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 #define SKB_MAX_RX_HEADROOM (U16_MAX / 2) 272 273 /* return minimum truesize of one skb containing X bytes of data */ 274 #define SKB_TRUESIZE(X) ((X) + \ 275 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 276 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 277 278 struct net_device; 279 struct scatterlist; 280 struct pipe_inode_info; 281 struct iov_iter; 282 struct napi_struct; 283 struct bpf_prog; 284 union bpf_attr; 285 struct skb_ext; 286 struct ts_config; 287 288 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 289 struct nf_bridge_info { 290 enum { 291 BRNF_PROTO_UNCHANGED, 292 BRNF_PROTO_8021Q, 293 BRNF_PROTO_PPPOE 294 } orig_proto:8; 295 u8 pkt_otherhost:1; 296 u8 in_prerouting:1; 297 u8 bridged_dnat:1; 298 u8 sabotage_in_done:1; 299 __u16 frag_max_size; 300 int physinif; 301 302 /* always valid & non-NULL from FORWARD on, for physdev match */ 303 struct net_device *physoutdev; 304 union { 305 /* prerouting: detect dnat in orig/reply direction */ 306 __be32 ipv4_daddr; 307 struct in6_addr ipv6_daddr; 308 309 /* after prerouting + nat detected: store original source 310 * mac since neigh resolution overwrites it, only used while 311 * skb is out in neigh layer. 312 */ 313 char neigh_header[8]; 314 }; 315 }; 316 #endif 317 318 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 319 /* Chain in tc_skb_ext will be used to share the tc chain with 320 * ovs recirc_id. It will be set to the current chain by tc 321 * and read by ovs to recirc_id. 322 */ 323 struct tc_skb_ext { 324 union { 325 u64 act_miss_cookie; 326 __u32 chain; 327 }; 328 __u16 mru; 329 __u16 zone; 330 u8 post_ct:1; 331 u8 post_ct_snat:1; 332 u8 post_ct_dnat:1; 333 u8 act_miss:1; /* Set if act_miss_cookie is used */ 334 u8 l2_miss:1; /* Set by bridge upon FDB or MDB miss */ 335 }; 336 #endif 337 338 struct sk_buff_head { 339 /* These two members must be first to match sk_buff. */ 340 struct_group_tagged(sk_buff_list, list, 341 struct sk_buff *next; 342 struct sk_buff *prev; 343 ); 344 345 __u32 qlen; 346 spinlock_t lock; 347 }; 348 349 struct sk_buff; 350 351 #ifndef CONFIG_MAX_SKB_FRAGS 352 # define CONFIG_MAX_SKB_FRAGS 17 353 #endif 354 355 #define MAX_SKB_FRAGS CONFIG_MAX_SKB_FRAGS 356 357 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 358 * segment using its current segmentation instead. 359 */ 360 #define GSO_BY_FRAGS 0xFFFF 361 362 /* All frags in an skb must either all be struct page-backed or all belong to 363 * the same memory provider instance (e.g. the same devmem binding or io_uring 364 * area). Mixing page and memory-provider frags, or mixing frags from different 365 * memory providers within a single skb (including via coalescing), is not 366 * allowed. 367 */ 368 typedef struct skb_frag { 369 netmem_ref netmem; 370 unsigned int len; 371 unsigned int offset; 372 } skb_frag_t; 373 374 /** 375 * skb_frag_size() - Returns the size of a skb fragment 376 * @frag: skb fragment 377 */ 378 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 379 { 380 return frag->len; 381 } 382 383 /** 384 * skb_frag_size_set() - Sets the size of a skb fragment 385 * @frag: skb fragment 386 * @size: size of fragment 387 */ 388 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 389 { 390 frag->len = size; 391 } 392 393 /** 394 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 395 * @frag: skb fragment 396 * @delta: value to add 397 */ 398 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 399 { 400 frag->len += delta; 401 } 402 403 /** 404 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 405 * @frag: skb fragment 406 * @delta: value to subtract 407 */ 408 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 409 { 410 frag->len -= delta; 411 } 412 413 /** 414 * skb_frag_must_loop - Test if %p is a high memory page 415 * @p: fragment's page 416 */ 417 static inline bool skb_frag_must_loop(struct page *p) 418 { 419 #if defined(CONFIG_HIGHMEM) 420 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 421 return true; 422 #endif 423 return false; 424 } 425 426 /** 427 * skb_frag_foreach_page - loop over pages in a fragment 428 * 429 * @f: skb frag to operate on 430 * @f_off: offset from start of f->netmem 431 * @f_len: length from f_off to loop over 432 * @p: (temp var) current page 433 * @p_off: (temp var) offset from start of current page, 434 * non-zero only on first page. 435 * @p_len: (temp var) length in current page, 436 * < PAGE_SIZE only on first and last page. 437 * @copied: (temp var) length so far, excluding current p_len. 438 * 439 * A fragment can hold a compound page, in which case per-page 440 * operations, notably kmap_atomic, must be called for each 441 * regular page. 442 */ 443 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 444 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 445 p_off = (f_off) & (PAGE_SIZE - 1), \ 446 p_len = skb_frag_must_loop(p) ? \ 447 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 448 copied = 0; \ 449 copied < f_len; \ 450 copied += p_len, p++, p_off = 0, \ 451 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 452 453 /** 454 * struct skb_shared_hwtstamps - hardware time stamps 455 * @hwtstamp: hardware time stamp transformed into duration 456 * since arbitrary point in time 457 * @netdev_data: address/cookie of network device driver used as 458 * reference to actual hardware time stamp 459 * 460 * Software time stamps generated by ktime_get_real() are stored in 461 * skb->tstamp. 462 * 463 * hwtstamps can only be compared against other hwtstamps from 464 * the same device. 465 * 466 * This structure is attached to packets as part of the 467 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 468 */ 469 struct skb_shared_hwtstamps { 470 union { 471 ktime_t hwtstamp; 472 void *netdev_data; 473 }; 474 }; 475 476 /* Definitions for tx_flags in struct skb_shared_info */ 477 enum { 478 /* generate hardware time stamp */ 479 SKBTX_HW_TSTAMP_NOBPF = 1 << 0, 480 481 /* generate software time stamp when queueing packet to NIC */ 482 SKBTX_SW_TSTAMP = 1 << 1, 483 484 /* device driver is going to provide hardware time stamp */ 485 SKBTX_IN_PROGRESS = 1 << 2, 486 487 /* generate software time stamp on packet tx completion */ 488 SKBTX_COMPLETION_TSTAMP = 1 << 3, 489 490 /* determine hardware time stamp based on time or cycles */ 491 SKBTX_HW_TSTAMP_NETDEV = 1 << 5, 492 493 /* generate software time stamp when entering packet scheduling */ 494 SKBTX_SCHED_TSTAMP = 1 << 6, 495 496 /* used for bpf extension when a bpf program is loaded */ 497 SKBTX_BPF = 1 << 7, 498 }; 499 500 #define SKBTX_HW_TSTAMP (SKBTX_HW_TSTAMP_NOBPF | SKBTX_BPF) 501 502 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 503 SKBTX_SCHED_TSTAMP | \ 504 SKBTX_BPF | \ 505 SKBTX_COMPLETION_TSTAMP) 506 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \ 507 SKBTX_ANY_SW_TSTAMP) 508 509 /* Definitions for flags in struct skb_shared_info */ 510 enum { 511 /* use zcopy routines */ 512 SKBFL_ZEROCOPY_ENABLE = BIT(0), 513 514 /* This indicates at least one fragment might be overwritten 515 * (as in vmsplice(), sendfile() ...) 516 * If we need to compute a TX checksum, we'll need to copy 517 * all frags to avoid possible bad checksum 518 */ 519 SKBFL_SHARED_FRAG = BIT(1), 520 521 /* segment contains only zerocopy data and should not be 522 * charged to the kernel memory. 523 */ 524 SKBFL_PURE_ZEROCOPY = BIT(2), 525 526 SKBFL_DONT_ORPHAN = BIT(3), 527 528 /* page references are managed by the ubuf_info, so it's safe to 529 * use frags only up until ubuf_info is released 530 */ 531 SKBFL_MANAGED_FRAG_REFS = BIT(4), 532 }; 533 534 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG) 535 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \ 536 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS) 537 538 struct ubuf_info_ops { 539 void (*complete)(struct sk_buff *, struct ubuf_info *, 540 bool zerocopy_success); 541 /* has to be compatible with skb_zcopy_set() */ 542 int (*link_skb)(struct sk_buff *skb, struct ubuf_info *uarg); 543 }; 544 545 /* 546 * The callback notifies userspace to release buffers when skb DMA is done in 547 * lower device, the skb last reference should be 0 when calling this. 548 * The zerocopy_success argument is true if zero copy transmit occurred, 549 * false on data copy or out of memory error caused by data copy attempt. 550 * The ctx field is used to track device context. 551 * The desc field is used to track userspace buffer index. 552 */ 553 struct ubuf_info { 554 const struct ubuf_info_ops *ops; 555 refcount_t refcnt; 556 u8 flags; 557 }; 558 559 struct ubuf_info_msgzc { 560 struct ubuf_info ubuf; 561 562 union { 563 struct { 564 unsigned long desc; 565 void *ctx; 566 }; 567 struct { 568 u32 id; 569 u16 len; 570 u16 zerocopy:1; 571 u32 bytelen; 572 }; 573 }; 574 575 struct mmpin { 576 struct user_struct *user; 577 unsigned int num_pg; 578 } mmp; 579 }; 580 581 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 582 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \ 583 ubuf) 584 585 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 586 void mm_unaccount_pinned_pages(struct mmpin *mmp); 587 588 /* Preserve some data across TX submission and completion. 589 * 590 * Note, this state is stored in the driver. Extending the layout 591 * might need some special care. 592 */ 593 struct xsk_tx_metadata_compl { 594 __u64 *tx_timestamp; 595 }; 596 597 /* This data is invariant across clones and lives at 598 * the end of the header data, ie. at skb->end. 599 */ 600 struct skb_shared_info { 601 __u8 flags; 602 __u8 meta_len; 603 __u8 nr_frags; 604 __u8 tx_flags; 605 unsigned short gso_size; 606 /* Warning: this field is not always filled in (UFO)! */ 607 unsigned short gso_segs; 608 struct sk_buff *frag_list; 609 union { 610 struct skb_shared_hwtstamps hwtstamps; 611 struct xsk_tx_metadata_compl xsk_meta; 612 }; 613 unsigned int gso_type; 614 u32 tskey; 615 616 /* 617 * Warning : all fields before dataref are cleared in __alloc_skb() 618 */ 619 atomic_t dataref; 620 621 union { 622 struct { 623 u32 xdp_frags_size; 624 u32 xdp_frags_truesize; 625 }; 626 627 /* 628 * Intermediate layers must ensure that destructor_arg 629 * remains valid until skb destructor. 630 */ 631 void *destructor_arg; 632 }; 633 634 /* must be last field, see pskb_expand_head() */ 635 skb_frag_t frags[MAX_SKB_FRAGS]; 636 }; 637 638 /** 639 * DOC: dataref and headerless skbs 640 * 641 * Transport layers send out clones of payload skbs they hold for 642 * retransmissions. To allow lower layers of the stack to prepend their headers 643 * we split &skb_shared_info.dataref into two parts. 644 * The lower 30 bits count the overall number of references. 645 * Bit 30 indicates whether one of the references is payload-only. 646 * skb_header_cloned() checks if skb is allowed to add / write the headers. 647 * 648 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr 649 * (via __skb_header_release()). Any clone created from marked skb will get 650 * &sk_buff.hdr_len populated with the available headroom. 651 * If there's the only clone in existence it's able to modify the headroom 652 * at will. The sequence of calls inside the transport layer is:: 653 * 654 * <alloc skb> 655 * skb_reserve() 656 * __skb_header_release() 657 * skb_clone() 658 * // send the clone down the stack 659 * 660 * This is not a very generic construct and it depends on the transport layers 661 * doing the right thing. In practice there's usually only one payload-only skb. 662 * Having multiple payload-only skbs with different lengths of hdr_len is not 663 * possible. The payload-only skbs should never leave their owner. 664 */ 665 #define SKB_DATAREF_SHIFT 30 666 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 667 668 669 enum { 670 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 671 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 672 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 673 }; 674 675 enum { 676 SKB_GSO_TCPV4 = 1 << 0, 677 678 /* This indicates the skb is from an untrusted source. */ 679 SKB_GSO_DODGY = 1 << 1, 680 681 /* This indicates the tcp segment has CWR set. */ 682 SKB_GSO_TCP_ECN = 1 << 2, 683 684 __SKB_GSO_TCP_FIXEDID = 1 << 3, 685 686 SKB_GSO_TCPV6 = 1 << 4, 687 688 SKB_GSO_FCOE = 1 << 5, 689 690 SKB_GSO_GRE = 1 << 6, 691 692 SKB_GSO_GRE_CSUM = 1 << 7, 693 694 SKB_GSO_IPXIP4 = 1 << 8, 695 696 SKB_GSO_IPXIP6 = 1 << 9, 697 698 SKB_GSO_UDP_TUNNEL = 1 << 10, 699 700 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 701 702 SKB_GSO_PARTIAL = 1 << 12, 703 704 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 705 706 SKB_GSO_SCTP = 1 << 14, 707 708 SKB_GSO_ESP = 1 << 15, 709 710 SKB_GSO_UDP = 1 << 16, 711 712 SKB_GSO_UDP_L4 = 1 << 17, 713 714 SKB_GSO_FRAGLIST = 1 << 18, 715 716 SKB_GSO_TCP_ACCECN = 1 << 19, 717 718 /* These indirectly map onto the same netdev feature. 719 * If NETIF_F_TSO_MANGLEID is set it may mangle both inner and outer IDs. 720 */ 721 SKB_GSO_TCP_FIXEDID = 1 << 30, 722 SKB_GSO_TCP_FIXEDID_INNER = 1 << 31, 723 }; 724 725 #if BITS_PER_LONG > 32 726 #define NET_SKBUFF_DATA_USES_OFFSET 1 727 #endif 728 729 #ifdef NET_SKBUFF_DATA_USES_OFFSET 730 typedef unsigned int sk_buff_data_t; 731 #else 732 typedef unsigned char *sk_buff_data_t; 733 #endif 734 735 enum skb_tstamp_type { 736 SKB_CLOCK_REALTIME, 737 SKB_CLOCK_MONOTONIC, 738 SKB_CLOCK_TAI, 739 __SKB_CLOCK_MAX = SKB_CLOCK_TAI, 740 }; 741 742 /** 743 * DOC: Basic sk_buff geometry 744 * 745 * struct sk_buff itself is a metadata structure and does not hold any packet 746 * data. All the data is held in associated buffers. 747 * 748 * &sk_buff.head points to the main "head" buffer. The head buffer is divided 749 * into two parts: 750 * 751 * - data buffer, containing headers and sometimes payload; 752 * this is the part of the skb operated on by the common helpers 753 * such as skb_put() or skb_pull(); 754 * - shared info (struct skb_shared_info) which holds an array of pointers 755 * to read-only data in the (page, offset, length) format. 756 * 757 * Optionally &skb_shared_info.frag_list may point to another skb. 758 * 759 * Basic diagram may look like this:: 760 * 761 * --------------- 762 * | sk_buff | 763 * --------------- 764 * ,--------------------------- + head 765 * / ,----------------- + data 766 * / / ,----------- + tail 767 * | | | , + end 768 * | | | | 769 * v v v v 770 * ----------------------------------------------- 771 * | headroom | data | tailroom | skb_shared_info | 772 * ----------------------------------------------- 773 * + [page frag] 774 * + [page frag] 775 * + [page frag] 776 * + [page frag] --------- 777 * + frag_list --> | sk_buff | 778 * --------- 779 * 780 */ 781 782 /** 783 * struct sk_buff - socket buffer 784 * @next: Next buffer in list 785 * @prev: Previous buffer in list 786 * @tstamp: Time we arrived/left 787 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 788 * for retransmit timer 789 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 790 * @list: queue head 791 * @ll_node: anchor in an llist (eg socket defer_list) 792 * @sk: Socket we are owned by 793 * @dev: Device we arrived on/are leaving by 794 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 795 * @cb: Control buffer. Free for use by every layer. Put private vars here 796 * @_skb_refdst: destination entry (with norefcount bit) 797 * @len: Length of actual data 798 * @data_len: Data length 799 * @mac_len: Length of link layer header 800 * @hdr_len: writable header length of cloned skb 801 * @csum: Checksum (must include start/offset pair) 802 * @csum_start: Offset from skb->head where checksumming should start 803 * @csum_offset: Offset from csum_start where checksum should be stored 804 * @priority: Packet queueing priority 805 * @ignore_df: allow local fragmentation 806 * @cloned: Head may be cloned (check refcnt to be sure) 807 * @ip_summed: Driver fed us an IP checksum 808 * @nohdr: Payload reference only, must not modify header 809 * @pkt_type: Packet class 810 * @fclone: skbuff clone status 811 * @ipvs_property: skbuff is owned by ipvs 812 * @inner_protocol_type: whether the inner protocol is 813 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 814 * @remcsum_offload: remote checksum offload is enabled 815 * @offload_fwd_mark: Packet was L2-forwarded in hardware 816 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 817 * @tc_skip_classify: do not classify packet. set by IFB device 818 * @tc_at_ingress: used within tc_classify to distinguish in/egress 819 * @redirected: packet was redirected by packet classifier 820 * @from_ingress: packet was redirected from the ingress path 821 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 822 * @peeked: this packet has been seen already, so stats have been 823 * done for it, don't do them again 824 * @nf_trace: netfilter packet trace flag 825 * @protocol: Packet protocol from driver 826 * @destructor: Destruct function 827 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 828 * @_sk_redir: socket redirection information for skmsg 829 * @_nfct: Associated connection, if any (with nfctinfo bits) 830 * @skb_iif: ifindex of device we arrived on 831 * @tc_depth: counter for packet duplication 832 * @tc_index: Traffic control index 833 * @hash: the packet hash 834 * @queue_mapping: Queue mapping for multiqueue devices 835 * @head_frag: skb was allocated from page fragments, 836 * not allocated by kmalloc() or vmalloc(). 837 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 838 * @pp_recycle: mark the packet for recycling instead of freeing (implies 839 * page_pool support on driver) 840 * @active_extensions: active extensions (skb_ext_id types) 841 * @ndisc_nodetype: router type (from link layer) 842 * @ooo_okay: allow the mapping of a socket to a queue to be changed 843 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 844 * ports. 845 * @sw_hash: indicates hash was computed in software stack 846 * @wifi_acked_valid: wifi_acked was set 847 * @wifi_acked: whether frame was acked on wifi or not 848 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 849 * @encapsulation: indicates the inner headers in the skbuff are valid 850 * @encap_hdr_csum: software checksum is needed 851 * @csum_valid: checksum is already valid 852 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 853 * @csum_complete_sw: checksum was completed by software 854 * @csum_level: indicates the number of consecutive checksums found in 855 * the packet minus one that have been verified as 856 * CHECKSUM_UNNECESSARY (max 3) 857 * @unreadable: indicates that at least 1 of the fragments in this skb is 858 * unreadable. 859 * @dst_pending_confirm: need to confirm neighbour 860 * @decrypted: Decrypted SKB 861 * @slow_gro: state present at GRO time, slower prepare step required 862 * @tstamp_type: When set, skb->tstamp has the 863 * delivery_time clock base of skb->tstamp. 864 * @napi_id: id of the NAPI struct this skb came from 865 * @sender_cpu: (aka @napi_id) source CPU in XPS 866 * @alloc_cpu: CPU which did the skb allocation. 