1 /* 2 * Definitions for the 'struct sk_buff' memory handlers. 3 * 4 * Authors: 5 * Alan Cox, <[email protected]> 6 * Florian La Roche, <[email protected]> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 */ 13 14 #ifndef _LINUX_SKBUFF_H 15 #define _LINUX_SKBUFF_H 16 17 #include <linux/kernel.h> 18 #include <linux/kmemcheck.h> 19 #include <linux/compiler.h> 20 #include <linux/time.h> 21 #include <linux/bug.h> 22 #include <linux/cache.h> 23 #include <linux/rbtree.h> 24 #include <linux/socket.h> 25 26 #include <linux/atomic.h> 27 #include <asm/types.h> 28 #include <linux/spinlock.h> 29 #include <linux/net.h> 30 #include <linux/textsearch.h> 31 #include <net/checksum.h> 32 #include <linux/rcupdate.h> 33 #include <linux/hrtimer.h> 34 #include <linux/dma-mapping.h> 35 #include <linux/netdev_features.h> 36 #include <linux/sched.h> 37 #include <net/flow_dissector.h> 38 #include <linux/splice.h> 39 #include <linux/in6.h> 40 #include <net/flow.h> 41 42 /* The interface for checksum offload between the stack and networking drivers 43 * is as follows... 44 * 45 * A. IP checksum related features 46 * 47 * Drivers advertise checksum offload capabilities in the features of a device. 48 * From the stack's point of view these are capabilities offered by the driver, 49 * a driver typically only advertises features that it is capable of offloading 50 * to its device. 51 * 52 * The checksum related features are: 53 * 54 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one 55 * IP (one's complement) checksum for any combination 56 * of protocols or protocol layering. The checksum is 57 * computed and set in a packet per the CHECKSUM_PARTIAL 58 * interface (see below). 59 * 60 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain 61 * TCP or UDP packets over IPv4. These are specifically 62 * unencapsulated packets of the form IPv4|TCP or 63 * IPv4|UDP where the Protocol field in the IPv4 header 64 * is TCP or UDP. The IPv4 header may contain IP options 65 * This feature cannot be set in features for a device 66 * with NETIF_F_HW_CSUM also set. This feature is being 67 * DEPRECATED (see below). 68 * 69 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain 70 * TCP or UDP packets over IPv6. These are specifically 71 * unencapsulated packets of the form IPv6|TCP or 72 * IPv4|UDP where the Next Header field in the IPv6 73 * header is either TCP or UDP. IPv6 extension headers 74 * are not supported with this feature. This feature 75 * cannot be set in features for a device with 76 * NETIF_F_HW_CSUM also set. This feature is being 77 * DEPRECATED (see below). 78 * 79 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload. 80 * This flag is used only used to disable the RX checksum 81 * feature for a device. The stack will accept receive 82 * checksum indication in packets received on a device 83 * regardless of whether NETIF_F_RXCSUM is set. 84 * 85 * B. Checksumming of received packets by device. Indication of checksum 86 * verification is in set skb->ip_summed. Possible values are: 87 * 88 * CHECKSUM_NONE: 89 * 90 * Device did not checksum this packet e.g. due to lack of capabilities. 91 * The packet contains full (though not verified) checksum in packet but 92 * not in skb->csum. Thus, skb->csum is undefined in this case. 93 * 94 * CHECKSUM_UNNECESSARY: 95 * 96 * The hardware you're dealing with doesn't calculate the full checksum 97 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 98 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY 99 * if their checksums are okay. skb->csum is still undefined in this case 100 * though. A driver or device must never modify the checksum field in the 101 * packet even if checksum is verified. 102 * 103 * CHECKSUM_UNNECESSARY is applicable to following protocols: 104 * TCP: IPv6 and IPv4. 105 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 106 * zero UDP checksum for either IPv4 or IPv6, the networking stack 107 * may perform further validation in this case. 108 * GRE: only if the checksum is present in the header. 109 * SCTP: indicates the CRC in SCTP header has been validated. 110 * 111 * skb->csum_level indicates the number of consecutive checksums found in 112 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY. 113 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 114 * and a device is able to verify the checksums for UDP (possibly zero), 115 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to 116 * two. If the device were only able to verify the UDP checksum and not 117 * GRE, either because it doesn't support GRE checksum of because GRE 118 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 119 * not considered in this case). 120 * 121 * CHECKSUM_COMPLETE: 122 * 123 * This is the most generic way. The device supplied checksum of the _whole_ 124 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the 125 * hardware doesn't need to parse L3/L4 headers to implement this. 126 * 127 * Note: Even if device supports only some protocols, but is able to produce 128 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 129 * 130 * CHECKSUM_PARTIAL: 131 * 132 * A checksum is set up to be offloaded to a device as described in the 133 * output description for CHECKSUM_PARTIAL. This may occur on a packet 134 * received directly from another Linux OS, e.g., a virtualized Linux kernel 135 * on the same host, or it may be set in the input path in GRO or remote 136 * checksum offload. For the purposes of checksum verification, the checksum 137 * referred to by skb->csum_start + skb->csum_offset and any preceding 138 * checksums in the packet are considered verified. Any checksums in the 139 * packet that are after the checksum being offloaded are not considered to 140 * be verified. 141 * 142 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload 143 * in the skb->ip_summed for a packet. Values are: 144 * 145 * CHECKSUM_PARTIAL: 146 * 147 * The driver is required to checksum the packet as seen by hard_start_xmit() 148 * from skb->csum_start up to the end, and to record/write the checksum at 149 * offset skb->csum_start + skb->csum_offset. A driver may verify that the 150 * csum_start and csum_offset values are valid values given the length and 151 * offset of the packet, however they should not attempt to validate that the 152 * checksum refers to a legitimate transport layer checksum-- it is the 153 * purview of the stack to validate that csum_start and csum_offset are set 154 * correctly. 155 * 156 * When the stack requests checksum offload for a packet, the driver MUST 157 * ensure that the checksum is set correctly. A driver can either offload the 158 * checksum calculation to the device, or call skb_checksum_help (in the case 159 * that the device does not support offload for a particular checksum). 160 * 161 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of 162 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate 163 * checksum offload capability. If a device has limited checksum capabilities 164 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as 165 * described above) a helper function can be called to resolve 166 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper 167 * function takes a spec argument that describes the protocol layer that is 168 * supported for checksum offload and can be called for each packet. If a 169 * packet does not match the specification for offload, skb_checksum_help 170 * is called to resolve the checksum. 171 * 172 * CHECKSUM_NONE: 173 * 174 * The skb was already checksummed by the protocol, or a checksum is not 175 * required. 176 * 177 * CHECKSUM_UNNECESSARY: 178 * 179 * This has the same meaning on as CHECKSUM_NONE for checksum offload on 180 * output. 181 * 182 * CHECKSUM_COMPLETE: 183 * Not used in checksum output. If a driver observes a packet with this value 184 * set in skbuff, if should treat as CHECKSUM_NONE being set. 185 * 186 * D. Non-IP checksum (CRC) offloads 187 * 188 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of 189 * offloading the SCTP CRC in a packet. To perform this offload the stack 190 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 191 * accordingly. Note the there is no indication in the skbuff that the 192 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports 193 * both IP checksum offload and SCTP CRC offload must verify which offload 194 * is configured for a packet presumably by inspecting packet headers. 195 * 196 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of 197 * offloading the FCOE CRC in a packet. To perform this offload the stack 198 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 199 * accordingly. Note the there is no indication in the skbuff that the 200 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports 201 * both IP checksum offload and FCOE CRC offload must verify which offload 202 * is configured for a packet presumably by inspecting packet headers. 203 * 204 * E. Checksumming on output with GSO. 205 * 206 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload 207 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 208 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as 209 * part of the GSO operation is implied. If a checksum is being offloaded 210 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset 211 * are set to refer to the outermost checksum being offload (two offloaded 212 * checksums are possible with UDP encapsulation). 213 */ 214 215 /* Don't change this without changing skb_csum_unnecessary! */ 216 #define CHECKSUM_NONE 0 217 #define CHECKSUM_UNNECESSARY 1 218 #define CHECKSUM_COMPLETE 2 219 #define CHECKSUM_PARTIAL 3 220 221 /* Maximum value in skb->csum_level */ 222 #define SKB_MAX_CSUM_LEVEL 3 223 224 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 225 #define SKB_WITH_OVERHEAD(X) \ 226 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 227 #define SKB_MAX_ORDER(X, ORDER) \ 228 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 229 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 230 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 231 232 /* return minimum truesize of one skb containing X bytes of data */ 233 #define SKB_TRUESIZE(X) ((X) + \ 234 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 235 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 236 237 struct net_device; 238 struct scatterlist; 239 struct pipe_inode_info; 240 struct iov_iter; 241 struct napi_struct; 242 243 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 244 struct nf_conntrack { 245 atomic_t use; 246 }; 247 #endif 248 249 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 250 struct nf_bridge_info { 251 atomic_t use; 252 enum { 253 BRNF_PROTO_UNCHANGED, 254 BRNF_PROTO_8021Q, 255 BRNF_PROTO_PPPOE 256 } orig_proto:8; 257 u8 pkt_otherhost:1; 258 u8 in_prerouting:1; 259 u8 bridged_dnat:1; 260 __u16 frag_max_size; 261 struct net_device *physindev; 262 263 /* always valid & non-NULL from FORWARD on, for physdev match */ 264 struct net_device *physoutdev; 265 union { 266 /* prerouting: detect dnat in orig/reply direction */ 267 __be32 ipv4_daddr; 268 struct in6_addr ipv6_daddr; 269 270 /* after prerouting + nat detected: store original source 271 * mac since neigh resolution overwrites it, only used while 272 * skb is out in neigh layer. 273 */ 274 char neigh_header[8]; 275 }; 276 }; 277 #endif 278 279 struct sk_buff_head { 280 /* These two members must be first. */ 281 struct sk_buff *next; 282 struct sk_buff *prev; 283 284 __u32 qlen; 285 spinlock_t lock; 286 }; 287 288 struct sk_buff; 289 290 /* To allow 64K frame to be packed as single skb without frag_list we 291 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 292 * buffers which do not start on a page boundary. 293 * 294 * Since GRO uses frags we allocate at least 16 regardless of page 295 * size. 296 */ 297 #if (65536/PAGE_SIZE + 1) < 16 298 #define MAX_SKB_FRAGS 16UL 299 #else 300 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 301 #endif 302 extern int sysctl_max_skb_frags; 303 304 typedef struct skb_frag_struct skb_frag_t; 305 306 struct skb_frag_struct { 307 struct { 308 struct page *p; 309 } page; 310 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 311 __u32 page_offset; 312 __u32 size; 313 #else 314 __u16 page_offset; 315 __u16 size; 316 #endif 317 }; 318 319 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 320 { 321 return frag->size; 322 } 323 324 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 325 { 326 frag->size = size; 327 } 328 329 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 330 { 331 frag->size += delta; 332 } 333 334 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 335 { 336 frag->size -= delta; 337 } 338 339 #define HAVE_HW_TIME_STAMP 340 341 /** 342 * struct skb_shared_hwtstamps - hardware time stamps 343 * @hwtstamp: hardware time stamp transformed into duration 344 * since arbitrary point in time 345 * 346 * Software time stamps generated by ktime_get_real() are stored in 347 * skb->tstamp. 348 * 349 * hwtstamps can only be compared against other hwtstamps from 350 * the same device. 351 * 352 * This structure is attached to packets as part of the 353 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 354 */ 355 struct skb_shared_hwtstamps { 356 ktime_t hwtstamp; 357 }; 358 359 /* Definitions for tx_flags in struct skb_shared_info */ 360 enum { 361 /* generate hardware time stamp */ 362 SKBTX_HW_TSTAMP = 1 << 0, 363 364 /* generate software time stamp when queueing packet to NIC */ 365 SKBTX_SW_TSTAMP = 1 << 1, 366 367 /* device driver is going to provide hardware time stamp */ 368 SKBTX_IN_PROGRESS = 1 << 2, 369 370 /* device driver supports TX zero-copy buffers */ 371 SKBTX_DEV_ZEROCOPY = 1 << 3, 372 373 /* generate wifi status information (where possible) */ 374 SKBTX_WIFI_STATUS = 1 << 4, 375 376 /* This indicates at least one fragment might be overwritten 377 * (as in vmsplice(), sendfile() ...) 