1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * Definitions for the 'struct sk_buff' memory handlers. 4 * 5 * Authors: 6 * Alan Cox, <gw4pts@gw4pts.ampr.org> 7 * Florian La Roche, <rzsfl@rz.uni-sb.de> 8 */ 9 10 #ifndef _LINUX_SKBUFF_H 11 #define _LINUX_SKBUFF_H 12 13 #include <linux/kernel.h> 14 #include <linux/compiler.h> 15 #include <linux/time.h> 16 #include <linux/bug.h> 17 #include <linux/bvec.h> 18 #include <linux/cache.h> 19 #include <linux/rbtree.h> 20 #include <linux/socket.h> 21 #include <linux/refcount.h> 22 23 #include <linux/atomic.h> 24 #include <asm/types.h> 25 #include <linux/spinlock.h> 26 #include <net/checksum.h> 27 #include <linux/rcupdate.h> 28 #include <linux/dma-mapping.h> 29 #include <linux/netdev_features.h> 30 #include <net/flow_dissector.h> 31 #include <linux/in6.h> 32 #include <linux/if_packet.h> 33 #include <linux/llist.h> 34 #include <net/flow.h> 35 #include <net/page_pool.h> 36 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 37 #include <linux/netfilter/nf_conntrack_common.h> 38 #endif 39 #include <net/net_debug.h> 40 #include <net/dropreason.h> 41 42 /** 43 * DOC: skb checksums 44 * 45 * The interface for checksum offload between the stack and networking drivers 46 * is as follows... 47 * 48 * IP checksum related features 49 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 50 * 51 * Drivers advertise checksum offload capabilities in the features of a device. 52 * From the stack's point of view these are capabilities offered by the driver. 53 * A driver typically only advertises features that it is capable of offloading 54 * to its device. 55 * 56 * .. flat-table:: Checksum related device features 57 * :widths: 1 10 58 * 59 * * - %NETIF_F_HW_CSUM 60 * - The driver (or its device) is able to compute one 61 * IP (one's complement) checksum for any combination 62 * of protocols or protocol layering. The checksum is 63 * computed and set in a packet per the CHECKSUM_PARTIAL 64 * interface (see below). 65 * 66 * * - %NETIF_F_IP_CSUM 67 * - Driver (device) is only able to checksum plain 68 * TCP or UDP packets over IPv4. These are specifically 69 * unencapsulated packets of the form IPv4|TCP or 70 * IPv4|UDP where the Protocol field in the IPv4 header 71 * is TCP or UDP. The IPv4 header may contain IP options. 72 * This feature cannot be set in features for a device 73 * with NETIF_F_HW_CSUM also set. This feature is being 74 * DEPRECATED (see below). 75 * 76 * * - %NETIF_F_IPV6_CSUM 77 * - Driver (device) is only able to checksum plain 78 * TCP or UDP packets over IPv6. These are specifically 79 * unencapsulated packets of the form IPv6|TCP or 80 * IPv6|UDP where the Next Header field in the IPv6 81 * header is either TCP or UDP. IPv6 extension headers 82 * are not supported with this feature. This feature 83 * cannot be set in features for a device with 84 * NETIF_F_HW_CSUM also set. This feature is being 85 * DEPRECATED (see below). 86 * 87 * * - %NETIF_F_RXCSUM 88 * - Driver (device) performs receive checksum offload. 89 * This flag is only used to disable the RX checksum 90 * feature for a device. The stack will accept receive 91 * checksum indication in packets received on a device 92 * regardless of whether NETIF_F_RXCSUM is set. 93 * 94 * Checksumming of received packets by device 95 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 96 * 97 * Indication of checksum verification is set in &sk_buff.ip_summed. 98 * Possible values are: 99 * 100 * - %CHECKSUM_NONE 101 * 102 * Device did not checksum this packet e.g. due to lack of capabilities. 103 * The packet contains full (though not verified) checksum in packet but 104 * not in skb->csum. Thus, skb->csum is undefined in this case. 105 * 106 * - %CHECKSUM_UNNECESSARY 107 * 108 * The hardware you're dealing with doesn't calculate the full checksum 109 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums 110 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY 111 * if their checksums are okay. &sk_buff.csum is still undefined in this case 112 * though. A driver or device must never modify the checksum field in the 113 * packet even if checksum is verified. 114 * 115 * %CHECKSUM_UNNECESSARY is applicable to following protocols: 116 * 117 * - TCP: IPv6 and IPv4. 118 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 119 * zero UDP checksum for either IPv4 or IPv6, the networking stack 120 * may perform further validation in this case. 121 * - GRE: only if the checksum is present in the header. 122 * - SCTP: indicates the CRC in SCTP header has been validated. 123 * - FCOE: indicates the CRC in FC frame has been validated. 124 * 125 * &sk_buff.csum_level indicates the number of consecutive checksums found in 126 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY. 127 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 128 * and a device is able to verify the checksums for UDP (possibly zero), 129 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to 130 * two. If the device were only able to verify the UDP checksum and not 131 * GRE, either because it doesn't support GRE checksum or because GRE 132 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 133 * not considered in this case). 134 * 135 * - %CHECKSUM_COMPLETE 136 * 137 * This is the most generic way. The device supplied checksum of the _whole_ 138 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the 139 * hardware doesn't need to parse L3/L4 headers to implement this. 140 * 141 * Notes: 142 * 143 * - Even if device supports only some protocols, but is able to produce 144 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 145 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 146 * 147 * - %CHECKSUM_PARTIAL 148 * 149 * A checksum is set up to be offloaded to a device as described in the 150 * output description for CHECKSUM_PARTIAL. This may occur on a packet 151 * received directly from another Linux OS, e.g., a virtualized Linux kernel 152 * on the same host, or it may be set in the input path in GRO or remote 153 * checksum offload. For the purposes of checksum verification, the checksum 154 * referred to by skb->csum_start + skb->csum_offset and any preceding 155 * checksums in the packet are considered verified. Any checksums in the 156 * packet that are after the checksum being offloaded are not considered to 157 * be verified. 158 * 159 * Checksumming on transmit for non-GSO 160 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 161 * 162 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet. 163 * Values are: 164 * 165 * - %CHECKSUM_PARTIAL 166 * 167 * The driver is required to checksum the packet as seen by hard_start_xmit() 168 * from &sk_buff.csum_start up to the end, and to record/write the checksum at 169 * offset &sk_buff.csum_start + &sk_buff.csum_offset. 170 * A driver may verify that the 171 * csum_start and csum_offset values are valid values given the length and 172 * offset of the packet, but it should not attempt to validate that the 173 * checksum refers to a legitimate transport layer checksum -- it is the 174 * purview of the stack to validate that csum_start and csum_offset are set 175 * correctly. 176 * 177 * When the stack requests checksum offload for a packet, the driver MUST 178 * ensure that the checksum is set correctly. A driver can either offload the 179 * checksum calculation to the device, or call skb_checksum_help (in the case 180 * that the device does not support offload for a particular checksum). 181 * 182 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of 183 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate 184 * checksum offload capability. 185 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based 186 * on network device checksumming capabilities: if a packet does not match 187 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of 188 * &sk_buff.csum_not_inet, see :ref:`crc`) 189 * is called to resolve the checksum. 190 * 191 * - %CHECKSUM_NONE 192 * 193 * The skb was already checksummed by the protocol, or a checksum is not 194 * required. 195 * 196 * - %CHECKSUM_UNNECESSARY 197 * 198 * This has the same meaning as CHECKSUM_NONE for checksum offload on 199 * output. 200 * 201 * - %CHECKSUM_COMPLETE 202 * 203 * Not used in checksum output. If a driver observes a packet with this value 204 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set. 205 * 206 * .. _crc: 207 * 208 * Non-IP checksum (CRC) offloads 209 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 210 * 211 * .. flat-table:: 212 * :widths: 1 10 213 * 214 * * - %NETIF_F_SCTP_CRC 215 * - This feature indicates that a device is capable of 216 * offloading the SCTP CRC in a packet. To perform this offload the stack 217 * will set csum_start and csum_offset accordingly, set ip_summed to 218 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication 219 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c. 220 * A driver that supports both IP checksum offload and SCTP CRC32c offload 221 * must verify which offload is configured for a packet by testing the 222 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to 223 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 224 * 225 * * - %NETIF_F_FCOE_CRC 226 * - This feature indicates that a device is capable of offloading the FCOE 227 * CRC in a packet. To perform this offload the stack will set ip_summed 228 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset 229 * accordingly. Note that there is no indication in the skbuff that the 230 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports 231 * both IP checksum offload and FCOE CRC offload must verify which offload 232 * is configured for a packet, presumably by inspecting packet headers. 233 * 234 * Checksumming on output with GSO 235 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 236 * 237 * In the case of a GSO packet (skb_is_gso() is true), checksum offload 238 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 239 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as 240 * part of the GSO operation is implied. If a checksum is being offloaded 241 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and 242 * csum_offset are set to refer to the outermost checksum being offloaded 243 * (two offloaded checksums are possible with UDP encapsulation). 244 */ 245 246 /* Don't change this without changing skb_csum_unnecessary! */ 247 #define CHECKSUM_NONE 0 248 #define CHECKSUM_UNNECESSARY 1 249 #define CHECKSUM_COMPLETE 2 250 #define CHECKSUM_PARTIAL 3 251 252 /* Maximum value in skb->csum_level */ 253 #define SKB_MAX_CSUM_LEVEL 3 254 255 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 256 #define SKB_WITH_OVERHEAD(X) \ 257 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 258 259 /* For X bytes available in skb->head, what is the minimal 260 * allocation needed, knowing struct skb_shared_info needs 261 * to be aligned. 262 */ 263 #define SKB_HEAD_ALIGN(X) (SKB_DATA_ALIGN(X) + \ 264 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 265 266 #define SKB_MAX_ORDER(X, ORDER) \ 267 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 268 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 269 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 270 271 /* return minimum truesize of one skb containing X bytes of data */ 272 #define SKB_TRUESIZE(X) ((X) + \ 273 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 274 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 275 276 struct ahash_request; 277 struct net_device; 278 struct scatterlist; 279 struct pipe_inode_info; 280 struct iov_iter; 281 struct napi_struct; 282 struct bpf_prog; 283 union bpf_attr; 284 struct skb_ext; 285 struct ts_config; 286 287 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 288 struct nf_bridge_info { 289 enum { 290 BRNF_PROTO_UNCHANGED, 291 BRNF_PROTO_8021Q, 292 BRNF_PROTO_PPPOE 293 } orig_proto:8; 294 u8 pkt_otherhost:1; 295 u8 in_prerouting:1; 296 u8 bridged_dnat:1; 297 __u16 frag_max_size; 298 struct net_device *physindev; 299 300 /* always valid & non-NULL from FORWARD on, for physdev match */ 301 struct net_device *physoutdev; 302 union { 303 /* prerouting: detect dnat in orig/reply direction */ 304 __be32 ipv4_daddr; 305 struct in6_addr ipv6_daddr; 306 307 /* after prerouting + nat detected: store original source 308 * mac since neigh resolution overwrites it, only used while 309 * skb is out in neigh layer. 310 */ 311 char neigh_header[8]; 312 }; 313 }; 314 #endif 315 316 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 317 /* Chain in tc_skb_ext will be used to share the tc chain with 318 * ovs recirc_id. It will be set to the current chain by tc 319 * and read by ovs to recirc_id. 320 */ 321 struct tc_skb_ext { 322 union { 323 u64 act_miss_cookie; 324 __u32 chain; 325 }; 326 __u16 mru; 327 __u16 zone; 328 u8 post_ct:1; 329 u8 post_ct_snat:1; 330 u8 post_ct_dnat:1; 331 u8 act_miss:1; /* Set if act_miss_cookie is used */ 332 }; 333 #endif 334 335 struct sk_buff_head { 336 /* These two members must be first to match sk_buff. */ 337 struct_group_tagged(sk_buff_list, list, 338 struct sk_buff *next; 339 struct sk_buff *prev; 340 ); 341 342 __u32 qlen; 343 spinlock_t lock; 344 }; 345 346 struct sk_buff; 347 348 /* To allow 64K frame to be packed as single skb without frag_list we 349 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 350 * buffers which do not start on a page boundary. 351 * 352 * Since GRO uses frags we allocate at least 16 regardless of page 353 * size. 354 */ 355 #if (65536/PAGE_SIZE + 1) < 16 356 #define MAX_SKB_FRAGS 16UL 357 #else 358 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 359 #endif 360 extern int sysctl_max_skb_frags; 361 362 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 363 * segment using its current segmentation instead. 364 */ 365 #define GSO_BY_FRAGS 0xFFFF 366 367 typedef struct bio_vec skb_frag_t; 368 369 /** 370 * skb_frag_size() - Returns the size of a skb fragment 371 * @frag: skb fragment 372 */ 373 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 374 { 375 return frag->bv_len; 376 } 377 378 /** 379 * skb_frag_size_set() - Sets the size of a skb fragment 380 * @frag: skb fragment 381 * @size: size of fragment 382 */ 383 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 384 { 385 frag->bv_len = size; 386 } 387 388 /** 389 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 390 * @frag: skb fragment 391 * @delta: value to add 392 */ 393 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 394 { 395 frag->bv_len += delta; 396 } 397 398 /** 399 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 400 * @frag: skb fragment 401 * @delta: value to subtract 402 */ 403 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 404 { 405 frag->bv_len -= delta; 406 } 407 408 /** 409 * skb_frag_must_loop - Test if %p is a high memory page 410 * @p: fragment's page 411 */ 412 static inline bool skb_frag_must_loop(struct page *p) 413 { 414 #if defined(CONFIG_HIGHMEM) 415 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 416 return true; 417 #endif 418 return false; 419 } 420 421 /** 422 * skb_frag_foreach_page - loop over pages in a fragment 423 * 424 * @f: skb frag to operate on 425 * @f_off: offset from start of f->bv_page 426 * @f_len: length from f_off to loop over 427 * @p: (temp var) current page 428 * @p_off: (temp var) offset from start of current page, 429 * non-zero only on first page. 430 * @p_len: (temp var) length in current page, 431 * < PAGE_SIZE only on first and last page. 432 * @copied: (temp var) length so far, excluding current p_len. 433 * 434 * A fragment can hold a compound page, in which case per-page 435 * operations, notably kmap_atomic, must be called for each 436 * regular page. 437 */ 438 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 439 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 440 p_off = (f_off) & (PAGE_SIZE - 1), \ 441 p_len = skb_frag_must_loop(p) ? \ 442 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 443 copied = 0; \ 444 copied < f_len; \ 445 copied += p_len, p++, p_off = 0, \ 446 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 447 448 #define HAVE_HW_TIME_STAMP 449 450 /** 451 * struct skb_shared_hwtstamps - hardware time stamps 452 * @hwtstamp: hardware time stamp transformed into duration 453 * since arbitrary point in time 454 * @netdev_data: address/cookie of network device driver used as 455 * reference to actual hardware time stamp 456 * 457 * Software time stamps generated by ktime_get_real() are stored in 458 * skb->tstamp. 459 * 460 * hwtstamps can only be compared against other hwtstamps from 461 * the same device. 462 * 463 * This structure is attached to packets as part of the 464 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 465 */ 466 struct skb_shared_hwtstamps { 467 union { 468 ktime_t hwtstamp; 469 void *netdev_data; 470 }; 471 }; 472 473 /* Definitions for tx_flags in struct skb_shared_info */ 474 enum { 475 /* generate hardware time stamp */ 476 SKBTX_HW_TSTAMP = 1 << 0, 477 478 /* generate software time stamp when queueing packet to NIC */ 479 SKBTX_SW_TSTAMP = 1 << 1, 480 481 /* device driver is going to provide hardware time stamp */ 482 SKBTX_IN_PROGRESS = 1 << 2, 483 484 /* generate hardware time stamp based on cycles if supported */ 485 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3, 486 487 /* generate wifi status information (where possible) */ 488 SKBTX_WIFI_STATUS = 1 << 4, 489 490 /* determine hardware time stamp based on time or cycles */ 491 SKBTX_HW_TSTAMP_NETDEV = 1 << 5, 492 493 /* generate software time stamp when entering packet scheduling */ 494 SKBTX_SCHED_TSTAMP = 1 << 6, 495 }; 496 497 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 498 SKBTX_SCHED_TSTAMP) 499 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \ 500 SKBTX_HW_TSTAMP_USE_CYCLES | \ 501 SKBTX_ANY_SW_TSTAMP) 502 503 /* Definitions for flags in struct skb_shared_info */ 504 enum { 505 /* use zcopy routines */ 506 SKBFL_ZEROCOPY_ENABLE = BIT(0), 507 508 /* This indicates at least one fragment might be overwritten 509 * (as in vmsplice(), sendfile() ...) 510 * If we need to compute a TX checksum, we'll need to copy 511 * all frags to avoid possible bad checksum 512 */ 513 SKBFL_SHARED_FRAG = BIT(1), 514 515 /* segment contains only zerocopy data and should not be 516 * charged to the kernel memory. 517 */ 518 SKBFL_PURE_ZEROCOPY = BIT(2), 519 520 SKBFL_DONT_ORPHAN = BIT(3), 521 522 /* page references are managed by the ubuf_info, so it's safe to 523 * use frags only up until ubuf_info is released 524 */ 525 SKBFL_MANAGED_FRAG_REFS = BIT(4), 526 }; 527 528 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG) 529 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \ 530 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS) 531 532 /* 533 * The callback notifies userspace to release buffers when skb DMA is done in 534 * lower device, the skb last reference should be 0 when calling this. 535 * The zerocopy_success argument is true if zero copy transmit occurred, 536 * false on data copy or out of memory error caused by data copy attempt. 537 * The ctx field is used to track device context. 538 * The desc field is used to track userspace buffer index. 