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 skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1838 { 1839 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1840 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1841 } 1842 1843 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1844 { 1845 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1846 } 1847 1848 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1849 { 1850 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1851 } 1852 1853 static inline void net_zcopy_put(struct ubuf_info *uarg) 1854 { 1855 if (uarg) 1856 uarg->ops->complete(NULL, uarg, true); 1857 } 1858 1859 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1860 { 1861 if (uarg) { 1862 if (uarg->ops == &msg_zerocopy_ubuf_ops) 1863 msg_zerocopy_put_abort(uarg, have_uref); 1864 else if (have_uref) 1865 net_zcopy_put(uarg); 1866 } 1867 } 1868 1869 /* Release a reference on a zerocopy structure */ 1870 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1871 { 1872 struct ubuf_info *uarg = skb_zcopy(skb); 1873 1874 if (uarg) { 1875 if (!skb_zcopy_is_nouarg(skb)) 1876 uarg->ops->complete(skb, uarg, zerocopy_success); 1877 1878 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1879 } 1880 } 1881 1882 void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1883 1884 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1885 { 1886 if (unlikely(skb_zcopy_managed(skb))) 1887 __skb_zcopy_downgrade_managed(skb); 1888 } 1889 1890 /* Return true if frags in this skb are readable by the host. */ 1891 static inline bool skb_frags_readable(const struct sk_buff *skb) 1892 { 1893 return !skb->unreadable; 1894 } 1895 1896 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1897 { 1898 skb->next = NULL; 1899 } 1900 1901 static inline void skb_poison_list(struct sk_buff *skb) 1902 { 1903 #ifdef CONFIG_DEBUG_NET 1904 skb->next = SKB_LIST_POISON_NEXT; 1905 #endif 1906 } 1907 1908 /* Iterate through singly-linked GSO fragments of an skb. */ 1909 #define skb_list_walk_safe(first, skb, next_skb) \ 1910 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1911 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1912 1913 static inline void skb_list_del_init(struct sk_buff *skb) 1914 { 1915 __list_del_entry(&skb->list); 1916 skb_mark_not_on_list(skb); 1917 } 1918 1919 /** 1920 * skb_queue_empty - check if a queue is empty 1921 * @list: queue head 1922 * 1923 * Returns true if the queue is empty, false otherwise. 1924 */ 1925 static inline int skb_queue_empty(const struct sk_buff_head *list) 1926 { 1927 return list->next == (const struct sk_buff *) list; 1928 } 1929 1930 /** 1931 * skb_queue_empty_lockless - check if a queue is empty 1932 * @list: queue head 1933 * 1934 * Returns true if the queue is empty, false otherwise. 1935 * This variant can be used in lockless contexts. 1936 */ 1937 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1938 { 1939 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1940 } 1941 1942 1943 /** 1944 * skb_queue_is_last - check if skb is the last entry in the queue 1945 * @list: queue head 1946 * @skb: buffer 1947 * 1948 * Returns true if @skb is the last buffer on the list. 1949 */ 1950 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1951 const struct sk_buff *skb) 1952 { 1953 return skb->next == (const struct sk_buff *) list; 1954 } 1955 1956 /** 1957 * skb_queue_is_first - check if skb is the first entry in the queue 1958 * @list: queue head 1959 * @skb: buffer 1960 * 1961 * Returns true if @skb is the first buffer on the list. 1962 */ 1963 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1964 const struct sk_buff *skb) 1965 { 1966 return skb->prev == (const struct sk_buff *) list; 1967 } 1968 1969 /** 1970 * skb_queue_next - return the next packet in the queue 1971 * @list: queue head 1972 * @skb: current buffer 1973 * 1974 * Return the next packet in @list after @skb. It is only valid to 1975 * call this if skb_queue_is_last() evaluates to false. 1976 */ 1977 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1978 const struct sk_buff *skb) 1979 { 1980 /* This BUG_ON may seem severe, but if we just return then we 1981 * are going to dereference garbage. 1982 */ 1983 BUG_ON(skb_queue_is_last(list, skb)); 1984 return skb->next; 1985 } 1986 1987 /** 1988 * skb_queue_prev - return the prev packet in the queue 1989 * @list: queue head 1990 * @skb: current buffer 1991 * 1992 * Return the prev packet in @list before @skb. It is only valid to 1993 * call this if skb_queue_is_first() evaluates to false. 1994 */ 1995 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1996 const struct sk_buff *skb) 1997 { 1998 /* This BUG_ON may seem severe, but if we just return then we 1999 * are going to dereference garbage. 2000 */ 2001 BUG_ON(skb_queue_is_first(list, skb)); 2002 return skb->prev; 2003 } 2004 2005 /** 2006 * skb_get - reference buffer 2007 * @skb: buffer to reference 2008 * 2009 * Makes another reference to a socket buffer and returns a pointer 2010 * to the buffer. 2011 */ 2012 static inline struct sk_buff *skb_get(struct sk_buff *skb) 2013 { 2014 refcount_inc(&skb->users); 2015 return skb; 2016 } 2017 2018 /* 2019 * If users == 1, we are the only owner and can avoid redundant atomic changes. 2020 */ 2021 2022 /** 2023 * skb_cloned - is the buffer a clone 2024 * @skb: buffer to check 2025 * 2026 * Returns true if the buffer was generated with skb_clone() and is 2027 * one of multiple shared copies of the buffer. Cloned buffers are 2028 * shared data so must not be written to under normal circumstances. 2029 */ 2030 static inline int skb_cloned(const struct sk_buff *skb) 2031 { 2032 return skb->cloned && 2033 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 2034 } 2035 2036 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 2037 { 2038 might_sleep_if(gfpflags_allow_blocking(pri)); 2039 2040 if (skb_cloned(skb)) 2041 return pskb_expand_head(skb, 0, 0, pri); 2042 2043 return 0; 2044 } 2045 2046 /* This variant of skb_unclone() makes sure skb->truesize 2047 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 2048 * 2049 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 2050 * when various debugging features are in place. 2051 */ 2052 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 2053 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 2054 { 2055 might_sleep_if(gfpflags_allow_blocking(pri)); 2056 2057 if (skb_cloned(skb)) 2058 return __skb_unclone_keeptruesize(skb, pri); 2059 return 0; 2060 } 2061 2062 /** 2063 * skb_header_cloned - is the header a clone 2064 * @skb: buffer to check 2065 * 2066 * Returns true if modifying the header part of the buffer requires 2067 * the data to be copied. 2068 */ 2069 static inline int skb_header_cloned(const struct sk_buff *skb) 2070 { 2071 int dataref; 2072 2073 if (!skb->cloned) 2074 return 0; 2075 2076 dataref = atomic_read(&skb_shinfo(skb)->dataref); 2077 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 2078 return dataref != 1; 2079 } 2080 2081 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 2082 { 2083 might_sleep_if(gfpflags_allow_blocking(pri)); 2084 2085 if (skb_header_cloned(skb)) 2086 return pskb_expand_head(skb, 0, 0, pri); 2087 2088 return 0; 2089 } 2090 2091 /** 2092 * __skb_header_release() - allow clones to use the headroom 2093 * @skb: buffer to operate on 2094 * 2095 * See "DOC: dataref and headerless skbs". 2096 */ 2097 static inline void __skb_header_release(struct sk_buff *skb) 2098 { 2099 skb->nohdr = 1; 2100 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 2101 } 2102 2103 2104 /** 2105 * skb_shared - is the buffer shared 2106 * @skb: buffer to check 2107 * 2108 * Returns true if more than one person has a reference to this 2109 * buffer. 2110 */ 2111 static inline int skb_shared(const struct sk_buff *skb) 2112 { 2113 return refcount_read(&skb->users) != 1; 2114 } 2115 2116 /** 2117 * skb_share_check - check if buffer is shared and if so clone it 2118 * @skb: buffer to check 2119 * @pri: priority for memory allocation 2120 * 2121 * If the buffer is shared the buffer is cloned and the old copy 2122 * drops a reference. A new clone with a single reference is returned. 2123 * If the buffer is not shared the original buffer is returned. When 2124 * being called from interrupt status or with spinlocks held pri must 2125 * be GFP_ATOMIC. 2126 * 2127 * NULL is returned on a memory allocation failure. 2128 */ 2129 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 2130 { 2131 might_sleep_if(gfpflags_allow_blocking(pri)); 2132 if (skb_shared(skb)) { 2133 struct sk_buff *nskb = skb_clone(skb, pri); 2134 2135 if (likely(nskb)) 2136 consume_skb(skb); 2137 else 2138 kfree_skb(skb); 2139 skb = nskb; 2140 } 2141 return skb; 2142 } 2143 2144 /* 2145 * Copy shared buffers into a new sk_buff. We effectively do COW on 2146 * packets to handle cases where we have a local reader and forward 2147 * and a couple of other messy ones. The normal one is tcpdumping 2148 * a packet that's being forwarded. 2149 */ 2150 2151 /** 2152 * skb_unshare - make a copy of a shared buffer 2153 * @skb: buffer to check 2154 * @pri: priority for memory allocation 2155 * 2156 * If the socket buffer is a clone then this function creates a new 2157 * copy of the data, drops a reference count on the old copy and returns 2158 * the new copy with the reference count at 1. If the buffer is not a clone 2159 * the original buffer is returned. When called with a spinlock held or 2160 * from interrupt state @pri must be %GFP_ATOMIC 2161 * 2162 * %NULL is returned on a memory allocation failure. 2163 */ 2164 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2165 gfp_t pri) 2166 { 2167 might_sleep_if(gfpflags_allow_blocking(pri)); 2168 if (skb_cloned(skb)) { 2169 struct sk_buff *nskb = skb_copy(skb, pri); 2170 2171 /* Free our shared copy */ 2172 if (likely(nskb)) 2173 consume_skb(skb); 2174 else 2175 kfree_skb(skb); 2176 skb = nskb; 2177 } 2178 return skb; 2179 } 2180 2181 /** 2182 * skb_peek - peek at the head of an &sk_buff_head 2183 * @list_: list to peek at 2184 * 2185 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2186 * be careful with this one. A peek leaves the buffer on the 2187 * list and someone else may run off with it. You must hold 2188 * the appropriate locks or have a private queue to do this. 2189 * 2190 * Returns %NULL for an empty list or a pointer to the head element. 2191 * The reference count is not incremented and the reference is therefore 2192 * volatile. Use with caution. 2193 */ 2194 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2195 { 2196 struct sk_buff *skb = list_->next; 2197 2198 if (skb == (struct sk_buff *)list_) 2199 skb = NULL; 2200 return skb; 2201 } 2202 2203 /** 2204 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2205 * @list_: list to peek at 2206 * 2207 * Like skb_peek(), but the caller knows that the list is not empty. 2208 */ 2209 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2210 { 2211 return list_->next; 2212 } 2213 2214 /** 2215 * skb_peek_next - peek skb following the given one from a queue 2216 * @skb: skb to start from 2217 * @list_: list to peek at 2218 * 2219 * Returns %NULL when the end of the list is met or a pointer to the 2220 * next element. The reference count is not incremented and the 2221 * reference is therefore volatile. Use with caution. 2222 */ 2223 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2224 const struct sk_buff_head *list_) 2225 { 2226 struct sk_buff *next = skb->next; 2227 2228 if (next == (struct sk_buff *)list_) 2229 next = NULL; 2230 return next; 2231 } 2232 2233 /** 2234 * skb_peek_tail - peek at the tail of an &sk_buff_head 2235 * @list_: list to peek at 2236 * 2237 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2238 * be careful with this one. A peek leaves the buffer on the 2239 * list and someone else may run off with it. You must hold 2240 * the appropriate locks or have a private queue to do this. 2241 * 2242 * Returns %NULL for an empty list or a pointer to the tail element. 2243 * The reference count is not incremented and the reference is therefore 2244 * volatile. Use with caution. 2245 */ 2246 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2247 { 2248 struct sk_buff *skb = READ_ONCE(list_->prev); 2249 2250 if (skb == (struct sk_buff *)list_) 2251 skb = NULL; 2252 return skb; 2253 2254 } 2255 2256 /** 2257 * skb_queue_len - get queue length 2258 * @list_: list to measure 2259 * 2260 * Return the length of an &sk_buff queue. 2261 */ 2262 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2263 { 2264 return list_->qlen; 2265 } 2266 2267 /** 2268 * skb_queue_len_lockless - get queue length 2269 * @list_: list to measure 2270 * 2271 * Return the length of an &sk_buff queue. 2272 * This variant can be used in lockless contexts. 2273 */ 2274 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2275 { 2276 return READ_ONCE(list_->qlen); 2277 } 2278 2279 /** 2280 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2281 * @list: queue to initialize 2282 * 2283 * This initializes only the list and queue length aspects of 2284 * an sk_buff_head object. This allows to initialize the list 2285 * aspects of an sk_buff_head without reinitializing things like 2286 * the spinlock. It can also be used for on-stack sk_buff_head 2287 * objects where the spinlock is known to not be used. 2288 */ 2289 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2290 { 2291 list->prev = list->next = (struct sk_buff *)list; 2292 list->qlen = 0; 2293 } 2294 2295 /* 2296 * This function creates a split out lock class for each invocation; 2297 * this is needed for now since a whole lot of users of the skb-queue 2298 * infrastructure in drivers have different locking usage (in hardirq) 2299 * than the networking core (in softirq only). In the long run either the 2300 * network layer or drivers should need annotation to consolidate the 2301 * main types of usage into 3 classes. 