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