867 * @secmark: security marking 868 * @mark: Generic packet mark 869 * @reserved_tailroom: (aka @mark) number of bytes of free space available 870 * at the tail of an sk_buff 871 * @vlan_all: vlan fields (proto & tci) 872 * @vlan_proto: vlan encapsulation protocol 873 * @vlan_tci: vlan tag control information 874 * @inner_protocol: Protocol (encapsulation) 875 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 876 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 877 * @inner_transport_header: Inner transport layer header (encapsulation) 878 * @inner_network_header: Network layer header (encapsulation) 879 * @inner_mac_header: Link layer header (encapsulation) 880 * @transport_header: Transport layer header 881 * @network_header: Network layer header 882 * @mac_header: Link layer header 883 * @kcov_handle: KCOV remote handle for remote coverage collection 884 * @tail: Tail pointer 885 * @end: End pointer 886 * @head: Head of buffer 887 * @data: Data head pointer 888 * @truesize: Buffer size 889 * @users: User count - see {datagram,tcp}.c 890 * @extensions: allocated extensions, valid if active_extensions is nonzero 891 */ 892 893 struct sk_buff { 894 union { 895 struct { 896 /* These two members must be first to match sk_buff_head. */ 897 struct sk_buff *next; 898 struct sk_buff *prev; 899 900 union { 901 struct net_device *dev; 902 /* Some protocols might use this space to store information, 903 * while device pointer would be NULL. 904 * UDP receive path is one user. 905 */ 906 unsigned long dev_scratch; 907 }; 908 }; 909 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 910 struct list_head list; 911 struct llist_node ll_node; 912 }; 913 914 struct sock *sk; 915 916 union { 917 ktime_t tstamp; 918 u64 skb_mstamp_ns; /* earliest departure time */ 919 }; 920 /* 921 * This is the control buffer. It is free to use for every 922 * layer. Please put your private variables there. If you 923 * want to keep them across layers you have to do a skb_clone() 924 * first. This is owned by whoever has the skb queued ATM. 925 */ 926 char cb[48] __aligned(8); 927 928 union { 929 struct { 930 unsigned long _skb_refdst; 931 void (*destructor)(struct sk_buff *skb); 932 }; 933 struct list_head tcp_tsorted_anchor; 934 #ifdef CONFIG_NET_SOCK_MSG 935 unsigned long _sk_redir; 936 #endif 937 }; 938 939 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 940 unsigned long _nfct; 941 #endif 942 unsigned int len, 943 data_len; 944 __u16 mac_len, 945 hdr_len; 946 947 /* Following fields are _not_ copied in __copy_skb_header() 948 * Note that queue_mapping is here mostly to fill a hole. 949 */ 950 __u16 queue_mapping; 951 952 /* if you move cloned around you also must adapt those constants */ 953 #ifdef __BIG_ENDIAN_BITFIELD 954 #define CLONED_MASK (1 << 7) 955 #else 956 #define CLONED_MASK 1 957 #endif 958 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 959 960 /* private: */ 961 __u8 __cloned_offset[0]; 962 /* public: */ 963 __u8 cloned:1, 964 nohdr:1, 965 fclone:2, 966 peeked:1, 967 head_frag:1, 968 pfmemalloc:1, 969 pp_recycle:1; /* page_pool recycle indicator */ 970 #ifdef CONFIG_SKB_EXTENSIONS 971 __u8 active_extensions; 972 #endif 973 974 /* Fields enclosed in headers group are copied 975 * using a single memcpy() in __copy_skb_header() 976 */ 977 struct_group(headers, 978 979 /* private: */ 980 __u8 __pkt_type_offset[0]; 981 /* public: */ 982 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 983 __u8 ignore_df:1; 984 __u8 dst_pending_confirm:1; 985 __u8 ip_summed:2; 986 __u8 ooo_okay:1; 987 988 /* private: */ 989 __u8 __mono_tc_offset[0]; 990 /* public: */ 991 __u8 tstamp_type:2; /* See skb_tstamp_type */ 992 #ifdef CONFIG_NET_XGRESS 993 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 994 __u8 tc_skip_classify:1; 995 #endif 996 __u8 remcsum_offload:1; 997 __u8 csum_complete_sw:1; 998 __u8 csum_level:2; 999 __u8 inner_protocol_type:1; 1000 1001 __u8 l4_hash:1; 1002 __u8 sw_hash:1; 1003 #ifdef CONFIG_WIRELESS 1004 __u8 wifi_acked_valid:1; 1005 __u8 wifi_acked:1; 1006 #endif 1007 __u8 no_fcs:1; 1008 /* Indicates the inner headers are valid in the skbuff. */ 1009 __u8 encapsulation:1; 1010 __u8 encap_hdr_csum:1; 1011 __u8 csum_valid:1; 1012 #ifdef CONFIG_IPV6_NDISC_NODETYPE 1013 __u8 ndisc_nodetype:2; 1014 #endif 1015 1016 #if IS_ENABLED(CONFIG_IP_VS) 1017 __u8 ipvs_property:1; 1018 #endif 1019 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 1020 __u8 nf_trace:1; 1021 #endif 1022 #ifdef CONFIG_NET_SWITCHDEV 1023 __u8 offload_fwd_mark:1; 1024 __u8 offload_l3_fwd_mark:1; 1025 #endif 1026 __u8 redirected:1; 1027 #ifdef CONFIG_NET_REDIRECT 1028 __u8 from_ingress:1; 1029 #endif 1030 #ifdef CONFIG_NET_EGRESS 1031 __u8 nf_skip_egress:1; 1032 #endif 1033 #ifdef CONFIG_SKB_DECRYPTED 1034 __u8 decrypted:1; 1035 #endif 1036 __u8 slow_gro:1; 1037 #if IS_ENABLED(CONFIG_IP_SCTP) 1038 __u8 csum_not_inet:1; 1039 #endif 1040 __u8 unreadable:1; 1041 __u8 tc_depth:2; 1042 #if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS) 1043 __u16 tc_index; /* traffic control index */ 1044 #endif 1045 1046 union { 1047 __wsum csum; 1048 struct { 1049 __u16 csum_start; 1050 __u16 csum_offset; 1051 }; 1052 }; 1053 __u32 priority; 1054 int skb_iif; 1055 __u32 hash; 1056 union { 1057 u32 vlan_all; 1058 struct { 1059 __be16 vlan_proto; 1060 __u16 vlan_tci; 1061 }; 1062 }; 1063 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1064 union { 1065 unsigned int napi_id; 1066 unsigned int sender_cpu; 1067 }; 1068 #endif 1069 #ifdef CONFIG_NETWORK_SECMARK 1070 __u32 secmark; 1071 #endif 1072 1073 union { 1074 __u32 mark; 1075 __u32 reserved_tailroom; 1076 }; 1077 1078 union { 1079 __be16 inner_protocol; 1080 __u8 inner_ipproto; 1081 }; 1082 1083 __u16 inner_transport_header; 1084 __u16 inner_network_header; 1085 __u16 inner_mac_header; 1086 1087 __be16 protocol; 1088 __u16 transport_header; 1089 __u16 network_header; 1090 __u16 mac_header; 1091 1092 struct kcov_common_handle_id kcov_handle; 1093 1094 ); /* end headers group */ 1095 1096 u16 alloc_cpu; 1097 1098 /* These elements must be at the end, see alloc_skb() for details. */ 1099 sk_buff_data_t tail; 1100 sk_buff_data_t end; 1101 unsigned char *head, 1102 *data; 1103 unsigned int truesize; 1104 refcount_t users; 1105 1106 #ifdef CONFIG_SKB_EXTENSIONS 1107 /* only usable after checking ->active_extensions != 0 */ 1108 struct skb_ext *extensions; 1109 #endif 1110 }; 1111 1112 /* if you move pkt_type around you also must adapt those constants */ 1113 #ifdef __BIG_ENDIAN_BITFIELD 1114 #define PKT_TYPE_MAX (7 << 5) 1115 #else 1116 #define PKT_TYPE_MAX 7 1117 #endif 1118 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1119 1120 /* if you move tc_at_ingress or tstamp_type 1121 * around, you also must adapt these constants. 1122 */ 1123 #ifdef __BIG_ENDIAN_BITFIELD 1124 #define SKB_TSTAMP_TYPE_MASK (3 << 6) 1125 #define SKB_TSTAMP_TYPE_RSHIFT (6) 1126 #define TC_AT_INGRESS_MASK (1 << 5) 1127 #else 1128 #define SKB_TSTAMP_TYPE_MASK (3) 1129 #define TC_AT_INGRESS_MASK (1 << 2) 1130 #endif 1131 #define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset) 1132 1133 #ifdef __KERNEL__ 1134 /* 1135 * Handling routines are only of interest to the kernel 1136 */ 1137 1138 #define SKB_ALLOC_FCLONE 0x01 1139 #define SKB_ALLOC_RX 0x02 1140 #define SKB_ALLOC_NAPI 0x04 1141 1142 /** 1143 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1144 * @skb: buffer 1145 */ 1146 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1147 { 1148 return unlikely(skb->pfmemalloc); 1149 } 1150 1151 /* 1152 * skb might have a dst pointer attached, refcounted or not. 1153 * _skb_refdst low order bit is set if refcount was _not_ taken 1154 */ 1155 #define SKB_DST_NOREF 1UL 1156 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1157 1158 /** 1159 * skb_dst - returns skb dst_entry 1160 * @skb: buffer 1161 * 1162 * Returns: skb dst_entry, regardless of reference taken or not. 1163 */ 1164 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1165 { 1166 /* If refdst was not refcounted, check we still are in a 1167 * rcu_read_lock section 1168 */ 1169 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1170 !rcu_read_lock_held() && 1171 !rcu_read_lock_bh_held()); 1172 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1173 } 1174 1175 static inline void skb_dst_check_unset(struct sk_buff *skb) 1176 { 1177 DEBUG_NET_WARN_ON_ONCE((skb->_skb_refdst & SKB_DST_PTRMASK) && 1178 !(skb->_skb_refdst & SKB_DST_NOREF)); 1179 } 1180 1181 /** 1182 * skb_dstref_steal() - return current dst_entry value and clear it 1183 * @skb: buffer 1184 * 1185 * Resets skb dst_entry without adjusting its reference count. Useful in 1186 * cases where dst_entry needs to be temporarily reset and restored. 1187 * Note that the returned value cannot be used directly because it 1188 * might contain SKB_DST_NOREF bit. 1189 * 1190 * When in doubt, prefer skb_dst_drop() over skb_dstref_steal() to correctly 1191 * handle dst_entry reference counting. 1192 * 1193 * Returns: original skb dst_entry. 1194 */ 1195 static inline unsigned long skb_dstref_steal(struct sk_buff *skb) 1196 { 1197 unsigned long refdst = skb->_skb_refdst; 1198 1199 skb->_skb_refdst = 0; 1200 return refdst; 1201 } 1202 1203 /** 1204 * skb_dstref_restore() - restore skb dst_entry removed via skb_dstref_steal() 1205 * @skb: buffer 1206 * @refdst: dst entry from a call to skb_dstref_steal() 1207 */ 1208 static inline void skb_dstref_restore(struct sk_buff *skb, unsigned long refdst) 1209 { 1210 skb_dst_check_unset(skb); 1211 skb->_skb_refdst = refdst; 1212 } 1213 1214 /** 1215 * skb_dst_set - sets skb dst 1216 * @skb: buffer 1217 * @dst: dst entry 1218 * 1219 * Sets skb dst, assuming a reference was taken on dst and should 1220 * be released by skb_dst_drop() 1221 */ 1222 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1223 { 1224 skb_dst_check_unset(skb); 1225 skb->slow_gro |= !!dst; 1226 skb->_skb_refdst = (unsigned long)dst; 1227 } 1228 1229 /** 1230 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1231 * @skb: buffer 1232 * @dst: dst entry 1233 * 1234 * Sets skb dst, assuming a reference was not taken on dst. 1235 * If dst entry is cached, we do not take reference and dst_release 1236 * will be avoided by refdst_drop. If dst entry is not cached, we take 1237 * reference, so that last dst_release can destroy the dst immediately. 1238 */ 1239 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1240 { 1241 skb_dst_check_unset(skb); 1242 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1243 skb->slow_gro |= !!dst; 1244 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1245 } 1246 1247 /** 1248 * skb_dst_is_noref - Test if skb dst isn't refcounted 1249 * @skb: buffer 1250 */ 1251 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1252 { 1253 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1254 } 1255 1256 /* For mangling skb->pkt_type from user space side from applications 1257 * such as nft, tc, etc, we only allow a conservative subset of 1258 * possible pkt_types to be set. 1259 */ 1260 static inline bool skb_pkt_type_ok(u32 ptype) 1261 { 1262 return ptype <= PACKET_OTHERHOST; 1263 } 1264 1265 /** 1266 * skb_napi_id - Returns the skb's NAPI id 1267 * @skb: buffer 1268 */ 1269 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1270 { 1271 #ifdef CONFIG_NET_RX_BUSY_POLL 1272 return skb->napi_id; 1273 #else 1274 return 0; 1275 #endif 1276 } 1277 1278 static inline bool skb_wifi_acked_valid(const struct sk_buff *skb) 1279 { 1280 #ifdef CONFIG_WIRELESS 1281 return skb->wifi_acked_valid; 1282 #else 1283 return 0; 1284 #endif 1285 } 1286 1287 /** 1288 * skb_unref - decrement the skb's reference count 1289 * @skb: buffer 1290 * 1291 * Returns: true if we can free the skb. 1292 */ 1293 static inline bool skb_unref(struct sk_buff *skb) 1294 { 1295 if (unlikely(!skb)) 1296 return false; 1297 if (!IS_ENABLED(CONFIG_DEBUG_NET) && likely(refcount_read(&skb->users) == 1)) 1298 smp_rmb(); 1299 else if (likely(!refcount_dec_and_test(&skb->users))) 1300 return false; 1301 1302 return true; 1303 } 1304 1305 static inline bool skb_data_unref(const struct sk_buff *skb, 1306 struct skb_shared_info *shinfo) 1307 { 1308 int bias; 1309 1310 if (!skb->cloned) 1311 return true; 1312 1313 bias = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1; 1314 1315 if (atomic_read(&shinfo->dataref) == bias) 1316 smp_rmb(); 1317 else if (atomic_sub_return(bias, &shinfo->dataref)) 1318 return false; 1319 1320 return true; 1321 } 1322 1323 void __fix_address sk_skb_reason_drop(const struct sock *sk, 1324 struct sk_buff *skb, 1325 enum skb_drop_reason reason); 1326 1327 static inline void 1328 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason) 1329 { 1330 sk_skb_reason_drop(NULL, skb, reason); 1331 } 1332 1333 /** 1334 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1335 * @skb: buffer to free 1336 */ 1337 static inline void kfree_skb(struct sk_buff *skb) 1338 { 1339 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1340 } 1341 1342 void skb_release_head_state(struct sk_buff *skb); 1343 void kfree_skb_list_reason(struct sk_buff *segs, 1344 enum skb_drop_reason reason); 1345 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1346 void skb_tx_error(struct sk_buff *skb); 1347 1348 static inline void kfree_skb_list(struct sk_buff *segs) 1349 { 1350 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1351 } 1352 1353 #ifdef CONFIG_TRACEPOINTS 1354 void consume_skb(struct sk_buff *skb); 1355 #else 1356 static inline void consume_skb(struct sk_buff *skb) 1357 { 1358 return kfree_skb(skb); 1359 } 1360 #endif 1361 1362 void __consume_stateless_skb(struct sk_buff *skb); 1363 void __kfree_skb(struct sk_buff *skb); 1364 1365 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1366 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1367 bool *fragstolen, int *delta_truesize); 1368 1369 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1370 int node); 1371 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1372 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1373 struct sk_buff *build_skb_around(struct sk_buff *skb, 1374 void *data, unsigned int frag_size); 1375 void skb_attempt_defer_free(struct sk_buff *skb); 1376 1377 u32 napi_skb_cache_get_bulk(void **skbs, u32 n); 1378 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1379 struct sk_buff *slab_build_skb(void *data); 1380 1381 /** 1382 * alloc_skb - allocate a network buffer 1383 * @size: size to allocate 1384 * @priority: allocation mask 1385 * 1386 * This function is a convenient wrapper around __alloc_skb(). 1387 */ 1388 static inline struct sk_buff *alloc_skb(unsigned int size, 1389 gfp_t priority) 1390 { 1391 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1392 } 1393 1394 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1395 unsigned long data_len, 1396 int max_page_order, 1397 int *errcode, 1398 gfp_t gfp_mask); 1399 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1400 1401 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1402 struct sk_buff_fclones { 1403 struct sk_buff skb1; 1404 1405 struct sk_buff skb2; 1406 1407 refcount_t fclone_ref; 1408 }; 1409 1410 /** 1411 * skb_fclone_busy - check if fclone is busy 1412 * @sk: socket 1413 * @skb: buffer 1414 * 1415 * Returns: true if skb is a fast clone, and its clone is not freed. 1416 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1417 * so we also check that didn't happen. 1418 */ 1419 static inline bool skb_fclone_busy(const struct sock *sk, 1420 const struct sk_buff *skb) 1421 { 1422 const struct sk_buff_fclones *fclones; 1423 1424 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1425 1426 return skb->fclone == SKB_FCLONE_ORIG && 1427 refcount_read(&fclones->fclone_ref) > 1 && 1428 READ_ONCE(fclones->skb2.sk) == sk; 1429 } 1430 1431 /** 1432 * alloc_skb_fclone - allocate a network buffer from fclone cache 1433 * @size: size to allocate 1434 * @priority: allocation mask 1435 * 1436 * This function is a convenient wrapper around __alloc_skb(). 1437 */ 1438 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1439 gfp_t priority) 1440 { 1441 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1442 } 1443 1444 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1445 void skb_headers_offset_update(struct sk_buff *skb, int off); 1446 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1447 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1448 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1449 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1450 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1451 gfp_t gfp_mask, bool fclone); 1452 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1453 gfp_t gfp_mask) 1454 { 1455 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1456 } 1457 1458 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1459 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1460 unsigned int headroom); 1461 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1462 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1463 int newtailroom, gfp_t priority); 1464 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1465 int offset, int len); 1466 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1467 int offset, int len); 1468 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1469 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1470 1471 /** 1472 * skb_pad - zero pad the tail of an skb 1473 * @skb: buffer to pad 1474 * @pad: space to pad 1475 * 1476 * Ensure that a buffer is followed by a padding area that is zero 1477 * filled. Used by network drivers which may DMA or transfer data 1478 * beyond the buffer end onto the wire. 1479 * 1480 * May return error in out of memory cases. The skb is freed on error. 1481 */ 1482 static inline int skb_pad(struct sk_buff *skb, int pad) 1483 { 1484 return __skb_pad(skb, pad, true); 1485 } 1486 #define dev_kfree_skb(a) consume_skb(a) 1487 1488 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1489 int offset, size_t size, size_t max_frags); 1490 1491 struct skb_seq_state { 1492 __u32 lower_offset; 1493 __u32 upper_offset; 1494 __u32 frag_idx; 1495 __u32 stepped_offset; 1496 struct sk_buff *root_skb; 1497 struct sk_buff *cur_skb; 1498 __u8 *frag_data; 1499 __u32 frag_off; 1500 }; 1501 1502 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1503 unsigned int to, struct skb_seq_state *st); 1504 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1505 struct skb_seq_state *st); 1506 void skb_abort_seq_read(struct skb_seq_state *st); 1507 int skb_copy_seq_read(struct skb_seq_state *st, int offset, void *to, int len); 1508 1509 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1510 unsigned int to, struct ts_config *config); 1511 1512 /* 1513 * Packet hash types specify the type of hash in skb_set_hash. 1514 * 1515 * Hash types refer to the protocol layer addresses which are used to 1516 * construct a packet's hash. The hashes are used to differentiate or identify 1517 * flows of the protocol layer for the hash type. Hash types are either 1518 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1519 * 1520 * Properties of hashes: 1521 * 1522 * 1) Two packets in different flows have different hash values 1523 * 2) Two packets in the same flow should have the same hash value 1524 * 1525 * A hash at a higher layer is considered to be more specific. A driver should 1526 * set the most specific hash possible. 