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_index: Traffic control index 825 * @hash: the packet hash 826 * @queue_mapping: Queue mapping for multiqueue devices 827 * @head_frag: skb was allocated from page fragments, 828 * not allocated by kmalloc() or vmalloc(). 829 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 830 * @pp_recycle: mark the packet for recycling instead of freeing (implies 831 * page_pool support on driver) 832 * @active_extensions: active extensions (skb_ext_id types) 833 * @ndisc_nodetype: router type (from link layer) 834 * @ooo_okay: allow the mapping of a socket to a queue to be changed 835 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 836 * ports. 837 * @sw_hash: indicates hash was computed in software stack 838 * @wifi_acked_valid: wifi_acked was set 839 * @wifi_acked: whether frame was acked on wifi or not 840 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 841 * @encapsulation: indicates the inner headers in the skbuff are valid 842 * @encap_hdr_csum: software checksum is needed 843 * @csum_valid: checksum is already valid 844 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 845 * @csum_complete_sw: checksum was completed by software 846 * @csum_level: indicates the number of consecutive checksums found in 847 * the packet minus one that have been verified as 848 * CHECKSUM_UNNECESSARY (max 3) 849 * @unreadable: indicates that at least 1 of the fragments in this skb is 850 * unreadable. 851 * @dst_pending_confirm: need to confirm neighbour 852 * @decrypted: Decrypted SKB 853 * @slow_gro: state present at GRO time, slower prepare step required 854 * @tstamp_type: When set, skb->tstamp has the 855 * delivery_time clock base of skb->tstamp. 856 * @napi_id: id of the NAPI struct this skb came from 857 * @sender_cpu: (aka @napi_id) source CPU in XPS 858 * @alloc_cpu: CPU which did the skb allocation. 859 * @secmark: security marking 860 * @mark: Generic packet mark 861 * @reserved_tailroom: (aka @mark) number of bytes of free space available 862 * at the tail of an sk_buff 863 * @vlan_all: vlan fields (proto & tci) 864 * @vlan_proto: vlan encapsulation protocol 865 * @vlan_tci: vlan tag control information 866 * @inner_protocol: Protocol (encapsulation) 867 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 868 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 869 * @inner_transport_header: Inner transport layer header (encapsulation) 870 * @inner_network_header: Network layer header (encapsulation) 871 * @inner_mac_header: Link layer header (encapsulation) 872 * @transport_header: Transport layer header 873 * @network_header: Network layer header 874 * @mac_header: Link layer header 875 * @kcov_handle: KCOV remote handle for remote coverage collection 876 * @tail: Tail pointer 877 * @end: End pointer 878 * @head: Head of buffer 879 * @data: Data head pointer 880 * @truesize: Buffer size 881 * @users: User count - see {datagram,tcp}.c 882 * @extensions: allocated extensions, valid if active_extensions is nonzero 883 */ 884 885 struct sk_buff { 886 union { 887 struct { 888 /* These two members must be first to match sk_buff_head. */ 889 struct sk_buff *next; 890 struct sk_buff *prev; 891 892 union { 893 struct net_device *dev; 894 /* Some protocols might use this space to store information, 895 * while device pointer would be NULL. 896 * UDP receive path is one user. 897 */ 898 unsigned long dev_scratch; 899 }; 900 }; 901 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 902 struct list_head list; 903 struct llist_node ll_node; 904 }; 905 906 struct sock *sk; 907 908 union { 909 ktime_t tstamp; 910 u64 skb_mstamp_ns; /* earliest departure time */ 911 }; 912 /* 913 * This is the control buffer. It is free to use for every 914 * layer. Please put your private variables there. If you 915 * want to keep them across layers you have to do a skb_clone() 916 * first. This is owned by whoever has the skb queued ATM. 917 */ 918 char cb[48] __aligned(8); 919 920 union { 921 struct { 922 unsigned long _skb_refdst; 923 void (*destructor)(struct sk_buff *skb); 924 }; 925 struct list_head tcp_tsorted_anchor; 926 #ifdef CONFIG_NET_SOCK_MSG 927 unsigned long _sk_redir; 928 #endif 929 }; 930 931 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 932 unsigned long _nfct; 933 #endif 934 unsigned int len, 935 data_len; 936 __u16 mac_len, 937 hdr_len; 938 939 /* Following fields are _not_ copied in __copy_skb_header() 940 * Note that queue_mapping is here mostly to fill a hole. 941 */ 942 __u16 queue_mapping; 943 944 /* if you move cloned around you also must adapt those constants */ 945 #ifdef __BIG_ENDIAN_BITFIELD 946 #define CLONED_MASK (1 << 7) 947 #else 948 #define CLONED_MASK 1 949 #endif 950 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 951 952 /* private: */ 953 __u8 __cloned_offset[0]; 954 /* public: */ 955 __u8 cloned:1, 956 nohdr:1, 957 fclone:2, 958 peeked:1, 959 head_frag:1, 960 pfmemalloc:1, 961 pp_recycle:1; /* page_pool recycle indicator */ 962 #ifdef CONFIG_SKB_EXTENSIONS 963 __u8 active_extensions; 964 #endif 965 966 /* Fields enclosed in headers group are copied 967 * using a single memcpy() in __copy_skb_header() 968 */ 969 struct_group(headers, 970 971 /* private: */ 972 __u8 __pkt_type_offset[0]; 973 /* public: */ 974 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 975 __u8 ignore_df:1; 976 __u8 dst_pending_confirm:1; 977 __u8 ip_summed:2; 978 __u8 ooo_okay:1; 979 980 /* private: */ 981 __u8 __mono_tc_offset[0]; 982 /* public: */ 983 __u8 tstamp_type:2; /* See skb_tstamp_type */ 984 #ifdef CONFIG_NET_XGRESS 985 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 986 __u8 tc_skip_classify:1; 987 #endif 988 __u8 remcsum_offload:1; 989 __u8 csum_complete_sw:1; 990 __u8 csum_level:2; 991 __u8 inner_protocol_type:1; 992 993 __u8 l4_hash:1; 994 __u8 sw_hash:1; 995 #ifdef CONFIG_WIRELESS 996 __u8 wifi_acked_valid:1; 997 __u8 wifi_acked:1; 998 #endif 999 __u8 no_fcs:1; 1000 /* Indicates the inner headers are valid in the skbuff. */ 1001 __u8 encapsulation:1; 1002 __u8 encap_hdr_csum:1; 1003 __u8 csum_valid:1; 1004 #ifdef CONFIG_IPV6_NDISC_NODETYPE 1005 __u8 ndisc_nodetype:2; 1006 #endif 1007 1008 #if IS_ENABLED(CONFIG_IP_VS) 1009 __u8 ipvs_property:1; 1010 #endif 1011 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 1012 __u8 nf_trace:1; 1013 #endif 1014 #ifdef CONFIG_NET_SWITCHDEV 1015 __u8 offload_fwd_mark:1; 1016 __u8 offload_l3_fwd_mark:1; 1017 #endif 1018 __u8 redirected:1; 1019 #ifdef CONFIG_NET_REDIRECT 1020 __u8 from_ingress:1; 1021 #endif 1022 #ifdef CONFIG_NETFILTER_SKIP_EGRESS 1023 __u8 nf_skip_egress:1; 1024 #endif 1025 #ifdef CONFIG_SKB_DECRYPTED 1026 __u8 decrypted:1; 1027 #endif 1028 __u8 slow_gro:1; 1029 #if IS_ENABLED(CONFIG_IP_SCTP) 1030 __u8 csum_not_inet:1; 1031 #endif 1032 __u8 unreadable:1; 1033 #if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS) 1034 __u16 tc_index; /* traffic control index */ 1035 #endif 1036 1037 u16 alloc_cpu; 1038 1039 union { 1040 __wsum csum; 1041 struct { 1042 __u16 csum_start; 1043 __u16 csum_offset; 1044 }; 1045 }; 1046 __u32 priority; 1047 int skb_iif; 1048 __u32 hash; 1049 union { 1050 u32 vlan_all; 1051 struct { 1052 __be16 vlan_proto; 1053 __u16 vlan_tci; 1054 }; 1055 }; 1056 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1057 union { 1058 unsigned int napi_id; 1059 unsigned int sender_cpu; 1060 }; 1061 #endif 1062 #ifdef CONFIG_NETWORK_SECMARK 1063 __u32 secmark; 1064 #endif 1065 1066 union { 1067 __u32 mark; 1068 __u32 reserved_tailroom; 1069 }; 1070 1071 union { 1072 __be16 inner_protocol; 1073 __u8 inner_ipproto; 1074 }; 1075 1076 __u16 inner_transport_header; 1077 __u16 inner_network_header; 1078 __u16 inner_mac_header; 1079 1080 __be16 protocol; 1081 __u16 transport_header; 1082 __u16 network_header; 1083 __u16 mac_header; 1084 1085 #ifdef CONFIG_KCOV 1086 u64 kcov_handle; 1087 #endif 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(struct sock *sk, struct sk_buff *skb, 1317 enum skb_drop_reason reason); 1318 1319 static inline void 1320 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason) 1321 { 1322 sk_skb_reason_drop(NULL, skb, reason); 1323 } 1324 1325 /** 1326 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1327 * @skb: buffer to free 1328 */ 1329 static inline void kfree_skb(struct sk_buff *skb) 1330 { 1331 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1332 } 1333 1334 void skb_release_head_state(struct sk_buff *skb); 1335 void kfree_skb_list_reason(struct sk_buff *segs, 1336 enum skb_drop_reason reason); 1337 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1338 void skb_tx_error(struct sk_buff *skb); 1339 1340 static inline void kfree_skb_list(struct sk_buff *segs) 1341 { 1342 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1343 } 1344 1345 #ifdef CONFIG_TRACEPOINTS 1346 void consume_skb(struct sk_buff *skb); 1347 #else 1348 static inline void consume_skb(struct sk_buff *skb) 1349 { 1350 return kfree_skb(skb); 1351 } 1352 #endif 1353 1354 void __consume_stateless_skb(struct sk_buff *skb); 1355 void __kfree_skb(struct sk_buff *skb); 1356 1357 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1358 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1359 bool *fragstolen, int *delta_truesize); 1360 1361 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1362 int node); 1363 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1364 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1365 struct sk_buff *build_skb_around(struct sk_buff *skb, 1366 void *data, unsigned int frag_size); 1367 void skb_attempt_defer_free(struct sk_buff *skb); 1368 1369 u32 napi_skb_cache_get_bulk(void **skbs, u32 n); 1370 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1371 struct sk_buff *slab_build_skb(void *data); 1372 1373 /** 1374 * alloc_skb - allocate a network buffer 1375 * @size: size to allocate 1376 * @priority: allocation mask 1377 * 1378 * This function is a convenient wrapper around __alloc_skb(). 1379 */ 1380 static inline struct sk_buff *alloc_skb(unsigned int size, 1381 gfp_t priority) 1382 { 1383 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1384 } 1385 1386 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1387 unsigned long data_len, 1388 int max_page_order, 1389 int *errcode, 1390 gfp_t gfp_mask); 1391 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1392 1393 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1394 struct sk_buff_fclones { 1395 struct sk_buff skb1; 1396 1397 struct sk_buff skb2; 1398 1399 refcount_t fclone_ref; 1400 }; 1401 1402 /** 1403 * skb_fclone_busy - check if fclone is busy 1404 * @sk: socket 1405 * @skb: buffer 1406 * 1407 * Returns: true if skb is a fast clone, and its clone is not freed. 1408 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1409 * so we also check that didn't happen. 1410 */ 1411 static inline bool skb_fclone_busy(const struct sock *sk, 1412 const struct sk_buff *skb) 1413 { 1414 const struct sk_buff_fclones *fclones; 1415 1416 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1417 1418 return skb->fclone == SKB_FCLONE_ORIG && 1419 refcount_read(&fclones->fclone_ref) > 1 && 1420 READ_ONCE(fclones->skb2.sk) == sk; 1421 } 1422 1423 /** 1424 * alloc_skb_fclone - allocate a network buffer from fclone cache 1425 * @size: size to allocate 1426 * @priority: allocation mask 1427 * 1428 * This function is a convenient wrapper around __alloc_skb(). 1429 */ 1430 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1431 gfp_t priority) 1432 { 1433 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1434 } 1435 1436 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1437 void skb_headers_offset_update(struct sk_buff *skb, int off); 1438 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1439 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1440 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1441 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1442 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1443 gfp_t gfp_mask, bool fclone); 1444 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1445 gfp_t gfp_mask) 1446 { 1447 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1448 } 1449 1450 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1451 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1452 unsigned int headroom); 1453 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1454 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1455 int newtailroom, gfp_t priority); 1456 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1457 int offset, int len); 1458 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1459 int offset, int len); 1460 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1461 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1462 1463 /** 1464 * skb_pad - zero pad the tail of an skb 1465 * @skb: buffer to pad 1466 * @pad: space to pad 1467 * 1468 * Ensure that a buffer is followed by a padding area that is zero 1469 * filled. Used by network drivers which may DMA or transfer data 1470 * beyond the buffer end onto the wire. 1471 * 1472 * May return error in out of memory cases. The skb is freed on error. 1473 */ 1474 static inline int skb_pad(struct sk_buff *skb, int pad) 1475 { 1476 return __skb_pad(skb, pad, true); 1477 } 1478 #define dev_kfree_skb(a) consume_skb(a) 1479 1480 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1481 int offset, size_t size, size_t max_frags); 1482 1483 struct skb_seq_state { 1484 __u32 lower_offset; 1485 __u32 upper_offset; 1486 __u32 frag_idx; 1487 __u32 stepped_offset; 1488 struct sk_buff *root_skb; 1489 struct sk_buff *cur_skb; 1490 __u8 *frag_data; 1491 __u32 frag_off; 1492 }; 1493 1494 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1495 unsigned int to, struct skb_seq_state *st); 1496 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1497 struct skb_seq_state *st); 1498 void skb_abort_seq_read(struct skb_seq_state *st); 1499 int skb_copy_seq_read(struct skb_seq_state *st, int offset, void *to, int len); 1500 1501 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1502 unsigned int to, struct ts_config *config); 1503 1504 /* 1505 * Packet hash types specify the type of hash in skb_set_hash. 1506 * 1507 * Hash types refer to the protocol layer addresses which are used to 1508 * construct a packet's hash. The hashes are used to differentiate or identify 1509 * flows of the protocol layer for the hash type. Hash types are either 1510 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1511 * 1512 * Properties of hashes: 1513 * 1514 * 1) Two packets in different flows have different hash values 1515 * 2) Two packets in the same flow should have the same hash value 1516 * 1517 * A hash at a higher layer is considered to be more specific. A driver should 1518 * set the most specific hash possible. 