1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * Definitions for the 'struct sk_buff' memory handlers. 4 * 5 * Authors: 6 * Alan Cox, <[email protected]> 7 * Florian La Roche, <[email protected]> 8 */ 9 10 #ifndef _LINUX_SKBUFF_H 11 #define _LINUX_SKBUFF_H 12 13 #include <linux/kernel.h> 14 #include <linux/compiler.h> 15 #include <linux/time.h> 16 #include <linux/bug.h> 17 #include <linux/bvec.h> 18 #include <linux/cache.h> 19 #include <linux/rbtree.h> 20 #include <linux/socket.h> 21 #include <linux/refcount.h> 22 23 #include <linux/atomic.h> 24 #include <asm/types.h> 25 #include <linux/spinlock.h> 26 #include <net/checksum.h> 27 #include <linux/rcupdate.h> 28 #include <linux/dma-mapping.h> 29 #include <linux/netdev_features.h> 30 #include <net/flow_dissector.h> 31 #include <linux/in6.h> 32 #include <linux/if_packet.h> 33 #include <linux/llist.h> 34 #include <linux/page_frag_cache.h> 35 #include <net/flow.h> 36 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 37 #include <linux/netfilter/nf_conntrack_common.h> 38 #endif 39 #include <net/net_debug.h> 40 #include <net/dropreason-core.h> 41 #include <net/netmem.h> 42 43 /** 44 * DOC: skb checksums 45 * 46 * The interface for checksum offload between the stack and networking drivers 47 * is as follows... 48 * 49 * IP checksum related features 50 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 51 * 52 * Drivers advertise checksum offload capabilities in the features of a device. 53 * From the stack's point of view these are capabilities offered by the driver. 54 * A driver typically only advertises features that it is capable of offloading 55 * to its device. 56 * 57 * .. flat-table:: Checksum related device features 58 * :widths: 1 10 59 * 60 * * - %NETIF_F_HW_CSUM 61 * - The driver (or its device) is able to compute one 62 * IP (one's complement) checksum for any combination 63 * of protocols or protocol layering. The checksum is 64 * computed and set in a packet per the CHECKSUM_PARTIAL 65 * interface (see below). 66 * 67 * * - %NETIF_F_IP_CSUM 68 * - Driver (device) is only able to checksum plain 69 * TCP or UDP packets over IPv4. These are specifically 70 * unencapsulated packets of the form IPv4|TCP or 71 * IPv4|UDP where the Protocol field in the IPv4 header 72 * is TCP or UDP. The IPv4 header may contain IP options. 73 * This feature cannot be set in features for a device 74 * with NETIF_F_HW_CSUM also set. This feature is being 75 * DEPRECATED (see below). 76 * 77 * * - %NETIF_F_IPV6_CSUM 78 * - Driver (device) is only able to checksum plain 79 * TCP or UDP packets over IPv6. These are specifically 80 * unencapsulated packets of the form IPv6|TCP or 81 * IPv6|UDP where the Next Header field in the IPv6 82 * header is either TCP or UDP. IPv6 extension headers 83 * are not supported with this feature. This feature 84 * cannot be set in features for a device with 85 * NETIF_F_HW_CSUM also set. This feature is being 86 * DEPRECATED (see below). 87 * 88 * * - %NETIF_F_RXCSUM 89 * - Driver (device) performs receive checksum offload. 90 * This flag is only used to disable the RX checksum 91 * feature for a device. The stack will accept receive 92 * checksum indication in packets received on a device 93 * regardless of whether NETIF_F_RXCSUM is set. 94 * 95 * Checksumming of received packets by device 96 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 97 * 98 * Indication of checksum verification is set in &sk_buff.ip_summed. 99 * Possible values are: 100 * 101 * - %CHECKSUM_NONE 102 * 103 * Device did not checksum this packet e.g. due to lack of capabilities. 104 * The packet contains full (though not verified) checksum in packet but 105 * not in skb->csum. Thus, skb->csum is undefined in this case. 106 * 107 * - %CHECKSUM_UNNECESSARY 108 * 109 * The hardware you're dealing with doesn't calculate the full checksum 110 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums 111 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY 112 * if their checksums are okay. &sk_buff.csum is still undefined in this case 113 * though. A driver or device must never modify the checksum field in the 114 * packet even if checksum is verified. 115 * 116 * %CHECKSUM_UNNECESSARY is applicable to following protocols: 117 * 118 * - TCP: IPv6 and IPv4. 119 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 120 * zero UDP checksum for either IPv4 or IPv6, the networking stack 121 * may perform further validation in this case. 122 * - GRE: only if the checksum is present in the header. 123 * - SCTP: indicates the CRC in SCTP header has been validated. 124 * - FCOE: indicates the CRC in FC frame has been validated. 125 * 126 * &sk_buff.csum_level indicates the number of consecutive checksums found in 127 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY. 128 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 129 * and a device is able to verify the checksums for UDP (possibly zero), 130 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to 131 * two. If the device were only able to verify the UDP checksum and not 132 * GRE, either because it doesn't support GRE checksum or because GRE 133 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 134 * not considered in this case). 135 * 136 * - %CHECKSUM_COMPLETE 137 * 138 * This is the most generic way. The device supplied checksum of the _whole_ 139 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the 140 * hardware doesn't need to parse L3/L4 headers to implement this. 141 * 142 * Notes: 143 * 144 * - Even if device supports only some protocols, but is able to produce 145 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 146 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 147 * 148 * - %CHECKSUM_PARTIAL 149 * 150 * A checksum is set up to be offloaded to a device as described in the 151 * output description for CHECKSUM_PARTIAL. This may occur on a packet 152 * received directly from another Linux OS, e.g., a virtualized Linux kernel 153 * on the same host, or it may be set in the input path in GRO or remote 154 * checksum offload. For the purposes of checksum verification, the checksum 155 * referred to by skb->csum_start + skb->csum_offset and any preceding 156 * checksums in the packet are considered verified. Any checksums in the 157 * packet that are after the checksum being offloaded are not considered to 158 * be verified. 159 * 160 * Checksumming on transmit for non-GSO 161 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 162 * 163 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet. 164 * Values are: 165 * 166 * - %CHECKSUM_PARTIAL 167 * 168 * The driver is required to checksum the packet as seen by hard_start_xmit() 169 * from &sk_buff.csum_start up to the end, and to record/write the checksum at 170 * offset &sk_buff.csum_start + &sk_buff.csum_offset. 171 * A driver may verify that the 172 * csum_start and csum_offset values are valid values given the length and 173 * offset of the packet, but it should not attempt to validate that the 174 * checksum refers to a legitimate transport layer checksum -- it is the 175 * purview of the stack to validate that csum_start and csum_offset are set 176 * correctly. 177 * 178 * When the stack requests checksum offload for a packet, the driver MUST 179 * ensure that the checksum is set correctly. A driver can either offload the 180 * checksum calculation to the device, or call skb_checksum_help (in the case 181 * that the device does not support offload for a particular checksum). 182 * 183 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of 184 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate 185 * checksum offload capability. 186 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based 187 * on network device checksumming capabilities: if a packet does not match 188 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of 189 * &sk_buff.csum_not_inet, see :ref:`crc`) 190 * is called to resolve the checksum. 191 * 192 * - %CHECKSUM_NONE 193 * 194 * The skb was already checksummed by the protocol, or a checksum is not 195 * required. 196 * 197 * - %CHECKSUM_UNNECESSARY 198 * 199 * This has the same meaning as CHECKSUM_NONE for checksum offload on 200 * output. 201 * 202 * - %CHECKSUM_COMPLETE 203 * 204 * Not used in checksum output. If a driver observes a packet with this value 205 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set. 206 * 207 * .. _crc: 208 * 209 * Non-IP checksum (CRC) offloads 210 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 211 * 212 * .. flat-table:: 213 * :widths: 1 10 214 * 215 * * - %NETIF_F_SCTP_CRC 216 * - This feature indicates that a device is capable of 217 * offloading the SCTP CRC in a packet. To perform this offload the stack 218 * will set csum_start and csum_offset accordingly, set ip_summed to 219 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication 220 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c. 221 * A driver that supports both IP checksum offload and SCTP CRC32c offload 222 * must verify which offload is configured for a packet by testing the 223 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to 224 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 225 * 226 * * - %NETIF_F_FCOE_CRC 227 * - This feature indicates that a device is capable of offloading the FCOE 228 * CRC in a packet. To perform this offload the stack will set ip_summed 229 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset 230 * accordingly. Note that there is no indication in the skbuff that the 231 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports 232 * both IP checksum offload and FCOE CRC offload must verify which offload 233 * is configured for a packet, presumably by inspecting packet headers. 234 * 235 * Checksumming on output with GSO 236 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 237 * 238 * In the case of a GSO packet (skb_is_gso() is true), checksum offload 239 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 240 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as 241 * part of the GSO operation is implied. If a checksum is being offloaded 242 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and 243 * csum_offset are set to refer to the outermost checksum being offloaded 244 * (two offloaded checksums are possible with UDP encapsulation). 245 */ 246 247 /* Don't change this without changing skb_csum_unnecessary! */ 248 #define CHECKSUM_NONE 0 249 #define CHECKSUM_UNNECESSARY 1 250 #define CHECKSUM_COMPLETE 2 251 #define CHECKSUM_PARTIAL 3 252 253 /* Maximum value in skb->csum_level */ 254 #define SKB_MAX_CSUM_LEVEL 3 255 256 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 257 #define SKB_WITH_OVERHEAD(X) \ 258 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 259 260 /* For X bytes available in skb->head, what is the minimal 261 * allocation needed, knowing struct skb_shared_info needs 262 * to be aligned. 263 */ 264 #define SKB_HEAD_ALIGN(X) (SKB_DATA_ALIGN(X) + \ 265 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 266 267 #define SKB_MAX_ORDER(X, ORDER) \ 268 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 269 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 270 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 271 272 /* return minimum truesize of one skb containing X bytes of data */ 273 #define SKB_TRUESIZE(X) ((X) + \ 274 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 275 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 276 277 struct ahash_request; 278 struct net_device; 279 struct scatterlist; 280 struct pipe_inode_info; 281 struct iov_iter; 282 struct napi_struct; 283 struct bpf_prog; 284 union bpf_attr; 285 struct skb_ext; 286 struct ts_config; 287 288 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 289 struct nf_bridge_info { 290 enum { 291 BRNF_PROTO_UNCHANGED, 292 BRNF_PROTO_8021Q, 293 BRNF_PROTO_PPPOE 294 } orig_proto:8; 295 u8 pkt_otherhost:1; 296 u8 in_prerouting:1; 297 u8 bridged_dnat:1; 298 u8 sabotage_in_done:1; 299 __u16 frag_max_size; 300 int physinif; 301 302 /* always valid & non-NULL from FORWARD on, for physdev match */ 303 struct net_device *physoutdev; 304 union { 305 /* prerouting: detect dnat in orig/reply direction */ 306 __be32 ipv4_daddr; 307 struct in6_addr ipv6_daddr; 308 309 /* after prerouting + nat detected: store original source 310 * mac since neigh resolution overwrites it, only used while 311 * skb is out in neigh layer. 312 */ 313 char neigh_header[8]; 314 }; 315 }; 316 #endif 317 318 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 319 /* Chain in tc_skb_ext will be used to share the tc chain with 320 * ovs recirc_id. It will be set to the current chain by tc 321 * and read by ovs to recirc_id. 322 */ 323 struct tc_skb_ext { 324 union { 325 u64 act_miss_cookie; 326 __u32 chain; 327 }; 328 __u16 mru; 329 __u16 zone; 330 u8 post_ct:1; 331 u8 post_ct_snat:1; 332 u8 post_ct_dnat:1; 333 u8 act_miss:1; /* Set if act_miss_cookie is used */ 334 u8 l2_miss:1; /* Set by bridge upon FDB or MDB miss */ 335 }; 336 #endif 337 338 struct sk_buff_head { 339 /* These two members must be first to match sk_buff. */ 340 struct_group_tagged(sk_buff_list, list, 341 struct sk_buff *next; 342 struct sk_buff *prev; 343 ); 344 345 __u32 qlen; 346 spinlock_t lock; 347 }; 348 349 struct sk_buff; 350 351 #ifndef CONFIG_MAX_SKB_FRAGS 352 # define CONFIG_MAX_SKB_FRAGS 17 353 #endif 354 355 #define MAX_SKB_FRAGS CONFIG_MAX_SKB_FRAGS 356 357 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 358 * segment using its current segmentation instead. 359 */ 360 #define GSO_BY_FRAGS 0xFFFF 361 362 typedef struct skb_frag { 363 netmem_ref netmem; 364 unsigned int len; 365 unsigned int offset; 366 } skb_frag_t; 367 368 /** 369 * skb_frag_size() - Returns the size of a skb fragment 370 * @frag: skb fragment 371 */ 372 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 373 { 374 return frag->len; 375 } 376 377 /** 378 * skb_frag_size_set() - Sets the size of a skb fragment 379 * @frag: skb fragment 380 * @size: size of fragment 381 */ 382 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 383 { 384 frag->len = size; 385 } 386 387 /** 388 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 389 * @frag: skb fragment 390 * @delta: value to add 391 */ 392 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 393 { 394 frag->len += delta; 395 } 396 397 /** 398 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 399 * @frag: skb fragment 400 * @delta: value to subtract 401 */ 402 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 403 { 404 frag->len -= delta; 405 } 406 407 /** 408 * skb_frag_must_loop - Test if %p is a high memory page 409 * @p: fragment's page 410 */ 411 static inline bool skb_frag_must_loop(struct page *p) 412 { 413 #if defined(CONFIG_HIGHMEM) 414 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 415 return true; 416 #endif 417 return false; 418 } 419 420 /** 421 * skb_frag_foreach_page - loop over pages in a fragment 422 * 423 * @f: skb frag to operate on 424 * @f_off: offset from start of f->netmem 425 * @f_len: length from f_off to loop over 426 * @p: (temp var) current page 427 * @p_off: (temp var) offset from start of current page, 428 * non-zero only on first page. 429 * @p_len: (temp var) length in current page, 430 * < PAGE_SIZE only on first and last page. 431 * @copied: (temp var) length so far, excluding current p_len. 432 * 433 * A fragment can hold a compound page, in which case per-page 434 * operations, notably kmap_atomic, must be called for each 435 * regular page. 436 */ 437 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 438 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 439 p_off = (f_off) & (PAGE_SIZE - 1), \ 440 p_len = skb_frag_must_loop(p) ? \ 441 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 442 copied = 0; \ 443 copied < f_len; \ 444 copied += p_len, p++, p_off = 0, \ 445 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 446 447 /** 448 * struct skb_shared_hwtstamps - hardware time stamps 449 * @hwtstamp: hardware time stamp transformed into duration 450 * since arbitrary point in time 451 * @netdev_data: address/cookie of network device driver used as 452 * reference to actual hardware time stamp 453 * 454 * Software time stamps generated by ktime_get_real() are stored in 455 * skb->tstamp. 456 * 457 * hwtstamps can only be compared against other hwtstamps from 458 * the same device. 459 * 460 * This structure is attached to packets as part of the 461 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 462 */ 463 struct skb_shared_hwtstamps { 464 union { 465 ktime_t hwtstamp; 466 void *netdev_data; 467 }; 468 }; 469 470 /* Definitions for tx_flags in struct skb_shared_info */ 471 enum { 472 /* generate hardware time stamp */ 473 SKBTX_HW_TSTAMP_NOBPF = 1 << 0, 474 475 /* generate software time stamp when queueing packet to NIC */ 476 SKBTX_SW_TSTAMP = 1 << 1, 477 478 /* device driver is going to provide hardware time stamp */ 479 SKBTX_IN_PROGRESS = 1 << 2, 480 481 /* reserved */ 482 SKBTX_RESERVED = 1 << 3, 483 484 /* generate wifi status information (where possible) */ 485 SKBTX_WIFI_STATUS = 1 << 4, 486 487 /* determine hardware time stamp based on time or cycles */ 488 SKBTX_HW_TSTAMP_NETDEV = 1 << 5, 489 490 /* generate software time stamp when entering packet scheduling */ 491 SKBTX_SCHED_TSTAMP = 1 << 6, 492 493 /* used for bpf extension when a bpf program is loaded */ 494 SKBTX_BPF = 1 << 7, 495 }; 496 497 #define SKBTX_HW_TSTAMP (SKBTX_HW_TSTAMP_NOBPF | SKBTX_BPF) 498 499 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 500 SKBTX_SCHED_TSTAMP | \ 501 SKBTX_BPF) 502 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \ 503 SKBTX_ANY_SW_TSTAMP) 504 505 /* Definitions for flags in struct skb_shared_info */ 506 enum { 507 /* use zcopy routines */ 508 SKBFL_ZEROCOPY_ENABLE = BIT(0), 509 510 /* This indicates at least one fragment might be overwritten 511 * (as in vmsplice(), sendfile() ...) 512 * If we need to compute a TX checksum, we'll need to copy 513 * all frags to avoid possible bad checksum 514 */ 515 SKBFL_SHARED_FRAG = BIT(1), 516 517 /* segment contains only zerocopy data and should not be 518 * charged to the kernel memory. 519 */ 520 SKBFL_PURE_ZEROCOPY = BIT(2), 521 522 SKBFL_DONT_ORPHAN = BIT(3), 523 524 /* page references are managed by the ubuf_info, so it's safe to 525 * use frags only up until ubuf_info is released 526 */ 527 SKBFL_MANAGED_FRAG_REFS = BIT(4), 528 }; 529 530 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG) 531 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \ 532 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS) 533 534 struct ubuf_info_ops { 535 void (*complete)(struct sk_buff *, struct ubuf_info *, 536 bool zerocopy_success); 537 /* has to be compatible with skb_zcopy_set() */ 538 int (*link_skb)(struct sk_buff *skb, struct ubuf_info *uarg); 539 }; 540 541 /* 542 * The callback notifies userspace to release buffers when skb DMA is done in 543 * lower device, the skb last reference should be 0 when calling this. 544 * The zerocopy_success argument is true if zero copy transmit occurred, 545 * false on data copy or out of memory error caused by data copy attempt. 546 * The ctx field is used to track device context. 547 * The desc field is used to track userspace buffer index. 548 */ 549 struct ubuf_info { 550 const struct ubuf_info_ops *ops; 551 refcount_t refcnt; 552 u8 flags; 553 }; 554 555 struct ubuf_info_msgzc { 556 struct ubuf_info ubuf; 557 558 union { 559 struct { 560 unsigned long desc; 561 void *ctx; 562 }; 563 struct { 564 u32 id; 565 u16 len; 566 u16 zerocopy:1; 567 u32 bytelen; 568 }; 569 }; 570 571 struct mmpin { 572 struct user_struct *user; 573 unsigned int num_pg; 574 } mmp; 575 }; 576 577 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 578 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \ 579 ubuf) 580 581 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 582 void mm_unaccount_pinned_pages(struct mmpin *mmp); 583 584 /* Preserve some data across TX submission and completion. 