378 * If we need to compute a TX checksum, we'll need to copy 379 * all frags to avoid possible bad checksum 380 */ 381 SKBTX_SHARED_FRAG = 1 << 5, 382 383 /* generate software time stamp when entering packet scheduling */ 384 SKBTX_SCHED_TSTAMP = 1 << 6, 385 386 /* generate software timestamp on peer data acknowledgment */ 387 SKBTX_ACK_TSTAMP = 1 << 7, 388 }; 389 390 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 391 SKBTX_SCHED_TSTAMP | \ 392 SKBTX_ACK_TSTAMP) 393 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 394 395 /* 396 * The callback notifies userspace to release buffers when skb DMA is done in 397 * lower device, the skb last reference should be 0 when calling this. 398 * The zerocopy_success argument is true if zero copy transmit occurred, 399 * false on data copy or out of memory error caused by data copy attempt. 400 * The ctx field is used to track device context. 401 * The desc field is used to track userspace buffer index. 402 */ 403 struct ubuf_info { 404 void (*callback)(struct ubuf_info *, bool zerocopy_success); 405 void *ctx; 406 unsigned long desc; 407 }; 408 409 /* This data is invariant across clones and lives at 410 * the end of the header data, ie. at skb->end. 411 */ 412 struct skb_shared_info { 413 unsigned char nr_frags; 414 __u8 tx_flags; 415 unsigned short gso_size; 416 /* Warning: this field is not always filled in (UFO)! */ 417 unsigned short gso_segs; 418 unsigned short gso_type; 419 struct sk_buff *frag_list; 420 struct skb_shared_hwtstamps hwtstamps; 421 u32 tskey; 422 __be32 ip6_frag_id; 423 424 /* 425 * Warning : all fields before dataref are cleared in __alloc_skb() 426 */ 427 atomic_t dataref; 428 429 /* Intermediate layers must ensure that destructor_arg 430 * remains valid until skb destructor */ 431 void * destructor_arg; 432 433 /* must be last field, see pskb_expand_head() */ 434 skb_frag_t frags[MAX_SKB_FRAGS]; 435 }; 436 437 /* We divide dataref into two halves. The higher 16 bits hold references 438 * to the payload part of skb->data. The lower 16 bits hold references to 439 * the entire skb->data. A clone of a headerless skb holds the length of 440 * the header in skb->hdr_len. 441 * 442 * All users must obey the rule that the skb->data reference count must be 443 * greater than or equal to the payload reference count. 444 * 445 * Holding a reference to the payload part means that the user does not 446 * care about modifications to the header part of skb->data. 447 */ 448 #define SKB_DATAREF_SHIFT 16 449 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 450 451 452 enum { 453 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 454 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 455 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 456 }; 457 458 enum { 459 SKB_GSO_TCPV4 = 1 << 0, 460 SKB_GSO_UDP = 1 << 1, 461 462 /* This indicates the skb is from an untrusted source. */ 463 SKB_GSO_DODGY = 1 << 2, 464 465 /* This indicates the tcp segment has CWR set. */ 466 SKB_GSO_TCP_ECN = 1 << 3, 467 468 SKB_GSO_TCPV6 = 1 << 4, 469 470 SKB_GSO_FCOE = 1 << 5, 471 472 SKB_GSO_GRE = 1 << 6, 473 474 SKB_GSO_GRE_CSUM = 1 << 7, 475 476 SKB_GSO_IPIP = 1 << 8, 477 478 SKB_GSO_SIT = 1 << 9, 479 480 SKB_GSO_UDP_TUNNEL = 1 << 10, 481 482 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 483 484 SKB_GSO_TUNNEL_REMCSUM = 1 << 12, 485 }; 486 487 #if BITS_PER_LONG > 32 488 #define NET_SKBUFF_DATA_USES_OFFSET 1 489 #endif 490 491 #ifdef NET_SKBUFF_DATA_USES_OFFSET 492 typedef unsigned int sk_buff_data_t; 493 #else 494 typedef unsigned char *sk_buff_data_t; 495 #endif 496 497 /** 498 * struct skb_mstamp - multi resolution time stamps 499 * @stamp_us: timestamp in us resolution 500 * @stamp_jiffies: timestamp in jiffies 501 */ 502 struct skb_mstamp { 503 union { 504 u64 v64; 505 struct { 506 u32 stamp_us; 507 u32 stamp_jiffies; 508 }; 509 }; 510 }; 511 512 /** 513 * skb_mstamp_get - get current timestamp 514 * @cl: place to store timestamps 515 */ 516 static inline void skb_mstamp_get(struct skb_mstamp *cl) 517 { 518 u64 val = local_clock(); 519 520 do_div(val, NSEC_PER_USEC); 521 cl->stamp_us = (u32)val; 522 cl->stamp_jiffies = (u32)jiffies; 523 } 524 525 /** 526 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp 527 * @t1: pointer to newest sample 528 * @t0: pointer to oldest sample 529 */ 530 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1, 531 const struct skb_mstamp *t0) 532 { 533 s32 delta_us = t1->stamp_us - t0->stamp_us; 534 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies; 535 536 /* If delta_us is negative, this might be because interval is too big, 537 * or local_clock() drift is too big : fallback using jiffies. 538 */ 539 if (delta_us <= 0 || 540 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ))) 541 542 delta_us = jiffies_to_usecs(delta_jiffies); 543 544 return delta_us; 545 } 546 547 static inline bool skb_mstamp_after(const struct skb_mstamp *t1, 548 const struct skb_mstamp *t0) 549 { 550 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies; 551 552 if (!diff) 553 diff = t1->stamp_us - t0->stamp_us; 554 return diff > 0; 555 } 556 557 /** 558 * struct sk_buff - socket buffer 559 * @next: Next buffer in list 560 * @prev: Previous buffer in list 561 * @tstamp: Time we arrived/left 562 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 563 * @sk: Socket we are owned by 564 * @dev: Device we arrived on/are leaving by 565 * @cb: Control buffer. Free for use by every layer. Put private vars here 566 * @_skb_refdst: destination entry (with norefcount bit) 567 * @sp: the security path, used for xfrm 568 * @len: Length of actual data 569 * @data_len: Data length 570 * @mac_len: Length of link layer header 571 * @hdr_len: writable header length of cloned skb 572 * @csum: Checksum (must include start/offset pair) 573 * @csum_start: Offset from skb->head where checksumming should start 574 * @csum_offset: Offset from csum_start where checksum should be stored 575 * @priority: Packet queueing priority 576 * @ignore_df: allow local fragmentation 577 * @cloned: Head may be cloned (check refcnt to be sure) 578 * @ip_summed: Driver fed us an IP checksum 579 * @nohdr: Payload reference only, must not modify header 580 * @nfctinfo: Relationship of this skb to the connection 581 * @pkt_type: Packet class 582 * @fclone: skbuff clone status 583 * @ipvs_property: skbuff is owned by ipvs 584 * @peeked: this packet has been seen already, so stats have been 585 * done for it, don't do them again 586 * @nf_trace: netfilter packet trace flag 587 * @protocol: Packet protocol from driver 588 * @destructor: Destruct function 589 * @nfct: Associated connection, if any 590 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 591 * @skb_iif: ifindex of device we arrived on 592 * @tc_index: Traffic control index 593 * @tc_verd: traffic control verdict 594 * @hash: the packet hash 595 * @queue_mapping: Queue mapping for multiqueue devices 596 * @xmit_more: More SKBs are pending for this queue 597 * @ndisc_nodetype: router type (from link layer) 598 * @ooo_okay: allow the mapping of a socket to a queue to be changed 599 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 600 * ports. 601 * @sw_hash: indicates hash was computed in software stack 602 * @wifi_acked_valid: wifi_acked was set 603 * @wifi_acked: whether frame was acked on wifi or not 604 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 605 * @napi_id: id of the NAPI struct this skb came from 606 * @secmark: security marking 607 * @offload_fwd_mark: fwding offload mark 608 * @mark: Generic packet mark 609 * @vlan_proto: vlan encapsulation protocol 610 * @vlan_tci: vlan tag control information 611 * @inner_protocol: Protocol (encapsulation) 612 * @inner_transport_header: Inner transport layer header (encapsulation) 613 * @inner_network_header: Network layer header (encapsulation) 614 * @inner_mac_header: Link layer header (encapsulation) 615 * @transport_header: Transport layer header 616 * @network_header: Network layer header 617 * @mac_header: Link layer header 618 * @tail: Tail pointer 619 * @end: End pointer 620 * @head: Head of buffer 621 * @data: Data head pointer 622 * @truesize: Buffer size 623 * @users: User count - see {datagram,tcp}.c 624 */ 625 626 struct sk_buff { 627 union { 628 struct { 629 /* These two members must be first. */ 630 struct sk_buff *next; 631 struct sk_buff *prev; 632 633 union { 634 ktime_t tstamp; 635 struct skb_mstamp skb_mstamp; 636 }; 637 }; 638 struct rb_node rbnode; /* used in netem & tcp stack */ 639 }; 640 struct sock *sk; 641 struct net_device *dev; 642 643 /* 644 * This is the control buffer. It is free to use for every 645 * layer. Please put your private variables there. If you 646 * want to keep them across layers you have to do a skb_clone() 647 * first. This is owned by whoever has the skb queued ATM. 648 */ 649 char cb[48] __aligned(8); 650 651 unsigned long _skb_refdst; 652 void (*destructor)(struct sk_buff *skb); 653 #ifdef CONFIG_XFRM 654 struct sec_path *sp; 655 #endif 656 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 657 struct nf_conntrack *nfct; 658 #endif 659 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 660 struct nf_bridge_info *nf_bridge; 661 #endif 662 unsigned int len, 663 data_len; 664 __u16 mac_len, 665 hdr_len; 666 667 /* Following fields are _not_ copied in __copy_skb_header() 668 * Note that queue_mapping is here mostly to fill a hole. 669 */ 670 kmemcheck_bitfield_begin(flags1); 671 __u16 queue_mapping; 672 __u8 cloned:1, 673 nohdr:1, 674 fclone:2, 675 peeked:1, 676 head_frag:1, 677 xmit_more:1; 678 /* one bit hole */ 679 kmemcheck_bitfield_end(flags1); 680 681 /* fields enclosed in headers_start/headers_end are copied 682 * using a single memcpy() in __copy_skb_header() 683 */ 684 /* private: */ 685 __u32 headers_start[0]; 686 /* public: */ 687 688 /* if you move pkt_type around you also must adapt those constants */ 689 #ifdef __BIG_ENDIAN_BITFIELD 690 #define PKT_TYPE_MAX (7 << 5) 691 #else 692 #define PKT_TYPE_MAX 7 693 #endif 694 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 695 696 __u8 __pkt_type_offset[0]; 697 __u8 pkt_type:3; 698 __u8 pfmemalloc:1; 699 __u8 ignore_df:1; 700 __u8 nfctinfo:3; 701 702 __u8 nf_trace:1; 703 __u8 ip_summed:2; 704 __u8 ooo_okay:1; 705 __u8 l4_hash:1; 706 __u8 sw_hash:1; 707 __u8 wifi_acked_valid:1; 708 __u8 wifi_acked:1; 709 710 __u8 no_fcs:1; 711 /* Indicates the inner headers are valid in the skbuff. */ 712 __u8 encapsulation:1; 713 __u8 encap_hdr_csum:1; 714 __u8 csum_valid:1; 715 __u8 csum_complete_sw:1; 716 __u8 csum_level:2; 717 __u8 csum_bad:1; 718 719 #ifdef CONFIG_IPV6_NDISC_NODETYPE 720 __u8 ndisc_nodetype:2; 721 #endif 722 __u8 ipvs_property:1; 723 __u8 inner_protocol_type:1; 724 __u8 remcsum_offload:1; 725 /* 3 or 5 bit hole */ 726 727 #ifdef CONFIG_NET_SCHED 728 __u16 tc_index; /* traffic control index */ 729 #ifdef CONFIG_NET_CLS_ACT 730 __u16 tc_verd; /* traffic control verdict */ 731 #endif 732 #endif 733 734 union { 735 __wsum csum; 736 struct { 737 __u16 csum_start; 738 __u16 csum_offset; 739 }; 740 }; 741 __u32 priority; 742 int skb_iif; 743 __u32 hash; 744 __be16 vlan_proto; 745 __u16 vlan_tci; 746 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 747 union { 748 unsigned int napi_id; 749 unsigned int sender_cpu; 750 }; 751 #endif 752 union { 753 #ifdef CONFIG_NETWORK_SECMARK 754 __u32 secmark; 755 #endif 756 #ifdef CONFIG_NET_SWITCHDEV 757 __u32 offload_fwd_mark; 758 #endif 759 }; 760 761 union { 762 __u32 mark; 763 __u32 reserved_tailroom; 764 }; 765 766 union { 767 __be16 inner_protocol; 768 __u8 inner_ipproto; 769 }; 770 771 __u16 inner_transport_header; 772 __u16 inner_network_header; 773 __u16 inner_mac_header; 774 775 __be16 protocol; 776 __u16 transport_header; 777 __u16 network_header; 778 __u16 mac_header; 779 780 /* private: */ 781 __u32 headers_end[0]; 782 /* public: */ 783 784 /* These elements must be at the end, see alloc_skb() for details. */ 785 sk_buff_data_t tail; 786 sk_buff_data_t end; 787 unsigned char *head, 788 *data; 789 unsigned int truesize; 790 atomic_t users; 791 }; 792 793 #ifdef __KERNEL__ 794 /* 795 * Handling routines are only of interest to the kernel 796 */ 797 #include <linux/slab.h> 798 799 800 #define SKB_ALLOC_FCLONE 0x01 801 #define SKB_ALLOC_RX 0x02 802 #define SKB_ALLOC_NAPI 0x04 803 804 /* Returns true if the skb was allocated from PFMEMALLOC reserves */ 805 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 806 { 807 return unlikely(skb->pfmemalloc); 808 } 809 810 /* 811 * skb might have a dst pointer attached, refcounted or not. 812 * _skb_refdst low order bit is set if refcount was _not_ taken 813 */ 814 #define SKB_DST_NOREF 1UL 815 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 816 817 /** 818 * skb_dst - returns skb dst_entry 819 * @skb: buffer 820 * 821 * Returns skb dst_entry, regardless of reference taken or not. 822 */ 823 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 824 { 825 /* If refdst was not refcounted, check we still are in a 826 * rcu_read_lock section 827 */ 828 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 829 !rcu_read_lock_held() && 830 !rcu_read_lock_bh_held()); 831 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 832 } 833 834 /** 835 * skb_dst_set - sets skb dst 836 * @skb: buffer 837 * @dst: dst entry 838 * 839 * Sets skb dst, assuming a reference was taken on dst and should 840 * be released by skb_dst_drop() 841 */ 842 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 843 { 844 skb->_skb_refdst = (unsigned long)dst; 845 } 846 847 /** 848 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 849 * @skb: buffer 850 * @dst: dst entry 851 * 852 * Sets skb dst, assuming a reference was not taken on dst. 853 * If dst entry is cached, we do not take reference and dst_release 854 * will be avoided by refdst_drop. If dst entry is not cached, we take 855 * reference, so that last dst_release can destroy the dst immediately. 