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 static inline struct sk_buff *alloc_skb(unsigned int size, 778 gfp_t priority) 779 { 780 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 781 } 782 783 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 784 unsigned long data_len, 785 int max_page_order, 786 int *errcode, 787 gfp_t gfp_mask); 788 789 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 790 struct sk_buff_fclones { 791 struct sk_buff skb1; 792 793 struct sk_buff skb2; 794 795 atomic_t fclone_ref; 796 }; 797 798 /** 799 * skb_fclone_busy - check if fclone is busy 800 * @skb: buffer 801 * 802 * Returns true is skb is a fast clone, and its clone is not freed. 803 * Some drivers call skb_orphan() in their ndo_start_xmit(), 804 * so we also check that this didnt happen. 805 */ 806 static inline bool skb_fclone_busy(const struct sock *sk, 807 const struct sk_buff *skb) 808 { 809 const struct sk_buff_fclones *fclones; 810 811 fclones = container_of(skb, struct sk_buff_fclones, skb1); 812 813 return skb->fclone == SKB_FCLONE_ORIG && 814 atomic_read(&fclones->fclone_ref) > 1 && 815 fclones->skb2.sk == sk; 816 } 817 818 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 819 gfp_t priority) 820 { 821 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 822 } 823 824 struct sk_buff *__alloc_skb_head(gfp_t priority, int node); 825 static inline struct sk_buff *alloc_skb_head(gfp_t priority) 826 { 827 return __alloc_skb_head(priority, -1); 828 } 829 830 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 831 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 832 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 833 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 834 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 835 gfp_t gfp_mask, bool fclone); 836 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 837 gfp_t gfp_mask) 838 { 839 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 840 } 841 842 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 843 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 844 unsigned int headroom); 845 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 846 int newtailroom, gfp_t priority); 847 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 848 int offset, int len); 849 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, 850 int len); 851 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 852 int skb_pad(struct sk_buff *skb, int pad); 853 #define dev_kfree_skb(a) consume_skb(a) 854 855 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 856 int getfrag(void *from, char *to, int offset, 857 int len, int odd, struct sk_buff *skb), 858 void *from, int length); 859 860 struct skb_seq_state { 861 __u32 lower_offset; 862 __u32 upper_offset; 863 __u32 frag_idx; 864 __u32 stepped_offset; 865 struct sk_buff *root_skb; 866 struct sk_buff *cur_skb; 867 __u8 *frag_data; 868 }; 869 870 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 871 unsigned int to, struct skb_seq_state *st); 872 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 873 struct skb_seq_state *st); 874 void skb_abort_seq_read(struct skb_seq_state *st); 875 876 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 877 unsigned int to, struct ts_config *config); 878 879 /* 880 * Packet hash types specify the type of hash in skb_set_hash. 881 * 882 * Hash types refer to the protocol layer addresses which are used to 883 * construct a packet's hash. The hashes are used to differentiate or identify 884 * flows of the protocol layer for the hash type. Hash types are either 885 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 886 * 887 * Properties of hashes: 888 * 889 * 1) Two packets in different flows have different hash values 890 * 2) Two packets in the same flow should have the same hash value 891 * 892 * A hash at a higher layer is considered to be more specific. A driver should 893 * set the most specific hash possible. 894 * 895 * A driver cannot indicate a more specific hash than the layer at which a hash 896 * was computed. For instance an L3 hash cannot be set as an L4 hash. 897 * 898 * A driver may indicate a hash level which is less specific than the 899 * actual layer the hash was computed on. For instance, a hash computed 900 * at L4 may be considered an L3 hash. This should only be done if the 901 * driver can't unambiguously determine that the HW computed the hash at 902 * the higher layer. Note that the "should" in the second property above 903 * permits this. 904 */ 905 enum pkt_hash_types { 906 PKT_HASH_TYPE_NONE, /* Undefined type */ 907 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 908 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 909 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 910 }; 911 912 static inline void 913 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 914 { 915 skb->l4_hash = (type == PKT_HASH_TYPE_L4); 916 skb->sw_hash = 0; 917 skb->hash = hash; 918 } 919 920 void __skb_get_hash(struct sk_buff *skb); 921 static inline __u32 skb_get_hash(struct sk_buff *skb) 922 { 923 if (!skb->l4_hash && !skb->sw_hash) 924 __skb_get_hash(skb); 925 926 return skb->hash; 927 } 928 929 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 930 { 931 return skb->hash; 932 } 933 934 static inline void skb_clear_hash(struct sk_buff *skb) 935 { 936 skb->hash = 0; 937 skb->sw_hash = 0; 938 skb->l4_hash = 0; 939 } 940 941 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 942 { 943 if (!skb->l4_hash) 944 skb_clear_hash(skb); 945 } 946 947 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 948 { 949 to->hash = from->hash; 950 to->sw_hash = from->sw_hash; 951 to->l4_hash = from->l4_hash; 952 }; 953 954 static inline void skb_sender_cpu_clear(struct sk_buff *skb) 955 { 956 #ifdef CONFIG_XPS 957 skb->sender_cpu = 0; 958 #endif 959 } 960 961 #ifdef NET_SKBUFF_DATA_USES_OFFSET 962 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 963 { 964 return skb->head + skb->end; 965 } 966 967 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 968 { 969 return skb->end; 970 } 971 #else 972 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 973 { 974 return skb->end; 975 } 976 977 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 978 { 979 return skb->end - skb->head; 980 } 981 #endif 982 983 /* Internal */ 984 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 985 986 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 987 { 988 return &skb_shinfo(skb)->hwtstamps; 989 } 990 991 /** 992 * skb_queue_empty - check if a queue is empty 993 * @list: queue head 994 * 995 * Returns true if the queue is empty, false otherwise. 996 */ 997 static inline int skb_queue_empty(const struct sk_buff_head *list) 998 { 999 return list->next == (const struct sk_buff *) list; 1000 } 1001 1002 /** 1003 * skb_queue_is_last - check if skb is the last entry in the queue 1004 * @list: queue head 1005 * @skb: buffer 1006 * 1007 * Returns true if @skb is the last buffer on the list. 1008 */ 1009 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1010 const struct sk_buff *skb) 1011 { 1012 return skb->next == (const struct sk_buff *) list; 1013 } 1014 1015 /** 1016 * skb_queue_is_first - check if skb is the first entry in the queue 1017 * @list: queue head 1018 * @skb: buffer 1019 * 1020 * Returns true if @skb is the first buffer on the list. 1021 */ 1022 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1023 const struct sk_buff *skb) 1024 { 1025 return skb->prev == (const struct sk_buff *) list; 1026 } 1027 1028 /** 1029 * skb_queue_next - return the next packet in the queue 1030 * @list: queue head 1031 * @skb: current buffer 1032 * 1033 * Return the next packet in @list after @skb. It is only valid to 1034 * call this if skb_queue_is_last() evaluates to false. 1035 */ 1036 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1037 const struct sk_buff *skb) 1038 { 1039 /* This BUG_ON may seem severe, but if we just return then we 1040 * are going to dereference garbage. 1041 */ 1042 BUG_ON(skb_queue_is_last(list, skb)); 1043 return skb->next; 1044 } 1045 1046 /** 1047 * skb_queue_prev - return the prev packet in the queue 1048 * @list: queue head 1049 * @skb: current buffer 1050 * 1051 * Return the prev packet in @list before @skb. It is only valid to 1052 * call this if skb_queue_is_first() evaluates to false. 1053 */ 1054 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1055 const struct sk_buff *skb) 1056 { 1057 /* This BUG_ON may seem severe, but if we just return then we 1058 * are going to dereference garbage. 1059 */ 1060 BUG_ON(skb_queue_is_first(list, skb)); 1061 return skb->prev; 1062 } 1063 1064 /** 1065 * skb_get - reference buffer 1066 * @skb: buffer to reference 1067 * 1068 * Makes another reference to a socket buffer and returns a pointer 1069 * to the buffer. 1070 */ 1071 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1072 { 1073 atomic_inc(&skb->users); 1074 return skb; 1075 } 1076 1077 /* 1078 * If users == 1, we are the only owner and are can avoid redundant 1079 * atomic change. 1080 */ 1081 1082 /** 1083 * skb_cloned - is the buffer a clone 1084 * @skb: buffer to check 1085 * 1086 * Returns true if the buffer was generated with skb_clone() and is 1087 * one of multiple shared copies of the buffer. Cloned buffers are 1088 * shared data so must not be written to under normal circumstances. 1089 */ 1090 static inline int skb_cloned(const struct sk_buff *skb) 1091 { 1092 return skb->cloned && 1093 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1094 } 1095 1096 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1097 { 1098 might_sleep_if(pri & __GFP_WAIT); 1099 1100 if (skb_cloned(skb)) 1101 return pskb_expand_head(skb, 0, 0, pri); 1102 1103 return 0; 1104 } 1105 1106 /** 1107 * skb_header_cloned - is the header a clone 1108 * @skb: buffer to check 1109 * 1110 * Returns true if modifying the header part of the buffer requires 1111 * the data to be copied. 