585 * 586 * Note, this state is stored in the driver. Extending the layout 587 * might need some special care. 588 */ 589 struct xsk_tx_metadata_compl { 590 __u64 *tx_timestamp; 591 }; 592 593 /* This data is invariant across clones and lives at 594 * the end of the header data, ie. at skb->end. 595 */ 596 struct skb_shared_info { 597 __u8 flags; 598 __u8 meta_len; 599 __u8 nr_frags; 600 __u8 tx_flags; 601 unsigned short gso_size; 602 /* Warning: this field is not always filled in (UFO)! */ 603 unsigned short gso_segs; 604 struct sk_buff *frag_list; 605 union { 606 struct skb_shared_hwtstamps hwtstamps; 607 struct xsk_tx_metadata_compl xsk_meta; 608 }; 609 unsigned int gso_type; 610 u32 tskey; 611 612 /* 613 * Warning : all fields before dataref are cleared in __alloc_skb() 614 */ 615 atomic_t dataref; 616 617 union { 618 struct { 619 u32 xdp_frags_size; 620 u32 xdp_frags_truesize; 621 }; 622 623 /* 624 * Intermediate layers must ensure that destructor_arg 625 * remains valid until skb destructor. 626 */ 627 void *destructor_arg; 628 }; 629 630 /* must be last field, see pskb_expand_head() */ 631 skb_frag_t frags[MAX_SKB_FRAGS]; 632 }; 633 634 /** 635 * DOC: dataref and headerless skbs 636 * 637 * Transport layers send out clones of payload skbs they hold for 638 * retransmissions. To allow lower layers of the stack to prepend their headers 639 * we split &skb_shared_info.dataref into two halves. 640 * The lower 16 bits count the overall number of references. 641 * The higher 16 bits indicate how many of the references are payload-only. 642 * skb_header_cloned() checks if skb is allowed to add / write the headers. 643 * 644 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr 645 * (via __skb_header_release()). Any clone created from marked skb will get 646 * &sk_buff.hdr_len populated with the available headroom. 647 * If there's the only clone in existence it's able to modify the headroom 648 * at will. The sequence of calls inside the transport layer is:: 649 * 650 * <alloc skb> 651 * skb_reserve() 652 * __skb_header_release() 653 * skb_clone() 654 * // send the clone down the stack 655 * 656 * This is not a very generic construct and it depends on the transport layers 657 * doing the right thing. In practice there's usually only one payload-only skb. 658 * Having multiple payload-only skbs with different lengths of hdr_len is not 659 * possible. The payload-only skbs should never leave their owner. 660 */ 661 #define SKB_DATAREF_SHIFT 16 662 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 663 664 665 enum { 666 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 667 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 668 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 669 }; 670 671 enum { 672 SKB_GSO_TCPV4 = 1 << 0, 673 674 /* This indicates the skb is from an untrusted source. */ 675 SKB_GSO_DODGY = 1 << 1, 676 677 /* This indicates the tcp segment has CWR set. */ 678 SKB_GSO_TCP_ECN = 1 << 2, 679 680 SKB_GSO_TCP_FIXEDID = 1 << 3, 681 682 SKB_GSO_TCPV6 = 1 << 4, 683 684 SKB_GSO_FCOE = 1 << 5, 685 686 SKB_GSO_GRE = 1 << 6, 687 688 SKB_GSO_GRE_CSUM = 1 << 7, 689 690 SKB_GSO_IPXIP4 = 1 << 8, 691 692 SKB_GSO_IPXIP6 = 1 << 9, 693 694 SKB_GSO_UDP_TUNNEL = 1 << 10, 695 696 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 697 698 SKB_GSO_PARTIAL = 1 << 12, 699 700 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 701 702 SKB_GSO_SCTP = 1 << 14, 703 704 SKB_GSO_ESP = 1 << 15, 705 706 SKB_GSO_UDP = 1 << 16, 707 708 SKB_GSO_UDP_L4 = 1 << 17, 709 710 SKB_GSO_FRAGLIST = 1 << 18, 711 712 SKB_GSO_TCP_ACCECN = 1 << 19, 713 }; 714 715 #if BITS_PER_LONG > 32 716 #define NET_SKBUFF_DATA_USES_OFFSET 1 717 #endif 718 719 #ifdef NET_SKBUFF_DATA_USES_OFFSET 720 typedef unsigned int sk_buff_data_t; 721 #else 722 typedef unsigned char *sk_buff_data_t; 723 #endif 724 725 enum skb_tstamp_type { 726 SKB_CLOCK_REALTIME, 727 SKB_CLOCK_MONOTONIC, 728 SKB_CLOCK_TAI, 729 __SKB_CLOCK_MAX = SKB_CLOCK_TAI, 730 }; 731 732 /** 733 * DOC: Basic sk_buff geometry 734 * 735 * struct sk_buff itself is a metadata structure and does not hold any packet 736 * data. All the data is held in associated buffers. 737 * 738 * &sk_buff.head points to the main "head" buffer. The head buffer is divided 739 * into two parts: 740 * 741 * - data buffer, containing headers and sometimes payload; 742 * this is the part of the skb operated on by the common helpers 743 * such as skb_put() or skb_pull(); 744 * - shared info (struct skb_shared_info) which holds an array of pointers 745 * to read-only data in the (page, offset, length) format. 746 * 747 * Optionally &skb_shared_info.frag_list may point to another skb. 748 * 749 * Basic diagram may look like this:: 750 * 751 * --------------- 752 * | sk_buff | 753 * --------------- 754 * ,--------------------------- + head 755 * / ,----------------- + data 756 * / / ,----------- + tail 757 * | | | , + end 758 * | | | | 759 * v v v v 760 * ----------------------------------------------- 761 * | headroom | data | tailroom | skb_shared_info | 762 * ----------------------------------------------- 763 * + [page frag] 764 * + [page frag] 765 * + [page frag] 766 * + [page frag] --------- 767 * + frag_list --> | sk_buff | 768 * --------- 769 * 770 */ 771 772 /** 773 * struct sk_buff - socket buffer 774 * @next: Next buffer in list 775 * @prev: Previous buffer in list 776 * @tstamp: Time we arrived/left 777 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 778 * for retransmit timer 779 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 780 * @list: queue head 781 * @ll_node: anchor in an llist (eg socket defer_list) 782 * @sk: Socket we are owned by 783 * @dev: Device we arrived on/are leaving by 784 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 785 * @cb: Control buffer. Free for use by every layer. Put private vars here 786 * @_skb_refdst: destination entry (with norefcount bit) 787 * @len: Length of actual data 788 * @data_len: Data length 789 * @mac_len: Length of link layer header 790 * @hdr_len: writable header length of cloned skb 791 * @csum: Checksum (must include start/offset pair) 792 * @csum_start: Offset from skb->head where checksumming should start 793 * @csum_offset: Offset from csum_start where checksum should be stored 794 * @priority: Packet queueing priority 795 * @ignore_df: allow local fragmentation 796 * @cloned: Head may be cloned (check refcnt to be sure) 797 * @ip_summed: Driver fed us an IP checksum 798 * @nohdr: Payload reference only, must not modify header 799 * @pkt_type: Packet class 800 * @fclone: skbuff clone status 801 * @ipvs_property: skbuff is owned by ipvs 802 * @inner_protocol_type: whether the inner protocol is 803 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 804 * @remcsum_offload: remote checksum offload is enabled 805 * @offload_fwd_mark: Packet was L2-forwarded in hardware 806 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 807 * @tc_skip_classify: do not classify packet. set by IFB device 808 * @tc_at_ingress: used within tc_classify to distinguish in/egress 809 * @redirected: packet was redirected by packet classifier 810 * @from_ingress: packet was redirected from the ingress path 811 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 812 * @peeked: this packet has been seen already, so stats have been 813 * done for it, don't do them again 814 * @nf_trace: netfilter packet trace flag 815 * @protocol: Packet protocol from driver 816 * @destructor: Destruct function 817 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 818 * @_sk_redir: socket redirection information for skmsg 819 * @_nfct: Associated connection, if any (with nfctinfo bits) 820 * @skb_iif: ifindex of device we arrived on 821 * @tc_index: Traffic control index 822 * @hash: the packet hash 823 * @queue_mapping: Queue mapping for multiqueue devices 824 * @head_frag: skb was allocated from page fragments, 825 * not allocated by kmalloc() or vmalloc(). 826 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 827 * @pp_recycle: mark the packet for recycling instead of freeing (implies 828 * page_pool support on driver) 829 * @active_extensions: active extensions (skb_ext_id types) 830 * @ndisc_nodetype: router type (from link layer) 831 * @ooo_okay: allow the mapping of a socket to a queue to be changed 832 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 833 * ports. 834 * @sw_hash: indicates hash was computed in software stack 835 * @wifi_acked_valid: wifi_acked was set 836 * @wifi_acked: whether frame was acked on wifi or not 837 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 838 * @encapsulation: indicates the inner headers in the skbuff are valid 839 * @encap_hdr_csum: software checksum is needed 840 * @csum_valid: checksum is already valid 841 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 842 * @csum_complete_sw: checksum was completed by software 843 * @csum_level: indicates the number of consecutive checksums found in 844 * the packet minus one that have been verified as 845 * CHECKSUM_UNNECESSARY (max 3) 846 * @unreadable: indicates that at least 1 of the fragments in this skb is 847 * unreadable. 848 * @dst_pending_confirm: need to confirm neighbour 849 * @decrypted: Decrypted SKB 850 * @slow_gro: state present at GRO time, slower prepare step required 851 * @tstamp_type: When set, skb->tstamp has the 852 * delivery_time clock base of skb->tstamp. 853 * @napi_id: id of the NAPI struct this skb came from 854 * @sender_cpu: (aka @napi_id) source CPU in XPS 855 * @alloc_cpu: CPU which did the skb allocation. 856 * @secmark: security marking 857 * @mark: Generic packet mark 858 * @reserved_tailroom: (aka @mark) number of bytes of free space available 859 * at the tail of an sk_buff 860 * @vlan_all: vlan fields (proto & tci) 861 * @vlan_proto: vlan encapsulation protocol 862 * @vlan_tci: vlan tag control information 863 * @inner_protocol: Protocol (encapsulation) 864 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 865 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 866 * @inner_transport_header: Inner transport layer header (encapsulation) 867 * @inner_network_header: Network layer header (encapsulation) 868 * @inner_mac_header: Link layer header (encapsulation) 869 * @transport_header: Transport layer header 870 * @network_header: Network layer header 871 * @mac_header: Link layer header 872 * @kcov_handle: KCOV remote handle for remote coverage collection 873 * @tail: Tail pointer 874 * @end: End pointer 875 * @head: Head of buffer 876 * @data: Data head pointer 877 * @truesize: Buffer size 878 * @users: User count - see {datagram,tcp}.c 879 * @extensions: allocated extensions, valid if active_extensions is nonzero 880 */ 881 882 struct sk_buff { 883 union { 884 struct { 885 /* These two members must be first to match sk_buff_head. */ 886 struct sk_buff *next; 887 struct sk_buff *prev; 888 889 union { 890 struct net_device *dev; 891 /* Some protocols might use this space to store information, 892 * while device pointer would be NULL. 893 * UDP receive path is one user. 894 */ 895 unsigned long dev_scratch; 896 }; 897 }; 898 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 899 struct list_head list; 900 struct llist_node ll_node; 901 }; 902 903 struct sock *sk; 904 905 union { 906 ktime_t tstamp; 907 u64 skb_mstamp_ns; /* earliest departure time */ 908 }; 909 /* 910 * This is the control buffer. It is free to use for every 911 * layer. Please put your private variables there. If you 912 * want to keep them across layers you have to do a skb_clone() 913 * first. This is owned by whoever has the skb queued ATM. 914 */ 915 char cb[48] __aligned(8); 916 917 union { 918 struct { 919 unsigned long _skb_refdst; 920 void (*destructor)(struct sk_buff *skb); 921 }; 922 struct list_head tcp_tsorted_anchor; 923 #ifdef CONFIG_NET_SOCK_MSG 924 unsigned long _sk_redir; 925 #endif 926 }; 927 928 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 929 unsigned long _nfct; 930 #endif 931 unsigned int len, 932 data_len; 933 __u16 mac_len, 934 hdr_len; 935 936 /* Following fields are _not_ copied in __copy_skb_header() 937 * Note that queue_mapping is here mostly to fill a hole. 938 */ 939 __u16 queue_mapping; 940 941 /* if you move cloned around you also must adapt those constants */ 942 #ifdef __BIG_ENDIAN_BITFIELD 943 #define CLONED_MASK (1 << 7) 944 #else 945 #define CLONED_MASK 1 946 #endif 947 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 948 949 /* private: */ 950 __u8 __cloned_offset[0]; 951 /* public: */ 952 __u8 cloned:1, 953 nohdr:1, 954 fclone:2, 955 peeked:1, 956 head_frag:1, 957 pfmemalloc:1, 958 pp_recycle:1; /* page_pool recycle indicator */ 959 #ifdef CONFIG_SKB_EXTENSIONS 960 __u8 active_extensions; 961 #endif 962 963 /* Fields enclosed in headers group are copied 964 * using a single memcpy() in __copy_skb_header() 965 */ 966 struct_group(headers, 967 968 /* private: */ 969 __u8 __pkt_type_offset[0]; 970 /* public: */ 971 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 972 __u8 ignore_df:1; 973 __u8 dst_pending_confirm:1; 974 __u8 ip_summed:2; 975 __u8 ooo_okay:1; 976 977 /* private: */ 978 __u8 __mono_tc_offset[0]; 979 /* public: */ 980 __u8 tstamp_type:2; /* See skb_tstamp_type */ 981 #ifdef CONFIG_NET_XGRESS 982 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 983 __u8 tc_skip_classify:1; 984 #endif 985 __u8 remcsum_offload:1; 986 __u8 csum_complete_sw:1; 987 __u8 csum_level:2; 988 __u8 inner_protocol_type:1; 989 990 __u8 l4_hash:1; 991 __u8 sw_hash:1; 992 #ifdef CONFIG_WIRELESS 993 __u8 wifi_acked_valid:1; 994 __u8 wifi_acked:1; 995 #endif 996 __u8 no_fcs:1; 997 /* Indicates the inner headers are valid in the skbuff. */ 998 __u8 encapsulation:1; 999 __u8 encap_hdr_csum:1; 1000 __u8 csum_valid:1; 1001 #ifdef CONFIG_IPV6_NDISC_NODETYPE 1002 __u8 ndisc_nodetype:2; 1003 #endif 1004 1005 #if IS_ENABLED(CONFIG_IP_VS) 1006 __u8 ipvs_property:1; 1007 #endif 1008 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 1009 __u8 nf_trace:1; 1010 #endif 1011 #ifdef CONFIG_NET_SWITCHDEV 1012 __u8 offload_fwd_mark:1; 1013 __u8 offload_l3_fwd_mark:1; 1014 #endif 1015 __u8 redirected:1; 1016 #ifdef CONFIG_NET_REDIRECT 1017 __u8 from_ingress:1; 1018 #endif 1019 #ifdef CONFIG_NETFILTER_SKIP_EGRESS 1020 __u8 nf_skip_egress:1; 1021 #endif 1022 #ifdef CONFIG_SKB_DECRYPTED 1023 __u8 decrypted:1; 1024 #endif 1025 __u8 slow_gro:1; 1026 #if IS_ENABLED(CONFIG_IP_SCTP) 1027 __u8 csum_not_inet:1; 1028 #endif 1029 __u8 unreadable:1; 1030 #if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS) 1031 __u16 tc_index; /* traffic control index */ 1032 #endif 1033 1034 u16 alloc_cpu; 1035 1036 union { 1037 __wsum csum; 1038 struct { 1039 __u16 csum_start; 1040 __u16 csum_offset; 1041 }; 1042 }; 1043 __u32 priority; 1044 int skb_iif; 1045 __u32 hash; 1046 union { 1047 u32 vlan_all; 1048 struct { 1049 __be16 vlan_proto; 1050 __u16 vlan_tci; 1051 }; 1052 }; 1053 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1054 union { 1055 unsigned int napi_id; 1056 unsigned int sender_cpu; 1057 }; 1058 #endif 1059 #ifdef CONFIG_NETWORK_SECMARK 1060 __u32 secmark; 1061 #endif 1062 1063 union { 1064 __u32 mark; 1065 __u32 reserved_tailroom; 1066 }; 1067 1068 union { 1069 __be16 inner_protocol; 1070 __u8 inner_ipproto; 1071 }; 1072 1073 __u16 inner_transport_header; 1074 __u16 inner_network_header; 1075 __u16 inner_mac_header; 1076 1077 __be16 protocol; 1078 __u16 transport_header; 1079 __u16 network_header; 1080 __u16 mac_header; 1081 1082 #ifdef CONFIG_KCOV 1083 u64 kcov_handle; 1084 #endif 1085 1086 ); /* end headers group */ 1087 1088 /* These elements must be at the end, see alloc_skb() for details. */ 1089 sk_buff_data_t tail; 1090 sk_buff_data_t end; 1091 unsigned char *head, 1092 *data; 1093 unsigned int truesize; 1094 refcount_t users; 1095 1096 #ifdef CONFIG_SKB_EXTENSIONS 1097 /* only usable after checking ->active_extensions != 0 */ 1098 struct skb_ext *extensions; 1099 #endif 1100 }; 1101 1102 /* if you move pkt_type around you also must adapt those constants */ 1103 #ifdef __BIG_ENDIAN_BITFIELD 1104 #define PKT_TYPE_MAX (7 << 5) 1105 #else 1106 #define PKT_TYPE_MAX 7 1107 #endif 1108 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1109 1110 /* if you move tc_at_ingress or tstamp_type 1111 * around, you also must adapt these constants. 1112 */ 1113 #ifdef __BIG_ENDIAN_BITFIELD 1114 #define SKB_TSTAMP_TYPE_MASK (3 << 6) 1115 #define SKB_TSTAMP_TYPE_RSHIFT (6) 1116 #define TC_AT_INGRESS_MASK (1 << 5) 1117 #else 1118 #define SKB_TSTAMP_TYPE_MASK (3) 1119 #define TC_AT_INGRESS_MASK (1 << 2) 1120 #endif 1121 #define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset) 1122 1123 #ifdef __KERNEL__ 1124 /* 1125 * Handling routines are only of interest to the kernel 1126 */ 1127 1128 #define SKB_ALLOC_FCLONE 0x01 1129 #define SKB_ALLOC_RX 0x02 1130 #define SKB_ALLOC_NAPI 0x04 1131 1132 /** 1133 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1134 * @skb: buffer 1135 */ 1136 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1137 { 1138 return unlikely(skb->pfmemalloc); 1139 } 1140 1141 /* 1142 * skb might have a dst pointer attached, refcounted or not. 1143 * _skb_refdst low order bit is set if refcount was _not_ taken 1144 */ 1145 #define SKB_DST_NOREF 1UL 1146 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1147 1148 /** 1149 * skb_dst - returns skb dst_entry 1150 * @skb: buffer 1151 * 1152 * Returns: skb dst_entry, regardless of reference taken or not. 1153 */ 1154 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1155 { 1156 /* If refdst was not refcounted, check we still are in a 1157 * rcu_read_lock section 1158 */ 1159 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1160 !rcu_read_lock_held() && 1161 !rcu_read_lock_bh_held()); 1162 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1163 } 1164 1165 /** 1166 * skb_dst_set - sets skb dst 1167 * @skb: buffer 1168 * @dst: dst entry 1169 * 1170 * Sets skb dst, assuming a reference was taken on dst and should 1171 * be released by skb_dst_drop() 1172 */ 1173 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1174 { 1175 skb->slow_gro |= !!dst; 1176 skb->_skb_refdst = (unsigned long)dst; 1177 } 1178 1179 /** 1180 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1181 * @skb: buffer 1182 * @dst: dst entry 1183 * 1184 * Sets skb dst, assuming a reference was not taken on dst. 1185 * If dst entry is cached, we do not take reference and dst_release 1186 * will be avoided by refdst_drop. If dst entry is not cached, we take 1187 * reference, so that last dst_release can destroy the dst immediately. 