856 */ 857 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 858 { 859 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 860 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 861 } 862 863 /** 864 * skb_dst_is_noref - Test if skb dst isn't refcounted 865 * @skb: buffer 866 */ 867 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 868 { 869 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 870 } 871 872 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 873 { 874 return (struct rtable *)skb_dst(skb); 875 } 876 877 void kfree_skb(struct sk_buff *skb); 878 void kfree_skb_list(struct sk_buff *segs); 879 void skb_tx_error(struct sk_buff *skb); 880 void consume_skb(struct sk_buff *skb); 881 void __kfree_skb(struct sk_buff *skb); 882 extern struct kmem_cache *skbuff_head_cache; 883 884 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 885 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 886 bool *fragstolen, int *delta_truesize); 887 888 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 889 int node); 890 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 891 struct sk_buff *build_skb(void *data, unsigned int frag_size); 892 static inline struct sk_buff *alloc_skb(unsigned int size, 893 gfp_t priority) 894 { 895 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 896 } 897 898 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 899 unsigned long data_len, 900 int max_page_order, 901 int *errcode, 902 gfp_t gfp_mask); 903 904 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 905 struct sk_buff_fclones { 906 struct sk_buff skb1; 907 908 struct sk_buff skb2; 909 910 atomic_t fclone_ref; 911 }; 912 913 /** 914 * skb_fclone_busy - check if fclone is busy 915 * @skb: buffer 916 * 917 * Returns true if skb is a fast clone, and its clone is not freed. 918 * Some drivers call skb_orphan() in their ndo_start_xmit(), 919 * so we also check that this didnt happen. 920 */ 921 static inline bool skb_fclone_busy(const struct sock *sk, 922 const struct sk_buff *skb) 923 { 924 const struct sk_buff_fclones *fclones; 925 926 fclones = container_of(skb, struct sk_buff_fclones, skb1); 927 928 return skb->fclone == SKB_FCLONE_ORIG && 929 atomic_read(&fclones->fclone_ref) > 1 && 930 fclones->skb2.sk == sk; 931 } 932 933 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 934 gfp_t priority) 935 { 936 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 937 } 938 939 struct sk_buff *__alloc_skb_head(gfp_t priority, int node); 940 static inline struct sk_buff *alloc_skb_head(gfp_t priority) 941 { 942 return __alloc_skb_head(priority, -1); 943 } 944 945 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 946 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 947 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 948 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 949 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 950 gfp_t gfp_mask, bool fclone); 951 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 952 gfp_t gfp_mask) 953 { 954 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 955 } 956 957 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 958 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 959 unsigned int headroom); 960 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 961 int newtailroom, gfp_t priority); 962 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 963 int offset, int len); 964 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, 965 int len); 966 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 967 int skb_pad(struct sk_buff *skb, int pad); 968 #define dev_kfree_skb(a) consume_skb(a) 969 970 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 971 int getfrag(void *from, char *to, int offset, 972 int len, int odd, struct sk_buff *skb), 973 void *from, int length); 974 975 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 976 int offset, size_t size); 977 978 struct skb_seq_state { 979 __u32 lower_offset; 980 __u32 upper_offset; 981 __u32 frag_idx; 982 __u32 stepped_offset; 983 struct sk_buff *root_skb; 984 struct sk_buff *cur_skb; 985 __u8 *frag_data; 986 }; 987 988 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 989 unsigned int to, struct skb_seq_state *st); 990 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 991 struct skb_seq_state *st); 992 void skb_abort_seq_read(struct skb_seq_state *st); 993 994 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 995 unsigned int to, struct ts_config *config); 996 997 /* 998 * Packet hash types specify the type of hash in skb_set_hash. 999 * 1000 * Hash types refer to the protocol layer addresses which are used to 1001 * construct a packet's hash. The hashes are used to differentiate or identify 1002 * flows of the protocol layer for the hash type. Hash types are either 1003 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1004 * 1005 * Properties of hashes: 1006 * 1007 * 1) Two packets in different flows have different hash values 1008 * 2) Two packets in the same flow should have the same hash value 1009 * 1010 * A hash at a higher layer is considered to be more specific. A driver should 1011 * set the most specific hash possible. 1012 * 1013 * A driver cannot indicate a more specific hash than the layer at which a hash 1014 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1015 * 1016 * A driver may indicate a hash level which is less specific than the 1017 * actual layer the hash was computed on. For instance, a hash computed 1018 * at L4 may be considered an L3 hash. This should only be done if the 1019 * driver can't unambiguously determine that the HW computed the hash at 1020 * the higher layer. Note that the "should" in the second property above 1021 * permits this. 1022 */ 1023 enum pkt_hash_types { 1024 PKT_HASH_TYPE_NONE, /* Undefined type */ 1025 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1026 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1027 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1028 }; 1029 1030 static inline void skb_clear_hash(struct sk_buff *skb) 1031 { 1032 skb->hash = 0; 1033 skb->sw_hash = 0; 1034 skb->l4_hash = 0; 1035 } 1036 1037 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1038 { 1039 if (!skb->l4_hash) 1040 skb_clear_hash(skb); 1041 } 1042 1043 static inline void 1044 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1045 { 1046 skb->l4_hash = is_l4; 1047 skb->sw_hash = is_sw; 1048 skb->hash = hash; 1049 } 1050 1051 static inline void 1052 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1053 { 1054 /* Used by drivers to set hash from HW */ 1055 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1056 } 1057 1058 static inline void 1059 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1060 { 1061 __skb_set_hash(skb, hash, true, is_l4); 1062 } 1063 1064 void __skb_get_hash(struct sk_buff *skb); 1065 u32 skb_get_poff(const struct sk_buff *skb); 1066 u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1067 const struct flow_keys *keys, int hlen); 1068 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1069 void *data, int hlen_proto); 1070 1071 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1072 int thoff, u8 ip_proto) 1073 { 1074 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1075 } 1076 1077 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1078 const struct flow_dissector_key *key, 1079 unsigned int key_count); 1080 1081 bool __skb_flow_dissect(const struct sk_buff *skb, 1082 struct flow_dissector *flow_dissector, 1083 void *target_container, 1084 void *data, __be16 proto, int nhoff, int hlen, 1085 unsigned int flags); 1086 1087 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1088 struct flow_dissector *flow_dissector, 1089 void *target_container, unsigned int flags) 1090 { 1091 return __skb_flow_dissect(skb, flow_dissector, target_container, 1092 NULL, 0, 0, 0, flags); 1093 } 1094 1095 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1096 struct flow_keys *flow, 1097 unsigned int flags) 1098 { 1099 memset(flow, 0, sizeof(*flow)); 1100 return __skb_flow_dissect(skb, &flow_keys_dissector, flow, 1101 NULL, 0, 0, 0, flags); 1102 } 1103 1104 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow, 1105 void *data, __be16 proto, 1106 int nhoff, int hlen, 1107 unsigned int flags) 1108 { 1109 memset(flow, 0, sizeof(*flow)); 1110 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow, 1111 data, proto, nhoff, hlen, flags); 1112 } 1113 1114 static inline __u32 skb_get_hash(struct sk_buff *skb) 1115 { 1116 if (!skb->l4_hash && !skb->sw_hash) 1117 __skb_get_hash(skb); 1118 1119 return skb->hash; 1120 } 1121 1122 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6); 1123 1124 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1125 { 1126 if (!skb->l4_hash && !skb->sw_hash) { 1127 struct flow_keys keys; 1128 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1129 1130 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1131 } 1132 1133 return skb->hash; 1134 } 1135 1136 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl); 1137 1138 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4) 1139 { 1140 if (!skb->l4_hash && !skb->sw_hash) { 1141 struct flow_keys keys; 1142 __u32 hash = __get_hash_from_flowi4(fl4, &keys); 1143 1144 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1145 } 1146 1147 return skb->hash; 1148 } 1149 1150 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1151 1152 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1153 { 1154 return skb->hash; 1155 } 1156 1157 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1158 { 1159 to->hash = from->hash; 1160 to->sw_hash = from->sw_hash; 1161 to->l4_hash = from->l4_hash; 1162 }; 1163 1164 static inline void skb_sender_cpu_clear(struct sk_buff *skb) 1165 { 1166 } 1167 1168 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1169 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1170 { 1171 return skb->head + skb->end; 1172 } 1173 1174 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1175 { 1176 return skb->end; 1177 } 1178 #else 1179 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1180 { 1181 return skb->end; 1182 } 1183 1184 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1185 { 1186 return skb->end - skb->head; 1187 } 1188 #endif 1189 1190 /* Internal */ 1191 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1192 1193 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1194 { 1195 return &skb_shinfo(skb)->hwtstamps; 1196 } 1197 1198 /** 1199 * skb_queue_empty - check if a queue is empty 1200 * @list: queue head 1201 * 1202 * Returns true if the queue is empty, false otherwise. 1203 */ 1204 static inline int skb_queue_empty(const struct sk_buff_head *list) 1205 { 1206 return list->next == (const struct sk_buff *) list; 1207 } 1208 1209 /** 1210 * skb_queue_is_last - check if skb is the last entry in the queue 1211 * @list: queue head 1212 * @skb: buffer 1213 * 1214 * Returns true if @skb is the last buffer on the list. 1215 */ 1216 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1217 const struct sk_buff *skb) 1218 { 1219 return skb->next == (const struct sk_buff *) list; 1220 } 1221 1222 /** 1223 * skb_queue_is_first - check if skb is the first entry in the queue 1224 * @list: queue head 1225 * @skb: buffer 1226 * 1227 * Returns true if @skb is the first buffer on the list. 1228 */ 1229 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1230 const struct sk_buff *skb) 1231 { 1232 return skb->prev == (const struct sk_buff *) list; 1233 } 1234 1235 /** 1236 * skb_queue_next - return the next packet in the queue 1237 * @list: queue head 1238 * @skb: current buffer 1239 * 1240 * Return the next packet in @list after @skb. It is only valid to 1241 * call this if skb_queue_is_last() evaluates to false. 1242 */ 1243 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1244 const struct sk_buff *skb) 1245 { 1246 /* This BUG_ON may seem severe, but if we just return then we 1247 * are going to dereference garbage. 1248 */ 1249 BUG_ON(skb_queue_is_last(list, skb)); 1250 return skb->next; 1251 } 1252 1253 /** 1254 * skb_queue_prev - return the prev packet in the queue 1255 * @list: queue head 1256 * @skb: current buffer 1257 * 1258 * Return the prev packet in @list before @skb. It is only valid to 1259 * call this if skb_queue_is_first() evaluates to false. 1260 */ 1261 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1262 const struct sk_buff *skb) 1263 { 1264 /* This BUG_ON may seem severe, but if we just return then we 1265 * are going to dereference garbage. 1266 */ 1267 BUG_ON(skb_queue_is_first(list, skb)); 1268 return skb->prev; 1269 } 1270 1271 /** 1272 * skb_get - reference buffer 1273 * @skb: buffer to reference 1274 * 1275 * Makes another reference to a socket buffer and returns a pointer 1276 * to the buffer. 1277 */ 1278 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1279 { 1280 atomic_inc(&skb->users); 1281 return skb; 1282 } 1283 1284 /* 1285 * If users == 1, we are the only owner and are can avoid redundant 1286 * atomic change. 1287 */ 1288 1289 /** 1290 * skb_cloned - is the buffer a clone 1291 * @skb: buffer to check 1292 * 1293 * Returns true if the buffer was generated with skb_clone() and is 1294 * one of multiple shared copies of the buffer. Cloned buffers are 1295 * shared data so must not be written to under normal circumstances. 1296 */ 1297 static inline int skb_cloned(const struct sk_buff *skb) 1298 { 1299 return skb->cloned && 1300 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1301 } 1302 1303 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1304 { 1305 might_sleep_if(gfpflags_allow_blocking(pri)); 1306 1307 if (skb_cloned(skb)) 1308 return pskb_expand_head(skb, 0, 0, pri); 1309 1310 return 0; 1311 } 1312 1313 /** 1314 * skb_header_cloned - is the header a clone 1315 * @skb: buffer to check 1316 * 1317 * Returns true if modifying the header part of the buffer requires 1318 * the data to be copied. 1319 */ 1320 static inline int skb_header_cloned(const struct sk_buff *skb) 1321 { 1322 int dataref; 1323 1324 if (!skb->cloned) 1325 return 0; 1326 1327 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1328 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1329 return dataref != 1; 1330 } 1331 1332 /** 1333 * skb_header_release - release reference to header 1334 * @skb: buffer to operate on 1335 * 1336 * Drop a reference to the header part of the buffer. This is done 1337 * by acquiring a payload reference. You must not read from the header 1338 * part of skb->data after this. 1339 * Note : Check if you can use __skb_header_release() instead. 1340 */ 1341 static inline void skb_header_release(struct sk_buff *skb) 1342 { 1343 BUG_ON(skb->nohdr); 1344 skb->nohdr = 1; 1345 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1346 } 1347 1348 /** 1349 * __skb_header_release - release reference to header 1350 * @skb: buffer to operate on 1351 * 1352 * Variant of skb_header_release() assuming skb is private to caller. 1353 * We can avoid one atomic operation. 