1 /* 2 * Definitions for the 'struct sk_buff' memory handlers. 3 * 4 * Authors: 5 * Alan Cox, <gw4pts@gw4pts.ampr.org> 6 * Florian La Roche, <rzsfl@rz.uni-sb.de> 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 __be32 ipv4_daddr; 180 }; 181 #endif 182 183 struct sk_buff_head { 184 /* These two members must be first. */ 185 struct sk_buff *next; 186 struct sk_buff *prev; 187 188 __u32 qlen; 189 spinlock_t lock; 190 }; 191 192 struct sk_buff; 193 194 /* To allow 64K frame to be packed as single skb without frag_list we 195 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 196 * buffers which do not start on a page boundary. 197 * 198 * Since GRO uses frags we allocate at least 16 regardless of page 199 * size. 200 */ 201 #if (65536/PAGE_SIZE + 1) < 16 202 #define MAX_SKB_FRAGS 16UL 203 #else 204 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 205 #endif 206 extern int sysctl_max_skb_frags; 207 208 typedef struct skb_frag_struct skb_frag_t; 209 210 struct skb_frag_struct { 211 struct { 212 struct page *p; 213 } page; 214 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 215 __u32 page_offset; 216 __u32 size; 217 #else 218 __u16 page_offset; 219 __u16 size; 220 #endif 221 }; 222 223 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 224 { 225 return frag->size; 226 } 227 228 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 229 { 230 frag->size = size; 231 } 232 233 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 234 { 235 frag->size += delta; 236 } 237 238 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 239 { 240 frag->size -= delta; 241 } 242 243 #define HAVE_HW_TIME_STAMP 244 245 /** 246 * struct skb_shared_hwtstamps - hardware time stamps 247 * @hwtstamp: hardware time stamp transformed into duration 248 * since arbitrary point in time 249 * 250 * Software time stamps generated by ktime_get_real() are stored in 251 * skb->tstamp. 252 * 253 * hwtstamps can only be compared against other hwtstamps from 254 * the same device. 255 * 256 * This structure is attached to packets as part of the 257 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 258 */ 259 struct skb_shared_hwtstamps { 260 ktime_t hwtstamp; 261 }; 262 263 /* Definitions for tx_flags in struct skb_shared_info */ 264 enum { 265 /* generate hardware time stamp */ 266 SKBTX_HW_TSTAMP = 1 << 0, 267 268 /* generate software time stamp when queueing packet to NIC */ 269 SKBTX_SW_TSTAMP = 1 << 1, 270 271 /* device driver is going to provide hardware time stamp */ 272 SKBTX_IN_PROGRESS = 1 << 2, 273 274 /* device driver supports TX zero-copy buffers */ 275 SKBTX_DEV_ZEROCOPY = 1 << 3, 276 277 /* generate wifi status information (where possible) */ 278 SKBTX_WIFI_STATUS = 1 << 4, 279 280 /* This indicates at least one fragment might be overwritten 281 * (as in vmsplice(), sendfile() ...) 282 * If we need to compute a TX checksum, we'll need to copy 283 * all frags to avoid possible bad checksum 284 */ 285 SKBTX_SHARED_FRAG = 1 << 5, 286 287 /* generate software time stamp when entering packet scheduling */ 288 SKBTX_SCHED_TSTAMP = 1 << 6, 289 290 /* generate software timestamp on peer data acknowledgment */ 291 SKBTX_ACK_TSTAMP = 1 << 7, 292 }; 293 294 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 295 SKBTX_SCHED_TSTAMP | \ 296 SKBTX_ACK_TSTAMP) 297 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 298 299 /* 300 * The callback notifies userspace to release buffers when skb DMA is done in 301 * lower device, the skb last reference should be 0 when calling this. 302 * The zerocopy_success argument is true if zero copy transmit occurred, 303 * false on data copy or out of memory error caused by data copy attempt. 304 * The ctx field is used to track device context. 305 * The desc field is used to track userspace buffer index. 306 */ 307 struct ubuf_info { 308 void (*callback)(struct ubuf_info *, bool zerocopy_success); 309 void *ctx; 310 unsigned long desc; 311 }; 312 313 /* This data is invariant across clones and lives at 314 * the end of the header data, ie. at skb->end. 315 */ 316 struct skb_shared_info { 317 unsigned char nr_frags; 318 __u8 tx_flags; 319 unsigned short gso_size; 320 /* Warning: this field is not always filled in (UFO)! */ 321 unsigned short gso_segs; 322 unsigned short gso_type; 323 struct sk_buff *frag_list; 324 struct skb_shared_hwtstamps hwtstamps; 325 u32 tskey; 326 __be32 ip6_frag_id; 327 328 /* 329 * Warning : all fields before dataref are cleared in __alloc_skb() 330 */ 331 atomic_t dataref; 332 333 /* Intermediate layers must ensure that destructor_arg 334 * remains valid until skb destructor */ 335 void * destructor_arg; 336 337 /* must be last field, see pskb_expand_head() */ 338 skb_frag_t frags[MAX_SKB_FRAGS]; 339 }; 340 341 /* We divide dataref into two halves. The higher 16 bits hold references 342 * to the payload part of skb->data. The lower 16 bits hold references to 343 * the entire skb->data. A clone of a headerless skb holds the length of 344 * the header in skb->hdr_len. 345 * 346 * All users must obey the rule that the skb->data reference count must be 347 * greater than or equal to the payload reference count. 348 * 349 * Holding a reference to the payload part means that the user does not 350 * care about modifications to the header part of skb->data. 351 */ 352 #define SKB_DATAREF_SHIFT 16 353 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 354 355 356 enum { 357 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 358 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 359 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 360 }; 361 362 enum { 363 SKB_GSO_TCPV4 = 1 << 0, 364 SKB_GSO_UDP = 1 << 1, 365 366 /* This indicates the skb is from an untrusted source. */ 367 SKB_GSO_DODGY = 1 << 2, 368 369 /* This indicates the tcp segment has CWR set. */ 370 SKB_GSO_TCP_ECN = 1 << 3, 371 372 SKB_GSO_TCPV6 = 1 << 4, 373 374 SKB_GSO_FCOE = 1 << 5, 375 376 SKB_GSO_GRE = 1 << 6, 377 378 SKB_GSO_GRE_CSUM = 1 << 7, 379 380 SKB_GSO_IPIP = 1 << 8, 381 382 SKB_GSO_SIT = 1 << 9, 383 384 SKB_GSO_UDP_TUNNEL = 1 << 10, 385 386 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 387 388 SKB_GSO_TUNNEL_REMCSUM = 1 << 12, 389 }; 390 391 #if BITS_PER_LONG > 32 392 #define NET_SKBUFF_DATA_USES_OFFSET 1 393 #endif 394 395 #ifdef NET_SKBUFF_DATA_USES_OFFSET 396 typedef unsigned int sk_buff_data_t; 397 #else 398 typedef unsigned char *sk_buff_data_t; 399 #endif 400 401 /** 402 * struct skb_mstamp - multi resolution time stamps 403 * @stamp_us: timestamp in us resolution 404 * @stamp_jiffies: timestamp in jiffies 405 */ 406 struct skb_mstamp { 407 union { 408 u64 v64; 409 struct { 410 u32 stamp_us; 411 u32 stamp_jiffies; 412 }; 413 }; 414 }; 415 416 /** 417 * skb_mstamp_get - get current timestamp 418 * @cl: place to store timestamps 419 */ 420 static inline void skb_mstamp_get(struct skb_mstamp *cl) 421 { 422 u64 val = local_clock(); 423 424 do_div(val, NSEC_PER_USEC); 425 cl->stamp_us = (u32)val; 426 cl->stamp_jiffies = (u32)jiffies; 427 } 428 429 /** 430 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp 431 * @t1: pointer to newest sample 432 * @t0: pointer to oldest sample 433 */ 434 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1, 435 const struct skb_mstamp *t0) 436 { 437 s32 delta_us = t1->stamp_us - t0->stamp_us; 438 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies; 439 440 /* If delta_us is negative, this might be because interval is too big, 441 * or local_clock() drift is too big : fallback using jiffies. 442 */ 443 if (delta_us <= 0 || 444 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ))) 445 446 delta_us = jiffies_to_usecs(delta_jiffies); 447 448 return delta_us; 449 } 450 451 452 /** 453 * struct sk_buff - socket buffer 454 * @next: Next buffer in list 455 * @prev: Previous buffer in list 456 * @tstamp: Time we arrived/left 457 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 458 * @sk: Socket we are owned by 459 * @dev: Device we arrived on/are leaving by 460 * @cb: Control buffer. Free for use by every layer. Put private vars here 461 * @_skb_refdst: destination entry (with norefcount bit) 462 * @sp: the security path, used for xfrm 463 * @len: Length of actual data 464 * @data_len: Data length 465 * @mac_len: Length of link layer header 466 * @hdr_len: writable header length of cloned skb 467 * @csum: Checksum (must include start/offset pair) 468 * @csum_start: Offset from skb->head where checksumming should start 469 * @csum_offset: Offset from csum_start where checksum should be stored 470 * @priority: Packet queueing priority 471 * @ignore_df: allow local fragmentation 472 * @cloned: Head may be cloned (check refcnt to be sure) 473 * @ip_summed: Driver fed us an IP checksum 474 * @nohdr: Payload reference only, must not modify header 475 * @nfctinfo: Relationship of this skb to the connection 476 * @pkt_type: Packet class 477 * @fclone: skbuff clone status 478 * @ipvs_property: skbuff is owned by ipvs 479 * @peeked: this packet has been seen already, so stats have been 480 * done for it, don't do them again 481 * @nf_trace: netfilter packet trace flag 482 * @protocol: Packet protocol from driver 483 * @destructor: Destruct function 484 * @nfct: Associated connection, if any 485 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 486 * @skb_iif: ifindex of device we arrived on 487 * @tc_index: Traffic control index 488 * @tc_verd: traffic control verdict 489 * @hash: the packet hash 490 * @queue_mapping: Queue mapping for multiqueue devices 491 * @xmit_more: More SKBs are pending for this queue 492 * @ndisc_nodetype: router type (from link layer) 493 * @ooo_okay: allow the mapping of a socket to a queue to be changed 494 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 495 * ports. 496 * @sw_hash: indicates hash was computed in software stack 497 * @wifi_acked_valid: wifi_acked was set 498 * @wifi_acked: whether frame was acked on wifi or not 499 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 500 * @napi_id: id of the NAPI struct this skb came from 501 * @secmark: security marking 502 * @mark: Generic packet mark 503 * @vlan_proto: vlan encapsulation protocol 504 * @vlan_tci: vlan tag control information 505 * @inner_protocol: Protocol (encapsulation) 506 * @inner_transport_header: Inner transport layer header (encapsulation) 507 * @inner_network_header: Network layer header (encapsulation) 508 * @inner_mac_header: Link layer header (encapsulation) 509 * @transport_header: Transport layer header 510 * @network_header: Network layer header 511 * @mac_header: Link layer header 512 * @tail: Tail pointer 513 * @end: End pointer 514 * @head: Head of buffer 515 * @data: Data head pointer 516 * @truesize: Buffer size 517 * @users: User count - see {datagram,tcp}.c 518 */ 519 520 struct sk_buff { 521 union { 522 struct { 523 /* These two members must be first. */ 524 struct sk_buff *next; 525 struct sk_buff *prev; 526 527 union { 528 ktime_t tstamp; 529 struct skb_mstamp skb_mstamp; 530 }; 531 }; 532 struct rb_node rbnode; /* used in netem & tcp stack */ 533 }; 534 struct sock *sk; 535 struct net_device *dev; 536 537 /* 538 * This is the control buffer. It is free to use for every 539 * layer. Please put your private variables there. If you 540 * want to keep them across layers you have to do a skb_clone() 541 * first. This is owned by whoever has the skb queued ATM. 542 */ 543 char cb[48] __aligned(8); 544 545 unsigned long _skb_refdst; 546 void (*destructor)(struct sk_buff *skb); 547 #ifdef CONFIG_XFRM 548 struct sec_path *sp; 549 #endif 550 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 551 struct nf_conntrack *nfct; 552 #endif 553 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 554 struct nf_bridge_info *nf_bridge; 555 #endif 556 unsigned int len, 557 data_len; 558 __u16 mac_len, 559 hdr_len; 560 561 /* Following fields are _not_ copied in __copy_skb_header() 562 * Note that queue_mapping is here mostly to fill a hole. 563 */ 564 kmemcheck_bitfield_begin(flags1); 565 __u16 queue_mapping; 566 __u8 cloned:1, 567 nohdr:1, 568 fclone:2, 569 peeked:1, 570 head_frag:1, 571 xmit_more:1; 572 /* one bit hole */ 573 kmemcheck_bitfield_end(flags1); 574 575 /* fields enclosed in headers_start/headers_end are copied 576 * using a single memcpy() in __copy_skb_header() 577 */ 578 /* private: */ 579 __u32 headers_start[0]; 580 /* public: */ 581 582 /* if you move pkt_type around you also must adapt those constants */ 583 #ifdef __BIG_ENDIAN_BITFIELD 584 #define PKT_TYPE_MAX (7 << 5) 585 #else 586 #define PKT_TYPE_MAX 7 587 #endif 588 