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