1354 */ 1355 static inline void __skb_header_release(struct sk_buff *skb) 1356 { 1357 skb->nohdr = 1; 1358 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1359 } 1360 1361 1362 /** 1363 * skb_shared - is the buffer shared 1364 * @skb: buffer to check 1365 * 1366 * Returns true if more than one person has a reference to this 1367 * buffer. 1368 */ 1369 static inline int skb_shared(const struct sk_buff *skb) 1370 { 1371 return atomic_read(&skb->users) != 1; 1372 } 1373 1374 /** 1375 * skb_share_check - check if buffer is shared and if so clone it 1376 * @skb: buffer to check 1377 * @pri: priority for memory allocation 1378 * 1379 * If the buffer is shared the buffer is cloned and the old copy 1380 * drops a reference. A new clone with a single reference is returned. 1381 * If the buffer is not shared the original buffer is returned. When 1382 * being called from interrupt status or with spinlocks held pri must 1383 * be GFP_ATOMIC. 1384 * 1385 * NULL is returned on a memory allocation failure. 1386 */ 1387 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1388 { 1389 might_sleep_if(gfpflags_allow_blocking(pri)); 1390 if (skb_shared(skb)) { 1391 struct sk_buff *nskb = skb_clone(skb, pri); 1392 1393 if (likely(nskb)) 1394 consume_skb(skb); 1395 else 1396 kfree_skb(skb); 1397 skb = nskb; 1398 } 1399 return skb; 1400 } 1401 1402 /* 1403 * Copy shared buffers into a new sk_buff. We effectively do COW on 1404 * packets to handle cases where we have a local reader and forward 1405 * and a couple of other messy ones. The normal one is tcpdumping 1406 * a packet thats being forwarded. 1407 */ 1408 1409 /** 1410 * skb_unshare - make a copy of a shared buffer 1411 * @skb: buffer to check 1412 * @pri: priority for memory allocation 1413 * 1414 * If the socket buffer is a clone then this function creates a new 1415 * copy of the data, drops a reference count on the old copy and returns 1416 * the new copy with the reference count at 1. If the buffer is not a clone 1417 * the original buffer is returned. When called with a spinlock held or 1418 * from interrupt state @pri must be %GFP_ATOMIC 1419 * 1420 * %NULL is returned on a memory allocation failure. 1421 */ 1422 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1423 gfp_t pri) 1424 { 1425 might_sleep_if(gfpflags_allow_blocking(pri)); 1426 if (skb_cloned(skb)) { 1427 struct sk_buff *nskb = skb_copy(skb, pri); 1428 1429 /* Free our shared copy */ 1430 if (likely(nskb)) 1431 consume_skb(skb); 1432 else 1433 kfree_skb(skb); 1434 skb = nskb; 1435 } 1436 return skb; 1437 } 1438 1439 /** 1440 * skb_peek - peek at the head of an &sk_buff_head 1441 * @list_: list to peek at 1442 * 1443 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1444 * be careful with this one. A peek leaves the buffer on the 1445 * list and someone else may run off with it. You must hold 1446 * the appropriate locks or have a private queue to do this. 1447 * 1448 * Returns %NULL for an empty list or a pointer to the head element. 1449 * The reference count is not incremented and the reference is therefore 1450 * volatile. Use with caution. 1451 */ 1452 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1453 { 1454 struct sk_buff *skb = list_->next; 1455 1456 if (skb == (struct sk_buff *)list_) 1457 skb = NULL; 1458 return skb; 1459 } 1460 1461 /** 1462 * skb_peek_next - peek skb following the given one from a queue 1463 * @skb: skb to start from 1464 * @list_: list to peek at 1465 * 1466 * Returns %NULL when the end of the list is met or a pointer to the 1467 * next element. The reference count is not incremented and the 1468 * reference is therefore volatile. Use with caution. 1469 */ 1470 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1471 const struct sk_buff_head *list_) 1472 { 1473 struct sk_buff *next = skb->next; 1474 1475 if (next == (struct sk_buff *)list_) 1476 next = NULL; 1477 return next; 1478 } 1479 1480 /** 1481 * skb_peek_tail - peek at the tail of an &sk_buff_head 1482 * @list_: list to peek at 1483 * 1484 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1485 * be careful with this one. A peek leaves the buffer on the 1486 * list and someone else may run off with it. You must hold 1487 * the appropriate locks or have a private queue to do this. 1488 * 1489 * Returns %NULL for an empty list or a pointer to the tail element. 1490 * The reference count is not incremented and the reference is therefore 1491 * volatile. Use with caution. 1492 */ 1493 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1494 { 1495 struct sk_buff *skb = list_->prev; 1496 1497 if (skb == (struct sk_buff *)list_) 1498 skb = NULL; 1499 return skb; 1500 1501 } 1502 1503 /** 1504 * skb_queue_len - get queue length 1505 * @list_: list to measure 1506 * 1507 * Return the length of an &sk_buff queue. 1508 */ 1509 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1510 { 1511 return list_->qlen; 1512 } 1513 1514 /** 1515 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1516 * @list: queue to initialize 1517 * 1518 * This initializes only the list and queue length aspects of 1519 * an sk_buff_head object. This allows to initialize the list 1520 * aspects of an sk_buff_head without reinitializing things like 1521 * the spinlock. It can also be used for on-stack sk_buff_head 1522 * objects where the spinlock is known to not be used. 1523 */ 1524 static inline void __skb_queue_head_init(struct sk_buff_head *list) 1525 { 1526 list->prev = list->next = (struct sk_buff *)list; 1527 list->qlen = 0; 1528 } 1529 1530 /* 1531 * This function creates a split out lock class for each invocation; 1532 * this is needed for now since a whole lot of users of the skb-queue 1533 * infrastructure in drivers have different locking usage (in hardirq) 1534 * than the networking core (in softirq only). In the long run either the 1535 * network layer or drivers should need annotation to consolidate the 1536 * main types of usage into 3 classes. 1537 */ 1538 static inline void skb_queue_head_init(struct sk_buff_head *list) 1539 { 1540 spin_lock_init(&list->lock); 1541 __skb_queue_head_init(list); 1542 } 1543 1544 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1545 struct lock_class_key *class) 1546 { 1547 skb_queue_head_init(list); 1548 lockdep_set_class(&list->lock, class); 1549 } 1550 1551 /* 1552 * Insert an sk_buff on a list. 1553 * 1554 * The "__skb_xxxx()" functions are the non-atomic ones that 1555 * can only be called with interrupts disabled. 1556 */ 1557 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1558 struct sk_buff_head *list); 1559 static inline void __skb_insert(struct sk_buff *newsk, 1560 struct sk_buff *prev, struct sk_buff *next, 1561 struct sk_buff_head *list) 1562 { 1563 newsk->next = next; 1564 newsk->prev = prev; 1565 next->prev = prev->next = newsk; 1566 list->qlen++; 1567 } 1568 1569 static inline void __skb_queue_splice(const struct sk_buff_head *list, 1570 struct sk_buff *prev, 1571 struct sk_buff *next) 1572 { 1573 struct sk_buff *first = list->next; 1574 struct sk_buff *last = list->prev; 1575 1576 first->prev = prev; 1577 prev->next = first; 1578 1579 last->next = next; 1580 next->prev = last; 1581 } 1582 1583 /** 1584 * skb_queue_splice - join two skb lists, this is designed for stacks 1585 * @list: the new list to add 1586 * @head: the place to add it in the first list 1587 */ 1588 static inline void skb_queue_splice(const struct sk_buff_head *list, 1589 struct sk_buff_head *head) 1590 { 1591 if (!skb_queue_empty(list)) { 1592 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1593 head->qlen += list->qlen; 1594 } 1595 } 1596 1597 /** 1598 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1599 * @list: the new list to add 1600 * @head: the place to add it in the first list 1601 * 1602 * The list at @list is reinitialised 1603 */ 1604 static inline void skb_queue_splice_init(struct sk_buff_head *list, 1605 struct sk_buff_head *head) 1606 { 1607 if (!skb_queue_empty(list)) { 1608 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1609 head->qlen += list->qlen; 1610 __skb_queue_head_init(list); 1611 } 1612 } 1613 1614 /** 1615 * skb_queue_splice_tail - join two skb lists, each list being a queue 1616 * @list: the new list to add 1617 * @head: the place to add it in the first list 1618 */ 1619 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1620 struct sk_buff_head *head) 1621 { 1622 if (!skb_queue_empty(list)) { 1623 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1624 head->qlen += list->qlen; 1625 } 1626 } 1627 1628 /** 1629 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1630 * @list: the new list to add 1631 * @head: the place to add it in the first list 1632 * 1633 * Each of the lists is a queue. 1634 * The list at @list is reinitialised 1635 */ 1636 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1637 struct sk_buff_head *head) 1638 { 1639 if (!skb_queue_empty(list)) { 1640 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1641 head->qlen += list->qlen; 1642 __skb_queue_head_init(list); 1643 } 1644 } 1645 1646 /** 1647 * __skb_queue_after - queue a buffer at the list head 1648 * @list: list to use 1649 * @prev: place after this buffer 1650 * @newsk: buffer to queue 1651 * 1652 * Queue a buffer int the middle of a list. This function takes no locks 1653 * and you must therefore hold required locks before calling it. 1654 * 1655 * A buffer cannot be placed on two lists at the same time. 1656 */ 1657 static inline void __skb_queue_after(struct sk_buff_head *list, 1658 struct sk_buff *prev, 1659 struct sk_buff *newsk) 1660 { 1661 __skb_insert(newsk, prev, prev->next, list); 1662 } 1663 1664 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1665 struct sk_buff_head *list); 1666 1667 static inline void __skb_queue_before(struct sk_buff_head *list, 1668 struct sk_buff *next, 1669 struct sk_buff *newsk) 1670 { 1671 __skb_insert(newsk, next->prev, next, list); 1672 } 1673 1674 /** 1675 * __skb_queue_head - queue a buffer at the list head 1676 * @list: list to use 1677 * @newsk: buffer to queue 1678 * 1679 * Queue a buffer at the start of a list. This function takes no locks 1680 * and you must therefore hold required locks before calling it. 1681 * 1682 * A buffer cannot be placed on two lists at the same time. 1683 */ 1684 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1685 static inline void __skb_queue_head(struct sk_buff_head *list, 1686 struct sk_buff *newsk) 1687 { 1688 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1689 } 1690 1691 /** 1692 * __skb_queue_tail - queue a buffer at the list tail 1693 * @list: list to use 1694 * @newsk: buffer to queue 1695 * 1696 * Queue a buffer at the end of a list. This function takes no locks 1697 * and you must therefore hold required locks before calling it. 1698 * 1699 * A buffer cannot be placed on two lists at the same time. 1700 */ 1701 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1702 static inline void __skb_queue_tail(struct sk_buff_head *list, 1703 struct sk_buff *newsk) 1704 { 1705 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1706 } 1707 1708 /* 1709 * remove sk_buff from list. _Must_ be called atomically, and with 1710 * the list known.. 1711 */ 1712 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1713 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1714 { 1715 struct sk_buff *next, *prev; 1716 1717 list->qlen--; 1718 next = skb->next; 1719 prev = skb->prev; 1720 skb->next = skb->prev = NULL; 1721 next->prev = prev; 1722 prev->next = next; 1723 } 1724 1725 /** 1726 * __skb_dequeue - remove from the head of the queue 1727 * @list: list to dequeue from 1728 * 1729 * Remove the head of the list. This function does not take any locks 1730 * so must be used with appropriate locks held only. The head item is 1731 * returned or %NULL if the list is empty. 1732 */ 1733 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1734 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1735 { 1736 struct sk_buff *skb = skb_peek(list); 1737 if (skb) 1738 __skb_unlink(skb, list); 1739 return skb; 1740 } 1741 1742 /** 1743 * __skb_dequeue_tail - remove from the tail of the queue 1744 * @list: list to dequeue from 1745 * 1746 * Remove the tail of the list. This function does not take any locks 1747 * so must be used with appropriate locks held only. The tail item is 1748 * returned or %NULL if the list is empty. 1749 */ 1750 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1751 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1752 { 1753 struct sk_buff *skb = skb_peek_tail(list); 1754 if (skb) 1755 __skb_unlink(skb, list); 1756 return skb; 1757 } 1758 1759 1760 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1761 { 1762 return skb->data_len; 1763 } 1764 1765 static inline unsigned int skb_headlen(const struct sk_buff *skb) 1766 { 1767 return skb->len - skb->data_len; 1768 } 1769 1770 static inline int skb_pagelen(const struct sk_buff *skb) 1771 { 1772 int i, len = 0; 1773 1774 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--) 1775 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1776 return len + skb_headlen(skb); 1777 } 1778 1779 /** 1780 * __skb_fill_page_desc - initialise a paged fragment in an skb 1781 * @skb: buffer containing fragment to be initialised 1782 * @i: paged fragment index to initialise 1783 * @page: the page to use for this fragment 1784 * @off: the offset to the data with @page 1785 * @size: the length of the data 1786 * 1787 * Initialises the @i'th fragment of @skb to point to &size bytes at 1788 * offset @off within @page. 1789 * 1790 * Does not take any additional reference on the fragment. 1791 */ 1792 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1793 struct page *page, int off, int size) 1794 { 1795 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1796 1797 /* 1798 * Propagate page pfmemalloc to the skb if we can. The problem is 1799 * that not all callers have unique ownership of the page but rely 1800 * on page_is_pfmemalloc doing the right thing(tm). 