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