1 /* 2 * Definitions for the 'struct sk_buff' memory handlers. 3 * 4 * Authors: 5 * Alan Cox, <[email protected]> 6 * Florian La Roche, <[email protected]> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 */ 13 14 #ifndef _LINUX_SKBUFF_H 15 #define _LINUX_SKBUFF_H 16 17 #include <linux/kernel.h> 18 #include <linux/kmemcheck.h> 19 #include <linux/compiler.h> 20 #include <linux/time.h> 21 #include <linux/bug.h> 22 #include <linux/cache.h> 23 #include <linux/rbtree.h> 24 #include <linux/socket.h> 25 26 #include <linux/atomic.h> 27 #include <asm/types.h> 28 #include <linux/spinlock.h> 29 #include <linux/net.h> 30 #include <linux/textsearch.h> 31 #include <net/checksum.h> 32 #include <linux/rcupdate.h> 33 #include <linux/hrtimer.h> 34 #include <linux/dma-mapping.h> 35 #include <linux/netdev_features.h> 36 #include <linux/sched.h> 37 #include <net/flow_keys.h> 38 39 /* A. Checksumming of received packets by device. 40 * 41 * CHECKSUM_NONE: 42 * 43 * Device failed to checksum this packet e.g. due to lack of capabilities. 44 * The packet contains full (though not verified) checksum in packet but 45 * not in skb->csum. Thus, skb->csum is undefined in this case. 46 * 47 * CHECKSUM_UNNECESSARY: 48 * 49 * The hardware you're dealing with doesn't calculate the full checksum 50 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 51 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY 52 * if their checksums are okay. skb->csum is still undefined in this case 53 * though. It is a bad option, but, unfortunately, nowadays most vendors do 54 * this. Apparently with the secret goal to sell you new devices, when you 55 * will add new protocol to your host, f.e. IPv6 8) 56 * 57 * CHECKSUM_UNNECESSARY is applicable to following protocols: 58 * TCP: IPv6 and IPv4. 59 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 60 * zero UDP checksum for either IPv4 or IPv6, the networking stack 61 * may perform further validation in this case. 62 * GRE: only if the checksum is present in the header. 63 * SCTP: indicates the CRC in SCTP header has been validated. 64 * 65 * skb->csum_level indicates the number of consecutive checksums found in 66 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY. 67 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 68 * and a device is able to verify the checksums for UDP (possibly zero), 69 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to 70 * two. If the device were only able to verify the UDP checksum and not 71 * GRE, either because it doesn't support GRE checksum of because GRE 72 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 73 * not considered in this case). 74 * 75 * CHECKSUM_COMPLETE: 76 * 77 * This is the most generic way. The device supplied checksum of the _whole_ 78 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the 79 * hardware doesn't need to parse L3/L4 headers to implement this. 80 * 81 * Note: Even if device supports only some protocols, but is able to produce 82 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 83 * 84 * CHECKSUM_PARTIAL: 85 * 86 * A checksum is set up to be offloaded to a device as described in the 87 * output description for CHECKSUM_PARTIAL. This may occur on a packet 88 * received directly from another Linux OS, e.g., a virtualized Linux kernel 89 * on the same host, or it may be set in the input path in GRO or remote 90 * checksum offload. For the purposes of checksum verification, the checksum 91 * referred to by skb->csum_start + skb->csum_offset and any preceding 92 * checksums in the packet are considered verified. Any checksums in the 93 * packet that are after the checksum being offloaded are not considered to 94 * be verified. 95 * 96 * B. Checksumming on output. 97 * 98 * CHECKSUM_NONE: 99 * 100 * The skb was already checksummed by the protocol, or a checksum is not 101 * required. 102 * 103 * CHECKSUM_PARTIAL: 104 * 105 * The device is required to checksum the packet as seen by hard_start_xmit() 106 * from skb->csum_start up to the end, and to record/write the checksum at 107 * offset skb->csum_start + skb->csum_offset. 108 * 109 * The device must show its capabilities in dev->features, set up at device 110 * setup time, e.g. netdev_features.h: 111 * 112 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything. 113 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over 114 * IPv4. Sigh. Vendors like this way for an unknown reason. 115 * Though, see comment above about CHECKSUM_UNNECESSARY. 8) 116 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead. 117 * NETIF_F_... - Well, you get the picture. 118 * 119 * CHECKSUM_UNNECESSARY: 120 * 121 * Normally, the device will do per protocol specific checksumming. Protocol 122 * implementations that do not want the NIC to perform the checksum 123 * calculation should use this flag in their outgoing skbs. 124 * 125 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC 126 * offload. Correspondingly, the FCoE protocol driver 127 * stack should use CHECKSUM_UNNECESSARY. 128 * 129 * Any questions? No questions, good. --ANK 130 */ 131 132 /* Don't change this without changing skb_csum_unnecessary! */ 133 #define CHECKSUM_NONE 0 134 #define CHECKSUM_UNNECESSARY 1 135 #define CHECKSUM_COMPLETE 2 136 #define CHECKSUM_PARTIAL 3 137 138 /* Maximum value in skb->csum_level */ 139 #define SKB_MAX_CSUM_LEVEL 3 140 141 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 142 #define SKB_WITH_OVERHEAD(X) \ 143 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 144 #define SKB_MAX_ORDER(X, ORDER) \ 145 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 146 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 147 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 148 149 /* return minimum truesize of one skb containing X bytes of data */ 150 #define SKB_TRUESIZE(X) ((X) + \ 151 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 152 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 153 154 struct net_device; 155 struct scatterlist; 156 struct pipe_inode_info; 157 struct iov_iter; 158 struct napi_struct; 159 160 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 161 struct nf_conntrack { 162 atomic_t use; 163 }; 164 #endif 165 166 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 167 struct nf_bridge_info { 168 atomic_t use; 169 enum { 170 BRNF_PROTO_UNCHANGED, 171 BRNF_PROTO_8021Q, 172 BRNF_PROTO_PPPOE 173 } orig_proto; 174 bool pkt_otherhost; 175 unsigned int mask; 176 struct net_device *physindev; 177 struct net_device *physoutdev; 178 char neigh_header[8]; 179 }; 180 #endif 181 182 struct sk_buff_head { 183 /* These two members must be first. */ 184 struct sk_buff *next; 185 struct sk_buff *prev; 186 187 __u32 qlen; 188 spinlock_t lock; 189 }; 190 191 struct sk_buff; 192 193 /* To allow 64K frame to be packed as single skb without frag_list we 194 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 195 * buffers which do not start on a page boundary. 196 * 197 * Since GRO uses frags we allocate at least 16 regardless of page 198 * size. 199 */ 200 #if (65536/PAGE_SIZE + 1) < 16 201 #define MAX_SKB_FRAGS 16UL 202 #else 203 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 204 #endif 205 206 typedef struct skb_frag_struct skb_frag_t; 207 208 struct skb_frag_struct { 209 struct { 210 struct page *p; 211 } page; 212 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 213 __u32 page_offset; 214 __u32 size; 215 #else 216 __u16 page_offset; 217 __u16 size; 218 #endif 219 }; 220 221 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 222 { 223 return frag->size; 224 } 225 226 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 227 { 228 frag->size = size; 229 } 230 231 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 232 { 233 frag->size += delta; 234 } 235 236 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 237 { 238 frag->size -= delta; 239 } 240 241 #define HAVE_HW_TIME_STAMP 242 243 /** 244 * struct skb_shared_hwtstamps - hardware time stamps 245 * @hwtstamp: hardware time stamp transformed into duration 246 * since arbitrary point in time 247 * 248 * Software time stamps generated by ktime_get_real() are stored in 249 * skb->tstamp. 250 * 251 * hwtstamps can only be compared against other hwtstamps from 252 * the same device. 253 * 254 * This structure is attached to packets as part of the 255 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 256 */ 257 struct skb_shared_hwtstamps { 258 ktime_t hwtstamp; 259 }; 260 261 /* Definitions for tx_flags in struct skb_shared_info */ 262 enum { 263 /* generate hardware time stamp */ 264 SKBTX_HW_TSTAMP = 1 << 0, 265 266 /* generate software time stamp when queueing packet to NIC */ 267 SKBTX_SW_TSTAMP = 1 << 1, 268 269 /* device driver is going to provide hardware time stamp */ 270 SKBTX_IN_PROGRESS = 1 << 2, 271 272 /* device driver supports TX zero-copy buffers */ 273 SKBTX_DEV_ZEROCOPY = 1 << 3, 274 275 /* generate wifi status information (where possible) */ 276 SKBTX_WIFI_STATUS = 1 << 4, 277 278 /* This indicates at least one fragment might be overwritten 279 * (as in vmsplice(), sendfile() ...) 280 * If we need to compute a TX checksum, we'll need to copy 281 * all frags to avoid possible bad checksum 282 */ 283 SKBTX_SHARED_FRAG = 1 << 5, 284 285 /* generate software time stamp when entering packet scheduling */ 286 SKBTX_SCHED_TSTAMP = 1 << 6, 287 288 /* generate software timestamp on peer data acknowledgment */ 289 SKBTX_ACK_TSTAMP = 1 << 7, 290 }; 291 292 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 293 SKBTX_SCHED_TSTAMP | \ 294 SKBTX_ACK_TSTAMP) 295 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 296 297 /* 298 * The callback notifies userspace to release buffers when skb DMA is done in 299 * lower device, the skb last reference should be 0 when calling this. 300 * The zerocopy_success argument is true if zero copy transmit occurred, 301 * false on data copy or out of memory error caused by data copy attempt. 302 * The ctx field is used to track device context. 303 * The desc field is used to track userspace buffer index. 304 */ 305 struct ubuf_info { 306 void (*callback)(struct ubuf_info *, bool zerocopy_success); 307 void *ctx; 308 unsigned long desc; 309 }; 310 311 /* This data is invariant across clones and lives at 312 * the end of the header data, ie. at skb->end. 313 */ 314 struct skb_shared_info { 315 unsigned char nr_frags; 316 __u8 tx_flags; 317 unsigned short gso_size; 318 /* Warning: this field is not always filled in (UFO)! */ 319 unsigned short gso_segs; 320 unsigned short gso_type; 321 struct sk_buff *frag_list; 322 struct skb_shared_hwtstamps hwtstamps; 323 u32 tskey; 324 __be32 ip6_frag_id; 325 326 /* 327 * Warning : all fields before dataref are cleared in __alloc_skb() 328 */ 329 atomic_t dataref; 330 331 /* Intermediate layers must ensure that destructor_arg 332 * remains valid until skb destructor */ 333 void * destructor_arg; 334 335 /* must be last field, see pskb_expand_head() */ 336 skb_frag_t frags[MAX_SKB_FRAGS]; 337 }; 338 339 /* We divide dataref into two halves. The higher 16 bits hold references 340 * to the payload part of skb->data. The lower 16 bits hold references to 341 * the entire skb->data. A clone of a headerless skb holds the length of 342 * the header in skb->hdr_len. 343 * 344 * All users must obey the rule that the skb->data reference count must be 345 * greater than or equal to the payload reference count. 346 * 347 * Holding a reference to the payload part means that the user does not 348 * care about modifications to the header part of skb->data. 349 */ 350 #define SKB_DATAREF_SHIFT 16 351 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 352 353 354 enum { 355 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 356 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 357 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 358 }; 359 360 enum { 361 SKB_GSO_TCPV4 = 1 << 0, 362 SKB_GSO_UDP = 1 << 1, 363 364 /* This indicates the skb is from an untrusted source. */ 365 SKB_GSO_DODGY = 1 << 2, 366 367 /* This indicates the tcp segment has CWR set. */ 368 SKB_GSO_TCP_ECN = 1 << 3, 369 370 SKB_GSO_TCPV6 = 1 << 4, 371 372 SKB_GSO_FCOE = 1 << 5, 373 374 SKB_GSO_GRE = 1 << 6, 375 376 SKB_GSO_GRE_CSUM = 1 << 7, 377 378 SKB_GSO_IPIP = 1 << 8, 379 380 SKB_GSO_SIT = 1 << 9, 381 382 SKB_GSO_UDP_TUNNEL = 1 << 10, 383 384 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 385 386 SKB_GSO_TUNNEL_REMCSUM = 1 << 12, 387 }; 388 389 #if BITS_PER_LONG > 32 390 #define NET_SKBUFF_DATA_USES_OFFSET 1 391 #endif 392 393 #ifdef NET_SKBUFF_DATA_USES_OFFSET 394 typedef unsigned int sk_buff_data_t; 395 #else 396 typedef unsigned char *sk_buff_data_t; 397 #endif 398 399 /** 400 * struct skb_mstamp - multi resolution time stamps 401 * @stamp_us: timestamp in us resolution 402 * @stamp_jiffies: timestamp in jiffies 403 */ 404 struct skb_mstamp { 405 union { 406 u64 v64; 407 struct { 408 u32 stamp_us; 409 u32 stamp_jiffies; 410 }; 411 }; 412 }; 413 414 /** 415 * skb_mstamp_get - get current timestamp 416 * @cl: place to store timestamps 417 */ 418 static inline void skb_mstamp_get(struct skb_mstamp *cl) 419 { 420 u64 val = local_clock(); 421 422 do_div(val, NSEC_PER_USEC); 423 cl->stamp_us = (u32)val; 424 cl->stamp_jiffies = (u32)jiffies; 425 } 426 427 /** 428 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp 429 * @t1: pointer to newest sample 430 * @t0: pointer to oldest sample 431 */ 432 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1, 433 const struct skb_mstamp *t0) 434 { 435 s32 delta_us = t1->stamp_us - t0->stamp_us; 436 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies; 437 438 /* If delta_us is negative, this might be because interval is too big, 439 * or local_clock() drift is too big : fallback using jiffies. 