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, <gw4pts@gw4pts.ampr.org> 7 * Florian La Roche, <rzsfl@rz.uni-sb.de> 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 net_device; 278 struct scatterlist; 279 struct pipe_inode_info; 280 struct iov_iter; 281 struct napi_struct; 282 struct bpf_prog; 283 union bpf_attr; 284 struct skb_ext; 285 struct ts_config; 286 287 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 288 struct nf_bridge_info { 289 enum { 290 BRNF_PROTO_UNCHANGED, 291 BRNF_PROTO_8021Q, 292 BRNF_PROTO_PPPOE 293 } orig_proto:8; 294 u8 pkt_otherhost:1; 295 u8 in_prerouting:1; 296 u8 bridged_dnat:1; 297 u8 sabotage_in_done:1; 298 __u16 frag_max_size; 299 int physinif; 300 301 /* always valid & non-NULL from FORWARD on, for physdev match */ 302 struct net_device *physoutdev; 303 union { 304 /* prerouting: detect dnat in orig/reply direction */ 305 __be32 ipv4_daddr; 306 struct in6_addr ipv6_daddr; 307 308 /* after prerouting + nat detected: store original source 309 * mac since neigh resolution overwrites it, only used while 310 * skb is out in neigh layer. 311 */ 312 char neigh_header[8]; 313 }; 314 }; 315 #endif 316 317 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 318 /* Chain in tc_skb_ext will be used to share the tc chain with 319 * ovs recirc_id. It will be set to the current chain by tc 320 * and read by ovs to recirc_id. 321 */ 322 struct tc_skb_ext { 323 union { 324 u64 act_miss_cookie; 325 __u32 chain; 326 }; 327 __u16 mru; 328 __u16 zone; 329 u8 post_ct:1; 330 u8 post_ct_snat:1; 331 u8 post_ct_dnat:1; 332 u8 act_miss:1; /* Set if act_miss_cookie is used */ 333 u8 l2_miss:1; /* Set by bridge upon FDB or MDB miss */ 334 }; 335 #endif 336 337 struct sk_buff_head { 338 /* These two members must be first to match sk_buff. */ 339 struct_group_tagged(sk_buff_list, list, 340 struct sk_buff *next; 341 struct sk_buff *prev; 342 ); 343 344 __u32 qlen; 345 spinlock_t lock; 346 }; 347 348 struct sk_buff; 349 350 #ifndef CONFIG_MAX_SKB_FRAGS 351 # define CONFIG_MAX_SKB_FRAGS 17 352 #endif 353 354 #define MAX_SKB_FRAGS CONFIG_MAX_SKB_FRAGS 355 356 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 357 * segment using its current segmentation instead. 358 */ 359 #define GSO_BY_FRAGS 0xFFFF 360 361 typedef struct skb_frag { 362 netmem_ref netmem; 363 unsigned int len; 364 unsigned int offset; 365 } skb_frag_t; 366 367 /** 368 * skb_frag_size() - Returns the size of a skb fragment 369 * @frag: skb fragment 370 */ 371 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 372 { 373 return frag->len; 374 } 375 376 /** 377 * skb_frag_size_set() - Sets the size of a skb fragment 378 * @frag: skb fragment 379 * @size: size of fragment 380 */ 381 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 382 { 383 frag->len = size; 384 } 385 386 /** 387 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 388 * @frag: skb fragment 389 * @delta: value to add 390 */ 391 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 392 { 393 frag->len += delta; 394 } 395 396 /** 397 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 398 * @frag: skb fragment 399 * @delta: value to subtract 400 */ 401 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 402 { 403 frag->len -= delta; 404 } 405 406 /** 407 * skb_frag_must_loop - Test if %p is a high memory page 408 * @p: fragment's page 409 */ 410 static inline bool skb_frag_must_loop(struct page *p) 411 { 412 #if defined(CONFIG_HIGHMEM) 413 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 414 return true; 415 #endif 416 return false; 417 } 418 419 /** 420 * skb_frag_foreach_page - loop over pages in a fragment 421 * 422 * @f: skb frag to operate on 423 * @f_off: offset from start of f->netmem 424 * @f_len: length from f_off to loop over 425 * @p: (temp var) current page 426 * @p_off: (temp var) offset from start of current page, 427 * non-zero only on first page. 428 * @p_len: (temp var) length in current page, 429 * < PAGE_SIZE only on first and last page. 430 * @copied: (temp var) length so far, excluding current p_len. 431 * 432 * A fragment can hold a compound page, in which case per-page 433 * operations, notably kmap_atomic, must be called for each 434 * regular page. 435 */ 436 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 437 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 438 p_off = (f_off) & (PAGE_SIZE - 1), \ 439 p_len = skb_frag_must_loop(p) ? \ 440 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 441 copied = 0; \ 442 copied < f_len; \ 443 copied += p_len, p++, p_off = 0, \ 444 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 445 446 /** 447 * struct skb_shared_hwtstamps - hardware time stamps 448 * @hwtstamp: hardware time stamp transformed into duration 449 * since arbitrary point in time 450 * @netdev_data: address/cookie of network device driver used as 451 * reference to actual hardware time stamp 452 * 453 * Software time stamps generated by ktime_get_real() are stored in 454 * skb->tstamp. 455 * 456 * hwtstamps can only be compared against other hwtstamps from 457 * the same device. 458 * 459 * This structure is attached to packets as part of the 460 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 461 */ 462 struct skb_shared_hwtstamps { 463 union { 464 ktime_t hwtstamp; 465 void *netdev_data; 466 }; 467 }; 468 469 /* Definitions for tx_flags in struct skb_shared_info */ 470 enum { 471 /* generate hardware time stamp */ 472 SKBTX_HW_TSTAMP_NOBPF = 1 << 0, 473 474 /* generate software time stamp when queueing packet to NIC */ 475 SKBTX_SW_TSTAMP = 1 << 1, 476 477 /* device driver is going to provide hardware time stamp */ 478 SKBTX_IN_PROGRESS = 1 << 2, 479 480 /* generate software time stamp on packet tx completion */ 481 SKBTX_COMPLETION_TSTAMP = 1 << 3, 482 483 /* determine hardware time stamp based on time or cycles */ 484 SKBTX_HW_TSTAMP_NETDEV = 1 << 5, 485 486 /* generate software time stamp when entering packet scheduling */ 487 SKBTX_SCHED_TSTAMP = 1 << 6, 488 489 /* used for bpf extension when a bpf program is loaded */ 490 SKBTX_BPF = 1 << 7, 491 }; 492 493 #define SKBTX_HW_TSTAMP (SKBTX_HW_TSTAMP_NOBPF | SKBTX_BPF) 494 495 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 496 SKBTX_SCHED_TSTAMP | \ 497 SKBTX_BPF | \ 498 SKBTX_COMPLETION_TSTAMP) 499 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \ 500 SKBTX_ANY_SW_TSTAMP) 501 502 /* Definitions for flags in struct skb_shared_info */ 503 enum { 504 /* use zcopy routines */ 505 SKBFL_ZEROCOPY_ENABLE = BIT(0), 506 507 /* This indicates at least one fragment might be overwritten 508 * (as in vmsplice(), sendfile() ...) 509 * If we need to compute a TX checksum, we'll need to copy 510 * all frags to avoid possible bad checksum 511 */ 512 SKBFL_SHARED_FRAG = BIT(1), 513 514 /* segment contains only zerocopy data and should not be 515 * charged to the kernel memory. 516 */ 517 SKBFL_PURE_ZEROCOPY = BIT(2), 518 519 SKBFL_DONT_ORPHAN = BIT(3), 520 521 /* page references are managed by the ubuf_info, so it's safe to 522 * use frags only up until ubuf_info is released 523 */ 524 SKBFL_MANAGED_FRAG_REFS = BIT(4), 525 }; 526 527 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG) 528 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \ 529 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS) 530 531 struct ubuf_info_ops { 532 void (*complete)(struct sk_buff *, struct ubuf_info *, 533 bool zerocopy_success); 534 /* has to be compatible with skb_zcopy_set() */ 535 int (*link_skb)(struct sk_buff *skb, struct ubuf_info *uarg); 536 }; 537 538 /* 539 * The callback notifies userspace to release buffers when skb DMA is done in 540 * lower device, the skb last reference should be 0 when calling this. 541 * The zerocopy_success argument is true if zero copy transmit occurred, 542 * false on data copy or out of memory error caused by data copy attempt. 543 * The ctx field is used to track device context. 544 * The desc field is used to track userspace buffer index. 545 */ 546 struct ubuf_info { 547 const struct ubuf_info_ops *ops; 548 refcount_t refcnt; 549 u8 flags; 550 }; 551 552 struct ubuf_info_msgzc { 553 struct ubuf_info ubuf; 554 555 union { 556 struct { 557 unsigned long desc; 558 void *ctx; 559 }; 560 struct { 561 u32 id; 562 u16 len; 563 u16 zerocopy:1; 564 u32 bytelen; 565 }; 566 }; 567 568 struct mmpin { 569 struct user_struct *user; 570 unsigned int num_pg; 571 } mmp; 572 }; 573 574 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 575 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \ 576 ubuf) 577 578 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 579 void mm_unaccount_pinned_pages(struct mmpin *mmp); 580 581 /* Preserve some data across TX submission and completion. 582 * 583 * Note, this state is stored in the driver. Extending the layout 584 * might need some special care. 585 */ 586 struct xsk_tx_metadata_compl { 587 __u64 *tx_timestamp; 588 }; 589 590 /* This data is invariant across clones and lives at 591 * the end of the header data, ie. at skb->end. 592 */ 593 struct skb_shared_info { 594 __u8 flags; 595 __u8 meta_len; 596 __u8 nr_frags; 597 __u8 tx_flags; 598 unsigned short gso_size; 599 /* Warning: this field is not always filled in (UFO)! */ 600 unsigned short gso_segs; 601 struct sk_buff *frag_list; 602 union { 603 struct skb_shared_hwtstamps hwtstamps; 604 struct xsk_tx_metadata_compl xsk_meta; 605 }; 606 unsigned int gso_type; 607 u32 tskey; 608 609 /* 610 * Warning : all fields before dataref are cleared in __alloc_skb() 611 */ 612 atomic_t dataref; 613 614 union { 615 struct { 616 u32 xdp_frags_size; 617 u32 xdp_frags_truesize; 618 }; 619 620 /* 621 * Intermediate layers must ensure that destructor_arg 622 * remains valid until skb destructor. 623 */ 624 void *destructor_arg; 625 }; 626 627 /* must be last field, see pskb_expand_head() */ 628 skb_frag_t frags[MAX_SKB_FRAGS]; 629 }; 630 631 /** 632 * DOC: dataref and headerless skbs 633 * 634 * Transport layers send out clones of payload skbs they hold for 635 * retransmissions. To allow lower layers of the stack to prepend their headers 636 * we split &skb_shared_info.dataref into two halves. 637 * The lower 16 bits count the overall number of references. 638 * The higher 16 bits indicate how many of the references are payload-only. 639 * skb_header_cloned() checks if skb is allowed to add / write the headers. 640 * 641 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr 642 * (via __skb_header_release()). Any clone created from marked skb will get 643 * &sk_buff.hdr_len populated with the available headroom. 644 * If there's the only clone in existence it's able to modify the headroom 645 * at will. The sequence of calls inside the transport layer is:: 646 * 647 * <alloc skb> 648 * skb_reserve() 649 * __skb_header_release() 650 * skb_clone() 651 * // send the clone down the stack 652 * 653 * This is not a very generic construct and it depends on the transport layers 654 * doing the right thing. In practice there's usually only one payload-only skb. 655 * Having multiple payload-only skbs with different lengths of hdr_len is not 656 * possible. The payload-only skbs should never leave their owner. 657 */ 658 #define SKB_DATAREF_SHIFT 16 659 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 660 661 662 enum { 663 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 664 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 665 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 666 }; 667 668 enum { 669 SKB_GSO_TCPV4 = 1 << 0, 670 671 /* This indicates the skb is from an untrusted source. */ 672 SKB_GSO_DODGY = 1 << 1, 673 674 /* This indicates the tcp segment has CWR set. */ 675 SKB_GSO_TCP_ECN = 1 << 2, 676 677 __SKB_GSO_TCP_FIXEDID = 1 << 3, 678 679 SKB_GSO_TCPV6 = 1 << 4, 680 681 SKB_GSO_FCOE = 1 << 5, 682 683 SKB_GSO_GRE = 1 << 6, 684 685 SKB_GSO_GRE_CSUM = 1 << 7, 686 687 SKB_GSO_IPXIP4 = 1 << 8, 688 689 SKB_GSO_IPXIP6 = 1 << 9, 690 691 SKB_GSO_UDP_TUNNEL = 1 << 10, 692 693 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 694 695 SKB_GSO_PARTIAL = 1 << 12, 696 697 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 698 699 SKB_GSO_SCTP = 1 << 14, 700 701 SKB_GSO_ESP = 1 << 15, 702 703 SKB_GSO_UDP = 1 << 16, 704 705 SKB_GSO_UDP_L4 = 1 << 17, 706 707 SKB_GSO_FRAGLIST = 1 << 18, 708 709 SKB_GSO_TCP_ACCECN = 1 << 19, 710 711 /* These indirectly map onto the same netdev feature. 712 * If NETIF_F_TSO_MANGLEID is set it may mangle both inner and outer IDs. 713 */ 714 SKB_GSO_TCP_FIXEDID = 1 << 30, 715 SKB_GSO_TCP_FIXEDID_INNER = 1 << 31, 716 }; 717 718 #if BITS_PER_LONG > 32 719 #define NET_SKBUFF_DATA_USES_OFFSET 1 720 #endif 721 722 #ifdef NET_SKBUFF_DATA_USES_OFFSET 723 typedef unsigned int sk_buff_data_t; 724 #else 725 typedef unsigned char *sk_buff_data_t; 726 #endif 727 728 enum skb_tstamp_type { 729 SKB_CLOCK_REALTIME, 730 SKB_CLOCK_MONOTONIC, 731 SKB_CLOCK_TAI, 732 __SKB_CLOCK_MAX = SKB_CLOCK_TAI, 733 }; 734 735 /** 736 * DOC: Basic sk_buff geometry 737 * 738 * struct sk_buff itself is a metadata structure and does not hold any packet 739 * data. All the data is held in associated buffers. 740 * 741 * &sk_buff.head points to the main "head" buffer. The head buffer is divided 742 * into two parts: 743 * 744 * - data buffer, containing headers and sometimes payload; 745 * this is the part of the skb operated on by the common helpers 746 * such as skb_put() or skb_pull(); 747 * - shared info (struct skb_shared_info) which holds an array of pointers 748 * to read-only data in the (page, offset, length) format. 749 * 750 * Optionally &skb_shared_info.frag_list may point to another skb. 751 * 752 * Basic diagram may look like this:: 753 * 754 * --------------- 755 * | sk_buff | 756 * --------------- 757 * ,--------------------------- + head 758 * / ,----------------- + data 759 * / / ,----------- + tail 760 * | | | , + end 761 * | | | | 762 * v v v v 763 * ----------------------------------------------- 764 * | headroom | data | tailroom | skb_shared_info | 765 * ----------------------------------------------- 766 * + [page frag] 767 * + [page frag] 768 * + [page frag] 769 * + [page frag] --------- 770 * + frag_list --> | sk_buff | 771 * --------- 772 * 773 */ 774 775 /** 776 * struct sk_buff - socket buffer 777 * @next: Next buffer in list 778 * @prev: Previous buffer in list 779 * @tstamp: Time we arrived/left 780 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 781 * for retransmit timer 782 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 783 * @list: queue head 784 * @ll_node: anchor in an llist (eg socket defer_list) 785 * @sk: Socket we are owned by 786 * @dev: Device we arrived on/are leaving by 787 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 788 * @cb: Control buffer. Free for use by every layer. Put private vars here 789 * @_skb_refdst: destination entry (with norefcount bit) 790 * @len: Length of actual data 791 * @data_len: Data length 792 * @mac_len: Length of link layer header 793 * @hdr_len: writable header length of cloned skb 794 * @csum: Checksum (must include start/offset pair) 795 * @csum_start: Offset from skb->head where checksumming should start 796 * @csum_offset: Offset from csum_start where checksum should be stored 797 * @priority: Packet queueing priority 798 * @ignore_df: allow local fragmentation 799 * @cloned: Head may be cloned (check refcnt to be sure) 800 * @ip_summed: Driver fed us an IP checksum 801 * @nohdr: Payload reference only, must not modify header 802 * @pkt_type: Packet class 803 * @fclone: skbuff clone status 804 * @ipvs_property: skbuff is owned by ipvs 805 * @inner_protocol_type: whether the inner protocol is 806 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 807 * @remcsum_offload: remote checksum offload is enabled 808 * @offload_fwd_mark: Packet was L2-forwarded in hardware 809 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 810 * @tc_skip_classify: do not classify packet. set by IFB device 811 * @tc_at_ingress: used within tc_classify to distinguish in/egress 812 * @redirected: packet was redirected by packet classifier 813 * @from_ingress: packet was redirected from the ingress path 814 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 815 * @peeked: this packet has been seen already, so stats have been 816 * done for it, don't do them again 817 * @nf_trace: netfilter packet trace flag 818 * @protocol: Packet protocol from driver 819 * @destructor: Destruct function 820 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 821 * @_sk_redir: socket redirection information for skmsg 822 * @_nfct: Associated connection, if any (with nfctinfo bits) 823 * @skb_iif: ifindex of device we arrived on 824 * @tc_depth: counter for packet duplication 825 * @tc_index: Traffic control index 826 * @hash: the packet hash 827 * @queue_mapping: Queue mapping for multiqueue devices 828 * @head_frag: skb was allocated from page fragments, 829 * not allocated by kmalloc() or vmalloc(). 830 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 831 * @pp_recycle: mark the packet for recycling instead of freeing (implies 832 * page_pool support on driver) 833 * @active_extensions: active extensions (skb_ext_id types) 834 * @ndisc_nodetype: router type (from link layer) 835 * @ooo_okay: allow the mapping of a socket to a queue to be changed 836 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 837 * ports. 838 * @sw_hash: indicates hash was computed in software stack 839 * @wifi_acked_valid: wifi_acked was set 840 * @wifi_acked: whether frame was acked on wifi or not 841 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 842 * @encapsulation: indicates the inner headers in the skbuff are valid 843 * @encap_hdr_csum: software checksum is needed 844 * @csum_valid: checksum is already valid 845 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 846 * @csum_complete_sw: checksum was completed by software 847 * @csum_level: indicates the number of consecutive checksums found in 848 * the packet minus one that have been verified as 849 * CHECKSUM_UNNECESSARY (max 3) 850 * @unreadable: indicates that at least 1 of the fragments in this skb is 851 * unreadable. 852 * @dst_pending_confirm: need to confirm neighbour 853 * @decrypted: Decrypted SKB 854 * @slow_gro: state present at GRO time, slower prepare step required 855 * @tstamp_type: When set, skb->tstamp has the 856 * delivery_time clock base of skb->tstamp. 857 * @napi_id: id of the NAPI struct this skb came from 858 * @sender_cpu: (aka @napi_id) source CPU in XPS 859 * @alloc_cpu: CPU which did the skb allocation. 