539 */ 540 struct ubuf_info { 541 void (*callback)(struct sk_buff *, struct ubuf_info *, 542 bool zerocopy_success); 543 refcount_t refcnt; 544 u8 flags; 545 }; 546 547 struct ubuf_info_msgzc { 548 struct ubuf_info ubuf; 549 550 union { 551 struct { 552 unsigned long desc; 553 void *ctx; 554 }; 555 struct { 556 u32 id; 557 u16 len; 558 u16 zerocopy:1; 559 u32 bytelen; 560 }; 561 }; 562 563 struct mmpin { 564 struct user_struct *user; 565 unsigned int num_pg; 566 } mmp; 567 }; 568 569 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 570 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \ 571 ubuf) 572 573 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 574 void mm_unaccount_pinned_pages(struct mmpin *mmp); 575 576 /* This data is invariant across clones and lives at 577 * the end of the header data, ie. at skb->end. 578 */ 579 struct skb_shared_info { 580 __u8 flags; 581 __u8 meta_len; 582 __u8 nr_frags; 583 __u8 tx_flags; 584 unsigned short gso_size; 585 /* Warning: this field is not always filled in (UFO)! */ 586 unsigned short gso_segs; 587 struct sk_buff *frag_list; 588 struct skb_shared_hwtstamps hwtstamps; 589 unsigned int gso_type; 590 u32 tskey; 591 592 /* 593 * Warning : all fields before dataref are cleared in __alloc_skb() 594 */ 595 atomic_t dataref; 596 unsigned int xdp_frags_size; 597 598 /* Intermediate layers must ensure that destructor_arg 599 * remains valid until skb destructor */ 600 void * destructor_arg; 601 602 /* must be last field, see pskb_expand_head() */ 603 skb_frag_t frags[MAX_SKB_FRAGS]; 604 }; 605 606 /** 607 * DOC: dataref and headerless skbs 608 * 609 * Transport layers send out clones of payload skbs they hold for 610 * retransmissions. To allow lower layers of the stack to prepend their headers 611 * we split &skb_shared_info.dataref into two halves. 612 * The lower 16 bits count the overall number of references. 613 * The higher 16 bits indicate how many of the references are payload-only. 614 * skb_header_cloned() checks if skb is allowed to add / write the headers. 615 * 616 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr 617 * (via __skb_header_release()). Any clone created from marked skb will get 618 * &sk_buff.hdr_len populated with the available headroom. 619 * If there's the only clone in existence it's able to modify the headroom 620 * at will. The sequence of calls inside the transport layer is:: 621 * 622 * <alloc skb> 623 * skb_reserve() 624 * __skb_header_release() 625 * skb_clone() 626 * // send the clone down the stack 627 * 628 * This is not a very generic construct and it depends on the transport layers 629 * doing the right thing. In practice there's usually only one payload-only skb. 630 * Having multiple payload-only skbs with different lengths of hdr_len is not 631 * possible. The payload-only skbs should never leave their owner. 632 */ 633 #define SKB_DATAREF_SHIFT 16 634 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 635 636 637 enum { 638 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 639 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 640 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 641 }; 642 643 enum { 644 SKB_GSO_TCPV4 = 1 << 0, 645 646 /* This indicates the skb is from an untrusted source. */ 647 SKB_GSO_DODGY = 1 << 1, 648 649 /* This indicates the tcp segment has CWR set. */ 650 SKB_GSO_TCP_ECN = 1 << 2, 651 652 SKB_GSO_TCP_FIXEDID = 1 << 3, 653 654 SKB_GSO_TCPV6 = 1 << 4, 655 656 SKB_GSO_FCOE = 1 << 5, 657 658 SKB_GSO_GRE = 1 << 6, 659 660 SKB_GSO_GRE_CSUM = 1 << 7, 661 662 SKB_GSO_IPXIP4 = 1 << 8, 663 664 SKB_GSO_IPXIP6 = 1 << 9, 665 666 SKB_GSO_UDP_TUNNEL = 1 << 10, 667 668 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 669 670 SKB_GSO_PARTIAL = 1 << 12, 671 672 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 673 674 SKB_GSO_SCTP = 1 << 14, 675 676 SKB_GSO_ESP = 1 << 15, 677 678 SKB_GSO_UDP = 1 << 16, 679 680 SKB_GSO_UDP_L4 = 1 << 17, 681 682 SKB_GSO_FRAGLIST = 1 << 18, 683 }; 684 685 #if BITS_PER_LONG > 32 686 #define NET_SKBUFF_DATA_USES_OFFSET 1 687 #endif 688 689 #ifdef NET_SKBUFF_DATA_USES_OFFSET 690 typedef unsigned int sk_buff_data_t; 691 #else 692 typedef unsigned char *sk_buff_data_t; 693 #endif 694 695 /** 696 * DOC: Basic sk_buff geometry 697 * 698 * struct sk_buff itself is a metadata structure and does not hold any packet 699 * data. All the data is held in associated buffers. 700 * 701 * &sk_buff.head points to the main "head" buffer. The head buffer is divided 702 * into two parts: 703 * 704 * - data buffer, containing headers and sometimes payload; 705 * this is the part of the skb operated on by the common helpers 706 * such as skb_put() or skb_pull(); 707 * - shared info (struct skb_shared_info) which holds an array of pointers 708 * to read-only data in the (page, offset, length) format. 709 * 710 * Optionally &skb_shared_info.frag_list may point to another skb. 711 * 712 * Basic diagram may look like this:: 713 * 714 * --------------- 715 * | sk_buff | 716 * --------------- 717 * ,--------------------------- + head 718 * / ,----------------- + data 719 * / / ,----------- + tail 720 * | | | , + end 721 * | | | | 722 * v v v v 723 * ----------------------------------------------- 724 * | headroom | data | tailroom | skb_shared_info | 725 * ----------------------------------------------- 726 * + [page frag] 727 * + [page frag] 728 * + [page frag] 729 * + [page frag] --------- 730 * + frag_list --> | sk_buff | 731 * --------- 732 * 733 */ 734 735 /** 736 * struct sk_buff - socket buffer 737 * @next: Next buffer in list 738 * @prev: Previous buffer in list 739 * @tstamp: Time we arrived/left 740 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 741 * for retransmit timer 742 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 743 * @list: queue head 744 * @ll_node: anchor in an llist (eg socket defer_list) 745 * @sk: Socket we are owned by 746 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in 747 * fragmentation management 748 * @dev: Device we arrived on/are leaving by 749 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 750 * @cb: Control buffer. Free for use by every layer. Put private vars here 751 * @_skb_refdst: destination entry (with norefcount bit) 752 * @sp: the security path, used for xfrm 753 * @len: Length of actual data 754 * @data_len: Data length 755 * @mac_len: Length of link layer header 756 * @hdr_len: writable header length of cloned skb 757 * @csum: Checksum (must include start/offset pair) 758 * @csum_start: Offset from skb->head where checksumming should start 759 * @csum_offset: Offset from csum_start where checksum should be stored 760 * @priority: Packet queueing priority 761 * @ignore_df: allow local fragmentation 762 * @cloned: Head may be cloned (check refcnt to be sure) 763 * @ip_summed: Driver fed us an IP checksum 764 * @nohdr: Payload reference only, must not modify header 765 * @pkt_type: Packet class 766 * @fclone: skbuff clone status 767 * @ipvs_property: skbuff is owned by ipvs 768 * @inner_protocol_type: whether the inner protocol is 769 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 770 * @remcsum_offload: remote checksum offload is enabled 771 * @offload_fwd_mark: Packet was L2-forwarded in hardware 772 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 773 * @tc_skip_classify: do not classify packet. set by IFB device 774 * @tc_at_ingress: used within tc_classify to distinguish in/egress 775 * @redirected: packet was redirected by packet classifier 776 * @from_ingress: packet was redirected from the ingress path 777 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 778 * @peeked: this packet has been seen already, so stats have been 779 * done for it, don't do them again 780 * @nf_trace: netfilter packet trace flag 781 * @protocol: Packet protocol from driver 782 * @destructor: Destruct function 783 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 784 * @_sk_redir: socket redirection information for skmsg 785 * @_nfct: Associated connection, if any (with nfctinfo bits) 786 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 787 * @skb_iif: ifindex of device we arrived on 788 * @tc_index: Traffic control index 789 * @hash: the packet hash 790 * @queue_mapping: Queue mapping for multiqueue devices 791 * @head_frag: skb was allocated from page fragments, 792 * not allocated by kmalloc() or vmalloc(). 793 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 794 * @pp_recycle: mark the packet for recycling instead of freeing (implies 795 * page_pool support on driver) 796 * @active_extensions: active extensions (skb_ext_id types) 797 * @ndisc_nodetype: router type (from link layer) 798 * @ooo_okay: allow the mapping of a socket to a queue to be changed 799 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 800 * ports. 801 * @sw_hash: indicates hash was computed in software stack 802 * @wifi_acked_valid: wifi_acked was set 803 * @wifi_acked: whether frame was acked on wifi or not 804 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 805 * @encapsulation: indicates the inner headers in the skbuff are valid 806 * @encap_hdr_csum: software checksum is needed 807 * @csum_valid: checksum is already valid 808 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 809 * @csum_complete_sw: checksum was completed by software 810 * @csum_level: indicates the number of consecutive checksums found in 811 * the packet minus one that have been verified as 812 * CHECKSUM_UNNECESSARY (max 3) 813 * @scm_io_uring: SKB holds io_uring registered files 814 * @dst_pending_confirm: need to confirm neighbour 815 * @decrypted: Decrypted SKB 816 * @slow_gro: state present at GRO time, slower prepare step required 817 * @mono_delivery_time: When set, skb->tstamp has the 818 * delivery_time in mono clock base (i.e. EDT). Otherwise, the 819 * skb->tstamp has the (rcv) timestamp at ingress and 820 * delivery_time at egress. 821 * @napi_id: id of the NAPI struct this skb came from 822 * @sender_cpu: (aka @napi_id) source CPU in XPS 823 * @alloc_cpu: CPU which did the skb allocation. 824 * @secmark: security marking 825 * @mark: Generic packet mark 826 * @reserved_tailroom: (aka @mark) number of bytes of free space available 827 * at the tail of an sk_buff 828 * @vlan_all: vlan fields (proto & tci) 829 * @vlan_proto: vlan encapsulation protocol 830 * @vlan_tci: vlan tag control information 831 * @inner_protocol: Protocol (encapsulation) 832 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 833 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 834 * @inner_transport_header: Inner transport layer header (encapsulation) 835 * @inner_network_header: Network layer header (encapsulation) 836 * @inner_mac_header: Link layer header (encapsulation) 837 * @transport_header: Transport layer header 838 * @network_header: Network layer header 839 * @mac_header: Link layer header 840 * @kcov_handle: KCOV remote handle for remote coverage collection 841 * @tail: Tail pointer 842 * @end: End pointer 843 * @head: Head of buffer 844 * @data: Data head pointer 845 * @truesize: Buffer size 846 * @users: User count - see {datagram,tcp}.c 847 * @extensions: allocated extensions, valid if active_extensions is nonzero 848 */ 849 850 struct sk_buff { 851 union { 852 struct { 853 /* These two members must be first to match sk_buff_head. */ 854 struct sk_buff *next; 855 struct sk_buff *prev; 856 857 union { 858 struct net_device *dev; 859 /* Some protocols might use this space to store information, 860 * while device pointer would be NULL. 861 * UDP receive path is one user. 862 */ 863 unsigned long dev_scratch; 864 }; 865 }; 866 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 867 struct list_head list; 868 struct llist_node ll_node; 869 }; 870 871 union { 872 struct sock *sk; 873 int ip_defrag_offset; 874 }; 875 876 union { 877 ktime_t tstamp; 878 u64 skb_mstamp_ns; /* earliest departure time */ 879 }; 880 /* 881 * This is the control buffer. It is free to use for every 882 * layer. Please put your private variables there. If you 883 * want to keep them across layers you have to do a skb_clone() 884 * first. This is owned by whoever has the skb queued ATM. 885 */ 886 char cb[48] __aligned(8); 887 888 union { 889 struct { 890 unsigned long _skb_refdst; 891 void (*destructor)(struct sk_buff *skb); 892 }; 893 struct list_head tcp_tsorted_anchor; 894 #ifdef CONFIG_NET_SOCK_MSG 895 unsigned long _sk_redir; 896 #endif 897 }; 898 899 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 900 unsigned long _nfct; 901 #endif 902 unsigned int len, 903 data_len; 904 __u16 mac_len, 905 hdr_len; 906 907 /* Following fields are _not_ copied in __copy_skb_header() 908 * Note that queue_mapping is here mostly to fill a hole. 909 */ 910 __u16 queue_mapping; 911 912 /* if you move cloned around you also must adapt those constants */ 913 #ifdef __BIG_ENDIAN_BITFIELD 914 #define CLONED_MASK (1 << 7) 915 #else 916 #define CLONED_MASK 1 917 #endif 918 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 919 920 /* private: */ 921 __u8 __cloned_offset[0]; 922 /* public: */ 923 __u8 cloned:1, 924 nohdr:1, 925 fclone:2, 926 peeked:1, 927 head_frag:1, 928 pfmemalloc:1, 929 pp_recycle:1; /* page_pool recycle indicator */ 930 #ifdef CONFIG_SKB_EXTENSIONS 931 __u8 active_extensions; 932 #endif 933 934 /* Fields enclosed in headers group are copied 935 * using a single memcpy() in __copy_skb_header() 936 */ 937 struct_group(headers, 938 939 /* private: */ 940 __u8 __pkt_type_offset[0]; 941 /* public: */ 942 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 943 __u8 ignore_df:1; 944 __u8 nf_trace:1; 945 __u8 ip_summed:2; 946 __u8 ooo_okay:1; 947 948 __u8 l4_hash:1; 949 __u8 sw_hash:1; 950 __u8 wifi_acked_valid:1; 951 __u8 wifi_acked:1; 952 __u8 no_fcs:1; 953 /* Indicates the inner headers are valid in the skbuff. */ 954 __u8 encapsulation:1; 955 __u8 encap_hdr_csum:1; 956 __u8 csum_valid:1; 957 958 /* private: */ 959 __u8 __pkt_vlan_present_offset[0]; 960 /* public: */ 961 __u8 remcsum_offload:1; 962 __u8 csum_complete_sw:1; 963 __u8 csum_level:2; 964 __u8 dst_pending_confirm:1; 965 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */ 966 #ifdef CONFIG_NET_CLS_ACT 967 __u8 tc_skip_classify:1; 968 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 969 #endif 970 #ifdef CONFIG_IPV6_NDISC_NODETYPE 971 __u8 ndisc_nodetype:2; 972 #endif 973 974 __u8 ipvs_property:1; 975 __u8 inner_protocol_type:1; 976 #ifdef CONFIG_NET_SWITCHDEV 977 __u8 offload_fwd_mark:1; 978 __u8 offload_l3_fwd_mark:1; 979 #endif 980 __u8 redirected:1; 981 #ifdef CONFIG_NET_REDIRECT 982 __u8 from_ingress:1; 983 #endif 984 #ifdef CONFIG_NETFILTER_SKIP_EGRESS 985 __u8 nf_skip_egress:1; 986 #endif 987 #ifdef CONFIG_TLS_DEVICE 988 __u8 decrypted:1; 989 #endif 990 __u8 slow_gro:1; 991 __u8 csum_not_inet:1; 992 __u8 scm_io_uring:1; 993 994 #ifdef CONFIG_NET_SCHED 995 __u16 tc_index; /* traffic control index */ 996 #endif 997 998 union { 999 __wsum csum; 1000 struct { 1001 __u16 csum_start; 1002 __u16 csum_offset; 1003 }; 1004 }; 1005 __u32 priority; 1006 int skb_iif; 1007 __u32 hash; 1008 union { 1009 u32 vlan_all; 1010 struct { 1011 __be16 vlan_proto; 1012 __u16 vlan_tci; 1013 }; 1014 }; 1015 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1016 union { 1017 unsigned int napi_id; 1018 unsigned int sender_cpu; 1019 }; 1020 #endif 1021 u16 alloc_cpu; 1022 #ifdef CONFIG_NETWORK_SECMARK 1023 __u32 secmark; 1024 #endif 1025 1026 union { 1027 __u32 mark; 1028 __u32 reserved_tailroom; 1029 }; 1030 1031 union { 1032 __be16 inner_protocol; 1033 __u8 inner_ipproto; 1034 }; 1035 1036 __u16 inner_transport_header; 1037 __u16 inner_network_header; 1038 __u16 inner_mac_header; 1039 1040 __be16 protocol; 1041 __u16 transport_header; 1042 __u16 network_header; 1043 __u16 mac_header; 1044 1045 #ifdef CONFIG_KCOV 1046 u64 kcov_handle; 1047 #endif 1048 1049 ); /* end headers group */ 1050 1051 /* These elements must be at the end, see alloc_skb() for details. */ 1052 sk_buff_data_t tail; 1053 sk_buff_data_t end; 1054 unsigned char *head, 1055 *data; 1056 unsigned int truesize; 1057 refcount_t users; 1058 1059 #ifdef CONFIG_SKB_EXTENSIONS 1060 /* only useable after checking ->active_extensions != 0 */ 1061 struct skb_ext *extensions; 1062 #endif 1063 }; 1064 1065 /* if you move pkt_type around you also must adapt those constants */ 1066 #ifdef __BIG_ENDIAN_BITFIELD 1067 #define PKT_TYPE_MAX (7 << 5) 1068 #else 1069 #define PKT_TYPE_MAX 7 1070 #endif 1071 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1072 1073 /* if you move tc_at_ingress or mono_delivery_time 1074 * around, you also must adapt these constants. 1075 */ 1076 #ifdef __BIG_ENDIAN_BITFIELD 1077 #define TC_AT_INGRESS_MASK (1 << 0) 1078 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 2) 1079 #else 1080 #define TC_AT_INGRESS_MASK (1 << 7) 1081 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 5) 1082 #endif 1083 #define PKT_VLAN_PRESENT_OFFSET offsetof(struct sk_buff, __pkt_vlan_present_offset) 1084 1085 #ifdef __KERNEL__ 1086 /* 1087 * Handling routines are only of interest to the kernel 1088 */ 1089 1090 #define SKB_ALLOC_FCLONE 0x01 1091 #define SKB_ALLOC_RX 0x02 1092 #define SKB_ALLOC_NAPI 0x04 1093 1094 /** 1095 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1096 * @skb: buffer 1097 */ 1098 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1099 { 1100 return unlikely(skb->pfmemalloc); 1101 } 1102 1103 /* 1104 * skb might have a dst pointer attached, refcounted or not. 1105 * _skb_refdst low order bit is set if refcount was _not_ taken 1106 */ 1107 #define SKB_DST_NOREF 1UL 1108 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1109 1110 /** 1111 * skb_dst - returns skb dst_entry 1112 * @skb: buffer 1113 * 1114 * Returns skb dst_entry, regardless of reference taken or not. 1115 */ 1116 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1117 { 1118 /* If refdst was not refcounted, check we still are in a 1119 * rcu_read_lock section 1120 */ 1121 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1122 !rcu_read_lock_held() && 1123 !rcu_read_lock_bh_held()); 1124 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1125 } 1126 1127 /** 1128 * skb_dst_set - sets skb dst 1129 * @skb: buffer 1130 * @dst: dst entry 1131 * 1132 * Sets skb dst, assuming a reference was taken on dst and should 1133 * be released by skb_dst_drop() 1134 */ 1135 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1136 { 1137 skb->slow_gro |= !!dst; 1138 skb->_skb_refdst = (unsigned long)dst; 1139 } 1140 1141 /** 1142 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1143 * @skb: buffer 1144 * @dst: dst entry 1145 * 1146 * Sets skb dst, assuming a reference was not taken on dst. 1147 * If dst entry is cached, we do not take reference and dst_release 1148 * will be avoided by refdst_drop. If dst entry is not cached, we take 1149 * reference, so that last dst_release can destroy the dst immediately. 