffer is shared and if so clone it 2072 * @skb: buffer to check 2073 * @pri: priority for memory allocation 2074 * 2075 * If the buffer is shared the buffer is cloned and the old copy 2076 * drops a reference. A new clone with a single reference is returned. 2077 * If the buffer is not shared the original buffer is returned. When 2078 * being called from interrupt status or with spinlocks held pri must 2079 * be GFP_ATOMIC. 2080 * 2081 * NULL is returned on a memory allocation failure. 2082 */ 2083 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 2084 { 2085 might_sleep_if(gfpflags_allow_blocking(pri)); 2086 if (skb_shared(skb)) { 2087 struct sk_buff *nskb = skb_clone(skb, pri); 2088 2089 if (likely(nskb)) 2090 consume_skb(skb); 2091 else 2092 kfree_skb(skb); 2093 skb = nskb; 2094 } 2095 return skb; 2096 } 2097 2098 /* 2099 * Copy shared buffers into a new sk_buff. We effectively do COW on 2100 * packets to handle cases where we have a local reader and forward 2101 * and a couple of other messy ones. The normal one is tcpdumping 2102 * a packet that's being forwarded. 2103 */ 2104 2105 /** 2106 * skb_unshare - make a copy of a shared buffer 2107 * @skb: buffer to check 2108 * @pri: priority for memory allocation 2109 * 2110 * If the socket buffer is a clone then this function creates a new 2111 * copy of the data, drops a reference count on the old copy and returns 2112 * the new copy with the reference count at 1. If the buffer is not a clone 2113 * the original buffer is returned. When called with a spinlock held or 2114 * from interrupt state @pri must be %GFP_ATOMIC 2115 * 2116 * %NULL is returned on a memory allocation failure. 2117 */ 2118 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2119 gfp_t pri) 2120 { 2121 might_sleep_if(gfpflags_allow_blocking(pri)); 2122 if (skb_cloned(skb)) { 2123 struct sk_buff *nskb = skb_copy(skb, pri); 2124 2125 /* Free our shared copy */ 2126 if (likely(nskb)) 2127 consume_skb(skb); 2128 else 2129 kfree_skb(skb); 2130 skb = nskb; 2131 } 2132 return skb; 2133 } 2134 2135 /** 2136 * skb_peek - peek at the head of an &sk_buff_head 2137 * @list_: list to peek at 2138 * 2139 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2140 * be careful with this one. A peek leaves the buffer on the 2141 * list and someone else may run off with it. You must hold 2142 * the appropriate locks or have a private queue to do this. 2143 * 2144 * Returns %NULL for an empty list or a pointer to the head element. 2145 * The reference count is not incremented and the reference is therefore 2146 * volatile. Use with caution. 2147 */ 2148 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2149 { 2150 struct sk_buff *skb = list_->next; 2151 2152 if (skb == (struct sk_buff *)list_) 2153 skb = NULL; 2154 return skb; 2155 } 2156 2157 /** 2158 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2159 * @list_: list to peek at 2160 * 2161 * Like skb_peek(), but the caller knows that the list is not empty. 2162 */ 2163 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2164 { 2165 return list_->next; 2166 } 2167 2168 /** 2169 * skb_peek_next - peek skb following the given one from a queue 2170 * @skb: skb to start from 2171 * @list_: list to peek at 2172 * 2173 * Returns %NULL when the end of the list is met or a pointer to the 2174 * next element. The reference count is not incremented and the 2175 * reference is therefore volatile. Use with caution. 2176 */ 2177 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2178 const struct sk_buff_head *list_) 2179 { 2180 struct sk_buff *next = skb->next; 2181 2182 if (next == (struct sk_buff *)list_) 2183 next = NULL; 2184 return next; 2185 } 2186 2187 /** 2188 * skb_peek_tail - peek at the tail of an &sk_buff_head 2189 * @list_: list to peek at 2190 * 2191 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2192 * be careful with this one. A peek leaves the buffer on the 2193 * list and someone else may run off with it. You must hold 2194 * the appropriate locks or have a private queue to do this. 2195 * 2196 * Returns %NULL for an empty list or a pointer to the tail element. 2197 * The reference count is not incremented and the reference is therefore 2198 * volatile. Use with caution. 2199 */ 2200 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2201 { 2202 struct sk_buff *skb = READ_ONCE(list_->prev); 2203 2204 if (skb == (struct sk_buff *)list_) 2205 skb = NULL; 2206 return skb; 2207 2208 } 2209 2210 /** 2211 * skb_queue_len - get queue length 2212 * @list_: list to measure 2213 * 2214 * Return the length of an &sk_buff queue. 2215 */ 2216 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2217 { 2218 return list_->qlen; 2219 } 2220 2221 /** 2222 * skb_queue_len_lockless - get queue length 2223 * @list_: list to measure 2224 * 2225 * Return the length of an &sk_buff queue. 2226 * This variant can be used in lockless contexts. 2227 */ 2228 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2229 { 2230 return READ_ONCE(list_->qlen); 2231 } 2232 2233 /** 2234 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2235 * @list: queue to initialize 2236 * 2237 * This initializes only the list and queue length aspects of 2238 * an sk_buff_head object. This allows to initialize the list 2239 * aspects of an sk_buff_head without reinitializing things like 2240 * the spinlock. It can also be used for on-stack sk_buff_head 2241 * objects where the spinlock is known to not be used. 2242 */ 2243 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2244 { 2245 list->prev = list->next = (struct sk_buff *)list; 2246 list->qlen = 0; 2247 } 2248 2249 /* 2250 * This function creates a split out lock class for each invocation; 2251 * this is needed for now since a whole lot of users of the skb-queue 2252 * infrastructure in drivers have different locking usage (in hardirq) 2253 * than the networking core (in softirq only). In the long run either the 2254 * network layer or drivers should need annotation to consolidate the 2255 * main types of usage into 3 classes. 2256 */ 2257 static inline void skb_queue_head_init(struct sk_buff_head *list) 2258 { 2259 spin_lock_init(&list->lock); 2260 __skb_queue_head_init(list); 2261 } 2262 2263 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2264 struct lock_class_key *class) 2265 { 2266 skb_queue_head_init(list); 2267 lockdep_set_class(&list->lock, class); 2268 } 2269 2270 /* 2271 * Insert an sk_buff on a list. 2272 * 2273 * The "__skb_xxxx()" functions are the non-atomic ones that 2274 * can only be called with interrupts disabled. 2275 */ 2276 static inline void __skb_insert(struct sk_buff *newsk, 2277 struct sk_buff *prev, struct sk_buff *next, 2278 struct sk_buff_head *list) 2279 { 2280 /* See skb_queue_empty_lockless() and skb_peek_tail() 2281 * for the opposite READ_ONCE() 2282 */ 2283 WRITE_ONCE(newsk->next, next); 2284 WRITE_ONCE(newsk->prev, prev); 2285 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2286 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2287 WRITE_ONCE(list->qlen, list->qlen + 1); 2288 } 2289 2290 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2291 struct sk_buff *prev, 2292 struct sk_buff *next) 2293 { 2294 struct sk_buff *first = list->next; 2295 struct sk_buff *last = list->prev; 2296 2297 WRITE_ONCE(first->prev, prev); 2298 WRITE_ONCE(prev->next, first); 2299 2300 WRITE_ONCE(last->next, next); 2301 WRITE_ONCE(next->prev, last); 2302 } 2303 2304 /** 2305 * skb_queue_splice - join two skb lists, this is designed for stacks 2306 * @list: the new list to add 2307 * @head: the place to add it in the first list 2308 */ 2309 static inline void skb_queue_splice(const struct sk_buff_head *list, 2310 struct sk_buff_head *head) 2311 { 2312 if (!skb_queue_empty(list)) { 2313 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2314 head->qlen += list->qlen; 2315 } 2316 } 2317 2318 /** 2319 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2320 * @list: the new list to add 2321 * @head: the place to add it in the first list 2322 * 2323 * The list at @list is reinitialised 2324 */ 2325 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2326 struct sk_buff_head *head) 2327 { 2328 if (!skb_queue_empty(list)) { 2329 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2330 head->qlen += list->qlen; 2331 __skb_queue_head_init(list); 2332 } 2333 } 2334 2335 /** 2336 * skb_queue_splice_tail - join two skb lists, each list being a queue 2337 * @list: the new list to add 2338 * @head: the place to add it in the first list 2339 */ 2340 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2341 struct sk_buff_head *head) 2342 { 2343 if (!skb_queue_empty(list)) { 2344 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2345 head->qlen += list->qlen; 2346 } 2347 } 2348 2349 /** 2350 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2351 * @list: the new list to add 2352 * @head: the place to add it in the first list 2353 * 2354 * Each of the lists is a queue. 2355 * The list at @list is reinitialised 2356 */ 2357 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2358 struct sk_buff_head *head) 2359 { 2360 if (!skb_queue_empty(list)) { 2361 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2362 head->qlen += list->qlen; 2363 __skb_queue_head_init(list); 2364 } 2365 } 2366 2367 /** 2368 * __skb_queue_after - queue a buffer at the list head 2369 * @list: list to use 2370 * @prev: place after this buffer 2371 * @newsk: buffer to queue 2372 * 2373 * Queue a buffer int the middle of a list. This function takes no locks 2374 * and you must therefore hold required locks before calling it. 2375 * 2376 * A buffer cannot be placed on two lists at the same time. 2377 */ 2378 static inline void __skb_queue_after(struct sk_buff_head *list, 2379 struct sk_buff *prev, 2380 struct sk_buff *newsk) 2381 { 2382 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2383 } 2384 2385 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2386 struct sk_buff_head *list); 2387 2388 static inline void __skb_queue_before(struct sk_buff_head *list, 2389 struct sk_buff *next, 2390 struct sk_buff *newsk) 2391 { 2392 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2393 } 2394 2395 /** 2396 * __skb_queue_head - queue a buffer at the list head 2397 * @list: list to use 2398 * @newsk: buffer to queue 2399 * 2400 * Queue a buffer at the start of a list. This function takes no locks 2401 * and you must therefore hold required locks before calling it. 2402 * 2403 * A buffer cannot be placed on two lists at the same time. 2404 */ 2405 static inline void __skb_queue_head(struct sk_buff_head *list, 2406 struct sk_buff *newsk) 2407 { 2408 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2409 } 2410 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2411 2412 /** 2413 * __skb_queue_tail - queue a buffer at the list tail 2414 * @list: list to use 2415 * @newsk: buffer to queue 2416 * 2417 * Queue a buffer at the end of a list. This function takes no locks 2418 * and you must therefore hold required locks before calling it. 2419 * 2420 * A buffer cannot be placed on two lists at the same time. 2421 */ 2422 static inline void __skb_queue_tail(struct sk_buff_head *list, 2423 struct sk_buff *newsk) 2424 { 2425 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2426 } 2427 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2428 2429 /* 2430 * remove sk_buff from list. _Must_ be called atomically, and with 2431 * the list known.. 2432 */ 2433 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2434 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2435 { 2436 struct sk_buff *next, *prev; 2437 2438 WRITE_ONCE(list->qlen, list->qlen - 1); 2439 next = skb->next; 2440 prev = skb->prev; 2441 skb->next = skb->prev = NULL; 2442 WRITE_ONCE(next->prev, prev); 2443 WRITE_ONCE(prev->next, next); 2444 } 2445 2446 /** 2447 * __skb_dequeue - remove from the head of the queue 2448 * @list: list to dequeue from 2449 * 2450 * Remove the head of the list. This function does not take any locks 2451 * so must be used with appropriate locks held only. The head item is 2452 * returned or %NULL if the list is empty. 