440 */ 441 if (delta_us <= 0 || 442 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ))) 443 444 delta_us = jiffies_to_usecs(delta_jiffies); 445 446 return delta_us; 447 } 448 449 450 /** 451 * struct sk_buff - socket buffer 452 * @next: Next buffer in list 453 * @prev: Previous buffer in list 454 * @tstamp: Time we arrived/left 455 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 456 * @sk: Socket we are owned by 457 * @dev: Device we arrived on/are leaving by 458 * @cb: Control buffer. Free for use by every layer. Put private vars here 459 * @_skb_refdst: destination entry (with norefcount bit) 460 * @sp: the security path, used for xfrm 461 * @len: Length of actual data 462 * @data_len: Data length 463 * @mac_len: Length of link layer header 464 * @hdr_len: writable header length of cloned skb 465 * @csum: Checksum (must include start/offset pair) 466 * @csum_start: Offset from skb->head where checksumming should start 467 * @csum_offset: Offset from csum_start where checksum should be stored 468 * @priority: Packet queueing priority 469 * @ignore_df: allow local fragmentation 470 * @cloned: Head may be cloned (check refcnt to be sure) 471 * @ip_summed: Driver fed us an IP checksum 472 * @nohdr: Payload reference only, must not modify header 473 * @nfctinfo: Relationship of this skb to the connection 474 * @pkt_type: Packet class 475 * @fclone: skbuff clone status 476 * @ipvs_property: skbuff is owned by ipvs 477 * @peeked: this packet has been seen already, so stats have been 478 * done for it, don't do them again 479 * @nf_trace: netfilter packet trace flag 480 * @protocol: Packet protocol from driver 481 * @destructor: Destruct function 482 * @nfct: Associated connection, if any 483 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 484 * @skb_iif: ifindex of device we arrived on 485 * @tc_index: Traffic control index 486 * @tc_verd: traffic control verdict 487 * @hash: the packet hash 488 * @queue_mapping: Queue mapping for multiqueue devices 489 * @xmit_more: More SKBs are pending for this queue 490 * @ndisc_nodetype: router type (from link layer) 491 * @ooo_okay: allow the mapping of a socket to a queue to be changed 492 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 493 * ports. 494 * @sw_hash: indicates hash was computed in software stack 495 * @wifi_acked_valid: wifi_acked was set 496 * @wifi_acked: whether frame was acked on wifi or not 497 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 498 * @napi_id: id of the NAPI struct this skb came from 499 * @secmark: security marking 500 * @mark: Generic packet mark 501 * @vlan_proto: vlan encapsulation protocol 502 * @vlan_tci: vlan tag control information 503 * @inner_protocol: Protocol (encapsulation) 504 * @inner_transport_header: Inner transport layer header (encapsulation) 505 * @inner_network_header: Network layer header (encapsulation) 506 * @inner_mac_header: Link layer header (encapsulation) 507 * @transport_header: Transport layer header 508 * @network_header: Network layer header 509 * @mac_header: Link layer header 510 * @tail: Tail pointer 511 * @end: End pointer 512 * @head: Head of buffer 513 * @data: Data head pointer 514 * @truesize: Buffer size 515 * @users: User count - see {datagram,tcp}.c 516 */ 517 518 struct sk_buff { 519 union { 520 struct { 521 /* These two members must be first. */ 522 struct sk_buff *next; 523 struct sk_buff *prev; 524 525 union { 526 ktime_t tstamp; 527 struct skb_mstamp skb_mstamp; 528 }; 529 }; 530 struct rb_node rbnode; /* used in netem & tcp stack */ 531 }; 532 struct sock *sk; 533 struct net_device *dev; 534 535 /* 536 * This is the control buffer. It is free to use for every 537 * layer. Please put your private variables there. If you 538 * want to keep them across layers you have to do a skb_clone() 539 * first. This is owned by whoever has the skb queued ATM. 540 */ 541 char cb[48] __aligned(8); 542 543 unsigned long _skb_refdst; 544 void (*destructor)(struct sk_buff *skb); 545 #ifdef CONFIG_XFRM 546 struct sec_path *sp; 547 #endif 548 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 549 struct nf_conntrack *nfct; 550 #endif 551 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 552 struct nf_bridge_info *nf_bridge; 553 #endif 554 unsigned int len, 555 data_len; 556 __u16 mac_len, 557 hdr_len; 558 559 /* Following fields are _not_ copied in __copy_skb_header() 560 * Note that queue_mapping is here mostly to fill a hole. 561 */ 562 kmemcheck_bitfield_begin(flags1); 563 __u16 queue_mapping; 564 __u8 cloned:1, 565 nohdr:1, 566 fclone:2, 567 peeked:1, 568 head_frag:1, 569 xmit_more:1; 570 /* one bit hole */ 571 kmemcheck_bitfield_end(flags1); 572 573 /* fields enclosed in headers_start/headers_end are copied 574 * using a single memcpy() in __copy_skb_header() 575 */ 576 /* private: */ 577 __u32 headers_start[0]; 578 /* public: */ 579 580 /* if you move pkt_type around you also must adapt those constants */ 581 #ifdef __BIG_ENDIAN_BITFIELD 582 #define PKT_TYPE_MAX (7 << 5) 583 #else 584 #define PKT_TYPE_MAX 7 585 #endif 586 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 587 588 __u8 __pkt_type_offset[0]; 589 __u8 pkt_type:3; 590 __u8 pfmemalloc:1; 591 __u8 ignore_df:1; 592 __u8 nfctinfo:3; 593 594 __u8 nf_trace:1; 595 __u8 ip_summed:2; 596 __u8 ooo_okay:1; 597 __u8 l4_hash:1; 598 __u8 sw_hash:1; 599 __u8 wifi_acked_valid:1; 600 __u8 wifi_acked:1; 601 602 __u8 no_fcs:1; 603 /* Indicates the inner headers are valid in the skbuff. */ 604 __u8 encapsulation:1; 605 __u8 encap_hdr_csum:1; 606 __u8 csum_valid:1; 607 __u8 csum_complete_sw:1; 608 __u8 csum_level:2; 609 __u8 csum_bad:1; 610 611 #ifdef CONFIG_IPV6_NDISC_NODETYPE 612 __u8 ndisc_nodetype:2; 613 #endif 614 __u8 ipvs_property:1; 615 __u8 inner_protocol_type:1; 616 __u8 remcsum_offload:1; 617 /* 3 or 5 bit hole */ 618 619 #ifdef CONFIG_NET_SCHED 620 __u16 tc_index; /* traffic control index */ 621 #ifdef CONFIG_NET_CLS_ACT 622 __u16 tc_verd; /* traffic control verdict */ 623 #endif 624 #endif 625 626 union { 627 __wsum csum; 628 struct { 629 __u16 csum_start; 630 __u16 csum_offset; 631 }; 632 }; 633 __u32 priority; 634 int skb_iif; 635 __u32 hash; 636 __be16 vlan_proto; 637 __u16 vlan_tci; 638 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 639 union { 640 unsigned int napi_id; 641 unsigned int sender_cpu; 642 }; 643 #endif 644 #ifdef CONFIG_NETWORK_SECMARK 645 __u32 secmark; 646 #endif 647 union { 648 __u32 mark; 649 __u32 reserved_tailroom; 650 }; 651 652 union { 653 __be16 inner_protocol; 654 __u8 inner_ipproto; 655 }; 656 657 __u16 inner_transport_header; 658 __u16 inner_network_header; 659 __u16 inner_mac_header; 660 661 __be16 protocol; 662 __u16 transport_header; 663 __u16 network_header; 664 __u16 mac_header; 665 666 /* private: */ 667 __u32 headers_end[0]; 668 /* public: */ 669 670 /* These elements must be at the end, see alloc_skb() for details. */ 671 sk_buff_data_t tail; 672 sk_buff_data_t end; 673 unsigned char *head, 674 *data; 675 unsigned int truesize; 676 atomic_t users; 677 }; 678 679 #ifdef __KERNEL__ 680 /* 681 * Handling routines are only of interest to the kernel 682 */ 683 #include <linux/slab.h> 684 685 686 #define SKB_ALLOC_FCLONE 0x01 687 #define SKB_ALLOC_RX 0x02 688 #define SKB_ALLOC_NAPI 0x04 689 690 /* Returns true if the skb was allocated from PFMEMALLOC reserves */ 691 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 692 { 693 return unlikely(skb->pfmemalloc); 694 } 695 696 /* 697 * skb might have a dst pointer attached, refcounted or not. 698 * _skb_refdst low order bit is set if refcount was _not_ taken 699 */ 700 #define SKB_DST_NOREF 1UL 701 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 702 703 /** 704 * skb_dst - returns skb dst_entry 705 * @skb: buffer 706 * 707 * Returns skb dst_entry, regardless of reference taken or not. 708 */ 709 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 710 { 711 /* If refdst was not refcounted, check we still are in a 712 * rcu_read_lock section 713 */ 714 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 715 !rcu_read_lock_held() && 716 !rcu_read_lock_bh_held()); 717 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 718 } 719 720 /** 721 * skb_dst_set - sets skb dst 722 * @skb: buffer 723 * @dst: dst entry 724 * 725 * Sets skb dst, assuming a reference was taken on dst and should 726 * be released by skb_dst_drop() 727 */ 728 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 729 { 730 skb->_skb_refdst = (unsigned long)dst; 731 } 732 733 /** 734 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 735 * @skb: buffer 736 * @dst: dst entry 737 * 738 * Sets skb dst, assuming a reference was not taken on dst. 739 * If dst entry is cached, we do not take reference and dst_release 740 * will be avoided by refdst_drop. If dst entry is not cached, we take 741 * reference, so that last dst_release can destroy the dst immediately. 742 */ 743 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 744 { 745 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 746 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 747 } 748 749 /** 750 * skb_dst_is_noref - Test if skb dst isn't refcounted 751 * @skb: buffer 752 */ 753 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 754 { 755 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 756 } 757 758 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 759 { 760 return (struct rtable *)skb_dst(skb); 761 } 762 763 void kfree_skb(struct sk_buff *skb); 764 void kfree_skb_list(struct sk_buff *segs); 765 void skb_tx_error(struct sk_buff *skb); 766 void consume_skb(struct sk_buff *skb); 767 void __kfree_skb(struct sk_buff *skb); 768 extern struct kmem_cache *skbuff_head_cache; 769 770 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 771 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 772 bool *fragstolen, int *delta_truesize); 773 774 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 775 int node); 776 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 777 struct sk_buff *build_skb(void *data, unsigned int frag_size); 778 static inline struct sk_buff *alloc_skb(unsigned int size, 779 gfp_t priority) 780 { 781 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 782 } 783 784 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 785 unsigned long data_len, 786 int max_page_order, 787 int *errcode, 788 gfp_t gfp_mask); 789 790 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 791 struct sk_buff_fclones { 792 struct sk_buff skb1; 793 794 struct sk_buff skb2; 795 796 atomic_t fclone_ref; 797 }; 798 799 /** 800 * skb_fclone_busy - check if fclone is busy 801 * @skb: buffer 802 * 803 * Returns true is skb is a fast clone, and its clone is not freed. 804 * Some drivers call skb_orphan() in their ndo_start_xmit(), 805 * so we also check that this didnt happen. 806 */ 807 static inline bool skb_fclone_busy(const struct sock *sk, 808 const struct sk_buff *skb) 809 { 810 const struct sk_buff_fclones *fclones; 811 812 fclones = container_of(skb, struct sk_buff_fclones, skb1); 813 814 return skb->fclone == SKB_FCLONE_ORIG && 815 atomic_read(&fclones->fclone_ref) > 1 && 816 fclones->skb2.sk == sk; 817 } 818 819 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 820 gfp_t priority) 821 { 822 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 823 } 824 825 struct sk_buff *__alloc_skb_head(gfp_t priority, int node); 826 static inline struct sk_buff *alloc_skb_head(gfp_t priority) 827 { 828 return __alloc_skb_head(priority, -1); 829 } 830 831 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 832 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 833 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 834 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 835 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 836 gfp_t gfp_mask, bool fclone); 837 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 838 gfp_t gfp_mask) 839 { 840 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 841 } 842 843 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 844 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 845 unsigned int headroom); 846 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 847 int newtailroom, gfp_t priority); 848 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 849 int offset, int len); 850 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, 851 int len); 852 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 853 int skb_pad(struct sk_buff *skb, int pad); 854 #define dev_kfree_skb(a) consume_skb(a) 855 856 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 857 int getfrag(void *from, char *to, int offset, 858 int len, int odd, struct sk_buff *skb), 859 void *from, int length); 860 861 struct skb_seq_state { 862 __u32 lower_offset; 863 __u32 upper_offset; 864 __u32 frag_idx; 865 __u32 stepped_offset; 866 struct sk_buff *root_skb; 867 struct sk_buff *cur_skb; 868 __u8 *frag_data; 869 }; 870 871 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 872 unsigned int to, struct skb_seq_state *st); 873 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 874 struct skb_seq_state *st); 875 void skb_abort_seq_read(struct skb_seq_state *st); 876 877 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 878 unsigned int to, struct ts_config *config); 879 880 /* 881 * Packet hash types specify the type of hash in skb_set_hash. 882 * 883 * Hash types refer to the protocol layer addresses which are used to 884 * construct a packet's hash. The hashes are used to differentiate or identify 885 * flows of the protocol layer for the hash type. Hash types are either 886 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 887 * 888 * Properties of hashes: 889 * 890 * 1) Two packets in different flows have different hash values 891 * 2) Two packets in the same flow should have the same hash value 892 * 893 * A hash at a higher layer is considered to be more specific. A driver should 894 * set the most specific hash possible. 