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 <linux/net.h> 27 #include <linux/textsearch.h> 28 #include <net/checksum.h> 29 #include <linux/rcupdate.h> 30 #include <linux/hrtimer.h> 31 #include <linux/dma-mapping.h> 32 #include <linux/netdev_features.h> 33 #include <linux/sched.h> 34 #include <linux/sched/clock.h> 35 #include <net/flow_dissector.h> 36 #include <linux/splice.h> 37 #include <linux/in6.h> 38 #include <linux/if_packet.h> 39 #include <linux/llist.h> 40 #include <net/flow.h> 41 #include <net/page_pool.h> 42 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 43 #include <linux/netfilter/nf_conntrack_common.h> 44 #endif 45 #include <net/net_debug.h> 46 #include <net/dropreason.h> 47 48 /** 49 * DOC: skb checksums 50 * 51 * The interface for checksum offload between the stack and networking drivers 52 * is as follows... 53 * 54 * IP checksum related features 55 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 56 * 57 * Drivers advertise checksum offload capabilities in the features of a device. 58 * From the stack's point of view these are capabilities offered by the driver. 59 * A driver typically only advertises features that it is capable of offloading 60 * to its device. 61 * 62 * .. flat-table:: Checksum related device features 63 * :widths: 1 10 64 * 65 * * - %NETIF_F_HW_CSUM 66 * - The driver (or its device) is able to compute one 67 * IP (one's complement) checksum for any combination 68 * of protocols or protocol layering. The checksum is 69 * computed and set in a packet per the CHECKSUM_PARTIAL 70 * interface (see below). 71 * 72 * * - %NETIF_F_IP_CSUM 73 * - Driver (device) is only able to checksum plain 74 * TCP or UDP packets over IPv4. These are specifically 75 * unencapsulated packets of the form IPv4|TCP or 76 * IPv4|UDP where the Protocol field in the IPv4 header 77 * is TCP or UDP. The IPv4 header may contain IP options. 78 * This feature cannot be set in features for a device 79 * with NETIF_F_HW_CSUM also set. This feature is being 80 * DEPRECATED (see below). 81 * 82 * * - %NETIF_F_IPV6_CSUM 83 * - Driver (device) is only able to checksum plain 84 * TCP or UDP packets over IPv6. These are specifically 85 * unencapsulated packets of the form IPv6|TCP or 86 * IPv6|UDP where the Next Header field in the IPv6 87 * header is either TCP or UDP. IPv6 extension headers 88 * are not supported with this feature. This feature 89 * cannot be set in features for a device with 90 * NETIF_F_HW_CSUM also set. This feature is being 91 * DEPRECATED (see below). 92 * 93 * * - %NETIF_F_RXCSUM 94 * - Driver (device) performs receive checksum offload. 95 * This flag is only used to disable the RX checksum 96 * feature for a device. The stack will accept receive 97 * checksum indication in packets received on a device 98 * regardless of whether NETIF_F_RXCSUM is set. 99 * 100 * Checksumming of received packets by device 101 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 102 * 103 * Indication of checksum verification is set in &sk_buff.ip_summed. 104 * Possible values are: 105 * 106 * - %CHECKSUM_NONE 107 * 108 * Device did not checksum this packet e.g. due to lack of capabilities. 109 * The packet contains full (though not verified) checksum in packet but 110 * not in skb->csum. Thus, skb->csum is undefined in this case. 111 * 112 * - %CHECKSUM_UNNECESSARY 113 * 114 * The hardware you're dealing with doesn't calculate the full checksum 115 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums 116 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY 117 * if their checksums are okay. &sk_buff.csum is still undefined in this case 118 * though. A driver or device must never modify the checksum field in the 119 * packet even if checksum is verified. 120 * 121 * %CHECKSUM_UNNECESSARY is applicable to following protocols: 122 * 123 * - TCP: IPv6 and IPv4. 124 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 125 * zero UDP checksum for either IPv4 or IPv6, the networking stack 126 * may perform further validation in this case. 127 * - GRE: only if the checksum is present in the header. 128 * - SCTP: indicates the CRC in SCTP header has been validated. 129 * - FCOE: indicates the CRC in FC frame has been validated. 130 * 131 * &sk_buff.csum_level indicates the number of consecutive checksums found in 132 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY. 133 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 134 * and a device is able to verify the checksums for UDP (possibly zero), 135 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to 136 * two. If the device were only able to verify the UDP checksum and not 137 * GRE, either because it doesn't support GRE checksum or because GRE 138 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 139 * not considered in this case). 140 * 141 * - %CHECKSUM_COMPLETE 142 * 143 * This is the most generic way. The device supplied checksum of the _whole_ 144 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the 145 * hardware doesn't need to parse L3/L4 headers to implement this. 146 * 147 * Notes: 148 * 149 * - Even if device supports only some protocols, but is able to produce 150 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 151 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 152 * 153 * - %CHECKSUM_PARTIAL 154 * 155 * A checksum is set up to be offloaded to a device as described in the 156 * output description for CHECKSUM_PARTIAL. This may occur on a packet 157 * received directly from another Linux OS, e.g., a virtualized Linux kernel 158 * on the same host, or it may be set in the input path in GRO or remote 159 * checksum offload. For the purposes of checksum verification, the checksum 160 * referred to by skb->csum_start + skb->csum_offset and any preceding 161 * checksums in the packet are considered verified. Any checksums in the 162 * packet that are after the checksum being offloaded are not considered to 163 * be verified. 164 * 165 * Checksumming on transmit for non-GSO 166 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 167 * 168 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet. 169 * Values are: 170 * 171 * - %CHECKSUM_PARTIAL 172 * 173 * The driver is required to checksum the packet as seen by hard_start_xmit() 174 * from &sk_buff.csum_start up to the end, and to record/write the checksum at 175 * offset &sk_buff.csum_start + &sk_buff.csum_offset. 176 * A driver may verify that the 177 * csum_start and csum_offset values are valid values given the length and 178 * offset of the packet, but it should not attempt to validate that the 179 * checksum refers to a legitimate transport layer checksum -- it is the 180 * purview of the stack to validate that csum_start and csum_offset are set 181 * correctly. 182 * 183 * When the stack requests checksum offload for a packet, the driver MUST 184 * ensure that the checksum is set correctly. A driver can either offload the 185 * checksum calculation to the device, or call skb_checksum_help (in the case 186 * that the device does not support offload for a particular checksum). 187 * 188 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of 189 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate 190 * checksum offload capability. 191 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based 192 * on network device checksumming capabilities: if a packet does not match 193 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of 194 * &sk_buff.csum_not_inet, see :ref:`crc`) 195 * is called to resolve the checksum. 196 * 197 * - %CHECKSUM_NONE 198 * 199 * The skb was already checksummed by the protocol, or a checksum is not 200 * required. 201 * 202 * - %CHECKSUM_UNNECESSARY 203 * 204 * This has the same meaning as CHECKSUM_NONE for checksum offload on 205 * output. 206 * 207 * - %CHECKSUM_COMPLETE 208 * 209 * Not used in checksum output. If a driver observes a packet with this value 210 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set. 211 * 212 * .. _crc: 213 * 214 * Non-IP checksum (CRC) offloads 215 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 216 * 217 * .. flat-table:: 218 * :widths: 1 10 219 * 220 * * - %NETIF_F_SCTP_CRC 221 * - This feature indicates that a device is capable of 222 * offloading the SCTP CRC in a packet. To perform this offload the stack 223 * will set csum_start and csum_offset accordingly, set ip_summed to 224 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication 225 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c. 226 * A driver that supports both IP checksum offload and SCTP CRC32c offload 227 * must verify which offload is configured for a packet by testing the 228 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to 229 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 230 * 231 * * - %NETIF_F_FCOE_CRC 232 * - This feature indicates that a device is capable of offloading the FCOE 233 * CRC in a packet. To perform this offload the stack will set ip_summed 234 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset 235 * accordingly. Note that there is no indication in the skbuff that the 236 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports 237 * both IP checksum offload and FCOE CRC offload must verify which offload 238 * is configured for a packet, presumably by inspecting packet headers. 239 * 240 * Checksumming on output with GSO 241 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 242 * 243 * In the case of a GSO packet (skb_is_gso() is true), checksum offload 244 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 245 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as 246 * part of the GSO operation is implied. If a checksum is being offloaded 247 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and 248 * csum_offset are set to refer to the outermost checksum being offloaded 249 * (two offloaded checksums are possible with UDP encapsulation). 250 */ 251 252 /* Don't change this without changing skb_csum_unnecessary! */ 253 #define CHECKSUM_NONE 0 254 #define CHECKSUM_UNNECESSARY 1 255 #define CHECKSUM_COMPLETE 2 256 #define CHECKSUM_PARTIAL 3 257 258 /* Maximum value in skb->csum_level */ 259 #define SKB_MAX_CSUM_LEVEL 3 260 261 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 262 #define SKB_WITH_OVERHEAD(X) \ 263 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 264 #define SKB_MAX_ORDER(X, ORDER) \ 265 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 266 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 267 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 268 269 /* return minimum truesize of one skb containing X bytes of data */ 270 #define SKB_TRUESIZE(X) ((X) + \ 271 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 272 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 273 274 struct ahash_request; 275 struct net_device; 276 struct scatterlist; 277 struct pipe_inode_info; 278 struct iov_iter; 279 struct napi_struct; 280 struct bpf_prog; 281 union bpf_attr; 282 struct skb_ext; 283 284 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 285 struct nf_bridge_info { 286 enum { 287 BRNF_PROTO_UNCHANGED, 288 BRNF_PROTO_8021Q, 289 BRNF_PROTO_PPPOE 290 } orig_proto:8; 291 u8 pkt_otherhost:1; 292 u8 in_prerouting:1; 293 u8 bridged_dnat:1; 294 __u16 frag_max_size; 295 struct net_device *physindev; 296 297 /* always valid & non-NULL from FORWARD on, for physdev match */ 298 struct net_device *physoutdev; 299 union { 300 /* prerouting: detect dnat in orig/reply direction */ 301 __be32 ipv4_daddr; 302 struct in6_addr ipv6_daddr; 303 304 /* after prerouting + nat detected: store original source 305 * mac since neigh resolution overwrites it, only used while 306 * skb is out in neigh layer. 307 */ 308 char neigh_header[8]; 309 }; 310 }; 311 #endif 312 313 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 314 /* Chain in tc_skb_ext will be used to share the tc chain with 315 * ovs recirc_id. It will be set to the current chain by tc 316 * and read by ovs to recirc_id. 317 */ 318 struct tc_skb_ext { 319 __u32 chain; 320 __u16 mru; 321 __u16 zone; 322 u8 post_ct:1; 323 u8 post_ct_snat:1; 324 u8 post_ct_dnat:1; 325 }; 326 #endif 327 328 struct sk_buff_head { 329 /* These two members must be first to match sk_buff. */ 330 struct_group_tagged(sk_buff_list, list, 331 struct sk_buff *next; 332 struct sk_buff *prev; 333 ); 334 335 __u32 qlen; 336 spinlock_t lock; 337 }; 338 339 struct sk_buff; 340 341 /* To allow 64K frame to be packed as single skb without frag_list we 342 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 343 * buffers which do not start on a page boundary. 344 * 345 * Since GRO uses frags we allocate at least 16 regardless of page 346 * size. 347 */ 348 #if (65536/PAGE_SIZE + 1) < 16 349 #define MAX_SKB_FRAGS 16UL 350 #else 351 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 352 #endif 353 extern int sysctl_max_skb_frags; 354 355 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 356 * segment using its current segmentation instead. 357 */ 358 #define GSO_BY_FRAGS 0xFFFF 359 360 typedef struct bio_vec skb_frag_t; 361 362 /** 363 * skb_frag_size() - Returns the size of a skb fragment 364 * @frag: skb fragment 365 */ 366 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 367 { 368 return frag->bv_len; 369 } 370 371 /** 372 * skb_frag_size_set() - Sets the size of a skb fragment 373 * @frag: skb fragment 374 * @size: size of fragment 375 */ 376 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 377 { 378 frag->bv_len = size; 379 } 380 381 /** 382 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 383 * @frag: skb fragment 384 * @delta: value to add 385 */ 386 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 387 { 388 frag->bv_len += delta; 389 } 390 391 /** 392 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 393 * @frag: skb fragment 394 * @delta: value to subtract 395 */ 396 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 397 { 398 frag->bv_len -= delta; 399 } 400 401 /** 402 * skb_frag_must_loop - Test if %p is a high memory page 403 * @p: fragment's page 404 */ 405 static inline bool skb_frag_must_loop(struct page *p) 406 { 407 #if defined(CONFIG_HIGHMEM) 408 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 409 return true; 410 #endif 411 return false; 412 } 413 414 /** 415 * skb_frag_foreach_page - loop over pages in a fragment 416 * 417 * @f: skb frag to operate on 418 * @f_off: offset from start of f->bv_page 419 * @f_len: length from f_off to loop over 420 * @p: (temp var) current page 421 * @p_off: (temp var) offset from start of current page, 422 * non-zero only on first page. 423 * @p_len: (temp var) length in current page, 424 * < PAGE_SIZE only on first and last page. 425 * @copied: (temp var) length so far, excluding current p_len. 426 * 427 * A fragment can hold a compound page, in which case per-page 428 * operations, notably kmap_atomic, must be called for each 429 * regular page. 430 */ 431 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 432 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 433 p_off = (f_off) & (PAGE_SIZE - 1), \ 434 p_len = skb_frag_must_loop(p) ? \ 435 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 436 copied = 0; \ 437 copied < f_len; \ 438 copied += p_len, p++, p_off = 0, \ 439 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 440 441 #define HAVE_HW_TIME_STAMP 442 443 /** 444 * struct skb_shared_hwtstamps - hardware time stamps 445 * @hwtstamp: hardware time stamp transformed into duration 446 * since arbitrary point in time 447 * @netdev_data: address/cookie of network device driver used as 448 * reference to actual hardware time stamp 449 * 450 * Software time stamps generated by ktime_get_real() are stored in 451 * skb->tstamp. 452 * 453 * hwtstamps can only be compared against other hwtstamps from 454 * the same device. 455 * 456 * This structure is attached to packets as part of the 457 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 458 */ 459 struct skb_shared_hwtstamps { 460 union { 461 ktime_t hwtstamp; 462 void *netdev_data; 463 }; 464 }; 465 466 /* Definitions for tx_flags in struct skb_shared_info */ 467 enum { 468 /* generate hardware time stamp */ 469 SKBTX_HW_TSTAMP = 1 << 0, 470 471 /* generate software time stamp when queueing packet to NIC */ 472 SKBTX_SW_TSTAMP = 1 << 1, 473 474 /* device driver is going to provide hardware time stamp */ 475 SKBTX_IN_PROGRESS = 1 << 2, 476 477 /* generate hardware time stamp based on cycles if supported */ 478 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3, 479 480 /* generate wifi status information (where possible) */ 481 SKBTX_WIFI_STATUS = 1 << 4, 482 483 /* determine hardware time stamp based on time or cycles */ 484 SKBTX_HW_TSTAMP_NETDEV = 1 << 5, 485 486 /* generate software time stamp when entering packet scheduling */ 487 SKBTX_SCHED_TSTAMP = 1 << 6, 488 }; 489 490 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 491 SKBTX_SCHED_TSTAMP) 492 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \ 493 SKBTX_HW_TSTAMP_USE_CYCLES | \ 494 SKBTX_ANY_SW_TSTAMP) 495 496 /* Definitions for flags in struct skb_shared_info */ 497 enum { 498 /* use zcopy routines */ 499 SKBFL_ZEROCOPY_ENABLE = BIT(0), 500 501 /* This indicates at least one fragment might be overwritten 502 * (as in vmsplice(), sendfile() ...) 503 * If we need to compute a TX checksum, we'll need to copy 504 * all frags to avoid possible bad checksum 505 */ 506 SKBFL_SHARED_FRAG = BIT(1), 507 508 /* segment contains only zerocopy data and should not be 509 * charged to the kernel memory. 510 */ 511 SKBFL_PURE_ZEROCOPY = BIT(2), 512 513 SKBFL_DONT_ORPHAN = BIT(3), 514 515 /* page references are managed by the ubuf_info, so it's safe to 516 * use frags only up until ubuf_info is released 517 */ 518 SKBFL_MANAGED_FRAG_REFS = BIT(4), 519 }; 520 521 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG) 522 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \ 523 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS) 524 525 /* 526 * The callback notifies userspace to release buffers when skb DMA is done in 527 * lower device, the skb last reference should be 0 when calling this. 528 * The zerocopy_success argument is true if zero copy transmit occurred, 529 * false on data copy or out of memory error caused by data copy attempt. 530 * The ctx field is used to track device context. 531 * The desc field is used to track userspace buffer index. 532 */ 533 struct ubuf_info { 534 void (*callback)(struct sk_buff *, struct ubuf_info *, 535 bool zerocopy_success); 536 refcount_t refcnt; 537 u8 flags; 538 }; 539 540 struct ubuf_info_msgzc { 541 struct ubuf_info ubuf; 542 543 union { 544 struct { 545 unsigned long desc; 546 void *ctx; 547 }; 548 struct { 549 u32 id; 550 u16 len; 551 u16 zerocopy:1; 552 u32 bytelen; 553 }; 554 }; 555 556 struct mmpin { 557 struct user_struct *user; 558 unsigned int num_pg; 559 } mmp; 560 }; 561 562 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 563 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \ 564 ubuf) 565 566 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 567 void mm_unaccount_pinned_pages(struct mmpin *mmp); 568 569 /* This data is invariant across clones and lives at 570 * the end of the header data, ie. at skb->end. 571 */ 572 struct skb_shared_info { 573 __u8 flags; 574 __u8 meta_len; 575 __u8 nr_frags; 576 __u8 tx_flags; 577 unsigned short gso_size; 578 /* Warning: this field is not always filled in (UFO)! */ 579 unsigned short gso_segs; 580 struct sk_buff *frag_list; 581 struct skb_shared_hwtstamps hwtstamps; 582 unsigned int gso_type; 583 u32 tskey; 584 585 /* 586 * Warning : all fields before dataref are cleared in __alloc_skb() 587 */ 588 atomic_t dataref; 589 unsigned int xdp_frags_size; 590 591 /* Intermediate layers must ensure that destructor_arg 592 * remains valid until skb destructor */ 593 void * destructor_arg; 594 595 /* must be last field, see pskb_expand_head() */ 596 skb_frag_t frags[MAX_SKB_FRAGS]; 597 }; 598 599 /** 600 * DOC: dataref and headerless skbs 601 * 602 * Transport layers send out clones of payload skbs they hold for 603 * retransmissions. To allow lower layers of the stack to prepend their headers 604 * we split &skb_shared_info.dataref into two halves. 