2453 */ 2454 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2455 { 2456 struct sk_buff *skb = skb_peek(list); 2457 if (skb) 2458 __skb_unlink(skb, list); 2459 return skb; 2460 } 2461 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2462 2463 /** 2464 * __skb_dequeue_tail - remove from the tail of the queue 2465 * @list: list to dequeue from 2466 * 2467 * Remove the tail of the list. This function does not take any locks 2468 * so must be used with appropriate locks held only. The tail item is 2469 * returned or %NULL if the list is empty. 2470 */ 2471 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2472 { 2473 struct sk_buff *skb = skb_peek_tail(list); 2474 if (skb) 2475 __skb_unlink(skb, list); 2476 return skb; 2477 } 2478 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2479 2480 2481 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2482 { 2483 return skb->data_len; 2484 } 2485 2486 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2487 { 2488 return skb->len - skb->data_len; 2489 } 2490 2491 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2492 { 2493 unsigned int i, len = 0; 2494 2495 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2496 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2497 return len; 2498 } 2499 2500 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2501 { 2502 return skb_headlen(skb) + __skb_pagelen(skb); 2503 } 2504 2505 static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag, 2506 netmem_ref netmem, int off, 2507 int size) 2508 { 2509 frag->netmem = netmem; 2510 frag->offset = off; 2511 skb_frag_size_set(frag, size); 2512 } 2513 2514 static inline void skb_frag_fill_page_desc(skb_frag_t *frag, 2515 struct page *page, 2516 int off, int size) 2517 { 2518 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size); 2519 } 2520 2521 static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo, 2522 int i, netmem_ref netmem, 2523 int off, int size) 2524 { 2525 skb_frag_t *frag = &shinfo->frags[i]; 2526 2527 skb_frag_fill_netmem_desc(frag, netmem, off, size); 2528 } 2529 2530 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2531 int i, struct page *page, 2532 int off, int size) 2533 { 2534 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off, 2535 size); 2536 } 2537 2538 /** 2539 * skb_len_add - adds a number to len fields of skb 2540 * @skb: buffer to add len to 2541 * @delta: number of bytes to add 2542 */ 2543 static inline void skb_len_add(struct sk_buff *skb, int delta) 2544 { 2545 skb->len += delta; 2546 skb->data_len += delta; 2547 skb->truesize += delta; 2548 } 2549 2550 /** 2551 * __skb_fill_netmem_desc - initialise a fragment in an skb 2552 * @skb: buffer containing fragment to be initialised 2553 * @i: fragment index to initialise 2554 * @netmem: the netmem to use for this fragment 2555 * @off: the offset to the data with @page 2556 * @size: the length of the data 2557 * 2558 * Initialises the @i'th fragment of @skb to point to &size bytes at 2559 * offset @off within @page. 2560 * 2561 * Does not take any additional reference on the fragment. 2562 */ 2563 static inline void __skb_fill_netmem_desc(struct sk_buff *skb, int i, 2564 netmem_ref netmem, int off, int size) 2565 { 2566 struct page *page; 2567 2568 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size); 2569 2570 if (netmem_is_net_iov(netmem)) { 2571 skb->unreadable = true; 2572 return; 2573 } 2574 2575 page = netmem_to_page(netmem); 2576 2577 /* Propagate page pfmemalloc to the skb if we can. The problem is 2578 * that not all callers have unique ownership of the page but rely 2579 * on page_is_pfmemalloc doing the right thing(tm). 2580 */ 2581 page = compound_head(page); 2582 if (page_is_pfmemalloc(page)) 2583 skb->pfmemalloc = true; 2584 } 2585 2586 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2587 struct page *page, int off, int size) 2588 { 2589 __skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2590 } 2591 2592 static inline void skb_fill_netmem_desc(struct sk_buff *skb, int i, 2593 netmem_ref netmem, int off, int size) 2594 { 2595 __skb_fill_netmem_desc(skb, i, netmem, off, size); 2596 skb_shinfo(skb)->nr_frags = i + 1; 2597 } 2598 2599 /** 2600 * skb_fill_page_desc - initialise a paged fragment in an skb 2601 * @skb: buffer containing fragment to be initialised 2602 * @i: paged fragment index to initialise 2603 * @page: the page to use for this fragment 2604 * @off: the offset to the data with @page 2605 * @size: the length of the data 2606 * 2607 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2608 * @skb to point to @size bytes at offset @off within @page. In 2609 * addition updates @skb such that @i is the last fragment. 2610 * 2611 * Does not take any additional reference on the fragment. 2612 */ 2613 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2614 struct page *page, int off, int size) 2615 { 2616 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2617 } 2618 2619 /** 2620 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2621 * @skb: buffer containing fragment to be initialised 2622 * @i: paged fragment index to initialise 2623 * @page: the page to use for this fragment 2624 * @off: the offset to the data with @page 2625 * @size: the length of the data 2626 * 2627 * Variant of skb_fill_page_desc() which does not deal with 2628 * pfmemalloc, if page is not owned by us. 2629 */ 2630 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2631 struct page *page, int off, 2632 int size) 2633 { 2634 struct skb_shared_info *shinfo = skb_shinfo(skb); 2635 2636 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2637 shinfo->nr_frags = i + 1; 2638 } 2639 2640 void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem, 2641 int off, int size, unsigned int truesize); 2642 2643 static inline void skb_add_rx_frag(struct sk_buff *skb, int i, 2644 struct page *page, int off, int size, 2645 unsigned int truesize) 2646 { 2647 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size, 2648 truesize); 2649 } 2650 2651 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2652 unsigned int truesize); 2653 2654 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2655 2656 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2657 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2658 { 2659 return skb->head + skb->tail; 2660 } 2661 2662 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2663 { 2664 skb->tail = skb->data - skb->head; 2665 } 2666 2667 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2668 { 2669 skb_reset_tail_pointer(skb); 2670 skb->tail += offset; 2671 } 2672 2673 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2674 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2675 { 2676 return skb->tail; 2677 } 2678 2679 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2680 { 2681 skb->tail = skb->data; 2682 } 2683 2684 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2685 { 2686 skb->tail = skb->data + offset; 2687 } 2688 2689 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2690 2691 static inline void skb_assert_len(struct sk_buff *skb) 2692 { 2693 #ifdef CONFIG_DEBUG_NET 2694 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2695 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2696 #endif /* CONFIG_DEBUG_NET */ 2697 } 2698 2699 #if defined(CONFIG_FAIL_SKB_REALLOC) 2700 void skb_might_realloc(struct sk_buff *skb); 2701 #else 2702 static inline void skb_might_realloc(struct sk_buff *skb) {} 2703 #endif 2704 2705 /* 2706 * Add data to an sk_buff 2707 */ 2708 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2709 void *skb_put(struct sk_buff *skb, unsigned int len); 2710 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2711 { 2712 void *tmp = skb_tail_pointer(skb); 2713 SKB_LINEAR_ASSERT(skb); 2714 skb->tail += len; 2715 skb->len += len; 2716 return tmp; 2717 } 2718 2719 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2720 { 2721 void *tmp = __skb_put(skb, len); 2722 2723 memset(tmp, 0, len); 2724 return tmp; 2725 } 2726 2727 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2728 unsigned int len) 2729 { 2730 void *tmp = __skb_put(skb, len); 2731 2732 memcpy(tmp, data, len); 2733 return tmp; 2734 } 2735 2736 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2737 { 2738 *(u8 *)__skb_put(skb, 1) = val; 2739 } 2740 2741 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2742 { 2743 void *tmp = skb_put(skb, len); 2744 2745 memset(tmp, 0, len); 2746 2747 return tmp; 2748 } 2749 2750 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2751 unsigned int len) 2752 { 2753 void *tmp = skb_put(skb, len); 2754 2755 memcpy(tmp, data, len); 2756 2757 return tmp; 2758 } 2759 2760 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2761 { 2762 *(u8 *)skb_put(skb, 1) = val; 2763 } 2764 2765 void *skb_push(struct sk_buff *skb, unsigned int len); 2766 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2767 { 2768 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2769 2770 skb->data -= len; 2771 skb->len += len; 2772 return skb->data; 2773 } 2774 2775 void *skb_pull(struct sk_buff *skb, unsigned int len); 2776 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2777 { 2778 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2779 2780 skb->len -= len; 2781 if (unlikely(skb->len < skb->data_len)) { 2782 #if defined(CONFIG_DEBUG_NET) 2783 skb->len += len; 2784 pr_err("__skb_pull(len=%u)\n", len); 2785 skb_dump(KERN_ERR, skb, false); 2786 #endif 2787 BUG(); 2788 } 2789 return skb->data += len; 2790 } 2791 2792 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2793 { 2794 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2795 } 2796 2797 void *skb_pull_data(struct sk_buff *skb, size_t len); 2798 2799 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2800 2801 static inline enum skb_drop_reason 2802 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len) 2803 { 2804 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2805 skb_might_realloc(skb); 2806 2807 if (likely(len <= skb_headlen(skb))) 2808 return SKB_NOT_DROPPED_YET; 2809 2810 if (unlikely(len > skb->len)) 2811 return SKB_DROP_REASON_PKT_TOO_SMALL; 2812 2813 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb)))) 2814 return SKB_DROP_REASON_NOMEM; 2815 2816 return SKB_NOT_DROPPED_YET; 2817 } 2818 2819 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2820 { 2821 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 2822 } 2823 2824 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2825 { 2826 if (!pskb_may_pull(skb, len)) 2827 return NULL; 2828 2829 skb->len -= len; 2830 return skb->data += len; 2831 } 2832 2833 void skb_condense(struct sk_buff *skb); 2834 2835 /** 2836 * skb_headroom - bytes at buffer head 2837 * @skb: buffer to check 2838 * 2839 * Return the number of bytes of free space at the head of an &sk_buff. 2840 */ 2841 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2842 { 2843 return skb->data - skb->head; 2844 } 2845 2846 /** 2847 * skb_tailroom - bytes at buffer end 2848 * @skb: buffer to check 2849 * 2850 * Return the number of bytes of free space at the tail of an sk_buff 2851 */ 2852 static inline int skb_tailroom(const struct sk_buff *skb) 2853 { 2854 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2855 } 2856 2857 /** 2858 * skb_availroom - bytes at buffer end 2859 * @skb: buffer to check 2860 * 2861 * Return the number of bytes of free space at the tail of an sk_buff 2862 * allocated by sk_stream_alloc() 2863 */ 2864 static inline int skb_availroom(const struct sk_buff *skb) 2865 { 2866 if (skb_is_nonlinear(skb)) 2867 return 0; 2868 2869 return skb->end - skb->tail - skb->reserved_tailroom; 2870 } 2871 2872 /** 2873 * skb_reserve - adjust headroom 2874 * @skb: buffer to alter 2875 * @len: bytes to move 2876 * 2877 * Increase the headroom of an empty &sk_buff by reducing the tail 2878 * room. This is only allowed for an empty buffer. 2879 */ 2880 static inline void skb_reserve(struct sk_buff *skb, int len) 2881 { 2882 skb->data += len; 2883 skb->tail += len; 2884 } 2885 2886 /** 2887 * skb_tailroom_reserve - adjust reserved_tailroom 2888 * @skb: buffer to alter 2889 * @mtu: maximum amount of headlen permitted 2890 * @needed_tailroom: minimum amount of reserved_tailroom 2891 * 2892 * Set reserved_tailroom so that headlen can be as large as possible but 2893 * not larger than mtu and tailroom cannot be smaller than 2894 * needed_tailroom. 