1112 */ 1113 static inline int skb_header_cloned(const struct sk_buff *skb) 1114 { 1115 int dataref; 1116 1117 if (!skb->cloned) 1118 return 0; 1119 1120 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1121 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1122 return dataref != 1; 1123 } 1124 1125 /** 1126 * skb_header_release - release reference to header 1127 * @skb: buffer to operate on 1128 * 1129 * Drop a reference to the header part of the buffer. This is done 1130 * by acquiring a payload reference. You must not read from the header 1131 * part of skb->data after this. 1132 * Note : Check if you can use __skb_header_release() instead. 1133 */ 1134 static inline void skb_header_release(struct sk_buff *skb) 1135 { 1136 BUG_ON(skb->nohdr); 1137 skb->nohdr = 1; 1138 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1139 } 1140 1141 /** 1142 * __skb_header_release - release reference to header 1143 * @skb: buffer to operate on 1144 * 1145 * Variant of skb_header_release() assuming skb is private to caller. 1146 * We can avoid one atomic operation. 1147 */ 1148 static inline void __skb_header_release(struct sk_buff *skb) 1149 { 1150 skb->nohdr = 1; 1151 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1152 } 1153 1154 1155 /** 1156 * skb_shared - is the buffer shared 1157 * @skb: buffer to check 1158 * 1159 * Returns true if more than one person has a reference to this 1160 * buffer. 1161 */ 1162 static inline int skb_shared(const struct sk_buff *skb) 1163 { 1164 return atomic_read(&skb->users) != 1; 1165 } 1166 1167 /** 1168 * skb_share_check - check if buffer is shared and if so clone it 1169 * @skb: buffer to check 1170 * @pri: priority for memory allocation 1171 * 1172 * If the buffer is shared the buffer is cloned and the old copy 1173 * drops a reference. A new clone with a single reference is returned. 1174 * If the buffer is not shared the original buffer is returned. When 1175 * being called from interrupt status or with spinlocks held pri must 1176 * be GFP_ATOMIC. 1177 * 1178 * NULL is returned on a memory allocation failure. 1179 */ 1180 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1181 { 1182 might_sleep_if(pri & __GFP_WAIT); 1183 if (skb_shared(skb)) { 1184 struct sk_buff *nskb = skb_clone(skb, pri); 1185 1186 if (likely(nskb)) 1187 consume_skb(skb); 1188 else 1189 kfree_skb(skb); 1190 skb = nskb; 1191 } 1192 return skb; 1193 } 1194 1195 /* 1196 * Copy shared buffers into a new sk_buff. We effectively do COW on 1197 * packets to handle cases where we have a local reader and forward 1198 * and a couple of other messy ones. The normal one is tcpdumping 1199 * a packet thats being forwarded. 1200 */ 1201 1202 /** 1203 * skb_unshare - make a copy of a shared buffer 1204 * @skb: buffer to check 1205 * @pri: priority for memory allocation 1206 * 1207 * If the socket buffer is a clone then this function creates a new 1208 * copy of the data, drops a reference count on the old copy and returns 1209 * the new copy with the reference count at 1. If the buffer is not a clone 1210 * the original buffer is returned. When called with a spinlock held or 1211 * from interrupt state @pri must be %GFP_ATOMIC 1212 * 1213 * %NULL is returned on a memory allocation failure. 1214 */ 1215 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1216 gfp_t pri) 1217 { 1218 might_sleep_if(pri & __GFP_WAIT); 1219 if (skb_cloned(skb)) { 1220 struct sk_buff *nskb = skb_copy(skb, pri); 1221 1222 /* Free our shared copy */ 1223 if (likely(nskb)) 1224 consume_skb(skb); 1225 else 1226 kfree_skb(skb); 1227 skb = nskb; 1228 } 1229 return skb; 1230 } 1231 1232 /** 1233 * skb_peek - peek at the head of an &sk_buff_head 1234 * @list_: list to peek at 1235 * 1236 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1237 * be careful with this one. A peek leaves the buffer on the 1238 * list and someone else may run off with it. You must hold 1239 * the appropriate locks or have a private queue to do this. 1240 * 1241 * Returns %NULL for an empty list or a pointer to the head element. 1242 * The reference count is not incremented and the reference is therefore 1243 * volatile. Use with caution. 1244 */ 1245 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1246 { 1247 struct sk_buff *skb = list_->next; 1248 1249 if (skb == (struct sk_buff *)list_) 1250 skb = NULL; 1251 return skb; 1252 } 1253 1254 /** 1255 * skb_peek_next - peek skb following the given one from a queue 1256 * @skb: skb to start from 1257 * @list_: list to peek at 1258 * 1259 * Returns %NULL when the end of the list is met or a pointer to the 1260 * next element. The reference count is not incremented and the 1261 * reference is therefore volatile. Use with caution. 1262 */ 1263 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1264 const struct sk_buff_head *list_) 1265 { 1266 struct sk_buff *next = skb->next; 1267 1268 if (next == (struct sk_buff *)list_) 1269 next = NULL; 1270 return next; 1271 } 1272 1273 /** 1274 * skb_peek_tail - peek at the tail of an &sk_buff_head 1275 * @list_: list to peek at 1276 * 1277 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1278 * be careful with this one. A peek leaves the buffer on the 1279 * list and someone else may run off with it. You must hold 1280 * the appropriate locks or have a private queue to do this. 1281 * 1282 * Returns %NULL for an empty list or a pointer to the tail element. 1283 * The reference count is not incremented and the reference is therefore 1284 * volatile. Use with caution. 1285 */ 1286 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1287 { 1288 struct sk_buff *skb = list_->prev; 1289 1290 if (skb == (struct sk_buff *)list_) 1291 skb = NULL; 1292 return skb; 1293 1294 } 1295 1296 /** 1297 * skb_queue_len - get queue length 1298 * @list_: list to measure 1299 * 1300 * Return the length of an &sk_buff queue. 1301 */ 1302 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1303 { 1304 return list_->qlen; 1305 } 1306 1307 /** 1308 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1309 * @list: queue to initialize 1310 * 1311 * This initializes only the list and queue length aspects of 1312 * an sk_buff_head object. This allows to initialize the list 1313 * aspects of an sk_buff_head without reinitializing things like 1314 * the spinlock. It can also be used for on-stack sk_buff_head 1315 * objects where the spinlock is known to not be used. 1316 */ 1317 static inline void __skb_queue_head_init(struct sk_buff_head *list) 1318 { 1319 list->prev = list->next = (struct sk_buff *)list; 1320 list->qlen = 0; 1321 } 1322 1323 /* 1324 * This function creates a split out lock class for each invocation; 1325 * this is needed for now since a whole lot of users of the skb-queue 1326 * infrastructure in drivers have different locking usage (in hardirq) 1327 * than the networking core (in softirq only). In the long run either the 1328 * network layer or drivers should need annotation to consolidate the 1329 * main types of usage into 3 classes. 1330 */ 1331 static inline void skb_queue_head_init(struct sk_buff_head *list) 1332 { 1333 spin_lock_init(&list->lock); 1334 __skb_queue_head_init(list); 1335 } 1336 1337 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1338 struct lock_class_key *class) 1339 { 1340 skb_queue_head_init(list); 1341 lockdep_set_class(&list->lock, class); 1342 } 1343 1344 /* 1345 * Insert an sk_buff on a list. 1346 * 1347 * The "__skb_xxxx()" functions are the non-atomic ones that 1348 * can only be called with interrupts disabled. 1349 */ 1350 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1351 struct sk_buff_head *list); 1352 static inline void __skb_insert(struct sk_buff *newsk, 1353 struct sk_buff *prev, struct sk_buff *next, 1354 struct sk_buff_head *list) 1355 { 1356 newsk->next = next; 1357 newsk->prev = prev; 1358 next->prev = prev->next = newsk; 1359 list->qlen++; 1360 } 1361 1362 static inline void __skb_queue_splice(const struct sk_buff_head *list, 1363 struct sk_buff *prev, 1364 struct sk_buff *next) 1365 { 1366 struct sk_buff *first = list->next; 1367 struct sk_buff *last = list->prev; 1368 1369 first->prev = prev; 1370 prev->next = first; 1371 1372 last->next = next; 1373 next->prev = last; 1374 } 1375 1376 /** 1377 * skb_queue_splice - join two skb lists, this is designed for stacks 1378 * @list: the new list to add 1379 * @head: the place to add it in the first list 1380 */ 1381 static inline void skb_queue_splice(const struct sk_buff_head *list, 1382 struct sk_buff_head *head) 1383 { 1384 if (!skb_queue_empty(list)) { 1385 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1386 head->qlen += list->qlen; 1387 } 1388 } 1389 1390 /** 1391 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1392 * @list: the new list to add 1393 * @head: the place to add it in the first list 1394 * 1395 * The list at @list is reinitialised 1396 */ 1397 static inline void skb_queue_splice_init(struct sk_buff_head *list, 1398 struct sk_buff_head *head) 1399 { 1400 if (!skb_queue_empty(list)) { 1401 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1402 head->qlen += list->qlen; 1403 __skb_queue_head_init(list); 1404 } 1405 } 1406 1407 /** 1408 * skb_queue_splice_tail - join two skb lists, each list being a queue 1409 * @list: the new list to add 1410 * @head: the place to add it in the first list 1411 */ 1412 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1413 struct sk_buff_head *head) 1414 { 1415 if (!skb_queue_empty(list)) { 1416 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1417 head->qlen += list->qlen; 1418 } 1419 } 1420 1421 /** 1422 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1423 * @list: the new list to add 1424 * @head: the place to add it in the first list 1425 * 1426 * Each of the lists is a queue. 