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 589 590 __u8 __pkt_type_offset[0]; 591 __u8 pkt_type:3; 592 __u8 pfmemalloc:1; 593 __u8 ignore_df:1; 594 __u8 nfctinfo:3; 595 596 __u8 nf_trace:1; 597 __u8 ip_summed:2; 598 __u8 ooo_okay:1; 599 __u8 l4_hash:1; 600 __u8 sw_hash:1; 601 __u8 wifi_acked_valid:1; 602 __u8 wifi_acked:1; 603 604 __u8 no_fcs:1; 605 /* Indicates the inner headers are valid in the skbuff. */ 606 __u8 encapsulation:1; 607 __u8 encap_hdr_csum:1; 608 __u8 csum_valid:1; 609 __u8 csum_complete_sw:1; 610 __u8 csum_level:2; 611 __u8 csum_bad:1; 612 613 #ifdef CONFIG_IPV6_NDISC_NODETYPE 614 __u8 ndisc_nodetype:2; 615 #endif 616 __u8 ipvs_property:1; 617 __u8 inner_protocol_type:1; 618 __u8 remcsum_offload:1; 619 /* 3 or 5 bit hole */ 620 621 #ifdef CONFIG_NET_SCHED 622 __u16 tc_index; /* traffic control index */ 623 #ifdef CONFIG_NET_CLS_ACT 624 __u16 tc_verd; /* traffic control verdict */ 625 #endif 626 #endif 627 628 union { 629 __wsum csum; 630 struct { 631 __u16 csum_start; 632 __u16 csum_offset; 633 }; 634 }; 635 __u32 priority; 636 int skb_iif; 637 __u32 hash; 638 __be16 vlan_proto; 639 __u16 vlan_tci; 640 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 641 union { 642 unsigned int napi_id; 643 unsigned int sender_cpu; 644 }; 645 #endif 646 #ifdef CONFIG_NETWORK_SECMARK 647 __u32 secmark; 648 #endif 649 union { 650 __u32 mark; 651 __u32 reserved_tailroom; 652 }; 653 654 union { 655 __be16 inner_protocol; 656 __u8 inner_ipproto; 657 }; 658 659 __u16 inner_transport_header; 660 __u16 inner_network_header; 661 __u16 inner_mac_header; 662 663 __be16 protocol; 664 __u16 transport_header; 665 __u16 network_header; 666 __u16 mac_header; 667 668 /* private: */ 669 __u32 headers_end[0]; 670 /* public: */ 671 672 /* These elements must be at the end, see alloc_skb() for details. */ 673 sk_buff_data_t tail; 674 sk_buff_data_t end; 675 unsigned char *head, 676 *data; 677 unsigned int truesize; 678 atomic_t users; 679 }; 680 681 #ifdef __KERNEL__ 682 /* 683 * Handling routines are only of interest to the kernel 684 */ 685 #include <linux/slab.h> 686 687 688 #define SKB_ALLOC_FCLONE 0x01 689 #define SKB_ALLOC_RX 0x02 690 #define SKB_ALLOC_NAPI 0x04 691 692 /* Returns true if the skb was allocated from PFMEMALLOC reserves */ 693 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 694 { 695 return unlikely(skb->pfmemalloc); 696 } 697 698 /* 699 * skb might have a dst pointer attached, refcounted or not. 700 * _skb_refdst low order bit is set if refcount was _not_ taken 701 */ 702 #define SKB_DST_NOREF 1UL 703 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 704 705 /** 706 * skb_dst - returns skb dst_entry 707 * @skb: buffer 708 * 709 * Returns skb dst_entry, regardless of reference taken or not. 710 */ 711 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 712 { 713 /* If refdst was not refcounted, check we still are in a 714 * rcu_read_lock section 715 */ 716 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 717 !rcu_read_lock_held() && 718 !rcu_read_lock_bh_held()); 719 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 720 } 721 722 /** 723 * skb_dst_set - sets skb dst 724 * @skb: buffer 725 * @dst: dst entry 726 * 727 * Sets skb dst, assuming a reference was taken on dst and should 728 * be released by skb_dst_drop() 729 */ 730 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 731 { 732 skb->_skb_refdst = (unsigned long)dst; 733 } 734 735 /** 736 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 737 * @skb: buffer 738 * @dst: dst entry 739 * 740 * Sets skb dst, assuming a reference was not taken on dst. 741 * If dst entry is cached, we do not take reference and dst_release 742 * will be avoided by refdst_drop. If dst entry is not cached, we take 743 * reference, so that last dst_release can destroy the dst immediately. 744 */ 745 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 746 { 747 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 748 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 749 } 750 751 /** 752 * skb_dst_is_noref - Test if skb dst isn't refcounted 753 * @skb: buffer 754 */ 755 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 756 { 757 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 758 } 759 760 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 761 { 762 return (struct rtable *)skb_dst(skb); 763 } 764 765 void kfree_skb(struct sk_buff *skb); 766 void kfree_skb_list(struct sk_buff *segs); 767 void skb_tx_error(struct sk_buff *skb); 768 void consume_skb(struct sk_buff *skb); 769 void __kfree_skb(struct sk_buff *skb); 770 extern struct kmem_cache *skbuff_head_cache; 771 772 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 773 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 774 bool *fragstolen, int *delta_truesize); 775 776 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 777 int node); 778 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 779 struct sk_buff *build_skb(void *data, unsigned int frag_size); 780 static inline struct sk_buff *alloc_skb(unsigned int size, 781 gfp_t priority) 782 { 783 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 784 } 785 786 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 787 unsigned long data_len, 788 int max_page_order, 789 int *errcode, 790 gfp_t gfp_mask); 791 792 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 793 struct sk_buff_fclones { 794 struct sk_buff skb1; 795 796 struct sk_buff skb2; 797 798 atomic_t fclone_ref; 799 }; 800 801 /** 802 * skb_fclone_busy - check if fclone is busy 803 * @skb: buffer 804 * 805 * Returns true is skb is a fast clone, and its clone is not freed. 806 * Some drivers call skb_orphan() in their ndo_start_xmit(), 807 * so we also check that this didnt happen. 808 */ 809 static inline bool skb_fclone_busy(const struct sock *sk, 810 const struct sk_buff *skb) 811 { 812 const struct sk_buff_fclones *fclones; 813 814 fclones = container_of(skb, struct sk_buff_fclones, skb1); 815 816 return skb->fclone == SKB_FCLONE_ORIG && 817 atomic_read(&fclones->fclone_ref) > 1 && 818 fclones->skb2.sk == sk; 819 } 820 821 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 822 gfp_t priority) 823 { 824 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 825 } 826 827 struct sk_buff *__alloc_skb_head(gfp_t priority, int node); 828 static inline struct sk_buff *alloc_skb_head(gfp_t priority) 829 { 830 return __alloc_skb_head(priority, -1); 831 } 832 833 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 834 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 835 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 836 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 837 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 838 gfp_t gfp_mask, bool fclone); 839 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 840 gfp_t gfp_mask) 841 { 842 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 843 } 844 845 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 846 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 847 unsigned int headroom); 848 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 849 int newtailroom, gfp_t priority); 850 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 851 int offset, int len); 852 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 853 int offset, int len); 854 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 855 int skb_pad(struct sk_buff *skb, int pad); 856 #define dev_kfree_skb(a) consume_skb(a) 857 858 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 859 int getfrag(void *from, char *to, int offset, 860 int len, int odd, struct sk_buff *skb), 861 void *from, int length); 862 863 struct skb_seq_state { 864 __u32 lower_offset; 865 __u32 upper_offset; 866 __u32 frag_idx; 867 __u32 stepped_offset; 868 struct sk_buff *root_skb; 869 struct sk_buff *cur_skb; 870 __u8 *frag_data; 871 }; 872 873 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 874 unsigned int to, struct skb_seq_state *st); 875 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 876 struct skb_seq_state *st); 877 void skb_abort_seq_read(struct skb_seq_state *st); 878 879 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 880 unsigned int to, struct ts_config *config); 881 882 /* 883 * Packet hash types specify the type of hash in skb_set_hash. 884 * 885 * Hash types refer to the protocol layer addresses which are used to 886 * construct a packet's hash. The hashes are used to differentiate or identify 887 * flows of the protocol layer for the hash type. Hash types are either 888 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 889 * 890 * Properties of hashes: 891 * 892 * 1) Two packets in different flows have different hash values 893 * 2) Two packets in the same flow should have the same hash value 894 * 895 * A hash at a higher layer is considered to be more specific. A driver should 896 * set the most specific hash possible. 897 * 898 * A driver cannot indicate a more specific hash than the layer at which a hash 899 * was computed. For instance an L3 hash cannot be set as an L4 hash. 900 * 901 * A driver may indicate a hash level which is less specific than the 902 * actual layer the hash was computed on. For instance, a hash computed 903 * at L4 may be considered an L3 hash. This should only be done if the 904 * driver can't unambiguously determine that the HW computed the hash at 905 * the higher layer. Note that the "should" in the second property above 906 * permits this. 907 */ 908 enum pkt_hash_types { 909 PKT_HASH_TYPE_NONE, /* Undefined type */ 910 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 911 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 912 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 913 }; 914 915 static inline void 916 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 917 { 918 skb->l4_hash = (type == PKT_HASH_TYPE_L4); 919 skb->sw_hash = 0; 920 skb->hash = hash; 921 } 922 923 void __skb_get_hash(struct sk_buff *skb); 924 static inline __u32 skb_get_hash(struct sk_buff *skb) 925 { 926 if (!skb->l4_hash && !skb->sw_hash) 927 __skb_get_hash(skb); 928 929 return skb->hash; 930 } 931 932 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 933 { 934 return skb->hash; 935 } 936 937 static inline void skb_clear_hash(struct sk_buff *skb) 938 { 939 skb->hash = 0; 940 skb->sw_hash = 0; 941 skb->l4_hash = 0; 942 } 943 944 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 945 { 946 if (!skb->l4_hash) 947 skb_clear_hash(skb); 948 } 949 950 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 951 { 952 to->hash = from->hash; 953 to->sw_hash = from->sw_hash; 954 to->l4_hash = from->l4_hash; 955 }; 956 957 static inline void skb_sender_cpu_clear(struct sk_buff *skb) 958 { 959 #ifdef CONFIG_XPS 960 skb->sender_cpu = 0; 961 #endif 962 } 963 964 #ifdef NET_SKBUFF_DATA_USES_OFFSET 965 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 966 { 967 return skb->head + skb->end; 968 } 969 970 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 971 { 972 return skb->end; 973 } 974 #else 975 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 976 { 977 return skb->end; 978 } 979 980 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 981 { 982 return skb->end - skb->head; 983 } 984 #endif 985 986 /* Internal */ 987 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 988 989 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 990 { 991 return &skb_shinfo(skb)->hwtstamps; 992 } 993 994 /** 995 * skb_queue_empty - check if a queue is empty 996 * @list: queue head 997 * 998 * Returns true if the queue is empty, false otherwise. 999 */ 1000 static inline int skb_queue_empty(const struct sk_buff_head *list) 1001 { 1002 return list->next == (const struct sk_buff *) list; 1003 } 1004 1005 /** 1006 * skb_queue_is_last - check if skb is the last entry in the queue 1007 * @list: queue head 1008 * @skb: buffer 1009 * 1010 * Returns true if @skb is the last buffer on the list. 1011 */ 1012 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1013 const struct sk_buff *skb) 1014 { 1015 return skb->next == (const struct sk_buff *) list; 1016 } 1017 1018 /** 1019 * skb_queue_is_first - check if skb is the first entry in the queue 1020 * @list: queue head 1021 * @skb: buffer 1022 * 1023 * Returns true if @skb is the first buffer on the list. 1024 */ 1025 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1026 const struct sk_buff *skb) 1027 { 1028 return skb->prev == (const struct sk_buff *) list; 1029 } 1030 1031 /** 1032 * skb_queue_next - return the next packet in the queue 1033 * @list: queue head 1034 * @skb: current buffer 1035 * 1036 * Return the next packet in @list after @skb. It is only valid to 1037 * call this if skb_queue_is_last() evaluates to false. 