1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * Definitions for the 'struct sk_buff' memory handlers. 4 * 5 * Authors: 6 * Alan Cox, <gw4pts@gw4pts.ampr.org> 7 * Florian La Roche, <rzsfl@rz.uni-sb.de> 8 */ 9 10 #ifndef _LINUX_SKBUFF_H 11 #define _LINUX_SKBUFF_H 12 13 #include <linux/kernel.h> 14 #include <linux/compiler.h> 15 #include <linux/time.h> 16 #include <linux/bug.h> 17 #include <linux/bvec.h> 18 #include <linux/cache.h> 19 #include <linux/rbtree.h> 20 #include <linux/socket.h> 21 #include <linux/refcount.h> 22 23 #include <linux/atomic.h> 24 #include <asm/types.h> 25 #include <linux/spinlock.h> 26 #include <net/checksum.h> 27 #include <linux/rcupdate.h> 28 #include <linux/dma-mapping.h> 29 #include <linux/netdev_features.h> 30 #include <net/flow_dissector.h> 31 #include <linux/in6.h> 32 #include <linux/if_packet.h> 33 #include <linux/llist.h> 34 #include <linux/page_frag_cache.h> 35 #include <net/flow.h> 36 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 37 #include <linux/netfilter/nf_conntrack_common.h> 38 #endif 39 #include <net/net_debug.h> 40 #include <net/dropreason-core.h> 41 #include <net/netmem.h> 42 43 /** 44 * DOC: skb checksums 45 * 46 * The interface for checksum offload between the stack and networking drivers 47 * is as follows... 48 * 49 * IP checksum related features 50 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 51 * 52 * Drivers advertise checksum offload capabilities in the features of a device. 53 * From the stack's point of view these are capabilities offered by the driver. 54 * A driver typically only advertises features that it is capable of offloading 55 * to its device. 56 * 57 * .. flat-table:: Checksum related device features 58 * :widths: 1 10 59 * 60 * * - %NETIF_F_HW_CSUM 61 * - The driver (or its device) is able to compute one 62 * IP (one's complement) checksum for any combination 63 * of protocols or protocol layering. The checksum is 64 * computed and set in a packet per the CHECKSUM_PARTIAL 65 * interface (see below). 66 * 67 * * - %NETIF_F_IP_CSUM 68 * - Driver (device) is only able to checksum plain 69 * TCP or UDP packets over IPv4. These are specifically 70 * unencapsulated packets of the form IPv4|TCP or 71 * IPv4|UDP where the Protocol field in the IPv4 header 72 * is TCP or UDP. The IPv4 header may contain IP options. 73 * This feature cannot be set in features for a device 74 * with NETIF_F_HW_CSUM also set. This feature is being 75 * DEPRECATED (see below). 76 * 77 * * - %NETIF_F_IPV6_CSUM 78 * - Driver (device) is only able to checksum plain 79 * TCP or UDP packets over IPv6. These are specifically 80 * unencapsulated packets of the form IPv6|TCP or 81 * IPv6|UDP where the Next Header field in the IPv6 82 * header is either TCP or UDP. IPv6 extension headers 83 * are not supported with this feature. This feature 84 * cannot be set in features for a device with 85 * NETIF_F_HW_CSUM also set. This feature is being 86 * DEPRECATED (see below). 87 * 88 * * - %NETIF_F_RXCSUM 89 * - Driver (device) performs receive checksum offload. 90 * This flag is only used to disable the RX checksum 91 * feature for a device. The stack will accept receive 92 * checksum indication in packets received on a device 93 * regardless of whether NETIF_F_RXCSUM is set. 94 * 95 * Checksumming of received packets by device 96 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 97 * 98 * Indication of checksum verification is set in &sk_buff.ip_summed. 99 * Possible values are: 100 * 101 * - %CHECKSUM_NONE 102 * 103 * Device did not checksum this packet e.g. due to lack of capabilities. 104 * The packet contains full (though not verified) checksum in packet but 105 * not in skb->csum. Thus, skb->csum is undefined in this case. 106 * 107 * - %CHECKSUM_UNNECESSARY 108 * 109 * The hardware you're dealing with doesn't calculate the full checksum 110 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums 111 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY 112 * if their checksums are okay. &sk_buff.csum is still undefined in this case 113 * though. A driver or device must never modify the checksum field in the 114 * packet even if checksum is verified. 115 * 116 * %CHECKSUM_UNNECESSARY is applicable to following protocols: 117 * 118 * - TCP: IPv6 and IPv4. 119 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 120 * zero UDP checksum for either IPv4 or IPv6, the networking stack 121 * may perform further validation in this case. 122 * - GRE: only if the checksum is present in the header. 123 * - SCTP: indicates the CRC in SCTP header has been validated. 124 * - FCOE: indicates the CRC in FC frame has been validated. 125 * 126 * &sk_buff.csum_level indicates the number of consecutive checksums found in 127 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY. 128 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 129 * and a device is able to verify the checksums for UDP (possibly zero), 130 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to 131 * two. If the device were only able to verify the UDP checksum and not 132 * GRE, either because it doesn't support GRE checksum or because GRE 133 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 134 * not considered in this case). 135 * 136 * - %CHECKSUM_COMPLETE 137 * 138 * This is the most generic way. The device supplied checksum of the _whole_ 139 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the 140 * hardware doesn't need to parse L3/L4 headers to implement this. 141 * 142 * Notes: 143 * 144 * - Even if device supports only some protocols, but is able to produce 145 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 146 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 147 * 148 * - %CHECKSUM_PARTIAL 149 * 150 * A checksum is set up to be offloaded to a device as described in the 151 * output description for CHECKSUM_PARTIAL. This may occur on a packet 152 * received directly from another Linux OS, e.g., a virtualized Linux kernel 153 * on the same host, or it may be set in the input path in GRO or remote 154 * checksum offload. For the purposes of checksum verification, the checksum 155 * referred to by skb->csum_start + skb->csum_offset and any preceding 156 * checksums in the packet are considered verified. Any checksums in the 157 * packet that are after the checksum being offloaded are not considered to 158 * be verified. 159 * 160 * Checksumming on transmit for non-GSO 161 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 162 * 163 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet. 164 * Values are: 165 * 166 * - %CHECKSUM_PARTIAL 167 * 168 * The driver is required to checksum the packet as seen by hard_start_xmit() 169 * from &sk_buff.csum_start up to the end, and to record/write the checksum at 170 * offset &sk_buff.csum_start + &sk_buff.csum_offset. 171 * A driver may verify that the 172 * csum_start and csum_offset values are valid values given the length and 173 * offset of the packet, but it should not attempt to validate that the 174 * checksum refers to a legitimate transport layer checksum -- it is the 175 * purview of the stack to validate that csum_start and csum_offset are set 176 * correctly. 177 * 178 * When the stack requests checksum offload for a packet, the driver MUST 179 * ensure that the checksum is set correctly. A driver can either offload the 180 * checksum calculation to the device, or call skb_checksum_help (in the case 181 * that the device does not support offload for a particular checksum). 182 * 183 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of 184 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate 185 * checksum offload capability. 186 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based 187 * on network device checksumming capabilities: if a packet does not match 188 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of 189 * &sk_buff.csum_not_inet, see :ref:`crc`) 190 * is called to resolve the checksum. 191 * 192 * - %CHECKSUM_NONE 193 * 194 * The skb was already checksummed by the protocol, or a checksum is not 195 * required. 196 * 197 * - %CHECKSUM_UNNECESSARY 198 * 199 * This has the same meaning as CHECKSUM_NONE for checksum offload on 200 * output. 201 * 202 * - %CHECKSUM_COMPLETE 203 * 204 * Not used in checksum output. If a driver observes a packet with this value 205 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set. 206 * 207 * .. _crc: 208 * 209 * Non-IP checksum (CRC) offloads 210 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 211 * 212 * .. flat-table:: 213 * :widths: 1 10 214 * 215 * * - %NETIF_F_SCTP_CRC 216 * - This feature indicates that a device is capable of 217 * offloading the SCTP CRC in a packet. To perform this offload the stack 218 * will set csum_start and csum_offset accordingly, set ip_summed to 219 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication 220 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c. 221 * A driver that supports both IP checksum offload and SCTP CRC32c offload 222 * must verify which offload is configured for a packet by testing the 223 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to 224 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 225 * 226 * * - %NETIF_F_FCOE_CRC 227 * - This feature indicates that a device is capable of offloading the FCOE 228 * CRC in a packet. To perform this offload the stack will set ip_summed 229 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset 230 * accordingly. Note that there is no indication in the skbuff that the 231 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports 232 * both IP checksum offload and FCOE CRC offload must verify which offload 233 * is configured for a packet, presumably by inspecting packet headers. 234 * 235 * Checksumming on output with GSO 236 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 237 * 238 * In the case of a GSO packet (skb_is_gso() is true), checksum offload 239 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 240 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as 241 * part of the GSO operation is implied. If a checksum is being offloaded 242 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and 243 * csum_offset are set to refer to the outermost checksum being offloaded 244 * (two offloaded checksums are possible with UDP encapsulation). 245 */ 246 247 /* Don't change this without changing skb_csum_unnecessary! */ 248 #define CHECKSUM_NONE 0 249 #define CHECKSUM_UNNECESSARY 1 250 #define CHECKSUM_COMPLETE 2 251 #define CHECKSUM_PARTIAL 3 252 253 /* Maximum value in skb->csum_level */ 254 #define SKB_MAX_CSUM_LEVEL 3 255 256 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 257 #define SKB_WITH_OVERHEAD(X) \ 258 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 259 260 /* For X bytes available in skb->head, what is the minimal 261 * allocation needed, knowing struct skb_shared_info needs 262 * to be aligned. 263 */ 264 #define SKB_HEAD_ALIGN(X) (SKB_DATA_ALIGN(X) + \ 265 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 266 267 #define SKB_MAX_ORDER(X, ORDER) \ 268 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 269 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 270 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 271 272 /* return minimum truesize of one skb containing X bytes of data */ 273 #define SKB_TRUESIZE(X) ((X) + \ 274 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 275 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 276 277 struct net_device; 278 struct scatterlist; 279 struct pipe_inode_info; 280 struct iov_iter; 281 struct napi_struct; 282 struct bpf_prog; 283 union bpf_attr; 284 struct skb_ext; 285 struct ts_config; 286 287 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 288 struct nf_bridge_info { 289 enum { 290 BRNF_PROTO_UNCHANGED, 291 BRNF_PROTO_8021Q, 292 BRNF_PROTO_PPPOE 293 } orig_proto:8; 294 u8 pkt_otherhost:1; 295 u8 in_prerouting:1; 296 u8 bridged_dnat:1; 297 u8 sabotage_in_done:1; 298 __u16 frag_max_size; 299 int physinif; 300 301 /* always valid & non-NULL from FORWARD on, for physdev match */ 302 struct net_device *physoutdev; 303 union { 304 /* prerouting: detect dnat in orig/reply direction */ 305 __be32 ipv4_daddr; 306 struct in6_addr ipv6_daddr; 307 308 /* after prerouting + nat detected: store original source 309 * mac since neigh resolution overwrites it, only used while 310 * skb is out in neigh layer. 311 */ 312 char neigh_header[8]; 313 }; 314 }; 315 #endif 316 317 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 318 /* Chain in tc_skb_ext will be used to share the tc chain with 319 * ovs recirc_id. It will be set to the current chain by tc 320 * and read by ovs to recirc_id. 321 */ 322 struct tc_skb_ext { 323 union { 324 u64 act_miss_cookie; 325 __u32 chain; 326 }; 327 __u16 mru; 328 __u16 zone; 329 u8 post_ct:1; 330 u8 post_ct_snat:1; 331 u8 post_ct_dnat:1; 332 u8 act_miss:1; /* Set if act_miss_cookie is used */ 333 u8 l2_miss:1; /* Set by bridge upon FDB or MDB miss */ 334 }; 335 #endif 336 337 struct sk_buff_head { 338 /* These two members must be first to match sk_buff. */ 339 struct_group_tagged(sk_buff_list, list, 340 struct sk_buff *next; 341 struct sk_buff *prev; 342 ); 343 344 __u32 qlen; 345 spinlock_t lock; 346 }; 347 348 struct sk_buff; 349 350 #ifndef CONFIG_MAX_SKB_FRAGS 351 # define CONFIG_MAX_SKB_FRAGS 17 352 #endif 353 354 #define MAX_SKB_FRAGS CONFIG_MAX_SKB_FRAGS 355 356 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 357 * segment using its current segmentation instead. 358 */ 359 #define GSO_BY_FRAGS 0xFFFF 360 361 typedef struct skb_frag { 362 netmem_ref netmem; 363 unsigned int len; 364 unsigned int offset; 365 } skb_frag_t; 366 367 /** 368 * skb_frag_size() - Returns the size of a skb fragment 369 * @frag: skb fragment 370 */ 371 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 372 { 373 return frag->len; 374 } 375 376 /** 377 * skb_frag_size_set() - Sets the size of a skb fragment 378 * @frag: skb fragment 379 * @size: size of fragment 380 */ 381 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 382 { 383 frag->len = size; 384 } 385 386 /** 387 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 388 * @frag: skb fragment 389 * @delta: value to add 390 */ 391 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 392 { 393 frag->len += delta; 394 } 395 396 /** 397 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 398 * @frag: skb fragment 399 * @delta: value to subtract 400 */ 401 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 402 { 403 frag->len -= delta; 404 } 405 406 /** 407 * skb_frag_must_loop - Test if %p is a high memory page 408 * @p: fragment's page 409 */ 410 static inline bool skb_frag_must_loop(struct page *p) 411 { 412 #if defined(CONFIG_HIGHMEM) 413 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 414 return true; 415 #endif 416 return false; 417 } 418 419 /** 420 * skb_frag_foreach_page - loop over pages in a fragment 421 * 422 * @f: skb frag to operate on 423 * @f_off: offset from start of f->netmem 424 * @f_len: length from f_off to loop over 425 * @p: (temp var) current page 426 * @p_off: (temp var) offset from start of current page, 427 * non-zero only on first page. 428 * @p_len: (temp var) length in current page, 429 * < PAGE_SIZE only on first and last page. 430 * @copied: (temp var) length so far, excluding current p_len. 431 * 432 * A fragment can hold a compound page, in which case per-page 433 * operations, notably kmap_atomic, must be called for each 434 * regular page. 435 */ 436 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 437 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 438 p_off = (f_off) & (PAGE_SIZE - 1), \ 439 p_len = skb_frag_must_loop(p) ? \ 440 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 441 copied = 0; \ 442 copied < f_len; \ 443 copied += p_len, p++, p_off = 0, \ 444 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 445 446 /** 447 * struct skb_shared_hwtstamps - hardware time stamps 448 * @hwtstamp: hardware time stamp transformed into duration 449 * since arbitrary point in time 450 * @netdev_data: address/cookie of network device driver used as 451 * reference to actual hardware time stamp 452 * 453 * Software time stamps generated by ktime_get_real() are stored in 454 * skb->tstamp. 455 * 456 * hwtstamps can only be compared against other hwtstamps from 457 * the same device. 458 * 459 * This structure is attached to packets as part of the 460 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 461 */ 462 struct skb_shared_hwtstamps { 463 union { 464 ktime_t hwtstamp; 465 void *netdev_data; 466 }; 467 }; 468 469 /* Definitions for tx_flags in struct skb_shared_info */ 470 enum { 471 /* generate hardware time stamp */ 472 SKBTX_HW_TSTAMP_NOBPF = 1 << 0, 473 474 /* generate software time stamp when queueing packet to NIC */ 475 SKBTX_SW_TSTAMP = 1 << 1, 476 477 /* device driver is going to provide hardware time stamp */ 478 SKBTX_IN_PROGRESS = 1 << 2, 479 480 /* generate software time stamp on packet tx completion */ 481 SKBTX_COMPLETION_TSTAMP = 1 << 3, 482 483 /* determine hardware time stamp based on time or cycles */ 484 SKBTX_HW_TSTAMP_NETDEV = 1 << 5, 485 486 /* generate software time stamp when entering packet scheduling */ 487 SKBTX_SCHED_TSTAMP = 1 << 6, 488 489 /* used for bpf extension when a bpf program is loaded */ 490 SKBTX_BPF = 1 << 7, 491 }; 492 493 #define SKBTX_HW_TSTAMP (SKBTX_HW_TSTAMP_NOBPF | SKBTX_BPF) 494 495 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 496 SKBTX_SCHED_TSTAMP | \ 497 SKBTX_BPF | \ 498 SKBTX_COMPLETION_TSTAMP) 499 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \ 500 SKBTX_ANY_SW_TSTAMP) 501 502 /* Definitions for flags in struct skb_shared_info */ 503 enum { 504 /* use zcopy routines */ 505 SKBFL_ZEROCOPY_ENABLE = BIT(0), 506 507 /* This indicates at least one fragment might be overwritten 508 * (as in vmsplice(), sendfile() ...) 509 * If we need to compute a TX checksum, we'll need to copy 510 * all frags to avoid possible bad checksum 511 */ 512 SKBFL_SHARED_FRAG = BIT(1), 513 514 /* segment contains only zerocopy data and should not be 515 * charged to the kernel memory. 516 */ 517 SKBFL_PURE_ZEROCOPY = BIT(2), 518 519 SKBFL_DONT_ORPHAN = BIT(3), 520 521 /* page references are managed by the ubuf_info, so it's safe to 522 * use frags only up until ubuf_info is released 523 */ 524 SKBFL_MANAGED_FRAG_REFS = BIT(4), 525 }; 526 527 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG) 528 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \ 529 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS) 530 531 struct ubuf_info_ops { 532 void (*complete)(struct sk_buff *, struct ubuf_info *, 533 bool zerocopy_success); 534 /* has to be compatible with skb_zcopy_set() */ 535 int (*link_skb)(struct sk_buff *skb, struct ubuf_info *uarg); 536 }; 537 538 /* 539 * The callback notifies userspace to release buffers when skb DMA is done in 540 * lower device, the skb last reference should be 0 when calling this. 541 * The zerocopy_success argument is true if zero copy transmit occurred, 542 * false on data copy or out of memory error caused by data copy attempt. 543 * The ctx field is used to track device context. 544 * The desc field is used to track userspace buffer index. 545 */ 546 struct ubuf_info { 547 const struct ubuf_info_ops *ops; 548 refcount_t refcnt; 549 u8 flags; 550 }; 551 552 struct ubuf_info_msgzc { 553 struct ubuf_info ubuf; 554 555 union { 556 struct { 557 unsigned long desc; 558 void *ctx; 559 }; 560 struct { 561 u32 id; 562 u16 len; 563 u16 zerocopy:1; 564 u32 bytelen; 565 }; 566 }; 567 568 struct mmpin { 569 struct user_struct *user; 570 unsigned int num_pg; 571 } mmp; 572 }; 573 574 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 575 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \ 576 ubuf) 577 578 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 579 void mm_unaccount_pinned_pages(struct mmpin *mmp); 580 581 /* Preserve some data across TX submission and completion. 