860 * @secmark: security marking 861 * @mark: Generic packet mark 862 * @reserved_tailroom: (aka @mark) number of bytes of free space available 863 * at the tail of an sk_buff 864 * @vlan_all: vlan fields (proto & tci) 865 * @vlan_proto: vlan encapsulation protocol 866 * @vlan_tci: vlan tag control information 867 * @inner_protocol: Protocol (encapsulation) 868 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 869 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 870 * @inner_transport_header: Inner transport layer header (encapsulation) 871 * @inner_network_header: Network layer header (encapsulation) 872 * @inner_mac_header: Link layer header (encapsulation) 873 * @transport_header: Transport layer header 874 * @network_header: Network layer header 875 * @mac_header: Link layer header 876 * @kcov_handle: KCOV remote handle for remote coverage collection 877 * @tail: Tail pointer 878 * @end: End pointer 879 * @head: Head of buffer 880 * @data: Data head pointer 881 * @truesize: Buffer size 882 * @users: User count - see {datagram,tcp}.c 883 * @extensions: allocated extensions, valid if active_extensions is nonzero 884 */ 885 886 struct sk_buff { 887 union { 888 struct { 889 /* These two members must be first to match sk_buff_head. */ 890 struct sk_buff *next; 891 struct sk_buff *prev; 892 893 union { 894 struct net_device *dev; 895 /* Some protocols might use this space to store information, 896 * while device pointer would be NULL. 897 * UDP receive path is one user. 898 */ 899 unsigned long dev_scratch; 900 }; 901 }; 902 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 903 struct list_head list; 904 struct llist_node ll_node; 905 }; 906 907 struct sock *sk; 908 909 union { 910 ktime_t tstamp; 911 u64 skb_mstamp_ns; /* earliest departure time */ 912 }; 913 /* 914 * This is the control buffer. It is free to use for every 915 * layer. Please put your private variables there. If you 916 * want to keep them across layers you have to do a skb_clone() 917 * first. This is owned by whoever has the skb queued ATM. 918 */ 919 char cb[48] __aligned(8); 920 921 union { 922 struct { 923 unsigned long _skb_refdst; 924 void (*destructor)(struct sk_buff *skb); 925 }; 926 struct list_head tcp_tsorted_anchor; 927 #ifdef CONFIG_NET_SOCK_MSG 928 unsigned long _sk_redir; 929 #endif 930 }; 931 932 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 933 unsigned long _nfct; 934 #endif 935 unsigned int len, 936 data_len; 937 __u16 mac_len, 938 hdr_len; 939 940 /* Following fields are _not_ copied in __copy_skb_header() 941 * Note that queue_mapping is here mostly to fill a hole. 942 */ 943 __u16 queue_mapping; 944 945 /* if you move cloned around you also must adapt those constants */ 946 #ifdef __BIG_ENDIAN_BITFIELD 947 #define CLONED_MASK (1 << 7) 948 #else 949 #define CLONED_MASK 1 950 #endif 951 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 952 953 /* private: */ 954 __u8 __cloned_offset[0]; 955 /* public: */ 956 __u8 cloned:1, 957 nohdr:1, 958 fclone:2, 959 peeked:1, 960 head_frag:1, 961 pfmemalloc:1, 962 pp_recycle:1; /* page_pool recycle indicator */ 963 #ifdef CONFIG_SKB_EXTENSIONS 964 __u8 active_extensions; 965 #endif 966 967 /* Fields enclosed in headers group are copied 968 * using a single memcpy() in __copy_skb_header() 969 */ 970 struct_group(headers, 971 972 /* private: */ 973 __u8 __pkt_type_offset[0]; 974 /* public: */ 975 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 976 __u8 ignore_df:1; 977 __u8 dst_pending_confirm:1; 978 __u8 ip_summed:2; 979 __u8 ooo_okay:1; 980 981 /* private: */ 982 __u8 __mono_tc_offset[0]; 983 /* public: */ 984 __u8 tstamp_type:2; /* See skb_tstamp_type */ 985 #ifdef CONFIG_NET_XGRESS 986 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 987 __u8 tc_skip_classify:1; 988 #endif 989 __u8 remcsum_offload:1; 990 __u8 csum_complete_sw:1; 991 __u8 csum_level:2; 992 __u8 inner_protocol_type:1; 993 994 __u8 l4_hash:1; 995 __u8 sw_hash:1; 996 #ifdef CONFIG_WIRELESS 997 __u8 wifi_acked_valid:1; 998 __u8 wifi_acked:1; 999 #endif 1000 __u8 no_fcs:1; 1001 /* Indicates the inner headers are valid in the skbuff. */ 1002 __u8 encapsulation:1; 1003 __u8 encap_hdr_csum:1; 1004 __u8 csum_valid:1; 1005 #ifdef CONFIG_IPV6_NDISC_NODETYPE 1006 __u8 ndisc_nodetype:2; 1007 #endif 1008 1009 #if IS_ENABLED(CONFIG_IP_VS) 1010 __u8 ipvs_property:1; 1011 #endif 1012 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 1013 __u8 nf_trace:1; 1014 #endif 1015 #ifdef CONFIG_NET_SWITCHDEV 1016 __u8 offload_fwd_mark:1; 1017 __u8 offload_l3_fwd_mark:1; 1018 #endif 1019 __u8 redirected:1; 1020 #ifdef CONFIG_NET_REDIRECT 1021 __u8 from_ingress:1; 1022 #endif 1023 #ifdef CONFIG_NETFILTER_SKIP_EGRESS 1024 __u8 nf_skip_egress:1; 1025 #endif 1026 #ifdef CONFIG_SKB_DECRYPTED 1027 __u8 decrypted:1; 1028 #endif 1029 __u8 slow_gro:1; 1030 #if IS_ENABLED(CONFIG_IP_SCTP) 1031 __u8 csum_not_inet:1; 1032 #endif 1033 __u8 unreadable:1; 1034 __u8 tc_depth:2; 1035 #if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS) 1036 __u16 tc_index; /* traffic control index */ 1037 #endif 1038 1039 u16 alloc_cpu; 1040 1041 union { 1042 __wsum csum; 1043 struct { 1044 __u16 csum_start; 1045 __u16 csum_offset; 1046 }; 1047 }; 1048 __u32 priority; 1049 int skb_iif; 1050 __u32 hash; 1051 union { 1052 u32 vlan_all; 1053 struct { 1054 __be16 vlan_proto; 1055 __u16 vlan_tci; 1056 }; 1057 }; 1058 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1059 union { 1060 unsigned int napi_id; 1061 unsigned int sender_cpu; 1062 }; 1063 #endif 1064 #ifdef CONFIG_NETWORK_SECMARK 1065 __u32 secmark; 1066 #endif 1067 1068 union { 1069 __u32 mark; 1070 __u32 reserved_tailroom; 1071 }; 1072 1073 union { 1074 __be16 inner_protocol; 1075 __u8 inner_ipproto; 1076 }; 1077 1078 __u16 inner_transport_header; 1079 __u16 inner_network_header; 1080 __u16 inner_mac_header; 1081 1082 __be16 protocol; 1083 __u16 transport_header; 1084 __u16 network_header; 1085 __u16 mac_header; 1086 1087 struct kcov_common_handle_id kcov_handle; 1088 1089 ); /* end headers group */ 1090 1091 /* These elements must be at the end, see alloc_skb() for details. */ 1092 sk_buff_data_t tail; 1093 sk_buff_data_t end; 1094 unsigned char *head, 1095 *data; 1096 unsigned int truesize; 1097 refcount_t users; 1098 1099 #ifdef CONFIG_SKB_EXTENSIONS 1100 /* only usable after checking ->active_extensions != 0 */ 1101 struct skb_ext *extensions; 1102 #endif 1103 }; 1104 1105 /* if you move pkt_type around you also must adapt those constants */ 1106 #ifdef __BIG_ENDIAN_BITFIELD 1107 #define PKT_TYPE_MAX (7 << 5) 1108 #else 1109 #define PKT_TYPE_MAX 7 1110 #endif 1111 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1112 1113 /* if you move tc_at_ingress or tstamp_type 1114 * around, you also must adapt these constants. 1115 */ 1116 #ifdef __BIG_ENDIAN_BITFIELD 1117 #define SKB_TSTAMP_TYPE_MASK (3 << 6) 1118 #define SKB_TSTAMP_TYPE_RSHIFT (6) 1119 #define TC_AT_INGRESS_MASK (1 << 5) 1120 #else 1121 #define SKB_TSTAMP_TYPE_MASK (3) 1122 #define TC_AT_INGRESS_MASK (1 << 2) 1123 #endif 1124 #define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset) 1125 1126 #ifdef __KERNEL__ 1127 /* 1128 * Handling routines are only of interest to the kernel 1129 */ 1130 1131 #define SKB_ALLOC_FCLONE 0x01 1132 #define SKB_ALLOC_RX 0x02 1133 #define SKB_ALLOC_NAPI 0x04 1134 1135 /** 1136 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1137 * @skb: buffer 1138 */ 1139 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1140 { 1141 return unlikely(skb->pfmemalloc); 1142 } 1143 1144 /* 1145 * skb might have a dst pointer attached, refcounted or not. 1146 * _skb_refdst low order bit is set if refcount was _not_ taken 1147 */ 1148 #define SKB_DST_NOREF 1UL 1149 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1150 1151 /** 1152 * skb_dst - returns skb dst_entry 1153 * @skb: buffer 1154 * 1155 * Returns: skb dst_entry, regardless of reference taken or not. 1156 */ 1157 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1158 { 1159 /* If refdst was not refcounted, check we still are in a 1160 * rcu_read_lock section 1161 */ 1162 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1163 !rcu_read_lock_held() && 1164 !rcu_read_lock_bh_held()); 1165 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1166 } 1167 1168 static inline void skb_dst_check_unset(struct sk_buff *skb) 1169 { 1170 DEBUG_NET_WARN_ON_ONCE((skb->_skb_refdst & SKB_DST_PTRMASK) && 1171 !(skb->_skb_refdst & SKB_DST_NOREF)); 1172 } 1173 1174 /** 1175 * skb_dstref_steal() - return current dst_entry value and clear it 1176 * @skb: buffer 1177 * 1178 * Resets skb dst_entry without adjusting its reference count. Useful in 1179 * cases where dst_entry needs to be temporarily reset and restored. 1180 * Note that the returned value cannot be used directly because it 1181 * might contain SKB_DST_NOREF bit. 1182 * 1183 * When in doubt, prefer skb_dst_drop() over skb_dstref_steal() to correctly 1184 * handle dst_entry reference counting. 1185 * 1186 * Returns: original skb dst_entry. 1187 */ 1188 static inline unsigned long skb_dstref_steal(struct sk_buff *skb) 1189 { 1190 unsigned long refdst = skb->_skb_refdst; 1191 1192 skb->_skb_refdst = 0; 1193 return refdst; 1194 } 1195 1196 /** 1197 * skb_dstref_restore() - restore skb dst_entry removed via skb_dstref_steal() 1198 * @skb: buffer 1199 * @refdst: dst entry from a call to skb_dstref_steal() 1200 */ 1201 static inline void skb_dstref_restore(struct sk_buff *skb, unsigned long refdst) 1202 { 1203 skb_dst_check_unset(skb); 1204 skb->_skb_refdst = refdst; 1205 } 1206 1207 /** 1208 * skb_dst_set - sets skb dst 1209 * @skb: buffer 1210 * @dst: dst entry 1211 * 1212 * Sets skb dst, assuming a reference was taken on dst and should 1213 * be released by skb_dst_drop() 1214 */ 1215 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1216 { 1217 skb_dst_check_unset(skb); 1218 skb->slow_gro |= !!dst; 1219 skb->_skb_refdst = (unsigned long)dst; 1220 } 1221 1222 /** 1223 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1224 * @skb: buffer 1225 * @dst: dst entry 1226 * 1227 * Sets skb dst, assuming a reference was not taken on dst. 1228 * If dst entry is cached, we do not take reference and dst_release 1229 * will be avoided by refdst_drop. If dst entry is not cached, we take 1230 * reference, so that last dst_release can destroy the dst immediately. 1231 */ 1232 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1233 { 1234 skb_dst_check_unset(skb); 1235 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1236 skb->slow_gro |= !!dst; 1237 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1238 } 1239 1240 /** 1241 * skb_dst_is_noref - Test if skb dst isn't refcounted 1242 * @skb: buffer 1243 */ 1244 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1245 { 1246 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1247 } 1248 1249 /* For mangling skb->pkt_type from user space side from applications 1250 * such as nft, tc, etc, we only allow a conservative subset of 1251 * possible pkt_types to be set. 1252 */ 1253 static inline bool skb_pkt_type_ok(u32 ptype) 1254 { 1255 return ptype <= PACKET_OTHERHOST; 1256 } 1257 1258 /** 1259 * skb_napi_id - Returns the skb's NAPI id 1260 * @skb: buffer 1261 */ 1262 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1263 { 1264 #ifdef CONFIG_NET_RX_BUSY_POLL 1265 return skb->napi_id; 1266 #else 1267 return 0; 1268 #endif 1269 } 1270 1271 static inline bool skb_wifi_acked_valid(const struct sk_buff *skb) 1272 { 1273 #ifdef CONFIG_WIRELESS 1274 return skb->wifi_acked_valid; 1275 #else 1276 return 0; 1277 #endif 1278 } 1279 1280 /** 1281 * skb_unref - decrement the skb's reference count 1282 * @skb: buffer 1283 * 1284 * Returns: true if we can free the skb. 1285 */ 1286 static inline bool skb_unref(struct sk_buff *skb) 1287 { 1288 if (unlikely(!skb)) 1289 return false; 1290 if (!IS_ENABLED(CONFIG_DEBUG_NET) && likely(refcount_read(&skb->users) == 1)) 1291 smp_rmb(); 1292 else if (likely(!refcount_dec_and_test(&skb->users))) 1293 return false; 1294 1295 return true; 1296 } 1297 1298 static inline bool skb_data_unref(const struct sk_buff *skb, 1299 struct skb_shared_info *shinfo) 1300 { 1301 int bias; 1302 1303 if (!skb->cloned) 1304 return true; 1305 1306 bias = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1; 1307 1308 if (atomic_read(&shinfo->dataref) == bias) 1309 smp_rmb(); 1310 else if (atomic_sub_return(bias, &shinfo->dataref)) 1311 return false; 1312 1313 return true; 1314 } 1315 1316 void __fix_address sk_skb_reason_drop(const struct sock *sk, 1317 struct sk_buff *skb, 1318 enum skb_drop_reason reason); 1319 1320 static inline void 1321 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason) 1322 { 1323 sk_skb_reason_drop(NULL, skb, reason); 1324 } 1325 1326 /** 1327 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1328 * @skb: buffer to free 1329 */ 1330 static inline void kfree_skb(struct sk_buff *skb) 1331 { 1332 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1333 } 1334 1335 void skb_release_head_state(struct sk_buff *skb); 1336 void kfree_skb_list_reason(struct sk_buff *segs, 1337 enum skb_drop_reason reason); 1338 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1339 void skb_tx_error(struct sk_buff *skb); 1340 1341 static inline void kfree_skb_list(struct sk_buff *segs) 1342 { 1343 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1344 } 1345 1346 #ifdef CONFIG_TRACEPOINTS 1347 void consume_skb(struct sk_buff *skb); 1348 #else 1349 static inline void consume_skb(struct sk_buff *skb) 1350 { 1351 return kfree_skb(skb); 1352 } 1353 #endif 1354 1355 void __consume_stateless_skb(struct sk_buff *skb); 1356 void __kfree_skb(struct sk_buff *skb); 1357 1358 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1359 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1360 bool *fragstolen, int *delta_truesize); 1361 1362 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1363 int node); 1364 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1365 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1366 struct sk_buff *build_skb_around(struct sk_buff *skb, 1367 void *data, unsigned int frag_size); 1368 void skb_attempt_defer_free(struct sk_buff *skb); 1369 1370 u32 napi_skb_cache_get_bulk(void **skbs, u32 n); 1371 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1372 struct sk_buff *slab_build_skb(void *data); 1373 1374 /** 1375 * alloc_skb - allocate a network buffer 1376 * @size: size to allocate 1377 * @priority: allocation mask 1378 * 1379 * This function is a convenient wrapper around __alloc_skb(). 1380 */ 1381 static inline struct sk_buff *alloc_skb(unsigned int size, 1382 gfp_t priority) 1383 { 1384 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1385 } 1386 1387 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1388 unsigned long data_len, 1389 int max_page_order, 1390 int *errcode, 1391 gfp_t gfp_mask); 1392 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1393 1394 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1395 struct sk_buff_fclones { 1396 struct sk_buff skb1; 1397 1398 struct sk_buff skb2; 1399 1400 refcount_t fclone_ref; 1401 }; 1402 1403 /** 1404 * skb_fclone_busy - check if fclone is busy 1405 * @sk: socket 1406 * @skb: buffer 1407 * 1408 * Returns: true if skb is a fast clone, and its clone is not freed. 1409 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1410 * so we also check that didn't happen. 1411 */ 1412 static inline bool skb_fclone_busy(const struct sock *sk, 1413 const struct sk_buff *skb) 1414 { 1415 const struct sk_buff_fclones *fclones; 1416 1417 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1418 1419 return skb->fclone == SKB_FCLONE_ORIG && 1420 refcount_read(&fclones->fclone_ref) > 1 && 1421 READ_ONCE(fclones->skb2.sk) == sk; 1422 } 1423 1424 /** 1425 * alloc_skb_fclone - allocate a network buffer from fclone cache 1426 * @size: size to allocate 1427 * @priority: allocation mask 1428 * 1429 * This function is a convenient wrapper around __alloc_skb(). 1430 */ 1431 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1432 gfp_t priority) 1433 { 1434 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1435 } 1436 1437 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1438 void skb_headers_offset_update(struct sk_buff *skb, int off); 1439 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1440 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1441 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1442 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1443 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1444 gfp_t gfp_mask, bool fclone); 1445 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1446 gfp_t gfp_mask) 1447 { 1448 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1449 } 1450 1451 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1452 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1453 unsigned int headroom); 1454 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1455 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1456 int newtailroom, gfp_t priority); 1457 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1458 int offset, int len); 1459 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1460 int offset, int len); 1461 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1462 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1463 1464 /** 1465 * skb_pad - zero pad the tail of an skb 1466 * @skb: buffer to pad 1467 * @pad: space to pad 1468 * 1469 * Ensure that a buffer is followed by a padding area that is zero 1470 * filled. Used by network drivers which may DMA or transfer data 1471 * beyond the buffer end onto the wire. 1472 * 1473 * May return error in out of memory cases. The skb is freed on error. 1474 */ 1475 static inline int skb_pad(struct sk_buff *skb, int pad) 1476 { 1477 return __skb_pad(skb, pad, true); 1478 } 1479 #define dev_kfree_skb(a) consume_skb(a) 1480 1481 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1482 int offset, size_t size, size_t max_frags); 1483 1484 struct skb_seq_state { 1485 __u32 lower_offset; 1486 __u32 upper_offset; 1487 __u32 frag_idx; 1488 __u32 stepped_offset; 1489 struct sk_buff *root_skb; 1490 struct sk_buff *cur_skb; 1491 __u8 *frag_data; 1492 __u32 frag_off; 1493 }; 1494 1495 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1496 unsigned int to, struct skb_seq_state *st); 1497 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1498 struct skb_seq_state *st); 1499 void skb_abort_seq_read(struct skb_seq_state *st); 1500 int skb_copy_seq_read(struct skb_seq_state *st, int offset, void *to, int len); 1501 1502 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1503 unsigned int to, struct ts_config *config); 1504 1505 /* 1506 * Packet hash types specify the type of hash in skb_set_hash. 1507 * 1508 * Hash types refer to the protocol layer addresses which are used to 1509 * construct a packet's hash. The hashes are used to differentiate or identify 1510 * flows of the protocol layer for the hash type. Hash types are either 1511 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1512 * 1513 * Properties of hashes: 1514 * 1515 * 1) Two packets in different flows have different hash values 1516 * 2) Two packets in the same flow should have the same hash value 1517 * 1518 * A hash at a higher layer is considered to be more specific. A driver should 1519 * set the most specific hash possible. 