1038 */ 1039 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1040 const struct sk_buff *skb) 1041 { 1042 /* This BUG_ON may seem severe, but if we just return then we 1043 * are going to dereference garbage. 1044 */ 1045 BUG_ON(skb_queue_is_last(list, skb)); 1046 return skb->next; 1047 } 1048 1049 /** 1050 * skb_queue_prev - return the prev packet in the queue 1051 * @list: queue head 1052 * @skb: current buffer 1053 * 1054 * Return the prev packet in @list before @skb. It is only valid to 1055 * call this if skb_queue_is_first() evaluates to false. 1056 */ 1057 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1058 const struct sk_buff *skb) 1059 { 1060 /* This BUG_ON may seem severe, but if we just return then we 1061 * are going to dereference garbage. 1062 */ 1063 BUG_ON(skb_queue_is_first(list, skb)); 1064 return skb->prev; 1065 } 1066 1067 /** 1068 * skb_get - reference buffer 1069 * @skb: buffer to reference 1070 * 1071 * Makes another reference to a socket buffer and returns a pointer 1072 * to the buffer. 1073 */ 1074 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1075 { 1076 atomic_inc(&skb->users); 1077 return skb; 1078 } 1079 1080 /* 1081 * If users == 1, we are the only owner and are can avoid redundant 1082 * atomic change. 1083 */ 1084 1085 /** 1086 * skb_cloned - is the buffer a clone 1087 * @skb: buffer to check 1088 * 1089 * Returns true if the buffer was generated with skb_clone() and is 1090 * one of multiple shared copies of the buffer. Cloned buffers are 1091 * shared data so must not be written to under normal circumstances. 1092 */ 1093 static inline int skb_cloned(const struct sk_buff *skb) 1094 { 1095 return skb->cloned && 1096 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1097 } 1098 1099 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1100 { 1101 might_sleep_if(pri & __GFP_WAIT); 1102 1103 if (skb_cloned(skb)) 1104 return pskb_expand_head(skb, 0, 0, pri); 1105 1106 return 0; 1107 } 1108 1109 /** 1110 * skb_header_cloned - is the header a clone 1111 * @skb: buffer to check 1112 * 1113 * Returns true if modifying the header part of the buffer requires 1114 * the data to be copied. 1115 */ 1116 static inline int skb_header_cloned(const struct sk_buff *skb) 1117 { 1118 int dataref; 1119 1120 if (!skb->cloned) 1121 return 0; 1122 1123 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1124 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1125 return dataref != 1; 1126 } 1127 1128 /** 1129 * skb_header_release - release reference to header 1130 * @skb: buffer to operate on 1131 * 1132 * Drop a reference to the header part of the buffer. This is done 1133 * by acquiring a payload reference. You must not read from the header 1134 * part of skb->data after this. 1135 * Note : Check if you can use __skb_header_release() instead. 1136 */ 1137 static inline void skb_header_release(struct sk_buff *skb) 1138 { 1139 BUG_ON(skb->nohdr); 1140 skb->nohdr = 1; 1141 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1142 } 1143 1144 /** 1145 * __skb_header_release - release reference to header 1146 * @skb: buffer to operate on 1147 * 1148 * Variant of skb_header_release() assuming skb is private to caller. 1149 * We can avoid one atomic operation. 1150 */ 1151 static inline void __skb_header_release(struct sk_buff *skb) 1152 { 1153 skb->nohdr = 1; 1154 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1155 } 1156 1157 1158 /** 1159 * skb_shared - is the buffer shared 1160 * @skb: buffer to check 1161 * 1162 * Returns true if more than one person has a reference to this 1163 * buffer. 1164 */ 1165 static inline int skb_shared(const struct sk_buff *skb) 1166 { 1167 return atomic_read(&skb->users) != 1; 1168 } 1169 1170 /** 1171 * skb_share_check - check if buffer is shared and if so clone it 1172 * @skb: buffer to check 1173 * @pri: priority for memory allocation 1174 * 1175 * If the buffer is shared the buffer is cloned and the old copy 1176 * drops a reference. A new clone with a single reference is returned. 1177 * If the buffer is not shared the original buffer is returned. When 1178 * being called from interrupt status or with spinlocks held pri must 1179 * be GFP_ATOMIC. 1180 * 1181 * NULL is returned on a memory allocation failure. 1182 */ 1183 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1184 { 1185 might_sleep_if(pri & __GFP_WAIT); 1186 if (skb_shared(skb)) { 1187 struct sk_buff *nskb = skb_clone(skb, pri); 1188 1189 if (likely(nskb)) 1190 consume_skb(skb); 1191 else 1192 kfree_skb(skb); 1193 skb = nskb; 1194 } 1195 return skb; 1196 } 1197 1198 /* 1199 * Copy shared buffers into a new sk_buff. We effectively do COW on 1200 * packets to handle cases where we have a local reader and forward 1201 * and a couple of other messy ones. The normal one is tcpdumping 1202 * a packet thats being forwarded. 1203 */ 1204 1205 /** 1206 * skb_unshare - make a copy of a shared buffer 1207 * @skb: buffer to check 1208 * @pri: priority for memory allocation 1209 * 1210 * If the socket buffer is a clone then this function creates a new 1211 * copy of the data, drops a reference count on the old copy and returns 1212 * the new copy with the reference count at 1. If the buffer is not a clone 1213 * the original buffer is returned. When called with a spinlock held or 1214 * from interrupt state @pri must be %GFP_ATOMIC 1215 * 1216 * %NULL is returned on a memory allocation failure. 1217 */ 1218 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1219 gfp_t pri) 1220 { 1221 might_sleep_if(pri & __GFP_WAIT); 1222 if (skb_cloned(skb)) { 1223 struct sk_buff *nskb = skb_copy(skb, pri); 1224 1225 /* Free our shared copy */ 1226 if (likely(nskb)) 1227 consume_skb(skb); 1228 else 1229 kfree_skb(skb); 1230 skb = nskb; 1231 } 1232 return skb; 1233 } 1234 1235 /** 1236 * skb_peek - peek at the head of an &sk_buff_head 1237 * @list_: list to peek at 1238 * 1239 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1240 * be careful with this one. A peek leaves the buffer on the 1241 * list and someone else may run off with it. You must hold 1242 * the appropriate locks or have a private queue to do this. 1243 * 1244 * Returns %NULL for an empty list or a pointer to the head element. 1245 * The reference count is not incremented and the reference is therefore 1246 * volatile. Use with caution. 1247 */ 1248 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1249 { 1250 struct sk_buff *skb = list_->next; 1251 1252 if (skb == (struct sk_buff *)list_) 1253 skb = NULL; 1254 return skb; 1255 } 1256 1257 /** 1258 * skb_peek_next - peek skb following the given one from a queue 1259 * @skb: skb to start from 1260 * @list_: list to peek at 1261 * 1262 * Returns %NULL when the end of the list is met or a pointer to the 1263 * next element. The reference count is not incremented and the 1264 * reference is therefore volatile. Use with caution. 1265 */ 1266 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1267 const struct sk_buff_head *list_) 1268 { 1269 struct sk_buff *next = skb->next; 1270 1271 if (next == (struct sk_buff *)list_) 1272 next = NULL; 1273 return next; 1274 } 1275 1276 /** 1277 * skb_peek_tail - peek at the tail of an &sk_buff_head 1278 * @list_: list to peek at 1279 * 1280 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1281 * be careful with this one. A peek leaves the buffer on the 1282 * list and someone else may run off with it. You must hold 1283 * the appropriate locks or have a private queue to do this. 1284 * 1285 * Returns %NULL for an empty list or a pointer to the tail element. 1286 * The reference count is not incremented and the reference is therefore 1287 * volatile. Use with caution. 1288 */ 1289 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1290 { 1291 struct sk_buff *skb = list_->prev; 1292 1293 if (skb == (struct sk_buff *)list_) 1294 skb = NULL; 1295 return skb; 1296 1297 } 1298 1299 /** 1300 * skb_queue_len - get queue length 1301 * @list_: list to measure 1302 * 1303 * Return the length of an &sk_buff queue. 1304 */ 1305 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1306 { 1307 return list_->qlen; 1308 } 1309 1310 /** 1311 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1312 * @list: queue to initialize 1313 * 1314 * This initializes only the list and queue length aspects of 1315 * an sk_buff_head object. This allows to initialize the list 1316 * aspects of an sk_buff_head without reinitializing things like 1317 * the spinlock. It can also be used for on-stack sk_buff_head 1318 * objects where the spinlock is known to not be used. 1319 */ 1320 static inline void __skb_queue_head_init(struct sk_buff_head *list) 1321 { 1322 list->prev = list->next = (struct sk_buff *)list; 1323 list->qlen = 0; 1324 } 1325 1326 /* 1327 * This function creates a split out lock class for each invocation; 1328 * this is needed for now since a whole lot of users of the skb-queue 1329 * infrastructure in drivers have different locking usage (in hardirq) 1330 * than the networking core (in softirq only). In the long run either the 1331 * network layer or drivers should need annotation to consolidate the 1332 * main types of usage into 3 classes. 1333 */ 1334 static inline void skb_queue_head_init(struct sk_buff_head *list) 1335 { 1336 spin_lock_init(&list->lock); 1337 __skb_queue_head_init(list); 1338 } 1339 1340 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1341 struct lock_class_key *class) 1342 { 1343 skb_queue_head_init(list); 1344 lockdep_set_class(&list->lock, class); 1345 } 1346 1347 /* 1348 * Insert an sk_buff on a list. 1349 * 1350 * The "__skb_xxxx()" functions are the non-atomic ones that 1351 * can only be called with interrupts disabled. 1352 */ 1353 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1354 struct sk_buff_head *list); 1355 static inline void __skb_insert(struct sk_buff *newsk, 1356 struct sk_buff *prev, struct sk_buff *next, 1357 struct sk_buff_head *list) 1358 { 1359 newsk->next = next; 1360 newsk->prev = prev; 1361 next->prev = prev->next = newsk; 1362 list->qlen++; 1363 } 1364 1365 static inline void __skb_queue_splice(const struct sk_buff_head *list, 1366 struct sk_buff *prev, 1367 struct sk_buff *next) 1368 { 1369 struct sk_buff *first = list->next; 1370 struct sk_buff *last = list->prev; 1371 1372 first->prev = prev; 1373 prev->next = first; 1374 1375 last->next = next; 1376 next->prev = last; 1377 } 1378 1379 /** 1380 * skb_queue_splice - join two skb lists, this is designed for stacks 1381 * @list: the new list to add 1382 * @head: the place to add it in the first list 1383 */ 1384 static inline void skb_queue_splice(const struct sk_buff_head *list, 1385 struct sk_buff_head *head) 1386 { 1387 if (!skb_queue_empty(list)) { 1388 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1389 head->qlen += list->qlen; 1390 } 1391 } 1392 1393 /** 1394 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1395 * @list: the new list to add 1396 * @head: the place to add it in the first list 1397 * 1398 * The list at @list is reinitialised 1399 */ 1400 static inline void skb_queue_splice_init(struct sk_buff_head *list, 1401 struct sk_buff_head *head) 1402 { 1403 if (!skb_queue_empty(list)) { 1404 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1405 head->qlen += list->qlen; 1406 __skb_queue_head_init(list); 1407 } 1408 } 1409 1410 /** 1411 * skb_queue_splice_tail - join two skb lists, each list being a queue 1412 * @list: the new list to add 1413 * @head: the place to add it in the first list 1414 */ 1415 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1416 struct sk_buff_head *head) 1417 { 1418 if (!skb_queue_empty(list)) { 1419 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1420 head->qlen += list->qlen; 1421 } 1422 } 1423 1424 /** 1425 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1426 * @list: the new list to add 1427 * @head: the place to add it in the first list 1428 * 1429 * Each of the lists is a queue. 