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