1188 */ 1189 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1190 { 1191 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1192 skb->slow_gro |= !!dst; 1193 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1194 } 1195 1196 /** 1197 * skb_dst_is_noref - Test if skb dst isn't refcounted 1198 * @skb: buffer 1199 */ 1200 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1201 { 1202 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1203 } 1204 1205 /* For mangling skb->pkt_type from user space side from applications 1206 * such as nft, tc, etc, we only allow a conservative subset of 1207 * possible pkt_types to be set. 1208 */ 1209 static inline bool skb_pkt_type_ok(u32 ptype) 1210 { 1211 return ptype <= PACKET_OTHERHOST; 1212 } 1213 1214 /** 1215 * skb_napi_id - Returns the skb's NAPI id 1216 * @skb: buffer 1217 */ 1218 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1219 { 1220 #ifdef CONFIG_NET_RX_BUSY_POLL 1221 return skb->napi_id; 1222 #else 1223 return 0; 1224 #endif 1225 } 1226 1227 static inline bool skb_wifi_acked_valid(const struct sk_buff *skb) 1228 { 1229 #ifdef CONFIG_WIRELESS 1230 return skb->wifi_acked_valid; 1231 #else 1232 return 0; 1233 #endif 1234 } 1235 1236 /** 1237 * skb_unref - decrement the skb's reference count 1238 * @skb: buffer 1239 * 1240 * Returns: true if we can free the skb. 1241 */ 1242 static inline bool skb_unref(struct sk_buff *skb) 1243 { 1244 if (unlikely(!skb)) 1245 return false; 1246 if (!IS_ENABLED(CONFIG_DEBUG_NET) && likely(refcount_read(&skb->users) == 1)) 1247 smp_rmb(); 1248 else if (likely(!refcount_dec_and_test(&skb->users))) 1249 return false; 1250 1251 return true; 1252 } 1253 1254 static inline bool skb_data_unref(const struct sk_buff *skb, 1255 struct skb_shared_info *shinfo) 1256 { 1257 int bias; 1258 1259 if (!skb->cloned) 1260 return true; 1261 1262 bias = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1; 1263 1264 if (atomic_read(&shinfo->dataref) == bias) 1265 smp_rmb(); 1266 else if (atomic_sub_return(bias, &shinfo->dataref)) 1267 return false; 1268 1269 return true; 1270 } 1271 1272 void __fix_address sk_skb_reason_drop(struct sock *sk, struct sk_buff *skb, 1273 enum skb_drop_reason reason); 1274 1275 static inline void 1276 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason) 1277 { 1278 sk_skb_reason_drop(NULL, skb, reason); 1279 } 1280 1281 /** 1282 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1283 * @skb: buffer to free 1284 */ 1285 static inline void kfree_skb(struct sk_buff *skb) 1286 { 1287 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1288 } 1289 1290 void skb_release_head_state(struct sk_buff *skb); 1291 void kfree_skb_list_reason(struct sk_buff *segs, 1292 enum skb_drop_reason reason); 1293 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1294 void skb_tx_error(struct sk_buff *skb); 1295 1296 static inline void kfree_skb_list(struct sk_buff *segs) 1297 { 1298 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1299 } 1300 1301 #ifdef CONFIG_TRACEPOINTS 1302 void consume_skb(struct sk_buff *skb); 1303 #else 1304 static inline void consume_skb(struct sk_buff *skb) 1305 { 1306 return kfree_skb(skb); 1307 } 1308 #endif 1309 1310 void __consume_stateless_skb(struct sk_buff *skb); 1311 void __kfree_skb(struct sk_buff *skb); 1312 1313 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1314 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1315 bool *fragstolen, int *delta_truesize); 1316 1317 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1318 int node); 1319 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1320 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1321 struct sk_buff *build_skb_around(struct sk_buff *skb, 1322 void *data, unsigned int frag_size); 1323 void skb_attempt_defer_free(struct sk_buff *skb); 1324 1325 u32 napi_skb_cache_get_bulk(void **skbs, u32 n); 1326 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1327 struct sk_buff *slab_build_skb(void *data); 1328 1329 /** 1330 * alloc_skb - allocate a network buffer 1331 * @size: size to allocate 1332 * @priority: allocation mask 1333 * 1334 * This function is a convenient wrapper around __alloc_skb(). 1335 */ 1336 static inline struct sk_buff *alloc_skb(unsigned int size, 1337 gfp_t priority) 1338 { 1339 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1340 } 1341 1342 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1343 unsigned long data_len, 1344 int max_page_order, 1345 int *errcode, 1346 gfp_t gfp_mask); 1347 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1348 1349 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1350 struct sk_buff_fclones { 1351 struct sk_buff skb1; 1352 1353 struct sk_buff skb2; 1354 1355 refcount_t fclone_ref; 1356 }; 1357 1358 /** 1359 * skb_fclone_busy - check if fclone is busy 1360 * @sk: socket 1361 * @skb: buffer 1362 * 1363 * Returns: true if skb is a fast clone, and its clone is not freed. 1364 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1365 * so we also check that didn't happen. 1366 */ 1367 static inline bool skb_fclone_busy(const struct sock *sk, 1368 const struct sk_buff *skb) 1369 { 1370 const struct sk_buff_fclones *fclones; 1371 1372 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1373 1374 return skb->fclone == SKB_FCLONE_ORIG && 1375 refcount_read(&fclones->fclone_ref) > 1 && 1376 READ_ONCE(fclones->skb2.sk) == sk; 1377 } 1378 1379 /** 1380 * alloc_skb_fclone - allocate a network buffer from fclone cache 1381 * @size: size to allocate 1382 * @priority: allocation mask 1383 * 1384 * This function is a convenient wrapper around __alloc_skb(). 1385 */ 1386 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1387 gfp_t priority) 1388 { 1389 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1390 } 1391 1392 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1393 void skb_headers_offset_update(struct sk_buff *skb, int off); 1394 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1395 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1396 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1397 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1398 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1399 gfp_t gfp_mask, bool fclone); 1400 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1401 gfp_t gfp_mask) 1402 { 1403 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1404 } 1405 1406 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1407 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1408 unsigned int headroom); 1409 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1410 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1411 int newtailroom, gfp_t priority); 1412 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1413 int offset, int len); 1414 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1415 int offset, int len); 1416 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1417 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1418 1419 /** 1420 * skb_pad - zero pad the tail of an skb 1421 * @skb: buffer to pad 1422 * @pad: space to pad 1423 * 1424 * Ensure that a buffer is followed by a padding area that is zero 1425 * filled. Used by network drivers which may DMA or transfer data 1426 * beyond the buffer end onto the wire. 1427 * 1428 * May return error in out of memory cases. The skb is freed on error. 1429 */ 1430 static inline int skb_pad(struct sk_buff *skb, int pad) 1431 { 1432 return __skb_pad(skb, pad, true); 1433 } 1434 #define dev_kfree_skb(a) consume_skb(a) 1435 1436 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1437 int offset, size_t size, size_t max_frags); 1438 1439 struct skb_seq_state { 1440 __u32 lower_offset; 1441 __u32 upper_offset; 1442 __u32 frag_idx; 1443 __u32 stepped_offset; 1444 struct sk_buff *root_skb; 1445 struct sk_buff *cur_skb; 1446 __u8 *frag_data; 1447 __u32 frag_off; 1448 }; 1449 1450 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1451 unsigned int to, struct skb_seq_state *st); 1452 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1453 struct skb_seq_state *st); 1454 void skb_abort_seq_read(struct skb_seq_state *st); 1455 int skb_copy_seq_read(struct skb_seq_state *st, int offset, void *to, int len); 1456 1457 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1458 unsigned int to, struct ts_config *config); 1459 1460 /* 1461 * Packet hash types specify the type of hash in skb_set_hash. 1462 * 1463 * Hash types refer to the protocol layer addresses which are used to 1464 * construct a packet's hash. The hashes are used to differentiate or identify 1465 * flows of the protocol layer for the hash type. Hash types are either 1466 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1467 * 1468 * Properties of hashes: 1469 * 1470 * 1) Two packets in different flows have different hash values 1471 * 2) Two packets in the same flow should have the same hash value 1472 * 1473 * A hash at a higher layer is considered to be more specific. A driver should 1474 * set the most specific hash possible. 1475 * 1476 * A driver cannot indicate a more specific hash than the layer at which a hash 1477 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1478 * 1479 * A driver may indicate a hash level which is less specific than the 1480 * actual layer the hash was computed on. For instance, a hash computed 1481 * at L4 may be considered an L3 hash. This should only be done if the 1482 * driver can't unambiguously determine that the HW computed the hash at 1483 * the higher layer. Note that the "should" in the second property above 1484 * permits this. 1485 */ 1486 enum pkt_hash_types { 1487 PKT_HASH_TYPE_NONE, /* Undefined type */ 1488 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1489 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1490 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1491 }; 1492 1493 static inline void skb_clear_hash(struct sk_buff *skb) 1494 { 1495 skb->hash = 0; 1496 skb->sw_hash = 0; 1497 skb->l4_hash = 0; 1498 } 1499 1500 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1501 { 1502 if (!skb->l4_hash) 1503 skb_clear_hash(skb); 1504 } 1505 1506 static inline void 1507 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1508 { 1509 skb->l4_hash = is_l4; 1510 skb->sw_hash = is_sw; 1511 skb->hash = hash; 1512 } 1513 1514 static inline void 1515 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1516 { 1517 /* Used by drivers to set hash from HW */ 1518 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1519 } 1520 1521 static inline void 1522 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1523 { 1524 __skb_set_hash(skb, hash, true, is_l4); 1525 } 1526 1527 u32 __skb_get_hash_symmetric_net(const struct net *net, const struct sk_buff *skb); 1528 1529 static inline u32 __skb_get_hash_symmetric(const struct sk_buff *skb) 1530 { 1531 return __skb_get_hash_symmetric_net(NULL, skb); 1532 } 1533 1534 void __skb_get_hash_net(const struct net *net, struct sk_buff *skb); 1535 u32 skb_get_poff(const struct sk_buff *skb); 1536 u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1537 const struct flow_keys_basic *keys, int hlen); 1538 __be32 skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1539 const void *data, int hlen_proto); 1540 1541 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1542 const struct flow_dissector_key *key, 1543 unsigned int key_count); 1544 1545 struct bpf_flow_dissector; 1546 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1547 __be16 proto, int nhoff, int hlen, unsigned int flags); 1548 1549 bool __skb_flow_dissect(const struct net *net, 1550 const struct sk_buff *skb, 1551 struct flow_dissector *flow_dissector, 1552 void *target_container, const void *data, 1553 __be16 proto, int nhoff, int hlen, unsigned int flags); 1554 1555 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1556 struct flow_dissector *flow_dissector, 1557 void *target_container, unsigned int flags) 1558 { 1559 return __skb_flow_dissect(NULL, skb, flow_dissector, 1560 target_container, NULL, 0, 0, 0, flags); 1561 } 1562 1563 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1564 struct flow_keys *flow, 1565 unsigned int flags) 1566 { 1567 memset(flow, 0, sizeof(*flow)); 1568 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1569 flow, NULL, 0, 0, 0, flags); 1570 } 1571 1572 static inline bool 1573 skb_flow_dissect_flow_keys_basic(const struct net *net, 1574 const struct sk_buff *skb, 1575 struct flow_keys_basic *flow, 1576 const void *data, __be16 proto, 1577 int nhoff, int hlen, unsigned int flags) 1578 { 1579 memset(flow, 0, sizeof(*flow)); 1580 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1581 data, proto, nhoff, hlen, flags); 1582 } 1583 1584 void skb_flow_dissect_meta(const struct sk_buff *skb, 1585 struct flow_dissector *flow_dissector, 1586 void *target_container); 1587 1588 /* Gets a skb connection tracking info, ctinfo map should be a 1589 * map of mapsize to translate enum ip_conntrack_info states 1590 * to user states. 1591 */ 1592 void 1593 skb_flow_dissect_ct(const struct sk_buff *skb, 1594 struct flow_dissector *flow_dissector, 1595 void *target_container, 1596 u16 *ctinfo_map, size_t mapsize, 1597 bool post_ct, u16 zone); 1598 void 1599 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1600 struct flow_dissector *flow_dissector, 1601 void *target_container); 1602 1603 void skb_flow_dissect_hash(const struct sk_buff *skb, 1604 struct flow_dissector *flow_dissector, 1605 void *target_container); 1606 1607 static inline __u32 skb_get_hash_net(const struct net *net, struct sk_buff *skb) 1608 { 1609 if (!skb->l4_hash && !skb->sw_hash) 1610 __skb_get_hash_net(net, skb); 1611 1612 return skb->hash; 1613 } 1614 1615 static inline __u32 skb_get_hash(struct sk_buff *skb) 1616 { 1617 if (!skb->l4_hash && !skb->sw_hash) 1618 __skb_get_hash_net(NULL, skb); 1619 1620 return skb->hash; 1621 } 1622 1623 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1624 { 1625 if (!skb->l4_hash && !skb->sw_hash) { 1626 struct flow_keys keys; 1627 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1628 1629 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1630 } 1631 1632 return skb->hash; 1633 } 1634 1635 __u32 skb_get_hash_perturb(const struct sk_buff *skb, 1636 const siphash_key_t *perturb); 1637 1638 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1639 { 1640 return skb->hash; 1641 } 1642 1643 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1644 { 1645 to->hash = from->hash; 1646 to->sw_hash = from->sw_hash; 1647 to->l4_hash = from->l4_hash; 1648 }; 1649 1650 static inline int skb_cmp_decrypted(const struct sk_buff *skb1, 1651 const struct sk_buff *skb2) 1652 { 1653 #ifdef CONFIG_SKB_DECRYPTED 1654 return skb2->decrypted - skb1->decrypted; 1655 #else 1656 return 0; 1657 #endif 1658 } 1659 1660 static inline bool skb_is_decrypted(const struct sk_buff *skb) 1661 { 1662 #ifdef CONFIG_SKB_DECRYPTED 1663 return skb->decrypted; 1664 #else 1665 return false; 1666 #endif 1667 } 1668 1669 static inline void skb_copy_decrypted(struct sk_buff *to, 1670 const struct sk_buff *from) 1671 { 1672 #ifdef CONFIG_SKB_DECRYPTED 1673 to->decrypted = from->decrypted; 1674 #endif 1675 } 1676 1677 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1678 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1679 { 1680 return skb->head + skb->end; 1681 } 1682 1683 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1684 { 1685 return skb->end; 1686 } 1687 1688 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1689 { 1690 skb->end = offset; 1691 } 1692 #else 1693 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1694 { 1695 return skb->end; 1696 } 1697 1698 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1699 { 1700 return skb->end - skb->head; 1701 } 1702 1703 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1704 { 1705 skb->end = skb->head + offset; 1706 } 1707 #endif 1708 1709 extern const struct ubuf_info_ops msg_zerocopy_ubuf_ops; 1710 1711 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1712 struct ubuf_info *uarg); 1713 1714 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1715 1716 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, 1717 struct sk_buff *skb, struct iov_iter *from, 1718 size_t length); 1719 1720 int zerocopy_fill_skb_from_iter(struct sk_buff *skb, 1721 struct iov_iter *from, size_t length); 1722 1723 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1724 struct msghdr *msg, int len) 1725 { 1726 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len); 1727 } 1728 1729 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1730 struct msghdr *msg, int len, 1731 struct ubuf_info *uarg); 1732 1733 /* Internal */ 1734 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1735 1736 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1737 { 1738 return &skb_shinfo(skb)->hwtstamps; 1739 } 1740 1741 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1742 { 1743 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1744 1745 return is_zcopy ? skb_uarg(skb) : NULL; 1746 } 1747 1748 static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1749 { 1750 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1751 } 1752 1753 static inline bool skb_zcopy_managed(const struct sk_buff *skb) 1754 { 1755 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS; 1756 } 1757 1758 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1759 const struct sk_buff *skb2) 1760 { 1761 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1762 } 1763 1764 static inline void net_zcopy_get(struct ubuf_info *uarg) 1765 { 1766 refcount_inc(&uarg->refcnt); 1767 } 1768 1769 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1770 { 1771 skb_shinfo(skb)->destructor_arg = uarg; 1772 skb_shinfo(skb)->flags |= uarg->flags; 1773 } 1774 1775 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1776 bool *have_ref) 1777 { 1778 if (skb && uarg && !skb_zcopy(skb)) { 1779 if (unlikely(have_ref && *have_ref)) 1780 *have_ref = false; 1781 else 1782 net_zcopy_get(uarg); 1783 skb_zcopy_init(skb, uarg); 1784 } 1785 } 1786 1787 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1788 { 1789 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1790 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1791 } 1792 1793 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1794 { 1795 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1796 } 1797 1798 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1799 { 1800 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1801 } 1802 1803 static inline void net_zcopy_put(struct ubuf_info *uarg) 1804 { 1805 if (uarg) 1806 uarg->ops->complete(NULL, uarg, true); 1807 } 1808 1809 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1810 { 1811 if (uarg) { 1812 if (uarg->ops == &msg_zerocopy_ubuf_ops) 1813 msg_zerocopy_put_abort(uarg, have_uref); 1814 else if (have_uref) 1815 net_zcopy_put(uarg); 1816 } 1817 } 1818 1819 /* Release a reference on a zerocopy structure */ 1820 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1821 { 1822 struct ubuf_info *uarg = skb_zcopy(skb); 1823 1824 if (uarg) { 1825 if (!skb_zcopy_is_nouarg(skb)) 1826 uarg->ops->complete(skb, uarg, zerocopy_success); 1827 1828 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1829 } 1830 } 1831 1832 void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1833 1834 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1835 { 1836 if (unlikely(skb_zcopy_managed(skb))) 1837 __skb_zcopy_downgrade_managed(skb); 1838 } 1839 1840 /* Return true if frags in this skb are readable by the host. */ 1841 static inline bool skb_frags_readable(const struct sk_buff *skb) 1842 { 1843 return !skb->unreadable; 1844 } 1845 1846 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1847 { 1848 skb->next = NULL; 1849 } 1850 1851 static inline void skb_poison_list(struct sk_buff *skb) 1852 { 1853 #ifdef CONFIG_DEBUG_NET 1854 skb->next = SKB_LIST_POISON_NEXT; 1855 #endif 1856 } 1857 1858 /* Iterate through singly-linked GSO fragments of an skb. */ 1859 #define skb_list_walk_safe(first, skb, next_skb) \ 1860 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1861 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1862 1863 static inline void skb_list_del_init(struct sk_buff *skb) 1864 { 1865 __list_del_entry(&skb->list); 1866 skb_mark_not_on_list(skb); 1867 } 1868 1869 /** 1870 * skb_queue_empty - check if a queue is empty 1871 * @list: queue head 1872 * 1873 * Returns true if the queue is empty, false otherwise. 1874 */ 1875 static inline int skb_queue_empty(const struct sk_buff_head *list) 1876 { 1877 return list->next == (const struct sk_buff *) list; 1878 } 1879 1880 /** 1881 * skb_queue_empty_lockless - check if a queue is empty 1882 * @list: queue head 1883 * 1884 * Returns true if the queue is empty, false otherwise. 1885 * This variant can be used in lockless contexts. 1886 */ 1887 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1888 { 1889 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1890 } 1891 1892 1893 /** 1894 * skb_queue_is_last - check if skb is the last entry in the queue 1895 * @list: queue head 1896 * @skb: buffer 1897 * 1898 * Returns true if @skb is the last buffer on the list. 1899 */ 1900 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1901 const struct sk_buff *skb) 1902 { 1903 return skb->next == (const struct sk_buff *) list; 1904 } 1905 1906 /** 1907 * skb_queue_is_first - check if skb is the first entry in the queue 1908 * @list: queue head 1909 * @skb: buffer 1910 * 1911 * Returns true if @skb is the first buffer on the list. 1912 */ 1913 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1914 const struct sk_buff *skb) 1915 { 1916 return skb->prev == (const struct sk_buff *) list; 1917 } 1918 1919 /** 1920 * skb_queue_next - return the next packet in the queue 1921 * @list: queue head 1922 * @skb: current buffer 1923 * 1924 * Return the next packet in @list after @skb. It is only valid to 1925 * call this if skb_queue_is_last() evaluates to false. 