582 * 583 * Note, this state is stored in the driver. Extending the layout 584 * might need some special care. 585 */ 586 struct xsk_tx_metadata_compl { 587 __u64 *tx_timestamp; 588 }; 589 590 /* This data is invariant across clones and lives at 591 * the end of the header data, ie. at skb->end. 592 */ 593 struct skb_shared_info { 594 __u8 flags; 595 __u8 meta_len; 596 __u8 nr_frags; 597 __u8 tx_flags; 598 unsigned short gso_size; 599 /* Warning: this field is not always filled in (UFO)! */ 600 unsigned short gso_segs; 601 struct sk_buff *frag_list; 602 union { 603 struct skb_shared_hwtstamps hwtstamps; 604 struct xsk_tx_metadata_compl xsk_meta; 605 }; 606 unsigned int gso_type; 607 u32 tskey; 608 609 /* 610 * Warning : all fields before dataref are cleared in __alloc_skb() 611 */ 612 atomic_t dataref; 613 614 union { 615 struct { 616 u32 xdp_frags_size; 617 u32 xdp_frags_truesize; 618 }; 619 620 /* 621 * Intermediate layers must ensure that destructor_arg 622 * remains valid until skb destructor. 623 */ 624 void *destructor_arg; 625 }; 626 627 /* must be last field, see pskb_expand_head() */ 628 skb_frag_t frags[MAX_SKB_FRAGS]; 629 }; 630 631 /** 632 * DOC: dataref and headerless skbs 633 * 634 * Transport layers send out clones of payload skbs they hold for 635 * retransmissions. To allow lower layers of the stack to prepend their headers 636 * we split &skb_shared_info.dataref into two halves. 637 * The lower 16 bits count the overall number of references. 638 * The higher 16 bits indicate how many of the references are payload-only. 639 * skb_header_cloned() checks if skb is allowed to add / write the headers. 640 * 641 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr 642 * (via __skb_header_release()). Any clone created from marked skb will get 643 * &sk_buff.hdr_len populated with the available headroom. 644 * If there's the only clone in existence it's able to modify the headroom 645 * at will. The sequence of calls inside the transport layer is:: 646 * 647 * <alloc skb> 648 * skb_reserve() 649 * __skb_header_release() 650 * skb_clone() 651 * // send the clone down the stack 652 * 653 * This is not a very generic construct and it depends on the transport layers 654 * doing the right thing. In practice there's usually only one payload-only skb. 655 * Having multiple payload-only skbs with different lengths of hdr_len is not 656 * possible. The payload-only skbs should never leave their owner. 657 */ 658 #define SKB_DATAREF_SHIFT 16 659 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 660 661 662 enum { 663 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 664 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 665 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 666 }; 667 668 enum { 669 SKB_GSO_TCPV4 = 1 << 0, 670 671 /* This indicates the skb is from an untrusted source. */ 672 SKB_GSO_DODGY = 1 << 1, 673 674 /* This indicates the tcp segment has CWR set. */ 675 SKB_GSO_TCP_ECN = 1 << 2, 676 677 SKB_GSO_TCP_FIXEDID = 1 << 3, 678 679 SKB_GSO_TCPV6 = 1 << 4, 680 681 SKB_GSO_FCOE = 1 << 5, 682 683 SKB_GSO_GRE = 1 << 6, 684 685 SKB_GSO_GRE_CSUM = 1 << 7, 686 687 SKB_GSO_IPXIP4 = 1 << 8, 688 689 SKB_GSO_IPXIP6 = 1 << 9, 690 691 SKB_GSO_UDP_TUNNEL = 1 << 10, 692 693 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 694 695 SKB_GSO_PARTIAL = 1 << 12, 696 697 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 698 699 SKB_GSO_SCTP = 1 << 14, 700 701 SKB_GSO_ESP = 1 << 15, 702 703 SKB_GSO_UDP = 1 << 16, 704 705 SKB_GSO_UDP_L4 = 1 << 17, 706 707 SKB_GSO_FRAGLIST = 1 << 18, 708 709 SKB_GSO_TCP_ACCECN = 1 << 19, 710 }; 711 712 #if BITS_PER_LONG > 32 713 #define NET_SKBUFF_DATA_USES_OFFSET 1 714 #endif 715 716 #ifdef NET_SKBUFF_DATA_USES_OFFSET 717 typedef unsigned int sk_buff_data_t; 718 #else 719 typedef unsigned char *sk_buff_data_t; 720 #endif 721 722 enum skb_tstamp_type { 723 SKB_CLOCK_REALTIME, 724 SKB_CLOCK_MONOTONIC, 725 SKB_CLOCK_TAI, 726 __SKB_CLOCK_MAX = SKB_CLOCK_TAI, 727 }; 728 729 /** 730 * DOC: Basic sk_buff geometry 731 * 732 * struct sk_buff itself is a metadata structure and does not hold any packet 733 * data. All the data is held in associated buffers. 734 * 735 * &sk_buff.head points to the main "head" buffer. The head buffer is divided 736 * into two parts: 737 * 738 * - data buffer, containing headers and sometimes payload; 739 * this is the part of the skb operated on by the common helpers 740 * such as skb_put() or skb_pull(); 741 * - shared info (struct skb_shared_info) which holds an array of pointers 742 * to read-only data in the (page, offset, length) format. 743 * 744 * Optionally &skb_shared_info.frag_list may point to another skb. 745 * 746 * Basic diagram may look like this:: 747 * 748 * --------------- 749 * | sk_buff | 750 * --------------- 751 * ,--------------------------- + head 752 * / ,----------------- + data 753 * / / ,----------- + tail 754 * | | | , + end 755 * | | | | 756 * v v v v 757 * ----------------------------------------------- 758 * | headroom | data | tailroom | skb_shared_info | 759 * ----------------------------------------------- 760 * + [page frag] 761 * + [page frag] 762 * + [page frag] 763 * + [page frag] --------- 764 * + frag_list --> | sk_buff | 765 * --------- 766 * 767 */ 768 769 /** 770 * struct sk_buff - socket buffer 771 * @next: Next buffer in list 772 * @prev: Previous buffer in list 773 * @tstamp: Time we arrived/left 774 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 775 * for retransmit timer 776 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 777 * @list: queue head 778 * @ll_node: anchor in an llist (eg socket defer_list) 779 * @sk: Socket we are owned by 780 * @dev: Device we arrived on/are leaving by 781 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 782 * @cb: Control buffer. Free for use by every layer. Put private vars here 783 * @_skb_refdst: destination entry (with norefcount bit) 784 * @len: Length of actual data 785 * @data_len: Data length 786 * @mac_len: Length of link layer header 787 * @hdr_len: writable header length of cloned skb 788 * @csum: Checksum (must include start/offset pair) 789 * @csum_start: Offset from skb->head where checksumming should start 790 * @csum_offset: Offset from csum_start where checksum should be stored 791 * @priority: Packet queueing priority 792 * @ignore_df: allow local fragmentation 793 * @cloned: Head may be cloned (check refcnt to be sure) 794 * @ip_summed: Driver fed us an IP checksum 795 * @nohdr: Payload reference only, must not modify header 796 * @pkt_type: Packet class 797 * @fclone: skbuff clone status 798 * @ipvs_property: skbuff is owned by ipvs 799 * @inner_protocol_type: whether the inner protocol is 800 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 801 * @remcsum_offload: remote checksum offload is enabled 802 * @offload_fwd_mark: Packet was L2-forwarded in hardware 803 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 804 * @tc_skip_classify: do not classify packet. set by IFB device 805 * @tc_at_ingress: used within tc_classify to distinguish in/egress 806 * @redirected: packet was redirected by packet classifier 807 * @from_ingress: packet was redirected from the ingress path 808 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 809 * @peeked: this packet has been seen already, so stats have been 810 * done for it, don't do them again 811 * @nf_trace: netfilter packet trace flag 812 * @protocol: Packet protocol from driver 813 * @destructor: Destruct function 814 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 815 * @_sk_redir: socket redirection information for skmsg 816 * @_nfct: Associated connection, if any (with nfctinfo bits) 817 * @skb_iif: ifindex of device we arrived on 818 * @tc_index: Traffic control index 819 * @hash: the packet hash 820 * @queue_mapping: Queue mapping for multiqueue devices 821 * @head_frag: skb was allocated from page fragments, 822 * not allocated by kmalloc() or vmalloc(). 823 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 824 * @pp_recycle: mark the packet for recycling instead of freeing (implies 825 * page_pool support on driver) 826 * @active_extensions: active extensions (skb_ext_id types) 827 * @ndisc_nodetype: router type (from link layer) 828 * @ooo_okay: allow the mapping of a socket to a queue to be changed 829 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 830 * ports. 831 * @sw_hash: indicates hash was computed in software stack 832 * @wifi_acked_valid: wifi_acked was set 833 * @wifi_acked: whether frame was acked on wifi or not 834 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 835 * @encapsulation: indicates the inner headers in the skbuff are valid 836 * @encap_hdr_csum: software checksum is needed 837 * @csum_valid: checksum is already valid 838 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 839 * @csum_complete_sw: checksum was completed by software 840 * @csum_level: indicates the number of consecutive checksums found in 841 * the packet minus one that have been verified as 842 * CHECKSUM_UNNECESSARY (max 3) 843 * @unreadable: indicates that at least 1 of the fragments in this skb is 844 * unreadable. 845 * @dst_pending_confirm: need to confirm neighbour 846 * @decrypted: Decrypted SKB 847 * @slow_gro: state present at GRO time, slower prepare step required 848 * @tstamp_type: When set, skb->tstamp has the 849 * delivery_time clock base of skb->tstamp. 850 * @napi_id: id of the NAPI struct this skb came from 851 * @sender_cpu: (aka @napi_id) source CPU in XPS 852 * @alloc_cpu: CPU which did the skb allocation. 853 * @secmark: security marking 854 * @mark: Generic packet mark 855 * @reserved_tailroom: (aka @mark) number of bytes of free space available 856 * at the tail of an sk_buff 857 * @vlan_all: vlan fields (proto & tci) 858 * @vlan_proto: vlan encapsulation protocol 859 * @vlan_tci: vlan tag control information 860 * @inner_protocol: Protocol (encapsulation) 861 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 862 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 863 * @inner_transport_header: Inner transport layer header (encapsulation) 864 * @inner_network_header: Network layer header (encapsulation) 865 * @inner_mac_header: Link layer header (encapsulation) 866 * @transport_header: Transport layer header 867 * @network_header: Network layer header 868 * @mac_header: Link layer header 869 * @kcov_handle: KCOV remote handle for remote coverage collection 870 * @tail: Tail pointer 871 * @end: End pointer 872 * @head: Head of buffer 873 * @data: Data head pointer 874 * @truesize: Buffer size 875 * @users: User count - see {datagram,tcp}.c 876 * @extensions: allocated extensions, valid if active_extensions is nonzero 877 */ 878 879 struct sk_buff { 880 union { 881 struct { 882 /* These two members must be first to match sk_buff_head. */ 883 struct sk_buff *next; 884 struct sk_buff *prev; 885 886 union { 887 struct net_device *dev; 888 /* Some protocols might use this space to store information, 889 * while device pointer would be NULL. 890 * UDP receive path is one user. 891 */ 892 unsigned long dev_scratch; 893 }; 894 }; 895 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 896 struct list_head list; 897 struct llist_node ll_node; 898 }; 899 900 struct sock *sk; 901 902 union { 903 ktime_t tstamp; 904 u64 skb_mstamp_ns; /* earliest departure time */ 905 }; 906 /* 907 * This is the control buffer. It is free to use for every 908 * layer. Please put your private variables there. If you 909 * want to keep them across layers you have to do a skb_clone() 910 * first. This is owned by whoever has the skb queued ATM. 911 */ 912 char cb[48] __aligned(8); 913 914 union { 915 struct { 916 unsigned long _skb_refdst; 917 void (*destructor)(struct sk_buff *skb); 918 }; 919 struct list_head tcp_tsorted_anchor; 920 #ifdef CONFIG_NET_SOCK_MSG 921 unsigned long _sk_redir; 922 #endif 923 }; 924 925 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 926 unsigned long _nfct; 927 #endif 928 unsigned int len, 929 data_len; 930 __u16 mac_len, 931 hdr_len; 932 933 /* Following fields are _not_ copied in __copy_skb_header() 934 * Note that queue_mapping is here mostly to fill a hole. 935 */ 936 __u16 queue_mapping; 937 938 /* if you move cloned around you also must adapt those constants */ 939 #ifdef __BIG_ENDIAN_BITFIELD 940 #define CLONED_MASK (1 << 7) 941 #else 942 #define CLONED_MASK 1 943 #endif 944 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 945 946 /* private: */ 947 __u8 __cloned_offset[0]; 948 /* public: */ 949 __u8 cloned:1, 950 nohdr:1, 951 fclone:2, 952 peeked:1, 953 head_frag:1, 954 pfmemalloc:1, 955 pp_recycle:1; /* page_pool recycle indicator */ 956 #ifdef CONFIG_SKB_EXTENSIONS 957 __u8 active_extensions; 958 #endif 959 960 /* Fields enclosed in headers group are copied 961 * using a single memcpy() in __copy_skb_header() 962 */ 963 struct_group(headers, 964 965 /* private: */ 966 __u8 __pkt_type_offset[0]; 967 /* public: */ 968 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 969 __u8 ignore_df:1; 970 __u8 dst_pending_confirm:1; 971 __u8 ip_summed:2; 972 __u8 ooo_okay:1; 973 974 /* private: */ 975 __u8 __mono_tc_offset[0]; 976 /* public: */ 977 __u8 tstamp_type:2; /* See skb_tstamp_type */ 978 #ifdef CONFIG_NET_XGRESS 979 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 980 __u8 tc_skip_classify:1; 981 #endif 982 __u8 remcsum_offload:1; 983 __u8 csum_complete_sw:1; 984 __u8 csum_level:2; 985 __u8 inner_protocol_type:1; 986 987 __u8 l4_hash:1; 988 __u8 sw_hash:1; 989 #ifdef CONFIG_WIRELESS 990 __u8 wifi_acked_valid:1; 991 __u8 wifi_acked:1; 992 #endif 993 __u8 no_fcs:1; 994 /* Indicates the inner headers are valid in the skbuff. */ 995 __u8 encapsulation:1; 996 __u8 encap_hdr_csum:1; 997 __u8 csum_valid:1; 998 #ifdef CONFIG_IPV6_NDISC_NODETYPE 999 __u8 ndisc_nodetype:2; 1000 #endif 1001 1002 #if IS_ENABLED(CONFIG_IP_VS) 1003 __u8 ipvs_property:1; 1004 #endif 1005 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 1006 __u8 nf_trace:1; 1007 #endif 1008 #ifdef CONFIG_NET_SWITCHDEV 1009 __u8 offload_fwd_mark:1; 1010 __u8 offload_l3_fwd_mark:1; 1011 #endif 1012 __u8 redirected:1; 1013 #ifdef CONFIG_NET_REDIRECT 1014 __u8 from_ingress:1; 1015 #endif 1016 #ifdef CONFIG_NETFILTER_SKIP_EGRESS 1017 __u8 nf_skip_egress:1; 1018 #endif 1019 #ifdef CONFIG_SKB_DECRYPTED 1020 __u8 decrypted:1; 1021 #endif 1022 __u8 slow_gro:1; 1023 #if IS_ENABLED(CONFIG_IP_SCTP) 1024 __u8 csum_not_inet:1; 1025 #endif 1026 __u8 unreadable:1; 1027 #if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS) 1028 __u16 tc_index; /* traffic control index */ 1029 #endif 1030 1031 u16 alloc_cpu; 1032 1033 union { 1034 __wsum csum; 1035 struct { 1036 __u16 csum_start; 1037 __u16 csum_offset; 1038 }; 1039 }; 1040 __u32 priority; 1041 int skb_iif; 1042 __u32 hash; 1043 union { 1044 u32 vlan_all; 1045 struct { 1046 __be16 vlan_proto; 1047 __u16 vlan_tci; 1048 }; 1049 }; 1050 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1051 union { 1052 unsigned int napi_id; 1053 unsigned int sender_cpu; 1054 }; 1055 #endif 1056 #ifdef CONFIG_NETWORK_SECMARK 1057 __u32 secmark; 1058 #endif 1059 1060 union { 1061 __u32 mark; 1062 __u32 reserved_tailroom; 1063 }; 1064 1065 union { 1066 __be16 inner_protocol; 1067 __u8 inner_ipproto; 1068 }; 1069 1070 __u16 inner_transport_header; 1071 __u16 inner_network_header; 1072 __u16 inner_mac_header; 1073 1074 __be16 protocol; 1075 __u16 transport_header; 1076 __u16 network_header; 1077 __u16 mac_header; 1078 1079 #ifdef CONFIG_KCOV 1080 u64 kcov_handle; 1081 #endif 1082 1083 ); /* end headers group */ 1084 1085 /* These elements must be at the end, see alloc_skb() for details. */ 1086 sk_buff_data_t tail; 1087 sk_buff_data_t end; 1088 unsigned char *head, 1089 *data; 1090 unsigned int truesize; 1091 refcount_t users; 1092 1093 #ifdef CONFIG_SKB_EXTENSIONS 1094 /* only usable after checking ->active_extensions != 0 */ 1095 struct skb_ext *extensions; 1096 #endif 1097 }; 1098 1099 /* if you move pkt_type around you also must adapt those constants */ 1100 #ifdef __BIG_ENDIAN_BITFIELD 1101 #define PKT_TYPE_MAX (7 << 5) 1102 #else 1103 #define PKT_TYPE_MAX 7 1104 #endif 1105 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1106 1107 /* if you move tc_at_ingress or tstamp_type 1108 * around, you also must adapt these constants. 1109 */ 1110 #ifdef __BIG_ENDIAN_BITFIELD 1111 #define SKB_TSTAMP_TYPE_MASK (3 << 6) 1112 #define SKB_TSTAMP_TYPE_RSHIFT (6) 1113 #define TC_AT_INGRESS_MASK (1 << 5) 1114 #else 1115 #define SKB_TSTAMP_TYPE_MASK (3) 1116 #define TC_AT_INGRESS_MASK (1 << 2) 1117 #endif 1118 #define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset) 1119 1120 #ifdef __KERNEL__ 1121 /* 1122 * Handling routines are only of interest to the kernel 1123 */ 1124 1125 #define SKB_ALLOC_FCLONE 0x01 1126 #define SKB_ALLOC_RX 0x02 1127 #define SKB_ALLOC_NAPI 0x04 1128 1129 /** 1130 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1131 * @skb: buffer 1132 */ 1133 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1134 { 1135 return unlikely(skb->pfmemalloc); 1136 } 1137 1138 /* 1139 * skb might have a dst pointer attached, refcounted or not. 1140 * _skb_refdst low order bit is set if refcount was _not_ taken 1141 */ 1142 #define SKB_DST_NOREF 1UL 1143 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1144 1145 /** 1146 * skb_dst - returns skb dst_entry 1147 * @skb: buffer 1148 * 1149 * Returns: skb dst_entry, regardless of reference taken or not. 1150 */ 1151 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1152 { 1153 /* If refdst was not refcounted, check we still are in a 1154 * rcu_read_lock section 1155 */ 1156 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1157 !rcu_read_lock_held() && 1158 !rcu_read_lock_bh_held()); 1159 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1160 } 1161 1162 /** 1163 * skb_dst_set - sets skb dst 1164 * @skb: buffer 1165 * @dst: dst entry 1166 * 1167 * Sets skb dst, assuming a reference was taken on dst and should 1168 * be released by skb_dst_drop() 1169 */ 1170 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1171 { 1172 skb->slow_gro |= !!dst; 1173 skb->_skb_refdst = (unsigned long)dst; 1174 } 1175 1176 /** 1177 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1178 * @skb: buffer 1179 * @dst: dst entry 1180 * 1181 * Sets skb dst, assuming a reference was not taken on dst. 1182 * If dst entry is cached, we do not take reference and dst_release 1183 * will be avoided by refdst_drop. If dst entry is not cached, we take 1184 * reference, so that last dst_release can destroy the dst immediately. 