1527 * 1528 * A driver cannot indicate a more specific hash than the layer at which a hash 1529 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1530 * 1531 * A driver may indicate a hash level which is less specific than the 1532 * actual layer the hash was computed on. For instance, a hash computed 1533 * at L4 may be considered an L3 hash. This should only be done if the 1534 * driver can't unambiguously determine that the HW computed the hash at 1535 * the higher layer. Note that the "should" in the second property above 1536 * permits this. 1537 */ 1538 enum pkt_hash_types { 1539 PKT_HASH_TYPE_NONE, /* Undefined type */ 1540 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1541 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1542 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1543 }; 1544 1545 static inline void skb_clear_hash(struct sk_buff *skb) 1546 { 1547 skb->hash = 0; 1548 skb->sw_hash = 0; 1549 skb->l4_hash = 0; 1550 } 1551 1552 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1553 { 1554 if (!skb->l4_hash) 1555 skb_clear_hash(skb); 1556 } 1557 1558 static inline void 1559 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1560 { 1561 skb->l4_hash = is_l4; 1562 skb->sw_hash = is_sw; 1563 skb->hash = hash; 1564 } 1565 1566 static inline void 1567 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1568 { 1569 /* Used by drivers to set hash from HW */ 1570 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1571 } 1572 1573 static inline void 1574 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1575 { 1576 __skb_set_hash(skb, hash, true, is_l4); 1577 } 1578 1579 u32 __skb_get_hash_symmetric_net(const struct net *net, const struct sk_buff *skb); 1580 1581 static inline u32 __skb_get_hash_symmetric(const struct sk_buff *skb) 1582 { 1583 return __skb_get_hash_symmetric_net(NULL, skb); 1584 } 1585 1586 void __skb_get_hash_net(const struct net *net, struct sk_buff *skb); 1587 u32 skb_get_poff(const struct sk_buff *skb); 1588 u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1589 const struct flow_keys_basic *keys, int hlen); 1590 __be32 skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1591 const void *data, int hlen_proto); 1592 1593 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1594 const struct flow_dissector_key *key, 1595 unsigned int key_count); 1596 1597 struct bpf_flow_dissector; 1598 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1599 __be16 proto, int nhoff, int hlen, unsigned int flags); 1600 1601 bool __skb_flow_dissect(const struct net *net, 1602 const struct sk_buff *skb, 1603 struct flow_dissector *flow_dissector, 1604 void *target_container, const void *data, 1605 __be16 proto, int nhoff, int hlen, unsigned int flags); 1606 1607 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1608 struct flow_dissector *flow_dissector, 1609 void *target_container, unsigned int flags) 1610 { 1611 return __skb_flow_dissect(NULL, skb, flow_dissector, 1612 target_container, NULL, 0, 0, 0, flags); 1613 } 1614 1615 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1616 struct flow_keys *flow, 1617 unsigned int flags) 1618 { 1619 memset(flow, 0, sizeof(*flow)); 1620 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1621 flow, NULL, 0, 0, 0, flags); 1622 } 1623 1624 static inline bool 1625 skb_flow_dissect_flow_keys_basic(const struct net *net, 1626 const struct sk_buff *skb, 1627 struct flow_keys_basic *flow, 1628 const void *data, __be16 proto, 1629 int nhoff, int hlen, unsigned int flags) 1630 { 1631 memset(flow, 0, sizeof(*flow)); 1632 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1633 data, proto, nhoff, hlen, flags); 1634 } 1635 1636 void skb_flow_dissect_meta(const struct sk_buff *skb, 1637 struct flow_dissector *flow_dissector, 1638 void *target_container); 1639 1640 /* Gets a skb connection tracking info, ctinfo map should be a 1641 * map of mapsize to translate enum ip_conntrack_info states 1642 * to user states. 1643 */ 1644 void 1645 skb_flow_dissect_ct(const struct sk_buff *skb, 1646 struct flow_dissector *flow_dissector, 1647 void *target_container, 1648 u16 *ctinfo_map, size_t mapsize, 1649 bool post_ct, u16 zone); 1650 void 1651 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1652 struct flow_dissector *flow_dissector, 1653 void *target_container); 1654 1655 void skb_flow_dissect_hash(const struct sk_buff *skb, 1656 struct flow_dissector *flow_dissector, 1657 void *target_container); 1658 1659 static inline __u32 skb_get_hash_net(const struct net *net, struct sk_buff *skb) 1660 { 1661 if (!skb->l4_hash && !skb->sw_hash) 1662 __skb_get_hash_net(net, skb); 1663 1664 return skb->hash; 1665 } 1666 1667 static inline __u32 skb_get_hash(struct sk_buff *skb) 1668 { 1669 if (!skb->l4_hash && !skb->sw_hash) 1670 __skb_get_hash_net(NULL, skb); 1671 1672 return skb->hash; 1673 } 1674 1675 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1676 { 1677 if (!skb->l4_hash && !skb->sw_hash) { 1678 struct flow_keys keys; 1679 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1680 1681 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1682 } 1683 1684 return skb->hash; 1685 } 1686 1687 __u32 skb_get_hash_perturb(const struct sk_buff *skb, 1688 const siphash_key_t *perturb); 1689 1690 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1691 { 1692 return skb->hash; 1693 } 1694 1695 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1696 { 1697 to->hash = from->hash; 1698 to->sw_hash = from->sw_hash; 1699 to->l4_hash = from->l4_hash; 1700 }; 1701 1702 static inline int skb_cmp_decrypted(const struct sk_buff *skb1, 1703 const struct sk_buff *skb2) 1704 { 1705 #ifdef CONFIG_SKB_DECRYPTED 1706 return skb2->decrypted - skb1->decrypted; 1707 #else 1708 return 0; 1709 #endif 1710 } 1711 1712 static inline bool skb_is_decrypted(const struct sk_buff *skb) 1713 { 1714 #ifdef CONFIG_SKB_DECRYPTED 1715 return skb->decrypted; 1716 #else 1717 return false; 1718 #endif 1719 } 1720 1721 static inline void skb_copy_decrypted(struct sk_buff *to, 1722 const struct sk_buff *from) 1723 { 1724 #ifdef CONFIG_SKB_DECRYPTED 1725 to->decrypted = from->decrypted; 1726 #endif 1727 } 1728 1729 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1730 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1731 { 1732 return skb->head + skb->end; 1733 } 1734 1735 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1736 { 1737 return skb->end; 1738 } 1739 1740 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1741 { 1742 skb->end = offset; 1743 } 1744 #else 1745 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1746 { 1747 return skb->end; 1748 } 1749 1750 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1751 { 1752 return skb->end - skb->head; 1753 } 1754 1755 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1756 { 1757 skb->end = skb->head + offset; 1758 } 1759 #endif 1760 1761 extern const struct ubuf_info_ops msg_zerocopy_ubuf_ops; 1762 1763 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1764 struct ubuf_info *uarg, bool devmem); 1765 1766 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1767 1768 struct net_devmem_dmabuf_binding; 1769 1770 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, 1771 struct sk_buff *skb, struct iov_iter *from, 1772 size_t length, 1773 struct net_devmem_dmabuf_binding *binding); 1774 1775 int zerocopy_fill_skb_from_iter(struct sk_buff *skb, 1776 struct iov_iter *from, size_t length); 1777 1778 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1779 struct msghdr *msg, int len) 1780 { 1781 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len, 1782 NULL); 1783 } 1784 1785 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1786 struct msghdr *msg, int len, 1787 struct ubuf_info *uarg, 1788 struct net_devmem_dmabuf_binding *binding); 1789 1790 /* Internal */ 1791 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1792 1793 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1794 { 1795 return &skb_shinfo(skb)->hwtstamps; 1796 } 1797 1798 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1799 { 1800 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1801 1802 return is_zcopy ? skb_uarg(skb) : NULL; 1803 } 1804 1805 static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1806 { 1807 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1808 } 1809 1810 static inline bool skb_zcopy_managed(const struct sk_buff *skb) 1811 { 1812 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS; 1813 } 1814 1815 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1816 const struct sk_buff *skb2) 1817 { 1818 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1819 } 1820 1821 static inline void net_zcopy_get(struct ubuf_info *uarg) 1822 { 1823 refcount_inc(&uarg->refcnt); 1824 } 1825 1826 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1827 { 1828 skb_shinfo(skb)->destructor_arg = uarg; 1829 skb_shinfo(skb)->flags |= uarg->flags; 1830 } 1831 1832 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1833 bool *have_ref) 1834 { 1835 if (skb && uarg && !skb_zcopy(skb)) { 1836 if (unlikely(have_ref && *have_ref)) 1837 *have_ref = false; 1838 else 1839 net_zcopy_get(uarg); 1840 skb_zcopy_init(skb, uarg); 1841 } 1842 } 1843 1844 static inline void net_zcopy_put(struct ubuf_info *uarg) 1845 { 1846 if (uarg) 1847 uarg->ops->complete(NULL, uarg, true); 1848 } 1849 1850 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1851 { 1852 if (uarg) { 1853 if (uarg->ops == &msg_zerocopy_ubuf_ops) 1854 msg_zerocopy_put_abort(uarg, have_uref); 1855 else if (have_uref) 1856 net_zcopy_put(uarg); 1857 } 1858 } 1859 1860 /* Release a reference on a zerocopy structure */ 1861 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1862 { 1863 struct ubuf_info *uarg = skb_zcopy(skb); 1864 1865 if (uarg) { 1866 uarg->ops->complete(skb, uarg, zerocopy_success); 1867 1868 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1869 } 1870 } 1871 1872 void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1873 1874 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1875 { 1876 if (unlikely(skb_zcopy_managed(skb))) 1877 __skb_zcopy_downgrade_managed(skb); 1878 } 1879 1880 /* Return true if frags in this skb are readable by the host. */ 1881 static inline bool skb_frags_readable(const struct sk_buff *skb) 1882 { 1883 return !skb->unreadable; 1884 } 1885 1886 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1887 { 1888 skb->next = NULL; 1889 } 1890 1891 static inline void skb_poison_list(struct sk_buff *skb) 1892 { 1893 #ifdef CONFIG_DEBUG_NET 1894 skb->next = SKB_LIST_POISON_NEXT; 1895 #endif 1896 } 1897 1898 /* Iterate through singly-linked GSO fragments of an skb. */ 1899 #define skb_list_walk_safe(first, skb, next_skb) \ 1900 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1901 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1902 1903 static inline void skb_list_del_init(struct sk_buff *skb) 1904 { 1905 __list_del_entry(&skb->list); 1906 skb_mark_not_on_list(skb); 1907 } 1908 1909 /** 1910 * skb_queue_empty - check if a queue is empty 1911 * @list: queue head 1912 * 1913 * Returns true if the queue is empty, false otherwise. 1914 */ 1915 static inline int skb_queue_empty(const struct sk_buff_head *list) 1916 { 1917 return list->next == (const struct sk_buff *) list; 1918 } 1919 1920 /** 1921 * skb_queue_empty_lockless - check if a queue is empty 1922 * @list: queue head 1923 * 1924 * Returns true if the queue is empty, false otherwise. 1925 * This variant can be used in lockless contexts. 1926 */ 1927 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1928 { 1929 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1930 } 1931 1932 1933 /** 1934 * skb_queue_is_last - check if skb is the last entry in the queue 1935 * @list: queue head 1936 * @skb: buffer 1937 * 1938 * Returns true if @skb is the last buffer on the list. 1939 */ 1940 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1941 const struct sk_buff *skb) 1942 { 1943 return skb->next == (const struct sk_buff *) list; 1944 } 1945 1946 /** 1947 * skb_queue_is_first - check if skb is the first entry in the queue 1948 * @list: queue head 1949 * @skb: buffer 1950 * 1951 * Returns true if @skb is the first buffer on the list. 1952 */ 1953 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1954 const struct sk_buff *skb) 1955 { 1956 return skb->prev == (const struct sk_buff *) list; 1957 } 1958 1959 /** 1960 * skb_queue_next - return the next packet in the queue 1961 * @list: queue head 1962 * @skb: current buffer 1963 * 1964 * Return the next packet in @list after @skb. It is only valid to 1965 * call this if skb_queue_is_last() evaluates to false. 1966 */ 1967 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1968 const struct sk_buff *skb) 1969 { 1970 /* This BUG_ON may seem severe, but if we just return then we 1971 * are going to dereference garbage. 1972 */ 1973 BUG_ON(skb_queue_is_last(list, skb)); 1974 return skb->next; 1975 } 1976 1977 /** 1978 * skb_queue_prev - return the prev packet in the queue 1979 * @list: queue head 1980 * @skb: current buffer 1981 * 1982 * Return the prev packet in @list before @skb. It is only valid to 1983 * call this if skb_queue_is_first() evaluates to false. 1984 */ 1985 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1986 const struct sk_buff *skb) 1987 { 1988 /* This BUG_ON may seem severe, but if we just return then we 1989 * are going to dereference garbage. 1990 */ 1991 BUG_ON(skb_queue_is_first(list, skb)); 1992 return skb->prev; 1993 } 1994 1995 /** 1996 * skb_get - reference buffer 1997 * @skb: buffer to reference 1998 * 1999 * Makes another reference to a socket buffer and returns a pointer 2000 * to the buffer. 2001 */ 2002 static inline struct sk_buff *skb_get(struct sk_buff *skb) 2003 { 2004 refcount_inc(&skb->users); 2005 return skb; 2006 } 2007 2008 /* 2009 * If users == 1, we are the only owner and can avoid redundant atomic changes. 2010 */ 2011 2012 /** 2013 * skb_cloned - is the buffer a clone 2014 * @skb: buffer to check 2015 * 2016 * Returns true if the buffer was generated with skb_clone() and is 2017 * one of multiple shared copies of the buffer. Cloned buffers are 2018 * shared data so must not be written to under normal circumstances. 2019 */ 2020 static inline int skb_cloned(const struct sk_buff *skb) 2021 { 2022 return skb->cloned && 2023 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 2024 } 2025 2026 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 2027 { 2028 might_sleep_if(gfpflags_allow_blocking(pri)); 2029 2030 if (skb_cloned(skb)) 2031 return pskb_expand_head(skb, 0, 0, pri); 2032 2033 return 0; 2034 } 2035 2036 /* This variant of skb_unclone() makes sure skb->truesize 2037 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 2038 * 2039 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 2040 * when various debugging features are in place. 2041 */ 2042 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 2043 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 2044 { 2045 might_sleep_if(gfpflags_allow_blocking(pri)); 2046 2047 if (skb_cloned(skb)) 2048 return __skb_unclone_keeptruesize(skb, pri); 2049 return 0; 2050 } 2051 2052 /** 2053 * skb_header_cloned - is the header a clone 2054 * @skb: buffer to check 2055 * 2056 * Returns true if modifying the header part of the buffer requires 2057 * the data to be copied. 2058 */ 2059 static inline int skb_header_cloned(const struct sk_buff *skb) 2060 { 2061 int dataref; 2062 2063 if (!skb->cloned) 2064 return 0; 2065 2066 dataref = atomic_read(&skb_shinfo(skb)->dataref); 2067 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 2068 return dataref != 1; 2069 } 2070 2071 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 2072 { 2073 might_sleep_if(gfpflags_allow_blocking(pri)); 2074 2075 if (skb_header_cloned(skb)) 2076 return pskb_expand_head(skb, 0, 0, pri); 2077 2078 return 0; 2079 } 2080 2081 /** 2082 * __skb_header_release() - allow clones to use the headroom 2083 * @skb: buffer to operate on 2084 * 2085 * See "DOC: dataref and headerless skbs". 2086 */ 2087 static inline void __skb_header_release(struct sk_buff *skb) 2088 { 2089 skb->nohdr = 1; 2090 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 2091 } 2092 2093 2094 /** 2095 * skb_shared - is the buffer shared 2096 * @skb: buffer to check 2097 * 2098 * Returns true if more than one person has a reference to this 2099 * buffer. 2100 */ 2101 static inline int skb_shared(const struct sk_buff *skb) 2102 { 2103 return refcount_read(&skb->users) != 1; 2104 } 2105 2106 /** 2107 * skb_share_check - check if buffer is shared and if so clone it 2108 * @skb: buffer to check 2109 * @pri: priority for memory allocation 2110 * 2111 * If the buffer is shared the buffer is cloned and the old copy 2112 * drops a reference. A new clone with a single reference is returned. 2113 * If the buffer is not shared the original buffer is returned. When 2114 * being called from interrupt status or with spinlocks held pri must 2115 * be GFP_ATOMIC. 2116 * 2117 * NULL is returned on a memory allocation failure. 2118 */ 2119 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 2120 { 2121 might_sleep_if(gfpflags_allow_blocking(pri)); 2122 if (skb_shared(skb)) { 2123 struct sk_buff *nskb = skb_clone(skb, pri); 2124 2125 if (likely(nskb)) 2126 consume_skb(skb); 2127 else 2128 kfree_skb(skb); 2129 skb = nskb; 2130 } 2131 return skb; 2132 } 2133 2134 /* 2135 * Copy cloned buffers into a new sk_buff. We effectively do COW on 2136 * packets to handle cases where we have a local reader and forward 2137 * and a couple of other messy ones. The normal one is tcpdumping 2138 * a packet that's being forwarded. 2139 */ 2140 2141 /** 2142 * skb_unclone_full - make a private copy of a cloned buffer 2143 * @skb: buffer to check 2144 * @pri: priority for memory allocation 2145 * 2146 * If the socket buffer is a clone then this function creates a new 2147 * copy of the data, drops a reference count on the old copy and returns 2148 * the new copy with the reference count at 1. If the buffer is not a clone 2149 * the original buffer is returned. When called with a spinlock held or 2150 * from interrupt state @pri must be %GFP_ATOMIC 2151 * 2152 * Unlike skb_unclone(), the paged data is copied as well. A buffer with 2153 * skb->users > 1 that is not a clone is returned as is, use 2154 * skb_share_check() for that case. 2155 * 2156 * %NULL is returned on a memory allocation failure. 2157 */ 2158 static inline struct sk_buff *skb_unclone_full(struct sk_buff *skb, gfp_t pri) 2159 { 2160 might_sleep_if(gfpflags_allow_blocking(pri)); 2161 if (skb_cloned(skb)) { 2162 struct sk_buff *nskb = skb_copy(skb, pri); 2163 2164 /* Free our shared copy */ 2165 if (likely(nskb)) 2166 consume_skb(skb); 2167 else 2168 kfree_skb(skb); 2169 skb = nskb; 2170 } 2171 return skb; 2172 } 2173 2174 /** 2175 * skb_peek - peek at the head of an &sk_buff_head 2176 * @list_: list to peek at 2177 * 2178 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2179 * be careful with this one. A peek leaves the buffer on the 2180 * list and someone else may run off with it. You must hold 2181 * the appropriate locks or have a private queue to do this. 2182 * 2183 * Returns %NULL for an empty list or a pointer to the head element. 2184 * The reference count is not incremented and the reference is therefore 2185 * volatile. Use with caution. 2186 */ 2187 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2188 { 2189 struct sk_buff *skb = list_->next; 2190 2191 if (skb == (struct sk_buff *)list_) 2192 skb = NULL; 2193 return skb; 2194 } 2195 2196 /** 2197 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2198 * @list_: list to peek at 2199 * 2200 * Like skb_peek(), but the caller knows that the list is not empty. 2201 */ 2202 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2203 { 2204 return list_->next; 2205 } 2206 2207 /** 2208 * skb_peek_next - peek skb following the given one from a queue 2209 * @skb: skb to start from 2210 * @list_: list to peek at 2211 * 2212 * Returns %NULL when the end of the list is met or a pointer to the 2213 * next element. The reference count is not incremented and the 2214 * reference is therefore volatile. Use with caution. 2215 */ 2216 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2217 const struct sk_buff_head *list_) 2218 { 2219 struct sk_buff *next = skb->next; 2220 2221 if (next == (struct sk_buff *)list_) 2222 next = NULL; 2223 return next; 2224 } 2225 2226 /** 2227 * skb_peek_tail - peek at the tail of an &sk_buff_head 2228 * @list_: list to peek at 2229 * 2230 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2231 * be careful with this one. A peek leaves the buffer on the 2232 * list and someone else may run off with it. You must hold 2233 * the appropriate locks or have a private queue to do this. 2234 * 2235 * Returns %NULL for an empty list or a pointer to the tail element. 