605 * The lower 16 bits count the overall number of references. 606 * The higher 16 bits indicate how many of the references are payload-only. 607 * skb_header_cloned() checks if skb is allowed to add / write the headers. 608 * 609 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr 610 * (via __skb_header_release()). Any clone created from marked skb will get 611 * &sk_buff.hdr_len populated with the available headroom. 612 * If there's the only clone in existence it's able to modify the headroom 613 * at will. The sequence of calls inside the transport layer is:: 614 * 615 * <alloc skb> 616 * skb_reserve() 617 * __skb_header_release() 618 * skb_clone() 619 * // send the clone down the stack 620 * 621 * This is not a very generic construct and it depends on the transport layers 622 * doing the right thing. In practice there's usually only one payload-only skb. 623 * Having multiple payload-only skbs with different lengths of hdr_len is not 624 * possible. The payload-only skbs should never leave their owner. 625 */ 626 #define SKB_DATAREF_SHIFT 16 627 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 628 629 630 enum { 631 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 632 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 633 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 634 }; 635 636 enum { 637 SKB_GSO_TCPV4 = 1 << 0, 638 639 /* This indicates the skb is from an untrusted source. */ 640 SKB_GSO_DODGY = 1 << 1, 641 642 /* This indicates the tcp segment has CWR set. */ 643 SKB_GSO_TCP_ECN = 1 << 2, 644 645 SKB_GSO_TCP_FIXEDID = 1 << 3, 646 647 SKB_GSO_TCPV6 = 1 << 4, 648 649 SKB_GSO_FCOE = 1 << 5, 650 651 SKB_GSO_GRE = 1 << 6, 652 653 SKB_GSO_GRE_CSUM = 1 << 7, 654 655 SKB_GSO_IPXIP4 = 1 << 8, 656 657 SKB_GSO_IPXIP6 = 1 << 9, 658 659 SKB_GSO_UDP_TUNNEL = 1 << 10, 660 661 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 662 663 SKB_GSO_PARTIAL = 1 << 12, 664 665 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 666 667 SKB_GSO_SCTP = 1 << 14, 668 669 SKB_GSO_ESP = 1 << 15, 670 671 SKB_GSO_UDP = 1 << 16, 672 673 SKB_GSO_UDP_L4 = 1 << 17, 674 675 SKB_GSO_FRAGLIST = 1 << 18, 676 }; 677 678 #if BITS_PER_LONG > 32 679 #define NET_SKBUFF_DATA_USES_OFFSET 1 680 #endif 681 682 #ifdef NET_SKBUFF_DATA_USES_OFFSET 683 typedef unsigned int sk_buff_data_t; 684 #else 685 typedef unsigned char *sk_buff_data_t; 686 #endif 687 688 /** 689 * DOC: Basic sk_buff geometry 690 * 691 * struct sk_buff itself is a metadata structure and does not hold any packet 692 * data. All the data is held in associated buffers. 693 * 694 * &sk_buff.head points to the main "head" buffer. The head buffer is divided 695 * into two parts: 696 * 697 * - data buffer, containing headers and sometimes payload; 698 * this is the part of the skb operated on by the common helpers 699 * such as skb_put() or skb_pull(); 700 * - shared info (struct skb_shared_info) which holds an array of pointers 701 * to read-only data in the (page, offset, length) format. 702 * 703 * Optionally &skb_shared_info.frag_list may point to another skb. 704 * 705 * Basic diagram may look like this:: 706 * 707 * --------------- 708 * | sk_buff | 709 * --------------- 710 * ,--------------------------- + head 711 * / ,----------------- + data 712 * / / ,----------- + tail 713 * | | | , + end 714 * | | | | 715 * v v v v 716 * ----------------------------------------------- 717 * | headroom | data | tailroom | skb_shared_info | 718 * ----------------------------------------------- 719 * + [page frag] 720 * + [page frag] 721 * + [page frag] 722 * + [page frag] --------- 723 * + frag_list --> | sk_buff | 724 * --------- 725 * 726 */ 727 728 /** 729 * struct sk_buff - socket buffer 730 * @next: Next buffer in list 731 * @prev: Previous buffer in list 732 * @tstamp: Time we arrived/left 733 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 734 * for retransmit timer 735 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 736 * @list: queue head 737 * @ll_node: anchor in an llist (eg socket defer_list) 738 * @sk: Socket we are owned by 739 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in 740 * fragmentation management 741 * @dev: Device we arrived on/are leaving by 742 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 743 * @cb: Control buffer. Free for use by every layer. Put private vars here 744 * @_skb_refdst: destination entry (with norefcount bit) 745 * @sp: the security path, used for xfrm 746 * @len: Length of actual data 747 * @data_len: Data length 748 * @mac_len: Length of link layer header 749 * @hdr_len: writable header length of cloned skb 750 * @csum: Checksum (must include start/offset pair) 751 * @csum_start: Offset from skb->head where checksumming should start 752 * @csum_offset: Offset from csum_start where checksum should be stored 753 * @priority: Packet queueing priority 754 * @ignore_df: allow local fragmentation 755 * @cloned: Head may be cloned (check refcnt to be sure) 756 * @ip_summed: Driver fed us an IP checksum 757 * @nohdr: Payload reference only, must not modify header 758 * @pkt_type: Packet class 759 * @fclone: skbuff clone status 760 * @ipvs_property: skbuff is owned by ipvs 761 * @inner_protocol_type: whether the inner protocol is 762 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 763 * @remcsum_offload: remote checksum offload is enabled 764 * @offload_fwd_mark: Packet was L2-forwarded in hardware 765 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 766 * @tc_skip_classify: do not classify packet. set by IFB device 767 * @tc_at_ingress: used within tc_classify to distinguish in/egress 768 * @redirected: packet was redirected by packet classifier 769 * @from_ingress: packet was redirected from the ingress path 770 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 771 * @peeked: this packet has been seen already, so stats have been 772 * done for it, don't do them again 773 * @nf_trace: netfilter packet trace flag 774 * @protocol: Packet protocol from driver 775 * @destructor: Destruct function 776 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 777 * @_sk_redir: socket redirection information for skmsg 778 * @_nfct: Associated connection, if any (with nfctinfo bits) 779 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 780 * @skb_iif: ifindex of device we arrived on 781 * @tc_index: Traffic control index 782 * @hash: the packet hash 783 * @queue_mapping: Queue mapping for multiqueue devices 784 * @head_frag: skb was allocated from page fragments, 785 * not allocated by kmalloc() or vmalloc(). 786 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 787 * @pp_recycle: mark the packet for recycling instead of freeing (implies 788 * page_pool support on driver) 789 * @active_extensions: active extensions (skb_ext_id types) 790 * @ndisc_nodetype: router type (from link layer) 791 * @ooo_okay: allow the mapping of a socket to a queue to be changed 792 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 793 * ports. 794 * @sw_hash: indicates hash was computed in software stack 795 * @wifi_acked_valid: wifi_acked was set 796 * @wifi_acked: whether frame was acked on wifi or not 797 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 798 * @encapsulation: indicates the inner headers in the skbuff are valid 799 * @encap_hdr_csum: software checksum is needed 800 * @csum_valid: checksum is already valid 801 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 802 * @csum_complete_sw: checksum was completed by software 803 * @csum_level: indicates the number of consecutive checksums found in 804 * the packet minus one that have been verified as 805 * CHECKSUM_UNNECESSARY (max 3) 806 * @scm_io_uring: SKB holds io_uring registered files 807 * @dst_pending_confirm: need to confirm neighbour 808 * @decrypted: Decrypted SKB 809 * @slow_gro: state present at GRO time, slower prepare step required 810 * @mono_delivery_time: When set, skb->tstamp has the 811 * delivery_time in mono clock base (i.e. EDT). Otherwise, the 812 * skb->tstamp has the (rcv) timestamp at ingress and 813 * delivery_time at egress. 814 * @napi_id: id of the NAPI struct this skb came from 815 * @sender_cpu: (aka @napi_id) source CPU in XPS 816 * @alloc_cpu: CPU which did the skb allocation. 817 * @secmark: security marking 818 * @mark: Generic packet mark 819 * @reserved_tailroom: (aka @mark) number of bytes of free space available 820 * at the tail of an sk_buff 821 * @vlan_all: vlan fields (proto & tci) 822 * @vlan_proto: vlan encapsulation protocol 823 * @vlan_tci: vlan tag control information 824 * @inner_protocol: Protocol (encapsulation) 825 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 826 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 827 * @inner_transport_header: Inner transport layer header (encapsulation) 828 * @inner_network_header: Network layer header (encapsulation) 829 * @inner_mac_header: Link layer header (encapsulation) 830 * @transport_header: Transport layer header 831 * @network_header: Network layer header 832 * @mac_header: Link layer header 833 * @kcov_handle: KCOV remote handle for remote coverage collection 834 * @tail: Tail pointer 835 * @end: End pointer 836 * @head: Head of buffer 837 * @data: Data head pointer 838 * @truesize: Buffer size 839 * @users: User count - see {datagram,tcp}.c 840 * @extensions: allocated extensions, valid if active_extensions is nonzero 841 */ 842 843 struct sk_buff { 844 union { 845 struct { 846 /* These two members must be first to match sk_buff_head. */ 847 struct sk_buff *next; 848 struct sk_buff *prev; 849 850 union { 851 struct net_device *dev; 852 /* Some protocols might use this space to store information, 853 * while device pointer would be NULL. 854 * UDP receive path is one user. 855 */ 856 unsigned long dev_scratch; 857 }; 858 }; 859 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 860 struct list_head list; 861 struct llist_node ll_node; 862 }; 863 864 union { 865 struct sock *sk; 866 int ip_defrag_offset; 867 }; 868 869 union { 870 ktime_t tstamp; 871 u64 skb_mstamp_ns; /* earliest departure time */ 872 }; 873 /* 874 * This is the control buffer. It is free to use for every 875 * layer. Please put your private variables there. If you 876 * want to keep them across layers you have to do a skb_clone() 877 * first. This is owned by whoever has the skb queued ATM. 878 */ 879 char cb[48] __aligned(8); 880 881 union { 882 struct { 883 unsigned long _skb_refdst; 884 void (*destructor)(struct sk_buff *skb); 885 }; 886 struct list_head tcp_tsorted_anchor; 887 #ifdef CONFIG_NET_SOCK_MSG 888 unsigned long _sk_redir; 889 #endif 890 }; 891 892 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 893 unsigned long _nfct; 894 #endif 895 unsigned int len, 896 data_len; 897 __u16 mac_len, 898 hdr_len; 899 900 /* Following fields are _not_ copied in __copy_skb_header() 901 * Note that queue_mapping is here mostly to fill a hole. 902 */ 903 __u16 queue_mapping; 904 905 /* if you move cloned around you also must adapt those constants */ 906 #ifdef __BIG_ENDIAN_BITFIELD 907 #define CLONED_MASK (1 << 7) 908 #else 909 #define CLONED_MASK 1 910 #endif 911 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 912 913 /* private: */ 914 __u8 __cloned_offset[0]; 915 /* public: */ 916 __u8 cloned:1, 917 nohdr:1, 918 fclone:2, 919 peeked:1, 920 head_frag:1, 921 pfmemalloc:1, 922 pp_recycle:1; /* page_pool recycle indicator */ 923 #ifdef CONFIG_SKB_EXTENSIONS 924 __u8 active_extensions; 925 #endif 926 927 /* Fields enclosed in headers group are copied 928 * using a single memcpy() in __copy_skb_header() 929 */ 930 struct_group(headers, 931 932 /* private: */ 933 __u8 __pkt_type_offset[0]; 934 /* public: */ 935 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 936 __u8 ignore_df:1; 937 __u8 nf_trace:1; 938 __u8 ip_summed:2; 939 __u8 ooo_okay:1; 940 941 __u8 l4_hash:1; 942 __u8 sw_hash:1; 943 __u8 wifi_acked_valid:1; 944 __u8 wifi_acked:1; 945 __u8 no_fcs:1; 946 /* Indicates the inner headers are valid in the skbuff. */ 947 __u8 encapsulation:1; 948 __u8 encap_hdr_csum:1; 949 __u8 csum_valid:1; 950 951 /* private: */ 952 __u8 __pkt_vlan_present_offset[0]; 953 /* public: */ 954 __u8 remcsum_offload:1; 955 __u8 csum_complete_sw:1; 956 __u8 csum_level:2; 957 __u8 dst_pending_confirm:1; 958 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */ 959 #ifdef CONFIG_NET_CLS_ACT 960 __u8 tc_skip_classify:1; 961 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 962 #endif 963 #ifdef CONFIG_IPV6_NDISC_NODETYPE 964 __u8 ndisc_nodetype:2; 965 #endif 966 967 __u8 ipvs_property:1; 968 __u8 inner_protocol_type:1; 969 #ifdef CONFIG_NET_SWITCHDEV 970 __u8 offload_fwd_mark:1; 971 __u8 offload_l3_fwd_mark:1; 972 #endif 973 __u8 redirected:1; 974 #ifdef CONFIG_NET_REDIRECT 975 __u8 from_ingress:1; 976 #endif 977 #ifdef CONFIG_NETFILTER_SKIP_EGRESS 978 __u8 nf_skip_egress:1; 979 #endif 980 #ifdef CONFIG_TLS_DEVICE 981 __u8 decrypted:1; 982 #endif 983 __u8 slow_gro:1; 984 __u8 csum_not_inet:1; 985 __u8 scm_io_uring:1; 986 987 #ifdef CONFIG_NET_SCHED 988 __u16 tc_index; /* traffic control index */ 989 #endif 990 991 union { 992 __wsum csum; 993 struct { 994 __u16 csum_start; 995 __u16 csum_offset; 996 }; 997 }; 998 __u32 priority; 999 int skb_iif; 1000 __u32 hash; 1001 union { 1002 u32 vlan_all; 1003 struct { 1004 __be16 vlan_proto; 1005 __u16 vlan_tci; 1006 }; 1007 }; 1008 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1009 union { 1010 unsigned int napi_id; 1011 unsigned int sender_cpu; 1012 }; 1013 #endif 1014 u16 alloc_cpu; 1015 #ifdef CONFIG_NETWORK_SECMARK 1016 __u32 secmark; 1017 #endif 1018 1019 union { 1020 __u32 mark; 1021 __u32 reserved_tailroom; 1022 }; 1023 1024 union { 1025 __be16 inner_protocol; 1026 __u8 inner_ipproto; 1027 }; 1028 1029 __u16 inner_transport_header; 1030 __u16 inner_network_header; 1031 __u16 inner_mac_header; 1032 1033 __be16 protocol; 1034 __u16 transport_header; 1035 __u16 network_header; 1036 __u16 mac_header; 1037 1038 #ifdef CONFIG_KCOV 1039 u64 kcov_handle; 1040 #endif 1041 1042 ); /* end headers group */ 1043 1044 /* These elements must be at the end, see alloc_skb() for details. */ 1045 sk_buff_data_t tail; 1046 sk_buff_data_t end; 1047 unsigned char *head, 1048 *data; 1049 unsigned int truesize; 1050 refcount_t users; 1051 1052 #ifdef CONFIG_SKB_EXTENSIONS 1053 /* only useable after checking ->active_extensions != 0 */ 1054 struct skb_ext *extensions; 1055 #endif 1056 }; 1057 1058 /* if you move pkt_type around you also must adapt those constants */ 1059 #ifdef __BIG_ENDIAN_BITFIELD 1060 #define PKT_TYPE_MAX (7 << 5) 1061 #else 1062 #define PKT_TYPE_MAX 7 1063 #endif 1064 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1065 1066 /* if you move tc_at_ingress or mono_delivery_time 1067 * around, you also must adapt these constants. 1068 */ 1069 #ifdef __BIG_ENDIAN_BITFIELD 1070 #define TC_AT_INGRESS_MASK (1 << 0) 1071 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 2) 1072 #else 1073 #define TC_AT_INGRESS_MASK (1 << 7) 1074 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 5) 1075 #endif 1076 #define PKT_VLAN_PRESENT_OFFSET offsetof(struct sk_buff, __pkt_vlan_present_offset) 1077 1078 #ifdef __KERNEL__ 1079 /* 1080 * Handling routines are only of interest to the kernel 1081 */ 1082 1083 #define SKB_ALLOC_FCLONE 0x01 1084 #define SKB_ALLOC_RX 0x02 1085 #define SKB_ALLOC_NAPI 0x04 1086 1087 /** 1088 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1089 * @skb: buffer 1090 */ 1091 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1092 { 1093 return unlikely(skb->pfmemalloc); 1094 } 1095 1096 /* 1097 * skb might have a dst pointer attached, refcounted or not. 1098 * _skb_refdst low order bit is set if refcount was _not_ taken 1099 */ 1100 #define SKB_DST_NOREF 1UL 1101 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1102 1103 /** 1104 * skb_dst - returns skb dst_entry 1105 * @skb: buffer 1106 * 1107 * Returns skb dst_entry, regardless of reference taken or not. 1108 */ 1109 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1110 { 1111 /* If refdst was not refcounted, check we still are in a 1112 * rcu_read_lock section 1113 */ 1114 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1115 !rcu_read_lock_held() && 1116 !rcu_read_lock_bh_held()); 1117 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1118 } 1119 1120 /** 1121 * skb_dst_set - sets skb dst 1122 * @skb: buffer 1123 * @dst: dst entry 1124 * 1125 * Sets skb dst, assuming a reference was taken on dst and should 1126 * be released by skb_dst_drop() 1127 */ 1128 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1129 { 1130 skb->slow_gro |= !!dst; 1131 skb->_skb_refdst = (unsigned long)dst; 1132 } 1133 1134 /** 1135 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1136 * @skb: buffer 1137 * @dst: dst entry 1138 * 1139 * Sets skb dst, assuming a reference was not taken on dst. 1140 * If dst entry is cached, we do not take reference and dst_release 1141 * will be avoided by refdst_drop. If dst entry is not cached, we take 1142 * reference, so that last dst_release can destroy the dst immediately. 