2302 */ 2303 static inline void skb_queue_head_init(struct sk_buff_head *list) 2304 { 2305 spin_lock_init(&list->lock); 2306 __skb_queue_head_init(list); 2307 } 2308 2309 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2310 struct lock_class_key *class) 2311 { 2312 skb_queue_head_init(list); 2313 lockdep_set_class(&list->lock, class); 2314 } 2315 2316 /* 2317 * Insert an sk_buff on a list. 2318 * 2319 * The "__skb_xxxx()" functions are the non-atomic ones that 2320 * can only be called with interrupts disabled. 2321 */ 2322 static inline void __skb_insert(struct sk_buff *newsk, 2323 struct sk_buff *prev, struct sk_buff *next, 2324 struct sk_buff_head *list) 2325 { 2326 /* See skb_queue_empty_lockless() and skb_peek_tail() 2327 * for the opposite READ_ONCE() 2328 */ 2329 WRITE_ONCE(newsk->next, next); 2330 WRITE_ONCE(newsk->prev, prev); 2331 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2332 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2333 WRITE_ONCE(list->qlen, list->qlen + 1); 2334 } 2335 2336 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2337 struct sk_buff *prev, 2338 struct sk_buff *next) 2339 { 2340 struct sk_buff *first = list->next; 2341 struct sk_buff *last = list->prev; 2342 2343 WRITE_ONCE(first->prev, prev); 2344 WRITE_ONCE(prev->next, first); 2345 2346 WRITE_ONCE(last->next, next); 2347 WRITE_ONCE(next->prev, last); 2348 } 2349 2350 /** 2351 * skb_queue_splice - join two skb lists, this is designed for stacks 2352 * @list: the new list to add 2353 * @head: the place to add it in the first list 2354 */ 2355 static inline void skb_queue_splice(const struct sk_buff_head *list, 2356 struct sk_buff_head *head) 2357 { 2358 if (!skb_queue_empty(list)) { 2359 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2360 head->qlen += list->qlen; 2361 } 2362 } 2363 2364 /** 2365 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2366 * @list: the new list to add 2367 * @head: the place to add it in the first list 2368 * 2369 * The list at @list is reinitialised 2370 */ 2371 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2372 struct sk_buff_head *head) 2373 { 2374 if (!skb_queue_empty(list)) { 2375 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2376 head->qlen += list->qlen; 2377 __skb_queue_head_init(list); 2378 } 2379 } 2380 2381 /** 2382 * skb_queue_splice_tail - join two skb lists, each list being a queue 2383 * @list: the new list to add 2384 * @head: the place to add it in the first list 2385 */ 2386 static inline void skb_queue_splice_tail(const 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 } 2393 } 2394 2395 /** 2396 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2397 * @list: the new list to add 2398 * @head: the place to add it in the first list 2399 * 2400 * Each of the lists is a queue. 2401 * The list at @list is reinitialised 2402 */ 2403 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2404 struct sk_buff_head *head) 2405 { 2406 if (!skb_queue_empty(list)) { 2407 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2408 head->qlen += list->qlen; 2409 __skb_queue_head_init(list); 2410 } 2411 } 2412 2413 /** 2414 * __skb_queue_after - queue a buffer at the list head 2415 * @list: list to use 2416 * @prev: place after this buffer 2417 * @newsk: buffer to queue 2418 * 2419 * Queue a buffer int the middle of a list. This function takes no locks 2420 * and you must therefore hold required locks before calling it. 2421 * 2422 * A buffer cannot be placed on two lists at the same time. 2423 */ 2424 static inline void __skb_queue_after(struct sk_buff_head *list, 2425 struct sk_buff *prev, 2426 struct sk_buff *newsk) 2427 { 2428 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2429 } 2430 2431 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2432 struct sk_buff_head *list); 2433 2434 static inline void __skb_queue_before(struct sk_buff_head *list, 2435 struct sk_buff *next, 2436 struct sk_buff *newsk) 2437 { 2438 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2439 } 2440 2441 /** 2442 * __skb_queue_head - queue a buffer at the list head 2443 * @list: list to use 2444 * @newsk: buffer to queue 2445 * 2446 * Queue a buffer at the start 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_head(struct sk_buff_head *list, 2452 struct sk_buff *newsk) 2453 { 2454 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2455 } 2456 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2457 2458 /** 2459 * __skb_queue_tail - queue a buffer at the list tail 2460 * @list: list to use 2461 * @newsk: buffer to queue 2462 * 2463 * Queue a buffer at the end of a list. This function takes no locks 2464 * and you must therefore hold required locks before calling it. 2465 * 2466 * A buffer cannot be placed on two lists at the same time. 2467 */ 2468 static inline void __skb_queue_tail(struct sk_buff_head *list, 2469 struct sk_buff *newsk) 2470 { 2471 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2472 } 2473 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2474 2475 /* 2476 * remove sk_buff from list. _Must_ be called atomically, and with 2477 * the list known.. 2478 */ 2479 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2480 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2481 { 2482 struct sk_buff *next, *prev; 2483 2484 WRITE_ONCE(list->qlen, list->qlen - 1); 2485 next = skb->next; 2486 prev = skb->prev; 2487 skb->next = skb->prev = NULL; 2488 WRITE_ONCE(next->prev, prev); 2489 WRITE_ONCE(prev->next, next); 2490 } 2491 2492 /** 2493 * __skb_dequeue - remove from the head of the queue 2494 * @list: list to dequeue from 2495 * 2496 * Remove the head of the list. This function does not take any locks 2497 * so must be used with appropriate locks held only. The head item is 2498 * returned or %NULL if the list is empty. 2499 */ 2500 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2501 { 2502 struct sk_buff *skb = skb_peek(list); 2503 if (skb) 2504 __skb_unlink(skb, list); 2505 return skb; 2506 } 2507 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2508 2509 /** 2510 * __skb_dequeue_tail - remove from the tail of the queue 2511 * @list: list to dequeue from 2512 * 2513 * Remove the tail of the list. This function does not take any locks 2514 * so must be used with appropriate locks held only. The tail item is 2515 * returned or %NULL if the list is empty. 2516 */ 2517 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2518 { 2519 struct sk_buff *skb = skb_peek_tail(list); 2520 if (skb) 2521 __skb_unlink(skb, list); 2522 return skb; 2523 } 2524 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2525 2526 2527 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2528 { 2529 return skb->data_len; 2530 } 2531 2532 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2533 { 2534 return skb->len - skb->data_len; 2535 } 2536 2537 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2538 { 2539 unsigned int i, len = 0; 2540 2541 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2542 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2543 return len; 2544 } 2545 2546 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2547 { 2548 return skb_headlen(skb) + __skb_pagelen(skb); 2549 } 2550 2551 static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag, 2552 netmem_ref netmem, int off, 2553 int size) 2554 { 2555 frag->netmem = netmem; 2556 frag->offset = off; 2557 skb_frag_size_set(frag, size); 2558 } 2559 2560 static inline void skb_frag_fill_page_desc(skb_frag_t *frag, 2561 struct page *page, 2562 int off, int size) 2563 { 2564 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size); 2565 } 2566 2567 static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo, 2568 int i, netmem_ref netmem, 2569 int off, int size) 2570 { 2571 skb_frag_t *frag = &shinfo->frags[i]; 2572 2573 skb_frag_fill_netmem_desc(frag, netmem, off, size); 2574 } 2575 2576 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2577 int i, struct page *page, 2578 int off, int size) 2579 { 2580 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off, 2581 size); 2582 } 2583 2584 /** 2585 * skb_len_add - adds a number to len fields of skb 2586 * @skb: buffer to add len to 2587 * @delta: number of bytes to add 2588 */ 2589 static inline void skb_len_add(struct sk_buff *skb, int delta) 2590 { 2591 skb->len += delta; 2592 skb->data_len += delta; 2593 skb->truesize += delta; 2594 } 2595 2596 /** 2597 * __skb_fill_netmem_desc - initialise a fragment in an skb 2598 * @skb: buffer containing fragment to be initialised 2599 * @i: fragment index to initialise 2600 * @netmem: the netmem to use for this fragment 2601 * @off: the offset to the data with @page 2602 * @size: the length of the data 2603 * 2604 * Initialises the @i'th fragment of @skb to point to &size bytes at 2605 * offset @off within @page. 2606 * 2607 * Does not take any additional reference on the fragment. 2608 */ 2609 static __always_inline void 2610 __skb_fill_netmem_desc(struct sk_buff *skb, int i, netmem_ref netmem, 2611 int off, int size) 2612 { 2613 struct page *page; 2614 2615 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size); 2616 2617 if (netmem_is_net_iov(netmem)) { 2618 skb->unreadable = true; 2619 return; 2620 } 2621 2622 page = netmem_to_page(netmem); 2623 2624 /* Propagate page pfmemalloc to the skb if we can. The problem is 2625 * that not all callers have unique ownership of the page but rely 2626 * on page_is_pfmemalloc doing the right thing(tm). 2627 */ 2628 page = compound_head(page); 2629 if (page_is_pfmemalloc(page)) 2630 skb->pfmemalloc = true; 2631 } 2632 2633 static __always_inline void 2634 __skb_fill_page_desc(struct sk_buff *skb, int i, struct page *page, 2635 int off, int size) 2636 { 2637 __skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2638 } 2639 2640 static __always_inline void 2641 skb_fill_netmem_desc(struct sk_buff *skb, int i, netmem_ref netmem, 2642 int off, int size) 2643 { 2644 __skb_fill_netmem_desc(skb, i, netmem, off, size); 2645 skb_shinfo(skb)->nr_frags = i + 1; 2646 } 2647 2648 /** 2649 * skb_fill_page_desc - initialise a paged fragment in an skb 2650 * @skb: buffer containing fragment to be initialised 2651 * @i: paged fragment index to initialise 2652 * @page: the page to use for this fragment 2653 * @off: the offset to the data with @page 2654 * @size: the length of the data 2655 * 2656 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2657 * @skb to point to @size bytes at offset @off within @page. In 2658 * addition updates @skb such that @i is the last fragment. 2659 * 2660 * Does not take any additional reference on the fragment. 2661 */ 2662 static __always_inline void 2663 skb_fill_page_desc(struct sk_buff *skb, int i, struct page *page, 2664 int off, int size) 2665 { 2666 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2667 } 2668 2669 /** 2670 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2671 * @skb: buffer containing fragment to be initialised 2672 * @i: paged fragment index to initialise 2673 * @page: the page to use for this fragment 2674 * @off: the offset to the data with @page 2675 * @size: the length of the data 2676 * 2677 * Variant of skb_fill_page_desc() which does not deal with 2678 * pfmemalloc, if page is not owned by us. 2679 */ 2680 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2681 struct page *page, int off, 2682 int size) 2683 { 2684 struct skb_shared_info *shinfo = skb_shinfo(skb); 2685 2686 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2687 shinfo->nr_frags = i + 1; 2688 } 2689 2690 static inline void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, 2691 netmem_ref netmem, int off, 2692 int size, unsigned int truesize) 2693 { 2694 DEBUG_NET_WARN_ON_ONCE(size > truesize); 2695 2696 skb_fill_netmem_desc(skb, i, netmem, off, size); 2697 skb->len += size; 2698 skb->data_len += size; 2699 skb->truesize += truesize; 2700 } 2701 2702 static inline void skb_add_rx_frag(struct sk_buff *skb, int i, 2703 struct page *page, int off, int size, 2704 unsigned int truesize) 2705 { 2706 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size, 2707 truesize); 2708 } 2709 2710 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2711 unsigned int truesize); 2712 2713 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2714 2715 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2716 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2717 { 2718 return skb->head + skb->tail; 2719 } 2720 2721 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2722 { 2723 skb->tail = skb->data - skb->head; 2724 } 2725 2726 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2727 { 2728 skb_reset_tail_pointer(skb); 2729 skb->tail += offset; 2730 } 2731 2732 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2733 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2734 { 2735 return skb->tail; 2736 } 2737 2738 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2739 { 2740 skb->tail = skb->data; 2741 } 2742 2743 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2744 { 2745 skb->tail = skb->data + offset; 2746 } 2747 2748 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2749 2750 static inline void skb_assert_len(struct sk_buff *skb) 2751 { 2752 #ifdef CONFIG_DEBUG_NET 2753 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2754 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2755 #endif /* CONFIG_DEBUG_NET */ 2756 } 2757 2758 #if