1520 * 1521 * A driver cannot indicate a more specific hash than the layer at which a hash 1522 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1523 * 1524 * A driver may indicate a hash level which is less specific than the 1525 * actual layer the hash was computed on. For instance, a hash computed 1526 * at L4 may be considered an L3 hash. This should only be done if the 1527 * driver can't unambiguously determine that the HW computed the hash at 1528 * the higher layer. Note that the "should" in the second property above 1529 * permits this. 1530 */ 1531 enum pkt_hash_types { 1532 PKT_HASH_TYPE_NONE, /* Undefined type */ 1533 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1534 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1535 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1536 }; 1537 1538 static inline void skb_clear_hash(struct sk_buff *skb) 1539 { 1540 skb->hash = 0; 1541 skb->sw_hash = 0; 1542 skb->l4_hash = 0; 1543 } 1544 1545 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1546 { 1547 if (!skb->l4_hash) 1548 skb_clear_hash(skb); 1549 } 1550 1551 static inline void 1552 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1553 { 1554 skb->l4_hash = is_l4; 1555 skb->sw_hash = is_sw; 1556 skb->hash = hash; 1557 } 1558 1559 static inline void 1560 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1561 { 1562 /* Used by drivers to set hash from HW */ 1563 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1564 } 1565 1566 static inline void 1567 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1568 { 1569 __skb_set_hash(skb, hash, true, is_l4); 1570 } 1571 1572 u32 __skb_get_hash_symmetric_net(const struct net *net, const struct sk_buff *skb); 1573 1574 static inline u32 __skb_get_hash_symmetric(const struct sk_buff *skb) 1575 { 1576 return __skb_get_hash_symmetric_net(NULL, skb); 1577 } 1578 1579 void __skb_get_hash_net(const struct net *net, struct sk_buff *skb); 1580 u32 skb_get_poff(const struct sk_buff *skb); 1581 u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1582 const struct flow_keys_basic *keys, int hlen); 1583 __be32 skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1584 const void *data, int hlen_proto); 1585 1586 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1587 const struct flow_dissector_key *key, 1588 unsigned int key_count); 1589 1590 struct bpf_flow_dissector; 1591 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1592 __be16 proto, int nhoff, int hlen, unsigned int flags); 1593 1594 bool __skb_flow_dissect(const struct net *net, 1595 const struct sk_buff *skb, 1596 struct flow_dissector *flow_dissector, 1597 void *target_container, const void *data, 1598 __be16 proto, int nhoff, int hlen, unsigned int flags); 1599 1600 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1601 struct flow_dissector *flow_dissector, 1602 void *target_container, unsigned int flags) 1603 { 1604 return __skb_flow_dissect(NULL, skb, flow_dissector, 1605 target_container, NULL, 0, 0, 0, flags); 1606 } 1607 1608 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1609 struct flow_keys *flow, 1610 unsigned int flags) 1611 { 1612 memset(flow, 0, sizeof(*flow)); 1613 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1614 flow, NULL, 0, 0, 0, flags); 1615 } 1616 1617 static inline bool 1618 skb_flow_dissect_flow_keys_basic(const struct net *net, 1619 const struct sk_buff *skb, 1620 struct flow_keys_basic *flow, 1621 const void *data, __be16 proto, 1622 int nhoff, int hlen, unsigned int flags) 1623 { 1624 memset(flow, 0, sizeof(*flow)); 1625 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1626 data, proto, nhoff, hlen, flags); 1627 } 1628 1629 void skb_flow_dissect_meta(const struct sk_buff *skb, 1630 struct flow_dissector *flow_dissector, 1631 void *target_container); 1632 1633 /* Gets a skb connection tracking info, ctinfo map should be a 1634 * map of mapsize to translate enum ip_conntrack_info states 1635 * to user states. 1636 */ 1637 void 1638 skb_flow_dissect_ct(const struct sk_buff *skb, 1639 struct flow_dissector *flow_dissector, 1640 void *target_container, 1641 u16 *ctinfo_map, size_t mapsize, 1642 bool post_ct, u16 zone); 1643 void 1644 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1645 struct flow_dissector *flow_dissector, 1646 void *target_container); 1647 1648 void skb_flow_dissect_hash(const struct sk_buff *skb, 1649 struct flow_dissector *flow_dissector, 1650 void *target_container); 1651 1652 static inline __u32 skb_get_hash_net(const struct net *net, struct sk_buff *skb) 1653 { 1654 if (!skb->l4_hash && !skb->sw_hash) 1655 __skb_get_hash_net(net, skb); 1656 1657 return skb->hash; 1658 } 1659 1660 static inline __u32 skb_get_hash(struct sk_buff *skb) 1661 { 1662 if (!skb->l4_hash && !skb->sw_hash) 1663 __skb_get_hash_net(NULL, skb); 1664 1665 return skb->hash; 1666 } 1667 1668 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1669 { 1670 if (!skb->l4_hash && !skb->sw_hash) { 1671 struct flow_keys keys; 1672 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1673 1674 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1675 } 1676 1677 return skb->hash; 1678 } 1679 1680 __u32 skb_get_hash_perturb(const struct sk_buff *skb, 1681 const siphash_key_t *perturb); 1682 1683 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1684 { 1685 return skb->hash; 1686 } 1687 1688 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1689 { 1690 to->hash = from->hash; 1691 to->sw_hash = from->sw_hash; 1692 to->l4_hash = from->l4_hash; 1693 }; 1694 1695 static inline int skb_cmp_decrypted(const struct sk_buff *skb1, 1696 const struct sk_buff *skb2) 1697 { 1698 #ifdef CONFIG_SKB_DECRYPTED 1699 return skb2->decrypted - skb1->decrypted; 1700 #else 1701 return 0; 1702 #endif 1703 } 1704 1705 static inline bool skb_is_decrypted(const struct sk_buff *skb) 1706 { 1707 #ifdef CONFIG_SKB_DECRYPTED 1708 return skb->decrypted; 1709 #else 1710 return false; 1711 #endif 1712 } 1713 1714 static inline void skb_copy_decrypted(struct sk_buff *to, 1715 const struct sk_buff *from) 1716 { 1717 #ifdef CONFIG_SKB_DECRYPTED 1718 to->decrypted = from->decrypted; 1719 #endif 1720 } 1721 1722 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1723 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1724 { 1725 return skb->head + skb->end; 1726 } 1727 1728 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1729 { 1730 return skb->end; 1731 } 1732 1733 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1734 { 1735 skb->end = offset; 1736 } 1737 #else 1738 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1739 { 1740 return skb->end; 1741 } 1742 1743 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1744 { 1745 return skb->end - skb->head; 1746 } 1747 1748 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1749 { 1750 skb->end = skb->head + offset; 1751 } 1752 #endif 1753 1754 extern const struct ubuf_info_ops msg_zerocopy_ubuf_ops; 1755 1756 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1757 struct ubuf_info *uarg, bool devmem); 1758 1759 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1760 1761 struct net_devmem_dmabuf_binding; 1762 1763 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, 1764 struct sk_buff *skb, struct iov_iter *from, 1765 size_t length, 1766 struct net_devmem_dmabuf_binding *binding); 1767 1768 int zerocopy_fill_skb_from_iter(struct sk_buff *skb, 1769 struct iov_iter *from, size_t length); 1770 1771 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1772 struct msghdr *msg, int len) 1773 { 1774 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len, 1775 NULL); 1776 } 1777 1778 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1779 struct msghdr *msg, int len, 1780 struct ubuf_info *uarg, 1781 struct net_devmem_dmabuf_binding *binding); 1782 1783 /* Internal */ 1784 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1785 1786 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1787 { 1788 return &skb_shinfo(skb)->hwtstamps; 1789 } 1790 1791 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1792 { 1793 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1794 1795 return is_zcopy ? skb_uarg(skb) : NULL; 1796 } 1797 1798 static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1799 { 1800 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1801 } 1802 1803 static inline bool skb_zcopy_managed(const struct sk_buff *skb) 1804 { 1805 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS; 1806 } 1807 1808 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1809 const struct sk_buff *skb2) 1810 { 1811 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1812 } 1813 1814 static inline void net_zcopy_get(struct ubuf_info *uarg) 1815 { 1816 refcount_inc(&uarg->refcnt); 1817 } 1818 1819 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1820 { 1821 skb_shinfo(skb)->destructor_arg = uarg; 1822 skb_shinfo(skb)->flags |= uarg->flags; 1823 } 1824 1825 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1826 bool *have_ref) 1827 { 1828 if (skb && uarg && !skb_zcopy(skb)) { 1829 if (unlikely(have_ref && *have_ref)) 1830 *have_ref = false; 1831 else 1832 net_zcopy_get(uarg); 1833 skb_zcopy_init(skb, uarg); 1834 } 1835 } 1836 1837 static inline void net_zcopy_put(struct ubuf_info *uarg) 1838 { 1839 if (uarg) 1840 uarg->ops->complete(NULL, uarg, true); 1841 } 1842 1843 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1844 { 1845 if (uarg) { 1846 if (uarg->ops == &msg_zerocopy_ubuf_ops) 1847 msg_zerocopy_put_abort(uarg, have_uref); 1848 else if (have_uref) 1849 net_zcopy_put(uarg); 1850 } 1851 } 1852 1853 /* Release a reference on a zerocopy structure */ 1854 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1855 { 1856 struct ubuf_info *uarg = skb_zcopy(skb); 1857 1858 if (uarg) { 1859 uarg->ops->complete(skb, uarg, zerocopy_success); 1860 1861 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1862 } 1863 } 1864 1865 void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1866 1867 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1868 { 1869 if (unlikely(skb_zcopy_managed(skb))) 1870 __skb_zcopy_downgrade_managed(skb); 1871 } 1872 1873 /* Return true if frags in this skb are readable by the host. */ 1874 static inline bool skb_frags_readable(const struct sk_buff *skb) 1875 { 1876 return !skb->unreadable; 1877 } 1878 1879 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1880 { 1881 skb->next = NULL; 1882 } 1883 1884 static inline void skb_poison_list(struct sk_buff *skb) 1885 { 1886 #ifdef CONFIG_DEBUG_NET 1887 skb->next = SKB_LIST_POISON_NEXT; 1888 #endif 1889 } 1890 1891 /* Iterate through singly-linked GSO fragments of an skb. */ 1892 #define skb_list_walk_safe(first, skb, next_skb) \ 1893 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1894 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1895 1896 static inline void skb_list_del_init(struct sk_buff *skb) 1897 { 1898 __list_del_entry(&skb->list); 1899 skb_mark_not_on_list(skb); 1900 } 1901 1902 /** 1903 * skb_queue_empty - check if a queue is empty 1904 * @list: queue head 1905 * 1906 * Returns true if the queue is empty, false otherwise. 1907 */ 1908 static inline int skb_queue_empty(const struct sk_buff_head *list) 1909 { 1910 return list->next == (const struct sk_buff *) list; 1911 } 1912 1913 /** 1914 * skb_queue_empty_lockless - check if a queue is empty 1915 * @list: queue head 1916 * 1917 * Returns true if the queue is empty, false otherwise. 1918 * This variant can be used in lockless contexts. 1919 */ 1920 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1921 { 1922 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1923 } 1924 1925 1926 /** 1927 * skb_queue_is_last - check if skb is the last entry in the queue 1928 * @list: queue head 1929 * @skb: buffer 1930 * 1931 * Returns true if @skb is the last buffer on the list. 1932 */ 1933 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1934 const struct sk_buff *skb) 1935 { 1936 return skb->next == (const struct sk_buff *) list; 1937 } 1938 1939 /** 1940 * skb_queue_is_first - check if skb is the first entry in the queue 1941 * @list: queue head 1942 * @skb: buffer 1943 * 1944 * Returns true if @skb is the first buffer on the list. 1945 */ 1946 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1947 const struct sk_buff *skb) 1948 { 1949 return skb->prev == (const struct sk_buff *) list; 1950 } 1951 1952 /** 1953 * skb_queue_next - return the next packet in the queue 1954 * @list: queue head 1955 * @skb: current buffer 1956 * 1957 * Return the next packet in @list after @skb. It is only valid to 1958 * call this if skb_queue_is_last() evaluates to false. 1959 */ 1960 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1961 const struct sk_buff *skb) 1962 { 1963 /* This BUG_ON may seem severe, but if we just return then we 1964 * are going to dereference garbage. 1965 */ 1966 BUG_ON(skb_queue_is_last(list, skb)); 1967 return skb->next; 1968 } 1969 1970 /** 1971 * skb_queue_prev - return the prev packet in the queue 1972 * @list: queue head 1973 * @skb: current buffer 1974 * 1975 * Return the prev packet in @list before @skb. It is only valid to 1976 * call this if skb_queue_is_first() evaluates to false. 1977 */ 1978 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1979 const struct sk_buff *skb) 1980 { 1981 /* This BUG_ON may seem severe, but if we just return then we 1982 * are going to dereference garbage. 1983 */ 1984 BUG_ON(skb_queue_is_first(list, skb)); 1985 return skb->prev; 1986 } 1987 1988 /** 1989 * skb_get - reference buffer 1990 * @skb: buffer to reference 1991 * 1992 * Makes another reference to a socket buffer and returns a pointer 1993 * to the buffer. 1994 */ 1995 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1996 { 1997 refcount_inc(&skb->users); 1998 return skb; 1999 } 2000 2001 /* 2002 * If users == 1, we are the only owner and can avoid redundant atomic changes. 2003 */ 2004 2005 /** 2006 * skb_cloned - is the buffer a clone 2007 * @skb: buffer to check 2008 * 2009 * Returns true if the buffer was generated with skb_clone() and is 2010 * one of multiple shared copies of the buffer. Cloned buffers are 2011 * shared data so must not be written to under normal circumstances. 2012 */ 2013 static inline int skb_cloned(const struct sk_buff *skb) 2014 { 2015 return skb->cloned && 2016 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 2017 } 2018 2019 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 2020 { 2021 might_sleep_if(gfpflags_allow_blocking(pri)); 2022 2023 if (skb_cloned(skb)) 2024 return pskb_expand_head(skb, 0, 0, pri); 2025 2026 return 0; 2027 } 2028 2029 /* This variant of skb_unclone() makes sure skb->truesize 2030 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 2031 * 2032 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 2033 * when various debugging features are in place. 2034 */ 2035 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 2036 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 2037 { 2038 might_sleep_if(gfpflags_allow_blocking(pri)); 2039 2040 if (skb_cloned(skb)) 2041 return __skb_unclone_keeptruesize(skb, pri); 2042 return 0; 2043 } 2044 2045 /** 2046 * skb_header_cloned - is the header a clone 2047 * @skb: buffer to check 2048 * 2049 * Returns true if modifying the header part of the buffer requires 2050 * the data to be copied. 2051 */ 2052 static inline int skb_header_cloned(const struct sk_buff *skb) 2053 { 2054 int dataref; 2055 2056 if (!skb->cloned) 2057 return 0; 2058 2059 dataref = atomic_read(&skb_shinfo(skb)->dataref); 2060 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 2061 return dataref != 1; 2062 } 2063 2064 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 2065 { 2066 might_sleep_if(gfpflags_allow_blocking(pri)); 2067 2068 if (skb_header_cloned(skb)) 2069 return pskb_expand_head(skb, 0, 0, pri); 2070 2071 return 0; 2072 } 2073 2074 /** 2075 * __skb_header_release() - allow clones to use the headroom 2076 * @skb: buffer to operate on 2077 * 2078 * See "DOC: dataref and headerless skbs". 2079 */ 2080 static inline void __skb_header_release(struct sk_buff *skb) 2081 { 2082 skb->nohdr = 1; 2083 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 2084 } 2085 2086 2087 /** 2088 * skb_shared - is the buffer shared 2089 * @skb: buffer to check 2090 * 2091 * Returns true if more than one person has a reference to this 2092 * buffer. 2093 */ 2094 static inline int skb_shared(const struct sk_buff *skb) 2095 { 2096 return refcount_read(&skb->users) != 1; 2097 } 2098 2099 /** 2100 * skb_share_check - check if buffer is shared and if so clone it 2101 * @skb: buffer to check 2102 * @pri: priority for memory allocation 2103 * 2104 * If the buffer is shared the buffer is cloned and the old copy 2105 * drops a reference. A new clone with a single reference is returned. 2106 * If the buffer is not shared the original buffer is returned. When 2107 * being called from interrupt status or with spinlocks held pri must 2108 * be GFP_ATOMIC. 2109 * 2110 * NULL is returned on a memory allocation failure. 2111 */ 2112 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 2113 { 2114 might_sleep_if(gfpflags_allow_blocking(pri)); 2115 if (skb_shared(skb)) { 2116 struct sk_buff *nskb = skb_clone(skb, pri); 2117 2118 if (likely(nskb)) 2119 consume_skb(skb); 2120 else 2121 kfree_skb(skb); 2122 skb = nskb; 2123 } 2124 return skb; 2125 } 2126 2127 /* 2128 * Copy shared buffers into a new sk_buff. We effectively do COW on 2129 * packets to handle cases where we have a local reader and forward 2130 * and a couple of other messy ones. The normal one is tcpdumping 2131 * a packet that's being forwarded. 2132 */ 2133 2134 /** 2135 * skb_unshare - make a copy of a shared buffer 2136 * @skb: buffer to check 2137 * @pri: priority for memory allocation 2138 * 2139 * If the socket buffer is a clone then this function creates a new 2140 * copy of the data, drops a reference count on the old copy and returns 2141 * the new copy with the reference count at 1. If the buffer is not a clone 2142 * the original buffer is returned. When called with a spinlock held or 2143 * from interrupt state @pri must be %GFP_ATOMIC 2144 * 2145 * %NULL is returned on a memory allocation failure. 2146 */ 2147 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2148 gfp_t pri) 2149 { 2150 might_sleep_if(gfpflags_allow_blocking(pri)); 2151 if (skb_cloned(skb)) { 2152 struct sk_buff *nskb = skb_copy(skb, pri); 2153 2154 /* Free our shared copy */ 2155 if (likely(nskb)) 2156 consume_skb(skb); 2157 else 2158 kfree_skb(skb); 2159 skb = nskb; 2160 } 2161 return skb; 2162 } 2163 2164 /** 2165 * skb_peek - peek at the head of an &sk_buff_head 2166 * @list_: list to peek at 2167 * 2168 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2169 * be careful with this one. A peek leaves the buffer on the 2170 * list and someone else may run off with it. You must hold 2171 * the appropriate locks or have a private queue to do this. 2172 * 2173 * Returns %NULL for an empty list or a pointer to the head element. 2174 * The reference count is not incremented and the reference is therefore 2175 * volatile. Use with caution. 2176 */ 2177 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2178 { 2179 struct sk_buff *skb = list_->next; 2180 2181 if (skb == (struct sk_buff *)list_) 2182 skb = NULL; 2183 return skb; 2184 } 2185 2186 /** 2187 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2188 * @list_: list to peek at 2189 * 2190 * Like skb_peek(), but the caller knows that the list is not empty. 2191 */ 2192 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2193 { 2194 return list_->next; 2195 } 2196 2197 /** 2198 * skb_peek_next - peek skb following the given one from a queue 2199 * @skb: skb to start from 2200 * @list_: list to peek at 2201 * 2202 * Returns %NULL when the end of the list is met or a pointer to the 2203 * next element. The reference count is not incremented and the 2204 * reference is therefore volatile. Use with caution. 2205 */ 2206 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2207 const struct sk_buff_head *list_) 2208 { 2209 struct sk_buff *next = skb->next; 2210 2211 if (next == (struct sk_buff *)list_) 2212 next = NULL; 2213 return next; 2214 } 2215 2216 /** 2217 * skb_peek_tail - peek at the tail of an &sk_buff_head 2218 * @list_: list to peek at 2219 * 2220 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2221 * be careful with this one. A peek leaves the buffer on the 2222 * list and someone else may run off with it. You must hold 2223 * the appropriate locks or have a private queue to do this. 2224 * 2225 * Returns %NULL for an empty list or a pointer to the tail element. 2226 * The reference count is not incremented and the reference is therefore 2227 * volatile. Use with caution. 2228 */ 2229 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2230 { 2231 struct sk_buff *skb = READ_ONCE(list_->prev); 2232 2233 if (skb == (struct sk_buff *)list_) 2234 skb = NULL; 2235 return skb; 2236 2237 } 2238 2239 /** 2240 * skb_queue_len - get queue length 2241 * @list_: list to measure 2242 * 2243 * Return the length of an &sk_buff queue. 