1143 */ 1144 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1145 { 1146 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1147 skb->slow_gro |= !!dst; 1148 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1149 } 1150 1151 /** 1152 * skb_dst_is_noref - Test if skb dst isn't refcounted 1153 * @skb: buffer 1154 */ 1155 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1156 { 1157 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1158 } 1159 1160 /** 1161 * skb_rtable - Returns the skb &rtable 1162 * @skb: buffer 1163 */ 1164 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 1165 { 1166 return (struct rtable *)skb_dst(skb); 1167 } 1168 1169 /* For mangling skb->pkt_type from user space side from applications 1170 * such as nft, tc, etc, we only allow a conservative subset of 1171 * possible pkt_types to be set. 1172 */ 1173 static inline bool skb_pkt_type_ok(u32 ptype) 1174 { 1175 return ptype <= PACKET_OTHERHOST; 1176 } 1177 1178 /** 1179 * skb_napi_id - Returns the skb's NAPI id 1180 * @skb: buffer 1181 */ 1182 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1183 { 1184 #ifdef CONFIG_NET_RX_BUSY_POLL 1185 return skb->napi_id; 1186 #else 1187 return 0; 1188 #endif 1189 } 1190 1191 /** 1192 * skb_unref - decrement the skb's reference count 1193 * @skb: buffer 1194 * 1195 * Returns true if we can free the skb. 1196 */ 1197 static inline bool skb_unref(struct sk_buff *skb) 1198 { 1199 if (unlikely(!skb)) 1200 return false; 1201 if (likely(refcount_read(&skb->users) == 1)) 1202 smp_rmb(); 1203 else if (likely(!refcount_dec_and_test(&skb->users))) 1204 return false; 1205 1206 return true; 1207 } 1208 1209 void __fix_address 1210 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason); 1211 1212 /** 1213 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1214 * @skb: buffer to free 1215 */ 1216 static inline void kfree_skb(struct sk_buff *skb) 1217 { 1218 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1219 } 1220 1221 void skb_release_head_state(struct sk_buff *skb); 1222 void kfree_skb_list_reason(struct sk_buff *segs, 1223 enum skb_drop_reason reason); 1224 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1225 void skb_tx_error(struct sk_buff *skb); 1226 1227 static inline void kfree_skb_list(struct sk_buff *segs) 1228 { 1229 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1230 } 1231 1232 #ifdef CONFIG_TRACEPOINTS 1233 void consume_skb(struct sk_buff *skb); 1234 #else 1235 static inline void consume_skb(struct sk_buff *skb) 1236 { 1237 return kfree_skb(skb); 1238 } 1239 #endif 1240 1241 void __consume_stateless_skb(struct sk_buff *skb); 1242 void __kfree_skb(struct sk_buff *skb); 1243 extern struct kmem_cache *skbuff_head_cache; 1244 1245 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1246 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1247 bool *fragstolen, int *delta_truesize); 1248 1249 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1250 int node); 1251 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1252 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1253 struct sk_buff *build_skb_around(struct sk_buff *skb, 1254 void *data, unsigned int frag_size); 1255 void skb_attempt_defer_free(struct sk_buff *skb); 1256 1257 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1258 1259 /** 1260 * alloc_skb - allocate a network buffer 1261 * @size: size to allocate 1262 * @priority: allocation mask 1263 * 1264 * This function is a convenient wrapper around __alloc_skb(). 1265 */ 1266 static inline struct sk_buff *alloc_skb(unsigned int size, 1267 gfp_t priority) 1268 { 1269 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1270 } 1271 1272 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1273 unsigned long data_len, 1274 int max_page_order, 1275 int *errcode, 1276 gfp_t gfp_mask); 1277 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1278 1279 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1280 struct sk_buff_fclones { 1281 struct sk_buff skb1; 1282 1283 struct sk_buff skb2; 1284 1285 refcount_t fclone_ref; 1286 }; 1287 1288 /** 1289 * skb_fclone_busy - check if fclone is busy 1290 * @sk: socket 1291 * @skb: buffer 1292 * 1293 * Returns true if skb is a fast clone, and its clone is not freed. 1294 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1295 * so we also check that this didnt happen. 1296 */ 1297 static inline bool skb_fclone_busy(const struct sock *sk, 1298 const struct sk_buff *skb) 1299 { 1300 const struct sk_buff_fclones *fclones; 1301 1302 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1303 1304 return skb->fclone == SKB_FCLONE_ORIG && 1305 refcount_read(&fclones->fclone_ref) > 1 && 1306 READ_ONCE(fclones->skb2.sk) == sk; 1307 } 1308 1309 /** 1310 * alloc_skb_fclone - allocate a network buffer from fclone cache 1311 * @size: size to allocate 1312 * @priority: allocation mask 1313 * 1314 * This function is a convenient wrapper around __alloc_skb(). 1315 */ 1316 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1317 gfp_t priority) 1318 { 1319 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1320 } 1321 1322 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1323 void skb_headers_offset_update(struct sk_buff *skb, int off); 1324 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1325 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1326 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1327 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1328 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1329 gfp_t gfp_mask, bool fclone); 1330 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1331 gfp_t gfp_mask) 1332 { 1333 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1334 } 1335 1336 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1337 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1338 unsigned int headroom); 1339 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1340 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1341 int newtailroom, gfp_t priority); 1342 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1343 int offset, int len); 1344 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1345 int offset, int len); 1346 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1347 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1348 1349 /** 1350 * skb_pad - zero pad the tail of an skb 1351 * @skb: buffer to pad 1352 * @pad: space to pad 1353 * 1354 * Ensure that a buffer is followed by a padding area that is zero 1355 * filled. Used by network drivers which may DMA or transfer data 1356 * beyond the buffer end onto the wire. 1357 * 1358 * May return error in out of memory cases. The skb is freed on error. 1359 */ 1360 static inline int skb_pad(struct sk_buff *skb, int pad) 1361 { 1362 return __skb_pad(skb, pad, true); 1363 } 1364 #define dev_kfree_skb(a) consume_skb(a) 1365 1366 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1367 int offset, size_t size); 1368 1369 struct skb_seq_state { 1370 __u32 lower_offset; 1371 __u32 upper_offset; 1372 __u32 frag_idx; 1373 __u32 stepped_offset; 1374 struct sk_buff *root_skb; 1375 struct sk_buff *cur_skb; 1376 __u8 *frag_data; 1377 __u32 frag_off; 1378 }; 1379 1380 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1381 unsigned int to, struct skb_seq_state *st); 1382 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1383 struct skb_seq_state *st); 1384 void skb_abort_seq_read(struct skb_seq_state *st); 1385 1386 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1387 unsigned int to, struct ts_config *config); 1388 1389 /* 1390 * Packet hash types specify the type of hash in skb_set_hash. 1391 * 1392 * Hash types refer to the protocol layer addresses which are used to 1393 * construct a packet's hash. The hashes are used to differentiate or identify 1394 * flows of the protocol layer for the hash type. Hash types are either 1395 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1396 * 1397 * Properties of hashes: 1398 * 1399 * 1) Two packets in different flows have different hash values 1400 * 2) Two packets in the same flow should have the same hash value 1401 * 1402 * A hash at a higher layer is considered to be more specific. A driver should 1403 * set the most specific hash possible. 1404 * 1405 * A driver cannot indicate a more specific hash than the layer at which a hash 1406 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1407 * 1408 * A driver may indicate a hash level which is less specific than the 1409 * actual layer the hash was computed on. For instance, a hash computed 1410 * at L4 may be considered an L3 hash. This should only be done if the 1411 * driver can't unambiguously determine that the HW computed the hash at 1412 * the higher layer. Note that the "should" in the second property above 1413 * permits this. 1414 */ 1415 enum pkt_hash_types { 1416 PKT_HASH_TYPE_NONE, /* Undefined type */ 1417 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1418 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1419 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1420 }; 1421 1422 static inline void skb_clear_hash(struct sk_buff *skb) 1423 { 1424 skb->hash = 0; 1425 skb->sw_hash = 0; 1426 skb->l4_hash = 0; 1427 } 1428 1429 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1430 { 1431 if (!skb->l4_hash) 1432 skb_clear_hash(skb); 1433 } 1434 1435 static inline void 1436 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1437 { 1438 skb->l4_hash = is_l4; 1439 skb->sw_hash = is_sw; 1440 skb->hash = hash; 1441 } 1442 1443 static inline void 1444 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1445 { 1446 /* Used by drivers to set hash from HW */ 1447 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1448 } 1449 1450 static inline void 1451 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1452 { 1453 __skb_set_hash(skb, hash, true, is_l4); 1454 } 1455 1456 void __skb_get_hash(struct sk_buff *skb); 1457 u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1458 u32 skb_get_poff(const struct sk_buff *skb); 1459 u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1460 const struct flow_keys_basic *keys, int hlen); 1461 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1462 const void *data, int hlen_proto); 1463 1464 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1465 int thoff, u8 ip_proto) 1466 { 1467 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1468 } 1469 1470 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1471 const struct flow_dissector_key *key, 1472 unsigned int key_count); 1473 1474 struct bpf_flow_dissector; 1475 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1476 __be16 proto, int nhoff, int hlen, unsigned int flags); 1477 1478 bool __skb_flow_dissect(const struct net *net, 1479 const struct sk_buff *skb, 1480 struct flow_dissector *flow_dissector, 1481 void *target_container, const void *data, 1482 __be16 proto, int nhoff, int hlen, unsigned int flags); 1483 1484 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1485 struct flow_dissector *flow_dissector, 1486 void *target_container, unsigned int flags) 1487 { 1488 return __skb_flow_dissect(NULL, skb, flow_dissector, 1489 target_container, NULL, 0, 0, 0, flags); 1490 } 1491 1492 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1493 struct flow_keys *flow, 1494 unsigned int flags) 1495 { 1496 memset(flow, 0, sizeof(*flow)); 1497 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1498 flow, NULL, 0, 0, 0, flags); 1499 } 1500 1501 static inline bool 1502 skb_flow_dissect_flow_keys_basic(const struct net *net, 1503 const struct sk_buff *skb, 1504 struct flow_keys_basic *flow, 1505 const void *data, __be16 proto, 1506 int nhoff, int hlen, unsigned int flags) 1507 { 1508 memset(flow, 0, sizeof(*flow)); 1509 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1510 data, proto, nhoff, hlen, flags); 1511 } 1512 1513 void skb_flow_dissect_meta(const struct sk_buff *skb, 1514 struct flow_dissector *flow_dissector, 1515 void *target_container); 1516 1517 /* Gets a skb connection tracking info, ctinfo map should be a 1518 * map of mapsize to translate enum ip_conntrack_info states 1519 * to user states. 1520 */ 1521 void 1522 skb_flow_dissect_ct(const struct sk_buff *skb, 1523 struct flow_dissector *flow_dissector, 1524 void *target_container, 1525 u16 *ctinfo_map, size_t mapsize, 1526 bool post_ct, u16 zone); 1527 void 1528 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1529 struct flow_dissector *flow_dissector, 1530 void *target_container); 1531 1532 void skb_flow_dissect_hash(const struct sk_buff *skb, 1533 struct flow_dissector *flow_dissector, 1534 void *target_container); 1535 1536 static inline __u32 skb_get_hash(struct sk_buff *skb) 1537 { 1538 if (!skb->l4_hash && !skb->sw_hash) 1539 __skb_get_hash(skb); 1540 1541 return skb->hash; 1542 } 1543 1544 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1545 { 1546 if (!skb->l4_hash && !skb->sw_hash) { 1547 struct flow_keys keys; 1548 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1549 1550 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1551 } 1552 1553 return skb->hash; 1554 } 1555 1556 __u32 skb_get_hash_perturb(const struct sk_buff *skb, 1557 const siphash_key_t *perturb); 1558 1559 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1560 { 1561 return skb->hash; 1562 } 1563 1564 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1565 { 1566 to->hash = from->hash; 1567 to->sw_hash = from->sw_hash; 1568 to->l4_hash = from->l4_hash; 1569 }; 1570 1571 static inline void skb_copy_decrypted(struct sk_buff *to, 1572 const struct sk_buff *from) 1573 { 1574 #ifdef CONFIG_TLS_DEVICE 1575 to->decrypted = from->decrypted; 1576 #endif 1577 } 1578 1579 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1580 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1581 { 1582 return skb->head + skb->end; 1583 } 1584 1585 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1586 { 1587 return skb->end; 1588 } 1589 1590 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1591 { 1592 skb->end = offset; 1593 } 1594 #else 1595 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1596 { 1597 return skb->end; 1598 } 1599 1600 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1601 { 1602 return skb->end - skb->head; 1603 } 1604 1605 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1606 { 1607 skb->end = skb->head + offset; 1608 } 1609 #endif 1610 1611 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1612 struct ubuf_info *uarg); 1613 1614 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1615 1616 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg, 1617 bool success); 1618 1619 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, 1620 struct sk_buff *skb, struct iov_iter *from, 1621 size_t length); 1622 1623 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1624 struct msghdr *msg, int len) 1625 { 1626 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len); 1627 } 1628 1629 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1630 struct msghdr *msg, int len, 1631 struct ubuf_info *uarg); 1632 1633 /* Internal */ 1634 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1635 1636 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1637 { 1638 return &skb_shinfo(skb)->hwtstamps; 1639 } 1640 1641 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1642 { 1643 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1644 1645 return is_zcopy ? skb_uarg(skb) : NULL; 1646 } 1647 1648 static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1649 { 1650 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1651 } 1652 1653 static inline bool skb_zcopy_managed(const struct sk_buff *skb) 1654 { 1655 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS; 1656 } 1657 1658 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1659 const struct sk_buff *skb2) 1660 { 1661 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1662 } 1663 1664 static inline void net_zcopy_get(struct ubuf_info *uarg) 1665 { 1666 refcount_inc(&uarg->refcnt); 1667 } 1668 1669 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1670 { 1671 skb_shinfo(skb)->destructor_arg = uarg; 1672 skb_shinfo(skb)->flags |= uarg->flags; 1673 } 1674 1675 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1676 bool *have_ref) 1677 { 1678 if (skb && uarg && !skb_zcopy(skb)) { 1679 if (unlikely(have_ref && *have_ref)) 1680 *have_ref = false; 1681 else 1682 net_zcopy_get(uarg); 1683 skb_zcopy_init(skb, uarg); 1684 } 1685 } 1686 1687 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1688 { 1689 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1690 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1691 } 1692 1693 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1694 { 1695 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1696 } 1697 1698 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1699 { 1700 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1701 } 1702 1703 static inline void net_zcopy_put(struct ubuf_info *uarg) 1704 { 1705 if (uarg) 1706 uarg->callback(NULL, uarg, true); 1707 } 1708 1709 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1710 { 1711 if (uarg) { 1712 if (uarg->callback == msg_zerocopy_callback) 1713 msg_zerocopy_put_abort(uarg, have_uref); 1714 else if (have_uref) 1715 net_zcopy_put(uarg); 1716 } 1717 } 1718 1719 /* Release a reference on a zerocopy structure */ 1720 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1721 { 1722 struct ubuf_info *uarg = skb_zcopy(skb); 1723 1724 if (uarg) { 1725 if (!skb_zcopy_is_nouarg(skb)) 1726 uarg->callback(skb, uarg, zerocopy_success); 1727 1728 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1729 } 1730 } 1731 1732 void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1733 1734 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1735 { 1736 if (unlikely(skb_zcopy_managed(skb))) 1737 __skb_zcopy_downgrade_managed(skb); 1738 } 1739 1740 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1741 { 1742 skb->next = NULL; 1743 } 1744 1745 /* Iterate through singly-linked GSO fragments of an skb. */ 1746 #define skb_list_walk_safe(first, skb, next_skb) \ 1747 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1748 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1749 1750 static inline void skb_list_del_init(struct sk_buff *skb) 1751 { 1752 __list_del_entry(&skb->list); 1753 skb_mark_not_on_list(skb); 1754 } 1755 1756 /** 1757 * skb_queue_empty - check if a queue is empty 1758 * @list: queue head 1759 * 1760 * Returns true if the queue is empty, false otherwise. 1761 */ 1762 static inline int skb_queue_empty(const struct sk_buff_head *list) 1763 { 1764 return list->next == (const struct sk_buff *) list; 1765 } 1766 1767 /** 1768 * skb_queue_empty_lockless - check if a queue is empty 1769 * @list: queue head 1770 * 1771 * Returns true if the queue is empty, false otherwise. 1772 * This variant can be used in lockless contexts. 1773 */ 1774 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1775 { 1776 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1777 } 1778 1779 1780 /** 1781 * skb_queue_is_last - check if skb is the last entry in the queue 1782 * @list: queue head 1783 * @skb: buffer 1784 * 1785 * Returns true if @skb is the last buffer on the list. 1786 */ 1787 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1788 const struct sk_buff *skb) 1789 { 1790 return skb->next == (const struct sk_buff *) list; 1791 } 1792 1793 /** 1794 * skb_queue_is_first - check if skb is the first entry in the queue 1795 * @list: queue head 1796 * @skb: buffer 1797 * 1798 * Returns true if @skb is the first buffer on the list. 1799 */ 1800 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1801 const struct sk_buff *skb) 1802 { 1803 return skb->prev == (const struct sk_buff *) list; 1804 } 1805 1806 /** 1807 * skb_queue_next - return the next packet in the queue 1808 * @list: queue head 1809 * @skb: current buffer 1810 * 1811 * Return the next packet in @list after @skb. It is only valid to 1812 * call this if skb_queue_is_last() evaluates to false. 1813 */ 1814 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1815 const struct sk_buff *skb) 1816 { 1817 /* This BUG_ON may seem severe, but if we just return then we 1818 * are going to dereference garbage. 1819 */ 1820 BUG_ON(skb_queue_is_last(list, skb)); 1821 return skb->next; 1822 } 1823 1824 /** 1825 * skb_queue_prev - return the prev packet in the queue 1826 * @list: queue head 1827 * @skb: current buffer 1828 * 1829 * Return the prev packet in @list before @skb. It is only valid to 1830 * call this if skb_queue_is_first() evaluates to false. 1831 */ 1832 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1833 const struct sk_buff *skb) 1834 { 1835 /* This BUG_ON may seem severe, but if we just return then we 1836 * are going to dereference garbage. 