1185 */ 1186 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1187 { 1188 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1189 skb->slow_gro |= !!dst; 1190 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1191 } 1192 1193 /** 1194 * skb_dst_is_noref - Test if skb dst isn't refcounted 1195 * @skb: buffer 1196 */ 1197 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1198 { 1199 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1200 } 1201 1202 /* For mangling skb->pkt_type from user space side from applications 1203 * such as nft, tc, etc, we only allow a conservative subset of 1204 * possible pkt_types to be set. 1205 */ 1206 static inline bool skb_pkt_type_ok(u32 ptype) 1207 { 1208 return ptype <= PACKET_OTHERHOST; 1209 } 1210 1211 /** 1212 * skb_napi_id - Returns the skb's NAPI id 1213 * @skb: buffer 1214 */ 1215 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1216 { 1217 #ifdef CONFIG_NET_RX_BUSY_POLL 1218 return skb->napi_id; 1219 #else 1220 return 0; 1221 #endif 1222 } 1223 1224 static inline bool skb_wifi_acked_valid(const struct sk_buff *skb) 1225 { 1226 #ifdef CONFIG_WIRELESS 1227 return skb->wifi_acked_valid; 1228 #else 1229 return 0; 1230 #endif 1231 } 1232 1233 /** 1234 * skb_unref - decrement the skb's reference count 1235 * @skb: buffer 1236 * 1237 * Returns: true if we can free the skb. 1238 */ 1239 static inline bool skb_unref(struct sk_buff *skb) 1240 { 1241 if (unlikely(!skb)) 1242 return false; 1243 if (!IS_ENABLED(CONFIG_DEBUG_NET) && likely(refcount_read(&skb->users) == 1)) 1244 smp_rmb(); 1245 else if (likely(!refcount_dec_and_test(&skb->users))) 1246 return false; 1247 1248 return true; 1249 } 1250 1251 static inline bool skb_data_unref(const struct sk_buff *skb, 1252 struct skb_shared_info *shinfo) 1253 { 1254 int bias; 1255 1256 if (!skb->cloned) 1257 return true; 1258 1259 bias = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1; 1260 1261 if (atomic_read(&shinfo->dataref) == bias) 1262 smp_rmb(); 1263 else if (atomic_sub_return(bias, &shinfo->dataref)) 1264 return false; 1265 1266 return true; 1267 } 1268 1269 void __fix_address sk_skb_reason_drop(struct sock *sk, struct sk_buff *skb, 1270 enum skb_drop_reason reason); 1271 1272 static inline void 1273 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason) 1274 { 1275 sk_skb_reason_drop(NULL, skb, reason); 1276 } 1277 1278 /** 1279 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1280 * @skb: buffer to free 1281 */ 1282 static inline void kfree_skb(struct sk_buff *skb) 1283 { 1284 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1285 } 1286 1287 void skb_release_head_state(struct sk_buff *skb); 1288 void kfree_skb_list_reason(struct sk_buff *segs, 1289 enum skb_drop_reason reason); 1290 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1291 void skb_tx_error(struct sk_buff *skb); 1292 1293 static inline void kfree_skb_list(struct sk_buff *segs) 1294 { 1295 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1296 } 1297 1298 #ifdef CONFIG_TRACEPOINTS 1299 void consume_skb(struct sk_buff *skb); 1300 #else 1301 static inline void consume_skb(struct sk_buff *skb) 1302 { 1303 return kfree_skb(skb); 1304 } 1305 #endif 1306 1307 void __consume_stateless_skb(struct sk_buff *skb); 1308 void __kfree_skb(struct sk_buff *skb); 1309 1310 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1311 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1312 bool *fragstolen, int *delta_truesize); 1313 1314 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1315 int node); 1316 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1317 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1318 struct sk_buff *build_skb_around(struct sk_buff *skb, 1319 void *data, unsigned int frag_size); 1320 void skb_attempt_defer_free(struct sk_buff *skb); 1321 1322 u32 napi_skb_cache_get_bulk(void **skbs, u32 n); 1323 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1324 struct sk_buff *slab_build_skb(void *data); 1325 1326 /** 1327 * alloc_skb - allocate a network buffer 1328 * @size: size to allocate 1329 * @priority: allocation mask 1330 * 1331 * This function is a convenient wrapper around __alloc_skb(). 1332 */ 1333 static inline struct sk_buff *alloc_skb(unsigned int size, 1334 gfp_t priority) 1335 { 1336 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1337 } 1338 1339 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1340 unsigned long data_len, 1341 int max_page_order, 1342 int *errcode, 1343 gfp_t gfp_mask); 1344 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1345 1346 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1347 struct sk_buff_fclones { 1348 struct sk_buff skb1; 1349 1350 struct sk_buff skb2; 1351 1352 refcount_t fclone_ref; 1353 }; 1354 1355 /** 1356 * skb_fclone_busy - check if fclone is busy 1357 * @sk: socket 1358 * @skb: buffer 1359 * 1360 * Returns: true if skb is a fast clone, and its clone is not freed. 1361 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1362 * so we also check that didn't happen. 1363 */ 1364 static inline bool skb_fclone_busy(const struct sock *sk, 1365 const struct sk_buff *skb) 1366 { 1367 const struct sk_buff_fclones *fclones; 1368 1369 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1370 1371 return skb->fclone == SKB_FCLONE_ORIG && 1372 refcount_read(&fclones->fclone_ref) > 1 && 1373 READ_ONCE(fclones->skb2.sk) == sk; 1374 } 1375 1376 /** 1377 * alloc_skb_fclone - allocate a network buffer from fclone cache 1378 * @size: size to allocate 1379 * @priority: allocation mask 1380 * 1381 * This function is a convenient wrapper around __alloc_skb(). 1382 */ 1383 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1384 gfp_t priority) 1385 { 1386 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1387 } 1388 1389 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1390 void skb_headers_offset_update(struct sk_buff *skb, int off); 1391 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1392 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1393 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1394 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1395 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1396 gfp_t gfp_mask, bool fclone); 1397 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1398 gfp_t gfp_mask) 1399 { 1400 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1401 } 1402 1403 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1404 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1405 unsigned int headroom); 1406 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1407 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1408 int newtailroom, gfp_t priority); 1409 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1410 int offset, int len); 1411 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1412 int offset, int len); 1413 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1414 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1415 1416 /** 1417 * skb_pad - zero pad the tail of an skb 1418 * @skb: buffer to pad 1419 * @pad: space to pad 1420 * 1421 * Ensure that a buffer is followed by a padding area that is zero 1422 * filled. Used by network drivers which may DMA or transfer data 1423 * beyond the buffer end onto the wire. 1424 * 1425 * May return error in out of memory cases. The skb is freed on error. 1426 */ 1427 static inline int skb_pad(struct sk_buff *skb, int pad) 1428 { 1429 return __skb_pad(skb, pad, true); 1430 } 1431 #define dev_kfree_skb(a) consume_skb(a) 1432 1433 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1434 int offset, size_t size, size_t max_frags); 1435 1436 struct skb_seq_state { 1437 __u32 lower_offset; 1438 __u32 upper_offset; 1439 __u32 frag_idx; 1440 __u32 stepped_offset; 1441 struct sk_buff *root_skb; 1442 struct sk_buff *cur_skb; 1443 __u8 *frag_data; 1444 __u32 frag_off; 1445 }; 1446 1447 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1448 unsigned int to, struct skb_seq_state *st); 1449 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1450 struct skb_seq_state *st); 1451 void skb_abort_seq_read(struct skb_seq_state *st); 1452 int skb_copy_seq_read(struct skb_seq_state *st, int offset, void *to, int len); 1453 1454 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1455 unsigned int to, struct ts_config *config); 1456 1457 /* 1458 * Packet hash types specify the type of hash in skb_set_hash. 1459 * 1460 * Hash types refer to the protocol layer addresses which are used to 1461 * construct a packet's hash. The hashes are used to differentiate or identify 1462 * flows of the protocol layer for the hash type. Hash types are either 1463 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1464 * 1465 * Properties of hashes: 1466 * 1467 * 1) Two packets in different flows have different hash values 1468 * 2) Two packets in the same flow should have the same hash value 1469 * 1470 * A hash at a higher layer is considered to be more specific. A driver should 1471 * set the most specific hash possible. 1472 * 1473 * A driver cannot indicate a more specific hash than the layer at which a hash 1474 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1475 * 1476 * A driver may indicate a hash level which is less specific than the 1477 * actual layer the hash was computed on. For instance, a hash computed 1478 * at L4 may be considered an L3 hash. This should only be done if the 1479 * driver can't unambiguously determine that the HW computed the hash at 1480 * the higher layer. Note that the "should" in the second property above 1481 * permits this. 1482 */ 1483 enum pkt_hash_types { 1484 PKT_HASH_TYPE_NONE, /* Undefined type */ 1485 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1486 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1487 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1488 }; 1489 1490 static inline void skb_clear_hash(struct sk_buff *skb) 1491 { 1492 skb->hash = 0; 1493 skb->sw_hash = 0; 1494 skb->l4_hash = 0; 1495 } 1496 1497 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1498 { 1499 if (!skb->l4_hash) 1500 skb_clear_hash(skb); 1501 } 1502 1503 static inline void 1504 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1505 { 1506 skb->l4_hash = is_l4; 1507 skb->sw_hash = is_sw; 1508 skb->hash = hash; 1509 } 1510 1511 static inline void 1512 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1513 { 1514 /* Used by drivers to set hash from HW */ 1515 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1516 } 1517 1518 static inline void 1519 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1520 { 1521 __skb_set_hash(skb, hash, true, is_l4); 1522 } 1523 1524 u32 __skb_get_hash_symmetric_net(const struct net *net, const struct sk_buff *skb); 1525 1526 static inline u32 __skb_get_hash_symmetric(const struct sk_buff *skb) 1527 { 1528 return __skb_get_hash_symmetric_net(NULL, skb); 1529 } 1530 1531 void __skb_get_hash_net(const struct net *net, struct sk_buff *skb); 1532 u32 skb_get_poff(const struct sk_buff *skb); 1533 u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1534 const struct flow_keys_basic *keys, int hlen); 1535 __be32 skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1536 const void *data, int hlen_proto); 1537 1538 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1539 const struct flow_dissector_key *key, 1540 unsigned int key_count); 1541 1542 struct bpf_flow_dissector; 1543 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1544 __be16 proto, int nhoff, int hlen, unsigned int flags); 1545 1546 bool __skb_flow_dissect(const struct net *net, 1547 const struct sk_buff *skb, 1548 struct flow_dissector *flow_dissector, 1549 void *target_container, const void *data, 1550 __be16 proto, int nhoff, int hlen, unsigned int flags); 1551 1552 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1553 struct flow_dissector *flow_dissector, 1554 void *target_container, unsigned int flags) 1555 { 1556 return __skb_flow_dissect(NULL, skb, flow_dissector, 1557 target_container, NULL, 0, 0, 0, flags); 1558 } 1559 1560 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1561 struct flow_keys *flow, 1562 unsigned int flags) 1563 { 1564 memset(flow, 0, sizeof(*flow)); 1565 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1566 flow, NULL, 0, 0, 0, flags); 1567 } 1568 1569 static inline bool 1570 skb_flow_dissect_flow_keys_basic(const struct net *net, 1571 const struct sk_buff *skb, 1572 struct flow_keys_basic *flow, 1573 const void *data, __be16 proto, 1574 int nhoff, int hlen, unsigned int flags) 1575 { 1576 memset(flow, 0, sizeof(*flow)); 1577 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1578 data, proto, nhoff, hlen, flags); 1579 } 1580 1581 void skb_flow_dissect_meta(const struct sk_buff *skb, 1582 struct flow_dissector *flow_dissector, 1583 void *target_container); 1584 1585 /* Gets a skb connection tracking info, ctinfo map should be a 1586 * map of mapsize to translate enum ip_conntrack_info states 1587 * to user states. 1588 */ 1589 void 1590 skb_flow_dissect_ct(const struct sk_buff *skb, 1591 struct flow_dissector *flow_dissector, 1592 void *target_container, 1593 u16 *ctinfo_map, size_t mapsize, 1594 bool post_ct, u16 zone); 1595 void 1596 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1597 struct flow_dissector *flow_dissector, 1598 void *target_container); 1599 1600 void skb_flow_dissect_hash(const struct sk_buff *skb, 1601 struct flow_dissector *flow_dissector, 1602 void *target_container); 1603 1604 static inline __u32 skb_get_hash_net(const struct net *net, struct sk_buff *skb) 1605 { 1606 if (!skb->l4_hash && !skb->sw_hash) 1607 __skb_get_hash_net(net, skb); 1608 1609 return skb->hash; 1610 } 1611 1612 static inline __u32 skb_get_hash(struct sk_buff *skb) 1613 { 1614 if (!skb->l4_hash && !skb->sw_hash) 1615 __skb_get_hash_net(NULL, skb); 1616 1617 return skb->hash; 1618 } 1619 1620 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1621 { 1622 if (!skb->l4_hash && !skb->sw_hash) { 1623 struct flow_keys keys; 1624 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1625 1626 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1627 } 1628 1629 return skb->hash; 1630 } 1631 1632 __u32 skb_get_hash_perturb(const struct sk_buff *skb, 1633 const siphash_key_t *perturb); 1634 1635 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1636 { 1637 return skb->hash; 1638 } 1639 1640 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1641 { 1642 to->hash = from->hash; 1643 to->sw_hash = from->sw_hash; 1644 to->l4_hash = from->l4_hash; 1645 }; 1646 1647 static inline int skb_cmp_decrypted(const struct sk_buff *skb1, 1648 const struct sk_buff *skb2) 1649 { 1650 #ifdef CONFIG_SKB_DECRYPTED 1651 return skb2->decrypted - skb1->decrypted; 1652 #else 1653 return 0; 1654 #endif 1655 } 1656 1657 static inline bool skb_is_decrypted(const struct sk_buff *skb) 1658 { 1659 #ifdef CONFIG_SKB_DECRYPTED 1660 return skb->decrypted; 1661 #else 1662 return false; 1663 #endif 1664 } 1665 1666 static inline void skb_copy_decrypted(struct sk_buff *to, 1667 const struct sk_buff *from) 1668 { 1669 #ifdef CONFIG_SKB_DECRYPTED 1670 to->decrypted = from->decrypted; 1671 #endif 1672 } 1673 1674 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1675 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1676 { 1677 return skb->head + skb->end; 1678 } 1679 1680 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1681 { 1682 return skb->end; 1683 } 1684 1685 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1686 { 1687 skb->end = offset; 1688 } 1689 #else 1690 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1691 { 1692 return skb->end; 1693 } 1694 1695 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1696 { 1697 return skb->end - skb->head; 1698 } 1699 1700 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1701 { 1702 skb->end = skb->head + offset; 1703 } 1704 #endif 1705 1706 extern const struct ubuf_info_ops msg_zerocopy_ubuf_ops; 1707 1708 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1709 struct ubuf_info *uarg, bool devmem); 1710 1711 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1712 1713 struct net_devmem_dmabuf_binding; 1714 1715 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, 1716 struct sk_buff *skb, struct iov_iter *from, 1717 size_t length, 1718 struct net_devmem_dmabuf_binding *binding); 1719 1720 int zerocopy_fill_skb_from_iter(struct sk_buff *skb, 1721 struct iov_iter *from, size_t length); 1722 1723 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1724 struct msghdr *msg, int len) 1725 { 1726 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len, 1727 NULL); 1728 } 1729 1730 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1731 struct msghdr *msg, int len, 1732 struct ubuf_info *uarg, 1733 struct net_devmem_dmabuf_binding *binding); 1734 1735 /* Internal */ 1736 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1737 1738 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1739 { 1740 return &skb_shinfo(skb)->hwtstamps; 1741 } 1742 1743 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1744 { 1745 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1746 1747 return is_zcopy ? skb_uarg(skb) : NULL; 1748 } 1749 1750 static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1751 { 1752 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1753 } 1754 1755 static inline bool skb_zcopy_managed(const struct sk_buff *skb) 1756 { 1757 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS; 1758 } 1759 1760 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1761 const struct sk_buff *skb2) 1762 { 1763 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1764 } 1765 1766 static inline void net_zcopy_get(struct ubuf_info *uarg) 1767 { 1768 refcount_inc(&uarg->refcnt); 1769 } 1770 1771 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1772 { 1773 skb_shinfo(skb)->destructor_arg = uarg; 1774 skb_shinfo(skb)->flags |= uarg->flags; 1775 } 1776 1777 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1778 bool *have_ref) 1779 { 1780 if (skb && uarg && !skb_zcopy(skb)) { 1781 if (unlikely(have_ref && *have_ref)) 1782 *have_ref = false; 1783 else 1784 net_zcopy_get(uarg); 1785 skb_zcopy_init(skb, uarg); 1786 } 1787 } 1788 1789 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1790 { 1791 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1792 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1793 } 1794 1795 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1796 { 1797 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1798 } 1799 1800 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1801 { 1802 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1803 } 1804 1805 static inline void net_zcopy_put(struct ubuf_info *uarg) 1806 { 1807 if (uarg) 1808 uarg->ops->complete(NULL, uarg, true); 1809 } 1810 1811 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1812 { 1813 if (uarg) { 1814 if (uarg->ops == &msg_zerocopy_ubuf_ops) 1815 msg_zerocopy_put_abort(uarg, have_uref); 1816 else if (have_uref) 1817 net_zcopy_put(uarg); 1818 } 1819 } 1820 1821 /* Release a reference on a zerocopy structure */ 1822 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1823 { 1824 struct ubuf_info *uarg = skb_zcopy(skb); 1825 1826 if (uarg) { 1827 if (!skb_zcopy_is_nouarg(skb)) 1828 uarg->ops->complete(skb, uarg, zerocopy_success); 1829 1830 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1831 } 1832 } 1833 1834 void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1835 1836 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1837 { 1838 if (unlikely(skb_zcopy_managed(skb))) 1839 __skb_zcopy_downgrade_managed(skb); 1840 } 1841 1842 /* Return true if frags in this skb are readable by the host. */ 1843 static inline bool skb_frags_readable(const struct sk_buff *skb) 1844 { 1845 return !skb->unreadable; 1846 } 1847 1848 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1849 { 1850 skb->next = NULL; 1851 } 1852 1853 static inline void skb_poison_list(struct sk_buff *skb) 1854 { 1855 #ifdef CONFIG_DEBUG_NET 1856 skb->next = SKB_LIST_POISON_NEXT; 1857 #endif 1858 } 1859 1860 /* Iterate through singly-linked GSO fragments of an skb. */ 1861 #define skb_list_walk_safe(first, skb, next_skb) \ 1862 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1863 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1864 1865 static inline void skb_list_del_init(struct sk_buff *skb) 1866 { 1867 __list_del_entry(&skb->list); 1868 skb_mark_not_on_list(skb); 1869 } 1870 1871 /** 1872 * skb_queue_empty - check if a queue is empty 1873 * @list: queue head 1874 * 1875 * Returns true if the queue is empty, false otherwise. 1876 */ 1877 static inline int skb_queue_empty(const struct sk_buff_head *list) 1878 { 1879 return list->next == (const struct sk_buff *) list; 1880 } 1881 1882 /** 1883 * skb_queue_empty_lockless - check if a queue is empty 1884 * @list: queue head 1885 * 1886 * Returns true if the queue is empty, false otherwise. 1887 * This variant can be used in lockless contexts. 1888 */ 1889 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1890 { 1891 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1892 } 1893 1894 1895 /** 1896 * skb_queue_is_last - check if skb is the last entry in the queue 1897 * @list: queue head 1898 * @skb: buffer 1899 * 1900 * Returns true if @skb is the last buffer on the list. 1901 */ 1902 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1903 const struct sk_buff *skb) 1904 { 1905 return skb->next == (const struct sk_buff *) list; 1906 } 1907 1908 /** 1909 * skb_queue_is_first - check if skb is the first entry in the queue 1910 * @list: queue head 1911 * @skb: buffer 1912 * 1913 * Returns true if @skb is the first buffer on the list. 1914 */ 1915 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1916 const struct sk_buff *skb) 1917 { 1918 return skb->prev == (const struct sk_buff *) list; 1919 } 1920 1921 /** 1922 * skb_queue_next - return the next packet in the queue 1923 * @list: queue head 1924 * @skb: current buffer 1925 * 1926 * Return the next packet in @list after @skb. It is only valid to 1927 * call this if skb_queue_is_last() evaluates to false. 