defined(CONFIG_FAIL_SKB_REALLOC) 2759 void skb_might_realloc(struct sk_buff *skb); 2760 #else 2761 static inline void skb_might_realloc(struct sk_buff *skb) {} 2762 #endif 2763 2764 /* 2765 * Add data to an sk_buff 2766 */ 2767 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2768 void *skb_put(struct sk_buff *skb, unsigned int len); 2769 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2770 { 2771 void *tmp = skb_tail_pointer(skb); 2772 SKB_LINEAR_ASSERT(skb); 2773 skb->tail += len; 2774 skb->len += len; 2775 return tmp; 2776 } 2777 2778 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2779 { 2780 void *tmp = __skb_put(skb, len); 2781 2782 memset(tmp, 0, len); 2783 return tmp; 2784 } 2785 2786 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2787 unsigned int len) 2788 { 2789 void *tmp = __skb_put(skb, len); 2790 2791 memcpy(tmp, data, len); 2792 return tmp; 2793 } 2794 2795 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2796 { 2797 *(u8 *)__skb_put(skb, 1) = val; 2798 } 2799 2800 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2801 { 2802 void *tmp = skb_put(skb, len); 2803 2804 memset(tmp, 0, len); 2805 2806 return tmp; 2807 } 2808 2809 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2810 unsigned int len) 2811 { 2812 void *tmp = skb_put(skb, len); 2813 2814 memcpy(tmp, data, len); 2815 2816 return tmp; 2817 } 2818 2819 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2820 { 2821 *(u8 *)skb_put(skb, 1) = val; 2822 } 2823 2824 void *skb_push(struct sk_buff *skb, unsigned int len); 2825 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2826 { 2827 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2828 2829 skb->data -= len; 2830 DEBUG_NET_WARN_ON_ONCE(skb->data < skb->head); 2831 skb->len += len; 2832 return skb->data; 2833 } 2834 2835 void *skb_pull(struct sk_buff *skb, unsigned int len); 2836 static __always_inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2837 { 2838 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2839 2840 skb->len -= len; 2841 if (unlikely(skb->len < skb->data_len)) { 2842 #if defined(CONFIG_DEBUG_NET) 2843 skb->len += len; 2844 pr_err("__skb_pull(len=%u)\n", len); 2845 skb_dump(KERN_ERR, skb, false); 2846 #endif 2847 BUG(); 2848 } 2849 return skb->data += len; 2850 } 2851 2852 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2853 { 2854 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2855 } 2856 2857 void *skb_pull_data(struct sk_buff *skb, size_t len); 2858 2859 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2860 2861 static __always_inline enum skb_drop_reason 2862 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len) 2863 { 2864 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2865 skb_might_realloc(skb); 2866 2867 if (likely(len <= skb_headlen(skb))) 2868 return SKB_NOT_DROPPED_YET; 2869 2870 if (unlikely(len > skb->len)) 2871 return SKB_DROP_REASON_PKT_TOO_SMALL; 2872 2873 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb)))) 2874 return SKB_DROP_REASON_NOMEM; 2875 2876 return SKB_NOT_DROPPED_YET; 2877 } 2878 2879 static __always_inline bool 2880 pskb_may_pull(struct sk_buff *skb, unsigned int len) 2881 { 2882 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 2883 } 2884 2885 static __always_inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2886 { 2887 if (!pskb_may_pull(skb, len)) 2888 return NULL; 2889 2890 skb->len -= len; 2891 return skb->data += len; 2892 } 2893 2894 void skb_condense(struct sk_buff *skb); 2895 2896 /** 2897 * skb_headroom - bytes at buffer head 2898 * @skb: buffer to check 2899 * 2900 * Return the number of bytes of free space at the head of an &sk_buff. 2901 */ 2902 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2903 { 2904 return skb->data - skb->head; 2905 } 2906 2907 /** 2908 * skb_tailroom - bytes at buffer end 2909 * @skb: buffer to check 2910 * 2911 * Return the number of bytes of free space at the tail of an sk_buff 2912 */ 2913 static inline int skb_tailroom(const struct sk_buff *skb) 2914 { 2915 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2916 } 2917 2918 /** 2919 * skb_availroom - bytes at buffer end 2920 * @skb: buffer to check 2921 * 2922 * Return the number of bytes of free space at the tail of an sk_buff 2923 * allocated by sk_stream_alloc() 2924 */ 2925 static inline int skb_availroom(const struct sk_buff *skb) 2926 { 2927 if (skb_is_nonlinear(skb)) 2928 return 0; 2929 2930 return skb->end - skb->tail - skb->reserved_tailroom; 2931 } 2932 2933 /** 2934 * skb_reserve - adjust headroom 2935 * @skb: buffer to alter 2936 * @len: bytes to move 2937 * 2938 * Increase the headroom of an empty &sk_buff by reducing the tail 2939 * room. This is only allowed for an empty buffer. 2940 */ 2941 static inline void skb_reserve(struct sk_buff *skb, int len) 2942 { 2943 skb->data += len; 2944 skb->tail += len; 2945 } 2946 2947 /** 2948 * skb_tailroom_reserve - adjust reserved_tailroom 2949 * @skb: buffer to alter 2950 * @mtu: maximum amount of headlen permitted 2951 * @needed_tailroom: minimum amount of reserved_tailroom 2952 * 2953 * Set reserved_tailroom so that headlen can be as large as possible but 2954 * not larger than mtu and tailroom cannot be smaller than 2955 * needed_tailroom. 2956 * The required headroom should already have been reserved before using 2957 * this function. 2958 */ 2959 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2960 unsigned int needed_tailroom) 2961 { 2962 SKB_LINEAR_ASSERT(skb); 2963 if (mtu < skb_tailroom(skb) - needed_tailroom) 2964 /* use at most mtu */ 2965 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2966 else 2967 /* use up to all available space */ 2968 skb->reserved_tailroom = needed_tailroom; 2969 } 2970 2971 #define ENCAP_TYPE_ETHER 0 2972 #define ENCAP_TYPE_IPPROTO 1 2973 2974 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2975 __be16 protocol) 2976 { 2977 skb->inner_protocol = protocol; 2978 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2979 } 2980 2981 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2982 __u8 ipproto) 2983 { 2984 skb->inner_ipproto = ipproto; 2985 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2986 } 2987 2988 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2989 { 2990 skb->inner_mac_header = skb->mac_header; 2991 skb->inner_network_header = skb->network_header; 2992 skb->inner_transport_header = skb->transport_header; 2993 } 2994 2995 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2996 { 2997 return skb->mac_header != (typeof(skb->mac_header))~0U; 2998 } 2999 3000 static inline void skb_reset_mac_len(struct sk_buff *skb) 3001 { 3002 if (!skb_mac_header_was_set(skb)) { 3003 DEBUG_NET_WARN_ON_ONCE(1); 3004 skb->mac_len = 0; 3005 } else { 3006 skb->mac_len = skb->network_header - skb->mac_header; 3007 } 3008 } 3009 3010 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 3011 *skb) 3012 { 3013 return skb->head + skb->inner_transport_header; 3014 } 3015 3016 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 3017 { 3018 return skb_inner_transport_header(skb) - skb->data; 3019 } 3020 3021 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 3022 { 3023 long offset = skb->data - skb->head; 3024 3025 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_transport_header))offset); 3026 skb->inner_transport_header = offset; 3027 } 3028 3029 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 3030 const int offset) 3031 { 3032 skb_reset_inner_transport_header(skb); 3033 skb->inner_transport_header += offset; 3034 } 3035 3036 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 3037 { 3038 return skb->head + skb->inner_network_header; 3039 } 3040 3041 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 3042 { 3043 long offset = skb->data - skb->head; 3044 3045 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_network_header))offset); 3046 skb->inner_network_header = offset; 3047 } 3048 3049 static inline void skb_set_inner_network_header(struct sk_buff *skb, 3050 const int offset) 3051 { 3052 skb_reset_inner_network_header(skb); 3053 skb->inner_network_header += offset; 3054 } 3055 3056 static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb) 3057 { 3058 return skb->inner_network_header > 0; 3059 } 3060 3061 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 3062 { 3063 return skb->head + skb->inner_mac_header; 3064 } 3065 3066 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 3067 { 3068 long offset = skb->data - skb->head; 3069 3070 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_mac_header))offset); 3071 skb->inner_mac_header = offset; 3072 } 3073 3074 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 3075 const int offset) 3076 { 3077 skb_reset_inner_mac_header(skb); 3078 skb->inner_mac_header += offset; 3079 } 3080 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 3081 { 3082 return skb->transport_header != (typeof(skb->transport_header))~0U; 3083 } 3084 3085 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 3086 { 3087 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3088 return skb->head + skb->transport_header; 3089 } 3090 3091 static inline void skb_reset_transport_header(struct sk_buff *skb) 3092 { 3093 long offset = skb->data - skb->head; 3094 3095 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->transport_header))offset); 3096 skb->transport_header = offset; 3097 } 3098 3099 /** 3100 * skb_reset_transport_header_careful - conditionally reset transport header 3101 * @skb: buffer to alter 3102 * 3103 * Hardened version of skb_reset_transport_header(). 3104 * 3105 * Returns: true if the operation was a success. 3106 */ 3107 static inline bool __must_check 3108 skb_reset_transport_header_careful(struct sk_buff *skb) 3109 { 3110 long offset = skb->data - skb->head; 3111 3112 if (unlikely(offset != (typeof(skb->transport_header))offset)) 3113 return false; 3114 3115 if (unlikely(offset == (typeof(skb->transport_header))~0U)) 3116 return false; 3117 3118 skb->transport_header = offset; 3119 return true; 3120 } 3121 3122 static inline void skb_set_transport_header(struct sk_buff *skb, 3123 const int offset) 3124 { 3125 skb_reset_transport_header(skb); 3126 skb->transport_header += offset; 3127 } 3128 3129 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 3130 { 3131 return skb->head + skb->network_header; 3132 } 3133 3134 static inline void skb_reset_network_header(struct sk_buff *skb) 3135 { 3136 long offset = skb->data - skb->head; 3137 3138 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->network_header))offset); 3139 skb->network_header = offset; 3140 } 3141 3142 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 3143 { 3144 skb_reset_network_header(skb); 3145 skb->network_header += offset; 3146 } 3147 3148 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 3149 { 3150 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3151 return skb->head + skb->mac_header; 3152 } 3153 3154 static inline int skb_mac_offset(const struct sk_buff *skb) 3155 { 3156 return skb_mac_header(skb) - skb->data; 3157 } 3158 3159 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 3160 { 3161 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3162 return skb->network_header - skb->mac_header; 3163 } 3164 3165 static inline void skb_unset_mac_header(struct sk_buff *skb) 3166 { 3167 skb->mac_header = (typeof(skb->mac_header))~0U; 3168 } 3169 3170 static inline void skb_reset_mac_header(struct sk_buff *skb) 3171 { 3172 long offset = skb->data - skb->head; 3173 3174 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->mac_header))offset); 3175 skb->mac_header = offset; 3176 } 3177 3178 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 3179 { 3180 skb_reset_mac_header(skb); 3181 skb->mac_header += offset; 3182 } 3183 3184 static inline void skb_pop_mac_header(struct sk_buff *skb) 3185 { 3186 skb->mac_header = skb->network_header; 3187 } 3188 3189 static inline void skb_probe_transport_header(struct sk_buff *skb) 3190 { 3191 struct flow_keys_basic keys; 3192 3193 if (skb_transport_header_was_set(skb)) 3194 return; 3195 3196 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 3197 NULL, 0, 0, 0, 0)) 3198 skb_set_transport_header(skb, keys.control.thoff); 3199 } 3200 3201 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 3202 { 3203 if (skb_mac_header_was_set(skb)) { 3204 const unsigned char *old_mac = skb_mac_header(skb); 3205 3206 skb_set_mac_header(skb, -skb->mac_len); 3207 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 3208 } 3209 } 3210 3211 /* Move the full mac header up to current network_header. 3212 * Leaves skb->data pointing at offset skb->mac_len into the mac_header. 3213 * Must be provided the complete mac header length. 3214 */ 3215 static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len) 3216 { 3217 if (skb_mac_header_was_set(skb)) { 3218 const unsigned char *old_mac = skb_mac_header(skb); 3219 3220 skb_set_mac_header(skb, -full_mac_len); 3221 memmove(skb_mac_header(skb), old_mac, full_mac_len); 3222 __skb_push(skb, full_mac_len - skb->mac_len); 3223 } 3224 } 3225 3226 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 3227 { 3228 return skb->csum_start - skb_headroom(skb); 3229 } 3230 3231 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 3232 { 3233 return skb->head + skb->csum_start; 3234 } 3235 3236 static inline int skb_transport_offset(const struct sk_buff *skb) 3237 { 3238 return skb_transport_header(skb) - skb->data; 3239 } 3240 3241 static inline u32 skb_network_header_len(const struct sk_buff *skb) 3242 { 3243 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3244 return skb->transport_header - skb->network_header; 3245 } 3246 3247 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 3248 { 3249 return skb->inner_transport_header - skb->inner_network_header; 3250 } 3251 3252 static inline int skb_network_offset(const struct sk_buff *skb) 3253 { 3254 return skb_network_header(skb) - skb->data; 3255 } 3256 3257 static inline int skb_inner_network_offset(const struct sk_buff *skb) 3258 { 3259 return skb_inner_network_header(skb) - skb->data; 3260 } 3261 3262 static inline enum skb_drop_reason 3263 pskb_network_may_pull_reason(struct sk_buff *skb, unsigned int len) 3264 { 3265 return pskb_may_pull_reason(skb, skb_network_offset(skb) + len); 3266 } 3267 3268 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 3269 { 3270 return pskb_network_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 3271 } 3272 3273 /* 3274 * CPUs often take a performance hit when accessing unaligned memory 3275 * locations. The actual performance hit varies, it can be small if the 3276 * hardware handles it or large if we have to take an exception and fix it 3277 * in software. 