1519 * 1520 * A driver cannot indicate a more specific hash than the layer at which a hash 1521 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1522 * 1523 * A driver may indicate a hash level which is less specific than the 1524 * actual layer the hash was computed on. For instance, a hash computed 1525 * at L4 may be considered an L3 hash. This should only be done if the 1526 * driver can't unambiguously determine that the HW computed the hash at 1527 * the higher layer. Note that the "should" in the second property above 1528 * permits this. 1529 */ 1530 enum pkt_hash_types { 1531 PKT_HASH_TYPE_NONE, /* Undefined type */ 1532 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1533 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1534 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1535 }; 1536 1537 static inline void skb_clear_hash(struct sk_buff *skb) 1538 { 1539 skb->hash = 0; 1540 skb->sw_hash = 0; 1541 skb->l4_hash = 0; 1542 } 1543 1544 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1545 { 1546 if (!skb->l4_hash) 1547 skb_clear_hash(skb); 1548 } 1549 1550 static inline void 1551 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1552 { 1553 skb->l4_hash = is_l4; 1554 skb->sw_hash = is_sw; 1555 skb->hash = hash; 1556 } 1557 1558 static inline void 1559 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1560 { 1561 /* Used by drivers to set hash from HW */ 1562 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1563 } 1564 1565 static inline void 1566 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1567 { 1568 __skb_set_hash(skb, hash, true, is_l4); 1569 } 1570 1571 u32 __skb_get_hash_symmetric_net(const struct net *net, const struct sk_buff *skb); 1572 1573 static inline u32 __skb_get_hash_symmetric(const struct sk_buff *skb) 1574 { 1575 return __skb_get_hash_symmetric_net(NULL, skb); 1576 } 1577 1578 void __skb_get_hash_net(const struct net *net, struct sk_buff *skb); 1579 u32 skb_get_poff(const struct sk_buff *skb); 1580 u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1581 const struct flow_keys_basic *keys, int hlen); 1582 __be32 skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1583 const void *data, int hlen_proto); 1584 1585 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1586 const struct flow_dissector_key *key, 1587 unsigned int key_count); 1588 1589 struct bpf_flow_dissector; 1590 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1591 __be16 proto, int nhoff, int hlen, unsigned int flags); 1592 1593 bool __skb_flow_dissect(const struct net *net, 1594 const struct sk_buff *skb, 1595 struct flow_dissector *flow_dissector, 1596 void *target_container, const void *data, 1597 __be16 proto, int nhoff, int hlen, unsigned int flags); 1598 1599 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1600 struct flow_dissector *flow_dissector, 1601 void *target_container, unsigned int flags) 1602 { 1603 return __skb_flow_dissect(NULL, skb, flow_dissector, 1604 target_container, NULL, 0, 0, 0, flags); 1605 } 1606 1607 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1608 struct flow_keys *flow, 1609 unsigned int flags) 1610 { 1611 memset(flow, 0, sizeof(*flow)); 1612 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1613 flow, NULL, 0, 0, 0, flags); 1614 } 1615 1616 static inline bool 1617 skb_flow_dissect_flow_keys_basic(const struct net *net, 1618 const struct sk_buff *skb, 1619 struct flow_keys_basic *flow, 1620 const void *data, __be16 proto, 1621 int nhoff, int hlen, unsigned int flags) 1622 { 1623 memset(flow, 0, sizeof(*flow)); 1624 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1625 data, proto, nhoff, hlen, flags); 1626 } 1627 1628 void skb_flow_dissect_meta(const struct sk_buff *skb, 1629 struct flow_dissector *flow_dissector, 1630 void *target_container); 1631 1632 /* Gets a skb connection tracking info, ctinfo map should be a 1633 * map of mapsize to translate enum ip_conntrack_info states 1634 * to user states. 1635 */ 1636 void 1637 skb_flow_dissect_ct(const struct sk_buff *skb, 1638 struct flow_dissector *flow_dissector, 1639 void *target_container, 1640 u16 *ctinfo_map, size_t mapsize, 1641 bool post_ct, u16 zone); 1642 void 1643 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1644 struct flow_dissector *flow_dissector, 1645 void *target_container); 1646 1647 void skb_flow_dissect_hash(const struct sk_buff *skb, 1648 struct flow_dissector *flow_dissector, 1649 void *target_container); 1650 1651 static inline __u32 skb_get_hash_net(const struct net *net, struct sk_buff *skb) 1652 { 1653 if (!skb->l4_hash && !skb->sw_hash) 1654 __skb_get_hash_net(net, skb); 1655 1656 return skb->hash; 1657 } 1658 1659 static inline __u32 skb_get_hash(struct sk_buff *skb) 1660 { 1661 if (!skb->l4_hash && !skb->sw_hash) 1662 __skb_get_hash_net(NULL, skb); 1663 1664 return skb->hash; 1665 } 1666 1667 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1668 { 1669 if (!skb->l4_hash && !skb->sw_hash) { 1670 struct flow_keys keys; 1671 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1672 1673 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1674 } 1675 1676 return skb->hash; 1677 } 1678 1679 __u32 skb_get_hash_perturb(const struct sk_buff *skb, 1680 const siphash_key_t *perturb); 1681 1682 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1683 { 1684 return skb->hash; 1685 } 1686 1687 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1688 { 1689 to->hash = from->hash; 1690 to->sw_hash = from->sw_hash; 1691 to->l4_hash = from->l4_hash; 1692 }; 1693 1694 static inline int skb_cmp_decrypted(const struct sk_buff *skb1, 1695 const struct sk_buff *skb2) 1696 { 1697 #ifdef CONFIG_SKB_DECRYPTED 1698 return skb2->decrypted - skb1->decrypted; 1699 #else 1700 return 0; 1701 #endif 1702 } 1703 1704 static inline bool skb_is_decrypted(const struct sk_buff *skb) 1705 { 1706 #ifdef CONFIG_SKB_DECRYPTED 1707 return skb->decrypted; 1708 #else 1709 return false; 1710 #endif 1711 } 1712 1713 static inline void skb_copy_decrypted(struct sk_buff *to, 1714 const struct sk_buff *from) 1715 { 1716 #ifdef CONFIG_SKB_DECRYPTED 1717 to->decrypted = from->decrypted; 1718 #endif 1719 } 1720 1721 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1722 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1723 { 1724 return skb->head + skb->end; 1725 } 1726 1727 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1728 { 1729 return skb->end; 1730 } 1731 1732 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1733 { 1734 skb->end = offset; 1735 } 1736 #else 1737 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1738 { 1739 return skb->end; 1740 } 1741 1742 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1743 { 1744 return skb->end - skb->head; 1745 } 1746 1747 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1748 { 1749 skb->end = skb->head + offset; 1750 } 1751 #endif 1752 1753 extern const struct ubuf_info_ops msg_zerocopy_ubuf_ops; 1754 1755 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1756 struct ubuf_info *uarg, bool devmem); 1757 1758 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1759 1760 struct net_devmem_dmabuf_binding; 1761 1762 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, 1763 struct sk_buff *skb, struct iov_iter *from, 1764 size_t length, 1765 struct net_devmem_dmabuf_binding *binding); 1766 1767 int zerocopy_fill_skb_from_iter(struct sk_buff *skb, 1768 struct iov_iter *from, size_t length); 1769 1770 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1771 struct msghdr *msg, int len) 1772 { 1773 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len, 1774 NULL); 1775 } 1776 1777 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1778 struct msghdr *msg, int len, 1779 struct ubuf_info *uarg, 1780 struct net_devmem_dmabuf_binding *binding); 1781 1782 /* Internal */ 1783 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1784 1785 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1786 { 1787 return &skb_shinfo(skb)->hwtstamps; 1788 } 1789 1790 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1791 { 1792 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1793 1794 return is_zcopy ? skb_uarg(skb) : NULL; 1795 } 1796 1797 static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1798 { 1799 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1800 } 1801 1802 static inline bool skb_zcopy_managed(const struct sk_buff *skb) 1803 { 1804 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS; 1805 } 1806 1807 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1808 const struct sk_buff *skb2) 1809 { 1810 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1811 } 1812 1813 static inline void net_zcopy_get(struct ubuf_info *uarg) 1814 { 1815 refcount_inc(&uarg->refcnt); 1816 } 1817 1818 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1819 { 1820 skb_shinfo(skb)->destructor_arg = uarg; 1821 skb_shinfo(skb)->flags |= uarg->flags; 1822 } 1823 1824 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1825 bool *have_ref) 1826 { 1827 if (skb && uarg && !skb_zcopy(skb)) { 1828 if (unlikely(have_ref && *have_ref)) 1829 *have_ref = false; 1830 else 1831 net_zcopy_get(uarg); 1832 skb_zcopy_init(skb, uarg); 1833 } 1834 } 1835 1836 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1837 { 1838 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1839 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1840 } 1841 1842 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1843 { 1844 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1845 } 1846 1847 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1848 { 1849 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1850 } 1851 1852 static inline void net_zcopy_put(struct ubuf_info *uarg) 1853 { 1854 if (uarg) 1855 uarg->ops->complete(NULL, uarg, true); 1856 } 1857 1858 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1859 { 1860 if (uarg) { 1861 if (uarg->ops == &msg_zerocopy_ubuf_ops) 1862 msg_zerocopy_put_abort(uarg, have_uref); 1863 else if (have_uref) 1864 net_zcopy_put(uarg); 1865 } 1866 } 1867 1868 /* Release a reference on a zerocopy structure */ 1869 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1870 { 1871 struct ubuf_info *uarg = skb_zcopy(skb); 1872 1873 if (uarg) { 1874 if (!skb_zcopy_is_nouarg(skb)) 1875 uarg->ops->complete(skb, uarg, zerocopy_success); 1876 1877 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1878 } 1879 } 1880 1881 void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1882 1883 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1884 { 1885 if (unlikely(skb_zcopy_managed(skb))) 1886 __skb_zcopy_downgrade_managed(skb); 1887 } 1888 1889 /* Return true if frags in this skb are readable by the host. */ 1890 static inline bool skb_frags_readable(const struct sk_buff *skb) 1891 { 1892 return !skb->unreadable; 1893 } 1894 1895 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1896 { 1897 skb->next = NULL; 1898 } 1899 1900 static inline void skb_poison_list(struct sk_buff *skb) 1901 { 1902 #ifdef CONFIG_DEBUG_NET 1903 skb->next = SKB_LIST_POISON_NEXT; 1904 #endif 1905 } 1906 1907 /* Iterate through singly-linked GSO fragments of an skb. */ 1908 #define skb_list_walk_safe(first, skb, next_skb) \ 1909 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1910 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1911 1912 static inline void skb_list_del_init(struct sk_buff *skb) 1913 { 1914 __list_del_entry(&skb->list); 1915 skb_mark_not_on_list(skb); 1916 } 1917 1918 /** 1919 * skb_queue_empty - check if a queue is empty 1920 * @list: queue head 1921 * 1922 * Returns true if the queue is empty, false otherwise. 1923 */ 1924 static inline int skb_queue_empty(const struct sk_buff_head *list) 1925 { 1926 return list->next == (const struct sk_buff *) list; 1927 } 1928 1929 /** 1930 * skb_queue_empty_lockless - check if a queue is empty 1931 * @list: queue head 1932 * 1933 * Returns true if the queue is empty, false otherwise. 1934 * This variant can be used in lockless contexts. 1935 */ 1936 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1937 { 1938 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1939 } 1940 1941 1942 /** 1943 * skb_queue_is_last - check if skb is the last entry in the queue 1944 * @list: queue head 1945 * @skb: buffer 1946 * 1947 * Returns true if @skb is the last buffer on the list. 1948 */ 1949 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1950 const struct sk_buff *skb) 1951 { 1952 return skb->next == (const struct sk_buff *) list; 1953 } 1954 1955 /** 1956 * skb_queue_is_first - check if skb is the first entry in the queue 1957 * @list: queue head 1958 * @skb: buffer 1959 * 1960 * Returns true if @skb is the first buffer on the list. 1961 */ 1962 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1963 const struct sk_buff *skb) 1964 { 1965 return skb->prev == (const struct sk_buff *) list; 1966 } 1967 1968 /** 1969 * skb_queue_next - return the next packet in the queue 1970 * @list: queue head 1971 * @skb: current buffer 1972 * 1973 * Return the next packet in @list after @skb. It is only valid to 1974 * call this if skb_queue_is_last() evaluates to false. 1975 */ 1976 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1977 const struct sk_buff *skb) 1978 { 1979 /* This BUG_ON may seem severe, but if we just return then we 1980 * are going to dereference garbage. 1981 */ 1982 BUG_ON(skb_queue_is_last(list, skb)); 1983 return skb->next; 1984 } 1985 1986 /** 1987 * skb_queue_prev - return the prev packet in the queue 1988 * @list: queue head 1989 * @skb: current buffer 1990 * 1991 * Return the prev packet in @list before @skb. It is only valid to 1992 * call this if skb_queue_is_first() evaluates to false. 1993 */ 1994 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1995 const struct sk_buff *skb) 1996 { 1997 /* This BUG_ON may seem severe, but if we just return then we 1998 * are going to dereference garbage. 1999 */ 2000 BUG_ON(skb_queue_is_first(list, skb)); 2001 return skb->prev; 2002 } 2003 2004 /** 2005 * skb_get - reference buffer 2006 * @skb: buffer to reference 2007 * 2008 * Makes another reference to a socket buffer and returns a pointer 2009 * to the buffer. 2010 */ 2011 static inline struct sk_buff *skb_get(struct sk_buff *skb) 2012 { 2013 refcount_inc(&skb->users); 2014 return skb; 2015 } 2016 2017 /* 2018 * If users == 1, we are the only owner and can avoid redundant atomic changes. 2019 */ 2020 2021 /** 2022 * skb_cloned - is the buffer a clone 2023 * @skb: buffer to check 2024 * 2025 * Returns true if the buffer was generated with skb_clone() and is 2026 * one of multiple shared copies of the buffer. Cloned buffers are 2027 * shared data so must not be written to under normal circumstances. 2028 */ 2029 static inline int skb_cloned(const struct sk_buff *skb) 2030 { 2031 return skb->cloned && 2032 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 2033 } 2034 2035 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 2036 { 2037 might_sleep_if(gfpflags_allow_blocking(pri)); 2038 2039 if (skb_cloned(skb)) 2040 return pskb_expand_head(skb, 0, 0, pri); 2041 2042 return 0; 2043 } 2044 2045 /* This variant of skb_unclone() makes sure skb->truesize 2046 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 2047 * 2048 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 2049 * when various debugging features are in place. 2050 */ 2051 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 2052 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 2053 { 2054 might_sleep_if(gfpflags_allow_blocking(pri)); 2055 2056 if (skb_cloned(skb)) 2057 return __skb_unclone_keeptruesize(skb, pri); 2058 return 0; 2059 } 2060 2061 /** 2062 * skb_header_cloned - is the header a clone 2063 * @skb: buffer to check 2064 * 2065 * Returns true if modifying the header part of the buffer requires 2066 * the data to be copied. 2067 */ 2068 static inline int skb_header_cloned(const struct sk_buff *skb) 2069 { 2070 int dataref; 2071 2072 if (!skb->cloned) 2073 return 0; 2074 2075 dataref = atomic_read(&skb_shinfo(skb)->dataref); 2076 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 2077 return dataref != 1; 2078 } 2079 2080 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 2081 { 2082 might_sleep_if(gfpflags_allow_blocking(pri)); 2083 2084 if (skb_header_cloned(skb)) 2085 return pskb_expand_head(skb, 0, 0, pri); 2086 2087 return 0; 2088 } 2089 2090 /** 2091 * __skb_header_release() - allow clones to use the headroom 2092 * @skb: buffer to operate on 2093 * 2094 * See "DOC: dataref and headerless skbs". 2095 */ 2096 static inline void __skb_header_release(struct sk_buff *skb) 2097 { 2098 skb->nohdr = 1; 2099 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 2100 } 2101 2102 2103 /** 2104 * skb_shared - is the buffer shared 2105 * @skb: buffer to check 2106 * 2107 * Returns true if more than one person has a reference to this 2108 * buffer. 2109 */ 2110 static inline int skb_shared(const struct sk_buff *skb) 2111 { 2112 return refcount_read(&skb->users) != 1; 2113 } 2114 2115 /** 2116 * skb_share_check - check if buffer is shared and if so clone it 2117 * @skb: buffer to check 2118 * @pri: priority for memory allocation 2119 * 2120 * If the buffer is shared the buffer is cloned and the old copy 2121 * drops a reference. A new clone with a single reference is returned. 2122 * If the buffer is not shared the original buffer is returned. When 2123 * being called from interrupt status or with spinlocks held pri must 2124 * be GFP_ATOMIC. 2125 * 2126 * NULL is returned on a memory allocation failure. 2127 */ 2128 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 2129 { 2130 might_sleep_if(gfpflags_allow_blocking(pri)); 2131 if (skb_shared(skb)) { 2132 struct sk_buff *nskb = skb_clone(skb, pri); 2133 2134 if (likely(nskb)) 2135 consume_skb(skb); 2136 else 2137 kfree_skb(skb); 2138 skb = nskb; 2139 } 2140 return skb; 2141 } 2142 2143 /* 2144 * Copy shared buffers into a new sk_buff. We effectively do COW on 2145 * packets to handle cases where we have a local reader and forward 2146 * and a couple of other messy ones. The normal one is tcpdumping 2147 * a packet that's being forwarded. 