1928 */ 1929 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1930 const struct sk_buff *skb) 1931 { 1932 /* This BUG_ON may seem severe, but if we just return then we 1933 * are going to dereference garbage. 1934 */ 1935 BUG_ON(skb_queue_is_last(list, skb)); 1936 return skb->next; 1937 } 1938 1939 /** 1940 * skb_queue_prev - return the prev packet in the queue 1941 * @list: queue head 1942 * @skb: current buffer 1943 * 1944 * Return the prev packet in @list before @skb. It is only valid to 1945 * call this if skb_queue_is_first() evaluates to false. 1946 */ 1947 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1948 const struct sk_buff *skb) 1949 { 1950 /* This BUG_ON may seem severe, but if we just return then we 1951 * are going to dereference garbage. 1952 */ 1953 BUG_ON(skb_queue_is_first(list, skb)); 1954 return skb->prev; 1955 } 1956 1957 /** 1958 * skb_get - reference buffer 1959 * @skb: buffer to reference 1960 * 1961 * Makes another reference to a socket buffer and returns a pointer 1962 * to the buffer. 1963 */ 1964 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1965 { 1966 refcount_inc(&skb->users); 1967 return skb; 1968 } 1969 1970 /* 1971 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1972 */ 1973 1974 /** 1975 * skb_cloned - is the buffer a clone 1976 * @skb: buffer to check 1977 * 1978 * Returns true if the buffer was generated with skb_clone() and is 1979 * one of multiple shared copies of the buffer. Cloned buffers are 1980 * shared data so must not be written to under normal circumstances. 1981 */ 1982 static inline int skb_cloned(const struct sk_buff *skb) 1983 { 1984 return skb->cloned && 1985 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1986 } 1987 1988 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1989 { 1990 might_sleep_if(gfpflags_allow_blocking(pri)); 1991 1992 if (skb_cloned(skb)) 1993 return pskb_expand_head(skb, 0, 0, pri); 1994 1995 return 0; 1996 } 1997 1998 /* This variant of skb_unclone() makes sure skb->truesize 1999 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 2000 * 2001 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 2002 * when various debugging features are in place. 2003 */ 2004 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 2005 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 2006 { 2007 might_sleep_if(gfpflags_allow_blocking(pri)); 2008 2009 if (skb_cloned(skb)) 2010 return __skb_unclone_keeptruesize(skb, pri); 2011 return 0; 2012 } 2013 2014 /** 2015 * skb_header_cloned - is the header a clone 2016 * @skb: buffer to check 2017 * 2018 * Returns true if modifying the header part of the buffer requires 2019 * the data to be copied. 2020 */ 2021 static inline int skb_header_cloned(const struct sk_buff *skb) 2022 { 2023 int dataref; 2024 2025 if (!skb->cloned) 2026 return 0; 2027 2028 dataref = atomic_read(&skb_shinfo(skb)->dataref); 2029 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 2030 return dataref != 1; 2031 } 2032 2033 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 2034 { 2035 might_sleep_if(gfpflags_allow_blocking(pri)); 2036 2037 if (skb_header_cloned(skb)) 2038 return pskb_expand_head(skb, 0, 0, pri); 2039 2040 return 0; 2041 } 2042 2043 /** 2044 * __skb_header_release() - allow clones to use the headroom 2045 * @skb: buffer to operate on 2046 * 2047 * See "DOC: dataref and headerless skbs". 2048 */ 2049 static inline void __skb_header_release(struct sk_buff *skb) 2050 { 2051 skb->nohdr = 1; 2052 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 2053 } 2054 2055 2056 /** 2057 * skb_shared - is the buffer shared 2058 * @skb: buffer to check 2059 * 2060 * Returns true if more than one person has a reference to this 2061 * buffer. 2062 */ 2063 static inline int skb_shared(const struct sk_buff *skb) 2064 { 2065 return refcount_read(&skb->users) != 1; 2066 } 2067 2068 /** 2069 * skb_share_check - check if buffer is shared and if so clone it 2070 * @skb: buffer to check 2071 * @pri: priority for memory allocation 2072 * 2073 * If the buffer is shared the buffer is cloned and the old copy 2074 * drops a reference. A new clone with a single reference is returned. 2075 * If the buffer is not shared the original buffer is returned. When 2076 * being called from interrupt status or with spinlocks held pri must 2077 * be GFP_ATOMIC. 2078 * 2079 * NULL is returned on a memory allocation failure. 2080 */ 2081 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 2082 { 2083 might_sleep_if(gfpflags_allow_blocking(pri)); 2084 if (skb_shared(skb)) { 2085 struct sk_buff *nskb = skb_clone(skb, pri); 2086 2087 if (likely(nskb)) 2088 consume_skb(skb); 2089 else 2090 kfree_skb(skb); 2091 skb = nskb; 2092 } 2093 return skb; 2094 } 2095 2096 /* 2097 * Copy shared buffers into a new sk_buff. We effectively do COW on 2098 * packets to handle cases where we have a local reader and forward 2099 * and a couple of other messy ones. The normal one is tcpdumping 2100 * a packet that's being forwarded. 2101 */ 2102 2103 /** 2104 * skb_unshare - make a copy of a shared buffer 2105 * @skb: buffer to check 2106 * @pri: priority for memory allocation 2107 * 2108 * If the socket buffer is a clone then this function creates a new 2109 * copy of the data, drops a reference count on the old copy and returns 2110 * the new copy with the reference count at 1. If the buffer is not a clone 2111 * the original buffer is returned. When called with a spinlock held or 2112 * from interrupt state @pri must be %GFP_ATOMIC 2113 * 2114 * %NULL is returned on a memory allocation failure. 2115 */ 2116 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2117 gfp_t pri) 2118 { 2119 might_sleep_if(gfpflags_allow_blocking(pri)); 2120 if (skb_cloned(skb)) { 2121 struct sk_buff *nskb = skb_copy(skb, pri); 2122 2123 /* Free our shared copy */ 2124 if (likely(nskb)) 2125 consume_skb(skb); 2126 else 2127 kfree_skb(skb); 2128 skb = nskb; 2129 } 2130 return skb; 2131 } 2132 2133 /** 2134 * skb_peek - peek at the head of an &sk_buff_head 2135 * @list_: list to peek at 2136 * 2137 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2138 * be careful with this one. A peek leaves the buffer on the 2139 * list and someone else may run off with it. You must hold 2140 * the appropriate locks or have a private queue to do this. 2141 * 2142 * Returns %NULL for an empty list or a pointer to the head element. 2143 * The reference count is not incremented and the reference is therefore 2144 * volatile. Use with caution. 2145 */ 2146 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2147 { 2148 struct sk_buff *skb = list_->next; 2149 2150 if (skb == (struct sk_buff *)list_) 2151 skb = NULL; 2152 return skb; 2153 } 2154 2155 /** 2156 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2157 * @list_: list to peek at 2158 * 2159 * Like skb_peek(), but the caller knows that the list is not empty. 2160 */ 2161 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2162 { 2163 return list_->next; 2164 } 2165 2166 /** 2167 * skb_peek_next - peek skb following the given one from a queue 2168 * @skb: skb to start from 2169 * @list_: list to peek at 2170 * 2171 * Returns %NULL when the end of the list is met or a pointer to the 2172 * next element. The reference count is not incremented and the 2173 * reference is therefore volatile. Use with caution. 2174 */ 2175 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2176 const struct sk_buff_head *list_) 2177 { 2178 struct sk_buff *next = skb->next; 2179 2180 if (next == (struct sk_buff *)list_) 2181 next = NULL; 2182 return next; 2183 } 2184 2185 /** 2186 * skb_peek_tail - peek at the tail of an &sk_buff_head 2187 * @list_: list to peek at 2188 * 2189 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2190 * be careful with this one. A peek leaves the buffer on the 2191 * list and someone else may run off with it. You must hold 2192 * the appropriate locks or have a private queue to do this. 2193 * 2194 * Returns %NULL for an empty list or a pointer to the tail element. 2195 * The reference count is not incremented and the reference is therefore 2196 * volatile. Use with caution. 2197 */ 2198 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2199 { 2200 struct sk_buff *skb = READ_ONCE(list_->prev); 2201 2202 if (skb == (struct sk_buff *)list_) 2203 skb = NULL; 2204 return skb; 2205 2206 } 2207 2208 /** 2209 * skb_queue_len - get queue length 2210 * @list_: list to measure 2211 * 2212 * Return the length of an &sk_buff queue. 2213 */ 2214 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2215 { 2216 return list_->qlen; 2217 } 2218 2219 /** 2220 * skb_queue_len_lockless - get queue length 2221 * @list_: list to measure 2222 * 2223 * Return the length of an &sk_buff queue. 2224 * This variant can be used in lockless contexts. 2225 */ 2226 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2227 { 2228 return READ_ONCE(list_->qlen); 2229 } 2230 2231 /** 2232 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2233 * @list: queue to initialize 2234 * 2235 * This initializes only the list and queue length aspects of 2236 * an sk_buff_head object. This allows to initialize the list 2237 * aspects of an sk_buff_head without reinitializing things like 2238 * the spinlock. It can also be used for on-stack sk_buff_head 2239 * objects where the spinlock is known to not be used. 2240 */ 2241 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2242 { 2243 list->prev = list->next = (struct sk_buff *)list; 2244 list->qlen = 0; 2245 } 2246 2247 /* 2248 * This function creates a split out lock class for each invocation; 2249 * this is needed for now since a whole lot of users of the skb-queue 2250 * infrastructure in drivers have different locking usage (in hardirq) 2251 * than the networking core (in softirq only). In the long run either the 2252 * network layer or drivers should need annotation to consolidate the 2253 * main types of usage into 3 classes. 2254 */ 2255 static inline void skb_queue_head_init(struct sk_buff_head *list) 2256 { 2257 spin_lock_init(&list->lock); 2258 __skb_queue_head_init(list); 2259 } 2260 2261 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2262 struct lock_class_key *class) 2263 { 2264 skb_queue_head_init(list); 2265 lockdep_set_class(&list->lock, class); 2266 } 2267 2268 /* 2269 * Insert an sk_buff on a list. 2270 * 2271 * The "__skb_xxxx()" functions are the non-atomic ones that 2272 * can only be called with interrupts disabled. 2273 */ 2274 static inline void __skb_insert(struct sk_buff *newsk, 2275 struct sk_buff *prev, struct sk_buff *next, 2276 struct sk_buff_head *list) 2277 { 2278 /* See skb_queue_empty_lockless() and skb_peek_tail() 2279 * for the opposite READ_ONCE() 2280 */ 2281 WRITE_ONCE(newsk->next, next); 2282 WRITE_ONCE(newsk->prev, prev); 2283 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2284 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2285 WRITE_ONCE(list->qlen, list->qlen + 1); 2286 } 2287 2288 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2289 struct sk_buff *prev, 2290 struct sk_buff *next) 2291 { 2292 struct sk_buff *first = list->next; 2293 struct sk_buff *last = list->prev; 2294 2295 WRITE_ONCE(first->prev, prev); 2296 WRITE_ONCE(prev->next, first); 2297 2298 WRITE_ONCE(last->next, next); 2299 WRITE_ONCE(next->prev, last); 2300 } 2301 2302 /** 2303 * skb_queue_splice - join two skb lists, this is designed for stacks 2304 * @list: the new list to add 2305 * @head: the place to add it in the first list 2306 */ 2307 static inline void skb_queue_splice(const struct sk_buff_head *list, 2308 struct sk_buff_head *head) 2309 { 2310 if (!skb_queue_empty(list)) { 2311 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2312 head->qlen += list->qlen; 2313 } 2314 } 2315 2316 /** 2317 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2318 * @list: the new list to add 2319 * @head: the place to add it in the first list 2320 * 2321 * The list at @list is reinitialised 2322 */ 2323 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2324 struct sk_buff_head *head) 2325 { 2326 if (!skb_queue_empty(list)) { 2327 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2328 head->qlen += list->qlen; 2329 __skb_queue_head_init(list); 2330 } 2331 } 2332 2333 /** 2334 * skb_queue_splice_tail - join two skb lists, each list being a queue 2335 * @list: the new list to add 2336 * @head: the place to add it in the first list 2337 */ 2338 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2339 struct sk_buff_head *head) 2340 { 2341 if (!skb_queue_empty(list)) { 2342 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2343 head->qlen += list->qlen; 2344 } 2345 } 2346 2347 /** 2348 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2349 * @list: the new list to add 2350 * @head: the place to add it in the first list 2351 * 2352 * Each of the lists is a queue. 2353 * The list at @list is reinitialised 2354 */ 2355 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2356 struct sk_buff_head *head) 2357 { 2358 if (!skb_queue_empty(list)) { 2359 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2360 head->qlen += list->qlen; 2361 __skb_queue_head_init(list); 2362 } 2363 } 2364 2365 /** 2366 * __skb_queue_after - queue a buffer at the list head 2367 * @list: list to use 2368 * @prev: place after this buffer 2369 * @newsk: buffer to queue 2370 * 2371 * Queue a buffer int the middle of a list. This function takes no locks 2372 * and you must therefore hold required locks before calling it. 2373 * 2374 * A buffer cannot be placed on two lists at the same time. 2375 */ 2376 static inline void __skb_queue_after(struct sk_buff_head *list, 2377 struct sk_buff *prev, 2378 struct sk_buff *newsk) 2379 { 2380 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2381 } 2382 2383 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2384 struct sk_buff_head *list); 2385 2386 static inline void __skb_queue_before(struct sk_buff_head *list, 2387 struct sk_buff *next, 2388 struct sk_buff *newsk) 2389 { 2390 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2391 } 2392 2393 /** 2394 * __skb_queue_head - queue a buffer at the list head 2395 * @list: list to use 2396 * @newsk: buffer to queue 2397 * 2398 * Queue a buffer at the start of a list. This function takes no locks 2399 * and you must therefore hold required locks before calling it. 2400 * 2401 * A buffer cannot be placed on two lists at the same time. 2402 */ 2403 static inline void __skb_queue_head(struct sk_buff_head *list, 2404 struct sk_buff *newsk) 2405 { 2406 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2407 } 2408 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2409 2410 /** 2411 * __skb_queue_tail - queue a buffer at the list tail 2412 * @list: list to use 2413 * @newsk: buffer to queue 2414 * 2415 * Queue a buffer at the end of a list. This function takes no locks 2416 * and you must therefore hold required locks before calling it. 2417 * 2418 * A buffer cannot be placed on two lists at the same time. 2419 */ 2420 static inline void __skb_queue_tail(struct sk_buff_head *list, 2421 struct sk_buff *newsk) 2422 { 2423 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2424 } 2425 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2426 2427 /* 2428 * remove sk_buff from list. _Must_ be called atomically, and with 2429 * the list known.. 2430 */ 2431 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2432 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2433 { 2434 struct sk_buff *next, *prev; 2435 2436 WRITE_ONCE(list->qlen, list->qlen - 1); 2437 next = skb->next; 2438 prev = skb->prev; 2439 skb->next = skb->prev = NULL; 2440 WRITE_ONCE(next->prev, prev); 2441 WRITE_ONCE(prev->next, next); 2442 } 2443 2444 /** 2445 * __skb_dequeue - remove from the head of the queue 2446 * @list: list to dequeue from 2447 * 2448 * Remove the head of the list. This function does not take any locks 2449 * so must be used with appropriate locks held only. The head item is 2450 * returned or %NULL if the list is empty. 2451 */ 2452 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2453 { 2454 struct sk_buff *skb = skb_peek(list); 2455 if (skb) 2456 __skb_unlink(skb, list); 2457 return skb; 2458 } 2459 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2460 2461 /** 2462 * __skb_dequeue_tail - remove from the tail of the queue 2463 * @list: list to dequeue from 2464 * 2465 * Remove the tail of the list. This function does not take any locks 2466 * so must be used with appropriate locks held only. The tail item is 2467 * returned or %NULL if the list is empty. 2468 */ 2469 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2470 { 2471 struct sk_buff *skb = skb_peek_tail(list); 2472 if (skb) 2473 __skb_unlink(skb, list); 2474 return skb; 2475 } 2476 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2477 2478 2479 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2480 { 2481 return skb->data_len; 2482 } 2483 2484 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2485 { 2486 return skb->len - skb->data_len; 2487 } 2488 2489 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2490 { 2491 unsigned int i, len = 0; 2492 2493 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2494 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2495 return len; 2496 } 2497 2498 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2499 { 2500 return skb_headlen(skb) + __skb_pagelen(skb); 2501 } 2502 2503 static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag, 2504 netmem_ref netmem, int off, 2505 int size) 2506 { 2507 frag->netmem = netmem; 2508 frag->offset = off; 2509 skb_frag_size_set(frag, size); 2510 } 2511 2512 static inline void skb_frag_fill_page_desc(skb_frag_t *frag, 2513 struct page *page, 2514 int off, int size) 2515 { 2516 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size); 2517 } 2518 2519 static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo, 2520 int i, netmem_ref netmem, 2521 int off, int size) 2522 { 2523 skb_frag_t *frag = &shinfo->frags[i]; 2524 2525 skb_frag_fill_netmem_desc(frag, netmem, off, size); 2526 } 2527 2528 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2529 int i, struct page *page, 2530 int off, int size) 2531 { 2532 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off, 2533 size); 2534 } 2535 2536 /** 2537 * skb_len_add - adds a number to len fields of skb 2538 * @skb: buffer to add len to 2539 * @delta: number of bytes to add 2540 */ 2541 static inline void skb_len_add(struct sk_buff *skb, int delta) 2542 { 2543 skb->len += delta; 2544 skb->data_len += delta; 2545 skb->truesize += delta; 2546 } 2547 2548 /** 2549 * __skb_fill_netmem_desc - initialise a fragment in an skb 2550 * @skb: buffer containing fragment to be initialised 2551 * @i: fragment index to initialise 2552 * @netmem: the netmem to use for this fragment 2553 * @off: the offset to the data with @page 2554 * @size: the length of the data 2555 * 2556 * Initialises the @i'th fragment of @skb to point to &size bytes at 2557 * offset @off within @page. 2558 * 2559 * Does not take any additional reference on the fragment. 2560 */ 2561 static inline void __skb_fill_netmem_desc(struct sk_buff *skb, int i, 2562 netmem_ref netmem, int off, int size) 2563 { 2564 struct page *page; 2565 2566 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size); 2567 2568 if (netmem_is_net_iov(netmem)) { 2569 skb->unreadable = true; 2570 return; 2571 } 2572 2573 page = netmem_to_page(netmem); 2574 2575 /* Propagate page pfmemalloc to the skb if we can. The problem is 2576 * that not all callers have unique ownership of the page but rely 2577 * on page_is_pfmemalloc doing the right thing(tm). 2578 */ 2579 page = compound_head(page); 2580 if (page_is_pfmemalloc(page)) 2581 skb->pfmemalloc = true; 2582 } 2583 2584 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2585 struct page *page, int off, int size) 2586 { 2587 __skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2588 } 2589 2590 static inline void skb_fill_netmem_desc(struct sk_buff *skb, int i, 2591 netmem_ref netmem, int off, int size) 2592 { 2593 __skb_fill_netmem_desc(skb, i, netmem, off, size); 2594 skb_shinfo(skb)->nr_frags = i + 1; 2595 } 2596 2597 /** 2598 * skb_fill_page_desc - initialise a paged fragment in an skb 2599 * @skb: buffer containing fragment to be initialised 2600 * @i: paged fragment index to initialise 2601 * @page: the page to use for this fragment 2602 * @off: the offset to the data with @page 2603 * @size: the length of the data 2604 * 2605 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2606 * @skb to point to @size bytes at offset @off within @page. In 2607 * addition updates @skb such that @i is the last fragment. 