3278 * 3279 * Since an ethernet header is 14 bytes network drivers often end up with 3280 * the IP header at an unaligned offset. The IP header can be aligned by 3281 * shifting the start of the packet by 2 bytes. Drivers should do this 3282 * with: 3283 * 3284 * skb_reserve(skb, NET_IP_ALIGN); 3285 * 3286 * The downside to this alignment of the IP header is that the DMA is now 3287 * unaligned. On some architectures the cost of an unaligned DMA is high 3288 * and this cost outweighs the gains made by aligning the IP header. 3289 * 3290 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 3291 * to be overridden. 3292 */ 3293 #ifndef NET_IP_ALIGN 3294 #define NET_IP_ALIGN 2 3295 #endif 3296 3297 /* 3298 * The networking layer reserves some headroom in skb data (via 3299 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 3300 * the header has to grow. In the default case, if the header has to grow 3301 * 32 bytes or less we avoid the reallocation. 3302 * 3303 * Unfortunately this headroom changes the DMA alignment of the resulting 3304 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 3305 * on some architectures. An architecture can override this value, 3306 * perhaps setting it to a cacheline in size (since that will maintain 3307 * cacheline alignment of the DMA). It must be a power of 2. 3308 * 3309 * Various parts of the networking layer expect at least 32 bytes of 3310 * headroom, you should not reduce this. 3311 * 3312 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 3313 * to reduce average number of cache lines per packet. 3314 * get_rps_cpu() for example only access one 64 bytes aligned block : 3315 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 3316 */ 3317 #ifndef NET_SKB_PAD 3318 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 3319 #endif 3320 3321 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 3322 3323 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3324 { 3325 if (WARN_ON(skb_is_nonlinear(skb))) 3326 return; 3327 skb->len = len; 3328 skb_set_tail_pointer(skb, len); 3329 } 3330 3331 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3332 { 3333 __skb_set_length(skb, len); 3334 } 3335 3336 void skb_trim(struct sk_buff *skb, unsigned int len); 3337 3338 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3339 { 3340 if (skb->data_len) 3341 return ___pskb_trim(skb, len); 3342 __skb_trim(skb, len); 3343 return 0; 3344 } 3345 3346 static __always_inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3347 { 3348 skb_might_realloc(skb); 3349 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3350 } 3351 3352 /** 3353 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3354 * @skb: buffer to alter 3355 * @len: new length 3356 * 3357 * This is identical to pskb_trim except that the caller knows that 3358 * the skb is not cloned so we should never get an error due to out- 3359 * of-memory. 3360 */ 3361 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3362 { 3363 int err = pskb_trim(skb, len); 3364 BUG_ON(err); 3365 } 3366 3367 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3368 { 3369 unsigned int diff = len - skb->len; 3370 3371 if (skb_tailroom(skb) < diff) { 3372 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3373 GFP_ATOMIC); 3374 if (ret) 3375 return ret; 3376 } 3377 __skb_set_length(skb, len); 3378 return 0; 3379 } 3380 3381 /** 3382 * skb_orphan - orphan a buffer 3383 * @skb: buffer to orphan 3384 * 3385 * If a buffer currently has an owner then we call the owner's 3386 * destructor function and make the @skb unowned. The buffer continues 3387 * to exist but is no longer charged to its former owner. 3388 */ 3389 static __always_inline void skb_orphan(struct sk_buff *skb) 3390 { 3391 if (skb->destructor) { 3392 skb->destructor(skb); 3393 skb->destructor = NULL; 3394 skb->sk = NULL; 3395 } else { 3396 BUG_ON(skb->sk); 3397 } 3398 } 3399 3400 /** 3401 * skb_orphan_frags - orphan the frags contained in a buffer 3402 * @skb: buffer to orphan frags from 3403 * @gfp_mask: allocation mask for replacement pages 3404 * 3405 * For each frag in the SKB which needs a destructor (i.e. has an 3406 * owner) create a copy of that frag and release the original 3407 * page by calling the destructor. 3408 */ 3409 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3410 { 3411 if (likely(!skb_zcopy(skb))) 3412 return 0; 3413 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3414 return 0; 3415 return skb_copy_ubufs(skb, gfp_mask); 3416 } 3417 3418 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3419 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3420 { 3421 if (likely(!skb_zcopy(skb))) 3422 return 0; 3423 return skb_copy_ubufs(skb, gfp_mask); 3424 } 3425 3426 /** 3427 * __skb_queue_purge_reason - empty a list 3428 * @list: list to empty 3429 * @reason: drop reason 3430 * 3431 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3432 * the list and one reference dropped. This function does not take the 3433 * list lock and the caller must hold the relevant locks to use it. 3434 */ 3435 static inline void __skb_queue_purge_reason(struct sk_buff_head *list, 3436 enum skb_drop_reason reason) 3437 { 3438 struct sk_buff *skb; 3439 3440 while ((skb = __skb_dequeue(list)) != NULL) 3441 kfree_skb_reason(skb, reason); 3442 } 3443 3444 static inline void __skb_queue_purge(struct sk_buff_head *list) 3445 { 3446 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3447 } 3448 3449 void skb_queue_purge_reason(struct sk_buff_head *list, 3450 enum skb_drop_reason reason); 3451 3452 static inline void skb_queue_purge(struct sk_buff_head *list) 3453 { 3454 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3455 } 3456 3457 unsigned int skb_rbtree_purge(struct rb_root *root); 3458 void skb_errqueue_purge(struct sk_buff_head *list); 3459 3460 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3461 3462 /** 3463 * netdev_alloc_frag - allocate a page fragment 3464 * @fragsz: fragment size 3465 * 3466 * Allocates a frag from a page for receive buffer. 3467 * Uses GFP_ATOMIC allocations. 3468 */ 3469 static inline void *netdev_alloc_frag(unsigned int fragsz) 3470 { 3471 return __netdev_alloc_frag_align(fragsz, ~0u); 3472 } 3473 3474 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3475 unsigned int align) 3476 { 3477 WARN_ON_ONCE(!is_power_of_2(align)); 3478 return __netdev_alloc_frag_align(fragsz, -align); 3479 } 3480 3481 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3482 gfp_t gfp_mask); 3483 3484 /** 3485 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3486 * @dev: network device to receive on 3487 * @length: length to allocate 3488 * 3489 * Allocate a new &sk_buff and assign it a usage count of one. The 3490 * buffer has unspecified headroom built in. Users should allocate 3491 * the headroom they think they need without accounting for the 3492 * built in space. The built in space is used for optimisations. 3493 * 3494 * %NULL is returned if there is no free memory. Although this function 3495 * allocates memory it can be called from an interrupt. 3496 */ 3497 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3498 unsigned int length) 3499 { 3500 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3501 } 3502 3503 /* legacy helper around __netdev_alloc_skb() */ 3504 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3505 gfp_t gfp_mask) 3506 { 3507 return __netdev_alloc_skb(NULL, length, gfp_mask); 3508 } 3509 3510 /* legacy helper around netdev_alloc_skb() */ 3511 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3512 { 3513 return netdev_alloc_skb(NULL, length); 3514 } 3515 3516 3517 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3518 unsigned int length, gfp_t gfp) 3519 { 3520 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3521 3522 if (NET_IP_ALIGN && skb) 3523 skb_reserve(skb, NET_IP_ALIGN); 3524 return skb; 3525 } 3526 3527 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3528 unsigned int length) 3529 { 3530 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3531 } 3532 3533 static inline void skb_free_frag(void *addr) 3534 { 3535 page_frag_free(addr); 3536 } 3537 3538 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3539 3540 static inline void *napi_alloc_frag(unsigned int fragsz) 3541 { 3542 return __napi_alloc_frag_align(fragsz, ~0u); 3543 } 3544 3545 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3546 unsigned int align) 3547 { 3548 WARN_ON_ONCE(!is_power_of_2(align)); 3549 return __napi_alloc_frag_align(fragsz, -align); 3550 } 3551 3552 struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int length); 3553 void napi_consume_skb(struct sk_buff *skb, int budget); 3554 3555 void napi_skb_free_stolen_head(struct sk_buff *skb); 3556 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason); 3557 3558 /** 3559 * __dev_alloc_pages - allocate page for network Rx 3560 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3561 * @order: size of the allocation 3562 * 3563 * Allocate a new page. 3564 * 3565 * %NULL is returned if there is no free memory. 3566 */ 3567 static inline struct page *__dev_alloc_pages_noprof(gfp_t gfp_mask, 3568 unsigned int order) 3569 { 3570 /* This piece of code contains several assumptions. 3571 * 1. This is for device Rx, therefore a cold page is preferred. 3572 * 2. The expectation is the user wants a compound page. 3573 * 3. If requesting a order 0 page it will not be compound 3574 * due to the check to see if order has a value in prep_new_page 3575 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3576 * code in gfp_to_alloc_flags that should be enforcing this. 3577 */ 3578 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3579 3580 return alloc_pages_node_noprof(NUMA_NO_NODE, gfp_mask, order); 3581 } 3582 #define __dev_alloc_pages(...) alloc_hooks(__dev_alloc_pages_noprof(__VA_ARGS__)) 3583 3584 /* 3585 * This specialized allocator has to be a macro for its allocations to be 3586 * accounted separately (to have a separate alloc_tag). 3587 */ 3588 #define dev_alloc_pages(_order) __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, _order) 3589 3590 /** 3591 * __dev_alloc_page - allocate a page for network Rx 3592 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3593 * 3594 * Allocate a new page. 3595 * 3596 * %NULL is returned if there is no free memory. 3597 */ 3598 static inline struct page *__dev_alloc_page_noprof(gfp_t gfp_mask) 3599 { 3600 return __dev_alloc_pages_noprof(gfp_mask, 0); 3601 } 3602 #define __dev_alloc_page(...) alloc_hooks(__dev_alloc_page_noprof(__VA_ARGS__)) 3603 3604 /* 3605 * This specialized allocator has to be a macro for its allocations to be 3606 * accounted separately (to have a separate alloc_tag). 3607 */ 3608 #define dev_alloc_page() dev_alloc_pages(0) 3609 3610 /** 3611 * dev_page_is_reusable - check whether a page can be reused for network Rx 3612 * @page: the page to test 3613 * 3614 * A page shouldn't be considered for reusing/recycling if it was allocated 3615 * under memory pressure or at a distant memory node. 3616 * 3617 * Returns: false if this page should be returned to page allocator, true 3618 * otherwise. 