1801 */ 1802 frag->page.p = page; 1803 frag->page_offset = off; 1804 skb_frag_size_set(frag, size); 1805 1806 page = compound_head(page); 1807 if (page_is_pfmemalloc(page)) 1808 skb->pfmemalloc = true; 1809 } 1810 1811 /** 1812 * skb_fill_page_desc - initialise a paged fragment in an skb 1813 * @skb: buffer containing fragment to be initialised 1814 * @i: paged fragment index to initialise 1815 * @page: the page to use for this fragment 1816 * @off: the offset to the data with @page 1817 * @size: the length of the data 1818 * 1819 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1820 * @skb to point to @size bytes at offset @off within @page. In 1821 * addition updates @skb such that @i is the last fragment. 1822 * 1823 * Does not take any additional reference on the fragment. 1824 */ 1825 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1826 struct page *page, int off, int size) 1827 { 1828 __skb_fill_page_desc(skb, i, page, off, size); 1829 skb_shinfo(skb)->nr_frags = i + 1; 1830 } 1831 1832 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1833 int size, unsigned int truesize); 1834 1835 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1836 unsigned int truesize); 1837 1838 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1839 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1840 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1841 1842 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1843 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1844 { 1845 return skb->head + skb->tail; 1846 } 1847 1848 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1849 { 1850 skb->tail = skb->data - skb->head; 1851 } 1852 1853 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1854 { 1855 skb_reset_tail_pointer(skb); 1856 skb->tail += offset; 1857 } 1858 1859 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 1860 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1861 { 1862 return skb->tail; 1863 } 1864 1865 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1866 { 1867 skb->tail = skb->data; 1868 } 1869 1870 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1871 { 1872 skb->tail = skb->data + offset; 1873 } 1874 1875 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1876 1877 /* 1878 * Add data to an sk_buff 1879 */ 1880 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1881 unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1882 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1883 { 1884 unsigned char *tmp = skb_tail_pointer(skb); 1885 SKB_LINEAR_ASSERT(skb); 1886 skb->tail += len; 1887 skb->len += len; 1888 return tmp; 1889 } 1890 1891 unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1892 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1893 { 1894 skb->data -= len; 1895 skb->len += len; 1896 return skb->data; 1897 } 1898 1899 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1900 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1901 { 1902 skb->len -= len; 1903 BUG_ON(skb->len < skb->data_len); 1904 return skb->data += len; 1905 } 1906 1907 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1908 { 1909 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1910 } 1911 1912 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1913 1914 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1915 { 1916 if (len > skb_headlen(skb) && 1917 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1918 return NULL; 1919 skb->len -= len; 1920 return skb->data += len; 1921 } 1922 1923 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1924 { 1925 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1926 } 1927 1928 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1929 { 1930 if (likely(len <= skb_headlen(skb))) 1931 return 1; 1932 if (unlikely(len > skb->len)) 1933 return 0; 1934 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1935 } 1936 1937 /** 1938 * skb_headroom - bytes at buffer head 1939 * @skb: buffer to check 1940 * 1941 * Return the number of bytes of free space at the head of an &sk_buff. 1942 */ 1943 static inline unsigned int skb_headroom(const struct sk_buff *skb) 1944 { 1945 return skb->data - skb->head; 1946 } 1947 1948 /** 1949 * skb_tailroom - bytes at buffer end 1950 * @skb: buffer to check 1951 * 1952 * Return the number of bytes of free space at the tail of an sk_buff 1953 */ 1954 static inline int skb_tailroom(const struct sk_buff *skb) 1955 { 1956 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 1957 } 1958 1959 /** 1960 * skb_availroom - bytes at buffer end 1961 * @skb: buffer to check 1962 * 1963 * Return the number of bytes of free space at the tail of an sk_buff 1964 * allocated by sk_stream_alloc() 1965 */ 1966 static inline int skb_availroom(const struct sk_buff *skb) 1967 { 1968 if (skb_is_nonlinear(skb)) 1969 return 0; 1970 1971 return skb->end - skb->tail - skb->reserved_tailroom; 1972 } 1973 1974 /** 1975 * skb_reserve - adjust headroom 1976 * @skb: buffer to alter 1977 * @len: bytes to move 1978 * 1979 * Increase the headroom of an empty &sk_buff by reducing the tail 1980 * room. This is only allowed for an empty buffer. 1981 */ 1982 static inline void skb_reserve(struct sk_buff *skb, int len) 1983 { 1984 skb->data += len; 1985 skb->tail += len; 1986 } 1987 1988 /** 1989 * skb_tailroom_reserve - adjust reserved_tailroom 1990 * @skb: buffer to alter 1991 * @mtu: maximum amount of headlen permitted 1992 * @needed_tailroom: minimum amount of reserved_tailroom 1993 * 1994 * Set reserved_tailroom so that headlen can be as large as possible but 1995 * not larger than mtu and tailroom cannot be smaller than 1996 * needed_tailroom. 1997 * The required headroom should already have been reserved before using 1998 * this function. 1999 */ 2000 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2001 unsigned int needed_tailroom) 2002 { 2003 SKB_LINEAR_ASSERT(skb); 2004 if (mtu < skb_tailroom(skb) - needed_tailroom) 2005 /* use at most mtu */ 2006 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2007 else 2008 /* use up to all available space */ 2009 skb->reserved_tailroom = needed_tailroom; 2010 } 2011 2012 #define ENCAP_TYPE_ETHER 0 2013 #define ENCAP_TYPE_IPPROTO 1 2014 2015 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2016 __be16 protocol) 2017 { 2018 skb->inner_protocol = protocol; 2019 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2020 } 2021 2022 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2023 __u8 ipproto) 2024 { 2025 skb->inner_ipproto = ipproto; 2026 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2027 } 2028 2029 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2030 { 2031 skb->inner_mac_header = skb->mac_header; 2032 skb->inner_network_header = skb->network_header; 2033 skb->inner_transport_header = skb->transport_header; 2034 } 2035 2036 static inline void skb_reset_mac_len(struct sk_buff *skb) 2037 { 2038 skb->mac_len = skb->network_header - skb->mac_header; 2039 } 2040 2041 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2042 *skb) 2043 { 2044 return skb->head + skb->inner_transport_header; 2045 } 2046 2047 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2048 { 2049 return skb_inner_transport_header(skb) - skb->data; 2050 } 2051 2052 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2053 { 2054 skb->inner_transport_header = skb->data - skb->head; 2055 } 2056 2057 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2058 const int offset) 2059 { 2060 skb_reset_inner_transport_header(skb); 2061 skb->inner_transport_header += offset; 2062 } 2063 2064 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2065 { 2066 return skb->head + skb->inner_network_header; 2067 } 2068 2069 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2070 { 2071 skb->inner_network_header = skb->data - skb->head; 2072 } 2073 2074 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2075 const int offset) 2076 { 2077 skb_reset_inner_network_header(skb); 2078 skb->inner_network_header += offset; 2079 } 2080 2081 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2082 { 2083 return skb->head + skb->inner_mac_header; 2084 } 2085 2086 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2087 { 2088 skb->inner_mac_header = skb->data - skb->head; 2089 } 2090 2091 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2092 const int offset) 2093 { 2094 skb_reset_inner_mac_header(skb); 2095 skb->inner_mac_header += offset; 2096 } 2097 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2098 { 2099 return skb->transport_header != (typeof(skb->transport_header))~0U; 2100 } 2101 2102 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2103 { 2104 return skb->head + skb->transport_header; 2105 } 2106 2107 static inline void skb_reset_transport_header(struct sk_buff *skb) 2108 { 2109 skb->transport_header = skb->data - skb->head; 2110 } 2111 2112 static inline void skb_set_transport_header(struct sk_buff *skb, 2113 const int offset) 2114 { 2115 skb_reset_transport_header(skb); 2116 skb->transport_header += offset; 2117 } 2118 2119 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2120 { 2121 return skb->head + skb->network_header; 2122 } 2123 2124 static inline void skb_reset_network_header(struct sk_buff *skb) 2125 { 2126 skb->network_header = skb->data - skb->head; 2127 } 2128 2129 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2130 { 2131 skb_reset_network_header(skb); 2132 skb->network_header += offset; 2133 } 2134 2135 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2136 { 2137 return skb->head + skb->mac_header; 2138 } 2139 2140 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2141 { 2142 return skb->mac_header != (typeof(skb->mac_header))~0U; 2143 } 2144 2145 static inline void skb_reset_mac_header(struct sk_buff *skb) 2146 { 2147 skb->mac_header = skb->data - skb->head; 2148 } 2149 2150 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2151 { 2152 skb_reset_mac_header(skb); 2153 skb->mac_header += offset; 2154 } 2155 2156 static inline void skb_pop_mac_header(struct sk_buff *skb) 2157 { 2158 skb->mac_header = skb->network_header; 2159 } 2160 2161 static inline void skb_probe_transport_header(struct sk_buff *skb, 2162 const int offset_hint) 2163 { 2164 struct flow_keys keys; 2165 2166 if (skb_transport_header_was_set(skb)) 2167 return; 2168 else if (skb_flow_dissect_flow_keys(skb, &keys, 0)) 2169 skb_set_transport_header(skb, keys.control.thoff); 2170 else 2171 skb_set_transport_header(skb, offset_hint); 2172 } 2173 2174 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2175 { 2176 if (skb_mac_header_was_set(skb)) { 2177 const unsigned char *old_mac = skb_mac_header(skb); 2178 2179 skb_set_mac_header(skb, -skb->mac_len); 2180 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2181 } 2182 } 2183 2184 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2185 { 2186 return skb->csum_start - skb_headroom(skb); 2187 } 2188 2189 static inline int skb_transport_offset(const struct sk_buff *skb) 2190 { 2191 return skb_transport_header(skb) - skb->data; 2192 } 2193 2194 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2195 { 2196 return skb->transport_header - skb->network_header; 2197 } 2198 2199 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2200 { 2201 return skb->inner_transport_header - skb->inner_network_header; 2202 } 2203 2204 static inline int skb_network_offset(const struct sk_buff *skb) 2205 { 2206 return skb_network_header(skb) - skb->data; 2207 } 2208 2209 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2210 { 2211 return skb_inner_network_header(skb) - skb->data; 2212 } 2213 2214 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2215 { 2216 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2217 } 2218 2219 /* 2220 * CPUs often take a performance hit when accessing unaligned memory 2221 * locations. The actual performance hit varies, it can be small if the 2222 * hardware handles it or large if we have to take an exception and fix it 2223 * in software. 2224 * 2225 * Since an ethernet header is 14 bytes network drivers often end up with 2226 * the IP header at an unaligned offset. The IP header can be aligned by 2227 * shifting the start of the packet by 2 bytes. Drivers should do this 2228 * with: 2229 * 2230 * skb_reserve(skb, NET_IP_ALIGN); 2231 * 2232 * The downside to this alignment of the IP header is that the DMA is now 2233 * unaligned. On some architectures the cost of an unaligned DMA is high 2234 * and this cost outweighs the gains made by aligning the IP header. 2235 * 2236 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2237 * to be overridden. 2238 */ 2239 #ifndef NET_IP_ALIGN 2240 #define NET_IP_ALIGN 2 2241 #endif 2242 2243 /* 2244 * The networking layer reserves some headroom in skb data (via 2245 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2246 * the header has to grow. In the default case, if the header has to grow 2247 * 32 bytes or less we avoid the reallocation. 2248 * 2249 * Unfortunately this headroom changes the DMA alignment of the resulting 2250 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2251 * on some architectures. An architecture can override this value, 2252 * perhaps setting it to a cacheline in size (since that will maintain 2253 * cacheline alignment of the DMA). It must be a power of 2. 2254 * 2255 * Various parts of the networking layer expect at least 32 bytes of 2256 * headroom, you should not reduce this. 2257 * 2258 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2259 * to reduce average number of cache lines per packet. 2260 * get_rps_cpus() for example only access one 64 bytes aligned block : 2261 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2262 */ 2263 #ifndef NET_SKB_PAD 2264 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2265 #endif 2266 2267 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2268 2269 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2270 { 2271 if (unlikely(skb_is_nonlinear(skb))) { 2272 WARN_ON(1); 2273 return; 2274 } 2275 skb->len = len; 2276 skb_set_tail_pointer(skb, len); 2277 } 2278 2279 void skb_trim(struct sk_buff *skb, unsigned int len); 2280 2281 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2282 { 2283 if (skb->data_len) 2284 return ___pskb_trim(skb, len); 2285 __skb_trim(skb, len); 2286 return 0; 2287 } 2288 2289 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2290 { 2291 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2292 } 2293 2294 /** 2295 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2296 * @skb: buffer to alter 2297 * @len: new length 2298 * 2299 * This is identical to pskb_trim except that the caller knows that 2300 * the skb is not cloned so we should never get an error due to out- 2301 * of-memory. 