2608 * 2609 * Does not take any additional reference on the fragment. 2610 */ 2611 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2612 struct page *page, int off, int size) 2613 { 2614 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2615 } 2616 2617 /** 2618 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2619 * @skb: buffer containing fragment to be initialised 2620 * @i: paged fragment index to initialise 2621 * @page: the page to use for this fragment 2622 * @off: the offset to the data with @page 2623 * @size: the length of the data 2624 * 2625 * Variant of skb_fill_page_desc() which does not deal with 2626 * pfmemalloc, if page is not owned by us. 2627 */ 2628 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2629 struct page *page, int off, 2630 int size) 2631 { 2632 struct skb_shared_info *shinfo = skb_shinfo(skb); 2633 2634 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2635 shinfo->nr_frags = i + 1; 2636 } 2637 2638 void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem, 2639 int off, int size, unsigned int truesize); 2640 2641 static inline void skb_add_rx_frag(struct sk_buff *skb, int i, 2642 struct page *page, int off, int size, 2643 unsigned int truesize) 2644 { 2645 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size, 2646 truesize); 2647 } 2648 2649 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2650 unsigned int truesize); 2651 2652 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2653 2654 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2655 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2656 { 2657 return skb->head + skb->tail; 2658 } 2659 2660 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2661 { 2662 skb->tail = skb->data - skb->head; 2663 } 2664 2665 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2666 { 2667 skb_reset_tail_pointer(skb); 2668 skb->tail += offset; 2669 } 2670 2671 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2672 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2673 { 2674 return skb->tail; 2675 } 2676 2677 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2678 { 2679 skb->tail = skb->data; 2680 } 2681 2682 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2683 { 2684 skb->tail = skb->data + offset; 2685 } 2686 2687 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2688 2689 static inline void skb_assert_len(struct sk_buff *skb) 2690 { 2691 #ifdef CONFIG_DEBUG_NET 2692 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2693 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2694 #endif /* CONFIG_DEBUG_NET */ 2695 } 2696 2697 #if defined(CONFIG_FAIL_SKB_REALLOC) 2698 void skb_might_realloc(struct sk_buff *skb); 2699 #else 2700 static inline void skb_might_realloc(struct sk_buff *skb) {} 2701 #endif 2702 2703 /* 2704 * Add data to an sk_buff 2705 */ 2706 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2707 void *skb_put(struct sk_buff *skb, unsigned int len); 2708 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2709 { 2710 void *tmp = skb_tail_pointer(skb); 2711 SKB_LINEAR_ASSERT(skb); 2712 skb->tail += len; 2713 skb->len += len; 2714 return tmp; 2715 } 2716 2717 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2718 { 2719 void *tmp = __skb_put(skb, len); 2720 2721 memset(tmp, 0, len); 2722 return tmp; 2723 } 2724 2725 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2726 unsigned int len) 2727 { 2728 void *tmp = __skb_put(skb, len); 2729 2730 memcpy(tmp, data, len); 2731 return tmp; 2732 } 2733 2734 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2735 { 2736 *(u8 *)__skb_put(skb, 1) = val; 2737 } 2738 2739 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2740 { 2741 void *tmp = skb_put(skb, len); 2742 2743 memset(tmp, 0, len); 2744 2745 return tmp; 2746 } 2747 2748 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2749 unsigned int len) 2750 { 2751 void *tmp = skb_put(skb, len); 2752 2753 memcpy(tmp, data, len); 2754 2755 return tmp; 2756 } 2757 2758 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2759 { 2760 *(u8 *)skb_put(skb, 1) = val; 2761 } 2762 2763 void *skb_push(struct sk_buff *skb, unsigned int len); 2764 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2765 { 2766 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2767 2768 skb->data -= len; 2769 skb->len += len; 2770 return skb->data; 2771 } 2772 2773 void *skb_pull(struct sk_buff *skb, unsigned int len); 2774 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2775 { 2776 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2777 2778 skb->len -= len; 2779 if (unlikely(skb->len < skb->data_len)) { 2780 #if defined(CONFIG_DEBUG_NET) 2781 skb->len += len; 2782 pr_err("__skb_pull(len=%u)\n", len); 2783 skb_dump(KERN_ERR, skb, false); 2784 #endif 2785 BUG(); 2786 } 2787 return skb->data += len; 2788 } 2789 2790 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2791 { 2792 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2793 } 2794 2795 void *skb_pull_data(struct sk_buff *skb, size_t len); 2796 2797 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2798 2799 static inline enum skb_drop_reason 2800 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len) 2801 { 2802 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2803 skb_might_realloc(skb); 2804 2805 if (likely(len <= skb_headlen(skb))) 2806 return SKB_NOT_DROPPED_YET; 2807 2808 if (unlikely(len > skb->len)) 2809 return SKB_DROP_REASON_PKT_TOO_SMALL; 2810 2811 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb)))) 2812 return SKB_DROP_REASON_NOMEM; 2813 2814 return SKB_NOT_DROPPED_YET; 2815 } 2816 2817 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2818 { 2819 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 2820 } 2821 2822 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2823 { 2824 if (!pskb_may_pull(skb, len)) 2825 return NULL; 2826 2827 skb->len -= len; 2828 return skb->data += len; 2829 } 2830 2831 void skb_condense(struct sk_buff *skb); 2832 2833 /** 2834 * skb_headroom - bytes at buffer head 2835 * @skb: buffer to check 2836 * 2837 * Return the number of bytes of free space at the head of an &sk_buff. 2838 */ 2839 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2840 { 2841 return skb->data - skb->head; 2842 } 2843 2844 /** 2845 * skb_tailroom - bytes at buffer end 2846 * @skb: buffer to check 2847 * 2848 * Return the number of bytes of free space at the tail of an sk_buff 2849 */ 2850 static inline int skb_tailroom(const struct sk_buff *skb) 2851 { 2852 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2853 } 2854 2855 /** 2856 * skb_availroom - bytes at buffer end 2857 * @skb: buffer to check 2858 * 2859 * Return the number of bytes of free space at the tail of an sk_buff 2860 * allocated by sk_stream_alloc() 2861 */ 2862 static inline int skb_availroom(const struct sk_buff *skb) 2863 { 2864 if (skb_is_nonlinear(skb)) 2865 return 0; 2866 2867 return skb->end - skb->tail - skb->reserved_tailroom; 2868 } 2869 2870 /** 2871 * skb_reserve - adjust headroom 2872 * @skb: buffer to alter 2873 * @len: bytes to move 2874 * 2875 * Increase the headroom of an empty &sk_buff by reducing the tail 2876 * room. This is only allowed for an empty buffer. 2877 */ 2878 static inline void skb_reserve(struct sk_buff *skb, int len) 2879 { 2880 skb->data += len; 2881 skb->tail += len; 2882 } 2883 2884 /** 2885 * skb_tailroom_reserve - adjust reserved_tailroom 2886 * @skb: buffer to alter 2887 * @mtu: maximum amount of headlen permitted 2888 * @needed_tailroom: minimum amount of reserved_tailroom 2889 * 2890 * Set reserved_tailroom so that headlen can be as large as possible but 2891 * not larger than mtu and tailroom cannot be smaller than 2892 * needed_tailroom. 2893 * The required headroom should already have been reserved before using 2894 * this function. 2895 */ 2896 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2897 unsigned int needed_tailroom) 2898 { 2899 SKB_LINEAR_ASSERT(skb); 2900 if (mtu < skb_tailroom(skb) - needed_tailroom) 2901 /* use at most mtu */ 2902 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2903 else 2904 /* use up to all available space */ 2905 skb->reserved_tailroom = needed_tailroom; 2906 } 2907 2908 #define ENCAP_TYPE_ETHER 0 2909 #define ENCAP_TYPE_IPPROTO 1 2910 2911 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2912 __be16 protocol) 2913 { 2914 skb->inner_protocol = protocol; 2915 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2916 } 2917 2918 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2919 __u8 ipproto) 2920 { 2921 skb->inner_ipproto = ipproto; 2922 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2923 } 2924 2925 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2926 { 2927 skb->inner_mac_header = skb->mac_header; 2928 skb->inner_network_header = skb->network_header; 2929 skb->inner_transport_header = skb->transport_header; 2930 } 2931 2932 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2933 { 2934 return skb->mac_header != (typeof(skb->mac_header))~0U; 2935 } 2936 2937 static inline void skb_reset_mac_len(struct sk_buff *skb) 2938 { 2939 if (!skb_mac_header_was_set(skb)) { 2940 DEBUG_NET_WARN_ON_ONCE(1); 2941 skb->mac_len = 0; 2942 } else { 2943 skb->mac_len = skb->network_header - skb->mac_header; 2944 } 2945 } 2946 2947 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2948 *skb) 2949 { 2950 return skb->head + skb->inner_transport_header; 2951 } 2952 2953 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2954 { 2955 return skb_inner_transport_header(skb) - skb->data; 2956 } 2957 2958 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2959 { 2960 long offset = skb->data - skb->head; 2961 2962 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_transport_header))offset); 2963 skb->inner_transport_header = offset; 2964 } 2965 2966 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2967 const int offset) 2968 { 2969 skb_reset_inner_transport_header(skb); 2970 skb->inner_transport_header += offset; 2971 } 2972 2973 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2974 { 2975 return skb->head + skb->inner_network_header; 2976 } 2977 2978 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2979 { 2980 long offset = skb->data - skb->head; 2981 2982 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_network_header))offset); 2983 skb->inner_network_header = offset; 2984 } 2985 2986 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2987 const int offset) 2988 { 2989 skb_reset_inner_network_header(skb); 2990 skb->inner_network_header += offset; 2991 } 2992 2993 static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb) 2994 { 2995 return skb->inner_network_header > 0; 2996 } 2997 2998 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2999 { 3000 return skb->head + skb->inner_mac_header; 3001 } 3002 3003 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 3004 { 3005 long offset = skb->data - skb->head; 3006 3007 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_mac_header))offset); 3008 skb->inner_mac_header = offset; 3009 } 3010 3011 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 3012 const int offset) 3013 { 3014 skb_reset_inner_mac_header(skb); 3015 skb->inner_mac_header += offset; 3016 } 3017 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 3018 { 3019 return skb->transport_header != (typeof(skb->transport_header))~0U; 3020 } 3021 3022 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 3023 { 3024 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3025 return skb->head + skb->transport_header; 3026 } 3027 3028 static inline void skb_reset_transport_header(struct sk_buff *skb) 3029 { 3030 long offset = skb->data - skb->head; 3031 3032 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->transport_header))offset); 3033 skb->transport_header = offset; 3034 } 3035 3036 /** 3037 * skb_reset_transport_header_careful - conditionally reset transport header 3038 * @skb: buffer to alter 3039 * 3040 * Hardened version of skb_reset_transport_header(). 3041 * 3042 * Returns: true if the operation was a success. 3043 */ 3044 static inline bool __must_check 3045 skb_reset_transport_header_careful(struct sk_buff *skb) 3046 { 3047 long offset = skb->data - skb->head; 3048 3049 if (unlikely(offset != (typeof(skb->transport_header))offset)) 3050 return false; 3051 3052 if (unlikely(offset == (typeof(skb->transport_header))~0U)) 3053 return false; 3054 3055 skb->transport_header = offset; 3056 return true; 3057 } 3058 3059 static inline void skb_set_transport_header(struct sk_buff *skb, 3060 const int offset) 3061 { 3062 skb_reset_transport_header(skb); 3063 skb->transport_header += offset; 3064 } 3065 3066 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 3067 { 3068 return skb->head + skb->network_header; 3069 } 3070 3071 static inline void skb_reset_network_header(struct sk_buff *skb) 3072 { 3073 long offset = skb->data - skb->head; 3074 3075 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->network_header))offset); 3076 skb->network_header = offset; 3077 } 3078 3079 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 3080 { 3081 skb_reset_network_header(skb); 3082 skb->network_header += offset; 3083 } 3084 3085 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 3086 { 3087 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3088 return skb->head + skb->mac_header; 3089 } 3090 3091 static inline int skb_mac_offset(const struct sk_buff *skb) 3092 { 3093 return skb_mac_header(skb) - skb->data; 3094 } 3095 3096 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 3097 { 3098 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3099 return skb->network_header - skb->mac_header; 3100 } 3101 3102 static inline void skb_unset_mac_header(struct sk_buff *skb) 3103 { 3104 skb->mac_header = (typeof(skb->mac_header))~0U; 3105 } 3106 3107 static inline void skb_reset_mac_header(struct sk_buff *skb) 3108 { 3109 long offset = skb->data - skb->head; 3110 3111 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->mac_header))offset); 3112 skb->mac_header = offset; 3113 } 3114 3115 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 3116 { 3117 skb_reset_mac_header(skb); 3118 skb->mac_header += offset; 3119 } 3120 3121 static inline void skb_pop_mac_header(struct sk_buff *skb) 3122 { 3123 skb->mac_header = skb->network_header; 3124 } 3125 3126 static inline void skb_probe_transport_header(struct sk_buff *skb) 3127 { 3128 struct flow_keys_basic keys; 3129 3130 if (skb_transport_header_was_set(skb)) 3131 return; 3132 3133 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 3134 NULL, 0, 0, 0, 0)) 3135 skb_set_transport_header(skb, keys.control.thoff); 3136 } 3137 3138 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 3139 { 3140 if (skb_mac_header_was_set(skb)) { 3141 const unsigned char *old_mac = skb_mac_header(skb); 3142 3143 skb_set_mac_header(skb, -skb->mac_len); 3144 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 3145 } 3146 } 3147 3148 /* Move the full mac header up to current network_header. 3149 * Leaves skb->data pointing at offset skb->mac_len into the mac_header. 3150 * Must be provided the complete mac header length. 3151 */ 3152 static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len) 3153 { 3154 if (skb_mac_header_was_set(skb)) { 3155 const unsigned char *old_mac = skb_mac_header(skb); 3156 3157 skb_set_mac_header(skb, -full_mac_len); 3158 memmove(skb_mac_header(skb), old_mac, full_mac_len); 3159 __skb_push(skb, full_mac_len - skb->mac_len); 3160 } 3161 } 3162 3163 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 3164 { 3165 return skb->csum_start - skb_headroom(skb); 3166 } 3167 3168 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 3169 { 3170 return skb->head + skb->csum_start; 3171 } 3172 3173 static inline int skb_transport_offset(const struct sk_buff *skb) 3174 { 3175 return skb_transport_header(skb) - skb->data; 3176 } 3177 3178 static inline u32 skb_network_header_len(const struct sk_buff *skb) 3179 { 3180 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3181 return skb->transport_header - skb->network_header; 3182 } 3183 3184 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 3185 { 3186 return skb->inner_transport_header - skb->inner_network_header; 3187 } 3188 3189 static inline int skb_network_offset(const struct sk_buff *skb) 3190 { 3191 return skb_network_header(skb) - skb->data; 3192 } 3193 3194 static inline int skb_inner_network_offset(const struct sk_buff *skb) 3195 { 3196 return skb_inner_network_header(skb) - skb->data; 3197 } 3198 3199 static inline enum skb_drop_reason 3200 pskb_network_may_pull_reason(struct sk_buff *skb, unsigned int len) 3201 { 3202 return pskb_may_pull_reason(skb, skb_network_offset(skb) + len); 3203 } 3204 3205 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 3206 { 3207 return pskb_network_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 3208 } 3209 3210 /* 3211 * CPUs often take a performance hit when accessing unaligned memory 3212 * locations. The actual performance hit varies, it can be small if the 3213 * hardware handles it or large if we have to take an exception and fix it 3214 * in software. 3215 * 3216 * Since an ethernet header is 14 bytes network drivers often end up with 3217 * the IP header at an unaligned offset. The IP header can be aligned by 3218 * shifting the start of the packet by 2 bytes. Drivers should do this 3219 * with: 3220 * 3221 * skb_reserve(skb, NET_IP_ALIGN); 3222 * 3223 * The downside to this alignment of the IP header is that the DMA is now 3224 * unaligned. On some architectures the cost of an unaligned DMA is high 3225 * and this cost outweighs the gains made by aligning the IP header. 3226 * 3227 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 3228 * to be overridden. 3229 */ 3230 #ifndef NET_IP_ALIGN 3231 #define NET_IP_ALIGN 2 3232 #endif 3233 3234 /* 3235 * The networking layer reserves some headroom in skb data (via 3236 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 3237 * the header has to grow. In the default case, if the header has to grow 3238 * 32 bytes or less we avoid the reallocation. 3239 * 3240 * Unfortunately this headroom changes the DMA alignment of the resulting 3241 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 3242 * on some architectures. An architecture can override this value, 3243 * perhaps setting it to a cacheline in size (since that will maintain 3244 * cacheline alignment of the DMA). It must be a power of 2. 3245 * 3246 * Various parts of the networking layer expect at least 32 bytes of 3247 * headroom, you should not reduce this. 3248 * 3249 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 3250 * to reduce average number of cache lines per packet. 3251 * get_rps_cpu() for example only access one 64 bytes aligned block : 3252 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 3253 */ 3254 #ifndef NET_SKB_PAD 3255 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 3256 #endif 3257 3258 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 3259 3260 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3261 { 3262 if (WARN_ON(skb_is_nonlinear(skb))) 3263 return; 3264 skb->len = len; 3265 skb_set_tail_pointer(skb, len); 3266 } 3267 3268 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3269 { 3270 __skb_set_length(skb, len); 3271 } 3272 3273 void skb_trim(struct sk_buff *skb, unsigned int len); 3274 3275 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3276 { 3277 if (skb->data_len) 3278 return ___pskb_trim(skb, len); 3279 __skb_trim(skb, len); 3280 return 0; 3281 } 3282 3283 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3284 { 3285 skb_might_realloc(skb); 3286 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3287 } 3288 3289 /** 3290 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3291 * @skb: buffer to alter 3292 * @len: new length 3293 * 3294 * This is identical to pskb_trim except that the caller knows that 3295 * the skb is not cloned so we should never get an error due to out- 3296 * of-memory. 