2236 * The reference count is not incremented and the reference is therefore 2237 * volatile. Use with caution. 2238 */ 2239 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2240 { 2241 struct sk_buff *skb = READ_ONCE(list_->prev); 2242 2243 if (skb == (struct sk_buff *)list_) 2244 skb = NULL; 2245 return skb; 2246 2247 } 2248 2249 /** 2250 * skb_queue_len - get queue length 2251 * @list_: list to measure 2252 * 2253 * Return the length of an &sk_buff queue. 2254 */ 2255 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2256 { 2257 return list_->qlen; 2258 } 2259 2260 /** 2261 * skb_queue_len_lockless - get queue length 2262 * @list_: list to measure 2263 * 2264 * Return the length of an &sk_buff queue. 2265 * This variant can be used in lockless contexts. 2266 */ 2267 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2268 { 2269 return READ_ONCE(list_->qlen); 2270 } 2271 2272 /** 2273 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2274 * @list: queue to initialize 2275 * 2276 * This initializes only the list and queue length aspects of 2277 * an sk_buff_head object. This allows to initialize the list 2278 * aspects of an sk_buff_head without reinitializing things like 2279 * the spinlock. It can also be used for on-stack sk_buff_head 2280 * objects where the spinlock is known to not be used. 2281 */ 2282 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2283 { 2284 list->prev = list->next = (struct sk_buff *)list; 2285 list->qlen = 0; 2286 } 2287 2288 /* 2289 * This function creates a split out lock class for each invocation; 2290 * this is needed for now since a whole lot of users of the skb-queue 2291 * infrastructure in drivers have different locking usage (in hardirq) 2292 * than the networking core (in softirq only). In the long run either the 2293 * network layer or drivers should need annotation to consolidate the 2294 * main types of usage into 3 classes. 2295 */ 2296 static inline void skb_queue_head_init(struct sk_buff_head *list) 2297 { 2298 spin_lock_init(&list->lock); 2299 __skb_queue_head_init(list); 2300 } 2301 2302 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2303 struct lock_class_key *class) 2304 { 2305 skb_queue_head_init(list); 2306 lockdep_set_class(&list->lock, class); 2307 } 2308 2309 /* 2310 * Insert an sk_buff on a list. 2311 * 2312 * The "__skb_xxxx()" functions are the non-atomic ones that 2313 * can only be called with interrupts disabled. 2314 */ 2315 static inline void __skb_insert(struct sk_buff *newsk, 2316 struct sk_buff *prev, struct sk_buff *next, 2317 struct sk_buff_head *list) 2318 { 2319 /* See skb_queue_empty_lockless() and skb_peek_tail() 2320 * for the opposite READ_ONCE() 2321 */ 2322 WRITE_ONCE(newsk->next, next); 2323 WRITE_ONCE(newsk->prev, prev); 2324 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2325 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2326 WRITE_ONCE(list->qlen, list->qlen + 1); 2327 } 2328 2329 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2330 struct sk_buff *prev, 2331 struct sk_buff *next) 2332 { 2333 struct sk_buff *first = list->next; 2334 struct sk_buff *last = list->prev; 2335 2336 WRITE_ONCE(first->prev, prev); 2337 WRITE_ONCE(prev->next, first); 2338 2339 WRITE_ONCE(last->next, next); 2340 WRITE_ONCE(next->prev, last); 2341 } 2342 2343 /** 2344 * skb_queue_splice - join two skb lists, this is designed for stacks 2345 * @list: the new list to add 2346 * @head: the place to add it in the first list 2347 */ 2348 static inline void skb_queue_splice(const struct sk_buff_head *list, 2349 struct sk_buff_head *head) 2350 { 2351 if (!skb_queue_empty(list)) { 2352 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2353 head->qlen += list->qlen; 2354 } 2355 } 2356 2357 /** 2358 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2359 * @list: the new list to add 2360 * @head: the place to add it in the first list 2361 * 2362 * The list at @list is reinitialised 2363 */ 2364 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2365 struct sk_buff_head *head) 2366 { 2367 if (!skb_queue_empty(list)) { 2368 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2369 head->qlen += list->qlen; 2370 __skb_queue_head_init(list); 2371 } 2372 } 2373 2374 /** 2375 * skb_queue_splice_tail - join two skb lists, each list being a queue 2376 * @list: the new list to add 2377 * @head: the place to add it in the first list 2378 */ 2379 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2380 struct sk_buff_head *head) 2381 { 2382 if (!skb_queue_empty(list)) { 2383 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2384 head->qlen += list->qlen; 2385 } 2386 } 2387 2388 /** 2389 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2390 * @list: the new list to add 2391 * @head: the place to add it in the first list 2392 * 2393 * Each of the lists is a queue. 2394 * The list at @list is reinitialised 2395 */ 2396 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2397 struct sk_buff_head *head) 2398 { 2399 if (!skb_queue_empty(list)) { 2400 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2401 head->qlen += list->qlen; 2402 __skb_queue_head_init(list); 2403 } 2404 } 2405 2406 /** 2407 * __skb_queue_after - queue a buffer at the list head 2408 * @list: list to use 2409 * @prev: place after this buffer 2410 * @newsk: buffer to queue 2411 * 2412 * Queue a buffer int the middle of a list. This function takes no locks 2413 * and you must therefore hold required locks before calling it. 2414 * 2415 * A buffer cannot be placed on two lists at the same time. 2416 */ 2417 static inline void __skb_queue_after(struct sk_buff_head *list, 2418 struct sk_buff *prev, 2419 struct sk_buff *newsk) 2420 { 2421 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2422 } 2423 2424 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2425 struct sk_buff_head *list); 2426 2427 static inline void __skb_queue_before(struct sk_buff_head *list, 2428 struct sk_buff *next, 2429 struct sk_buff *newsk) 2430 { 2431 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2432 } 2433 2434 /** 2435 * __skb_queue_head - queue a buffer at the list head 2436 * @list: list to use 2437 * @newsk: buffer to queue 2438 * 2439 * Queue a buffer at the start of a list. This function takes no locks 2440 * and you must therefore hold required locks before calling it. 2441 * 2442 * A buffer cannot be placed on two lists at the same time. 2443 */ 2444 static inline void __skb_queue_head(struct sk_buff_head *list, 2445 struct sk_buff *newsk) 2446 { 2447 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2448 } 2449 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2450 2451 /** 2452 * __skb_queue_tail - queue a buffer at the list tail 2453 * @list: list to use 2454 * @newsk: buffer to queue 2455 * 2456 * Queue a buffer at the end of a list. This function takes no locks 2457 * and you must therefore hold required locks before calling it. 2458 * 2459 * A buffer cannot be placed on two lists at the same time. 2460 */ 2461 static inline void __skb_queue_tail(struct sk_buff_head *list, 2462 struct sk_buff *newsk) 2463 { 2464 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2465 } 2466 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2467 2468 /* 2469 * remove sk_buff from list. _Must_ be called atomically, and with 2470 * the list known.. 2471 */ 2472 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2473 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2474 { 2475 struct sk_buff *next, *prev; 2476 2477 WRITE_ONCE(list->qlen, list->qlen - 1); 2478 next = skb->next; 2479 prev = skb->prev; 2480 skb->next = skb->prev = NULL; 2481 WRITE_ONCE(next->prev, prev); 2482 WRITE_ONCE(prev->next, next); 2483 } 2484 2485 /** 2486 * __skb_dequeue - remove from the head of the queue 2487 * @list: list to dequeue from 2488 * 2489 * Remove the head of the list. This function does not take any locks 2490 * so must be used with appropriate locks held only. The head item is 2491 * returned or %NULL if the list is empty. 2492 */ 2493 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2494 { 2495 struct sk_buff *skb = skb_peek(list); 2496 if (skb) 2497 __skb_unlink(skb, list); 2498 return skb; 2499 } 2500 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2501 2502 /** 2503 * __skb_dequeue_tail - remove from the tail of the queue 2504 * @list: list to dequeue from 2505 * 2506 * Remove the tail of the list. This function does not take any locks 2507 * so must be used with appropriate locks held only. The tail item is 2508 * returned or %NULL if the list is empty. 2509 */ 2510 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2511 { 2512 struct sk_buff *skb = skb_peek_tail(list); 2513 if (skb) 2514 __skb_unlink(skb, list); 2515 return skb; 2516 } 2517 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2518 2519 2520 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2521 { 2522 return skb->data_len; 2523 } 2524 2525 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2526 { 2527 return skb->len - skb->data_len; 2528 } 2529 2530 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2531 { 2532 unsigned int i, len = 0; 2533 2534 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2535 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2536 return len; 2537 } 2538 2539 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2540 { 2541 return skb_headlen(skb) + __skb_pagelen(skb); 2542 } 2543 2544 static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag, 2545 netmem_ref netmem, int off, 2546 int size) 2547 { 2548 frag->netmem = netmem; 2549 frag->offset = off; 2550 skb_frag_size_set(frag, size); 2551 } 2552 2553 static inline void skb_frag_fill_page_desc(skb_frag_t *frag, 2554 struct page *page, 2555 int off, int size) 2556 { 2557 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size); 2558 } 2559 2560 static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo, 2561 int i, netmem_ref netmem, 2562 int off, int size) 2563 { 2564 skb_frag_t *frag = &shinfo->frags[i]; 2565 2566 skb_frag_fill_netmem_desc(frag, netmem, off, size); 2567 } 2568 2569 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2570 int i, struct page *page, 2571 int off, int size) 2572 { 2573 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off, 2574 size); 2575 } 2576 2577 /** 2578 * skb_len_add - adds a number to len fields of skb 2579 * @skb: buffer to add len to 2580 * @delta: number of bytes to add 2581 */ 2582 static inline void skb_len_add(struct sk_buff *skb, int delta) 2583 { 2584 skb->len += delta; 2585 skb->data_len += delta; 2586 skb->truesize += delta; 2587 } 2588 2589 /** 2590 * __skb_fill_netmem_desc - initialise a fragment in an skb 2591 * @skb: buffer containing fragment to be initialised 2592 * @i: fragment index to initialise 2593 * @netmem: the netmem to use for this fragment 2594 * @off: the offset to the data with @page 2595 * @size: the length of the data 2596 * 2597 * Initialises the @i'th fragment of @skb to point to &size bytes at 2598 * offset @off within @page. 2599 * 2600 * Does not take any additional reference on the fragment. 2601 */ 2602 static __always_inline void 2603 __skb_fill_netmem_desc(struct sk_buff *skb, int i, netmem_ref netmem, 2604 int off, int size) 2605 { 2606 struct page *page; 2607 2608 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size); 2609 2610 if (netmem_is_net_iov(netmem)) { 2611 skb->unreadable = true; 2612 return; 2613 } 2614 2615 page = netmem_to_page(netmem); 2616 2617 /* Propagate page pfmemalloc to the skb if we can. The problem is 2618 * that not all callers have unique ownership of the page but rely 2619 * on page_is_pfmemalloc doing the right thing(tm). 2620 */ 2621 page = compound_head(page); 2622 if (page_is_pfmemalloc(page)) 2623 skb->pfmemalloc = true; 2624 } 2625 2626 static __always_inline void 2627 __skb_fill_page_desc(struct sk_buff *skb, int i, struct page *page, 2628 int off, int size) 2629 { 2630 __skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2631 } 2632 2633 static __always_inline void 2634 skb_fill_netmem_desc(struct sk_buff *skb, int i, netmem_ref netmem, 2635 int off, int size) 2636 { 2637 __skb_fill_netmem_desc(skb, i, netmem, off, size); 2638 skb_shinfo(skb)->nr_frags = i + 1; 2639 } 2640 2641 /** 2642 * skb_fill_page_desc - initialise a paged fragment in an skb 2643 * @skb: buffer containing fragment to be initialised 2644 * @i: paged fragment index to initialise 2645 * @page: the page to use for this fragment 2646 * @off: the offset to the data with @page 2647 * @size: the length of the data 2648 * 2649 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2650 * @skb to point to @size bytes at offset @off within @page. In 2651 * addition updates @skb such that @i is the last fragment. 2652 * 2653 * Does not take any additional reference on the fragment. 2654 */ 2655 static __always_inline void 2656 skb_fill_page_desc(struct sk_buff *skb, int i, struct page *page, 2657 int off, int size) 2658 { 2659 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2660 } 2661 2662 /** 2663 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2664 * @skb: buffer containing fragment to be initialised 2665 * @i: paged fragment index to initialise 2666 * @page: the page to use for this fragment 2667 * @off: the offset to the data with @page 2668 * @size: the length of the data 2669 * 2670 * Variant of skb_fill_page_desc() which does not deal with 2671 * pfmemalloc, if page is not owned by us. 2672 */ 2673 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2674 struct page *page, int off, 2675 int size) 2676 { 2677 struct skb_shared_info *shinfo = skb_shinfo(skb); 2678 2679 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2680 shinfo->nr_frags = i + 1; 2681 } 2682 2683 static inline void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, 2684 netmem_ref netmem, int off, 2685 int size, unsigned int truesize) 2686 { 2687 DEBUG_NET_WARN_ON_ONCE(size > truesize); 2688 2689 skb_fill_netmem_desc(skb, i, netmem, off, size); 2690 skb->len += size; 2691 skb->data_len += size; 2692 skb->truesize += truesize; 2693 } 2694 2695 static inline void skb_add_rx_frag(struct sk_buff *skb, int i, 2696 struct page *page, int off, int size, 2697 unsigned int truesize) 2698 { 2699 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size, 2700 truesize); 2701 } 2702 2703 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2704 unsigned int truesize); 2705 2706 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2707 2708 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2709 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2710 { 2711 return skb->head + skb->tail; 2712 } 2713 2714 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2715 { 2716 skb->tail = skb->data - skb->head; 2717 } 2718 2719 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2720 { 2721 skb_reset_tail_pointer(skb); 2722 skb->tail += offset; 2723 } 2724 2725 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2726 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2727 { 2728 return skb->tail; 2729 } 2730 2731 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2732 { 2733 skb->tail = skb->data; 2734 } 2735 2736 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2737 { 2738 skb->tail = skb->data + offset; 2739 } 2740 2741 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2742 2743 static inline void skb_assert_len(struct sk_buff *skb) 2744 { 2745 #ifdef CONFIG_DEBUG_NET 2746 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2747 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2748 #endif /* CONFIG_DEBUG_NET */ 2749 } 2750 2751 #if defined(CONFIG_FAIL_SKB_REALLOC) 2752 void skb_might_realloc(struct sk_buff *skb); 2753 #else 2754 static inline void skb_might_realloc(struct sk_buff *skb) {} 2755 #endif 2756 2757 /* 2758 * Add data to an sk_buff 2759 */ 2760 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2761 void *skb_put(struct sk_buff *skb, unsigned int len); 2762 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2763 { 2764 void *tmp = skb_tail_pointer(skb); 2765 SKB_LINEAR_ASSERT(skb); 2766 skb->tail += len; 2767 skb->len += len; 2768 return tmp; 2769 } 2770 2771 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2772 { 2773 void *tmp = __skb_put(skb, len); 2774 2775 memset(tmp, 0, len); 2776 return tmp; 2777 } 2778 2779 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2780 unsigned int len) 2781 { 2782 void *tmp = __skb_put(skb, len); 2783 2784 memcpy(tmp, data, len); 2785 return tmp; 2786 } 2787 2788 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2789 { 2790 *(u8 *)__skb_put(skb, 1) = val; 2791 } 2792 2793 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2794 { 2795 void *tmp = skb_put(skb, len); 2796 2797 memset(tmp, 0, len); 2798 2799 return tmp; 2800 } 2801 2802 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2803 unsigned int len) 2804 { 2805 void *tmp = skb_put(skb, len); 2806 2807 memcpy(tmp, data, len); 2808 2809 return tmp; 2810 } 2811 2812 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2813 { 2814 *(u8 *)skb_put(skb, 1) = val; 2815 } 2816 2817 void *skb_push(struct sk_buff *skb, unsigned int len); 2818 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2819 { 2820 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2821 2822 skb->data -= len; 2823 DEBUG_NET_WARN_ON_ONCE(skb->data < skb->head); 2824 skb->len += len; 2825 return skb->data; 2826 } 2827 2828 void *skb_pull(struct sk_buff *skb, unsigned int len); 2829 static __always_inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2830 { 2831 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2832 2833 skb->len -= len; 2834 if (unlikely(skb->len < skb->data_len)) { 2835 #if defined(CONFIG_DEBUG_NET) 2836 skb->len += len; 2837 pr_err("__skb_pull(len=%u)\n", len); 2838 skb_dump(KERN_ERR, skb, false); 2839 #endif 2840 BUG(); 2841 } 2842 return skb->data += len; 2843 } 2844 2845 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2846 { 2847 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2848 } 2849 2850 void *skb_pull_data(struct sk_buff *skb, size_t len); 2851 2852 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2853 2854 static __always_inline enum skb_drop_reason 2855 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len) 2856 { 2857 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2858 skb_might_realloc(skb); 2859 2860 if (likely(len <= skb_headlen(skb))) 2861 return SKB_NOT_DROPPED_YET; 2862 2863 if (unlikely(len > skb->len)) 2864 return SKB_DROP_REASON_PKT_TOO_SMALL; 2865 2866 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb)))) 2867 return SKB_DROP_REASON_NOMEM; 2868 2869 return SKB_NOT_DROPPED_YET; 2870 } 2871 2872 static __always_inline bool 2873 pskb_may_pull(struct sk_buff *skb, unsigned int len) 2874 { 2875 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 2876 } 2877 2878 static __always_inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2879 { 2880 if (!pskb_may_pull(skb, len)) 2881 return NULL; 2882 2883 skb->len -= len; 2884 return skb->data += len; 2885 } 2886 2887 void skb_condense(struct sk_buff *skb); 2888 2889 /** 2890 * skb_headroom - bytes at buffer head 2891 * @skb: buffer to check 2892 * 2893 * Return the number of bytes of free space at the head of an &sk_buff. 2894 */ 2895 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2896 { 2897 return skb->data - skb->head; 2898 } 2899 2900 /** 2901 * skb_tailroom - bytes at buffer end 2902 * @skb: buffer to check 2903 * 2904 * Return the number of bytes of free space at the tail of an sk_buff 2905 */ 2906 static inline int skb_tailroom(const struct sk_buff *skb) 2907 { 2908 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2909 } 2910 2911 /** 2912 * skb_availroom - bytes at buffer end 2913 * @skb: buffer to check 2914 * 2915 * Return the number of bytes of free space at the tail of an sk_buff 2916 * allocated by sk_stream_alloc() 2917 */ 2918 static inline int skb_availroom(const struct sk_buff *skb) 2919 { 2920 if (skb_is_nonlinear(skb)) 2921 return 0; 2922 2923 return skb->end - skb->tail - skb->reserved_tailroom; 2924 } 2925 2926 /** 2927 * skb_reserve - adjust headroom 2928 * @skb: buffer to alter 2929 * @len: bytes to move 2930 * 2931 * Increase the headroom of an empty &sk_buff by reducing the tail 2932 * room. This is only allowed for an empty buffer. 