1427 * The list at @list is reinitialised 1428 */ 1429 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1430 struct sk_buff_head *head) 1431 { 1432 if (!skb_queue_empty(list)) { 1433 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1434 head->qlen += list->qlen; 1435 __skb_queue_head_init(list); 1436 } 1437 } 1438 1439 /** 1440 * __skb_queue_after - queue a buffer at the list head 1441 * @list: list to use 1442 * @prev: place after this buffer 1443 * @newsk: buffer to queue 1444 * 1445 * Queue a buffer int the middle of a list. This function takes no locks 1446 * and you must therefore hold required locks before calling it. 1447 * 1448 * A buffer cannot be placed on two lists at the same time. 1449 */ 1450 static inline void __skb_queue_after(struct sk_buff_head *list, 1451 struct sk_buff *prev, 1452 struct sk_buff *newsk) 1453 { 1454 __skb_insert(newsk, prev, prev->next, list); 1455 } 1456 1457 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1458 struct sk_buff_head *list); 1459 1460 static inline void __skb_queue_before(struct sk_buff_head *list, 1461 struct sk_buff *next, 1462 struct sk_buff *newsk) 1463 { 1464 __skb_insert(newsk, next->prev, next, list); 1465 } 1466 1467 /** 1468 * __skb_queue_head - queue a buffer at the list head 1469 * @list: list to use 1470 * @newsk: buffer to queue 1471 * 1472 * Queue a buffer at the start of a list. This function takes no locks 1473 * and you must therefore hold required locks before calling it. 1474 * 1475 * A buffer cannot be placed on two lists at the same time. 1476 */ 1477 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1478 static inline void __skb_queue_head(struct sk_buff_head *list, 1479 struct sk_buff *newsk) 1480 { 1481 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1482 } 1483 1484 /** 1485 * __skb_queue_tail - queue a buffer at the list tail 1486 * @list: list to use 1487 * @newsk: buffer to queue 1488 * 1489 * Queue a buffer at the end of a list. This function takes no locks 1490 * and you must therefore hold required locks before calling it. 1491 * 1492 * A buffer cannot be placed on two lists at the same time. 1493 */ 1494 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1495 static inline void __skb_queue_tail(struct sk_buff_head *list, 1496 struct sk_buff *newsk) 1497 { 1498 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1499 } 1500 1501 /* 1502 * remove sk_buff from list. _Must_ be called atomically, and with 1503 * the list known.. 1504 */ 1505 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1506 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1507 { 1508 struct sk_buff *next, *prev; 1509 1510 list->qlen--; 1511 next = skb->next; 1512 prev = skb->prev; 1513 skb->next = skb->prev = NULL; 1514 next->prev = prev; 1515 prev->next = next; 1516 } 1517 1518 /** 1519 * __skb_dequeue - remove from the head of the queue 1520 * @list: list to dequeue from 1521 * 1522 * Remove the head of the list. This function does not take any locks 1523 * so must be used with appropriate locks held only. The head item is 1524 * returned or %NULL if the list is empty. 1525 */ 1526 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1527 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1528 { 1529 struct sk_buff *skb = skb_peek(list); 1530 if (skb) 1531 __skb_unlink(skb, list); 1532 return skb; 1533 } 1534 1535 /** 1536 * __skb_dequeue_tail - remove from the tail of the queue 1537 * @list: list to dequeue from 1538 * 1539 * Remove the tail of the list. This function does not take any locks 1540 * so must be used with appropriate locks held only. The tail item is 1541 * returned or %NULL if the list is empty. 1542 */ 1543 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1544 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1545 { 1546 struct sk_buff *skb = skb_peek_tail(list); 1547 if (skb) 1548 __skb_unlink(skb, list); 1549 return skb; 1550 } 1551 1552 1553 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1554 { 1555 return skb->data_len; 1556 } 1557 1558 static inline unsigned int skb_headlen(const struct sk_buff *skb) 1559 { 1560 return skb->len - skb->data_len; 1561 } 1562 1563 static inline int skb_pagelen(const struct sk_buff *skb) 1564 { 1565 int i, len = 0; 1566 1567 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--) 1568 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1569 return len + skb_headlen(skb); 1570 } 1571 1572 /** 1573 * __skb_fill_page_desc - initialise a paged fragment in an skb 1574 * @skb: buffer containing fragment to be initialised 1575 * @i: paged fragment index to initialise 1576 * @page: the page to use for this fragment 1577 * @off: the offset to the data with @page 1578 * @size: the length of the data 1579 * 1580 * Initialises the @i'th fragment of @skb to point to &size bytes at 1581 * offset @off within @page. 1582 * 1583 * Does not take any additional reference on the fragment. 1584 */ 1585 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1586 struct page *page, int off, int size) 1587 { 1588 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1589 1590 /* 1591 * Propagate page->pfmemalloc to the skb if we can. The problem is 1592 * that not all callers have unique ownership of the page. If 1593 * pfmemalloc is set, we check the mapping as a mapping implies 1594 * page->index is set (index and pfmemalloc share space). 1595 * If it's a valid mapping, we cannot use page->pfmemalloc but we 1596 * do not lose pfmemalloc information as the pages would not be 1597 * allocated using __GFP_MEMALLOC. 1598 */ 1599 frag->page.p = page; 1600 frag->page_offset = off; 1601 skb_frag_size_set(frag, size); 1602 1603 page = compound_head(page); 1604 if (page->pfmemalloc && !page->mapping) 1605 skb->pfmemalloc = true; 1606 } 1607 1608 /** 1609 * skb_fill_page_desc - initialise a paged fragment in an skb 1610 * @skb: buffer containing fragment to be initialised 1611 * @i: paged fragment index to initialise 1612 * @page: the page to use for this fragment 1613 * @off: the offset to the data with @page 1614 * @size: the length of the data 1615 * 1616 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1617 * @skb to point to @size bytes at offset @off within @page. In 1618 * addition updates @skb such that @i is the last fragment. 1619 * 1620 * Does not take any additional reference on the fragment. 1621 */ 1622 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1623 struct page *page, int off, int size) 1624 { 1625 __skb_fill_page_desc(skb, i, page, off, size); 1626 skb_shinfo(skb)->nr_frags = i + 1; 1627 } 1628 1629 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1630 int size, unsigned int truesize); 1631 1632 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1633 unsigned int truesize); 1634 1635 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1636 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1637 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1638 1639 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1640 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1641 { 1642 return skb->head + skb->tail; 1643 } 1644 1645 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1646 { 1647 skb->tail = skb->data - skb->head; 1648 } 1649 1650 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1651 { 1652 skb_reset_tail_pointer(skb); 1653 skb->tail += offset; 1654 } 1655 1656 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 1657 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1658 { 1659 return skb->tail; 1660 } 1661 1662 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1663 { 1664 skb->tail = skb->data; 1665 } 1666 1667 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1668 { 1669 skb->tail = skb->data + offset; 1670 } 1671 1672 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1673 1674 /* 1675 * Add data to an sk_buff 1676 */ 1677 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1678 unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1679 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1680 { 1681 unsigned char *tmp = skb_tail_pointer(skb); 1682 SKB_LINEAR_ASSERT(skb); 1683 skb->tail += len; 1684 skb->len += len; 1685 return tmp; 1686 } 1687 1688 unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1689 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1690 { 1691 skb->data -= len; 1692 skb->len += len; 1693 return skb->data; 1694 } 1695 1696 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1697 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1698 { 1699 skb->len -= len; 1700 BUG_ON(skb->len < skb->data_len); 1701 return skb->data += len; 1702 } 1703 1704 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1705 { 1706 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1707 } 1708 1709 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1710 1711 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1712 { 1713 if (len > skb_headlen(skb) && 1714 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1715 return NULL; 1716 skb->len -= len; 1717 return skb->data += len; 1718 } 1719 1720 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1721 { 1722 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1723 } 1724 1725 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1726 { 1727 if (likely(len <= skb_headlen(skb))) 1728 return 1; 1729 if (unlikely(len > skb->len)) 1730 return 0; 1731 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1732 } 1733 1734 /** 1735 * skb_headroom - bytes at buffer head 1736 * @skb: buffer to check 1737 * 1738 * Return the number of bytes of free space at the head of an &sk_buff. 1739 */ 1740 static inline unsigned int skb_headroom(const struct sk_buff *skb) 1741 { 1742 return skb->data - skb->head; 1743 } 1744 1745 /** 1746 * skb_tailroom - bytes at buffer end 1747 * @skb: buffer to check 1748 * 1749 * Return the number of bytes of free space at the tail of an sk_buff 1750 */ 1751 static inline int skb_tailroom(const struct sk_buff *skb) 1752 { 1753 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 1754 } 1755 1756 /** 1757 * skb_availroom - bytes at buffer end 1758 * @skb: buffer to check 1759 * 1760 * Return the number of bytes of free space at the tail of an sk_buff 1761 * allocated by sk_stream_alloc() 1762 */ 1763 static inline int skb_availroom(const struct sk_buff *skb) 1764 { 1765 if (skb_is_nonlinear(skb)) 1766 return 0; 1767 1768 return skb->end - skb->tail - skb->reserved_tailroom; 1769 } 1770 1771 /** 1772 * skb_reserve - adjust headroom 1773 * @skb: buffer to alter 1774 * @len: bytes to move 1775 * 1776 * Increase the headroom of an empty &sk_buff by reducing the tail 1777 * room. This is only allowed for an empty buffer. 