3619 */ 3620 static inline bool dev_page_is_reusable(const struct page *page) 3621 { 3622 return likely(page_to_nid(page) == numa_mem_id() && 3623 !page_is_pfmemalloc(page)); 3624 } 3625 3626 /** 3627 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3628 * @page: The page that was allocated from skb_alloc_page 3629 * @skb: The skb that may need pfmemalloc set 3630 */ 3631 static inline void skb_propagate_pfmemalloc(const struct page *page, 3632 struct sk_buff *skb) 3633 { 3634 if (page_is_pfmemalloc(page)) 3635 skb->pfmemalloc = true; 3636 } 3637 3638 /** 3639 * skb_frag_off() - Returns the offset of a skb fragment 3640 * @frag: the paged fragment 3641 */ 3642 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3643 { 3644 return frag->offset; 3645 } 3646 3647 /** 3648 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3649 * @frag: skb fragment 3650 * @delta: value to add 3651 */ 3652 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3653 { 3654 frag->offset += delta; 3655 } 3656 3657 /** 3658 * skb_frag_off_set() - Sets the offset of a skb fragment 3659 * @frag: skb fragment 3660 * @offset: offset of fragment 3661 */ 3662 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3663 { 3664 frag->offset = offset; 3665 } 3666 3667 /** 3668 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3669 * @fragto: skb fragment where offset is set 3670 * @fragfrom: skb fragment offset is copied from 3671 */ 3672 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3673 const skb_frag_t *fragfrom) 3674 { 3675 fragto->offset = fragfrom->offset; 3676 } 3677 3678 /* Return: true if the skb_frag contains a net_iov. */ 3679 static inline bool skb_frag_is_net_iov(const skb_frag_t *frag) 3680 { 3681 return netmem_is_net_iov(frag->netmem); 3682 } 3683 3684 /** 3685 * skb_frag_net_iov - retrieve the net_iov referred to by fragment 3686 * @frag: the fragment 3687 * 3688 * Return: the &struct net_iov associated with @frag. Returns NULL if this 3689 * frag has no associated net_iov. 3690 */ 3691 static inline struct net_iov *skb_frag_net_iov(const skb_frag_t *frag) 3692 { 3693 if (!skb_frag_is_net_iov(frag)) 3694 return NULL; 3695 3696 return netmem_to_net_iov(frag->netmem); 3697 } 3698 3699 /** 3700 * skb_frag_page - retrieve the page referred to by a paged fragment 3701 * @frag: the paged fragment 3702 * 3703 * Return: the &struct page associated with @frag. Returns NULL if this frag 3704 * has no associated page. 3705 */ 3706 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3707 { 3708 if (skb_frag_is_net_iov(frag)) 3709 return NULL; 3710 3711 return netmem_to_page(frag->netmem); 3712 } 3713 3714 /** 3715 * skb_frag_netmem - retrieve the netmem referred to by a fragment 3716 * @frag: the fragment 3717 * 3718 * Return: the &netmem_ref associated with @frag. 3719 */ 3720 static inline netmem_ref skb_frag_netmem(const skb_frag_t *frag) 3721 { 3722 return frag->netmem; 3723 } 3724 3725 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb, 3726 unsigned int headroom); 3727 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb, 3728 const struct bpf_prog *prog); 3729 3730 /** 3731 * skb_frag_address - gets the address of the data contained in a paged fragment 3732 * @frag: the paged fragment buffer 3733 * 3734 * Returns: the address of the data within @frag. The page must already 3735 * be mapped. 3736 */ 3737 static inline void *skb_frag_address(const skb_frag_t *frag) 3738 { 3739 if (!skb_frag_page(frag)) 3740 return NULL; 3741 3742 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3743 } 3744 3745 /** 3746 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3747 * @frag: the paged fragment buffer 3748 * 3749 * Returns: the address of the data within @frag. Checks that the page 3750 * is mapped and returns %NULL otherwise. 3751 */ 3752 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3753 { 3754 struct page *page = skb_frag_page(frag); 3755 void *ptr; 3756 3757 if (!page) 3758 return NULL; 3759 3760 ptr = page_address(page); 3761 if (unlikely(!ptr)) 3762 return NULL; 3763 3764 return ptr + skb_frag_off(frag); 3765 } 3766 3767 /** 3768 * skb_frag_phys - gets the physical address of the data in a paged fragment 3769 * @frag: the paged fragment buffer 3770 * 3771 * Returns: the physical address of the data within @frag. 3772 */ 3773 static inline phys_addr_t skb_frag_phys(const skb_frag_t *frag) 3774 { 3775 return page_to_phys(skb_frag_page(frag)) + skb_frag_off(frag); 3776 } 3777 3778 /** 3779 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3780 * @fragto: skb fragment where page is set 3781 * @fragfrom: skb fragment page is copied from 3782 */ 3783 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3784 const skb_frag_t *fragfrom) 3785 { 3786 fragto->netmem = fragfrom->netmem; 3787 } 3788 3789 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3790 3791 /** 3792 * __skb_frag_dma_map - maps a paged fragment via the DMA API 3793 * @dev: the device to map the fragment to 3794 * @frag: the paged fragment to map 3795 * @offset: the offset within the fragment (starting at the 3796 * fragment's own offset) 3797 * @size: the number of bytes to map 3798 * @dir: the direction of the mapping (``PCI_DMA_*``) 3799 * 3800 * Maps the page associated with @frag to @device. 3801 */ 3802 static inline dma_addr_t __skb_frag_dma_map(struct device *dev, 3803 const skb_frag_t *frag, 3804 size_t offset, size_t size, 3805 enum dma_data_direction dir) 3806 { 3807 if (skb_frag_is_net_iov(frag)) { 3808 return netmem_to_net_iov(frag->netmem)->desc.dma_addr + 3809 offset + frag->offset; 3810 } 3811 return dma_map_page(dev, skb_frag_page(frag), 3812 skb_frag_off(frag) + offset, size, dir); 3813 } 3814 3815 #define skb_frag_dma_map(dev, frag, ...) \ 3816 CONCATENATE(_skb_frag_dma_map, \ 3817 COUNT_ARGS(__VA_ARGS__))(dev, frag, ##__VA_ARGS__) 3818 3819 #define __skb_frag_dma_map1(dev, frag, offset, uf, uo) ({ \ 3820 const skb_frag_t *uf = (frag); \ 3821 size_t uo = (offset); \ 3822 \ 3823 __skb_frag_dma_map(dev, uf, uo, skb_frag_size(uf) - uo, \ 3824 DMA_TO_DEVICE); \ 3825 }) 3826 #define _skb_frag_dma_map1(dev, frag, offset) \ 3827 __skb_frag_dma_map1(dev, frag, offset, __UNIQUE_ID(frag_), \ 3828 __UNIQUE_ID(offset_)) 3829 #define _skb_frag_dma_map0(dev, frag) \ 3830 _skb_frag_dma_map1(dev, frag, 0) 3831 #define _skb_frag_dma_map2(dev, frag, offset, size) \ 3832 __skb_frag_dma_map(dev, frag, offset, size, DMA_TO_DEVICE) 3833 #define _skb_frag_dma_map3(dev, frag, offset, size, dir) \ 3834 __skb_frag_dma_map(dev, frag, offset, size, dir) 3835 3836 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3837 gfp_t gfp_mask) 3838 { 3839 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3840 } 3841 3842 3843 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3844 gfp_t gfp_mask) 3845 { 3846 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3847 } 3848 3849 3850 /** 3851 * skb_clone_writable - is the header of a clone writable 3852 * @skb: buffer to check 3853 * @len: length up to which to write 3854 * 3855 * Returns true if modifying the header part of the cloned buffer 3856 * does not requires the data to be copied. 3857 */ 3858 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3859 { 3860 return !skb_header_cloned(skb) && 3861 skb_headroom(skb) + len <= skb->hdr_len; 3862 } 3863 3864 static inline int skb_try_make_writable(struct sk_buff *skb, 3865 unsigned int write_len) 3866 { 3867 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3868 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3869 } 3870 3871 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3872 int cloned) 3873 { 3874 int delta = 0; 3875 3876 if (headroom > skb_headroom(skb)) 3877 delta = headroom - skb_headroom(skb); 3878 3879 if (delta || cloned) 3880 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3881 GFP_ATOMIC); 3882 return 0; 3883 } 3884 3885 /** 3886 * skb_cow - copy header of skb when it is required 3887 * @skb: buffer to cow 3888 * @headroom: needed headroom 3889 * 3890 * If the skb passed lacks sufficient headroom or its data part 3891 * is shared, data is reallocated. If reallocation fails, an error 3892 * is returned and original skb is not changed. 3893 * 3894 * The result is skb with writable area skb->head...skb->tail 3895 * and at least @headroom of space at head. 3896 */ 3897 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3898 { 3899 return __skb_cow(skb, headroom, skb_cloned(skb)); 3900 } 3901 3902 /** 3903 * skb_cow_head - skb_cow but only making the head writable 3904 * @skb: buffer to cow 3905 * @headroom: needed headroom 3906 * 3907 * This function is identical to skb_cow except that we replace the 3908 * skb_cloned check by skb_header_cloned. It should be used when 3909 * you only need to push on some header and do not need to modify 3910 * the data. 3911 */ 3912 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3913 { 3914 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3915 } 3916 3917 /** 3918 * skb_padto - pad an skbuff up to a minimal size 3919 * @skb: buffer to pad 3920 * @len: minimal length 3921 * 3922 * Pads up a buffer to ensure the trailing bytes exist and are 3923 * blanked. If the buffer already contains sufficient data it 3924 * is untouched. Otherwise it is extended. Returns zero on 3925 * success. The skb is freed on error. 3926 */ 3927 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3928 { 3929 unsigned int size = skb->len; 3930 if (likely(size >= len)) 3931 return 0; 3932 return skb_pad(skb, len - size); 3933 } 3934 3935 /** 3936 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3937 * @skb: buffer to pad 3938 * @len: minimal length 3939 * @free_on_error: free buffer on error 3940 * 3941 * Pads up a buffer to ensure the trailing bytes exist and are 3942 * blanked. If the buffer already contains sufficient data it 3943 * is untouched. Otherwise it is extended. Returns zero on 3944 * success. The skb is freed on error if @free_on_error is true. 3945 */ 3946 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3947 unsigned int len, 3948 bool free_on_error) 3949 { 3950 unsigned int size = skb->len; 3951 3952 if (unlikely(size < len)) { 3953 len -= size; 3954 if (__skb_pad(skb, len, free_on_error)) 3955 return -ENOMEM; 3956 __skb_put(skb, len); 3957 } 3958 return 0; 3959 } 3960 3961 /** 3962 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3963 * @skb: buffer to pad 3964 * @len: minimal length 3965 * 3966 * Pads up a buffer to ensure the trailing bytes exist and are 3967 * blanked. If the buffer already contains sufficient data it 3968 * is untouched. Otherwise it is extended. Returns zero on 3969 * success. The skb is freed on error. 3970 */ 3971 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3972 { 3973 return __skb_put_padto(skb, len, true); 3974 } 3975 3976 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i) 3977 __must_check; 3978 3979 static inline bool skb_can_coalesce_netmem(struct sk_buff *skb, int i, 3980 netmem_ref netmem, int off) 3981 { 3982 if (skb_zcopy(skb)) 3983 return false; 3984 if (i) { 3985 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3986 3987 return netmem == skb_frag_netmem(frag) && 3988 off == skb_frag_off(frag) + skb_frag_size(frag); 3989 } 3990 return false; 3991 } 3992 3993 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3994 const struct page *page, int off) 3995 { 3996 return skb_can_coalesce_netmem(skb, i, page_to_netmem(page), off); 3997 } 3998 3999 static inline int __skb_linearize(struct sk_buff *skb) 4000 { 4001 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 4002 } 4003 4004 /** 4005 * skb_linearize - convert paged skb to linear one 4006 * @skb: buffer to linarize 4007 * 4008 * If there is no free memory -ENOMEM is returned, otherwise zero 4009 * is returned and the old skb data released. 4010 */ 4011 static inline int skb_linearize(struct sk_buff *skb) 4012 { 4013 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 4014 } 4015 4016 /** 4017 * skb_has_shared_frag - can any frag be overwritten 4018 * @skb: buffer to test 4019 * 4020 * Return: true if the skb has at least one frag that might be modified 4021 * by an external entity (as in vmsplice()/sendfile()) 4022 */ 4023 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 4024 { 4025 return skb_is_nonlinear(skb) && 4026 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 4027 } 4028 4029 /** 4030 * skb_linearize_cow - make sure skb is linear and writable 4031 * @skb: buffer to process 4032 * 4033 * If there is no free memory -ENOMEM is returned, otherwise zero 4034 * is returned and the old skb data released. 4035 */ 4036 static inline int skb_linearize_cow(struct sk_buff *skb) 4037 { 4038 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 4039 __skb_linearize(skb) : 0; 4040 } 4041 4042 static __always_inline void 4043 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 4044 unsigned int off) 4045 { 4046 if (skb->ip_summed == CHECKSUM_COMPLETE) 4047 skb->csum = csum_block_sub(skb->csum, 4048 csum_partial(start, len, 0), off); 4049 else if (skb->ip_summed == CHECKSUM_PARTIAL && 4050 skb_checksum_start_offset(skb) < 0) 4051 skb->ip_summed = CHECKSUM_NONE; 4052 } 4053 4054 /** 4055 * skb_postpull_rcsum - update checksum for received skb after pull 4056 * @skb: buffer to update 4057 * @start: start of data before pull 4058 * @len: length of data pulled 4059 * 4060 * After doing a pull on a received packet, you need to call this to 4061 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 4062 * CHECKSUM_NONE so that it can be recomputed from scratch. 4063 */ 4064 static __always_inline void 4065 skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len) 4066 { 4067 if (skb->ip_summed == CHECKSUM_COMPLETE) 4068 skb->csum = wsum_negate(csum_partial(start, len, 4069 wsum_negate(skb->csum))); 4070 else if (skb->ip_summed == CHECKSUM_PARTIAL && 4071 skb_checksum_start_offset(skb) < 0) 4072 skb->ip_summed = CHECKSUM_NONE; 4073 } 4074 4075 static __always_inline void 4076 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 4077 unsigned int off) 4078 { 4079 if (skb->ip_summed == CHECKSUM_COMPLETE) 4080 skb->csum = csum_block_add(skb->csum, 4081 csum_partial(start, len, 0), off); 4082 } 4083 4084 /** 4085 * skb_postpush_rcsum - update checksum for received skb after push 4086 * @skb: buffer to update 4087 * @start: start of data after push 4088 * @len: length of data pushed 4089 * 4090 * After doing a push on a received packet, you need to call this to 4091 * update the CHECKSUM_COMPLETE checksum. 4092 */ 4093 static inline void skb_postpush_rcsum(struct sk_buff *skb, 4094 const void *start, unsigned int len) 4095 { 4096 __skb_postpush_rcsum(skb, start, len, 0); 4097 } 4098 4099 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 4100 4101 /** 4102 * skb_push_rcsum - push skb and update receive checksum 4103 * @skb: buffer to update 4104 * @len: length of data pulled 4105 * 4106 * This function performs an skb_push on the packet and updates 4107 * the CHECKSUM_COMPLETE checksum. It should be used on 4108 * receive path processing instead of skb_push unless you know 4109 * that the checksum difference is zero (e.g., a valid IP header) 4110 * or you are setting ip_summed to CHECKSUM_NONE. 4111 */ 4112 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 4113 { 4114 skb_push(skb, len); 4115 skb_postpush_rcsum(skb, skb->data, len); 4116 return skb->data; 4117 } 4118 4119 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 4120 /** 4121 * pskb_trim_rcsum - trim received skb and update checksum 4122 * @skb: buffer to trim 4123 * @len: new length 4124 * 4125 * This is exactly the same as pskb_trim except that it ensures the 4126 * checksum of received packets are still valid after the operation. 