1430 * The list at @list is reinitialised 1431 */ 1432 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1433 struct sk_buff_head *head) 1434 { 1435 if (!skb_queue_empty(list)) { 1436 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1437 head->qlen += list->qlen; 1438 __skb_queue_head_init(list); 1439 } 1440 } 1441 1442 /** 1443 * __skb_queue_after - queue a buffer at the list head 1444 * @list: list to use 1445 * @prev: place after this buffer 1446 * @newsk: buffer to queue 1447 * 1448 * Queue a buffer int the middle of a list. This function takes no locks 1449 * and you must therefore hold required locks before calling it. 1450 * 1451 * A buffer cannot be placed on two lists at the same time. 1452 */ 1453 static inline void __skb_queue_after(struct sk_buff_head *list, 1454 struct sk_buff *prev, 1455 struct sk_buff *newsk) 1456 { 1457 __skb_insert(newsk, prev, prev->next, list); 1458 } 1459 1460 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1461 struct sk_buff_head *list); 1462 1463 static inline void __skb_queue_before(struct sk_buff_head *list, 1464 struct sk_buff *next, 1465 struct sk_buff *newsk) 1466 { 1467 __skb_insert(newsk, next->prev, next, list); 1468 } 1469 1470 /** 1471 * __skb_queue_head - queue a buffer at the list head 1472 * @list: list to use 1473 * @newsk: buffer to queue 1474 * 1475 * Queue a buffer at the start of a list. This function takes no locks 1476 * and you must therefore hold required locks before calling it. 1477 * 1478 * A buffer cannot be placed on two lists at the same time. 1479 */ 1480 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1481 static inline void __skb_queue_head(struct sk_buff_head *list, 1482 struct sk_buff *newsk) 1483 { 1484 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1485 } 1486 1487 /** 1488 * __skb_queue_tail - queue a buffer at the list tail 1489 * @list: list to use 1490 * @newsk: buffer to queue 1491 * 1492 * Queue a buffer at the end of a list. This function takes no locks 1493 * and you must therefore hold required locks before calling it. 1494 * 1495 * A buffer cannot be placed on two lists at the same time. 1496 */ 1497 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1498 static inline void __skb_queue_tail(struct sk_buff_head *list, 1499 struct sk_buff *newsk) 1500 { 1501 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1502 } 1503 1504 /* 1505 * remove sk_buff from list. _Must_ be called atomically, and with 1506 * the list known.. 1507 */ 1508 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1509 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1510 { 1511 struct sk_buff *next, *prev; 1512 1513 list->qlen--; 1514 next = skb->next; 1515 prev = skb->prev; 1516 skb->next = skb->prev = NULL; 1517 next->prev = prev; 1518 prev->next = next; 1519 } 1520 1521 /** 1522 * __skb_dequeue - remove from the head of the queue 1523 * @list: list to dequeue from 1524 * 1525 * Remove the head of the list. This function does not take any locks 1526 * so must be used with appropriate locks held only. The head item is 1527 * returned or %NULL if the list is empty. 1528 */ 1529 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1530 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1531 { 1532 struct sk_buff *skb = skb_peek(list); 1533 if (skb) 1534 __skb_unlink(skb, list); 1535 return skb; 1536 } 1537 1538 /** 1539 * __skb_dequeue_tail - remove from the tail of the queue 1540 * @list: list to dequeue from 1541 * 1542 * Remove the tail of the list. This function does not take any locks 1543 * so must be used with appropriate locks held only. The tail item is 1544 * returned or %NULL if the list is empty. 1545 */ 1546 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1547 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1548 { 1549 struct sk_buff *skb = skb_peek_tail(list); 1550 if (skb) 1551 __skb_unlink(skb, list); 1552 return skb; 1553 } 1554 1555 1556 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1557 { 1558 return skb->data_len; 1559 } 1560 1561 static inline unsigned int skb_headlen(const struct sk_buff *skb) 1562 { 1563 return skb->len - skb->data_len; 1564 } 1565 1566 static inline int skb_pagelen(const struct sk_buff *skb) 1567 { 1568 int i, len = 0; 1569 1570 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--) 1571 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1572 return len + skb_headlen(skb); 1573 } 1574 1575 /** 1576 * __skb_fill_page_desc - initialise a paged fragment in an skb 1577 * @skb: buffer containing fragment to be initialised 1578 * @i: paged fragment index to initialise 1579 * @page: the page to use for this fragment 1580 * @off: the offset to the data with @page 1581 * @size: the length of the data 1582 * 1583 * Initialises the @i'th fragment of @skb to point to &size bytes at 1584 * offset @off within @page. 1585 * 1586 * Does not take any additional reference on the fragment. 1587 */ 1588 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1589 struct page *page, int off, int size) 1590 { 1591 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1592 1593 /* 1594 * Propagate page pfmemalloc to the skb if we can. The problem is 1595 * that not all callers have unique ownership of the page but rely 1596 * on page_is_pfmemalloc doing the right thing(tm). 1597 */ 1598 frag->page.p = page; 1599 frag->page_offset = off; 1600 skb_frag_size_set(frag, size); 1601 1602 page = compound_head(page); 1603 if (page_is_pfmemalloc(page)) 1604 skb->pfmemalloc = true; 1605 } 1606 1607 /** 1608 * skb_fill_page_desc - initialise a paged fragment in an skb 1609 * @skb: buffer containing fragment to be initialised 1610 * @i: paged fragment index to initialise 1611 * @page: the page to use for this fragment 1612 * @off: the offset to the data with @page 1613 * @size: the length of the data 1614 * 1615 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1616 * @skb to point to @size bytes at offset @off within @page. In 1617 * addition updates @skb such that @i is the last fragment. 1618 * 1619 * Does not take any additional reference on the fragment. 1620 */ 1621 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1622 struct page *page, int off, int size) 1623 { 1624 __skb_fill_page_desc(skb, i, page, off, size); 1625 skb_shinfo(skb)->nr_frags = i + 1; 1626 } 1627 1628 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1629 int size, unsigned int truesize); 1630 1631 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1632 unsigned int truesize); 1633 1634 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1635 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1636 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1637 1638 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1639 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1640 { 1641 return skb->head + skb->tail; 1642 } 1643 1644 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1645 { 1646 skb->tail = skb->data - skb->head; 1647 } 1648 1649 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1650 { 1651 skb_reset_tail_pointer(skb); 1652 skb->tail += offset; 1653 } 1654 1655 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 1656 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1657 { 1658 return skb->tail; 1659 } 1660 1661 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1662 { 1663 skb->tail = skb->data; 1664 } 1665 1666 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1667 { 1668 skb->tail = skb->data + offset; 1669 } 1670 1671 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1672 1673 /* 1674 * Add data to an sk_buff 1675 */ 1676 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1677 unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1678 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1679 { 1680 unsigned char *tmp = skb_tail_pointer(skb); 1681 SKB_LINEAR_ASSERT(skb); 1682 skb->tail += len; 1683 skb->len += len; 1684 return tmp; 1685 } 1686 1687 unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1688 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1689 { 1690 skb->data -= len; 1691 skb->len += len; 1692 return skb->data; 1693 } 1694 1695 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1696 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1697 { 1698 skb->len -= len; 1699 BUG_ON(skb->len < skb->data_len); 1700 return skb->data += len; 1701 } 1702 1703 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1704 { 1705 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1706 } 1707 1708 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1709 1710 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1711 { 1712 if (len > skb_headlen(skb) && 1713 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1714 return NULL; 1715 skb->len -= len; 1716 return skb->data += len; 1717 } 1718 1719 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1720 { 1721 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1722 } 1723 1724 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1725 { 1726 if (likely(len <= skb_headlen(skb))) 1727 return 1; 1728 if (unlikely(len > skb->len)) 1729 return 0; 1730 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1731 } 1732 1733 /** 1734 * skb_headroom - bytes at buffer head 1735 * @skb: buffer to check 1736 * 1737 * Return the number of bytes of free space at the head of an &sk_buff. 1738 */ 1739 static inline unsigned int skb_headroom(const struct sk_buff *skb) 1740 { 1741 return skb->data - skb->head; 1742 } 1743 1744 /** 1745 * skb_tailroom - bytes at buffer end 1746 * @skb: buffer to check 1747 * 1748 * Return the number of bytes of free space at the tail of an sk_buff 1749 */ 1750 static inline int skb_tailroom(const struct sk_buff *skb) 1751 { 1752 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 1753 } 1754 1755 /** 1756 * skb_availroom - bytes at buffer end 1757 * @skb: buffer to check 1758 * 1759 * Return the number of bytes of free space at the tail of an sk_buff 1760 * allocated by sk_stream_alloc() 1761 */ 1762 static inline int skb_availroom(const struct sk_buff *skb) 1763 { 1764 if (skb_is_nonlinear(skb)) 1765 return 0; 1766 1767 return skb->end - skb->tail - skb->reserved_tailroom; 1768 } 1769 1770 /** 1771 * skb_reserve - adjust headroom 1772 * @skb: buffer to alter 1773 * @len: bytes to move 1774 * 1775 * Increase the headroom of an empty &sk_buff by reducing the tail 1776 * room. This is only allowed for an empty buffer. 1777 */ 1778 static inline void skb_reserve(struct sk_buff *skb, int len) 1779 { 1780 skb->data += len; 1781 skb->tail += len; 1782 } 1783 1784 /** 1785 * skb_tailroom_reserve - adjust reserved_tailroom 1786 * @skb: buffer to alter 1787 * @mtu: maximum amount of headlen permitted 1788 * @needed_tailroom: minimum amount of reserved_tailroom 1789 * 1790 * Set reserved_tailroom so that headlen can be as large as possible but 1791 * not larger than mtu and tailroom cannot be smaller than 1792 * needed_tailroom. 1793 * The required headroom should already have been reserved before using 1794 * this function. 1795 */ 1796 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 1797 unsigned int needed_tailroom) 1798 { 1799 SKB_LINEAR_ASSERT(skb); 1800 if (mtu < skb_tailroom(skb) - needed_tailroom) 1801 /* use at most mtu */ 1802 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 1803 else 1804 /* use up to all available space */ 1805 skb->reserved_tailroom = needed_tailroom; 1806 } 1807 1808 #define ENCAP_TYPE_ETHER 0 1809 #define ENCAP_TYPE_IPPROTO 1 1810 1811 static inline void skb_set_inner_protocol(struct sk_buff *skb, 1812 __be16 protocol) 1813 { 1814 skb->inner_protocol = protocol; 1815 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 1816 } 1817 1818 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 1819 __u8 ipproto) 1820 { 1821 skb->inner_ipproto = ipproto; 1822 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 1823 } 1824 1825 static inline void skb_reset_inner_headers(struct sk_buff *skb) 1826 { 1827 skb->inner_mac_header = skb->mac_header; 1828 skb->inner_network_header = skb->network_header; 1829 skb->inner_transport_header = skb->transport_header; 1830 } 1831 1832 static inline void skb_reset_mac_len(struct sk_buff *skb) 1833 { 1834 skb->mac_len = skb->network_header - skb->mac_header; 1835 } 1836 1837 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 1838 *skb) 1839 { 1840 return skb->head + skb->inner_transport_header; 1841 } 1842 1843 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 1844 { 1845 skb->inner_transport_header = skb->data - skb->head; 1846 } 1847 1848 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 1849 const int offset) 1850 { 1851 skb_reset_inner_transport_header(skb); 1852 skb->inner_transport_header += offset; 1853 } 1854 1855 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 1856 { 1857 return skb->head + skb->inner_network_header; 1858 } 1859 1860 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 1861 { 1862 skb->inner_network_header = skb->data - skb->head; 1863 } 1864 1865 static inline void skb_set_inner_network_header(struct sk_buff *skb, 1866 const int offset) 