2302 */ 2303 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2304 { 2305 int err = pskb_trim(skb, len); 2306 BUG_ON(err); 2307 } 2308 2309 /** 2310 * skb_orphan - orphan a buffer 2311 * @skb: buffer to orphan 2312 * 2313 * If a buffer currently has an owner then we call the owner's 2314 * destructor function and make the @skb unowned. The buffer continues 2315 * to exist but is no longer charged to its former owner. 2316 */ 2317 static inline void skb_orphan(struct sk_buff *skb) 2318 { 2319 if (skb->destructor) { 2320 skb->destructor(skb); 2321 skb->destructor = NULL; 2322 skb->sk = NULL; 2323 } else { 2324 BUG_ON(skb->sk); 2325 } 2326 } 2327 2328 /** 2329 * skb_orphan_frags - orphan the frags contained in a buffer 2330 * @skb: buffer to orphan frags from 2331 * @gfp_mask: allocation mask for replacement pages 2332 * 2333 * For each frag in the SKB which needs a destructor (i.e. has an 2334 * owner) create a copy of that frag and release the original 2335 * page by calling the destructor. 2336 */ 2337 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2338 { 2339 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 2340 return 0; 2341 return skb_copy_ubufs(skb, gfp_mask); 2342 } 2343 2344 /** 2345 * __skb_queue_purge - empty a list 2346 * @list: list to empty 2347 * 2348 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2349 * the list and one reference dropped. This function does not take the 2350 * list lock and the caller must hold the relevant locks to use it. 2351 */ 2352 void skb_queue_purge(struct sk_buff_head *list); 2353 static inline void __skb_queue_purge(struct sk_buff_head *list) 2354 { 2355 struct sk_buff *skb; 2356 while ((skb = __skb_dequeue(list)) != NULL) 2357 kfree_skb(skb); 2358 } 2359 2360 void *netdev_alloc_frag(unsigned int fragsz); 2361 2362 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2363 gfp_t gfp_mask); 2364 2365 /** 2366 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2367 * @dev: network device to receive on 2368 * @length: length to allocate 2369 * 2370 * Allocate a new &sk_buff and assign it a usage count of one. The 2371 * buffer has unspecified headroom built in. Users should allocate 2372 * the headroom they think they need without accounting for the 2373 * built in space. The built in space is used for optimisations. 2374 * 2375 * %NULL is returned if there is no free memory. Although this function 2376 * allocates memory it can be called from an interrupt. 2377 */ 2378 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2379 unsigned int length) 2380 { 2381 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2382 } 2383 2384 /* legacy helper around __netdev_alloc_skb() */ 2385 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2386 gfp_t gfp_mask) 2387 { 2388 return __netdev_alloc_skb(NULL, length, gfp_mask); 2389 } 2390 2391 /* legacy helper around netdev_alloc_skb() */ 2392 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2393 { 2394 return netdev_alloc_skb(NULL, length); 2395 } 2396 2397 2398 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2399 unsigned int length, gfp_t gfp) 2400 { 2401 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2402 2403 if (NET_IP_ALIGN && skb) 2404 skb_reserve(skb, NET_IP_ALIGN); 2405 return skb; 2406 } 2407 2408 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2409 unsigned int length) 2410 { 2411 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2412 } 2413 2414 static inline void skb_free_frag(void *addr) 2415 { 2416 __free_page_frag(addr); 2417 } 2418 2419 void *napi_alloc_frag(unsigned int fragsz); 2420 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2421 unsigned int length, gfp_t gfp_mask); 2422 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2423 unsigned int length) 2424 { 2425 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2426 } 2427 2428 /** 2429 * __dev_alloc_pages - allocate page for network Rx 2430 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2431 * @order: size of the allocation 2432 * 2433 * Allocate a new page. 2434 * 2435 * %NULL is returned if there is no free memory. 2436 */ 2437 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2438 unsigned int order) 2439 { 2440 /* This piece of code contains several assumptions. 2441 * 1. This is for device Rx, therefor a cold page is preferred. 2442 * 2. The expectation is the user wants a compound page. 2443 * 3. If requesting a order 0 page it will not be compound 2444 * due to the check to see if order has a value in prep_new_page 2445 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2446 * code in gfp_to_alloc_flags that should be enforcing this. 2447 */ 2448 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2449 2450 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2451 } 2452 2453 static inline struct page *dev_alloc_pages(unsigned int order) 2454 { 2455 return __dev_alloc_pages(GFP_ATOMIC, order); 2456 } 2457 2458 /** 2459 * __dev_alloc_page - allocate a page for network Rx 2460 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2461 * 2462 * Allocate a new page. 2463 * 2464 * %NULL is returned if there is no free memory. 2465 */ 2466 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2467 { 2468 return __dev_alloc_pages(gfp_mask, 0); 2469 } 2470 2471 static inline struct page *dev_alloc_page(void) 2472 { 2473 return __dev_alloc_page(GFP_ATOMIC); 2474 } 2475 2476 /** 2477 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2478 * @page: The page that was allocated from skb_alloc_page 2479 * @skb: The skb that may need pfmemalloc set 2480 */ 2481 static inline void skb_propagate_pfmemalloc(struct page *page, 2482 struct sk_buff *skb) 2483 { 2484 if (page_is_pfmemalloc(page)) 2485 skb->pfmemalloc = true; 2486 } 2487 2488 /** 2489 * skb_frag_page - retrieve the page referred to by a paged fragment 2490 * @frag: the paged fragment 2491 * 2492 * Returns the &struct page associated with @frag. 2493 */ 2494 static inline struct page *skb_frag_page(const skb_frag_t *frag) 2495 { 2496 return frag->page.p; 2497 } 2498 2499 /** 2500 * __skb_frag_ref - take an addition reference on a paged fragment. 2501 * @frag: the paged fragment 2502 * 2503 * Takes an additional reference on the paged fragment @frag. 2504 */ 2505 static inline void __skb_frag_ref(skb_frag_t *frag) 2506 { 2507 get_page(skb_frag_page(frag)); 2508 } 2509 2510 /** 2511 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2512 * @skb: the buffer 2513 * @f: the fragment offset. 2514 * 2515 * Takes an additional reference on the @f'th paged fragment of @skb. 2516 */ 2517 static inline void skb_frag_ref(struct sk_buff *skb, int f) 2518 { 2519 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2520 } 2521 2522 /** 2523 * __skb_frag_unref - release a reference on a paged fragment. 2524 * @frag: the paged fragment 2525 * 2526 * Releases a reference on the paged fragment @frag. 2527 */ 2528 static inline void __skb_frag_unref(skb_frag_t *frag) 2529 { 2530 put_page(skb_frag_page(frag)); 2531 } 2532 2533 /** 2534 * skb_frag_unref - release a reference on a paged fragment of an skb. 2535 * @skb: the buffer 2536 * @f: the fragment offset 2537 * 2538 * Releases a reference on the @f'th paged fragment of @skb. 2539 */ 2540 static inline void skb_frag_unref(struct sk_buff *skb, int f) 2541 { 2542 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2543 } 2544 2545 /** 2546 * skb_frag_address - gets the address of the data contained in a paged fragment 2547 * @frag: the paged fragment buffer 2548 * 2549 * Returns the address of the data within @frag. The page must already 2550 * be mapped. 2551 */ 2552 static inline void *skb_frag_address(const skb_frag_t *frag) 2553 { 2554 return page_address(skb_frag_page(frag)) + frag->page_offset; 2555 } 2556 2557 /** 2558 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2559 * @frag: the paged fragment buffer 2560 * 2561 * Returns the address of the data within @frag. Checks that the page 2562 * is mapped and returns %NULL otherwise. 2563 */ 2564 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2565 { 2566 void *ptr = page_address(skb_frag_page(frag)); 2567 if (unlikely(!ptr)) 2568 return NULL; 2569 2570 return ptr + frag->page_offset; 2571 } 2572 2573 /** 2574 * __skb_frag_set_page - sets the page contained in a paged fragment 2575 * @frag: the paged fragment 2576 * @page: the page to set 2577 * 2578 * Sets the fragment @frag to contain @page. 2579 */ 2580 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2581 { 2582 frag->page.p = page; 2583 } 2584 2585 /** 2586 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2587 * @skb: the buffer 2588 * @f: the fragment offset 2589 * @page: the page to set 2590 * 2591 * Sets the @f'th fragment of @skb to contain @page. 2592 */ 2593 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2594 struct page *page) 2595 { 2596 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2597 } 2598 2599 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2600 2601 /** 2602 * skb_frag_dma_map - maps a paged fragment via the DMA API 2603 * @dev: the device to map the fragment to 2604 * @frag: the paged fragment to map 2605 * @offset: the offset within the fragment (starting at the 2606 * fragment's own offset) 2607 * @size: the number of bytes to map 2608 * @dir: the direction of the mapping (%PCI_DMA_*) 2609 * 2610 * Maps the page associated with @frag to @device. 2611 */ 2612 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2613 const skb_frag_t *frag, 2614 size_t offset, size_t size, 2615 enum dma_data_direction dir) 2616 { 2617 return dma_map_page(dev, skb_frag_page(frag), 2618 frag->page_offset + offset, size, dir); 2619 } 2620 2621 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2622 gfp_t gfp_mask) 2623 { 2624 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2625 } 2626 2627 2628 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2629 gfp_t gfp_mask) 2630 { 2631 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2632 } 2633 2634 2635 /** 2636 * skb_clone_writable - is the header of a clone writable 2637 * @skb: buffer to check 2638 * @len: length up to which to write 2639 * 2640 * Returns true if modifying the header part of the cloned buffer 2641 * does not requires the data to be copied. 2642 */ 2643 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2644 { 2645 return !skb_header_cloned(skb) && 2646 skb_headroom(skb) + len <= skb->hdr_len; 2647 } 2648 2649 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2650 int cloned) 2651 { 2652 int delta = 0; 2653 2654 if (headroom > skb_headroom(skb)) 2655 delta = headroom - skb_headroom(skb); 2656 2657 if (delta || cloned) 2658 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2659 GFP_ATOMIC); 2660 return 0; 2661 } 2662 2663 /** 2664 * skb_cow - copy header of skb when it is required 2665 * @skb: buffer to cow 2666 * @headroom: needed headroom 2667 * 2668 * If the skb passed lacks sufficient headroom or its data part 2669 * is shared, data is reallocated. If reallocation fails, an error 2670 * is returned and original skb is not changed. 2671 * 2672 * The result is skb with writable area skb->head...skb->tail 2673 * and at least @headroom of space at head. 2674 */ 2675 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2676 { 2677 return __skb_cow(skb, headroom, skb_cloned(skb)); 2678 } 2679 2680 /** 2681 * skb_cow_head - skb_cow but only making the head writable 2682 * @skb: buffer to cow 2683 * @headroom: needed headroom 2684 * 2685 * This function is identical to skb_cow except that we replace the 2686 * skb_cloned check by skb_header_cloned. It should be used when 2687 * you only need to push on some header and do not need to modify 2688 * the data. 2689 */ 2690 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2691 { 2692 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2693 } 2694 2695 /** 2696 * skb_padto - pad an skbuff up to a minimal size 2697 * @skb: buffer to pad 2698 * @len: minimal length 2699 * 2700 * Pads up a buffer to ensure the trailing bytes exist and are 2701 * blanked. If the buffer already contains sufficient data it 2702 * is untouched. Otherwise it is extended. Returns zero on 2703 * success. The skb is freed on error. 2704 */ 2705 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2706 { 2707 unsigned int size = skb->len; 2708 if (likely(size >= len)) 2709 return 0; 2710 return skb_pad(skb, len - size); 2711 } 2712 2713 /** 2714 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2715 * @skb: buffer to pad 2716 * @len: minimal length 2717 * 2718 * Pads up a buffer to ensure the trailing bytes exist and are 2719 * blanked. If the buffer already contains sufficient data it 2720 * is untouched. Otherwise it is extended. Returns zero on 2721 * success. The skb is freed on error. 2722 */ 2723 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2724 { 2725 unsigned int size = skb->len; 2726 2727 if (unlikely(size < len)) { 2728 len -= size; 2729 if (skb_pad(skb, len)) 2730 return -ENOMEM; 2731 __skb_put(skb, len); 2732 } 2733 return 0; 2734 } 2735 2736 static inline int skb_add_data(struct sk_buff *skb, 2737 struct iov_iter *from, int copy) 2738 { 2739 const int off = skb->len; 2740 2741 if (skb->ip_summed == CHECKSUM_NONE) { 2742 __wsum csum = 0; 2743 if (csum_and_copy_from_iter(skb_put(skb, copy), copy, 2744 &csum, from) == copy) { 2745 skb->csum = csum_block_add(skb->csum, csum, off); 2746 return 0; 2747 } 2748 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy) 2749 return 0; 2750 2751 __skb_trim(skb, off); 2752 return -EFAULT; 2753 } 2754 2755 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2756 const struct page *page, int off) 2757 { 2758 if (i) { 2759 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2760 2761 return page == skb_frag_page(frag) && 2762 off == frag->page_offset + skb_frag_size(frag); 2763 } 2764 return false; 2765 } 2766 2767 static inline int __skb_linearize(struct sk_buff *skb) 2768 { 2769 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2770 } 2771 2772 /** 2773 * skb_linearize - convert paged skb to linear one 2774 * @skb: buffer to linarize 2775 * 2776 * If there is no free memory -ENOMEM is returned, otherwise zero 2777 * is returned and the old skb data released. 