2148 */ 2149 2150 /** 2151 * skb_unshare - make a copy of a shared buffer 2152 * @skb: buffer to check 2153 * @pri: priority for memory allocation 2154 * 2155 * If the socket buffer is a clone then this function creates a new 2156 * copy of the data, drops a reference count on the old copy and returns 2157 * the new copy with the reference count at 1. If the buffer is not a clone 2158 * the original buffer is returned. When called with a spinlock held or 2159 * from interrupt state @pri must be %GFP_ATOMIC 2160 * 2161 * %NULL is returned on a memory allocation failure. 2162 */ 2163 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2164 gfp_t pri) 2165 { 2166 might_sleep_if(gfpflags_allow_blocking(pri)); 2167 if (skb_cloned(skb)) { 2168 struct sk_buff *nskb = skb_copy(skb, pri); 2169 2170 /* Free our shared copy */ 2171 if (likely(nskb)) 2172 consume_skb(skb); 2173 else 2174 kfree_skb(skb); 2175 skb = nskb; 2176 } 2177 return skb; 2178 } 2179 2180 /** 2181 * skb_peek - peek at the head of an &sk_buff_head 2182 * @list_: list to peek at 2183 * 2184 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2185 * be careful with this one. A peek leaves the buffer on the 2186 * list and someone else may run off with it. You must hold 2187 * the appropriate locks or have a private queue to do this. 2188 * 2189 * Returns %NULL for an empty list or a pointer to the head element. 2190 * The reference count is not incremented and the reference is therefore 2191 * volatile. Use with caution. 2192 */ 2193 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2194 { 2195 struct sk_buff *skb = list_->next; 2196 2197 if (skb == (struct sk_buff *)list_) 2198 skb = NULL; 2199 return skb; 2200 } 2201 2202 /** 2203 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2204 * @list_: list to peek at 2205 * 2206 * Like skb_peek(), but the caller knows that the list is not empty. 2207 */ 2208 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2209 { 2210 return list_->next; 2211 } 2212 2213 /** 2214 * skb_peek_next - peek skb following the given one from a queue 2215 * @skb: skb to start from 2216 * @list_: list to peek at 2217 * 2218 * Returns %NULL when the end of the list is met or a pointer to the 2219 * next element. The reference count is not incremented and the 2220 * reference is therefore volatile. Use with caution. 2221 */ 2222 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2223 const struct sk_buff_head *list_) 2224 { 2225 struct sk_buff *next = skb->next; 2226 2227 if (next == (struct sk_buff *)list_) 2228 next = NULL; 2229 return next; 2230 } 2231 2232 /** 2233 * skb_peek_tail - peek at the tail of an &sk_buff_head 2234 * @list_: list to peek at 2235 * 2236 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2237 * be careful with this one. A peek leaves the buffer on the 2238 * list and someone else may run off with it. You must hold 2239 * the appropriate locks or have a private queue to do this. 2240 * 2241 * Returns %NULL for an empty list or a pointer to the tail element. 2242 * The reference count is not incremented and the reference is therefore 2243 * volatile. Use with caution. 2244 */ 2245 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2246 { 2247 struct sk_buff *skb = READ_ONCE(list_->prev); 2248 2249 if (skb == (struct sk_buff *)list_) 2250 skb = NULL; 2251 return skb; 2252 2253 } 2254 2255 /** 2256 * skb_queue_len - get queue length 2257 * @list_: list to measure 2258 * 2259 * Return the length of an &sk_buff queue. 2260 */ 2261 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2262 { 2263 return list_->qlen; 2264 } 2265 2266 /** 2267 * skb_queue_len_lockless - get queue length 2268 * @list_: list to measure 2269 * 2270 * Return the length of an &sk_buff queue. 2271 * This variant can be used in lockless contexts. 2272 */ 2273 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2274 { 2275 return READ_ONCE(list_->qlen); 2276 } 2277 2278 /** 2279 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2280 * @list: queue to initialize 2281 * 2282 * This initializes only the list and queue length aspects of 2283 * an sk_buff_head object. This allows to initialize the list 2284 * aspects of an sk_buff_head without reinitializing things like 2285 * the spinlock. It can also be used for on-stack sk_buff_head 2286 * objects where the spinlock is known to not be used. 2287 */ 2288 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2289 { 2290 list->prev = list->next = (struct sk_buff *)list; 2291 list->qlen = 0; 2292 } 2293 2294 /* 2295 * This function creates a split out lock class for each invocation; 2296 * this is needed for now since a whole lot of users of the skb-queue 2297 * infrastructure in drivers have different locking usage (in hardirq) 2298 * than the networking core (in softirq only). In the long run either the 2299 * network layer or drivers should need annotation to consolidate the 2300 * main types of usage into 3 classes. 2301 */ 2302 static inline void skb_queue_head_init(struct sk_buff_head *list) 2303 { 2304 spin_lock_init(&list->lock); 2305 __skb_queue_head_init(list); 2306 } 2307 2308 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2309 struct lock_class_key *class) 2310 { 2311 skb_queue_head_init(list); 2312 lockdep_set_class(&list->lock, class); 2313 } 2314 2315 /* 2316 * Insert an sk_buff on a list. 2317 * 2318 * The "__skb_xxxx()" functions are the non-atomic ones that 2319 * can only be called with interrupts disabled. 2320 */ 2321 static inline void __skb_insert(struct sk_buff *newsk, 2322 struct sk_buff *prev, struct sk_buff *next, 2323 struct sk_buff_head *list) 2324 { 2325 /* See skb_queue_empty_lockless() and skb_peek_tail() 2326 * for the opposite READ_ONCE() 2327 */ 2328 WRITE_ONCE(newsk->next, next); 2329 WRITE_ONCE(newsk->prev, prev); 2330 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2331 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2332 WRITE_ONCE(list->qlen, list->qlen + 1); 2333 } 2334 2335 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2336 struct sk_buff *prev, 2337 struct sk_buff *next) 2338 { 2339 struct sk_buff *first = list->next; 2340 struct sk_buff *last = list->prev; 2341 2342 WRITE_ONCE(first->prev, prev); 2343 WRITE_ONCE(prev->next, first); 2344 2345 WRITE_ONCE(last->next, next); 2346 WRITE_ONCE(next->prev, last); 2347 } 2348 2349 /** 2350 * skb_queue_splice - join two skb lists, this is designed for stacks 2351 * @list: the new list to add 2352 * @head: the place to add it in the first list 2353 */ 2354 static inline void skb_queue_splice(const struct sk_buff_head *list, 2355 struct sk_buff_head *head) 2356 { 2357 if (!skb_queue_empty(list)) { 2358 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2359 head->qlen += list->qlen; 2360 } 2361 } 2362 2363 /** 2364 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2365 * @list: the new list to add 2366 * @head: the place to add it in the first list 2367 * 2368 * The list at @list is reinitialised 2369 */ 2370 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2371 struct sk_buff_head *head) 2372 { 2373 if (!skb_queue_empty(list)) { 2374 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2375 head->qlen += list->qlen; 2376 __skb_queue_head_init(list); 2377 } 2378 } 2379 2380 /** 2381 * skb_queue_splice_tail - join two skb lists, each list being a queue 2382 * @list: the new list to add 2383 * @head: the place to add it in the first list 2384 */ 2385 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2386 struct sk_buff_head *head) 2387 { 2388 if (!skb_queue_empty(list)) { 2389 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2390 head->qlen += list->qlen; 2391 } 2392 } 2393 2394 /** 2395 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2396 * @list: the new list to add 2397 * @head: the place to add it in the first list 2398 * 2399 * Each of the lists is a queue. 2400 * The list at @list is reinitialised 2401 */ 2402 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2403 struct sk_buff_head *head) 2404 { 2405 if (!skb_queue_empty(list)) { 2406 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2407 head->qlen += list->qlen; 2408 __skb_queue_head_init(list); 2409 } 2410 } 2411 2412 /** 2413 * __skb_queue_after - queue a buffer at the list head 2414 * @list: list to use 2415 * @prev: place after this buffer 2416 * @newsk: buffer to queue 2417 * 2418 * Queue a buffer int the middle of a list. This function takes no locks 2419 * and you must therefore hold required locks before calling it. 2420 * 2421 * A buffer cannot be placed on two lists at the same time. 2422 */ 2423 static inline void __skb_queue_after(struct sk_buff_head *list, 2424 struct sk_buff *prev, 2425 struct sk_buff *newsk) 2426 { 2427 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2428 } 2429 2430 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2431 struct sk_buff_head *list); 2432 2433 static inline void __skb_queue_before(struct sk_buff_head *list, 2434 struct sk_buff *next, 2435 struct sk_buff *newsk) 2436 { 2437 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2438 } 2439 2440 /** 2441 * __skb_queue_head - queue a buffer at the list head 2442 * @list: list to use 2443 * @newsk: buffer to queue 2444 * 2445 * Queue a buffer at the start of a list. This function takes no locks 2446 * and you must therefore hold required locks before calling it. 2447 * 2448 * A buffer cannot be placed on two lists at the same time. 2449 */ 2450 static inline void __skb_queue_head(struct sk_buff_head *list, 2451 struct sk_buff *newsk) 2452 { 2453 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2454 } 2455 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2456 2457 /** 2458 * __skb_queue_tail - queue a buffer at the list tail 2459 * @list: list to use 2460 * @newsk: buffer to queue 2461 * 2462 * Queue a buffer at the end of a list. This function takes no locks 2463 * and you must therefore hold required locks before calling it. 2464 * 2465 * A buffer cannot be placed on two lists at the same time. 2466 */ 2467 static inline void __skb_queue_tail(struct sk_buff_head *list, 2468 struct sk_buff *newsk) 2469 { 2470 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2471 } 2472 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2473 2474 /* 2475 * remove sk_buff from list. _Must_ be called atomically, and with 2476 * the list known.. 2477 */ 2478 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2479 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2480 { 2481 struct sk_buff *next, *prev; 2482 2483 WRITE_ONCE(list->qlen, list->qlen - 1); 2484 next = skb->next; 2485 prev = skb->prev; 2486 skb->next = skb->prev = NULL; 2487 WRITE_ONCE(next->prev, prev); 2488 WRITE_ONCE(prev->next, next); 2489 } 2490 2491 /** 2492 * __skb_dequeue - remove from the head of the queue 2493 * @list: list to dequeue from 2494 * 2495 * Remove the head of the list. This function does not take any locks 2496 * so must be used with appropriate locks held only. The head item is 2497 * returned or %NULL if the list is empty. 2498 */ 2499 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2500 { 2501 struct sk_buff *skb = skb_peek(list); 2502 if (skb) 2503 __skb_unlink(skb, list); 2504 return skb; 2505 } 2506 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2507 2508 /** 2509 * __skb_dequeue_tail - remove from the tail of the queue 2510 * @list: list to dequeue from 2511 * 2512 * Remove the tail of the list. This function does not take any locks 2513 * so must be used with appropriate locks held only. The tail item is 2514 * returned or %NULL if the list is empty. 2515 */ 2516 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2517 { 2518 struct sk_buff *skb = skb_peek_tail(list); 2519 if (skb) 2520 __skb_unlink(skb, list); 2521 return skb; 2522 } 2523 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2524 2525 2526 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2527 { 2528 return skb->data_len; 2529 } 2530 2531 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2532 { 2533 return skb->len - skb->data_len; 2534 } 2535 2536 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2537 { 2538 unsigned int i, len = 0; 2539 2540 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2541 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2542 return len; 2543 } 2544 2545 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2546 { 2547 return skb_headlen(skb) + __skb_pagelen(skb); 2548 } 2549 2550 static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag, 2551 netmem_ref netmem, int off, 2552 int size) 2553 { 2554 frag->netmem = netmem; 2555 frag->offset = off; 2556 skb_frag_size_set(frag, size); 2557 } 2558 2559 static inline void skb_frag_fill_page_desc(skb_frag_t *frag, 2560 struct page *page, 2561 int off, int size) 2562 { 2563 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size); 2564 } 2565 2566 static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo, 2567 int i, netmem_ref netmem, 2568 int off, int size) 2569 { 2570 skb_frag_t *frag = &shinfo->frags[i]; 2571 2572 skb_frag_fill_netmem_desc(frag, netmem, off, size); 2573 } 2574 2575 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2576 int i, struct page *page, 2577 int off, int size) 2578 { 2579 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off, 2580 size); 2581 } 2582 2583 /** 2584 * skb_len_add - adds a number to len fields of skb 2585 * @skb: buffer to add len to 2586 * @delta: number of bytes to add 2587 */ 2588 static inline void skb_len_add(struct sk_buff *skb, int delta) 2589 { 2590 skb->len += delta; 2591 skb->data_len += delta; 2592 skb->truesize += delta; 2593 } 2594 2595 /** 2596 * __skb_fill_netmem_desc - initialise a fragment in an skb 2597 * @skb: buffer containing fragment to be initialised 2598 * @i: fragment index to initialise 2599 * @netmem: the netmem to use for this fragment 2600 * @off: the offset to the data with @page 2601 * @size: the length of the data 2602 * 2603 * Initialises the @i'th fragment of @skb to point to &size bytes at 2604 * offset @off within @page. 2605 * 2606 * Does not take any additional reference on the fragment. 2607 */ 2608 static inline void __skb_fill_netmem_desc(struct sk_buff *skb, int i, 2609 netmem_ref netmem, int off, int size) 2610 { 2611 struct page *page; 2612 2613 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size); 2614 2615 if (netmem_is_net_iov(netmem)) { 2616 skb->unreadable = true; 2617 return; 2618 } 2619 2620 page = netmem_to_page(netmem); 2621 2622 /* Propagate page pfmemalloc to the skb if we can. The problem is 2623 * that not all callers have unique ownership of the page but rely 2624 * on page_is_pfmemalloc doing the right thing(tm). 2625 */ 2626 page = compound_head(page); 2627 if (page_is_pfmemalloc(page)) 2628 skb->pfmemalloc = true; 2629 } 2630 2631 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2632 struct page *page, int off, int size) 2633 { 2634 __skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2635 } 2636 2637 static inline void skb_fill_netmem_desc(struct sk_buff *skb, int i, 2638 netmem_ref netmem, int off, int size) 2639 { 2640 __skb_fill_netmem_desc(skb, i, netmem, off, size); 2641 skb_shinfo(skb)->nr_frags = i + 1; 2642 } 2643 2644 /** 2645 * skb_fill_page_desc - initialise a paged fragment in an skb 2646 * @skb: buffer containing fragment to be initialised 2647 * @i: paged fragment index to initialise 2648 * @page: the page to use for this fragment 2649 * @off: the offset to the data with @page 2650 * @size: the length of the data 2651 * 2652 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2653 * @skb to point to @size bytes at offset @off within @page. In 2654 * addition updates @skb such that @i is the last fragment. 2655 * 2656 * Does not take any additional reference on the fragment. 