895 * 896 * A driver cannot indicate a more specific hash than the layer at which a hash 897 * was computed. For instance an L3 hash cannot be set as an L4 hash. 898 * 899 * A driver may indicate a hash level which is less specific than the 900 * actual layer the hash was computed on. For instance, a hash computed 901 * at L4 may be considered an L3 hash. This should only be done if the 902 * driver can't unambiguously determine that the HW computed the hash at 903 * the higher layer. Note that the "should" in the second property above 904 * permits this. 905 */ 906 enum pkt_hash_types { 907 PKT_HASH_TYPE_NONE, /* Undefined type */ 908 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 909 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 910 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 911 }; 912 913 static inline void 914 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 915 { 916 skb->l4_hash = (type == PKT_HASH_TYPE_L4); 917 skb->sw_hash = 0; 918 skb->hash = hash; 919 } 920 921 void __skb_get_hash(struct sk_buff *skb); 922 static inline __u32 skb_get_hash(struct sk_buff *skb) 923 { 924 if (!skb->l4_hash && !skb->sw_hash) 925 __skb_get_hash(skb); 926 927 return skb->hash; 928 } 929 930 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 931 { 932 return skb->hash; 933 } 934 935 static inline void skb_clear_hash(struct sk_buff *skb) 936 { 937 skb->hash = 0; 938 skb->sw_hash = 0; 939 skb->l4_hash = 0; 940 } 941 942 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 943 { 944 if (!skb->l4_hash) 945 skb_clear_hash(skb); 946 } 947 948 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 949 { 950 to->hash = from->hash; 951 to->sw_hash = from->sw_hash; 952 to->l4_hash = from->l4_hash; 953 }; 954 955 static inline void skb_sender_cpu_clear(struct sk_buff *skb) 956 { 957 #ifdef CONFIG_XPS 958 skb->sender_cpu = 0; 959 #endif 960 } 961 962 #ifdef NET_SKBUFF_DATA_USES_OFFSET 963 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 964 { 965 return skb->head + skb->end; 966 } 967 968 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 969 { 970 return skb->end; 971 } 972 #else 973 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 974 { 975 return skb->end; 976 } 977 978 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 979 { 980 return skb->end - skb->head; 981 } 982 #endif 983 984 /* Internal */ 985 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 986 987 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 988 { 989 return &skb_shinfo(skb)->hwtstamps; 990 } 991 992 /** 993 * skb_queue_empty - check if a queue is empty 994 * @list: queue head 995 * 996 * Returns true if the queue is empty, false otherwise. 997 */ 998 static inline int skb_queue_empty(const struct sk_buff_head *list) 999 { 1000 return list->next == (const struct sk_buff *) list; 1001 } 1002 1003 /** 1004 * skb_queue_is_last - check if skb is the last entry in the queue 1005 * @list: queue head 1006 * @skb: buffer 1007 * 1008 * Returns true if @skb is the last buffer on the list. 1009 */ 1010 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1011 const struct sk_buff *skb) 1012 { 1013 return skb->next == (const struct sk_buff *) list; 1014 } 1015 1016 /** 1017 * skb_queue_is_first - check if skb is the first entry in the queue 1018 * @list: queue head 1019 * @skb: buffer 1020 * 1021 * Returns true if @skb is the first buffer on the list. 1022 */ 1023 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1024 const struct sk_buff *skb) 1025 { 1026 return skb->prev == (const struct sk_buff *) list; 1027 } 1028 1029 /** 1030 * skb_queue_next - return the next packet in the queue 1031 * @list: queue head 1032 * @skb: current buffer 1033 * 1034 * Return the next packet in @list after @skb. It is only valid to 1035 * call this if skb_queue_is_last() evaluates to false. 1036 */ 1037 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1038 const struct sk_buff *skb) 1039 { 1040 /* This BUG_ON may seem severe, but if we just return then we 1041 * are going to dereference garbage. 1042 */ 1043 BUG_ON(skb_queue_is_last(list, skb)); 1044 return skb->next; 1045 } 1046 1047 /** 1048 * skb_queue_prev - return the prev packet in the queue 1049 * @list: queue head 1050 * @skb: current buffer 1051 * 1052 * Return the prev packet in @list before @skb. It is only valid to 1053 * call this if skb_queue_is_first() evaluates to false. 1054 */ 1055 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1056 const struct sk_buff *skb) 1057 { 1058 /* This BUG_ON may seem severe, but if we just return then we 1059 * are going to dereference garbage. 1060 */ 1061 BUG_ON(skb_queue_is_first(list, skb)); 1062 return skb->prev; 1063 } 1064 1065 /** 1066 * skb_get - reference buffer 1067 * @skb: buffer to reference 1068 * 1069 * Makes another reference to a socket buffer and returns a pointer 1070 * to the buffer. 1071 */ 1072 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1073 { 1074 atomic_inc(&skb->users); 1075 return skb; 1076 } 1077 1078 /* 1079 * If users == 1, we are the only owner and are can avoid redundant 1080 * atomic change. 1081 */ 1082 1083 /** 1084 * skb_cloned - is the buffer a clone 1085 * @skb: buffer to check 1086 * 1087 * Returns true if the buffer was generated with skb_clone() and is 1088 * one of multiple shared copies of the buffer. Cloned buffers are 1089 * shared data so must not be written to under normal circumstances. 1090 */ 1091 static inline int skb_cloned(const struct sk_buff *skb) 1092 { 1093 return skb->cloned && 1094 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1095 } 1096 1097 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1098 { 1099 might_sleep_if(pri & __GFP_WAIT); 1100 1101 if (skb_cloned(skb)) 1102 return pskb_expand_head(skb, 0, 0, pri); 1103 1104 return 0; 1105 } 1106 1107 /** 1108 * skb_header_cloned - is the header a clone 1109 * @skb: buffer to check 1110 * 1111 * Returns true if modifying the header part of the buffer requires 1112 * the data to be copied. 1113 */ 1114 static inline int skb_header_cloned(const struct sk_buff *skb) 1115 { 1116 int dataref; 1117 1118 if (!skb->cloned) 1119 return 0; 1120 1121 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1122 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1123 return dataref != 1; 1124 } 1125 1126 /** 1127 * skb_header_release - release reference to header 1128 * @skb: buffer to operate on 1129 * 1130 * Drop a reference to the header part of the buffer. This is done 1131 * by acquiring a payload reference. You must not read from the header 1132 * part of skb->data after this. 1133 * Note : Check if you can use __skb_header_release() instead. 1134 */ 1135 static inline void skb_header_release(struct sk_buff *skb) 1136 { 1137 BUG_ON(skb->nohdr); 1138 skb->nohdr = 1; 1139 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1140 } 1141 1142 /** 1143 * __skb_header_release - release reference to header 1144 * @skb: buffer to operate on 1145 * 1146 * Variant of skb_header_release() assuming skb is private to caller. 1147 * We can avoid one atomic operation. 1148 */ 1149 static inline void __skb_header_release(struct sk_buff *skb) 1150 { 1151 skb->nohdr = 1; 1152 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1153 } 1154 1155 1156 /** 1157 * skb_shared - is the buffer shared 1158 * @skb: buffer to check 1159 * 1160 * Returns true if more than one person has a reference to this 1161 * buffer. 1162 */ 1163 static inline int skb_shared(const struct sk_buff *skb) 1164 { 1165 return atomic_read(&skb->users) != 1; 1166 } 1167 1168 /** 1169 * skb_share_check - check if buffer is shared and if so clone it 1170 * @skb: buffer to check 1171 * @pri: priority for memory allocation 1172 * 1173 * If the buffer is shared the buffer is cloned and the old copy 1174 * drops a reference. A new clone with a single reference is returned. 1175 * If the buffer is not shared the original buffer is returned. When 1176 * being called from interrupt status or with spinlocks held pri must 1177 * be GFP_ATOMIC. 1178 * 1179 * NULL is returned on a memory allocation failure. 1180 */ 1181 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1182 { 1183 might_sleep_if(pri & __GFP_WAIT); 1184 if (skb_shared(skb)) { 1185 struct sk_buff *nskb = skb_clone(skb, pri); 1186 1187 if (likely(nskb)) 1188 consume_skb(skb); 1189 else 1190 kfree_skb(skb); 1191 skb = nskb; 1192 } 1193 return skb; 1194 } 1195 1196 /* 1197 * Copy shared buffers into a new sk_buff. We effectively do COW on 1198 * packets to handle cases where we have a local reader and forward 1199 * and a couple of other messy ones. The normal one is tcpdumping 1200 * a packet thats being forwarded. 1201 */ 1202 1203 /** 1204 * skb_unshare - make a copy of a shared buffer 1205 * @skb: buffer to check 1206 * @pri: priority for memory allocation 1207 * 1208 * If the socket buffer is a clone then this function creates a new 1209 * copy of the data, drops a reference count on the old copy and returns 1210 * the new copy with the reference count at 1. If the buffer is not a clone 1211 * the original buffer is returned. When called with a spinlock held or 1212 * from interrupt state @pri must be %GFP_ATOMIC 1213 * 1214 * %NULL is returned on a memory allocation failure. 1215 */ 1216 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1217 gfp_t pri) 1218 { 1219 might_sleep_if(pri & __GFP_WAIT); 1220 if (skb_cloned(skb)) { 1221 struct sk_buff *nskb = skb_copy(skb, pri); 1222 1223 /* Free our shared copy */ 1224 if (likely(nskb)) 1225 consume_skb(skb); 1226 else 1227 kfree_skb(skb); 1228 skb = nskb; 1229 } 1230 return skb; 1231 } 1232 1233 /** 1234 * skb_peek - peek at the head of an &sk_buff_head 1235 * @list_: list to peek at 1236 * 1237 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1238 * be careful with this one. A peek leaves the buffer on the 1239 * list and someone else may run off with it. You must hold 1240 * the appropriate locks or have a private queue to do this. 1241 * 1242 * Returns %NULL for an empty list or a pointer to the head element. 1243 * The reference count is not incremented and the reference is therefore 1244 * volatile. Use with caution. 1245 */ 1246 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1247 { 1248 struct sk_buff *skb = list_->next; 1249 1250 if (skb == (struct sk_buff *)list_) 1251 skb = NULL; 1252 return skb; 1253 } 1254 1255 /** 1256 * skb_peek_next - peek skb following the given one from a queue 1257 * @skb: skb to start from 1258 * @list_: list to peek at 1259 * 1260 * Returns %NULL when the end of the list is met or a pointer to the 1261 * next element. The reference count is not incremented and the 1262 * reference is therefore volatile. Use with caution. 1263 */ 1264 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1265 const struct sk_buff_head *list_) 1266 { 1267 struct sk_buff *next = skb->next; 1268 1269 if (next == (struct sk_buff *)list_) 1270 next = NULL; 1271 return next; 1272 } 1273 1274 /** 1275 * skb_peek_tail - peek at the tail of an &sk_buff_head 1276 * @list_: list to peek at 1277 * 1278 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1279 * be careful with this one. A peek leaves the buffer on the 1280 * list and someone else may run off with it. You must hold 1281 * the appropriate locks or have a private queue to do this. 1282 * 1283 * Returns %NULL for an empty list or a pointer to the tail element. 1284 * The reference count is not incremented and the reference is therefore 1285 * volatile. Use with caution. 1286 */ 1287 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1288 { 1289 struct sk_buff *skb = list_->prev; 1290 1291 if (skb == (struct sk_buff *)list_) 1292 skb = NULL; 1293 return skb; 1294 1295 } 1296 1297 /** 1298 * skb_queue_len - get queue length 1299 * @list_: list to measure 1300 * 1301 * Return the length of an &sk_buff queue. 