1778 */ 1779 static inline void skb_reserve(struct sk_buff *skb, int len) 1780 { 1781 skb->data += len; 1782 skb->tail += len; 1783 } 1784 1785 #define ENCAP_TYPE_ETHER 0 1786 #define ENCAP_TYPE_IPPROTO 1 1787 1788 static inline void skb_set_inner_protocol(struct sk_buff *skb, 1789 __be16 protocol) 1790 { 1791 skb->inner_protocol = protocol; 1792 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 1793 } 1794 1795 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 1796 __u8 ipproto) 1797 { 1798 skb->inner_ipproto = ipproto; 1799 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 1800 } 1801 1802 static inline void skb_reset_inner_headers(struct sk_buff *skb) 1803 { 1804 skb->inner_mac_header = skb->mac_header; 1805 skb->inner_network_header = skb->network_header; 1806 skb->inner_transport_header = skb->transport_header; 1807 } 1808 1809 static inline void skb_reset_mac_len(struct sk_buff *skb) 1810 { 1811 skb->mac_len = skb->network_header - skb->mac_header; 1812 } 1813 1814 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 1815 *skb) 1816 { 1817 return skb->head + skb->inner_transport_header; 1818 } 1819 1820 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 1821 { 1822 skb->inner_transport_header = skb->data - skb->head; 1823 } 1824 1825 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 1826 const int offset) 1827 { 1828 skb_reset_inner_transport_header(skb); 1829 skb->inner_transport_header += offset; 1830 } 1831 1832 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 1833 { 1834 return skb->head + skb->inner_network_header; 1835 } 1836 1837 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 1838 { 1839 skb->inner_network_header = skb->data - skb->head; 1840 } 1841 1842 static inline void skb_set_inner_network_header(struct sk_buff *skb, 1843 const int offset) 1844 { 1845 skb_reset_inner_network_header(skb); 1846 skb->inner_network_header += offset; 1847 } 1848 1849 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 1850 { 1851 return skb->head + skb->inner_mac_header; 1852 } 1853 1854 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 1855 { 1856 skb->inner_mac_header = skb->data - skb->head; 1857 } 1858 1859 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 1860 const int offset) 1861 { 1862 skb_reset_inner_mac_header(skb); 1863 skb->inner_mac_header += offset; 1864 } 1865 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 1866 { 1867 return skb->transport_header != (typeof(skb->transport_header))~0U; 1868 } 1869 1870 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 1871 { 1872 return skb->head + skb->transport_header; 1873 } 1874 1875 static inline void skb_reset_transport_header(struct sk_buff *skb) 1876 { 1877 skb->transport_header = skb->data - skb->head; 1878 } 1879 1880 static inline void skb_set_transport_header(struct sk_buff *skb, 1881 const int offset) 1882 { 1883 skb_reset_transport_header(skb); 1884 skb->transport_header += offset; 1885 } 1886 1887 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 1888 { 1889 return skb->head + skb->network_header; 1890 } 1891 1892 static inline void skb_reset_network_header(struct sk_buff *skb) 1893 { 1894 skb->network_header = skb->data - skb->head; 1895 } 1896 1897 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 1898 { 1899 skb_reset_network_header(skb); 1900 skb->network_header += offset; 1901 } 1902 1903 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 1904 { 1905 return skb->head + skb->mac_header; 1906 } 1907 1908 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 1909 { 1910 return skb->mac_header != (typeof(skb->mac_header))~0U; 1911 } 1912 1913 static inline void skb_reset_mac_header(struct sk_buff *skb) 1914 { 1915 skb->mac_header = skb->data - skb->head; 1916 } 1917 1918 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 1919 { 1920 skb_reset_mac_header(skb); 1921 skb->mac_header += offset; 1922 } 1923 1924 static inline void skb_pop_mac_header(struct sk_buff *skb) 1925 { 1926 skb->mac_header = skb->network_header; 1927 } 1928 1929 static inline void skb_probe_transport_header(struct sk_buff *skb, 1930 const int offset_hint) 1931 { 1932 struct flow_keys keys; 1933 1934 if (skb_transport_header_was_set(skb)) 1935 return; 1936 else if (skb_flow_dissect(skb, &keys)) 1937 skb_set_transport_header(skb, keys.thoff); 1938 else 1939 skb_set_transport_header(skb, offset_hint); 1940 } 1941 1942 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 1943 { 1944 if (skb_mac_header_was_set(skb)) { 1945 const unsigned char *old_mac = skb_mac_header(skb); 1946 1947 skb_set_mac_header(skb, -skb->mac_len); 1948 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 1949 } 1950 } 1951 1952 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 1953 { 1954 return skb->csum_start - skb_headroom(skb); 1955 } 1956 1957 static inline int skb_transport_offset(const struct sk_buff *skb) 1958 { 1959 return skb_transport_header(skb) - skb->data; 1960 } 1961 1962 static inline u32 skb_network_header_len(const struct sk_buff *skb) 1963 { 1964 return skb->transport_header - skb->network_header; 1965 } 1966 1967 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 1968 { 1969 return skb->inner_transport_header - skb->inner_network_header; 1970 } 1971 1972 static inline int skb_network_offset(const struct sk_buff *skb) 1973 { 1974 return skb_network_header(skb) - skb->data; 1975 } 1976 1977 static inline int skb_inner_network_offset(const struct sk_buff *skb) 1978 { 1979 return skb_inner_network_header(skb) - skb->data; 1980 } 1981 1982 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 1983 { 1984 return pskb_may_pull(skb, skb_network_offset(skb) + len); 1985 } 1986 1987 /* 1988 * CPUs often take a performance hit when accessing unaligned memory 1989 * locations. The actual performance hit varies, it can be small if the 1990 * hardware handles it or large if we have to take an exception and fix it 1991 * in software. 1992 * 1993 * Since an ethernet header is 14 bytes network drivers often end up with 1994 * the IP header at an unaligned offset. The IP header can be aligned by 1995 * shifting the start of the packet by 2 bytes. Drivers should do this 1996 * with: 1997 * 1998 * skb_reserve(skb, NET_IP_ALIGN); 1999 * 2000 * The downside to this alignment of the IP header is that the DMA is now 2001 * unaligned. On some architectures the cost of an unaligned DMA is high 2002 * and this cost outweighs the gains made by aligning the IP header. 2003 * 2004 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2005 * to be overridden. 2006 */ 2007 #ifndef NET_IP_ALIGN 2008 #define NET_IP_ALIGN 2 2009 #endif 2010 2011 /* 2012 * The networking layer reserves some headroom in skb data (via 2013 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2014 * the header has to grow. In the default case, if the header has to grow 2015 * 32 bytes or less we avoid the reallocation. 2016 * 2017 * Unfortunately this headroom changes the DMA alignment of the resulting 2018 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2019 * on some architectures. An architecture can override this value, 2020 * perhaps setting it to a cacheline in size (since that will maintain 2021 * cacheline alignment of the DMA). It must be a power of 2. 2022 * 2023 * Various parts of the networking layer expect at least 32 bytes of 2024 * headroom, you should not reduce this. 2025 * 2026 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2027 * to reduce average number of cache lines per packet. 2028 * get_rps_cpus() for example only access one 64 bytes aligned block : 2029 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2030 */ 2031 #ifndef NET_SKB_PAD 2032 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2033 #endif 2034 2035 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2036 2037 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2038 { 2039 if (unlikely(skb_is_nonlinear(skb))) { 2040 WARN_ON(1); 2041 return; 2042 } 2043 skb->len = len; 2044 skb_set_tail_pointer(skb, len); 2045 } 2046 2047 void skb_trim(struct sk_buff *skb, unsigned int len); 2048 2049 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2050 { 2051 if (skb->data_len) 2052 return ___pskb_trim(skb, len); 2053 __skb_trim(skb, len); 2054 return 0; 2055 } 2056 2057 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2058 { 2059 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2060 } 2061 2062 /** 2063 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2064 * @skb: buffer to alter 2065 * @len: new length 2066 * 2067 * This is identical to pskb_trim except that the caller knows that 2068 * the skb is not cloned so we should never get an error due to out- 2069 * of-memory. 2070 */ 2071 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2072 { 2073 int err = pskb_trim(skb, len); 2074 BUG_ON(err); 2075 } 2076 2077 /** 2078 * skb_orphan - orphan a buffer 2079 * @skb: buffer to orphan 2080 * 2081 * If a buffer currently has an owner then we call the owner's 2082 * destructor function and make the @skb unowned. The buffer continues 2083 * to exist but is no longer charged to its former owner. 2084 */ 2085 static inline void skb_orphan(struct sk_buff *skb) 2086 { 2087 if (skb->destructor) { 2088 skb->destructor(skb); 2089 skb->destructor = NULL; 2090 skb->sk = NULL; 2091 } else { 2092 BUG_ON(skb->sk); 2093 } 2094 } 2095 2096 /** 2097 * skb_orphan_frags - orphan the frags contained in a buffer 2098 * @skb: buffer to orphan frags from 2099 * @gfp_mask: allocation mask for replacement pages 2100 * 2101 * For each frag in the SKB which needs a destructor (i.e. has an 2102 * owner) create a copy of that frag and release the original 2103 * page by calling the destructor. 2104 */ 2105 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2106 { 2107 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 2108 return 0; 2109 return skb_copy_ubufs(skb, gfp_mask); 2110 } 2111 2112 /** 2113 * __skb_queue_purge - empty a list 2114 * @list: list to empty 2115 * 2116 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2117 * the list and one reference dropped. This function does not take the 2118 * list lock and the caller must hold the relevant locks to use it. 2119 */ 2120 void skb_queue_purge(struct sk_buff_head *list); 2121 static inline void __skb_queue_purge(struct sk_buff_head *list) 2122 { 2123 struct sk_buff *skb; 2124 while ((skb = __skb_dequeue(list)) != NULL) 2125 kfree_skb(skb); 2126 } 2127 2128 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768) 2129 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER) 2130 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE 2131 2132 void *netdev_alloc_frag(unsigned int fragsz); 2133 2134 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2135 gfp_t gfp_mask); 2136 2137 /** 2138 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2139 * @dev: network device to receive on 2140 * @length: length to allocate 2141 * 2142 * Allocate a new &sk_buff and assign it a usage count of one. The 2143 * buffer has unspecified headroom built in. Users should allocate 2144 * the headroom they think they need without accounting for the 2145 * built in space. The built in space is used for optimisations. 