1926 */ 1927 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1928 const struct sk_buff *skb) 1929 { 1930 /* This BUG_ON may seem severe, but if we just return then we 1931 * are going to dereference garbage. 1932 */ 1933 BUG_ON(skb_queue_is_last(list, skb)); 1934 return skb->next; 1935 } 1936 1937 /** 1938 * skb_queue_prev - return the prev packet in the queue 1939 * @list: queue head 1940 * @skb: current buffer 1941 * 1942 * Return the prev packet in @list before @skb. It is only valid to 1943 * call this if skb_queue_is_first() evaluates to false. 1944 */ 1945 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1946 const struct sk_buff *skb) 1947 { 1948 /* This BUG_ON may seem severe, but if we just return then we 1949 * are going to dereference garbage. 1950 */ 1951 BUG_ON(skb_queue_is_first(list, skb)); 1952 return skb->prev; 1953 } 1954 1955 /** 1956 * skb_get - reference buffer 1957 * @skb: buffer to reference 1958 * 1959 * Makes another reference to a socket buffer and returns a pointer 1960 * to the buffer. 1961 */ 1962 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1963 { 1964 refcount_inc(&skb->users); 1965 return skb; 1966 } 1967 1968 /* 1969 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1970 */ 1971 1972 /** 1973 * skb_cloned - is the buffer a clone 1974 * @skb: buffer to check 1975 * 1976 * Returns true if the buffer was generated with skb_clone() and is 1977 * one of multiple shared copies of the buffer. Cloned buffers are 1978 * shared data so must not be written to under normal circumstances. 1979 */ 1980 static inline int skb_cloned(const struct sk_buff *skb) 1981 { 1982 return skb->cloned && 1983 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1984 } 1985 1986 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1987 { 1988 might_sleep_if(gfpflags_allow_blocking(pri)); 1989 1990 if (skb_cloned(skb)) 1991 return pskb_expand_head(skb, 0, 0, pri); 1992 1993 return 0; 1994 } 1995 1996 /* This variant of skb_unclone() makes sure skb->truesize 1997 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 1998 * 1999 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 2000 * when various debugging features are in place. 2001 */ 2002 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 2003 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 2004 { 2005 might_sleep_if(gfpflags_allow_blocking(pri)); 2006 2007 if (skb_cloned(skb)) 2008 return __skb_unclone_keeptruesize(skb, pri); 2009 return 0; 2010 } 2011 2012 /** 2013 * skb_header_cloned - is the header a clone 2014 * @skb: buffer to check 2015 * 2016 * Returns true if modifying the header part of the buffer requires 2017 * the data to be copied. 2018 */ 2019 static inline int skb_header_cloned(const struct sk_buff *skb) 2020 { 2021 int dataref; 2022 2023 if (!skb->cloned) 2024 return 0; 2025 2026 dataref = atomic_read(&skb_shinfo(skb)->dataref); 2027 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 2028 return dataref != 1; 2029 } 2030 2031 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 2032 { 2033 might_sleep_if(gfpflags_allow_blocking(pri)); 2034 2035 if (skb_header_cloned(skb)) 2036 return pskb_expand_head(skb, 0, 0, pri); 2037 2038 return 0; 2039 } 2040 2041 /** 2042 * __skb_header_release() - allow clones to use the headroom 2043 * @skb: buffer to operate on 2044 * 2045 * See "DOC: dataref and headerless skbs". 2046 */ 2047 static inline void __skb_header_release(struct sk_buff *skb) 2048 { 2049 skb->nohdr = 1; 2050 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 2051 } 2052 2053 2054 /** 2055 * skb_shared - is the buffer shared 2056 * @skb: buffer to check 2057 * 2058 * Returns true if more than one person has a reference to this 2059 * buffer. 2060 */ 2061 static inline int skb_shared(const struct sk_buff *skb) 2062 { 2063 return refcount_read(&skb->users) != 1; 2064 } 2065 2066 /** 2067 * skb_share_check - check if buffer is shared and if so clone it 2068 * @skb: buffer to check 2069 * @pri: priority for memory allocation 2070 * 2071 * If the buffer is shared the buffer is cloned and the old copy 2072 * drops a reference. A new clone with a single reference is returned. 2073 * If the buffer is not shared the original buffer is returned. When 2074 * being called from interrupt status or with spinlocks held pri must 2075 * be GFP_ATOMIC. 2076 * 2077 * NULL is returned on a memory allocation failure. 2078 */ 2079 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 2080 { 2081 might_sleep_if(gfpflags_allow_blocking(pri)); 2082 if (skb_shared(skb)) { 2083 struct sk_buff *nskb = skb_clone(skb, pri); 2084 2085 if (likely(nskb)) 2086 consume_skb(skb); 2087 else 2088 kfree_skb(skb); 2089 skb = nskb; 2090 } 2091 return skb; 2092 } 2093 2094 /* 2095 * Copy shared buffers into a new sk_buff. We effectively do COW on 2096 * packets to handle cases where we have a local reader and forward 2097 * and a couple of other messy ones. The normal one is tcpdumping 2098 * a packet that's being forwarded. 2099 */ 2100 2101 /** 2102 * skb_unshare - make a copy of a shared buffer 2103 * @skb: buffer to check 2104 * @pri: priority for memory allocation 2105 * 2106 * If the socket buffer is a clone then this function creates a new 2107 * copy of the data, drops a reference count on the old copy and returns 2108 * the new copy with the reference count at 1. If the buffer is not a clone 2109 * the original buffer is returned. When called with a spinlock held or 2110 * from interrupt state @pri must be %GFP_ATOMIC 2111 * 2112 * %NULL is returned on a memory allocation failure. 2113 */ 2114 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2115 gfp_t pri) 2116 { 2117 might_sleep_if(gfpflags_allow_blocking(pri)); 2118 if (skb_cloned(skb)) { 2119 struct sk_buff *nskb = skb_copy(skb, pri); 2120 2121 /* Free our shared copy */ 2122 if (likely(nskb)) 2123 consume_skb(skb); 2124 else 2125 kfree_skb(skb); 2126 skb = nskb; 2127 } 2128 return skb; 2129 } 2130 2131 /** 2132 * skb_peek - peek at the head of an &sk_buff_head 2133 * @list_: list to peek at 2134 * 2135 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2136 * be careful with this one. A peek leaves the buffer on the 2137 * list and someone else may run off with it. You must hold 2138 * the appropriate locks or have a private queue to do this. 2139 * 2140 * Returns %NULL for an empty list or a pointer to the head element. 2141 * The reference count is not incremented and the reference is therefore 2142 * volatile. Use with caution. 2143 */ 2144 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2145 { 2146 struct sk_buff *skb = list_->next; 2147 2148 if (skb == (struct sk_buff *)list_) 2149 skb = NULL; 2150 return skb; 2151 } 2152 2153 /** 2154 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2155 * @list_: list to peek at 2156 * 2157 * Like skb_peek(), but the caller knows that the list is not empty. 2158 */ 2159 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2160 { 2161 return list_->next; 2162 } 2163 2164 /** 2165 * skb_peek_next - peek skb following the given one from a queue 2166 * @skb: skb to start from 2167 * @list_: list to peek at 2168 * 2169 * Returns %NULL when the end of the list is met or a pointer to the 2170 * next element. The reference count is not incremented and the 2171 * reference is therefore volatile. Use with caution. 2172 */ 2173 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2174 const struct sk_buff_head *list_) 2175 { 2176 struct sk_buff *next = skb->next; 2177 2178 if (next == (struct sk_buff *)list_) 2179 next = NULL; 2180 return next; 2181 } 2182 2183 /** 2184 * skb_peek_tail - peek at the tail of an &sk_buff_head 2185 * @list_: list to peek at 2186 * 2187 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2188 * be careful with this one. A peek leaves the buffer on the 2189 * list and someone else may run off with it. You must hold 2190 * the appropriate locks or have a private queue to do this. 2191 * 2192 * Returns %NULL for an empty list or a pointer to the tail element. 2193 * The reference count is not incremented and the reference is therefore 2194 * volatile. Use with caution. 2195 */ 2196 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2197 { 2198 struct sk_buff *skb = READ_ONCE(list_->prev); 2199 2200 if (skb == (struct sk_buff *)list_) 2201 skb = NULL; 2202 return skb; 2203 2204 } 2205 2206 /** 2207 * skb_queue_len - get queue length 2208 * @list_: list to measure 2209 * 2210 * Return the length of an &sk_buff queue. 2211 */ 2212 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2213 { 2214 return list_->qlen; 2215 } 2216 2217 /** 2218 * skb_queue_len_lockless - get queue length 2219 * @list_: list to measure 2220 * 2221 * Return the length of an &sk_buff queue. 2222 * This variant can be used in lockless contexts. 2223 */ 2224 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2225 { 2226 return READ_ONCE(list_->qlen); 2227 } 2228 2229 /** 2230 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2231 * @list: queue to initialize 2232 * 2233 * This initializes only the list and queue length aspects of 2234 * an sk_buff_head object. This allows to initialize the list 2235 * aspects of an sk_buff_head without reinitializing things like 2236 * the spinlock. It can also be used for on-stack sk_buff_head 2237 * objects where the spinlock is known to not be used. 2238 */ 2239 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2240 { 2241 list->prev = list->next = (struct sk_buff *)list; 2242 list->qlen = 0; 2243 } 2244 2245 /* 2246 * This function creates a split out lock class for each invocation; 2247 * this is needed for now since a whole lot of users of the skb-queue 2248 * infrastructure in drivers have different locking usage (in hardirq) 2249 * than the networking core (in softirq only). In the long run either the 2250 * network layer or drivers should need annotation to consolidate the 2251 * main types of usage into 3 classes. 2252 */ 2253 static inline void skb_queue_head_init(struct sk_buff_head *list) 2254 { 2255 spin_lock_init(&list->lock); 2256 __skb_queue_head_init(list); 2257 } 2258 2259 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2260 struct lock_class_key *class) 2261 { 2262 skb_queue_head_init(list); 2263 lockdep_set_class(&list->lock, class); 2264 } 2265 2266 /* 2267 * Insert an sk_buff on a list. 2268 * 2269 * The "__skb_xxxx()" functions are the non-atomic ones that 2270 * can only be called with interrupts disabled. 2271 */ 2272 static inline void __skb_insert(struct sk_buff *newsk, 2273 struct sk_buff *prev, struct sk_buff *next, 2274 struct sk_buff_head *list) 2275 { 2276 /* See skb_queue_empty_lockless() and skb_peek_tail() 2277 * for the opposite READ_ONCE() 2278 */ 2279 WRITE_ONCE(newsk->next, next); 2280 WRITE_ONCE(newsk->prev, prev); 2281 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2282 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2283 WRITE_ONCE(list->qlen, list->qlen + 1); 2284 } 2285 2286 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2287 struct sk_buff *prev, 2288 struct sk_buff *next) 2289 { 2290 struct sk_buff *first = list->next; 2291 struct sk_buff *last = list->prev; 2292 2293 WRITE_ONCE(first->prev, prev); 2294 WRITE_ONCE(prev->next, first); 2295 2296 WRITE_ONCE(last->next, next); 2297 WRITE_ONCE(next->prev, last); 2298 } 2299 2300 /** 2301 * skb_queue_splice - join two skb lists, this is designed for stacks 2302 * @list: the new list to add 2303 * @head: the place to add it in the first list 2304 */ 2305 static inline void skb_queue_splice(const struct sk_buff_head *list, 2306 struct sk_buff_head *head) 2307 { 2308 if (!skb_queue_empty(list)) { 2309 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2310 head->qlen += list->qlen; 2311 } 2312 } 2313 2314 /** 2315 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2316 * @list: the new list to add 2317 * @head: the place to add it in the first list 2318 * 2319 * The list at @list is reinitialised 2320 */ 2321 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2322 struct sk_buff_head *head) 2323 { 2324 if (!skb_queue_empty(list)) { 2325 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2326 head->qlen += list->qlen; 2327 __skb_queue_head_init(list); 2328 } 2329 } 2330 2331 /** 2332 * skb_queue_splice_tail - join two skb lists, each list being a queue 2333 * @list: the new list to add 2334 * @head: the place to add it in the first list 2335 */ 2336 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2337 struct sk_buff_head *head) 2338 { 2339 if (!skb_queue_empty(list)) { 2340 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2341 head->qlen += list->qlen; 2342 } 2343 } 2344 2345 /** 2346 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2347 * @list: the new list to add 2348 * @head: the place to add it in the first list 2349 * 2350 * Each of the lists is a queue. 2351 * The list at @list is reinitialised 2352 */ 2353 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2354 struct sk_buff_head *head) 2355 { 2356 if (!skb_queue_empty(list)) { 2357 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2358 head->qlen += list->qlen; 2359 __skb_queue_head_init(list); 2360 } 2361 } 2362 2363 /** 2364 * __skb_queue_after - queue a buffer at the list head 2365 * @list: list to use 2366 * @prev: place after this buffer 2367 * @newsk: buffer to queue 2368 * 2369 * Queue a buffer int the middle of a list. This function takes no locks 2370 * and you must therefore hold required locks before calling it. 2371 * 2372 * A buffer cannot be placed on two lists at the same time. 2373 */ 2374 static inline void __skb_queue_after(struct sk_buff_head *list, 2375 struct sk_buff *prev, 2376 struct sk_buff *newsk) 2377 { 2378 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2379 } 2380 2381 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2382 struct sk_buff_head *list); 2383 2384 static inline void __skb_queue_before(struct sk_buff_head *list, 2385 struct sk_buff *next, 2386 struct sk_buff *newsk) 2387 { 2388 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2389 } 2390 2391 /** 2392 * __skb_queue_head - queue a buffer at the list head 2393 * @list: list to use 2394 * @newsk: buffer to queue 2395 * 2396 * Queue a buffer at the start of a list. This function takes no locks 2397 * and you must therefore hold required locks before calling it. 2398 * 2399 * A buffer cannot be placed on two lists at the same time. 2400 */ 2401 static inline void __skb_queue_head(struct sk_buff_head *list, 2402 struct sk_buff *newsk) 2403 { 2404 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2405 } 2406 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2407 2408 /** 2409 * __skb_queue_tail - queue a buffer at the list tail 2410 * @list: list to use 2411 * @newsk: buffer to queue 2412 * 2413 * Queue a buffer at the end of a list. This function takes no locks 2414 * and you must therefore hold required locks before calling it. 2415 * 2416 * A buffer cannot be placed on two lists at the same time. 2417 */ 2418 static inline void __skb_queue_tail(struct sk_buff_head *list, 2419 struct sk_buff *newsk) 2420 { 2421 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2422 } 2423 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2424 2425 /* 2426 * remove sk_buff from list. _Must_ be called atomically, and with 2427 * the list known.. 2428 */ 2429 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2430 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2431 { 2432 struct sk_buff *next, *prev; 2433 2434 WRITE_ONCE(list->qlen, list->qlen - 1); 2435 next = skb->next; 2436 prev = skb->prev; 2437 skb->next = skb->prev = NULL; 2438 WRITE_ONCE(next->prev, prev); 2439 WRITE_ONCE(prev->next, next); 2440 } 2441 2442 /** 2443 * __skb_dequeue - remove from the head of the queue 2444 * @list: list to dequeue from 2445 * 2446 * Remove the head of the list. This function does not take any locks 2447 * so must be used with appropriate locks held only. The head item is 2448 * returned or %NULL if the list is empty. 2449 */ 2450 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2451 { 2452 struct sk_buff *skb = skb_peek(list); 2453 if (skb) 2454 __skb_unlink(skb, list); 2455 return skb; 2456 } 2457 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2458 2459 /** 2460 * __skb_dequeue_tail - remove from the tail of the queue 2461 * @list: list to dequeue from 2462 * 2463 * Remove the tail of the list. This function does not take any locks 2464 * so must be used with appropriate locks held only. The tail item is 2465 * returned or %NULL if the list is empty. 2466 */ 2467 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2468 { 2469 struct sk_buff *skb = skb_peek_tail(list); 2470 if (skb) 2471 __skb_unlink(skb, list); 2472 return skb; 2473 } 2474 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2475 2476 2477 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2478 { 2479 return skb->data_len; 2480 } 2481 2482 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2483 { 2484 return skb->len - skb->data_len; 2485 } 2486 2487 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2488 { 2489 unsigned int i, len = 0; 2490 2491 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2492 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2493 return len; 2494 } 2495 2496 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2497 { 2498 return skb_headlen(skb) + __skb_pagelen(skb); 2499 } 2500 2501 static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag, 2502 netmem_ref netmem, int off, 2503 int size) 2504 { 2505 frag->netmem = netmem; 2506 frag->offset = off; 2507 skb_frag_size_set(frag, size); 2508 } 2509 2510 static inline void skb_frag_fill_page_desc(skb_frag_t *frag, 2511 struct page *page, 2512 int off, int size) 2513 { 2514 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size); 2515 } 2516 2517 static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo, 2518 int i, netmem_ref netmem, 2519 int off, int size) 2520 { 2521 skb_frag_t *frag = &shinfo->frags[i]; 2522 2523 skb_frag_fill_netmem_desc(frag, netmem, off, size); 2524 } 2525 2526 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2527 int i, struct page *page, 2528 int off, int size) 2529 { 2530 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off, 2531 size); 2532 } 2533 2534 /** 2535 * skb_len_add - adds a number to len fields of skb 2536 * @skb: buffer to add len to 2537 * @delta: number of bytes to add 2538 */ 2539 static inline void skb_len_add(struct sk_buff *skb, int delta) 2540 { 2541 skb->len += delta; 2542 skb->data_len += delta; 2543 skb->truesize += delta; 2544 } 2545 2546 /** 2547 * __skb_fill_netmem_desc - initialise a fragment in an skb 2548 * @skb: buffer containing fragment to be initialised 2549 * @i: fragment index to initialise 2550 * @netmem: the netmem to use for this fragment 2551 * @off: the offset to the data with @page 2552 * @size: the length of the data 2553 * 2554 * Initialises the @i'th fragment of @skb to point to &size bytes at 2555 * offset @off within @page. 2556 * 2557 * Does not take any additional reference on the fragment. 2558 */ 2559 static inline void __skb_fill_netmem_desc(struct sk_buff *skb, int i, 2560 netmem_ref netmem, int off, int size) 2561 { 2562 struct page *page; 2563 2564 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size); 2565 2566 if (netmem_is_net_iov(netmem)) { 2567 skb->unreadable = true; 2568 return; 2569 } 2570 2571 page = netmem_to_page(netmem); 2572 2573 /* Propagate page pfmemalloc to the skb if we can. The problem is 2574 * that not all callers have unique ownership of the page but rely 2575 * on page_is_pfmemalloc doing the right thing(tm). 