1150 */ 1151 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1152 { 1153 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1154 skb->slow_gro |= !!dst; 1155 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1156 } 1157 1158 /** 1159 * skb_dst_is_noref - Test if skb dst isn't refcounted 1160 * @skb: buffer 1161 */ 1162 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1163 { 1164 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1165 } 1166 1167 /** 1168 * skb_rtable - Returns the skb &rtable 1169 * @skb: buffer 1170 */ 1171 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 1172 { 1173 return (struct rtable *)skb_dst(skb); 1174 } 1175 1176 /* For mangling skb->pkt_type from user space side from applications 1177 * such as nft, tc, etc, we only allow a conservative subset of 1178 * possible pkt_types to be set. 1179 */ 1180 static inline bool skb_pkt_type_ok(u32 ptype) 1181 { 1182 return ptype <= PACKET_OTHERHOST; 1183 } 1184 1185 /** 1186 * skb_napi_id - Returns the skb's NAPI id 1187 * @skb: buffer 1188 */ 1189 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1190 { 1191 #ifdef CONFIG_NET_RX_BUSY_POLL 1192 return skb->napi_id; 1193 #else 1194 return 0; 1195 #endif 1196 } 1197 1198 /** 1199 * skb_unref - decrement the skb's reference count 1200 * @skb: buffer 1201 * 1202 * Returns true if we can free the skb. 1203 */ 1204 static inline bool skb_unref(struct sk_buff *skb) 1205 { 1206 if (unlikely(!skb)) 1207 return false; 1208 if (likely(refcount_read(&skb->users) == 1)) 1209 smp_rmb(); 1210 else if (likely(!refcount_dec_and_test(&skb->users))) 1211 return false; 1212 1213 return true; 1214 } 1215 1216 void __fix_address 1217 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason); 1218 1219 /** 1220 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1221 * @skb: buffer to free 1222 */ 1223 static inline void kfree_skb(struct sk_buff *skb) 1224 { 1225 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1226 } 1227 1228 void skb_release_head_state(struct sk_buff *skb); 1229 void kfree_skb_list_reason(struct sk_buff *segs, 1230 enum skb_drop_reason reason); 1231 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1232 void skb_tx_error(struct sk_buff *skb); 1233 1234 static inline void kfree_skb_list(struct sk_buff *segs) 1235 { 1236 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1237 } 1238 1239 #ifdef CONFIG_TRACEPOINTS 1240 void consume_skb(struct sk_buff *skb); 1241 #else 1242 static inline void consume_skb(struct sk_buff *skb) 1243 { 1244 return kfree_skb(skb); 1245 } 1246 #endif 1247 1248 void __consume_stateless_skb(struct sk_buff *skb); 1249 void __kfree_skb(struct sk_buff *skb); 1250 extern struct kmem_cache *skbuff_cache; 1251 1252 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1253 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1254 bool *fragstolen, int *delta_truesize); 1255 1256 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1257 int node); 1258 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1259 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1260 struct sk_buff *build_skb_around(struct sk_buff *skb, 1261 void *data, unsigned int frag_size); 1262 void skb_attempt_defer_free(struct sk_buff *skb); 1263 1264 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1265 struct sk_buff *slab_build_skb(void *data); 1266 1267 /** 1268 * alloc_skb - allocate a network buffer 1269 * @size: size to allocate 1270 * @priority: allocation mask 1271 * 1272 * This function is a convenient wrapper around __alloc_skb(). 1273 */ 1274 static inline struct sk_buff *alloc_skb(unsigned int size, 1275 gfp_t priority) 1276 { 1277 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1278 } 1279 1280 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1281 unsigned long data_len, 1282 int max_page_order, 1283 int *errcode, 1284 gfp_t gfp_mask); 1285 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1286 1287 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1288 struct sk_buff_fclones { 1289 struct sk_buff skb1; 1290 1291 struct sk_buff skb2; 1292 1293 refcount_t fclone_ref; 1294 }; 1295 1296 /** 1297 * skb_fclone_busy - check if fclone is busy 1298 * @sk: socket 1299 * @skb: buffer 1300 * 1301 * Returns true if skb is a fast clone, and its clone is not freed. 1302 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1303 * so we also check that this didnt happen. 1304 */ 1305 static inline bool skb_fclone_busy(const struct sock *sk, 1306 const struct sk_buff *skb) 1307 { 1308 const struct sk_buff_fclones *fclones; 1309 1310 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1311 1312 return skb->fclone == SKB_FCLONE_ORIG && 1313 refcount_read(&fclones->fclone_ref) > 1 && 1314 READ_ONCE(fclones->skb2.sk) == sk; 1315 } 1316 1317 /** 1318 * alloc_skb_fclone - allocate a network buffer from fclone cache 1319 * @size: size to allocate 1320 * @priority: allocation mask 1321 * 1322 * This function is a convenient wrapper around __alloc_skb(). 1323 */ 1324 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1325 gfp_t priority) 1326 { 1327 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1328 } 1329 1330 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1331 void skb_headers_offset_update(struct sk_buff *skb, int off); 1332 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1333 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1334 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1335 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1336 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1337 gfp_t gfp_mask, bool fclone); 1338 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1339 gfp_t gfp_mask) 1340 { 1341 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1342 } 1343 1344 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1345 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1346 unsigned int headroom); 1347 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1348 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1349 int newtailroom, gfp_t priority); 1350 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1351 int offset, int len); 1352 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1353 int offset, int len); 1354 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1355 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1356 1357 /** 1358 * skb_pad - zero pad the tail of an skb 1359 * @skb: buffer to pad 1360 * @pad: space to pad 1361 * 1362 * Ensure that a buffer is followed by a padding area that is zero 1363 * filled. Used by network drivers which may DMA or transfer data 1364 * beyond the buffer end onto the wire. 1365 * 1366 * May return error in out of memory cases. The skb is freed on error. 1367 */ 1368 static inline int skb_pad(struct sk_buff *skb, int pad) 1369 { 1370 return __skb_pad(skb, pad, true); 1371 } 1372 #define dev_kfree_skb(a) consume_skb(a) 1373 1374 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1375 int offset, size_t size); 1376 1377 struct skb_seq_state { 1378 __u32 lower_offset; 1379 __u32 upper_offset; 1380 __u32 frag_idx; 1381 __u32 stepped_offset; 1382 struct sk_buff *root_skb; 1383 struct sk_buff *cur_skb; 1384 __u8 *frag_data; 1385 __u32 frag_off; 1386 }; 1387 1388 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1389 unsigned int to, struct skb_seq_state *st); 1390 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1391 struct skb_seq_state *st); 1392 void skb_abort_seq_read(struct skb_seq_state *st); 1393 1394 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1395 unsigned int to, struct ts_config *config); 1396 1397 /* 1398 * Packet hash types specify the type of hash in skb_set_hash. 1399 * 1400 * Hash types refer to the protocol layer addresses which are used to 1401 * construct a packet's hash. The hashes are used to differentiate or identify 1402 * flows of the protocol layer for the hash type. Hash types are either 1403 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1404 * 1405 * Properties of hashes: 1406 * 1407 * 1) Two packets in different flows have different hash values 1408 * 2) Two packets in the same flow should have the same hash value 1409 * 1410 * A hash at a higher layer is considered to be more specific. A driver should 1411 * set the most specific hash possible. 1412 * 1413 * A driver cannot indicate a more specific hash than the layer at which a hash 1414 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1415 * 1416 * A driver may indicate a hash level which is less specific than the 1417 * actual layer the hash was computed on. For instance, a hash computed 1418 * at L4 may be considered an L3 hash. This should only be done if the 1419 * driver can't unambiguously determine that the HW computed the hash at 1420 * the higher layer. Note that the "should" in the second property above 1421 * permits this. 1422 */ 1423 enum pkt_hash_types { 1424 PKT_HASH_TYPE_NONE, /* Undefined type */ 1425 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1426 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1427 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1428 }; 1429 1430 static inline void skb_clear_hash(struct sk_buff *skb) 1431 { 1432 skb->hash = 0; 1433 skb->sw_hash = 0; 1434 skb->l4_hash = 0; 1435 } 1436 1437 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1438 { 1439 if (!skb->l4_hash) 1440 skb_clear_hash(skb); 1441 } 1442 1443 static inline void 1444 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1445 { 1446 skb->l4_hash = is_l4; 1447 skb->sw_hash = is_sw; 1448 skb->hash = hash; 1449 } 1450 1451 static inline void 1452 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1453 { 1454 /* Used by drivers to set hash from HW */ 1455 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1456 } 1457 1458 static inline void 1459 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1460 { 1461 __skb_set_hash(skb, hash, true, is_l4); 1462 } 1463 1464 void __skb_get_hash(struct sk_buff *skb); 1465 u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1466 u32 skb_get_poff(const struct sk_buff *skb); 1467 u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1468 const struct flow_keys_basic *keys, int hlen); 1469 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1470 const void *data, int hlen_proto); 1471 1472 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1473 int thoff, u8 ip_proto) 1474 { 1475 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1476 } 1477 1478 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1479 const struct flow_dissector_key *key, 1480 unsigned int key_count); 1481 1482 struct bpf_flow_dissector; 1483 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1484 __be16 proto, int nhoff, int hlen, unsigned int flags); 1485 1486 bool __skb_flow_dissect(const struct net *net, 1487 const struct sk_buff *skb, 1488 struct flow_dissector *flow_dissector, 1489 void *target_container, const void *data, 1490 __be16 proto, int nhoff, int hlen, unsigned int flags); 1491 1492 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1493 struct flow_dissector *flow_dissector, 1494 void *target_container, unsigned int flags) 1495 { 1496 return __skb_flow_dissect(NULL, skb, flow_dissector, 1497 target_container, NULL, 0, 0, 0, flags); 1498 } 1499 1500 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1501 struct flow_keys *flow, 1502 unsigned int flags) 1503 { 1504 memset(flow, 0, sizeof(*flow)); 1505 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1506 flow, NULL, 0, 0, 0, flags); 1507 } 1508 1509 static inline bool 1510 skb_flow_dissect_flow_keys_basic(const struct net *net, 1511 const struct sk_buff *skb, 1512 struct flow_keys_basic *flow, 1513 const void *data, __be16 proto, 1514 int nhoff, int hlen, unsigned int flags) 1515 { 1516 memset(flow, 0, sizeof(*flow)); 1517 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1518 data, proto, nhoff, hlen, flags); 1519 } 1520 1521 void skb_flow_dissect_meta(const struct sk_buff *skb, 1522 struct flow_dissector *flow_dissector, 1523 void *target_container); 1524 1525 /* Gets a skb connection tracking info, ctinfo map should be a 1526 * map of mapsize to translate enum ip_conntrack_info states 1527 * to user states. 1528 */ 1529 void 1530 skb_flow_dissect_ct(const struct sk_buff *skb, 1531 struct flow_dissector *flow_dissector, 1532 void *target_container, 1533 u16 *ctinfo_map, size_t mapsize, 1534 bool post_ct, u16 zone); 1535 void 1536 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1537 struct flow_dissector *flow_dissector, 1538 void *target_container); 1539 1540 void skb_flow_dissect_hash(const struct sk_buff *skb, 1541 struct flow_dissector *flow_dissector, 1542 void *target_container); 1543 1544 static inline __u32 skb_get_hash(struct sk_buff *skb) 1545 { 1546 if (!skb->l4_hash && !skb->sw_hash) 1547 __skb_get_hash(skb); 1548 1549 return skb->hash; 1550 } 1551 1552 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1553 { 1554 if (!skb->l4_hash && !skb->sw_hash) { 1555 struct flow_keys keys; 1556 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1557 1558 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1559 } 1560 1561 return skb->hash; 1562 } 1563 1564 __u32 skb_get_hash_perturb(const struct sk_buff *skb, 1565 const siphash_key_t *perturb); 1566 1567 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1568 { 1569 return skb->hash; 1570 } 1571 1572 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1573 { 1574 to->hash = from->hash; 1575 to->sw_hash = from->sw_hash; 1576 to->l4_hash = from->l4_hash; 1577 }; 1578 1579 static inline void skb_copy_decrypted(struct sk_buff *to, 1580 const struct sk_buff *from) 1581 { 1582 #ifdef CONFIG_TLS_DEVICE 1583 to->decrypted = from->decrypted; 1584 #endif 1585 } 1586 1587 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1588 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1589 { 1590 return skb->head + skb->end; 1591 } 1592 1593 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1594 { 1595 return skb->end; 1596 } 1597 1598 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1599 { 1600 skb->end = offset; 1601 } 1602 #else 1603 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1604 { 1605 return skb->end; 1606 } 1607 1608 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1609 { 1610 return skb->end - skb->head; 1611 } 1612 1613 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1614 { 1615 skb->end = skb->head + offset; 1616 } 1617 #endif 1618 1619 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1620 struct ubuf_info *uarg); 1621 1622 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1623 1624 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg, 1625 bool success); 1626 1627 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, 1628 struct sk_buff *skb, struct iov_iter *from, 1629 size_t length); 1630 1631 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1632 struct msghdr *msg, int len) 1633 { 1634 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len); 1635 } 1636 1637 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1638 struct msghdr *msg, int len, 1639 struct ubuf_info *uarg); 1640 1641 /* Internal */ 1642 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1643 1644 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1645 { 1646 return &skb_shinfo(skb)->hwtstamps; 1647 } 1648 1649 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1650 { 1651 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1652 1653 return is_zcopy ? skb_uarg(skb) : NULL; 1654 } 1655 1656 static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1657 { 1658 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1659 } 1660 1661 static inline bool skb_zcopy_managed(const struct sk_buff *skb) 1662 { 1663 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS; 1664 } 1665 1666 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1667 const struct sk_buff *skb2) 1668 { 1669 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1670 } 1671 1672 static inline void net_zcopy_get(struct ubuf_info *uarg) 1673 { 1674 refcount_inc(&uarg->refcnt); 1675 } 1676 1677 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1678 { 1679 skb_shinfo(skb)->destructor_arg = uarg; 1680 skb_shinfo(skb)->flags |= uarg->flags; 1681 } 1682 1683 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1684 bool *have_ref) 1685 { 1686 if (skb && uarg && !skb_zcopy(skb)) { 1687 if (unlikely(have_ref && *have_ref)) 1688 *have_ref = false; 1689 else 1690 net_zcopy_get(uarg); 1691 skb_zcopy_init(skb, uarg); 1692 } 1693 } 1694 1695 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1696 { 1697 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1698 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1699 } 1700 1701 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1702 { 1703 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1704 } 1705 1706 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1707 { 1708 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1709 } 1710 1711 static inline void net_zcopy_put(struct ubuf_info *uarg) 1712 { 1713 if (uarg) 1714 uarg->callback(NULL, uarg, true); 1715 } 1716 1717 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1718 { 1719 if (uarg) { 1720 if (uarg->callback == msg_zerocopy_callback) 1721 msg_zerocopy_put_abort(uarg, have_uref); 1722 else if (have_uref) 1723 net_zcopy_put(uarg); 1724 } 1725 } 1726 1727 /* Release a reference on a zerocopy structure */ 1728 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1729 { 1730 struct ubuf_info *uarg = skb_zcopy(skb); 1731 1732 if (uarg) { 1733 if (!skb_zcopy_is_nouarg(skb)) 1734 uarg->callback(skb, uarg, zerocopy_success); 1735 1736 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1737 } 1738 } 1739 1740 void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1741 1742 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1743 { 1744 if (unlikely(skb_zcopy_managed(skb))) 1745 __skb_zcopy_downgrade_managed(skb); 1746 } 1747 1748 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1749 { 1750 skb->next = NULL; 1751 } 1752 1753 static inline void skb_poison_list(struct sk_buff *skb) 1754 { 1755 #ifdef CONFIG_DEBUG_NET 1756 skb->next = SKB_LIST_POISON_NEXT; 1757 #endif 1758 } 1759 1760 /* Iterate through singly-linked GSO fragments of an skb. */ 1761 #define skb_list_walk_safe(first, skb, next_skb) \ 1762 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1763 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1764 1765 static inline void skb_list_del_init(struct sk_buff *skb) 1766 { 1767 __list_del_entry(&skb->list); 1768 skb_mark_not_on_list(skb); 1769 } 1770 1771 /** 1772 * skb_queue_empty - check if a queue is empty 1773 * @list: queue head 1774 * 1775 * Returns true if the queue is empty, false otherwise. 1776 */ 1777 static inline int skb_queue_empty(const struct sk_buff_head *list) 1778 { 1779 return list->next == (const struct sk_buff *) list; 1780 } 1781 1782 /** 1783 * skb_queue_empty_lockless - check if a queue is empty 1784 * @list: queue head 1785 * 1786 * Returns true if the queue is empty, false otherwise. 1787 * This variant can be used in lockless contexts. 1788 */ 1789 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1790 { 1791 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1792 } 1793 1794 1795 /** 1796 * skb_queue_is_last - check if skb is the last entry in the queue 1797 * @list: queue head 1798 * @skb: buffer 1799 * 1800 * Returns true if @skb is the last buffer on the list. 1801 */ 1802 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1803 const struct sk_buff *skb) 1804 { 1805 return skb->next == (const struct sk_buff *) list; 1806 } 1807 1808 /** 1809 * skb_queue_is_first - check if skb is the first entry in the queue 1810 * @list: queue head 1811 * @skb: buffer 1812 * 1813 * Returns true if @skb is the first buffer on the list. 1814 */ 1815 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1816 const struct sk_buff *skb) 1817 { 1818 return skb->prev == (const struct sk_buff *) list; 1819 } 1820 1821 /** 1822 * skb_queue_next - return the next packet in the queue 1823 * @list: queue head 1824 * @skb: current buffer 1825 * 1826 * Return the next packet in @list after @skb. It is only valid to 1827 * call this if skb_queue_is_last() evaluates to false. 1828 */ 1829 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1830 const struct sk_buff *skb) 1831 { 1832 /* This BUG_ON may seem severe, but if we just return then we 1833 * are going to dereference garbage. 