2895 * The required headroom should already have been reserved before using 2896 * this function. 2897 */ 2898 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2899 unsigned int needed_tailroom) 2900 { 2901 SKB_LINEAR_ASSERT(skb); 2902 if (mtu < skb_tailroom(skb) - needed_tailroom) 2903 /* use at most mtu */ 2904 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2905 else 2906 /* use up to all available space */ 2907 skb->reserved_tailroom = needed_tailroom; 2908 } 2909 2910 #define ENCAP_TYPE_ETHER 0 2911 #define ENCAP_TYPE_IPPROTO 1 2912 2913 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2914 __be16 protocol) 2915 { 2916 skb->inner_protocol = protocol; 2917 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2918 } 2919 2920 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2921 __u8 ipproto) 2922 { 2923 skb->inner_ipproto = ipproto; 2924 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2925 } 2926 2927 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2928 { 2929 skb->inner_mac_header = skb->mac_header; 2930 skb->inner_network_header = skb->network_header; 2931 skb->inner_transport_header = skb->transport_header; 2932 } 2933 2934 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2935 { 2936 return skb->mac_header != (typeof(skb->mac_header))~0U; 2937 } 2938 2939 static inline void skb_reset_mac_len(struct sk_buff *skb) 2940 { 2941 if (!skb_mac_header_was_set(skb)) { 2942 DEBUG_NET_WARN_ON_ONCE(1); 2943 skb->mac_len = 0; 2944 } else { 2945 skb->mac_len = skb->network_header - skb->mac_header; 2946 } 2947 } 2948 2949 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2950 *skb) 2951 { 2952 return skb->head + skb->inner_transport_header; 2953 } 2954 2955 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2956 { 2957 return skb_inner_transport_header(skb) - skb->data; 2958 } 2959 2960 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2961 { 2962 long offset = skb->data - skb->head; 2963 2964 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_transport_header))offset); 2965 skb->inner_transport_header = offset; 2966 } 2967 2968 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2969 const int offset) 2970 { 2971 skb_reset_inner_transport_header(skb); 2972 skb->inner_transport_header += offset; 2973 } 2974 2975 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2976 { 2977 return skb->head + skb->inner_network_header; 2978 } 2979 2980 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2981 { 2982 long offset = skb->data - skb->head; 2983 2984 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_network_header))offset); 2985 skb->inner_network_header = offset; 2986 } 2987 2988 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2989 const int offset) 2990 { 2991 skb_reset_inner_network_header(skb); 2992 skb->inner_network_header += offset; 2993 } 2994 2995 static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb) 2996 { 2997 return skb->inner_network_header > 0; 2998 } 2999 3000 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 3001 { 3002 return skb->head + skb->inner_mac_header; 3003 } 3004 3005 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 3006 { 3007 long offset = skb->data - skb->head; 3008 3009 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_mac_header))offset); 3010 skb->inner_mac_header = offset; 3011 } 3012 3013 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 3014 const int offset) 3015 { 3016 skb_reset_inner_mac_header(skb); 3017 skb->inner_mac_header += offset; 3018 } 3019 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 3020 { 3021 return skb->transport_header != (typeof(skb->transport_header))~0U; 3022 } 3023 3024 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 3025 { 3026 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3027 return skb->head + skb->transport_header; 3028 } 3029 3030 static inline void skb_reset_transport_header(struct sk_buff *skb) 3031 { 3032 long offset = skb->data - skb->head; 3033 3034 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->transport_header))offset); 3035 skb->transport_header = offset; 3036 } 3037 3038 static inline void skb_set_transport_header(struct sk_buff *skb, 3039 const int offset) 3040 { 3041 skb_reset_transport_header(skb); 3042 skb->transport_header += offset; 3043 } 3044 3045 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 3046 { 3047 return skb->head + skb->network_header; 3048 } 3049 3050 static inline void skb_reset_network_header(struct sk_buff *skb) 3051 { 3052 long offset = skb->data - skb->head; 3053 3054 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->network_header))offset); 3055 skb->network_header = offset; 3056 } 3057 3058 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 3059 { 3060 skb_reset_network_header(skb); 3061 skb->network_header += offset; 3062 } 3063 3064 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 3065 { 3066 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3067 return skb->head + skb->mac_header; 3068 } 3069 3070 static inline int skb_mac_offset(const struct sk_buff *skb) 3071 { 3072 return skb_mac_header(skb) - skb->data; 3073 } 3074 3075 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 3076 { 3077 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3078 return skb->network_header - skb->mac_header; 3079 } 3080 3081 static inline void skb_unset_mac_header(struct sk_buff *skb) 3082 { 3083 skb->mac_header = (typeof(skb->mac_header))~0U; 3084 } 3085 3086 static inline void skb_reset_mac_header(struct sk_buff *skb) 3087 { 3088 long offset = skb->data - skb->head; 3089 3090 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->mac_header))offset); 3091 skb->mac_header = offset; 3092 } 3093 3094 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 3095 { 3096 skb_reset_mac_header(skb); 3097 skb->mac_header += offset; 3098 } 3099 3100 static inline void skb_pop_mac_header(struct sk_buff *skb) 3101 { 3102 skb->mac_header = skb->network_header; 3103 } 3104 3105 static inline void skb_probe_transport_header(struct sk_buff *skb) 3106 { 3107 struct flow_keys_basic keys; 3108 3109 if (skb_transport_header_was_set(skb)) 3110 return; 3111 3112 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 3113 NULL, 0, 0, 0, 0)) 3114 skb_set_transport_header(skb, keys.control.thoff); 3115 } 3116 3117 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 3118 { 3119 if (skb_mac_header_was_set(skb)) { 3120 const unsigned char *old_mac = skb_mac_header(skb); 3121 3122 skb_set_mac_header(skb, -skb->mac_len); 3123 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 3124 } 3125 } 3126 3127 /* Move the full mac header up to current network_header. 3128 * Leaves skb->data pointing at offset skb->mac_len into the mac_header. 3129 * Must be provided the complete mac header length. 3130 */ 3131 static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len) 3132 { 3133 if (skb_mac_header_was_set(skb)) { 3134 const unsigned char *old_mac = skb_mac_header(skb); 3135 3136 skb_set_mac_header(skb, -full_mac_len); 3137 memmove(skb_mac_header(skb), old_mac, full_mac_len); 3138 __skb_push(skb, full_mac_len - skb->mac_len); 3139 } 3140 } 3141 3142 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 3143 { 3144 return skb->csum_start - skb_headroom(skb); 3145 } 3146 3147 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 3148 { 3149 return skb->head + skb->csum_start; 3150 } 3151 3152 static inline int skb_transport_offset(const struct sk_buff *skb) 3153 { 3154 return skb_transport_header(skb) - skb->data; 3155 } 3156 3157 static inline u32 skb_network_header_len(const struct sk_buff *skb) 3158 { 3159 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3160 return skb->transport_header - skb->network_header; 3161 } 3162 3163 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 3164 { 3165 return skb->inner_transport_header - skb->inner_network_header; 3166 } 3167 3168 static inline int skb_network_offset(const struct sk_buff *skb) 3169 { 3170 return skb_network_header(skb) - skb->data; 3171 } 3172 3173 static inline int skb_inner_network_offset(const struct sk_buff *skb) 3174 { 3175 return skb_inner_network_header(skb) - skb->data; 3176 } 3177 3178 static inline enum skb_drop_reason 3179 pskb_network_may_pull_reason(struct sk_buff *skb, unsigned int len) 3180 { 3181 return pskb_may_pull_reason(skb, skb_network_offset(skb) + len); 3182 } 3183 3184 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 3185 { 3186 return pskb_network_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 3187 } 3188 3189 /* 3190 * CPUs often take a performance hit when accessing unaligned memory 3191 * locations. The actual performance hit varies, it can be small if the 3192 * hardware handles it or large if we have to take an exception and fix it 3193 * in software. 3194 * 3195 * Since an ethernet header is 14 bytes network drivers often end up with 3196 * the IP header at an unaligned offset. The IP header can be aligned by 3197 * shifting the start of the packet by 2 bytes. Drivers should do this 3198 * with: 3199 * 3200 * skb_reserve(skb, NET_IP_ALIGN); 3201 * 3202 * The downside to this alignment of the IP header is that the DMA is now 3203 * unaligned. On some architectures the cost of an unaligned DMA is high 3204 * and this cost outweighs the gains made by aligning the IP header. 3205 * 3206 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 3207 * to be overridden. 3208 */ 3209 #ifndef NET_IP_ALIGN 3210 #define NET_IP_ALIGN 2 3211 #endif 3212 3213 /* 3214 * The networking layer reserves some headroom in skb data (via 3215 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 3216 * the header has to grow. In the default case, if the header has to grow 3217 * 32 bytes or less we avoid the reallocation. 3218 * 3219 * Unfortunately this headroom changes the DMA alignment of the resulting 3220 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 3221 * on some architectures. An architecture can override this value, 3222 * perhaps setting it to a cacheline in size (since that will maintain 3223 * cacheline alignment of the DMA). It must be a power of 2. 3224 * 3225 * Various parts of the networking layer expect at least 32 bytes of 3226 * headroom, you should not reduce this. 3227 * 3228 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 3229 * to reduce average number of cache lines per packet. 3230 * get_rps_cpu() for example only access one 64 bytes aligned block : 3231 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 3232 */ 3233 #ifndef NET_SKB_PAD 3234 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 3235 #endif 3236 3237 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 3238 3239 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3240 { 3241 if (WARN_ON(skb_is_nonlinear(skb))) 3242 return; 3243 skb->len = len; 3244 skb_set_tail_pointer(skb, len); 3245 } 3246 3247 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3248 { 3249 __skb_set_length(skb, len); 3250 } 3251 3252 void skb_trim(struct sk_buff *skb, unsigned int len); 3253 3254 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3255 { 3256 if (skb->data_len) 3257 return ___pskb_trim(skb, len); 3258 __skb_trim(skb, len); 3259 return 0; 3260 } 3261 3262 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3263 { 3264 skb_might_realloc(skb); 3265 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3266 } 3267 3268 /** 3269 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3270 * @skb: buffer to alter 3271 * @len: new length 3272 * 3273 * This is identical to pskb_trim except that the caller knows that 3274 * the skb is not cloned so we should never get an error due to out- 3275 * of-memory. 