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