2244 */ 2245 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2246 { 2247 return list_->qlen; 2248 } 2249 2250 /** 2251 * skb_queue_len_lockless - get queue length 2252 * @list_: list to measure 2253 * 2254 * Return the length of an &sk_buff queue. 2255 * This variant can be used in lockless contexts. 2256 */ 2257 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2258 { 2259 return READ_ONCE(list_->qlen); 2260 } 2261 2262 /** 2263 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2264 * @list: queue to initialize 2265 * 2266 * This initializes only the list and queue length aspects of 2267 * an sk_buff_head object. This allows to initialize the list 2268 * aspects of an sk_buff_head without reinitializing things like 2269 * the spinlock. It can also be used for on-stack sk_buff_head 2270 * objects where the spinlock is known to not be used. 2271 */ 2272 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2273 { 2274 list->prev = list->next = (struct sk_buff *)list; 2275 list->qlen = 0; 2276 } 2277 2278 /* 2279 * This function creates a split out lock class for each invocation; 2280 * this is needed for now since a whole lot of users of the skb-queue 2281 * infrastructure in drivers have different locking usage (in hardirq) 2282 * than the networking core (in softirq only). In the long run either the 2283 * network layer or drivers should need annotation to consolidate the 2284 * main types of usage into 3 classes. 2285 */ 2286 static inline void skb_queue_head_init(struct sk_buff_head *list) 2287 { 2288 spin_lock_init(&list->lock); 2289 __skb_queue_head_init(list); 2290 } 2291 2292 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2293 struct lock_class_key *class) 2294 { 2295 skb_queue_head_init(list); 2296 lockdep_set_class(&list->lock, class); 2297 } 2298 2299 /* 2300 * Insert an sk_buff on a list. 2301 * 2302 * The "__skb_xxxx()" functions are the non-atomic ones that 2303 * can only be called with interrupts disabled. 2304 */ 2305 static inline void __skb_insert(struct sk_buff *newsk, 2306 struct sk_buff *prev, struct sk_buff *next, 2307 struct sk_buff_head *list) 2308 { 2309 /* See skb_queue_empty_lockless() and skb_peek_tail() 2310 * for the opposite READ_ONCE() 2311 */ 2312 WRITE_ONCE(newsk->next, next); 2313 WRITE_ONCE(newsk->prev, prev); 2314 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2315 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2316 WRITE_ONCE(list->qlen, list->qlen + 1); 2317 } 2318 2319 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2320 struct sk_buff *prev, 2321 struct sk_buff *next) 2322 { 2323 struct sk_buff *first = list->next; 2324 struct sk_buff *last = list->prev; 2325 2326 WRITE_ONCE(first->prev, prev); 2327 WRITE_ONCE(prev->next, first); 2328 2329 WRITE_ONCE(last->next, next); 2330 WRITE_ONCE(next->prev, last); 2331 } 2332 2333 /** 2334 * skb_queue_splice - join two skb lists, this is designed for stacks 2335 * @list: the new list to add 2336 * @head: the place to add it in the first list 2337 */ 2338 static inline void skb_queue_splice(const struct sk_buff_head *list, 2339 struct sk_buff_head *head) 2340 { 2341 if (!skb_queue_empty(list)) { 2342 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2343 head->qlen += list->qlen; 2344 } 2345 } 2346 2347 /** 2348 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2349 * @list: the new list to add 2350 * @head: the place to add it in the first list 2351 * 2352 * The list at @list is reinitialised 2353 */ 2354 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2355 struct sk_buff_head *head) 2356 { 2357 if (!skb_queue_empty(list)) { 2358 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2359 head->qlen += list->qlen; 2360 __skb_queue_head_init(list); 2361 } 2362 } 2363 2364 /** 2365 * skb_queue_splice_tail - join two skb lists, each list being a queue 2366 * @list: the new list to add 2367 * @head: the place to add it in the first list 2368 */ 2369 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2370 struct sk_buff_head *head) 2371 { 2372 if (!skb_queue_empty(list)) { 2373 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2374 head->qlen += list->qlen; 2375 } 2376 } 2377 2378 /** 2379 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2380 * @list: the new list to add 2381 * @head: the place to add it in the first list 2382 * 2383 * Each of the lists is a queue. 2384 * The list at @list is reinitialised 2385 */ 2386 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2387 struct sk_buff_head *head) 2388 { 2389 if (!skb_queue_empty(list)) { 2390 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2391 head->qlen += list->qlen; 2392 __skb_queue_head_init(list); 2393 } 2394 } 2395 2396 /** 2397 * __skb_queue_after - queue a buffer at the list head 2398 * @list: list to use 2399 * @prev: place after this buffer 2400 * @newsk: buffer to queue 2401 * 2402 * Queue a buffer int the middle of a list. This function takes no locks 2403 * and you must therefore hold required locks before calling it. 2404 * 2405 * A buffer cannot be placed on two lists at the same time. 2406 */ 2407 static inline void __skb_queue_after(struct sk_buff_head *list, 2408 struct sk_buff *prev, 2409 struct sk_buff *newsk) 2410 { 2411 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2412 } 2413 2414 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2415 struct sk_buff_head *list); 2416 2417 static inline void __skb_queue_before(struct sk_buff_head *list, 2418 struct sk_buff *next, 2419 struct sk_buff *newsk) 2420 { 2421 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2422 } 2423 2424 /** 2425 * __skb_queue_head - queue a buffer at the list head 2426 * @list: list to use 2427 * @newsk: buffer to queue 2428 * 2429 * Queue a buffer at the start of a list. This function takes no locks 2430 * and you must therefore hold required locks before calling it. 2431 * 2432 * A buffer cannot be placed on two lists at the same time. 2433 */ 2434 static inline void __skb_queue_head(struct sk_buff_head *list, 2435 struct sk_buff *newsk) 2436 { 2437 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2438 } 2439 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2440 2441 /** 2442 * __skb_queue_tail - queue a buffer at the list tail 2443 * @list: list to use 2444 * @newsk: buffer to queue 2445 * 2446 * Queue a buffer at the end of a list. This function takes no locks 2447 * and you must therefore hold required locks before calling it. 2448 * 2449 * A buffer cannot be placed on two lists at the same time. 2450 */ 2451 static inline void __skb_queue_tail(struct sk_buff_head *list, 2452 struct sk_buff *newsk) 2453 { 2454 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2455 } 2456 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2457 2458 /* 2459 * remove sk_buff from list. _Must_ be called atomically, and with 2460 * the list known.. 2461 */ 2462 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2463 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2464 { 2465 struct sk_buff *next, *prev; 2466 2467 WRITE_ONCE(list->qlen, list->qlen - 1); 2468 next = skb->next; 2469 prev = skb->prev; 2470 skb->next = skb->prev = NULL; 2471 WRITE_ONCE(next->prev, prev); 2472 WRITE_ONCE(prev->next, next); 2473 } 2474 2475 /** 2476 * __skb_dequeue - remove from the head of the queue 2477 * @list: list to dequeue from 2478 * 2479 * Remove the head of the list. This function does not take any locks 2480 * so must be used with appropriate locks held only. The head item is 2481 * returned or %NULL if the list is empty. 2482 */ 2483 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2484 { 2485 struct sk_buff *skb = skb_peek(list); 2486 if (skb) 2487 __skb_unlink(skb, list); 2488 return skb; 2489 } 2490 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2491 2492 /** 2493 * __skb_dequeue_tail - remove from the tail of the queue 2494 * @list: list to dequeue from 2495 * 2496 * Remove the tail of the list. This function does not take any locks 2497 * so must be used with appropriate locks held only. The tail item is 2498 * returned or %NULL if the list is empty. 2499 */ 2500 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2501 { 2502 struct sk_buff *skb = skb_peek_tail(list); 2503 if (skb) 2504 __skb_unlink(skb, list); 2505 return skb; 2506 } 2507 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2508 2509 2510 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2511 { 2512 return skb->data_len; 2513 } 2514 2515 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2516 { 2517 return skb->len - skb->data_len; 2518 } 2519 2520 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2521 { 2522 unsigned int i, len = 0; 2523 2524 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2525 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2526 return len; 2527 } 2528 2529 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2530 { 2531 return skb_headlen(skb) + __skb_pagelen(skb); 2532 } 2533 2534 static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag, 2535 netmem_ref netmem, int off, 2536 int size) 2537 { 2538 frag->netmem = netmem; 2539 frag->offset = off; 2540 skb_frag_size_set(frag, size); 2541 } 2542 2543 static inline void skb_frag_fill_page_desc(skb_frag_t *frag, 2544 struct page *page, 2545 int off, int size) 2546 { 2547 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size); 2548 } 2549 2550 static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo, 2551 int i, netmem_ref netmem, 2552 int off, int size) 2553 { 2554 skb_frag_t *frag = &shinfo->frags[i]; 2555 2556 skb_frag_fill_netmem_desc(frag, netmem, off, size); 2557 } 2558 2559 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2560 int i, struct page *page, 2561 int off, int size) 2562 { 2563 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off, 2564 size); 2565 } 2566 2567 /** 2568 * skb_len_add - adds a number to len fields of skb 2569 * @skb: buffer to add len to 2570 * @delta: number of bytes to add 2571 */ 2572 static inline void skb_len_add(struct sk_buff *skb, int delta) 2573 { 2574 skb->len += delta; 2575 skb->data_len += delta; 2576 skb->truesize += delta; 2577 } 2578 2579 /** 2580 * __skb_fill_netmem_desc - initialise a fragment in an skb 2581 * @skb: buffer containing fragment to be initialised 2582 * @i: fragment index to initialise 2583 * @netmem: the netmem to use for this fragment 2584 * @off: the offset to the data with @page 2585 * @size: the length of the data 2586 * 2587 * Initialises the @i'th fragment of @skb to point to &size bytes at 2588 * offset @off within @page. 2589 * 2590 * Does not take any additional reference on the fragment. 2591 */ 2592 static __always_inline void 2593 __skb_fill_netmem_desc(struct sk_buff *skb, int i, netmem_ref netmem, 2594 int off, int size) 2595 { 2596 struct page *page; 2597 2598 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size); 2599 2600 if (netmem_is_net_iov(netmem)) { 2601 skb->unreadable = true; 2602 return; 2603 } 2604 2605 page = netmem_to_page(netmem); 2606 2607 /* Propagate page pfmemalloc to the skb if we can. The problem is 2608 * that not all callers have unique ownership of the page but rely 2609 * on page_is_pfmemalloc doing the right thing(tm). 2610 */ 2611 page = compound_head(page); 2612 if (page_is_pfmemalloc(page)) 2613 skb->pfmemalloc = true; 2614 } 2615 2616 static __always_inline void 2617 __skb_fill_page_desc(struct sk_buff *skb, int i, struct page *page, 2618 int off, int size) 2619 { 2620 __skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2621 } 2622 2623 static __always_inline void 2624 skb_fill_netmem_desc(struct sk_buff *skb, int i, netmem_ref netmem, 2625 int off, int size) 2626 { 2627 __skb_fill_netmem_desc(skb, i, netmem, off, size); 2628 skb_shinfo(skb)->nr_frags = i + 1; 2629 } 2630 2631 /** 2632 * skb_fill_page_desc - initialise a paged fragment in an skb 2633 * @skb: buffer containing fragment to be initialised 2634 * @i: paged fragment index to initialise 2635 * @page: the page to use for this fragment 2636 * @off: the offset to the data with @page 2637 * @size: the length of the data 2638 * 2639 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2640 * @skb to point to @size bytes at offset @off within @page. In 2641 * addition updates @skb such that @i is the last fragment. 2642 * 2643 * Does not take any additional reference on the fragment. 2644 */ 2645 static __always_inline void 2646 skb_fill_page_desc(struct sk_buff *skb, int i, struct page *page, 2647 int off, int size) 2648 { 2649 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2650 } 2651 2652 /** 2653 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2654 * @skb: buffer containing fragment to be initialised 2655 * @i: paged fragment index to initialise 2656 * @page: the page to use for this fragment 2657 * @off: the offset to the data with @page 2658 * @size: the length of the data 2659 * 2660 * Variant of skb_fill_page_desc() which does not deal with 2661 * pfmemalloc, if page is not owned by us. 2662 */ 2663 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2664 struct page *page, int off, 2665 int size) 2666 { 2667 struct skb_shared_info *shinfo = skb_shinfo(skb); 2668 2669 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2670 shinfo->nr_frags = i + 1; 2671 } 2672 2673 static inline void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, 2674 netmem_ref netmem, int off, 2675 int size, unsigned int truesize) 2676 { 2677 DEBUG_NET_WARN_ON_ONCE(size > truesize); 2678 2679 skb_fill_netmem_desc(skb, i, netmem, off, size); 2680 skb->len += size; 2681 skb->data_len += size; 2682 skb->truesize += truesize; 2683 } 2684 2685 static inline void skb_add_rx_frag(struct sk_buff *skb, int i, 2686 struct page *page, int off, int size, 2687 unsigned int truesize) 2688 { 2689 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size, 2690 truesize); 2691 } 2692 2693 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2694 unsigned int truesize); 2695 2696 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2697 2698 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2699 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2700 { 2701 return skb->head + skb->tail; 2702 } 2703 2704 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2705 { 2706 skb->tail = skb->data - skb->head; 2707 } 2708 2709 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2710 { 2711 skb_reset_tail_pointer(skb); 2712 skb->tail += offset; 2713 } 2714 2715 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2716 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2717 { 2718 return skb->tail; 2719 } 2720 2721 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2722 { 2723 skb->tail = skb->data; 2724 } 2725 2726 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2727 { 2728 skb->tail = skb->data + offset; 2729 } 2730 2731 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2732 2733 static inline void skb_assert_len(struct sk_buff *skb) 2734 { 2735 #ifdef CONFIG_DEBUG_NET 2736 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2737 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2738 #endif /* CONFIG_DEBUG_NET */ 2739 } 2740 2741 #if defined(CONFIG_FAIL_SKB_REALLOC) 2742 void skb_might_realloc(struct sk_buff *skb); 2743 #else 2744 static inline void skb_might_realloc(struct sk_buff *skb) {} 2745 #endif 2746 2747 /* 2748 * Add data to an sk_buff 2749 */ 2750 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2751 void *skb_put(struct sk_buff *skb, unsigned int len); 2752 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2753 { 2754 void *tmp = skb_tail_pointer(skb); 2755 SKB_LINEAR_ASSERT(skb); 2756 skb->tail += len; 2757 skb->len += len; 2758 return tmp; 2759 } 2760 2761 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2762 { 2763 void *tmp = __skb_put(skb, len); 2764 2765 memset(tmp, 0, len); 2766 return tmp; 2767 } 2768 2769 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2770 unsigned int len) 2771 { 2772 void *tmp = __skb_put(skb, len); 2773 2774 memcpy(tmp, data, len); 2775 return tmp; 2776 } 2777 2778 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2779 { 2780 *(u8 *)__skb_put(skb, 1) = val; 2781 } 2782 2783 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2784 { 2785 void *tmp = skb_put(skb, len); 2786 2787 memset(tmp, 0, len); 2788 2789 return tmp; 2790 } 2791 2792 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2793 unsigned int len) 2794 { 2795 void *tmp = skb_put(skb, len); 2796 2797 memcpy(tmp, data, len); 2798 2799 return tmp; 2800 } 2801 2802 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2803 { 2804 *(u8 *)skb_put(skb, 1) = val; 2805 } 2806 2807 void *skb_push(struct sk_buff *skb, unsigned int len); 2808 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2809 { 2810 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2811 2812 skb->data -= len; 2813 DEBUG_NET_WARN_ON_ONCE(skb->data < skb->head); 2814 skb->len += len; 2815 return skb->data; 2816 } 2817 2818 void *skb_pull(struct sk_buff *skb, unsigned int len); 2819 static __always_inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2820 { 2821 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2822 2823 skb->len -= len; 2824 if (unlikely(skb->len < skb->data_len)) { 2825 #if defined(CONFIG_DEBUG_NET) 2826 skb->len += len; 2827 pr_err("__skb_pull(len=%u)\n", len); 2828 skb_dump(KERN_ERR, skb, false); 2829 #endif 2830 BUG(); 2831 } 2832 return skb->data += len; 2833 } 2834 2835 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2836 { 2837 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2838 } 2839 2840 void *skb_pull_data(struct sk_buff *skb, size_t len); 2841 2842 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2843 2844 static __always_inline enum skb_drop_reason 2845 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len) 2846 { 2847 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2848 skb_might_realloc(skb); 2849 2850 if (likely(len <= skb_headlen(skb))) 2851 return SKB_NOT_DROPPED_YET; 2852 2853 if (unlikely(len > skb->len)) 2854 return SKB_DROP_REASON_PKT_TOO_SMALL; 2855 2856 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb)))) 2857 return SKB_DROP_REASON_NOMEM; 2858 2859 return SKB_NOT_DROPPED_YET; 2860 } 2861 2862 static __always_inline bool 2863 pskb_may_pull(struct sk_buff *skb, unsigned int len) 2864 { 2865 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 2866 } 2867 2868 static __always_inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2869 { 2870 if (!pskb_may_pull(skb, len)) 2871 return NULL; 2872 2873 skb->len -= len; 2874 return skb->data += len; 2875 } 2876 2877 void skb_condense(struct sk_buff *skb); 2878 2879 /** 2880 * skb_headroom - bytes at buffer head 2881 * @skb: buffer to check 2882 * 2883 * Return the number of bytes of free space at the head of an &sk_buff. 2884 */ 2885 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2886 { 2887 return skb->data - skb->head; 2888 } 2889 2890 /** 2891 * skb_tailroom - bytes at buffer end 2892 * @skb: buffer to check 2893 * 2894 * Return the number of bytes of free space at the tail of an sk_buff 2895 */ 2896 static inline int skb_tailroom(const struct sk_buff *skb) 2897 { 2898 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2899 } 2900 2901 /** 2902 * skb_availroom - bytes at buffer end 2903 * @skb: buffer to check 2904 * 2905 * Return the number of bytes of free space at the tail of an sk_buff 2906 * allocated by sk_stream_alloc() 2907 */ 2908 static inline int skb_availroom(const struct sk_buff *skb) 2909 { 2910 if (skb_is_nonlinear(skb)) 2911 return 0; 2912 2913 return skb->end - skb->tail - skb->reserved_tailroom; 2914 } 2915 2916 /** 2917 * skb_reserve - adjust headroom 2918 * @skb: buffer to alter 2919 * @len: bytes to move 2920 * 2921 * Increase the headroom of an empty &sk_buff by reducing the tail 2922 * room. This is only allowed for an empty buffer. 