2657 */ 2658 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2659 struct page *page, int off, int size) 2660 { 2661 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2662 } 2663 2664 /** 2665 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2666 * @skb: buffer containing fragment to be initialised 2667 * @i: paged fragment index to initialise 2668 * @page: the page to use for this fragment 2669 * @off: the offset to the data with @page 2670 * @size: the length of the data 2671 * 2672 * Variant of skb_fill_page_desc() which does not deal with 2673 * pfmemalloc, if page is not owned by us. 2674 */ 2675 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2676 struct page *page, int off, 2677 int size) 2678 { 2679 struct skb_shared_info *shinfo = skb_shinfo(skb); 2680 2681 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2682 shinfo->nr_frags = i + 1; 2683 } 2684 2685 void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem, 2686 int off, int size, unsigned int truesize); 2687 2688 static inline void skb_add_rx_frag(struct sk_buff *skb, int i, 2689 struct page *page, int off, int size, 2690 unsigned int truesize) 2691 { 2692 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size, 2693 truesize); 2694 } 2695 2696 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2697 unsigned int truesize); 2698 2699 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2700 2701 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2702 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2703 { 2704 return skb->head + skb->tail; 2705 } 2706 2707 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2708 { 2709 skb->tail = skb->data - skb->head; 2710 } 2711 2712 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2713 { 2714 skb_reset_tail_pointer(skb); 2715 skb->tail += offset; 2716 } 2717 2718 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2719 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2720 { 2721 return skb->tail; 2722 } 2723 2724 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2725 { 2726 skb->tail = skb->data; 2727 } 2728 2729 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2730 { 2731 skb->tail = skb->data + offset; 2732 } 2733 2734 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2735 2736 static inline void skb_assert_len(struct sk_buff *skb) 2737 { 2738 #ifdef CONFIG_DEBUG_NET 2739 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2740 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2741 #endif /* CONFIG_DEBUG_NET */ 2742 } 2743 2744 #if defined(CONFIG_FAIL_SKB_REALLOC) 2745 void skb_might_realloc(struct sk_buff *skb); 2746 #else 2747 static inline void skb_might_realloc(struct sk_buff *skb) {} 2748 #endif 2749 2750 /* 2751 * Add data to an sk_buff 2752 */ 2753 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2754 void *skb_put(struct sk_buff *skb, unsigned int len); 2755 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2756 { 2757 void *tmp = skb_tail_pointer(skb); 2758 SKB_LINEAR_ASSERT(skb); 2759 skb->tail += len; 2760 skb->len += len; 2761 return tmp; 2762 } 2763 2764 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2765 { 2766 void *tmp = __skb_put(skb, len); 2767 2768 memset(tmp, 0, len); 2769 return tmp; 2770 } 2771 2772 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2773 unsigned int len) 2774 { 2775 void *tmp = __skb_put(skb, len); 2776 2777 memcpy(tmp, data, len); 2778 return tmp; 2779 } 2780 2781 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2782 { 2783 *(u8 *)__skb_put(skb, 1) = val; 2784 } 2785 2786 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2787 { 2788 void *tmp = skb_put(skb, len); 2789 2790 memset(tmp, 0, len); 2791 2792 return tmp; 2793 } 2794 2795 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2796 unsigned int len) 2797 { 2798 void *tmp = skb_put(skb, len); 2799 2800 memcpy(tmp, data, len); 2801 2802 return tmp; 2803 } 2804 2805 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2806 { 2807 *(u8 *)skb_put(skb, 1) = val; 2808 } 2809 2810 void *skb_push(struct sk_buff *skb, unsigned int len); 2811 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2812 { 2813 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2814 2815 skb->data -= len; 2816 DEBUG_NET_WARN_ON_ONCE(skb->data < skb->head); 2817 skb->len += len; 2818 return skb->data; 2819 } 2820 2821 void *skb_pull(struct sk_buff *skb, unsigned int len); 2822 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2823 { 2824 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2825 2826 skb->len -= len; 2827 if (unlikely(skb->len < skb->data_len)) { 2828 #if defined(CONFIG_DEBUG_NET) 2829 skb->len += len; 2830 pr_err("__skb_pull(len=%u)\n", len); 2831 skb_dump(KERN_ERR, skb, false); 2832 #endif 2833 BUG(); 2834 } 2835 return skb->data += len; 2836 } 2837 2838 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2839 { 2840 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2841 } 2842 2843 void *skb_pull_data(struct sk_buff *skb, size_t len); 2844 2845 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2846 2847 static inline enum skb_drop_reason 2848 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len) 2849 { 2850 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2851 skb_might_realloc(skb); 2852 2853 if (likely(len <= skb_headlen(skb))) 2854 return SKB_NOT_DROPPED_YET; 2855 2856 if (unlikely(len > skb->len)) 2857 return SKB_DROP_REASON_PKT_TOO_SMALL; 2858 2859 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb)))) 2860 return SKB_DROP_REASON_NOMEM; 2861 2862 return SKB_NOT_DROPPED_YET; 2863 } 2864 2865 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2866 { 2867 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 2868 } 2869 2870 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2871 { 2872 if (!pskb_may_pull(skb, len)) 2873 return NULL; 2874 2875 skb->len -= len; 2876 return skb->data += len; 2877 } 2878 2879 void skb_condense(struct sk_buff *skb); 2880 2881 /** 2882 * skb_headroom - bytes at buffer head 2883 * @skb: buffer to check 2884 * 2885 * Return the number of bytes of free space at the head of an &sk_buff. 2886 */ 2887 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2888 { 2889 return skb->data - skb->head; 2890 } 2891 2892 /** 2893 * skb_tailroom - bytes at buffer end 2894 * @skb: buffer to check 2895 * 2896 * Return the number of bytes of free space at the tail of an sk_buff 2897 */ 2898 static inline int skb_tailroom(const struct sk_buff *skb) 2899 { 2900 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2901 } 2902 2903 /** 2904 * skb_availroom - bytes at buffer end 2905 * @skb: buffer to check 2906 * 2907 * Return the number of bytes of free space at the tail of an sk_buff 2908 * allocated by sk_stream_alloc() 2909 */ 2910 static inline int skb_availroom(const struct sk_buff *skb) 2911 { 2912 if (skb_is_nonlinear(skb)) 2913 return 0; 2914 2915 return skb->end - skb->tail - skb->reserved_tailroom; 2916 } 2917 2918 /** 2919 * skb_reserve - adjust headroom 2920 * @skb: buffer to alter 2921 * @len: bytes to move 2922 * 2923 * Increase the headroom of an empty &sk_buff by reducing the tail 2924 * room. This is only allowed for an empty buffer. 2925 */ 2926 static inline void skb_reserve(struct sk_buff *skb, int len) 2927 { 2928 skb->data += len; 2929 skb->tail += len; 2930 } 2931 2932 /** 2933 * skb_tailroom_reserve - adjust reserved_tailroom 2934 * @skb: buffer to alter 2935 * @mtu: maximum amount of headlen permitted 2936 * @needed_tailroom: minimum amount of reserved_tailroom 2937 * 2938 * Set reserved_tailroom so that headlen can be as large as possible but 2939 * not larger than mtu and tailroom cannot be smaller than 2940 * needed_tailroom. 2941 * The required headroom should already have been reserved before using 2942 * this function. 2943 */ 2944 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2945 unsigned int needed_tailroom) 2946 { 2947 SKB_LINEAR_ASSERT(skb); 2948 if (mtu < skb_tailroom(skb) - needed_tailroom) 2949 /* use at most mtu */ 2950 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2951 else 2952 /* use up to all available space */ 2953 skb->reserved_tailroom = needed_tailroom; 2954 } 2955 2956 #define ENCAP_TYPE_ETHER 0 2957 #define ENCAP_TYPE_IPPROTO 1 2958 2959 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2960 __be16 protocol) 2961 { 2962 skb->inner_protocol = protocol; 2963 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2964 } 2965 2966 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2967 __u8 ipproto) 2968 { 2969 skb->inner_ipproto = ipproto; 2970 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2971 } 2972 2973 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2974 { 2975 skb->inner_mac_header = skb->mac_header; 2976 skb->inner_network_header = skb->network_header; 2977 skb->inner_transport_header = skb->transport_header; 2978 } 2979 2980 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2981 { 2982 return skb->mac_header != (typeof(skb->mac_header))~0U; 2983 } 2984 2985 static inline void skb_reset_mac_len(struct sk_buff *skb) 2986 { 2987 if (!skb_mac_header_was_set(skb)) { 2988 DEBUG_NET_WARN_ON_ONCE(1); 2989 skb->mac_len = 0; 2990 } else { 2991 skb->mac_len = skb->network_header - skb->mac_header; 2992 } 2993 } 2994 2995 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2996 *skb) 2997 { 2998 return skb->head + skb->inner_transport_header; 2999 } 3000 3001 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 3002 { 3003 return skb_inner_transport_header(skb) - skb->data; 3004 } 3005 3006 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 3007 { 3008 long offset = skb->data - skb->head; 3009 3010 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_transport_header))offset); 3011 skb->inner_transport_header = offset; 3012 } 3013 3014 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 3015 const int offset) 3016 { 3017 skb_reset_inner_transport_header(skb); 3018 skb->inner_transport_header += offset; 3019 } 3020 3021 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 3022 { 3023 return skb->head + skb->inner_network_header; 3024 } 3025 3026 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 3027 { 3028 long offset = skb->data - skb->head; 3029 3030 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_network_header))offset); 3031 skb->inner_network_header = offset; 3032 } 3033 3034 static inline void skb_set_inner_network_header(struct sk_buff *skb, 3035 const int offset) 3036 { 3037 skb_reset_inner_network_header(skb); 3038 skb->inner_network_header += offset; 3039 } 3040 3041 static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb) 3042 { 3043 return skb->inner_network_header > 0; 3044 } 3045 3046 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 3047 { 3048 return skb->head + skb->inner_mac_header; 3049 } 3050 3051 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 3052 { 3053 long offset = skb->data - skb->head; 3054 3055 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_mac_header))offset); 3056 skb->inner_mac_header = offset; 3057 } 3058 3059 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 3060 const int offset) 3061 { 3062 skb_reset_inner_mac_header(skb); 3063 skb->inner_mac_header += offset; 3064 } 3065 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 3066 { 3067 return skb->transport_header != (typeof(skb->transport_header))~0U; 3068 } 3069 3070 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 3071 { 3072 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3073 return skb->head + skb->transport_header; 3074 } 3075 3076 static inline void skb_reset_transport_header(struct sk_buff *skb) 3077 { 3078 long offset = skb->data - skb->head; 3079 3080 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->transport_header))offset); 3081 skb->transport_header = offset; 3082 } 3083 3084 /** 3085 * skb_reset_transport_header_careful - conditionally reset transport header 3086 * @skb: buffer to alter 3087 * 3088 * Hardened version of skb_reset_transport_header(). 3089 * 3090 * Returns: true if the operation was a success. 3091 */ 3092 static inline bool __must_check 3093 skb_reset_transport_header_careful(struct sk_buff *skb) 3094 { 3095 long offset = skb->data - skb->head; 3096 3097 if (unlikely(offset != (typeof(skb->transport_header))offset)) 3098 return false; 3099 3100 if (unlikely(offset == (typeof(skb->transport_header))~0U)) 3101 return false; 3102 3103 skb->transport_header = offset; 3104 return true; 3105 } 3106 3107 static inline void skb_set_transport_header(struct sk_buff *skb, 3108 const int offset) 3109 { 3110 skb_reset_transport_header(skb); 3111 skb->transport_header += offset; 3112 } 3113 3114 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 3115 { 3116 return skb->head + skb->network_header; 3117 } 3118 3119 static inline void skb_reset_network_header(struct sk_buff *skb) 3120 { 3121 long offset = skb->data - skb->head; 3122 3123 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->network_header))offset); 3124 skb->network_header = offset; 3125 } 3126 3127 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 3128 { 3129 skb_reset_network_header(skb); 3130 skb->network_header += offset; 3131 } 3132 3133 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 3134 { 3135 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3136 return skb->head + skb->mac_header; 3137 } 3138 3139 static inline int skb_mac_offset(const struct sk_buff *skb) 3140 { 3141 return skb_mac_header(skb) - skb->data; 3142 } 3143 3144 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 3145 { 3146 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3147 return skb->network_header - skb->mac_header; 3148 } 3149 3150 static inline void skb_unset_mac_header(struct sk_buff *skb) 3151 { 3152 skb->mac_header = (typeof(skb->mac_header))~0U; 3153 } 3154 3155 static inline void skb_reset_mac_header(struct sk_buff *skb) 3156 { 3157 long offset = skb->data - skb->head; 3158 3159 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->mac_header))offset); 3160 skb->mac_header = offset; 3161 } 3162 3163 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 3164 { 3165 skb_reset_mac_header(skb); 3166 skb->mac_header += offset; 3167 } 3168 3169 static inline void skb_pop_mac_header(struct sk_buff *skb) 3170 { 3171 skb->mac_header = skb->network_header; 3172 } 3173 3174 static inline void skb_probe_transport_header(struct sk_buff *skb) 3175 { 3176 struct flow_keys_basic keys; 3177 3178 if (skb_transport_header_was_set(skb)) 3179 return; 3180 3181 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 3182 NULL, 0, 0, 0, 0)) 3183 skb_set_transport_header(skb, keys.control.thoff); 3184 } 3185 3186 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 3187 { 3188 if (skb_mac_header_was_set(skb)) { 3189 const unsigned char *old_mac = skb_mac_header(skb); 3190 3191 skb_set_mac_header(skb, -skb->mac_len); 3192 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 3193 } 3194 } 3195 3196 /* Move the full mac header up to current network_header. 3197 * Leaves skb->data pointing at offset skb->mac_len into the mac_header. 3198 * Must be provided the complete mac header length. 3199 */ 3200 static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len) 3201 { 3202 if (skb_mac_header_was_set(skb)) { 3203 const unsigned char *old_mac = skb_mac_header(skb); 3204 3205 skb_set_mac_header(skb, -full_mac_len); 3206 memmove(skb_mac_header(skb), old_mac, full_mac_len); 3207 __skb_push(skb, full_mac_len - skb->mac_len); 3208 } 3209 } 3210 3211 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 3212 { 3213 return skb->csum_start - skb_headroom(skb); 3214 } 3215 3216 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 3217 { 3218 return skb->head + skb->csum_start; 3219 } 3220 3221 static inline int skb_transport_offset(const struct sk_buff *skb) 3222 { 3223 return skb_transport_header(skb) - skb->data; 3224 } 3225 3226 static inline u32 skb_network_header_len(const struct sk_buff *skb) 3227 { 3228 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3229 return skb->transport_header - skb->network_header; 3230 } 3231 3232 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 3233 { 3234 return skb->inner_transport_header - skb->inner_network_header; 3235 } 3236 3237 static inline int skb_network_offset(const struct sk_buff *skb) 3238 { 3239 return skb_network_header(skb) - skb->data; 3240 } 3241 3242 static inline int skb_inner_network_offset(const struct sk_buff *skb) 3243 { 3244 return skb_inner_network_header(skb) - skb->data; 3245 } 3246 3247 static inline enum skb_drop_reason 3248 pskb_network_may_pull_reason(struct sk_buff *skb, unsigned int len) 3249 { 3250 return pskb_may_pull_reason(skb, skb_network_offset(skb) + len); 3251 } 3252 3253 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 3254 { 3255 return pskb_network_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 3256 } 3257 3258 /* 3259 * CPUs often take a performance hit when accessing unaligned memory 3260 * locations. The actual performance hit varies, it can be small if the 3261 * hardware handles it or large if we have to take an exception and fix it 3262 * in software. 3263 * 3264 * Since an ethernet header is 14 bytes network drivers often end up with 3265 * the IP header at an unaligned offset. The IP header can be aligned by 3266 * shifting the start of the packet by 2 bytes. Drivers should do this 3267 * with: 3268 * 3269 * skb_reserve(skb, NET_IP_ALIGN); 3270 * 3271 * The downside to this alignment of the IP header is that the DMA is now 3272 * unaligned. On some architectures the cost of an unaligned DMA is high 3273 * and this cost outweighs the gains made by aligning the IP header. 