2146 * 2147 * %NULL is returned if there is no free memory. Although this function 2148 * allocates memory it can be called from an interrupt. 2149 */ 2150 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2151 unsigned int length) 2152 { 2153 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2154 } 2155 2156 /* legacy helper around __netdev_alloc_skb() */ 2157 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2158 gfp_t gfp_mask) 2159 { 2160 return __netdev_alloc_skb(NULL, length, gfp_mask); 2161 } 2162 2163 /* legacy helper around netdev_alloc_skb() */ 2164 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2165 { 2166 return netdev_alloc_skb(NULL, length); 2167 } 2168 2169 2170 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2171 unsigned int length, gfp_t gfp) 2172 { 2173 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2174 2175 if (NET_IP_ALIGN && skb) 2176 skb_reserve(skb, NET_IP_ALIGN); 2177 return skb; 2178 } 2179 2180 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2181 unsigned int length) 2182 { 2183 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2184 } 2185 2186 void *napi_alloc_frag(unsigned int fragsz); 2187 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2188 unsigned int length, gfp_t gfp_mask); 2189 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2190 unsigned int length) 2191 { 2192 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2193 } 2194 2195 /** 2196 * __dev_alloc_pages - allocate page for network Rx 2197 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2198 * @order: size of the allocation 2199 * 2200 * Allocate a new page. 2201 * 2202 * %NULL is returned if there is no free memory. 2203 */ 2204 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2205 unsigned int order) 2206 { 2207 /* This piece of code contains several assumptions. 2208 * 1. This is for device Rx, therefor a cold page is preferred. 2209 * 2. The expectation is the user wants a compound page. 2210 * 3. If requesting a order 0 page it will not be compound 2211 * due to the check to see if order has a value in prep_new_page 2212 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2213 * code in gfp_to_alloc_flags that should be enforcing this. 2214 */ 2215 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2216 2217 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2218 } 2219 2220 static inline struct page *dev_alloc_pages(unsigned int order) 2221 { 2222 return __dev_alloc_pages(GFP_ATOMIC, order); 2223 } 2224 2225 /** 2226 * __dev_alloc_page - allocate a page for network Rx 2227 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2228 * 2229 * Allocate a new page. 2230 * 2231 * %NULL is returned if there is no free memory. 2232 */ 2233 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2234 { 2235 return __dev_alloc_pages(gfp_mask, 0); 2236 } 2237 2238 static inline struct page *dev_alloc_page(void) 2239 { 2240 return __dev_alloc_page(GFP_ATOMIC); 2241 } 2242 2243 /** 2244 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2245 * @page: The page that was allocated from skb_alloc_page 2246 * @skb: The skb that may need pfmemalloc set 2247 */ 2248 static inline void skb_propagate_pfmemalloc(struct page *page, 2249 struct sk_buff *skb) 2250 { 2251 if (page && page->pfmemalloc) 2252 skb->pfmemalloc = true; 2253 } 2254 2255 /** 2256 * skb_frag_page - retrieve the page referred to by a paged fragment 2257 * @frag: the paged fragment 2258 * 2259 * Returns the &struct page associated with @frag. 2260 */ 2261 static inline struct page *skb_frag_page(const skb_frag_t *frag) 2262 { 2263 return frag->page.p; 2264 } 2265 2266 /** 2267 * __skb_frag_ref - take an addition reference on a paged fragment. 2268 * @frag: the paged fragment 2269 * 2270 * Takes an additional reference on the paged fragment @frag. 2271 */ 2272 static inline void __skb_frag_ref(skb_frag_t *frag) 2273 { 2274 get_page(skb_frag_page(frag)); 2275 } 2276 2277 /** 2278 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2279 * @skb: the buffer 2280 * @f: the fragment offset. 2281 * 2282 * Takes an additional reference on the @f'th paged fragment of @skb. 2283 */ 2284 static inline void skb_frag_ref(struct sk_buff *skb, int f) 2285 { 2286 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2287 } 2288 2289 /** 2290 * __skb_frag_unref - release a reference on a paged fragment. 2291 * @frag: the paged fragment 2292 * 2293 * Releases a reference on the paged fragment @frag. 2294 */ 2295 static inline void __skb_frag_unref(skb_frag_t *frag) 2296 { 2297 put_page(skb_frag_page(frag)); 2298 } 2299 2300 /** 2301 * skb_frag_unref - release a reference on a paged fragment of an skb. 2302 * @skb: the buffer 2303 * @f: the fragment offset 2304 * 2305 * Releases a reference on the @f'th paged fragment of @skb. 2306 */ 2307 static inline void skb_frag_unref(struct sk_buff *skb, int f) 2308 { 2309 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2310 } 2311 2312 /** 2313 * skb_frag_address - gets the address of the data contained in a paged fragment 2314 * @frag: the paged fragment buffer 2315 * 2316 * Returns the address of the data within @frag. The page must already 2317 * be mapped. 2318 */ 2319 static inline void *skb_frag_address(const skb_frag_t *frag) 2320 { 2321 return page_address(skb_frag_page(frag)) + frag->page_offset; 2322 } 2323 2324 /** 2325 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2326 * @frag: the paged fragment buffer 2327 * 2328 * Returns the address of the data within @frag. Checks that the page 2329 * is mapped and returns %NULL otherwise. 2330 */ 2331 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2332 { 2333 void *ptr = page_address(skb_frag_page(frag)); 2334 if (unlikely(!ptr)) 2335 return NULL; 2336 2337 return ptr + frag->page_offset; 2338 } 2339 2340 /** 2341 * __skb_frag_set_page - sets the page contained in a paged fragment 2342 * @frag: the paged fragment 2343 * @page: the page to set 2344 * 2345 * Sets the fragment @frag to contain @page. 2346 */ 2347 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2348 { 2349 frag->page.p = page; 2350 } 2351 2352 /** 2353 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2354 * @skb: the buffer 2355 * @f: the fragment offset 2356 * @page: the page to set 2357 * 2358 * Sets the @f'th fragment of @skb to contain @page. 2359 */ 2360 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2361 struct page *page) 2362 { 2363 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2364 } 2365 2366 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2367 2368 /** 2369 * skb_frag_dma_map - maps a paged fragment via the DMA API 2370 * @dev: the device to map the fragment to 2371 * @frag: the paged fragment to map 2372 * @offset: the offset within the fragment (starting at the 2373 * fragment's own offset) 2374 * @size: the number of bytes to map 2375 * @dir: the direction of the mapping (%PCI_DMA_*) 2376 * 2377 * Maps the page associated with @frag to @device. 2378 */ 2379 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2380 const skb_frag_t *frag, 2381 size_t offset, size_t size, 2382 enum dma_data_direction dir) 2383 { 2384 return dma_map_page(dev, skb_frag_page(frag), 2385 frag->page_offset + offset, size, dir); 2386 } 2387 2388 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2389 gfp_t gfp_mask) 2390 { 2391 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2392 } 2393 2394 2395 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2396 gfp_t gfp_mask) 2397 { 2398 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2399 } 2400 2401 2402 /** 2403 * skb_clone_writable - is the header of a clone writable 2404 * @skb: buffer to check 2405 * @len: length up to which to write 2406 * 2407 * Returns true if modifying the header part of the cloned buffer 2408 * does not requires the data to be copied. 2409 */ 2410 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2411 { 2412 return !skb_header_cloned(skb) && 2413 skb_headroom(skb) + len <= skb->hdr_len; 2414 } 2415 2416 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2417 int cloned) 2418 { 2419 int delta = 0; 2420 2421 if (headroom > skb_headroom(skb)) 2422 delta = headroom - skb_headroom(skb); 2423 2424 if (delta || cloned) 2425 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2426 GFP_ATOMIC); 2427 return 0; 2428 } 2429 2430 /** 2431 * skb_cow - copy header of skb when it is required 2432 * @skb: buffer to cow 2433 * @headroom: needed headroom 2434 * 2435 * If the skb passed lacks sufficient headroom or its data part 2436 * is shared, data is reallocated. If reallocation fails, an error 2437 * is returned and original skb is not changed. 2438 * 2439 * The result is skb with writable area skb->head...skb->tail 2440 * and at least @headroom of space at head. 2441 */ 2442 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2443 { 2444 return __skb_cow(skb, headroom, skb_cloned(skb)); 2445 } 2446 2447 /** 2448 * skb_cow_head - skb_cow but only making the head writable 2449 * @skb: buffer to cow 2450 * @headroom: needed headroom 2451 * 2452 * This function is identical to skb_cow except that we replace the 2453 * skb_cloned check by skb_header_cloned. It should be used when 2454 * you only need to push on some header and do not need to modify 2455 * the data. 2456 */ 2457 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2458 { 2459 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2460 } 2461 2462 /** 2463 * skb_padto - pad an skbuff up to a minimal size 2464 * @skb: buffer to pad 2465 * @len: minimal length 2466 * 2467 * Pads up a buffer to ensure the trailing bytes exist and are 2468 * blanked. If the buffer already contains sufficient data it 2469 * is untouched. Otherwise it is extended. Returns zero on 2470 * success. The skb is freed on error. 