2933 */ 2934 static inline void skb_reserve(struct sk_buff *skb, int len) 2935 { 2936 skb->data += len; 2937 skb->tail += len; 2938 } 2939 2940 /** 2941 * skb_tailroom_reserve - adjust reserved_tailroom 2942 * @skb: buffer to alter 2943 * @mtu: maximum amount of headlen permitted 2944 * @needed_tailroom: minimum amount of reserved_tailroom 2945 * 2946 * Set reserved_tailroom so that headlen can be as large as possible but 2947 * not larger than mtu and tailroom cannot be smaller than 2948 * needed_tailroom. 2949 * The required headroom should already have been reserved before using 2950 * this function. 2951 */ 2952 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2953 unsigned int needed_tailroom) 2954 { 2955 SKB_LINEAR_ASSERT(skb); 2956 if (mtu < skb_tailroom(skb) - needed_tailroom) 2957 /* use at most mtu */ 2958 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2959 else 2960 /* use up to all available space */ 2961 skb->reserved_tailroom = needed_tailroom; 2962 } 2963 2964 #define ENCAP_TYPE_ETHER 0 2965 #define ENCAP_TYPE_IPPROTO 1 2966 2967 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2968 __be16 protocol) 2969 { 2970 skb->inner_protocol = protocol; 2971 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2972 } 2973 2974 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2975 __u8 ipproto) 2976 { 2977 skb->inner_ipproto = ipproto; 2978 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2979 } 2980 2981 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2982 { 2983 skb->inner_mac_header = skb->mac_header; 2984 skb->inner_network_header = skb->network_header; 2985 skb->inner_transport_header = skb->transport_header; 2986 } 2987 2988 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2989 { 2990 return skb->mac_header != (typeof(skb->mac_header))~0U; 2991 } 2992 2993 static inline void skb_reset_mac_len(struct sk_buff *skb) 2994 { 2995 if (!skb_mac_header_was_set(skb)) { 2996 DEBUG_NET_WARN_ON_ONCE(1); 2997 skb->mac_len = 0; 2998 } else { 2999 skb->mac_len = skb->network_header - skb->mac_header; 3000 } 3001 } 3002 3003 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 3004 *skb) 3005 { 3006 return skb->head + skb->inner_transport_header; 3007 } 3008 3009 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 3010 { 3011 return skb_inner_transport_header(skb) - skb->data; 3012 } 3013 3014 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 3015 { 3016 long offset = skb->data - skb->head; 3017 3018 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_transport_header))offset); 3019 skb->inner_transport_header = offset; 3020 } 3021 3022 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 3023 const int offset) 3024 { 3025 skb_reset_inner_transport_header(skb); 3026 skb->inner_transport_header += offset; 3027 } 3028 3029 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 3030 { 3031 return skb->head + skb->inner_network_header; 3032 } 3033 3034 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 3035 { 3036 long offset = skb->data - skb->head; 3037 3038 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_network_header))offset); 3039 skb->inner_network_header = offset; 3040 } 3041 3042 static inline void skb_set_inner_network_header(struct sk_buff *skb, 3043 const int offset) 3044 { 3045 skb_reset_inner_network_header(skb); 3046 skb->inner_network_header += offset; 3047 } 3048 3049 static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb) 3050 { 3051 return skb->inner_network_header > 0; 3052 } 3053 3054 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 3055 { 3056 return skb->head + skb->inner_mac_header; 3057 } 3058 3059 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 3060 { 3061 long offset = skb->data - skb->head; 3062 3063 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_mac_header))offset); 3064 skb->inner_mac_header = offset; 3065 } 3066 3067 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 3068 const int offset) 3069 { 3070 skb_reset_inner_mac_header(skb); 3071 skb->inner_mac_header += offset; 3072 } 3073 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 3074 { 3075 return skb->transport_header != (typeof(skb->transport_header))~0U; 3076 } 3077 3078 static inline void skb_unset_transport_header(struct sk_buff *skb) 3079 { 3080 skb->transport_header = (typeof(skb->transport_header))~0U; 3081 } 3082 3083 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 3084 { 3085 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3086 return skb->head + skb->transport_header; 3087 } 3088 3089 static inline void skb_reset_transport_header(struct sk_buff *skb) 3090 { 3091 long offset = skb->data - skb->head; 3092 3093 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->transport_header))offset); 3094 skb->transport_header = offset; 3095 } 3096 3097 /** 3098 * skb_reset_transport_header_careful - conditionally reset transport header 3099 * @skb: buffer to alter 3100 * 3101 * Hardened version of skb_reset_transport_header(). 3102 * 3103 * Returns: true if the operation was a success. 3104 */ 3105 static inline bool __must_check 3106 skb_reset_transport_header_careful(struct sk_buff *skb) 3107 { 3108 long offset = skb->data - skb->head; 3109 3110 if (unlikely(offset != (typeof(skb->transport_header))offset)) 3111 return false; 3112 3113 if (unlikely(offset == (typeof(skb->transport_header))~0U)) 3114 return false; 3115 3116 skb->transport_header = offset; 3117 return true; 3118 } 3119 3120 static inline void skb_set_transport_header(struct sk_buff *skb, 3121 const int offset) 3122 { 3123 skb_reset_transport_header(skb); 3124 skb->transport_header += offset; 3125 } 3126 3127 /** 3128 * skb_set_transport_header_careful - conditionally set transport header 3129 * @skb: buffer to alter 3130 * @offset: offset to add to skb->data 3131 * 3132 * Hardened version of skb_set_transport_header(). 3133 * 3134 * Returns: true if the operation was a success. 3135 */ 3136 static inline bool __must_check 3137 skb_set_transport_header_careful(struct sk_buff *skb, const int offset) 3138 { 3139 long thoff = skb->data - skb->head + offset; 3140 3141 if (unlikely(thoff != (typeof(skb->transport_header))thoff)) 3142 return false; 3143 3144 if (unlikely(thoff == (typeof(skb->transport_header))~0U)) 3145 return false; 3146 3147 skb->transport_header = thoff; 3148 return true; 3149 } 3150 3151 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 3152 { 3153 return skb->head + skb->network_header; 3154 } 3155 3156 static inline void skb_reset_network_header(struct sk_buff *skb) 3157 { 3158 long offset = skb->data - skb->head; 3159 3160 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->network_header))offset); 3161 skb->network_header = offset; 3162 } 3163 3164 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 3165 { 3166 skb_reset_network_header(skb); 3167 skb->network_header += offset; 3168 } 3169 3170 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 3171 { 3172 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3173 return skb->head + skb->mac_header; 3174 } 3175 3176 static inline int skb_mac_offset(const struct sk_buff *skb) 3177 { 3178 return skb_mac_header(skb) - skb->data; 3179 } 3180 3181 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 3182 { 3183 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 3184 return skb->network_header - skb->mac_header; 3185 } 3186 3187 static inline void skb_unset_mac_header(struct sk_buff *skb) 3188 { 3189 skb->mac_header = (typeof(skb->mac_header))~0U; 3190 } 3191 3192 static inline void skb_reset_mac_header(struct sk_buff *skb) 3193 { 3194 long offset = skb->data - skb->head; 3195 3196 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->mac_header))offset); 3197 skb->mac_header = offset; 3198 } 3199 3200 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 3201 { 3202 skb_reset_mac_header(skb); 3203 skb->mac_header += offset; 3204 } 3205 3206 static inline void skb_pop_mac_header(struct sk_buff *skb) 3207 { 3208 skb->mac_header = skb->network_header; 3209 } 3210 3211 static inline void skb_probe_transport_header(struct sk_buff *skb) 3212 { 3213 struct flow_keys_basic keys; 3214 3215 if (skb_transport_header_was_set(skb)) 3216 return; 3217 3218 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 3219 NULL, 0, 0, 0, 0)) 3220 skb_set_transport_header(skb, keys.control.thoff); 3221 } 3222 3223 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 3224 { 3225 if (skb_mac_header_was_set(skb)) { 3226 const unsigned char *old_mac = skb_mac_header(skb); 3227 3228 skb_set_mac_header(skb, -skb->mac_len); 3229 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 3230 } 3231 } 3232 3233 /* Move the full mac header up to current network_header. 3234 * Leaves skb->data pointing at offset skb->mac_len into the mac_header. 3235 * Must be provided the complete mac header length. 3236 */ 3237 static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len) 3238 { 3239 if (skb_mac_header_was_set(skb)) { 3240 const unsigned char *old_mac = skb_mac_header(skb); 3241 3242 skb_set_mac_header(skb, -full_mac_len); 3243 memmove(skb_mac_header(skb), old_mac, full_mac_len); 3244 __skb_push(skb, full_mac_len - skb->mac_len); 3245 } 3246 } 3247 3248 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 3249 { 3250 return skb->csum_start - skb_headroom(skb); 3251 } 3252 3253 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 3254 { 3255 return skb->head + skb->csum_start; 3256 } 3257 3258 static inline int skb_transport_offset(const struct sk_buff *skb) 3259 { 3260 return skb_transport_header(skb) - skb->data; 3261 } 3262 3263 static inline u32 skb_network_header_len(const struct sk_buff *skb) 3264 { 3265 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3266 return skb->transport_header - skb->network_header; 3267 } 3268 3269 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 3270 { 3271 return skb->inner_transport_header - skb->inner_network_header; 3272 } 3273 3274 static inline int skb_network_offset(const struct sk_buff *skb) 3275 { 3276 return skb_network_header(skb) - skb->data; 3277 } 3278 3279 static inline int skb_inner_network_offset(const struct sk_buff *skb) 3280 { 3281 return skb_inner_network_header(skb) - skb->data; 3282 } 3283 3284 static inline enum skb_drop_reason 3285 pskb_network_may_pull_reason(struct sk_buff *skb, unsigned int len) 3286 { 3287 return pskb_may_pull_reason(skb, skb_network_offset(skb) + len); 3288 } 3289 3290 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 3291 { 3292 return pskb_network_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 3293 } 3294 3295 /* 3296 * CPUs often take a performance hit when accessing unaligned memory 3297 * locations. The actual performance hit varies, it can be small if the 3298 * hardware handles it or large if we have to take an exception and fix it 3299 * in software. 3300 * 3301 * Since an ethernet header is 14 bytes network drivers often end up with 3302 * the IP header at an unaligned offset. The IP header can be aligned by 3303 * shifting the start of the packet by 2 bytes. Drivers should do this 3304 * with: 3305 * 3306 * skb_reserve(skb, NET_IP_ALIGN); 3307 * 3308 * The downside to this alignment of the IP header is that the DMA is now 3309 * unaligned. On some architectures the cost of an unaligned DMA is high 3310 * and this cost outweighs the gains made by aligning the IP header. 3311 * 3312 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 3313 * to be overridden. 3314 */ 3315 #ifndef NET_IP_ALIGN 3316 #define NET_IP_ALIGN 2 3317 #endif 3318 3319 /* 3320 * The networking layer reserves some headroom in skb data (via 3321 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 3322 * the header has to grow. In the default case, if the header has to grow 3323 * 32 bytes or less we avoid the reallocation. 3324 * 3325 * Unfortunately this headroom changes the DMA alignment of the resulting 3326 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 3327 * on some architectures. An architecture can override this value, 3328 * perhaps setting it to a cacheline in size (since that will maintain 3329 * cacheline alignment of the DMA). It must be a power of 2. 3330 * 3331 * Various parts of the networking layer expect at least 32 bytes of 3332 * headroom, you should not reduce this. 3333 * 3334 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 3335 * to reduce average number of cache lines per packet. 3336 * get_rps_cpu() for example only access one 64 bytes aligned block : 3337 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 3338 */ 3339 #ifndef NET_SKB_PAD 3340 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 3341 #endif 3342 3343 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 3344 3345 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3346 { 3347 if (WARN_ON(skb_is_nonlinear(skb))) 3348 return; 3349 skb->len = len; 3350 skb_set_tail_pointer(skb, len); 3351 } 3352 3353 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3354 { 3355 __skb_set_length(skb, len); 3356 } 3357 3358 void skb_trim(struct sk_buff *skb, unsigned int len); 3359 3360 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3361 { 3362 if (skb->data_len) 3363 return ___pskb_trim(skb, len); 3364 __skb_trim(skb, len); 3365 return 0; 3366 } 3367 3368 static __always_inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3369 { 3370 skb_might_realloc(skb); 3371 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3372 } 3373 3374 /** 3375 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3376 * @skb: buffer to alter 3377 * @len: new length 3378 * 3379 * This is identical to pskb_trim except that the caller knows that 3380 * the skb is not cloned so we should never get an error due to out- 3381 * of-memory. 3382 */ 3383 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3384 { 3385 int err = pskb_trim(skb, len); 3386 BUG_ON(err); 3387 } 3388 3389 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3390 { 3391 unsigned int diff = len - skb->len; 3392 3393 if (skb_tailroom(skb) < diff) { 3394 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3395 GFP_ATOMIC); 3396 if (ret) 3397 return ret; 3398 } 3399 __skb_set_length(skb, len); 3400 return 0; 3401 } 3402 3403 /** 3404 * skb_orphan - orphan a buffer 3405 * @skb: buffer to orphan 3406 * 3407 * If a buffer currently has an owner then we call the owner's 3408 * destructor function and make the @skb unowned. The buffer continues 3409 * to exist but is no longer charged to its former owner. 3410 */ 3411 static __always_inline void skb_orphan(struct sk_buff *skb) 3412 { 3413 if (skb->destructor) { 3414 skb->destructor(skb); 3415 skb->destructor = NULL; 3416 skb->sk = NULL; 3417 } else { 3418 BUG_ON(skb->sk); 3419 } 3420 } 3421 3422 /** 3423 * skb_orphan_frags - orphan the frags contained in a buffer 3424 * @skb: buffer to orphan frags from 3425 * @gfp_mask: allocation mask for replacement pages 3426 * 3427 * For each frag in the SKB which needs a destructor (i.e. has an 3428 * owner) create a copy of that frag and release the original 3429 * page by calling the destructor. 3430 */ 3431 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3432 { 3433 if (likely(!skb_zcopy(skb))) 3434 return 0; 3435 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3436 return 0; 3437 return skb_copy_ubufs(skb, gfp_mask); 3438 } 3439 3440 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3441 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3442 { 3443 if (likely(!skb_zcopy(skb))) 3444 return 0; 3445 return skb_copy_ubufs(skb, gfp_mask); 3446 } 3447 3448 /** 3449 * __skb_queue_purge_reason - empty a list 3450 * @list: list to empty 3451 * @reason: drop reason 3452 * 3453 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3454 * the list and one reference dropped. This function does not take the 3455 * list lock and the caller must hold the relevant locks to use it. 3456 */ 3457 static inline void __skb_queue_purge_reason(struct sk_buff_head *list, 3458 enum skb_drop_reason reason) 3459 { 3460 struct sk_buff *skb; 3461 3462 while ((skb = __skb_dequeue(list)) != NULL) 3463 kfree_skb_reason(skb, reason); 3464 } 3465 3466 static inline void __skb_queue_purge(struct sk_buff_head *list) 3467 { 3468 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3469 } 3470 3471 void skb_queue_purge_reason(struct sk_buff_head *list, 3472 enum skb_drop_reason reason); 3473 3474 static inline void skb_queue_purge(struct sk_buff_head *list) 3475 { 3476 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3477 } 3478 3479 unsigned int skb_rbtree_purge(struct rb_root *root); 3480 void skb_errqueue_purge(struct sk_buff_head *list); 3481 3482 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3483 3484 /** 3485 * netdev_alloc_frag - allocate a page fragment 3486 * @fragsz: fragment size 3487 * 3488 * Allocates a frag from a page for receive buffer. 3489 * Uses GFP_ATOMIC allocations. 3490 */ 3491 static inline void *netdev_alloc_frag(unsigned int fragsz) 3492 { 3493 return __netdev_alloc_frag_align(fragsz, ~0u); 3494 } 3495 3496 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3497 unsigned int align) 3498 { 3499 WARN_ON_ONCE(!is_power_of_2(align)); 3500 return __netdev_alloc_frag_align(fragsz, -align); 3501 } 3502 3503 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3504 gfp_t gfp_mask); 3505 3506 /** 3507 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3508 * @dev: network device to receive on 3509 * @length: length to allocate 3510 * 3511 * Allocate a new &sk_buff and assign it a usage count of one. The 3512 * buffer has unspecified headroom built in. Users should allocate 3513 * the headroom they think they need without accounting for the 3514 * built in space. The built in space is used for optimisations. 3515 * 3516 * %NULL is returned if there is no free memory. Although this function 3517 * allocates memory it can be called from an interrupt. 3518 */ 3519 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3520 unsigned int length) 3521 { 3522 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3523 } 3524 3525 /* legacy helper around __netdev_alloc_skb() */ 3526 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3527 gfp_t gfp_mask) 3528 { 3529 return __netdev_alloc_skb(NULL, length, gfp_mask); 3530 } 3531 3532 /* legacy helper around netdev_alloc_skb() */ 3533 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3534 { 3535 return netdev_alloc_skb(NULL, length); 3536 } 3537 3538 3539 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3540 unsigned int length, gfp_t gfp) 3541 { 3542 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3543 3544 if (NET_IP_ALIGN && skb) 3545 skb_reserve(skb, NET_IP_ALIGN); 3546 return skb; 3547 } 3548 3549 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3550 unsigned int length) 3551 { 3552 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3553 } 3554 3555 static inline void skb_free_frag(void *addr) 3556 { 3557 page_frag_free(addr); 3558 } 3559 3560 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3561 3562 static inline void *napi_alloc_frag(unsigned int fragsz) 3563 { 3564 return __napi_alloc_frag_align(fragsz, ~0u); 3565 } 3566 3567 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3568 unsigned int align) 3569 { 3570 WARN_ON_ONCE(!is_power_of_2(align)); 3571 return __napi_alloc_frag_align(fragsz, -align); 3572 } 3573 3574 struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int length); 3575 void napi_consume_skb(struct sk_buff *skb, int budget); 3576 3577 void napi_skb_free_stolen_head(struct sk_buff *skb); 3578 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason); 3579 3580 /** 3581 * __dev_alloc_pages - allocate page for network Rx 3582 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3583 * @order: size of the allocation 3584 * 3585 * Allocate a new page. 3586 * 3587 * %NULL is returned if there is no free memory. 