1837 */ 1838 BUG_ON(skb_queue_is_first(list, skb)); 1839 return skb->prev; 1840 } 1841 1842 /** 1843 * skb_get - reference buffer 1844 * @skb: buffer to reference 1845 * 1846 * Makes another reference to a socket buffer and returns a pointer 1847 * to the buffer. 1848 */ 1849 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1850 { 1851 refcount_inc(&skb->users); 1852 return skb; 1853 } 1854 1855 /* 1856 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1857 */ 1858 1859 /** 1860 * skb_cloned - is the buffer a clone 1861 * @skb: buffer to check 1862 * 1863 * Returns true if the buffer was generated with skb_clone() and is 1864 * one of multiple shared copies of the buffer. Cloned buffers are 1865 * shared data so must not be written to under normal circumstances. 1866 */ 1867 static inline int skb_cloned(const struct sk_buff *skb) 1868 { 1869 return skb->cloned && 1870 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1871 } 1872 1873 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1874 { 1875 might_sleep_if(gfpflags_allow_blocking(pri)); 1876 1877 if (skb_cloned(skb)) 1878 return pskb_expand_head(skb, 0, 0, pri); 1879 1880 return 0; 1881 } 1882 1883 /* This variant of skb_unclone() makes sure skb->truesize 1884 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 1885 * 1886 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 1887 * when various debugging features are in place. 1888 */ 1889 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 1890 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 1891 { 1892 might_sleep_if(gfpflags_allow_blocking(pri)); 1893 1894 if (skb_cloned(skb)) 1895 return __skb_unclone_keeptruesize(skb, pri); 1896 return 0; 1897 } 1898 1899 /** 1900 * skb_header_cloned - is the header a clone 1901 * @skb: buffer to check 1902 * 1903 * Returns true if modifying the header part of the buffer requires 1904 * the data to be copied. 1905 */ 1906 static inline int skb_header_cloned(const struct sk_buff *skb) 1907 { 1908 int dataref; 1909 1910 if (!skb->cloned) 1911 return 0; 1912 1913 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1914 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1915 return dataref != 1; 1916 } 1917 1918 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1919 { 1920 might_sleep_if(gfpflags_allow_blocking(pri)); 1921 1922 if (skb_header_cloned(skb)) 1923 return pskb_expand_head(skb, 0, 0, pri); 1924 1925 return 0; 1926 } 1927 1928 /** 1929 * __skb_header_release() - allow clones to use the headroom 1930 * @skb: buffer to operate on 1931 * 1932 * See "DOC: dataref and headerless skbs". 1933 */ 1934 static inline void __skb_header_release(struct sk_buff *skb) 1935 { 1936 skb->nohdr = 1; 1937 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1938 } 1939 1940 1941 /** 1942 * skb_shared - is the buffer shared 1943 * @skb: buffer to check 1944 * 1945 * Returns true if more than one person has a reference to this 1946 * buffer. 1947 */ 1948 static inline int skb_shared(const struct sk_buff *skb) 1949 { 1950 return refcount_read(&skb->users) != 1; 1951 } 1952 1953 /** 1954 * skb_share_check - check if buffer is shared and if so clone it 1955 * @skb: buffer to check 1956 * @pri: priority for memory allocation 1957 * 1958 * If the buffer is shared the buffer is cloned and the old copy 1959 * drops a reference. A new clone with a single reference is returned. 1960 * If the buffer is not shared the original buffer is returned. When 1961 * being called from interrupt status or with spinlocks held pri must 1962 * be GFP_ATOMIC. 1963 * 1964 * NULL is returned on a memory allocation failure. 1965 */ 1966 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1967 { 1968 might_sleep_if(gfpflags_allow_blocking(pri)); 1969 if (skb_shared(skb)) { 1970 struct sk_buff *nskb = skb_clone(skb, pri); 1971 1972 if (likely(nskb)) 1973 consume_skb(skb); 1974 else 1975 kfree_skb(skb); 1976 skb = nskb; 1977 } 1978 return skb; 1979 } 1980 1981 /* 1982 * Copy shared buffers into a new sk_buff. We effectively do COW on 1983 * packets to handle cases where we have a local reader and forward 1984 * and a couple of other messy ones. The normal one is tcpdumping 1985 * a packet thats being forwarded. 1986 */ 1987 1988 /** 1989 * skb_unshare - make a copy of a shared buffer 1990 * @skb: buffer to check 1991 * @pri: priority for memory allocation 1992 * 1993 * If the socket buffer is a clone then this function creates a new 1994 * copy of the data, drops a reference count on the old copy and returns 1995 * the new copy with the reference count at 1. If the buffer is not a clone 1996 * the original buffer is returned. When called with a spinlock held or 1997 * from interrupt state @pri must be %GFP_ATOMIC 1998 * 1999 * %NULL is returned on a memory allocation failure. 2000 */ 2001 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2002 gfp_t pri) 2003 { 2004 might_sleep_if(gfpflags_allow_blocking(pri)); 2005 if (skb_cloned(skb)) { 2006 struct sk_buff *nskb = skb_copy(skb, pri); 2007 2008 /* Free our shared copy */ 2009 if (likely(nskb)) 2010 consume_skb(skb); 2011 else 2012 kfree_skb(skb); 2013 skb = nskb; 2014 } 2015 return skb; 2016 } 2017 2018 /** 2019 * skb_peek - peek at the head of an &sk_buff_head 2020 * @list_: list to peek at 2021 * 2022 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2023 * be careful with this one. A peek leaves the buffer on the 2024 * list and someone else may run off with it. You must hold 2025 * the appropriate locks or have a private queue to do this. 2026 * 2027 * Returns %NULL for an empty list or a pointer to the head element. 2028 * The reference count is not incremented and the reference is therefore 2029 * volatile. Use with caution. 2030 */ 2031 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2032 { 2033 struct sk_buff *skb = list_->next; 2034 2035 if (skb == (struct sk_buff *)list_) 2036 skb = NULL; 2037 return skb; 2038 } 2039 2040 /** 2041 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2042 * @list_: list to peek at 2043 * 2044 * Like skb_peek(), but the caller knows that the list is not empty. 2045 */ 2046 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2047 { 2048 return list_->next; 2049 } 2050 2051 /** 2052 * skb_peek_next - peek skb following the given one from a queue 2053 * @skb: skb to start from 2054 * @list_: list to peek at 2055 * 2056 * Returns %NULL when the end of the list is met or a pointer to the 2057 * next element. The reference count is not incremented and the 2058 * reference is therefore volatile. Use with caution. 2059 */ 2060 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2061 const struct sk_buff_head *list_) 2062 { 2063 struct sk_buff *next = skb->next; 2064 2065 if (next == (struct sk_buff *)list_) 2066 next = NULL; 2067 return next; 2068 } 2069 2070 /** 2071 * skb_peek_tail - peek at the tail of an &sk_buff_head 2072 * @list_: list to peek at 2073 * 2074 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2075 * be careful with this one. A peek leaves the buffer on the 2076 * list and someone else may run off with it. You must hold 2077 * the appropriate locks or have a private queue to do this. 2078 * 2079 * Returns %NULL for an empty list or a pointer to the tail element. 2080 * The reference count is not incremented and the reference is therefore 2081 * volatile. Use with caution. 2082 */ 2083 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2084 { 2085 struct sk_buff *skb = READ_ONCE(list_->prev); 2086 2087 if (skb == (struct sk_buff *)list_) 2088 skb = NULL; 2089 return skb; 2090 2091 } 2092 2093 /** 2094 * skb_queue_len - get queue length 2095 * @list_: list to measure 2096 * 2097 * Return the length of an &sk_buff queue. 2098 */ 2099 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2100 { 2101 return list_->qlen; 2102 } 2103 2104 /** 2105 * skb_queue_len_lockless - get queue length 2106 * @list_: list to measure 2107 * 2108 * Return the length of an &sk_buff queue. 2109 * This variant can be used in lockless contexts. 2110 */ 2111 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2112 { 2113 return READ_ONCE(list_->qlen); 2114 } 2115 2116 /** 2117 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2118 * @list: queue to initialize 2119 * 2120 * This initializes only the list and queue length aspects of 2121 * an sk_buff_head object. This allows to initialize the list 2122 * aspects of an sk_buff_head without reinitializing things like 2123 * the spinlock. It can also be used for on-stack sk_buff_head 2124 * objects where the spinlock is known to not be used. 2125 */ 2126 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2127 { 2128 list->prev = list->next = (struct sk_buff *)list; 2129 list->qlen = 0; 2130 } 2131 2132 /* 2133 * This function creates a split out lock class for each invocation; 2134 * this is needed for now since a whole lot of users of the skb-queue 2135 * infrastructure in drivers have different locking usage (in hardirq) 2136 * than the networking core (in softirq only). In the long run either the 2137 * network layer or drivers should need annotation to consolidate the 2138 * main types of usage into 3 classes. 2139 */ 2140 static inline void skb_queue_head_init(struct sk_buff_head *list) 2141 { 2142 spin_lock_init(&list->lock); 2143 __skb_queue_head_init(list); 2144 } 2145 2146 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2147 struct lock_class_key *class) 2148 { 2149 skb_queue_head_init(list); 2150 lockdep_set_class(&list->lock, class); 2151 } 2152 2153 /* 2154 * Insert an sk_buff on a list. 2155 * 2156 * The "__skb_xxxx()" functions are the non-atomic ones that 2157 * can only be called with interrupts disabled. 2158 */ 2159 static inline void __skb_insert(struct sk_buff *newsk, 2160 struct sk_buff *prev, struct sk_buff *next, 2161 struct sk_buff_head *list) 2162 { 2163 /* See skb_queue_empty_lockless() and skb_peek_tail() 2164 * for the opposite READ_ONCE() 2165 */ 2166 WRITE_ONCE(newsk->next, next); 2167 WRITE_ONCE(newsk->prev, prev); 2168 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2169 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2170 WRITE_ONCE(list->qlen, list->qlen + 1); 2171 } 2172 2173 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2174 struct sk_buff *prev, 2175 struct sk_buff *next) 2176 { 2177 struct sk_buff *first = list->next; 2178 struct sk_buff *last = list->prev; 2179 2180 WRITE_ONCE(first->prev, prev); 2181 WRITE_ONCE(prev->next, first); 2182 2183 WRITE_ONCE(last->next, next); 2184 WRITE_ONCE(next->prev, last); 2185 } 2186 2187 /** 2188 * skb_queue_splice - join two skb lists, this is designed for stacks 2189 * @list: the new list to add 2190 * @head: the place to add it in the first list 2191 */ 2192 static inline void skb_queue_splice(const struct sk_buff_head *list, 2193 struct sk_buff_head *head) 2194 { 2195 if (!skb_queue_empty(list)) { 2196 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2197 head->qlen += list->qlen; 2198 } 2199 } 2200 2201 /** 2202 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2203 * @list: the new list to add 2204 * @head: the place to add it in the first list 2205 * 2206 * The list at @list is reinitialised 2207 */ 2208 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2209 struct sk_buff_head *head) 2210 { 2211 if (!skb_queue_empty(list)) { 2212 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2213 head->qlen += list->qlen; 2214 __skb_queue_head_init(list); 2215 } 2216 } 2217 2218 /** 2219 * skb_queue_splice_tail - join two skb lists, each list being a queue 2220 * @list: the new list to add 2221 * @head: the place to add it in the first list 2222 */ 2223 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2224 struct sk_buff_head *head) 2225 { 2226 if (!skb_queue_empty(list)) { 2227 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2228 head->qlen += list->qlen; 2229 } 2230 } 2231 2232 /** 2233 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2234 * @list: the new list to add 2235 * @head: the place to add it in the first list 2236 * 2237 * Each of the lists is a queue. 2238 * The list at @list is reinitialised 2239 */ 2240 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2241 struct sk_buff_head *head) 2242 { 2243 if (!skb_queue_empty(list)) { 2244 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2245 head->qlen += list->qlen; 2246 __skb_queue_head_init(list); 2247 } 2248 } 2249 2250 /** 2251 * __skb_queue_after - queue a buffer at the list head 2252 * @list: list to use 2253 * @prev: place after this buffer 2254 * @newsk: buffer to queue 2255 * 2256 * Queue a buffer int the middle of a list. This function takes no locks 2257 * and you must therefore hold required locks before calling it. 2258 * 2259 * A buffer cannot be placed on two lists at the same time. 2260 */ 2261 static inline void __skb_queue_after(struct sk_buff_head *list, 2262 struct sk_buff *prev, 2263 struct sk_buff *newsk) 2264 { 2265 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2266 } 2267 2268 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2269 struct sk_buff_head *list); 2270 2271 static inline void __skb_queue_before(struct sk_buff_head *list, 2272 struct sk_buff *next, 2273 struct sk_buff *newsk) 2274 { 2275 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2276 } 2277 2278 /** 2279 * __skb_queue_head - queue a buffer at the list head 2280 * @list: list to use 2281 * @newsk: buffer to queue 2282 * 2283 * Queue a buffer at the start of a list. This function takes no locks 2284 * and you must therefore hold required locks before calling it. 2285 * 2286 * A buffer cannot be placed on two lists at the same time. 2287 */ 2288 static inline void __skb_queue_head(struct sk_buff_head *list, 2289 struct sk_buff *newsk) 2290 { 2291 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2292 } 2293 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2294 2295 /** 2296 * __skb_queue_tail - queue a buffer at the list tail 2297 * @list: list to use 2298 * @newsk: buffer to queue 2299 * 2300 * Queue a buffer at the end of a list. This function takes no locks 2301 * and you must therefore hold required locks before calling it. 2302 * 2303 * A buffer cannot be placed on two lists at the same time. 2304 */ 2305 static inline void __skb_queue_tail(struct sk_buff_head *list, 2306 struct sk_buff *newsk) 2307 { 2308 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2309 } 2310 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2311 2312 /* 2313 * remove sk_buff from list. _Must_ be called atomically, and with 2314 * the list known.. 2315 */ 2316 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2317 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2318 { 2319 struct sk_buff *next, *prev; 2320 2321 WRITE_ONCE(list->qlen, list->qlen - 1); 2322 next = skb->next; 2323 prev = skb->prev; 2324 skb->next = skb->prev = NULL; 2325 WRITE_ONCE(next->prev, prev); 2326 WRITE_ONCE(prev->next, next); 2327 } 2328 2329 /** 2330 * __skb_dequeue - remove from the head of the queue 2331 * @list: list to dequeue from 2332 * 2333 * Remove the head of the list. This function does not take any locks 2334 * so must be used with appropriate locks held only. The head item is 2335 * returned or %NULL if the list is empty. 2336 */ 2337 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2338 { 2339 struct sk_buff *skb = skb_peek(list); 2340 if (skb) 2341 __skb_unlink(skb, list); 2342 return skb; 2343 } 2344 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2345 2346 /** 2347 * __skb_dequeue_tail - remove from the tail of the queue 2348 * @list: list to dequeue from 2349 * 2350 * Remove the tail of the list. This function does not take any locks 2351 * so must be used with appropriate locks held only. The tail item is 2352 * returned or %NULL if the list is empty. 2353 */ 2354 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2355 { 2356 struct sk_buff *skb = skb_peek_tail(list); 2357 if (skb) 2358 __skb_unlink(skb, list); 2359 return skb; 2360 } 2361 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2362 2363 2364 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2365 { 2366 return skb->data_len; 2367 } 2368 2369 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2370 { 2371 return skb->len - skb->data_len; 2372 } 2373 2374 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2375 { 2376 unsigned int i, len = 0; 2377 2378 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2379 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2380 return len; 2381 } 2382 2383 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2384 { 2385 return skb_headlen(skb) + __skb_pagelen(skb); 2386 } 2387 2388 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2389 int i, struct page *page, 2390 int off, int size) 2391 { 2392 skb_frag_t *frag = &shinfo->frags[i]; 2393 2394 /* 2395 * Propagate page pfmemalloc to the skb if we can. The problem is 2396 * that not all callers have unique ownership of the page but rely 2397 * on page_is_pfmemalloc doing the right thing(tm). 2398 */ 2399 frag->bv_page = page; 2400 frag->bv_offset = off; 2401 skb_frag_size_set(frag, size); 2402 } 2403 2404 /** 2405 * skb_len_add - adds a number to len fields of skb 2406 * @skb: buffer to add len to 2407 * @delta: number of bytes to add 2408 */ 2409 static inline void skb_len_add(struct sk_buff *skb, int delta) 2410 { 2411 skb->len += delta; 2412 skb->data_len += delta; 2413 skb->truesize += delta; 2414 } 2415 2416 /** 2417 * __skb_fill_page_desc - initialise a paged fragment in an skb 2418 * @skb: buffer containing fragment to be initialised 2419 * @i: paged fragment index to initialise 2420 * @page: the page to use for this fragment 2421 * @off: the offset to the data with @page 2422 * @size: the length of the data 2423 * 2424 * Initialises the @i'th fragment of @skb to point to &size bytes at 2425 * offset @off within @page. 2426 * 2427 * Does not take any additional reference on the fragment. 2428 */ 2429 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2430 struct page *page, int off, int size) 2431 { 2432 __skb_fill_page_desc_noacc(skb_shinfo(skb), i, page, off, size); 2433 page = compound_head(page); 2434 if (page_is_pfmemalloc(page)) 2435 skb->pfmemalloc = true; 2436 } 2437 2438 /** 2439 * skb_fill_page_desc - initialise a paged fragment in an skb 2440 * @skb: buffer containing fragment to be initialised 2441 * @i: paged fragment index to initialise 2442 * @page: the page to use for this fragment 2443 * @off: the offset to the data with @page 2444 * @size: the length of the data 2445 * 2446 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2447 * @skb to point to @size bytes at offset @off within @page. In 2448 * addition updates @skb such that @i is the last fragment. 2449 * 2450 * Does not take any additional reference on the fragment. 