2778 */ 2779 static inline int skb_linearize(struct sk_buff *skb) 2780 { 2781 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2782 } 2783 2784 /** 2785 * skb_has_shared_frag - can any frag be overwritten 2786 * @skb: buffer to test 2787 * 2788 * Return true if the skb has at least one frag that might be modified 2789 * by an external entity (as in vmsplice()/sendfile()) 2790 */ 2791 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 2792 { 2793 return skb_is_nonlinear(skb) && 2794 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2795 } 2796 2797 /** 2798 * skb_linearize_cow - make sure skb is linear and writable 2799 * @skb: buffer to process 2800 * 2801 * If there is no free memory -ENOMEM is returned, otherwise zero 2802 * is returned and the old skb data released. 2803 */ 2804 static inline int skb_linearize_cow(struct sk_buff *skb) 2805 { 2806 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2807 __skb_linearize(skb) : 0; 2808 } 2809 2810 /** 2811 * skb_postpull_rcsum - update checksum for received skb after pull 2812 * @skb: buffer to update 2813 * @start: start of data before pull 2814 * @len: length of data pulled 2815 * 2816 * After doing a pull on a received packet, you need to call this to 2817 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2818 * CHECKSUM_NONE so that it can be recomputed from scratch. 2819 */ 2820 2821 static inline void skb_postpull_rcsum(struct sk_buff *skb, 2822 const void *start, unsigned int len) 2823 { 2824 if (skb->ip_summed == CHECKSUM_COMPLETE) 2825 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0)); 2826 else if (skb->ip_summed == CHECKSUM_PARTIAL && 2827 skb_checksum_start_offset(skb) < 0) 2828 skb->ip_summed = CHECKSUM_NONE; 2829 } 2830 2831 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2832 2833 static inline void skb_postpush_rcsum(struct sk_buff *skb, 2834 const void *start, unsigned int len) 2835 { 2836 /* For performing the reverse operation to skb_postpull_rcsum(), 2837 * we can instead of ... 2838 * 2839 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0)); 2840 * 2841 * ... just use this equivalent version here to save a few 2842 * instructions. Feeding csum of 0 in csum_partial() and later 2843 * on adding skb->csum is equivalent to feed skb->csum in the 2844 * first place. 2845 */ 2846 if (skb->ip_summed == CHECKSUM_COMPLETE) 2847 skb->csum = csum_partial(start, len, skb->csum); 2848 } 2849 2850 /** 2851 * pskb_trim_rcsum - trim received skb and update checksum 2852 * @skb: buffer to trim 2853 * @len: new length 2854 * 2855 * This is exactly the same as pskb_trim except that it ensures the 2856 * checksum of received packets are still valid after the operation. 2857 */ 2858 2859 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2860 { 2861 if (likely(len >= skb->len)) 2862 return 0; 2863 if (skb->ip_summed == CHECKSUM_COMPLETE) 2864 skb->ip_summed = CHECKSUM_NONE; 2865 return __pskb_trim(skb, len); 2866 } 2867 2868 #define skb_queue_walk(queue, skb) \ 2869 for (skb = (queue)->next; \ 2870 skb != (struct sk_buff *)(queue); \ 2871 skb = skb->next) 2872 2873 #define skb_queue_walk_safe(queue, skb, tmp) \ 2874 for (skb = (queue)->next, tmp = skb->next; \ 2875 skb != (struct sk_buff *)(queue); \ 2876 skb = tmp, tmp = skb->next) 2877 2878 #define skb_queue_walk_from(queue, skb) \ 2879 for (; skb != (struct sk_buff *)(queue); \ 2880 skb = skb->next) 2881 2882 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 2883 for (tmp = skb->next; \ 2884 skb != (struct sk_buff *)(queue); \ 2885 skb = tmp, tmp = skb->next) 2886 2887 #define skb_queue_reverse_walk(queue, skb) \ 2888 for (skb = (queue)->prev; \ 2889 skb != (struct sk_buff *)(queue); \ 2890 skb = skb->prev) 2891 2892 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 2893 for (skb = (queue)->prev, tmp = skb->prev; \ 2894 skb != (struct sk_buff *)(queue); \ 2895 skb = tmp, tmp = skb->prev) 2896 2897 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 2898 for (tmp = skb->prev; \ 2899 skb != (struct sk_buff *)(queue); \ 2900 skb = tmp, tmp = skb->prev) 2901 2902 static inline bool skb_has_frag_list(const struct sk_buff *skb) 2903 { 2904 return skb_shinfo(skb)->frag_list != NULL; 2905 } 2906 2907 static inline void skb_frag_list_init(struct sk_buff *skb) 2908 { 2909 skb_shinfo(skb)->frag_list = NULL; 2910 } 2911 2912 #define skb_walk_frags(skb, iter) \ 2913 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 2914 2915 2916 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, 2917 const struct sk_buff *skb); 2918 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags, 2919 int *peeked, int *off, int *err, 2920 struct sk_buff **last); 2921 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 2922 int *peeked, int *off, int *err); 2923 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 2924 int *err); 2925 unsigned int datagram_poll(struct file *file, struct socket *sock, 2926 struct poll_table_struct *wait); 2927 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 2928 struct iov_iter *to, int size); 2929 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 2930 struct msghdr *msg, int size) 2931 { 2932 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 2933 } 2934 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 2935 struct msghdr *msg); 2936 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 2937 struct iov_iter *from, int len); 2938 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 2939 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 2940 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb); 2941 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 2942 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 2943 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 2944 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 2945 int len, __wsum csum); 2946 ssize_t skb_socket_splice(struct sock *sk, 2947 struct pipe_inode_info *pipe, 2948 struct splice_pipe_desc *spd); 2949 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 2950 struct pipe_inode_info *pipe, unsigned int len, 2951 unsigned int flags, 2952 ssize_t (*splice_cb)(struct sock *, 2953 struct pipe_inode_info *, 2954 struct splice_pipe_desc *)); 2955 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 2956 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 2957 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 2958 int len, int hlen); 2959 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 2960 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 2961 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 2962 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 2963 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 2964 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 2965 int skb_ensure_writable(struct sk_buff *skb, int write_len); 2966 int skb_vlan_pop(struct sk_buff *skb); 2967 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 2968 2969 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 2970 { 2971 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2972 } 2973 2974 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 2975 { 2976 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2977 } 2978 2979 struct skb_checksum_ops { 2980 __wsum (*update)(const void *mem, int len, __wsum wsum); 2981 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 2982 }; 2983 2984 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 2985 __wsum csum, const struct skb_checksum_ops *ops); 2986 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 2987 __wsum csum); 2988 2989 static inline void * __must_check 2990 __skb_header_pointer(const struct sk_buff *skb, int offset, 2991 int len, void *data, int hlen, void *buffer) 2992 { 2993 if (hlen - offset >= len) 2994 return data + offset; 2995 2996 if (!skb || 2997 skb_copy_bits(skb, offset, buffer, len) < 0) 2998 return NULL; 2999 3000 return buffer; 3001 } 3002 3003 static inline void * __must_check 3004 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 3005 { 3006 return __skb_header_pointer(skb, offset, len, skb->data, 3007 skb_headlen(skb), buffer); 3008 } 3009 3010 /** 3011 * skb_needs_linearize - check if we need to linearize a given skb 3012 * depending on the given device features. 3013 * @skb: socket buffer to check 3014 * @features: net device features 3015 * 3016 * Returns true if either: 3017 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3018 * 2. skb is fragmented and the device does not support SG. 3019 */ 3020 static inline bool skb_needs_linearize(struct sk_buff *skb, 3021 netdev_features_t features) 3022 { 3023 return skb_is_nonlinear(skb) && 3024 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3025 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3026 } 3027 3028 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3029 void *to, 3030 const unsigned int len) 3031 { 3032 memcpy(to, skb->data, len); 3033 } 3034 3035 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3036 const int offset, void *to, 3037 const unsigned int len) 3038 { 3039 memcpy(to, skb->data + offset, len); 3040 } 3041 3042 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3043 const void *from, 3044 const unsigned int len) 3045 { 3046 memcpy(skb->data, from, len); 3047 } 3048 3049 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3050 const int offset, 3051 const void *from, 3052 const unsigned int len) 3053 { 3054 memcpy(skb->data + offset, from, len); 3055 } 3056 3057 void skb_init(void); 3058 3059 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3060 { 3061 return skb->tstamp; 3062 } 3063 3064 /** 3065 * skb_get_timestamp - get timestamp from a skb 3066 * @skb: skb to get stamp from 3067 * @stamp: pointer to struct timeval to store stamp in 3068 * 3069 * Timestamps are stored in the skb as offsets to a base timestamp. 3070 * This function converts the offset back to a struct timeval and stores 3071 * it in stamp. 3072 */ 3073 static inline void skb_get_timestamp(const struct sk_buff *skb, 3074 struct timeval *stamp) 3075 { 3076 *stamp = ktime_to_timeval(skb->tstamp); 3077 } 3078 3079 static inline void skb_get_timestampns(const struct sk_buff *skb, 3080 struct timespec *stamp) 3081 { 3082 *stamp = ktime_to_timespec(skb->tstamp); 3083 } 3084 3085 static inline void __net_timestamp(struct sk_buff *skb) 3086 { 3087 skb->tstamp = ktime_get_real(); 3088 } 3089 3090 static inline ktime_t net_timedelta(ktime_t t) 3091 { 3092 return ktime_sub(ktime_get_real(), t); 3093 } 3094 3095 static inline ktime_t net_invalid_timestamp(void) 3096 { 3097 return ktime_set(0, 0); 3098 } 3099 3100 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3101 3102 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3103 3104 void skb_clone_tx_timestamp(struct sk_buff *skb); 3105 bool skb_defer_rx_timestamp(struct sk_buff *skb); 3106 3107 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3108 3109 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3110 { 3111 } 3112 3113 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3114 { 3115 return false; 3116 } 3117 3118 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3119 3120 /** 3121 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3122 * 3123 * PHY drivers may accept clones of transmitted packets for 3124 * timestamping via their phy_driver.txtstamp method. These drivers 3125 * must call this function to return the skb back to the stack with a 3126 * timestamp. 3127 * 3128 * @skb: clone of the the original outgoing packet 3129 * @hwtstamps: hardware time stamps 3130 * 3131 */ 3132 void skb_complete_tx_timestamp(struct sk_buff *skb, 3133 struct skb_shared_hwtstamps *hwtstamps); 3134 3135 void __skb_tstamp_tx(struct sk_buff *orig_skb, 3136 struct skb_shared_hwtstamps *hwtstamps, 3137 struct sock *sk, int tstype); 3138 3139 /** 3140 * skb_tstamp_tx - queue clone of skb with send time stamps 3141 * @orig_skb: the original outgoing packet 3142 * @hwtstamps: hardware time stamps, may be NULL if not available 3143 * 3144 * If the skb has a socket associated, then this function clones the 3145 * skb (thus sharing the actual data and optional structures), stores 3146 * the optional hardware time stamping information (if non NULL) or 3147 * generates a software time stamp (otherwise), then queues the clone 3148 * to the error queue of the socket. Errors are silently ignored. 3149 */ 3150 void skb_tstamp_tx(struct sk_buff *orig_skb, 3151 struct skb_shared_hwtstamps *hwtstamps); 3152 3153 static inline void sw_tx_timestamp(struct sk_buff *skb) 3154 { 3155 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 3156 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 3157 skb_tstamp_tx(skb, NULL); 3158 } 3159 3160 /** 3161 * skb_tx_timestamp() - Driver hook for transmit timestamping 3162 * 3163 * Ethernet MAC Drivers should call this function in their hard_xmit() 3164 * function immediately before giving the sk_buff to the MAC hardware. 3165 * 3166 * Specifically, one should make absolutely sure that this function is 3167 * called before TX completion of this packet can trigger. Otherwise 3168 * the packet could potentially already be freed. 3169 * 3170 * @skb: A socket buffer. 3171 */ 3172 static inline void skb_tx_timestamp(struct sk_buff *skb) 3173 { 3174 skb_clone_tx_timestamp(skb); 3175 sw_tx_timestamp(skb); 3176 } 3177 3178 /** 3179 * skb_complete_wifi_ack - deliver skb with wifi status 3180 * 3181 * @skb: the original outgoing packet 3182 * @acked: ack status 3183 * 3184 */ 3185 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3186 3187 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3188 __sum16 __skb_checksum_complete(struct sk_buff *skb); 3189 3190 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3191 { 3192 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3193 skb->csum_valid || 3194 (skb->ip_summed == CHECKSUM_PARTIAL && 3195 skb_checksum_start_offset(skb) >= 0)); 3196 } 3197 3198 /** 3199 * skb_checksum_complete - Calculate checksum of an entire packet 3200 * @skb: packet to process 3201 * 3202 * This function calculates the checksum over the entire packet plus 3203 * the value of skb->csum. The latter can be used to supply the 3204 * checksum of a pseudo header as used by TCP/UDP. It returns the 3205 * checksum. 