1834 */ 1835 BUG_ON(skb_queue_is_last(list, skb)); 1836 return skb->next; 1837 } 1838 1839 /** 1840 * skb_queue_prev - return the prev packet in the queue 1841 * @list: queue head 1842 * @skb: current buffer 1843 * 1844 * Return the prev packet in @list before @skb. It is only valid to 1845 * call this if skb_queue_is_first() evaluates to false. 1846 */ 1847 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1848 const struct sk_buff *skb) 1849 { 1850 /* This BUG_ON may seem severe, but if we just return then we 1851 * are going to dereference garbage. 1852 */ 1853 BUG_ON(skb_queue_is_first(list, skb)); 1854 return skb->prev; 1855 } 1856 1857 /** 1858 * skb_get - reference buffer 1859 * @skb: buffer to reference 1860 * 1861 * Makes another reference to a socket buffer and returns a pointer 1862 * to the buffer. 1863 */ 1864 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1865 { 1866 refcount_inc(&skb->users); 1867 return skb; 1868 } 1869 1870 /* 1871 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1872 */ 1873 1874 /** 1875 * skb_cloned - is the buffer a clone 1876 * @skb: buffer to check 1877 * 1878 * Returns true if the buffer was generated with skb_clone() and is 1879 * one of multiple shared copies of the buffer. Cloned buffers are 1880 * shared data so must not be written to under normal circumstances. 1881 */ 1882 static inline int skb_cloned(const struct sk_buff *skb) 1883 { 1884 return skb->cloned && 1885 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1886 } 1887 1888 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1889 { 1890 might_sleep_if(gfpflags_allow_blocking(pri)); 1891 1892 if (skb_cloned(skb)) 1893 return pskb_expand_head(skb, 0, 0, pri); 1894 1895 return 0; 1896 } 1897 1898 /* This variant of skb_unclone() makes sure skb->truesize 1899 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 1900 * 1901 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 1902 * when various debugging features are in place. 1903 */ 1904 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 1905 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 1906 { 1907 might_sleep_if(gfpflags_allow_blocking(pri)); 1908 1909 if (skb_cloned(skb)) 1910 return __skb_unclone_keeptruesize(skb, pri); 1911 return 0; 1912 } 1913 1914 /** 1915 * skb_header_cloned - is the header a clone 1916 * @skb: buffer to check 1917 * 1918 * Returns true if modifying the header part of the buffer requires 1919 * the data to be copied. 1920 */ 1921 static inline int skb_header_cloned(const struct sk_buff *skb) 1922 { 1923 int dataref; 1924 1925 if (!skb->cloned) 1926 return 0; 1927 1928 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1929 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1930 return dataref != 1; 1931 } 1932 1933 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1934 { 1935 might_sleep_if(gfpflags_allow_blocking(pri)); 1936 1937 if (skb_header_cloned(skb)) 1938 return pskb_expand_head(skb, 0, 0, pri); 1939 1940 return 0; 1941 } 1942 1943 /** 1944 * __skb_header_release() - allow clones to use the headroom 1945 * @skb: buffer to operate on 1946 * 1947 * See "DOC: dataref and headerless skbs". 1948 */ 1949 static inline void __skb_header_release(struct sk_buff *skb) 1950 { 1951 skb->nohdr = 1; 1952 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1953 } 1954 1955 1956 /** 1957 * skb_shared - is the buffer shared 1958 * @skb: buffer to check 1959 * 1960 * Returns true if more than one person has a reference to this 1961 * buffer. 1962 */ 1963 static inline int skb_shared(const struct sk_buff *skb) 1964 { 1965 return refcount_read(&skb->users) != 1; 1966 } 1967 1968 /** 1969 * skb_share_check - check if buffer is shared and if so clone it 1970 * @skb: buffer to check 1971 * @pri: priority for memory allocation 1972 * 1973 * If the buffer is shared the buffer is cloned and the old copy 1974 * drops a reference. A new clone with a single reference is returned. 1975 * If the buffer is not shared the original buffer is returned. When 1976 * being called from interrupt status or with spinlocks held pri must 1977 * be GFP_ATOMIC. 1978 * 1979 * NULL is returned on a memory allocation failure. 1980 */ 1981 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1982 { 1983 might_sleep_if(gfpflags_allow_blocking(pri)); 1984 if (skb_shared(skb)) { 1985 struct sk_buff *nskb = skb_clone(skb, pri); 1986 1987 if (likely(nskb)) 1988 consume_skb(skb); 1989 else 1990 kfree_skb(skb); 1991 skb = nskb; 1992 } 1993 return skb; 1994 } 1995 1996 /* 1997 * Copy shared buffers into a new sk_buff. We effectively do COW on 1998 * packets to handle cases where we have a local reader and forward 1999 * and a couple of other messy ones. The normal one is tcpdumping 2000 * a packet thats being forwarded. 2001 */ 2002 2003 /** 2004 * skb_unshare - make a copy of a shared buffer 2005 * @skb: buffer to check 2006 * @pri: priority for memory allocation 2007 * 2008 * If the socket buffer is a clone then this function creates a new 2009 * copy of the data, drops a reference count on the old copy and returns 2010 * the new copy with the reference count at 1. If the buffer is not a clone 2011 * the original buffer is returned. When called with a spinlock held or 2012 * from interrupt state @pri must be %GFP_ATOMIC 2013 * 2014 * %NULL is returned on a memory allocation failure. 2015 */ 2016 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2017 gfp_t pri) 2018 { 2019 might_sleep_if(gfpflags_allow_blocking(pri)); 2020 if (skb_cloned(skb)) { 2021 struct sk_buff *nskb = skb_copy(skb, pri); 2022 2023 /* Free our shared copy */ 2024 if (likely(nskb)) 2025 consume_skb(skb); 2026 else 2027 kfree_skb(skb); 2028 skb = nskb; 2029 } 2030 return skb; 2031 } 2032 2033 /** 2034 * skb_peek - peek at the head of an &sk_buff_head 2035 * @list_: list to peek at 2036 * 2037 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2038 * be careful with this one. A peek leaves the buffer on the 2039 * list and someone else may run off with it. You must hold 2040 * the appropriate locks or have a private queue to do this. 2041 * 2042 * Returns %NULL for an empty list or a pointer to the head element. 2043 * The reference count is not incremented and the reference is therefore 2044 * volatile. Use with caution. 2045 */ 2046 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2047 { 2048 struct sk_buff *skb = list_->next; 2049 2050 if (skb == (struct sk_buff *)list_) 2051 skb = NULL; 2052 return skb; 2053 } 2054 2055 /** 2056 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2057 * @list_: list to peek at 2058 * 2059 * Like skb_peek(), but the caller knows that the list is not empty. 2060 */ 2061 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2062 { 2063 return list_->next; 2064 } 2065 2066 /** 2067 * skb_peek_next - peek skb following the given one from a queue 2068 * @skb: skb to start from 2069 * @list_: list to peek at 2070 * 2071 * Returns %NULL when the end of the list is met or a pointer to the 2072 * next element. The reference count is not incremented and the 2073 * reference is therefore volatile. Use with caution. 2074 */ 2075 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2076 const struct sk_buff_head *list_) 2077 { 2078 struct sk_buff *next = skb->next; 2079 2080 if (next == (struct sk_buff *)list_) 2081 next = NULL; 2082 return next; 2083 } 2084 2085 /** 2086 * skb_peek_tail - peek at the tail of an &sk_buff_head 2087 * @list_: list to peek at 2088 * 2089 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2090 * be careful with this one. A peek leaves the buffer on the 2091 * list and someone else may run off with it. You must hold 2092 * the appropriate locks or have a private queue to do this. 2093 * 2094 * Returns %NULL for an empty list or a pointer to the tail element. 2095 * The reference count is not incremented and the reference is therefore 2096 * volatile. Use with caution. 2097 */ 2098 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2099 { 2100 struct sk_buff *skb = READ_ONCE(list_->prev); 2101 2102 if (skb == (struct sk_buff *)list_) 2103 skb = NULL; 2104 return skb; 2105 2106 } 2107 2108 /** 2109 * skb_queue_len - get queue length 2110 * @list_: list to measure 2111 * 2112 * Return the length of an &sk_buff queue. 2113 */ 2114 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2115 { 2116 return list_->qlen; 2117 } 2118 2119 /** 2120 * skb_queue_len_lockless - get queue length 2121 * @list_: list to measure 2122 * 2123 * Return the length of an &sk_buff queue. 2124 * This variant can be used in lockless contexts. 2125 */ 2126 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2127 { 2128 return READ_ONCE(list_->qlen); 2129 } 2130 2131 /** 2132 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2133 * @list: queue to initialize 2134 * 2135 * This initializes only the list and queue length aspects of 2136 * an sk_buff_head object. This allows to initialize the list 2137 * aspects of an sk_buff_head without reinitializing things like 2138 * the spinlock. It can also be used for on-stack sk_buff_head 2139 * objects where the spinlock is known to not be used. 2140 */ 2141 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2142 { 2143 list->prev = list->next = (struct sk_buff *)list; 2144 list->qlen = 0; 2145 } 2146 2147 /* 2148 * This function creates a split out lock class for each invocation; 2149 * this is needed for now since a whole lot of users of the skb-queue 2150 * infrastructure in drivers have different locking usage (in hardirq) 2151 * than the networking core (in softirq only). In the long run either the 2152 * network layer or drivers should need annotation to consolidate the 2153 * main types of usage into 3 classes. 2154 */ 2155 static inline void skb_queue_head_init(struct sk_buff_head *list) 2156 { 2157 spin_lock_init(&list->lock); 2158 __skb_queue_head_init(list); 2159 } 2160 2161 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2162 struct lock_class_key *class) 2163 { 2164 skb_queue_head_init(list); 2165 lockdep_set_class(&list->lock, class); 2166 } 2167 2168 /* 2169 * Insert an sk_buff on a list. 2170 * 2171 * The "__skb_xxxx()" functions are the non-atomic ones that 2172 * can only be called with interrupts disabled. 2173 */ 2174 static inline void __skb_insert(struct sk_buff *newsk, 2175 struct sk_buff *prev, struct sk_buff *next, 2176 struct sk_buff_head *list) 2177 { 2178 /* See skb_queue_empty_lockless() and skb_peek_tail() 2179 * for the opposite READ_ONCE() 2180 */ 2181 WRITE_ONCE(newsk->next, next); 2182 WRITE_ONCE(newsk->prev, prev); 2183 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2184 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2185 WRITE_ONCE(list->qlen, list->qlen + 1); 2186 } 2187 2188 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2189 struct sk_buff *prev, 2190 struct sk_buff *next) 2191 { 2192 struct sk_buff *first = list->next; 2193 struct sk_buff *last = list->prev; 2194 2195 WRITE_ONCE(first->prev, prev); 2196 WRITE_ONCE(prev->next, first); 2197 2198 WRITE_ONCE(last->next, next); 2199 WRITE_ONCE(next->prev, last); 2200 } 2201 2202 /** 2203 * skb_queue_splice - join two skb lists, this is designed for stacks 2204 * @list: the new list to add 2205 * @head: the place to add it in the first list 2206 */ 2207 static inline void skb_queue_splice(const struct sk_buff_head *list, 2208 struct sk_buff_head *head) 2209 { 2210 if (!skb_queue_empty(list)) { 2211 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2212 head->qlen += list->qlen; 2213 } 2214 } 2215 2216 /** 2217 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2218 * @list: the new list to add 2219 * @head: the place to add it in the first list 2220 * 2221 * The list at @list is reinitialised 2222 */ 2223 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2224 struct sk_buff_head *head) 2225 { 2226 if (!skb_queue_empty(list)) { 2227 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2228 head->qlen += list->qlen; 2229 __skb_queue_head_init(list); 2230 } 2231 } 2232 2233 /** 2234 * skb_queue_splice_tail - join two skb lists, each list being a queue 2235 * @list: the new list to add 2236 * @head: the place to add it in the first list 2237 */ 2238 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2239 struct sk_buff_head *head) 2240 { 2241 if (!skb_queue_empty(list)) { 2242 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2243 head->qlen += list->qlen; 2244 } 2245 } 2246 2247 /** 2248 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2249 * @list: the new list to add 2250 * @head: the place to add it in the first list 2251 * 2252 * Each of the lists is a queue. 2253 * The list at @list is reinitialised 2254 */ 2255 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2256 struct sk_buff_head *head) 2257 { 2258 if (!skb_queue_empty(list)) { 2259 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2260 head->qlen += list->qlen; 2261 __skb_queue_head_init(list); 2262 } 2263 } 2264 2265 /** 2266 * __skb_queue_after - queue a buffer at the list head 2267 * @list: list to use 2268 * @prev: place after this buffer 2269 * @newsk: buffer to queue 2270 * 2271 * Queue a buffer int the middle of a list. This function takes no locks 2272 * and you must therefore hold required locks before calling it. 2273 * 2274 * A buffer cannot be placed on two lists at the same time. 2275 */ 2276 static inline void __skb_queue_after(struct sk_buff_head *list, 2277 struct sk_buff *prev, 2278 struct sk_buff *newsk) 2279 { 2280 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2281 } 2282 2283 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2284 struct sk_buff_head *list); 2285 2286 static inline void __skb_queue_before(struct sk_buff_head *list, 2287 struct sk_buff *next, 2288 struct sk_buff *newsk) 2289 { 2290 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2291 } 2292 2293 /** 2294 * __skb_queue_head - queue a buffer at the list head 2295 * @list: list to use 2296 * @newsk: buffer to queue 2297 * 2298 * Queue a buffer at the start of a list. This function takes no locks 2299 * and you must therefore hold required locks before calling it. 2300 * 2301 * A buffer cannot be placed on two lists at the same time. 2302 */ 2303 static inline void __skb_queue_head(struct sk_buff_head *list, 2304 struct sk_buff *newsk) 2305 { 2306 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2307 } 2308 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2309 2310 /** 2311 * __skb_queue_tail - queue a buffer at the list tail 2312 * @list: list to use 2313 * @newsk: buffer to queue 2314 * 2315 * Queue a buffer at the end of a list. This function takes no locks 2316 * and you must therefore hold required locks before calling it. 2317 * 2318 * A buffer cannot be placed on two lists at the same time. 2319 */ 2320 static inline void __skb_queue_tail(struct sk_buff_head *list, 2321 struct sk_buff *newsk) 2322 { 2323 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2324 } 2325 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2326 2327 /* 2328 * remove sk_buff from list. _Must_ be called atomically, and with 2329 * the list known.. 2330 */ 2331 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2332 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2333 { 2334 struct sk_buff *next, *prev; 2335 2336 WRITE_ONCE(list->qlen, list->qlen - 1); 2337 next = skb->next; 2338 prev = skb->prev; 2339 skb->next = skb->prev = NULL; 2340 WRITE_ONCE(next->prev, prev); 2341 WRITE_ONCE(prev->next, next); 2342 } 2343 2344 /** 2345 * __skb_dequeue - remove from the head of the queue 2346 * @list: list to dequeue from 2347 * 2348 * Remove the head of the list. This function does not take any locks 2349 * so must be used with appropriate locks held only. The head item is 2350 * returned or %NULL if the list is empty. 2351 */ 2352 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2353 { 2354 struct sk_buff *skb = skb_peek(list); 2355 if (skb) 2356 __skb_unlink(skb, list); 2357 return skb; 2358 } 2359 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2360 2361 /** 2362 * __skb_dequeue_tail - remove from the tail of the queue 2363 * @list: list to dequeue from 2364 * 2365 * Remove the tail of the list. This function does not take any locks 2366 * so must be used with appropriate locks held only. The tail item is 2367 * returned or %NULL if the list is empty. 2368 */ 2369 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2370 { 2371 struct sk_buff *skb = skb_peek_tail(list); 2372 if (skb) 2373 __skb_unlink(skb, list); 2374 return skb; 2375 } 2376 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2377 2378 2379 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2380 { 2381 return skb->data_len; 2382 } 2383 2384 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2385 { 2386 return skb->len - skb->data_len; 2387 } 2388 2389 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2390 { 2391 unsigned int i, len = 0; 2392 2393 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2394 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2395 return len; 2396 } 2397 2398 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2399 { 2400 return skb_headlen(skb) + __skb_pagelen(skb); 2401 } 2402 2403 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2404 int i, struct page *page, 2405 int off, int size) 2406 { 2407 skb_frag_t *frag = &shinfo->frags[i]; 2408 2409 /* 2410 * Propagate page pfmemalloc to the skb if we can. The problem is 2411 * that not all callers have unique ownership of the page but rely 2412 * on page_is_pfmemalloc doing the right thing(tm). 2413 */ 2414 frag->bv_page = page; 2415 frag->bv_offset = off; 2416 skb_frag_size_set(frag, size); 2417 } 2418 2419 /** 2420 * skb_len_add - adds a number to len fields of skb 2421 * @skb: buffer to add len to 2422 * @delta: number of bytes to add 2423 */ 2424 static inline void skb_len_add(struct sk_buff *skb, int delta) 2425 { 2426 skb->len += delta; 2427 skb->data_len += delta; 2428 skb->truesize += delta; 2429 } 2430 2431 /** 2432 * __skb_fill_page_desc - initialise a paged fragment in an skb 2433 * @skb: buffer containing fragment to be initialised 2434 * @i: paged fragment index to initialise 2435 * @page: the page to use for this fragment 2436 * @off: the offset to the data with @page 2437 * @size: the length of the data 2438 * 2439 * Initialises the @i'th fragment of @skb to point to &size bytes at 2440 * offset @off within @page. 2441 * 2442 * Does not take any additional reference on the fragment. 2443 */ 2444 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2445 struct page *page, int off, int size) 2446 { 2447 __skb_fill_page_desc_noacc(skb_shinfo(skb), i, page, off, size); 2448 page = compound_head(page); 2449 if (page_is_pfmemalloc(page)) 2450 skb->pfmemalloc = true; 2451 } 2452 2453 /** 2454 * skb_fill_page_desc - initialise a paged fragment in an skb 2455 * @skb: buffer containing fragment to be initialised 2456 * @i: paged fragment index to initialise 2457 * @page: the page to use for this fragment 2458 * @off: the offset to the data with @page 2459 * @size: the length of the data 2460 * 2461 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2462 * @skb to point to @size bytes at offset @off within @page. In 2463 * addition updates @skb such that @i is the last fragment. 2464 * 2465 * Does not take any additional reference on the fragment. 