3588 */ 3589 static inline struct page *__dev_alloc_pages_noprof(gfp_t gfp_mask, 3590 unsigned int order) 3591 { 3592 /* This piece of code contains several assumptions. 3593 * 1. This is for device Rx, therefore a cold page is preferred. 3594 * 2. The expectation is the user wants a compound page. 3595 * 3. If requesting a order 0 page it will not be compound 3596 * due to the check to see if order has a value in prep_new_page 3597 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3598 * code in alloc_flags_slowpath() that should be enforcing this. 3599 */ 3600 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3601 3602 return alloc_pages_node_noprof(NUMA_NO_NODE, gfp_mask, order); 3603 } 3604 #define __dev_alloc_pages(...) alloc_hooks(__dev_alloc_pages_noprof(__VA_ARGS__)) 3605 3606 /* 3607 * This specialized allocator has to be a macro for its allocations to be 3608 * accounted separately (to have a separate alloc_tag). 3609 */ 3610 #define dev_alloc_pages(_order) __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, _order) 3611 3612 /** 3613 * __dev_alloc_page - allocate a page for network Rx 3614 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3615 * 3616 * Allocate a new page. 3617 * 3618 * %NULL is returned if there is no free memory. 3619 */ 3620 static inline struct page *__dev_alloc_page_noprof(gfp_t gfp_mask) 3621 { 3622 return __dev_alloc_pages_noprof(gfp_mask, 0); 3623 } 3624 #define __dev_alloc_page(...) alloc_hooks(__dev_alloc_page_noprof(__VA_ARGS__)) 3625 3626 /* 3627 * This specialized allocator has to be a macro for its allocations to be 3628 * accounted separately (to have a separate alloc_tag). 3629 */ 3630 #define dev_alloc_page() dev_alloc_pages(0) 3631 3632 /** 3633 * dev_page_is_reusable - check whether a page can be reused for network Rx 3634 * @page: the page to test 3635 * 3636 * A page shouldn't be considered for reusing/recycling if it was allocated 3637 * under memory pressure or at a distant memory node. 3638 * 3639 * Returns: false if this page should be returned to page allocator, true 3640 * otherwise. 3641 */ 3642 static inline bool dev_page_is_reusable(const struct page *page) 3643 { 3644 return likely(page_to_nid(page) == numa_mem_id() && 3645 !page_is_pfmemalloc(page)); 3646 } 3647 3648 /** 3649 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3650 * @page: The page that was allocated from skb_alloc_page 3651 * @skb: The skb that may need pfmemalloc set 3652 */ 3653 static inline void skb_propagate_pfmemalloc(const struct page *page, 3654 struct sk_buff *skb) 3655 { 3656 if (page_is_pfmemalloc(page)) 3657 skb->pfmemalloc = true; 3658 } 3659 3660 /** 3661 * skb_frag_off() - Returns the offset of a skb fragment 3662 * @frag: the paged fragment 3663 */ 3664 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3665 { 3666 return frag->offset; 3667 } 3668 3669 /** 3670 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3671 * @frag: skb fragment 3672 * @delta: value to add 3673 */ 3674 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3675 { 3676 frag->offset += delta; 3677 } 3678 3679 /** 3680 * skb_frag_off_set() - Sets the offset of a skb fragment 3681 * @frag: skb fragment 3682 * @offset: offset of fragment 3683 */ 3684 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3685 { 3686 frag->offset = offset; 3687 } 3688 3689 /** 3690 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3691 * @fragto: skb fragment where offset is set 3692 * @fragfrom: skb fragment offset is copied from 3693 */ 3694 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3695 const skb_frag_t *fragfrom) 3696 { 3697 fragto->offset = fragfrom->offset; 3698 } 3699 3700 /* Return: true if the skb_frag contains a net_iov. */ 3701 static inline bool skb_frag_is_net_iov(const skb_frag_t *frag) 3702 { 3703 return netmem_is_net_iov(frag->netmem); 3704 } 3705 3706 /** 3707 * skb_frag_net_iov - retrieve the net_iov referred to by fragment 3708 * @frag: the fragment 3709 * 3710 * Return: the &struct net_iov associated with @frag. Returns NULL if this 3711 * frag has no associated net_iov. 3712 */ 3713 static inline struct net_iov *skb_frag_net_iov(const skb_frag_t *frag) 3714 { 3715 if (!skb_frag_is_net_iov(frag)) 3716 return NULL; 3717 3718 return netmem_to_net_iov(frag->netmem); 3719 } 3720 3721 /** 3722 * skb_frag_page - retrieve the page referred to by a paged fragment 3723 * @frag: the paged fragment 3724 * 3725 * Return: the &struct page associated with @frag. Returns NULL if this frag 3726 * has no associated page. 3727 */ 3728 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3729 { 3730 if (skb_frag_is_net_iov(frag)) 3731 return NULL; 3732 3733 return netmem_to_page(frag->netmem); 3734 } 3735 3736 /** 3737 * skb_frag_netmem - retrieve the netmem referred to by a fragment 3738 * @frag: the fragment 3739 * 3740 * Return: the &netmem_ref associated with @frag. 3741 */ 3742 static inline netmem_ref skb_frag_netmem(const skb_frag_t *frag) 3743 { 3744 return frag->netmem; 3745 } 3746 3747 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb, 3748 unsigned int headroom); 3749 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb, 3750 const struct bpf_prog *prog); 3751 3752 /** 3753 * skb_frag_address - gets the address of the data contained in a paged fragment 3754 * @frag: the paged fragment buffer 3755 * 3756 * Returns: the address of the data within @frag. The page must already 3757 * be mapped. 3758 */ 3759 static inline void *skb_frag_address(const skb_frag_t *frag) 3760 { 3761 if (!skb_frag_page(frag)) 3762 return NULL; 3763 3764 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3765 } 3766 3767 /** 3768 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3769 * @frag: the paged fragment buffer 3770 * 3771 * Returns: the address of the data within @frag. Checks that the page 3772 * is mapped and returns %NULL otherwise. 3773 */ 3774 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3775 { 3776 struct page *page = skb_frag_page(frag); 3777 void *ptr; 3778 3779 if (!page) 3780 return NULL; 3781 3782 ptr = page_address(page); 3783 if (unlikely(!ptr)) 3784 return NULL; 3785 3786 return ptr + skb_frag_off(frag); 3787 } 3788 3789 /** 3790 * skb_frag_phys - gets the physical address of the data in a paged fragment 3791 * @frag: the paged fragment buffer 3792 * 3793 * Returns: the physical address of the data within @frag. 3794 */ 3795 static inline phys_addr_t skb_frag_phys(const skb_frag_t *frag) 3796 { 3797 return page_to_phys(skb_frag_page(frag)) + skb_frag_off(frag); 3798 } 3799 3800 /** 3801 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3802 * @fragto: skb fragment where page is set 3803 * @fragfrom: skb fragment page is copied from 3804 */ 3805 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3806 const skb_frag_t *fragfrom) 3807 { 3808 fragto->netmem = fragfrom->netmem; 3809 } 3810 3811 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3812 3813 /** 3814 * __skb_frag_dma_map - maps a paged fragment via the DMA API 3815 * @dev: the device to map the fragment to 3816 * @frag: the paged fragment to map 3817 * @offset: the offset within the fragment (starting at the 3818 * fragment's own offset) 3819 * @size: the number of bytes to map 3820 * @dir: the direction of the mapping (``PCI_DMA_*``) 3821 * 3822 * Maps the page associated with @frag to @device. 3823 */ 3824 static inline dma_addr_t __skb_frag_dma_map(struct device *dev, 3825 const skb_frag_t *frag, 3826 size_t offset, size_t size, 3827 enum dma_data_direction dir) 3828 { 3829 dma_addr_t addr; 3830 3831 if (skb_frag_is_net_iov(frag)) { 3832 addr = netmem_dma_addr_decode( 3833 netmem_get_dma_addr(frag->netmem)); 3834 return addr + offset + frag->offset; 3835 } 3836 return dma_map_page(dev, skb_frag_page(frag), 3837 skb_frag_off(frag) + offset, size, dir); 3838 } 3839 3840 #define skb_frag_dma_map(dev, frag, ...) \ 3841 CONCATENATE(_skb_frag_dma_map, \ 3842 COUNT_ARGS(__VA_ARGS__))(dev, frag, ##__VA_ARGS__) 3843 3844 #define __skb_frag_dma_map1(dev, frag, offset, uf, uo) ({ \ 3845 const skb_frag_t *uf = (frag); \ 3846 size_t uo = (offset); \ 3847 \ 3848 __skb_frag_dma_map(dev, uf, uo, skb_frag_size(uf) - uo, \ 3849 DMA_TO_DEVICE); \ 3850 }) 3851 #define _skb_frag_dma_map1(dev, frag, offset) \ 3852 __skb_frag_dma_map1(dev, frag, offset, __UNIQUE_ID(frag_), \ 3853 __UNIQUE_ID(offset_)) 3854 #define _skb_frag_dma_map0(dev, frag) \ 3855 _skb_frag_dma_map1(dev, frag, 0) 3856 #define _skb_frag_dma_map2(dev, frag, offset, size) \ 3857 __skb_frag_dma_map(dev, frag, offset, size, DMA_TO_DEVICE) 3858 #define _skb_frag_dma_map3(dev, frag, offset, size, dir) \ 3859 __skb_frag_dma_map(dev, frag, offset, size, dir) 3860 3861 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3862 gfp_t gfp_mask) 3863 { 3864 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3865 } 3866 3867 3868 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3869 gfp_t gfp_mask) 3870 { 3871 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3872 } 3873 3874 3875 /** 3876 * skb_clone_writable - is the header of a clone writable 3877 * @skb: buffer to check 3878 * @len: length up to which to write 3879 * 3880 * Returns true if modifying the header part of the cloned buffer 3881 * does not requires the data to be copied. 3882 */ 3883 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3884 { 3885 return !skb_header_cloned(skb) && 3886 skb_headroom(skb) + len <= skb->hdr_len; 3887 } 3888 3889 static inline int skb_try_make_writable(struct sk_buff *skb, 3890 unsigned int write_len) 3891 { 3892 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3893 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3894 } 3895 3896 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3897 int cloned) 3898 { 3899 int delta = 0; 3900 3901 if (headroom > skb_headroom(skb)) 3902 delta = headroom - skb_headroom(skb); 3903 3904 if (delta || cloned) 3905 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3906 GFP_ATOMIC); 3907 return 0; 3908 } 3909 3910 /** 3911 * skb_cow - copy header of skb when it is required 3912 * @skb: buffer to cow 3913 * @headroom: needed headroom 3914 * 3915 * If the skb passed lacks sufficient headroom or its data part 3916 * is shared, data is reallocated. If reallocation fails, an error 3917 * is returned and original skb is not changed. 3918 * 3919 * The result is skb with writable area skb->head...skb->tail 3920 * and at least @headroom of space at head. 3921 */ 3922 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3923 { 3924 return __skb_cow(skb, headroom, skb_cloned(skb)); 3925 } 3926 3927 /** 3928 * skb_cow_head - skb_cow but only making the head writable 3929 * @skb: buffer to cow 3930 * @headroom: needed headroom 3931 * 3932 * This function is identical to skb_cow except that we replace the 3933 * skb_cloned check by skb_header_cloned. It should be used when 3934 * you only need to push on some header and do not need to modify 3935 * the data. 3936 */ 3937 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3938 { 3939 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3940 } 3941 3942 /** 3943 * skb_padto - pad an skbuff up to a minimal size 3944 * @skb: buffer to pad 3945 * @len: minimal length 3946 * 3947 * Pads up a buffer to ensure the trailing bytes exist and are 3948 * blanked. If the buffer already contains sufficient data it 3949 * is untouched. Otherwise it is extended. Returns zero on 3950 * success. The skb is freed on error. 3951 */ 3952 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3953 { 3954 unsigned int size = skb->len; 3955 if (likely(size >= len)) 3956 return 0; 3957 return skb_pad(skb, len - size); 3958 } 3959 3960 /** 3961 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3962 * @skb: buffer to pad 3963 * @len: minimal length 3964 * @free_on_error: free buffer on error 3965 * 3966 * Pads up a buffer to ensure the trailing bytes exist and are 3967 * blanked. If the buffer already contains sufficient data it 3968 * is untouched. Otherwise it is extended. Returns zero on 3969 * success. The skb is freed on error if @free_on_error is true. 3970 */ 3971 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3972 unsigned int len, 3973 bool free_on_error) 3974 { 3975 unsigned int size = skb->len; 3976 3977 if (unlikely(size < len)) { 3978 len -= size; 3979 if (__skb_pad(skb, len, free_on_error)) 3980 return -ENOMEM; 3981 __skb_put(skb, len); 3982 } 3983 return 0; 3984 } 3985 3986 /** 3987 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3988 * @skb: buffer to pad 3989 * @len: minimal length 3990 * 3991 * Pads up a buffer to ensure the trailing bytes exist and are 3992 * blanked. If the buffer already contains sufficient data it 3993 * is untouched. Otherwise it is extended. Returns zero on 3994 * success. The skb is freed on error. 3995 */ 3996 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3997 { 3998 return __skb_put_padto(skb, len, true); 3999 } 4000 4001 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i) 4002 __must_check; 4003 4004 static inline bool skb_can_coalesce_netmem(struct sk_buff *skb, int i, 4005 netmem_ref netmem, int off) 4006 { 4007 if (skb_zcopy(skb)) 4008 return false; 4009 if (i) { 4010 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 4011 4012 return netmem == skb_frag_netmem(frag) && 4013 off == skb_frag_off(frag) + skb_frag_size(frag); 4014 } 4015 return false; 4016 } 4017 4018 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 4019 const struct page *page, int off) 4020 { 4021 return skb_can_coalesce_netmem(skb, i, page_to_netmem(page), off); 4022 } 4023 4024 static inline int __skb_linearize(struct sk_buff *skb) 4025 { 4026 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 4027 } 4028 4029 /** 4030 * skb_linearize - convert paged skb to linear one 4031 * @skb: buffer to linarize 4032 * 4033 * If there is no free memory -ENOMEM is returned, otherwise zero 4034 * is returned and the old skb data released. 4035 */ 4036 static inline int skb_linearize(struct sk_buff *skb) 4037 { 4038 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 4039 } 4040 4041 /** 4042 * skb_has_shared_frag - can any frag be overwritten 4043 * @skb: buffer to test 4044 * 4045 * Return: true if the skb has at least one frag that might be modified 4046 * by an external entity (as in vmsplice()/sendfile()) 4047 */ 4048 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 4049 { 4050 return skb_is_nonlinear(skb) && 4051 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 4052 } 4053 4054 /** 4055 * skb_linearize_cow - make sure skb is linear and writable 4056 * @skb: buffer to process 4057 * 4058 * If there is no free memory -ENOMEM is returned, otherwise zero 4059 * is returned and the old skb data released. 4060 */ 4061 static inline int skb_linearize_cow(struct sk_buff *skb) 4062 { 4063 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 4064 __skb_linearize(skb) : 0; 4065 } 4066 4067 static __always_inline void 4068 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 4069 unsigned int off) 4070 { 4071 if (skb->ip_summed == CHECKSUM_COMPLETE) 4072 skb->csum = csum_block_sub(skb->csum, 4073 csum_partial(start, len, 0), off); 4074 else if (skb->ip_summed == CHECKSUM_PARTIAL && 4075 skb_checksum_start_offset(skb) < 0) 4076 skb->ip_summed = CHECKSUM_NONE; 4077 } 4078 4079 /** 4080 * skb_postpull_rcsum - update checksum for received skb after pull 4081 * @skb: buffer to update 4082 * @start: start of data before pull 4083 * @len: length of data pulled 4084 * 4085 * After doing a pull on a received packet, you need to call this to 4086 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 4087 * CHECKSUM_NONE so that it can be recomputed from scratch. 4088 */ 4089 static __always_inline void 4090 skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len) 4091 { 4092 if (skb->ip_summed == CHECKSUM_COMPLETE) 4093 skb->csum = wsum_negate(csum_partial(start, len, 4094 wsum_negate(skb->csum))); 4095 else if (skb->ip_summed == CHECKSUM_PARTIAL && 4096 skb_checksum_start_offset(skb) < 0) 4097 skb->ip_summed = CHECKSUM_NONE; 4098 } 4099 4100 static __always_inline void 4101 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 4102 unsigned int off) 4103 { 4104 if (skb->ip_summed == CHECKSUM_COMPLETE) 4105 skb->csum = csum_block_add(skb->csum, 4106 csum_partial(start, len, 0), off); 4107 } 4108 4109 /** 4110 * skb_postpush_rcsum - update checksum for received skb after push 4111 * @skb: buffer to update 4112 * @start: start of data after push 4113 * @len: length of data pushed 4114 * 4115 * After doing a push on a received packet, you need to call this to 4116 * update the CHECKSUM_COMPLETE checksum. 4117 */ 4118 static inline void skb_postpush_rcsum(struct sk_buff *skb, 4119 const void *start, unsigned int len) 4120 { 4121 __skb_postpush_rcsum(skb, start, len, 0); 4122 } 4123 4124 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 4125 4126 /** 4127 * skb_push_rcsum - push skb and update receive checksum 4128 * @skb: buffer to update 4129 * @len: length of data pulled 4130 * 4131 * This function performs an skb_push on the packet and updates 4132 * the CHECKSUM_COMPLETE checksum. It should be used on 4133 * receive path processing instead of skb_push unless you know 4134 * that the checksum difference is zero (e.g., a valid IP header) 4135 * or you are setting ip_summed to CHECKSUM_NONE. 4136 */ 4137 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 4138 { 4139 skb_push(skb, len); 4140 skb_postpush_rcsum(skb, skb->data, len); 4141 return skb->data; 4142 } 4143 4144 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 4145 /** 4146 * pskb_trim_rcsum - trim received skb and update checksum 4147 * @skb: buffer to trim 4148 * @len: new length 4149 * 4150 * This is exactly the same as pskb_trim except that it ensures the 4151 * checksum of received packets are still valid after the operation. 4152 * It can change skb pointers. 