3274 * 3275 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 3276 * to be overridden. 3277 */ 3278 #ifndef NET_IP_ALIGN 3279 #define NET_IP_ALIGN 2 3280 #endif 3281 3282 /* 3283 * The networking layer reserves some headroom in skb data (via 3284 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 3285 * the header has to grow. In the default case, if the header has to grow 3286 * 32 bytes or less we avoid the reallocation. 3287 * 3288 * Unfortunately this headroom changes the DMA alignment of the resulting 3289 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 3290 * on some architectures. An architecture can override this value, 3291 * perhaps setting it to a cacheline in size (since that will maintain 3292 * cacheline alignment of the DMA). It must be a power of 2. 3293 * 3294 * Various parts of the networking layer expect at least 32 bytes of 3295 * headroom, you should not reduce this. 3296 * 3297 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 3298 * to reduce average number of cache lines per packet. 3299 * get_rps_cpu() for example only access one 64 bytes aligned block : 3300 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 3301 */ 3302 #ifndef NET_SKB_PAD 3303 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 3304 #endif 3305 3306 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 3307 3308 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3309 { 3310 if (WARN_ON(skb_is_nonlinear(skb))) 3311 return; 3312 skb->len = len; 3313 skb_set_tail_pointer(skb, len); 3314 } 3315 3316 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3317 { 3318 __skb_set_length(skb, len); 3319 } 3320 3321 void skb_trim(struct sk_buff *skb, unsigned int len); 3322 3323 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3324 { 3325 if (skb->data_len) 3326 return ___pskb_trim(skb, len); 3327 __skb_trim(skb, len); 3328 return 0; 3329 } 3330 3331 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3332 { 3333 skb_might_realloc(skb); 3334 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3335 } 3336 3337 /** 3338 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3339 * @skb: buffer to alter 3340 * @len: new length 3341 * 3342 * This is identical to pskb_trim except that the caller knows that 3343 * the skb is not cloned so we should never get an error due to out- 3344 * of-memory. 3345 */ 3346 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3347 { 3348 int err = pskb_trim(skb, len); 3349 BUG_ON(err); 3350 } 3351 3352 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3353 { 3354 unsigned int diff = len - skb->len; 3355 3356 if (skb_tailroom(skb) < diff) { 3357 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3358 GFP_ATOMIC); 3359 if (ret) 3360 return ret; 3361 } 3362 __skb_set_length(skb, len); 3363 return 0; 3364 } 3365 3366 /** 3367 * skb_orphan - orphan a buffer 3368 * @skb: buffer to orphan 3369 * 3370 * If a buffer currently has an owner then we call the owner's 3371 * destructor function and make the @skb unowned. The buffer continues 3372 * to exist but is no longer charged to its former owner. 3373 */ 3374 static inline void skb_orphan(struct sk_buff *skb) 3375 { 3376 if (skb->destructor) { 3377 skb->destructor(skb); 3378 skb->destructor = NULL; 3379 skb->sk = NULL; 3380 } else { 3381 BUG_ON(skb->sk); 3382 } 3383 } 3384 3385 /** 3386 * skb_orphan_frags - orphan the frags contained in a buffer 3387 * @skb: buffer to orphan frags from 3388 * @gfp_mask: allocation mask for replacement pages 3389 * 3390 * For each frag in the SKB which needs a destructor (i.e. has an 3391 * owner) create a copy of that frag and release the original 3392 * page by calling the destructor. 3393 */ 3394 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3395 { 3396 if (likely(!skb_zcopy(skb))) 3397 return 0; 3398 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3399 return 0; 3400 return skb_copy_ubufs(skb, gfp_mask); 3401 } 3402 3403 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3404 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3405 { 3406 if (likely(!skb_zcopy(skb))) 3407 return 0; 3408 return skb_copy_ubufs(skb, gfp_mask); 3409 } 3410 3411 /** 3412 * __skb_queue_purge_reason - empty a list 3413 * @list: list to empty 3414 * @reason: drop reason 3415 * 3416 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3417 * the list and one reference dropped. This function does not take the 3418 * list lock and the caller must hold the relevant locks to use it. 3419 */ 3420 static inline void __skb_queue_purge_reason(struct sk_buff_head *list, 3421 enum skb_drop_reason reason) 3422 { 3423 struct sk_buff *skb; 3424 3425 while ((skb = __skb_dequeue(list)) != NULL) 3426 kfree_skb_reason(skb, reason); 3427 } 3428 3429 static inline void __skb_queue_purge(struct sk_buff_head *list) 3430 { 3431 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3432 } 3433 3434 void skb_queue_purge_reason(struct sk_buff_head *list, 3435 enum skb_drop_reason reason); 3436 3437 static inline void skb_queue_purge(struct sk_buff_head *list) 3438 { 3439 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3440 } 3441 3442 unsigned int skb_rbtree_purge(struct rb_root *root); 3443 void skb_errqueue_purge(struct sk_buff_head *list); 3444 3445 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3446 3447 /** 3448 * netdev_alloc_frag - allocate a page fragment 3449 * @fragsz: fragment size 3450 * 3451 * Allocates a frag from a page for receive buffer. 3452 * Uses GFP_ATOMIC allocations. 3453 */ 3454 static inline void *netdev_alloc_frag(unsigned int fragsz) 3455 { 3456 return __netdev_alloc_frag_align(fragsz, ~0u); 3457 } 3458 3459 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3460 unsigned int align) 3461 { 3462 WARN_ON_ONCE(!is_power_of_2(align)); 3463 return __netdev_alloc_frag_align(fragsz, -align); 3464 } 3465 3466 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3467 gfp_t gfp_mask); 3468 3469 /** 3470 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3471 * @dev: network device to receive on 3472 * @length: length to allocate 3473 * 3474 * Allocate a new &sk_buff and assign it a usage count of one. The 3475 * buffer has unspecified headroom built in. Users should allocate 3476 * the headroom they think they need without accounting for the 3477 * built in space. The built in space is used for optimisations. 3478 * 3479 * %NULL is returned if there is no free memory. Although this function 3480 * allocates memory it can be called from an interrupt. 3481 */ 3482 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3483 unsigned int length) 3484 { 3485 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3486 } 3487 3488 /* legacy helper around __netdev_alloc_skb() */ 3489 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3490 gfp_t gfp_mask) 3491 { 3492 return __netdev_alloc_skb(NULL, length, gfp_mask); 3493 } 3494 3495 /* legacy helper around netdev_alloc_skb() */ 3496 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3497 { 3498 return netdev_alloc_skb(NULL, length); 3499 } 3500 3501 3502 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3503 unsigned int length, gfp_t gfp) 3504 { 3505 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3506 3507 if (NET_IP_ALIGN && skb) 3508 skb_reserve(skb, NET_IP_ALIGN); 3509 return skb; 3510 } 3511 3512 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3513 unsigned int length) 3514 { 3515 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3516 } 3517 3518 static inline void skb_free_frag(void *addr) 3519 { 3520 page_frag_free(addr); 3521 } 3522 3523 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3524 3525 static inline void *napi_alloc_frag(unsigned int fragsz) 3526 { 3527 return __napi_alloc_frag_align(fragsz, ~0u); 3528 } 3529 3530 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3531 unsigned int align) 3532 { 3533 WARN_ON_ONCE(!is_power_of_2(align)); 3534 return __napi_alloc_frag_align(fragsz, -align); 3535 } 3536 3537 struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int length); 3538 void napi_consume_skb(struct sk_buff *skb, int budget); 3539 3540 void napi_skb_free_stolen_head(struct sk_buff *skb); 3541 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason); 3542 3543 /** 3544 * __dev_alloc_pages - allocate page for network Rx 3545 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3546 * @order: size of the allocation 3547 * 3548 * Allocate a new page. 3549 * 3550 * %NULL is returned if there is no free memory. 3551 */ 3552 static inline struct page *__dev_alloc_pages_noprof(gfp_t gfp_mask, 3553 unsigned int order) 3554 { 3555 /* This piece of code contains several assumptions. 3556 * 1. This is for device Rx, therefore a cold page is preferred. 3557 * 2. The expectation is the user wants a compound page. 3558 * 3. If requesting a order 0 page it will not be compound 3559 * due to the check to see if order has a value in prep_new_page 3560 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3561 * code in gfp_to_alloc_flags that should be enforcing this. 3562 */ 3563 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3564 3565 return alloc_pages_node_noprof(NUMA_NO_NODE, gfp_mask, order); 3566 } 3567 #define __dev_alloc_pages(...) alloc_hooks(__dev_alloc_pages_noprof(__VA_ARGS__)) 3568 3569 /* 3570 * This specialized allocator has to be a macro for its allocations to be 3571 * accounted separately (to have a separate alloc_tag). 3572 */ 3573 #define dev_alloc_pages(_order) __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, _order) 3574 3575 /** 3576 * __dev_alloc_page - allocate a page for network Rx 3577 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3578 * 3579 * Allocate a new page. 3580 * 3581 * %NULL is returned if there is no free memory. 3582 */ 3583 static inline struct page *__dev_alloc_page_noprof(gfp_t gfp_mask) 3584 { 3585 return __dev_alloc_pages_noprof(gfp_mask, 0); 3586 } 3587 #define __dev_alloc_page(...) alloc_hooks(__dev_alloc_page_noprof(__VA_ARGS__)) 3588 3589 /* 3590 * This specialized allocator has to be a macro for its allocations to be 3591 * accounted separately (to have a separate alloc_tag). 3592 */ 3593 #define dev_alloc_page() dev_alloc_pages(0) 3594 3595 /** 3596 * dev_page_is_reusable - check whether a page can be reused for network Rx 3597 * @page: the page to test 3598 * 3599 * A page shouldn't be considered for reusing/recycling if it was allocated 3600 * under memory pressure or at a distant memory node. 3601 * 3602 * Returns: false if this page should be returned to page allocator, true 3603 * otherwise. 3604 */ 3605 static inline bool dev_page_is_reusable(const struct page *page) 3606 { 3607 return likely(page_to_nid(page) == numa_mem_id() && 3608 !page_is_pfmemalloc(page)); 3609 } 3610 3611 /** 3612 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3613 * @page: The page that was allocated from skb_alloc_page 3614 * @skb: The skb that may need pfmemalloc set 3615 */ 3616 static inline void skb_propagate_pfmemalloc(const struct page *page, 3617 struct sk_buff *skb) 3618 { 3619 if (page_is_pfmemalloc(page)) 3620 skb->pfmemalloc = true; 3621 } 3622 3623 /** 3624 * skb_frag_off() - Returns the offset of a skb fragment 3625 * @frag: the paged fragment 3626 */ 3627 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3628 { 3629 return frag->offset; 3630 } 3631 3632 /** 3633 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3634 * @frag: skb fragment 3635 * @delta: value to add 3636 */ 3637 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3638 { 3639 frag->offset += delta; 3640 } 3641 3642 /** 3643 * skb_frag_off_set() - Sets the offset of a skb fragment 3644 * @frag: skb fragment 3645 * @offset: offset of fragment 3646 */ 3647 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3648 { 3649 frag->offset = offset; 3650 } 3651 3652 /** 3653 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3654 * @fragto: skb fragment where offset is set 3655 * @fragfrom: skb fragment offset is copied from 3656 */ 3657 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3658 const skb_frag_t *fragfrom) 3659 { 3660 fragto->offset = fragfrom->offset; 3661 } 3662 3663 /* Return: true if the skb_frag contains a net_iov. */ 3664 static inline bool skb_frag_is_net_iov(const skb_frag_t *frag) 3665 { 3666 return netmem_is_net_iov(frag->netmem); 3667 } 3668 3669 /** 3670 * skb_frag_net_iov - retrieve the net_iov referred to by fragment 3671 * @frag: the fragment 3672 * 3673 * Return: the &struct net_iov associated with @frag. Returns NULL if this 3674 * frag has no associated net_iov. 3675 */ 3676 static inline struct net_iov *skb_frag_net_iov(const skb_frag_t *frag) 3677 { 3678 if (!skb_frag_is_net_iov(frag)) 3679 return NULL; 3680 3681 return netmem_to_net_iov(frag->netmem); 3682 } 3683 3684 /** 3685 * skb_frag_page - retrieve the page referred to by a paged fragment 3686 * @frag: the paged fragment 3687 * 3688 * Return: the &struct page associated with @frag. Returns NULL if this frag 3689 * has no associated page. 3690 */ 3691 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3692 { 3693 if (skb_frag_is_net_iov(frag)) 3694 return NULL; 3695 3696 return netmem_to_page(frag->netmem); 3697 } 3698 3699 /** 3700 * skb_frag_netmem - retrieve the netmem referred to by a fragment 3701 * @frag: the fragment 3702 * 3703 * Return: the &netmem_ref associated with @frag. 3704 */ 3705 static inline netmem_ref skb_frag_netmem(const skb_frag_t *frag) 3706 { 3707 return frag->netmem; 3708 } 3709 3710 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb, 3711 unsigned int headroom); 3712 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb, 3713 const struct bpf_prog *prog); 3714 3715 /** 3716 * skb_frag_address - gets the address of the data contained in a paged fragment 3717 * @frag: the paged fragment buffer 3718 * 3719 * Returns: the address of the data within @frag. The page must already 3720 * be mapped. 3721 */ 3722 static inline void *skb_frag_address(const skb_frag_t *frag) 3723 { 3724 if (!skb_frag_page(frag)) 3725 return NULL; 3726 3727 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3728 } 3729 3730 /** 3731 * skb_frag_address_safe - 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. Checks that the page 3735 * is mapped and returns %NULL otherwise. 3736 */ 3737 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3738 { 3739 struct page *page = skb_frag_page(frag); 3740 void *ptr; 3741 3742 if (!page) 3743 return NULL; 3744 3745 ptr = page_address(page); 3746 if (unlikely(!ptr)) 3747 return NULL; 3748 3749 return ptr + skb_frag_off(frag); 3750 } 3751 3752 /** 3753 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3754 * @fragto: skb fragment where page is set 3755 * @fragfrom: skb fragment page is copied from 3756 */ 3757 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3758 const skb_frag_t *fragfrom) 3759 { 3760 fragto->netmem = fragfrom->netmem; 3761 } 3762 3763 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3764 3765 /** 3766 * __skb_frag_dma_map - maps a paged fragment via the DMA API 3767 * @dev: the device to map the fragment to 3768 * @frag: the paged fragment to map 3769 * @offset: the offset within the fragment (starting at the 3770 * fragment's own offset) 3771 * @size: the number of bytes to map 3772 * @dir: the direction of the mapping (``PCI_DMA_*``) 3773 * 3774 * Maps the page associated with @frag to @device. 3775 */ 3776 static inline dma_addr_t __skb_frag_dma_map(struct device *dev, 3777 const skb_frag_t *frag, 3778 size_t offset, size_t size, 3779 enum dma_data_direction dir) 3780 { 3781 if (skb_frag_is_net_iov(frag)) { 3782 return netmem_to_net_iov(frag->netmem)->desc.dma_addr + 3783 offset + frag->offset; 3784 } 3785 return dma_map_page(dev, skb_frag_page(frag), 3786 skb_frag_off(frag) + offset, size, dir); 3787 } 3788 3789 #define skb_frag_dma_map(dev, frag, ...) \ 3790 CONCATENATE(_skb_frag_dma_map, \ 3791 COUNT_ARGS(__VA_ARGS__))(dev, frag, ##__VA_ARGS__) 3792 3793 #define __skb_frag_dma_map1(dev, frag, offset, uf, uo) ({ \ 3794 const skb_frag_t *uf = (frag); \ 3795 size_t uo = (offset); \ 3796 \ 3797 __skb_frag_dma_map(dev, uf, uo, skb_frag_size(uf) - uo, \ 3798 DMA_TO_DEVICE); \ 3799 }) 3800 #define _skb_frag_dma_map1(dev, frag, offset) \ 3801 __skb_frag_dma_map1(dev, frag, offset, __UNIQUE_ID(frag_), \ 3802 __UNIQUE_ID(offset_)) 3803 #define _skb_frag_dma_map0(dev, frag) \ 3804 _skb_frag_dma_map1(dev, frag, 0) 3805 #define _skb_frag_dma_map2(dev, frag, offset, size) \ 3806 __skb_frag_dma_map(dev, frag, offset, size, DMA_TO_DEVICE) 3807 #define _skb_frag_dma_map3(dev, frag, offset, size, dir) \ 3808 __skb_frag_dma_map(dev, frag, offset, size, dir) 3809 3810 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3811 gfp_t gfp_mask) 3812 { 3813 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3814 } 3815 3816 3817 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3818 gfp_t gfp_mask) 3819 { 3820 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3821 } 3822 3823 3824 /** 3825 * skb_clone_writable - is the header of a clone writable 3826 * @skb: buffer to check 3827 * @len: length up to which to write 3828 * 3829 * Returns true if modifying the header part of the cloned buffer 3830 * does not requires the data to be copied. 