1302 */ 1303 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1304 { 1305 return list_->qlen; 1306 } 1307 1308 /** 1309 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1310 * @list: queue to initialize 1311 * 1312 * This initializes only the list and queue length aspects of 1313 * an sk_buff_head object. This allows to initialize the list 1314 * aspects of an sk_buff_head without reinitializing things like 1315 * the spinlock. It can also be used for on-stack sk_buff_head 1316 * objects where the spinlock is known to not be used. 1317 */ 1318 static inline void __skb_queue_head_init(struct sk_buff_head *list) 1319 { 1320 list->prev = list->next = (struct sk_buff *)list; 1321 list->qlen = 0; 1322 } 1323 1324 /* 1325 * This function creates a split out lock class for each invocation; 1326 * this is needed for now since a whole lot of users of the skb-queue 1327 * infrastructure in drivers have different locking usage (in hardirq) 1328 * than the networking core (in softirq only). In the long run either the 1329 * network layer or drivers should need annotation to consolidate the 1330 * main types of usage into 3 classes. 1331 */ 1332 static inline void skb_queue_head_init(struct sk_buff_head *list) 1333 { 1334 spin_lock_init(&list->lock); 1335 __skb_queue_head_init(list); 1336 } 1337 1338 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1339 struct lock_class_key *class) 1340 { 1341 skb_queue_head_init(list); 1342 lockdep_set_class(&list->lock, class); 1343 } 1344 1345 /* 1346 * Insert an sk_buff on a list. 1347 * 1348 * The "__skb_xxxx()" functions are the non-atomic ones that 1349 * can only be called with interrupts disabled. 1350 */ 1351 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1352 struct sk_buff_head *list); 1353 static inline void __skb_insert(struct sk_buff *newsk, 1354 struct sk_buff *prev, struct sk_buff *next, 1355 struct sk_buff_head *list) 1356 { 1357 newsk->next = next; 1358 newsk->prev = prev; 1359 next->prev = prev->next = newsk; 1360 list->qlen++; 1361 } 1362 1363 static inline void __skb_queue_splice(const struct sk_buff_head *list, 1364 struct sk_buff *prev, 1365 struct sk_buff *next) 1366 { 1367 struct sk_buff *first = list->next; 1368 struct sk_buff *last = list->prev; 1369 1370 first->prev = prev; 1371 prev->next = first; 1372 1373 last->next = next; 1374 next->prev = last; 1375 } 1376 1377 /** 1378 * skb_queue_splice - join two skb lists, this is designed for stacks 1379 * @list: the new list to add 1380 * @head: the place to add it in the first list 1381 */ 1382 static inline void skb_queue_splice(const struct sk_buff_head *list, 1383 struct sk_buff_head *head) 1384 { 1385 if (!skb_queue_empty(list)) { 1386 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1387 head->qlen += list->qlen; 1388 } 1389 } 1390 1391 /** 1392 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1393 * @list: the new list to add 1394 * @head: the place to add it in the first list 1395 * 1396 * The list at @list is reinitialised 1397 */ 1398 static inline void skb_queue_splice_init(struct sk_buff_head *list, 1399 struct sk_buff_head *head) 1400 { 1401 if (!skb_queue_empty(list)) { 1402 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1403 head->qlen += list->qlen; 1404 __skb_queue_head_init(list); 1405 } 1406 } 1407 1408 /** 1409 * skb_queue_splice_tail - join two skb lists, each list being a queue 1410 * @list: the new list to add 1411 * @head: the place to add it in the first list 1412 */ 1413 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1414 struct sk_buff_head *head) 1415 { 1416 if (!skb_queue_empty(list)) { 1417 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1418 head->qlen += list->qlen; 1419 } 1420 } 1421 1422 /** 1423 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1424 * @list: the new list to add 1425 * @head: the place to add it in the first list 1426 * 1427 * Each of the lists is a queue. 1428 * The list at @list is reinitialised 1429 */ 1430 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1431 struct sk_buff_head *head) 1432 { 1433 if (!skb_queue_empty(list)) { 1434 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1435 head->qlen += list->qlen; 1436 __skb_queue_head_init(list); 1437 } 1438 } 1439 1440 /** 1441 * __skb_queue_after - queue a buffer at the list head 1442 * @list: list to use 1443 * @prev: place after this buffer 1444 * @newsk: buffer to queue 1445 * 1446 * Queue a buffer int the middle of a list. This function takes no locks 1447 * and you must therefore hold required locks before calling it. 1448 * 1449 * A buffer cannot be placed on two lists at the same time. 1450 */ 1451 static inline void __skb_queue_after(struct sk_buff_head *list, 1452 struct sk_buff *prev, 1453 struct sk_buff *newsk) 1454 { 1455 __skb_insert(newsk, prev, prev->next, list); 1456 } 1457 1458 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1459 struct sk_buff_head *list); 1460 1461 static inline void __skb_queue_before(struct sk_buff_head *list, 1462 struct sk_buff *next, 1463 struct sk_buff *newsk) 1464 { 1465 __skb_insert(newsk, next->prev, next, list); 1466 } 1467 1468 /** 1469 * __skb_queue_head - queue a buffer at the list head 1470 * @list: list to use 1471 * @newsk: buffer to queue 1472 * 1473 * Queue a buffer at the start of a list. This function takes no locks 1474 * and you must therefore hold required locks before calling it. 1475 * 1476 * A buffer cannot be placed on two lists at the same time. 1477 */ 1478 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1479 static inline void __skb_queue_head(struct sk_buff_head *list, 1480 struct sk_buff *newsk) 1481 { 1482 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1483 } 1484 1485 /** 1486 * __skb_queue_tail - queue a buffer at the list tail 1487 * @list: list to use 1488 * @newsk: buffer to queue 1489 * 1490 * Queue a buffer at the end of a list. This function takes no locks 1491 * and you must therefore hold required locks before calling it. 1492 * 1493 * A buffer cannot be placed on two lists at the same time. 1494 */ 1495 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1496 static inline void __skb_queue_tail(struct sk_buff_head *list, 1497 struct sk_buff *newsk) 1498 { 1499 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1500 } 1501 1502 /* 1503 * remove sk_buff from list. _Must_ be called atomically, and with 1504 * the list known.. 1505 */ 1506 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1507 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1508 { 1509 struct sk_buff *next, *prev; 1510 1511 list->qlen--; 1512 next = skb->next; 1513 prev = skb->prev; 1514 skb->next = skb->prev = NULL; 1515 next->prev = prev; 1516 prev->next = next; 1517 } 1518 1519 /** 1520 * __skb_dequeue - remove from the head of the queue 1521 * @list: list to dequeue from 1522 * 1523 * Remove the head of the list. This function does not take any locks 1524 * so must be used with appropriate locks held only. The head item is 1525 * returned or %NULL if the list is empty. 1526 */ 1527 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1528 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1529 { 1530 struct sk_buff *skb = skb_peek(list); 1531 if (skb) 1532 __skb_unlink(skb, list); 1533 return skb; 1534 } 1535 1536 /** 1537 * __skb_dequeue_tail - remove from the tail of the queue 1538 * @list: list to dequeue from 1539 * 1540 * Remove the tail of the list. This function does not take any locks 1541 * so must be used with appropriate locks held only. The tail item is 1542 * returned or %NULL if the list is empty. 1543 */ 1544 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1545 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1546 { 1547 struct sk_buff *skb = skb_peek_tail(list); 1548 if (skb) 1549 __skb_unlink(skb, list); 1550 return skb; 1551 } 1552 1553 1554 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1555 { 1556 return skb->data_len; 1557 } 1558 1559 static inline unsigned int skb_headlen(const struct sk_buff *skb) 1560 { 1561 return skb->len - skb->data_len; 1562 } 1563 1564 static inline int skb_pagelen(const struct sk_buff *skb) 1565 { 1566 int i, len = 0; 1567 1568 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--) 1569 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1570 return len + skb_headlen(skb); 1571 } 1572 1573 /** 1574 * __skb_fill_page_desc - initialise a paged fragment in an skb 1575 * @skb: buffer containing fragment to be initialised 1576 * @i: paged fragment index to initialise 1577 * @page: the page to use for this fragment 1578 * @off: the offset to the data with @page 1579 * @size: the length of the data 1580 * 1581 * Initialises the @i'th fragment of @skb to point to &size bytes at 1582 * offset @off within @page. 1583 * 1584 * Does not take any additional reference on the fragment. 1585 */ 1586 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1587 struct page *page, int off, int size) 1588 { 1589 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1590 1591 /* 1592 * Propagate page->pfmemalloc to the skb if we can. The problem is 1593 * that not all callers have unique ownership of the page. If 1594 * pfmemalloc is set, we check the mapping as a mapping implies 1595 * page->index is set (index and pfmemalloc share space). 1596 * If it's a valid mapping, we cannot use page->pfmemalloc but we 1597 * do not lose pfmemalloc information as the pages would not be 1598 * allocated using __GFP_MEMALLOC. 1599 */ 1600 frag->page.p = page; 1601 frag->page_offset = off; 1602 skb_frag_size_set(frag, size); 1603 1604 page = compound_head(page); 1605 if (page->pfmemalloc && !page->mapping) 1606 skb->pfmemalloc = true; 1607 } 1608 1609 /** 1610 * skb_fill_page_desc - initialise a paged fragment in an skb 1611 * @skb: buffer containing fragment to be initialised 1612 * @i: paged fragment index to initialise 1613 * @page: the page to use for this fragment 1614 * @off: the offset to the data with @page 1615 * @size: the length of the data 1616 * 1617 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1618 * @skb to point to @size bytes at offset @off within @page. In 1619 * addition updates @skb such that @i is the last fragment. 1620 * 1621 * Does not take any additional reference on the fragment. 1622 */ 1623 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1624 struct page *page, int off, int size) 1625 { 1626 __skb_fill_page_desc(skb, i, page, off, size); 1627 skb_shinfo(skb)->nr_frags = i + 1; 1628 } 1629 1630 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1631 int size, unsigned int truesize); 1632 1633 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1634 unsigned int truesize); 1635 1636 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1637 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1638 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1639 1640 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1641 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1642 { 1643 return skb->head + skb->tail; 1644 } 1645 1646 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1647 { 1648 skb->tail = skb->data - skb->head; 1649 } 1650 1651 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1652 { 1653 skb_reset_tail_pointer(skb); 1654 skb->tail += offset; 1655 } 1656 1657 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 1658 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1659 { 1660 return skb->tail; 1661 } 1662 1663 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1664 { 1665 skb->tail = skb->data; 1666 } 1667 1668 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1669 { 1670 skb->tail = skb->data + offset; 1671 } 1672 1673 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1674 1675 /* 1676 * Add data to an sk_buff 1677 */ 1678 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1679 unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1680 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1681 { 1682 unsigned char *tmp = skb_tail_pointer(skb); 1683 SKB_LINEAR_ASSERT(skb); 1684 skb->tail += len; 1685 skb->len += len; 1686 return tmp; 1687 } 1688 1689 unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1690 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1691 { 1692 skb->data -= len; 1693 skb->len += len; 1694 return skb->data; 1695 } 1696 1697 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1698 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1699 { 1700 skb->len -= len; 1701 BUG_ON(skb->len < skb->data_len); 1702 return skb->data += len; 1703 } 1704 1705 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1706 { 1707 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1708 } 1709 1710 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1711 1712 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1713 { 1714 if (len > skb_headlen(skb) && 1715 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1716 return NULL; 1717 skb->len -= len; 1718 return skb->data += len; 1719 } 1720 1721 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1722 { 1723 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1724 } 1725 1726 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1727 { 1728 if (likely(len <= skb_headlen(skb))) 1729 return 1; 1730 if (unlikely(len > skb->len)) 1731 return 0; 1732 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1733 } 1734 1735 /** 1736 * skb_headroom - bytes at buffer head 1737 * @skb: buffer to check 1738 * 1739 * Return the number of bytes of free space at the head of an &sk_buff. 1740 */ 1741 static inline unsigned int skb_headroom(const struct sk_buff *skb) 1742 { 1743 return skb->data - skb->head; 1744 } 1745 1746 /** 1747 * skb_tailroom - bytes at buffer end 1748 * @skb: buffer to check 1749 * 1750 * Return the number of bytes of free space at the tail of an sk_buff 1751 */ 1752 static inline int skb_tailroom(const struct sk_buff *skb) 1753 { 1754 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 1755 } 1756 1757 /** 1758 * skb_availroom - bytes at buffer end 1759 * @skb: buffer to check 1760 * 1761 * Return the number of bytes of free space at the tail of an sk_buff 1762 * allocated by sk_stream_alloc() 1763 */ 1764 static inline int skb_availroom(const struct sk_buff *skb) 1765 { 1766 if (skb_is_nonlinear(skb)) 1767 return 0; 1768 1769 return skb->end - skb->tail - skb->reserved_tailroom; 1770 } 1771 1772 /** 1773 * skb_reserve - adjust headroom 1774 * @skb: buffer to alter 1775 * @len: bytes to move 1776 * 1777 * Increase the headroom of an empty &sk_buff by reducing the tail 1778 * room. This is only allowed for an empty buffer. 