3297 */ 3298 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3299 { 3300 int err = pskb_trim(skb, len); 3301 BUG_ON(err); 3302 } 3303 3304 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3305 { 3306 unsigned int diff = len - skb->len; 3307 3308 if (skb_tailroom(skb) < diff) { 3309 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3310 GFP_ATOMIC); 3311 if (ret) 3312 return ret; 3313 } 3314 __skb_set_length(skb, len); 3315 return 0; 3316 } 3317 3318 /** 3319 * skb_orphan - orphan a buffer 3320 * @skb: buffer to orphan 3321 * 3322 * If a buffer currently has an owner then we call the owner's 3323 * destructor function and make the @skb unowned. The buffer continues 3324 * to exist but is no longer charged to its former owner. 3325 */ 3326 static inline void skb_orphan(struct sk_buff *skb) 3327 { 3328 if (skb->destructor) { 3329 skb->destructor(skb); 3330 skb->destructor = NULL; 3331 skb->sk = NULL; 3332 } else { 3333 BUG_ON(skb->sk); 3334 } 3335 } 3336 3337 /** 3338 * skb_orphan_frags - orphan the frags contained in a buffer 3339 * @skb: buffer to orphan frags from 3340 * @gfp_mask: allocation mask for replacement pages 3341 * 3342 * For each frag in the SKB which needs a destructor (i.e. has an 3343 * owner) create a copy of that frag and release the original 3344 * page by calling the destructor. 3345 */ 3346 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3347 { 3348 if (likely(!skb_zcopy(skb))) 3349 return 0; 3350 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3351 return 0; 3352 return skb_copy_ubufs(skb, gfp_mask); 3353 } 3354 3355 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3356 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3357 { 3358 if (likely(!skb_zcopy(skb))) 3359 return 0; 3360 return skb_copy_ubufs(skb, gfp_mask); 3361 } 3362 3363 /** 3364 * __skb_queue_purge_reason - empty a list 3365 * @list: list to empty 3366 * @reason: drop reason 3367 * 3368 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3369 * the list and one reference dropped. This function does not take the 3370 * list lock and the caller must hold the relevant locks to use it. 3371 */ 3372 static inline void __skb_queue_purge_reason(struct sk_buff_head *list, 3373 enum skb_drop_reason reason) 3374 { 3375 struct sk_buff *skb; 3376 3377 while ((skb = __skb_dequeue(list)) != NULL) 3378 kfree_skb_reason(skb, reason); 3379 } 3380 3381 static inline void __skb_queue_purge(struct sk_buff_head *list) 3382 { 3383 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3384 } 3385 3386 void skb_queue_purge_reason(struct sk_buff_head *list, 3387 enum skb_drop_reason reason); 3388 3389 static inline void skb_queue_purge(struct sk_buff_head *list) 3390 { 3391 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3392 } 3393 3394 unsigned int skb_rbtree_purge(struct rb_root *root); 3395 void skb_errqueue_purge(struct sk_buff_head *list); 3396 3397 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3398 3399 /** 3400 * netdev_alloc_frag - allocate a page fragment 3401 * @fragsz: fragment size 3402 * 3403 * Allocates a frag from a page for receive buffer. 3404 * Uses GFP_ATOMIC allocations. 3405 */ 3406 static inline void *netdev_alloc_frag(unsigned int fragsz) 3407 { 3408 return __netdev_alloc_frag_align(fragsz, ~0u); 3409 } 3410 3411 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3412 unsigned int align) 3413 { 3414 WARN_ON_ONCE(!is_power_of_2(align)); 3415 return __netdev_alloc_frag_align(fragsz, -align); 3416 } 3417 3418 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3419 gfp_t gfp_mask); 3420 3421 /** 3422 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3423 * @dev: network device to receive on 3424 * @length: length to allocate 3425 * 3426 * Allocate a new &sk_buff and assign it a usage count of one. The 3427 * buffer has unspecified headroom built in. Users should allocate 3428 * the headroom they think they need without accounting for the 3429 * built in space. The built in space is used for optimisations. 3430 * 3431 * %NULL is returned if there is no free memory. Although this function 3432 * allocates memory it can be called from an interrupt. 3433 */ 3434 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3435 unsigned int length) 3436 { 3437 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3438 } 3439 3440 /* legacy helper around __netdev_alloc_skb() */ 3441 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3442 gfp_t gfp_mask) 3443 { 3444 return __netdev_alloc_skb(NULL, length, gfp_mask); 3445 } 3446 3447 /* legacy helper around netdev_alloc_skb() */ 3448 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3449 { 3450 return netdev_alloc_skb(NULL, length); 3451 } 3452 3453 3454 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3455 unsigned int length, gfp_t gfp) 3456 { 3457 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3458 3459 if (NET_IP_ALIGN && skb) 3460 skb_reserve(skb, NET_IP_ALIGN); 3461 return skb; 3462 } 3463 3464 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3465 unsigned int length) 3466 { 3467 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3468 } 3469 3470 static inline void skb_free_frag(void *addr) 3471 { 3472 page_frag_free(addr); 3473 } 3474 3475 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3476 3477 static inline void *napi_alloc_frag(unsigned int fragsz) 3478 { 3479 return __napi_alloc_frag_align(fragsz, ~0u); 3480 } 3481 3482 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3483 unsigned int align) 3484 { 3485 WARN_ON_ONCE(!is_power_of_2(align)); 3486 return __napi_alloc_frag_align(fragsz, -align); 3487 } 3488 3489 struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int length); 3490 void napi_consume_skb(struct sk_buff *skb, int budget); 3491 3492 void napi_skb_free_stolen_head(struct sk_buff *skb); 3493 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason); 3494 3495 /** 3496 * __dev_alloc_pages - allocate page for network Rx 3497 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3498 * @order: size of the allocation 3499 * 3500 * Allocate a new page. 3501 * 3502 * %NULL is returned if there is no free memory. 3503 */ 3504 static inline struct page *__dev_alloc_pages_noprof(gfp_t gfp_mask, 3505 unsigned int order) 3506 { 3507 /* This piece of code contains several assumptions. 3508 * 1. This is for device Rx, therefore a cold page is preferred. 3509 * 2. The expectation is the user wants a compound page. 3510 * 3. If requesting a order 0 page it will not be compound 3511 * due to the check to see if order has a value in prep_new_page 3512 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3513 * code in gfp_to_alloc_flags that should be enforcing this. 3514 */ 3515 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3516 3517 return alloc_pages_node_noprof(NUMA_NO_NODE, gfp_mask, order); 3518 } 3519 #define __dev_alloc_pages(...) alloc_hooks(__dev_alloc_pages_noprof(__VA_ARGS__)) 3520 3521 /* 3522 * This specialized allocator has to be a macro for its allocations to be 3523 * accounted separately (to have a separate alloc_tag). 3524 */ 3525 #define dev_alloc_pages(_order) __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, _order) 3526 3527 /** 3528 * __dev_alloc_page - allocate a page for network Rx 3529 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3530 * 3531 * Allocate a new page. 3532 * 3533 * %NULL is returned if there is no free memory. 3534 */ 3535 static inline struct page *__dev_alloc_page_noprof(gfp_t gfp_mask) 3536 { 3537 return __dev_alloc_pages_noprof(gfp_mask, 0); 3538 } 3539 #define __dev_alloc_page(...) alloc_hooks(__dev_alloc_page_noprof(__VA_ARGS__)) 3540 3541 /* 3542 * This specialized allocator has to be a macro for its allocations to be 3543 * accounted separately (to have a separate alloc_tag). 3544 */ 3545 #define dev_alloc_page() dev_alloc_pages(0) 3546 3547 /** 3548 * dev_page_is_reusable - check whether a page can be reused for network Rx 3549 * @page: the page to test 3550 * 3551 * A page shouldn't be considered for reusing/recycling if it was allocated 3552 * under memory pressure or at a distant memory node. 3553 * 3554 * Returns: false if this page should be returned to page allocator, true 3555 * otherwise. 3556 */ 3557 static inline bool dev_page_is_reusable(const struct page *page) 3558 { 3559 return likely(page_to_nid(page) == numa_mem_id() && 3560 !page_is_pfmemalloc(page)); 3561 } 3562 3563 /** 3564 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3565 * @page: The page that was allocated from skb_alloc_page 3566 * @skb: The skb that may need pfmemalloc set 3567 */ 3568 static inline void skb_propagate_pfmemalloc(const struct page *page, 3569 struct sk_buff *skb) 3570 { 3571 if (page_is_pfmemalloc(page)) 3572 skb->pfmemalloc = true; 3573 } 3574 3575 /** 3576 * skb_frag_off() - Returns the offset of a skb fragment 3577 * @frag: the paged fragment 3578 */ 3579 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3580 { 3581 return frag->offset; 3582 } 3583 3584 /** 3585 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3586 * @frag: skb fragment 3587 * @delta: value to add 3588 */ 3589 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3590 { 3591 frag->offset += delta; 3592 } 3593 3594 /** 3595 * skb_frag_off_set() - Sets the offset of a skb fragment 3596 * @frag: skb fragment 3597 * @offset: offset of fragment 3598 */ 3599 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3600 { 3601 frag->offset = offset; 3602 } 3603 3604 /** 3605 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3606 * @fragto: skb fragment where offset is set 3607 * @fragfrom: skb fragment offset is copied from 3608 */ 3609 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3610 const skb_frag_t *fragfrom) 3611 { 3612 fragto->offset = fragfrom->offset; 3613 } 3614 3615 /* Return: true if the skb_frag contains a net_iov. */ 3616 static inline bool skb_frag_is_net_iov(const skb_frag_t *frag) 3617 { 3618 return netmem_is_net_iov(frag->netmem); 3619 } 3620 3621 /** 3622 * skb_frag_net_iov - retrieve the net_iov referred to by fragment 3623 * @frag: the fragment 3624 * 3625 * Return: the &struct net_iov associated with @frag. Returns NULL if this 3626 * frag has no associated net_iov. 3627 */ 3628 static inline struct net_iov *skb_frag_net_iov(const skb_frag_t *frag) 3629 { 3630 if (!skb_frag_is_net_iov(frag)) 3631 return NULL; 3632 3633 return netmem_to_net_iov(frag->netmem); 3634 } 3635 3636 /** 3637 * skb_frag_page - retrieve the page referred to by a paged fragment 3638 * @frag: the paged fragment 3639 * 3640 * Return: the &struct page associated with @frag. Returns NULL if this frag 3641 * has no associated page. 3642 */ 3643 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3644 { 3645 if (skb_frag_is_net_iov(frag)) 3646 return NULL; 3647 3648 return netmem_to_page(frag->netmem); 3649 } 3650 3651 /** 3652 * skb_frag_netmem - retrieve the netmem referred to by a fragment 3653 * @frag: the fragment 3654 * 3655 * Return: the &netmem_ref associated with @frag. 3656 */ 3657 static inline netmem_ref skb_frag_netmem(const skb_frag_t *frag) 3658 { 3659 return frag->netmem; 3660 } 3661 3662 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb, 3663 unsigned int headroom); 3664 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb, 3665 const struct bpf_prog *prog); 3666 3667 /** 3668 * skb_frag_address - gets the address of the data contained in a paged fragment 3669 * @frag: the paged fragment buffer 3670 * 3671 * Returns: the address of the data within @frag. The page must already 3672 * be mapped. 3673 */ 3674 static inline void *skb_frag_address(const skb_frag_t *frag) 3675 { 3676 if (!skb_frag_page(frag)) 3677 return NULL; 3678 3679 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3680 } 3681 3682 /** 3683 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3684 * @frag: the paged fragment buffer 3685 * 3686 * Returns: the address of the data within @frag. Checks that the page 3687 * is mapped and returns %NULL otherwise. 3688 */ 3689 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3690 { 3691 struct page *page = skb_frag_page(frag); 3692 void *ptr; 3693 3694 if (!page) 3695 return NULL; 3696 3697 ptr = page_address(page); 3698 if (unlikely(!ptr)) 3699 return NULL; 3700 3701 return ptr + skb_frag_off(frag); 3702 } 3703 3704 /** 3705 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3706 * @fragto: skb fragment where page is set 3707 * @fragfrom: skb fragment page is copied from 3708 */ 3709 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3710 const skb_frag_t *fragfrom) 3711 { 3712 fragto->netmem = fragfrom->netmem; 3713 } 3714 3715 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3716 3717 /** 3718 * __skb_frag_dma_map - maps a paged fragment via the DMA API 3719 * @dev: the device to map the fragment to 3720 * @frag: the paged fragment to map 3721 * @offset: the offset within the fragment (starting at the 3722 * fragment's own offset) 3723 * @size: the number of bytes to map 3724 * @dir: the direction of the mapping (``PCI_DMA_*``) 3725 * 3726 * Maps the page associated with @frag to @device. 3727 */ 3728 static inline dma_addr_t __skb_frag_dma_map(struct device *dev, 3729 const skb_frag_t *frag, 3730 size_t offset, size_t size, 3731 enum dma_data_direction dir) 3732 { 3733 if (skb_frag_is_net_iov(frag)) { 3734 return netmem_to_net_iov(frag->netmem)->dma_addr + offset + 3735 frag->offset; 3736 } 3737 return dma_map_page(dev, skb_frag_page(frag), 3738 skb_frag_off(frag) + offset, size, dir); 3739 } 3740 3741 #define skb_frag_dma_map(dev, frag, ...) \ 3742 CONCATENATE(_skb_frag_dma_map, \ 3743 COUNT_ARGS(__VA_ARGS__))(dev, frag, ##__VA_ARGS__) 3744 3745 #define __skb_frag_dma_map1(dev, frag, offset, uf, uo) ({ \ 3746 const skb_frag_t *uf = (frag); \ 3747 size_t uo = (offset); \ 3748 \ 3749 __skb_frag_dma_map(dev, uf, uo, skb_frag_size(uf) - uo, \ 3750 DMA_TO_DEVICE); \ 3751 }) 3752 #define _skb_frag_dma_map1(dev, frag, offset) \ 3753 __skb_frag_dma_map1(dev, frag, offset, __UNIQUE_ID(frag_), \ 3754 __UNIQUE_ID(offset_)) 3755 #define _skb_frag_dma_map0(dev, frag) \ 3756 _skb_frag_dma_map1(dev, frag, 0) 3757 #define _skb_frag_dma_map2(dev, frag, offset, size) \ 3758 __skb_frag_dma_map(dev, frag, offset, size, DMA_TO_DEVICE) 3759 #define _skb_frag_dma_map3(dev, frag, offset, size, dir) \ 3760 __skb_frag_dma_map(dev, frag, offset, size, dir) 3761 3762 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3763 gfp_t gfp_mask) 3764 { 3765 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3766 } 3767 3768 3769 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3770 gfp_t gfp_mask) 3771 { 3772 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3773 } 3774 3775 3776 /** 3777 * skb_clone_writable - is the header of a clone writable 3778 * @skb: buffer to check 3779 * @len: length up to which to write 3780 * 3781 * Returns true if modifying the header part of the cloned buffer 3782 * does not requires the data to be copied. 3783 */ 3784 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3785 { 3786 return !skb_header_cloned(skb) && 3787 skb_headroom(skb) + len <= skb->hdr_len; 3788 } 3789 3790 static inline int skb_try_make_writable(struct sk_buff *skb, 3791 unsigned int write_len) 3792 { 3793 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3794 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3795 } 3796 3797 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3798 int cloned) 3799 { 3800 int delta = 0; 3801 3802 if (headroom > skb_headroom(skb)) 3803 delta = headroom - skb_headroom(skb); 3804 3805 if (delta || cloned) 3806 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3807 GFP_ATOMIC); 3808 return 0; 3809 } 3810 3811 /** 3812 * skb_cow - copy header of skb when it is required 3813 * @skb: buffer to cow 3814 * @headroom: needed headroom 3815 * 3816 * If the skb passed lacks sufficient headroom or its data part 3817 * is shared, data is reallocated. If reallocation fails, an error 3818 * is returned and original skb is not changed. 3819 * 3820 * The result is skb with writable area skb->head...skb->tail 3821 * and at least @headroom of space at head. 3822 */ 3823 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3824 { 3825 return __skb_cow(skb, headroom, skb_cloned(skb)); 3826 } 3827 3828 /** 3829 * skb_cow_head - skb_cow but only making the head writable 3830 * @skb: buffer to cow 3831 * @headroom: needed headroom 3832 * 3833 * This function is identical to skb_cow except that we replace the 3834 * skb_cloned check by skb_header_cloned. It should be used when 3835 * you only need to push on some header and do not need to modify 3836 * the data. 3837 */ 3838 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3839 { 3840 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3841 } 3842 3843 /** 3844 * skb_padto - pad an skbuff up to a minimal size 3845 * @skb: buffer to pad 3846 * @len: minimal length 3847 * 3848 * Pads up a buffer to ensure the trailing bytes exist and are 3849 * blanked. If the buffer already contains sufficient data it 3850 * is untouched. Otherwise it is extended. Returns zero on 3851 * success. The skb is freed on error. 3852 */ 3853 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3854 { 3855 unsigned int size = skb->len; 3856 if (likely(size >= len)) 3857 return 0; 3858 return skb_pad(skb, len - size); 3859 } 3860 3861 /** 3862 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3863 * @skb: buffer to pad 3864 * @len: minimal length 3865 * @free_on_error: free buffer on error 3866 * 3867 * Pads up a buffer to ensure the trailing bytes exist and are 3868 * blanked. If the buffer already contains sufficient data it 3869 * is untouched. Otherwise it is extended. Returns zero on 3870 * success. The skb is freed on error if @free_on_error is true. 3871 */ 3872 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3873 unsigned int len, 3874 bool free_on_error) 3875 { 3876 unsigned int size = skb->len; 3877 3878 if (unlikely(size < len)) { 3879 len -= size; 3880 if (__skb_pad(skb, len, free_on_error)) 3881 return -ENOMEM; 3882 __skb_put(skb, len); 3883 } 3884 return 0; 3885 } 3886 3887 /** 3888 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3889 * @skb: buffer to pad 3890 * @len: minimal length 3891 * 3892 * Pads up a buffer to ensure the trailing bytes exist and are 3893 * blanked. If the buffer already contains sufficient data it 3894 * is untouched. Otherwise it is extended. Returns zero on 3895 * success. The skb is freed on error. 3896 */ 3897 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3898 { 3899 return __skb_put_padto(skb, len, true); 3900 } 3901 3902 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i) 3903 __must_check; 3904 3905 static inline bool skb_can_coalesce_netmem(struct sk_buff *skb, int i, 3906 netmem_ref netmem, int off) 3907 { 3908 if (skb_zcopy(skb)) 3909 return false; 3910 if (i) { 3911 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3912 3913 return netmem == skb_frag_netmem(frag) && 3914 off == skb_frag_off(frag) + skb_frag_size(frag); 3915 } 3916 return false; 3917 } 3918 3919 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3920 const struct page *page, int off) 3921 { 3922 return skb_can_coalesce_netmem(skb, i, page_to_netmem(page), off); 3923 } 3924 3925 static inline int __skb_linearize(struct sk_buff *skb) 3926 { 3927 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3928 } 3929 3930 /** 3931 * skb_linearize - convert paged skb to linear one 3932 * @skb: buffer to linarize 3933 * 3934 * If there is no free memory -ENOMEM is returned, otherwise zero 3935 * is returned and the old skb data released. 3936 */ 3937 static inline int skb_linearize(struct sk_buff *skb) 3938 { 3939 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3940 } 3941 3942 /** 3943 * skb_has_shared_frag - can any frag be overwritten 3944 * @skb: buffer to test 3945 * 3946 * Return: true if the skb has at least one frag that might be modified 3947 * by an external entity (as in vmsplice()/sendfile()) 3948 */ 3949 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3950 { 3951 return skb_is_nonlinear(skb) && 3952 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 3953 } 3954 3955 /** 3956 * skb_linearize_cow - make sure skb is linear and writable 3957 * @skb: buffer to process 3958 * 3959 * If there is no free memory -ENOMEM is returned, otherwise zero 3960 * is returned and the old skb data released. 