2466 */ 2467 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2468 struct page *page, int off, int size) 2469 { 2470 __skb_fill_page_desc(skb, i, page, off, size); 2471 skb_shinfo(skb)->nr_frags = i + 1; 2472 } 2473 2474 /** 2475 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2476 * @skb: buffer containing fragment to be initialised 2477 * @i: paged fragment index to initialise 2478 * @page: the page to use for this fragment 2479 * @off: the offset to the data with @page 2480 * @size: the length of the data 2481 * 2482 * Variant of skb_fill_page_desc() which does not deal with 2483 * pfmemalloc, if page is not owned by us. 2484 */ 2485 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2486 struct page *page, int off, 2487 int size) 2488 { 2489 struct skb_shared_info *shinfo = skb_shinfo(skb); 2490 2491 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2492 shinfo->nr_frags = i + 1; 2493 } 2494 2495 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 2496 int size, unsigned int truesize); 2497 2498 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2499 unsigned int truesize); 2500 2501 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2502 2503 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2504 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2505 { 2506 return skb->head + skb->tail; 2507 } 2508 2509 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2510 { 2511 skb->tail = skb->data - skb->head; 2512 } 2513 2514 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2515 { 2516 skb_reset_tail_pointer(skb); 2517 skb->tail += offset; 2518 } 2519 2520 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2521 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2522 { 2523 return skb->tail; 2524 } 2525 2526 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2527 { 2528 skb->tail = skb->data; 2529 } 2530 2531 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2532 { 2533 skb->tail = skb->data + offset; 2534 } 2535 2536 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2537 2538 static inline void skb_assert_len(struct sk_buff *skb) 2539 { 2540 #ifdef CONFIG_DEBUG_NET 2541 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2542 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2543 #endif /* CONFIG_DEBUG_NET */ 2544 } 2545 2546 /* 2547 * Add data to an sk_buff 2548 */ 2549 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2550 void *skb_put(struct sk_buff *skb, unsigned int len); 2551 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2552 { 2553 void *tmp = skb_tail_pointer(skb); 2554 SKB_LINEAR_ASSERT(skb); 2555 skb->tail += len; 2556 skb->len += len; 2557 return tmp; 2558 } 2559 2560 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2561 { 2562 void *tmp = __skb_put(skb, len); 2563 2564 memset(tmp, 0, len); 2565 return tmp; 2566 } 2567 2568 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2569 unsigned int len) 2570 { 2571 void *tmp = __skb_put(skb, len); 2572 2573 memcpy(tmp, data, len); 2574 return tmp; 2575 } 2576 2577 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2578 { 2579 *(u8 *)__skb_put(skb, 1) = val; 2580 } 2581 2582 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2583 { 2584 void *tmp = skb_put(skb, len); 2585 2586 memset(tmp, 0, len); 2587 2588 return tmp; 2589 } 2590 2591 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2592 unsigned int len) 2593 { 2594 void *tmp = skb_put(skb, len); 2595 2596 memcpy(tmp, data, len); 2597 2598 return tmp; 2599 } 2600 2601 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2602 { 2603 *(u8 *)skb_put(skb, 1) = val; 2604 } 2605 2606 void *skb_push(struct sk_buff *skb, unsigned int len); 2607 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2608 { 2609 skb->data -= len; 2610 skb->len += len; 2611 return skb->data; 2612 } 2613 2614 void *skb_pull(struct sk_buff *skb, unsigned int len); 2615 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2616 { 2617 skb->len -= len; 2618 if (unlikely(skb->len < skb->data_len)) { 2619 #if defined(CONFIG_DEBUG_NET) 2620 skb->len += len; 2621 pr_err("__skb_pull(len=%u)\n", len); 2622 skb_dump(KERN_ERR, skb, false); 2623 #endif 2624 BUG(); 2625 } 2626 return skb->data += len; 2627 } 2628 2629 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2630 { 2631 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2632 } 2633 2634 void *skb_pull_data(struct sk_buff *skb, size_t len); 2635 2636 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2637 2638 static inline enum skb_drop_reason 2639 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len) 2640 { 2641 if (likely(len <= skb_headlen(skb))) 2642 return SKB_NOT_DROPPED_YET; 2643 2644 if (unlikely(len > skb->len)) 2645 return SKB_DROP_REASON_PKT_TOO_SMALL; 2646 2647 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb)))) 2648 return SKB_DROP_REASON_NOMEM; 2649 2650 return SKB_NOT_DROPPED_YET; 2651 } 2652 2653 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2654 { 2655 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 2656 } 2657 2658 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2659 { 2660 if (!pskb_may_pull(skb, len)) 2661 return NULL; 2662 2663 skb->len -= len; 2664 return skb->data += len; 2665 } 2666 2667 void skb_condense(struct sk_buff *skb); 2668 2669 /** 2670 * skb_headroom - bytes at buffer head 2671 * @skb: buffer to check 2672 * 2673 * Return the number of bytes of free space at the head of an &sk_buff. 2674 */ 2675 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2676 { 2677 return skb->data - skb->head; 2678 } 2679 2680 /** 2681 * skb_tailroom - bytes at buffer end 2682 * @skb: buffer to check 2683 * 2684 * Return the number of bytes of free space at the tail of an sk_buff 2685 */ 2686 static inline int skb_tailroom(const struct sk_buff *skb) 2687 { 2688 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2689 } 2690 2691 /** 2692 * skb_availroom - bytes at buffer end 2693 * @skb: buffer to check 2694 * 2695 * Return the number of bytes of free space at the tail of an sk_buff 2696 * allocated by sk_stream_alloc() 2697 */ 2698 static inline int skb_availroom(const struct sk_buff *skb) 2699 { 2700 if (skb_is_nonlinear(skb)) 2701 return 0; 2702 2703 return skb->end - skb->tail - skb->reserved_tailroom; 2704 } 2705 2706 /** 2707 * skb_reserve - adjust headroom 2708 * @skb: buffer to alter 2709 * @len: bytes to move 2710 * 2711 * Increase the headroom of an empty &sk_buff by reducing the tail 2712 * room. This is only allowed for an empty buffer. 2713 */ 2714 static inline void skb_reserve(struct sk_buff *skb, int len) 2715 { 2716 skb->data += len; 2717 skb->tail += len; 2718 } 2719 2720 /** 2721 * skb_tailroom_reserve - adjust reserved_tailroom 2722 * @skb: buffer to alter 2723 * @mtu: maximum amount of headlen permitted 2724 * @needed_tailroom: minimum amount of reserved_tailroom 2725 * 2726 * Set reserved_tailroom so that headlen can be as large as possible but 2727 * not larger than mtu and tailroom cannot be smaller than 2728 * needed_tailroom. 2729 * The required headroom should already have been reserved before using 2730 * this function. 2731 */ 2732 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2733 unsigned int needed_tailroom) 2734 { 2735 SKB_LINEAR_ASSERT(skb); 2736 if (mtu < skb_tailroom(skb) - needed_tailroom) 2737 /* use at most mtu */ 2738 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2739 else 2740 /* use up to all available space */ 2741 skb->reserved_tailroom = needed_tailroom; 2742 } 2743 2744 #define ENCAP_TYPE_ETHER 0 2745 #define ENCAP_TYPE_IPPROTO 1 2746 2747 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2748 __be16 protocol) 2749 { 2750 skb->inner_protocol = protocol; 2751 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2752 } 2753 2754 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2755 __u8 ipproto) 2756 { 2757 skb->inner_ipproto = ipproto; 2758 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2759 } 2760 2761 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2762 { 2763 skb->inner_mac_header = skb->mac_header; 2764 skb->inner_network_header = skb->network_header; 2765 skb->inner_transport_header = skb->transport_header; 2766 } 2767 2768 static inline void skb_reset_mac_len(struct sk_buff *skb) 2769 { 2770 skb->mac_len = skb->network_header - skb->mac_header; 2771 } 2772 2773 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2774 *skb) 2775 { 2776 return skb->head + skb->inner_transport_header; 2777 } 2778 2779 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2780 { 2781 return skb_inner_transport_header(skb) - skb->data; 2782 } 2783 2784 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2785 { 2786 skb->inner_transport_header = skb->data - skb->head; 2787 } 2788 2789 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2790 const int offset) 2791 { 2792 skb_reset_inner_transport_header(skb); 2793 skb->inner_transport_header += offset; 2794 } 2795 2796 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2797 { 2798 return skb->head + skb->inner_network_header; 2799 } 2800 2801 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2802 { 2803 skb->inner_network_header = skb->data - skb->head; 2804 } 2805 2806 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2807 const int offset) 2808 { 2809 skb_reset_inner_network_header(skb); 2810 skb->inner_network_header += offset; 2811 } 2812 2813 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2814 { 2815 return skb->head + skb->inner_mac_header; 2816 } 2817 2818 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2819 { 2820 skb->inner_mac_header = skb->data - skb->head; 2821 } 2822 2823 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2824 const int offset) 2825 { 2826 skb_reset_inner_mac_header(skb); 2827 skb->inner_mac_header += offset; 2828 } 2829 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2830 { 2831 return skb->transport_header != (typeof(skb->transport_header))~0U; 2832 } 2833 2834 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2835 { 2836 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 2837 return skb->head + skb->transport_header; 2838 } 2839 2840 static inline void skb_reset_transport_header(struct sk_buff *skb) 2841 { 2842 skb->transport_header = skb->data - skb->head; 2843 } 2844 2845 static inline void skb_set_transport_header(struct sk_buff *skb, 2846 const int offset) 2847 { 2848 skb_reset_transport_header(skb); 2849 skb->transport_header += offset; 2850 } 2851 2852 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2853 { 2854 return skb->head + skb->network_header; 2855 } 2856 2857 static inline void skb_reset_network_header(struct sk_buff *skb) 2858 { 2859 skb->network_header = skb->data - skb->head; 2860 } 2861 2862 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2863 { 2864 skb_reset_network_header(skb); 2865 skb->network_header += offset; 2866 } 2867 2868 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2869 { 2870 return skb->mac_header != (typeof(skb->mac_header))~0U; 2871 } 2872 2873 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2874 { 2875 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 2876 return skb->head + skb->mac_header; 2877 } 2878 2879 static inline int skb_mac_offset(const struct sk_buff *skb) 2880 { 2881 return skb_mac_header(skb) - skb->data; 2882 } 2883 2884 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2885 { 2886 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 2887 return skb->network_header - skb->mac_header; 2888 } 2889 2890 static inline void skb_unset_mac_header(struct sk_buff *skb) 2891 { 2892 skb->mac_header = (typeof(skb->mac_header))~0U; 2893 } 2894 2895 static inline void skb_reset_mac_header(struct sk_buff *skb) 2896 { 2897 skb->mac_header = skb->data - skb->head; 2898 } 2899 2900 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2901 { 2902 skb_reset_mac_header(skb); 2903 skb->mac_header += offset; 2904 } 2905 2906 static inline void skb_pop_mac_header(struct sk_buff *skb) 2907 { 2908 skb->mac_header = skb->network_header; 2909 } 2910 2911 static inline void skb_probe_transport_header(struct sk_buff *skb) 2912 { 2913 struct flow_keys_basic keys; 2914 2915 if (skb_transport_header_was_set(skb)) 2916 return; 2917 2918 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 2919 NULL, 0, 0, 0, 0)) 2920 skb_set_transport_header(skb, keys.control.thoff); 2921 } 2922 2923 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2924 { 2925 if (skb_mac_header_was_set(skb)) { 2926 const unsigned char *old_mac = skb_mac_header(skb); 2927 2928 skb_set_mac_header(skb, -skb->mac_len); 2929 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2930 } 2931 } 2932 2933 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2934 { 2935 return skb->csum_start - skb_headroom(skb); 2936 } 2937 2938 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2939 { 2940 return skb->head + skb->csum_start; 2941 } 2942 2943 static inline int skb_transport_offset(const struct sk_buff *skb) 2944 { 2945 return skb_transport_header(skb) - skb->data; 2946 } 2947 2948 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2949 { 2950 return skb->transport_header - skb->network_header; 2951 } 2952 2953 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2954 { 2955 return skb->inner_transport_header - skb->inner_network_header; 2956 } 2957 2958 static inline int skb_network_offset(const struct sk_buff *skb) 2959 { 2960 return skb_network_header(skb) - skb->data; 2961 } 2962 2963 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2964 { 2965 return skb_inner_network_header(skb) - skb->data; 2966 } 2967 2968 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2969 { 2970 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2971 } 2972 2973 /* 2974 * CPUs often take a performance hit when accessing unaligned memory 2975 * locations. The actual performance hit varies, it can be small if the 2976 * hardware handles it or large if we have to take an exception and fix it 2977 * in software. 2978 * 2979 * Since an ethernet header is 14 bytes network drivers often end up with 2980 * the IP header at an unaligned offset. The IP header can be aligned by 2981 * shifting the start of the packet by 2 bytes. Drivers should do this 2982 * with: 2983 * 2984 * skb_reserve(skb, NET_IP_ALIGN); 2985 * 2986 * The downside to this alignment of the IP header is that the DMA is now 2987 * unaligned. On some architectures the cost of an unaligned DMA is high 2988 * and this cost outweighs the gains made by aligning the IP header. 2989 * 2990 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2991 * to be overridden. 2992 */ 2993 #ifndef NET_IP_ALIGN 2994 #define NET_IP_ALIGN 2 2995 #endif 2996 2997 /* 2998 * The networking layer reserves some headroom in skb data (via 2999 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 3000 * the header has to grow. In the default case, if the header has to grow 3001 * 32 bytes or less we avoid the reallocation. 3002 * 3003 * Unfortunately this headroom changes the DMA alignment of the resulting 3004 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 3005 * on some architectures. An architecture can override this value, 3006 * perhaps setting it to a cacheline in size (since that will maintain 3007 * cacheline alignment of the DMA). It must be a power of 2. 3008 * 3009 * Various parts of the networking layer expect at least 32 bytes of 3010 * headroom, you should not reduce this. 3011 * 3012 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 3013 * to reduce average number of cache lines per packet. 3014 * get_rps_cpu() for example only access one 64 bytes aligned block : 3015 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 3016 */ 3017 #ifndef NET_SKB_PAD 3018 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 3019 #endif 3020 3021 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 3022 3023 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3024 { 3025 if (WARN_ON(skb_is_nonlinear(skb))) 3026 return; 3027 skb->len = len; 3028 skb_set_tail_pointer(skb, len); 3029 } 3030 3031 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3032 { 3033 __skb_set_length(skb, len); 3034 } 3035 3036 void skb_trim(struct sk_buff *skb, unsigned int len); 3037 3038 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3039 { 3040 if (skb->data_len) 3041 return ___pskb_trim(skb, len); 3042 __skb_trim(skb, len); 3043 return 0; 3044 } 3045 3046 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3047 { 3048 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3049 } 3050 3051 /** 3052 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3053 * @skb: buffer to alter 3054 * @len: new length 3055 * 3056 * This is identical to pskb_trim except that the caller knows that 3057 * the skb is not cloned so we should never get an error due to out- 3058 * of-memory. 3059 */ 3060 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3061 { 3062 int err = pskb_trim(skb, len); 3063 BUG_ON(err); 3064 } 3065 3066 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3067 { 3068 unsigned int diff = len - skb->len; 3069 3070 if (skb_tailroom(skb) < diff) { 3071 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3072 GFP_ATOMIC); 3073 if (ret) 3074 return ret; 3075 } 3076 __skb_set_length(skb, len); 3077 return 0; 3078 } 3079 3080 /** 3081 * skb_orphan - orphan a buffer 3082 * @skb: buffer to orphan 3083 * 3084 * If a buffer currently has an owner then we call the owner's 3085 * destructor function and make the @skb unowned. The buffer continues 3086 * to exist but is no longer charged to its former owner. 3087 */ 3088 static inline void skb_orphan(struct sk_buff *skb) 3089 { 3090 if (skb->destructor) { 3091 skb->destructor(skb); 3092 skb->destructor = NULL; 3093 skb->sk = NULL; 3094 } else { 3095 BUG_ON(skb->sk); 3096 } 3097 } 3098 3099 /** 3100 * skb_orphan_frags - orphan the frags contained in a buffer 3101 * @skb: buffer to orphan frags from 3102 * @gfp_mask: allocation mask for replacement pages 3103 * 3104 * For each frag in the SKB which needs a destructor (i.e. has an 3105 * owner) create a copy of that frag and release the original 3106 * page by calling the destructor. 3107 */ 3108 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3109 { 3110 if (likely(!skb_zcopy(skb))) 3111 return 0; 3112 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3113 return 0; 3114 return skb_copy_ubufs(skb, gfp_mask); 3115 } 3116 3117 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3118 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3119 { 3120 if (likely(!skb_zcopy(skb))) 3121 return 0; 3122 return skb_copy_ubufs(skb, gfp_mask); 3123 } 3124 3125 /** 3126 * __skb_queue_purge - empty a list 3127 * @list: list to empty 3128 * 3129 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3130 * the list and one reference dropped. This function does not take the 3131 * list lock and the caller must hold the relevant locks to use it. 3132 */ 3133 static inline void __skb_queue_purge(struct sk_buff_head *list) 3134 { 3135 struct sk_buff *skb; 3136 while ((skb = __skb_dequeue(list)) != NULL) 3137 kfree_skb(skb); 3138 } 3139 void skb_queue_purge(struct sk_buff_head *list); 3140 3141 unsigned int skb_rbtree_purge(struct rb_root *root); 3142 3143 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3144 3145 /** 3146 * netdev_alloc_frag - allocate a page fragment 3147 * @fragsz: fragment size 3148 * 3149 * Allocates a frag from a page for receive buffer. 3150 * Uses GFP_ATOMIC allocations. 