1867 { 1868 skb_reset_inner_network_header(skb); 1869 skb->inner_network_header += offset; 1870 } 1871 1872 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 1873 { 1874 return skb->head + skb->inner_mac_header; 1875 } 1876 1877 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 1878 { 1879 skb->inner_mac_header = skb->data - skb->head; 1880 } 1881 1882 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 1883 const int offset) 1884 { 1885 skb_reset_inner_mac_header(skb); 1886 skb->inner_mac_header += offset; 1887 } 1888 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 1889 { 1890 return skb->transport_header != (typeof(skb->transport_header))~0U; 1891 } 1892 1893 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 1894 { 1895 return skb->head + skb->transport_header; 1896 } 1897 1898 static inline void skb_reset_transport_header(struct sk_buff *skb) 1899 { 1900 skb->transport_header = skb->data - skb->head; 1901 } 1902 1903 static inline void skb_set_transport_header(struct sk_buff *skb, 1904 const int offset) 1905 { 1906 skb_reset_transport_header(skb); 1907 skb->transport_header += offset; 1908 } 1909 1910 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 1911 { 1912 return skb->head + skb->network_header; 1913 } 1914 1915 static inline void skb_reset_network_header(struct sk_buff *skb) 1916 { 1917 skb->network_header = skb->data - skb->head; 1918 } 1919 1920 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 1921 { 1922 skb_reset_network_header(skb); 1923 skb->network_header += offset; 1924 } 1925 1926 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 1927 { 1928 return skb->head + skb->mac_header; 1929 } 1930 1931 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 1932 { 1933 return skb->mac_header != (typeof(skb->mac_header))~0U; 1934 } 1935 1936 static inline void skb_reset_mac_header(struct sk_buff *skb) 1937 { 1938 skb->mac_header = skb->data - skb->head; 1939 } 1940 1941 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 1942 { 1943 skb_reset_mac_header(skb); 1944 skb->mac_header += offset; 1945 } 1946 1947 static inline void skb_pop_mac_header(struct sk_buff *skb) 1948 { 1949 skb->mac_header = skb->network_header; 1950 } 1951 1952 static inline void skb_probe_transport_header(struct sk_buff *skb, 1953 const int offset_hint) 1954 { 1955 struct flow_keys keys; 1956 1957 if (skb_transport_header_was_set(skb)) 1958 return; 1959 else if (skb_flow_dissect(skb, &keys)) 1960 skb_set_transport_header(skb, keys.thoff); 1961 else 1962 skb_set_transport_header(skb, offset_hint); 1963 } 1964 1965 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 1966 { 1967 if (skb_mac_header_was_set(skb)) { 1968 const unsigned char *old_mac = skb_mac_header(skb); 1969 1970 skb_set_mac_header(skb, -skb->mac_len); 1971 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 1972 } 1973 } 1974 1975 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 1976 { 1977 return skb->csum_start - skb_headroom(skb); 1978 } 1979 1980 static inline int skb_transport_offset(const struct sk_buff *skb) 1981 { 1982 return skb_transport_header(skb) - skb->data; 1983 } 1984 1985 static inline u32 skb_network_header_len(const struct sk_buff *skb) 1986 { 1987 return skb->transport_header - skb->network_header; 1988 } 1989 1990 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 1991 { 1992 return skb->inner_transport_header - skb->inner_network_header; 1993 } 1994 1995 static inline int skb_network_offset(const struct sk_buff *skb) 1996 { 1997 return skb_network_header(skb) - skb->data; 1998 } 1999 2000 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2001 { 2002 return skb_inner_network_header(skb) - skb->data; 2003 } 2004 2005 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2006 { 2007 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2008 } 2009 2010 /* 2011 * CPUs often take a performance hit when accessing unaligned memory 2012 * locations. The actual performance hit varies, it can be small if the 2013 * hardware handles it or large if we have to take an exception and fix it 2014 * in software. 2015 * 2016 * Since an ethernet header is 14 bytes network drivers often end up with 2017 * the IP header at an unaligned offset. The IP header can be aligned by 2018 * shifting the start of the packet by 2 bytes. Drivers should do this 2019 * with: 2020 * 2021 * skb_reserve(skb, NET_IP_ALIGN); 2022 * 2023 * The downside to this alignment of the IP header is that the DMA is now 2024 * unaligned. On some architectures the cost of an unaligned DMA is high 2025 * and this cost outweighs the gains made by aligning the IP header. 2026 * 2027 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2028 * to be overridden. 2029 */ 2030 #ifndef NET_IP_ALIGN 2031 #define NET_IP_ALIGN 2 2032 #endif 2033 2034 /* 2035 * The networking layer reserves some headroom in skb data (via 2036 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2037 * the header has to grow. In the default case, if the header has to grow 2038 * 32 bytes or less we avoid the reallocation. 2039 * 2040 * Unfortunately this headroom changes the DMA alignment of the resulting 2041 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2042 * on some architectures. An architecture can override this value, 2043 * perhaps setting it to a cacheline in size (since that will maintain 2044 * cacheline alignment of the DMA). It must be a power of 2. 2045 * 2046 * Various parts of the networking layer expect at least 32 bytes of 2047 * headroom, you should not reduce this. 2048 * 2049 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2050 * to reduce average number of cache lines per packet. 2051 * get_rps_cpus() for example only access one 64 bytes aligned block : 2052 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2053 */ 2054 #ifndef NET_SKB_PAD 2055 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2056 #endif 2057 2058 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2059 2060 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2061 { 2062 if (unlikely(skb_is_nonlinear(skb))) { 2063 WARN_ON(1); 2064 return; 2065 } 2066 skb->len = len; 2067 skb_set_tail_pointer(skb, len); 2068 } 2069 2070 void skb_trim(struct sk_buff *skb, unsigned int len); 2071 2072 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2073 { 2074 if (skb->data_len) 2075 return ___pskb_trim(skb, len); 2076 __skb_trim(skb, len); 2077 return 0; 2078 } 2079 2080 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2081 { 2082 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2083 } 2084 2085 /** 2086 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2087 * @skb: buffer to alter 2088 * @len: new length 2089 * 2090 * This is identical to pskb_trim except that the caller knows that 2091 * the skb is not cloned so we should never get an error due to out- 2092 * of-memory. 2093 */ 2094 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2095 { 2096 int err = pskb_trim(skb, len); 2097 BUG_ON(err); 2098 } 2099 2100 /** 2101 * skb_orphan - orphan a buffer 2102 * @skb: buffer to orphan 2103 * 2104 * If a buffer currently has an owner then we call the owner's 2105 * destructor function and make the @skb unowned. The buffer continues 2106 * to exist but is no longer charged to its former owner. 2107 */ 2108 static inline void skb_orphan(struct sk_buff *skb) 2109 { 2110 if (skb->destructor) { 2111 skb->destructor(skb); 2112 skb->destructor = NULL; 2113 skb->sk = NULL; 2114 } else { 2115 BUG_ON(skb->sk); 2116 } 2117 } 2118 2119 /** 2120 * skb_orphan_frags - orphan the frags contained in a buffer 2121 * @skb: buffer to orphan frags from 2122 * @gfp_mask: allocation mask for replacement pages 2123 * 2124 * For each frag in the SKB which needs a destructor (i.e. has an 2125 * owner) create a copy of that frag and release the original 2126 * page by calling the destructor. 2127 */ 2128 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2129 { 2130 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 2131 return 0; 2132 return skb_copy_ubufs(skb, gfp_mask); 2133 } 2134 2135 /** 2136 * __skb_queue_purge - empty a list 2137 * @list: list to empty 2138 * 2139 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2140 * the list and one reference dropped. This function does not take the 2141 * list lock and the caller must hold the relevant locks to use it. 2142 */ 2143 void skb_queue_purge(struct sk_buff_head *list); 2144 static inline void __skb_queue_purge(struct sk_buff_head *list) 2145 { 2146 struct sk_buff *skb; 2147 while ((skb = __skb_dequeue(list)) != NULL) 2148 kfree_skb(skb); 2149 } 2150 2151 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768) 2152 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER) 2153 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE 2154 2155 void *netdev_alloc_frag(unsigned int fragsz); 2156 2157 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2158 gfp_t gfp_mask); 2159 2160 /** 2161 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2162 * @dev: network device to receive on 2163 * @length: length to allocate 2164 * 2165 * Allocate a new &sk_buff and assign it a usage count of one. The 2166 * buffer has unspecified headroom built in. Users should allocate 2167 * the headroom they think they need without accounting for the 2168 * built in space. The built in space is used for optimisations. 2169 * 2170 * %NULL is returned if there is no free memory. Although this function 2171 * allocates memory it can be called from an interrupt. 2172 */ 2173 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2174 unsigned int length) 2175 { 2176 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2177 } 2178 2179 /* legacy helper around __netdev_alloc_skb() */ 2180 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2181 gfp_t gfp_mask) 2182 { 2183 return __netdev_alloc_skb(NULL, length, gfp_mask); 2184 } 2185 2186 /* legacy helper around netdev_alloc_skb() */ 2187 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2188 { 2189 return netdev_alloc_skb(NULL, length); 2190 } 2191 2192 2193 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2194 unsigned int length, gfp_t gfp) 2195 { 2196 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2197 2198 if (NET_IP_ALIGN && skb) 2199 skb_reserve(skb, NET_IP_ALIGN); 2200 return skb; 2201 } 2202 2203 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2204 unsigned int length) 2205 { 2206 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2207 } 2208 2209 void *napi_alloc_frag(unsigned int fragsz); 2210 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2211 unsigned int length, gfp_t gfp_mask); 2212 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2213 unsigned int length) 2214 { 2215 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2216 } 2217 2218 /** 2219 * __dev_alloc_pages - allocate page for network Rx 2220 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2221 * @order: size of the allocation 2222 * 2223 * Allocate a new page. 2224 * 2225 * %NULL is returned if there is no free memory. 2226 */ 2227 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2228 unsigned int order) 2229 { 2230 /* This piece of code contains several assumptions. 2231 * 1. This is for device Rx, therefor a cold page is preferred. 2232 * 2. The expectation is the user wants a compound page. 2233 * 3. If requesting a order 0 page it will not be compound 2234 * due to the check to see if order has a value in prep_new_page 2235 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2236 * code in gfp_to_alloc_flags that should be enforcing this. 2237 */ 2238 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2239 2240 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2241 } 2242 2243 static inline struct page *dev_alloc_pages(unsigned int order) 2244 { 2245 return __dev_alloc_pages(GFP_ATOMIC, order); 2246 } 2247 2248 /** 2249 * __dev_alloc_page - allocate a page for network Rx 2250 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2251 * 2252 * Allocate a new page. 2253 * 2254 * %NULL is returned if there is no free memory. 2255 */ 2256 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2257 { 2258 return __dev_alloc_pages(gfp_mask, 0); 2259 } 2260 2261 static inline struct page *dev_alloc_page(void) 2262 { 2263 return __dev_alloc_page(GFP_ATOMIC); 2264 } 2265 2266 /** 2267 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2268 * @page: The page that was allocated from skb_alloc_page 2269 * @skb: The skb that may need pfmemalloc set 2270 */ 2271 static inline void skb_propagate_pfmemalloc(struct page *page, 2272 struct sk_buff *skb) 2273 { 2274 if (page_is_pfmemalloc(page)) 2275 skb->pfmemalloc = true; 2276 } 2277 2278 /** 2279 * skb_frag_page - retrieve the page referred to by a paged fragment 2280 * @frag: the paged fragment 2281 * 2282 * Returns the &struct page associated with @frag. 2283 */ 2284 static inline struct page *skb_frag_page(const skb_frag_t *frag) 2285 { 2286 return frag->page.p; 2287 } 2288 2289 /** 2290 * __skb_frag_ref - take an addition reference on a paged fragment. 