3961 */ 3962 static inline int skb_linearize_cow(struct sk_buff *skb) 3963 { 3964 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3965 __skb_linearize(skb) : 0; 3966 } 3967 3968 static __always_inline void 3969 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3970 unsigned int off) 3971 { 3972 if (skb->ip_summed == CHECKSUM_COMPLETE) 3973 skb->csum = csum_block_sub(skb->csum, 3974 csum_partial(start, len, 0), off); 3975 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3976 skb_checksum_start_offset(skb) < 0) 3977 skb->ip_summed = CHECKSUM_NONE; 3978 } 3979 3980 /** 3981 * skb_postpull_rcsum - update checksum for received skb after pull 3982 * @skb: buffer to update 3983 * @start: start of data before pull 3984 * @len: length of data pulled 3985 * 3986 * After doing a pull on a received packet, you need to call this to 3987 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3988 * CHECKSUM_NONE so that it can be recomputed from scratch. 3989 */ 3990 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3991 const void *start, unsigned int len) 3992 { 3993 if (skb->ip_summed == CHECKSUM_COMPLETE) 3994 skb->csum = wsum_negate(csum_partial(start, len, 3995 wsum_negate(skb->csum))); 3996 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3997 skb_checksum_start_offset(skb) < 0) 3998 skb->ip_summed = CHECKSUM_NONE; 3999 } 4000 4001 static __always_inline void 4002 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 4003 unsigned int off) 4004 { 4005 if (skb->ip_summed == CHECKSUM_COMPLETE) 4006 skb->csum = csum_block_add(skb->csum, 4007 csum_partial(start, len, 0), off); 4008 } 4009 4010 /** 4011 * skb_postpush_rcsum - update checksum for received skb after push 4012 * @skb: buffer to update 4013 * @start: start of data after push 4014 * @len: length of data pushed 4015 * 4016 * After doing a push on a received packet, you need to call this to 4017 * update the CHECKSUM_COMPLETE checksum. 4018 */ 4019 static inline void skb_postpush_rcsum(struct sk_buff *skb, 4020 const void *start, unsigned int len) 4021 { 4022 __skb_postpush_rcsum(skb, start, len, 0); 4023 } 4024 4025 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 4026 4027 /** 4028 * skb_push_rcsum - push skb and update receive checksum 4029 * @skb: buffer to update 4030 * @len: length of data pulled 4031 * 4032 * This function performs an skb_push on the packet and updates 4033 * the CHECKSUM_COMPLETE checksum. It should be used on 4034 * receive path processing instead of skb_push unless you know 4035 * that the checksum difference is zero (e.g., a valid IP header) 4036 * or you are setting ip_summed to CHECKSUM_NONE. 4037 */ 4038 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 4039 { 4040 skb_push(skb, len); 4041 skb_postpush_rcsum(skb, skb->data, len); 4042 return skb->data; 4043 } 4044 4045 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 4046 /** 4047 * pskb_trim_rcsum - trim received skb and update checksum 4048 * @skb: buffer to trim 4049 * @len: new length 4050 * 4051 * This is exactly the same as pskb_trim except that it ensures the 4052 * checksum of received packets are still valid after the operation. 4053 * It can change skb pointers. 4054 */ 4055 4056 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4057 { 4058 skb_might_realloc(skb); 4059 if (likely(len >= skb->len)) 4060 return 0; 4061 return pskb_trim_rcsum_slow(skb, len); 4062 } 4063 4064 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4065 { 4066 if (skb->ip_summed == CHECKSUM_COMPLETE) 4067 skb->ip_summed = CHECKSUM_NONE; 4068 __skb_trim(skb, len); 4069 return 0; 4070 } 4071 4072 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 4073 { 4074 if (skb->ip_summed == CHECKSUM_COMPLETE) 4075 skb->ip_summed = CHECKSUM_NONE; 4076 return __skb_grow(skb, len); 4077 } 4078 4079 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 4080 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 4081 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 4082 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 4083 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 4084 4085 #define skb_queue_walk(queue, skb) \ 4086 for (skb = (queue)->next; \ 4087 skb != (struct sk_buff *)(queue); \ 4088 skb = skb->next) 4089 4090 #define skb_queue_walk_safe(queue, skb, tmp) \ 4091 for (skb = (queue)->next, tmp = skb->next; \ 4092 skb != (struct sk_buff *)(queue); \ 4093 skb = tmp, tmp = skb->next) 4094 4095 #define skb_queue_walk_from(queue, skb) \ 4096 for (; skb != (struct sk_buff *)(queue); \ 4097 skb = skb->next) 4098 4099 #define skb_rbtree_walk(skb, root) \ 4100 for (skb = skb_rb_first(root); skb != NULL; \ 4101 skb = skb_rb_next(skb)) 4102 4103 #define skb_rbtree_walk_from(skb) \ 4104 for (; skb != NULL; \ 4105 skb = skb_rb_next(skb)) 4106 4107 #define skb_rbtree_walk_from_safe(skb, tmp) \ 4108 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 4109 skb = tmp) 4110 4111 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 4112 for (tmp = skb->next; \ 4113 skb != (struct sk_buff *)(queue); \ 4114 skb = tmp, tmp = skb->next) 4115 4116 #define skb_queue_reverse_walk(queue, skb) \ 4117 for (skb = (queue)->prev; \ 4118 skb != (struct sk_buff *)(queue); \ 4119 skb = skb->prev) 4120 4121 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 4122 for (skb = (queue)->prev, tmp = skb->prev; \ 4123 skb != (struct sk_buff *)(queue); \ 4124 skb = tmp, tmp = skb->prev) 4125 4126 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 4127 for (tmp = skb->prev; \ 4128 skb != (struct sk_buff *)(queue); \ 4129 skb = tmp, tmp = skb->prev) 4130 4131 static inline bool skb_has_frag_list(const struct sk_buff *skb) 4132 { 4133 return skb_shinfo(skb)->frag_list != NULL; 4134 } 4135 4136 static inline void skb_frag_list_init(struct sk_buff *skb) 4137 { 4138 skb_shinfo(skb)->frag_list = NULL; 4139 } 4140 4141 #define skb_walk_frags(skb, iter) \ 4142 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 4143 4144 4145 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 4146 int *err, long *timeo_p, 4147 const struct sk_buff *skb); 4148 struct sk_buff *__skb_try_recv_from_queue(struct sk_buff_head *queue, 4149 unsigned int flags, 4150 int *off, int *err, 4151 struct sk_buff **last); 4152 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 4153 struct sk_buff_head *queue, 4154 unsigned int flags, int *off, int *err, 4155 struct sk_buff **last); 4156 struct sk_buff *__skb_recv_datagram(struct sock *sk, 4157 struct sk_buff_head *sk_queue, 4158 unsigned int flags, int *off, int *err); 4159 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 4160 __poll_t datagram_poll(struct file *file, struct socket *sock, 4161 struct poll_table_struct *wait); 4162 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 4163 struct iov_iter *to, int size); 4164 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 4165 struct msghdr *msg, int size) 4166 { 4167 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 4168 } 4169 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 4170 struct msghdr *msg); 4171 int skb_copy_and_crc32c_datagram_iter(const struct sk_buff *skb, int offset, 4172 struct iov_iter *to, int len, u32 *crcp); 4173 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 4174 struct iov_iter *from, int len); 4175 int skb_copy_datagram_from_iter_full(struct sk_buff *skb, int offset, 4176 struct iov_iter *from, int len); 4177 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 4178 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 4179 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 4180 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 4181 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 4182 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 4183 int len); 4184 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 4185 struct pipe_inode_info *pipe, unsigned int len, 4186 unsigned int flags); 4187 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 4188 int len); 4189 int skb_send_sock_locked_with_flags(struct sock *sk, struct sk_buff *skb, 4190 int offset, int len, int flags); 4191 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 4192 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 4193 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 4194 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 4195 int len, int hlen); 4196 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 4197 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 4198 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 4199 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 4200 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 4201 unsigned int offset); 4202 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 4203 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 4204 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev); 4205 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 4206 int skb_vlan_pop(struct sk_buff *skb); 4207 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 4208 int skb_eth_pop(struct sk_buff *skb); 4209 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 4210 const unsigned char *src); 4211 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 4212 int mac_len, bool ethernet); 4213 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 4214 bool ethernet); 4215 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 4216 int skb_mpls_dec_ttl(struct sk_buff *skb); 4217 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 4218 gfp_t gfp); 4219 4220 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 4221 { 4222 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 4223 } 4224 4225 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 4226 { 4227 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 4228 } 4229 4230 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 4231 __wsum csum); 4232 u32 skb_crc32c(const struct sk_buff *skb, int offset, int len, u32 crc); 4233 4234 static inline void * __must_check 4235 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 4236 const void *data, int hlen, void *buffer) 4237 { 4238 if (likely(hlen - offset >= len)) 4239 return (void *)data + offset; 4240 4241 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 4242 return NULL; 4243 4244 return buffer; 4245 } 4246 4247 static inline void * __must_check 4248 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4249 { 4250 return __skb_header_pointer(skb, offset, len, skb->data, 4251 skb_headlen(skb), buffer); 4252 } 4253 4254 static inline void * __must_check 4255 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len) 4256 { 4257 if (likely(skb_headlen(skb) - offset >= len)) 4258 return skb->data + offset; 4259 return NULL; 4260 } 4261 4262 /** 4263 * skb_needs_linearize - check if we need to linearize a given skb 4264 * depending on the given device features. 4265 * @skb: socket buffer to check 4266 * @features: net device features 4267 * 4268 * Returns true if either: 4269 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4270 * 2. skb is fragmented and the device does not support SG. 4271 */ 4272 static inline bool skb_needs_linearize(struct sk_buff *skb, 4273 netdev_features_t features) 4274 { 4275 return skb_is_nonlinear(skb) && 4276 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4277 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4278 } 4279 4280 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4281 void *to, 4282 const unsigned int len) 4283 { 4284 memcpy(to, skb->data, len); 4285 } 4286 4287 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4288 const int offset, void *to, 4289 const unsigned int len) 4290 { 4291 memcpy(to, skb->data + offset, len); 4292 } 4293 4294 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4295 const void *from, 4296 const unsigned int len) 4297 { 4298 memcpy(skb->data, from, len); 4299 } 4300 4301 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4302 const int offset, 4303 const void *from, 4304 const unsigned int len) 4305 { 4306 memcpy(skb->data + offset, from, len); 4307 } 4308 4309 void skb_init(void); 4310 4311 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4312 { 4313 return skb->tstamp; 4314 } 4315 4316 /** 4317 * skb_get_timestamp - get timestamp from a skb 4318 * @skb: skb to get stamp from 4319 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4320 * 4321 * Timestamps are stored in the skb as offsets to a base timestamp. 4322 * This function converts the offset back to a struct timeval and stores 4323 * it in stamp. 4324 */ 4325 static inline void skb_get_timestamp(const struct sk_buff *skb, 4326 struct __kernel_old_timeval *stamp) 4327 { 4328 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4329 } 4330 4331 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4332 struct __kernel_sock_timeval *stamp) 4333 { 4334 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4335 4336 stamp->tv_sec = ts.tv_sec; 4337 stamp->tv_usec = ts.tv_nsec / 1000; 4338 } 4339 4340 static inline void skb_get_timestampns(const struct sk_buff *skb, 4341 struct __kernel_old_timespec *stamp) 4342 { 4343 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4344 4345 stamp->tv_sec = ts.tv_sec; 4346 stamp->tv_nsec = ts.tv_nsec; 4347 } 4348 4349 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4350 struct __kernel_timespec *stamp) 4351 { 4352 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4353 4354 stamp->tv_sec = ts.tv_sec; 4355 stamp->tv_nsec = ts.tv_nsec; 4356 } 4357 4358 static inline void __net_timestamp(struct sk_buff *skb) 4359 { 4360 skb->tstamp = ktime_get_real(); 4361 skb->tstamp_type = SKB_CLOCK_REALTIME; 4362 } 4363 4364 static inline ktime_t net_timedelta(ktime_t t) 4365 { 4366 return ktime_sub(ktime_get_real(), t); 4367 } 4368 4369 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4370 u8 tstamp_type) 4371 { 4372 skb->tstamp = kt; 4373 4374 if (kt) 4375 skb->tstamp_type = tstamp_type; 4376 else 4377 skb->tstamp_type = SKB_CLOCK_REALTIME; 4378 } 4379 4380 static inline void skb_set_delivery_type_by_clockid(struct sk_buff *skb, 4381 ktime_t kt, clockid_t clockid) 4382 { 4383 u8 tstamp_type = SKB_CLOCK_REALTIME; 4384 4385 switch (clockid) { 4386 case CLOCK_REALTIME: 4387 break; 4388 case CLOCK_MONOTONIC: 4389 tstamp_type = SKB_CLOCK_MONOTONIC; 4390 break; 4391 case CLOCK_TAI: 4392 tstamp_type = SKB_CLOCK_TAI; 4393 break; 4394 default: 4395 WARN_ON_ONCE(1); 4396 kt = 0; 4397 } 4398 4399 skb_set_delivery_time(skb, kt, tstamp_type); 4400 } 4401 4402 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4403 4404 /* It is used in the ingress path to clear the delivery_time. 4405 * If needed, set the skb->tstamp to the (rcv) timestamp. 4406 */ 4407 static inline void skb_clear_delivery_time(struct sk_buff *skb) 4408 { 4409 if (skb->tstamp_type) { 4410 skb->tstamp_type = SKB_CLOCK_REALTIME; 4411 if (static_branch_unlikely(&netstamp_needed_key)) 4412 skb->tstamp = ktime_get_real(); 4413 else 4414 skb->tstamp = 0; 4415 } 4416 } 4417 4418 static inline void skb_clear_tstamp(struct sk_buff *skb) 4419 { 4420 if (skb->tstamp_type) 4421 return; 4422 4423 skb->tstamp = 0; 4424 } 4425 4426 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4427 { 4428 if (skb->tstamp_type) 4429 return 0; 4430 4431 return skb->tstamp; 4432 } 4433 4434 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4435 { 4436 if (skb->tstamp_type != SKB_CLOCK_MONOTONIC && skb->tstamp) 4437 return skb->tstamp; 4438 4439 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4440 return ktime_get_real(); 4441 4442 return 0; 4443 } 4444 4445 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4446 { 4447 return skb_shinfo(skb)->meta_len; 4448 } 4449 4450 static inline void *skb_metadata_end(const struct sk_buff *skb) 4451 { 4452 return skb_mac_header(skb); 4453 } 4454 4455 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4456 const struct sk_buff *skb_b, 4457 u8 meta_len) 4458 { 4459 const void *a = skb_metadata_end(skb_a); 4460 const void *b = skb_metadata_end(skb_b); 4461 u64 diffs = 0; 4462 4463 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) || 4464 BITS_PER_LONG != 64) 4465 goto slow; 4466 4467 /* Using more efficient variant than plain call to memcmp(). */ 4468 switch (meta_len) { 4469 #define __it(x, op) (x -= sizeof(u##op)) 4470 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4471 case 32: diffs |= __it_diff(a, b, 64); 4472 fallthrough; 4473 case 24: diffs |= __it_diff(a, b, 64); 4474 fallthrough; 4475 case 16: diffs |= __it_diff(a, b, 64); 4476 fallthrough; 4477 case 8: diffs |= __it_diff(a, b, 64); 4478 break; 4479 case 28: diffs |= __it_diff(a, b, 64); 4480 fallthrough; 4481 case 20: diffs |= __it_diff(a, b, 64); 4482 fallthrough; 4483 case 12: diffs |= __it_diff(a, b, 64); 4484 fallthrough; 4485 case 4: diffs |= __it_diff(a, b, 32); 4486 break; 4487 default: 4488 slow: 4489 return memcmp(a - meta_len, b - meta_len, meta_len); 4490 } 4491 return diffs; 4492 } 4493 4494 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4495 const struct sk_buff *skb_b) 4496 { 4497 u8 len_a = skb_metadata_len(skb_a); 4498 u8 len_b = skb_metadata_len(skb_b); 4499 4500 if (!(len_a | len_b)) 4501 return false; 4502 4503 return len_a != len_b ? 4504 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4505 } 4506 4507 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4508 { 4509 skb_shinfo(skb)->meta_len = meta_len; 4510 } 4511 4512 static inline void skb_metadata_clear(struct sk_buff *skb) 4513 { 4514 skb_metadata_set(skb, 0); 4515 } 4516 4517 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4518 4519 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4520 4521 void skb_clone_tx_timestamp(struct sk_buff *skb); 4522 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4523 4524 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4525 4526 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4527 { 4528 } 4529 4530 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4531 { 4532 return false; 4533 } 4534 4535 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4536 4537 /** 4538 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4539 * 4540 * PHY drivers may accept clones of transmitted packets for 4541 * timestamping via their phy_driver.txtstamp method. These drivers 4542 * must call this function to return the skb back to the stack with a 4543 * timestamp. 4544 * 4545 * @skb: clone of the original outgoing packet 4546 * @hwtstamps: hardware time stamps 4547 * 4548 */ 4549 void skb_complete_tx_timestamp(struct sk_buff *skb, 4550 struct skb_shared_hwtstamps *hwtstamps); 4551 4552 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4553 struct skb_shared_hwtstamps *hwtstamps, 4554 struct sock *sk, int tstype); 4555 4556 /** 4557 * skb_tstamp_tx - queue clone of skb with send time stamps 4558 * @orig_skb: the original outgoing packet 4559 * @hwtstamps: hardware time stamps, may be NULL if not available 4560 * 4561 * If the skb has a socket associated, then this function clones the 4562 * skb (thus sharing the actual data and optional structures), stores 4563 * the optional hardware time stamping information (if non NULL) or 4564 * generates a software time stamp (otherwise), then queues the clone 4565 * to the error queue of the socket. Errors are silently ignored. 4566 */ 4567 void skb_tstamp_tx(struct sk_buff *orig_skb, 4568 struct skb_shared_hwtstamps *hwtstamps); 4569 4570 /** 4571 * skb_tx_timestamp() - Driver hook for transmit timestamping 4572 * 4573 * Ethernet MAC Drivers should call this function in their hard_xmit() 4574 * function immediately before giving the sk_buff to the MAC hardware. 4575 * 4576 * Specifically, one should make absolutely sure that this function is 4577 * called before TX completion of this packet can trigger. Otherwise 4578 * the packet could potentially already be freed. 4579 * 4580 * @skb: A socket buffer. 