3151 */ 3152 static inline void *netdev_alloc_frag(unsigned int fragsz) 3153 { 3154 return __netdev_alloc_frag_align(fragsz, ~0u); 3155 } 3156 3157 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3158 unsigned int align) 3159 { 3160 WARN_ON_ONCE(!is_power_of_2(align)); 3161 return __netdev_alloc_frag_align(fragsz, -align); 3162 } 3163 3164 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3165 gfp_t gfp_mask); 3166 3167 /** 3168 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3169 * @dev: network device to receive on 3170 * @length: length to allocate 3171 * 3172 * Allocate a new &sk_buff and assign it a usage count of one. The 3173 * buffer has unspecified headroom built in. Users should allocate 3174 * the headroom they think they need without accounting for the 3175 * built in space. The built in space is used for optimisations. 3176 * 3177 * %NULL is returned if there is no free memory. Although this function 3178 * allocates memory it can be called from an interrupt. 3179 */ 3180 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3181 unsigned int length) 3182 { 3183 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3184 } 3185 3186 /* legacy helper around __netdev_alloc_skb() */ 3187 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3188 gfp_t gfp_mask) 3189 { 3190 return __netdev_alloc_skb(NULL, length, gfp_mask); 3191 } 3192 3193 /* legacy helper around netdev_alloc_skb() */ 3194 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3195 { 3196 return netdev_alloc_skb(NULL, length); 3197 } 3198 3199 3200 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3201 unsigned int length, gfp_t gfp) 3202 { 3203 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3204 3205 if (NET_IP_ALIGN && skb) 3206 skb_reserve(skb, NET_IP_ALIGN); 3207 return skb; 3208 } 3209 3210 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3211 unsigned int length) 3212 { 3213 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3214 } 3215 3216 static inline void skb_free_frag(void *addr) 3217 { 3218 page_frag_free(addr); 3219 } 3220 3221 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3222 3223 static inline void *napi_alloc_frag(unsigned int fragsz) 3224 { 3225 return __napi_alloc_frag_align(fragsz, ~0u); 3226 } 3227 3228 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3229 unsigned int align) 3230 { 3231 WARN_ON_ONCE(!is_power_of_2(align)); 3232 return __napi_alloc_frag_align(fragsz, -align); 3233 } 3234 3235 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 3236 unsigned int length, gfp_t gfp_mask); 3237 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 3238 unsigned int length) 3239 { 3240 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 3241 } 3242 void napi_consume_skb(struct sk_buff *skb, int budget); 3243 3244 void napi_skb_free_stolen_head(struct sk_buff *skb); 3245 void __kfree_skb_defer(struct sk_buff *skb); 3246 3247 /** 3248 * __dev_alloc_pages - allocate page for network Rx 3249 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3250 * @order: size of the allocation 3251 * 3252 * Allocate a new page. 3253 * 3254 * %NULL is returned if there is no free memory. 3255 */ 3256 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 3257 unsigned int order) 3258 { 3259 /* This piece of code contains several assumptions. 3260 * 1. This is for device Rx, therefor a cold page is preferred. 3261 * 2. The expectation is the user wants a compound page. 3262 * 3. If requesting a order 0 page it will not be compound 3263 * due to the check to see if order has a value in prep_new_page 3264 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3265 * code in gfp_to_alloc_flags that should be enforcing this. 3266 */ 3267 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3268 3269 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 3270 } 3271 3272 static inline struct page *dev_alloc_pages(unsigned int order) 3273 { 3274 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 3275 } 3276 3277 /** 3278 * __dev_alloc_page - allocate a page for network Rx 3279 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3280 * 3281 * Allocate a new page. 3282 * 3283 * %NULL is returned if there is no free memory. 3284 */ 3285 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 3286 { 3287 return __dev_alloc_pages(gfp_mask, 0); 3288 } 3289 3290 static inline struct page *dev_alloc_page(void) 3291 { 3292 return dev_alloc_pages(0); 3293 } 3294 3295 /** 3296 * dev_page_is_reusable - check whether a page can be reused for network Rx 3297 * @page: the page to test 3298 * 3299 * A page shouldn't be considered for reusing/recycling if it was allocated 3300 * under memory pressure or at a distant memory node. 3301 * 3302 * Returns false if this page should be returned to page allocator, true 3303 * otherwise. 3304 */ 3305 static inline bool dev_page_is_reusable(const struct page *page) 3306 { 3307 return likely(page_to_nid(page) == numa_mem_id() && 3308 !page_is_pfmemalloc(page)); 3309 } 3310 3311 /** 3312 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3313 * @page: The page that was allocated from skb_alloc_page 3314 * @skb: The skb that may need pfmemalloc set 3315 */ 3316 static inline void skb_propagate_pfmemalloc(const struct page *page, 3317 struct sk_buff *skb) 3318 { 3319 if (page_is_pfmemalloc(page)) 3320 skb->pfmemalloc = true; 3321 } 3322 3323 /** 3324 * skb_frag_off() - Returns the offset of a skb fragment 3325 * @frag: the paged fragment 3326 */ 3327 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3328 { 3329 return frag->bv_offset; 3330 } 3331 3332 /** 3333 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3334 * @frag: skb fragment 3335 * @delta: value to add 3336 */ 3337 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3338 { 3339 frag->bv_offset += delta; 3340 } 3341 3342 /** 3343 * skb_frag_off_set() - Sets the offset of a skb fragment 3344 * @frag: skb fragment 3345 * @offset: offset of fragment 3346 */ 3347 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3348 { 3349 frag->bv_offset = offset; 3350 } 3351 3352 /** 3353 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3354 * @fragto: skb fragment where offset is set 3355 * @fragfrom: skb fragment offset is copied from 3356 */ 3357 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3358 const skb_frag_t *fragfrom) 3359 { 3360 fragto->bv_offset = fragfrom->bv_offset; 3361 } 3362 3363 /** 3364 * skb_frag_page - retrieve the page referred to by a paged fragment 3365 * @frag: the paged fragment 3366 * 3367 * Returns the &struct page associated with @frag. 3368 */ 3369 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3370 { 3371 return frag->bv_page; 3372 } 3373 3374 /** 3375 * __skb_frag_ref - take an addition reference on a paged fragment. 3376 * @frag: the paged fragment 3377 * 3378 * Takes an additional reference on the paged fragment @frag. 3379 */ 3380 static inline void __skb_frag_ref(skb_frag_t *frag) 3381 { 3382 get_page(skb_frag_page(frag)); 3383 } 3384 3385 /** 3386 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 3387 * @skb: the buffer 3388 * @f: the fragment offset. 3389 * 3390 * Takes an additional reference on the @f'th paged fragment of @skb. 3391 */ 3392 static inline void skb_frag_ref(struct sk_buff *skb, int f) 3393 { 3394 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 3395 } 3396 3397 /** 3398 * __skb_frag_unref - release a reference on a paged fragment. 3399 * @frag: the paged fragment 3400 * @recycle: recycle the page if allocated via page_pool 3401 * 3402 * Releases a reference on the paged fragment @frag 3403 * or recycles the page via the page_pool API. 3404 */ 3405 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle) 3406 { 3407 struct page *page = skb_frag_page(frag); 3408 3409 #ifdef CONFIG_PAGE_POOL 3410 if (recycle && page_pool_return_skb_page(page)) 3411 return; 3412 #endif 3413 put_page(page); 3414 } 3415 3416 /** 3417 * skb_frag_unref - release a reference on a paged fragment of an skb. 3418 * @skb: the buffer 3419 * @f: the fragment offset 3420 * 3421 * Releases a reference on the @f'th paged fragment of @skb. 3422 */ 3423 static inline void skb_frag_unref(struct sk_buff *skb, int f) 3424 { 3425 struct skb_shared_info *shinfo = skb_shinfo(skb); 3426 3427 if (!skb_zcopy_managed(skb)) 3428 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle); 3429 } 3430 3431 /** 3432 * skb_frag_address - gets the address of the data contained in a paged fragment 3433 * @frag: the paged fragment buffer 3434 * 3435 * Returns the address of the data within @frag. The page must already 3436 * be mapped. 3437 */ 3438 static inline void *skb_frag_address(const skb_frag_t *frag) 3439 { 3440 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3441 } 3442 3443 /** 3444 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3445 * @frag: the paged fragment buffer 3446 * 3447 * Returns the address of the data within @frag. Checks that the page 3448 * is mapped and returns %NULL otherwise. 3449 */ 3450 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3451 { 3452 void *ptr = page_address(skb_frag_page(frag)); 3453 if (unlikely(!ptr)) 3454 return NULL; 3455 3456 return ptr + skb_frag_off(frag); 3457 } 3458 3459 /** 3460 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3461 * @fragto: skb fragment where page is set 3462 * @fragfrom: skb fragment page is copied from 3463 */ 3464 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3465 const skb_frag_t *fragfrom) 3466 { 3467 fragto->bv_page = fragfrom->bv_page; 3468 } 3469 3470 /** 3471 * __skb_frag_set_page - sets the page contained in a paged fragment 3472 * @frag: the paged fragment 3473 * @page: the page to set 3474 * 3475 * Sets the fragment @frag to contain @page. 3476 */ 3477 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 3478 { 3479 frag->bv_page = page; 3480 } 3481 3482 /** 3483 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 3484 * @skb: the buffer 3485 * @f: the fragment offset 3486 * @page: the page to set 3487 * 3488 * Sets the @f'th fragment of @skb to contain @page. 3489 */ 3490 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 3491 struct page *page) 3492 { 3493 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 3494 } 3495 3496 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3497 3498 /** 3499 * skb_frag_dma_map - maps a paged fragment via the DMA API 3500 * @dev: the device to map the fragment to 3501 * @frag: the paged fragment to map 3502 * @offset: the offset within the fragment (starting at the 3503 * fragment's own offset) 3504 * @size: the number of bytes to map 3505 * @dir: the direction of the mapping (``PCI_DMA_*``) 3506 * 3507 * Maps the page associated with @frag to @device. 3508 */ 3509 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 3510 const skb_frag_t *frag, 3511 size_t offset, size_t size, 3512 enum dma_data_direction dir) 3513 { 3514 return dma_map_page(dev, skb_frag_page(frag), 3515 skb_frag_off(frag) + offset, size, dir); 3516 } 3517 3518 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3519 gfp_t gfp_mask) 3520 { 3521 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3522 } 3523 3524 3525 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3526 gfp_t gfp_mask) 3527 { 3528 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3529 } 3530 3531 3532 /** 3533 * skb_clone_writable - is the header of a clone writable 3534 * @skb: buffer to check 3535 * @len: length up to which to write 3536 * 3537 * Returns true if modifying the header part of the cloned buffer 3538 * does not requires the data to be copied. 3539 */ 3540 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3541 { 3542 return !skb_header_cloned(skb) && 3543 skb_headroom(skb) + len <= skb->hdr_len; 3544 } 3545 3546 static inline int skb_try_make_writable(struct sk_buff *skb, 3547 unsigned int write_len) 3548 { 3549 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3550 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3551 } 3552 3553 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3554 int cloned) 3555 { 3556 int delta = 0; 3557 3558 if (headroom > skb_headroom(skb)) 3559 delta = headroom - skb_headroom(skb); 3560 3561 if (delta || cloned) 3562 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3563 GFP_ATOMIC); 3564 return 0; 3565 } 3566 3567 /** 3568 * skb_cow - copy header of skb when it is required 3569 * @skb: buffer to cow 3570 * @headroom: needed headroom 3571 * 3572 * If the skb passed lacks sufficient headroom or its data part 3573 * is shared, data is reallocated. If reallocation fails, an error 3574 * is returned and original skb is not changed. 3575 * 3576 * The result is skb with writable area skb->head...skb->tail 3577 * and at least @headroom of space at head. 3578 */ 3579 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3580 { 3581 return __skb_cow(skb, headroom, skb_cloned(skb)); 3582 } 3583 3584 /** 3585 * skb_cow_head - skb_cow but only making the head writable 3586 * @skb: buffer to cow 3587 * @headroom: needed headroom 3588 * 3589 * This function is identical to skb_cow except that we replace the 3590 * skb_cloned check by skb_header_cloned. It should be used when 3591 * you only need to push on some header and do not need to modify 3592 * the data. 3593 */ 3594 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3595 { 3596 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3597 } 3598 3599 /** 3600 * skb_padto - pad an skbuff up to a minimal size 3601 * @skb: buffer to pad 3602 * @len: minimal length 3603 * 3604 * Pads up a buffer to ensure the trailing bytes exist and are 3605 * blanked. If the buffer already contains sufficient data it 3606 * is untouched. Otherwise it is extended. Returns zero on 3607 * success. The skb is freed on error. 3608 */ 3609 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3610 { 3611 unsigned int size = skb->len; 3612 if (likely(size >= len)) 3613 return 0; 3614 return skb_pad(skb, len - size); 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 * @free_on_error: free buffer on error 3622 * 3623 * Pads up a buffer to ensure the trailing bytes exist and are 3624 * blanked. If the buffer already contains sufficient data it 3625 * is untouched. Otherwise it is extended. Returns zero on 3626 * success. The skb is freed on error if @free_on_error is true. 3627 */ 3628 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3629 unsigned int len, 3630 bool free_on_error) 3631 { 3632 unsigned int size = skb->len; 3633 3634 if (unlikely(size < len)) { 3635 len -= size; 3636 if (__skb_pad(skb, len, free_on_error)) 3637 return -ENOMEM; 3638 __skb_put(skb, len); 3639 } 3640 return 0; 3641 } 3642 3643 /** 3644 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3645 * @skb: buffer to pad 3646 * @len: minimal length 3647 * 3648 * Pads up a buffer to ensure the trailing bytes exist and are 3649 * blanked. If the buffer already contains sufficient data it 3650 * is untouched. Otherwise it is extended. Returns zero on 3651 * success. The skb is freed on error. 3652 */ 3653 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3654 { 3655 return __skb_put_padto(skb, len, true); 3656 } 3657 3658 static inline int skb_add_data(struct sk_buff *skb, 3659 struct iov_iter *from, int copy) 3660 { 3661 const int off = skb->len; 3662 3663 if (skb->ip_summed == CHECKSUM_NONE) { 3664 __wsum csum = 0; 3665 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 3666 &csum, from)) { 3667 skb->csum = csum_block_add(skb->csum, csum, off); 3668 return 0; 3669 } 3670 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3671 return 0; 3672 3673 __skb_trim(skb, off); 3674 return -EFAULT; 3675 } 3676 3677 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3678 const struct page *page, int off) 3679 { 3680 if (skb_zcopy(skb)) 3681 return false; 3682 if (i) { 3683 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3684 3685 return page == skb_frag_page(frag) && 3686 off == skb_frag_off(frag) + skb_frag_size(frag); 3687 } 3688 return false; 3689 } 3690 3691 static inline int __skb_linearize(struct sk_buff *skb) 3692 { 3693 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3694 } 3695 3696 /** 3697 * skb_linearize - convert paged skb to linear one 3698 * @skb: buffer to linarize 3699 * 3700 * If there is no free memory -ENOMEM is returned, otherwise zero 3701 * is returned and the old skb data released. 3702 */ 3703 static inline int skb_linearize(struct sk_buff *skb) 3704 { 3705 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3706 } 3707 3708 /** 3709 * skb_has_shared_frag - can any frag be overwritten 3710 * @skb: buffer to test 3711 * 3712 * Return true if the skb has at least one frag that might be modified 3713 * by an external entity (as in vmsplice()/sendfile()) 3714 */ 3715 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3716 { 3717 return skb_is_nonlinear(skb) && 3718 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 3719 } 3720 3721 /** 3722 * skb_linearize_cow - make sure skb is linear and writable 3723 * @skb: buffer to process 3724 * 3725 * If there is no free memory -ENOMEM is returned, otherwise zero 3726 * is returned and the old skb data released. 3727 */ 3728 static inline int skb_linearize_cow(struct sk_buff *skb) 3729 { 3730 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3731 __skb_linearize(skb) : 0; 3732 } 3733 3734 static __always_inline void 3735 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3736 unsigned int off) 3737 { 3738 if (skb->ip_summed == CHECKSUM_COMPLETE) 3739 skb->csum = csum_block_sub(skb->csum, 3740 csum_partial(start, len, 0), off); 3741 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3742 skb_checksum_start_offset(skb) < 0) 3743 skb->ip_summed = CHECKSUM_NONE; 3744 } 3745 3746 /** 3747 * skb_postpull_rcsum - update checksum for received skb after pull 3748 * @skb: buffer to update 3749 * @start: start of data before pull 3750 * @len: length of data pulled 3751 * 3752 * After doing a pull on a received packet, you need to call this to 3753 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3754 * CHECKSUM_NONE so that it can be recomputed from scratch. 3755 */ 3756 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3757 const void *start, unsigned int len) 3758 { 3759 if (skb->ip_summed == CHECKSUM_COMPLETE) 3760 skb->csum = wsum_negate(csum_partial(start, len, 3761 wsum_negate(skb->csum))); 3762 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3763 skb_checksum_start_offset(skb) < 0) 3764 skb->ip_summed = CHECKSUM_NONE; 3765 } 3766 3767 static __always_inline void 3768 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3769 unsigned int off) 3770 { 3771 if (skb->ip_summed == CHECKSUM_COMPLETE) 3772 skb->csum = csum_block_add(skb->csum, 3773 csum_partial(start, len, 0), off); 3774 } 3775 3776 /** 3777 * skb_postpush_rcsum - update checksum for received skb after push 3778 * @skb: buffer to update 3779 * @start: start of data after push 3780 * @len: length of data pushed 3781 * 3782 * After doing a push on a received packet, you need to call this to 3783 * update the CHECKSUM_COMPLETE checksum. 