4127 * It can change skb pointers. 4128 */ 4129 4130 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4131 { 4132 skb_might_realloc(skb); 4133 if (likely(len >= skb->len)) 4134 return 0; 4135 return pskb_trim_rcsum_slow(skb, len); 4136 } 4137 4138 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4139 { 4140 if (skb->ip_summed == CHECKSUM_COMPLETE) 4141 skb->ip_summed = CHECKSUM_NONE; 4142 __skb_trim(skb, len); 4143 return 0; 4144 } 4145 4146 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 4147 { 4148 if (skb->ip_summed == CHECKSUM_COMPLETE) 4149 skb->ip_summed = CHECKSUM_NONE; 4150 return __skb_grow(skb, len); 4151 } 4152 4153 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 4154 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 4155 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 4156 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 4157 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 4158 4159 #define skb_queue_walk(queue, skb) \ 4160 for (skb = (queue)->next; \ 4161 skb != (struct sk_buff *)(queue); \ 4162 skb = skb->next) 4163 4164 #define skb_queue_walk_safe(queue, skb, tmp) \ 4165 for (skb = (queue)->next, tmp = skb->next; \ 4166 skb != (struct sk_buff *)(queue); \ 4167 skb = tmp, tmp = skb->next) 4168 4169 #define skb_queue_walk_from(queue, skb) \ 4170 for (; skb != (struct sk_buff *)(queue); \ 4171 skb = skb->next) 4172 4173 #define skb_rbtree_walk(skb, root) \ 4174 for (skb = skb_rb_first(root); skb != NULL; \ 4175 skb = skb_rb_next(skb)) 4176 4177 #define skb_rbtree_walk_from(skb) \ 4178 for (; skb != NULL; \ 4179 skb = skb_rb_next(skb)) 4180 4181 #define skb_rbtree_walk_from_safe(skb, tmp) \ 4182 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 4183 skb = tmp) 4184 4185 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 4186 for (tmp = skb->next; \ 4187 skb != (struct sk_buff *)(queue); \ 4188 skb = tmp, tmp = skb->next) 4189 4190 #define skb_queue_reverse_walk(queue, skb) \ 4191 for (skb = (queue)->prev; \ 4192 skb != (struct sk_buff *)(queue); \ 4193 skb = skb->prev) 4194 4195 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 4196 for (skb = (queue)->prev, tmp = skb->prev; \ 4197 skb != (struct sk_buff *)(queue); \ 4198 skb = tmp, tmp = skb->prev) 4199 4200 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 4201 for (tmp = skb->prev; \ 4202 skb != (struct sk_buff *)(queue); \ 4203 skb = tmp, tmp = skb->prev) 4204 4205 static inline bool skb_has_frag_list(const struct sk_buff *skb) 4206 { 4207 return skb_shinfo(skb)->frag_list != NULL; 4208 } 4209 4210 static inline void skb_frag_list_init(struct sk_buff *skb) 4211 { 4212 skb_shinfo(skb)->frag_list = NULL; 4213 } 4214 4215 #define skb_walk_frags(skb, iter) \ 4216 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 4217 4218 4219 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 4220 int *err, long *timeo_p, 4221 const struct sk_buff *skb); 4222 struct sk_buff *__skb_try_recv_from_queue(struct sk_buff_head *queue, 4223 unsigned int flags, 4224 int *off, int *err, 4225 struct sk_buff **last); 4226 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 4227 struct sk_buff_head *queue, 4228 unsigned int flags, int *off, int *err, 4229 struct sk_buff **last); 4230 struct sk_buff *__skb_recv_datagram(struct sock *sk, 4231 struct sk_buff_head *sk_queue, 4232 unsigned int flags, int *off, int *err); 4233 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 4234 __poll_t datagram_poll_queue(struct file *file, struct socket *sock, 4235 struct poll_table_struct *wait, 4236 struct sk_buff_head *rcv_queue); 4237 __poll_t datagram_poll(struct file *file, struct socket *sock, 4238 struct poll_table_struct *wait); 4239 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 4240 struct iov_iter *to, int size); 4241 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 4242 struct msghdr *msg, int size) 4243 { 4244 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 4245 } 4246 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 4247 struct msghdr *msg); 4248 int skb_copy_and_crc32c_datagram_iter(const struct sk_buff *skb, int offset, 4249 struct iov_iter *to, int len, u32 *crcp); 4250 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 4251 struct iov_iter *from, int len); 4252 int skb_copy_datagram_from_iter_full(struct sk_buff *skb, int offset, 4253 struct iov_iter *from, int len); 4254 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 4255 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 4256 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 4257 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 4258 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 4259 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 4260 int len); 4261 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 4262 struct pipe_inode_info *pipe, unsigned int len, 4263 unsigned int flags); 4264 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 4265 int len); 4266 int skb_send_sock_locked_with_flags(struct sock *sk, struct sk_buff *skb, 4267 int offset, int len, int flags); 4268 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 4269 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 4270 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 4271 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 4272 int len, int hlen); 4273 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 4274 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 4275 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 4276 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 4277 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 4278 unsigned int offset); 4279 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 4280 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 4281 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev); 4282 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 4283 int skb_vlan_pop(struct sk_buff *skb); 4284 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 4285 int skb_eth_pop(struct sk_buff *skb); 4286 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 4287 const unsigned char *src); 4288 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 4289 int mac_len, bool ethernet); 4290 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 4291 bool ethernet); 4292 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 4293 int skb_mpls_dec_ttl(struct sk_buff *skb); 4294 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 4295 gfp_t gfp); 4296 4297 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 4298 { 4299 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 4300 } 4301 4302 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 4303 { 4304 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 4305 } 4306 4307 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 4308 __wsum csum); 4309 u32 skb_crc32c(const struct sk_buff *skb, int offset, int len, u32 crc); 4310 4311 static inline void * __must_check 4312 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 4313 const void *data, int hlen, void *buffer) 4314 { 4315 if (likely(hlen - offset >= len)) 4316 return (void *)data + offset; 4317 4318 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 4319 return NULL; 4320 4321 return buffer; 4322 } 4323 4324 static __always_inline void * __must_check 4325 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4326 { 4327 return __skb_header_pointer(skb, offset, len, skb->data, 4328 skb_headlen(skb), buffer); 4329 } 4330 4331 /* Variant of skb_header_pointer() where @offset is user-controlled 4332 * and potentially negative. 4333 */ 4334 static inline void * __must_check 4335 skb_header_pointer_careful(const struct sk_buff *skb, int offset, 4336 int len, void *buffer) 4337 { 4338 if (unlikely(offset < 0 && -offset > skb_headroom(skb))) 4339 return NULL; 4340 return skb_header_pointer(skb, offset, len, buffer); 4341 } 4342 4343 static inline void * __must_check 4344 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len) 4345 { 4346 if (likely(skb_headlen(skb) - offset >= len)) 4347 return skb->data + offset; 4348 return NULL; 4349 } 4350 4351 /** 4352 * skb_needs_linearize - check if we need to linearize a given skb 4353 * depending on the given device features. 4354 * @skb: socket buffer to check 4355 * @features: net device features 4356 * 4357 * Returns true if either: 4358 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4359 * 2. skb is fragmented and the device does not support SG. 4360 */ 4361 static inline bool skb_needs_linearize(struct sk_buff *skb, 4362 netdev_features_t features) 4363 { 4364 return skb_is_nonlinear(skb) && 4365 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4366 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4367 } 4368 4369 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4370 void *to, 4371 const unsigned int len) 4372 { 4373 memcpy(to, skb->data, len); 4374 } 4375 4376 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4377 const int offset, void *to, 4378 const unsigned int len) 4379 { 4380 memcpy(to, skb->data + offset, len); 4381 } 4382 4383 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4384 const void *from, 4385 const unsigned int len) 4386 { 4387 memcpy(skb->data, from, len); 4388 } 4389 4390 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4391 const int offset, 4392 const void *from, 4393 const unsigned int len) 4394 { 4395 memcpy(skb->data + offset, from, len); 4396 } 4397 4398 void skb_init(void); 4399 4400 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4401 { 4402 return skb->tstamp; 4403 } 4404 4405 /** 4406 * skb_get_timestamp - get timestamp from a skb 4407 * @skb: skb to get stamp from 4408 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4409 * 4410 * Timestamps are stored in the skb as offsets to a base timestamp. 4411 * This function converts the offset back to a struct timeval and stores 4412 * it in stamp. 4413 */ 4414 static inline void skb_get_timestamp(const struct sk_buff *skb, 4415 struct __kernel_old_timeval *stamp) 4416 { 4417 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4418 } 4419 4420 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4421 struct __kernel_sock_timeval *stamp) 4422 { 4423 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4424 4425 stamp->tv_sec = ts.tv_sec; 4426 stamp->tv_usec = ts.tv_nsec / 1000; 4427 } 4428 4429 static inline void skb_get_timestampns(const struct sk_buff *skb, 4430 struct __kernel_old_timespec *stamp) 4431 { 4432 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4433 4434 stamp->tv_sec = ts.tv_sec; 4435 stamp->tv_nsec = ts.tv_nsec; 4436 } 4437 4438 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4439 struct __kernel_timespec *stamp) 4440 { 4441 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4442 4443 stamp->tv_sec = ts.tv_sec; 4444 stamp->tv_nsec = ts.tv_nsec; 4445 } 4446 4447 static inline void __net_timestamp(struct sk_buff *skb) 4448 { 4449 skb->tstamp = ktime_get_real(); 4450 skb->tstamp_type = SKB_CLOCK_REALTIME; 4451 } 4452 4453 static inline ktime_t net_timedelta(ktime_t t) 4454 { 4455 return ktime_sub(ktime_get_real(), t); 4456 } 4457 4458 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4459 u8 tstamp_type) 4460 { 4461 skb->tstamp = kt; 4462 4463 if (kt) 4464 skb->tstamp_type = tstamp_type; 4465 else 4466 skb->tstamp_type = SKB_CLOCK_REALTIME; 4467 } 4468 4469 static inline void skb_set_delivery_type_by_clockid(struct sk_buff *skb, 4470 ktime_t kt, clockid_t clockid) 4471 { 4472 u8 tstamp_type = SKB_CLOCK_REALTIME; 4473 4474 switch (clockid) { 4475 case CLOCK_REALTIME: 4476 break; 4477 case CLOCK_MONOTONIC: 4478 tstamp_type = SKB_CLOCK_MONOTONIC; 4479 break; 4480 case CLOCK_TAI: 4481 tstamp_type = SKB_CLOCK_TAI; 4482 break; 4483 default: 4484 WARN_ON_ONCE(1); 4485 kt = 0; 4486 } 4487 4488 skb_set_delivery_time(skb, kt, tstamp_type); 4489 } 4490 4491 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4492 4493 /* It is used in the ingress path to clear the delivery_time. 4494 * If needed, set the skb->tstamp to the (rcv) timestamp. 