2291 * @frag: the paged fragment 2292 * 2293 * Takes an additional reference on the paged fragment @frag. 2294 */ 2295 static inline void __skb_frag_ref(skb_frag_t *frag) 2296 { 2297 get_page(skb_frag_page(frag)); 2298 } 2299 2300 /** 2301 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2302 * @skb: the buffer 2303 * @f: the fragment offset. 2304 * 2305 * Takes an additional reference on the @f'th paged fragment of @skb. 2306 */ 2307 static inline void skb_frag_ref(struct sk_buff *skb, int f) 2308 { 2309 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2310 } 2311 2312 /** 2313 * __skb_frag_unref - release a reference on a paged fragment. 2314 * @frag: the paged fragment 2315 * 2316 * Releases a reference on the paged fragment @frag. 2317 */ 2318 static inline void __skb_frag_unref(skb_frag_t *frag) 2319 { 2320 put_page(skb_frag_page(frag)); 2321 } 2322 2323 /** 2324 * skb_frag_unref - release a reference on a paged fragment of an skb. 2325 * @skb: the buffer 2326 * @f: the fragment offset 2327 * 2328 * Releases a reference on the @f'th paged fragment of @skb. 2329 */ 2330 static inline void skb_frag_unref(struct sk_buff *skb, int f) 2331 { 2332 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2333 } 2334 2335 /** 2336 * skb_frag_address - gets the address of the data contained in a paged fragment 2337 * @frag: the paged fragment buffer 2338 * 2339 * Returns the address of the data within @frag. The page must already 2340 * be mapped. 2341 */ 2342 static inline void *skb_frag_address(const skb_frag_t *frag) 2343 { 2344 return page_address(skb_frag_page(frag)) + frag->page_offset; 2345 } 2346 2347 /** 2348 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2349 * @frag: the paged fragment buffer 2350 * 2351 * Returns the address of the data within @frag. Checks that the page 2352 * is mapped and returns %NULL otherwise. 2353 */ 2354 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2355 { 2356 void *ptr = page_address(skb_frag_page(frag)); 2357 if (unlikely(!ptr)) 2358 return NULL; 2359 2360 return ptr + frag->page_offset; 2361 } 2362 2363 /** 2364 * __skb_frag_set_page - sets the page contained in a paged fragment 2365 * @frag: the paged fragment 2366 * @page: the page to set 2367 * 2368 * Sets the fragment @frag to contain @page. 2369 */ 2370 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2371 { 2372 frag->page.p = page; 2373 } 2374 2375 /** 2376 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2377 * @skb: the buffer 2378 * @f: the fragment offset 2379 * @page: the page to set 2380 * 2381 * Sets the @f'th fragment of @skb to contain @page. 2382 */ 2383 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2384 struct page *page) 2385 { 2386 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2387 } 2388 2389 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2390 2391 /** 2392 * skb_frag_dma_map - maps a paged fragment via the DMA API 2393 * @dev: the device to map the fragment to 2394 * @frag: the paged fragment to map 2395 * @offset: the offset within the fragment (starting at the 2396 * fragment's own offset) 2397 * @size: the number of bytes to map 2398 * @dir: the direction of the mapping (%PCI_DMA_*) 2399 * 2400 * Maps the page associated with @frag to @device. 2401 */ 2402 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2403 const skb_frag_t *frag, 2404 size_t offset, size_t size, 2405 enum dma_data_direction dir) 2406 { 2407 return dma_map_page(dev, skb_frag_page(frag), 2408 frag->page_offset + offset, size, dir); 2409 } 2410 2411 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2412 gfp_t gfp_mask) 2413 { 2414 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2415 } 2416 2417 2418 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2419 gfp_t gfp_mask) 2420 { 2421 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2422 } 2423 2424 2425 /** 2426 * skb_clone_writable - is the header of a clone writable 2427 * @skb: buffer to check 2428 * @len: length up to which to write 2429 * 2430 * Returns true if modifying the header part of the cloned buffer 2431 * does not requires the data to be copied. 2432 */ 2433 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2434 { 2435 return !skb_header_cloned(skb) && 2436 skb_headroom(skb) + len <= skb->hdr_len; 2437 } 2438 2439 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2440 int cloned) 2441 { 2442 int delta = 0; 2443 2444 if (headroom > skb_headroom(skb)) 2445 delta = headroom - skb_headroom(skb); 2446 2447 if (delta || cloned) 2448 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2449 GFP_ATOMIC); 2450 return 0; 2451 } 2452 2453 /** 2454 * skb_cow - copy header of skb when it is required 2455 * @skb: buffer to cow 2456 * @headroom: needed headroom 2457 * 2458 * If the skb passed lacks sufficient headroom or its data part 2459 * is shared, data is reallocated. If reallocation fails, an error 2460 * is returned and original skb is not changed. 2461 * 2462 * The result is skb with writable area skb->head...skb->tail 2463 * and at least @headroom of space at head. 2464 */ 2465 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2466 { 2467 return __skb_cow(skb, headroom, skb_cloned(skb)); 2468 } 2469 2470 /** 2471 * skb_cow_head - skb_cow but only making the head writable 2472 * @skb: buffer to cow 2473 * @headroom: needed headroom 2474 * 2475 * This function is identical to skb_cow except that we replace the 2476 * skb_cloned check by skb_header_cloned. It should be used when 2477 * you only need to push on some header and do not need to modify 2478 * the data. 2479 */ 2480 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2481 { 2482 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2483 } 2484 2485 /** 2486 * skb_padto - pad an skbuff up to a minimal size 2487 * @skb: buffer to pad 2488 * @len: minimal length 2489 * 2490 * Pads up a buffer to ensure the trailing bytes exist and are 2491 * blanked. If the buffer already contains sufficient data it 2492 * is untouched. Otherwise it is extended. Returns zero on 2493 * success. The skb is freed on error. 2494 */ 2495 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2496 { 2497 unsigned int size = skb->len; 2498 if (likely(size >= len)) 2499 return 0; 2500 return skb_pad(skb, len - size); 2501 } 2502 2503 /** 2504 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2505 * @skb: buffer to pad 2506 * @len: minimal length 2507 * 2508 * Pads up a buffer to ensure the trailing bytes exist and are 2509 * blanked. If the buffer already contains sufficient data it 2510 * is untouched. Otherwise it is extended. Returns zero on 2511 * success. The skb is freed on error. 2512 */ 2513 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2514 { 2515 unsigned int size = skb->len; 2516 2517 if (unlikely(size < len)) { 2518 len -= size; 2519 if (skb_pad(skb, len)) 2520 return -ENOMEM; 2521 __skb_put(skb, len); 2522 } 2523 return 0; 2524 } 2525 2526 static inline int skb_add_data(struct sk_buff *skb, 2527 struct iov_iter *from, int copy) 2528 { 2529 const int off = skb->len; 2530 2531 if (skb->ip_summed == CHECKSUM_NONE) { 2532 __wsum csum = 0; 2533 if (csum_and_copy_from_iter(skb_put(skb, copy), copy, 2534 &csum, from) == copy) { 2535 skb->csum = csum_block_add(skb->csum, csum, off); 2536 return 0; 2537 } 2538 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy) 2539 return 0; 2540 2541 __skb_trim(skb, off); 2542 return -EFAULT; 2543 } 2544 2545 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2546 const struct page *page, int off) 2547 { 2548 if (i) { 2549 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2550 2551 return page == skb_frag_page(frag) && 2552 off == frag->page_offset + skb_frag_size(frag); 2553 } 2554 return false; 2555 } 2556 2557 static inline int __skb_linearize(struct sk_buff *skb) 2558 { 2559 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2560 } 2561 2562 /** 2563 * skb_linearize - convert paged skb to linear one 2564 * @skb: buffer to linarize 2565 * 2566 * If there is no free memory -ENOMEM is returned, otherwise zero 2567 * is returned and the old skb data released. 2568 */ 2569 static inline int skb_linearize(struct sk_buff *skb) 2570 { 2571 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2572 } 2573 2574 /** 2575 * skb_has_shared_frag - can any frag be overwritten 2576 * @skb: buffer to test 2577 * 2578 * Return true if the skb has at least one frag that might be modified 2579 * by an external entity (as in vmsplice()/sendfile()) 2580 */ 2581 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 2582 { 2583 return skb_is_nonlinear(skb) && 2584 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2585 } 2586 2587 /** 2588 * skb_linearize_cow - make sure skb is linear and writable 2589 * @skb: buffer to process 2590 * 2591 * If there is no free memory -ENOMEM is returned, otherwise zero 2592 * is returned and the old skb data released. 2593 */ 2594 static inline int skb_linearize_cow(struct sk_buff *skb) 2595 { 2596 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2597 __skb_linearize(skb) : 0; 2598 } 2599 2600 /** 2601 * skb_postpull_rcsum - update checksum for received skb after pull 2602 * @skb: buffer to update 2603 * @start: start of data before pull 2604 * @len: length of data pulled 2605 * 2606 * After doing a pull on a received packet, you need to call this to 2607 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2608 * CHECKSUM_NONE so that it can be recomputed from scratch. 2609 */ 2610 2611 static inline void skb_postpull_rcsum(struct sk_buff *skb, 2612 const void *start, unsigned int len) 2613 { 2614 if (skb->ip_summed == CHECKSUM_COMPLETE) 2615 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0)); 2616 else if (skb->ip_summed == CHECKSUM_PARTIAL && 2617 skb_checksum_start_offset(skb) < 0) 2618 skb->ip_summed = CHECKSUM_NONE; 2619 } 2620 2621 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2622 2623 /** 2624 * pskb_trim_rcsum - trim received skb and update checksum 2625 * @skb: buffer to trim 2626 * @len: new length 2627 * 2628 * This is exactly the same as pskb_trim except that it ensures the 2629 * checksum of received packets are still valid after the operation. 2630 */ 2631 2632 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2633 { 2634 if (likely(len >= skb->len)) 2635 return 0; 2636 if (skb->ip_summed == CHECKSUM_COMPLETE) 2637 skb->ip_summed = CHECKSUM_NONE; 2638 return __pskb_trim(skb, len); 2639 } 2640 2641 #define skb_queue_walk(queue, skb) \ 2642 for (skb = (queue)->next; \ 2643 skb != (struct sk_buff *)(queue); \ 2644 skb = skb->next) 2645 2646 #define skb_queue_walk_safe(queue, skb, tmp) \ 2647 for (skb = (queue)->next, tmp = skb->next; \ 2648 skb != (struct sk_buff *)(queue); \ 2649 skb = tmp, tmp = skb->next) 2650 2651 #define skb_queue_walk_from(queue, skb) \ 2652 for (; skb != (struct sk_buff *)(queue); \ 2653 skb = skb->next) 2654 2655 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 2656 for (tmp = skb->next; \ 2657 skb != (struct sk_buff *)(queue); \ 2658 skb = tmp, tmp = skb->next) 2659 2660 #define skb_queue_reverse_walk(queue, skb) \ 2661 for (skb = (queue)->prev; \ 2662 skb != (struct sk_buff *)(queue); \ 2663 skb = skb->prev) 2664 2665 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 2666 for (skb = (queue)->prev, tmp = skb->prev; \ 2667 skb != (struct sk_buff *)(queue); \ 2668 skb = tmp, tmp = skb->prev) 2669 2670 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 2671 for (tmp = skb->prev; \ 2672 skb != (struct sk_buff *)(queue); \ 2673 skb = tmp, tmp = skb->prev) 2674 2675 static inline bool skb_has_frag_list(const struct sk_buff *skb) 2676 { 2677 return skb_shinfo(skb)->frag_list != NULL; 2678 } 2679 2680 static inline void skb_frag_list_init(struct sk_buff *skb) 2681 { 2682 skb_shinfo(skb)->frag_list = NULL; 2683 } 2684 2685 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag) 2686 { 2687 frag->next = skb_shinfo(skb)->frag_list; 2688 skb_shinfo(skb)->frag_list = frag; 2689 } 2690 2691 #define skb_walk_frags(skb, iter) \ 2692 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 2693 2694 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 