3276 */ 3277 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3278 { 3279 int err = pskb_trim(skb, len); 3280 BUG_ON(err); 3281 } 3282 3283 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3284 { 3285 unsigned int diff = len - skb->len; 3286 3287 if (skb_tailroom(skb) < diff) { 3288 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3289 GFP_ATOMIC); 3290 if (ret) 3291 return ret; 3292 } 3293 __skb_set_length(skb, len); 3294 return 0; 3295 } 3296 3297 /** 3298 * skb_orphan - orphan a buffer 3299 * @skb: buffer to orphan 3300 * 3301 * If a buffer currently has an owner then we call the owner's 3302 * destructor function and make the @skb unowned. The buffer continues 3303 * to exist but is no longer charged to its former owner. 3304 */ 3305 static inline void skb_orphan(struct sk_buff *skb) 3306 { 3307 if (skb->destructor) { 3308 skb->destructor(skb); 3309 skb->destructor = NULL; 3310 skb->sk = NULL; 3311 } else { 3312 BUG_ON(skb->sk); 3313 } 3314 } 3315 3316 /** 3317 * skb_orphan_frags - orphan the frags contained in a buffer 3318 * @skb: buffer to orphan frags from 3319 * @gfp_mask: allocation mask for replacement pages 3320 * 3321 * For each frag in the SKB which needs a destructor (i.e. has an 3322 * owner) create a copy of that frag and release the original 3323 * page by calling the destructor. 3324 */ 3325 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3326 { 3327 if (likely(!skb_zcopy(skb))) 3328 return 0; 3329 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3330 return 0; 3331 return skb_copy_ubufs(skb, gfp_mask); 3332 } 3333 3334 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3335 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3336 { 3337 if (likely(!skb_zcopy(skb))) 3338 return 0; 3339 return skb_copy_ubufs(skb, gfp_mask); 3340 } 3341 3342 /** 3343 * __skb_queue_purge_reason - empty a list 3344 * @list: list to empty 3345 * @reason: drop reason 3346 * 3347 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3348 * the list and one reference dropped. This function does not take the 3349 * list lock and the caller must hold the relevant locks to use it. 3350 */ 3351 static inline void __skb_queue_purge_reason(struct sk_buff_head *list, 3352 enum skb_drop_reason reason) 3353 { 3354 struct sk_buff *skb; 3355 3356 while ((skb = __skb_dequeue(list)) != NULL) 3357 kfree_skb_reason(skb, reason); 3358 } 3359 3360 static inline void __skb_queue_purge(struct sk_buff_head *list) 3361 { 3362 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3363 } 3364 3365 void skb_queue_purge_reason(struct sk_buff_head *list, 3366 enum skb_drop_reason reason); 3367 3368 static inline void skb_queue_purge(struct sk_buff_head *list) 3369 { 3370 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3371 } 3372 3373 unsigned int skb_rbtree_purge(struct rb_root *root); 3374 void skb_errqueue_purge(struct sk_buff_head *list); 3375 3376 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3377 3378 /** 3379 * netdev_alloc_frag - allocate a page fragment 3380 * @fragsz: fragment size 3381 * 3382 * Allocates a frag from a page for receive buffer. 3383 * Uses GFP_ATOMIC allocations. 3384 */ 3385 static inline void *netdev_alloc_frag(unsigned int fragsz) 3386 { 3387 return __netdev_alloc_frag_align(fragsz, ~0u); 3388 } 3389 3390 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3391 unsigned int align) 3392 { 3393 WARN_ON_ONCE(!is_power_of_2(align)); 3394 return __netdev_alloc_frag_align(fragsz, -align); 3395 } 3396 3397 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3398 gfp_t gfp_mask); 3399 3400 /** 3401 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3402 * @dev: network device to receive on 3403 * @length: length to allocate 3404 * 3405 * Allocate a new &sk_buff and assign it a usage count of one. The 3406 * buffer has unspecified headroom built in. Users should allocate 3407 * the headroom they think they need without accounting for the 3408 * built in space. The built in space is used for optimisations. 3409 * 3410 * %NULL is returned if there is no free memory. Although this function 3411 * allocates memory it can be called from an interrupt. 3412 */ 3413 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3414 unsigned int length) 3415 { 3416 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3417 } 3418 3419 /* legacy helper around __netdev_alloc_skb() */ 3420 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3421 gfp_t gfp_mask) 3422 { 3423 return __netdev_alloc_skb(NULL, length, gfp_mask); 3424 } 3425 3426 /* legacy helper around netdev_alloc_skb() */ 3427 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3428 { 3429 return netdev_alloc_skb(NULL, length); 3430 } 3431 3432 3433 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3434 unsigned int length, gfp_t gfp) 3435 { 3436 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3437 3438 if (NET_IP_ALIGN && skb) 3439 skb_reserve(skb, NET_IP_ALIGN); 3440 return skb; 3441 } 3442 3443 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3444 unsigned int length) 3445 { 3446 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3447 } 3448 3449 static inline void skb_free_frag(void *addr) 3450 { 3451 page_frag_free(addr); 3452 } 3453 3454 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3455 3456 static inline void *napi_alloc_frag(unsigned int fragsz) 3457 { 3458 return __napi_alloc_frag_align(fragsz, ~0u); 3459 } 3460 3461 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3462 unsigned int align) 3463 { 3464 WARN_ON_ONCE(!is_power_of_2(align)); 3465 return __napi_alloc_frag_align(fragsz, -align); 3466 } 3467 3468 struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int length); 3469 void napi_consume_skb(struct sk_buff *skb, int budget); 3470 3471 void napi_skb_free_stolen_head(struct sk_buff *skb); 3472 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason); 3473 3474 /** 3475 * __dev_alloc_pages - allocate page for network Rx 3476 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3477 * @order: size of the allocation 3478 * 3479 * Allocate a new page. 3480 * 3481 * %NULL is returned if there is no free memory. 3482 */ 3483 static inline struct page *__dev_alloc_pages_noprof(gfp_t gfp_mask, 3484 unsigned int order) 3485 { 3486 /* This piece of code contains several assumptions. 3487 * 1. This is for device Rx, therefore a cold page is preferred. 3488 * 2. The expectation is the user wants a compound page. 3489 * 3. If requesting a order 0 page it will not be compound 3490 * due to the check to see if order has a value in prep_new_page 3491 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3492 * code in gfp_to_alloc_flags that should be enforcing this. 3493 */ 3494 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3495 3496 return alloc_pages_node_noprof(NUMA_NO_NODE, gfp_mask, order); 3497 } 3498 #define __dev_alloc_pages(...) alloc_hooks(__dev_alloc_pages_noprof(__VA_ARGS__)) 3499 3500 /* 3501 * This specialized allocator has to be a macro for its allocations to be 3502 * accounted separately (to have a separate alloc_tag). 3503 */ 3504 #define dev_alloc_pages(_order) __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, _order) 3505 3506 /** 3507 * __dev_alloc_page - allocate a page for network Rx 3508 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3509 * 3510 * Allocate a new page. 3511 * 3512 * %NULL is returned if there is no free memory. 3513 */ 3514 static inline struct page *__dev_alloc_page_noprof(gfp_t gfp_mask) 3515 { 3516 return __dev_alloc_pages_noprof(gfp_mask, 0); 3517 } 3518 #define __dev_alloc_page(...) alloc_hooks(__dev_alloc_page_noprof(__VA_ARGS__)) 3519 3520 /* 3521 * This specialized allocator has to be a macro for its allocations to be 3522 * accounted separately (to have a separate alloc_tag). 3523 */ 3524 #define dev_alloc_page() dev_alloc_pages(0) 3525 3526 /** 3527 * dev_page_is_reusable - check whether a page can be reused for network Rx 3528 * @page: the page to test 3529 * 3530 * A page shouldn't be considered for reusing/recycling if it was allocated 3531 * under memory pressure or at a distant memory node. 3532 * 3533 * Returns: false if this page should be returned to page allocator, true 3534 * otherwise. 3535 */ 3536 static inline bool dev_page_is_reusable(const struct page *page) 3537 { 3538 return likely(page_to_nid(page) == numa_mem_id() && 3539 !page_is_pfmemalloc(page)); 3540 } 3541 3542 /** 3543 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3544 * @page: The page that was allocated from skb_alloc_page 3545 * @skb: The skb that may need pfmemalloc set 3546 */ 3547 static inline void skb_propagate_pfmemalloc(const struct page *page, 3548 struct sk_buff *skb) 3549 { 3550 if (page_is_pfmemalloc(page)) 3551 skb->pfmemalloc = true; 3552 } 3553 3554 /** 3555 * skb_frag_off() - Returns the offset of a skb fragment 3556 * @frag: the paged fragment 3557 */ 3558 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3559 { 3560 return frag->offset; 3561 } 3562 3563 /** 3564 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3565 * @frag: skb fragment 3566 * @delta: value to add 3567 */ 3568 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3569 { 3570 frag->offset += delta; 3571 } 3572 3573 /** 3574 * skb_frag_off_set() - Sets the offset of a skb fragment 3575 * @frag: skb fragment 3576 * @offset: offset of fragment 3577 */ 3578 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3579 { 3580 frag->offset = offset; 3581 } 3582 3583 /** 3584 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3585 * @fragto: skb fragment where offset is set 3586 * @fragfrom: skb fragment offset is copied from 3587 */ 3588 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3589 const skb_frag_t *fragfrom) 3590 { 3591 fragto->offset = fragfrom->offset; 3592 } 3593 3594 /* Return: true if the skb_frag contains a net_iov. */ 3595 static inline bool skb_frag_is_net_iov(const skb_frag_t *frag) 3596 { 3597 return netmem_is_net_iov(frag->netmem); 3598 } 3599 3600 /** 3601 * skb_frag_net_iov - retrieve the net_iov referred to by fragment 3602 * @frag: the fragment 3603 * 3604 * Return: the &struct net_iov associated with @frag. Returns NULL if this 3605 * frag has no associated net_iov. 3606 */ 3607 static inline struct net_iov *skb_frag_net_iov(const skb_frag_t *frag) 3608 { 3609 if (!skb_frag_is_net_iov(frag)) 3610 return NULL; 3611 3612 return netmem_to_net_iov(frag->netmem); 3613 } 3614 3615 /** 3616 * skb_frag_page - retrieve the page referred to by a paged fragment 3617 * @frag: the paged fragment 3618 * 3619 * Return: the &struct page associated with @frag. Returns NULL if this frag 3620 * has no associated page. 3621 */ 3622 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3623 { 3624 if (skb_frag_is_net_iov(frag)) 3625 return NULL; 3626 3627 return netmem_to_page(frag->netmem); 3628 } 3629 3630 /** 3631 * skb_frag_netmem - retrieve the netmem referred to by a fragment 3632 * @frag: the fragment 3633 * 3634 * Return: the &netmem_ref associated with @frag. 3635 */ 3636 static inline netmem_ref skb_frag_netmem(const skb_frag_t *frag) 3637 { 3638 return frag->netmem; 3639 } 3640 3641 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb, 3642 unsigned int headroom); 3643 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb, 3644 const struct bpf_prog *prog); 3645 3646 /** 3647 * skb_frag_address - gets the address of the data contained in a paged fragment 3648 * @frag: the paged fragment buffer 3649 * 3650 * Returns: the address of the data within @frag. The page must already 3651 * be mapped. 3652 */ 3653 static inline void *skb_frag_address(const skb_frag_t *frag) 3654 { 3655 if (!skb_frag_page(frag)) 3656 return NULL; 3657 3658 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3659 } 3660 3661 /** 3662 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3663 * @frag: the paged fragment buffer 3664 * 3665 * Returns: the address of the data within @frag. Checks that the page 3666 * is mapped and returns %NULL otherwise. 