2471 */ 2472 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2473 { 2474 unsigned int size = skb->len; 2475 if (likely(size >= len)) 2476 return 0; 2477 return skb_pad(skb, len - size); 2478 } 2479 2480 /** 2481 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2482 * @skb: buffer to pad 2483 * @len: minimal length 2484 * 2485 * Pads up a buffer to ensure the trailing bytes exist and are 2486 * blanked. If the buffer already contains sufficient data it 2487 * is untouched. Otherwise it is extended. Returns zero on 2488 * success. The skb is freed on error. 2489 */ 2490 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2491 { 2492 unsigned int size = skb->len; 2493 2494 if (unlikely(size < len)) { 2495 len -= size; 2496 if (skb_pad(skb, len)) 2497 return -ENOMEM; 2498 __skb_put(skb, len); 2499 } 2500 return 0; 2501 } 2502 2503 static inline int skb_add_data(struct sk_buff *skb, 2504 struct iov_iter *from, int copy) 2505 { 2506 const int off = skb->len; 2507 2508 if (skb->ip_summed == CHECKSUM_NONE) { 2509 __wsum csum = 0; 2510 if (csum_and_copy_from_iter(skb_put(skb, copy), copy, 2511 &csum, from) == copy) { 2512 skb->csum = csum_block_add(skb->csum, csum, off); 2513 return 0; 2514 } 2515 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy) 2516 return 0; 2517 2518 __skb_trim(skb, off); 2519 return -EFAULT; 2520 } 2521 2522 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2523 const struct page *page, int off) 2524 { 2525 if (i) { 2526 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2527 2528 return page == skb_frag_page(frag) && 2529 off == frag->page_offset + skb_frag_size(frag); 2530 } 2531 return false; 2532 } 2533 2534 static inline int __skb_linearize(struct sk_buff *skb) 2535 { 2536 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2537 } 2538 2539 /** 2540 * skb_linearize - convert paged skb to linear one 2541 * @skb: buffer to linarize 2542 * 2543 * If there is no free memory -ENOMEM is returned, otherwise zero 2544 * is returned and the old skb data released. 2545 */ 2546 static inline int skb_linearize(struct sk_buff *skb) 2547 { 2548 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2549 } 2550 2551 /** 2552 * skb_has_shared_frag - can any frag be overwritten 2553 * @skb: buffer to test 2554 * 2555 * Return true if the skb has at least one frag that might be modified 2556 * by an external entity (as in vmsplice()/sendfile()) 2557 */ 2558 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 2559 { 2560 return skb_is_nonlinear(skb) && 2561 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2562 } 2563 2564 /** 2565 * skb_linearize_cow - make sure skb is linear and writable 2566 * @skb: buffer to process 2567 * 2568 * If there is no free memory -ENOMEM is returned, otherwise zero 2569 * is returned and the old skb data released. 2570 */ 2571 static inline int skb_linearize_cow(struct sk_buff *skb) 2572 { 2573 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2574 __skb_linearize(skb) : 0; 2575 } 2576 2577 /** 2578 * skb_postpull_rcsum - update checksum for received skb after pull 2579 * @skb: buffer to update 2580 * @start: start of data before pull 2581 * @len: length of data pulled 2582 * 2583 * After doing a pull on a received packet, you need to call this to 2584 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2585 * CHECKSUM_NONE so that it can be recomputed from scratch. 2586 */ 2587 2588 static inline void skb_postpull_rcsum(struct sk_buff *skb, 2589 const void *start, unsigned int len) 2590 { 2591 if (skb->ip_summed == CHECKSUM_COMPLETE) 2592 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0)); 2593 } 2594 2595 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2596 2597 /** 2598 * pskb_trim_rcsum - trim received skb and update checksum 2599 * @skb: buffer to trim 2600 * @len: new length 2601 * 2602 * This is exactly the same as pskb_trim except that it ensures the 2603 * checksum of received packets are still valid after the operation. 2604 */ 2605 2606 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2607 { 2608 if (likely(len >= skb->len)) 2609 return 0; 2610 if (skb->ip_summed == CHECKSUM_COMPLETE) 2611 skb->ip_summed = CHECKSUM_NONE; 2612 return __pskb_trim(skb, len); 2613 } 2614 2615 #define skb_queue_walk(queue, skb) \ 2616 for (skb = (queue)->next; \ 2617 skb != (struct sk_buff *)(queue); \ 2618 skb = skb->next) 2619 2620 #define skb_queue_walk_safe(queue, skb, tmp) \ 2621 for (skb = (queue)->next, tmp = skb->next; \ 2622 skb != (struct sk_buff *)(queue); \ 2623 skb = tmp, tmp = skb->next) 2624 2625 #define skb_queue_walk_from(queue, skb) \ 2626 for (; skb != (struct sk_buff *)(queue); \ 2627 skb = skb->next) 2628 2629 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 2630 for (tmp = skb->next; \ 2631 skb != (struct sk_buff *)(queue); \ 2632 skb = tmp, tmp = skb->next) 2633 2634 #define skb_queue_reverse_walk(queue, skb) \ 2635 for (skb = (queue)->prev; \ 2636 skb != (struct sk_buff *)(queue); \ 2637 skb = skb->prev) 2638 2639 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 2640 for (skb = (queue)->prev, tmp = skb->prev; \ 2641 skb != (struct sk_buff *)(queue); \ 2642 skb = tmp, tmp = skb->prev) 2643 2644 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 2645 for (tmp = skb->prev; \ 2646 skb != (struct sk_buff *)(queue); \ 2647 skb = tmp, tmp = skb->prev) 2648 2649 static inline bool skb_has_frag_list(const struct sk_buff *skb) 2650 { 2651 return skb_shinfo(skb)->frag_list != NULL; 2652 } 2653 2654 static inline void skb_frag_list_init(struct sk_buff *skb) 2655 { 2656 skb_shinfo(skb)->frag_list = NULL; 2657 } 2658 2659 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag) 2660 { 2661 frag->next = skb_shinfo(skb)->frag_list; 2662 skb_shinfo(skb)->frag_list = frag; 2663 } 2664 2665 #define skb_walk_frags(skb, iter) \ 2666 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 2667 2668 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 2669 int *peeked, int *off, int *err); 2670 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 2671 int *err); 2672 unsigned int datagram_poll(struct file *file, struct socket *sock, 2673 struct poll_table_struct *wait); 2674 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 2675 struct iov_iter *to, int size); 2676 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 2677 struct msghdr *msg, int size) 2678 { 2679 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 2680 } 2681 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 2682 struct msghdr *msg); 2683 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 2684 struct iov_iter *from, int len); 2685 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 2686 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 2687 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb); 2688 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 2689 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 2690 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 2691 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 2692 int len, __wsum csum); 2693 int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 2694 struct pipe_inode_info *pipe, unsigned int len, 2695 unsigned int flags); 2696 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 2697 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 2698 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 2699 int len, int hlen); 2700 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 2701 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 2702 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 2703 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 2704 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 2705 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 2706 int skb_ensure_writable(struct sk_buff *skb, int write_len); 2707 int skb_vlan_pop(struct sk_buff *skb); 2708 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 2709 2710 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 2711 { 2712 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2713 } 2714 2715 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 2716 { 2717 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2718 } 2719 2720 struct skb_checksum_ops { 2721 __wsum (*update)(const void *mem, int len, __wsum wsum); 2722 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 2723 }; 2724 2725 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 2726 __wsum csum, const struct skb_checksum_ops *ops); 2727 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 2728 __wsum csum); 2729 2730 static inline void *__skb_header_pointer(const struct sk_buff *skb, int offset, 2731 int len, void *data, int hlen, void *buffer) 2732 { 2733 if (hlen - offset >= len) 2734 return data + offset; 2735 2736 if (!skb || 2737 skb_copy_bits(skb, offset, buffer, len) < 0) 2738 return NULL; 2739 2740 return buffer; 2741 } 2742 2743 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset, 2744 int len, void *buffer) 2745 { 2746 return __skb_header_pointer(skb, offset, len, skb->data, 2747 skb_headlen(skb), buffer); 2748 } 2749 2750 /** 2751 * skb_needs_linearize - check if we need to linearize a given skb 2752 * depending on the given device features. 2753 * @skb: socket buffer to check 2754 * @features: net device features 2755 * 2756 * Returns true if either: 2757 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 2758 * 2. skb is fragmented and the device does not support SG. 2759 */ 2760 static inline bool skb_needs_linearize(struct sk_buff *skb, 2761 netdev_features_t features) 2762 { 2763 return skb_is_nonlinear(skb) && 2764 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 2765 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 2766 } 2767 2768 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 2769 void *to, 2770 const unsigned int len) 2771 { 2772 memcpy(to, skb->data, len); 2773 } 2774 2775 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 2776 const int offset, void *to, 2777 const unsigned int len) 2778 { 2779 memcpy(to, skb->data + offset, len); 2780 } 2781 2782 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 2783 const void *from, 2784 const unsigned int len) 2785 { 2786 memcpy(skb->data, from, len); 2787 } 2788 2789 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 2790 const int offset, 2791 const void *from, 2792 const unsigned int len) 2793 { 2794 memcpy(skb->data + offset, from, len); 2795 } 2796 2797 void skb_init(void); 2798 2799 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 2800 { 2801 return skb->tstamp; 2802 } 2803 2804 /** 2805 * skb_get_timestamp - get timestamp from a skb 2806 * @skb: skb to get stamp from 2807 * @stamp: pointer to struct timeval to store stamp in 2808 * 2809 * Timestamps are stored in the skb as offsets to a base timestamp. 