3831 */ 3832 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3833 { 3834 return !skb_header_cloned(skb) && 3835 skb_headroom(skb) + len <= skb->hdr_len; 3836 } 3837 3838 static inline int skb_try_make_writable(struct sk_buff *skb, 3839 unsigned int write_len) 3840 { 3841 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3842 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3843 } 3844 3845 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3846 int cloned) 3847 { 3848 int delta = 0; 3849 3850 if (headroom > skb_headroom(skb)) 3851 delta = headroom - skb_headroom(skb); 3852 3853 if (delta || cloned) 3854 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3855 GFP_ATOMIC); 3856 return 0; 3857 } 3858 3859 /** 3860 * skb_cow - copy header of skb when it is required 3861 * @skb: buffer to cow 3862 * @headroom: needed headroom 3863 * 3864 * If the skb passed lacks sufficient headroom or its data part 3865 * is shared, data is reallocated. If reallocation fails, an error 3866 * is returned and original skb is not changed. 3867 * 3868 * The result is skb with writable area skb->head...skb->tail 3869 * and at least @headroom of space at head. 3870 */ 3871 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3872 { 3873 return __skb_cow(skb, headroom, skb_cloned(skb)); 3874 } 3875 3876 /** 3877 * skb_cow_head - skb_cow but only making the head writable 3878 * @skb: buffer to cow 3879 * @headroom: needed headroom 3880 * 3881 * This function is identical to skb_cow except that we replace the 3882 * skb_cloned check by skb_header_cloned. It should be used when 3883 * you only need to push on some header and do not need to modify 3884 * the data. 3885 */ 3886 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3887 { 3888 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3889 } 3890 3891 /** 3892 * skb_padto - pad an skbuff up to a minimal size 3893 * @skb: buffer to pad 3894 * @len: minimal length 3895 * 3896 * Pads up a buffer to ensure the trailing bytes exist and are 3897 * blanked. If the buffer already contains sufficient data it 3898 * is untouched. Otherwise it is extended. Returns zero on 3899 * success. The skb is freed on error. 3900 */ 3901 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3902 { 3903 unsigned int size = skb->len; 3904 if (likely(size >= len)) 3905 return 0; 3906 return skb_pad(skb, len - size); 3907 } 3908 3909 /** 3910 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3911 * @skb: buffer to pad 3912 * @len: minimal length 3913 * @free_on_error: free buffer on error 3914 * 3915 * Pads up a buffer to ensure the trailing bytes exist and are 3916 * blanked. If the buffer already contains sufficient data it 3917 * is untouched. Otherwise it is extended. Returns zero on 3918 * success. The skb is freed on error if @free_on_error is true. 3919 */ 3920 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3921 unsigned int len, 3922 bool free_on_error) 3923 { 3924 unsigned int size = skb->len; 3925 3926 if (unlikely(size < len)) { 3927 len -= size; 3928 if (__skb_pad(skb, len, free_on_error)) 3929 return -ENOMEM; 3930 __skb_put(skb, len); 3931 } 3932 return 0; 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 * 3940 * Pads up a buffer to ensure the trailing bytes exist and are 3941 * blanked. If the buffer already contains sufficient data it 3942 * is untouched. Otherwise it is extended. Returns zero on 3943 * success. The skb is freed on error. 3944 */ 3945 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3946 { 3947 return __skb_put_padto(skb, len, true); 3948 } 3949 3950 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i) 3951 __must_check; 3952 3953 static inline bool skb_can_coalesce_netmem(struct sk_buff *skb, int i, 3954 netmem_ref netmem, int off) 3955 { 3956 if (skb_zcopy(skb)) 3957 return false; 3958 if (i) { 3959 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3960 3961 return netmem == skb_frag_netmem(frag) && 3962 off == skb_frag_off(frag) + skb_frag_size(frag); 3963 } 3964 return false; 3965 } 3966 3967 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3968 const struct page *page, int off) 3969 { 3970 return skb_can_coalesce_netmem(skb, i, page_to_netmem(page), off); 3971 } 3972 3973 static inline int __skb_linearize(struct sk_buff *skb) 3974 { 3975 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3976 } 3977 3978 /** 3979 * skb_linearize - convert paged skb to linear one 3980 * @skb: buffer to linarize 3981 * 3982 * If there is no free memory -ENOMEM is returned, otherwise zero 3983 * is returned and the old skb data released. 3984 */ 3985 static inline int skb_linearize(struct sk_buff *skb) 3986 { 3987 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3988 } 3989 3990 /** 3991 * skb_has_shared_frag - can any frag be overwritten 3992 * @skb: buffer to test 3993 * 3994 * Return: true if the skb has at least one frag that might be modified 3995 * by an external entity (as in vmsplice()/sendfile()) 3996 */ 3997 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3998 { 3999 return skb_is_nonlinear(skb) && 4000 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 4001 } 4002 4003 /** 4004 * skb_linearize_cow - make sure skb is linear and writable 4005 * @skb: buffer to process 4006 * 4007 * If there is no free memory -ENOMEM is returned, otherwise zero 4008 * is returned and the old skb data released. 4009 */ 4010 static inline int skb_linearize_cow(struct sk_buff *skb) 4011 { 4012 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 4013 __skb_linearize(skb) : 0; 4014 } 4015 4016 static __always_inline void 4017 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 4018 unsigned int off) 4019 { 4020 if (skb->ip_summed == CHECKSUM_COMPLETE) 4021 skb->csum = csum_block_sub(skb->csum, 4022 csum_partial(start, len, 0), off); 4023 else if (skb->ip_summed == CHECKSUM_PARTIAL && 4024 skb_checksum_start_offset(skb) < 0) 4025 skb->ip_summed = CHECKSUM_NONE; 4026 } 4027 4028 /** 4029 * skb_postpull_rcsum - update checksum for received skb after pull 4030 * @skb: buffer to update 4031 * @start: start of data before pull 4032 * @len: length of data pulled 4033 * 4034 * After doing a pull on a received packet, you need to call this to 4035 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 4036 * CHECKSUM_NONE so that it can be recomputed from scratch. 4037 */ 4038 static inline void skb_postpull_rcsum(struct sk_buff *skb, 4039 const void *start, unsigned int len) 4040 { 4041 if (skb->ip_summed == CHECKSUM_COMPLETE) 4042 skb->csum = wsum_negate(csum_partial(start, len, 4043 wsum_negate(skb->csum))); 4044 else if (skb->ip_summed == CHECKSUM_PARTIAL && 4045 skb_checksum_start_offset(skb) < 0) 4046 skb->ip_summed = CHECKSUM_NONE; 4047 } 4048 4049 static __always_inline void 4050 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 4051 unsigned int off) 4052 { 4053 if (skb->ip_summed == CHECKSUM_COMPLETE) 4054 skb->csum = csum_block_add(skb->csum, 4055 csum_partial(start, len, 0), off); 4056 } 4057 4058 /** 4059 * skb_postpush_rcsum - update checksum for received skb after push 4060 * @skb: buffer to update 4061 * @start: start of data after push 4062 * @len: length of data pushed 4063 * 4064 * After doing a push on a received packet, you need to call this to 4065 * update the CHECKSUM_COMPLETE checksum. 4066 */ 4067 static inline void skb_postpush_rcsum(struct sk_buff *skb, 4068 const void *start, unsigned int len) 4069 { 4070 __skb_postpush_rcsum(skb, start, len, 0); 4071 } 4072 4073 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 4074 4075 /** 4076 * skb_push_rcsum - push skb and update receive checksum 4077 * @skb: buffer to update 4078 * @len: length of data pulled 4079 * 4080 * This function performs an skb_push on the packet and updates 4081 * the CHECKSUM_COMPLETE checksum. It should be used on 4082 * receive path processing instead of skb_push unless you know 4083 * that the checksum difference is zero (e.g., a valid IP header) 4084 * or you are setting ip_summed to CHECKSUM_NONE. 4085 */ 4086 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 4087 { 4088 skb_push(skb, len); 4089 skb_postpush_rcsum(skb, skb->data, len); 4090 return skb->data; 4091 } 4092 4093 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 4094 /** 4095 * pskb_trim_rcsum - trim received skb and update checksum 4096 * @skb: buffer to trim 4097 * @len: new length 4098 * 4099 * This is exactly the same as pskb_trim except that it ensures the 4100 * checksum of received packets are still valid after the operation. 4101 * It can change skb pointers. 4102 */ 4103 4104 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4105 { 4106 skb_might_realloc(skb); 4107 if (likely(len >= skb->len)) 4108 return 0; 4109 return pskb_trim_rcsum_slow(skb, len); 4110 } 4111 4112 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4113 { 4114 if (skb->ip_summed == CHECKSUM_COMPLETE) 4115 skb->ip_summed = CHECKSUM_NONE; 4116 __skb_trim(skb, len); 4117 return 0; 4118 } 4119 4120 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 4121 { 4122 if (skb->ip_summed == CHECKSUM_COMPLETE) 4123 skb->ip_summed = CHECKSUM_NONE; 4124 return __skb_grow(skb, len); 4125 } 4126 4127 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 4128 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 4129 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 4130 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 4131 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 4132 4133 #define skb_queue_walk(queue, skb) \ 4134 for (skb = (queue)->next; \ 4135 skb != (struct sk_buff *)(queue); \ 4136 skb = skb->next) 4137 4138 #define skb_queue_walk_safe(queue, skb, tmp) \ 4139 for (skb = (queue)->next, tmp = skb->next; \ 4140 skb != (struct sk_buff *)(queue); \ 4141 skb = tmp, tmp = skb->next) 4142 4143 #define skb_queue_walk_from(queue, skb) \ 4144 for (; skb != (struct sk_buff *)(queue); \ 4145 skb = skb->next) 4146 4147 #define skb_rbtree_walk(skb, root) \ 4148 for (skb = skb_rb_first(root); skb != NULL; \ 4149 skb = skb_rb_next(skb)) 4150 4151 #define skb_rbtree_walk_from(skb) \ 4152 for (; skb != NULL; \ 4153 skb = skb_rb_next(skb)) 4154 4155 #define skb_rbtree_walk_from_safe(skb, tmp) \ 4156 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 4157 skb = tmp) 4158 4159 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 4160 for (tmp = skb->next; \ 4161 skb != (struct sk_buff *)(queue); \ 4162 skb = tmp, tmp = skb->next) 4163 4164 #define skb_queue_reverse_walk(queue, skb) \ 4165 for (skb = (queue)->prev; \ 4166 skb != (struct sk_buff *)(queue); \ 4167 skb = skb->prev) 4168 4169 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 4170 for (skb = (queue)->prev, tmp = skb->prev; \ 4171 skb != (struct sk_buff *)(queue); \ 4172 skb = tmp, tmp = skb->prev) 4173 4174 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 4175 for (tmp = skb->prev; \ 4176 skb != (struct sk_buff *)(queue); \ 4177 skb = tmp, tmp = skb->prev) 4178 4179 static inline bool skb_has_frag_list(const struct sk_buff *skb) 4180 { 4181 return skb_shinfo(skb)->frag_list != NULL; 4182 } 4183 4184 static inline void skb_frag_list_init(struct sk_buff *skb) 4185 { 4186 skb_shinfo(skb)->frag_list = NULL; 4187 } 4188 4189 #define skb_walk_frags(skb, iter) \ 4190 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 4191 4192 4193 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 4194 int *err, long *timeo_p, 4195 const struct sk_buff *skb); 4196 struct sk_buff *__skb_try_recv_from_queue(struct sk_buff_head *queue, 4197 unsigned int flags, 4198 int *off, int *err, 4199 struct sk_buff **last); 4200 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 4201 struct sk_buff_head *queue, 4202 unsigned int flags, int *off, int *err, 4203 struct sk_buff **last); 4204 struct sk_buff *__skb_recv_datagram(struct sock *sk, 4205 struct sk_buff_head *sk_queue, 4206 unsigned int flags, int *off, int *err); 4207 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 4208 __poll_t datagram_poll_queue(struct file *file, struct socket *sock, 4209 struct poll_table_struct *wait, 4210 struct sk_buff_head *rcv_queue); 4211 __poll_t datagram_poll(struct file *file, struct socket *sock, 4212 struct poll_table_struct *wait); 4213 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 4214 struct iov_iter *to, int size); 4215 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 4216 struct msghdr *msg, int size) 4217 { 4218 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 4219 } 4220 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 4221 struct msghdr *msg); 4222 int skb_copy_and_crc32c_datagram_iter(const struct sk_buff *skb, int offset, 4223 struct iov_iter *to, int len, u32 *crcp); 4224 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 4225 struct iov_iter *from, int len); 4226 int skb_copy_datagram_from_iter_full(struct sk_buff *skb, int offset, 4227 struct iov_iter *from, int len); 4228 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 4229 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 4230 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 4231 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 4232 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 4233 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 4234 int len); 4235 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 4236 struct pipe_inode_info *pipe, unsigned int len, 4237 unsigned int flags); 4238 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 4239 int len); 4240 int skb_send_sock_locked_with_flags(struct sock *sk, struct sk_buff *skb, 4241 int offset, int len, int flags); 4242 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 4243 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 4244 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 4245 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 4246 int len, int hlen); 4247 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 4248 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 4249 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 4250 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 4251 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 4252 unsigned int offset); 4253 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 4254 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 4255 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev); 4256 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 4257 int skb_vlan_pop(struct sk_buff *skb); 4258 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 4259 int skb_eth_pop(struct sk_buff *skb); 4260 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 4261 const unsigned char *src); 4262 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 4263 int mac_len, bool ethernet); 4264 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 4265 bool ethernet); 4266 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 4267 int skb_mpls_dec_ttl(struct sk_buff *skb); 4268 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 4269 gfp_t gfp); 4270 4271 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 4272 { 4273 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 4274 } 4275 4276 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 4277 { 4278 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 4279 } 4280 4281 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 4282 __wsum csum); 4283 u32 skb_crc32c(const struct sk_buff *skb, int offset, int len, u32 crc); 4284 4285 static inline void * __must_check 4286 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 4287 const void *data, int hlen, void *buffer) 4288 { 4289 if (likely(hlen - offset >= len)) 4290 return (void *)data + offset; 4291 4292 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 4293 return NULL; 4294 4295 return buffer; 4296 } 4297 4298 static inline void * __must_check 4299 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4300 { 4301 return __skb_header_pointer(skb, offset, len, skb->data, 4302 skb_headlen(skb), buffer); 4303 } 4304 4305 /* Variant of skb_header_pointer() where @offset is user-controlled 4306 * and potentially negative. 