2923 */ 2924 static inline void skb_reserve(struct sk_buff *skb, int len) 2925 { 2926 skb->data += len; 2927 skb->tail += len; 2928 } 2929 2930 /** 2931 * skb_tailroom_reserve - adjust reserved_tailroom 2932 * @skb: buffer to alter 2933 * @mtu: maximum amount of headlen permitted 2934 * @needed_tailroom: minimum amount of reserved_tailroom 2935 * 2936 * Set reserved_tailroom so that headlen can be as large as possible but 2937 * not larger than mtu and tailroom cannot be smaller than 2938 * needed_tailroom. 2939 * The required headroom should already have been reserved before using 2940 * this function. 2941 */ 2942 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2943 unsigned int needed_tailroom) 2944 { 2945 SKB_LINEAR_ASSERT(skb); 2946 if (mtu < skb_tailroom(skb) - needed_tailroom) 2947 /* use at most mtu */ 2948 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2949 else 2950 /* use up to all available space */ 2951 skb->reserved_tailroom = needed_tailroom; 2952 } 2953 2954 #define ENCAP_TYPE_ETHER 0 2955 #define ENCAP_TYPE_IPPROTO 1 2956 2957 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2958 __be16 protocol) 2959 { 2960 skb->inner_protocol = protocol; 2961 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2962 } 2963 2964 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2965 __u8 ipproto) 2966 { 2967 skb->inner_ipproto = ipproto; 2968 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2969 } 2970 2971 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2972 { 2973 skb->inner_mac_header = skb->mac_header; 2974 skb->inner_network_header = skb->network_header; 2975 skb->inner_transport_header = skb->transport_header; 2976 } 2977 2978 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2979 { 2980 return skb->mac_header != (typeof(skb->mac_header))~0U; 2981 } 2982 2983 static inline void skb_reset_mac_len(struct sk_buff *skb) 2984 { 2985 if (!skb_mac_header_was_set(skb)) { 2986 DEBUG_NET_WARN_ON_ONCE(1); 2987 skb->mac_len = 0; 2988 } else { 2989 skb->mac_len = skb->network_header - skb->mac_header; 2990 } 2991 } 2992 2993 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2994 *skb) 2995 { 2996 return skb->head + skb->inner_transport_header; 2997 } 2998 2999 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 3000 { 3001 return skb_inner_transport_header(skb) - skb->data; 3002 } 3003 3004 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 3005 { 3006 long offset = skb->data - skb->head; 3007 3008 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_transport_header))offset); 3009 skb->inner_transport_header = offset; 3010 } 3011 3012 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 3013 const int offset) 3014 { 3015 skb_reset_inner_transport_header(skb); 3016 skb->inner_transport_header += offset; 3017 } 3018 3019 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 3020 { 3021 return skb->head + skb->inner_network_header; 3022 } 3023 3024 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 3025 { 3026 long offset = skb->data - skb->head; 3027 3028 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_network_header))offset); 3029 skb->inner_network_header = offset; 3030 } 3031 3032 static inline void skb_set_inner_network_header(struct sk_buff *skb, 3033 const int offset) 3034 { 3035 skb_reset_inner_network_header(skb); 3036 skb->inner_network_header += offset; 3037 } 3038 3039 static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb) 3040 { 3041 return skb->inner_network_header > 0; 3042 } 3043 3044 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 3045 { 3046 return skb->head + skb->inner_mac_header; 3047 } 3048 3049 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 3050 { 3051 long offset = skb->data - skb->head; 3052 3053 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_mac_header))offset); 3054 skb->inner_mac_header = offset; 3055 } 3056 3057 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 3058 const int offset) 3059 { 3060 skb_reset_inner_mac_header(skb); 3061 skb->inner_mac_header += offset; 3062 } 3063 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 3064 { 3065 return skb->transport_header != (typeof(skb->transport_header))~0U; 3066 } 3067 3068 static inline void skb_unset_transport_header(struct sk_buff *skb) 3069 { 3070 skb->transport_header = (typeof(skb->transport_header))~0U; 3071 } 3072 3073 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 3074 { 3075 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3076 return skb->head + skb->transport_header; 3077 } 3078 3079 static inline void skb_reset_transport_header(struct sk_buff *skb) 3080 { 3081 long offset = skb->data - skb->head; 3082 3083 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->transport_header))offset); 3084 skb->transport_header = offset; 3085 } 3086 3087 /** 3088 * skb_reset_transport_header_careful - conditionally reset transport header 3089 * @skb: buffer to alter 3090 * 3091 * Hardened version of skb_reset_transport_header(). 3092 * 3093 * Returns: true if the operation was a success. 3094 */ 3095 static inline bool __must_check 3096 skb_reset_transport_header_careful(struct sk_buff *skb) 3097 { 3098 long offset = skb->data - skb->head; 3099 3100 if (unlikely(offset != (typeof(skb->transport_header))offset)) 3101 return false; 3102 3103 if (unlikely(offset == (typeof(skb->transport_header))~0U)) 3104 return false; 3105 3106 skb->transport_header = offset; 3107 return true; 3108 } 3109 3110 static inline void skb_set_transport_header(struct sk_buff *skb, 3111 const int offset) 3112 { 3113 skb_reset_transport_header(skb); 3114 skb->transport_header += offset; 3115 } 3116 3117 /** 3118 * skb_set_transport_header_careful - conditionally set transport header 3119 * @skb: buffer to alter 3120 * @offset: offset to add to skb->data 3121 * 3122 * Hardened version of skb_set_transport_header(). 3123 * 3124 * Returns: true if the operation was a success. 3125 */ 3126 static inline bool __must_check 3127 skb_set_transport_header_careful(struct sk_buff *skb, const int offset) 3128 { 3129 long thoff = skb->data - skb->head + offset; 3130 3131 if (unlikely(thoff != (typeof(skb->transport_header))thoff)) 3132 return false; 3133 3134 if (unlikely(thoff == (typeof(skb->transport_header))~0U)) 3135 return false; 3136 3137 skb->transport_header = thoff; 3138 return true; 3139 } 3140 3141 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 3142 { 3143 return skb->head + skb->network_header; 3144 } 3145 3146 static inline void skb_reset_network_header(struct sk_buff *skb) 3147 { 3148 long offset = skb->data - skb->head; 3149 3150 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->network_header))offset); 3151 skb->network_header = offset; 3152 } 3153 3154 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 3155 { 3156 skb_reset_network_header(skb); 3157 skb->network_header += offset; 3158 } 3159 3160 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 3161 { 3162 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3163 return skb->head + skb->mac_header; 3164 } 3165 3166 static inline int skb_mac_offset(const struct sk_buff *skb) 3167 { 3168 return skb_mac_header(skb) - skb->data; 3169 } 3170 3171 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 3172 { 3173 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3174 return skb->network_header - skb->mac_header; 3175 } 3176 3177 static inline void skb_unset_mac_header(struct sk_buff *skb) 3178 { 3179 skb->mac_header = (typeof(skb->mac_header))~0U; 3180 } 3181 3182 static inline void skb_reset_mac_header(struct sk_buff *skb) 3183 { 3184 long offset = skb->data - skb->head; 3185 3186 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->mac_header))offset); 3187 skb->mac_header = offset; 3188 } 3189 3190 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 3191 { 3192 skb_reset_mac_header(skb); 3193 skb->mac_header += offset; 3194 } 3195 3196 static inline void skb_pop_mac_header(struct sk_buff *skb) 3197 { 3198 skb->mac_header = skb->network_header; 3199 } 3200 3201 static inline void skb_probe_transport_header(struct sk_buff *skb) 3202 { 3203 struct flow_keys_basic keys; 3204 3205 if (skb_transport_header_was_set(skb)) 3206 return; 3207 3208 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 3209 NULL, 0, 0, 0, 0)) 3210 skb_set_transport_header(skb, keys.control.thoff); 3211 } 3212 3213 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 3214 { 3215 if (skb_mac_header_was_set(skb)) { 3216 const unsigned char *old_mac = skb_mac_header(skb); 3217 3218 skb_set_mac_header(skb, -skb->mac_len); 3219 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 3220 } 3221 } 3222 3223 /* Move the full mac header up to current network_header. 3224 * Leaves skb->data pointing at offset skb->mac_len into the mac_header. 3225 * Must be provided the complete mac header length. 3226 */ 3227 static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len) 3228 { 3229 if (skb_mac_header_was_set(skb)) { 3230 const unsigned char *old_mac = skb_mac_header(skb); 3231 3232 skb_set_mac_header(skb, -full_mac_len); 3233 memmove(skb_mac_header(skb), old_mac, full_mac_len); 3234 __skb_push(skb, full_mac_len - skb->mac_len); 3235 } 3236 } 3237 3238 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 3239 { 3240 return skb->csum_start - skb_headroom(skb); 3241 } 3242 3243 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 3244 { 3245 return skb->head + skb->csum_start; 3246 } 3247 3248 static inline int skb_transport_offset(const struct sk_buff *skb) 3249 { 3250 return skb_transport_header(skb) - skb->data; 3251 } 3252 3253 static inline u32 skb_network_header_len(const struct sk_buff *skb) 3254 { 3255 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3256 return skb->transport_header - skb->network_header; 3257 } 3258 3259 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 3260 { 3261 return skb->inner_transport_header - skb->inner_network_header; 3262 } 3263 3264 static inline int skb_network_offset(const struct sk_buff *skb) 3265 { 3266 return skb_network_header(skb) - skb->data; 3267 } 3268 3269 static inline int skb_inner_network_offset(const struct sk_buff *skb) 3270 { 3271 return skb_inner_network_header(skb) - skb->data; 3272 } 3273 3274 static inline enum skb_drop_reason 3275 pskb_network_may_pull_reason(struct sk_buff *skb, unsigned int len) 3276 { 3277 return pskb_may_pull_reason(skb, skb_network_offset(skb) + len); 3278 } 3279 3280 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 3281 { 3282 return pskb_network_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 3283 } 3284 3285 /* 3286 * CPUs often take a performance hit when accessing unaligned memory 3287 * locations. The actual performance hit varies, it can be small if the 3288 * hardware handles it or large if we have to take an exception and fix it 3289 * in software. 3290 * 3291 * Since an ethernet header is 14 bytes network drivers often end up with 3292 * the IP header at an unaligned offset. The IP header can be aligned by 3293 * shifting the start of the packet by 2 bytes. Drivers should do this 3294 * with: 3295 * 3296 * skb_reserve(skb, NET_IP_ALIGN); 3297 * 3298 * The downside to this alignment of the IP header is that the DMA is now 3299 * unaligned. On some architectures the cost of an unaligned DMA is high 3300 * and this cost outweighs the gains made by aligning the IP header. 3301 * 3302 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 3303 * to be overridden. 3304 */ 3305 #ifndef NET_IP_ALIGN 3306 #define NET_IP_ALIGN 2 3307 #endif 3308 3309 /* 3310 * The networking layer reserves some headroom in skb data (via 3311 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 3312 * the header has to grow. In the default case, if the header has to grow 3313 * 32 bytes or less we avoid the reallocation. 3314 * 3315 * Unfortunately this headroom changes the DMA alignment of the resulting 3316 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 3317 * on some architectures. An architecture can override this value, 3318 * perhaps setting it to a cacheline in size (since that will maintain 3319 * cacheline alignment of the DMA). It must be a power of 2. 3320 * 3321 * Various parts of the networking layer expect at least 32 bytes of 3322 * headroom, you should not reduce this. 3323 * 3324 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 3325 * to reduce average number of cache lines per packet. 3326 * get_rps_cpu() for example only access one 64 bytes aligned block : 3327 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 3328 */ 3329 #ifndef NET_SKB_PAD 3330 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 3331 #endif 3332 3333 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 3334 3335 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3336 { 3337 if (WARN_ON(skb_is_nonlinear(skb))) 3338 return; 3339 skb->len = len; 3340 skb_set_tail_pointer(skb, len); 3341 } 3342 3343 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3344 { 3345 __skb_set_length(skb, len); 3346 } 3347 3348 void skb_trim(struct sk_buff *skb, unsigned int len); 3349 3350 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3351 { 3352 if (skb->data_len) 3353 return ___pskb_trim(skb, len); 3354 __skb_trim(skb, len); 3355 return 0; 3356 } 3357 3358 static __always_inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3359 { 3360 skb_might_realloc(skb); 3361 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3362 } 3363 3364 /** 3365 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3366 * @skb: buffer to alter 3367 * @len: new length 3368 * 3369 * This is identical to pskb_trim except that the caller knows that 3370 * the skb is not cloned so we should never get an error due to out- 3371 * of-memory. 3372 */ 3373 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3374 { 3375 int err = pskb_trim(skb, len); 3376 BUG_ON(err); 3377 } 3378 3379 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3380 { 3381 unsigned int diff = len - skb->len; 3382 3383 if (skb_tailroom(skb) < diff) { 3384 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3385 GFP_ATOMIC); 3386 if (ret) 3387 return ret; 3388 } 3389 __skb_set_length(skb, len); 3390 return 0; 3391 } 3392 3393 /** 3394 * skb_orphan - orphan a buffer 3395 * @skb: buffer to orphan 3396 * 3397 * If a buffer currently has an owner then we call the owner's 3398 * destructor function and make the @skb unowned. The buffer continues 3399 * to exist but is no longer charged to its former owner. 3400 */ 3401 static __always_inline void skb_orphan(struct sk_buff *skb) 3402 { 3403 if (skb->destructor) { 3404 skb->destructor(skb); 3405 skb->destructor = NULL; 3406 skb->sk = NULL; 3407 } else { 3408 BUG_ON(skb->sk); 3409 } 3410 } 3411 3412 /** 3413 * skb_orphan_frags - orphan the frags contained in a buffer 3414 * @skb: buffer to orphan frags from 3415 * @gfp_mask: allocation mask for replacement pages 3416 * 3417 * For each frag in the SKB which needs a destructor (i.e. has an 3418 * owner) create a copy of that frag and release the original 3419 * page by calling the destructor. 3420 */ 3421 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3422 { 3423 if (likely(!skb_zcopy(skb))) 3424 return 0; 3425 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3426 return 0; 3427 return skb_copy_ubufs(skb, gfp_mask); 3428 } 3429 3430 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3431 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3432 { 3433 if (likely(!skb_zcopy(skb))) 3434 return 0; 3435 return skb_copy_ubufs(skb, gfp_mask); 3436 } 3437 3438 /** 3439 * __skb_queue_purge_reason - empty a list 3440 * @list: list to empty 3441 * @reason: drop reason 3442 * 3443 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3444 * the list and one reference dropped. This function does not take the 3445 * list lock and the caller must hold the relevant locks to use it. 3446 */ 3447 static inline void __skb_queue_purge_reason(struct sk_buff_head *list, 3448 enum skb_drop_reason reason) 3449 { 3450 struct sk_buff *skb; 3451 3452 while ((skb = __skb_dequeue(list)) != NULL) 3453 kfree_skb_reason(skb, reason); 3454 } 3455 3456 static inline void __skb_queue_purge(struct sk_buff_head *list) 3457 { 3458 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3459 } 3460 3461 void skb_queue_purge_reason(struct sk_buff_head *list, 3462 enum skb_drop_reason reason); 3463 3464 static inline void skb_queue_purge(struct sk_buff_head *list) 3465 { 3466 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3467 } 3468 3469 unsigned int skb_rbtree_purge(struct rb_root *root); 3470 void skb_errqueue_purge(struct sk_buff_head *list); 3471 3472 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3473 3474 /** 3475 * netdev_alloc_frag - allocate a page fragment 3476 * @fragsz: fragment size 3477 * 3478 * Allocates a frag from a page for receive buffer. 3479 * Uses GFP_ATOMIC allocations. 3480 */ 3481 static inline void *netdev_alloc_frag(unsigned int fragsz) 3482 { 3483 return __netdev_alloc_frag_align(fragsz, ~0u); 3484 } 3485 3486 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3487 unsigned int align) 3488 { 3489 WARN_ON_ONCE(!is_power_of_2(align)); 3490 return __netdev_alloc_frag_align(fragsz, -align); 3491 } 3492 3493 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3494 gfp_t gfp_mask); 3495 3496 /** 3497 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3498 * @dev: network device to receive on 3499 * @length: length to allocate 3500 * 3501 * Allocate a new &sk_buff and assign it a usage count of one. The 3502 * buffer has unspecified headroom built in. Users should allocate 3503 * the headroom they think they need without accounting for the 3504 * built in space. The built in space is used for optimisations. 3505 * 3506 * %NULL is returned if there is no free memory. Although this function 3507 * allocates memory it can be called from an interrupt. 3508 */ 3509 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3510 unsigned int length) 3511 { 3512 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3513 } 3514 3515 /* legacy helper around __netdev_alloc_skb() */ 3516 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3517 gfp_t gfp_mask) 3518 { 3519 return __netdev_alloc_skb(NULL, length, gfp_mask); 3520 } 3521 3522 /* legacy helper around netdev_alloc_skb() */ 3523 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3524 { 3525 return netdev_alloc_skb(NULL, length); 3526 } 3527 3528 3529 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3530 unsigned int length, gfp_t gfp) 3531 { 3532 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3533 3534 if (NET_IP_ALIGN && skb) 3535 skb_reserve(skb, NET_IP_ALIGN); 3536 return skb; 3537 } 3538 3539 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3540 unsigned int length) 3541 { 3542 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3543 } 3544 3545 static inline void skb_free_frag(void *addr) 3546 { 3547 page_frag_free(addr); 3548 } 3549 3550 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3551 3552 static inline void *napi_alloc_frag(unsigned int fragsz) 3553 { 3554 return __napi_alloc_frag_align(fragsz, ~0u); 3555 } 3556 3557 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3558 unsigned int align) 3559 { 3560 WARN_ON_ONCE(!is_power_of_2(align)); 3561 return __napi_alloc_frag_align(fragsz, -align); 3562 } 3563 3564 struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int length); 3565 void napi_consume_skb(struct sk_buff *skb, int budget); 3566 3567 void napi_skb_free_stolen_head(struct sk_buff *skb); 3568 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason); 3569 3570 /** 3571 * __dev_alloc_pages - allocate page for network Rx 3572 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3573 * @order: size of the allocation 3574 * 3575 * Allocate a new page. 3576 * 3577 * %NULL is returned if there is no free memory. 