4495 */ 4496 static __always_inline void skb_clear_delivery_time(struct sk_buff *skb) 4497 { 4498 if (skb->tstamp_type) { 4499 skb->tstamp_type = SKB_CLOCK_REALTIME; 4500 if (static_branch_unlikely(&netstamp_needed_key)) 4501 skb->tstamp = ktime_get_real(); 4502 else 4503 skb->tstamp = 0; 4504 } 4505 } 4506 4507 static inline void skb_clear_tstamp(struct sk_buff *skb) 4508 { 4509 if (skb->tstamp_type) 4510 return; 4511 4512 skb->tstamp = 0; 4513 } 4514 4515 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4516 { 4517 if (skb->tstamp_type) 4518 return 0; 4519 4520 return skb->tstamp; 4521 } 4522 4523 static __always_inline ktime_t 4524 skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4525 { 4526 if (skb->tstamp_type != SKB_CLOCK_MONOTONIC && skb->tstamp) 4527 return skb->tstamp; 4528 4529 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4530 return ktime_get_real(); 4531 4532 return 0; 4533 } 4534 4535 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4536 { 4537 return skb_shinfo(skb)->meta_len; 4538 } 4539 4540 static inline void *skb_metadata_end(const struct sk_buff *skb) 4541 { 4542 return skb_mac_header(skb); 4543 } 4544 4545 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4546 const struct sk_buff *skb_b, 4547 u8 meta_len) 4548 { 4549 const void *a = skb_metadata_end(skb_a); 4550 const void *b = skb_metadata_end(skb_b); 4551 u64 diffs = 0; 4552 4553 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) || 4554 BITS_PER_LONG != 64) 4555 goto slow; 4556 4557 /* Using more efficient variant than plain call to memcmp(). */ 4558 switch (meta_len) { 4559 #define __it(x, op) (x -= sizeof(u##op)) 4560 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4561 case 32: diffs |= __it_diff(a, b, 64); 4562 fallthrough; 4563 case 24: diffs |= __it_diff(a, b, 64); 4564 fallthrough; 4565 case 16: diffs |= __it_diff(a, b, 64); 4566 fallthrough; 4567 case 8: diffs |= __it_diff(a, b, 64); 4568 break; 4569 case 28: diffs |= __it_diff(a, b, 64); 4570 fallthrough; 4571 case 20: diffs |= __it_diff(a, b, 64); 4572 fallthrough; 4573 case 12: diffs |= __it_diff(a, b, 64); 4574 fallthrough; 4575 case 4: diffs |= __it_diff(a, b, 32); 4576 break; 4577 default: 4578 slow: 4579 return memcmp(a - meta_len, b - meta_len, meta_len); 4580 } 4581 return diffs; 4582 } 4583 4584 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4585 const struct sk_buff *skb_b) 4586 { 4587 u8 len_a = skb_metadata_len(skb_a); 4588 u8 len_b = skb_metadata_len(skb_b); 4589 4590 if (!(len_a | len_b)) 4591 return false; 4592 4593 return len_a != len_b ? 4594 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4595 } 4596 4597 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4598 { 4599 skb_shinfo(skb)->meta_len = meta_len; 4600 } 4601 4602 static inline void skb_metadata_clear(struct sk_buff *skb) 4603 { 4604 skb_metadata_set(skb, 0); 4605 } 4606 4607 /** 4608 * skb_data_move - Move packet data and metadata after skb_push() or skb_pull(). 4609 * @skb: packet to operate on 4610 * @len: number of bytes pushed or pulled from &sk_buff->data 4611 * @n: number of bytes to memmove() from pre-push/pull &sk_buff->data 4612 * 4613 * Moves @n bytes of packet data, can be zero, and all bytes of skb metadata. 4614 * 4615 * Assumes metadata is located immediately before &sk_buff->data prior to the 4616 * push/pull, and that sufficient headroom exists to hold it after an 4617 * skb_push(). Otherwise, metadata is cleared and a one-time warning is issued. 4618 * 4619 * Prefer skb_postpull_data_move() or skb_postpush_data_move() to calling this 4620 * helper directly. 4621 */ 4622 static inline void skb_data_move(struct sk_buff *skb, const int len, 4623 const unsigned int n) 4624 { 4625 const u8 meta_len = skb_metadata_len(skb); 4626 u8 *meta, *meta_end; 4627 4628 if (!len || (!n && !meta_len)) 4629 return; 4630 4631 if (!meta_len) 4632 goto no_metadata; 4633 4634 meta_end = skb_metadata_end(skb); 4635 meta = meta_end - meta_len; 4636 4637 if (WARN_ON_ONCE(meta_end + len != skb->data || 4638 meta_len > skb_headroom(skb))) { 4639 skb_metadata_clear(skb); 4640 goto no_metadata; 4641 } 4642 4643 memmove(meta + len, meta, meta_len + n); 4644 return; 4645 4646 no_metadata: 4647 memmove(skb->data, skb->data - len, n); 4648 } 4649 4650 /** 4651 * skb_postpull_data_move - Move packet data and metadata after skb_pull(). 4652 * @skb: packet to operate on 4653 * @len: number of bytes pulled from &sk_buff->data 4654 * @n: number of bytes to memmove() from pre-pull &sk_buff->data 4655 * 4656 * See skb_data_move() for details. 4657 */ 4658 static inline void skb_postpull_data_move(struct sk_buff *skb, 4659 const unsigned int len, 4660 const unsigned int n) 4661 { 4662 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 4663 skb_data_move(skb, len, n); 4664 } 4665 4666 /** 4667 * skb_postpush_data_move - Move packet data and metadata after skb_push(). 4668 * @skb: packet to operate on 4669 * @len: number of bytes pushed onto &sk_buff->data 4670 * @n: number of bytes to memmove() from pre-push &sk_buff->data 4671 * 4672 * See skb_data_move() for details. 4673 */ 4674 static inline void skb_postpush_data_move(struct sk_buff *skb, 4675 const unsigned int len, 4676 const unsigned int n) 4677 { 4678 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 4679 skb_data_move(skb, -len, n); 4680 } 4681 4682 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4683 4684 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4685 4686 void skb_clone_tx_timestamp(struct sk_buff *skb); 4687 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4688 4689 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4690 4691 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4692 { 4693 } 4694 4695 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4696 { 4697 return false; 4698 } 4699 4700 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4701 4702 /** 4703 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4704 * 4705 * PHY drivers may accept clones of transmitted packets for 4706 * timestamping via their phy_driver.txtstamp method. These drivers 4707 * must call this function to return the skb back to the stack with a 4708 * timestamp. 4709 * 4710 * @skb: clone of the original outgoing packet 4711 * @hwtstamps: hardware time stamps 4712 * 4713 */ 4714 void skb_complete_tx_timestamp(struct sk_buff *skb, 4715 struct skb_shared_hwtstamps *hwtstamps); 4716 4717 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4718 struct skb_shared_hwtstamps *hwtstamps, 4719 struct sock *sk, int tstype); 4720 4721 /** 4722 * skb_tstamp_tx - queue clone of skb with send time stamps 4723 * @orig_skb: the original outgoing packet 4724 * @hwtstamps: hardware time stamps, may be NULL if not available 4725 * 4726 * If the skb has a socket associated, then this function clones the 4727 * skb (thus sharing the actual data and optional structures), stores 4728 * the optional hardware time stamping information (if non NULL) or 4729 * generates a software time stamp (otherwise), then queues the clone 4730 * to the error queue of the socket. Errors are silently ignored. 4731 */ 4732 void skb_tstamp_tx(struct sk_buff *orig_skb, 4733 struct skb_shared_hwtstamps *hwtstamps); 4734 4735 /** 4736 * skb_tx_timestamp() - Driver hook for transmit timestamping 4737 * 4738 * Ethernet MAC Drivers should call this function in their hard_xmit() 4739 * function immediately before giving the sk_buff to the MAC hardware. 4740 * 4741 * Specifically, one should make absolutely sure that this function is 4742 * called before TX completion of this packet can trigger. Otherwise 4743 * the packet could potentially already be freed. 4744 * 4745 * @skb: A socket buffer. 4746 */ 4747 static inline void skb_tx_timestamp(struct sk_buff *skb) 4748 { 4749 skb_clone_tx_timestamp(skb); 4750 if (skb_shinfo(skb)->tx_flags & (SKBTX_SW_TSTAMP | SKBTX_BPF)) 4751 skb_tstamp_tx(skb, NULL); 4752 } 4753 4754 /** 4755 * skb_complete_wifi_ack - deliver skb with wifi status 4756 * 4757 * @skb: the original outgoing packet 4758 * @acked: ack status 4759 * 4760 */ 4761 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4762 4763 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4764 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4765 4766 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4767 { 4768 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4769 skb->csum_valid || 4770 (skb->ip_summed == CHECKSUM_PARTIAL && 4771 skb_checksum_start_offset(skb) >= 0)); 4772 } 4773 4774 /** 4775 * skb_checksum_complete - Calculate checksum of an entire packet 4776 * @skb: packet to process 4777 * 4778 * This function calculates the checksum over the entire packet plus 4779 * the value of skb->csum. The latter can be used to supply the 4780 * checksum of a pseudo header as used by TCP/UDP. It returns the 4781 * checksum. 4782 * 4783 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4784 * this function can be used to verify that checksum on received 4785 * packets. In that case the function should return zero if the 4786 * checksum is correct. In particular, this function will return zero 4787 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4788 * hardware has already verified the correctness of the checksum. 4789 */ 4790 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4791 { 4792 return skb_csum_unnecessary(skb) ? 4793 0 : __skb_checksum_complete(skb); 4794 } 4795 4796 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4797 { 4798 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4799 if (skb->csum_level == 0) 4800 skb->ip_summed = CHECKSUM_NONE; 4801 else 4802 skb->csum_level--; 4803 } 4804 } 4805 4806 static __always_inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4807 { 4808 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4809 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4810 skb->csum_level++; 4811 } else if (skb->ip_summed == CHECKSUM_NONE) { 4812 skb->ip_summed = CHECKSUM_UNNECESSARY; 4813 skb->csum_level = 0; 4814 } 4815 } 4816 4817 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4818 { 4819 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4820 skb->ip_summed = CHECKSUM_NONE; 4821 skb->csum_level = 0; 4822 } 4823 } 4824 4825 /* Check if we need to perform checksum complete validation. 4826 * 4827 * Returns: true if checksum complete is needed, false otherwise 4828 * (either checksum is unnecessary or zero checksum is allowed). 4829 */ 4830 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4831 bool zero_okay, 4832 __sum16 check) 4833 { 4834 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4835 skb->csum_valid = 1; 4836 __skb_decr_checksum_unnecessary(skb); 4837 return false; 4838 } 4839 4840 return true; 4841 } 4842 4843 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4844 * in checksum_init. 4845 */ 4846 #define CHECKSUM_BREAK 76 4847 4848 /* Unset checksum-complete 4849 * 4850 * Unset checksum complete can be done when packet is being modified 4851 * (uncompressed for instance) and checksum-complete value is 4852 * invalidated. 4853 */ 4854 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4855 { 4856 if (skb->ip_summed == CHECKSUM_COMPLETE) 4857 skb->ip_summed = CHECKSUM_NONE; 4858 } 4859 4860 /* Validate (init) checksum based on checksum complete. 4861 * 4862 * Return values: 4863 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4864 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4865 * checksum is stored in skb->csum for use in __skb_checksum_complete 4866 * non-zero: value of invalid checksum 4867 * 4868 */ 4869 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4870 bool complete, 4871 __wsum psum) 4872 { 4873 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4874 if (!csum_fold(csum_add(psum, skb->csum))) { 4875 skb->csum_valid = 1; 4876 return 0; 4877 } 4878 } 4879 4880 skb->csum = psum; 4881 4882 if (complete || skb->len <= CHECKSUM_BREAK) { 4883 __sum16 csum; 4884 4885 csum = __skb_checksum_complete(skb); 4886 skb->csum_valid = !csum; 4887 return csum; 4888 } 4889 4890 return 0; 4891 } 4892 4893 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4894 { 4895 return 0; 4896 } 4897 4898 /* Perform checksum validate (init). Note that this is a macro since we only 4899 * want to calculate the pseudo header which is an input function if necessary. 4900 * First we try to validate without any computation (checksum unnecessary) and 4901 * then calculate based on checksum complete calling the function to compute 4902 * pseudo header. 4903 * 4904 * Return values: 4905 * 0: checksum is validated or try to in skb_checksum_complete 4906 * non-zero: value of invalid checksum 4907 */ 4908 #define __skb_checksum_validate(skb, proto, complete, \ 4909 zero_okay, check, compute_pseudo) \ 4910 ({ \ 4911 __sum16 __ret = 0; \ 4912 skb->csum_valid = 0; \ 4913 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4914 __ret = __skb_checksum_validate_complete(skb, \ 4915 complete, compute_pseudo(skb, proto)); \ 4916 __ret; \ 4917 }) 4918 4919 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4920 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4921 4922 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4923 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4924 4925 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4926 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4927 4928 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4929 compute_pseudo) \ 4930 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4931 4932 #define skb_checksum_simple_validate(skb) \ 4933 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4934 4935 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4936 { 4937 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4938 } 4939 4940 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4941 { 4942 skb->csum = ~pseudo; 4943 skb->ip_summed = CHECKSUM_COMPLETE; 4944 } 4945 4946 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4947 do { \ 4948 if (__skb_checksum_convert_check(skb)) \ 4949 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4950 } while (0) 4951 4952 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4953 u16 start, u16 offset) 4954 { 4955 skb->ip_summed = CHECKSUM_PARTIAL; 4956 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4957 skb->csum_offset = offset - start; 4958 } 4959 4960 /* Update skbuf and packet to reflect the remote checksum offload operation. 