4581 */ 4582 static inline void skb_tx_timestamp(struct sk_buff *skb) 4583 { 4584 skb_clone_tx_timestamp(skb); 4585 if (skb_shinfo(skb)->tx_flags & (SKBTX_SW_TSTAMP | SKBTX_BPF)) 4586 skb_tstamp_tx(skb, NULL); 4587 } 4588 4589 /** 4590 * skb_complete_wifi_ack - deliver skb with wifi status 4591 * 4592 * @skb: the original outgoing packet 4593 * @acked: ack status 4594 * 4595 */ 4596 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4597 4598 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4599 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4600 4601 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4602 { 4603 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4604 skb->csum_valid || 4605 (skb->ip_summed == CHECKSUM_PARTIAL && 4606 skb_checksum_start_offset(skb) >= 0)); 4607 } 4608 4609 /** 4610 * skb_checksum_complete - Calculate checksum of an entire packet 4611 * @skb: packet to process 4612 * 4613 * This function calculates the checksum over the entire packet plus 4614 * the value of skb->csum. The latter can be used to supply the 4615 * checksum of a pseudo header as used by TCP/UDP. It returns the 4616 * checksum. 4617 * 4618 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4619 * this function can be used to verify that checksum on received 4620 * packets. In that case the function should return zero if the 4621 * checksum is correct. In particular, this function will return zero 4622 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4623 * hardware has already verified the correctness of the checksum. 4624 */ 4625 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4626 { 4627 return skb_csum_unnecessary(skb) ? 4628 0 : __skb_checksum_complete(skb); 4629 } 4630 4631 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4632 { 4633 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4634 if (skb->csum_level == 0) 4635 skb->ip_summed = CHECKSUM_NONE; 4636 else 4637 skb->csum_level--; 4638 } 4639 } 4640 4641 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4642 { 4643 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4644 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4645 skb->csum_level++; 4646 } else if (skb->ip_summed == CHECKSUM_NONE) { 4647 skb->ip_summed = CHECKSUM_UNNECESSARY; 4648 skb->csum_level = 0; 4649 } 4650 } 4651 4652 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4653 { 4654 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4655 skb->ip_summed = CHECKSUM_NONE; 4656 skb->csum_level = 0; 4657 } 4658 } 4659 4660 /* Check if we need to perform checksum complete validation. 4661 * 4662 * Returns: true if checksum complete is needed, false otherwise 4663 * (either checksum is unnecessary or zero checksum is allowed). 4664 */ 4665 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4666 bool zero_okay, 4667 __sum16 check) 4668 { 4669 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4670 skb->csum_valid = 1; 4671 __skb_decr_checksum_unnecessary(skb); 4672 return false; 4673 } 4674 4675 return true; 4676 } 4677 4678 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4679 * in checksum_init. 4680 */ 4681 #define CHECKSUM_BREAK 76 4682 4683 /* Unset checksum-complete 4684 * 4685 * Unset checksum complete can be done when packet is being modified 4686 * (uncompressed for instance) and checksum-complete value is 4687 * invalidated. 4688 */ 4689 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4690 { 4691 if (skb->ip_summed == CHECKSUM_COMPLETE) 4692 skb->ip_summed = CHECKSUM_NONE; 4693 } 4694 4695 /* Validate (init) checksum based on checksum complete. 4696 * 4697 * Return values: 4698 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4699 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4700 * checksum is stored in skb->csum for use in __skb_checksum_complete 4701 * non-zero: value of invalid checksum 4702 * 4703 */ 4704 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4705 bool complete, 4706 __wsum psum) 4707 { 4708 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4709 if (!csum_fold(csum_add(psum, skb->csum))) { 4710 skb->csum_valid = 1; 4711 return 0; 4712 } 4713 } 4714 4715 skb->csum = psum; 4716 4717 if (complete || skb->len <= CHECKSUM_BREAK) { 4718 __sum16 csum; 4719 4720 csum = __skb_checksum_complete(skb); 4721 skb->csum_valid = !csum; 4722 return csum; 4723 } 4724 4725 return 0; 4726 } 4727 4728 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4729 { 4730 return 0; 4731 } 4732 4733 /* Perform checksum validate (init). Note that this is a macro since we only 4734 * want to calculate the pseudo header which is an input function if necessary. 4735 * First we try to validate without any computation (checksum unnecessary) and 4736 * then calculate based on checksum complete calling the function to compute 4737 * pseudo header. 4738 * 4739 * Return values: 4740 * 0: checksum is validated or try to in skb_checksum_complete 4741 * non-zero: value of invalid checksum 4742 */ 4743 #define __skb_checksum_validate(skb, proto, complete, \ 4744 zero_okay, check, compute_pseudo) \ 4745 ({ \ 4746 __sum16 __ret = 0; \ 4747 skb->csum_valid = 0; \ 4748 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4749 __ret = __skb_checksum_validate_complete(skb, \ 4750 complete, compute_pseudo(skb, proto)); \ 4751 __ret; \ 4752 }) 4753 4754 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4755 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4756 4757 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4758 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4759 4760 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4761 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4762 4763 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4764 compute_pseudo) \ 4765 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4766 4767 #define skb_checksum_simple_validate(skb) \ 4768 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4769 4770 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4771 { 4772 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4773 } 4774 4775 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4776 { 4777 skb->csum = ~pseudo; 4778 skb->ip_summed = CHECKSUM_COMPLETE; 4779 } 4780 4781 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4782 do { \ 4783 if (__skb_checksum_convert_check(skb)) \ 4784 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4785 } while (0) 4786 4787 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4788 u16 start, u16 offset) 4789 { 4790 skb->ip_summed = CHECKSUM_PARTIAL; 4791 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4792 skb->csum_offset = offset - start; 4793 } 4794 4795 /* Update skbuf and packet to reflect the remote checksum offload operation. 4796 * When called, ptr indicates the starting point for skb->csum when 4797 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4798 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4799 */ 4800 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4801 int start, int offset, bool nopartial) 4802 { 4803 __wsum delta; 4804 4805 if (!nopartial) { 4806 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4807 return; 4808 } 4809 4810 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4811 __skb_checksum_complete(skb); 4812 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4813 } 4814 4815 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4816 4817 /* Adjust skb->csum since we changed the packet */ 4818 skb->csum = csum_add(skb->csum, delta); 4819 } 4820 4821 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4822 { 4823 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4824 return (void *)(skb->_nfct & NFCT_PTRMASK); 4825 #else 4826 return NULL; 4827 #endif 4828 } 4829 4830 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4831 { 4832 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4833 return skb->_nfct; 4834 #else 4835 return 0UL; 4836 #endif 4837 } 4838 4839 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4840 { 4841 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4842 skb->slow_gro |= !!nfct; 4843 skb->_nfct = nfct; 4844 #endif 4845 } 4846 4847 #ifdef CONFIG_SKB_EXTENSIONS 4848 enum skb_ext_id { 4849 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4850 SKB_EXT_BRIDGE_NF, 4851 #endif 4852 #ifdef CONFIG_XFRM 4853 SKB_EXT_SEC_PATH, 4854 #endif 4855 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4856 TC_SKB_EXT, 4857 #endif 4858 #if IS_ENABLED(CONFIG_MPTCP) 4859 SKB_EXT_MPTCP, 4860 #endif 4861 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 4862 SKB_EXT_MCTP, 4863 #endif 4864 SKB_EXT_NUM, /* must be last */ 4865 }; 4866 4867 /** 4868 * struct skb_ext - sk_buff extensions 4869 * @refcnt: 1 on allocation, deallocated on 0 4870 * @offset: offset to add to @data to obtain extension address 4871 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4872 * @data: start of extension data, variable sized 4873 * 4874 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4875 * to use 'u8' types while allowing up to 2kb worth of extension data. 4876 */ 4877 struct skb_ext { 4878 refcount_t refcnt; 4879 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4880 u8 chunks; /* same */ 4881 char data[] __aligned(8); 4882 }; 4883 4884 struct skb_ext *__skb_ext_alloc(gfp_t flags); 4885 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 4886 struct skb_ext *ext); 4887 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4888 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4889 void __skb_ext_put(struct skb_ext *ext); 4890 4891 static inline void skb_ext_put(struct sk_buff *skb) 4892 { 4893 if (skb->active_extensions) 4894 __skb_ext_put(skb->extensions); 4895 } 4896 4897 static inline void __skb_ext_copy(struct sk_buff *dst, 4898 const struct sk_buff *src) 4899 { 4900 dst->active_extensions = src->active_extensions; 4901 4902 if (src->active_extensions) { 4903 struct skb_ext *ext = src->extensions; 4904 4905 refcount_inc(&ext->refcnt); 4906 dst->extensions = ext; 4907 } 4908 } 4909 4910 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4911 { 4912 skb_ext_put(dst); 4913 __skb_ext_copy(dst, src); 4914 } 4915 4916 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4917 { 4918 return !!ext->offset[i]; 4919 } 4920 4921 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4922 { 4923 return skb->active_extensions & (1 << id); 4924 } 4925 4926 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4927 { 4928 if (skb_ext_exist(skb, id)) 4929 __skb_ext_del(skb, id); 4930 } 4931 4932 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4933 { 4934 if (skb_ext_exist(skb, id)) { 4935 struct skb_ext *ext = skb->extensions; 4936 4937 return (void *)ext + (ext->offset[id] << 3); 4938 } 4939 4940 return NULL; 4941 } 4942 4943 static inline void skb_ext_reset(struct sk_buff *skb) 4944 { 4945 if (unlikely(skb->active_extensions)) { 4946 __skb_ext_put(skb->extensions); 4947 skb->active_extensions = 0; 4948 } 4949 } 4950 4951 static inline bool skb_has_extensions(struct sk_buff *skb) 4952 { 4953 return unlikely(skb->active_extensions); 4954 } 4955 #else 4956 static inline void skb_ext_put(struct sk_buff *skb) {} 4957 static inline void skb_ext_reset(struct sk_buff *skb) {} 4958 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4959 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4960 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4961 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 4962 #endif /* CONFIG_SKB_EXTENSIONS */ 4963 4964 static inline void nf_reset_ct(struct sk_buff *skb) 4965 { 4966 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4967 nf_conntrack_put(skb_nfct(skb)); 4968 skb->_nfct = 0; 4969 #endif 4970 } 4971 4972 static inline void nf_reset_trace(struct sk_buff *skb) 4973 { 4974 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4975 skb->nf_trace = 0; 4976 #endif 4977 } 4978 4979 static inline void ipvs_reset(struct sk_buff *skb) 4980 { 4981 #if IS_ENABLED(CONFIG_IP_VS) 4982 skb->ipvs_property = 0; 4983 #endif 4984 } 4985 4986 /* Note: This doesn't put any conntrack info in dst. */ 4987 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4988 bool copy) 4989 { 4990 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4991 dst->_nfct = src->_nfct; 4992 nf_conntrack_get(skb_nfct(src)); 4993 #endif 4994 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4995 if (copy) 4996 dst->nf_trace = src->nf_trace; 4997 #endif 4998 } 4999 5000 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 5001 { 5002 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5003 nf_conntrack_put(skb_nfct(dst)); 5004 #endif 5005 dst->slow_gro = src->slow_gro; 5006 __nf_copy(dst, src, true); 5007 } 5008 5009 #ifdef CONFIG_NETWORK_SECMARK 5010 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 5011 { 5012 to->secmark = from->secmark; 5013 } 5014 5015 static inline void skb_init_secmark(struct sk_buff *skb) 5016 { 5017 skb->secmark = 0; 5018 } 5019 #else 5020 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 5021 { } 5022 5023 static inline void skb_init_secmark(struct sk_buff *skb) 5024 { } 5025 #endif 5026 5027 static inline int secpath_exists(const struct sk_buff *skb) 5028 { 5029 #ifdef CONFIG_XFRM 5030 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 5031 #else 5032 return 0; 5033 #endif 5034 } 5035 5036 static inline bool skb_irq_freeable(const struct sk_buff *skb) 5037 { 5038 return !skb->destructor && 5039 !secpath_exists(skb) && 5040 !skb_nfct(skb) && 5041 !skb->_skb_refdst && 5042 !skb_has_frag_list(skb); 5043 } 5044 5045 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 5046 { 5047 skb->queue_mapping = queue_mapping; 5048 } 5049 5050 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 5051 { 5052 return skb->queue_mapping; 5053 } 5054 5055 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 5056 { 5057 to->queue_mapping = from->queue_mapping; 5058 } 5059 5060 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 5061 { 5062 skb->queue_mapping = rx_queue + 1; 5063 } 5064 5065 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 5066 { 5067 return skb->queue_mapping - 1; 5068 } 5069 5070 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 5071 { 5072 return skb->queue_mapping != 0; 5073 } 5074 5075 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 5076 { 5077 skb->dst_pending_confirm = val; 5078 } 5079 5080 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 5081 { 5082 return skb->dst_pending_confirm != 0; 5083 } 5084 5085 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 5086 { 5087 #ifdef CONFIG_XFRM 5088 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 5089 #else 5090 return NULL; 5091 #endif 5092 } 5093 5094 static inline bool skb_is_gso(const struct sk_buff *skb) 5095 { 5096 return skb_shinfo(skb)->gso_size; 5097 } 5098 5099 /* Note: Should be called only if skb_is_gso(skb) is true */ 5100 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 5101 { 5102 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 5103 } 5104 5105 /* Note: Should be called only if skb_is_gso(skb) is true */ 5106 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 5107 { 5108 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 5109 } 5110 5111 /* Note: Should be called only if skb_is_gso(skb) is true */ 5112 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 5113 { 5114 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 5115 } 5116 5117 static inline void skb_gso_reset(struct sk_buff *skb) 5118 { 5119 skb_shinfo(skb)->gso_size = 0; 5120 skb_shinfo(skb)->gso_segs = 0; 5121 skb_shinfo(skb)->gso_type = 0; 5122 } 5123 5124 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 5125 u16 increment) 5126 { 5127 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5128 return; 5129 shinfo->gso_size += increment; 5130 } 5131 5132 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 5133 u16 decrement) 5134 { 5135 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5136 return; 5137 shinfo->gso_size -= decrement; 5138 } 5139 5140 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 5141 5142 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 5143 { 5144 /* LRO sets gso_size but not gso_type, whereas if GSO is really 5145 * wanted then gso_type will be set. */ 5146 const struct skb_shared_info *shinfo = skb_shinfo(skb); 5147 5148 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 5149 unlikely(shinfo->gso_type == 0)) { 5150 __skb_warn_lro_forwarding(skb); 5151 return true; 5152 } 5153 return false; 5154 } 5155 5156 static inline void skb_forward_csum(struct sk_buff *skb) 5157 { 5158 /* Unfortunately we don't support this one. Any brave souls? */ 5159 if (skb->ip_summed == CHECKSUM_COMPLETE) 5160 skb->ip_summed = CHECKSUM_NONE; 5161 } 5162 5163 /** 5164 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 5165 * @skb: skb to check 5166 * 5167 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 5168 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 5169 * use this helper, to document places where we make this assertion. 5170 */ 5171 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 5172 { 5173 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 5174 } 5175 5176 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 5177 5178 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 5179 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 5180 unsigned int transport_len, 5181 __sum16(*skb_chkf)(struct sk_buff *skb)); 5182 5183 /** 5184 * skb_head_is_locked - Determine if the skb->head is locked down 5185 * @skb: skb to check 5186 * 5187 * The head on skbs build around a head frag can be removed if they are 5188 * not cloned. This function returns true if the skb head is locked down 5189 * due to either being allocated via kmalloc, or by being a clone with 5190 * multiple references to the head. 5191 */ 5192 static inline bool skb_head_is_locked(const struct sk_buff *skb) 5193 { 5194 return !skb->head_frag || skb_cloned(skb); 5195 } 5196 5197 /* Local Checksum Offload. 5198 * Compute outer checksum based on the assumption that the 5199 * inner checksum will be offloaded later. 5200 * See Documentation/networking/checksum-offloads.rst for 5201 * explanation of how this works. 5202 * Fill in outer checksum adjustment (e.g. with sum of outer 5203 * pseudo-header) before calling. 5204 * Also ensure that inner checksum is in linear data area. 5205 */ 5206 static inline __wsum lco_csum(struct sk_buff *skb) 5207 { 5208 unsigned char *csum_start = skb_checksum_start(skb); 5209 unsigned char *l4_hdr = skb_transport_header(skb); 5210 __wsum partial; 5211 5212 /* Start with complement of inner checksum adjustment */ 5213 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 5214 skb->csum_offset)); 5215 5216 /* Add in checksum of our headers (incl. outer checksum 5217 * adjustment filled in by caller) and return result. 5218 */ 5219 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5220 } 5221 5222 static inline bool skb_is_redirected(const struct sk_buff *skb) 5223 { 5224 return skb->redirected; 5225 } 5226 5227 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5228 { 5229 skb->redirected = 1; 5230 #ifdef CONFIG_NET_REDIRECT 5231 skb->from_ingress = from_ingress; 5232 if (skb->from_ingress) 5233 skb_clear_tstamp(skb); 5234 #endif 5235 } 5236 5237 static inline void skb_reset_redirect(struct sk_buff *skb) 5238 { 5239 skb->redirected = 0; 5240 } 5241 5242 static inline void skb_set_redirected_noclear(struct sk_buff *skb, 5243 bool from_ingress) 5244 { 5245 skb->redirected = 1; 5246 #ifdef CONFIG_NET_REDIRECT 5247 skb->from_ingress = from_ingress; 5248 #endif 5249 } 5250 5251 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5252 { 5253 #if IS_ENABLED(CONFIG_IP_SCTP) 5254 return skb->csum_not_inet; 5255 #else 5256 return 0; 5257 #endif 5258 } 5259 5260 static inline void skb_reset_csum_not_inet(struct sk_buff *skb) 5261 { 5262 skb->ip_summed = CHECKSUM_NONE; 5263 #if IS_ENABLED(CONFIG_IP_SCTP) 5264 skb->csum_not_inet = 0; 5265 #endif 5266 } 5267 5268 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5269 const u64 kcov_handle) 5270 { 5271 #ifdef CONFIG_KCOV 5272 skb->kcov_handle = kcov_handle; 5273 #endif 5274 } 5275 5276 static inline u64 skb_get_kcov_handle(struct sk_buff *skb) 5277 { 5278 #ifdef CONFIG_KCOV 5279 return skb->kcov_handle; 5280 #else 5281 return 0; 5282 #endif 5283 } 5284 5285 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5286 { 5287 #ifdef CONFIG_PAGE_POOL 5288 skb->pp_recycle = 1; 5289 #endif 5290 } 5291 5292 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter, 5293 ssize_t maxsize); 5294 5295 #endif /* __KERNEL__ */ 5296 #endif /* _LINUX_SKBUFF_H */ 5297