3667 */ 3668 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3669 { 3670 void *ptr = page_address(skb_frag_page(frag)); 3671 if (unlikely(!ptr)) 3672 return NULL; 3673 3674 return ptr + skb_frag_off(frag); 3675 } 3676 3677 /** 3678 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3679 * @fragto: skb fragment where page is set 3680 * @fragfrom: skb fragment page is copied from 3681 */ 3682 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3683 const skb_frag_t *fragfrom) 3684 { 3685 fragto->netmem = fragfrom->netmem; 3686 } 3687 3688 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3689 3690 /** 3691 * __skb_frag_dma_map - maps a paged fragment via the DMA API 3692 * @dev: the device to map the fragment to 3693 * @frag: the paged fragment to map 3694 * @offset: the offset within the fragment (starting at the 3695 * fragment's own offset) 3696 * @size: the number of bytes to map 3697 * @dir: the direction of the mapping (``PCI_DMA_*``) 3698 * 3699 * Maps the page associated with @frag to @device. 3700 */ 3701 static inline dma_addr_t __skb_frag_dma_map(struct device *dev, 3702 const skb_frag_t *frag, 3703 size_t offset, size_t size, 3704 enum dma_data_direction dir) 3705 { 3706 return dma_map_page(dev, skb_frag_page(frag), 3707 skb_frag_off(frag) + offset, size, dir); 3708 } 3709 3710 #define skb_frag_dma_map(dev, frag, ...) \ 3711 CONCATENATE(_skb_frag_dma_map, \ 3712 COUNT_ARGS(__VA_ARGS__))(dev, frag, ##__VA_ARGS__) 3713 3714 #define __skb_frag_dma_map1(dev, frag, offset, uf, uo) ({ \ 3715 const skb_frag_t *uf = (frag); \ 3716 size_t uo = (offset); \ 3717 \ 3718 __skb_frag_dma_map(dev, uf, uo, skb_frag_size(uf) - uo, \ 3719 DMA_TO_DEVICE); \ 3720 }) 3721 #define _skb_frag_dma_map1(dev, frag, offset) \ 3722 __skb_frag_dma_map1(dev, frag, offset, __UNIQUE_ID(frag_), \ 3723 __UNIQUE_ID(offset_)) 3724 #define _skb_frag_dma_map0(dev, frag) \ 3725 _skb_frag_dma_map1(dev, frag, 0) 3726 #define _skb_frag_dma_map2(dev, frag, offset, size) \ 3727 __skb_frag_dma_map(dev, frag, offset, size, DMA_TO_DEVICE) 3728 #define _skb_frag_dma_map3(dev, frag, offset, size, dir) \ 3729 __skb_frag_dma_map(dev, frag, offset, size, dir) 3730 3731 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3732 gfp_t gfp_mask) 3733 { 3734 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3735 } 3736 3737 3738 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3739 gfp_t gfp_mask) 3740 { 3741 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3742 } 3743 3744 3745 /** 3746 * skb_clone_writable - is the header of a clone writable 3747 * @skb: buffer to check 3748 * @len: length up to which to write 3749 * 3750 * Returns true if modifying the header part of the cloned buffer 3751 * does not requires the data to be copied. 3752 */ 3753 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3754 { 3755 return !skb_header_cloned(skb) && 3756 skb_headroom(skb) + len <= skb->hdr_len; 3757 } 3758 3759 static inline int skb_try_make_writable(struct sk_buff *skb, 3760 unsigned int write_len) 3761 { 3762 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3763 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3764 } 3765 3766 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3767 int cloned) 3768 { 3769 int delta = 0; 3770 3771 if (headroom > skb_headroom(skb)) 3772 delta = headroom - skb_headroom(skb); 3773 3774 if (delta || cloned) 3775 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3776 GFP_ATOMIC); 3777 return 0; 3778 } 3779 3780 /** 3781 * skb_cow - copy header of skb when it is required 3782 * @skb: buffer to cow 3783 * @headroom: needed headroom 3784 * 3785 * If the skb passed lacks sufficient headroom or its data part 3786 * is shared, data is reallocated. If reallocation fails, an error 3787 * is returned and original skb is not changed. 3788 * 3789 * The result is skb with writable area skb->head...skb->tail 3790 * and at least @headroom of space at head. 3791 */ 3792 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3793 { 3794 return __skb_cow(skb, headroom, skb_cloned(skb)); 3795 } 3796 3797 /** 3798 * skb_cow_head - skb_cow but only making the head writable 3799 * @skb: buffer to cow 3800 * @headroom: needed headroom 3801 * 3802 * This function is identical to skb_cow except that we replace the 3803 * skb_cloned check by skb_header_cloned. It should be used when 3804 * you only need to push on some header and do not need to modify 3805 * the data. 3806 */ 3807 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3808 { 3809 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3810 } 3811 3812 /** 3813 * skb_padto - pad an skbuff up to a minimal size 3814 * @skb: buffer to pad 3815 * @len: minimal length 3816 * 3817 * Pads up a buffer to ensure the trailing bytes exist and are 3818 * blanked. If the buffer already contains sufficient data it 3819 * is untouched. Otherwise it is extended. Returns zero on 3820 * success. The skb is freed on error. 3821 */ 3822 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3823 { 3824 unsigned int size = skb->len; 3825 if (likely(size >= len)) 3826 return 0; 3827 return skb_pad(skb, len - size); 3828 } 3829 3830 /** 3831 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3832 * @skb: buffer to pad 3833 * @len: minimal length 3834 * @free_on_error: free buffer on error 3835 * 3836 * Pads up a buffer to ensure the trailing bytes exist and are 3837 * blanked. If the buffer already contains sufficient data it 3838 * is untouched. Otherwise it is extended. Returns zero on 3839 * success. The skb is freed on error if @free_on_error is true. 3840 */ 3841 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3842 unsigned int len, 3843 bool free_on_error) 3844 { 3845 unsigned int size = skb->len; 3846 3847 if (unlikely(size < len)) { 3848 len -= size; 3849 if (__skb_pad(skb, len, free_on_error)) 3850 return -ENOMEM; 3851 __skb_put(skb, len); 3852 } 3853 return 0; 3854 } 3855 3856 /** 3857 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3858 * @skb: buffer to pad 3859 * @len: minimal length 3860 * 3861 * Pads up a buffer to ensure the trailing bytes exist and are 3862 * blanked. If the buffer already contains sufficient data it 3863 * is untouched. Otherwise it is extended. Returns zero on 3864 * success. The skb is freed on error. 3865 */ 3866 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3867 { 3868 return __skb_put_padto(skb, len, true); 3869 } 3870 3871 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i) 3872 __must_check; 3873 3874 static inline int skb_add_data(struct sk_buff *skb, 3875 struct iov_iter *from, int copy) 3876 { 3877 const int off = skb->len; 3878 3879 if (skb->ip_summed == CHECKSUM_NONE) { 3880 __wsum csum = 0; 3881 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 3882 &csum, from)) { 3883 skb->csum = csum_block_add(skb->csum, csum, off); 3884 return 0; 3885 } 3886 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3887 return 0; 3888 3889 __skb_trim(skb, off); 3890 return -EFAULT; 3891 } 3892 3893 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3894 const struct page *page, int off) 3895 { 3896 if (skb_zcopy(skb)) 3897 return false; 3898 if (i) { 3899 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3900 3901 return page == skb_frag_page(frag) && 3902 off == skb_frag_off(frag) + skb_frag_size(frag); 3903 } 3904 return false; 3905 } 3906 3907 static inline int __skb_linearize(struct sk_buff *skb) 3908 { 3909 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3910 } 3911 3912 /** 3913 * skb_linearize - convert paged skb to linear one 3914 * @skb: buffer to linarize 3915 * 3916 * If there is no free memory -ENOMEM is returned, otherwise zero 3917 * is returned and the old skb data released. 3918 */ 3919 static inline int skb_linearize(struct sk_buff *skb) 3920 { 3921 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3922 } 3923 3924 /** 3925 * skb_has_shared_frag - can any frag be overwritten 3926 * @skb: buffer to test 3927 * 3928 * Return: true if the skb has at least one frag that might be modified 3929 * by an external entity (as in vmsplice()/sendfile()) 3930 */ 3931 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3932 { 3933 return skb_is_nonlinear(skb) && 3934 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 3935 } 3936 3937 /** 3938 * skb_linearize_cow - make sure skb is linear and writable 3939 * @skb: buffer to process 3940 * 3941 * If there is no free memory -ENOMEM is returned, otherwise zero 3942 * is returned and the old skb data released. 3943 */ 3944 static inline int skb_linearize_cow(struct sk_buff *skb) 3945 { 3946 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3947 __skb_linearize(skb) : 0; 3948 } 3949 3950 static __always_inline void 3951 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3952 unsigned int off) 3953 { 3954 if (skb->ip_summed == CHECKSUM_COMPLETE) 3955 skb->csum = csum_block_sub(skb->csum, 3956 csum_partial(start, len, 0), off); 3957 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3958 skb_checksum_start_offset(skb) < 0) 3959 skb->ip_summed = CHECKSUM_NONE; 3960 } 3961 3962 /** 3963 * skb_postpull_rcsum - update checksum for received skb after pull 3964 * @skb: buffer to update 3965 * @start: start of data before pull 3966 * @len: length of data pulled 3967 * 3968 * After doing a pull on a received packet, you need to call this to 3969 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3970 * CHECKSUM_NONE so that it can be recomputed from scratch. 3971 */ 3972 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3973 const void *start, unsigned int len) 3974 { 3975 if (skb->ip_summed == CHECKSUM_COMPLETE) 3976 skb->csum = wsum_negate(csum_partial(start, len, 3977 wsum_negate(skb->csum))); 3978 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3979 skb_checksum_start_offset(skb) < 0) 3980 skb->ip_summed = CHECKSUM_NONE; 3981 } 3982 3983 static __always_inline void 3984 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3985 unsigned int off) 3986 { 3987 if (skb->ip_summed == CHECKSUM_COMPLETE) 3988 skb->csum = csum_block_add(skb->csum, 3989 csum_partial(start, len, 0), off); 3990 } 3991 3992 /** 3993 * skb_postpush_rcsum - update checksum for received skb after push 3994 * @skb: buffer to update 3995 * @start: start of data after push 3996 * @len: length of data pushed 3997 * 3998 * After doing a push on a received packet, you need to call this to 3999 * update the CHECKSUM_COMPLETE checksum. 4000 */ 4001 static inline void skb_postpush_rcsum(struct sk_buff *skb, 4002 const void *start, unsigned int len) 4003 { 4004 __skb_postpush_rcsum(skb, start, len, 0); 4005 } 4006 4007 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 4008 4009 /** 4010 * skb_push_rcsum - push skb and update receive checksum 4011 * @skb: buffer to update 4012 * @len: length of data pulled 4013 * 4014 * This function performs an skb_push on the packet and updates 4015 * the CHECKSUM_COMPLETE checksum. It should be used on 4016 * receive path processing instead of skb_push unless you know 4017 * that the checksum difference is zero (e.g., a valid IP header) 4018 * or you are setting ip_summed to CHECKSUM_NONE. 4019 */ 4020 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 4021 { 4022 skb_push(skb, len); 4023 skb_postpush_rcsum(skb, skb->data, len); 4024 return skb->data; 4025 } 4026 4027 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 4028 /** 4029 * pskb_trim_rcsum - trim received skb and update checksum 4030 * @skb: buffer to trim 4031 * @len: new length 4032 * 4033 * This is exactly the same as pskb_trim except that it ensures the 4034 * checksum of received packets are still valid after the operation. 4035 * It can change skb pointers. 