3784 */ 3785 static inline void skb_postpush_rcsum(struct sk_buff *skb, 3786 const void *start, unsigned int len) 3787 { 3788 __skb_postpush_rcsum(skb, start, len, 0); 3789 } 3790 3791 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3792 3793 /** 3794 * skb_push_rcsum - push skb and update receive checksum 3795 * @skb: buffer to update 3796 * @len: length of data pulled 3797 * 3798 * This function performs an skb_push on the packet and updates 3799 * the CHECKSUM_COMPLETE checksum. It should be used on 3800 * receive path processing instead of skb_push unless you know 3801 * that the checksum difference is zero (e.g., a valid IP header) 3802 * or you are setting ip_summed to CHECKSUM_NONE. 3803 */ 3804 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3805 { 3806 skb_push(skb, len); 3807 skb_postpush_rcsum(skb, skb->data, len); 3808 return skb->data; 3809 } 3810 3811 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 3812 /** 3813 * pskb_trim_rcsum - trim received skb and update checksum 3814 * @skb: buffer to trim 3815 * @len: new length 3816 * 3817 * This is exactly the same as pskb_trim except that it ensures the 3818 * checksum of received packets are still valid after the operation. 3819 * It can change skb pointers. 3820 */ 3821 3822 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3823 { 3824 if (likely(len >= skb->len)) 3825 return 0; 3826 return pskb_trim_rcsum_slow(skb, len); 3827 } 3828 3829 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3830 { 3831 if (skb->ip_summed == CHECKSUM_COMPLETE) 3832 skb->ip_summed = CHECKSUM_NONE; 3833 __skb_trim(skb, len); 3834 return 0; 3835 } 3836 3837 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3838 { 3839 if (skb->ip_summed == CHECKSUM_COMPLETE) 3840 skb->ip_summed = CHECKSUM_NONE; 3841 return __skb_grow(skb, len); 3842 } 3843 3844 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3845 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 3846 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 3847 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3848 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3849 3850 #define skb_queue_walk(queue, skb) \ 3851 for (skb = (queue)->next; \ 3852 skb != (struct sk_buff *)(queue); \ 3853 skb = skb->next) 3854 3855 #define skb_queue_walk_safe(queue, skb, tmp) \ 3856 for (skb = (queue)->next, tmp = skb->next; \ 3857 skb != (struct sk_buff *)(queue); \ 3858 skb = tmp, tmp = skb->next) 3859 3860 #define skb_queue_walk_from(queue, skb) \ 3861 for (; skb != (struct sk_buff *)(queue); \ 3862 skb = skb->next) 3863 3864 #define skb_rbtree_walk(skb, root) \ 3865 for (skb = skb_rb_first(root); skb != NULL; \ 3866 skb = skb_rb_next(skb)) 3867 3868 #define skb_rbtree_walk_from(skb) \ 3869 for (; skb != NULL; \ 3870 skb = skb_rb_next(skb)) 3871 3872 #define skb_rbtree_walk_from_safe(skb, tmp) \ 3873 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3874 skb = tmp) 3875 3876 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 3877 for (tmp = skb->next; \ 3878 skb != (struct sk_buff *)(queue); \ 3879 skb = tmp, tmp = skb->next) 3880 3881 #define skb_queue_reverse_walk(queue, skb) \ 3882 for (skb = (queue)->prev; \ 3883 skb != (struct sk_buff *)(queue); \ 3884 skb = skb->prev) 3885 3886 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3887 for (skb = (queue)->prev, tmp = skb->prev; \ 3888 skb != (struct sk_buff *)(queue); \ 3889 skb = tmp, tmp = skb->prev) 3890 3891 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3892 for (tmp = skb->prev; \ 3893 skb != (struct sk_buff *)(queue); \ 3894 skb = tmp, tmp = skb->prev) 3895 3896 static inline bool skb_has_frag_list(const struct sk_buff *skb) 3897 { 3898 return skb_shinfo(skb)->frag_list != NULL; 3899 } 3900 3901 static inline void skb_frag_list_init(struct sk_buff *skb) 3902 { 3903 skb_shinfo(skb)->frag_list = NULL; 3904 } 3905 3906 #define skb_walk_frags(skb, iter) \ 3907 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3908 3909 3910 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 3911 int *err, long *timeo_p, 3912 const struct sk_buff *skb); 3913 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3914 struct sk_buff_head *queue, 3915 unsigned int flags, 3916 int *off, int *err, 3917 struct sk_buff **last); 3918 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 3919 struct sk_buff_head *queue, 3920 unsigned int flags, int *off, int *err, 3921 struct sk_buff **last); 3922 struct sk_buff *__skb_recv_datagram(struct sock *sk, 3923 struct sk_buff_head *sk_queue, 3924 unsigned int flags, int *off, int *err); 3925 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 3926 __poll_t datagram_poll(struct file *file, struct socket *sock, 3927 struct poll_table_struct *wait); 3928 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3929 struct iov_iter *to, int size); 3930 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3931 struct msghdr *msg, int size) 3932 { 3933 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3934 } 3935 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3936 struct msghdr *msg); 3937 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 3938 struct iov_iter *to, int len, 3939 struct ahash_request *hash); 3940 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3941 struct iov_iter *from, int len); 3942 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3943 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3944 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3945 static inline void skb_free_datagram_locked(struct sock *sk, 3946 struct sk_buff *skb) 3947 { 3948 __skb_free_datagram_locked(sk, skb, 0); 3949 } 3950 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3951 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3952 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3953 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3954 int len); 3955 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3956 struct pipe_inode_info *pipe, unsigned int len, 3957 unsigned int flags); 3958 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 3959 int len); 3960 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 3961 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3962 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3963 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3964 int len, int hlen); 3965 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3966 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3967 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3968 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu); 3969 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len); 3970 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3971 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 3972 unsigned int offset); 3973 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3974 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 3975 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3976 int skb_vlan_pop(struct sk_buff *skb); 3977 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3978 int skb_eth_pop(struct sk_buff *skb); 3979 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 3980 const unsigned char *src); 3981 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 3982 int mac_len, bool ethernet); 3983 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 3984 bool ethernet); 3985 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 3986 int skb_mpls_dec_ttl(struct sk_buff *skb); 3987 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3988 gfp_t gfp); 3989 3990 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3991 { 3992 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 3993 } 3994 3995 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3996 { 3997 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3998 } 3999 4000 struct skb_checksum_ops { 4001 __wsum (*update)(const void *mem, int len, __wsum wsum); 4002 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 4003 }; 4004 4005 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 4006 4007 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 4008 __wsum csum, const struct skb_checksum_ops *ops); 4009 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 4010 __wsum csum); 4011 4012 static inline void * __must_check 4013 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 4014 const void *data, int hlen, void *buffer) 4015 { 4016 if (likely(hlen - offset >= len)) 4017 return (void *)data + offset; 4018 4019 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 4020 return NULL; 4021 4022 return buffer; 4023 } 4024 4025 static inline void * __must_check 4026 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4027 { 4028 return __skb_header_pointer(skb, offset, len, skb->data, 4029 skb_headlen(skb), buffer); 4030 } 4031 4032 /** 4033 * skb_needs_linearize - check if we need to linearize a given skb 4034 * depending on the given device features. 4035 * @skb: socket buffer to check 4036 * @features: net device features 4037 * 4038 * Returns true if either: 4039 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4040 * 2. skb is fragmented and the device does not support SG. 4041 */ 4042 static inline bool skb_needs_linearize(struct sk_buff *skb, 4043 netdev_features_t features) 4044 { 4045 return skb_is_nonlinear(skb) && 4046 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4047 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4048 } 4049 4050 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4051 void *to, 4052 const unsigned int len) 4053 { 4054 memcpy(to, skb->data, len); 4055 } 4056 4057 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4058 const int offset, void *to, 4059 const unsigned int len) 4060 { 4061 memcpy(to, skb->data + offset, len); 4062 } 4063 4064 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4065 const void *from, 4066 const unsigned int len) 4067 { 4068 memcpy(skb->data, from, len); 4069 } 4070 4071 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4072 const int offset, 4073 const void *from, 4074 const unsigned int len) 4075 { 4076 memcpy(skb->data + offset, from, len); 4077 } 4078 4079 void skb_init(void); 4080 4081 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4082 { 4083 return skb->tstamp; 4084 } 4085 4086 /** 4087 * skb_get_timestamp - get timestamp from a skb 4088 * @skb: skb to get stamp from 4089 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4090 * 4091 * Timestamps are stored in the skb as offsets to a base timestamp. 4092 * This function converts the offset back to a struct timeval and stores 4093 * it in stamp. 4094 */ 4095 static inline void skb_get_timestamp(const struct sk_buff *skb, 4096 struct __kernel_old_timeval *stamp) 4097 { 4098 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4099 } 4100 4101 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4102 struct __kernel_sock_timeval *stamp) 4103 { 4104 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4105 4106 stamp->tv_sec = ts.tv_sec; 4107 stamp->tv_usec = ts.tv_nsec / 1000; 4108 } 4109 4110 static inline void skb_get_timestampns(const struct sk_buff *skb, 4111 struct __kernel_old_timespec *stamp) 4112 { 4113 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4114 4115 stamp->tv_sec = ts.tv_sec; 4116 stamp->tv_nsec = ts.tv_nsec; 4117 } 4118 4119 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4120 struct __kernel_timespec *stamp) 4121 { 4122 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4123 4124 stamp->tv_sec = ts.tv_sec; 4125 stamp->tv_nsec = ts.tv_nsec; 4126 } 4127 4128 static inline void __net_timestamp(struct sk_buff *skb) 4129 { 4130 skb->tstamp = ktime_get_real(); 4131 skb->mono_delivery_time = 0; 4132 } 4133 4134 static inline ktime_t net_timedelta(ktime_t t) 4135 { 4136 return ktime_sub(ktime_get_real(), t); 4137 } 4138 4139 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4140 bool mono) 4141 { 4142 skb->tstamp = kt; 4143 skb->mono_delivery_time = kt && mono; 4144 } 4145 4146 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4147 4148 /* It is used in the ingress path to clear the delivery_time. 4149 * If needed, set the skb->tstamp to the (rcv) timestamp. 4150 */ 4151 static inline void skb_clear_delivery_time(struct sk_buff *skb) 4152 { 4153 if (skb->mono_delivery_time) { 4154 skb->mono_delivery_time = 0; 4155 if (static_branch_unlikely(&netstamp_needed_key)) 4156 skb->tstamp = ktime_get_real(); 4157 else 4158 skb->tstamp = 0; 4159 } 4160 } 4161 4162 static inline void skb_clear_tstamp(struct sk_buff *skb) 4163 { 4164 if (skb->mono_delivery_time) 4165 return; 4166 4167 skb->tstamp = 0; 4168 } 4169 4170 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4171 { 4172 if (skb->mono_delivery_time) 4173 return 0; 4174 4175 return skb->tstamp; 4176 } 4177 4178 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4179 { 4180 if (!skb->mono_delivery_time && skb->tstamp) 4181 return skb->tstamp; 4182 4183 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4184 return ktime_get_real(); 4185 4186 return 0; 4187 } 4188 4189 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4190 { 4191 return skb_shinfo(skb)->meta_len; 4192 } 4193 4194 static inline void *skb_metadata_end(const struct sk_buff *skb) 4195 { 4196 return skb_mac_header(skb); 4197 } 4198 4199 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4200 const struct sk_buff *skb_b, 4201 u8 meta_len) 4202 { 4203 const void *a = skb_metadata_end(skb_a); 4204 const void *b = skb_metadata_end(skb_b); 4205 /* Using more efficient varaiant than plain call to memcmp(). */ 4206 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64 4207 u64 diffs = 0; 4208 4209 switch (meta_len) { 4210 #define __it(x, op) (x -= sizeof(u##op)) 4211 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4212 case 32: diffs |= __it_diff(a, b, 64); 4213 fallthrough; 4214 case 24: diffs |= __it_diff(a, b, 64); 4215 fallthrough; 4216 case 16: diffs |= __it_diff(a, b, 64); 4217 fallthrough; 4218 case 8: diffs |= __it_diff(a, b, 64); 4219 break; 4220 case 28: diffs |= __it_diff(a, b, 64); 4221 fallthrough; 4222 case 20: diffs |= __it_diff(a, b, 64); 4223 fallthrough; 4224 case 12: diffs |= __it_diff(a, b, 64); 4225 fallthrough; 4226 case 4: diffs |= __it_diff(a, b, 32); 4227 break; 4228 } 4229 return diffs; 4230 #else 4231 return memcmp(a - meta_len, b - meta_len, meta_len); 4232 #endif 4233 } 4234 4235 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4236 const struct sk_buff *skb_b) 4237 { 4238 u8 len_a = skb_metadata_len(skb_a); 4239 u8 len_b = skb_metadata_len(skb_b); 4240 4241 if (!(len_a | len_b)) 4242 return false; 4243 4244 return len_a != len_b ? 4245 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4246 } 4247 4248 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4249 { 4250 skb_shinfo(skb)->meta_len = meta_len; 4251 } 4252 4253 static inline void skb_metadata_clear(struct sk_buff *skb) 4254 { 4255 skb_metadata_set(skb, 0); 4256 } 4257 4258 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4259 4260 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4261 4262 void skb_clone_tx_timestamp(struct sk_buff *skb); 4263 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4264 4265 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4266 4267 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4268 { 4269 } 4270 4271 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4272 { 4273 return false; 4274 } 4275 4276 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4277 4278 /** 4279 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4280 * 4281 * PHY drivers may accept clones of transmitted packets for 4282 * timestamping via their phy_driver.txtstamp method. These drivers 4283 * must call this function to return the skb back to the stack with a 4284 * timestamp. 4285 * 4286 * @skb: clone of the original outgoing packet 4287 * @hwtstamps: hardware time stamps 4288 * 4289 */ 4290 void skb_complete_tx_timestamp(struct sk_buff *skb, 4291 struct skb_shared_hwtstamps *hwtstamps); 4292 4293 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4294 struct skb_shared_hwtstamps *hwtstamps, 4295 struct sock *sk, int tstype); 4296 4297 /** 4298 * skb_tstamp_tx - queue clone of skb with send time stamps 4299 * @orig_skb: the original outgoing packet 4300 * @hwtstamps: hardware time stamps, may be NULL if not available 4301 * 4302 * If the skb has a socket associated, then this function clones the 4303 * skb (thus sharing the actual data and optional structures), stores 4304 * the optional hardware time stamping information (if non NULL) or 4305 * generates a software time stamp (otherwise), then queues the clone 4306 * to the error queue of the socket. Errors are silently ignored. 4307 */ 4308 void skb_tstamp_tx(struct sk_buff *orig_skb, 4309 struct skb_shared_hwtstamps *hwtstamps); 4310 4311 /** 4312 * skb_tx_timestamp() - Driver hook for transmit timestamping 4313 * 4314 * Ethernet MAC Drivers should call this function in their hard_xmit() 4315 * function immediately before giving the sk_buff to the MAC hardware. 4316 * 4317 * Specifically, one should make absolutely sure that this function is 4318 * called before TX completion of this packet can trigger. Otherwise 4319 * the packet could potentially already be freed. 4320 * 4321 * @skb: A socket buffer. 4322 */ 4323 static inline void skb_tx_timestamp(struct sk_buff *skb) 4324 { 4325 skb_clone_tx_timestamp(skb); 4326 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 4327 skb_tstamp_tx(skb, NULL); 4328 } 4329 4330 /** 4331 * skb_complete_wifi_ack - deliver skb with wifi status 4332 * 4333 * @skb: the original outgoing packet 4334 * @acked: ack status 4335 * 4336 */ 4337 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4338 4339 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4340 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4341 4342 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4343 { 4344 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4345 skb->csum_valid || 4346 (skb->ip_summed == CHECKSUM_PARTIAL && 4347 skb_checksum_start_offset(skb) >= 0)); 4348 } 4349 4350 /** 4351 * skb_checksum_complete - Calculate checksum of an entire packet 4352 * @skb: packet to process 4353 * 4354 * This function calculates the checksum over the entire packet plus 4355 * the value of skb->csum. The latter can be used to supply the 4356 * checksum of a pseudo header as used by TCP/UDP. It returns the 4357 * checksum. 4358 * 4359 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4360 * this function can be used to verify that checksum on received 4361 * packets. In that case the function should return zero if the 4362 * checksum is correct. In particular, this function will return zero 4363 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4364 * hardware has already verified the correctness of the checksum. 4365 */ 4366 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4367 { 4368 return skb_csum_unnecessary(skb) ? 4369 0 : __skb_checksum_complete(skb); 4370 } 4371 4372 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4373 { 4374 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4375 if (skb->csum_level == 0) 4376 skb->ip_summed = CHECKSUM_NONE; 4377 else 4378 skb->csum_level--; 4379 } 4380 } 4381 4382 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4383 { 4384 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4385 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4386 skb->csum_level++; 4387 } else if (skb->ip_summed == CHECKSUM_NONE) { 4388 skb->ip_summed = CHECKSUM_UNNECESSARY; 4389 skb->csum_level = 0; 4390 } 4391 } 4392 4393 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4394 { 4395 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4396 skb->ip_summed = CHECKSUM_NONE; 4397 skb->csum_level = 0; 4398 } 4399 } 4400 4401 /* Check if we need to perform checksum complete validation. 4402 * 4403 * Returns true if checksum complete is needed, false otherwise 4404 * (either checksum is unnecessary or zero checksum is allowed). 