4307 */ 4308 static inline void * __must_check 4309 skb_header_pointer_careful(const struct sk_buff *skb, int offset, 4310 int len, void *buffer) 4311 { 4312 if (unlikely(offset < 0 && -offset > skb_headroom(skb))) 4313 return NULL; 4314 return skb_header_pointer(skb, offset, len, buffer); 4315 } 4316 4317 static inline void * __must_check 4318 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len) 4319 { 4320 if (likely(skb_headlen(skb) - offset >= len)) 4321 return skb->data + offset; 4322 return NULL; 4323 } 4324 4325 /** 4326 * skb_needs_linearize - check if we need to linearize a given skb 4327 * depending on the given device features. 4328 * @skb: socket buffer to check 4329 * @features: net device features 4330 * 4331 * Returns true if either: 4332 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4333 * 2. skb is fragmented and the device does not support SG. 4334 */ 4335 static inline bool skb_needs_linearize(struct sk_buff *skb, 4336 netdev_features_t features) 4337 { 4338 return skb_is_nonlinear(skb) && 4339 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4340 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4341 } 4342 4343 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4344 void *to, 4345 const unsigned int len) 4346 { 4347 memcpy(to, skb->data, len); 4348 } 4349 4350 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4351 const int offset, void *to, 4352 const unsigned int len) 4353 { 4354 memcpy(to, skb->data + offset, len); 4355 } 4356 4357 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4358 const void *from, 4359 const unsigned int len) 4360 { 4361 memcpy(skb->data, from, len); 4362 } 4363 4364 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4365 const int offset, 4366 const void *from, 4367 const unsigned int len) 4368 { 4369 memcpy(skb->data + offset, from, len); 4370 } 4371 4372 void skb_init(void); 4373 4374 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4375 { 4376 return skb->tstamp; 4377 } 4378 4379 /** 4380 * skb_get_timestamp - get timestamp from a skb 4381 * @skb: skb to get stamp from 4382 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4383 * 4384 * Timestamps are stored in the skb as offsets to a base timestamp. 4385 * This function converts the offset back to a struct timeval and stores 4386 * it in stamp. 4387 */ 4388 static inline void skb_get_timestamp(const struct sk_buff *skb, 4389 struct __kernel_old_timeval *stamp) 4390 { 4391 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4392 } 4393 4394 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4395 struct __kernel_sock_timeval *stamp) 4396 { 4397 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4398 4399 stamp->tv_sec = ts.tv_sec; 4400 stamp->tv_usec = ts.tv_nsec / 1000; 4401 } 4402 4403 static inline void skb_get_timestampns(const struct sk_buff *skb, 4404 struct __kernel_old_timespec *stamp) 4405 { 4406 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4407 4408 stamp->tv_sec = ts.tv_sec; 4409 stamp->tv_nsec = ts.tv_nsec; 4410 } 4411 4412 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4413 struct __kernel_timespec *stamp) 4414 { 4415 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4416 4417 stamp->tv_sec = ts.tv_sec; 4418 stamp->tv_nsec = ts.tv_nsec; 4419 } 4420 4421 static inline void __net_timestamp(struct sk_buff *skb) 4422 { 4423 skb->tstamp = ktime_get_real(); 4424 skb->tstamp_type = SKB_CLOCK_REALTIME; 4425 } 4426 4427 static inline ktime_t net_timedelta(ktime_t t) 4428 { 4429 return ktime_sub(ktime_get_real(), t); 4430 } 4431 4432 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4433 u8 tstamp_type) 4434 { 4435 skb->tstamp = kt; 4436 4437 if (kt) 4438 skb->tstamp_type = tstamp_type; 4439 else 4440 skb->tstamp_type = SKB_CLOCK_REALTIME; 4441 } 4442 4443 static inline void skb_set_delivery_type_by_clockid(struct sk_buff *skb, 4444 ktime_t kt, clockid_t clockid) 4445 { 4446 u8 tstamp_type = SKB_CLOCK_REALTIME; 4447 4448 switch (clockid) { 4449 case CLOCK_REALTIME: 4450 break; 4451 case CLOCK_MONOTONIC: 4452 tstamp_type = SKB_CLOCK_MONOTONIC; 4453 break; 4454 case CLOCK_TAI: 4455 tstamp_type = SKB_CLOCK_TAI; 4456 break; 4457 default: 4458 WARN_ON_ONCE(1); 4459 kt = 0; 4460 } 4461 4462 skb_set_delivery_time(skb, kt, tstamp_type); 4463 } 4464 4465 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4466 4467 /* It is used in the ingress path to clear the delivery_time. 4468 * If needed, set the skb->tstamp to the (rcv) timestamp. 4469 */ 4470 static inline void skb_clear_delivery_time(struct sk_buff *skb) 4471 { 4472 if (skb->tstamp_type) { 4473 skb->tstamp_type = SKB_CLOCK_REALTIME; 4474 if (static_branch_unlikely(&netstamp_needed_key)) 4475 skb->tstamp = ktime_get_real(); 4476 else 4477 skb->tstamp = 0; 4478 } 4479 } 4480 4481 static inline void skb_clear_tstamp(struct sk_buff *skb) 4482 { 4483 if (skb->tstamp_type) 4484 return; 4485 4486 skb->tstamp = 0; 4487 } 4488 4489 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4490 { 4491 if (skb->tstamp_type) 4492 return 0; 4493 4494 return skb->tstamp; 4495 } 4496 4497 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4498 { 4499 if (skb->tstamp_type != SKB_CLOCK_MONOTONIC && skb->tstamp) 4500 return skb->tstamp; 4501 4502 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4503 return ktime_get_real(); 4504 4505 return 0; 4506 } 4507 4508 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4509 { 4510 return skb_shinfo(skb)->meta_len; 4511 } 4512 4513 static inline void *skb_metadata_end(const struct sk_buff *skb) 4514 { 4515 return skb_mac_header(skb); 4516 } 4517 4518 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4519 const struct sk_buff *skb_b, 4520 u8 meta_len) 4521 { 4522 const void *a = skb_metadata_end(skb_a); 4523 const void *b = skb_metadata_end(skb_b); 4524 u64 diffs = 0; 4525 4526 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) || 4527 BITS_PER_LONG != 64) 4528 goto slow; 4529 4530 /* Using more efficient variant than plain call to memcmp(). */ 4531 switch (meta_len) { 4532 #define __it(x, op) (x -= sizeof(u##op)) 4533 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4534 case 32: diffs |= __it_diff(a, b, 64); 4535 fallthrough; 4536 case 24: diffs |= __it_diff(a, b, 64); 4537 fallthrough; 4538 case 16: diffs |= __it_diff(a, b, 64); 4539 fallthrough; 4540 case 8: diffs |= __it_diff(a, b, 64); 4541 break; 4542 case 28: diffs |= __it_diff(a, b, 64); 4543 fallthrough; 4544 case 20: diffs |= __it_diff(a, b, 64); 4545 fallthrough; 4546 case 12: diffs |= __it_diff(a, b, 64); 4547 fallthrough; 4548 case 4: diffs |= __it_diff(a, b, 32); 4549 break; 4550 default: 4551 slow: 4552 return memcmp(a - meta_len, b - meta_len, meta_len); 4553 } 4554 return diffs; 4555 } 4556 4557 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4558 const struct sk_buff *skb_b) 4559 { 4560 u8 len_a = skb_metadata_len(skb_a); 4561 u8 len_b = skb_metadata_len(skb_b); 4562 4563 if (!(len_a | len_b)) 4564 return false; 4565 4566 return len_a != len_b ? 4567 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4568 } 4569 4570 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4571 { 4572 skb_shinfo(skb)->meta_len = meta_len; 4573 } 4574 4575 static inline void skb_metadata_clear(struct sk_buff *skb) 4576 { 4577 skb_metadata_set(skb, 0); 4578 } 4579 4580 /** 4581 * skb_data_move - Move packet data and metadata after skb_push() or skb_pull(). 4582 * @skb: packet to operate on 4583 * @len: number of bytes pushed or pulled from &sk_buff->data 4584 * @n: number of bytes to memmove() from pre-push/pull &sk_buff->data 4585 * 4586 * Moves @n bytes of packet data, can be zero, and all bytes of skb metadata. 4587 * 4588 * Assumes metadata is located immediately before &sk_buff->data prior to the 4589 * push/pull, and that sufficient headroom exists to hold it after an 4590 * skb_push(). Otherwise, metadata is cleared and a one-time warning is issued. 4591 * 4592 * Prefer skb_postpull_data_move() or skb_postpush_data_move() to calling this 4593 * helper directly. 4594 */ 4595 static inline void skb_data_move(struct sk_buff *skb, const int len, 4596 const unsigned int n) 4597 { 4598 const u8 meta_len = skb_metadata_len(skb); 4599 u8 *meta, *meta_end; 4600 4601 if (!len || (!n && !meta_len)) 4602 return; 4603 4604 if (!meta_len) 4605 goto no_metadata; 4606 4607 meta_end = skb_metadata_end(skb); 4608 meta = meta_end - meta_len; 4609 4610 if (WARN_ON_ONCE(meta_end + len != skb->data || 4611 meta_len > skb_headroom(skb))) { 4612 skb_metadata_clear(skb); 4613 goto no_metadata; 4614 } 4615 4616 memmove(meta + len, meta, meta_len + n); 4617 return; 4618 4619 no_metadata: 4620 memmove(skb->data, skb->data - len, n); 4621 } 4622 4623 /** 4624 * skb_postpull_data_move - Move packet data and metadata after skb_pull(). 4625 * @skb: packet to operate on 4626 * @len: number of bytes pulled from &sk_buff->data 4627 * @n: number of bytes to memmove() from pre-pull &sk_buff->data 4628 * 4629 * See skb_data_move() for details. 4630 */ 4631 static inline void skb_postpull_data_move(struct sk_buff *skb, 4632 const unsigned int len, 4633 const unsigned int n) 4634 { 4635 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 4636 skb_data_move(skb, len, n); 4637 } 4638 4639 /** 4640 * skb_postpush_data_move - Move packet data and metadata after skb_push(). 4641 * @skb: packet to operate on 4642 * @len: number of bytes pushed onto &sk_buff->data 4643 * @n: number of bytes to memmove() from pre-push &sk_buff->data 4644 * 4645 * See skb_data_move() for details. 4646 */ 4647 static inline void skb_postpush_data_move(struct sk_buff *skb, 4648 const unsigned int len, 4649 const unsigned int n) 4650 { 4651 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 4652 skb_data_move(skb, -len, n); 4653 } 4654 4655 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4656 4657 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4658 4659 void skb_clone_tx_timestamp(struct sk_buff *skb); 4660 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4661 4662 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4663 4664 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4665 { 4666 } 4667 4668 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4669 { 4670 return false; 4671 } 4672 4673 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4674 4675 /** 4676 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4677 * 4678 * PHY drivers may accept clones of transmitted packets for 4679 * timestamping via their phy_driver.txtstamp method. These drivers 4680 * must call this function to return the skb back to the stack with a 4681 * timestamp. 4682 * 4683 * @skb: clone of the original outgoing packet 4684 * @hwtstamps: hardware time stamps 4685 * 4686 */ 4687 void skb_complete_tx_timestamp(struct sk_buff *skb, 4688 struct skb_shared_hwtstamps *hwtstamps); 4689 4690 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4691 struct skb_shared_hwtstamps *hwtstamps, 4692 struct sock *sk, int tstype); 4693 4694 /** 4695 * skb_tstamp_tx - queue clone of skb with send time stamps 4696 * @orig_skb: the original outgoing packet 4697 * @hwtstamps: hardware time stamps, may be NULL if not available 4698 * 4699 * If the skb has a socket associated, then this function clones the 4700 * skb (thus sharing the actual data and optional structures), stores 4701 * the optional hardware time stamping information (if non NULL) or 4702 * generates a software time stamp (otherwise), then queues the clone 4703 * to the error queue of the socket. Errors are silently ignored. 4704 */ 4705 void skb_tstamp_tx(struct sk_buff *orig_skb, 4706 struct skb_shared_hwtstamps *hwtstamps); 4707 4708 /** 4709 * skb_tx_timestamp() - Driver hook for transmit timestamping 4710 * 4711 * Ethernet MAC Drivers should call this function in their hard_xmit() 4712 * function immediately before giving the sk_buff to the MAC hardware. 4713 * 4714 * Specifically, one should make absolutely sure that this function is 4715 * called before TX completion of this packet can trigger. Otherwise 4716 * the packet could potentially already be freed. 4717 * 4718 * @skb: A socket buffer. 4719 */ 4720 static inline void skb_tx_timestamp(struct sk_buff *skb) 4721 { 4722 skb_clone_tx_timestamp(skb); 4723 if (skb_shinfo(skb)->tx_flags & (SKBTX_SW_TSTAMP | SKBTX_BPF)) 4724 skb_tstamp_tx(skb, NULL); 4725 } 4726 4727 /** 4728 * skb_complete_wifi_ack - deliver skb with wifi status 4729 * 4730 * @skb: the original outgoing packet 4731 * @acked: ack status 4732 * 4733 */ 4734 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4735 4736 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4737 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4738 4739 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4740 { 4741 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4742 skb->csum_valid || 4743 (skb->ip_summed == CHECKSUM_PARTIAL && 4744 skb_checksum_start_offset(skb) >= 0)); 4745 } 4746 4747 /** 4748 * skb_checksum_complete - Calculate checksum of an entire packet 4749 * @skb: packet to process 4750 * 4751 * This function calculates the checksum over the entire packet plus 4752 * the value of skb->csum. The latter can be used to supply the 4753 * checksum of a pseudo header as used by TCP/UDP. It returns the 4754 * checksum. 4755 * 4756 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4757 * this function can be used to verify that checksum on received 4758 * packets. In that case the function should return zero if the 4759 * checksum is correct. In particular, this function will return zero 4760 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4761 * hardware has already verified the correctness of the checksum. 4762 */ 4763 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4764 { 4765 return skb_csum_unnecessary(skb) ? 4766 0 : __skb_checksum_complete(skb); 4767 } 4768 4769 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4770 { 4771 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4772 if (skb->csum_level == 0) 4773 skb->ip_summed = CHECKSUM_NONE; 4774 else 4775 skb->csum_level--; 4776 } 4777 } 4778 4779 static __always_inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4780 { 4781 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4782 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4783 skb->csum_level++; 4784 } else if (skb->ip_summed == CHECKSUM_NONE) { 4785 skb->ip_summed = CHECKSUM_UNNECESSARY; 4786 skb->csum_level = 0; 4787 } 4788 } 4789 4790 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4791 { 4792 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4793 skb->ip_summed = CHECKSUM_NONE; 4794 skb->csum_level = 0; 4795 } 4796 } 4797 4798 /* Check if we need to perform checksum complete validation. 4799 * 4800 * Returns: true if checksum complete is needed, false otherwise 4801 * (either checksum is unnecessary or zero checksum is allowed). 4802 */ 4803 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4804 bool zero_okay, 4805 __sum16 check) 4806 { 4807 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4808 skb->csum_valid = 1; 4809 __skb_decr_checksum_unnecessary(skb); 4810 return false; 4811 } 4812 4813 return true; 4814 } 4815 4816 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4817 * in checksum_init. 4818 */ 4819 #define CHECKSUM_BREAK 76 4820 4821 /* Unset checksum-complete 4822 * 4823 * Unset checksum complete can be done when packet is being modified 4824 * (uncompressed for instance) and checksum-complete value is 4825 * invalidated. 4826 */ 4827 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4828 { 4829 if (skb->ip_summed == CHECKSUM_COMPLETE) 4830 skb->ip_summed = CHECKSUM_NONE; 4831 } 4832 4833 /* Validate (init) checksum based on checksum complete. 4834 * 4835 * Return values: 4836 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4837 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4838 * checksum is stored in skb->csum for use in __skb_checksum_complete 4839 * non-zero: value of invalid checksum 4840 * 4841 */ 4842 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4843 bool complete, 4844 __wsum psum) 4845 { 4846 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4847 if (!csum_fold(csum_add(psum, skb->csum))) { 4848 skb->csum_valid = 1; 4849 return 0; 4850 } 4851 } 4852 4853 skb->csum = psum; 4854 4855 if (complete || skb->len <= CHECKSUM_BREAK) { 4856 __sum16 csum; 4857 4858 csum = __skb_checksum_complete(skb); 4859 skb->csum_valid = !csum; 4860 return csum; 4861 } 4862 4863 return 0; 4864 } 4865 4866 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4867 { 4868 return 0; 4869 } 4870 4871 /* Perform checksum validate (init). Note that this is a macro since we only 4872 * want to calculate the pseudo header which is an input function if necessary. 