1779 */ 1780 static inline void skb_reserve(struct sk_buff *skb, int len) 1781 { 1782 skb->data += len; 1783 skb->tail += len; 1784 } 1785 1786 #define ENCAP_TYPE_ETHER 0 1787 #define ENCAP_TYPE_IPPROTO 1 1788 1789 static inline void skb_set_inner_protocol(struct sk_buff *skb, 1790 __be16 protocol) 1791 { 1792 skb->inner_protocol = protocol; 1793 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 1794 } 1795 1796 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 1797 __u8 ipproto) 1798 { 1799 skb->inner_ipproto = ipproto; 1800 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 1801 } 1802 1803 static inline void skb_reset_inner_headers(struct sk_buff *skb) 1804 { 1805 skb->inner_mac_header = skb->mac_header; 1806 skb->inner_network_header = skb->network_header; 1807 skb->inner_transport_header = skb->transport_header; 1808 } 1809 1810 static inline void skb_reset_mac_len(struct sk_buff *skb) 1811 { 1812 skb->mac_len = skb->network_header - skb->mac_header; 1813 } 1814 1815 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 1816 *skb) 1817 { 1818 return skb->head + skb->inner_transport_header; 1819 } 1820 1821 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 1822 { 1823 skb->inner_transport_header = skb->data - skb->head; 1824 } 1825 1826 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 1827 const int offset) 1828 { 1829 skb_reset_inner_transport_header(skb); 1830 skb->inner_transport_header += offset; 1831 } 1832 1833 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 1834 { 1835 return skb->head + skb->inner_network_header; 1836 } 1837 1838 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 1839 { 1840 skb->inner_network_header = skb->data - skb->head; 1841 } 1842 1843 static inline void skb_set_inner_network_header(struct sk_buff *skb, 1844 const int offset) 1845 { 1846 skb_reset_inner_network_header(skb); 1847 skb->inner_network_header += offset; 1848 } 1849 1850 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 1851 { 1852 return skb->head + skb->inner_mac_header; 1853 } 1854 1855 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 1856 { 1857 skb->inner_mac_header = skb->data - skb->head; 1858 } 1859 1860 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 1861 const int offset) 1862 { 1863 skb_reset_inner_mac_header(skb); 1864 skb->inner_mac_header += offset; 1865 } 1866 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 1867 { 1868 return skb->transport_header != (typeof(skb->transport_header))~0U; 1869 } 1870 1871 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 1872 { 1873 return skb->head + skb->transport_header; 1874 } 1875 1876 static inline void skb_reset_transport_header(struct sk_buff *skb) 1877 { 1878 skb->transport_header = skb->data - skb->head; 1879 } 1880 1881 static inline void skb_set_transport_header(struct sk_buff *skb, 1882 const int offset) 1883 { 1884 skb_reset_transport_header(skb); 1885 skb->transport_header += offset; 1886 } 1887 1888 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 1889 { 1890 return skb->head + skb->network_header; 1891 } 1892 1893 static inline void skb_reset_network_header(struct sk_buff *skb) 1894 { 1895 skb->network_header = skb->data - skb->head; 1896 } 1897 1898 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 1899 { 1900 skb_reset_network_header(skb); 1901 skb->network_header += offset; 1902 } 1903 1904 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 1905 { 1906 return skb->head + skb->mac_header; 1907 } 1908 1909 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 1910 { 1911 return skb->mac_header != (typeof(skb->mac_header))~0U; 1912 } 1913 1914 static inline void skb_reset_mac_header(struct sk_buff *skb) 1915 { 1916 skb->mac_header = skb->data - skb->head; 1917 } 1918 1919 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 1920 { 1921 skb_reset_mac_header(skb); 1922 skb->mac_header += offset; 1923 } 1924 1925 static inline void skb_pop_mac_header(struct sk_buff *skb) 1926 { 1927 skb->mac_header = skb->network_header; 1928 } 1929 1930 static inline void skb_probe_transport_header(struct sk_buff *skb, 1931 const int offset_hint) 1932 { 1933 struct flow_keys keys; 1934 1935 if (skb_transport_header_was_set(skb)) 1936 return; 1937 else if (skb_flow_dissect(skb, &keys)) 1938 skb_set_transport_header(skb, keys.thoff); 1939 else 1940 skb_set_transport_header(skb, offset_hint); 1941 } 1942 1943 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 1944 { 1945 if (skb_mac_header_was_set(skb)) { 1946 const unsigned char *old_mac = skb_mac_header(skb); 1947 1948 skb_set_mac_header(skb, -skb->mac_len); 1949 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 1950 } 1951 } 1952 1953 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 1954 { 1955 return skb->csum_start - skb_headroom(skb); 1956 } 1957 1958 static inline int skb_transport_offset(const struct sk_buff *skb) 1959 { 1960 return skb_transport_header(skb) - skb->data; 1961 } 1962 1963 static inline u32 skb_network_header_len(const struct sk_buff *skb) 1964 { 1965 return skb->transport_header - skb->network_header; 1966 } 1967 1968 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 1969 { 1970 return skb->inner_transport_header - skb->inner_network_header; 1971 } 1972 1973 static inline int skb_network_offset(const struct sk_buff *skb) 1974 { 1975 return skb_network_header(skb) - skb->data; 1976 } 1977 1978 static inline int skb_inner_network_offset(const struct sk_buff *skb) 1979 { 1980 return skb_inner_network_header(skb) - skb->data; 1981 } 1982 1983 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 1984 { 1985 return pskb_may_pull(skb, skb_network_offset(skb) + len); 1986 } 1987 1988 /* 1989 * CPUs often take a performance hit when accessing unaligned memory 1990 * locations. The actual performance hit varies, it can be small if the 1991 * hardware handles it or large if we have to take an exception and fix it 1992 * in software. 1993 * 1994 * Since an ethernet header is 14 bytes network drivers often end up with 1995 * the IP header at an unaligned offset. The IP header can be aligned by 1996 * shifting the start of the packet by 2 bytes. Drivers should do this 1997 * with: 1998 * 1999 * skb_reserve(skb, NET_IP_ALIGN); 2000 * 2001 * The downside to this alignment of the IP header is that the DMA is now 2002 * unaligned. On some architectures the cost of an unaligned DMA is high 2003 * and this cost outweighs the gains made by aligning the IP header. 2004 * 2005 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2006 * to be overridden. 2007 */ 2008 #ifndef NET_IP_ALIGN 2009 #define NET_IP_ALIGN 2 2010 #endif 2011 2012 /* 2013 * The networking layer reserves some headroom in skb data (via 2014 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2015 * the header has to grow. In the default case, if the header has to grow 2016 * 32 bytes or less we avoid the reallocation. 2017 * 2018 * Unfortunately this headroom changes the DMA alignment of the resulting 2019 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2020 * on some architectures. An architecture can override this value, 2021 * perhaps setting it to a cacheline in size (since that will maintain 2022 * cacheline alignment of the DMA). It must be a power of 2. 2023 * 2024 * Various parts of the networking layer expect at least 32 bytes of 2025 * headroom, you should not reduce this. 2026 * 2027 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2028 * to reduce average number of cache lines per packet. 2029 * get_rps_cpus() for example only access one 64 bytes aligned block : 2030 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2031 */ 2032 #ifndef NET_SKB_PAD 2033 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2034 #endif 2035 2036 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2037 2038 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2039 { 2040 if (unlikely(skb_is_nonlinear(skb))) { 2041 WARN_ON(1); 2042 return; 2043 } 2044 skb->len = len; 2045 skb_set_tail_pointer(skb, len); 2046 } 2047 2048 void skb_trim(struct sk_buff *skb, unsigned int len); 2049 2050 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2051 { 2052 if (skb->data_len) 2053 return ___pskb_trim(skb, len); 2054 __skb_trim(skb, len); 2055 return 0; 2056 } 2057 2058 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2059 { 2060 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2061 } 2062 2063 /** 2064 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2065 * @skb: buffer to alter 2066 * @len: new length 2067 * 2068 * This is identical to pskb_trim except that the caller knows that 2069 * the skb is not cloned so we should never get an error due to out- 2070 * of-memory. 2071 */ 2072 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2073 { 2074 int err = pskb_trim(skb, len); 2075 BUG_ON(err); 2076 } 2077 2078 /** 2079 * skb_orphan - orphan a buffer 2080 * @skb: buffer to orphan 2081 * 2082 * If a buffer currently has an owner then we call the owner's 2083 * destructor function and make the @skb unowned. The buffer continues 2084 * to exist but is no longer charged to its former owner. 2085 */ 2086 static inline void skb_orphan(struct sk_buff *skb) 2087 { 2088 if (skb->destructor) { 2089 skb->destructor(skb); 2090 skb->destructor = NULL; 2091 skb->sk = NULL; 2092 } else { 2093 BUG_ON(skb->sk); 2094 } 2095 } 2096 2097 /** 2098 * skb_orphan_frags - orphan the frags contained in a buffer 2099 * @skb: buffer to orphan frags from 2100 * @gfp_mask: allocation mask for replacement pages 2101 * 2102 * For each frag in the SKB which needs a destructor (i.e. has an 2103 * owner) create a copy of that frag and release the original 2104 * page by calling the destructor. 2105 */ 2106 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2107 { 2108 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 2109 return 0; 2110 return skb_copy_ubufs(skb, gfp_mask); 2111 } 2112 2113 /** 2114 * __skb_queue_purge - empty a list 2115 * @list: list to empty 2116 * 2117 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2118 * the list and one reference dropped. This function does not take the 2119 * list lock and the caller must hold the relevant locks to use it. 2120 */ 2121 void skb_queue_purge(struct sk_buff_head *list); 2122 static inline void __skb_queue_purge(struct sk_buff_head *list) 2123 { 2124 struct sk_buff *skb; 2125 while ((skb = __skb_dequeue(list)) != NULL) 2126 kfree_skb(skb); 2127 } 2128 2129 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768) 2130 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER) 2131 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE 2132 2133 void *netdev_alloc_frag(unsigned int fragsz); 2134 2135 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2136 gfp_t gfp_mask); 2137 2138 /** 2139 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2140 * @dev: network device to receive on 2141 * @length: length to allocate 2142 * 2143 * Allocate a new &sk_buff and assign it a usage count of one. The 2144 * buffer has unspecified headroom built in. Users should allocate 2145 * the headroom they think they need without accounting for the 2146 * built in space. The built in space is used for optimisations. 2147 * 2148 * %NULL is returned if there is no free memory. Although this function 2149 * allocates memory it can be called from an interrupt. 