3578 */ 3579 static inline struct page *__dev_alloc_pages_noprof(gfp_t gfp_mask, 3580 unsigned int order) 3581 { 3582 /* This piece of code contains several assumptions. 3583 * 1. This is for device Rx, therefore a cold page is preferred. 3584 * 2. The expectation is the user wants a compound page. 3585 * 3. If requesting a order 0 page it will not be compound 3586 * due to the check to see if order has a value in prep_new_page 3587 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3588 * code in alloc_flags_slowpath() that should be enforcing this. 3589 */ 3590 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3591 3592 return alloc_pages_node_noprof(NUMA_NO_NODE, gfp_mask, order); 3593 } 3594 #define __dev_alloc_pages(...) alloc_hooks(__dev_alloc_pages_noprof(__VA_ARGS__)) 3595 3596 /* 3597 * This specialized allocator has to be a macro for its allocations to be 3598 * accounted separately (to have a separate alloc_tag). 3599 */ 3600 #define dev_alloc_pages(_order) __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, _order) 3601 3602 /** 3603 * __dev_alloc_page - allocate a page for network Rx 3604 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3605 * 3606 * Allocate a new page. 3607 * 3608 * %NULL is returned if there is no free memory. 3609 */ 3610 static inline struct page *__dev_alloc_page_noprof(gfp_t gfp_mask) 3611 { 3612 return __dev_alloc_pages_noprof(gfp_mask, 0); 3613 } 3614 #define __dev_alloc_page(...) alloc_hooks(__dev_alloc_page_noprof(__VA_ARGS__)) 3615 3616 /* 3617 * This specialized allocator has to be a macro for its allocations to be 3618 * accounted separately (to have a separate alloc_tag). 3619 */ 3620 #define dev_alloc_page() dev_alloc_pages(0) 3621 3622 /** 3623 * dev_page_is_reusable - check whether a page can be reused for network Rx 3624 * @page: the page to test 3625 * 3626 * A page shouldn't be considered for reusing/recycling if it was allocated 3627 * under memory pressure or at a distant memory node. 3628 * 3629 * Returns: false if this page should be returned to page allocator, true 3630 * otherwise. 3631 */ 3632 static inline bool dev_page_is_reusable(const struct page *page) 3633 { 3634 return likely(page_to_nid(page) == numa_mem_id() && 3635 !page_is_pfmemalloc(page)); 3636 } 3637 3638 /** 3639 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3640 * @page: The page that was allocated from skb_alloc_page 3641 * @skb: The skb that may need pfmemalloc set 3642 */ 3643 static inline void skb_propagate_pfmemalloc(const struct page *page, 3644 struct sk_buff *skb) 3645 { 3646 if (page_is_pfmemalloc(page)) 3647 skb->pfmemalloc = true; 3648 } 3649 3650 /** 3651 * skb_frag_off() - Returns the offset of a skb fragment 3652 * @frag: the paged fragment 3653 */ 3654 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3655 { 3656 return frag->offset; 3657 } 3658 3659 /** 3660 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3661 * @frag: skb fragment 3662 * @delta: value to add 3663 */ 3664 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3665 { 3666 frag->offset += delta; 3667 } 3668 3669 /** 3670 * skb_frag_off_set() - Sets the offset of a skb fragment 3671 * @frag: skb fragment 3672 * @offset: offset of fragment 3673 */ 3674 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3675 { 3676 frag->offset = offset; 3677 } 3678 3679 /** 3680 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3681 * @fragto: skb fragment where offset is set 3682 * @fragfrom: skb fragment offset is copied from 3683 */ 3684 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3685 const skb_frag_t *fragfrom) 3686 { 3687 fragto->offset = fragfrom->offset; 3688 } 3689 3690 /* Return: true if the skb_frag contains a net_iov. */ 3691 static inline bool skb_frag_is_net_iov(const skb_frag_t *frag) 3692 { 3693 return netmem_is_net_iov(frag->netmem); 3694 } 3695 3696 /** 3697 * skb_frag_net_iov - retrieve the net_iov referred to by fragment 3698 * @frag: the fragment 3699 * 3700 * Return: the &struct net_iov associated with @frag. Returns NULL if this 3701 * frag has no associated net_iov. 3702 */ 3703 static inline struct net_iov *skb_frag_net_iov(const skb_frag_t *frag) 3704 { 3705 if (!skb_frag_is_net_iov(frag)) 3706 return NULL; 3707 3708 return netmem_to_net_iov(frag->netmem); 3709 } 3710 3711 /** 3712 * skb_frag_page - retrieve the page referred to by a paged fragment 3713 * @frag: the paged fragment 3714 * 3715 * Return: the &struct page associated with @frag. Returns NULL if this frag 3716 * has no associated page. 3717 */ 3718 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3719 { 3720 if (skb_frag_is_net_iov(frag)) 3721 return NULL; 3722 3723 return netmem_to_page(frag->netmem); 3724 } 3725 3726 /** 3727 * skb_frag_netmem - retrieve the netmem referred to by a fragment 3728 * @frag: the fragment 3729 * 3730 * Return: the &netmem_ref associated with @frag. 3731 */ 3732 static inline netmem_ref skb_frag_netmem(const skb_frag_t *frag) 3733 { 3734 return frag->netmem; 3735 } 3736 3737 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb, 3738 unsigned int headroom); 3739 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb, 3740 const struct bpf_prog *prog); 3741 3742 /** 3743 * skb_frag_address - gets the address of the data contained in a paged fragment 3744 * @frag: the paged fragment buffer 3745 * 3746 * Returns: the address of the data within @frag. The page must already 3747 * be mapped. 3748 */ 3749 static inline void *skb_frag_address(const skb_frag_t *frag) 3750 { 3751 if (!skb_frag_page(frag)) 3752 return NULL; 3753 3754 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3755 } 3756 3757 /** 3758 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3759 * @frag: the paged fragment buffer 3760 * 3761 * Returns: the address of the data within @frag. Checks that the page 3762 * is mapped and returns %NULL otherwise. 3763 */ 3764 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3765 { 3766 struct page *page = skb_frag_page(frag); 3767 void *ptr; 3768 3769 if (!page) 3770 return NULL; 3771 3772 ptr = page_address(page); 3773 if (unlikely(!ptr)) 3774 return NULL; 3775 3776 return ptr + skb_frag_off(frag); 3777 } 3778 3779 /** 3780 * skb_frag_phys - gets the physical address of the data in a paged fragment 3781 * @frag: the paged fragment buffer 3782 * 3783 * Returns: the physical address of the data within @frag. 3784 */ 3785 static inline phys_addr_t skb_frag_phys(const skb_frag_t *frag) 3786 { 3787 return page_to_phys(skb_frag_page(frag)) + skb_frag_off(frag); 3788 } 3789 3790 /** 3791 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3792 * @fragto: skb fragment where page is set 3793 * @fragfrom: skb fragment page is copied from 3794 */ 3795 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3796 const skb_frag_t *fragfrom) 3797 { 3798 fragto->netmem = fragfrom->netmem; 3799 } 3800 3801 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3802 3803 /** 3804 * __skb_frag_dma_map - maps a paged fragment via the DMA API 3805 * @dev: the device to map the fragment to 3806 * @frag: the paged fragment to map 3807 * @offset: the offset within the fragment (starting at the 3808 * fragment's own offset) 3809 * @size: the number of bytes to map 3810 * @dir: the direction of the mapping (``PCI_DMA_*``) 3811 * 3812 * Maps the page associated with @frag to @device. 3813 */ 3814 static inline dma_addr_t __skb_frag_dma_map(struct device *dev, 3815 const skb_frag_t *frag, 3816 size_t offset, size_t size, 3817 enum dma_data_direction dir) 3818 { 3819 if (skb_frag_is_net_iov(frag)) { 3820 return netmem_to_net_iov(frag->netmem)->desc.dma_addr + 3821 offset + frag->offset; 3822 } 3823 return dma_map_page(dev, skb_frag_page(frag), 3824 skb_frag_off(frag) + offset, size, dir); 3825 } 3826 3827 #define skb_frag_dma_map(dev, frag, ...) \ 3828 CONCATENATE(_skb_frag_dma_map, \ 3829 COUNT_ARGS(__VA_ARGS__))(dev, frag, ##__VA_ARGS__) 3830 3831 #define __skb_frag_dma_map1(dev, frag, offset, uf, uo) ({ \ 3832 const skb_frag_t *uf = (frag); \ 3833 size_t uo = (offset); \ 3834 \ 3835 __skb_frag_dma_map(dev, uf, uo, skb_frag_size(uf) - uo, \ 3836 DMA_TO_DEVICE); \ 3837 }) 3838 #define _skb_frag_dma_map1(dev, frag, offset) \ 3839 __skb_frag_dma_map1(dev, frag, offset, __UNIQUE_ID(frag_), \ 3840 __UNIQUE_ID(offset_)) 3841 #define _skb_frag_dma_map0(dev, frag) \ 3842 _skb_frag_dma_map1(dev, frag, 0) 3843 #define _skb_frag_dma_map2(dev, frag, offset, size) \ 3844 __skb_frag_dma_map(dev, frag, offset, size, DMA_TO_DEVICE) 3845 #define _skb_frag_dma_map3(dev, frag, offset, size, dir) \ 3846 __skb_frag_dma_map(dev, frag, offset, size, dir) 3847 3848 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3849 gfp_t gfp_mask) 3850 { 3851 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3852 } 3853 3854 3855 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3856 gfp_t gfp_mask) 3857 { 3858 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3859 } 3860 3861 3862 /** 3863 * skb_clone_writable - is the header of a clone writable 3864 * @skb: buffer to check 3865 * @len: length up to which to write 3866 * 3867 * Returns true if modifying the header part of the cloned buffer 3868 * does not requires the data to be copied. 3869 */ 3870 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3871 { 3872 return !skb_header_cloned(skb) && 3873 skb_headroom(skb) + len <= skb->hdr_len; 3874 } 3875 3876 static inline int skb_try_make_writable(struct sk_buff *skb, 3877 unsigned int write_len) 3878 { 3879 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3880 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3881 } 3882 3883 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3884 int cloned) 3885 { 3886 int delta = 0; 3887 3888 if (headroom > skb_headroom(skb)) 3889 delta = headroom - skb_headroom(skb); 3890 3891 if (delta || cloned) 3892 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3893 GFP_ATOMIC); 3894 return 0; 3895 } 3896 3897 /** 3898 * skb_cow - copy header of skb when it is required 3899 * @skb: buffer to cow 3900 * @headroom: needed headroom 3901 * 3902 * If the skb passed lacks sufficient headroom or its data part 3903 * is shared, data is reallocated. If reallocation fails, an error 3904 * is returned and original skb is not changed. 3905 * 3906 * The result is skb with writable area skb->head...skb->tail 3907 * and at least @headroom of space at head. 3908 */ 3909 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3910 { 3911 return __skb_cow(skb, headroom, skb_cloned(skb)); 3912 } 3913 3914 /** 3915 * skb_cow_head - skb_cow but only making the head writable 3916 * @skb: buffer to cow 3917 * @headroom: needed headroom 3918 * 3919 * This function is identical to skb_cow except that we replace the 3920 * skb_cloned check by skb_header_cloned. It should be used when 3921 * you only need to push on some header and do not need to modify 3922 * the data. 3923 */ 3924 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3925 { 3926 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3927 } 3928 3929 /** 3930 * skb_padto - pad an skbuff up to a minimal size 3931 * @skb: buffer to pad 3932 * @len: minimal length 3933 * 3934 * Pads up a buffer to ensure the trailing bytes exist and are 3935 * blanked. If the buffer already contains sufficient data it 3936 * is untouched. Otherwise it is extended. Returns zero on 3937 * success. The skb is freed on error. 3938 */ 3939 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3940 { 3941 unsigned int size = skb->len; 3942 if (likely(size >= len)) 3943 return 0; 3944 return skb_pad(skb, len - size); 3945 } 3946 3947 /** 3948 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3949 * @skb: buffer to pad 3950 * @len: minimal length 3951 * @free_on_error: free buffer on error 3952 * 3953 * Pads up a buffer to ensure the trailing bytes exist and are 3954 * blanked. If the buffer already contains sufficient data it 3955 * is untouched. Otherwise it is extended. Returns zero on 3956 * success. The skb is freed on error if @free_on_error is true. 3957 */ 3958 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3959 unsigned int len, 3960 bool free_on_error) 3961 { 3962 unsigned int size = skb->len; 3963 3964 if (unlikely(size < len)) { 3965 len -= size; 3966 if (__skb_pad(skb, len, free_on_error)) 3967 return -ENOMEM; 3968 __skb_put(skb, len); 3969 } 3970 return 0; 3971 } 3972 3973 /** 3974 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3975 * @skb: buffer to pad 3976 * @len: minimal length 3977 * 3978 * Pads up a buffer to ensure the trailing bytes exist and are 3979 * blanked. If the buffer already contains sufficient data it 3980 * is untouched. Otherwise it is extended. Returns zero on 3981 * success. The skb is freed on error. 3982 */ 3983 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3984 { 3985 return __skb_put_padto(skb, len, true); 3986 } 3987 3988 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i) 3989 __must_check; 3990 3991 static inline bool skb_can_coalesce_netmem(struct sk_buff *skb, int i, 3992 netmem_ref netmem, int off) 3993 { 3994 if (skb_zcopy(skb)) 3995 return false; 3996 if (i) { 3997 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3998 3999 return netmem == skb_frag_netmem(frag) && 4000 off == skb_frag_off(frag) + skb_frag_size(frag); 4001 } 4002 return false; 4003 } 4004 4005 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 4006 const struct page *page, int off) 4007 { 4008 return skb_can_coalesce_netmem(skb, i, page_to_netmem(page), off); 4009 } 4010 4011 static inline int __skb_linearize(struct sk_buff *skb) 4012 { 4013 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 4014 } 4015 4016 /** 4017 * skb_linearize - convert paged skb to linear one 4018 * @skb: buffer to linarize 4019 * 4020 * If there is no free memory -ENOMEM is returned, otherwise zero 4021 * is returned and the old skb data released. 4022 */ 4023 static inline int skb_linearize(struct sk_buff *skb) 4024 { 4025 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 4026 } 4027 4028 /** 4029 * skb_has_shared_frag - can any frag be overwritten 4030 * @skb: buffer to test 4031 * 4032 * Return: true if the skb has at least one frag that might be modified 4033 * by an external entity (as in vmsplice()/sendfile()) 4034 */ 4035 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 4036 { 4037 return skb_is_nonlinear(skb) && 4038 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 4039 } 4040 4041 /** 4042 * skb_linearize_cow - make sure skb is linear and writable 4043 * @skb: buffer to process 4044 * 4045 * If there is no free memory -ENOMEM is returned, otherwise zero 4046 * is returned and the old skb data released. 4047 */ 4048 static inline int skb_linearize_cow(struct sk_buff *skb) 4049 { 4050 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 4051 __skb_linearize(skb) : 0; 4052 } 4053 4054 static __always_inline void 4055 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 4056 unsigned int off) 4057 { 4058 if (skb->ip_summed == CHECKSUM_COMPLETE) 4059 skb->csum = csum_block_sub(skb->csum, 4060 csum_partial(start, len, 0), off); 4061 else if (skb->ip_summed == CHECKSUM_PARTIAL && 4062 skb_checksum_start_offset(skb) < 0) 4063 skb->ip_summed = CHECKSUM_NONE; 4064 } 4065 4066 /** 4067 * skb_postpull_rcsum - update checksum for received skb after pull 4068 * @skb: buffer to update 4069 * @start: start of data before pull 4070 * @len: length of data pulled 4071 * 4072 * After doing a pull on a received packet, you need to call this to 4073 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 4074 * CHECKSUM_NONE so that it can be recomputed from scratch. 4075 */ 4076 static __always_inline void 4077 skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len) 4078 { 4079 if (skb->ip_summed == CHECKSUM_COMPLETE) 4080 skb->csum = wsum_negate(csum_partial(start, len, 4081 wsum_negate(skb->csum))); 4082 else if (skb->ip_summed == CHECKSUM_PARTIAL && 4083 skb_checksum_start_offset(skb) < 0) 4084 skb->ip_summed = CHECKSUM_NONE; 4085 } 4086 4087 static __always_inline void 4088 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 4089 unsigned int off) 4090 { 4091 if (skb->ip_summed == CHECKSUM_COMPLETE) 4092 skb->csum = csum_block_add(skb->csum, 4093 csum_partial(start, len, 0), off); 4094 } 4095 4096 /** 4097 * skb_postpush_rcsum - update checksum for received skb after push 4098 * @skb: buffer to update 4099 * @start: start of data after push 4100 * @len: length of data pushed 4101 * 4102 * After doing a push on a received packet, you need to call this to 4103 * update the CHECKSUM_COMPLETE checksum. 4104 */ 4105 static inline void skb_postpush_rcsum(struct sk_buff *skb, 4106 const void *start, unsigned int len) 4107 { 4108 __skb_postpush_rcsum(skb, start, len, 0); 4109 } 4110 4111 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 4112 4113 /** 4114 * skb_push_rcsum - push skb and update receive checksum 4115 * @skb: buffer to update 4116 * @len: length of data pulled 4117 * 4118 * This function performs an skb_push on the packet and updates 4119 * the CHECKSUM_COMPLETE checksum. It should be used on 4120 * receive path processing instead of skb_push unless you know 4121 * that the checksum difference is zero (e.g., a valid IP header) 4122 * or you are setting ip_summed to CHECKSUM_NONE. 4123 */ 4124 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 4125 { 4126 skb_push(skb, len); 4127 skb_postpush_rcsum(skb, skb->data, len); 4128 return skb->data; 4129 } 4130 4131 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 4132 /** 4133 * pskb_trim_rcsum - trim received skb and update checksum 4134 * @skb: buffer to trim 4135 * @len: new length 4136 * 4137 * This is exactly the same as pskb_trim except that it ensures the 4138 * checksum of received packets are still valid after the operation. 4139 * It can change skb pointers. 