2451 */ 2452 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2453 struct page *page, int off, int size) 2454 { 2455 __skb_fill_page_desc(skb, i, page, off, size); 2456 skb_shinfo(skb)->nr_frags = i + 1; 2457 } 2458 2459 /** 2460 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2461 * @skb: buffer containing fragment to be initialised 2462 * @i: paged fragment index to initialise 2463 * @page: the page to use for this fragment 2464 * @off: the offset to the data with @page 2465 * @size: the length of the data 2466 * 2467 * Variant of skb_fill_page_desc() which does not deal with 2468 * pfmemalloc, if page is not owned by us. 2469 */ 2470 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2471 struct page *page, int off, 2472 int size) 2473 { 2474 struct skb_shared_info *shinfo = skb_shinfo(skb); 2475 2476 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2477 shinfo->nr_frags = i + 1; 2478 } 2479 2480 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 2481 int size, unsigned int truesize); 2482 2483 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2484 unsigned int truesize); 2485 2486 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2487 2488 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2489 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2490 { 2491 return skb->head + skb->tail; 2492 } 2493 2494 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2495 { 2496 skb->tail = skb->data - skb->head; 2497 } 2498 2499 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2500 { 2501 skb_reset_tail_pointer(skb); 2502 skb->tail += offset; 2503 } 2504 2505 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2506 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2507 { 2508 return skb->tail; 2509 } 2510 2511 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2512 { 2513 skb->tail = skb->data; 2514 } 2515 2516 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2517 { 2518 skb->tail = skb->data + offset; 2519 } 2520 2521 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2522 2523 static inline void skb_assert_len(struct sk_buff *skb) 2524 { 2525 #ifdef CONFIG_DEBUG_NET 2526 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2527 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2528 #endif /* CONFIG_DEBUG_NET */ 2529 } 2530 2531 /* 2532 * Add data to an sk_buff 2533 */ 2534 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2535 void *skb_put(struct sk_buff *skb, unsigned int len); 2536 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2537 { 2538 void *tmp = skb_tail_pointer(skb); 2539 SKB_LINEAR_ASSERT(skb); 2540 skb->tail += len; 2541 skb->len += len; 2542 return tmp; 2543 } 2544 2545 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2546 { 2547 void *tmp = __skb_put(skb, len); 2548 2549 memset(tmp, 0, len); 2550 return tmp; 2551 } 2552 2553 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2554 unsigned int len) 2555 { 2556 void *tmp = __skb_put(skb, len); 2557 2558 memcpy(tmp, data, len); 2559 return tmp; 2560 } 2561 2562 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2563 { 2564 *(u8 *)__skb_put(skb, 1) = val; 2565 } 2566 2567 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2568 { 2569 void *tmp = skb_put(skb, len); 2570 2571 memset(tmp, 0, len); 2572 2573 return tmp; 2574 } 2575 2576 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2577 unsigned int len) 2578 { 2579 void *tmp = skb_put(skb, len); 2580 2581 memcpy(tmp, data, len); 2582 2583 return tmp; 2584 } 2585 2586 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2587 { 2588 *(u8 *)skb_put(skb, 1) = val; 2589 } 2590 2591 void *skb_push(struct sk_buff *skb, unsigned int len); 2592 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2593 { 2594 skb->data -= len; 2595 skb->len += len; 2596 return skb->data; 2597 } 2598 2599 void *skb_pull(struct sk_buff *skb, unsigned int len); 2600 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2601 { 2602 skb->len -= len; 2603 if (unlikely(skb->len < skb->data_len)) { 2604 #if defined(CONFIG_DEBUG_NET) 2605 skb->len += len; 2606 pr_err("__skb_pull(len=%u)\n", len); 2607 skb_dump(KERN_ERR, skb, false); 2608 #endif 2609 BUG(); 2610 } 2611 return skb->data += len; 2612 } 2613 2614 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2615 { 2616 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2617 } 2618 2619 void *skb_pull_data(struct sk_buff *skb, size_t len); 2620 2621 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2622 2623 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2624 { 2625 if (likely(len <= skb_headlen(skb))) 2626 return true; 2627 if (unlikely(len > skb->len)) 2628 return false; 2629 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 2630 } 2631 2632 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2633 { 2634 if (!pskb_may_pull(skb, len)) 2635 return NULL; 2636 2637 skb->len -= len; 2638 return skb->data += len; 2639 } 2640 2641 void skb_condense(struct sk_buff *skb); 2642 2643 /** 2644 * skb_headroom - bytes at buffer head 2645 * @skb: buffer to check 2646 * 2647 * Return the number of bytes of free space at the head of an &sk_buff. 2648 */ 2649 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2650 { 2651 return skb->data - skb->head; 2652 } 2653 2654 /** 2655 * skb_tailroom - bytes at buffer end 2656 * @skb: buffer to check 2657 * 2658 * Return the number of bytes of free space at the tail of an sk_buff 2659 */ 2660 static inline int skb_tailroom(const struct sk_buff *skb) 2661 { 2662 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2663 } 2664 2665 /** 2666 * skb_availroom - bytes at buffer end 2667 * @skb: buffer to check 2668 * 2669 * Return the number of bytes of free space at the tail of an sk_buff 2670 * allocated by sk_stream_alloc() 2671 */ 2672 static inline int skb_availroom(const struct sk_buff *skb) 2673 { 2674 if (skb_is_nonlinear(skb)) 2675 return 0; 2676 2677 return skb->end - skb->tail - skb->reserved_tailroom; 2678 } 2679 2680 /** 2681 * skb_reserve - adjust headroom 2682 * @skb: buffer to alter 2683 * @len: bytes to move 2684 * 2685 * Increase the headroom of an empty &sk_buff by reducing the tail 2686 * room. This is only allowed for an empty buffer. 2687 */ 2688 static inline void skb_reserve(struct sk_buff *skb, int len) 2689 { 2690 skb->data += len; 2691 skb->tail += len; 2692 } 2693 2694 /** 2695 * skb_tailroom_reserve - adjust reserved_tailroom 2696 * @skb: buffer to alter 2697 * @mtu: maximum amount of headlen permitted 2698 * @needed_tailroom: minimum amount of reserved_tailroom 2699 * 2700 * Set reserved_tailroom so that headlen can be as large as possible but 2701 * not larger than mtu and tailroom cannot be smaller than 2702 * needed_tailroom. 2703 * The required headroom should already have been reserved before using 2704 * this function. 2705 */ 2706 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2707 unsigned int needed_tailroom) 2708 { 2709 SKB_LINEAR_ASSERT(skb); 2710 if (mtu < skb_tailroom(skb) - needed_tailroom) 2711 /* use at most mtu */ 2712 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2713 else 2714 /* use up to all available space */ 2715 skb->reserved_tailroom = needed_tailroom; 2716 } 2717 2718 #define ENCAP_TYPE_ETHER 0 2719 #define ENCAP_TYPE_IPPROTO 1 2720 2721 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2722 __be16 protocol) 2723 { 2724 skb->inner_protocol = protocol; 2725 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2726 } 2727 2728 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2729 __u8 ipproto) 2730 { 2731 skb->inner_ipproto = ipproto; 2732 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2733 } 2734 2735 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2736 { 2737 skb->inner_mac_header = skb->mac_header; 2738 skb->inner_network_header = skb->network_header; 2739 skb->inner_transport_header = skb->transport_header; 2740 } 2741 2742 static inline void skb_reset_mac_len(struct sk_buff *skb) 2743 { 2744 skb->mac_len = skb->network_header - skb->mac_header; 2745 } 2746 2747 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2748 *skb) 2749 { 2750 return skb->head + skb->inner_transport_header; 2751 } 2752 2753 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2754 { 2755 return skb_inner_transport_header(skb) - skb->data; 2756 } 2757 2758 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2759 { 2760 skb->inner_transport_header = skb->data - skb->head; 2761 } 2762 2763 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2764 const int offset) 2765 { 2766 skb_reset_inner_transport_header(skb); 2767 skb->inner_transport_header += offset; 2768 } 2769 2770 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2771 { 2772 return skb->head + skb->inner_network_header; 2773 } 2774 2775 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2776 { 2777 skb->inner_network_header = skb->data - skb->head; 2778 } 2779 2780 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2781 const int offset) 2782 { 2783 skb_reset_inner_network_header(skb); 2784 skb->inner_network_header += offset; 2785 } 2786 2787 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2788 { 2789 return skb->head + skb->inner_mac_header; 2790 } 2791 2792 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2793 { 2794 skb->inner_mac_header = skb->data - skb->head; 2795 } 2796 2797 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2798 const int offset) 2799 { 2800 skb_reset_inner_mac_header(skb); 2801 skb->inner_mac_header += offset; 2802 } 2803 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2804 { 2805 return skb->transport_header != (typeof(skb->transport_header))~0U; 2806 } 2807 2808 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2809 { 2810 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 2811 return skb->head + skb->transport_header; 2812 } 2813 2814 static inline void skb_reset_transport_header(struct sk_buff *skb) 2815 { 2816 skb->transport_header = skb->data - skb->head; 2817 } 2818 2819 static inline void skb_set_transport_header(struct sk_buff *skb, 2820 const int offset) 2821 { 2822 skb_reset_transport_header(skb); 2823 skb->transport_header += offset; 2824 } 2825 2826 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2827 { 2828 return skb->head + skb->network_header; 2829 } 2830 2831 static inline void skb_reset_network_header(struct sk_buff *skb) 2832 { 2833 skb->network_header = skb->data - skb->head; 2834 } 2835 2836 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2837 { 2838 skb_reset_network_header(skb); 2839 skb->network_header += offset; 2840 } 2841 2842 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2843 { 2844 return skb->mac_header != (typeof(skb->mac_header))~0U; 2845 } 2846 2847 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2848 { 2849 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 2850 return skb->head + skb->mac_header; 2851 } 2852 2853 static inline int skb_mac_offset(const struct sk_buff *skb) 2854 { 2855 return skb_mac_header(skb) - skb->data; 2856 } 2857 2858 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2859 { 2860 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 2861 return skb->network_header - skb->mac_header; 2862 } 2863 2864 static inline void skb_unset_mac_header(struct sk_buff *skb) 2865 { 2866 skb->mac_header = (typeof(skb->mac_header))~0U; 2867 } 2868 2869 static inline void skb_reset_mac_header(struct sk_buff *skb) 2870 { 2871 skb->mac_header = skb->data - skb->head; 2872 } 2873 2874 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2875 { 2876 skb_reset_mac_header(skb); 2877 skb->mac_header += offset; 2878 } 2879 2880 static inline void skb_pop_mac_header(struct sk_buff *skb) 2881 { 2882 skb->mac_header = skb->network_header; 2883 } 2884 2885 static inline void skb_probe_transport_header(struct sk_buff *skb) 2886 { 2887 struct flow_keys_basic keys; 2888 2889 if (skb_transport_header_was_set(skb)) 2890 return; 2891 2892 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 2893 NULL, 0, 0, 0, 0)) 2894 skb_set_transport_header(skb, keys.control.thoff); 2895 } 2896 2897 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2898 { 2899 if (skb_mac_header_was_set(skb)) { 2900 const unsigned char *old_mac = skb_mac_header(skb); 2901 2902 skb_set_mac_header(skb, -skb->mac_len); 2903 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2904 } 2905 } 2906 2907 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2908 { 2909 return skb->csum_start - skb_headroom(skb); 2910 } 2911 2912 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2913 { 2914 return skb->head + skb->csum_start; 2915 } 2916 2917 static inline int skb_transport_offset(const struct sk_buff *skb) 2918 { 2919 return skb_transport_header(skb) - skb->data; 2920 } 2921 2922 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2923 { 2924 return skb->transport_header - skb->network_header; 2925 } 2926 2927 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2928 { 2929 return skb->inner_transport_header - skb->inner_network_header; 2930 } 2931 2932 static inline int skb_network_offset(const struct sk_buff *skb) 2933 { 2934 return skb_network_header(skb) - skb->data; 2935 } 2936 2937 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2938 { 2939 return skb_inner_network_header(skb) - skb->data; 2940 } 2941 2942 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2943 { 2944 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2945 } 2946 2947 /* 2948 * CPUs often take a performance hit when accessing unaligned memory 2949 * locations. The actual performance hit varies, it can be small if the 2950 * hardware handles it or large if we have to take an exception and fix it 2951 * in software. 2952 * 2953 * Since an ethernet header is 14 bytes network drivers often end up with 2954 * the IP header at an unaligned offset. The IP header can be aligned by 2955 * shifting the start of the packet by 2 bytes. Drivers should do this 2956 * with: 2957 * 2958 * skb_reserve(skb, NET_IP_ALIGN); 2959 * 2960 * The downside to this alignment of the IP header is that the DMA is now 2961 * unaligned. On some architectures the cost of an unaligned DMA is high 2962 * and this cost outweighs the gains made by aligning the IP header. 2963 * 2964 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2965 * to be overridden. 2966 */ 2967 #ifndef NET_IP_ALIGN 2968 #define NET_IP_ALIGN 2 2969 #endif 2970 2971 /* 2972 * The networking layer reserves some headroom in skb data (via 2973 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2974 * the header has to grow. In the default case, if the header has to grow 2975 * 32 bytes or less we avoid the reallocation. 2976 * 2977 * Unfortunately this headroom changes the DMA alignment of the resulting 2978 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2979 * on some architectures. An architecture can override this value, 2980 * perhaps setting it to a cacheline in size (since that will maintain 2981 * cacheline alignment of the DMA). It must be a power of 2. 2982 * 2983 * Various parts of the networking layer expect at least 32 bytes of 2984 * headroom, you should not reduce this. 2985 * 2986 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2987 * to reduce average number of cache lines per packet. 2988 * get_rps_cpu() for example only access one 64 bytes aligned block : 2989 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2990 */ 2991 #ifndef NET_SKB_PAD 2992 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2993 #endif 2994 2995 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2996 2997 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 2998 { 2999 if (WARN_ON(skb_is_nonlinear(skb))) 3000 return; 3001 skb->len = len; 3002 skb_set_tail_pointer(skb, len); 3003 } 3004 3005 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3006 { 3007 __skb_set_length(skb, len); 3008 } 3009 3010 void skb_trim(struct sk_buff *skb, unsigned int len); 3011 3012 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3013 { 3014 if (skb->data_len) 3015 return ___pskb_trim(skb, len); 3016 __skb_trim(skb, len); 3017 return 0; 3018 } 3019 3020 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3021 { 3022 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3023 } 3024 3025 /** 3026 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3027 * @skb: buffer to alter 3028 * @len: new length 3029 * 3030 * This is identical to pskb_trim except that the caller knows that 3031 * the skb is not cloned so we should never get an error due to out- 3032 * of-memory. 3033 */ 3034 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3035 { 3036 int err = pskb_trim(skb, len); 3037 BUG_ON(err); 3038 } 3039 3040 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3041 { 3042 unsigned int diff = len - skb->len; 3043 3044 if (skb_tailroom(skb) < diff) { 3045 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3046 GFP_ATOMIC); 3047 if (ret) 3048 return ret; 3049 } 3050 __skb_set_length(skb, len); 3051 return 0; 3052 } 3053 3054 /** 3055 * skb_orphan - orphan a buffer 3056 * @skb: buffer to orphan 3057 * 3058 * If a buffer currently has an owner then we call the owner's 3059 * destructor function and make the @skb unowned. The buffer continues 3060 * to exist but is no longer charged to its former owner. 3061 */ 3062 static inline void skb_orphan(struct sk_buff *skb) 3063 { 3064 if (skb->destructor) { 3065 skb->destructor(skb); 3066 skb->destructor = NULL; 3067 skb->sk = NULL; 3068 } else { 3069 BUG_ON(skb->sk); 3070 } 3071 } 3072 3073 /** 3074 * skb_orphan_frags - orphan the frags contained in a buffer 3075 * @skb: buffer to orphan frags from 3076 * @gfp_mask: allocation mask for replacement pages 3077 * 3078 * For each frag in the SKB which needs a destructor (i.e. has an 3079 * owner) create a copy of that frag and release the original 3080 * page by calling the destructor. 3081 */ 3082 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3083 { 3084 if (likely(!skb_zcopy(skb))) 3085 return 0; 3086 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3087 return 0; 3088 return skb_copy_ubufs(skb, gfp_mask); 3089 } 3090 3091 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3092 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3093 { 3094 if (likely(!skb_zcopy(skb))) 3095 return 0; 3096 return skb_copy_ubufs(skb, gfp_mask); 3097 } 3098 3099 /** 3100 * __skb_queue_purge - empty a list 3101 * @list: list to empty 3102 * 3103 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3104 * the list and one reference dropped. This function does not take the 3105 * list lock and the caller must hold the relevant locks to use it. 3106 */ 3107 static inline void __skb_queue_purge(struct sk_buff_head *list) 3108 { 3109 struct sk_buff *skb; 3110 while ((skb = __skb_dequeue(list)) != NULL) 3111 kfree_skb(skb); 3112 } 3113 void skb_queue_purge(struct sk_buff_head *list); 3114 3115 unsigned int skb_rbtree_purge(struct rb_root *root); 3116 3117 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3118 3119 /** 3120 * netdev_alloc_frag - allocate a page fragment 3121 * @fragsz: fragment size 3122 * 3123 * Allocates a frag from a page for receive buffer. 3124 * Uses GFP_ATOMIC allocations. 3125 */ 3126 static inline void *netdev_alloc_frag(unsigned int fragsz) 3127 { 3128 return __netdev_alloc_frag_align(fragsz, ~0u); 3129 } 3130 3131 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3132 unsigned int align) 3133 { 3134 WARN_ON_ONCE(!is_power_of_2(align)); 3135 return __netdev_alloc_frag_align(fragsz, -align); 3136 } 3137 3138 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3139 gfp_t gfp_mask); 3140 3141 /** 3142 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3143 * @dev: network device to receive on 3144 * @length: length to allocate 3145 * 3146 * Allocate a new &sk_buff and assign it a usage count of one. The 3147 * buffer has unspecified headroom built in. Users should allocate 3148 * the headroom they think they need without accounting for the 3149 * built in space. The built in space is used for optimisations. 