2576 */ 2577 page = compound_head(page); 2578 if (page_is_pfmemalloc(page)) 2579 skb->pfmemalloc = true; 2580 } 2581 2582 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2583 struct page *page, int off, int size) 2584 { 2585 __skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2586 } 2587 2588 static inline void skb_fill_netmem_desc(struct sk_buff *skb, int i, 2589 netmem_ref netmem, int off, int size) 2590 { 2591 __skb_fill_netmem_desc(skb, i, netmem, off, size); 2592 skb_shinfo(skb)->nr_frags = i + 1; 2593 } 2594 2595 /** 2596 * skb_fill_page_desc - initialise a paged fragment in an skb 2597 * @skb: buffer containing fragment to be initialised 2598 * @i: paged fragment index to initialise 2599 * @page: the page to use for this fragment 2600 * @off: the offset to the data with @page 2601 * @size: the length of the data 2602 * 2603 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2604 * @skb to point to @size bytes at offset @off within @page. In 2605 * addition updates @skb such that @i is the last fragment. 2606 * 2607 * Does not take any additional reference on the fragment. 2608 */ 2609 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2610 struct page *page, int off, int size) 2611 { 2612 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2613 } 2614 2615 /** 2616 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2617 * @skb: buffer containing fragment to be initialised 2618 * @i: paged fragment index to initialise 2619 * @page: the page to use for this fragment 2620 * @off: the offset to the data with @page 2621 * @size: the length of the data 2622 * 2623 * Variant of skb_fill_page_desc() which does not deal with 2624 * pfmemalloc, if page is not owned by us. 2625 */ 2626 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2627 struct page *page, int off, 2628 int size) 2629 { 2630 struct skb_shared_info *shinfo = skb_shinfo(skb); 2631 2632 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2633 shinfo->nr_frags = i + 1; 2634 } 2635 2636 void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem, 2637 int off, int size, unsigned int truesize); 2638 2639 static inline void skb_add_rx_frag(struct sk_buff *skb, int i, 2640 struct page *page, int off, int size, 2641 unsigned int truesize) 2642 { 2643 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size, 2644 truesize); 2645 } 2646 2647 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2648 unsigned int truesize); 2649 2650 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2651 2652 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2653 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2654 { 2655 return skb->head + skb->tail; 2656 } 2657 2658 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2659 { 2660 skb->tail = skb->data - skb->head; 2661 } 2662 2663 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2664 { 2665 skb_reset_tail_pointer(skb); 2666 skb->tail += offset; 2667 } 2668 2669 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2670 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2671 { 2672 return skb->tail; 2673 } 2674 2675 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2676 { 2677 skb->tail = skb->data; 2678 } 2679 2680 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2681 { 2682 skb->tail = skb->data + offset; 2683 } 2684 2685 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2686 2687 static inline void skb_assert_len(struct sk_buff *skb) 2688 { 2689 #ifdef CONFIG_DEBUG_NET 2690 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2691 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2692 #endif /* CONFIG_DEBUG_NET */ 2693 } 2694 2695 #if defined(CONFIG_FAIL_SKB_REALLOC) 2696 void skb_might_realloc(struct sk_buff *skb); 2697 #else 2698 static inline void skb_might_realloc(struct sk_buff *skb) {} 2699 #endif 2700 2701 /* 2702 * Add data to an sk_buff 2703 */ 2704 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2705 void *skb_put(struct sk_buff *skb, unsigned int len); 2706 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2707 { 2708 void *tmp = skb_tail_pointer(skb); 2709 SKB_LINEAR_ASSERT(skb); 2710 skb->tail += len; 2711 skb->len += len; 2712 return tmp; 2713 } 2714 2715 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2716 { 2717 void *tmp = __skb_put(skb, len); 2718 2719 memset(tmp, 0, len); 2720 return tmp; 2721 } 2722 2723 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2724 unsigned int len) 2725 { 2726 void *tmp = __skb_put(skb, len); 2727 2728 memcpy(tmp, data, len); 2729 return tmp; 2730 } 2731 2732 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2733 { 2734 *(u8 *)__skb_put(skb, 1) = val; 2735 } 2736 2737 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2738 { 2739 void *tmp = skb_put(skb, len); 2740 2741 memset(tmp, 0, len); 2742 2743 return tmp; 2744 } 2745 2746 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2747 unsigned int len) 2748 { 2749 void *tmp = skb_put(skb, len); 2750 2751 memcpy(tmp, data, len); 2752 2753 return tmp; 2754 } 2755 2756 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2757 { 2758 *(u8 *)skb_put(skb, 1) = val; 2759 } 2760 2761 void *skb_push(struct sk_buff *skb, unsigned int len); 2762 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2763 { 2764 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2765 2766 skb->data -= len; 2767 skb->len += len; 2768 return skb->data; 2769 } 2770 2771 void *skb_pull(struct sk_buff *skb, unsigned int len); 2772 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2773 { 2774 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2775 2776 skb->len -= len; 2777 if (unlikely(skb->len < skb->data_len)) { 2778 #if defined(CONFIG_DEBUG_NET) 2779 skb->len += len; 2780 pr_err("__skb_pull(len=%u)\n", len); 2781 skb_dump(KERN_ERR, skb, false); 2782 #endif 2783 BUG(); 2784 } 2785 return skb->data += len; 2786 } 2787 2788 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2789 { 2790 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2791 } 2792 2793 void *skb_pull_data(struct sk_buff *skb, size_t len); 2794 2795 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2796 2797 static inline enum skb_drop_reason 2798 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len) 2799 { 2800 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2801 skb_might_realloc(skb); 2802 2803 if (likely(len <= skb_headlen(skb))) 2804 return SKB_NOT_DROPPED_YET; 2805 2806 if (unlikely(len > skb->len)) 2807 return SKB_DROP_REASON_PKT_TOO_SMALL; 2808 2809 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb)))) 2810 return SKB_DROP_REASON_NOMEM; 2811 2812 return SKB_NOT_DROPPED_YET; 2813 } 2814 2815 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2816 { 2817 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 2818 } 2819 2820 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2821 { 2822 if (!pskb_may_pull(skb, len)) 2823 return NULL; 2824 2825 skb->len -= len; 2826 return skb->data += len; 2827 } 2828 2829 void skb_condense(struct sk_buff *skb); 2830 2831 /** 2832 * skb_headroom - bytes at buffer head 2833 * @skb: buffer to check 2834 * 2835 * Return the number of bytes of free space at the head of an &sk_buff. 2836 */ 2837 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2838 { 2839 return skb->data - skb->head; 2840 } 2841 2842 /** 2843 * skb_tailroom - bytes at buffer end 2844 * @skb: buffer to check 2845 * 2846 * Return the number of bytes of free space at the tail of an sk_buff 2847 */ 2848 static inline int skb_tailroom(const struct sk_buff *skb) 2849 { 2850 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2851 } 2852 2853 /** 2854 * skb_availroom - bytes at buffer end 2855 * @skb: buffer to check 2856 * 2857 * Return the number of bytes of free space at the tail of an sk_buff 2858 * allocated by sk_stream_alloc() 2859 */ 2860 static inline int skb_availroom(const struct sk_buff *skb) 2861 { 2862 if (skb_is_nonlinear(skb)) 2863 return 0; 2864 2865 return skb->end - skb->tail - skb->reserved_tailroom; 2866 } 2867 2868 /** 2869 * skb_reserve - adjust headroom 2870 * @skb: buffer to alter 2871 * @len: bytes to move 2872 * 2873 * Increase the headroom of an empty &sk_buff by reducing the tail 2874 * room. This is only allowed for an empty buffer. 2875 */ 2876 static inline void skb_reserve(struct sk_buff *skb, int len) 2877 { 2878 skb->data += len; 2879 skb->tail += len; 2880 } 2881 2882 /** 2883 * skb_tailroom_reserve - adjust reserved_tailroom 2884 * @skb: buffer to alter 2885 * @mtu: maximum amount of headlen permitted 2886 * @needed_tailroom: minimum amount of reserved_tailroom 2887 * 2888 * Set reserved_tailroom so that headlen can be as large as possible but 2889 * not larger than mtu and tailroom cannot be smaller than 2890 * needed_tailroom. 2891 * The required headroom should already have been reserved before using 2892 * this function. 2893 */ 2894 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2895 unsigned int needed_tailroom) 2896 { 2897 SKB_LINEAR_ASSERT(skb); 2898 if (mtu < skb_tailroom(skb) - needed_tailroom) 2899 /* use at most mtu */ 2900 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2901 else 2902 /* use up to all available space */ 2903 skb->reserved_tailroom = needed_tailroom; 2904 } 2905 2906 #define ENCAP_TYPE_ETHER 0 2907 #define ENCAP_TYPE_IPPROTO 1 2908 2909 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2910 __be16 protocol) 2911 { 2912 skb->inner_protocol = protocol; 2913 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2914 } 2915 2916 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2917 __u8 ipproto) 2918 { 2919 skb->inner_ipproto = ipproto; 2920 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2921 } 2922 2923 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2924 { 2925 skb->inner_mac_header = skb->mac_header; 2926 skb->inner_network_header = skb->network_header; 2927 skb->inner_transport_header = skb->transport_header; 2928 } 2929 2930 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2931 { 2932 return skb->mac_header != (typeof(skb->mac_header))~0U; 2933 } 2934 2935 static inline void skb_reset_mac_len(struct sk_buff *skb) 2936 { 2937 if (!skb_mac_header_was_set(skb)) { 2938 DEBUG_NET_WARN_ON_ONCE(1); 2939 skb->mac_len = 0; 2940 } else { 2941 skb->mac_len = skb->network_header - skb->mac_header; 2942 } 2943 } 2944 2945 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2946 *skb) 2947 { 2948 return skb->head + skb->inner_transport_header; 2949 } 2950 2951 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2952 { 2953 return skb_inner_transport_header(skb) - skb->data; 2954 } 2955 2956 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2957 { 2958 long offset = skb->data - skb->head; 2959 2960 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_transport_header))offset); 2961 skb->inner_transport_header = offset; 2962 } 2963 2964 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2965 const int offset) 2966 { 2967 skb_reset_inner_transport_header(skb); 2968 skb->inner_transport_header += offset; 2969 } 2970 2971 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2972 { 2973 return skb->head + skb->inner_network_header; 2974 } 2975 2976 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2977 { 2978 long offset = skb->data - skb->head; 2979 2980 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_network_header))offset); 2981 skb->inner_network_header = offset; 2982 } 2983 2984 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2985 const int offset) 2986 { 2987 skb_reset_inner_network_header(skb); 2988 skb->inner_network_header += offset; 2989 } 2990 2991 static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb) 2992 { 2993 return skb->inner_network_header > 0; 2994 } 2995 2996 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2997 { 2998 return skb->head + skb->inner_mac_header; 2999 } 3000 3001 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 3002 { 3003 long offset = skb->data - skb->head; 3004 3005 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_mac_header))offset); 3006 skb->inner_mac_header = offset; 3007 } 3008 3009 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 3010 const int offset) 3011 { 3012 skb_reset_inner_mac_header(skb); 3013 skb->inner_mac_header += offset; 3014 } 3015 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 3016 { 3017 return skb->transport_header != (typeof(skb->transport_header))~0U; 3018 } 3019 3020 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 3021 { 3022 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3023 return skb->head + skb->transport_header; 3024 } 3025 3026 static inline void skb_reset_transport_header(struct sk_buff *skb) 3027 { 3028 long offset = skb->data - skb->head; 3029 3030 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->transport_header))offset); 3031 skb->transport_header = offset; 3032 } 3033 3034 static inline void skb_set_transport_header(struct sk_buff *skb, 3035 const int offset) 3036 { 3037 skb_reset_transport_header(skb); 3038 skb->transport_header += offset; 3039 } 3040 3041 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 3042 { 3043 return skb->head + skb->network_header; 3044 } 3045 3046 static inline void skb_reset_network_header(struct sk_buff *skb) 3047 { 3048 long offset = skb->data - skb->head; 3049 3050 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->network_header))offset); 3051 skb->network_header = offset; 3052 } 3053 3054 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 3055 { 3056 skb_reset_network_header(skb); 3057 skb->network_header += offset; 3058 } 3059 3060 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 3061 { 3062 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3063 return skb->head + skb->mac_header; 3064 } 3065 3066 static inline int skb_mac_offset(const struct sk_buff *skb) 3067 { 3068 return skb_mac_header(skb) - skb->data; 3069 } 3070 3071 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 3072 { 3073 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3074 return skb->network_header - skb->mac_header; 3075 } 3076 3077 static inline void skb_unset_mac_header(struct sk_buff *skb) 3078 { 3079 skb->mac_header = (typeof(skb->mac_header))~0U; 3080 } 3081 3082 static inline void skb_reset_mac_header(struct sk_buff *skb) 3083 { 3084 long offset = skb->data - skb->head; 3085 3086 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->mac_header))offset); 3087 skb->mac_header = offset; 3088 } 3089 3090 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 3091 { 3092 skb_reset_mac_header(skb); 3093 skb->mac_header += offset; 3094 } 3095 3096 static inline void skb_pop_mac_header(struct sk_buff *skb) 3097 { 3098 skb->mac_header = skb->network_header; 3099 } 3100 3101 static inline void skb_probe_transport_header(struct sk_buff *skb) 3102 { 3103 struct flow_keys_basic keys; 3104 3105 if (skb_transport_header_was_set(skb)) 3106 return; 3107 3108 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 3109 NULL, 0, 0, 0, 0)) 3110 skb_set_transport_header(skb, keys.control.thoff); 3111 } 3112 3113 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 3114 { 3115 if (skb_mac_header_was_set(skb)) { 3116 const unsigned char *old_mac = skb_mac_header(skb); 3117 3118 skb_set_mac_header(skb, -skb->mac_len); 3119 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 3120 } 3121 } 3122 3123 /* Move the full mac header up to current network_header. 3124 * Leaves skb->data pointing at offset skb->mac_len into the mac_header. 3125 * Must be provided the complete mac header length. 3126 */ 3127 static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len) 3128 { 3129 if (skb_mac_header_was_set(skb)) { 3130 const unsigned char *old_mac = skb_mac_header(skb); 3131 3132 skb_set_mac_header(skb, -full_mac_len); 3133 memmove(skb_mac_header(skb), old_mac, full_mac_len); 3134 __skb_push(skb, full_mac_len - skb->mac_len); 3135 } 3136 } 3137 3138 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 3139 { 3140 return skb->csum_start - skb_headroom(skb); 3141 } 3142 3143 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 3144 { 3145 return skb->head + skb->csum_start; 3146 } 3147 3148 static inline int skb_transport_offset(const struct sk_buff *skb) 3149 { 3150 return skb_transport_header(skb) - skb->data; 3151 } 3152 3153 static inline u32 skb_network_header_len(const struct sk_buff *skb) 3154 { 3155 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3156 return skb->transport_header - skb->network_header; 3157 } 3158 3159 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 3160 { 3161 return skb->inner_transport_header - skb->inner_network_header; 3162 } 3163 3164 static inline int skb_network_offset(const struct sk_buff *skb) 3165 { 3166 return skb_network_header(skb) - skb->data; 3167 } 3168 3169 static inline int skb_inner_network_offset(const struct sk_buff *skb) 3170 { 3171 return skb_inner_network_header(skb) - skb->data; 3172 } 3173 3174 static inline enum skb_drop_reason 3175 pskb_network_may_pull_reason(struct sk_buff *skb, unsigned int len) 3176 { 3177 return pskb_may_pull_reason(skb, skb_network_offset(skb) + len); 3178 } 3179 3180 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 3181 { 3182 return pskb_network_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 3183 } 3184 3185 /* 3186 * CPUs often take a performance hit when accessing unaligned memory 3187 * locations. The actual performance hit varies, it can be small if the 3188 * hardware handles it or large if we have to take an exception and fix it 3189 * in software. 3190 * 3191 * Since an ethernet header is 14 bytes network drivers often end up with 3192 * the IP header at an unaligned offset. The IP header can be aligned by 3193 * shifting the start of the packet by 2 bytes. Drivers should do this 3194 * with: 3195 * 3196 * skb_reserve(skb, NET_IP_ALIGN); 3197 * 3198 * The downside to this alignment of the IP header is that the DMA is now 3199 * unaligned. On some architectures the cost of an unaligned DMA is high 3200 * and this cost outweighs the gains made by aligning the IP header. 3201 * 3202 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 3203 * to be overridden. 3204 */ 3205 #ifndef NET_IP_ALIGN 3206 #define NET_IP_ALIGN 2 3207 #endif 3208 3209 /* 3210 * The networking layer reserves some headroom in skb data (via 3211 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 3212 * the header has to grow. In the default case, if the header has to grow 3213 * 32 bytes or less we avoid the reallocation. 3214 * 3215 * Unfortunately this headroom changes the DMA alignment of the resulting 3216 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 3217 * on some architectures. An architecture can override this value, 3218 * perhaps setting it to a cacheline in size (since that will maintain 3219 * cacheline alignment of the DMA). It must be a power of 2. 3220 * 3221 * Various parts of the networking layer expect at least 32 bytes of 3222 * headroom, you should not reduce this. 3223 * 3224 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 3225 * to reduce average number of cache lines per packet. 3226 * get_rps_cpu() for example only access one 64 bytes aligned block : 3227 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 3228 */ 3229 #ifndef NET_SKB_PAD 3230 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 3231 #endif 3232 3233 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 3234 3235 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3236 { 3237 if (WARN_ON(skb_is_nonlinear(skb))) 3238 return; 3239 skb->len = len; 3240 skb_set_tail_pointer(skb, len); 3241 } 3242 3243 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3244 { 3245 __skb_set_length(skb, len); 3246 } 3247 3248 void skb_trim(struct sk_buff *skb, unsigned int len); 3249 3250 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3251 { 3252 if (skb->data_len) 3253 return ___pskb_trim(skb, len); 3254 __skb_trim(skb, len); 3255 return 0; 3256 } 3257 3258 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3259 { 3260 skb_might_realloc(skb); 3261 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3262 } 3263 3264 /** 3265 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3266 * @skb: buffer to alter 3267 * @len: new length 3268 * 3269 * This is identical to pskb_trim except that the caller knows that 3270 * the skb is not cloned so we should never get an error due to out- 3271 * of-memory. 