4140 */ 4141 4142 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4143 { 4144 skb_might_realloc(skb); 4145 if (likely(len >= skb->len)) 4146 return 0; 4147 return pskb_trim_rcsum_slow(skb, len); 4148 } 4149 4150 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4151 { 4152 if (skb->ip_summed == CHECKSUM_COMPLETE) 4153 skb->ip_summed = CHECKSUM_NONE; 4154 __skb_trim(skb, len); 4155 return 0; 4156 } 4157 4158 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 4159 { 4160 if (skb->ip_summed == CHECKSUM_COMPLETE) 4161 skb->ip_summed = CHECKSUM_NONE; 4162 return __skb_grow(skb, len); 4163 } 4164 4165 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 4166 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 4167 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 4168 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 4169 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 4170 4171 #define skb_queue_walk(queue, skb) \ 4172 for (skb = (queue)->next; \ 4173 skb != (struct sk_buff *)(queue); \ 4174 skb = skb->next) 4175 4176 #define skb_queue_walk_safe(queue, skb, tmp) \ 4177 for (skb = (queue)->next, tmp = skb->next; \ 4178 skb != (struct sk_buff *)(queue); \ 4179 skb = tmp, tmp = skb->next) 4180 4181 #define skb_queue_walk_from(queue, skb) \ 4182 for (; skb != (struct sk_buff *)(queue); \ 4183 skb = skb->next) 4184 4185 #define skb_rbtree_walk(skb, root) \ 4186 for (skb = skb_rb_first(root); skb != NULL; \ 4187 skb = skb_rb_next(skb)) 4188 4189 #define skb_rbtree_walk_from(skb) \ 4190 for (; skb != NULL; \ 4191 skb = skb_rb_next(skb)) 4192 4193 #define skb_rbtree_walk_from_safe(skb, tmp) \ 4194 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 4195 skb = tmp) 4196 4197 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 4198 for (tmp = skb->next; \ 4199 skb != (struct sk_buff *)(queue); \ 4200 skb = tmp, tmp = skb->next) 4201 4202 #define skb_queue_reverse_walk(queue, skb) \ 4203 for (skb = (queue)->prev; \ 4204 skb != (struct sk_buff *)(queue); \ 4205 skb = skb->prev) 4206 4207 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 4208 for (skb = (queue)->prev, tmp = skb->prev; \ 4209 skb != (struct sk_buff *)(queue); \ 4210 skb = tmp, tmp = skb->prev) 4211 4212 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 4213 for (tmp = skb->prev; \ 4214 skb != (struct sk_buff *)(queue); \ 4215 skb = tmp, tmp = skb->prev) 4216 4217 static inline bool skb_has_frag_list(const struct sk_buff *skb) 4218 { 4219 return skb_shinfo(skb)->frag_list != NULL; 4220 } 4221 4222 static inline void skb_frag_list_init(struct sk_buff *skb) 4223 { 4224 skb_shinfo(skb)->frag_list = NULL; 4225 } 4226 4227 #define skb_walk_frags(skb, iter) \ 4228 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 4229 4230 4231 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 4232 int *err, long *timeo_p, 4233 const struct sk_buff *skb); 4234 struct sk_buff *__skb_try_recv_from_queue(struct sk_buff_head *queue, 4235 unsigned int flags, 4236 int *off, int *err, 4237 struct sk_buff **last); 4238 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 4239 struct sk_buff_head *queue, 4240 unsigned int flags, int *off, int *err, 4241 struct sk_buff **last); 4242 struct sk_buff *__skb_recv_datagram(struct sock *sk, 4243 struct sk_buff_head *sk_queue, 4244 unsigned int flags, int *off, int *err); 4245 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 4246 __poll_t datagram_poll_queue(struct file *file, struct socket *sock, 4247 struct poll_table_struct *wait, 4248 struct sk_buff_head *rcv_queue); 4249 __poll_t datagram_poll(struct file *file, struct socket *sock, 4250 struct poll_table_struct *wait); 4251 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 4252 struct iov_iter *to, int size); 4253 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 4254 struct msghdr *msg, int size) 4255 { 4256 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 4257 } 4258 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 4259 struct msghdr *msg); 4260 int skb_copy_and_crc32c_datagram_iter(const struct sk_buff *skb, int offset, 4261 struct iov_iter *to, int len, u32 *crcp); 4262 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 4263 struct iov_iter *from, int len); 4264 int skb_copy_datagram_from_iter_full(struct sk_buff *skb, int offset, 4265 struct iov_iter *from, int len); 4266 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 4267 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 4268 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 4269 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 4270 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 4271 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 4272 int len); 4273 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 4274 struct pipe_inode_info *pipe, unsigned int len, 4275 unsigned int flags); 4276 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 4277 int len); 4278 int skb_send_sock_locked_with_flags(struct sock *sk, struct sk_buff *skb, 4279 int offset, int len, int flags); 4280 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 4281 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 4282 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 4283 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 4284 int len, int hlen); 4285 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 4286 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 4287 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 4288 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 4289 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 4290 unsigned int offset); 4291 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 4292 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 4293 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev); 4294 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 4295 int skb_vlan_pop(struct sk_buff *skb); 4296 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 4297 int skb_eth_pop(struct sk_buff *skb); 4298 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 4299 const unsigned char *src); 4300 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 4301 int mac_len, bool ethernet); 4302 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 4303 bool ethernet); 4304 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 4305 int skb_mpls_dec_ttl(struct sk_buff *skb); 4306 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 4307 gfp_t gfp); 4308 4309 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 4310 { 4311 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 4312 } 4313 4314 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 4315 { 4316 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 4317 } 4318 4319 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 4320 __wsum csum); 4321 u32 skb_crc32c(const struct sk_buff *skb, int offset, int len, u32 crc); 4322 4323 static inline void * __must_check 4324 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 4325 const void *data, int hlen, void *buffer) 4326 { 4327 if (likely(hlen - offset >= len)) 4328 return (void *)data + offset; 4329 4330 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 4331 return NULL; 4332 4333 return buffer; 4334 } 4335 4336 static __always_inline void * __must_check 4337 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4338 { 4339 return __skb_header_pointer(skb, offset, len, skb->data, 4340 skb_headlen(skb), buffer); 4341 } 4342 4343 /* Variant of skb_header_pointer() where @offset is user-controlled 4344 * and potentially negative. 4345 */ 4346 static inline void * __must_check 4347 skb_header_pointer_careful(const struct sk_buff *skb, int offset, 4348 int len, void *buffer) 4349 { 4350 if (unlikely(offset < 0 && -offset > skb_headroom(skb))) 4351 return NULL; 4352 return skb_header_pointer(skb, offset, len, buffer); 4353 } 4354 4355 static inline void * __must_check 4356 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len) 4357 { 4358 unsigned int uoffset = (unsigned int)offset; 4359 4360 if (likely(uoffset <= skb_headlen(skb) && 4361 (unsigned int)len <= skb_headlen(skb) - uoffset)) 4362 return skb->data + offset; 4363 return NULL; 4364 } 4365 4366 /** 4367 * skb_needs_linearize - check if we need to linearize a given skb 4368 * depending on the given device features. 4369 * @skb: socket buffer to check 4370 * @features: net device features 4371 * 4372 * Returns true if either: 4373 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4374 * 2. skb is fragmented and the device does not support SG. 4375 */ 4376 static inline bool skb_needs_linearize(struct sk_buff *skb, 4377 netdev_features_t features) 4378 { 4379 return skb_is_nonlinear(skb) && 4380 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4381 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4382 } 4383 4384 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4385 void *to, 4386 const unsigned int len) 4387 { 4388 memcpy(to, skb->data, len); 4389 } 4390 4391 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4392 const int offset, void *to, 4393 const unsigned int len) 4394 { 4395 memcpy(to, skb->data + offset, len); 4396 } 4397 4398 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4399 const void *from, 4400 const unsigned int len) 4401 { 4402 memcpy(skb->data, from, len); 4403 } 4404 4405 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4406 const int offset, 4407 const void *from, 4408 const unsigned int len) 4409 { 4410 memcpy(skb->data + offset, from, len); 4411 } 4412 4413 void skb_init(void); 4414 4415 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4416 { 4417 return skb->tstamp; 4418 } 4419 4420 /** 4421 * skb_get_timestamp - get timestamp from a skb 4422 * @skb: skb to get stamp from 4423 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4424 * 4425 * Timestamps are stored in the skb as offsets to a base timestamp. 4426 * This function converts the offset back to a struct timeval and stores 4427 * it in stamp. 4428 */ 4429 static inline void skb_get_timestamp(const struct sk_buff *skb, 4430 struct __kernel_old_timeval *stamp) 4431 { 4432 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4433 } 4434 4435 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4436 struct __kernel_sock_timeval *stamp) 4437 { 4438 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4439 4440 stamp->tv_sec = ts.tv_sec; 4441 stamp->tv_usec = ts.tv_nsec / 1000; 4442 } 4443 4444 static inline void skb_get_timestampns(const struct sk_buff *skb, 4445 struct __kernel_old_timespec *stamp) 4446 { 4447 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4448 4449 stamp->tv_sec = ts.tv_sec; 4450 stamp->tv_nsec = ts.tv_nsec; 4451 } 4452 4453 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4454 struct __kernel_timespec *stamp) 4455 { 4456 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4457 4458 stamp->tv_sec = ts.tv_sec; 4459 stamp->tv_nsec = ts.tv_nsec; 4460 } 4461 4462 static inline void __net_timestamp(struct sk_buff *skb) 4463 { 4464 skb->tstamp = ktime_get_real(); 4465 skb->tstamp_type = SKB_CLOCK_REALTIME; 4466 } 4467 4468 static inline ktime_t net_timedelta(ktime_t t) 4469 { 4470 return ktime_sub(ktime_get_real(), t); 4471 } 4472 4473 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4474 u8 tstamp_type) 4475 { 4476 skb->tstamp = kt; 4477 4478 if (kt) 4479 skb->tstamp_type = tstamp_type; 4480 else 4481 skb->tstamp_type = SKB_CLOCK_REALTIME; 4482 } 4483 4484 static inline void skb_set_delivery_type_by_clockid(struct sk_buff *skb, 4485 ktime_t kt, clockid_t clockid) 4486 { 4487 u8 tstamp_type = SKB_CLOCK_REALTIME; 4488 4489 switch (clockid) { 4490 case CLOCK_REALTIME: 4491 break; 4492 case CLOCK_MONOTONIC: 4493 tstamp_type = SKB_CLOCK_MONOTONIC; 4494 break; 4495 case CLOCK_TAI: 4496 tstamp_type = SKB_CLOCK_TAI; 4497 break; 4498 default: 4499 WARN_ON_ONCE(1); 4500 kt = 0; 4501 } 4502 4503 skb_set_delivery_time(skb, kt, tstamp_type); 4504 } 4505 4506 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4507 4508 /* It is used in the ingress path to clear the delivery_time. 4509 * If needed, set the skb->tstamp to the (rcv) timestamp. 4510 */ 4511 static __always_inline void skb_clear_delivery_time(struct sk_buff *skb) 4512 { 4513 if (skb->tstamp_type) { 4514 skb->tstamp_type = SKB_CLOCK_REALTIME; 4515 if (static_branch_unlikely(&netstamp_needed_key)) 4516 skb->tstamp = ktime_get_real(); 4517 else 4518 skb->tstamp = 0; 4519 } 4520 } 4521 4522 static inline void skb_clear_tstamp(struct sk_buff *skb) 4523 { 4524 if (skb->tstamp_type) 4525 return; 4526 4527 skb->tstamp = 0; 4528 } 4529 4530 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4531 { 4532 if (skb->tstamp_type) 4533 return 0; 4534 4535 return skb->tstamp; 4536 } 4537 4538 static __always_inline ktime_t 4539 skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4540 { 4541 if (skb->tstamp_type != SKB_CLOCK_MONOTONIC && skb->tstamp) 4542 return skb->tstamp; 4543 4544 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4545 return ktime_get_real(); 4546 4547 return 0; 4548 } 4549 4550 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4551 { 4552 return skb_shinfo(skb)->meta_len; 4553 } 4554 4555 static inline void *skb_metadata_end(const struct sk_buff *skb) 4556 { 4557 return skb_mac_header(skb); 4558 } 4559 4560 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4561 const struct sk_buff *skb_b, 4562 u8 meta_len) 4563 { 4564 const void *a = skb_metadata_end(skb_a); 4565 const void *b = skb_metadata_end(skb_b); 4566 u64 diffs = 0; 4567 4568 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) || 4569 BITS_PER_LONG != 64) 4570 goto slow; 4571 4572 /* Using more efficient variant than plain call to memcmp(). */ 4573 switch (meta_len) { 4574 #define __it(x, op) (x -= sizeof(u##op)) 4575 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4576 case 32: diffs |= __it_diff(a, b, 64); 4577 fallthrough; 4578 case 24: diffs |= __it_diff(a, b, 64); 4579 fallthrough; 4580 case 16: diffs |= __it_diff(a, b, 64); 4581 fallthrough; 4582 case 8: diffs |= __it_diff(a, b, 64); 4583 break; 4584 case 28: diffs |= __it_diff(a, b, 64); 4585 fallthrough; 4586 case 20: diffs |= __it_diff(a, b, 64); 4587 fallthrough; 4588 case 12: diffs |= __it_diff(a, b, 64); 4589 fallthrough; 4590 case 4: diffs |= __it_diff(a, b, 32); 4591 break; 4592 default: 4593 slow: 4594 return memcmp(a - meta_len, b - meta_len, meta_len); 4595 } 4596 return diffs; 4597 } 4598 4599 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4600 const struct sk_buff *skb_b) 4601 { 4602 u8 len_a = skb_metadata_len(skb_a); 4603 u8 len_b = skb_metadata_len(skb_b); 4604 4605 if (!(len_a | len_b)) 4606 return false; 4607 4608 return len_a != len_b ? 4609 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4610 } 4611 4612 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4613 { 4614 skb_shinfo(skb)->meta_len = meta_len; 4615 } 4616 4617 static inline void skb_metadata_clear(struct sk_buff *skb) 4618 { 4619 skb_metadata_set(skb, 0); 4620 } 4621 4622 /** 4623 * skb_data_move - Move packet data and metadata after skb_push() or skb_pull(). 4624 * @skb: packet to operate on 4625 * @len: number of bytes pushed or pulled from &sk_buff->data 4626 * @n: number of bytes to memmove() from pre-push/pull &sk_buff->data 4627 * 4628 * Moves @n bytes of packet data, can be zero, and all bytes of skb metadata. 4629 * 4630 * Assumes metadata is located immediately before &sk_buff->data prior to the 4631 * push/pull, and that sufficient headroom exists to hold it after an 4632 * skb_push(). Otherwise, metadata is cleared and a one-time warning is issued. 4633 * 4634 * Prefer skb_postpull_data_move() or skb_postpush_data_move() to calling this 4635 * helper directly. 4636 */ 4637 static inline void skb_data_move(struct sk_buff *skb, const int len, 4638 const unsigned int n) 4639 { 4640 const u8 meta_len = skb_metadata_len(skb); 4641 u8 *meta, *meta_end; 4642 4643 if (!len || (!n && !meta_len)) 4644 return; 4645 4646 if (!meta_len) 4647 goto no_metadata; 4648 4649 meta_end = skb_metadata_end(skb); 4650 meta = meta_end - meta_len; 4651 4652 if (WARN_ON_ONCE(meta_end + len != skb->data || 4653 meta_len > skb_headroom(skb))) { 4654 skb_metadata_clear(skb); 4655 goto no_metadata; 4656 } 4657 4658 memmove(meta + len, meta, meta_len + n); 4659 return; 4660 4661 no_metadata: 4662 memmove(skb->data, skb->data - len, n); 4663 } 4664 4665 /** 4666 * skb_postpull_data_move - Move packet data and metadata after skb_pull(). 4667 * @skb: packet to operate on 4668 * @len: number of bytes pulled from &sk_buff->data 4669 * @n: number of bytes to memmove() from pre-pull &sk_buff->data 4670 * 4671 * See skb_data_move() for details. 4672 */ 4673 static inline void skb_postpull_data_move(struct sk_buff *skb, 4674 const unsigned int len, 4675 const unsigned int n) 4676 { 4677 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 4678 skb_data_move(skb, len, n); 4679 } 4680 4681 /** 4682 * skb_postpush_data_move - Move packet data and metadata after skb_push(). 4683 * @skb: packet to operate on 4684 * @len: number of bytes pushed onto &sk_buff->data 4685 * @n: number of bytes to memmove() from pre-push &sk_buff->data 4686 * 4687 * See skb_data_move() for details. 4688 */ 4689 static inline void skb_postpush_data_move(struct sk_buff *skb, 4690 const unsigned int len, 4691 const unsigned int n) 4692 { 4693 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 4694 skb_data_move(skb, -len, n); 4695 } 4696 4697 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4698 4699 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4700 4701 void skb_clone_tx_timestamp(struct sk_buff *skb); 4702 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4703 4704 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4705 4706 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4707 { 4708 } 4709 4710 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4711 { 4712 return false; 4713 } 4714 4715 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4716 4717 /** 4718 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4719 * 4720 * PHY drivers may accept clones of transmitted packets for 4721 * timestamping via their phy_driver.txtstamp method. These drivers 4722 * must call this function to return the skb back to the stack with a 4723 * timestamp. 4724 * 4725 * @skb: clone of the original outgoing packet 4726 * @hwtstamps: hardware time stamps 4727 * 4728 */ 4729 void skb_complete_tx_timestamp(struct sk_buff *skb, 4730 struct skb_shared_hwtstamps *hwtstamps); 4731 4732 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4733 struct skb_shared_hwtstamps *hwtstamps, 4734 struct sock *sk, int tstype); 4735 4736 /** 4737 * skb_tstamp_tx - queue clone of skb with send time stamps 4738 * @orig_skb: the original outgoing packet 4739 * @hwtstamps: hardware time stamps, may be NULL if not available 4740 * 4741 * If the skb has a socket associated, then this function clones the 4742 * skb (thus sharing the actual data and optional structures), stores 4743 * the optional hardware time stamping information (if non NULL) or 4744 * generates a software time stamp (otherwise), then queues the clone 4745 * to the error queue of the socket. Errors are silently ignored. 4746 */ 4747 void skb_tstamp_tx(struct sk_buff *orig_skb, 4748 struct skb_shared_hwtstamps *hwtstamps); 4749 4750 /** 4751 * skb_tx_timestamp() - Driver hook for transmit timestamping 4752 * 4753 * Ethernet MAC Drivers should call this function in their hard_xmit() 4754 * function immediately before giving the sk_buff to the MAC hardware. 4755 * 4756 * Specifically, one should make absolutely sure that this function is 4757 * called before TX completion of this packet can trigger. Otherwise 4758 * the packet could potentially already be freed. 4759 * 4760 * @skb: A socket buffer. 4761 */ 4762 static inline void skb_tx_timestamp(struct sk_buff *skb) 4763 { 4764 skb_clone_tx_timestamp(skb); 4765 if (skb_shinfo(skb)->tx_flags & (SKBTX_SW_TSTAMP | SKBTX_BPF)) 4766 skb_tstamp_tx(skb, NULL); 4767 } 4768 4769 /** 4770 * skb_complete_wifi_ack - deliver skb with wifi status 4771 * 4772 * @skb: the original outgoing packet 4773 * @acked: ack status 4774 * 4775 */ 4776 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4777 4778 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4779 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4780 4781 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4782 { 4783 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4784 skb->csum_valid || 4785 (skb->ip_summed == CHECKSUM_PARTIAL && 4786 skb_checksum_start_offset(skb) >= 0)); 4787 } 4788 4789 /** 4790 * skb_checksum_complete - Calculate checksum of an entire packet 4791 * @skb: packet to process 4792 * 4793 * This function calculates the checksum over the entire packet plus 4794 * the value of skb->csum. The latter can be used to supply the 4795 * checksum of a pseudo header as used by TCP/UDP. It returns the 4796 * checksum. 