3150 * 3151 * %NULL is returned if there is no free memory. Although this function 3152 * allocates memory it can be called from an interrupt. 3153 */ 3154 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3155 unsigned int length) 3156 { 3157 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3158 } 3159 3160 /* legacy helper around __netdev_alloc_skb() */ 3161 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3162 gfp_t gfp_mask) 3163 { 3164 return __netdev_alloc_skb(NULL, length, gfp_mask); 3165 } 3166 3167 /* legacy helper around netdev_alloc_skb() */ 3168 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3169 { 3170 return netdev_alloc_skb(NULL, length); 3171 } 3172 3173 3174 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3175 unsigned int length, gfp_t gfp) 3176 { 3177 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3178 3179 if (NET_IP_ALIGN && skb) 3180 skb_reserve(skb, NET_IP_ALIGN); 3181 return skb; 3182 } 3183 3184 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3185 unsigned int length) 3186 { 3187 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3188 } 3189 3190 static inline void skb_free_frag(void *addr) 3191 { 3192 page_frag_free(addr); 3193 } 3194 3195 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3196 3197 static inline void *napi_alloc_frag(unsigned int fragsz) 3198 { 3199 return __napi_alloc_frag_align(fragsz, ~0u); 3200 } 3201 3202 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3203 unsigned int align) 3204 { 3205 WARN_ON_ONCE(!is_power_of_2(align)); 3206 return __napi_alloc_frag_align(fragsz, -align); 3207 } 3208 3209 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 3210 unsigned int length, gfp_t gfp_mask); 3211 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 3212 unsigned int length) 3213 { 3214 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 3215 } 3216 void napi_consume_skb(struct sk_buff *skb, int budget); 3217 3218 void napi_skb_free_stolen_head(struct sk_buff *skb); 3219 void __kfree_skb_defer(struct sk_buff *skb); 3220 3221 /** 3222 * __dev_alloc_pages - allocate page for network Rx 3223 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3224 * @order: size of the allocation 3225 * 3226 * Allocate a new page. 3227 * 3228 * %NULL is returned if there is no free memory. 3229 */ 3230 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 3231 unsigned int order) 3232 { 3233 /* This piece of code contains several assumptions. 3234 * 1. This is for device Rx, therefor a cold page is preferred. 3235 * 2. The expectation is the user wants a compound page. 3236 * 3. If requesting a order 0 page it will not be compound 3237 * due to the check to see if order has a value in prep_new_page 3238 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3239 * code in gfp_to_alloc_flags that should be enforcing this. 3240 */ 3241 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3242 3243 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 3244 } 3245 3246 static inline struct page *dev_alloc_pages(unsigned int order) 3247 { 3248 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 3249 } 3250 3251 /** 3252 * __dev_alloc_page - allocate a page for network Rx 3253 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3254 * 3255 * Allocate a new page. 3256 * 3257 * %NULL is returned if there is no free memory. 3258 */ 3259 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 3260 { 3261 return __dev_alloc_pages(gfp_mask, 0); 3262 } 3263 3264 static inline struct page *dev_alloc_page(void) 3265 { 3266 return dev_alloc_pages(0); 3267 } 3268 3269 /** 3270 * dev_page_is_reusable - check whether a page can be reused for network Rx 3271 * @page: the page to test 3272 * 3273 * A page shouldn't be considered for reusing/recycling if it was allocated 3274 * under memory pressure or at a distant memory node. 3275 * 3276 * Returns false if this page should be returned to page allocator, true 3277 * otherwise. 3278 */ 3279 static inline bool dev_page_is_reusable(const struct page *page) 3280 { 3281 return likely(page_to_nid(page) == numa_mem_id() && 3282 !page_is_pfmemalloc(page)); 3283 } 3284 3285 /** 3286 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3287 * @page: The page that was allocated from skb_alloc_page 3288 * @skb: The skb that may need pfmemalloc set 3289 */ 3290 static inline void skb_propagate_pfmemalloc(const struct page *page, 3291 struct sk_buff *skb) 3292 { 3293 if (page_is_pfmemalloc(page)) 3294 skb->pfmemalloc = true; 3295 } 3296 3297 /** 3298 * skb_frag_off() - Returns the offset of a skb fragment 3299 * @frag: the paged fragment 3300 */ 3301 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3302 { 3303 return frag->bv_offset; 3304 } 3305 3306 /** 3307 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3308 * @frag: skb fragment 3309 * @delta: value to add 3310 */ 3311 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3312 { 3313 frag->bv_offset += delta; 3314 } 3315 3316 /** 3317 * skb_frag_off_set() - Sets the offset of a skb fragment 3318 * @frag: skb fragment 3319 * @offset: offset of fragment 3320 */ 3321 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3322 { 3323 frag->bv_offset = offset; 3324 } 3325 3326 /** 3327 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3328 * @fragto: skb fragment where offset is set 3329 * @fragfrom: skb fragment offset is copied from 3330 */ 3331 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3332 const skb_frag_t *fragfrom) 3333 { 3334 fragto->bv_offset = fragfrom->bv_offset; 3335 } 3336 3337 /** 3338 * skb_frag_page - retrieve the page referred to by a paged fragment 3339 * @frag: the paged fragment 3340 * 3341 * Returns the &struct page associated with @frag. 3342 */ 3343 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3344 { 3345 return frag->bv_page; 3346 } 3347 3348 /** 3349 * __skb_frag_ref - take an addition reference on a paged fragment. 3350 * @frag: the paged fragment 3351 * 3352 * Takes an additional reference on the paged fragment @frag. 3353 */ 3354 static inline void __skb_frag_ref(skb_frag_t *frag) 3355 { 3356 get_page(skb_frag_page(frag)); 3357 } 3358 3359 /** 3360 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 3361 * @skb: the buffer 3362 * @f: the fragment offset. 3363 * 3364 * Takes an additional reference on the @f'th paged fragment of @skb. 3365 */ 3366 static inline void skb_frag_ref(struct sk_buff *skb, int f) 3367 { 3368 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 3369 } 3370 3371 /** 3372 * __skb_frag_unref - release a reference on a paged fragment. 3373 * @frag: the paged fragment 3374 * @recycle: recycle the page if allocated via page_pool 3375 * 3376 * Releases a reference on the paged fragment @frag 3377 * or recycles the page via the page_pool API. 3378 */ 3379 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle) 3380 { 3381 struct page *page = skb_frag_page(frag); 3382 3383 #ifdef CONFIG_PAGE_POOL 3384 if (recycle && page_pool_return_skb_page(page)) 3385 return; 3386 #endif 3387 put_page(page); 3388 } 3389 3390 /** 3391 * skb_frag_unref - release a reference on a paged fragment of an skb. 3392 * @skb: the buffer 3393 * @f: the fragment offset 3394 * 3395 * Releases a reference on the @f'th paged fragment of @skb. 3396 */ 3397 static inline void skb_frag_unref(struct sk_buff *skb, int f) 3398 { 3399 struct skb_shared_info *shinfo = skb_shinfo(skb); 3400 3401 if (!skb_zcopy_managed(skb)) 3402 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle); 3403 } 3404 3405 /** 3406 * skb_frag_address - gets the address of the data contained in a paged fragment 3407 * @frag: the paged fragment buffer 3408 * 3409 * Returns the address of the data within @frag. The page must already 3410 * be mapped. 3411 */ 3412 static inline void *skb_frag_address(const skb_frag_t *frag) 3413 { 3414 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3415 } 3416 3417 /** 3418 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3419 * @frag: the paged fragment buffer 3420 * 3421 * Returns the address of the data within @frag. Checks that the page 3422 * is mapped and returns %NULL otherwise. 3423 */ 3424 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3425 { 3426 void *ptr = page_address(skb_frag_page(frag)); 3427 if (unlikely(!ptr)) 3428 return NULL; 3429 3430 return ptr + skb_frag_off(frag); 3431 } 3432 3433 /** 3434 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3435 * @fragto: skb fragment where page is set 3436 * @fragfrom: skb fragment page is copied from 3437 */ 3438 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3439 const skb_frag_t *fragfrom) 3440 { 3441 fragto->bv_page = fragfrom->bv_page; 3442 } 3443 3444 /** 3445 * __skb_frag_set_page - sets the page contained in a paged fragment 3446 * @frag: the paged fragment 3447 * @page: the page to set 3448 * 3449 * Sets the fragment @frag to contain @page. 3450 */ 3451 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 3452 { 3453 frag->bv_page = page; 3454 } 3455 3456 /** 3457 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 3458 * @skb: the buffer 3459 * @f: the fragment offset 3460 * @page: the page to set 3461 * 3462 * Sets the @f'th fragment of @skb to contain @page. 3463 */ 3464 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 3465 struct page *page) 3466 { 3467 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 3468 } 3469 3470 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3471 3472 /** 3473 * skb_frag_dma_map - maps a paged fragment via the DMA API 3474 * @dev: the device to map the fragment to 3475 * @frag: the paged fragment to map 3476 * @offset: the offset within the fragment (starting at the 3477 * fragment's own offset) 3478 * @size: the number of bytes to map 3479 * @dir: the direction of the mapping (``PCI_DMA_*``) 3480 * 3481 * Maps the page associated with @frag to @device. 3482 */ 3483 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 3484 const skb_frag_t *frag, 3485 size_t offset, size_t size, 3486 enum dma_data_direction dir) 3487 { 3488 return dma_map_page(dev, skb_frag_page(frag), 3489 skb_frag_off(frag) + offset, size, dir); 3490 } 3491 3492 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3493 gfp_t gfp_mask) 3494 { 3495 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3496 } 3497 3498 3499 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3500 gfp_t gfp_mask) 3501 { 3502 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3503 } 3504 3505 3506 /** 3507 * skb_clone_writable - is the header of a clone writable 3508 * @skb: buffer to check 3509 * @len: length up to which to write 3510 * 3511 * Returns true if modifying the header part of the cloned buffer 3512 * does not requires the data to be copied. 3513 */ 3514 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3515 { 3516 return !skb_header_cloned(skb) && 3517 skb_headroom(skb) + len <= skb->hdr_len; 3518 } 3519 3520 static inline int skb_try_make_writable(struct sk_buff *skb, 3521 unsigned int write_len) 3522 { 3523 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3524 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3525 } 3526 3527 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3528 int cloned) 3529 { 3530 int delta = 0; 3531 3532 if (headroom > skb_headroom(skb)) 3533 delta = headroom - skb_headroom(skb); 3534 3535 if (delta || cloned) 3536 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3537 GFP_ATOMIC); 3538 return 0; 3539 } 3540 3541 /** 3542 * skb_cow - copy header of skb when it is required 3543 * @skb: buffer to cow 3544 * @headroom: needed headroom 3545 * 3546 * If the skb passed lacks sufficient headroom or its data part 3547 * is shared, data is reallocated. If reallocation fails, an error 3548 * is returned and original skb is not changed. 3549 * 3550 * The result is skb with writable area skb->head...skb->tail 3551 * and at least @headroom of space at head. 3552 */ 3553 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3554 { 3555 return __skb_cow(skb, headroom, skb_cloned(skb)); 3556 } 3557 3558 /** 3559 * skb_cow_head - skb_cow but only making the head writable 3560 * @skb: buffer to cow 3561 * @headroom: needed headroom 3562 * 3563 * This function is identical to skb_cow except that we replace the 3564 * skb_cloned check by skb_header_cloned. It should be used when 3565 * you only need to push on some header and do not need to modify 3566 * the data. 3567 */ 3568 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3569 { 3570 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3571 } 3572 3573 /** 3574 * skb_padto - pad an skbuff up to a minimal size 3575 * @skb: buffer to pad 3576 * @len: minimal length 3577 * 3578 * Pads up a buffer to ensure the trailing bytes exist and are 3579 * blanked. If the buffer already contains sufficient data it 3580 * is untouched. Otherwise it is extended. Returns zero on 3581 * success. The skb is freed on error. 3582 */ 3583 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3584 { 3585 unsigned int size = skb->len; 3586 if (likely(size >= len)) 3587 return 0; 3588 return skb_pad(skb, len - size); 3589 } 3590 3591 /** 3592 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3593 * @skb: buffer to pad 3594 * @len: minimal length 3595 * @free_on_error: free buffer on error 3596 * 3597 * Pads up a buffer to ensure the trailing bytes exist and are 3598 * blanked. If the buffer already contains sufficient data it 3599 * is untouched. Otherwise it is extended. Returns zero on 3600 * success. The skb is freed on error if @free_on_error is true. 3601 */ 3602 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3603 unsigned int len, 3604 bool free_on_error) 3605 { 3606 unsigned int size = skb->len; 3607 3608 if (unlikely(size < len)) { 3609 len -= size; 3610 if (__skb_pad(skb, len, free_on_error)) 3611 return -ENOMEM; 3612 __skb_put(skb, len); 3613 } 3614 return 0; 3615 } 3616 3617 /** 3618 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3619 * @skb: buffer to pad 3620 * @len: minimal length 3621 * 3622 * Pads up a buffer to ensure the trailing bytes exist and are 3623 * blanked. If the buffer already contains sufficient data it 3624 * is untouched. Otherwise it is extended. Returns zero on 3625 * success. The skb is freed on error. 3626 */ 3627 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3628 { 3629 return __skb_put_padto(skb, len, true); 3630 } 3631 3632 static inline int skb_add_data(struct sk_buff *skb, 3633 struct iov_iter *from, int copy) 3634 { 3635 const int off = skb->len; 3636 3637 if (skb->ip_summed == CHECKSUM_NONE) { 3638 __wsum csum = 0; 3639 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 3640 &csum, from)) { 3641 skb->csum = csum_block_add(skb->csum, csum, off); 3642 return 0; 3643 } 3644 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3645 return 0; 3646 3647 __skb_trim(skb, off); 3648 return -EFAULT; 3649 } 3650 3651 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3652 const struct page *page, int off) 3653 { 3654 if (skb_zcopy(skb)) 3655 return false; 3656 if (i) { 3657 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3658 3659 return page == skb_frag_page(frag) && 3660 off == skb_frag_off(frag) + skb_frag_size(frag); 3661 } 3662 return false; 3663 } 3664 3665 static inline int __skb_linearize(struct sk_buff *skb) 3666 { 3667 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3668 } 3669 3670 /** 3671 * skb_linearize - convert paged skb to linear one 3672 * @skb: buffer to linarize 3673 * 3674 * If there is no free memory -ENOMEM is returned, otherwise zero 3675 * is returned and the old skb data released. 3676 */ 3677 static inline int skb_linearize(struct sk_buff *skb) 3678 { 3679 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3680 } 3681 3682 /** 3683 * skb_has_shared_frag - can any frag be overwritten 3684 * @skb: buffer to test 3685 * 3686 * Return true if the skb has at least one frag that might be modified 3687 * by an external entity (as in vmsplice()/sendfile()) 3688 */ 3689 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3690 { 3691 return skb_is_nonlinear(skb) && 3692 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 3693 } 3694 3695 /** 3696 * skb_linearize_cow - make sure skb is linear and writable 3697 * @skb: buffer to process 3698 * 3699 * If there is no free memory -ENOMEM is returned, otherwise zero 3700 * is returned and the old skb data released. 3701 */ 3702 static inline int skb_linearize_cow(struct sk_buff *skb) 3703 { 3704 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3705 __skb_linearize(skb) : 0; 3706 } 3707 3708 static __always_inline void 3709 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3710 unsigned int off) 3711 { 3712 if (skb->ip_summed == CHECKSUM_COMPLETE) 3713 skb->csum = csum_block_sub(skb->csum, 3714 csum_partial(start, len, 0), off); 3715 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3716 skb_checksum_start_offset(skb) < 0) 3717 skb->ip_summed = CHECKSUM_NONE; 3718 } 3719 3720 /** 3721 * skb_postpull_rcsum - update checksum for received skb after pull 3722 * @skb: buffer to update 3723 * @start: start of data before pull 3724 * @len: length of data pulled 3725 * 3726 * After doing a pull on a received packet, you need to call this to 3727 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3728 * CHECKSUM_NONE so that it can be recomputed from scratch. 3729 */ 3730 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3731 const void *start, unsigned int len) 3732 { 3733 if (skb->ip_summed == CHECKSUM_COMPLETE) 3734 skb->csum = wsum_negate(csum_partial(start, len, 3735 wsum_negate(skb->csum))); 3736 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3737 skb_checksum_start_offset(skb) < 0) 3738 skb->ip_summed = CHECKSUM_NONE; 3739 } 3740 3741 static __always_inline void 3742 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3743 unsigned int off) 3744 { 3745 if (skb->ip_summed == CHECKSUM_COMPLETE) 3746 skb->csum = csum_block_add(skb->csum, 3747 csum_partial(start, len, 0), off); 3748 } 3749 3750 /** 3751 * skb_postpush_rcsum - update checksum for received skb after push 3752 * @skb: buffer to update 3753 * @start: start of data after push 3754 * @len: length of data pushed 3755 * 3756 * After doing a push on a received packet, you need to call this to 3757 * update the CHECKSUM_COMPLETE checksum. 3758 */ 3759 static inline void skb_postpush_rcsum(struct sk_buff *skb, 3760 const void *start, unsigned int len) 3761 { 3762 __skb_postpush_rcsum(skb, start, len, 0); 3763 } 3764 3765 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3766 3767 /** 3768 * skb_push_rcsum - push skb and update receive checksum 3769 * @skb: buffer to update 3770 * @len: length of data pulled 3771 * 3772 * This function performs an skb_push on the packet and updates 3773 * the CHECKSUM_COMPLETE checksum. It should be used on 3774 * receive path processing instead of skb_push unless you know 3775 * that the checksum difference is zero (e.g., a valid IP header) 3776 * or you are setting ip_summed to CHECKSUM_NONE. 3777 */ 3778 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3779 { 3780 skb_push(skb, len); 3781 skb_postpush_rcsum(skb, skb->data, len); 3782 return skb->data; 3783 } 3784 3785 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 3786 /** 3787 * pskb_trim_rcsum - trim received skb and update checksum 3788 * @skb: buffer to trim 3789 * @len: new length 3790 * 3791 * This is exactly the same as pskb_trim except that it ensures the 3792 * checksum of received packets are still valid after the operation. 