4036 */ 4037 4038 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4039 { 4040 skb_might_realloc(skb); 4041 if (likely(len >= skb->len)) 4042 return 0; 4043 return pskb_trim_rcsum_slow(skb, len); 4044 } 4045 4046 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4047 { 4048 if (skb->ip_summed == CHECKSUM_COMPLETE) 4049 skb->ip_summed = CHECKSUM_NONE; 4050 __skb_trim(skb, len); 4051 return 0; 4052 } 4053 4054 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 4055 { 4056 if (skb->ip_summed == CHECKSUM_COMPLETE) 4057 skb->ip_summed = CHECKSUM_NONE; 4058 return __skb_grow(skb, len); 4059 } 4060 4061 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 4062 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 4063 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 4064 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 4065 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 4066 4067 #define skb_queue_walk(queue, skb) \ 4068 for (skb = (queue)->next; \ 4069 skb != (struct sk_buff *)(queue); \ 4070 skb = skb->next) 4071 4072 #define skb_queue_walk_safe(queue, skb, tmp) \ 4073 for (skb = (queue)->next, tmp = skb->next; \ 4074 skb != (struct sk_buff *)(queue); \ 4075 skb = tmp, tmp = skb->next) 4076 4077 #define skb_queue_walk_from(queue, skb) \ 4078 for (; skb != (struct sk_buff *)(queue); \ 4079 skb = skb->next) 4080 4081 #define skb_rbtree_walk(skb, root) \ 4082 for (skb = skb_rb_first(root); skb != NULL; \ 4083 skb = skb_rb_next(skb)) 4084 4085 #define skb_rbtree_walk_from(skb) \ 4086 for (; skb != NULL; \ 4087 skb = skb_rb_next(skb)) 4088 4089 #define skb_rbtree_walk_from_safe(skb, tmp) \ 4090 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 4091 skb = tmp) 4092 4093 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 4094 for (tmp = skb->next; \ 4095 skb != (struct sk_buff *)(queue); \ 4096 skb = tmp, tmp = skb->next) 4097 4098 #define skb_queue_reverse_walk(queue, skb) \ 4099 for (skb = (queue)->prev; \ 4100 skb != (struct sk_buff *)(queue); \ 4101 skb = skb->prev) 4102 4103 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 4104 for (skb = (queue)->prev, tmp = skb->prev; \ 4105 skb != (struct sk_buff *)(queue); \ 4106 skb = tmp, tmp = skb->prev) 4107 4108 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 4109 for (tmp = skb->prev; \ 4110 skb != (struct sk_buff *)(queue); \ 4111 skb = tmp, tmp = skb->prev) 4112 4113 static inline bool skb_has_frag_list(const struct sk_buff *skb) 4114 { 4115 return skb_shinfo(skb)->frag_list != NULL; 4116 } 4117 4118 static inline void skb_frag_list_init(struct sk_buff *skb) 4119 { 4120 skb_shinfo(skb)->frag_list = NULL; 4121 } 4122 4123 #define skb_walk_frags(skb, iter) \ 4124 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 4125 4126 4127 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 4128 int *err, long *timeo_p, 4129 const struct sk_buff *skb); 4130 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 4131 struct sk_buff_head *queue, 4132 unsigned int flags, 4133 int *off, int *err, 4134 struct sk_buff **last); 4135 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 4136 struct sk_buff_head *queue, 4137 unsigned int flags, int *off, int *err, 4138 struct sk_buff **last); 4139 struct sk_buff *__skb_recv_datagram(struct sock *sk, 4140 struct sk_buff_head *sk_queue, 4141 unsigned int flags, int *off, int *err); 4142 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 4143 __poll_t datagram_poll(struct file *file, struct socket *sock, 4144 struct poll_table_struct *wait); 4145 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 4146 struct iov_iter *to, int size); 4147 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 4148 struct msghdr *msg, int size) 4149 { 4150 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 4151 } 4152 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 4153 struct msghdr *msg); 4154 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 4155 struct iov_iter *to, int len, 4156 struct ahash_request *hash); 4157 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 4158 struct iov_iter *from, int len); 4159 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 4160 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 4161 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 4162 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 4163 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 4164 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 4165 int len); 4166 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 4167 struct pipe_inode_info *pipe, unsigned int len, 4168 unsigned int flags); 4169 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 4170 int len); 4171 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 4172 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 4173 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 4174 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 4175 int len, int hlen); 4176 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 4177 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 4178 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 4179 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 4180 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 4181 unsigned int offset); 4182 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 4183 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 4184 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev); 4185 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 4186 int skb_vlan_pop(struct sk_buff *skb); 4187 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 4188 int skb_eth_pop(struct sk_buff *skb); 4189 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 4190 const unsigned char *src); 4191 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 4192 int mac_len, bool ethernet); 4193 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 4194 bool ethernet); 4195 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 4196 int skb_mpls_dec_ttl(struct sk_buff *skb); 4197 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 4198 gfp_t gfp); 4199 4200 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 4201 { 4202 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 4203 } 4204 4205 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 4206 { 4207 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 4208 } 4209 4210 struct skb_checksum_ops { 4211 __wsum (*update)(const void *mem, int len, __wsum wsum); 4212 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 4213 }; 4214 4215 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 4216 4217 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 4218 __wsum csum, const struct skb_checksum_ops *ops); 4219 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 4220 __wsum csum); 4221 4222 static inline void * __must_check 4223 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 4224 const void *data, int hlen, void *buffer) 4225 { 4226 if (likely(hlen - offset >= len)) 4227 return (void *)data + offset; 4228 4229 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 4230 return NULL; 4231 4232 return buffer; 4233 } 4234 4235 static inline void * __must_check 4236 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4237 { 4238 return __skb_header_pointer(skb, offset, len, skb->data, 4239 skb_headlen(skb), buffer); 4240 } 4241 4242 static inline void * __must_check 4243 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len) 4244 { 4245 if (likely(skb_headlen(skb) - offset >= len)) 4246 return skb->data + offset; 4247 return NULL; 4248 } 4249 4250 /** 4251 * skb_needs_linearize - check if we need to linearize a given skb 4252 * depending on the given device features. 4253 * @skb: socket buffer to check 4254 * @features: net device features 4255 * 4256 * Returns true if either: 4257 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4258 * 2. skb is fragmented and the device does not support SG. 4259 */ 4260 static inline bool skb_needs_linearize(struct sk_buff *skb, 4261 netdev_features_t features) 4262 { 4263 return skb_is_nonlinear(skb) && 4264 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4265 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4266 } 4267 4268 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4269 void *to, 4270 const unsigned int len) 4271 { 4272 memcpy(to, skb->data, len); 4273 } 4274 4275 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4276 const int offset, void *to, 4277 const unsigned int len) 4278 { 4279 memcpy(to, skb->data + offset, len); 4280 } 4281 4282 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4283 const void *from, 4284 const unsigned int len) 4285 { 4286 memcpy(skb->data, from, len); 4287 } 4288 4289 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4290 const int offset, 4291 const void *from, 4292 const unsigned int len) 4293 { 4294 memcpy(skb->data + offset, from, len); 4295 } 4296 4297 void skb_init(void); 4298 4299 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4300 { 4301 return skb->tstamp; 4302 } 4303 4304 /** 4305 * skb_get_timestamp - get timestamp from a skb 4306 * @skb: skb to get stamp from 4307 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4308 * 4309 * Timestamps are stored in the skb as offsets to a base timestamp. 4310 * This function converts the offset back to a struct timeval and stores 4311 * it in stamp. 4312 */ 4313 static inline void skb_get_timestamp(const struct sk_buff *skb, 4314 struct __kernel_old_timeval *stamp) 4315 { 4316 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4317 } 4318 4319 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4320 struct __kernel_sock_timeval *stamp) 4321 { 4322 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4323 4324 stamp->tv_sec = ts.tv_sec; 4325 stamp->tv_usec = ts.tv_nsec / 1000; 4326 } 4327 4328 static inline void skb_get_timestampns(const struct sk_buff *skb, 4329 struct __kernel_old_timespec *stamp) 4330 { 4331 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4332 4333 stamp->tv_sec = ts.tv_sec; 4334 stamp->tv_nsec = ts.tv_nsec; 4335 } 4336 4337 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4338 struct __kernel_timespec *stamp) 4339 { 4340 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4341 4342 stamp->tv_sec = ts.tv_sec; 4343 stamp->tv_nsec = ts.tv_nsec; 4344 } 4345 4346 static inline void __net_timestamp(struct sk_buff *skb) 4347 { 4348 skb->tstamp = ktime_get_real(); 4349 skb->tstamp_type = SKB_CLOCK_REALTIME; 4350 } 4351 4352 static inline ktime_t net_timedelta(ktime_t t) 4353 { 4354 return ktime_sub(ktime_get_real(), t); 4355 } 4356 4357 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4358 u8 tstamp_type) 4359 { 4360 skb->tstamp = kt; 4361 4362 if (kt) 4363 skb->tstamp_type = tstamp_type; 4364 else 4365 skb->tstamp_type = SKB_CLOCK_REALTIME; 4366 } 4367 4368 static inline void skb_set_delivery_type_by_clockid(struct sk_buff *skb, 4369 ktime_t kt, clockid_t clockid) 4370 { 4371 u8 tstamp_type = SKB_CLOCK_REALTIME; 4372 4373 switch (clockid) { 4374 case CLOCK_REALTIME: 4375 break; 4376 case CLOCK_MONOTONIC: 4377 tstamp_type = SKB_CLOCK_MONOTONIC; 4378 break; 4379 case CLOCK_TAI: 4380 tstamp_type = SKB_CLOCK_TAI; 4381 break; 4382 default: 4383 WARN_ON_ONCE(1); 4384 kt = 0; 4385 } 4386 4387 skb_set_delivery_time(skb, kt, tstamp_type); 4388 } 4389 4390 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4391 4392 /* It is used in the ingress path to clear the delivery_time. 4393 * If needed, set the skb->tstamp to the (rcv) timestamp. 4394 */ 4395 static inline void skb_clear_delivery_time(struct sk_buff *skb) 4396 { 4397 if (skb->tstamp_type) { 4398 skb->tstamp_type = SKB_CLOCK_REALTIME; 4399 if (static_branch_unlikely(&netstamp_needed_key)) 4400 skb->tstamp = ktime_get_real(); 4401 else 4402 skb->tstamp = 0; 4403 } 4404 } 4405 4406 static inline void skb_clear_tstamp(struct sk_buff *skb) 4407 { 4408 if (skb->tstamp_type) 4409 return; 4410 4411 skb->tstamp = 0; 4412 } 4413 4414 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4415 { 4416 if (skb->tstamp_type) 4417 return 0; 4418 4419 return skb->tstamp; 4420 } 4421 4422 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4423 { 4424 if (skb->tstamp_type != SKB_CLOCK_MONOTONIC && skb->tstamp) 4425 return skb->tstamp; 4426 4427 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4428 return ktime_get_real(); 4429 4430 return 0; 4431 } 4432 4433 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4434 { 4435 return skb_shinfo(skb)->meta_len; 4436 } 4437 4438 static inline void *skb_metadata_end(const struct sk_buff *skb) 4439 { 4440 return skb_mac_header(skb); 4441 } 4442 4443 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4444 const struct sk_buff *skb_b, 4445 u8 meta_len) 4446 { 4447 const void *a = skb_metadata_end(skb_a); 4448 const void *b = skb_metadata_end(skb_b); 4449 u64 diffs = 0; 4450 4451 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) || 4452 BITS_PER_LONG != 64) 4453 goto slow; 4454 4455 /* Using more efficient variant than plain call to memcmp(). */ 4456 switch (meta_len) { 4457 #define __it(x, op) (x -= sizeof(u##op)) 4458 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4459 case 32: diffs |= __it_diff(a, b, 64); 4460 fallthrough; 4461 case 24: diffs |= __it_diff(a, b, 64); 4462 fallthrough; 4463 case 16: diffs |= __it_diff(a, b, 64); 4464 fallthrough; 4465 case 8: diffs |= __it_diff(a, b, 64); 4466 break; 4467 case 28: diffs |= __it_diff(a, b, 64); 4468 fallthrough; 4469 case 20: diffs |= __it_diff(a, b, 64); 4470 fallthrough; 4471 case 12: diffs |= __it_diff(a, b, 64); 4472 fallthrough; 4473 case 4: diffs |= __it_diff(a, b, 32); 4474 break; 4475 default: 4476 slow: 4477 return memcmp(a - meta_len, b - meta_len, meta_len); 4478 } 4479 return diffs; 4480 } 4481 4482 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4483 const struct sk_buff *skb_b) 4484 { 4485 u8 len_a = skb_metadata_len(skb_a); 4486 u8 len_b = skb_metadata_len(skb_b); 4487 4488 if (!