4797 * 4798 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4799 * this function can be used to verify that checksum on received 4800 * packets. In that case the function should return zero if the 4801 * checksum is correct. In particular, this function will return zero 4802 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4803 * hardware has already verified the correctness of the checksum. 4804 */ 4805 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4806 { 4807 return skb_csum_unnecessary(skb) ? 4808 0 : __skb_checksum_complete(skb); 4809 } 4810 4811 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4812 { 4813 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4814 if (skb->csum_level == 0) 4815 skb->ip_summed = CHECKSUM_NONE; 4816 else 4817 skb->csum_level--; 4818 } 4819 } 4820 4821 static __always_inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4822 { 4823 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4824 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4825 skb->csum_level++; 4826 } else if (skb->ip_summed == CHECKSUM_NONE) { 4827 skb->ip_summed = CHECKSUM_UNNECESSARY; 4828 skb->csum_level = 0; 4829 } 4830 } 4831 4832 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4833 { 4834 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4835 skb->ip_summed = CHECKSUM_NONE; 4836 skb->csum_level = 0; 4837 } 4838 } 4839 4840 /* Check if we need to perform checksum complete validation. 4841 * 4842 * Returns: true if checksum complete is needed, false otherwise 4843 * (either checksum is unnecessary or zero checksum is allowed). 4844 */ 4845 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4846 bool zero_okay, 4847 __sum16 check) 4848 { 4849 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4850 skb->csum_valid = 1; 4851 __skb_decr_checksum_unnecessary(skb); 4852 return false; 4853 } 4854 4855 return true; 4856 } 4857 4858 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4859 * in checksum_init. 4860 */ 4861 #define CHECKSUM_BREAK 76 4862 4863 /* Unset checksum-complete 4864 * 4865 * Unset checksum complete can be done when packet is being modified 4866 * (uncompressed for instance) and checksum-complete value is 4867 * invalidated. 4868 */ 4869 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4870 { 4871 if (skb->ip_summed == CHECKSUM_COMPLETE) 4872 skb->ip_summed = CHECKSUM_NONE; 4873 } 4874 4875 /* Validate (init) checksum based on checksum complete. 4876 * 4877 * Return values: 4878 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4879 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4880 * checksum is stored in skb->csum for use in __skb_checksum_complete 4881 * non-zero: value of invalid checksum 4882 * 4883 */ 4884 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4885 bool complete, 4886 __wsum psum) 4887 { 4888 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4889 if (!csum_fold(csum_add(psum, skb->csum))) { 4890 skb->csum_valid = 1; 4891 return 0; 4892 } 4893 } 4894 4895 skb->csum = psum; 4896 4897 if (complete || skb->len <= CHECKSUM_BREAK) { 4898 __sum16 csum; 4899 4900 csum = __skb_checksum_complete(skb); 4901 skb->csum_valid = !csum; 4902 return csum; 4903 } 4904 4905 return 0; 4906 } 4907 4908 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4909 { 4910 return 0; 4911 } 4912 4913 /* Perform checksum validate (init). Note that this is a macro since we only 4914 * want to calculate the pseudo header which is an input function if necessary. 4915 * First we try to validate without any computation (checksum unnecessary) and 4916 * then calculate based on checksum complete calling the function to compute 4917 * pseudo header. 4918 * 4919 * Return values: 4920 * 0: checksum is validated or try to in skb_checksum_complete 4921 * non-zero: value of invalid checksum 4922 */ 4923 #define __skb_checksum_validate(skb, proto, complete, \ 4924 zero_okay, check, compute_pseudo) \ 4925 ({ \ 4926 __sum16 __ret = 0; \ 4927 skb->csum_valid = 0; \ 4928 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4929 __ret = __skb_checksum_validate_complete(skb, \ 4930 complete, compute_pseudo(skb, proto)); \ 4931 __ret; \ 4932 }) 4933 4934 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4935 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4936 4937 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4938 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4939 4940 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4941 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4942 4943 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4944 compute_pseudo) \ 4945 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4946 4947 #define skb_checksum_simple_validate(skb) \ 4948 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4949 4950 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4951 { 4952 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4953 } 4954 4955 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4956 { 4957 skb->csum = ~pseudo; 4958 skb->ip_summed = CHECKSUM_COMPLETE; 4959 } 4960 4961 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4962 do { \ 4963 if (__skb_checksum_convert_check(skb)) \ 4964 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4965 } while (0) 4966 4967 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4968 u16 start, u16 offset) 4969 { 4970 skb->ip_summed = CHECKSUM_PARTIAL; 4971 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4972 skb->csum_offset = offset - start; 4973 } 4974 4975 /* Update skbuf and packet to reflect the remote checksum offload operation. 4976 * When called, ptr indicates the starting point for skb->csum when 4977 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4978 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4979 */ 4980 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4981 int start, int offset, bool nopartial) 4982 { 4983 __wsum delta; 4984 4985 if (!nopartial) { 4986 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4987 return; 4988 } 4989 4990 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4991 __skb_checksum_complete(skb); 4992 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4993 } 4994 4995 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4996 4997 /* Adjust skb->csum since we changed the packet */ 4998 skb->csum = csum_add(skb->csum, delta); 4999 } 5000 5001 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 5002 { 5003 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 5004 return (void *)(skb->_nfct & NFCT_PTRMASK); 5005 #else 5006 return NULL; 5007 #endif 5008 } 5009 5010 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 5011 { 5012 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 5013 return skb->_nfct; 5014 #else 5015 return 0UL; 5016 #endif 5017 } 5018 5019 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 5020 { 5021 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 5022 DEBUG_NET_WARN_ON_ONCE(skb->_nfct & NFCT_PTRMASK); 5023 skb->slow_gro |= !!nfct; 5024 skb->_nfct = nfct; 5025 #endif 5026 } 5027 5028 #ifdef CONFIG_SKB_EXTENSIONS 5029 enum skb_ext_id { 5030 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 5031 SKB_EXT_BRIDGE_NF, 5032 #endif 5033 #ifdef CONFIG_XFRM 5034 SKB_EXT_SEC_PATH, 5035 #endif 5036 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5037 TC_SKB_EXT, 5038 #endif 5039 #if IS_ENABLED(CONFIG_MPTCP) 5040 SKB_EXT_MPTCP, 5041 #endif 5042 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 5043 SKB_EXT_MCTP, 5044 #endif 5045 #if IS_ENABLED(CONFIG_INET_PSP) 5046 SKB_EXT_PSP, 5047 #endif 5048 #if IS_ENABLED(CONFIG_CAN) 5049 SKB_EXT_CAN, 5050 #endif 5051 SKB_EXT_NUM, /* must be last */ 5052 }; 5053 5054 /** 5055 * struct skb_ext - sk_buff extensions 5056 * @refcnt: 1 on allocation, deallocated on 0 5057 * @offset: offset to add to @data to obtain extension address 5058 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 5059 * @data: start of extension data, variable sized 5060 * 5061 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 5062 * to use 'u8' types while allowing up to 2kb worth of extension data. 5063 */ 5064 struct skb_ext { 5065 refcount_t refcnt; 5066 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 5067 u8 chunks; /* same */ 5068 char data[] __aligned(8); 5069 }; 5070 5071 struct skb_ext *__skb_ext_alloc(gfp_t flags); 5072 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 5073 struct skb_ext *ext); 5074 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 5075 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 5076 void __skb_ext_put(struct skb_ext *ext); 5077 5078 static inline void skb_ext_put(struct sk_buff *skb) 5079 { 5080 if (skb->active_extensions) 5081 __skb_ext_put(skb->extensions); 5082 } 5083 5084 static inline void __skb_ext_copy(struct sk_buff *dst, 5085 const struct sk_buff *src) 5086 { 5087 dst->active_extensions = src->active_extensions; 5088 5089 if (src->active_extensions) { 5090 struct skb_ext *ext = src->extensions; 5091 5092 refcount_inc(&ext->refcnt); 5093 dst->extensions = ext; 5094 } 5095 } 5096 5097 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 5098 { 5099 skb_ext_put(dst); 5100 __skb_ext_copy(dst, src); 5101 } 5102 5103 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 5104 { 5105 return !!ext->offset[i]; 5106 } 5107 5108 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 5109 { 5110 return skb->active_extensions & (1 << id); 5111 } 5112 5113 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 5114 { 5115 if (skb_ext_exist(skb, id)) 5116 __skb_ext_del(skb, id); 5117 } 5118 5119 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 5120 { 5121 if (skb_ext_exist(skb, id)) { 5122 struct skb_ext *ext = skb->extensions; 5123 5124 return (void *)ext + (ext->offset[id] << 3); 5125 } 5126 5127 return NULL; 5128 } 5129 5130 static inline void skb_ext_reset(struct sk_buff *skb) 5131 { 5132 if (unlikely(skb->active_extensions)) { 5133 __skb_ext_put(skb->extensions); 5134 skb->active_extensions = 0; 5135 } 5136 } 5137 5138 static inline bool skb_has_extensions(struct sk_buff *skb) 5139 { 5140 return unlikely(skb->active_extensions); 5141 } 5142 #else 5143 static inline void __skb_ext_put(struct skb_ext *ext) {} 5144 static inline void skb_ext_put(struct sk_buff *skb) {} 5145 static inline void skb_ext_reset(struct sk_buff *skb) {} 5146 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 5147 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 5148 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 5149 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 5150 #endif /* CONFIG_SKB_EXTENSIONS */ 5151 5152 static inline void nf_reset_ct(struct sk_buff *skb) 5153 { 5154 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5155 nf_conntrack_put(skb_nfct(skb)); 5156 skb->_nfct = 0; 5157 #endif 5158 } 5159 5160 static inline void nf_reset_trace(struct sk_buff *skb) 5161 { 5162 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 5163 skb->nf_trace = 0; 5164 #endif 5165 } 5166 5167 static inline void ipvs_reset(struct sk_buff *skb) 5168 { 5169 #if IS_ENABLED(CONFIG_IP_VS) 5170 skb->ipvs_property = 0; 5171 #endif 5172 } 5173 5174 /* Note: This doesn't put any conntrack info in dst. */ 5175 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 5176 bool copy) 5177 { 5178 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5179 dst->_nfct = src->_nfct; 5180 nf_conntrack_get(skb_nfct(src)); 5181 #endif 5182 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 5183 if (copy) 5184 dst->nf_trace = src->nf_trace; 5185 #endif 5186 } 5187 5188 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 5189 { 5190 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5191 nf_conntrack_put(skb_nfct(dst)); 5192 #endif 5193 dst->slow_gro = src->slow_gro; 5194 __nf_copy(dst, src, true); 5195 } 5196 5197 #ifdef CONFIG_NETWORK_SECMARK 5198 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 5199 { 5200 to->secmark = from->secmark; 5201 } 5202 5203 static inline void skb_init_secmark(struct sk_buff *skb) 5204 { 5205 skb->secmark = 0; 5206 } 5207 #else 5208 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 5209 { } 5210 5211 static inline void skb_init_secmark(struct sk_buff *skb) 5212 { } 5213 #endif 5214 5215 static inline int secpath_exists(const struct sk_buff *skb) 5216 { 5217 #ifdef CONFIG_XFRM 5218 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 5219 #else 5220 return 0; 5221 #endif 5222 } 5223 5224 static inline bool skb_irq_freeable(const struct sk_buff *skb) 5225 { 5226 return !skb->destructor && 5227 !secpath_exists(skb) && 5228 !skb_nfct(skb) && 5229 !skb->_skb_refdst && 5230 !skb_has_frag_list(skb); 5231 } 5232 5233 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 5234 { 5235 skb->queue_mapping = queue_mapping; 5236 } 5237 5238 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 5239 { 5240 return skb->queue_mapping; 5241 } 5242 5243 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 5244 { 5245 to->queue_mapping = from->queue_mapping; 5246 } 5247 5248 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 5249 { 5250 skb->queue_mapping = rx_queue + 1; 5251 } 5252 5253 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 5254 { 5255 return skb->queue_mapping - 1; 5256 } 5257 5258 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 5259 { 5260 return skb->queue_mapping != 0; 5261 } 5262 5263 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 5264 { 5265 skb->dst_pending_confirm = val; 5266 } 5267 5268 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 5269 { 5270 return skb->dst_pending_confirm != 0; 5271 } 5272 5273 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 5274 { 5275 #ifdef CONFIG_XFRM 5276 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 5277 #else 5278 return NULL; 5279 #endif 5280 } 5281 5282 static inline bool skb_is_gso(const struct sk_buff *skb) 5283 { 5284 return skb_shinfo(skb)->gso_size; 5285 } 5286 5287 /* Note: Should be called only if skb_is_gso(skb) is true */ 5288 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 5289 { 5290 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 5291 } 5292 5293 /* Note: Should be called only if skb_is_gso(skb) is true */ 5294 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 5295 { 5296 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 5297 } 5298 5299 /* Note: Should be called only if skb_is_gso(skb) is true */ 5300 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 5301 { 5302 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 5303 } 5304 5305 static inline void skb_gso_reset(struct sk_buff *skb) 5306 { 5307 skb_shinfo(skb)->gso_size = 0; 5308 skb_shinfo(skb)->gso_segs = 0; 5309 skb_shinfo(skb)->gso_type = 0; 5310 } 5311 5312 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 5313 u16 increment) 5314 { 5315 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5316 return; 5317 shinfo->gso_size += increment; 5318 } 5319 5320 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 5321 u16 decrement) 5322 { 5323 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5324 return; 5325 shinfo->gso_size -= decrement; 5326 } 5327 5328 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 5329 5330 static __always_inline bool skb_warn_if_lro(const struct sk_buff *skb) 5331 { 5332 /* LRO sets gso_size but not gso_type, whereas if GSO is really 5333 * wanted then gso_type will be set. */ 5334 const struct skb_shared_info *shinfo = skb_shinfo(skb); 5335 5336 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 5337 unlikely(shinfo->gso_type == 0)) { 5338 __skb_warn_lro_forwarding(skb); 5339 return true; 5340 } 5341 return false; 5342 } 5343 5344 static inline void skb_forward_csum(struct sk_buff *skb) 5345 { 5346 /* Unfortunately we don't support this one. Any brave souls? */ 5347 if (skb->ip_summed == CHECKSUM_COMPLETE) 5348 skb->ip_summed = CHECKSUM_NONE; 5349 } 5350 5351 /** 5352 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 5353 * @skb: skb to check 5354 * 5355 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 5356 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 5357 * use this helper, to document places where we make this assertion. 5358 */ 5359 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 5360 { 5361 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 5362 } 5363 5364 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 5365 5366 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 5367 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 5368 unsigned int transport_len, 5369 __sum16(*skb_chkf)(struct sk_buff *skb)); 5370 5371 /** 5372 * skb_head_is_locked - Determine if the skb->head is locked down 5373 * @skb: skb to check 5374 * 5375 * The head on skbs build around a head frag can be removed if they are 5376 * not cloned. This function returns true if the skb head is locked down 5377 * due to either being allocated via kmalloc, or by being a clone with 5378 * multiple references to the head. 5379 */ 5380 static inline bool skb_head_is_locked(const struct sk_buff *skb) 5381 { 5382 return !skb->head_frag || skb_cloned(skb); 5383 } 5384 5385 /* Local Checksum Offload. 5386 * Compute outer checksum based on the assumption that the 5387 * inner checksum will be offloaded later. 5388 * See Documentation/networking/checksum-offloads.rst for 5389 * explanation of how this works. 5390 * Fill in outer checksum adjustment (e.g. with sum of outer 5391 * pseudo-header) before calling. 5392 * Also ensure that inner checksum is in linear data area. 5393 */ 5394 static inline __wsum lco_csum(struct sk_buff *skb) 5395 { 5396 unsigned char *csum_start = skb_checksum_start(skb); 5397 unsigned char *l4_hdr = skb_transport_header(skb); 5398 __wsum partial; 5399 5400 /* Start with complement of inner checksum adjustment */ 5401 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 5402 skb->csum_offset)); 5403 5404 /* Add in checksum of our headers (incl. outer checksum 5405 * adjustment filled in by caller) and return result. 5406 */ 5407 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5408 } 5409 5410 static inline bool skb_is_redirected(const struct sk_buff *skb) 5411 { 5412 return skb->redirected; 5413 } 5414 5415 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5416 { 5417 skb->redirected = 1; 5418 #ifdef CONFIG_NET_REDIRECT 5419 skb->from_ingress = from_ingress; 5420 if (skb->from_ingress) 5421 skb_clear_tstamp(skb); 5422 #endif 5423 } 5424 5425 static inline void skb_reset_redirect(struct sk_buff *skb) 5426 { 5427 skb->redirected = 0; 5428 } 5429 5430 static inline void skb_set_redirected_noclear(struct sk_buff *skb, 5431 bool from_ingress) 5432 { 5433 skb->redirected = 1; 5434 #ifdef CONFIG_NET_REDIRECT 5435 skb->from_ingress = from_ingress; 5436 #endif 5437 } 5438 5439 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5440 { 5441 #if IS_ENABLED(CONFIG_IP_SCTP) 5442 return skb->csum_not_inet; 5443 #else 5444 return 0; 5445 #endif 5446 } 5447 5448 static inline void skb_reset_csum_not_inet(struct sk_buff *skb) 5449 { 5450 skb->ip_summed = CHECKSUM_NONE; 5451 #if IS_ENABLED(CONFIG_IP_SCTP) 5452 skb->csum_not_inet = 0; 5453 #endif 5454 } 5455 5456 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5457 struct kcov_common_handle_id kcov_handle) 5458 { 5459 skb->kcov_handle = kcov_handle; 5460 } 5461 5462 static inline struct kcov_common_handle_id skb_get_kcov_handle(struct sk_buff *skb) 5463 { 5464 return skb->kcov_handle; 5465 } 5466 5467 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5468 { 5469 #ifdef CONFIG_PAGE_POOL 5470 skb->pp_recycle = 1; 5471 #endif 5472 } 5473 5474 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter, 5475 ssize_t maxsize); 5476 5477 #endif /* __KERNEL__ */ 5478 #endif /* _LINUX_SKBUFF_H */ 5479