3793 * It can change skb pointers. 3794 */ 3795 3796 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3797 { 3798 if (likely(len >= skb->len)) 3799 return 0; 3800 return pskb_trim_rcsum_slow(skb, len); 3801 } 3802 3803 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3804 { 3805 if (skb->ip_summed == CHECKSUM_COMPLETE) 3806 skb->ip_summed = CHECKSUM_NONE; 3807 __skb_trim(skb, len); 3808 return 0; 3809 } 3810 3811 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3812 { 3813 if (skb->ip_summed == CHECKSUM_COMPLETE) 3814 skb->ip_summed = CHECKSUM_NONE; 3815 return __skb_grow(skb, len); 3816 } 3817 3818 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3819 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 3820 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 3821 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3822 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3823 3824 #define skb_queue_walk(queue, skb) \ 3825 for (skb = (queue)->next; \ 3826 skb != (struct sk_buff *)(queue); \ 3827 skb = skb->next) 3828 3829 #define skb_queue_walk_safe(queue, skb, tmp) \ 3830 for (skb = (queue)->next, tmp = skb->next; \ 3831 skb != (struct sk_buff *)(queue); \ 3832 skb = tmp, tmp = skb->next) 3833 3834 #define skb_queue_walk_from(queue, skb) \ 3835 for (; skb != (struct sk_buff *)(queue); \ 3836 skb = skb->next) 3837 3838 #define skb_rbtree_walk(skb, root) \ 3839 for (skb = skb_rb_first(root); skb != NULL; \ 3840 skb = skb_rb_next(skb)) 3841 3842 #define skb_rbtree_walk_from(skb) \ 3843 for (; skb != NULL; \ 3844 skb = skb_rb_next(skb)) 3845 3846 #define skb_rbtree_walk_from_safe(skb, tmp) \ 3847 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3848 skb = tmp) 3849 3850 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 3851 for (tmp = skb->next; \ 3852 skb != (struct sk_buff *)(queue); \ 3853 skb = tmp, tmp = skb->next) 3854 3855 #define skb_queue_reverse_walk(queue, skb) \ 3856 for (skb = (queue)->prev; \ 3857 skb != (struct sk_buff *)(queue); \ 3858 skb = skb->prev) 3859 3860 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3861 for (skb = (queue)->prev, tmp = skb->prev; \ 3862 skb != (struct sk_buff *)(queue); \ 3863 skb = tmp, tmp = skb->prev) 3864 3865 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3866 for (tmp = skb->prev; \ 3867 skb != (struct sk_buff *)(queue); \ 3868 skb = tmp, tmp = skb->prev) 3869 3870 static inline bool skb_has_frag_list(const struct sk_buff *skb) 3871 { 3872 return skb_shinfo(skb)->frag_list != NULL; 3873 } 3874 3875 static inline void skb_frag_list_init(struct sk_buff *skb) 3876 { 3877 skb_shinfo(skb)->frag_list = NULL; 3878 } 3879 3880 #define skb_walk_frags(skb, iter) \ 3881 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3882 3883 3884 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 3885 int *err, long *timeo_p, 3886 const struct sk_buff *skb); 3887 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3888 struct sk_buff_head *queue, 3889 unsigned int flags, 3890 int *off, int *err, 3891 struct sk_buff **last); 3892 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 3893 struct sk_buff_head *queue, 3894 unsigned int flags, int *off, int *err, 3895 struct sk_buff **last); 3896 struct sk_buff *__skb_recv_datagram(struct sock *sk, 3897 struct sk_buff_head *sk_queue, 3898 unsigned int flags, int *off, int *err); 3899 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 3900 __poll_t datagram_poll(struct file *file, struct socket *sock, 3901 struct poll_table_struct *wait); 3902 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3903 struct iov_iter *to, int size); 3904 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3905 struct msghdr *msg, int size) 3906 { 3907 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3908 } 3909 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3910 struct msghdr *msg); 3911 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 3912 struct iov_iter *to, int len, 3913 struct ahash_request *hash); 3914 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3915 struct iov_iter *from, int len); 3916 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3917 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3918 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3919 static inline void skb_free_datagram_locked(struct sock *sk, 3920 struct sk_buff *skb) 3921 { 3922 __skb_free_datagram_locked(sk, skb, 0); 3923 } 3924 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3925 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3926 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3927 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3928 int len); 3929 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3930 struct pipe_inode_info *pipe, unsigned int len, 3931 unsigned int flags); 3932 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 3933 int len); 3934 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 3935 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3936 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3937 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3938 int len, int hlen); 3939 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3940 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3941 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3942 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu); 3943 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len); 3944 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3945 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 3946 unsigned int offset); 3947 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3948 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 3949 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3950 int skb_vlan_pop(struct sk_buff *skb); 3951 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3952 int skb_eth_pop(struct sk_buff *skb); 3953 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 3954 const unsigned char *src); 3955 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 3956 int mac_len, bool ethernet); 3957 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 3958 bool ethernet); 3959 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 3960 int skb_mpls_dec_ttl(struct sk_buff *skb); 3961 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3962 gfp_t gfp); 3963 3964 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3965 { 3966 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 3967 } 3968 3969 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3970 { 3971 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3972 } 3973 3974 struct skb_checksum_ops { 3975 __wsum (*update)(const void *mem, int len, __wsum wsum); 3976 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3977 }; 3978 3979 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 3980 3981 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3982 __wsum csum, const struct skb_checksum_ops *ops); 3983 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3984 __wsum csum); 3985 3986 static inline void * __must_check 3987 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 3988 const void *data, int hlen, void *buffer) 3989 { 3990 if (likely(hlen - offset >= len)) 3991 return (void *)data + offset; 3992 3993 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 3994 return NULL; 3995 3996 return buffer; 3997 } 3998 3999 static inline void * __must_check 4000 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4001 { 4002 return __skb_header_pointer(skb, offset, len, skb->data, 4003 skb_headlen(skb), buffer); 4004 } 4005 4006 /** 4007 * skb_needs_linearize - check if we need to linearize a given skb 4008 * depending on the given device features. 4009 * @skb: socket buffer to check 4010 * @features: net device features 4011 * 4012 * Returns true if either: 4013 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4014 * 2. skb is fragmented and the device does not support SG. 4015 */ 4016 static inline bool skb_needs_linearize(struct sk_buff *skb, 4017 netdev_features_t features) 4018 { 4019 return skb_is_nonlinear(skb) && 4020 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4021 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4022 } 4023 4024 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4025 void *to, 4026 const unsigned int len) 4027 { 4028 memcpy(to, skb->data, len); 4029 } 4030 4031 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4032 const int offset, void *to, 4033 const unsigned int len) 4034 { 4035 memcpy(to, skb->data + offset, len); 4036 } 4037 4038 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4039 const void *from, 4040 const unsigned int len) 4041 { 4042 memcpy(skb->data, from, len); 4043 } 4044 4045 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4046 const int offset, 4047 const void *from, 4048 const unsigned int len) 4049 { 4050 memcpy(skb->data + offset, from, len); 4051 } 4052 4053 void skb_init(void); 4054 4055 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4056 { 4057 return skb->tstamp; 4058 } 4059 4060 /** 4061 * skb_get_timestamp - get timestamp from a skb 4062 * @skb: skb to get stamp from 4063 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4064 * 4065 * Timestamps are stored in the skb as offsets to a base timestamp. 4066 * This function converts the offset back to a struct timeval and stores 4067 * it in stamp. 4068 */ 4069 static inline void skb_get_timestamp(const struct sk_buff *skb, 4070 struct __kernel_old_timeval *stamp) 4071 { 4072 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4073 } 4074 4075 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4076 struct __kernel_sock_timeval *stamp) 4077 { 4078 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4079 4080 stamp->tv_sec = ts.tv_sec; 4081 stamp->tv_usec = ts.tv_nsec / 1000; 4082 } 4083 4084 static inline void skb_get_timestampns(const struct sk_buff *skb, 4085 struct __kernel_old_timespec *stamp) 4086 { 4087 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4088 4089 stamp->tv_sec = ts.tv_sec; 4090 stamp->tv_nsec = ts.tv_nsec; 4091 } 4092 4093 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4094 struct __kernel_timespec *stamp) 4095 { 4096 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4097 4098 stamp->tv_sec = ts.tv_sec; 4099 stamp->tv_nsec = ts.tv_nsec; 4100 } 4101 4102 static inline void __net_timestamp(struct sk_buff *skb) 4103 { 4104 skb->tstamp = ktime_get_real(); 4105 skb->mono_delivery_time = 0; 4106 } 4107 4108 static inline ktime_t net_timedelta(ktime_t t) 4109 { 4110 return ktime_sub(ktime_get_real(), t); 4111 } 4112 4113 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4114 bool mono) 4115 { 4116 skb->tstamp = kt; 4117 skb->mono_delivery_time = kt && mono; 4118 } 4119 4120 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4121 4122 /* It is used in the ingress path to clear the delivery_time. 4123 * If needed, set the skb->tstamp to the (rcv) timestamp. 4124 */ 4125 static inline void skb_clear_delivery_time(struct sk_buff *skb) 4126 { 4127 if (skb->mono_delivery_time) { 4128 skb->mono_delivery_time = 0; 4129 if (static_branch_unlikely(&netstamp_needed_key)) 4130 skb->tstamp = ktime_get_real(); 4131 else 4132 skb->tstamp = 0; 4133 } 4134 } 4135 4136 static inline void skb_clear_tstamp(struct sk_buff *skb) 4137 { 4138 if (skb->mono_delivery_time) 4139 return; 4140 4141 skb->tstamp = 0; 4142 } 4143 4144 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4145 { 4146 if (skb->mono_delivery_time) 4147 return 0; 4148 4149 return skb->tstamp; 4150 } 4151 4152 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4153 { 4154 if (!skb->mono_delivery_time && skb->tstamp) 4155 return skb->tstamp; 4156 4157 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4158 return ktime_get_real(); 4159 4160 return 0; 4161 } 4162 4163 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4164 { 4165 return skb_shinfo(skb)->meta_len; 4166 } 4167 4168 static inline void *skb_metadata_end(const struct sk_buff *skb) 4169 { 4170 return skb_mac_header(skb); 4171 } 4172 4173 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4174 const struct sk_buff *skb_b, 4175 u8 meta_len) 4176 { 4177 const void *a = skb_metadata_end(skb_a); 4178 const void *b = skb_metadata_end(skb_b); 4179 /* Using more efficient varaiant than plain call to memcmp(). */ 4180 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64 4181 u64 diffs = 0; 4182 4183 switch (meta_len) { 4184 #define __it(x, op) (x -= sizeof(u##op)) 4185 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4186 case 32: diffs |= __it_diff(a, b, 64); 4187 fallthrough; 4188 case 24: diffs |= __it_diff(a, b, 64); 4189 fallthrough; 4190 case 16: diffs |= __it_diff(a, b, 64); 4191 fallthrough; 4192 case 8: diffs |= __it_diff(a, b, 64); 4193 break; 4194 case 28: diffs |= __it_diff(a, b, 64); 4195 fallthrough; 4196 case 20: diffs |= __it_diff(a, b, 64); 4197 fallthrough; 4198 case 12: diffs |= __it_diff(a, b, 64); 4199 fallthrough; 4200 case 4: diffs |= __it_diff(a, b, 32); 4201 break; 4202 } 4203 return diffs; 4204 #else 4205 return memcmp(a - meta_len, b - meta_len, meta_len); 4206 #endif 4207 } 4208 4209 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4210 const struct sk_buff *skb_b) 4211 { 4212 u8 len_a = skb_metadata_len(skb_a); 4213 u8 len_b = skb_metadata_len(skb_b); 4214 4215 if (!(len_a | len_b)) 4216 return false; 4217 4218 return len_a != len_b ? 4219 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4220 } 4221 4222 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4223 { 4224 skb_shinfo(skb)->meta_len = meta_len; 4225 } 4226 4227 static inline void skb_metadata_clear(struct sk_buff *skb) 4228 { 4229 skb_metadata_set(skb, 0); 4230 } 4231 4232 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4233 4234 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4235 4236 void skb_clone_tx_timestamp(struct sk_buff *skb); 4237 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4238 4239 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4240 4241 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4242 { 4243 } 4244 4245 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4246 { 4247 return false; 4248 } 4249 4250 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4251 4252 /** 4253 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4254 * 4255 * PHY drivers may accept clones of transmitted packets for 4256 * timestamping via their phy_driver.txtstamp method. These drivers 4257 * must call this function to return the skb back to the stack with a 4258 * timestamp. 4259 * 4260 * @skb: clone of the original outgoing packet 4261 * @hwtstamps: hardware time stamps 4262 * 4263 */ 4264 void skb_complete_tx_timestamp(struct sk_buff *skb, 4265 struct skb_shared_hwtstamps *hwtstamps); 4266 4267 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4268 struct skb_shared_hwtstamps *hwtstamps, 4269 struct sock *sk, int tstype); 4270 4271 /** 4272 * skb_tstamp_tx - queue clone of skb with send time stamps 4273 * @orig_skb: the original outgoing packet 4274 * @hwtstamps: hardware time stamps, may be NULL if not available 4275 * 4276 * If the skb has a socket associated, then this function clones the 4277 * skb (thus sharing the actual data and optional structures), stores 4278 * the optional hardware time stamping information (if non NULL) or 4279 * generates a software time stamp (otherwise), then queues the clone 4280 * to the error queue of the socket. Errors are silently ignored. 4281 */ 4282 void skb_tstamp_tx(struct sk_buff *orig_skb, 4283 struct skb_shared_hwtstamps *hwtstamps); 4284 4285 /** 4286 * skb_tx_timestamp() - Driver hook for transmit timestamping 4287 * 4288 * Ethernet MAC Drivers should call this function in their hard_xmit() 4289 * function immediately before giving the sk_buff to the MAC hardware. 4290 * 4291 * Specifically, one should make absolutely sure that this function is 4292 * called before TX completion of this packet can trigger. Otherwise 4293 * the packet could potentially already be freed. 4294 * 4295 * @skb: A socket buffer. 4296 */ 4297 static inline void skb_tx_timestamp(struct sk_buff *skb) 4298 { 4299 skb_clone_tx_timestamp(skb); 4300 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 4301 skb_tstamp_tx(skb, NULL); 4302 } 4303 4304 /** 4305 * skb_complete_wifi_ack - deliver skb with wifi status 4306 * 4307 * @skb: the original outgoing packet 4308 * @acked: ack status 4309 * 4310 */ 4311 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4312 4313 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4314 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4315 4316 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4317 { 4318 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4319 skb->csum_valid || 4320 (skb->ip_summed == CHECKSUM_PARTIAL && 4321 skb_checksum_start_offset(skb) >= 0)); 4322 } 4323 4324 /** 4325 * skb_checksum_complete - Calculate checksum of an entire packet 4326 * @skb: packet to process 4327 * 4328 * This function calculates the checksum over the entire packet plus 4329 * the value of skb->csum. The latter can be used to supply the 4330 * checksum of a pseudo header as used by TCP/UDP. It returns the 4331 * checksum. 4332 * 4333 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4334 * this function can be used to verify that checksum on received 4335 * packets. In that case the function should return zero if the 4336 * checksum is correct. In particular, this function will return zero 4337 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4338 * hardware has already verified the correctness of the checksum. 4339 */ 4340 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4341 { 4342 return skb_csum_unnecessary(skb) ? 4343 0 : __skb_checksum_complete(skb); 4344 } 4345 4346 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4347 { 4348 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4349 if (skb->csum_level == 0) 4350 skb->ip_summed = CHECKSUM_NONE; 4351 else 4352 skb->csum_level--; 4353 } 4354 } 4355 4356 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4357 { 4358 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4359 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4360 skb->csum_level++; 4361 } else if (skb->ip_summed == CHECKSUM_NONE) { 4362 skb->ip_summed = CHECKSUM_UNNECESSARY; 4363 skb->csum_level = 0; 4364 } 4365 } 4366 4367 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4368 { 4369 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4370 skb->ip_summed = CHECKSUM_NONE; 4371 skb->csum_level = 0; 4372 } 4373 } 4374 4375 /* Check if we need to perform checksum complete validation. 4376 * 4377 * Returns true if checksum complete is needed, false otherwise 4378 * (either checksum is unnecessary or zero checksum is allowed). 