(len_a | len_b)) 4489 return false; 4490 4491 return len_a != len_b ? 4492 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4493 } 4494 4495 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4496 { 4497 skb_shinfo(skb)->meta_len = meta_len; 4498 } 4499 4500 static inline void skb_metadata_clear(struct sk_buff *skb) 4501 { 4502 skb_metadata_set(skb, 0); 4503 } 4504 4505 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4506 4507 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4508 4509 void skb_clone_tx_timestamp(struct sk_buff *skb); 4510 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4511 4512 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4513 4514 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4515 { 4516 } 4517 4518 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4519 { 4520 return false; 4521 } 4522 4523 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4524 4525 /** 4526 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4527 * 4528 * PHY drivers may accept clones of transmitted packets for 4529 * timestamping via their phy_driver.txtstamp method. These drivers 4530 * must call this function to return the skb back to the stack with a 4531 * timestamp. 4532 * 4533 * @skb: clone of the original outgoing packet 4534 * @hwtstamps: hardware time stamps 4535 * 4536 */ 4537 void skb_complete_tx_timestamp(struct sk_buff *skb, 4538 struct skb_shared_hwtstamps *hwtstamps); 4539 4540 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4541 struct skb_shared_hwtstamps *hwtstamps, 4542 struct sock *sk, int tstype); 4543 4544 /** 4545 * skb_tstamp_tx - queue clone of skb with send time stamps 4546 * @orig_skb: the original outgoing packet 4547 * @hwtstamps: hardware time stamps, may be NULL if not available 4548 * 4549 * If the skb has a socket associated, then this function clones the 4550 * skb (thus sharing the actual data and optional structures), stores 4551 * the optional hardware time stamping information (if non NULL) or 4552 * generates a software time stamp (otherwise), then queues the clone 4553 * to the error queue of the socket. Errors are silently ignored. 4554 */ 4555 void skb_tstamp_tx(struct sk_buff *orig_skb, 4556 struct skb_shared_hwtstamps *hwtstamps); 4557 4558 /** 4559 * skb_tx_timestamp() - Driver hook for transmit timestamping 4560 * 4561 * Ethernet MAC Drivers should call this function in their hard_xmit() 4562 * function immediately before giving the sk_buff to the MAC hardware. 4563 * 4564 * Specifically, one should make absolutely sure that this function is 4565 * called before TX completion of this packet can trigger. Otherwise 4566 * the packet could potentially already be freed. 4567 * 4568 * @skb: A socket buffer. 4569 */ 4570 static inline void skb_tx_timestamp(struct sk_buff *skb) 4571 { 4572 skb_clone_tx_timestamp(skb); 4573 if (skb_shinfo(skb)->tx_flags & (SKBTX_SW_TSTAMP | SKBTX_BPF)) 4574 skb_tstamp_tx(skb, NULL); 4575 } 4576 4577 /** 4578 * skb_complete_wifi_ack - deliver skb with wifi status 4579 * 4580 * @skb: the original outgoing packet 4581 * @acked: ack status 4582 * 4583 */ 4584 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4585 4586 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4587 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4588 4589 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4590 { 4591 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4592 skb->csum_valid || 4593 (skb->ip_summed == CHECKSUM_PARTIAL && 4594 skb_checksum_start_offset(skb) >= 0)); 4595 } 4596 4597 /** 4598 * skb_checksum_complete - Calculate checksum of an entire packet 4599 * @skb: packet to process 4600 * 4601 * This function calculates the checksum over the entire packet plus 4602 * the value of skb->csum. The latter can be used to supply the 4603 * checksum of a pseudo header as used by TCP/UDP. It returns the 4604 * checksum. 4605 * 4606 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4607 * this function can be used to verify that checksum on received 4608 * packets. In that case the function should return zero if the 4609 * checksum is correct. In particular, this function will return zero 4610 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4611 * hardware has already verified the correctness of the checksum. 4612 */ 4613 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4614 { 4615 return skb_csum_unnecessary(skb) ? 4616 0 : __skb_checksum_complete(skb); 4617 } 4618 4619 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4620 { 4621 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4622 if (skb->csum_level == 0) 4623 skb->ip_summed = CHECKSUM_NONE; 4624 else 4625 skb->csum_level--; 4626 } 4627 } 4628 4629 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4630 { 4631 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4632 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4633 skb->csum_level++; 4634 } else if (skb->ip_summed == CHECKSUM_NONE) { 4635 skb->ip_summed = CHECKSUM_UNNECESSARY; 4636 skb->csum_level = 0; 4637 } 4638 } 4639 4640 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4641 { 4642 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4643 skb->ip_summed = CHECKSUM_NONE; 4644 skb->csum_level = 0; 4645 } 4646 } 4647 4648 /* Check if we need to perform checksum complete validation. 4649 * 4650 * Returns: true if checksum complete is needed, false otherwise 4651 * (either checksum is unnecessary or zero checksum is allowed). 4652 */ 4653 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4654 bool zero_okay, 4655 __sum16 check) 4656 { 4657 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4658 skb->csum_valid = 1; 4659 __skb_decr_checksum_unnecessary(skb); 4660 return false; 4661 } 4662 4663 return true; 4664 } 4665 4666 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4667 * in checksum_init. 4668 */ 4669 #define CHECKSUM_BREAK 76 4670 4671 /* Unset checksum-complete 4672 * 4673 * Unset checksum complete can be done when packet is being modified 4674 * (uncompressed for instance) and checksum-complete value is 4675 * invalidated. 4676 */ 4677 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4678 { 4679 if (skb->ip_summed == CHECKSUM_COMPLETE) 4680 skb->ip_summed = CHECKSUM_NONE; 4681 } 4682 4683 /* Validate (init) checksum based on checksum complete. 4684 * 4685 * Return values: 4686 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4687 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4688 * checksum is stored in skb->csum for use in __skb_checksum_complete 4689 * non-zero: value of invalid checksum 4690 * 4691 */ 4692 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4693 bool complete, 4694 __wsum psum) 4695 { 4696 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4697 if (!csum_fold(csum_add(psum, skb->csum))) { 4698 skb->csum_valid = 1; 4699 return 0; 4700 } 4701 } 4702 4703 skb->csum = psum; 4704 4705 if (complete || skb->len <= CHECKSUM_BREAK) { 4706 __sum16 csum; 4707 4708 csum = __skb_checksum_complete(skb); 4709 skb->csum_valid = !csum; 4710 return csum; 4711 } 4712 4713 return 0; 4714 } 4715 4716 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4717 { 4718 return 0; 4719 } 4720 4721 /* Perform checksum validate (init). Note that this is a macro since we only 4722 * want to calculate the pseudo header which is an input function if necessary. 4723 * First we try to validate without any computation (checksum unnecessary) and 4724 * then calculate based on checksum complete calling the function to compute 4725 * pseudo header. 4726 * 4727 * Return values: 4728 * 0: checksum is validated or try to in skb_checksum_complete 4729 * non-zero: value of invalid checksum 4730 */ 4731 #define __skb_checksum_validate(skb, proto, complete, \ 4732 zero_okay, check, compute_pseudo) \ 4733 ({ \ 4734 __sum16 __ret = 0; \ 4735 skb->csum_valid = 0; \ 4736 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4737 __ret = __skb_checksum_validate_complete(skb, \ 4738 complete, compute_pseudo(skb, proto)); \ 4739 __ret; \ 4740 }) 4741 4742 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4743 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4744 4745 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4746 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4747 4748 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4749 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4750 4751 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4752 compute_pseudo) \ 4753 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4754 4755 #define skb_checksum_simple_validate(skb) \ 4756 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4757 4758 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4759 { 4760 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4761 } 4762 4763 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4764 { 4765 skb->csum = ~pseudo; 4766 skb->ip_summed = CHECKSUM_COMPLETE; 4767 } 4768 4769 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4770 do { \ 4771 if (__skb_checksum_convert_check(skb)) \ 4772 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4773 } while (0) 4774 4775 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4776 u16 start, u16 offset) 4777 { 4778 skb->ip_summed = CHECKSUM_PARTIAL; 4779 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4780 skb->csum_offset = offset - start; 4781 } 4782 4783 /* Update skbuf and packet to reflect the remote checksum offload operation. 