3272 */ 3273 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3274 { 3275 int err = pskb_trim(skb, len); 3276 BUG_ON(err); 3277 } 3278 3279 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3280 { 3281 unsigned int diff = len - skb->len; 3282 3283 if (skb_tailroom(skb) < diff) { 3284 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3285 GFP_ATOMIC); 3286 if (ret) 3287 return ret; 3288 } 3289 __skb_set_length(skb, len); 3290 return 0; 3291 } 3292 3293 /** 3294 * skb_orphan - orphan a buffer 3295 * @skb: buffer to orphan 3296 * 3297 * If a buffer currently has an owner then we call the owner's 3298 * destructor function and make the @skb unowned. The buffer continues 3299 * to exist but is no longer charged to its former owner. 3300 */ 3301 static inline void skb_orphan(struct sk_buff *skb) 3302 { 3303 if (skb->destructor) { 3304 skb->destructor(skb); 3305 skb->destructor = NULL; 3306 skb->sk = NULL; 3307 } else { 3308 BUG_ON(skb->sk); 3309 } 3310 } 3311 3312 /** 3313 * skb_orphan_frags - orphan the frags contained in a buffer 3314 * @skb: buffer to orphan frags from 3315 * @gfp_mask: allocation mask for replacement pages 3316 * 3317 * For each frag in the SKB which needs a destructor (i.e. has an 3318 * owner) create a copy of that frag and release the original 3319 * page by calling the destructor. 3320 */ 3321 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3322 { 3323 if (likely(!skb_zcopy(skb))) 3324 return 0; 3325 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3326 return 0; 3327 return skb_copy_ubufs(skb, gfp_mask); 3328 } 3329 3330 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3331 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3332 { 3333 if (likely(!skb_zcopy(skb))) 3334 return 0; 3335 return skb_copy_ubufs(skb, gfp_mask); 3336 } 3337 3338 /** 3339 * __skb_queue_purge_reason - empty a list 3340 * @list: list to empty 3341 * @reason: drop reason 3342 * 3343 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3344 * the list and one reference dropped. This function does not take the 3345 * list lock and the caller must hold the relevant locks to use it. 3346 */ 3347 static inline void __skb_queue_purge_reason(struct sk_buff_head *list, 3348 enum skb_drop_reason reason) 3349 { 3350 struct sk_buff *skb; 3351 3352 while ((skb = __skb_dequeue(list)) != NULL) 3353 kfree_skb_reason(skb, reason); 3354 } 3355 3356 static inline void __skb_queue_purge(struct sk_buff_head *list) 3357 { 3358 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3359 } 3360 3361 void skb_queue_purge_reason(struct sk_buff_head *list, 3362 enum skb_drop_reason reason); 3363 3364 static inline void skb_queue_purge(struct sk_buff_head *list) 3365 { 3366 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3367 } 3368 3369 unsigned int skb_rbtree_purge(struct rb_root *root); 3370 void skb_errqueue_purge(struct sk_buff_head *list); 3371 3372 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3373 3374 /** 3375 * netdev_alloc_frag - allocate a page fragment 3376 * @fragsz: fragment size 3377 * 3378 * Allocates a frag from a page for receive buffer. 3379 * Uses GFP_ATOMIC allocations. 3380 */ 3381 static inline void *netdev_alloc_frag(unsigned int fragsz) 3382 { 3383 return __netdev_alloc_frag_align(fragsz, ~0u); 3384 } 3385 3386 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3387 unsigned int align) 3388 { 3389 WARN_ON_ONCE(!is_power_of_2(align)); 3390 return __netdev_alloc_frag_align(fragsz, -align); 3391 } 3392 3393 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3394 gfp_t gfp_mask); 3395 3396 /** 3397 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3398 * @dev: network device to receive on 3399 * @length: length to allocate 3400 * 3401 * Allocate a new &sk_buff and assign it a usage count of one. The 3402 * buffer has unspecified headroom built in. Users should allocate 3403 * the headroom they think they need without accounting for the 3404 * built in space. The built in space is used for optimisations. 3405 * 3406 * %NULL is returned if there is no free memory. Although this function 3407 * allocates memory it can be called from an interrupt. 3408 */ 3409 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3410 unsigned int length) 3411 { 3412 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3413 } 3414 3415 /* legacy helper around __netdev_alloc_skb() */ 3416 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3417 gfp_t gfp_mask) 3418 { 3419 return __netdev_alloc_skb(NULL, length, gfp_mask); 3420 } 3421 3422 /* legacy helper around netdev_alloc_skb() */ 3423 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3424 { 3425 return netdev_alloc_skb(NULL, length); 3426 } 3427 3428 3429 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3430 unsigned int length, gfp_t gfp) 3431 { 3432 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3433 3434 if (NET_IP_ALIGN && skb) 3435 skb_reserve(skb, NET_IP_ALIGN); 3436 return skb; 3437 } 3438 3439 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3440 unsigned int length) 3441 { 3442 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3443 } 3444 3445 static inline void skb_free_frag(void *addr) 3446 { 3447 page_frag_free(addr); 3448 } 3449 3450 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3451 3452 static inline void *napi_alloc_frag(unsigned int fragsz) 3453 { 3454 return __napi_alloc_frag_align(fragsz, ~0u); 3455 } 3456 3457 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3458 unsigned int align) 3459 { 3460 WARN_ON_ONCE(!is_power_of_2(align)); 3461 return __napi_alloc_frag_align(fragsz, -align); 3462 } 3463 3464 struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int length); 3465 void napi_consume_skb(struct sk_buff *skb, int budget); 3466 3467 void napi_skb_free_stolen_head(struct sk_buff *skb); 3468 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason); 3469 3470 /** 3471 * __dev_alloc_pages - allocate page for network Rx 3472 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3473 * @order: size of the allocation 3474 * 3475 * Allocate a new page. 3476 * 3477 * %NULL is returned if there is no free memory. 3478 */ 3479 static inline struct page *__dev_alloc_pages_noprof(gfp_t gfp_mask, 3480 unsigned int order) 3481 { 3482 /* This piece of code contains several assumptions. 3483 * 1. This is for device Rx, therefore a cold page is preferred. 3484 * 2. The expectation is the user wants a compound page. 3485 * 3. If requesting a order 0 page it will not be compound 3486 * due to the check to see if order has a value in prep_new_page 3487 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3488 * code in gfp_to_alloc_flags that should be enforcing this. 3489 */ 3490 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3491 3492 return alloc_pages_node_noprof(NUMA_NO_NODE, gfp_mask, order); 3493 } 3494 #define __dev_alloc_pages(...) alloc_hooks(__dev_alloc_pages_noprof(__VA_ARGS__)) 3495 3496 /* 3497 * This specialized allocator has to be a macro for its allocations to be 3498 * accounted separately (to have a separate alloc_tag). 3499 */ 3500 #define dev_alloc_pages(_order) __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, _order) 3501 3502 /** 3503 * __dev_alloc_page - allocate a page for network Rx 3504 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3505 * 3506 * Allocate a new page. 3507 * 3508 * %NULL is returned if there is no free memory. 3509 */ 3510 static inline struct page *__dev_alloc_page_noprof(gfp_t gfp_mask) 3511 { 3512 return __dev_alloc_pages_noprof(gfp_mask, 0); 3513 } 3514 #define __dev_alloc_page(...) alloc_hooks(__dev_alloc_page_noprof(__VA_ARGS__)) 3515 3516 /* 3517 * This specialized allocator has to be a macro for its allocations to be 3518 * accounted separately (to have a separate alloc_tag). 3519 */ 3520 #define dev_alloc_page() dev_alloc_pages(0) 3521 3522 /** 3523 * dev_page_is_reusable - check whether a page can be reused for network Rx 3524 * @page: the page to test 3525 * 3526 * A page shouldn't be considered for reusing/recycling if it was allocated 3527 * under memory pressure or at a distant memory node. 3528 * 3529 * Returns: false if this page should be returned to page allocator, true 3530 * otherwise. 3531 */ 3532 static inline bool dev_page_is_reusable(const struct page *page) 3533 { 3534 return likely(page_to_nid(page) == numa_mem_id() && 3535 !page_is_pfmemalloc(page)); 3536 } 3537 3538 /** 3539 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3540 * @page: The page that was allocated from skb_alloc_page 3541 * @skb: The skb that may need pfmemalloc set 3542 */ 3543 static inline void skb_propagate_pfmemalloc(const struct page *page, 3544 struct sk_buff *skb) 3545 { 3546 if (page_is_pfmemalloc(page)) 3547 skb->pfmemalloc = true; 3548 } 3549 3550 /** 3551 * skb_frag_off() - Returns the offset of a skb fragment 3552 * @frag: the paged fragment 3553 */ 3554 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3555 { 3556 return frag->offset; 3557 } 3558 3559 /** 3560 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3561 * @frag: skb fragment 3562 * @delta: value to add 3563 */ 3564 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3565 { 3566 frag->offset += delta; 3567 } 3568 3569 /** 3570 * skb_frag_off_set() - Sets the offset of a skb fragment 3571 * @frag: skb fragment 3572 * @offset: offset of fragment 3573 */ 3574 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3575 { 3576 frag->offset = offset; 3577 } 3578 3579 /** 3580 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3581 * @fragto: skb fragment where offset is set 3582 * @fragfrom: skb fragment offset is copied from 3583 */ 3584 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3585 const skb_frag_t *fragfrom) 3586 { 3587 fragto->offset = fragfrom->offset; 3588 } 3589 3590 /* Return: true if the skb_frag contains a net_iov. */ 3591 static inline bool skb_frag_is_net_iov(const skb_frag_t *frag) 3592 { 3593 return netmem_is_net_iov(frag->netmem); 3594 } 3595 3596 /** 3597 * skb_frag_net_iov - retrieve the net_iov referred to by fragment 3598 * @frag: the fragment 3599 * 3600 * Return: the &struct net_iov associated with @frag. Returns NULL if this 3601 * frag has no associated net_iov. 3602 */ 3603 static inline struct net_iov *skb_frag_net_iov(const skb_frag_t *frag) 3604 { 3605 if (!skb_frag_is_net_iov(frag)) 3606 return NULL; 3607 3608 return netmem_to_net_iov(frag->netmem); 3609 } 3610 3611 /** 3612 * skb_frag_page - retrieve the page referred to by a paged fragment 3613 * @frag: the paged fragment 3614 * 3615 * Return: the &struct page associated with @frag. Returns NULL if this frag 3616 * has no associated page. 3617 */ 3618 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3619 { 3620 if (skb_frag_is_net_iov(frag)) 3621 return NULL; 3622 3623 return netmem_to_page(frag->netmem); 3624 } 3625 3626 /** 3627 * skb_frag_netmem - retrieve the netmem referred to by a fragment 3628 * @frag: the fragment 3629 * 3630 * Return: the &netmem_ref associated with @frag. 3631 */ 3632 static inline netmem_ref skb_frag_netmem(const skb_frag_t *frag) 3633 { 3634 return frag->netmem; 3635 } 3636 3637 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb, 3638 unsigned int headroom); 3639 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb, 3640 const struct bpf_prog *prog); 3641 3642 /** 3643 * skb_frag_address - gets the address of the data contained in a paged fragment 3644 * @frag: the paged fragment buffer 3645 * 3646 * Returns: the address of the data within @frag. The page must already 3647 * be mapped. 3648 */ 3649 static inline void *skb_frag_address(const skb_frag_t *frag) 3650 { 3651 if (!skb_frag_page(frag)) 3652 return NULL; 3653 3654 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3655 } 3656 3657 /** 3658 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3659 * @frag: the paged fragment buffer 3660 * 3661 * Returns: the address of the data within @frag. Checks that the page 3662 * is mapped and returns %NULL otherwise. 3663 */ 3664 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3665 { 3666 void *ptr = page_address(skb_frag_page(frag)); 3667 if (unlikely(!ptr)) 3668 return NULL; 3669 3670 return ptr + skb_frag_off(frag); 3671 } 3672 3673 /** 3674 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3675 * @fragto: skb fragment where page is set 3676 * @fragfrom: skb fragment page is copied from 3677 */ 3678 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3679 const skb_frag_t *fragfrom) 3680 { 3681 fragto->netmem = fragfrom->netmem; 3682 } 3683 3684 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3685 3686 /** 3687 * __skb_frag_dma_map - maps a paged fragment via the DMA API 3688 * @dev: the device to map the fragment to 3689 * @frag: the paged fragment to map 3690 * @offset: the offset within the fragment (starting at the 3691 * fragment's own offset) 3692 * @size: the number of bytes to map 3693 * @dir: the direction of the mapping (``PCI_DMA_*``) 3694 * 3695 * Maps the page associated with @frag to @device. 3696 */ 3697 static inline dma_addr_t __skb_frag_dma_map(struct device *dev, 3698 const skb_frag_t *frag, 3699 size_t offset, size_t size, 3700 enum dma_data_direction dir) 3701 { 3702 return dma_map_page(dev, skb_frag_page(frag), 3703 skb_frag_off(frag) + offset, size, dir); 3704 } 3705 3706 #define skb_frag_dma_map(dev, frag, ...) \ 3707 CONCATENATE(_skb_frag_dma_map, \ 3708 COUNT_ARGS(__VA_ARGS__))(dev, frag, ##__VA_ARGS__) 3709 3710 #define __skb_frag_dma_map1(dev, frag, offset, uf, uo) ({ \ 3711 const skb_frag_t *uf = (frag); \ 3712 size_t uo = (offset); \ 3713 \ 3714 __skb_frag_dma_map(dev, uf, uo, skb_frag_size(uf) - uo, \ 3715 DMA_TO_DEVICE); \ 3716 }) 3717 #define _skb_frag_dma_map1(dev, frag, offset) \ 3718 __skb_frag_dma_map1(dev, frag, offset, __UNIQUE_ID(frag_), \ 3719 __UNIQUE_ID(offset_)) 3720 #define _skb_frag_dma_map0(dev, frag) \ 3721 _skb_frag_dma_map1(dev, frag, 0) 3722 #define _skb_frag_dma_map2(dev, frag, offset, size) \ 3723 __skb_frag_dma_map(dev, frag, offset, size, DMA_TO_DEVICE) 3724 #define _skb_frag_dma_map3(dev, frag, offset, size, dir) \ 3725 __skb_frag_dma_map(dev, frag, offset, size, dir) 3726 3727 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3728 gfp_t gfp_mask) 3729 { 3730 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3731 } 3732 3733 3734 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3735 gfp_t gfp_mask) 3736 { 3737 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3738 } 3739 3740 3741 /** 3742 * skb_clone_writable - is the header of a clone writable 3743 * @skb: buffer to check 3744 * @len: length up to which to write 3745 * 3746 * Returns true if modifying the header part of the cloned buffer 3747 * does not requires the data to be copied. 3748 */ 3749 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3750 { 3751 return !skb_header_cloned(skb) && 3752 skb_headroom(skb) + len <= skb->hdr_len; 3753 } 3754 3755 static inline int skb_try_make_writable(struct sk_buff *skb, 3756 unsigned int write_len) 3757 { 3758 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3759 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3760 } 3761 3762 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3763 int cloned) 3764 { 3765 int delta = 0; 3766 3767 if (headroom > skb_headroom(skb)) 3768 delta = headroom - skb_headroom(skb); 3769 3770 if (delta || cloned) 3771 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3772 GFP_ATOMIC); 3773 return 0; 3774 } 3775 3776 /** 3777 * skb_cow - copy header of skb when it is required 3778 * @skb: buffer to cow 3779 * @headroom: needed headroom 3780 * 3781 * If the skb passed lacks sufficient headroom or its data part 3782 * is shared, data is reallocated. If reallocation fails, an error 3783 * is returned and original skb is not changed. 3784 * 3785 * The result is skb with writable area skb->head...skb->tail 3786 * and at least @headroom of space at head. 3787 */ 3788 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3789 { 3790 return __skb_cow(skb, headroom, skb_cloned(skb)); 3791 } 3792 3793 /** 3794 * skb_cow_head - skb_cow but only making the head writable 3795 * @skb: buffer to cow 3796 * @headroom: needed headroom 3797 * 3798 * This function is identical to skb_cow except that we replace the 3799 * skb_cloned check by skb_header_cloned. It should be used when 3800 * you only need to push on some header and do not need to modify 3801 * the data. 3802 */ 3803 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3804 { 3805 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3806 } 3807 3808 /** 3809 * skb_padto - pad an skbuff up to a minimal size 3810 * @skb: buffer to pad 3811 * @len: minimal length 3812 * 3813 * Pads up a buffer to ensure the trailing bytes exist and are 3814 * blanked. If the buffer already contains sufficient data it 3815 * is untouched. Otherwise it is extended. Returns zero on 3816 * success. The skb is freed on error. 3817 */ 3818 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3819 { 3820 unsigned int size = skb->len; 3821 if (likely(size >= len)) 3822 return 0; 3823 return skb_pad(skb, len - size); 3824 } 3825 3826 /** 3827 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3828 * @skb: buffer to pad 3829 * @len: minimal length 3830 * @free_on_error: free buffer on error 3831 * 3832 * Pads up a buffer to ensure the trailing bytes exist and are 3833 * blanked. If the buffer already contains sufficient data it 3834 * is untouched. Otherwise it is extended. Returns zero on 3835 * success. The skb is freed on error if @free_on_error is true. 3836 */ 3837 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3838 unsigned int len, 3839 bool free_on_error) 3840 { 3841 unsigned int size = skb->len; 3842 3843 if (unlikely(size < len)) { 3844 len -= size; 3845 if (__skb_pad(skb, len, free_on_error)) 3846 return -ENOMEM; 3847 __skb_put(skb, len); 3848 } 3849 return 0; 3850 } 3851 3852 /** 3853 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3854 * @skb: buffer to pad 3855 * @len: minimal length 3856 * 3857 * Pads up a buffer to ensure the trailing bytes exist and are 3858 * blanked. If the buffer already contains sufficient data it 3859 * is untouched. Otherwise it is extended. Returns zero on 3860 * success. The skb is freed on error. 3861 */ 3862 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3863 { 3864 return __skb_put_padto(skb, len, true); 3865 } 3866 3867 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i) 3868 __must_check; 3869 3870 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3871 const struct page *page, int off) 3872 { 3873 if (skb_zcopy(skb)) 3874 return false; 3875 if (i) { 3876 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3877 3878 return page == skb_frag_page(frag) && 3879 off == skb_frag_off(frag) + skb_frag_size(frag); 3880 } 3881 return false; 3882 } 3883 3884 static inline int __skb_linearize(struct sk_buff *skb) 3885 { 3886 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3887 } 3888 3889 /** 3890 * skb_linearize - convert paged skb to linear one 3891 * @skb: buffer to linarize 3892 * 3893 * If there is no free memory -ENOMEM is returned, otherwise zero 3894 * is returned and the old skb data released. 3895 */ 3896 static inline int skb_linearize(struct sk_buff *skb) 3897 { 3898 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3899 } 3900 3901 /** 3902 * skb_has_shared_frag - can any frag be overwritten 3903 * @skb: buffer to test 3904 * 3905 * Return: true if the skb has at least one frag that might be modified 3906 * by an external entity (as in vmsplice()/sendfile()) 3907 */ 3908 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3909 { 3910 return skb_is_nonlinear(skb) && 3911 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 3912 } 3913 3914 /** 3915 * skb_linearize_cow - make sure skb is linear and writable 3916 * @skb: buffer to process 3917 * 3918 * If there is no free memory -ENOMEM is returned, otherwise zero 3919 * is returned and the old skb data released. 