4379 */ 4380 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4381 bool zero_okay, 4382 __sum16 check) 4383 { 4384 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4385 skb->csum_valid = 1; 4386 __skb_decr_checksum_unnecessary(skb); 4387 return false; 4388 } 4389 4390 return true; 4391 } 4392 4393 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4394 * in checksum_init. 4395 */ 4396 #define CHECKSUM_BREAK 76 4397 4398 /* Unset checksum-complete 4399 * 4400 * Unset checksum complete can be done when packet is being modified 4401 * (uncompressed for instance) and checksum-complete value is 4402 * invalidated. 4403 */ 4404 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4405 { 4406 if (skb->ip_summed == CHECKSUM_COMPLETE) 4407 skb->ip_summed = CHECKSUM_NONE; 4408 } 4409 4410 /* Validate (init) checksum based on checksum complete. 4411 * 4412 * Return values: 4413 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4414 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4415 * checksum is stored in skb->csum for use in __skb_checksum_complete 4416 * non-zero: value of invalid checksum 4417 * 4418 */ 4419 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4420 bool complete, 4421 __wsum psum) 4422 { 4423 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4424 if (!csum_fold(csum_add(psum, skb->csum))) { 4425 skb->csum_valid = 1; 4426 return 0; 4427 } 4428 } 4429 4430 skb->csum = psum; 4431 4432 if (complete || skb->len <= CHECKSUM_BREAK) { 4433 __sum16 csum; 4434 4435 csum = __skb_checksum_complete(skb); 4436 skb->csum_valid = !csum; 4437 return csum; 4438 } 4439 4440 return 0; 4441 } 4442 4443 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4444 { 4445 return 0; 4446 } 4447 4448 /* Perform checksum validate (init). Note that this is a macro since we only 4449 * want to calculate the pseudo header which is an input function if necessary. 4450 * First we try to validate without any computation (checksum unnecessary) and 4451 * then calculate based on checksum complete calling the function to compute 4452 * pseudo header. 4453 * 4454 * Return values: 4455 * 0: checksum is validated or try to in skb_checksum_complete 4456 * non-zero: value of invalid checksum 4457 */ 4458 #define __skb_checksum_validate(skb, proto, complete, \ 4459 zero_okay, check, compute_pseudo) \ 4460 ({ \ 4461 __sum16 __ret = 0; \ 4462 skb->csum_valid = 0; \ 4463 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4464 __ret = __skb_checksum_validate_complete(skb, \ 4465 complete, compute_pseudo(skb, proto)); \ 4466 __ret; \ 4467 }) 4468 4469 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4470 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4471 4472 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4473 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4474 4475 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4476 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4477 4478 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4479 compute_pseudo) \ 4480 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4481 4482 #define skb_checksum_simple_validate(skb) \ 4483 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4484 4485 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4486 { 4487 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4488 } 4489 4490 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4491 { 4492 skb->csum = ~pseudo; 4493 skb->ip_summed = CHECKSUM_COMPLETE; 4494 } 4495 4496 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4497 do { \ 4498 if (__skb_checksum_convert_check(skb)) \ 4499 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4500 } while (0) 4501 4502 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4503 u16 start, u16 offset) 4504 { 4505 skb->ip_summed = CHECKSUM_PARTIAL; 4506 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4507 skb->csum_offset = offset - start; 4508 } 4509 4510 /* Update skbuf and packet to reflect the remote checksum offload operation. 4511 * When called, ptr indicates the starting point for skb->csum when 4512 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4513 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4514 */ 4515 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4516 int start, int offset, bool nopartial) 4517 { 4518 __wsum delta; 4519 4520 if (!nopartial) { 4521 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4522 return; 4523 } 4524 4525 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4526 __skb_checksum_complete(skb); 4527 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4528 } 4529 4530 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4531 4532 /* Adjust skb->csum since we changed the packet */ 4533 skb->csum = csum_add(skb->csum, delta); 4534 } 4535 4536 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4537 { 4538 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4539 return (void *)(skb->_nfct & NFCT_PTRMASK); 4540 #else 4541 return NULL; 4542 #endif 4543 } 4544 4545 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4546 { 4547 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4548 return skb->_nfct; 4549 #else 4550 return 0UL; 4551 #endif 4552 } 4553 4554 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4555 { 4556 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4557 skb->slow_gro |= !!nfct; 4558 skb->_nfct = nfct; 4559 #endif 4560 } 4561 4562 #ifdef CONFIG_SKB_EXTENSIONS 4563 enum skb_ext_id { 4564 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4565 SKB_EXT_BRIDGE_NF, 4566 #endif 4567 #ifdef CONFIG_XFRM 4568 SKB_EXT_SEC_PATH, 4569 #endif 4570 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4571 TC_SKB_EXT, 4572 #endif 4573 #if IS_ENABLED(CONFIG_MPTCP) 4574 SKB_EXT_MPTCP, 4575 #endif 4576 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 4577 SKB_EXT_MCTP, 4578 #endif 4579 SKB_EXT_NUM, /* must be last */ 4580 }; 4581 4582 /** 4583 * struct skb_ext - sk_buff extensions 4584 * @refcnt: 1 on allocation, deallocated on 0 4585 * @offset: offset to add to @data to obtain extension address 4586 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4587 * @data: start of extension data, variable sized 4588 * 4589 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4590 * to use 'u8' types while allowing up to 2kb worth of extension data. 4591 */ 4592 struct skb_ext { 4593 refcount_t refcnt; 4594 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4595 u8 chunks; /* same */ 4596 char data[] __aligned(8); 4597 }; 4598 4599 struct skb_ext *__skb_ext_alloc(gfp_t flags); 4600 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 4601 struct skb_ext *ext); 4602 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4603 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4604 void __skb_ext_put(struct skb_ext *ext); 4605 4606 static inline void skb_ext_put(struct sk_buff *skb) 4607 { 4608 if (skb->active_extensions) 4609 __skb_ext_put(skb->extensions); 4610 } 4611 4612 static inline void __skb_ext_copy(struct sk_buff *dst, 4613 const struct sk_buff *src) 4614 { 4615 dst->active_extensions = src->active_extensions; 4616 4617 if (src->active_extensions) { 4618 struct skb_ext *ext = src->extensions; 4619 4620 refcount_inc(&ext->refcnt); 4621 dst->extensions = ext; 4622 } 4623 } 4624 4625 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4626 { 4627 skb_ext_put(dst); 4628 __skb_ext_copy(dst, src); 4629 } 4630 4631 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4632 { 4633 return !!ext->offset[i]; 4634 } 4635 4636 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4637 { 4638 return skb->active_extensions & (1 << id); 4639 } 4640 4641 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4642 { 4643 if (skb_ext_exist(skb, id)) 4644 __skb_ext_del(skb, id); 4645 } 4646 4647 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4648 { 4649 if (skb_ext_exist(skb, id)) { 4650 struct skb_ext *ext = skb->extensions; 4651 4652 return (void *)ext + (ext->offset[id] << 3); 4653 } 4654 4655 return NULL; 4656 } 4657 4658 static inline void skb_ext_reset(struct sk_buff *skb) 4659 { 4660 if (unlikely(skb->active_extensions)) { 4661 __skb_ext_put(skb->extensions); 4662 skb->active_extensions = 0; 4663 } 4664 } 4665 4666 static inline bool skb_has_extensions(struct sk_buff *skb) 4667 { 4668 return unlikely(skb->active_extensions); 4669 } 4670 #else 4671 static inline void skb_ext_put(struct sk_buff *skb) {} 4672 static inline void skb_ext_reset(struct sk_buff *skb) {} 4673 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4674 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4675 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4676 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 4677 #endif /* CONFIG_SKB_EXTENSIONS */ 4678 4679 static inline void nf_reset_ct(struct sk_buff *skb) 4680 { 4681 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4682 nf_conntrack_put(skb_nfct(skb)); 4683 skb->_nfct = 0; 4684 #endif 4685 } 4686 4687 static inline void nf_reset_trace(struct sk_buff *skb) 4688 { 4689 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4690 skb->nf_trace = 0; 4691 #endif 4692 } 4693 4694 static inline void ipvs_reset(struct sk_buff *skb) 4695 { 4696 #if IS_ENABLED(CONFIG_IP_VS) 4697 skb->ipvs_property = 0; 4698 #endif 4699 } 4700 4701 /* Note: This doesn't put any conntrack info in dst. */ 4702 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4703 bool copy) 4704 { 4705 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4706 dst->_nfct = src->_nfct; 4707 nf_conntrack_get(skb_nfct(src)); 4708 #endif 4709 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4710 if (copy) 4711 dst->nf_trace = src->nf_trace; 4712 #endif 4713 } 4714 4715 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 4716 { 4717 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4718 nf_conntrack_put(skb_nfct(dst)); 4719 #endif 4720 dst->slow_gro = src->slow_gro; 4721 __nf_copy(dst, src, true); 4722 } 4723 4724 #ifdef CONFIG_NETWORK_SECMARK 4725 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4726 { 4727 to->secmark = from->secmark; 4728 } 4729 4730 static inline void skb_init_secmark(struct sk_buff *skb) 4731 { 4732 skb->secmark = 0; 4733 } 4734 #else 4735 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4736 { } 4737 4738 static inline void skb_init_secmark(struct sk_buff *skb) 4739 { } 4740 #endif 4741 4742 static inline int secpath_exists(const struct sk_buff *skb) 4743 { 4744 #ifdef CONFIG_XFRM 4745 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 4746 #else 4747 return 0; 4748 #endif 4749 } 4750 4751 static inline bool skb_irq_freeable(const struct sk_buff *skb) 4752 { 4753 return !skb->destructor && 4754 !secpath_exists(skb) && 4755 !skb_nfct(skb) && 4756 !skb->_skb_refdst && 4757 !skb_has_frag_list(skb); 4758 } 4759 4760 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 4761 { 4762 skb->queue_mapping = queue_mapping; 4763 } 4764 4765 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 4766 { 4767 return skb->queue_mapping; 4768 } 4769 4770 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 4771 { 4772 to->queue_mapping = from->queue_mapping; 4773 } 4774 4775 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 4776 { 4777 skb->queue_mapping = rx_queue + 1; 4778 } 4779 4780 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 4781 { 4782 return skb->queue_mapping - 1; 4783 } 4784 4785 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 4786 { 4787 return skb->queue_mapping != 0; 4788 } 4789 4790 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 4791 { 4792 skb->dst_pending_confirm = val; 4793 } 4794 4795 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 4796 { 4797 return skb->dst_pending_confirm != 0; 4798 } 4799 4800 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 4801 { 4802 #ifdef CONFIG_XFRM 4803 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 4804 #else 4805 return NULL; 4806 #endif 4807 } 4808 4809 /* Keeps track of mac header offset relative to skb->head. 4810 * It is useful for TSO of Tunneling protocol. e.g. GRE. 4811 * For non-tunnel skb it points to skb_mac_header() and for 4812 * tunnel skb it points to outer mac header. 4813 * Keeps track of level of encapsulation of network headers. 4814 */ 4815 struct skb_gso_cb { 4816 union { 4817 int mac_offset; 4818 int data_offset; 4819 }; 4820 int encap_level; 4821 __wsum csum; 4822 __u16 csum_start; 4823 }; 4824 #define SKB_GSO_CB_OFFSET 32 4825 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET)) 4826 4827 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 4828 { 4829 return (skb_mac_header(inner_skb) - inner_skb->head) - 4830 SKB_GSO_CB(inner_skb)->mac_offset; 4831 } 4832 4833 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 4834 { 4835 int new_headroom, headroom; 4836 int ret; 4837 4838 headroom = skb_headroom(skb); 4839 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 4840 if (ret) 4841 return ret; 4842 4843 new_headroom = skb_headroom(skb); 4844 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 4845 return 0; 4846 } 4847 4848 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 4849 { 4850 /* Do not update partial checksums if remote checksum is enabled. */ 4851 if (skb->remcsum_offload) 4852 return; 4853 4854 SKB_GSO_CB(skb)->csum = res; 4855 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 4856 } 4857 4858 /* Compute the checksum for a gso segment. First compute the checksum value 4859 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 4860 * then add in skb->csum (checksum from csum_start to end of packet). 4861 * skb->csum and csum_start are then updated to reflect the checksum of the 4862 * resultant packet starting from the transport header-- the resultant checksum 4863 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 4864 * header. 4865 */ 4866 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 4867 { 4868 unsigned char *csum_start = skb_transport_header(skb); 4869 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 4870 __wsum partial = SKB_GSO_CB(skb)->csum; 4871 4872 SKB_GSO_CB(skb)->csum = res; 4873 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 4874 4875 return csum_fold(csum_partial(csum_start, plen, partial)); 4876 } 4877 4878 static inline bool skb_is_gso(const struct sk_buff *skb) 4879 { 4880 return skb_shinfo(skb)->gso_size; 4881 } 4882 4883 /* Note: Should be called only if skb_is_gso(skb) is true */ 4884 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4885 { 4886 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4887 } 4888 4889 /* Note: Should be called only if skb_is_gso(skb) is true */ 4890 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 4891 { 4892 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 4893 } 4894 4895 /* Note: Should be called only if skb_is_gso(skb) is true */ 4896 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 4897 { 4898 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 4899 } 4900 4901 static inline void skb_gso_reset(struct sk_buff *skb) 4902 { 4903 skb_shinfo(skb)->gso_size = 0; 4904 skb_shinfo(skb)->gso_segs = 0; 4905 skb_shinfo(skb)->gso_type = 0; 4906 } 4907 4908 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 4909 u16 increment) 4910 { 4911 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4912 return; 4913 shinfo->gso_size += increment; 4914 } 4915 4916 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 4917 u16 decrement) 4918 { 4919 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4920 return; 4921 shinfo->gso_size -= decrement; 4922 } 4923 4924 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 4925 4926 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 4927 { 4928 /* LRO sets gso_size but not gso_type, whereas if GSO is really 4929 * wanted then gso_type will be set. */ 4930 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4931 4932 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 4933 unlikely(shinfo->gso_type == 0)) { 4934 __skb_warn_lro_forwarding(skb); 4935 return true; 4936 } 4937 return false; 4938 } 4939 4940 static inline void skb_forward_csum(struct sk_buff *skb) 4941 { 4942 /* Unfortunately we don't support this one. Any brave souls? */ 4943 if (skb->ip_summed == CHECKSUM_COMPLETE) 4944 skb->ip_summed = CHECKSUM_NONE; 4945 } 4946 4947 /** 4948 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 4949 * @skb: skb to check 4950 * 4951 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 4952 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 4953 * use this helper, to document places where we make this assertion. 4954 */ 4955 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 4956 { 4957 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 4958 } 4959 4960 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 4961 4962 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 4963 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4964 unsigned int transport_len, 4965 __sum16(*skb_chkf)(struct sk_buff *skb)); 4966 4967 /** 4968 * skb_head_is_locked - Determine if the skb->head is locked down 4969 * @skb: skb to check 4970 * 4971 * The head on skbs build around a head frag can be removed if they are 4972 * not cloned. This function returns true if the skb head is locked down 4973 * due to either being allocated via kmalloc, or by being a clone with 4974 * multiple references to the head. 4975 */ 4976 static inline bool skb_head_is_locked(const struct sk_buff *skb) 4977 { 4978 return !skb->head_frag || skb_cloned(skb); 4979 } 4980 4981 /* Local Checksum Offload. 4982 * Compute outer checksum based on the assumption that the 4983 * inner checksum will be offloaded later. 4984 * See Documentation/networking/checksum-offloads.rst for 4985 * explanation of how this works. 4986 * Fill in outer checksum adjustment (e.g. with sum of outer 4987 * pseudo-header) before calling. 4988 * Also ensure that inner checksum is in linear data area. 4989 */ 4990 static inline __wsum lco_csum(struct sk_buff *skb) 4991 { 4992 unsigned char *csum_start = skb_checksum_start(skb); 4993 unsigned char *l4_hdr = skb_transport_header(skb); 4994 __wsum partial; 4995 4996 /* Start with complement of inner checksum adjustment */ 4997 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 4998 skb->csum_offset)); 4999 5000 /* Add in checksum of our headers (incl. outer checksum 5001 * adjustment filled in by caller) and return result. 5002 */ 5003 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5004 } 5005 5006 static inline bool skb_is_redirected(const struct sk_buff *skb) 5007 { 5008 return skb->redirected; 5009 } 5010 5011 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5012 { 5013 skb->redirected = 1; 5014 #ifdef CONFIG_NET_REDIRECT 5015 skb->from_ingress = from_ingress; 5016 if (skb->from_ingress) 5017 skb_clear_tstamp(skb); 5018 #endif 5019 } 5020 5021 static inline void skb_reset_redirect(struct sk_buff *skb) 5022 { 5023 skb->redirected = 0; 5024 } 5025 5026 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5027 { 5028 return skb->csum_not_inet; 5029 } 5030 5031 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5032 const u64 kcov_handle) 5033 { 5034 #ifdef CONFIG_KCOV 5035 skb->kcov_handle = kcov_handle; 5036 #endif 5037 } 5038 5039 static inline u64 skb_get_kcov_handle(struct sk_buff *skb) 5040 { 5041 #ifdef CONFIG_KCOV 5042 return skb->kcov_handle; 5043 #else 5044 return 0; 5045 #endif 5046 } 5047 5048 #ifdef CONFIG_PAGE_POOL 5049 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5050 { 5051 skb->pp_recycle = 1; 5052 } 5053 #endif 5054 5055 #endif /* __KERNEL__ */ 5056 #endif /* _LINUX_SKBUFF_H */ 5057