4153 */ 4154 4155 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4156 { 4157 skb_might_realloc(skb); 4158 if (likely(len >= skb->len)) 4159 return 0; 4160 return pskb_trim_rcsum_slow(skb, len); 4161 } 4162 4163 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 4164 { 4165 if (skb->ip_summed == CHECKSUM_COMPLETE) 4166 skb->ip_summed = CHECKSUM_NONE; 4167 __skb_trim(skb, len); 4168 return 0; 4169 } 4170 4171 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 4172 { 4173 if (skb->ip_summed == CHECKSUM_COMPLETE) 4174 skb->ip_summed = CHECKSUM_NONE; 4175 return __skb_grow(skb, len); 4176 } 4177 4178 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 4179 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 4180 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 4181 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 4182 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 4183 4184 #define skb_queue_walk(queue, skb) \ 4185 for (skb = (queue)->next; \ 4186 skb != (struct sk_buff *)(queue); \ 4187 skb = skb->next) 4188 4189 #define skb_queue_walk_safe(queue, skb, tmp) \ 4190 for (skb = (queue)->next, tmp = skb->next; \ 4191 skb != (struct sk_buff *)(queue); \ 4192 skb = tmp, tmp = skb->next) 4193 4194 #define skb_queue_walk_from(queue, skb) \ 4195 for (; skb != (struct sk_buff *)(queue); \ 4196 skb = skb->next) 4197 4198 #define skb_rbtree_walk(skb, root) \ 4199 for (skb = skb_rb_first(root); skb != NULL; \ 4200 skb = skb_rb_next(skb)) 4201 4202 #define skb_rbtree_walk_from(skb) \ 4203 for (; skb != NULL; \ 4204 skb = skb_rb_next(skb)) 4205 4206 #define skb_rbtree_walk_from_safe(skb, tmp) \ 4207 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 4208 skb = tmp) 4209 4210 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 4211 for (tmp = skb->next; \ 4212 skb != (struct sk_buff *)(queue); \ 4213 skb = tmp, tmp = skb->next) 4214 4215 #define skb_queue_reverse_walk(queue, skb) \ 4216 for (skb = (queue)->prev; \ 4217 skb != (struct sk_buff *)(queue); \ 4218 skb = skb->prev) 4219 4220 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 4221 for (skb = (queue)->prev, tmp = skb->prev; \ 4222 skb != (struct sk_buff *)(queue); \ 4223 skb = tmp, tmp = skb->prev) 4224 4225 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 4226 for (tmp = skb->prev; \ 4227 skb != (struct sk_buff *)(queue); \ 4228 skb = tmp, tmp = skb->prev) 4229 4230 static inline bool skb_has_frag_list(const struct sk_buff *skb) 4231 { 4232 return skb_shinfo(skb)->frag_list != NULL; 4233 } 4234 4235 static inline void skb_frag_list_init(struct sk_buff *skb) 4236 { 4237 skb_shinfo(skb)->frag_list = NULL; 4238 } 4239 4240 #define skb_walk_frags(skb, iter) \ 4241 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 4242 4243 4244 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 4245 int *err, long *timeo_p, 4246 const struct sk_buff *skb); 4247 struct sk_buff *__skb_try_recv_from_queue(struct sk_buff_head *queue, 4248 unsigned int flags, 4249 int *off, int *err, 4250 struct sk_buff **last); 4251 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 4252 struct sk_buff_head *queue, 4253 unsigned int flags, int *off, int *err, 4254 struct sk_buff **last); 4255 struct sk_buff *__skb_recv_datagram(struct sock *sk, 4256 struct sk_buff_head *sk_queue, 4257 unsigned int flags, int *off, int *err); 4258 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 4259 __poll_t datagram_poll_queue(struct file *file, struct socket *sock, 4260 struct poll_table_struct *wait, 4261 struct sk_buff_head *rcv_queue); 4262 __poll_t datagram_poll(struct file *file, struct socket *sock, 4263 struct poll_table_struct *wait); 4264 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 4265 struct iov_iter *to, int size); 4266 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 4267 struct msghdr *msg, int size) 4268 { 4269 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 4270 } 4271 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 4272 struct msghdr *msg); 4273 int skb_copy_and_crc32c_datagram_iter(const struct sk_buff *skb, int offset, 4274 struct iov_iter *to, int len, u32 *crcp); 4275 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 4276 struct iov_iter *from, int len); 4277 int skb_copy_datagram_from_iter_full(struct sk_buff *skb, int offset, 4278 struct iov_iter *from, int len); 4279 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 4280 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 4281 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 4282 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 4283 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 4284 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 4285 int len); 4286 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 4287 struct pipe_inode_info *pipe, unsigned int len, 4288 unsigned int flags); 4289 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 4290 int len); 4291 int skb_send_sock_locked_with_flags(struct sock *sk, struct sk_buff *skb, 4292 int offset, int len, int flags); 4293 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 4294 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 4295 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 4296 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 4297 int len, int hlen); 4298 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 4299 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 4300 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 4301 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 4302 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 4303 unsigned int offset); 4304 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 4305 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 4306 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev); 4307 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 4308 int skb_vlan_pop(struct sk_buff *skb); 4309 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 4310 int skb_eth_pop(struct sk_buff *skb); 4311 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 4312 const unsigned char *src); 4313 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 4314 int mac_len, bool ethernet); 4315 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 4316 bool ethernet); 4317 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 4318 int skb_mpls_dec_ttl(struct sk_buff *skb); 4319 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 4320 gfp_t gfp); 4321 4322 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 4323 { 4324 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 4325 } 4326 4327 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 4328 { 4329 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 4330 } 4331 4332 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 4333 __wsum csum); 4334 u32 skb_crc32c(const struct sk_buff *skb, int offset, int len, u32 crc); 4335 4336 static inline void * __must_check 4337 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 4338 const void *data, int hlen, void *buffer) 4339 { 4340 if (likely(hlen - offset >= len)) 4341 return (void *)data + offset; 4342 4343 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 4344 return NULL; 4345 4346 return buffer; 4347 } 4348 4349 static __always_inline void * __must_check 4350 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4351 { 4352 return __skb_header_pointer(skb, offset, len, skb->data, 4353 skb_headlen(skb), buffer); 4354 } 4355 4356 /* Variant of skb_header_pointer() where @offset is user-controlled 4357 * and potentially negative. 4358 */ 4359 static inline void * __must_check 4360 skb_header_pointer_careful(const struct sk_buff *skb, int offset, 4361 int len, void *buffer) 4362 { 4363 if (unlikely(offset < 0 && -offset > skb_headroom(skb))) 4364 return NULL; 4365 return skb_header_pointer(skb, offset, len, buffer); 4366 } 4367 4368 static inline void * __must_check 4369 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len) 4370 { 4371 unsigned int uoffset = (unsigned int)offset; 4372 4373 if (likely(uoffset <= skb_headlen(skb) && 4374 (unsigned int)len <= skb_headlen(skb) - uoffset)) 4375 return skb->data + offset; 4376 return NULL; 4377 } 4378 4379 /** 4380 * skb_needs_linearize - check if we need to linearize a given skb 4381 * depending on the given device features. 4382 * @skb: socket buffer to check 4383 * @features: net device features 4384 * 4385 * Returns true if either: 4386 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4387 * 2. skb is fragmented and the device does not support SG. 4388 */ 4389 static inline bool skb_needs_linearize(struct sk_buff *skb, 4390 netdev_features_t features) 4391 { 4392 return skb_is_nonlinear(skb) && 4393 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4394 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4395 } 4396 4397 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4398 void *to, 4399 const unsigned int len) 4400 { 4401 memcpy(to, skb->data, len); 4402 } 4403 4404 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4405 const int offset, void *to, 4406 const unsigned int len) 4407 { 4408 memcpy(to, skb->data + offset, len); 4409 } 4410 4411 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4412 const void *from, 4413 const unsigned int len) 4414 { 4415 memcpy(skb->data, from, len); 4416 } 4417 4418 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4419 const int offset, 4420 const void *from, 4421 const unsigned int len) 4422 { 4423 memcpy(skb->data + offset, from, len); 4424 } 4425 4426 void skb_init(void); 4427 4428 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4429 { 4430 return skb->tstamp; 4431 } 4432 4433 /** 4434 * skb_get_timestamp - get timestamp from a skb 4435 * @skb: skb to get stamp from 4436 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4437 * 4438 * Timestamps are stored in the skb as offsets to a base timestamp. 4439 * This function converts the offset back to a struct timeval and stores 4440 * it in stamp. 4441 */ 4442 static inline void skb_get_timestamp(const struct sk_buff *skb, 4443 struct __kernel_old_timeval *stamp) 4444 { 4445 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4446 } 4447 4448 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4449 struct __kernel_sock_timeval *stamp) 4450 { 4451 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4452 4453 stamp->tv_sec = ts.tv_sec; 4454 stamp->tv_usec = ts.tv_nsec / 1000; 4455 } 4456 4457 static inline void skb_get_timestampns(const struct sk_buff *skb, 4458 struct __kernel_old_timespec *stamp) 4459 { 4460 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4461 4462 stamp->tv_sec = ts.tv_sec; 4463 stamp->tv_nsec = ts.tv_nsec; 4464 } 4465 4466 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4467 struct __kernel_timespec *stamp) 4468 { 4469 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4470 4471 stamp->tv_sec = ts.tv_sec; 4472 stamp->tv_nsec = ts.tv_nsec; 4473 } 4474 4475 static inline void __net_timestamp(struct sk_buff *skb) 4476 { 4477 skb->tstamp = ktime_get_real(); 4478 skb->tstamp_type = SKB_CLOCK_REALTIME; 4479 } 4480 4481 static inline ktime_t net_timedelta(ktime_t t) 4482 { 4483 return ktime_sub(ktime_get_real(), t); 4484 } 4485 4486 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4487 u8 tstamp_type) 4488 { 4489 skb->tstamp = kt; 4490 4491 if (kt) 4492 skb->tstamp_type = tstamp_type; 4493 else 4494 skb->tstamp_type = SKB_CLOCK_REALTIME; 4495 } 4496 4497 static inline void skb_set_delivery_type_by_clockid(struct sk_buff *skb, 4498 ktime_t kt, clockid_t clockid) 4499 { 4500 u8 tstamp_type = SKB_CLOCK_REALTIME; 4501 4502 switch (clockid) { 4503 case CLOCK_REALTIME: 4504 break; 4505 case CLOCK_MONOTONIC: 4506 tstamp_type = SKB_CLOCK_MONOTONIC; 4507 break; 4508 case CLOCK_TAI: 4509 tstamp_type = SKB_CLOCK_TAI; 4510 break; 4511 default: 4512 WARN_ON_ONCE(1); 4513 kt = 0; 4514 } 4515 4516 skb_set_delivery_time(skb, kt, tstamp_type); 4517 } 4518 4519 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4520 4521 /* It is used in the ingress path to clear the delivery_time. 4522 * If needed, set the skb->tstamp to the (rcv) timestamp. 4523 */ 4524 static __always_inline void skb_clear_delivery_time(struct sk_buff *skb) 4525 { 4526 if (skb->tstamp_type) { 4527 skb->tstamp_type = SKB_CLOCK_REALTIME; 4528 if (static_branch_unlikely(&netstamp_needed_key)) 4529 skb->tstamp = ktime_get_real(); 4530 else 4531 skb->tstamp = 0; 4532 } 4533 } 4534 4535 static inline void skb_clear_tstamp(struct sk_buff *skb) 4536 { 4537 if (skb->tstamp_type) 4538 return; 4539 4540 skb->tstamp = 0; 4541 } 4542 4543 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4544 { 4545 if (skb->tstamp_type) 4546 return 0; 4547 4548 return skb->tstamp; 4549 } 4550 4551 static __always_inline ktime_t 4552 skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4553 { 4554 if (skb->tstamp_type != SKB_CLOCK_MONOTONIC && skb->tstamp) 4555 return skb->tstamp; 4556 4557 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4558 return ktime_get_real(); 4559 4560 return 0; 4561 } 4562 4563 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4564 { 4565 return skb_shinfo(skb)->meta_len; 4566 } 4567 4568 static inline void *skb_metadata_end(const struct sk_buff *skb) 4569 { 4570 return skb_mac_header(skb); 4571 } 4572 4573 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4574 const struct sk_buff *skb_b, 4575 u8 meta_len) 4576 { 4577 const void *a = skb_metadata_end(skb_a); 4578 const void *b = skb_metadata_end(skb_b); 4579 u64 diffs = 0; 4580 4581 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) || 4582 BITS_PER_LONG != 64) 4583 goto slow; 4584 4585 /* Using more efficient variant than plain call to memcmp(). */ 4586 switch (meta_len) { 4587 #define __it(x, op) (x -= sizeof(u##op)) 4588 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4589 case 32: diffs |= __it_diff(a, b, 64); 4590 fallthrough; 4591 case 24: diffs |= __it_diff(a, b, 64); 4592 fallthrough; 4593 case 16: diffs |= __it_diff(a, b, 64); 4594 fallthrough; 4595 case 8: diffs |= __it_diff(a, b, 64); 4596 break; 4597 case 28: diffs |= __it_diff(a, b, 64); 4598 fallthrough; 4599 case 20: diffs |= __it_diff(a, b, 64); 4600 fallthrough; 4601 case 12: diffs |= __it_diff(a, b, 64); 4602 fallthrough; 4603 case 4: diffs |= __it_diff(a, b, 32); 4604 break; 4605 default: 4606 slow: 4607 return memcmp(a - meta_len, b - meta_len, meta_len); 4608 } 4609 return diffs; 4610 } 4611 4612 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4613 const struct sk_buff *skb_b) 4614 { 4615 u8 len_a = skb_metadata_len(skb_a); 4616 u8 len_b = skb_metadata_len(skb_b); 4617 4618 if (!(len_a | len_b)) 4619 return false; 4620 4621 return len_a != len_b ? 4622 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4623 } 4624 4625 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4626 { 4627 skb_shinfo(skb)->meta_len = meta_len; 4628 } 4629 4630 static inline void skb_metadata_clear(struct sk_buff *skb) 4631 { 4632 skb_metadata_set(skb, 0); 4633 } 4634 4635 /** 4636 * skb_data_move - Move packet data and metadata after skb_push() or skb_pull(). 4637 * @skb: packet to operate on 4638 * @len: number of bytes pushed or pulled from &sk_buff->data 4639 * @n: number of bytes to memmove() from pre-push/pull &sk_buff->data 4640 * 4641 * Moves @n bytes of packet data, can be zero, and all bytes of skb metadata. 4642 * 4643 * Assumes metadata is located immediately before &sk_buff->data prior to the 4644 * push/pull, and that sufficient headroom exists to hold it after an 4645 * skb_push(). Otherwise, metadata is cleared and a one-time warning is issued. 4646 * 4647 * Prefer skb_postpull_data_move() or skb_postpush_data_move() to calling this 4648 * helper directly. 4649 */ 4650 static inline void skb_data_move(struct sk_buff *skb, const int len, 4651 const unsigned int n) 4652 { 4653 const u8 meta_len = skb_metadata_len(skb); 4654 u8 *meta, *meta_end; 4655 4656 if (!len || (!n && !meta_len)) 4657 return; 4658 4659 if (!meta_len) 4660 goto no_metadata; 4661 4662 meta_end = skb_metadata_end(skb); 4663 meta = meta_end - meta_len; 4664 4665 if (WARN_ON_ONCE(meta_end + len != skb->data || 4666 meta_len > skb_headroom(skb))) { 4667 skb_metadata_clear(skb); 4668 goto no_metadata; 4669 } 4670 4671 memmove(meta + len, meta, meta_len + n); 4672 return; 4673 4674 no_metadata: 4675 memmove(skb->data, skb->data - len, n); 4676 } 4677 4678 /** 4679 * skb_postpull_data_move - Move packet data and metadata after skb_pull(). 4680 * @skb: packet to operate on 4681 * @len: number of bytes pulled from &sk_buff->data 4682 * @n: number of bytes to memmove() from pre-pull &sk_buff->data 4683 * 4684 * See skb_data_move() for details. 4685 */ 4686 static inline void skb_postpull_data_move(struct sk_buff *skb, 4687 const unsigned int len, 4688 const unsigned int n) 4689 { 4690 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 4691 skb_data_move(skb, len, n); 4692 } 4693 4694 /** 4695 * skb_postpush_data_move - Move packet data and metadata after skb_push(). 4696 * @skb: packet to operate on 4697 * @len: number of bytes pushed onto &sk_buff->data 4698 * @n: number of bytes to memmove() from pre-push &sk_buff->data 4699 * 4700 * See skb_data_move() for details. 4701 */ 4702 static inline void skb_postpush_data_move(struct sk_buff *skb, 4703 const unsigned int len, 4704 const unsigned int n) 4705 { 4706 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 4707 skb_data_move(skb, -len, n); 4708 } 4709 4710 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4711 4712 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4713 4714 void skb_clone_tx_timestamp(struct sk_buff *skb); 4715 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4716 4717 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4718 4719 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4720 { 4721 } 4722 4723 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4724 { 4725 return false; 4726 } 4727 4728 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4729 4730 /** 4731 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4732 * 4733 * PHY drivers may accept clones of transmitted packets for 4734 * timestamping via their phy_driver.txtstamp method. These drivers 4735 * must call this function to return the skb back to the stack with a 4736 * timestamp. 4737 * 4738 * @skb: clone of the original outgoing packet 4739 * @hwtstamps: hardware time stamps 4740 * 4741 */ 4742 void skb_complete_tx_timestamp(struct sk_buff *skb, 4743 struct skb_shared_hwtstamps *hwtstamps); 4744 4745 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4746 struct skb_shared_hwtstamps *hwtstamps, 4747 struct sock *sk, int tstype); 4748 4749 /** 4750 * skb_tstamp_tx - queue clone of skb with send time stamps 4751 * @orig_skb: the original outgoing packet 4752 * @hwtstamps: hardware time stamps, may be NULL if not available 4753 * 4754 * If the skb has a socket associated, then this function clones the 4755 * skb (thus sharing the actual data and optional structures), stores 4756 * the optional hardware time stamping information (if non NULL) or 4757 * generates a software time stamp (otherwise), then queues the clone 4758 * to the error queue of the socket. Errors are silently ignored. 4759 */ 4760 void skb_tstamp_tx(struct sk_buff *orig_skb, 4761 struct skb_shared_hwtstamps *hwtstamps); 4762 4763 /** 4764 * skb_tx_timestamp() - Driver hook for transmit timestamping 4765 * 4766 * Ethernet MAC Drivers should call this function in their hard_xmit() 4767 * function immediately before giving the sk_buff to the MAC hardware. 4768 * 4769 * Specifically, one should make absolutely sure that this function is 4770 * called before TX completion of this packet can trigger. Otherwise 4771 * the packet could potentially already be freed. 4772 * 4773 * @skb: A socket buffer. 4774 */ 4775 static inline void skb_tx_timestamp(struct sk_buff *skb) 4776 { 4777 skb_clone_tx_timestamp(skb); 4778 if (skb_shinfo(skb)->tx_flags & (SKBTX_SW_TSTAMP | SKBTX_BPF)) 4779 skb_tstamp_tx(skb, NULL); 4780 } 4781 4782 /** 4783 * skb_complete_wifi_ack - deliver skb with wifi status 4784 * 4785 * @skb: the original outgoing packet 4786 * @acked: ack status 4787 * 4788 */ 4789 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4790 4791 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4792 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4793 4794 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4795 { 4796 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4797 skb->csum_valid || 4798 (skb->ip_summed == CHECKSUM_PARTIAL && 4799 skb_checksum_start_offset(skb) >= 0)); 4800 } 4801 4802 /** 4803 * skb_checksum_complete - Calculate checksum of an entire packet 4804 * @skb: packet to process 4805 * 4806 * This function calculates the checksum over the entire packet plus 4807 * the value of skb->csum. The latter can be used to supply the 4808 * checksum of a pseudo header as used by TCP/UDP. It returns the 4809 * checksum. 