3920 */ 3921 static inline int skb_linearize_cow(struct sk_buff *skb) 3922 { 3923 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3924 __skb_linearize(skb) : 0; 3925 } 3926 3927 static __always_inline void 3928 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3929 unsigned int off) 3930 { 3931 if (skb->ip_summed == CHECKSUM_COMPLETE) 3932 skb->csum = csum_block_sub(skb->csum, 3933 csum_partial(start, len, 0), off); 3934 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3935 skb_checksum_start_offset(skb) < 0) 3936 skb->ip_summed = CHECKSUM_NONE; 3937 } 3938 3939 /** 3940 * skb_postpull_rcsum - update checksum for received skb after pull 3941 * @skb: buffer to update 3942 * @start: start of data before pull 3943 * @len: length of data pulled 3944 * 3945 * After doing a pull on a received packet, you need to call this to 3946 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3947 * CHECKSUM_NONE so that it can be recomputed from scratch. 3948 */ 3949 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3950 const void *start, unsigned int len) 3951 { 3952 if (skb->ip_summed == CHECKSUM_COMPLETE) 3953 skb->csum = wsum_negate(csum_partial(start, len, 3954 wsum_negate(skb->csum))); 3955 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3956 skb_checksum_start_offset(skb) < 0) 3957 skb->ip_summed = CHECKSUM_NONE; 3958 } 3959 3960 static __always_inline void 3961 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3962 unsigned int off) 3963 { 3964 if (skb->ip_summed == CHECKSUM_COMPLETE) 3965 skb->csum = csum_block_add(skb->csum, 3966 csum_partial(start, len, 0), off); 3967 } 3968 3969 /** 3970 * skb_postpush_rcsum - update checksum for received skb after push 3971 * @skb: buffer to update 3972 * @start: start of data after push 3973 * @len: length of data pushed 3974 * 3975 * After doing a push on a received packet, you need to call this to 3976 * update the CHECKSUM_COMPLETE checksum. 3977 */ 3978 static inline void skb_postpush_rcsum(struct sk_buff *skb, 3979 const void *start, unsigned int len) 3980 { 3981 __skb_postpush_rcsum(skb, start, len, 0); 3982 } 3983 3984 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3985 3986 /** 3987 * skb_push_rcsum - push skb and update receive checksum 3988 * @skb: buffer to update 3989 * @len: length of data pulled 3990 * 3991 * This function performs an skb_push on the packet and updates 3992 * the CHECKSUM_COMPLETE checksum. It should be used on 3993 * receive path processing instead of skb_push unless you know 3994 * that the checksum difference is zero (e.g., a valid IP header) 3995 * or you are setting ip_summed to CHECKSUM_NONE. 3996 */ 3997 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3998 { 3999 skb_push(skb, len); 4000 skb_postpush_rcsum(skb, skb->data, len); 4001 return skb->data; 4002 } 4003 4004 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 4005 /** 4006 * pskb_trim_rcsum - trim received skb and update checksum 4007 * @skb: buffer to trim 4008 * @len: new length 4009 * 4010 * This is exactly the same as pskb_trim except that it ensures the 4011 * checksum of received packets are still valid after the operation. 4012 * It can change skb pointers. 4013 */ 4014 4015 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4016 { 4017 skb_might_realloc(skb); 4018 if (likely(len >= skb->len)) 4019 return 0; 4020 return pskb_trim_rcsum_slow(skb, len); 4021 } 4022 4023 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4024 { 4025 if (skb->ip_summed == CHECKSUM_COMPLETE) 4026 skb->ip_summed = CHECKSUM_NONE; 4027 __skb_trim(skb, len); 4028 return 0; 4029 } 4030 4031 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 4032 { 4033 if (skb->ip_summed == CHECKSUM_COMPLETE) 4034 skb->ip_summed = CHECKSUM_NONE; 4035 return __skb_grow(skb, len); 4036 } 4037 4038 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 4039 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 4040 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 4041 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 4042 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 4043 4044 #define skb_queue_walk(queue, skb) \ 4045 for (skb = (queue)->next; \ 4046 skb != (struct sk_buff *)(queue); \ 4047 skb = skb->next) 4048 4049 #define skb_queue_walk_safe(queue, skb, tmp) \ 4050 for (skb = (queue)->next, tmp = skb->next; \ 4051 skb != (struct sk_buff *)(queue); \ 4052 skb = tmp, tmp = skb->next) 4053 4054 #define skb_queue_walk_from(queue, skb) \ 4055 for (; skb != (struct sk_buff *)(queue); \ 4056 skb = skb->next) 4057 4058 #define skb_rbtree_walk(skb, root) \ 4059 for (skb = skb_rb_first(root); skb != NULL; \ 4060 skb = skb_rb_next(skb)) 4061 4062 #define skb_rbtree_walk_from(skb) \ 4063 for (; skb != NULL; \ 4064 skb = skb_rb_next(skb)) 4065 4066 #define skb_rbtree_walk_from_safe(skb, tmp) \ 4067 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 4068 skb = tmp) 4069 4070 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 4071 for (tmp = skb->next; \ 4072 skb != (struct sk_buff *)(queue); \ 4073 skb = tmp, tmp = skb->next) 4074 4075 #define skb_queue_reverse_walk(queue, skb) \ 4076 for (skb = (queue)->prev; \ 4077 skb != (struct sk_buff *)(queue); \ 4078 skb = skb->prev) 4079 4080 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 4081 for (skb = (queue)->prev, tmp = skb->prev; \ 4082 skb != (struct sk_buff *)(queue); \ 4083 skb = tmp, tmp = skb->prev) 4084 4085 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 4086 for (tmp = skb->prev; \ 4087 skb != (struct sk_buff *)(queue); \ 4088 skb = tmp, tmp = skb->prev) 4089 4090 static inline bool skb_has_frag_list(const struct sk_buff *skb) 4091 { 4092 return skb_shinfo(skb)->frag_list != NULL; 4093 } 4094 4095 static inline void skb_frag_list_init(struct sk_buff *skb) 4096 { 4097 skb_shinfo(skb)->frag_list = NULL; 4098 } 4099 4100 #define skb_walk_frags(skb, iter) \ 4101 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 4102 4103 4104 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 4105 int *err, long *timeo_p, 4106 const struct sk_buff *skb); 4107 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 4108 struct sk_buff_head *queue, 4109 unsigned int flags, 4110 int *off, int *err, 4111 struct sk_buff **last); 4112 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 4113 struct sk_buff_head *queue, 4114 unsigned int flags, int *off, int *err, 4115 struct sk_buff **last); 4116 struct sk_buff *__skb_recv_datagram(struct sock *sk, 4117 struct sk_buff_head *sk_queue, 4118 unsigned int flags, int *off, int *err); 4119 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 4120 __poll_t datagram_poll(struct file *file, struct socket *sock, 4121 struct poll_table_struct *wait); 4122 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 4123 struct iov_iter *to, int size); 4124 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 4125 struct msghdr *msg, int size) 4126 { 4127 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 4128 } 4129 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 4130 struct msghdr *msg); 4131 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 4132 struct iov_iter *to, int len, 4133 struct ahash_request *hash); 4134 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 4135 struct iov_iter *from, int len); 4136 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 4137 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 4138 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 4139 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 4140 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 4141 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 4142 int len); 4143 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 4144 struct pipe_inode_info *pipe, unsigned int len, 4145 unsigned int flags); 4146 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 4147 int len); 4148 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 4149 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 4150 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 4151 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 4152 int len, int hlen); 4153 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 4154 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 4155 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 4156 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 4157 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 4158 unsigned int offset); 4159 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 4160 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 4161 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev); 4162 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 4163 int skb_vlan_pop(struct sk_buff *skb); 4164 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 4165 int skb_eth_pop(struct sk_buff *skb); 4166 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 4167 const unsigned char *src); 4168 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 4169 int mac_len, bool ethernet); 4170 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 4171 bool ethernet); 4172 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 4173 int skb_mpls_dec_ttl(struct sk_buff *skb); 4174 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 4175 gfp_t gfp); 4176 4177 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 4178 { 4179 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 4180 } 4181 4182 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 4183 { 4184 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 4185 } 4186 4187 struct skb_checksum_ops { 4188 __wsum (*update)(const void *mem, int len, __wsum wsum); 4189 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 4190 }; 4191 4192 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 4193 4194 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 4195 __wsum csum, const struct skb_checksum_ops *ops); 4196 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 4197 __wsum csum); 4198 4199 static inline void * __must_check 4200 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 4201 const void *data, int hlen, void *buffer) 4202 { 4203 if (likely(hlen - offset >= len)) 4204 return (void *)data + offset; 4205 4206 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 4207 return NULL; 4208 4209 return buffer; 4210 } 4211 4212 static inline void * __must_check 4213 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4214 { 4215 return __skb_header_pointer(skb, offset, len, skb->data, 4216 skb_headlen(skb), buffer); 4217 } 4218 4219 static inline void * __must_check 4220 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len) 4221 { 4222 if (likely(skb_headlen(skb) - offset >= len)) 4223 return skb->data + offset; 4224 return NULL; 4225 } 4226 4227 /** 4228 * skb_needs_linearize - check if we need to linearize a given skb 4229 * depending on the given device features. 4230 * @skb: socket buffer to check 4231 * @features: net device features 4232 * 4233 * Returns true if either: 4234 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4235 * 2. skb is fragmented and the device does not support SG. 4236 */ 4237 static inline bool skb_needs_linearize(struct sk_buff *skb, 4238 netdev_features_t features) 4239 { 4240 return skb_is_nonlinear(skb) && 4241 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4242 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4243 } 4244 4245 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4246 void *to, 4247 const unsigned int len) 4248 { 4249 memcpy(to, skb->data, len); 4250 } 4251 4252 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4253 const int offset, void *to, 4254 const unsigned int len) 4255 { 4256 memcpy(to, skb->data + offset, len); 4257 } 4258 4259 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4260 const void *from, 4261 const unsigned int len) 4262 { 4263 memcpy(skb->data, from, len); 4264 } 4265 4266 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4267 const int offset, 4268 const void *from, 4269 const unsigned int len) 4270 { 4271 memcpy(skb->data + offset, from, len); 4272 } 4273 4274 void skb_init(void); 4275 4276 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4277 { 4278 return skb->tstamp; 4279 } 4280 4281 /** 4282 * skb_get_timestamp - get timestamp from a skb 4283 * @skb: skb to get stamp from 4284 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4285 * 4286 * Timestamps are stored in the skb as offsets to a base timestamp. 4287 * This function converts the offset back to a struct timeval and stores 4288 * it in stamp. 4289 */ 4290 static inline void skb_get_timestamp(const struct sk_buff *skb, 4291 struct __kernel_old_timeval *stamp) 4292 { 4293 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4294 } 4295 4296 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4297 struct __kernel_sock_timeval *stamp) 4298 { 4299 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4300 4301 stamp->tv_sec = ts.tv_sec; 4302 stamp->tv_usec = ts.tv_nsec / 1000; 4303 } 4304 4305 static inline void skb_get_timestampns(const struct sk_buff *skb, 4306 struct __kernel_old_timespec *stamp) 4307 { 4308 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4309 4310 stamp->tv_sec = ts.tv_sec; 4311 stamp->tv_nsec = ts.tv_nsec; 4312 } 4313 4314 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4315 struct __kernel_timespec *stamp) 4316 { 4317 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4318 4319 stamp->tv_sec = ts.tv_sec; 4320 stamp->tv_nsec = ts.tv_nsec; 4321 } 4322 4323 static inline void __net_timestamp(struct sk_buff *skb) 4324 { 4325 skb->tstamp = ktime_get_real(); 4326 skb->tstamp_type = SKB_CLOCK_REALTIME; 4327 } 4328 4329 static inline ktime_t net_timedelta(ktime_t t) 4330 { 4331 return ktime_sub(ktime_get_real(), t); 4332 } 4333 4334 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4335 u8 tstamp_type) 4336 { 4337 skb->tstamp = kt; 4338 4339 if (kt) 4340 skb->tstamp_type = tstamp_type; 4341 else 4342 skb->tstamp_type = SKB_CLOCK_REALTIME; 4343 } 4344 4345 static inline void skb_set_delivery_type_by_clockid(struct sk_buff *skb, 4346 ktime_t kt, clockid_t clockid) 4347 { 4348 u8 tstamp_type = SKB_CLOCK_REALTIME; 4349 4350 switch (clockid) { 4351 case CLOCK_REALTIME: 4352 break; 4353 case CLOCK_MONOTONIC: 4354 tstamp_type = SKB_CLOCK_MONOTONIC; 4355 break; 4356 case CLOCK_TAI: 4357 tstamp_type = SKB_CLOCK_TAI; 4358 break; 4359 default: 4360 WARN_ON_ONCE(1); 4361 kt = 0; 4362 } 4363 4364 skb_set_delivery_time(skb, kt, tstamp_type); 4365 } 4366 4367 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4368 4369 /* It is used in the ingress path to clear the delivery_time. 4370 * If needed, set the skb->tstamp to the (rcv) timestamp. 4371 */ 4372 static inline void skb_clear_delivery_time(struct sk_buff *skb) 4373 { 4374 if (skb->tstamp_type) { 4375 skb->tstamp_type = SKB_CLOCK_REALTIME; 4376 if (static_branch_unlikely(&netstamp_needed_key)) 4377 skb->tstamp = ktime_get_real(); 4378 else 4379 skb->tstamp = 0; 4380 } 4381 } 4382 4383 static inline void skb_clear_tstamp(struct sk_buff *skb) 4384 { 4385 if (skb->tstamp_type) 4386 return; 4387 4388 skb->tstamp = 0; 4389 } 4390 4391 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4392 { 4393 if (skb->tstamp_type) 4394 return 0; 4395 4396 return skb->tstamp; 4397 } 4398 4399 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4400 { 4401 if (skb->tstamp_type != SKB_CLOCK_MONOTONIC && skb->tstamp) 4402 return skb->tstamp; 4403 4404 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4405 return ktime_get_real(); 4406 4407 return 0; 4408 } 4409 4410 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4411 { 4412 return skb_shinfo(skb)->meta_len; 4413 } 4414 4415 static inline void *skb_metadata_end(const struct sk_buff *skb) 4416 { 4417 return skb_mac_header(skb); 4418 } 4419 4420 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4421 const struct sk_buff *skb_b, 4422 u8 meta_len) 4423 { 4424 const void *a = skb_metadata_end(skb_a); 4425 const void *b = skb_metadata_end(skb_b); 4426 u64 diffs = 0; 4427 4428 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) || 4429 BITS_PER_LONG != 64) 4430 goto slow; 4431 4432 /* Using more efficient variant than plain call to memcmp(). */ 4433 switch (meta_len) { 4434 #define __it(x, op) (x -= sizeof(u##op)) 4435 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4436 case 32: diffs |= __it_diff(a, b, 64); 4437 fallthrough; 4438 case 24: diffs |= __it_diff(a, b, 64); 4439 fallthrough; 4440 case 16: diffs |= __it_diff(a, b, 64); 4441 fallthrough; 4442 case 8: diffs |= __it_diff(a, b, 64); 4443 break; 4444 case 28: diffs |= __it_diff(a, b, 64); 4445 fallthrough; 4446 case 20: diffs |= __it_diff(a, b, 64); 4447 fallthrough; 4448 case 12: diffs |= __it_diff(a, b, 64); 4449 fallthrough; 4450 case 4: diffs |= __it_diff(a, b, 32); 4451 break; 4452 default: 4453 slow: 4454 return memcmp(a - meta_len, b - meta_len, meta_len); 4455 } 4456 return diffs; 4457 } 4458 4459 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4460 const struct sk_buff *skb_b) 4461 { 4462 u8 len_a = skb_metadata_len(skb_a); 4463 u8 len_b = skb_metadata_len(skb_b); 4464 4465 if (!(len_a | len_b)) 4466 return false; 4467 4468 return len_a != len_b ? 4469 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4470 } 4471 4472 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4473 { 4474 skb_shinfo(skb)->meta_len = meta_len; 4475 } 4476 4477 static inline void skb_metadata_clear(struct sk_buff *skb) 4478 { 4479 skb_metadata_set(skb, 0); 4480 } 4481 4482 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4483 4484 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4485 4486 void skb_clone_tx_timestamp(struct sk_buff *skb); 4487 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4488 4489 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4490 4491 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4492 { 4493 } 4494 4495 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4496 { 4497 return false; 4498 } 4499 4500 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4501 4502 /** 4503 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4504 * 4505 * PHY drivers may accept clones of transmitted packets for 4506 * timestamping via their phy_driver.txtstamp method. These drivers 4507 * must call this function to return the skb back to the stack with a 4508 * timestamp. 4509 * 4510 * @skb: clone of the original outgoing packet 4511 * @hwtstamps: hardware time stamps 4512 * 4513 */ 4514 void skb_complete_tx_timestamp(struct sk_buff *skb, 4515 struct skb_shared_hwtstamps *hwtstamps); 4516 4517 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4518 struct skb_shared_hwtstamps *hwtstamps, 4519 struct sock *sk, int tstype); 4520 4521 /** 4522 * skb_tstamp_tx - queue clone of skb with send time stamps 4523 * @orig_skb: the original outgoing packet 4524 * @hwtstamps: hardware time stamps, may be NULL if not available 4525 * 4526 * If the skb has a socket associated, then this function clones the 4527 * skb (thus sharing the actual data and optional structures), stores 4528 * the optional hardware time stamping information (if non NULL) or 4529 * generates a software time stamp (otherwise), then queues the clone 4530 * to the error queue of the socket. Errors are silently ignored. 4531 */ 4532 void skb_tstamp_tx(struct sk_buff *orig_skb, 4533 struct skb_shared_hwtstamps *hwtstamps); 4534 4535 /** 4536 * skb_tx_timestamp() - Driver hook for transmit timestamping 4537 * 4538 * Ethernet MAC Drivers should call this function in their hard_xmit() 4539 * function immediately before giving the sk_buff to the MAC hardware. 4540 * 4541 * Specifically, one should make absolutely sure that this function is 4542 * called before TX completion of this packet can trigger. Otherwise 4543 * the packet could potentially already be freed. 4544 * 4545 * @skb: A socket buffer. 