2150 */ 2151 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2152 unsigned int length) 2153 { 2154 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2155 } 2156 2157 /* legacy helper around __netdev_alloc_skb() */ 2158 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2159 gfp_t gfp_mask) 2160 { 2161 return __netdev_alloc_skb(NULL, length, gfp_mask); 2162 } 2163 2164 /* legacy helper around netdev_alloc_skb() */ 2165 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2166 { 2167 return netdev_alloc_skb(NULL, length); 2168 } 2169 2170 2171 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2172 unsigned int length, gfp_t gfp) 2173 { 2174 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2175 2176 if (NET_IP_ALIGN && skb) 2177 skb_reserve(skb, NET_IP_ALIGN); 2178 return skb; 2179 } 2180 2181 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2182 unsigned int length) 2183 { 2184 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2185 } 2186 2187 void *napi_alloc_frag(unsigned int fragsz); 2188 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2189 unsigned int length, gfp_t gfp_mask); 2190 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2191 unsigned int length) 2192 { 2193 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2194 } 2195 2196 /** 2197 * __dev_alloc_pages - allocate page for network Rx 2198 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2199 * @order: size of the allocation 2200 * 2201 * Allocate a new page. 2202 * 2203 * %NULL is returned if there is no free memory. 2204 */ 2205 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2206 unsigned int order) 2207 { 2208 /* This piece of code contains several assumptions. 2209 * 1. This is for device Rx, therefor a cold page is preferred. 2210 * 2. The expectation is the user wants a compound page. 2211 * 3. If requesting a order 0 page it will not be compound 2212 * due to the check to see if order has a value in prep_new_page 2213 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2214 * code in gfp_to_alloc_flags that should be enforcing this. 2215 */ 2216 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2217 2218 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2219 } 2220 2221 static inline struct page *dev_alloc_pages(unsigned int order) 2222 { 2223 return __dev_alloc_pages(GFP_ATOMIC, order); 2224 } 2225 2226 /** 2227 * __dev_alloc_page - allocate a page for network Rx 2228 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2229 * 2230 * Allocate a new page. 2231 * 2232 * %NULL is returned if there is no free memory. 2233 */ 2234 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2235 { 2236 return __dev_alloc_pages(gfp_mask, 0); 2237 } 2238 2239 static inline struct page *dev_alloc_page(void) 2240 { 2241 return __dev_alloc_page(GFP_ATOMIC); 2242 } 2243 2244 /** 2245 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2246 * @page: The page that was allocated from skb_alloc_page 2247 * @skb: The skb that may need pfmemalloc set 2248 */ 2249 static inline void skb_propagate_pfmemalloc(struct page *page, 2250 struct sk_buff *skb) 2251 { 2252 if (page && page->pfmemalloc) 2253 skb->pfmemalloc = true; 2254 } 2255 2256 /** 2257 * skb_frag_page - retrieve the page referred to by a paged fragment 2258 * @frag: the paged fragment 2259 * 2260 * Returns the &struct page associated with @frag. 2261 */ 2262 static inline struct page *skb_frag_page(const skb_frag_t *frag) 2263 { 2264 return frag->page.p; 2265 } 2266 2267 /** 2268 * __skb_frag_ref - take an addition reference on a paged fragment. 2269 * @frag: the paged fragment 2270 * 2271 * Takes an additional reference on the paged fragment @frag. 2272 */ 2273 static inline void __skb_frag_ref(skb_frag_t *frag) 2274 { 2275 get_page(skb_frag_page(frag)); 2276 } 2277 2278 /** 2279 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2280 * @skb: the buffer 2281 * @f: the fragment offset. 2282 * 2283 * Takes an additional reference on the @f'th paged fragment of @skb. 2284 */ 2285 static inline void skb_frag_ref(struct sk_buff *skb, int f) 2286 { 2287 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2288 } 2289 2290 /** 2291 * __skb_frag_unref - release a reference on a paged fragment. 2292 * @frag: the paged fragment 2293 * 2294 * Releases a reference on the paged fragment @frag. 2295 */ 2296 static inline void __skb_frag_unref(skb_frag_t *frag) 2297 { 2298 put_page(skb_frag_page(frag)); 2299 } 2300 2301 /** 2302 * skb_frag_unref - release a reference on a paged fragment of an skb. 2303 * @skb: the buffer 2304 * @f: the fragment offset 2305 * 2306 * Releases a reference on the @f'th paged fragment of @skb. 2307 */ 2308 static inline void skb_frag_unref(struct sk_buff *skb, int f) 2309 { 2310 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2311 } 2312 2313 /** 2314 * skb_frag_address - gets the address of the data contained in a paged fragment 2315 * @frag: the paged fragment buffer 2316 * 2317 * Returns the address of the data within @frag. The page must already 2318 * be mapped. 2319 */ 2320 static inline void *skb_frag_address(const skb_frag_t *frag) 2321 { 2322 return page_address(skb_frag_page(frag)) + frag->page_offset; 2323 } 2324 2325 /** 2326 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2327 * @frag: the paged fragment buffer 2328 * 2329 * Returns the address of the data within @frag. Checks that the page 2330 * is mapped and returns %NULL otherwise. 2331 */ 2332 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2333 { 2334 void *ptr = page_address(skb_frag_page(frag)); 2335 if (unlikely(!ptr)) 2336 return NULL; 2337 2338 return ptr + frag->page_offset; 2339 } 2340 2341 /** 2342 * __skb_frag_set_page - sets the page contained in a paged fragment 2343 * @frag: the paged fragment 2344 * @page: the page to set 2345 * 2346 * Sets the fragment @frag to contain @page. 2347 */ 2348 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2349 { 2350 frag->page.p = page; 2351 } 2352 2353 /** 2354 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2355 * @skb: the buffer 2356 * @f: the fragment offset 2357 * @page: the page to set 2358 * 2359 * Sets the @f'th fragment of @skb to contain @page. 2360 */ 2361 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2362 struct page *page) 2363 { 2364 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2365 } 2366 2367 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2368 2369 /** 2370 * skb_frag_dma_map - maps a paged fragment via the DMA API 2371 * @dev: the device to map the fragment to 2372 * @frag: the paged fragment to map 2373 * @offset: the offset within the fragment (starting at the 2374 * fragment's own offset) 2375 * @size: the number of bytes to map 2376 * @dir: the direction of the mapping (%PCI_DMA_*) 2377 * 2378 * Maps the page associated with @frag to @device. 2379 */ 2380 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2381 const skb_frag_t *frag, 2382 size_t offset, size_t size, 2383 enum dma_data_direction dir) 2384 { 2385 return dma_map_page(dev, skb_frag_page(frag), 2386 frag->page_offset + offset, size, dir); 2387 } 2388 2389 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2390 gfp_t gfp_mask) 2391 { 2392 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2393 } 2394 2395 2396 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2397 gfp_t gfp_mask) 2398 { 2399 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2400 } 2401 2402 2403 /** 2404 * skb_clone_writable - is the header of a clone writable 2405 * @skb: buffer to check 2406 * @len: length up to which to write 2407 * 2408 * Returns true if modifying the header part of the cloned buffer 2409 * does not requires the data to be copied. 2410 */ 2411 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2412 { 2413 return !skb_header_cloned(skb) && 2414 skb_headroom(skb) + len <= skb->hdr_len; 2415 } 2416 2417 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2418 int cloned) 2419 { 2420 int delta = 0; 2421 2422 if (headroom > skb_headroom(skb)) 2423 delta = headroom - skb_headroom(skb); 2424 2425 if (delta || cloned) 2426 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2427 GFP_ATOMIC); 2428 return 0; 2429 } 2430 2431 /** 2432 * skb_cow - copy header of skb when it is required 2433 * @skb: buffer to cow 2434 * @headroom: needed headroom 2435 * 2436 * If the skb passed lacks sufficient headroom or its data part 2437 * is shared, data is reallocated. If reallocation fails, an error 2438 * is returned and original skb is not changed. 2439 * 2440 * The result is skb with writable area skb->head...skb->tail 2441 * and at least @headroom of space at head. 2442 */ 2443 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2444 { 2445 return __skb_cow(skb, headroom, skb_cloned(skb)); 2446 } 2447 2448 /** 2449 * skb_cow_head - skb_cow but only making the head writable 2450 * @skb: buffer to cow 2451 * @headroom: needed headroom 2452 * 2453 * This function is identical to skb_cow except that we replace the 2454 * skb_cloned check by skb_header_cloned. It should be used when 2455 * you only need to push on some header and do not need to modify 2456 * the data. 2457 */ 2458 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2459 { 2460 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2461 } 2462 2463 /** 2464 * skb_padto - pad an skbuff up to a minimal size 2465 * @skb: buffer to pad 2466 * @len: minimal length 2467 * 2468 * Pads up a buffer to ensure the trailing bytes exist and are 2469 * blanked. If the buffer already contains sufficient data it 2470 * is untouched. Otherwise it is extended. Returns zero on 2471 * success. The skb is freed on error. 2472 */ 2473 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2474 { 2475 unsigned int size = skb->len; 2476 if (likely(size >= len)) 2477 return 0; 2478 return skb_pad(skb, len - size); 2479 } 2480 2481 /** 2482 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2483 * @skb: buffer to pad 2484 * @len: minimal length 2485 * 2486 * Pads up a buffer to ensure the trailing bytes exist and are 2487 * blanked. If the buffer already contains sufficient data it 2488 * is untouched. Otherwise it is extended. Returns zero on 2489 * success. The skb is freed on error. 2490 */ 2491 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2492 { 2493 unsigned int size = skb->len; 2494 2495 if (unlikely(size < len)) { 2496 len -= size; 2497 if (skb_pad(skb, len)) 2498 return -ENOMEM; 2499 __skb_put(skb, len); 2500 } 2501 return 0; 2502 } 2503 2504 static inline int skb_add_data(struct sk_buff *skb, 2505 struct iov_iter *from, int copy) 2506 { 2507 const int off = skb->len; 2508 2509 if (skb->ip_summed == CHECKSUM_NONE) { 2510 __wsum csum = 0; 2511 if (csum_and_copy_from_iter(skb_put(skb, copy), copy, 2512 &csum, from) == copy) { 2513 skb->csum = csum_block_add(skb->csum, csum, off); 2514 return 0; 2515 } 2516 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy) 2517 return 0; 2518 2519 __skb_trim(skb, off); 2520 return -EFAULT; 2521 } 2522 2523 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2524 const struct page *page, int off) 2525 { 2526 if (i) { 2527 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2528 2529 return page == skb_frag_page(frag) && 2530 off == frag->page_offset + skb_frag_size(frag); 2531 } 2532 return false; 2533 } 2534 2535 static inline int __skb_linearize(struct sk_buff *skb) 2536 { 2537 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2538 } 2539 2540 /** 2541 * skb_linearize - convert paged skb to linear one 2542 * @skb: buffer to linarize 2543 * 2544 * If there is no free memory -ENOMEM is returned, otherwise zero 2545 * is returned and the old skb data released. 2546 */ 2547 static inline int skb_linearize(struct sk_buff *skb) 2548 { 2549 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2550 } 2551 2552 /** 2553 * skb_has_shared_frag - can any frag be overwritten 2554 * @skb: buffer to test 2555 * 2556 * Return true if the skb has at least one frag that might be modified 2557 * by an external entity (as in vmsplice()/sendfile()) 2558 */ 2559 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 2560 { 2561 return skb_is_nonlinear(skb) && 2562 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2563 } 2564 2565 /** 2566 * skb_linearize_cow - make sure skb is linear and writable 2567 * @skb: buffer to process 2568 * 2569 * If there is no free memory -ENOMEM is returned, otherwise zero 2570 * is returned and the old skb data released. 2571 */ 2572 static inline int skb_linearize_cow(struct sk_buff *skb) 2573 { 2574 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2575 __skb_linearize(skb) : 0; 2576 } 2577 2578 /** 2579 * skb_postpull_rcsum - update checksum for received skb after pull 2580 * @skb: buffer to update 2581 * @start: start of data before pull 2582 * @len: length of data pulled 2583 * 2584 * After doing a pull on a received packet, you need to call this to 2585 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2586 * CHECKSUM_NONE so that it can be recomputed from scratch. 2587 */ 2588 2589 static inline void skb_postpull_rcsum(struct sk_buff *skb, 2590 const void *start, unsigned int len) 2591 { 2592 if (skb->ip_summed == CHECKSUM_COMPLETE) 2593 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0)); 2594 } 2595 2596 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2597 2598 /** 2599 * pskb_trim_rcsum - trim received skb and update checksum 2600 * @skb: buffer to trim 2601 * @len: new length 2602 * 2603 * This is exactly the same as pskb_trim except that it ensures the 2604 * checksum of received packets are still valid after the operation. 