2695 int *peeked, int *off, int *err); 2696 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 2697 int *err); 2698 unsigned int datagram_poll(struct file *file, struct socket *sock, 2699 struct poll_table_struct *wait); 2700 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 2701 struct iov_iter *to, int size); 2702 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 2703 struct msghdr *msg, int size) 2704 { 2705 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 2706 } 2707 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 2708 struct msghdr *msg); 2709 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 2710 struct iov_iter *from, int len); 2711 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 2712 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 2713 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb); 2714 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 2715 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 2716 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 2717 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 2718 int len, __wsum csum); 2719 int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 2720 struct pipe_inode_info *pipe, unsigned int len, 2721 unsigned int flags); 2722 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 2723 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 2724 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 2725 int len, int hlen); 2726 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 2727 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 2728 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 2729 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 2730 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 2731 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 2732 int skb_ensure_writable(struct sk_buff *skb, int write_len); 2733 int skb_vlan_pop(struct sk_buff *skb); 2734 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 2735 2736 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 2737 { 2738 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2739 } 2740 2741 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 2742 { 2743 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2744 } 2745 2746 struct skb_checksum_ops { 2747 __wsum (*update)(const void *mem, int len, __wsum wsum); 2748 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 2749 }; 2750 2751 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 2752 __wsum csum, const struct skb_checksum_ops *ops); 2753 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 2754 __wsum csum); 2755 2756 static inline void *__skb_header_pointer(const struct sk_buff *skb, int offset, 2757 int len, void *data, int hlen, void *buffer) 2758 { 2759 if (hlen - offset >= len) 2760 return data + offset; 2761 2762 if (!skb || 2763 skb_copy_bits(skb, offset, buffer, len) < 0) 2764 return NULL; 2765 2766 return buffer; 2767 } 2768 2769 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset, 2770 int len, void *buffer) 2771 { 2772 return __skb_header_pointer(skb, offset, len, skb->data, 2773 skb_headlen(skb), buffer); 2774 } 2775 2776 /** 2777 * skb_needs_linearize - check if we need to linearize a given skb 2778 * depending on the given device features. 2779 * @skb: socket buffer to check 2780 * @features: net device features 2781 * 2782 * Returns true if either: 2783 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 2784 * 2. skb is fragmented and the device does not support SG. 2785 */ 2786 static inline bool skb_needs_linearize(struct sk_buff *skb, 2787 netdev_features_t features) 2788 { 2789 return skb_is_nonlinear(skb) && 2790 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 2791 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 2792 } 2793 2794 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 2795 void *to, 2796 const unsigned int len) 2797 { 2798 memcpy(to, skb->data, len); 2799 } 2800 2801 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 2802 const int offset, void *to, 2803 const unsigned int len) 2804 { 2805 memcpy(to, skb->data + offset, len); 2806 } 2807 2808 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 2809 const void *from, 2810 const unsigned int len) 2811 { 2812 memcpy(skb->data, from, len); 2813 } 2814 2815 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 2816 const int offset, 2817 const void *from, 2818 const unsigned int len) 2819 { 2820 memcpy(skb->data + offset, from, len); 2821 } 2822 2823 void skb_init(void); 2824 2825 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 2826 { 2827 return skb->tstamp; 2828 } 2829 2830 /** 2831 * skb_get_timestamp - get timestamp from a skb 2832 * @skb: skb to get stamp from 2833 * @stamp: pointer to struct timeval to store stamp in 2834 * 2835 * Timestamps are stored in the skb as offsets to a base timestamp. 2836 * This function converts the offset back to a struct timeval and stores 2837 * it in stamp. 2838 */ 2839 static inline void skb_get_timestamp(const struct sk_buff *skb, 2840 struct timeval *stamp) 2841 { 2842 *stamp = ktime_to_timeval(skb->tstamp); 2843 } 2844 2845 static inline void skb_get_timestampns(const struct sk_buff *skb, 2846 struct timespec *stamp) 2847 { 2848 *stamp = ktime_to_timespec(skb->tstamp); 2849 } 2850 2851 static inline void __net_timestamp(struct sk_buff *skb) 2852 { 2853 skb->tstamp = ktime_get_real(); 2854 } 2855 2856 static inline ktime_t net_timedelta(ktime_t t) 2857 { 2858 return ktime_sub(ktime_get_real(), t); 2859 } 2860 2861 static inline ktime_t net_invalid_timestamp(void) 2862 { 2863 return ktime_set(0, 0); 2864 } 2865 2866 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 2867 2868 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 2869 2870 void skb_clone_tx_timestamp(struct sk_buff *skb); 2871 bool skb_defer_rx_timestamp(struct sk_buff *skb); 2872 2873 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 2874 2875 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 2876 { 2877 } 2878 2879 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 2880 { 2881 return false; 2882 } 2883 2884 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 2885 2886 /** 2887 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 2888 * 2889 * PHY drivers may accept clones of transmitted packets for 2890 * timestamping via their phy_driver.txtstamp method. These drivers 2891 * must call this function to return the skb back to the stack, with 2892 * or without a timestamp. 2893 * 2894 * @skb: clone of the the original outgoing packet 2895 * @hwtstamps: hardware time stamps, may be NULL if not available 2896 * 2897 */ 2898 void skb_complete_tx_timestamp(struct sk_buff *skb, 2899 struct skb_shared_hwtstamps *hwtstamps); 2900 2901 void __skb_tstamp_tx(struct sk_buff *orig_skb, 2902 struct skb_shared_hwtstamps *hwtstamps, 2903 struct sock *sk, int tstype); 2904 2905 /** 2906 * skb_tstamp_tx - queue clone of skb with send time stamps 2907 * @orig_skb: the original outgoing packet 2908 * @hwtstamps: hardware time stamps, may be NULL if not available 2909 * 2910 * If the skb has a socket associated, then this function clones the 2911 * skb (thus sharing the actual data and optional structures), stores 2912 * the optional hardware time stamping information (if non NULL) or 2913 * generates a software time stamp (otherwise), then queues the clone 2914 * to the error queue of the socket. Errors are silently ignored. 2915 */ 2916 void skb_tstamp_tx(struct sk_buff *orig_skb, 2917 struct skb_shared_hwtstamps *hwtstamps); 2918 2919 static inline void sw_tx_timestamp(struct sk_buff *skb) 2920 { 2921 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 2922 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 2923 skb_tstamp_tx(skb, NULL); 2924 } 2925 2926 /** 2927 * skb_tx_timestamp() - Driver hook for transmit timestamping 2928 * 2929 * Ethernet MAC Drivers should call this function in their hard_xmit() 2930 * function immediately before giving the sk_buff to the MAC hardware. 2931 * 2932 * Specifically, one should make absolutely sure that this function is 2933 * called before TX completion of this packet can trigger. Otherwise 2934 * the packet could potentially already be freed. 2935 * 2936 * @skb: A socket buffer. 2937 */ 2938 static inline void skb_tx_timestamp(struct sk_buff *skb) 2939 { 2940 skb_clone_tx_timestamp(skb); 2941 sw_tx_timestamp(skb); 2942 } 2943 2944 /** 2945 * skb_complete_wifi_ack - deliver skb with wifi status 2946 * 2947 * @skb: the original outgoing packet 2948 * @acked: ack status 2949 * 2950 */ 2951 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 2952 2953 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 2954 __sum16 __skb_checksum_complete(struct sk_buff *skb); 2955 2956 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 2957 { 2958 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 2959 skb->csum_valid || 2960 (skb->ip_summed == CHECKSUM_PARTIAL && 2961 skb_checksum_start_offset(skb) >= 0)); 2962 } 2963 2964 /** 2965 * skb_checksum_complete - Calculate checksum of an entire packet 2966 * @skb: packet to process 2967 * 2968 * This function calculates the checksum over the entire packet plus 2969 * the value of skb->csum. The latter can be used to supply the 2970 * checksum of a pseudo header as used by TCP/UDP. It returns the 2971 * checksum. 2972 * 2973 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 2974 * this function can be used to verify that checksum on received 2975 * packets. In that case the function should return zero if the 2976 * checksum is correct. In particular, this function will return zero 2977 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 2978 * hardware has already verified the correctness of the checksum. 2979 */ 2980 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 2981 { 2982 return skb_csum_unnecessary(skb) ? 2983 0 : __skb_checksum_complete(skb); 2984 } 2985 2986 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 2987 { 2988 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 2989 if (skb->csum_level == 0) 2990 skb->ip_summed = CHECKSUM_NONE; 2991 else 2992 skb->csum_level--; 2993 } 2994 } 2995 2996 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 2997 { 2998 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 2999 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3000 skb->csum_level++; 3001 } else if (skb->ip_summed == CHECKSUM_NONE) { 3002 skb->ip_summed = CHECKSUM_UNNECESSARY; 3003 skb->csum_level = 0; 3004 } 3005 } 3006 3007 static inline void __skb_mark_checksum_bad(struct sk_buff *skb) 3008 { 3009 /* Mark current checksum as bad (typically called from GRO 3010 * path). In the case that ip_summed is CHECKSUM_NONE 3011 * this must be the first checksum encountered in the packet. 3012 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first 3013 * checksum after the last one validated. For UDP, a zero 3014 * checksum can not be marked as bad. 3015 */ 3016 3017 if (skb->ip_summed == CHECKSUM_NONE || 3018 skb->ip_summed == CHECKSUM_UNNECESSARY) 3019 skb->csum_bad = 1; 3020 } 3021 3022 /* Check if we need to perform checksum complete validation. 3023 * 3024 * Returns true if checksum complete is needed, false otherwise 3025 * (either checksum is unnecessary or zero checksum is allowed). 3026 */ 3027 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3028 bool zero_okay, 3029 __sum16 check) 3030 { 3031 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3032 skb->csum_valid = 1; 3033 __skb_decr_checksum_unnecessary(skb); 3034 return false; 3035 } 3036 3037 return true; 3038 } 3039 3040 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3041 * in checksum_init. 3042 */ 3043 #define CHECKSUM_BREAK 76 3044 3045 /* Unset checksum-complete 3046 * 3047 * Unset checksum complete can be done when packet is being modified 3048 * (uncompressed for instance) and checksum-complete value is 3049 * invalidated. 3050 */ 3051 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3052 { 3053 if (skb->ip_summed == CHECKSUM_COMPLETE) 3054 skb->ip_summed = CHECKSUM_NONE; 3055 } 3056 3057 /* Validate (init) checksum based on checksum complete. 