4546 */ 4547 static inline void skb_tx_timestamp(struct sk_buff *skb) 4548 { 4549 skb_clone_tx_timestamp(skb); 4550 if (skb_shinfo(skb)->tx_flags & (SKBTX_SW_TSTAMP | SKBTX_BPF)) 4551 skb_tstamp_tx(skb, NULL); 4552 } 4553 4554 /** 4555 * skb_complete_wifi_ack - deliver skb with wifi status 4556 * 4557 * @skb: the original outgoing packet 4558 * @acked: ack status 4559 * 4560 */ 4561 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4562 4563 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4564 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4565 4566 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4567 { 4568 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4569 skb->csum_valid || 4570 (skb->ip_summed == CHECKSUM_PARTIAL && 4571 skb_checksum_start_offset(skb) >= 0)); 4572 } 4573 4574 /** 4575 * skb_checksum_complete - Calculate checksum of an entire packet 4576 * @skb: packet to process 4577 * 4578 * This function calculates the checksum over the entire packet plus 4579 * the value of skb->csum. The latter can be used to supply the 4580 * checksum of a pseudo header as used by TCP/UDP. It returns the 4581 * checksum. 4582 * 4583 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4584 * this function can be used to verify that checksum on received 4585 * packets. In that case the function should return zero if the 4586 * checksum is correct. In particular, this function will return zero 4587 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4588 * hardware has already verified the correctness of the checksum. 4589 */ 4590 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4591 { 4592 return skb_csum_unnecessary(skb) ? 4593 0 : __skb_checksum_complete(skb); 4594 } 4595 4596 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4597 { 4598 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4599 if (skb->csum_level == 0) 4600 skb->ip_summed = CHECKSUM_NONE; 4601 else 4602 skb->csum_level--; 4603 } 4604 } 4605 4606 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4607 { 4608 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4609 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4610 skb->csum_level++; 4611 } else if (skb->ip_summed == CHECKSUM_NONE) { 4612 skb->ip_summed = CHECKSUM_UNNECESSARY; 4613 skb->csum_level = 0; 4614 } 4615 } 4616 4617 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4618 { 4619 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4620 skb->ip_summed = CHECKSUM_NONE; 4621 skb->csum_level = 0; 4622 } 4623 } 4624 4625 /* Check if we need to perform checksum complete validation. 4626 * 4627 * Returns: true if checksum complete is needed, false otherwise 4628 * (either checksum is unnecessary or zero checksum is allowed). 4629 */ 4630 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4631 bool zero_okay, 4632 __sum16 check) 4633 { 4634 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4635 skb->csum_valid = 1; 4636 __skb_decr_checksum_unnecessary(skb); 4637 return false; 4638 } 4639 4640 return true; 4641 } 4642 4643 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4644 * in checksum_init. 4645 */ 4646 #define CHECKSUM_BREAK 76 4647 4648 /* Unset checksum-complete 4649 * 4650 * Unset checksum complete can be done when packet is being modified 4651 * (uncompressed for instance) and checksum-complete value is 4652 * invalidated. 4653 */ 4654 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4655 { 4656 if (skb->ip_summed == CHECKSUM_COMPLETE) 4657 skb->ip_summed = CHECKSUM_NONE; 4658 } 4659 4660 /* Validate (init) checksum based on checksum complete. 4661 * 4662 * Return values: 4663 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4664 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4665 * checksum is stored in skb->csum for use in __skb_checksum_complete 4666 * non-zero: value of invalid checksum 4667 * 4668 */ 4669 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4670 bool complete, 4671 __wsum psum) 4672 { 4673 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4674 if (!csum_fold(csum_add(psum, skb->csum))) { 4675 skb->csum_valid = 1; 4676 return 0; 4677 } 4678 } 4679 4680 skb->csum = psum; 4681 4682 if (complete || skb->len <= CHECKSUM_BREAK) { 4683 __sum16 csum; 4684 4685 csum = __skb_checksum_complete(skb); 4686 skb->csum_valid = !csum; 4687 return csum; 4688 } 4689 4690 return 0; 4691 } 4692 4693 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4694 { 4695 return 0; 4696 } 4697 4698 /* Perform checksum validate (init). Note that this is a macro since we only 4699 * want to calculate the pseudo header which is an input function if necessary. 4700 * First we try to validate without any computation (checksum unnecessary) and 4701 * then calculate based on checksum complete calling the function to compute 4702 * pseudo header. 4703 * 4704 * Return values: 4705 * 0: checksum is validated or try to in skb_checksum_complete 4706 * non-zero: value of invalid checksum 4707 */ 4708 #define __skb_checksum_validate(skb, proto, complete, \ 4709 zero_okay, check, compute_pseudo) \ 4710 ({ \ 4711 __sum16 __ret = 0; \ 4712 skb->csum_valid = 0; \ 4713 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4714 __ret = __skb_checksum_validate_complete(skb, \ 4715 complete, compute_pseudo(skb, proto)); \ 4716 __ret; \ 4717 }) 4718 4719 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4720 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4721 4722 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4723 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4724 4725 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4726 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4727 4728 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4729 compute_pseudo) \ 4730 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4731 4732 #define skb_checksum_simple_validate(skb) \ 4733 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4734 4735 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4736 { 4737 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4738 } 4739 4740 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4741 { 4742 skb->csum = ~pseudo; 4743 skb->ip_summed = CHECKSUM_COMPLETE; 4744 } 4745 4746 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4747 do { \ 4748 if (__skb_checksum_convert_check(skb)) \ 4749 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4750 } while (0) 4751 4752 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4753 u16 start, u16 offset) 4754 { 4755 skb->ip_summed = CHECKSUM_PARTIAL; 4756 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4757 skb->csum_offset = offset - start; 4758 } 4759 4760 /* Update skbuf and packet to reflect the remote checksum offload operation. 4761 * When called, ptr indicates the starting point for skb->csum when 4762 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4763 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4764 */ 4765 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4766 int start, int offset, bool nopartial) 4767 { 4768 __wsum delta; 4769 4770 if (!nopartial) { 4771 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4772 return; 4773 } 4774 4775 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4776 __skb_checksum_complete(skb); 4777 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4778 } 4779 4780 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4781 4782 /* Adjust skb->csum since we changed the packet */ 4783 skb->csum = csum_add(skb->csum, delta); 4784 } 4785 4786 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4787 { 4788 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4789 return (void *)(skb->_nfct & NFCT_PTRMASK); 4790 #else 4791 return NULL; 4792 #endif 4793 } 4794 4795 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4796 { 4797 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4798 return skb->_nfct; 4799 #else 4800 return 0UL; 4801 #endif 4802 } 4803 4804 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4805 { 4806 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4807 skb->slow_gro |= !!nfct; 4808 skb->_nfct = nfct; 4809 #endif 4810 } 4811 4812 #ifdef CONFIG_SKB_EXTENSIONS 4813 enum skb_ext_id { 4814 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4815 SKB_EXT_BRIDGE_NF, 4816 #endif 4817 #ifdef CONFIG_XFRM 4818 SKB_EXT_SEC_PATH, 4819 #endif 4820 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4821 TC_SKB_EXT, 4822 #endif 4823 #if IS_ENABLED(CONFIG_MPTCP) 4824 SKB_EXT_MPTCP, 4825 #endif 4826 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 4827 SKB_EXT_MCTP, 4828 #endif 4829 SKB_EXT_NUM, /* must be last */ 4830 }; 4831 4832 /** 4833 * struct skb_ext - sk_buff extensions 4834 * @refcnt: 1 on allocation, deallocated on 0 4835 * @offset: offset to add to @data to obtain extension address 4836 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4837 * @data: start of extension data, variable sized 4838 * 4839 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4840 * to use 'u8' types while allowing up to 2kb worth of extension data. 4841 */ 4842 struct skb_ext { 4843 refcount_t refcnt; 4844 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4845 u8 chunks; /* same */ 4846 char data[] __aligned(8); 4847 }; 4848 4849 struct skb_ext *__skb_ext_alloc(gfp_t flags); 4850 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 4851 struct skb_ext *ext); 4852 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4853 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4854 void __skb_ext_put(struct skb_ext *ext); 4855 4856 static inline void skb_ext_put(struct sk_buff *skb) 4857 { 4858 if (skb->active_extensions) 4859 __skb_ext_put(skb->extensions); 4860 } 4861 4862 static inline void __skb_ext_copy(struct sk_buff *dst, 4863 const struct sk_buff *src) 4864 { 4865 dst->active_extensions = src->active_extensions; 4866 4867 if (src->active_extensions) { 4868 struct skb_ext *ext = src->extensions; 4869 4870 refcount_inc(&ext->refcnt); 4871 dst->extensions = ext; 4872 } 4873 } 4874 4875 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4876 { 4877 skb_ext_put(dst); 4878 __skb_ext_copy(dst, src); 4879 } 4880 4881 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4882 { 4883 return !!ext->offset[i]; 4884 } 4885 4886 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4887 { 4888 return skb->active_extensions & (1 << id); 4889 } 4890 4891 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4892 { 4893 if (skb_ext_exist(skb, id)) 4894 __skb_ext_del(skb, id); 4895 } 4896 4897 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4898 { 4899 if (skb_ext_exist(skb, id)) { 4900 struct skb_ext *ext = skb->extensions; 4901 4902 return (void *)ext + (ext->offset[id] << 3); 4903 } 4904 4905 return NULL; 4906 } 4907 4908 static inline void skb_ext_reset(struct sk_buff *skb) 4909 { 4910 if (unlikely(skb->active_extensions)) { 4911 __skb_ext_put(skb->extensions); 4912 skb->active_extensions = 0; 4913 } 4914 } 4915 4916 static inline bool skb_has_extensions(struct sk_buff *skb) 4917 { 4918 return unlikely(skb->active_extensions); 4919 } 4920 #else 4921 static inline void skb_ext_put(struct sk_buff *skb) {} 4922 static inline void skb_ext_reset(struct sk_buff *skb) {} 4923 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4924 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4925 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4926 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 4927 #endif /* CONFIG_SKB_EXTENSIONS */ 4928 4929 static inline void nf_reset_ct(struct sk_buff *skb) 4930 { 4931 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4932 nf_conntrack_put(skb_nfct(skb)); 4933 skb->_nfct = 0; 4934 #endif 4935 } 4936 4937 static inline void nf_reset_trace(struct sk_buff *skb) 4938 { 4939 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4940 skb->nf_trace = 0; 4941 #endif 4942 } 4943 4944 static inline void ipvs_reset(struct sk_buff *skb) 4945 { 4946 #if IS_ENABLED(CONFIG_IP_VS) 4947 skb->ipvs_property = 0; 4948 #endif 4949 } 4950 4951 /* Note: This doesn't put any conntrack info in dst. */ 4952 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4953 bool copy) 4954 { 4955 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4956 dst->_nfct = src->_nfct; 4957 nf_conntrack_get(skb_nfct(src)); 4958 #endif 4959 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4960 if (copy) 4961 dst->nf_trace = src->nf_trace; 4962 #endif 4963 } 4964 4965 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 4966 { 4967 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4968 nf_conntrack_put(skb_nfct(dst)); 4969 #endif 4970 dst->slow_gro = src->slow_gro; 4971 __nf_copy(dst, src, true); 4972 } 4973 4974 #ifdef CONFIG_NETWORK_SECMARK 4975 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4976 { 4977 to->secmark = from->secmark; 4978 } 4979 4980 static inline void skb_init_secmark(struct sk_buff *skb) 4981 { 4982 skb->secmark = 0; 4983 } 4984 #else 4985 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4986 { } 4987 4988 static inline void skb_init_secmark(struct sk_buff *skb) 4989 { } 4990 #endif 4991 4992 static inline int secpath_exists(const struct sk_buff *skb) 4993 { 4994 #ifdef CONFIG_XFRM 4995 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 4996 #else 4997 return 0; 4998 #endif 4999 } 5000 5001 static inline bool skb_irq_freeable(const struct sk_buff *skb) 5002 { 5003 return !skb->destructor && 5004 !secpath_exists(skb) && 5005 !skb_nfct(skb) && 5006 !skb->_skb_refdst && 5007 !skb_has_frag_list(skb); 5008 } 5009 5010 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 5011 { 5012 skb->queue_mapping = queue_mapping; 5013 } 5014 5015 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 5016 { 5017 return skb->queue_mapping; 5018 } 5019 5020 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 5021 { 5022 to->queue_mapping = from->queue_mapping; 5023 } 5024 5025 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 5026 { 5027 skb->queue_mapping = rx_queue + 1; 5028 } 5029 5030 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 5031 { 5032 return skb->queue_mapping - 1; 5033 } 5034 5035 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 5036 { 5037 return skb->queue_mapping != 0; 5038 } 5039 5040 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 5041 { 5042 skb->dst_pending_confirm = val; 5043 } 5044 5045 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 5046 { 5047 return skb->dst_pending_confirm != 0; 5048 } 5049 5050 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 5051 { 5052 #ifdef CONFIG_XFRM 5053 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 5054 #else 5055 return NULL; 5056 #endif 5057 } 5058 5059 static inline bool skb_is_gso(const struct sk_buff *skb) 5060 { 5061 return skb_shinfo(skb)->gso_size; 5062 } 5063 5064 /* Note: Should be called only if skb_is_gso(skb) is true */ 5065 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 5066 { 5067 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 5068 } 5069 5070 /* Note: Should be called only if skb_is_gso(skb) is true */ 5071 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 5072 { 5073 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 5074 } 5075 5076 /* Note: Should be called only if skb_is_gso(skb) is true */ 5077 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 5078 { 5079 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 5080 } 5081 5082 static inline void skb_gso_reset(struct sk_buff *skb) 5083 { 5084 skb_shinfo(skb)->gso_size = 0; 5085 skb_shinfo(skb)->gso_segs = 0; 5086 skb_shinfo(skb)->gso_type = 0; 5087 } 5088 5089 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 5090 u16 increment) 5091 { 5092 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5093 return; 5094 shinfo->gso_size += increment; 5095 } 5096 5097 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 5098 u16 decrement) 5099 { 5100 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5101 return; 5102 shinfo->gso_size -= decrement; 5103 } 5104 5105 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 5106 5107 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 5108 { 5109 /* LRO sets gso_size but not gso_type, whereas if GSO is really 5110 * wanted then gso_type will be set. */ 5111 const struct skb_shared_info *shinfo = skb_shinfo(skb); 5112 5113 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 5114 unlikely(shinfo->gso_type == 0)) { 5115 __skb_warn_lro_forwarding(skb); 5116 return true; 5117 } 5118 return false; 5119 } 5120 5121 static inline void skb_forward_csum(struct sk_buff *skb) 5122 { 5123 /* Unfortunately we don't support this one. Any brave souls? */ 5124 if (skb->ip_summed == CHECKSUM_COMPLETE) 5125 skb->ip_summed = CHECKSUM_NONE; 5126 } 5127 5128 /** 5129 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 5130 * @skb: skb to check 5131 * 5132 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 5133 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 5134 * use this helper, to document places where we make this assertion. 5135 */ 5136 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 5137 { 5138 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 5139 } 5140 5141 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 5142 5143 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 5144 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 5145 unsigned int transport_len, 5146 __sum16(*skb_chkf)(struct sk_buff *skb)); 5147 5148 /** 5149 * skb_head_is_locked - Determine if the skb->head is locked down 5150 * @skb: skb to check 5151 * 5152 * The head on skbs build around a head frag can be removed if they are 5153 * not cloned. This function returns true if the skb head is locked down 5154 * due to either being allocated via kmalloc, or by being a clone with 5155 * multiple references to the head. 5156 */ 5157 static inline bool skb_head_is_locked(const struct sk_buff *skb) 5158 { 5159 return !skb->head_frag || skb_cloned(skb); 5160 } 5161 5162 /* Local Checksum Offload. 5163 * Compute outer checksum based on the assumption that the 5164 * inner checksum will be offloaded later. 5165 * See Documentation/networking/checksum-offloads.rst for 5166 * explanation of how this works. 5167 * Fill in outer checksum adjustment (e.g. with sum of outer 5168 * pseudo-header) before calling. 5169 * Also ensure that inner checksum is in linear data area. 5170 */ 5171 static inline __wsum lco_csum(struct sk_buff *skb) 5172 { 5173 unsigned char *csum_start = skb_checksum_start(skb); 5174 unsigned char *l4_hdr = skb_transport_header(skb); 5175 __wsum partial; 5176 5177 /* Start with complement of inner checksum adjustment */ 5178 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 5179 skb->csum_offset)); 5180 5181 /* Add in checksum of our headers (incl. outer checksum 5182 * adjustment filled in by caller) and return result. 5183 */ 5184 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5185 } 5186 5187 static inline bool skb_is_redirected(const struct sk_buff *skb) 5188 { 5189 return skb->redirected; 5190 } 5191 5192 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5193 { 5194 skb->redirected = 1; 5195 #ifdef CONFIG_NET_REDIRECT 5196 skb->from_ingress = from_ingress; 5197 if (skb->from_ingress) 5198 skb_clear_tstamp(skb); 5199 #endif 5200 } 5201 5202 static inline void skb_reset_redirect(struct sk_buff *skb) 5203 { 5204 skb->redirected = 0; 5205 } 5206 5207 static inline void skb_set_redirected_noclear(struct sk_buff *skb, 5208 bool from_ingress) 5209 { 5210 skb->redirected = 1; 5211 #ifdef CONFIG_NET_REDIRECT 5212 skb->from_ingress = from_ingress; 5213 #endif 5214 } 5215 5216 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5217 { 5218 #if IS_ENABLED(CONFIG_IP_SCTP) 5219 return skb->csum_not_inet; 5220 #else 5221 return 0; 5222 #endif 5223 } 5224 5225 static inline void skb_reset_csum_not_inet(struct sk_buff *skb) 5226 { 5227 skb->ip_summed = CHECKSUM_NONE; 5228 #if IS_ENABLED(CONFIG_IP_SCTP) 5229 skb->csum_not_inet = 0; 5230 #endif 5231 } 5232 5233 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5234 const u64 kcov_handle) 5235 { 5236 #ifdef CONFIG_KCOV 5237 skb->kcov_handle = kcov_handle; 5238 #endif 5239 } 5240 5241 static inline u64 skb_get_kcov_handle(struct sk_buff *skb) 5242 { 5243 #ifdef CONFIG_KCOV 5244 return skb->kcov_handle; 5245 #else 5246 return 0; 5247 #endif 5248 } 5249 5250 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5251 { 5252 #ifdef CONFIG_PAGE_POOL 5253 skb->pp_recycle = 1; 5254 #endif 5255 } 5256 5257 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter, 5258 ssize_t maxsize, gfp_t gfp); 5259 5260 #endif /* __KERNEL__ */ 5261 #endif /* _LINUX_SKBUFF_H */ 5262