4810 * 4811 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4812 * this function can be used to verify that checksum on received 4813 * packets. In that case the function should return zero if the 4814 * checksum is correct. In particular, this function will return zero 4815 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4816 * hardware has already verified the correctness of the checksum. 4817 */ 4818 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4819 { 4820 return skb_csum_unnecessary(skb) ? 4821 0 : __skb_checksum_complete(skb); 4822 } 4823 4824 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4825 { 4826 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4827 if (skb->csum_level == 0) 4828 skb->ip_summed = CHECKSUM_NONE; 4829 else 4830 skb->csum_level--; 4831 } 4832 } 4833 4834 static __always_inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4835 { 4836 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4837 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4838 skb->csum_level++; 4839 } else if (skb->ip_summed == CHECKSUM_NONE) { 4840 skb->ip_summed = CHECKSUM_UNNECESSARY; 4841 skb->csum_level = 0; 4842 } 4843 } 4844 4845 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4846 { 4847 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4848 skb->ip_summed = CHECKSUM_NONE; 4849 skb->csum_level = 0; 4850 } 4851 } 4852 4853 /* Check if we need to perform checksum complete validation. 4854 * 4855 * Returns: true if checksum complete is needed, false otherwise 4856 * (either checksum is unnecessary or zero checksum is allowed). 4857 */ 4858 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4859 bool zero_okay, 4860 __sum16 check) 4861 { 4862 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4863 skb->csum_valid = 1; 4864 __skb_decr_checksum_unnecessary(skb); 4865 return false; 4866 } 4867 4868 return true; 4869 } 4870 4871 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4872 * in checksum_init. 4873 */ 4874 #define CHECKSUM_BREAK 76 4875 4876 /* Unset checksum-complete 4877 * 4878 * Unset checksum complete can be done when packet is being modified 4879 * (uncompressed for instance) and checksum-complete value is 4880 * invalidated. 4881 */ 4882 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4883 { 4884 if (skb->ip_summed == CHECKSUM_COMPLETE) 4885 skb->ip_summed = CHECKSUM_NONE; 4886 } 4887 4888 /* Validate (init) checksum based on checksum complete. 4889 * 4890 * Return values: 4891 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4892 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4893 * checksum is stored in skb->csum for use in __skb_checksum_complete 4894 * non-zero: value of invalid checksum 4895 * 4896 */ 4897 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4898 bool complete, 4899 __wsum psum) 4900 { 4901 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4902 if (!csum_fold(csum_add(psum, skb->csum))) { 4903 skb->csum_valid = 1; 4904 return 0; 4905 } 4906 } 4907 4908 skb->csum = psum; 4909 4910 if (complete || skb->len <= CHECKSUM_BREAK) { 4911 __sum16 csum; 4912 4913 csum = __skb_checksum_complete(skb); 4914 skb->csum_valid = !csum; 4915 return csum; 4916 } 4917 4918 return 0; 4919 } 4920 4921 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4922 { 4923 return 0; 4924 } 4925 4926 /* Perform checksum validate (init). Note that this is a macro since we only 4927 * want to calculate the pseudo header which is an input function if necessary. 4928 * First we try to validate without any computation (checksum unnecessary) and 4929 * then calculate based on checksum complete calling the function to compute 4930 * pseudo header. 4931 * 4932 * Return values: 4933 * 0: checksum is validated or try to in skb_checksum_complete 4934 * non-zero: value of invalid checksum 4935 */ 4936 #define __skb_checksum_validate(skb, proto, complete, \ 4937 zero_okay, check, compute_pseudo) \ 4938 ({ \ 4939 __sum16 __ret = 0; \ 4940 skb->csum_valid = 0; \ 4941 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4942 __ret = __skb_checksum_validate_complete(skb, \ 4943 complete, compute_pseudo(skb, proto)); \ 4944 __ret; \ 4945 }) 4946 4947 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4948 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4949 4950 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4951 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4952 4953 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4954 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4955 4956 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4957 compute_pseudo) \ 4958 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4959 4960 #define skb_checksum_simple_validate(skb) \ 4961 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4962 4963 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4964 { 4965 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4966 } 4967 4968 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4969 { 4970 skb->csum = ~pseudo; 4971 skb->ip_summed = CHECKSUM_COMPLETE; 4972 } 4973 4974 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4975 do { \ 4976 if (__skb_checksum_convert_check(skb)) \ 4977 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4978 } while (0) 4979 4980 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4981 u16 start, u16 offset) 4982 { 4983 skb->ip_summed = CHECKSUM_PARTIAL; 4984 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4985 skb->csum_offset = offset - start; 4986 } 4987 4988 /* Update skbuf and packet to reflect the remote checksum offload operation. 4989 * When called, ptr indicates the starting point for skb->csum when 4990 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4991 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4992 */ 4993 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4994 int start, int offset, bool nopartial) 4995 { 4996 __wsum delta; 4997 4998 if (!nopartial) { 4999 skb_remcsum_adjust_partial(skb, ptr, start, offset); 5000 return; 5001 } 5002 5003 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 5004 __skb_checksum_complete(skb); 5005 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 5006 } 5007 5008 delta = remcsum_adjust(ptr, skb->csum, start, offset); 5009 5010 /* Adjust skb->csum since we changed the packet */ 5011 skb->csum = csum_add(skb->csum, delta); 5012 } 5013 5014 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 5015 { 5016 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 5017 return (void *)(skb->_nfct & NFCT_PTRMASK); 5018 #else 5019 return NULL; 5020 #endif 5021 } 5022 5023 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 5024 { 5025 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 5026 return skb->_nfct; 5027 #else 5028 return 0UL; 5029 #endif 5030 } 5031 5032 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 5033 { 5034 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 5035 DEBUG_NET_WARN_ON_ONCE(skb->_nfct & NFCT_PTRMASK); 5036 skb->slow_gro |= !!nfct; 5037 skb->_nfct = nfct; 5038 #endif 5039 } 5040 5041 #ifdef CONFIG_SKB_EXTENSIONS 5042 enum skb_ext_id { 5043 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 5044 SKB_EXT_BRIDGE_NF, 5045 #endif 5046 #ifdef CONFIG_XFRM 5047 SKB_EXT_SEC_PATH, 5048 #endif 5049 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5050 TC_SKB_EXT, 5051 #endif 5052 #if IS_ENABLED(CONFIG_MPTCP) 5053 SKB_EXT_MPTCP, 5054 #endif 5055 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 5056 SKB_EXT_MCTP, 5057 #endif 5058 #if IS_ENABLED(CONFIG_INET_PSP) 5059 SKB_EXT_PSP, 5060 #endif 5061 #if IS_ENABLED(CONFIG_CAN) 5062 SKB_EXT_CAN, 5063 #endif 5064 SKB_EXT_NUM, /* must be last */ 5065 }; 5066 5067 /** 5068 * struct skb_ext - sk_buff extensions 5069 * @refcnt: 1 on allocation, deallocated on 0 5070 * @offset: offset to add to @data to obtain extension address 5071 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 5072 * @data: start of extension data, variable sized 5073 * 5074 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 5075 * to use 'u8' types while allowing up to 2kb worth of extension data. 5076 */ 5077 struct skb_ext { 5078 refcount_t refcnt; 5079 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 5080 u8 chunks; /* same */ 5081 char data[] __aligned(8); 5082 }; 5083 5084 struct skb_ext *__skb_ext_alloc(gfp_t flags); 5085 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 5086 struct skb_ext *ext); 5087 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 5088 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 5089 void __skb_ext_put(struct skb_ext *ext); 5090 5091 static inline void skb_ext_put(struct sk_buff *skb) 5092 { 5093 if (skb->active_extensions) 5094 __skb_ext_put(skb->extensions); 5095 } 5096 5097 static inline void __skb_ext_copy(struct sk_buff *dst, 5098 const struct sk_buff *src) 5099 { 5100 dst->active_extensions = src->active_extensions; 5101 5102 if (src->active_extensions) { 5103 struct skb_ext *ext = src->extensions; 5104 5105 refcount_inc(&ext->refcnt); 5106 dst->extensions = ext; 5107 } 5108 } 5109 5110 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 5111 { 5112 skb_ext_put(dst); 5113 __skb_ext_copy(dst, src); 5114 } 5115 5116 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 5117 { 5118 return !!ext->offset[i]; 5119 } 5120 5121 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 5122 { 5123 return skb->active_extensions & (1 << id); 5124 } 5125 5126 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 5127 { 5128 if (skb_ext_exist(skb, id)) 5129 __skb_ext_del(skb, id); 5130 } 5131 5132 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 5133 { 5134 if (skb_ext_exist(skb, id)) { 5135 struct skb_ext *ext = skb->extensions; 5136 5137 return (void *)ext + (ext->offset[id] << 3); 5138 } 5139 5140 return NULL; 5141 } 5142 5143 static inline void skb_ext_reset(struct sk_buff *skb) 5144 { 5145 if (unlikely(skb->active_extensions)) { 5146 __skb_ext_put(skb->extensions); 5147 skb->active_extensions = 0; 5148 } 5149 } 5150 5151 static inline bool skb_has_extensions(struct sk_buff *skb) 5152 { 5153 return unlikely(skb->active_extensions); 5154 } 5155 #else 5156 static inline void __skb_ext_put(struct skb_ext *ext) {} 5157 static inline void skb_ext_put(struct sk_buff *skb) {} 5158 static inline void skb_ext_reset(struct sk_buff *skb) {} 5159 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 5160 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 5161 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 5162 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 5163 #endif /* CONFIG_SKB_EXTENSIONS */ 5164 5165 static inline void nf_reset_ct(struct sk_buff *skb) 5166 { 5167 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5168 nf_conntrack_put(skb_nfct(skb)); 5169 skb->_nfct = 0; 5170 #endif 5171 } 5172 5173 static inline void nf_reset_trace(struct sk_buff *skb) 5174 { 5175 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 5176 skb->nf_trace = 0; 5177 #endif 5178 } 5179 5180 static inline void ipvs_reset(struct sk_buff *skb) 5181 { 5182 #if IS_ENABLED(CONFIG_IP_VS) 5183 skb->ipvs_property = 0; 5184 #endif 5185 } 5186 5187 /* Note: This doesn't put any conntrack info in dst. */ 5188 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 5189 bool copy) 5190 { 5191 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5192 dst->_nfct = src->_nfct; 5193 nf_conntrack_get(skb_nfct(src)); 5194 #endif 5195 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 5196 if (copy) 5197 dst->nf_trace = src->nf_trace; 5198 #endif 5199 } 5200 5201 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 5202 { 5203 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5204 nf_conntrack_put(skb_nfct(dst)); 5205 #endif 5206 dst->slow_gro = src->slow_gro; 5207 __nf_copy(dst, src, true); 5208 } 5209 5210 #ifdef CONFIG_NETWORK_SECMARK 5211 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 5212 { 5213 to->secmark = from->secmark; 5214 } 5215 5216 static inline void skb_init_secmark(struct sk_buff *skb) 5217 { 5218 skb->secmark = 0; 5219 } 5220 #else 5221 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 5222 { } 5223 5224 static inline void skb_init_secmark(struct sk_buff *skb) 5225 { } 5226 #endif 5227 5228 static inline int secpath_exists(const struct sk_buff *skb) 5229 { 5230 #ifdef CONFIG_XFRM 5231 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 5232 #else 5233 return 0; 5234 #endif 5235 } 5236 5237 static inline bool skb_irq_freeable(const struct sk_buff *skb) 5238 { 5239 return !skb->destructor && 5240 !secpath_exists(skb) && 5241 !skb_nfct(skb) && 5242 !skb->_skb_refdst && 5243 !skb_has_frag_list(skb); 5244 } 5245 5246 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 5247 { 5248 skb->queue_mapping = queue_mapping; 5249 } 5250 5251 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 5252 { 5253 return skb->queue_mapping; 5254 } 5255 5256 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 5257 { 5258 to->queue_mapping = from->queue_mapping; 5259 } 5260 5261 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 5262 { 5263 skb->queue_mapping = rx_queue + 1; 5264 } 5265 5266 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 5267 { 5268 return skb->queue_mapping - 1; 5269 } 5270 5271 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 5272 { 5273 return skb->queue_mapping != 0; 5274 } 5275 5276 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 5277 { 5278 skb->dst_pending_confirm = val; 5279 } 5280 5281 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 5282 { 5283 return skb->dst_pending_confirm != 0; 5284 } 5285 5286 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 5287 { 5288 #ifdef CONFIG_XFRM 5289 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 5290 #else 5291 return NULL; 5292 #endif 5293 } 5294 5295 static inline bool skb_is_gso(const struct sk_buff *skb) 5296 { 5297 return skb_shinfo(skb)->gso_size; 5298 } 5299 5300 /* Note: Should be called only if skb_is_gso(skb) is true */ 5301 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 5302 { 5303 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 5304 } 5305 5306 /* Note: Should be called only if skb_is_gso(skb) is true */ 5307 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 5308 { 5309 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 5310 } 5311 5312 /* Note: Should be called only if skb_is_gso(skb) is true */ 5313 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 5314 { 5315 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 5316 } 5317 5318 static inline void skb_gso_reset(struct sk_buff *skb) 5319 { 5320 skb_shinfo(skb)->gso_size = 0; 5321 skb_shinfo(skb)->gso_segs = 0; 5322 skb_shinfo(skb)->gso_type = 0; 5323 } 5324 5325 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 5326 u16 increment) 5327 { 5328 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5329 return; 5330 shinfo->gso_size += increment; 5331 } 5332 5333 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 5334 u16 decrement) 5335 { 5336 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5337 return; 5338 shinfo->gso_size -= decrement; 5339 } 5340 5341 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 5342 5343 static __always_inline bool skb_warn_if_lro(const struct sk_buff *skb) 5344 { 5345 /* LRO sets gso_size but not gso_type, whereas if GSO is really 5346 * wanted then gso_type will be set. */ 5347 const struct skb_shared_info *shinfo = skb_shinfo(skb); 5348 5349 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 5350 unlikely(shinfo->gso_type == 0)) { 5351 __skb_warn_lro_forwarding(skb); 5352 return true; 5353 } 5354 return false; 5355 } 5356 5357 static inline void skb_forward_csum(struct sk_buff *skb) 5358 { 5359 /* Unfortunately we don't support this one. Any brave souls? */ 5360 if (skb->ip_summed == CHECKSUM_COMPLETE) 5361 skb->ip_summed = CHECKSUM_NONE; 5362 } 5363 5364 /** 5365 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 5366 * @skb: skb to check 5367 * 5368 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 5369 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 5370 * use this helper, to document places where we make this assertion. 5371 */ 5372 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 5373 { 5374 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 5375 } 5376 5377 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 5378 5379 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 5380 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 5381 unsigned int transport_len, 5382 __sum16(*skb_chkf)(struct sk_buff *skb)); 5383 5384 /** 5385 * skb_head_is_locked - Determine if the skb->head is locked down 5386 * @skb: skb to check 5387 * 5388 * The head on skbs build around a head frag can be removed if they are 5389 * not cloned. This function returns true if the skb head is locked down 5390 * due to either being allocated via kmalloc, or by being a clone with 5391 * multiple references to the head. 5392 */ 5393 static inline bool skb_head_is_locked(const struct sk_buff *skb) 5394 { 5395 return !skb->head_frag || skb_cloned(skb); 5396 } 5397 5398 /* Local Checksum Offload. 5399 * Compute outer checksum based on the assumption that the 5400 * inner checksum will be offloaded later. 5401 * See Documentation/networking/checksum-offloads.rst for 5402 * explanation of how this works. 5403 * Fill in outer checksum adjustment (e.g. with sum of outer 5404 * pseudo-header) before calling. 5405 * Also ensure that inner checksum is in linear data area. 5406 */ 5407 static inline __wsum lco_csum(struct sk_buff *skb) 5408 { 5409 unsigned char *csum_start = skb_checksum_start(skb); 5410 unsigned char *l4_hdr = skb_transport_header(skb); 5411 __wsum partial; 5412 5413 /* Start with complement of inner checksum adjustment */ 5414 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 5415 skb->csum_offset)); 5416 5417 /* Add in checksum of our headers (incl. outer checksum 5418 * adjustment filled in by caller) and return result. 5419 */ 5420 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5421 } 5422 5423 static inline bool skb_is_redirected(const struct sk_buff *skb) 5424 { 5425 return skb->redirected; 5426 } 5427 5428 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5429 { 5430 skb->redirected = 1; 5431 #ifdef CONFIG_NET_REDIRECT 5432 skb->from_ingress = from_ingress; 5433 if (skb->from_ingress) 5434 skb_clear_tstamp(skb); 5435 #endif 5436 } 5437 5438 static inline void skb_reset_redirect(struct sk_buff *skb) 5439 { 5440 skb->redirected = 0; 5441 } 5442 5443 static inline void skb_set_redirected_noclear(struct sk_buff *skb, 5444 bool from_ingress) 5445 { 5446 skb->redirected = 1; 5447 #ifdef CONFIG_NET_REDIRECT 5448 skb->from_ingress = from_ingress; 5449 #endif 5450 } 5451 5452 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5453 { 5454 #if IS_ENABLED(CONFIG_IP_SCTP) 5455 return skb->csum_not_inet; 5456 #else 5457 return 0; 5458 #endif 5459 } 5460 5461 static inline void skb_reset_csum_not_inet(struct sk_buff *skb) 5462 { 5463 skb->ip_summed = CHECKSUM_NONE; 5464 #if IS_ENABLED(CONFIG_IP_SCTP) 5465 skb->csum_not_inet = 0; 5466 #endif 5467 } 5468 5469 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5470 struct kcov_common_handle_id kcov_handle) 5471 { 5472 skb->kcov_handle = kcov_handle; 5473 } 5474 5475 static inline struct kcov_common_handle_id skb_get_kcov_handle(struct sk_buff *skb) 5476 { 5477 return skb->kcov_handle; 5478 } 5479 5480 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5481 { 5482 #ifdef CONFIG_PAGE_POOL 5483 skb->pp_recycle = 1; 5484 #endif 5485 } 5486 5487 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter, 5488 ssize_t maxsize); 5489 5490 #endif /* __KERNEL__ */ 5491 #endif /* _LINUX_SKBUFF_H */ 5492