4405 */ 4406 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4407 bool zero_okay, 4408 __sum16 check) 4409 { 4410 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4411 skb->csum_valid = 1; 4412 __skb_decr_checksum_unnecessary(skb); 4413 return false; 4414 } 4415 4416 return true; 4417 } 4418 4419 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4420 * in checksum_init. 4421 */ 4422 #define CHECKSUM_BREAK 76 4423 4424 /* Unset checksum-complete 4425 * 4426 * Unset checksum complete can be done when packet is being modified 4427 * (uncompressed for instance) and checksum-complete value is 4428 * invalidated. 4429 */ 4430 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4431 { 4432 if (skb->ip_summed == CHECKSUM_COMPLETE) 4433 skb->ip_summed = CHECKSUM_NONE; 4434 } 4435 4436 /* Validate (init) checksum based on checksum complete. 4437 * 4438 * Return values: 4439 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4440 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4441 * checksum is stored in skb->csum for use in __skb_checksum_complete 4442 * non-zero: value of invalid checksum 4443 * 4444 */ 4445 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4446 bool complete, 4447 __wsum psum) 4448 { 4449 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4450 if (!csum_fold(csum_add(psum, skb->csum))) { 4451 skb->csum_valid = 1; 4452 return 0; 4453 } 4454 } 4455 4456 skb->csum = psum; 4457 4458 if (complete || skb->len <= CHECKSUM_BREAK) { 4459 __sum16 csum; 4460 4461 csum = __skb_checksum_complete(skb); 4462 skb->csum_valid = !csum; 4463 return csum; 4464 } 4465 4466 return 0; 4467 } 4468 4469 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4470 { 4471 return 0; 4472 } 4473 4474 /* Perform checksum validate (init). Note that this is a macro since we only 4475 * want to calculate the pseudo header which is an input function if necessary. 4476 * First we try to validate without any computation (checksum unnecessary) and 4477 * then calculate based on checksum complete calling the function to compute 4478 * pseudo header. 4479 * 4480 * Return values: 4481 * 0: checksum is validated or try to in skb_checksum_complete 4482 * non-zero: value of invalid checksum 4483 */ 4484 #define __skb_checksum_validate(skb, proto, complete, \ 4485 zero_okay, check, compute_pseudo) \ 4486 ({ \ 4487 __sum16 __ret = 0; \ 4488 skb->csum_valid = 0; \ 4489 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4490 __ret = __skb_checksum_validate_complete(skb, \ 4491 complete, compute_pseudo(skb, proto)); \ 4492 __ret; \ 4493 }) 4494 4495 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4496 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4497 4498 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4499 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4500 4501 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4502 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4503 4504 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4505 compute_pseudo) \ 4506 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4507 4508 #define skb_checksum_simple_validate(skb) \ 4509 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4510 4511 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4512 { 4513 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4514 } 4515 4516 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4517 { 4518 skb->csum = ~pseudo; 4519 skb->ip_summed = CHECKSUM_COMPLETE; 4520 } 4521 4522 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4523 do { \ 4524 if (__skb_checksum_convert_check(skb)) \ 4525 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4526 } while (0) 4527 4528 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4529 u16 start, u16 offset) 4530 { 4531 skb->ip_summed = CHECKSUM_PARTIAL; 4532 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4533 skb->csum_offset = offset - start; 4534 } 4535 4536 /* Update skbuf and packet to reflect the remote checksum offload operation. 4537 * When called, ptr indicates the starting point for skb->csum when 4538 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4539 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4540 */ 4541 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4542 int start, int offset, bool nopartial) 4543 { 4544 __wsum delta; 4545 4546 if (!nopartial) { 4547 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4548 return; 4549 } 4550 4551 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4552 __skb_checksum_complete(skb); 4553 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4554 } 4555 4556 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4557 4558 /* Adjust skb->csum since we changed the packet */ 4559 skb->csum = csum_add(skb->csum, delta); 4560 } 4561 4562 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4563 { 4564 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4565 return (void *)(skb->_nfct & NFCT_PTRMASK); 4566 #else 4567 return NULL; 4568 #endif 4569 } 4570 4571 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4572 { 4573 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4574 return skb->_nfct; 4575 #else 4576 return 0UL; 4577 #endif 4578 } 4579 4580 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4581 { 4582 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4583 skb->slow_gro |= !!nfct; 4584 skb->_nfct = nfct; 4585 #endif 4586 } 4587 4588 #ifdef CONFIG_SKB_EXTENSIONS 4589 enum skb_ext_id { 4590 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4591 SKB_EXT_BRIDGE_NF, 4592 #endif 4593 #ifdef CONFIG_XFRM 4594 SKB_EXT_SEC_PATH, 4595 #endif 4596 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4597 TC_SKB_EXT, 4598 #endif 4599 #if IS_ENABLED(CONFIG_MPTCP) 4600 SKB_EXT_MPTCP, 4601 #endif 4602 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 4603 SKB_EXT_MCTP, 4604 #endif 4605 SKB_EXT_NUM, /* must be last */ 4606 }; 4607 4608 /** 4609 * struct skb_ext - sk_buff extensions 4610 * @refcnt: 1 on allocation, deallocated on 0 4611 * @offset: offset to add to @data to obtain extension address 4612 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4613 * @data: start of extension data, variable sized 4614 * 4615 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4616 * to use 'u8' types while allowing up to 2kb worth of extension data. 4617 */ 4618 struct skb_ext { 4619 refcount_t refcnt; 4620 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4621 u8 chunks; /* same */ 4622 char data[] __aligned(8); 4623 }; 4624 4625 struct skb_ext *__skb_ext_alloc(gfp_t flags); 4626 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 4627 struct skb_ext *ext); 4628 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4629 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4630 void __skb_ext_put(struct skb_ext *ext); 4631 4632 static inline void skb_ext_put(struct sk_buff *skb) 4633 { 4634 if (skb->active_extensions) 4635 __skb_ext_put(skb->extensions); 4636 } 4637 4638 static inline void __skb_ext_copy(struct sk_buff *dst, 4639 const struct sk_buff *src) 4640 { 4641 dst->active_extensions = src->active_extensions; 4642 4643 if (src->active_extensions) { 4644 struct skb_ext *ext = src->extensions; 4645 4646 refcount_inc(&ext->refcnt); 4647 dst->extensions = ext; 4648 } 4649 } 4650 4651 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4652 { 4653 skb_ext_put(dst); 4654 __skb_ext_copy(dst, src); 4655 } 4656 4657 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4658 { 4659 return !!ext->offset[i]; 4660 } 4661 4662 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4663 { 4664 return skb->active_extensions & (1 << id); 4665 } 4666 4667 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4668 { 4669 if (skb_ext_exist(skb, id)) 4670 __skb_ext_del(skb, id); 4671 } 4672 4673 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4674 { 4675 if (skb_ext_exist(skb, id)) { 4676 struct skb_ext *ext = skb->extensions; 4677 4678 return (void *)ext + (ext->offset[id] << 3); 4679 } 4680 4681 return NULL; 4682 } 4683 4684 static inline void skb_ext_reset(struct sk_buff *skb) 4685 { 4686 if (unlikely(skb->active_extensions)) { 4687 __skb_ext_put(skb->extensions); 4688 skb->active_extensions = 0; 4689 } 4690 } 4691 4692 static inline bool skb_has_extensions(struct sk_buff *skb) 4693 { 4694 return unlikely(skb->active_extensions); 4695 } 4696 #else 4697 static inline void skb_ext_put(struct sk_buff *skb) {} 4698 static inline void skb_ext_reset(struct sk_buff *skb) {} 4699 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4700 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4701 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4702 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 4703 #endif /* CONFIG_SKB_EXTENSIONS */ 4704 4705 static inline void nf_reset_ct(struct sk_buff *skb) 4706 { 4707 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4708 nf_conntrack_put(skb_nfct(skb)); 4709 skb->_nfct = 0; 4710 #endif 4711 } 4712 4713 static inline void nf_reset_trace(struct sk_buff *skb) 4714 { 4715 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4716 skb->nf_trace = 0; 4717 #endif 4718 } 4719 4720 static inline void ipvs_reset(struct sk_buff *skb) 4721 { 4722 #if IS_ENABLED(CONFIG_IP_VS) 4723 skb->ipvs_property = 0; 4724 #endif 4725 } 4726 4727 /* Note: This doesn't put any conntrack info in dst. */ 4728 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4729 bool copy) 4730 { 4731 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4732 dst->_nfct = src->_nfct; 4733 nf_conntrack_get(skb_nfct(src)); 4734 #endif 4735 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4736 if (copy) 4737 dst->nf_trace = src->nf_trace; 4738 #endif 4739 } 4740 4741 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 4742 { 4743 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4744 nf_conntrack_put(skb_nfct(dst)); 4745 #endif 4746 dst->slow_gro = src->slow_gro; 4747 __nf_copy(dst, src, true); 4748 } 4749 4750 #ifdef CONFIG_NETWORK_SECMARK 4751 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4752 { 4753 to->secmark = from->secmark; 4754 } 4755 4756 static inline void skb_init_secmark(struct sk_buff *skb) 4757 { 4758 skb->secmark = 0; 4759 } 4760 #else 4761 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4762 { } 4763 4764 static inline void skb_init_secmark(struct sk_buff *skb) 4765 { } 4766 #endif 4767 4768 static inline int secpath_exists(const struct sk_buff *skb) 4769 { 4770 #ifdef CONFIG_XFRM 4771 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 4772 #else 4773 return 0; 4774 #endif 4775 } 4776 4777 static inline bool skb_irq_freeable(const struct sk_buff *skb) 4778 { 4779 return !skb->destructor && 4780 !secpath_exists(skb) && 4781 !skb_nfct(skb) && 4782 !skb->_skb_refdst && 4783 !skb_has_frag_list(skb); 4784 } 4785 4786 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 4787 { 4788 skb->queue_mapping = queue_mapping; 4789 } 4790 4791 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 4792 { 4793 return skb->queue_mapping; 4794 } 4795 4796 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 4797 { 4798 to->queue_mapping = from->queue_mapping; 4799 } 4800 4801 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 4802 { 4803 skb->queue_mapping = rx_queue + 1; 4804 } 4805 4806 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 4807 { 4808 return skb->queue_mapping - 1; 4809 } 4810 4811 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 4812 { 4813 return skb->queue_mapping != 0; 4814 } 4815 4816 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 4817 { 4818 skb->dst_pending_confirm = val; 4819 } 4820 4821 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 4822 { 4823 return skb->dst_pending_confirm != 0; 4824 } 4825 4826 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 4827 { 4828 #ifdef CONFIG_XFRM 4829 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 4830 #else 4831 return NULL; 4832 #endif 4833 } 4834 4835 /* Keeps track of mac header offset relative to skb->head. 4836 * It is useful for TSO of Tunneling protocol. e.g. GRE. 4837 * For non-tunnel skb it points to skb_mac_header() and for 4838 * tunnel skb it points to outer mac header. 4839 * Keeps track of level of encapsulation of network headers. 4840 */ 4841 struct skb_gso_cb { 4842 union { 4843 int mac_offset; 4844 int data_offset; 4845 }; 4846 int encap_level; 4847 __wsum csum; 4848 __u16 csum_start; 4849 }; 4850 #define SKB_GSO_CB_OFFSET 32 4851 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET)) 4852 4853 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 4854 { 4855 return (skb_mac_header(inner_skb) - inner_skb->head) - 4856 SKB_GSO_CB(inner_skb)->mac_offset; 4857 } 4858 4859 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 4860 { 4861 int new_headroom, headroom; 4862 int ret; 4863 4864 headroom = skb_headroom(skb); 4865 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 4866 if (ret) 4867 return ret; 4868 4869 new_headroom = skb_headroom(skb); 4870 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 4871 return 0; 4872 } 4873 4874 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 4875 { 4876 /* Do not update partial checksums if remote checksum is enabled. */ 4877 if (skb->remcsum_offload) 4878 return; 4879 4880 SKB_GSO_CB(skb)->csum = res; 4881 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 4882 } 4883 4884 /* Compute the checksum for a gso segment. First compute the checksum value 4885 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 4886 * then add in skb->csum (checksum from csum_start to end of packet). 4887 * skb->csum and csum_start are then updated to reflect the checksum of the 4888 * resultant packet starting from the transport header-- the resultant checksum 4889 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 4890 * header. 4891 */ 4892 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 4893 { 4894 unsigned char *csum_start = skb_transport_header(skb); 4895 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 4896 __wsum partial = SKB_GSO_CB(skb)->csum; 4897 4898 SKB_GSO_CB(skb)->csum = res; 4899 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 4900 4901 return csum_fold(csum_partial(csum_start, plen, partial)); 4902 } 4903 4904 static inline bool skb_is_gso(const struct sk_buff *skb) 4905 { 4906 return skb_shinfo(skb)->gso_size; 4907 } 4908 4909 /* Note: Should be called only if skb_is_gso(skb) is true */ 4910 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4911 { 4912 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4913 } 4914 4915 /* Note: Should be called only if skb_is_gso(skb) is true */ 4916 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 4917 { 4918 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 4919 } 4920 4921 /* Note: Should be called only if skb_is_gso(skb) is true */ 4922 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 4923 { 4924 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 4925 } 4926 4927 static inline void skb_gso_reset(struct sk_buff *skb) 4928 { 4929 skb_shinfo(skb)->gso_size = 0; 4930 skb_shinfo(skb)->gso_segs = 0; 4931 skb_shinfo(skb)->gso_type = 0; 4932 } 4933 4934 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 4935 u16 increment) 4936 { 4937 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4938 return; 4939 shinfo->gso_size += increment; 4940 } 4941 4942 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 4943 u16 decrement) 4944 { 4945 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4946 return; 4947 shinfo->gso_size -= decrement; 4948 } 4949 4950 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 4951 4952 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 4953 { 4954 /* LRO sets gso_size but not gso_type, whereas if GSO is really 4955 * wanted then gso_type will be set. */ 4956 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4957 4958 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 4959 unlikely(shinfo->gso_type == 0)) { 4960 __skb_warn_lro_forwarding(skb); 4961 return true; 4962 } 4963 return false; 4964 } 4965 4966 static inline void skb_forward_csum(struct sk_buff *skb) 4967 { 4968 /* Unfortunately we don't support this one. Any brave souls? */ 4969 if (skb->ip_summed == CHECKSUM_COMPLETE) 4970 skb->ip_summed = CHECKSUM_NONE; 4971 } 4972 4973 /** 4974 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 4975 * @skb: skb to check 4976 * 4977 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 4978 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 4979 * use this helper, to document places where we make this assertion. 4980 */ 4981 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 4982 { 4983 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 4984 } 4985 4986 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 4987 4988 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 4989 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4990 unsigned int transport_len, 4991 __sum16(*skb_chkf)(struct sk_buff *skb)); 4992 4993 /** 4994 * skb_head_is_locked - Determine if the skb->head is locked down 4995 * @skb: skb to check 4996 * 4997 * The head on skbs build around a head frag can be removed if they are 4998 * not cloned. This function returns true if the skb head is locked down 4999 * due to either being allocated via kmalloc, or by being a clone with 5000 * multiple references to the head. 5001 */ 5002 static inline bool skb_head_is_locked(const struct sk_buff *skb) 5003 { 5004 return !skb->head_frag || skb_cloned(skb); 5005 } 5006 5007 /* Local Checksum Offload. 5008 * Compute outer checksum based on the assumption that the 5009 * inner checksum will be offloaded later. 5010 * See Documentation/networking/checksum-offloads.rst for 5011 * explanation of how this works. 5012 * Fill in outer checksum adjustment (e.g. with sum of outer 5013 * pseudo-header) before calling. 5014 * Also ensure that inner checksum is in linear data area. 5015 */ 5016 static inline __wsum lco_csum(struct sk_buff *skb) 5017 { 5018 unsigned char *csum_start = skb_checksum_start(skb); 5019 unsigned char *l4_hdr = skb_transport_header(skb); 5020 __wsum partial; 5021 5022 /* Start with complement of inner checksum adjustment */ 5023 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 5024 skb->csum_offset)); 5025 5026 /* Add in checksum of our headers (incl. outer checksum 5027 * adjustment filled in by caller) and return result. 5028 */ 5029 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5030 } 5031 5032 static inline bool skb_is_redirected(const struct sk_buff *skb) 5033 { 5034 return skb->redirected; 5035 } 5036 5037 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5038 { 5039 skb->redirected = 1; 5040 #ifdef CONFIG_NET_REDIRECT 5041 skb->from_ingress = from_ingress; 5042 if (skb->from_ingress) 5043 skb_clear_tstamp(skb); 5044 #endif 5045 } 5046 5047 static inline void skb_reset_redirect(struct sk_buff *skb) 5048 { 5049 skb->redirected = 0; 5050 } 5051 5052 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5053 { 5054 return skb->csum_not_inet; 5055 } 5056 5057 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5058 const u64 kcov_handle) 5059 { 5060 #ifdef CONFIG_KCOV 5061 skb->kcov_handle = kcov_handle; 5062 #endif 5063 } 5064 5065 static inline u64 skb_get_kcov_handle(struct sk_buff *skb) 5066 { 5067 #ifdef CONFIG_KCOV 5068 return skb->kcov_handle; 5069 #else 5070 return 0; 5071 #endif 5072 } 5073 5074 #ifdef CONFIG_PAGE_POOL 5075 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5076 { 5077 skb->pp_recycle = 1; 5078 } 5079 #endif 5080 5081 #endif /* __KERNEL__ */ 5082 #endif /* _LINUX_SKBUFF_H */ 5083