3058 * 3059 * Return values: 3060 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3061 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3062 * checksum is stored in skb->csum for use in __skb_checksum_complete 3063 * non-zero: value of invalid checksum 3064 * 3065 */ 3066 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3067 bool complete, 3068 __wsum psum) 3069 { 3070 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3071 if (!csum_fold(csum_add(psum, skb->csum))) { 3072 skb->csum_valid = 1; 3073 return 0; 3074 } 3075 } else if (skb->csum_bad) { 3076 /* ip_summed == CHECKSUM_NONE in this case */ 3077 return 1; 3078 } 3079 3080 skb->csum = psum; 3081 3082 if (complete || skb->len <= CHECKSUM_BREAK) { 3083 __sum16 csum; 3084 3085 csum = __skb_checksum_complete(skb); 3086 skb->csum_valid = !csum; 3087 return csum; 3088 } 3089 3090 return 0; 3091 } 3092 3093 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3094 { 3095 return 0; 3096 } 3097 3098 /* Perform checksum validate (init). Note that this is a macro since we only 3099 * want to calculate the pseudo header which is an input function if necessary. 3100 * First we try to validate without any computation (checksum unnecessary) and 3101 * then calculate based on checksum complete calling the function to compute 3102 * pseudo header. 3103 * 3104 * Return values: 3105 * 0: checksum is validated or try to in skb_checksum_complete 3106 * non-zero: value of invalid checksum 3107 */ 3108 #define __skb_checksum_validate(skb, proto, complete, \ 3109 zero_okay, check, compute_pseudo) \ 3110 ({ \ 3111 __sum16 __ret = 0; \ 3112 skb->csum_valid = 0; \ 3113 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3114 __ret = __skb_checksum_validate_complete(skb, \ 3115 complete, compute_pseudo(skb, proto)); \ 3116 __ret; \ 3117 }) 3118 3119 #define skb_checksum_init(skb, proto, compute_pseudo) \ 3120 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3121 3122 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3123 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3124 3125 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 3126 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3127 3128 #define skb_checksum_validate_zero_check(skb, proto, check, \ 3129 compute_pseudo) \ 3130 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3131 3132 #define skb_checksum_simple_validate(skb) \ 3133 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3134 3135 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3136 { 3137 return (skb->ip_summed == CHECKSUM_NONE && 3138 skb->csum_valid && !skb->csum_bad); 3139 } 3140 3141 static inline void __skb_checksum_convert(struct sk_buff *skb, 3142 __sum16 check, __wsum pseudo) 3143 { 3144 skb->csum = ~pseudo; 3145 skb->ip_summed = CHECKSUM_COMPLETE; 3146 } 3147 3148 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3149 do { \ 3150 if (__skb_checksum_convert_check(skb)) \ 3151 __skb_checksum_convert(skb, check, \ 3152 compute_pseudo(skb, proto)); \ 3153 } while (0) 3154 3155 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3156 u16 start, u16 offset) 3157 { 3158 skb->ip_summed = CHECKSUM_PARTIAL; 3159 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3160 skb->csum_offset = offset - start; 3161 } 3162 3163 /* Update skbuf and packet to reflect the remote checksum offload operation. 3164 * When called, ptr indicates the starting point for skb->csum when 3165 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3166 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3167 */ 3168 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3169 int start, int offset, bool nopartial) 3170 { 3171 __wsum delta; 3172 3173 if (!nopartial) { 3174 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3175 return; 3176 } 3177 3178 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3179 __skb_checksum_complete(skb); 3180 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3181 } 3182 3183 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3184 3185 /* Adjust skb->csum since we changed the packet */ 3186 skb->csum = csum_add(skb->csum, delta); 3187 } 3188 3189 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3190 void nf_conntrack_destroy(struct nf_conntrack *nfct); 3191 static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3192 { 3193 if (nfct && atomic_dec_and_test(&nfct->use)) 3194 nf_conntrack_destroy(nfct); 3195 } 3196 static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3197 { 3198 if (nfct) 3199 atomic_inc(&nfct->use); 3200 } 3201 #endif 3202 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3203 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3204 { 3205 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 3206 kfree(nf_bridge); 3207 } 3208 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3209 { 3210 if (nf_bridge) 3211 atomic_inc(&nf_bridge->use); 3212 } 3213 #endif /* CONFIG_BRIDGE_NETFILTER */ 3214 static inline void nf_reset(struct sk_buff *skb) 3215 { 3216 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3217 nf_conntrack_put(skb->nfct); 3218 skb->nfct = NULL; 3219 #endif 3220 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3221 nf_bridge_put(skb->nf_bridge); 3222 skb->nf_bridge = NULL; 3223 #endif 3224 } 3225 3226 static inline void nf_reset_trace(struct sk_buff *skb) 3227 { 3228 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3229 skb->nf_trace = 0; 3230 #endif 3231 } 3232 3233 static inline void ipvs_reset(struct sk_buff *skb) 3234 { 3235 #if IS_ENABLED(CONFIG_IP_VS) 3236 skb->ipvs_property = 0; 3237 #endif 3238 } 3239 3240 /* Note: This doesn't put any conntrack and bridge info in dst. */ 3241 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3242 bool copy) 3243 { 3244 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3245 dst->nfct = src->nfct; 3246 nf_conntrack_get(src->nfct); 3247 if (copy) 3248 dst->nfctinfo = src->nfctinfo; 3249 #endif 3250 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3251 dst->nf_bridge = src->nf_bridge; 3252 nf_bridge_get(src->nf_bridge); 3253 #endif 3254 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3255 if (copy) 3256 dst->nf_trace = src->nf_trace; 3257 #endif 3258 } 3259 3260 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3261 { 3262 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3263 nf_conntrack_put(dst->nfct); 3264 #endif 3265 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3266 nf_bridge_put(dst->nf_bridge); 3267 #endif 3268 __nf_copy(dst, src, true); 3269 } 3270 3271 #ifdef CONFIG_NETWORK_SECMARK 3272 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3273 { 3274 to->secmark = from->secmark; 3275 } 3276 3277 static inline void skb_init_secmark(struct sk_buff *skb) 3278 { 3279 skb->secmark = 0; 3280 } 3281 #else 3282 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3283 { } 3284 3285 static inline void skb_init_secmark(struct sk_buff *skb) 3286 { } 3287 #endif 3288 3289 static inline bool skb_irq_freeable(const struct sk_buff *skb) 3290 { 3291 return !skb->destructor && 3292 #if IS_ENABLED(CONFIG_XFRM) 3293 !skb->sp && 3294 #endif 3295 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 3296 !skb->nfct && 3297 #endif 3298 !skb->_skb_refdst && 3299 !skb_has_frag_list(skb); 3300 } 3301 3302 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3303 { 3304 skb->queue_mapping = queue_mapping; 3305 } 3306 3307 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3308 { 3309 return skb->queue_mapping; 3310 } 3311 3312 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3313 { 3314 to->queue_mapping = from->queue_mapping; 3315 } 3316 3317 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3318 { 3319 skb->queue_mapping = rx_queue + 1; 3320 } 3321 3322 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3323 { 3324 return skb->queue_mapping - 1; 3325 } 3326 3327 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3328 { 3329 return skb->queue_mapping != 0; 3330 } 3331 3332 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, 3333 unsigned int num_tx_queues); 3334 3335 static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3336 { 3337 #ifdef CONFIG_XFRM 3338 return skb->sp; 3339 #else 3340 return NULL; 3341 #endif 3342 } 3343 3344 /* Keeps track of mac header offset relative to skb->head. 3345 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3346 * For non-tunnel skb it points to skb_mac_header() and for 3347 * tunnel skb it points to outer mac header. 3348 * Keeps track of level of encapsulation of network headers. 3349 */ 3350 struct skb_gso_cb { 3351 int mac_offset; 3352 int encap_level; 3353 __u16 csum_start; 3354 }; 3355 #define SKB_SGO_CB_OFFSET 32 3356 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 3357 3358 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3359 { 3360 return (skb_mac_header(inner_skb) - inner_skb->head) - 3361 SKB_GSO_CB(inner_skb)->mac_offset; 3362 } 3363 3364 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3365 { 3366 int new_headroom, headroom; 3367 int ret; 3368 3369 headroom = skb_headroom(skb); 3370 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3371 if (ret) 3372 return ret; 3373 3374 new_headroom = skb_headroom(skb); 3375 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3376 return 0; 3377 } 3378 3379 /* Compute the checksum for a gso segment. First compute the checksum value 3380 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3381 * then add in skb->csum (checksum from csum_start to end of packet). 3382 * skb->csum and csum_start are then updated to reflect the checksum of the 3383 * resultant packet starting from the transport header-- the resultant checksum 3384 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3385 * header. 3386 */ 3387 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3388 { 3389 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) - 3390 skb_transport_offset(skb); 3391 __u16 csum; 3392 3393 csum = csum_fold(csum_partial(skb_transport_header(skb), 3394 plen, skb->csum)); 3395 skb->csum = res; 3396 SKB_GSO_CB(skb)->csum_start -= plen; 3397 3398 return csum; 3399 } 3400 3401 static inline bool skb_is_gso(const struct sk_buff *skb) 3402 { 3403 return skb_shinfo(skb)->gso_size; 3404 } 3405 3406 /* Note: Should be called only if skb_is_gso(skb) is true */ 3407 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3408 { 3409 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3410 } 3411 3412 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3413 3414 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3415 { 3416 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3417 * wanted then gso_type will be set. */ 3418 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3419 3420 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3421 unlikely(shinfo->gso_type == 0)) { 3422 __skb_warn_lro_forwarding(skb); 3423 return true; 3424 } 3425 return false; 3426 } 3427 3428 static inline void skb_forward_csum(struct sk_buff *skb) 3429 { 3430 /* Unfortunately we don't support this one. Any brave souls? */ 3431 if (skb->ip_summed == CHECKSUM_COMPLETE) 3432 skb->ip_summed = CHECKSUM_NONE; 3433 } 3434 3435 /** 3436 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3437 * @skb: skb to check 3438 * 3439 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3440 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3441 * use this helper, to document places where we make this assertion. 3442 */ 3443 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3444 { 3445 #ifdef DEBUG 3446 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3447 #endif 3448 } 3449 3450 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3451 3452 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3453 3454 u32 skb_get_poff(const struct sk_buff *skb); 3455 u32 __skb_get_poff(const struct sk_buff *skb, void *data, 3456 const struct flow_keys *keys, int hlen); 3457 3458 /** 3459 * skb_head_is_locked - Determine if the skb->head is locked down 3460 * @skb: skb to check 3461 * 3462 * The head on skbs build around a head frag can be removed if they are 3463 * not cloned. This function returns true if the skb head is locked down 3464 * due to either being allocated via kmalloc, or by being a clone with 3465 * multiple references to the head. 3466 */ 3467 static inline bool skb_head_is_locked(const struct sk_buff *skb) 3468 { 3469 return !skb->head_frag || skb_cloned(skb); 3470 } 3471 3472 /** 3473 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3474 * 3475 * @skb: GSO skb 3476 * 3477 * skb_gso_network_seglen is used to determine the real size of the 3478 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3479 * 3480 * The MAC/L2 header is not accounted for. 3481 */ 3482 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 3483 { 3484 unsigned int hdr_len = skb_transport_header(skb) - 3485 skb_network_header(skb); 3486 return hdr_len + skb_gso_transport_seglen(skb); 3487 } 3488 #endif /* __KERNEL__ */ 3489 #endif /* _LINUX_SKBUFF_H */ 3490
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.