4961 * When called, ptr indicates the starting point for skb->csum when 4962 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4963 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4964 */ 4965 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4966 int start, int offset, bool nopartial) 4967 { 4968 __wsum delta; 4969 4970 if (!nopartial) { 4971 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4972 return; 4973 } 4974 4975 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4976 __skb_checksum_complete(skb); 4977 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4978 } 4979 4980 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4981 4982 /* Adjust skb->csum since we changed the packet */ 4983 skb->csum = csum_add(skb->csum, delta); 4984 } 4985 4986 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4987 { 4988 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4989 return (void *)(skb->_nfct & NFCT_PTRMASK); 4990 #else 4991 return NULL; 4992 #endif 4993 } 4994 4995 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4996 { 4997 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4998 return skb->_nfct; 4999 #else 5000 return 0UL; 5001 #endif 5002 } 5003 5004 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 5005 { 5006 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 5007 skb->slow_gro |= !!nfct; 5008 skb->_nfct = nfct; 5009 #endif 5010 } 5011 5012 #ifdef CONFIG_SKB_EXTENSIONS 5013 enum skb_ext_id { 5014 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 5015 SKB_EXT_BRIDGE_NF, 5016 #endif 5017 #ifdef CONFIG_XFRM 5018 SKB_EXT_SEC_PATH, 5019 #endif 5020 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5021 TC_SKB_EXT, 5022 #endif 5023 #if IS_ENABLED(CONFIG_MPTCP) 5024 SKB_EXT_MPTCP, 5025 #endif 5026 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 5027 SKB_EXT_MCTP, 5028 #endif 5029 #if IS_ENABLED(CONFIG_INET_PSP) 5030 SKB_EXT_PSP, 5031 #endif 5032 #if IS_ENABLED(CONFIG_CAN) 5033 SKB_EXT_CAN, 5034 #endif 5035 SKB_EXT_NUM, /* must be last */ 5036 }; 5037 5038 /** 5039 * struct skb_ext - sk_buff extensions 5040 * @refcnt: 1 on allocation, deallocated on 0 5041 * @offset: offset to add to @data to obtain extension address 5042 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 5043 * @data: start of extension data, variable sized 5044 * 5045 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 5046 * to use 'u8' types while allowing up to 2kb worth of extension data. 5047 */ 5048 struct skb_ext { 5049 refcount_t refcnt; 5050 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 5051 u8 chunks; /* same */ 5052 char data[] __aligned(8); 5053 }; 5054 5055 struct skb_ext *__skb_ext_alloc(gfp_t flags); 5056 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 5057 struct skb_ext *ext); 5058 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 5059 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 5060 void __skb_ext_put(struct skb_ext *ext); 5061 5062 static inline void skb_ext_put(struct sk_buff *skb) 5063 { 5064 if (skb->active_extensions) 5065 __skb_ext_put(skb->extensions); 5066 } 5067 5068 static inline void __skb_ext_copy(struct sk_buff *dst, 5069 const struct sk_buff *src) 5070 { 5071 dst->active_extensions = src->active_extensions; 5072 5073 if (src->active_extensions) { 5074 struct skb_ext *ext = src->extensions; 5075 5076 refcount_inc(&ext->refcnt); 5077 dst->extensions = ext; 5078 } 5079 } 5080 5081 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 5082 { 5083 skb_ext_put(dst); 5084 __skb_ext_copy(dst, src); 5085 } 5086 5087 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 5088 { 5089 return !!ext->offset[i]; 5090 } 5091 5092 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 5093 { 5094 return skb->active_extensions & (1 << id); 5095 } 5096 5097 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 5098 { 5099 if (skb_ext_exist(skb, id)) 5100 __skb_ext_del(skb, id); 5101 } 5102 5103 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 5104 { 5105 if (skb_ext_exist(skb, id)) { 5106 struct skb_ext *ext = skb->extensions; 5107 5108 return (void *)ext + (ext->offset[id] << 3); 5109 } 5110 5111 return NULL; 5112 } 5113 5114 static inline void skb_ext_reset(struct sk_buff *skb) 5115 { 5116 if (unlikely(skb->active_extensions)) { 5117 __skb_ext_put(skb->extensions); 5118 skb->active_extensions = 0; 5119 } 5120 } 5121 5122 static inline bool skb_has_extensions(struct sk_buff *skb) 5123 { 5124 return unlikely(skb->active_extensions); 5125 } 5126 #else 5127 static inline void __skb_ext_put(struct skb_ext *ext) {} 5128 static inline void skb_ext_put(struct sk_buff *skb) {} 5129 static inline void skb_ext_reset(struct sk_buff *skb) {} 5130 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 5131 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 5132 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 5133 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 5134 #endif /* CONFIG_SKB_EXTENSIONS */ 5135 5136 static inline void nf_reset_ct(struct sk_buff *skb) 5137 { 5138 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5139 nf_conntrack_put(skb_nfct(skb)); 5140 skb->_nfct = 0; 5141 #endif 5142 } 5143 5144 static inline void nf_reset_trace(struct sk_buff *skb) 5145 { 5146 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 5147 skb->nf_trace = 0; 5148 #endif 5149 } 5150 5151 static inline void ipvs_reset(struct sk_buff *skb) 5152 { 5153 #if IS_ENABLED(CONFIG_IP_VS) 5154 skb->ipvs_property = 0; 5155 #endif 5156 } 5157 5158 /* Note: This doesn't put any conntrack info in dst. */ 5159 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 5160 bool copy) 5161 { 5162 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5163 dst->_nfct = src->_nfct; 5164 nf_conntrack_get(skb_nfct(src)); 5165 #endif 5166 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 5167 if (copy) 5168 dst->nf_trace = src->nf_trace; 5169 #endif 5170 } 5171 5172 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 5173 { 5174 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5175 nf_conntrack_put(skb_nfct(dst)); 5176 #endif 5177 dst->slow_gro = src->slow_gro; 5178 __nf_copy(dst, src, true); 5179 } 5180 5181 #ifdef CONFIG_NETWORK_SECMARK 5182 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 5183 { 5184 to->secmark = from->secmark; 5185 } 5186 5187 static inline void skb_init_secmark(struct sk_buff *skb) 5188 { 5189 skb->secmark = 0; 5190 } 5191 #else 5192 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 5193 { } 5194 5195 static inline void skb_init_secmark(struct sk_buff *skb) 5196 { } 5197 #endif 5198 5199 static inline int secpath_exists(const struct sk_buff *skb) 5200 { 5201 #ifdef CONFIG_XFRM 5202 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 5203 #else 5204 return 0; 5205 #endif 5206 } 5207 5208 static inline bool skb_irq_freeable(const struct sk_buff *skb) 5209 { 5210 return !skb->destructor && 5211 !secpath_exists(skb) && 5212 !skb_nfct(skb) && 5213 !skb->_skb_refdst && 5214 !skb_has_frag_list(skb); 5215 } 5216 5217 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 5218 { 5219 skb->queue_mapping = queue_mapping; 5220 } 5221 5222 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 5223 { 5224 return skb->queue_mapping; 5225 } 5226 5227 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 5228 { 5229 to->queue_mapping = from->queue_mapping; 5230 } 5231 5232 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 5233 { 5234 skb->queue_mapping = rx_queue + 1; 5235 } 5236 5237 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 5238 { 5239 return skb->queue_mapping - 1; 5240 } 5241 5242 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 5243 { 5244 return skb->queue_mapping != 0; 5245 } 5246 5247 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 5248 { 5249 skb->dst_pending_confirm = val; 5250 } 5251 5252 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 5253 { 5254 return skb->dst_pending_confirm != 0; 5255 } 5256 5257 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 5258 { 5259 #ifdef CONFIG_XFRM 5260 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 5261 #else 5262 return NULL; 5263 #endif 5264 } 5265 5266 static inline bool skb_is_gso(const struct sk_buff *skb) 5267 { 5268 return skb_shinfo(skb)->gso_size; 5269 } 5270 5271 /* Note: Should be called only if skb_is_gso(skb) is true */ 5272 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 5273 { 5274 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 5275 } 5276 5277 /* Note: Should be called only if skb_is_gso(skb) is true */ 5278 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 5279 { 5280 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 5281 } 5282 5283 /* Note: Should be called only if skb_is_gso(skb) is true */ 5284 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 5285 { 5286 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 5287 } 5288 5289 static inline void skb_gso_reset(struct sk_buff *skb) 5290 { 5291 skb_shinfo(skb)->gso_size = 0; 5292 skb_shinfo(skb)->gso_segs = 0; 5293 skb_shinfo(skb)->gso_type = 0; 5294 } 5295 5296 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 5297 u16 increment) 5298 { 5299 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5300 return; 5301 shinfo->gso_size += increment; 5302 } 5303 5304 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 5305 u16 decrement) 5306 { 5307 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5308 return; 5309 shinfo->gso_size -= decrement; 5310 } 5311 5312 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 5313 5314 static __always_inline bool skb_warn_if_lro(const struct sk_buff *skb) 5315 { 5316 /* LRO sets gso_size but not gso_type, whereas if GSO is really 5317 * wanted then gso_type will be set. */ 5318 const struct skb_shared_info *shinfo = skb_shinfo(skb); 5319 5320 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 5321 unlikely(shinfo->gso_type == 0)) { 5322 __skb_warn_lro_forwarding(skb); 5323 return true; 5324 } 5325 return false; 5326 } 5327 5328 static inline void skb_forward_csum(struct sk_buff *skb) 5329 { 5330 /* Unfortunately we don't support this one. Any brave souls? */ 5331 if (skb->ip_summed == CHECKSUM_COMPLETE) 5332 skb->ip_summed = CHECKSUM_NONE; 5333 } 5334 5335 /** 5336 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 5337 * @skb: skb to check 5338 * 5339 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 5340 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 5341 * use this helper, to document places where we make this assertion. 5342 */ 5343 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 5344 { 5345 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 5346 } 5347 5348 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 5349 5350 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 5351 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 5352 unsigned int transport_len, 5353 __sum16(*skb_chkf)(struct sk_buff *skb)); 5354 5355 /** 5356 * skb_head_is_locked - Determine if the skb->head is locked down 5357 * @skb: skb to check 5358 * 5359 * The head on skbs build around a head frag can be removed if they are 5360 * not cloned. This function returns true if the skb head is locked down 5361 * due to either being allocated via kmalloc, or by being a clone with 5362 * multiple references to the head. 5363 */ 5364 static inline bool skb_head_is_locked(const struct sk_buff *skb) 5365 { 5366 return !skb->head_frag || skb_cloned(skb); 5367 } 5368 5369 /* Local Checksum Offload. 5370 * Compute outer checksum based on the assumption that the 5371 * inner checksum will be offloaded later. 5372 * See Documentation/networking/checksum-offloads.rst for 5373 * explanation of how this works. 5374 * Fill in outer checksum adjustment (e.g. with sum of outer 5375 * pseudo-header) before calling. 5376 * Also ensure that inner checksum is in linear data area. 5377 */ 5378 static inline __wsum lco_csum(struct sk_buff *skb) 5379 { 5380 unsigned char *csum_start = skb_checksum_start(skb); 5381 unsigned char *l4_hdr = skb_transport_header(skb); 5382 __wsum partial; 5383 5384 /* Start with complement of inner checksum adjustment */ 5385 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 5386 skb->csum_offset)); 5387 5388 /* Add in checksum of our headers (incl. outer checksum 5389 * adjustment filled in by caller) and return result. 5390 */ 5391 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5392 } 5393 5394 static inline bool skb_is_redirected(const struct sk_buff *skb) 5395 { 5396 return skb->redirected; 5397 } 5398 5399 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5400 { 5401 skb->redirected = 1; 5402 #ifdef CONFIG_NET_REDIRECT 5403 skb->from_ingress = from_ingress; 5404 if (skb->from_ingress) 5405 skb_clear_tstamp(skb); 5406 #endif 5407 } 5408 5409 static inline void skb_reset_redirect(struct sk_buff *skb) 5410 { 5411 skb->redirected = 0; 5412 } 5413 5414 static inline void skb_set_redirected_noclear(struct sk_buff *skb, 5415 bool from_ingress) 5416 { 5417 skb->redirected = 1; 5418 #ifdef CONFIG_NET_REDIRECT 5419 skb->from_ingress = from_ingress; 5420 #endif 5421 } 5422 5423 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5424 { 5425 #if IS_ENABLED(CONFIG_IP_SCTP) 5426 return skb->csum_not_inet; 5427 #else 5428 return 0; 5429 #endif 5430 } 5431 5432 static inline void skb_reset_csum_not_inet(struct sk_buff *skb) 5433 { 5434 skb->ip_summed = CHECKSUM_NONE; 5435 #if IS_ENABLED(CONFIG_IP_SCTP) 5436 skb->csum_not_inet = 0; 5437 #endif 5438 } 5439 5440 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5441 struct kcov_common_handle_id kcov_handle) 5442 { 5443 skb->kcov_handle = kcov_handle; 5444 } 5445 5446 static inline struct kcov_common_handle_id skb_get_kcov_handle(struct sk_buff *skb) 5447 { 5448 return skb->kcov_handle; 5449 } 5450 5451 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5452 { 5453 #ifdef CONFIG_PAGE_POOL 5454 skb->pp_recycle = 1; 5455 #endif 5456 } 5457 5458 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter, 5459 ssize_t maxsize); 5460 5461 #endif /* __KERNEL__ */ 5462 #endif /* _LINUX_SKBUFF_H */ 5463