2810 * This function converts the offset back to a struct timeval and stores 2811 * it in stamp. 2812 */ 2813 static inline void skb_get_timestamp(const struct sk_buff *skb, 2814 struct timeval *stamp) 2815 { 2816 *stamp = ktime_to_timeval(skb->tstamp); 2817 } 2818 2819 static inline void skb_get_timestampns(const struct sk_buff *skb, 2820 struct timespec *stamp) 2821 { 2822 *stamp = ktime_to_timespec(skb->tstamp); 2823 } 2824 2825 static inline void __net_timestamp(struct sk_buff *skb) 2826 { 2827 skb->tstamp = ktime_get_real(); 2828 } 2829 2830 static inline ktime_t net_timedelta(ktime_t t) 2831 { 2832 return ktime_sub(ktime_get_real(), t); 2833 } 2834 2835 static inline ktime_t net_invalid_timestamp(void) 2836 { 2837 return ktime_set(0, 0); 2838 } 2839 2840 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 2841 2842 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 2843 2844 void skb_clone_tx_timestamp(struct sk_buff *skb); 2845 bool skb_defer_rx_timestamp(struct sk_buff *skb); 2846 2847 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 2848 2849 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 2850 { 2851 } 2852 2853 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 2854 { 2855 return false; 2856 } 2857 2858 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 2859 2860 /** 2861 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 2862 * 2863 * PHY drivers may accept clones of transmitted packets for 2864 * timestamping via their phy_driver.txtstamp method. These drivers 2865 * must call this function to return the skb back to the stack, with 2866 * or without a timestamp. 2867 * 2868 * @skb: clone of the the original outgoing packet 2869 * @hwtstamps: hardware time stamps, may be NULL if not available 2870 * 2871 */ 2872 void skb_complete_tx_timestamp(struct sk_buff *skb, 2873 struct skb_shared_hwtstamps *hwtstamps); 2874 2875 void __skb_tstamp_tx(struct sk_buff *orig_skb, 2876 struct skb_shared_hwtstamps *hwtstamps, 2877 struct sock *sk, int tstype); 2878 2879 /** 2880 * skb_tstamp_tx - queue clone of skb with send time stamps 2881 * @orig_skb: the original outgoing packet 2882 * @hwtstamps: hardware time stamps, may be NULL if not available 2883 * 2884 * If the skb has a socket associated, then this function clones the 2885 * skb (thus sharing the actual data and optional structures), stores 2886 * the optional hardware time stamping information (if non NULL) or 2887 * generates a software time stamp (otherwise), then queues the clone 2888 * to the error queue of the socket. Errors are silently ignored. 2889 */ 2890 void skb_tstamp_tx(struct sk_buff *orig_skb, 2891 struct skb_shared_hwtstamps *hwtstamps); 2892 2893 static inline void sw_tx_timestamp(struct sk_buff *skb) 2894 { 2895 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 2896 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 2897 skb_tstamp_tx(skb, NULL); 2898 } 2899 2900 /** 2901 * skb_tx_timestamp() - Driver hook for transmit timestamping 2902 * 2903 * Ethernet MAC Drivers should call this function in their hard_xmit() 2904 * function immediately before giving the sk_buff to the MAC hardware. 2905 * 2906 * Specifically, one should make absolutely sure that this function is 2907 * called before TX completion of this packet can trigger. Otherwise 2908 * the packet could potentially already be freed. 2909 * 2910 * @skb: A socket buffer. 2911 */ 2912 static inline void skb_tx_timestamp(struct sk_buff *skb) 2913 { 2914 skb_clone_tx_timestamp(skb); 2915 sw_tx_timestamp(skb); 2916 } 2917 2918 /** 2919 * skb_complete_wifi_ack - deliver skb with wifi status 2920 * 2921 * @skb: the original outgoing packet 2922 * @acked: ack status 2923 * 2924 */ 2925 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 2926 2927 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 2928 __sum16 __skb_checksum_complete(struct sk_buff *skb); 2929 2930 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 2931 { 2932 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 2933 skb->csum_valid || 2934 (skb->ip_summed == CHECKSUM_PARTIAL && 2935 skb_checksum_start_offset(skb) >= 0)); 2936 } 2937 2938 /** 2939 * skb_checksum_complete - Calculate checksum of an entire packet 2940 * @skb: packet to process 2941 * 2942 * This function calculates the checksum over the entire packet plus 2943 * the value of skb->csum. The latter can be used to supply the 2944 * checksum of a pseudo header as used by TCP/UDP. It returns the 2945 * checksum. 2946 * 2947 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 2948 * this function can be used to verify that checksum on received 2949 * packets. In that case the function should return zero if the 2950 * checksum is correct. In particular, this function will return zero 2951 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 2952 * hardware has already verified the correctness of the checksum. 2953 */ 2954 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 2955 { 2956 return skb_csum_unnecessary(skb) ? 2957 0 : __skb_checksum_complete(skb); 2958 } 2959 2960 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 2961 { 2962 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 2963 if (skb->csum_level == 0) 2964 skb->ip_summed = CHECKSUM_NONE; 2965 else 2966 skb->csum_level--; 2967 } 2968 } 2969 2970 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 2971 { 2972 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 2973 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 2974 skb->csum_level++; 2975 } else if (skb->ip_summed == CHECKSUM_NONE) { 2976 skb->ip_summed = CHECKSUM_UNNECESSARY; 2977 skb->csum_level = 0; 2978 } 2979 } 2980 2981 static inline void __skb_mark_checksum_bad(struct sk_buff *skb) 2982 { 2983 /* Mark current checksum as bad (typically called from GRO 2984 * path). In the case that ip_summed is CHECKSUM_NONE 2985 * this must be the first checksum encountered in the packet. 2986 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first 2987 * checksum after the last one validated. For UDP, a zero 2988 * checksum can not be marked as bad. 2989 */ 2990 2991 if (skb->ip_summed == CHECKSUM_NONE || 2992 skb->ip_summed == CHECKSUM_UNNECESSARY) 2993 skb->csum_bad = 1; 2994 } 2995 2996 /* Check if we need to perform checksum complete validation. 2997 * 2998 * Returns true if checksum complete is needed, false otherwise 2999 * (either checksum is unnecessary or zero checksum is allowed). 3000 */ 3001 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3002 bool zero_okay, 3003 __sum16 check) 3004 { 3005 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3006 skb->csum_valid = 1; 3007 __skb_decr_checksum_unnecessary(skb); 3008 return false; 3009 } 3010 3011 return true; 3012 } 3013 3014 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3015 * in checksum_init. 3016 */ 3017 #define CHECKSUM_BREAK 76 3018 3019 /* Unset checksum-complete 3020 * 3021 * Unset checksum complete can be done when packet is being modified 3022 * (uncompressed for instance) and checksum-complete value is 3023 * invalidated. 3024 */ 3025 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3026 { 3027 if (skb->ip_summed == CHECKSUM_COMPLETE) 3028 skb->ip_summed = CHECKSUM_NONE; 3029 } 3030 3031 /* Validate (init) checksum based on checksum complete. 3032 * 3033 * Return values: 3034 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3035 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3036 * checksum is stored in skb->csum for use in __skb_checksum_complete 3037 * non-zero: value of invalid checksum 3038 * 3039 */ 3040 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3041 bool complete, 3042 __wsum psum) 3043 { 3044 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3045 if (!csum_fold(csum_add(psum, skb->csum))) { 3046 skb->csum_valid = 1; 3047 return 0; 3048 } 3049 } else if (skb->csum_bad) { 3050 /* ip_summed == CHECKSUM_NONE in this case */ 3051 return 1; 3052 } 3053 3054 skb->csum = psum; 3055 3056 if (complete || skb->len <= CHECKSUM_BREAK) { 3057 __sum16 csum; 3058 3059 csum = __skb_checksum_complete(skb); 3060 skb->csum_valid = !csum; 3061 return csum; 3062 } 3063 3064 return 0; 3065 } 3066 3067 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3068 { 3069 return 0; 3070 } 3071 3072 /* Perform checksum validate (init). Note that this is a macro since we only 3073 * want to calculate the pseudo header which is an input function if necessary. 3074 * First we try to validate without any computation (checksum unnecessary) and 3075 * then calculate based on checksum complete calling the function to compute 3076 * pseudo header. 3077 * 3078 * Return values: 3079 * 0: checksum is validated or try to in skb_checksum_complete 3080 * non-zero: value of invalid checksum 3081 */ 3082 #define __skb_checksum_validate(skb, proto, complete, \ 3083 zero_okay, check, compute_pseudo) \ 3084 ({ \ 3085 __sum16 __ret = 0; \ 3086 skb->csum_valid = 0; \ 3087 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3088 __ret = __skb_checksum_validate_complete(skb, \ 3089 complete, compute_pseudo(skb, proto)); \ 3090 __ret; \ 3091 }) 3092 3093 #define skb_checksum_init(skb, proto, compute_pseudo) \ 3094 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3095 3096 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3097 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3098 3099 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 3100 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3101 3102 #define skb_checksum_validate_zero_check(skb, proto, check, \ 3103 compute_pseudo) \ 3104 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3105 3106 #define skb_checksum_simple_validate(skb) \ 3107 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3108 3109 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3110 { 3111 return (skb->ip_summed == CHECKSUM_NONE && 3112 skb->csum_valid && !skb->csum_bad); 3113 } 3114 3115 static inline void __skb_checksum_convert(struct sk_buff *skb, 3116 __sum16 check, __wsum pseudo) 3117 { 3118 skb->csum = ~pseudo; 3119 skb->ip_summed = CHECKSUM_COMPLETE; 3120 } 3121 3122 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3123 do { \ 3124 if (__skb_checksum_convert_check(skb)) \ 3125 __skb_checksum_convert(skb, check, \ 3126 compute_pseudo(skb, proto)); \ 3127 } while (0) 3128 3129 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3130 u16 start, u16 offset) 3131 { 3132 skb->ip_summed = CHECKSUM_PARTIAL; 3133 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3134 skb->csum_offset = offset - start; 3135 } 3136 3137 /* Update skbuf and packet to reflect the remote checksum offload operation. 