2605 */ 2606 2607 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2608 { 2609 if (likely(len >= skb->len)) 2610 return 0; 2611 if (skb->ip_summed == CHECKSUM_COMPLETE) 2612 skb->ip_summed = CHECKSUM_NONE; 2613 return __pskb_trim(skb, len); 2614 } 2615 2616 #define skb_queue_walk(queue, skb) \ 2617 for (skb = (queue)->next; \ 2618 skb != (struct sk_buff *)(queue); \ 2619 skb = skb->next) 2620 2621 #define skb_queue_walk_safe(queue, skb, tmp) \ 2622 for (skb = (queue)->next, tmp = skb->next; \ 2623 skb != (struct sk_buff *)(queue); \ 2624 skb = tmp, tmp = skb->next) 2625 2626 #define skb_queue_walk_from(queue, skb) \ 2627 for (; skb != (struct sk_buff *)(queue); \ 2628 skb = skb->next) 2629 2630 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 2631 for (tmp = skb->next; \ 2632 skb != (struct sk_buff *)(queue); \ 2633 skb = tmp, tmp = skb->next) 2634 2635 #define skb_queue_reverse_walk(queue, skb) \ 2636 for (skb = (queue)->prev; \ 2637 skb != (struct sk_buff *)(queue); \ 2638 skb = skb->prev) 2639 2640 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 2641 for (skb = (queue)->prev, tmp = skb->prev; \ 2642 skb != (struct sk_buff *)(queue); \ 2643 skb = tmp, tmp = skb->prev) 2644 2645 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 2646 for (tmp = skb->prev; \ 2647 skb != (struct sk_buff *)(queue); \ 2648 skb = tmp, tmp = skb->prev) 2649 2650 static inline bool skb_has_frag_list(const struct sk_buff *skb) 2651 { 2652 return skb_shinfo(skb)->frag_list != NULL; 2653 } 2654 2655 static inline void skb_frag_list_init(struct sk_buff *skb) 2656 { 2657 skb_shinfo(skb)->frag_list = NULL; 2658 } 2659 2660 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag) 2661 { 2662 frag->next = skb_shinfo(skb)->frag_list; 2663 skb_shinfo(skb)->frag_list = frag; 2664 } 2665 2666 #define skb_walk_frags(skb, iter) \ 2667 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 2668 2669 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 2670 int *peeked, int *off, int *err); 2671 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 2672 int *err); 2673 unsigned int datagram_poll(struct file *file, struct socket *sock, 2674 struct poll_table_struct *wait); 2675 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 2676 struct iov_iter *to, int size); 2677 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 2678 struct msghdr *msg, int size) 2679 { 2680 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 2681 } 2682 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 2683 struct msghdr *msg); 2684 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 2685 struct iov_iter *from, int len); 2686 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 2687 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 2688 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb); 2689 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 2690 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 2691 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 2692 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 2693 int len, __wsum csum); 2694 int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 2695 struct pipe_inode_info *pipe, unsigned int len, 2696 unsigned int flags); 2697 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 2698 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 2699 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 2700 int len, int hlen); 2701 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 2702 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 2703 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 2704 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 2705 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 2706 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 2707 int skb_ensure_writable(struct sk_buff *skb, int write_len); 2708 int skb_vlan_pop(struct sk_buff *skb); 2709 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 2710 2711 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 2712 { 2713 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2714 } 2715 2716 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 2717 { 2718 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2719 } 2720 2721 struct skb_checksum_ops { 2722 __wsum (*update)(const void *mem, int len, __wsum wsum); 2723 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 2724 }; 2725 2726 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 2727 __wsum csum, const struct skb_checksum_ops *ops); 2728 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 2729 __wsum csum); 2730 2731 static inline void *__skb_header_pointer(const struct sk_buff *skb, int offset, 2732 int len, void *data, int hlen, void *buffer) 2733 { 2734 if (hlen - offset >= len) 2735 return data + offset; 2736 2737 if (!skb || 2738 skb_copy_bits(skb, offset, buffer, len) < 0) 2739 return NULL; 2740 2741 return buffer; 2742 } 2743 2744 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset, 2745 int len, void *buffer) 2746 { 2747 return __skb_header_pointer(skb, offset, len, skb->data, 2748 skb_headlen(skb), buffer); 2749 } 2750 2751 /** 2752 * skb_needs_linearize - check if we need to linearize a given skb 2753 * depending on the given device features. 2754 * @skb: socket buffer to check 2755 * @features: net device features 2756 * 2757 * Returns true if either: 2758 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 2759 * 2. skb is fragmented and the device does not support SG. 2760 */ 2761 static inline bool skb_needs_linearize(struct sk_buff *skb, 2762 netdev_features_t features) 2763 { 2764 return skb_is_nonlinear(skb) && 2765 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 2766 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 2767 } 2768 2769 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 2770 void *to, 2771 const unsigned int len) 2772 { 2773 memcpy(to, skb->data, len); 2774 } 2775 2776 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 2777 const int offset, void *to, 2778 const unsigned int len) 2779 { 2780 memcpy(to, skb->data + offset, len); 2781 } 2782 2783 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 2784 const void *from, 2785 const unsigned int len) 2786 { 2787 memcpy(skb->data, from, len); 2788 } 2789 2790 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 2791 const int offset, 2792 const void *from, 2793 const unsigned int len) 2794 { 2795 memcpy(skb->data + offset, from, len); 2796 } 2797 2798 void skb_init(void); 2799 2800 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 2801 { 2802 return skb->tstamp; 2803 } 2804 2805 /** 2806 * skb_get_timestamp - get timestamp from a skb 2807 * @skb: skb to get stamp from 2808 * @stamp: pointer to struct timeval to store stamp in 2809 * 2810 * Timestamps are stored in the skb as offsets to a base timestamp. 2811 * This function converts the offset back to a struct timeval and stores 2812 * it in stamp. 2813 */ 2814 static inline void skb_get_timestamp(const struct sk_buff *skb, 2815 struct timeval *stamp) 2816 { 2817 *stamp = ktime_to_timeval(skb->tstamp); 2818 } 2819 2820 static inline void skb_get_timestampns(const struct sk_buff *skb, 2821 struct timespec *stamp) 2822 { 2823 *stamp = ktime_to_timespec(skb->tstamp); 2824 } 2825 2826 static inline void __net_timestamp(struct sk_buff *skb) 2827 { 2828 skb->tstamp = ktime_get_real(); 2829 } 2830 2831 static inline ktime_t net_timedelta(ktime_t t) 2832 { 2833 return ktime_sub(ktime_get_real(), t); 2834 } 2835 2836 static inline ktime_t net_invalid_timestamp(void) 2837 { 2838 return ktime_set(0, 0); 2839 } 2840 2841 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 2842 2843 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 2844 2845 void skb_clone_tx_timestamp(struct sk_buff *skb); 2846 bool skb_defer_rx_timestamp(struct sk_buff *skb); 2847 2848 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 2849 2850 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 2851 { 2852 } 2853 2854 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 2855 { 2856 return false; 2857 } 2858 2859 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 2860 2861 /** 2862 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 2863 * 2864 * PHY drivers may accept clones of transmitted packets for 2865 * timestamping via their phy_driver.txtstamp method. These drivers 2866 * must call this function to return the skb back to the stack, with 2867 * or without a timestamp. 2868 * 2869 * @skb: clone of the the original outgoing packet 2870 * @hwtstamps: hardware time stamps, may be NULL if not available 2871 * 2872 */ 2873 void skb_complete_tx_timestamp(struct sk_buff *skb, 2874 struct skb_shared_hwtstamps *hwtstamps); 2875 2876 void __skb_tstamp_tx(struct sk_buff *orig_skb, 2877 struct skb_shared_hwtstamps *hwtstamps, 2878 struct sock *sk, int tstype); 2879 2880 /** 2881 * skb_tstamp_tx - queue clone of skb with send time stamps 2882 * @orig_skb: the original outgoing packet 2883 * @hwtstamps: hardware time stamps, may be NULL if not available 2884 * 2885 * If the skb has a socket associated, then this function clones the 2886 * skb (thus sharing the actual data and optional structures), stores 2887 * the optional hardware time stamping information (if non NULL) or 2888 * generates a software time stamp (otherwise), then queues the clone 2889 * to the error queue of the socket. Errors are silently ignored. 2890 */ 2891 void skb_tstamp_tx(struct sk_buff *orig_skb, 2892 struct skb_shared_hwtstamps *hwtstamps); 2893 2894 static inline void sw_tx_timestamp(struct sk_buff *skb) 2895 { 2896 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 2897 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 2898 skb_tstamp_tx(skb, NULL); 2899 } 2900 2901 /** 2902 * skb_tx_timestamp() - Driver hook for transmit timestamping 2903 * 2904 * Ethernet MAC Drivers should call this function in their hard_xmit() 2905 * function immediately before giving the sk_buff to the MAC hardware. 2906 * 2907 * Specifically, one should make absolutely sure that this function is 2908 * called before TX completion of this packet can trigger. Otherwise 2909 * the packet could potentially already be freed. 2910 * 2911 * @skb: A socket buffer. 2912 */ 2913 static inline void skb_tx_timestamp(struct sk_buff *skb) 2914 { 2915 skb_clone_tx_timestamp(skb); 2916 sw_tx_timestamp(skb); 2917 } 2918 2919 /** 2920 * skb_complete_wifi_ack - deliver skb with wifi status 2921 * 2922 * @skb: the original outgoing packet 2923 * @acked: ack status 2924 * 2925 */ 2926 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 2927 2928 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 2929 __sum16 __skb_checksum_complete(struct sk_buff *skb); 2930 2931 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 2932 { 2933 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 2934 skb->csum_valid || 2935 (skb->ip_summed == CHECKSUM_PARTIAL && 2936 skb_checksum_start_offset(skb) >= 0)); 2937 } 2938 2939 /** 2940 * skb_checksum_complete - Calculate checksum of an entire packet 2941 * @skb: packet to process 2942 * 2943 * This function calculates the checksum over the entire packet plus 2944 * the value of skb->csum. The latter can be used to supply the 2945 * checksum of a pseudo header as used by TCP/UDP. It returns the 2946 * checksum. 2947 * 2948 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 2949 * this function can be used to verify that checksum on received 2950 * packets. In that case the function should return zero if the 2951 * checksum is correct. In particular, this function will return zero 2952 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 2953 * hardware has already verified the correctness of the checksum. 2954 */ 2955 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 2956 { 2957 return skb_csum_unnecessary(skb) ? 2958 0 : __skb_checksum_complete(skb); 2959 } 2960 2961 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 2962 { 2963 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 2964 if (skb->csum_level == 0) 2965 skb->ip_summed = CHECKSUM_NONE; 2966 else 2967 skb->csum_level--; 2968 } 2969 } 2970 2971 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 2972 { 2973 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 2974 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 2975 skb->csum_level++; 2976 } else if (skb->ip_summed == CHECKSUM_NONE) { 2977 skb->ip_summed = CHECKSUM_UNNECESSARY; 2978 skb->csum_level = 0; 2979 } 2980 } 2981 2982 static inline void __skb_mark_checksum_bad(struct sk_buff *skb) 2983 { 2984 /* Mark current checksum as bad (typically called from GRO 2985 * path). In the case that ip_summed is CHECKSUM_NONE 2986 * this must be the first checksum encountered in the packet. 2987 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first 2988 * checksum after the last one validated. For UDP, a zero 2989 * checksum can not be marked as bad. 2990 */ 2991 2992 if (skb->ip_summed == CHECKSUM_NONE || 2993 skb->ip_summed == CHECKSUM_UNNECESSARY) 2994 skb->csum_bad = 1; 2995 } 2996 2997 /* Check if we need to perform checksum complete validation. 2998 * 2999 * Returns true if checksum complete is needed, false otherwise 3000 * (either checksum is unnecessary or zero checksum is allowed). 3001 */ 3002 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3003 bool zero_okay, 3004 __sum16 check) 3005 { 3006 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3007 skb->csum_valid = 1; 3008 __skb_decr_checksum_unnecessary(skb); 3009 return false; 3010 } 3011 3012 return true; 3013 } 3014 3015 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3016 * in checksum_init. 3017 */ 3018 #define CHECKSUM_BREAK 76 3019 3020 /* Unset checksum-complete 3021 * 3022 * Unset checksum complete can be done when packet is being modified 3023 * (uncompressed for instance) and checksum-complete value is 3024 * invalidated. 