4873 * First we try to validate without any computation (checksum unnecessary) and 4874 * then calculate based on checksum complete calling the function to compute 4875 * pseudo header. 4876 * 4877 * Return values: 4878 * 0: checksum is validated or try to in skb_checksum_complete 4879 * non-zero: value of invalid checksum 4880 */ 4881 #define __skb_checksum_validate(skb, proto, complete, \ 4882 zero_okay, check, compute_pseudo) \ 4883 ({ \ 4884 __sum16 __ret = 0; \ 4885 skb->csum_valid = 0; \ 4886 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4887 __ret = __skb_checksum_validate_complete(skb, \ 4888 complete, compute_pseudo(skb, proto)); \ 4889 __ret; \ 4890 }) 4891 4892 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4893 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4894 4895 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4896 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4897 4898 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4899 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4900 4901 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4902 compute_pseudo) \ 4903 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4904 4905 #define skb_checksum_simple_validate(skb) \ 4906 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4907 4908 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4909 { 4910 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4911 } 4912 4913 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4914 { 4915 skb->csum = ~pseudo; 4916 skb->ip_summed = CHECKSUM_COMPLETE; 4917 } 4918 4919 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4920 do { \ 4921 if (__skb_checksum_convert_check(skb)) \ 4922 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4923 } while (0) 4924 4925 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4926 u16 start, u16 offset) 4927 { 4928 skb->ip_summed = CHECKSUM_PARTIAL; 4929 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4930 skb->csum_offset = offset - start; 4931 } 4932 4933 /* Update skbuf and packet to reflect the remote checksum offload operation. 4934 * When called, ptr indicates the starting point for skb->csum when 4935 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4936 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4937 */ 4938 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4939 int start, int offset, bool nopartial) 4940 { 4941 __wsum delta; 4942 4943 if (!nopartial) { 4944 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4945 return; 4946 } 4947 4948 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4949 __skb_checksum_complete(skb); 4950 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4951 } 4952 4953 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4954 4955 /* Adjust skb->csum since we changed the packet */ 4956 skb->csum = csum_add(skb->csum, delta); 4957 } 4958 4959 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4960 { 4961 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4962 return (void *)(skb->_nfct & NFCT_PTRMASK); 4963 #else 4964 return NULL; 4965 #endif 4966 } 4967 4968 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4969 { 4970 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4971 return skb->_nfct; 4972 #else 4973 return 0UL; 4974 #endif 4975 } 4976 4977 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4978 { 4979 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4980 skb->slow_gro |= !!nfct; 4981 skb->_nfct = nfct; 4982 #endif 4983 } 4984 4985 #ifdef CONFIG_SKB_EXTENSIONS 4986 enum skb_ext_id { 4987 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4988 SKB_EXT_BRIDGE_NF, 4989 #endif 4990 #ifdef CONFIG_XFRM 4991 SKB_EXT_SEC_PATH, 4992 #endif 4993 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4994 TC_SKB_EXT, 4995 #endif 4996 #if IS_ENABLED(CONFIG_MPTCP) 4997 SKB_EXT_MPTCP, 4998 #endif 4999 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 5000 SKB_EXT_MCTP, 5001 #endif 5002 #if IS_ENABLED(CONFIG_INET_PSP) 5003 SKB_EXT_PSP, 5004 #endif 5005 #if IS_ENABLED(CONFIG_CAN) 5006 SKB_EXT_CAN, 5007 #endif 5008 SKB_EXT_NUM, /* must be last */ 5009 }; 5010 5011 /** 5012 * struct skb_ext - sk_buff extensions 5013 * @refcnt: 1 on allocation, deallocated on 0 5014 * @offset: offset to add to @data to obtain extension address 5015 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 5016 * @data: start of extension data, variable sized 5017 * 5018 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 5019 * to use 'u8' types while allowing up to 2kb worth of extension data. 5020 */ 5021 struct skb_ext { 5022 refcount_t refcnt; 5023 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 5024 u8 chunks; /* same */ 5025 char data[] __aligned(8); 5026 }; 5027 5028 struct skb_ext *__skb_ext_alloc(gfp_t flags); 5029 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 5030 struct skb_ext *ext); 5031 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 5032 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 5033 void __skb_ext_put(struct skb_ext *ext); 5034 5035 static inline void skb_ext_put(struct sk_buff *skb) 5036 { 5037 if (skb->active_extensions) 5038 __skb_ext_put(skb->extensions); 5039 } 5040 5041 static inline void __skb_ext_copy(struct sk_buff *dst, 5042 const struct sk_buff *src) 5043 { 5044 dst->active_extensions = src->active_extensions; 5045 5046 if (src->active_extensions) { 5047 struct skb_ext *ext = src->extensions; 5048 5049 refcount_inc(&ext->refcnt); 5050 dst->extensions = ext; 5051 } 5052 } 5053 5054 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 5055 { 5056 skb_ext_put(dst); 5057 __skb_ext_copy(dst, src); 5058 } 5059 5060 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 5061 { 5062 return !!ext->offset[i]; 5063 } 5064 5065 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 5066 { 5067 return skb->active_extensions & (1 << id); 5068 } 5069 5070 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 5071 { 5072 if (skb_ext_exist(skb, id)) 5073 __skb_ext_del(skb, id); 5074 } 5075 5076 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 5077 { 5078 if (skb_ext_exist(skb, id)) { 5079 struct skb_ext *ext = skb->extensions; 5080 5081 return (void *)ext + (ext->offset[id] << 3); 5082 } 5083 5084 return NULL; 5085 } 5086 5087 static inline void skb_ext_reset(struct sk_buff *skb) 5088 { 5089 if (unlikely(skb->active_extensions)) { 5090 __skb_ext_put(skb->extensions); 5091 skb->active_extensions = 0; 5092 } 5093 } 5094 5095 static inline bool skb_has_extensions(struct sk_buff *skb) 5096 { 5097 return unlikely(skb->active_extensions); 5098 } 5099 #else 5100 static inline void skb_ext_put(struct sk_buff *skb) {} 5101 static inline void skb_ext_reset(struct sk_buff *skb) {} 5102 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 5103 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 5104 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 5105 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 5106 #endif /* CONFIG_SKB_EXTENSIONS */ 5107 5108 static inline void nf_reset_ct(struct sk_buff *skb) 5109 { 5110 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5111 nf_conntrack_put(skb_nfct(skb)); 5112 skb->_nfct = 0; 5113 #endif 5114 } 5115 5116 static inline void nf_reset_trace(struct sk_buff *skb) 5117 { 5118 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 5119 skb->nf_trace = 0; 5120 #endif 5121 } 5122 5123 static inline void ipvs_reset(struct sk_buff *skb) 5124 { 5125 #if IS_ENABLED(CONFIG_IP_VS) 5126 skb->ipvs_property = 0; 5127 #endif 5128 } 5129 5130 /* Note: This doesn't put any conntrack info in dst. */ 5131 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 5132 bool copy) 5133 { 5134 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5135 dst->_nfct = src->_nfct; 5136 nf_conntrack_get(skb_nfct(src)); 5137 #endif 5138 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 5139 if (copy) 5140 dst->nf_trace = src->nf_trace; 5141 #endif 5142 } 5143 5144 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 5145 { 5146 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5147 nf_conntrack_put(skb_nfct(dst)); 5148 #endif 5149 dst->slow_gro = src->slow_gro; 5150 __nf_copy(dst, src, true); 5151 } 5152 5153 #ifdef CONFIG_NETWORK_SECMARK 5154 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 5155 { 5156 to->secmark = from->secmark; 5157 } 5158 5159 static inline void skb_init_secmark(struct sk_buff *skb) 5160 { 5161 skb->secmark = 0; 5162 } 5163 #else 5164 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 5165 { } 5166 5167 static inline void skb_init_secmark(struct sk_buff *skb) 5168 { } 5169 #endif 5170 5171 static inline int secpath_exists(const struct sk_buff *skb) 5172 { 5173 #ifdef CONFIG_XFRM 5174 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 5175 #else 5176 return 0; 5177 #endif 5178 } 5179 5180 static inline bool skb_irq_freeable(const struct sk_buff *skb) 5181 { 5182 return !skb->destructor && 5183 !secpath_exists(skb) && 5184 !skb_nfct(skb) && 5185 !skb->_skb_refdst && 5186 !skb_has_frag_list(skb); 5187 } 5188 5189 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 5190 { 5191 skb->queue_mapping = queue_mapping; 5192 } 5193 5194 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 5195 { 5196 return skb->queue_mapping; 5197 } 5198 5199 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 5200 { 5201 to->queue_mapping = from->queue_mapping; 5202 } 5203 5204 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 5205 { 5206 skb->queue_mapping = rx_queue + 1; 5207 } 5208 5209 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 5210 { 5211 return skb->queue_mapping - 1; 5212 } 5213 5214 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 5215 { 5216 return skb->queue_mapping != 0; 5217 } 5218 5219 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 5220 { 5221 skb->dst_pending_confirm = val; 5222 } 5223 5224 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 5225 { 5226 return skb->dst_pending_confirm != 0; 5227 } 5228 5229 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 5230 { 5231 #ifdef CONFIG_XFRM 5232 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 5233 #else 5234 return NULL; 5235 #endif 5236 } 5237 5238 static inline bool skb_is_gso(const struct sk_buff *skb) 5239 { 5240 return skb_shinfo(skb)->gso_size; 5241 } 5242 5243 /* Note: Should be called only if skb_is_gso(skb) is true */ 5244 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 5245 { 5246 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 5247 } 5248 5249 /* Note: Should be called only if skb_is_gso(skb) is true */ 5250 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 5251 { 5252 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 5253 } 5254 5255 /* Note: Should be called only if skb_is_gso(skb) is true */ 5256 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 5257 { 5258 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 5259 } 5260 5261 static inline void skb_gso_reset(struct sk_buff *skb) 5262 { 5263 skb_shinfo(skb)->gso_size = 0; 5264 skb_shinfo(skb)->gso_segs = 0; 5265 skb_shinfo(skb)->gso_type = 0; 5266 } 5267 5268 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 5269 u16 increment) 5270 { 5271 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5272 return; 5273 shinfo->gso_size += increment; 5274 } 5275 5276 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 5277 u16 decrement) 5278 { 5279 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5280 return; 5281 shinfo->gso_size -= decrement; 5282 } 5283 5284 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 5285 5286 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 5287 { 5288 /* LRO sets gso_size but not gso_type, whereas if GSO is really 5289 * wanted then gso_type will be set. */ 5290 const struct skb_shared_info *shinfo = skb_shinfo(skb); 5291 5292 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 5293 unlikely(shinfo->gso_type == 0)) { 5294 __skb_warn_lro_forwarding(skb); 5295 return true; 5296 } 5297 return false; 5298 } 5299 5300 static inline void skb_forward_csum(struct sk_buff *skb) 5301 { 5302 /* Unfortunately we don't support this one. Any brave souls? */ 5303 if (skb->ip_summed == CHECKSUM_COMPLETE) 5304 skb->ip_summed = CHECKSUM_NONE; 5305 } 5306 5307 /** 5308 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 5309 * @skb: skb to check 5310 * 5311 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 5312 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 5313 * use this helper, to document places where we make this assertion. 5314 */ 5315 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 5316 { 5317 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 5318 } 5319 5320 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 5321 5322 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 5323 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 5324 unsigned int transport_len, 5325 __sum16(*skb_chkf)(struct sk_buff *skb)); 5326 5327 /** 5328 * skb_head_is_locked - Determine if the skb->head is locked down 5329 * @skb: skb to check 5330 * 5331 * The head on skbs build around a head frag can be removed if they are 5332 * not cloned. This function returns true if the skb head is locked down 5333 * due to either being allocated via kmalloc, or by being a clone with 5334 * multiple references to the head. 5335 */ 5336 static inline bool skb_head_is_locked(const struct sk_buff *skb) 5337 { 5338 return !skb->head_frag || skb_cloned(skb); 5339 } 5340 5341 /* Local Checksum Offload. 5342 * Compute outer checksum based on the assumption that the 5343 * inner checksum will be offloaded later. 5344 * See Documentation/networking/checksum-offloads.rst for 5345 * explanation of how this works. 5346 * Fill in outer checksum adjustment (e.g. with sum of outer 5347 * pseudo-header) before calling. 5348 * Also ensure that inner checksum is in linear data area. 5349 */ 5350 static inline __wsum lco_csum(struct sk_buff *skb) 5351 { 5352 unsigned char *csum_start = skb_checksum_start(skb); 5353 unsigned char *l4_hdr = skb_transport_header(skb); 5354 __wsum partial; 5355 5356 /* Start with complement of inner checksum adjustment */ 5357 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 5358 skb->csum_offset)); 5359 5360 /* Add in checksum of our headers (incl. outer checksum 5361 * adjustment filled in by caller) and return result. 5362 */ 5363 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5364 } 5365 5366 static inline bool skb_is_redirected(const struct sk_buff *skb) 5367 { 5368 return skb->redirected; 5369 } 5370 5371 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5372 { 5373 skb->redirected = 1; 5374 #ifdef CONFIG_NET_REDIRECT 5375 skb->from_ingress = from_ingress; 5376 if (skb->from_ingress) 5377 skb_clear_tstamp(skb); 5378 #endif 5379 } 5380 5381 static inline void skb_reset_redirect(struct sk_buff *skb) 5382 { 5383 skb->redirected = 0; 5384 } 5385 5386 static inline void skb_set_redirected_noclear(struct sk_buff *skb, 5387 bool from_ingress) 5388 { 5389 skb->redirected = 1; 5390 #ifdef CONFIG_NET_REDIRECT 5391 skb->from_ingress = from_ingress; 5392 #endif 5393 } 5394 5395 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5396 { 5397 #if IS_ENABLED(CONFIG_IP_SCTP) 5398 return skb->csum_not_inet; 5399 #else 5400 return 0; 5401 #endif 5402 } 5403 5404 static inline void skb_reset_csum_not_inet(struct sk_buff *skb) 5405 { 5406 skb->ip_summed = CHECKSUM_NONE; 5407 #if IS_ENABLED(CONFIG_IP_SCTP) 5408 skb->csum_not_inet = 0; 5409 #endif 5410 } 5411 5412 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5413 const u64 kcov_handle) 5414 { 5415 #ifdef CONFIG_KCOV 5416 skb->kcov_handle = kcov_handle; 5417 #endif 5418 } 5419 5420 static inline u64 skb_get_kcov_handle(struct sk_buff *skb) 5421 { 5422 #ifdef CONFIG_KCOV 5423 return skb->kcov_handle; 5424 #else 5425 return 0; 5426 #endif 5427 } 5428 5429 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5430 { 5431 #ifdef CONFIG_PAGE_POOL 5432 skb->pp_recycle = 1; 5433 #endif 5434 } 5435 5436 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter, 5437 ssize_t maxsize); 5438 5439 #endif /* __KERNEL__ */ 5440 #endif /* _LINUX_SKBUFF_H */ 5441