3138 * When called, ptr indicates the starting point for skb->csum when 3139 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3140 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3141 */ 3142 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3143 int start, int offset, bool nopartial) 3144 { 3145 __wsum delta; 3146 3147 if (!nopartial) { 3148 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3149 return; 3150 } 3151 3152 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3153 __skb_checksum_complete(skb); 3154 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3155 } 3156 3157 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3158 3159 /* Adjust skb->csum since we changed the packet */ 3160 skb->csum = csum_add(skb->csum, delta); 3161 } 3162 3163 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3164 void nf_conntrack_destroy(struct nf_conntrack *nfct); 3165 static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3166 { 3167 if (nfct && atomic_dec_and_test(&nfct->use)) 3168 nf_conntrack_destroy(nfct); 3169 } 3170 static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3171 { 3172 if (nfct) 3173 atomic_inc(&nfct->use); 3174 } 3175 #endif 3176 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3177 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3178 { 3179 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 3180 kfree(nf_bridge); 3181 } 3182 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3183 { 3184 if (nf_bridge) 3185 atomic_inc(&nf_bridge->use); 3186 } 3187 #endif /* CONFIG_BRIDGE_NETFILTER */ 3188 static inline void nf_reset(struct sk_buff *skb) 3189 { 3190 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3191 nf_conntrack_put(skb->nfct); 3192 skb->nfct = NULL; 3193 #endif 3194 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3195 nf_bridge_put(skb->nf_bridge); 3196 skb->nf_bridge = NULL; 3197 #endif 3198 } 3199 3200 static inline void nf_reset_trace(struct sk_buff *skb) 3201 { 3202 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3203 skb->nf_trace = 0; 3204 #endif 3205 } 3206 3207 /* Note: This doesn't put any conntrack and bridge info in dst. */ 3208 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3209 bool copy) 3210 { 3211 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3212 dst->nfct = src->nfct; 3213 nf_conntrack_get(src->nfct); 3214 if (copy) 3215 dst->nfctinfo = src->nfctinfo; 3216 #endif 3217 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3218 dst->nf_bridge = src->nf_bridge; 3219 nf_bridge_get(src->nf_bridge); 3220 #endif 3221 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3222 if (copy) 3223 dst->nf_trace = src->nf_trace; 3224 #endif 3225 } 3226 3227 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3228 { 3229 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3230 nf_conntrack_put(dst->nfct); 3231 #endif 3232 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3233 nf_bridge_put(dst->nf_bridge); 3234 #endif 3235 __nf_copy(dst, src, true); 3236 } 3237 3238 #ifdef CONFIG_NETWORK_SECMARK 3239 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3240 { 3241 to->secmark = from->secmark; 3242 } 3243 3244 static inline void skb_init_secmark(struct sk_buff *skb) 3245 { 3246 skb->secmark = 0; 3247 } 3248 #else 3249 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3250 { } 3251 3252 static inline void skb_init_secmark(struct sk_buff *skb) 3253 { } 3254 #endif 3255 3256 static inline bool skb_irq_freeable(const struct sk_buff *skb) 3257 { 3258 return !skb->destructor && 3259 #if IS_ENABLED(CONFIG_XFRM) 3260 !skb->sp && 3261 #endif 3262 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 3263 !skb->nfct && 3264 #endif 3265 !skb->_skb_refdst && 3266 !skb_has_frag_list(skb); 3267 } 3268 3269 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3270 { 3271 skb->queue_mapping = queue_mapping; 3272 } 3273 3274 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3275 { 3276 return skb->queue_mapping; 3277 } 3278 3279 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3280 { 3281 to->queue_mapping = from->queue_mapping; 3282 } 3283 3284 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3285 { 3286 skb->queue_mapping = rx_queue + 1; 3287 } 3288 3289 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3290 { 3291 return skb->queue_mapping - 1; 3292 } 3293 3294 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3295 { 3296 return skb->queue_mapping != 0; 3297 } 3298 3299 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, 3300 unsigned int num_tx_queues); 3301 3302 static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3303 { 3304 #ifdef CONFIG_XFRM 3305 return skb->sp; 3306 #else 3307 return NULL; 3308 #endif 3309 } 3310 3311 /* Keeps track of mac header offset relative to skb->head. 3312 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3313 * For non-tunnel skb it points to skb_mac_header() and for 3314 * tunnel skb it points to outer mac header. 3315 * Keeps track of level of encapsulation of network headers. 3316 */ 3317 struct skb_gso_cb { 3318 int mac_offset; 3319 int encap_level; 3320 __u16 csum_start; 3321 }; 3322 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb) 3323 3324 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3325 { 3326 return (skb_mac_header(inner_skb) - inner_skb->head) - 3327 SKB_GSO_CB(inner_skb)->mac_offset; 3328 } 3329 3330 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3331 { 3332 int new_headroom, headroom; 3333 int ret; 3334 3335 headroom = skb_headroom(skb); 3336 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3337 if (ret) 3338 return ret; 3339 3340 new_headroom = skb_headroom(skb); 3341 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3342 return 0; 3343 } 3344 3345 /* Compute the checksum for a gso segment. First compute the checksum value 3346 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3347 * then add in skb->csum (checksum from csum_start to end of packet). 3348 * skb->csum and csum_start are then updated to reflect the checksum of the 3349 * resultant packet starting from the transport header-- the resultant checksum 3350 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3351 * header. 3352 */ 3353 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3354 { 3355 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) - 3356 skb_transport_offset(skb); 3357 __u16 csum; 3358 3359 csum = csum_fold(csum_partial(skb_transport_header(skb), 3360 plen, skb->csum)); 3361 skb->csum = res; 3362 SKB_GSO_CB(skb)->csum_start -= plen; 3363 3364 return csum; 3365 } 3366 3367 static inline bool skb_is_gso(const struct sk_buff *skb) 3368 { 3369 return skb_shinfo(skb)->gso_size; 3370 } 3371 3372 /* Note: Should be called only if skb_is_gso(skb) is true */ 3373 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3374 { 3375 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3376 } 3377 3378 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3379 3380 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3381 { 3382 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3383 * wanted then gso_type will be set. */ 3384 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3385 3386 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3387 unlikely(shinfo->gso_type == 0)) { 3388 __skb_warn_lro_forwarding(skb); 3389 return true; 3390 } 3391 return false; 3392 } 3393 3394 static inline void skb_forward_csum(struct sk_buff *skb) 3395 { 3396 /* Unfortunately we don't support this one. Any brave souls? */ 3397 if (skb->ip_summed == CHECKSUM_COMPLETE) 3398 skb->ip_summed = CHECKSUM_NONE; 3399 } 3400 3401 /** 3402 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3403 * @skb: skb to check 3404 * 3405 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3406 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3407 * use this helper, to document places where we make this assertion. 3408 */ 3409 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3410 { 3411 #ifdef DEBUG 3412 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3413 #endif 3414 } 3415 3416 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3417 3418 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3419 3420 u32 skb_get_poff(const struct sk_buff *skb); 3421 u32 __skb_get_poff(const struct sk_buff *skb, void *data, 3422 const struct flow_keys *keys, int hlen); 3423 3424 /** 3425 * skb_head_is_locked - Determine if the skb->head is locked down 3426 * @skb: skb to check 3427 * 3428 * The head on skbs build around a head frag can be removed if they are 3429 * not cloned. This function returns true if the skb head is locked down 3430 * due to either being allocated via kmalloc, or by being a clone with 3431 * multiple references to the head. 3432 */ 3433 static inline bool skb_head_is_locked(const struct sk_buff *skb) 3434 { 3435 return !skb->head_frag || skb_cloned(skb); 3436 } 3437 3438 /** 3439 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3440 * 3441 * @skb: GSO skb 3442 * 3443 * skb_gso_network_seglen is used to determine the real size of the 3444 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3445 * 3446 * The MAC/L2 header is not accounted for. 3447 */ 3448 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 3449 { 3450 unsigned int hdr_len = skb_transport_header(skb) - 3451 skb_network_header(skb); 3452 return hdr_len + skb_gso_transport_seglen(skb); 3453 } 3454 #endif /* __KERNEL__ */ 3455 #endif /* _LINUX_SKBUFF_H */ 3456