3025 */ 3026 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3027 { 3028 if (skb->ip_summed == CHECKSUM_COMPLETE) 3029 skb->ip_summed = CHECKSUM_NONE; 3030 } 3031 3032 /* Validate (init) checksum based on checksum complete. 3033 * 3034 * Return values: 3035 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3036 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3037 * checksum is stored in skb->csum for use in __skb_checksum_complete 3038 * non-zero: value of invalid checksum 3039 * 3040 */ 3041 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3042 bool complete, 3043 __wsum psum) 3044 { 3045 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3046 if (!csum_fold(csum_add(psum, skb->csum))) { 3047 skb->csum_valid = 1; 3048 return 0; 3049 } 3050 } else if (skb->csum_bad) { 3051 /* ip_summed == CHECKSUM_NONE in this case */ 3052 return 1; 3053 } 3054 3055 skb->csum = psum; 3056 3057 if (complete || skb->len <= CHECKSUM_BREAK) { 3058 __sum16 csum; 3059 3060 csum = __skb_checksum_complete(skb); 3061 skb->csum_valid = !csum; 3062 return csum; 3063 } 3064 3065 return 0; 3066 } 3067 3068 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3069 { 3070 return 0; 3071 } 3072 3073 /* Perform checksum validate (init). Note that this is a macro since we only 3074 * want to calculate the pseudo header which is an input function if necessary. 3075 * First we try to validate without any computation (checksum unnecessary) and 3076 * then calculate based on checksum complete calling the function to compute 3077 * pseudo header. 3078 * 3079 * Return values: 3080 * 0: checksum is validated or try to in skb_checksum_complete 3081 * non-zero: value of invalid checksum 3082 */ 3083 #define __skb_checksum_validate(skb, proto, complete, \ 3084 zero_okay, check, compute_pseudo) \ 3085 ({ \ 3086 __sum16 __ret = 0; \ 3087 skb->csum_valid = 0; \ 3088 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3089 __ret = __skb_checksum_validate_complete(skb, \ 3090 complete, compute_pseudo(skb, proto)); \ 3091 __ret; \ 3092 }) 3093 3094 #define skb_checksum_init(skb, proto, compute_pseudo) \ 3095 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3096 3097 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3098 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3099 3100 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 3101 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3102 3103 #define skb_checksum_validate_zero_check(skb, proto, check, \ 3104 compute_pseudo) \ 3105 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3106 3107 #define skb_checksum_simple_validate(skb) \ 3108 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3109 3110 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3111 { 3112 return (skb->ip_summed == CHECKSUM_NONE && 3113 skb->csum_valid && !skb->csum_bad); 3114 } 3115 3116 static inline void __skb_checksum_convert(struct sk_buff *skb, 3117 __sum16 check, __wsum pseudo) 3118 { 3119 skb->csum = ~pseudo; 3120 skb->ip_summed = CHECKSUM_COMPLETE; 3121 } 3122 3123 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3124 do { \ 3125 if (__skb_checksum_convert_check(skb)) \ 3126 __skb_checksum_convert(skb, check, \ 3127 compute_pseudo(skb, proto)); \ 3128 } while (0) 3129 3130 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3131 u16 start, u16 offset) 3132 { 3133 skb->ip_summed = CHECKSUM_PARTIAL; 3134 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3135 skb->csum_offset = offset - start; 3136 } 3137 3138 /* Update skbuf and packet to reflect the remote checksum offload operation. 3139 * When called, ptr indicates the starting point for skb->csum when 3140 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3141 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3142 */ 3143 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3144 int start, int offset, bool nopartial) 3145 { 3146 __wsum delta; 3147 3148 if (!nopartial) { 3149 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3150 return; 3151 } 3152 3153 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3154 __skb_checksum_complete(skb); 3155 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3156 } 3157 3158 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3159 3160 /* Adjust skb->csum since we changed the packet */ 3161 skb->csum = csum_add(skb->csum, delta); 3162 } 3163 3164 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3165 void nf_conntrack_destroy(struct nf_conntrack *nfct); 3166 static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3167 { 3168 if (nfct && atomic_dec_and_test(&nfct->use)) 3169 nf_conntrack_destroy(nfct); 3170 } 3171 static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3172 { 3173 if (nfct) 3174 atomic_inc(&nfct->use); 3175 } 3176 #endif 3177 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3178 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3179 { 3180 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 3181 kfree(nf_bridge); 3182 } 3183 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3184 { 3185 if (nf_bridge) 3186 atomic_inc(&nf_bridge->use); 3187 } 3188 #endif /* CONFIG_BRIDGE_NETFILTER */ 3189 static inline void nf_reset(struct sk_buff *skb) 3190 { 3191 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3192 nf_conntrack_put(skb->nfct); 3193 skb->nfct = NULL; 3194 #endif 3195 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3196 nf_bridge_put(skb->nf_bridge); 3197 skb->nf_bridge = NULL; 3198 #endif 3199 } 3200 3201 static inline void nf_reset_trace(struct sk_buff *skb) 3202 { 3203 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3204 skb->nf_trace = 0; 3205 #endif 3206 } 3207 3208 /* Note: This doesn't put any conntrack and bridge info in dst. */ 3209 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3210 bool copy) 3211 { 3212 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3213 dst->nfct = src->nfct; 3214 nf_conntrack_get(src->nfct); 3215 if (copy) 3216 dst->nfctinfo = src->nfctinfo; 3217 #endif 3218 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3219 dst->nf_bridge = src->nf_bridge; 3220 nf_bridge_get(src->nf_bridge); 3221 #endif 3222 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3223 if (copy) 3224 dst->nf_trace = src->nf_trace; 3225 #endif 3226 } 3227 3228 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3229 { 3230 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3231 nf_conntrack_put(dst->nfct); 3232 #endif 3233 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3234 nf_bridge_put(dst->nf_bridge); 3235 #endif 3236 __nf_copy(dst, src, true); 3237 } 3238 3239 #ifdef CONFIG_NETWORK_SECMARK 3240 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3241 { 3242 to->secmark = from->secmark; 3243 } 3244 3245 static inline void skb_init_secmark(struct sk_buff *skb) 3246 { 3247 skb->secmark = 0; 3248 } 3249 #else 3250 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3251 { } 3252 3253 static inline void skb_init_secmark(struct sk_buff *skb) 3254 { } 3255 #endif 3256 3257 static inline bool skb_irq_freeable(const struct sk_buff *skb) 3258 { 3259 return !skb->destructor && 3260 #if IS_ENABLED(CONFIG_XFRM) 3261 !skb->sp && 3262 #endif 3263 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 3264 !skb->nfct && 3265 #endif 3266 !skb->_skb_refdst && 3267 !skb_has_frag_list(skb); 3268 } 3269 3270 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3271 { 3272 skb->queue_mapping = queue_mapping; 3273 } 3274 3275 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3276 { 3277 return skb->queue_mapping; 3278 } 3279 3280 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3281 { 3282 to->queue_mapping = from->queue_mapping; 3283 } 3284 3285 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3286 { 3287 skb->queue_mapping = rx_queue + 1; 3288 } 3289 3290 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3291 { 3292 return skb->queue_mapping - 1; 3293 } 3294 3295 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3296 { 3297 return skb->queue_mapping != 0; 3298 } 3299 3300 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, 3301 unsigned int num_tx_queues); 3302 3303 static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3304 { 3305 #ifdef CONFIG_XFRM 3306 return skb->sp; 3307 #else 3308 return NULL; 3309 #endif 3310 } 3311 3312 /* Keeps track of mac header offset relative to skb->head. 3313 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3314 * For non-tunnel skb it points to skb_mac_header() and for 3315 * tunnel skb it points to outer mac header. 3316 * Keeps track of level of encapsulation of network headers. 3317 */ 3318 struct skb_gso_cb { 3319 int mac_offset; 3320 int encap_level; 3321 __u16 csum_start; 3322 }; 3323 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb) 3324 3325 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3326 { 3327 return (skb_mac_header(inner_skb) - inner_skb->head) - 3328 SKB_GSO_CB(inner_skb)->mac_offset; 3329 } 3330 3331 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3332 { 3333 int new_headroom, headroom; 3334 int ret; 3335 3336 headroom = skb_headroom(skb); 3337 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3338 if (ret) 3339 return ret; 3340 3341 new_headroom = skb_headroom(skb); 3342 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3343 return 0; 3344 } 3345 3346 /* Compute the checksum for a gso segment. First compute the checksum value 3347 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3348 * then add in skb->csum (checksum from csum_start to end of packet). 3349 * skb->csum and csum_start are then updated to reflect the checksum of the 3350 * resultant packet starting from the transport header-- the resultant checksum 3351 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3352 * header. 3353 */ 3354 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3355 { 3356 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) - 3357 skb_transport_offset(skb); 3358 __u16 csum; 3359 3360 csum = csum_fold(csum_partial(skb_transport_header(skb), 3361 plen, skb->csum)); 3362 skb->csum = res; 3363 SKB_GSO_CB(skb)->csum_start -= plen; 3364 3365 return csum; 3366 } 3367 3368 static inline bool skb_is_gso(const struct sk_buff *skb) 3369 { 3370 return skb_shinfo(skb)->gso_size; 3371 } 3372 3373 /* Note: Should be called only if skb_is_gso(skb) is true */ 3374 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3375 { 3376 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3377 } 3378 3379 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3380 3381 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3382 { 3383 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3384 * wanted then gso_type will be set. */ 3385 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3386 3387 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3388 unlikely(shinfo->gso_type == 0)) { 3389 __skb_warn_lro_forwarding(skb); 3390 return true; 3391 } 3392 return false; 3393 } 3394 3395 static inline void skb_forward_csum(struct sk_buff *skb) 3396 { 3397 /* Unfortunately we don't support this one. Any brave souls? */ 3398 if (skb->ip_summed == CHECKSUM_COMPLETE) 3399 skb->ip_summed = CHECKSUM_NONE; 3400 } 3401 3402 /** 3403 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3404 * @skb: skb to check 3405 * 3406 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3407 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3408 * use this helper, to document places where we make this assertion. 3409 */ 3410 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3411 { 3412 #ifdef DEBUG 3413 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3414 #endif 3415 } 3416 3417 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3418 3419 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3420 3421 u32 skb_get_poff(const struct sk_buff *skb); 3422 u32 __skb_get_poff(const struct sk_buff *skb, void *data, 3423 const struct flow_keys *keys, int hlen); 3424 3425 /** 3426 * skb_head_is_locked - Determine if the skb->head is locked down 3427 * @skb: skb to check 3428 * 3429 * The head on skbs build around a head frag can be removed if they are 3430 * not cloned. This function returns true if the skb head is locked down 3431 * due to either being allocated via kmalloc, or by being a clone with 3432 * multiple references to the head. 3433 */ 3434 static inline bool skb_head_is_locked(const struct sk_buff *skb) 3435 { 3436 return !skb->head_frag || skb_cloned(skb); 3437 } 3438 3439 /** 3440 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3441 * 3442 * @skb: GSO skb 3443 * 3444 * skb_gso_network_seglen is used to determine the real size of the 3445 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3446 * 3447 * The MAC/L2 header is not accounted for. 3448 */ 3449 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 3450 { 3451 unsigned int hdr_len = skb_transport_header(skb) - 3452 skb_network_header(skb); 3453 return hdr_len + skb_gso_transport_seglen(skb); 3454 } 3455 #endif /* __KERNEL__ */ 3456 #endif /* _LINUX_SKBUFF_H */ 3457