4784 * When called, ptr indicates the starting point for skb->csum when 4785 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4786 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4787 */ 4788 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4789 int start, int offset, bool nopartial) 4790 { 4791 __wsum delta; 4792 4793 if (!nopartial) { 4794 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4795 return; 4796 } 4797 4798 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4799 __skb_checksum_complete(skb); 4800 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4801 } 4802 4803 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4804 4805 /* Adjust skb->csum since we changed the packet */ 4806 skb->csum = csum_add(skb->csum, delta); 4807 } 4808 4809 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4810 { 4811 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4812 return (void *)(skb->_nfct & NFCT_PTRMASK); 4813 #else 4814 return NULL; 4815 #endif 4816 } 4817 4818 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4819 { 4820 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4821 return skb->_nfct; 4822 #else 4823 return 0UL; 4824 #endif 4825 } 4826 4827 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4828 { 4829 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4830 skb->slow_gro |= !!nfct; 4831 skb->_nfct = nfct; 4832 #endif 4833 } 4834 4835 #ifdef CONFIG_SKB_EXTENSIONS 4836 enum skb_ext_id { 4837 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4838 SKB_EXT_BRIDGE_NF, 4839 #endif 4840 #ifdef CONFIG_XFRM 4841 SKB_EXT_SEC_PATH, 4842 #endif 4843 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4844 TC_SKB_EXT, 4845 #endif 4846 #if IS_ENABLED(CONFIG_MPTCP) 4847 SKB_EXT_MPTCP, 4848 #endif 4849 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 4850 SKB_EXT_MCTP, 4851 #endif 4852 SKB_EXT_NUM, /* must be last */ 4853 }; 4854 4855 /** 4856 * struct skb_ext - sk_buff extensions 4857 * @refcnt: 1 on allocation, deallocated on 0 4858 * @offset: offset to add to @data to obtain extension address 4859 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4860 * @data: start of extension data, variable sized 4861 * 4862 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4863 * to use 'u8' types while allowing up to 2kb worth of extension data. 4864 */ 4865 struct skb_ext { 4866 refcount_t refcnt; 4867 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4868 u8 chunks; /* same */ 4869 char data[] __aligned(8); 4870 }; 4871 4872 struct skb_ext *__skb_ext_alloc(gfp_t flags); 4873 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 4874 struct skb_ext *ext); 4875 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4876 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4877 void __skb_ext_put(struct skb_ext *ext); 4878 4879 static inline void skb_ext_put(struct sk_buff *skb) 4880 { 4881 if (skb->active_extensions) 4882 __skb_ext_put(skb->extensions); 4883 } 4884 4885 static inline void __skb_ext_copy(struct sk_buff *dst, 4886 const struct sk_buff *src) 4887 { 4888 dst->active_extensions = src->active_extensions; 4889 4890 if (src->active_extensions) { 4891 struct skb_ext *ext = src->extensions; 4892 4893 refcount_inc(&ext->refcnt); 4894 dst->extensions = ext; 4895 } 4896 } 4897 4898 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4899 { 4900 skb_ext_put(dst); 4901 __skb_ext_copy(dst, src); 4902 } 4903 4904 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4905 { 4906 return !!ext->offset[i]; 4907 } 4908 4909 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4910 { 4911 return skb->active_extensions & (1 << id); 4912 } 4913 4914 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4915 { 4916 if (skb_ext_exist(skb, id)) 4917 __skb_ext_del(skb, id); 4918 } 4919 4920 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4921 { 4922 if (skb_ext_exist(skb, id)) { 4923 struct skb_ext *ext = skb->extensions; 4924 4925 return (void *)ext + (ext->offset[id] << 3); 4926 } 4927 4928 return NULL; 4929 } 4930 4931 static inline void skb_ext_reset(struct sk_buff *skb) 4932 { 4933 if (unlikely(skb->active_extensions)) { 4934 __skb_ext_put(skb->extensions); 4935 skb->active_extensions = 0; 4936 } 4937 } 4938 4939 static inline bool skb_has_extensions(struct sk_buff *skb) 4940 { 4941 return unlikely(skb->active_extensions); 4942 } 4943 #else 4944 static inline void skb_ext_put(struct sk_buff *skb) {} 4945 static inline void skb_ext_reset(struct sk_buff *skb) {} 4946 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4947 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4948 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4949 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 4950 #endif /* CONFIG_SKB_EXTENSIONS */ 4951 4952 static inline void nf_reset_ct(struct sk_buff *skb) 4953 { 4954 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4955 nf_conntrack_put(skb_nfct(skb)); 4956 skb->_nfct = 0; 4957 #endif 4958 } 4959 4960 static inline void nf_reset_trace(struct sk_buff *skb) 4961 { 4962 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4963 skb->nf_trace = 0; 4964 #endif 4965 } 4966 4967 static inline void ipvs_reset(struct sk_buff *skb) 4968 { 4969 #if IS_ENABLED(CONFIG_IP_VS) 4970 skb->ipvs_property = 0; 4971 #endif 4972 } 4973 4974 /* Note: This doesn't put any conntrack info in dst. */ 4975 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4976 bool copy) 4977 { 4978 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4979 dst->_nfct = src->_nfct; 4980 nf_conntrack_get(skb_nfct(src)); 4981 #endif 4982 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4983 if (copy) 4984 dst->nf_trace = src->nf_trace; 4985 #endif 4986 } 4987 4988 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 4989 { 4990 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4991 nf_conntrack_put(skb_nfct(dst)); 4992 #endif 4993 dst->slow_gro = src->slow_gro; 4994 __nf_copy(dst, src, true); 4995 } 4996 4997 #ifdef CONFIG_NETWORK_SECMARK 4998 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4999 { 5000 to->secmark = from->secmark; 5001 } 5002 5003 static inline void skb_init_secmark(struct sk_buff *skb) 5004 { 5005 skb->secmark = 0; 5006 } 5007 #else 5008 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 5009 { } 5010 5011 static inline void skb_init_secmark(struct sk_buff *skb) 5012 { } 5013 #endif 5014 5015 static inline int secpath_exists(const struct sk_buff *skb) 5016 { 5017 #ifdef CONFIG_XFRM 5018 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 5019 #else 5020 return 0; 5021 #endif 5022 } 5023 5024 static inline bool skb_irq_freeable(const struct sk_buff *skb) 5025 { 5026 return !skb->destructor && 5027 !secpath_exists(skb) && 5028 !skb_nfct(skb) && 5029 !skb->_skb_refdst && 5030 !skb_has_frag_list(skb); 5031 } 5032 5033 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 5034 { 5035 skb->queue_mapping = queue_mapping; 5036 } 5037 5038 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 5039 { 5040 return skb->queue_mapping; 5041 } 5042 5043 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 5044 { 5045 to->queue_mapping = from->queue_mapping; 5046 } 5047 5048 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 5049 { 5050 skb->queue_mapping = rx_queue + 1; 5051 } 5052 5053 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 5054 { 5055 return skb->queue_mapping - 1; 5056 } 5057 5058 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 5059 { 5060 return skb->queue_mapping != 0; 5061 } 5062 5063 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 5064 { 5065 skb->dst_pending_confirm = val; 5066 } 5067 5068 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 5069 { 5070 return skb->dst_pending_confirm != 0; 5071 } 5072 5073 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 5074 { 5075 #ifdef CONFIG_XFRM 5076 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 5077 #else 5078 return NULL; 5079 #endif 5080 } 5081 5082 static inline bool skb_is_gso(const struct sk_buff *skb) 5083 { 5084 return skb_shinfo(skb)->gso_size; 5085 } 5086 5087 /* Note: Should be called only if skb_is_gso(skb) is true */ 5088 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 5089 { 5090 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 5091 } 5092 5093 /* Note: Should be called only if skb_is_gso(skb) is true */ 5094 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 5095 { 5096 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 5097 } 5098 5099 /* Note: Should be called only if skb_is_gso(skb) is true */ 5100 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 5101 { 5102 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 5103 } 5104 5105 static inline void skb_gso_reset(struct sk_buff *skb) 5106 { 5107 skb_shinfo(skb)->gso_size = 0; 5108 skb_shinfo(skb)->gso_segs = 0; 5109 skb_shinfo(skb)->gso_type = 0; 5110 } 5111 5112 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 5113 u16 increment) 5114 { 5115 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5116 return; 5117 shinfo->gso_size += increment; 5118 } 5119 5120 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 5121 u16 decrement) 5122 { 5123 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5124 return; 5125 shinfo->gso_size -= decrement; 5126 } 5127 5128 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 5129 5130 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 5131 { 5132 /* LRO sets gso_size but not gso_type, whereas if GSO is really 5133 * wanted then gso_type will be set. */ 5134 const struct skb_shared_info *shinfo = skb_shinfo(skb); 5135 5136 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 5137 unlikely(shinfo->gso_type == 0)) { 5138 __skb_warn_lro_forwarding(skb); 5139 return true; 5140 } 5141 return false; 5142 } 5143 5144 static inline void skb_forward_csum(struct sk_buff *skb) 5145 { 5146 /* Unfortunately we don't support this one. Any brave souls? */ 5147 if (skb->ip_summed == CHECKSUM_COMPLETE) 5148 skb->ip_summed = CHECKSUM_NONE; 5149 } 5150 5151 /** 5152 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 5153 * @skb: skb to check 5154 * 5155 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 5156 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 5157 * use this helper, to document places where we make this assertion. 5158 */ 5159 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 5160 { 5161 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 5162 } 5163 5164 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 5165 5166 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 5167 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 5168 unsigned int transport_len, 5169 __sum16(*skb_chkf)(struct sk_buff *skb)); 5170 5171 /** 5172 * skb_head_is_locked - Determine if the skb->head is locked down 5173 * @skb: skb to check 5174 * 5175 * The head on skbs build around a head frag can be removed if they are 5176 * not cloned. This function returns true if the skb head is locked down 5177 * due to either being allocated via kmalloc, or by being a clone with 5178 * multiple references to the head. 5179 */ 5180 static inline bool skb_head_is_locked(const struct sk_buff *skb) 5181 { 5182 return !skb->head_frag || skb_cloned(skb); 5183 } 5184 5185 /* Local Checksum Offload. 5186 * Compute outer checksum based on the assumption that the 5187 * inner checksum will be offloaded later. 5188 * See Documentation/networking/checksum-offloads.rst for 5189 * explanation of how this works. 5190 * Fill in outer checksum adjustment (e.g. with sum of outer 5191 * pseudo-header) before calling. 5192 * Also ensure that inner checksum is in linear data area. 5193 */ 5194 static inline __wsum lco_csum(struct sk_buff *skb) 5195 { 5196 unsigned char *csum_start = skb_checksum_start(skb); 5197 unsigned char *l4_hdr = skb_transport_header(skb); 5198 __wsum partial; 5199 5200 /* Start with complement of inner checksum adjustment */ 5201 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 5202 skb->csum_offset)); 5203 5204 /* Add in checksum of our headers (incl. outer checksum 5205 * adjustment filled in by caller) and return result. 5206 */ 5207 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5208 } 5209 5210 static inline bool skb_is_redirected(const struct sk_buff *skb) 5211 { 5212 return skb->redirected; 5213 } 5214 5215 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5216 { 5217 skb->redirected = 1; 5218 #ifdef CONFIG_NET_REDIRECT 5219 skb->from_ingress = from_ingress; 5220 if (skb->from_ingress) 5221 skb_clear_tstamp(skb); 5222 #endif 5223 } 5224 5225 static inline void skb_reset_redirect(struct sk_buff *skb) 5226 { 5227 skb->redirected = 0; 5228 } 5229 5230 static inline void skb_set_redirected_noclear(struct sk_buff *skb, 5231 bool from_ingress) 5232 { 5233 skb->redirected = 1; 5234 #ifdef CONFIG_NET_REDIRECT 5235 skb->from_ingress = from_ingress; 5236 #endif 5237 } 5238 5239 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5240 { 5241 #if IS_ENABLED(CONFIG_IP_SCTP) 5242 return skb->csum_not_inet; 5243 #else 5244 return 0; 5245 #endif 5246 } 5247 5248 static inline void skb_reset_csum_not_inet(struct sk_buff *skb) 5249 { 5250 skb->ip_summed = CHECKSUM_NONE; 5251 #if IS_ENABLED(CONFIG_IP_SCTP) 5252 skb->csum_not_inet = 0; 5253 #endif 5254 } 5255 5256 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5257 const u64 kcov_handle) 5258 { 5259 #ifdef CONFIG_KCOV 5260 skb->kcov_handle = kcov_handle; 5261 #endif 5262 } 5263 5264 static inline u64 skb_get_kcov_handle(struct sk_buff *skb) 5265 { 5266 #ifdef CONFIG_KCOV 5267 return skb->kcov_handle; 5268 #else 5269 return 0; 5270 #endif 5271 } 5272 5273 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5274 { 5275 #ifdef CONFIG_PAGE_POOL 5276 skb->pp_recycle = 1; 5277 #endif 5278 } 5279 5280 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter, 5281 ssize_t maxsize, gfp_t gfp); 5282 5283 #endif /* __KERNEL__ */ 5284 #endif /* _LINUX_SKBUFF_H */ 5285