3206 * 3207 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3208 * this function can be used to verify that checksum on received 3209 * packets. In that case the function should return zero if the 3210 * checksum is correct. In particular, this function will return zero 3211 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3212 * hardware has already verified the correctness of the checksum. 3213 */ 3214 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3215 { 3216 return skb_csum_unnecessary(skb) ? 3217 0 : __skb_checksum_complete(skb); 3218 } 3219 3220 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3221 { 3222 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3223 if (skb->csum_level == 0) 3224 skb->ip_summed = CHECKSUM_NONE; 3225 else 3226 skb->csum_level--; 3227 } 3228 } 3229 3230 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3231 { 3232 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3233 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3234 skb->csum_level++; 3235 } else if (skb->ip_summed == CHECKSUM_NONE) { 3236 skb->ip_summed = CHECKSUM_UNNECESSARY; 3237 skb->csum_level = 0; 3238 } 3239 } 3240 3241 static inline void __skb_mark_checksum_bad(struct sk_buff *skb) 3242 { 3243 /* Mark current checksum as bad (typically called from GRO 3244 * path). In the case that ip_summed is CHECKSUM_NONE 3245 * this must be the first checksum encountered in the packet. 3246 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first 3247 * checksum after the last one validated. For UDP, a zero 3248 * checksum can not be marked as bad. 3249 */ 3250 3251 if (skb->ip_summed == CHECKSUM_NONE || 3252 skb->ip_summed == CHECKSUM_UNNECESSARY) 3253 skb->csum_bad = 1; 3254 } 3255 3256 /* Check if we need to perform checksum complete validation. 3257 * 3258 * Returns true if checksum complete is needed, false otherwise 3259 * (either checksum is unnecessary or zero checksum is allowed). 3260 */ 3261 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3262 bool zero_okay, 3263 __sum16 check) 3264 { 3265 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3266 skb->csum_valid = 1; 3267 __skb_decr_checksum_unnecessary(skb); 3268 return false; 3269 } 3270 3271 return true; 3272 } 3273 3274 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3275 * in checksum_init. 3276 */ 3277 #define CHECKSUM_BREAK 76 3278 3279 /* Unset checksum-complete 3280 * 3281 * Unset checksum complete can be done when packet is being modified 3282 * (uncompressed for instance) and checksum-complete value is 3283 * invalidated. 3284 */ 3285 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3286 { 3287 if (skb->ip_summed == CHECKSUM_COMPLETE) 3288 skb->ip_summed = CHECKSUM_NONE; 3289 } 3290 3291 /* Validate (init) checksum based on checksum complete. 3292 * 3293 * Return values: 3294 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3295 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3296 * checksum is stored in skb->csum for use in __skb_checksum_complete 3297 * non-zero: value of invalid checksum 3298 * 3299 */ 3300 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3301 bool complete, 3302 __wsum psum) 3303 { 3304 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3305 if (!csum_fold(csum_add(psum, skb->csum))) { 3306 skb->csum_valid = 1; 3307 return 0; 3308 } 3309 } else if (skb->csum_bad) { 3310 /* ip_summed == CHECKSUM_NONE in this case */ 3311 return (__force __sum16)1; 3312 } 3313 3314 skb->csum = psum; 3315 3316 if (complete || skb->len <= CHECKSUM_BREAK) { 3317 __sum16 csum; 3318 3319 csum = __skb_checksum_complete(skb); 3320 skb->csum_valid = !csum; 3321 return csum; 3322 } 3323 3324 return 0; 3325 } 3326 3327 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3328 { 3329 return 0; 3330 } 3331 3332 /* Perform checksum validate (init). Note that this is a macro since we only 3333 * want to calculate the pseudo header which is an input function if necessary. 3334 * First we try to validate without any computation (checksum unnecessary) and 3335 * then calculate based on checksum complete calling the function to compute 3336 * pseudo header. 3337 * 3338 * Return values: 3339 * 0: checksum is validated or try to in skb_checksum_complete 3340 * non-zero: value of invalid checksum 3341 */ 3342 #define __skb_checksum_validate(skb, proto, complete, \ 3343 zero_okay, check, compute_pseudo) \ 3344 ({ \ 3345 __sum16 __ret = 0; \ 3346 skb->csum_valid = 0; \ 3347 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3348 __ret = __skb_checksum_validate_complete(skb, \ 3349 complete, compute_pseudo(skb, proto)); \ 3350 __ret; \ 3351 }) 3352 3353 #define skb_checksum_init(skb, proto, compute_pseudo) \ 3354 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3355 3356 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3357 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3358 3359 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 3360 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3361 3362 #define skb_checksum_validate_zero_check(skb, proto, check, \ 3363 compute_pseudo) \ 3364 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3365 3366 #define skb_checksum_simple_validate(skb) \ 3367 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3368 3369 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3370 { 3371 return (skb->ip_summed == CHECKSUM_NONE && 3372 skb->csum_valid && !skb->csum_bad); 3373 } 3374 3375 static inline void __skb_checksum_convert(struct sk_buff *skb, 3376 __sum16 check, __wsum pseudo) 3377 { 3378 skb->csum = ~pseudo; 3379 skb->ip_summed = CHECKSUM_COMPLETE; 3380 } 3381 3382 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3383 do { \ 3384 if (__skb_checksum_convert_check(skb)) \ 3385 __skb_checksum_convert(skb, check, \ 3386 compute_pseudo(skb, proto)); \ 3387 } while (0) 3388 3389 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3390 u16 start, u16 offset) 3391 { 3392 skb->ip_summed = CHECKSUM_PARTIAL; 3393 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3394 skb->csum_offset = offset - start; 3395 } 3396 3397 /* Update skbuf and packet to reflect the remote checksum offload operation. 3398 * When called, ptr indicates the starting point for skb->csum when 3399 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3400 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3401 */ 3402 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3403 int start, int offset, bool nopartial) 3404 { 3405 __wsum delta; 3406 3407 if (!nopartial) { 3408 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3409 return; 3410 } 3411 3412 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3413 __skb_checksum_complete(skb); 3414 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3415 } 3416 3417 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3418 3419 /* Adjust skb->csum since we changed the packet */ 3420 skb->csum = csum_add(skb->csum, delta); 3421 } 3422 3423 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3424 void nf_conntrack_destroy(struct nf_conntrack *nfct); 3425 static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3426 { 3427 if (nfct && atomic_dec_and_test(&nfct->use)) 3428 nf_conntrack_destroy(nfct); 3429 } 3430 static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3431 { 3432 if (nfct) 3433 atomic_inc(&nfct->use); 3434 } 3435 #endif 3436 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3437 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3438 { 3439 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 3440 kfree(nf_bridge); 3441 } 3442 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3443 { 3444 if (nf_bridge) 3445 atomic_inc(&nf_bridge->use); 3446 } 3447 #endif /* CONFIG_BRIDGE_NETFILTER */ 3448 static inline void nf_reset(struct sk_buff *skb) 3449 { 3450 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3451 nf_conntrack_put(skb->nfct); 3452 skb->nfct = NULL; 3453 #endif 3454 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3455 nf_bridge_put(skb->nf_bridge); 3456 skb->nf_bridge = NULL; 3457 #endif 3458 } 3459 3460 static inline void nf_reset_trace(struct sk_buff *skb) 3461 { 3462 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3463 skb->nf_trace = 0; 3464 #endif 3465 } 3466 3467 /* Note: This doesn't put any conntrack and bridge info in dst. */ 3468 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3469 bool copy) 3470 { 3471 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3472 dst->nfct = src->nfct; 3473 nf_conntrack_get(src->nfct); 3474 if (copy) 3475 dst->nfctinfo = src->nfctinfo; 3476 #endif 3477 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3478 dst->nf_bridge = src->nf_bridge; 3479 nf_bridge_get(src->nf_bridge); 3480 #endif 3481 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3482 if (copy) 3483 dst->nf_trace = src->nf_trace; 3484 #endif 3485 } 3486 3487 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3488 { 3489 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3490 nf_conntrack_put(dst->nfct); 3491 #endif 3492 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3493 nf_bridge_put(dst->nf_bridge); 3494 #endif 3495 __nf_copy(dst, src, true); 3496 } 3497 3498 #ifdef CONFIG_NETWORK_SECMARK 3499 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3500 { 3501 to->secmark = from->secmark; 3502 } 3503 3504 static inline void skb_init_secmark(struct sk_buff *skb) 3505 { 3506 skb->secmark = 0; 3507 } 3508 #else 3509 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3510 { } 3511 3512 static inline void skb_init_secmark(struct sk_buff *skb) 3513 { } 3514 #endif 3515 3516 static inline bool skb_irq_freeable(const struct sk_buff *skb) 3517 { 3518 return !skb->destructor && 3519 #if IS_ENABLED(CONFIG_XFRM) 3520 !skb->sp && 3521 #endif 3522 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 3523 !skb->nfct && 3524 #endif 3525 !skb->_skb_refdst && 3526 !skb_has_frag_list(skb); 3527 } 3528 3529 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3530 { 3531 skb->queue_mapping = queue_mapping; 3532 } 3533 3534 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3535 { 3536 return skb->queue_mapping; 3537 } 3538 3539 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3540 { 3541 to->queue_mapping = from->queue_mapping; 3542 } 3543 3544 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3545 { 3546 skb->queue_mapping = rx_queue + 1; 3547 } 3548 3549 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3550 { 3551 return skb->queue_mapping - 1; 3552 } 3553 3554 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3555 { 3556 return skb->queue_mapping != 0; 3557 } 3558 3559 static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3560 { 3561 #ifdef CONFIG_XFRM 3562 return skb->sp; 3563 #else 3564 return NULL; 3565 #endif 3566 } 3567 3568 /* Keeps track of mac header offset relative to skb->head. 3569 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3570 * For non-tunnel skb it points to skb_mac_header() and for 3571 * tunnel skb it points to outer mac header. 3572 * Keeps track of level of encapsulation of network headers. 3573 */ 3574 struct skb_gso_cb { 3575 int mac_offset; 3576 int encap_level; 3577 __u16 csum_start; 3578 }; 3579 #define SKB_SGO_CB_OFFSET 32 3580 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 3581 3582 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3583 { 3584 return (skb_mac_header(inner_skb) - inner_skb->head) - 3585 SKB_GSO_CB(inner_skb)->mac_offset; 3586 } 3587 3588 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3589 { 3590 int new_headroom, headroom; 3591 int ret; 3592 3593 headroom = skb_headroom(skb); 3594 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3595 if (ret) 3596 return ret; 3597 3598 new_headroom = skb_headroom(skb); 3599 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3600 return 0; 3601 } 3602 3603 /* Compute the checksum for a gso segment. First compute the checksum value 3604 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3605 * then add in skb->csum (checksum from csum_start to end of packet). 3606 * skb->csum and csum_start are then updated to reflect the checksum of the 3607 * resultant packet starting from the transport header-- the resultant checksum 3608 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3609 * header. 3610 */ 3611 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3612 { 3613 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) - 3614 skb_transport_offset(skb); 3615 __wsum partial; 3616 3617 partial = csum_partial(skb_transport_header(skb), plen, skb->csum); 3618 skb->csum = res; 3619 SKB_GSO_CB(skb)->csum_start -= plen; 3620 3621 return csum_fold(partial); 3622 } 3623 3624 static inline bool skb_is_gso(const struct sk_buff *skb) 3625 { 3626 return skb_shinfo(skb)->gso_size; 3627 } 3628 3629 /* Note: Should be called only if skb_is_gso(skb) is true */ 3630 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3631 { 3632 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3633 } 3634 3635 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3636 3637 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3638 { 3639 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3640 * wanted then gso_type will be set. */ 3641 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3642 3643 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3644 unlikely(shinfo->gso_type == 0)) { 3645 __skb_warn_lro_forwarding(skb); 3646 return true; 3647 } 3648 return false; 3649 } 3650 3651 static inline void skb_forward_csum(struct sk_buff *skb) 3652 { 3653 /* Unfortunately we don't support this one. Any brave souls? */ 3654 if (skb->ip_summed == CHECKSUM_COMPLETE) 3655 skb->ip_summed = CHECKSUM_NONE; 3656 } 3657 3658 /** 3659 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3660 * @skb: skb to check 3661 * 3662 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3663 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3664 * use this helper, to document places where we make this assertion. 3665 */ 3666 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3667 { 3668 #ifdef DEBUG 3669 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3670 #endif 3671 } 3672 3673 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3674 3675 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3676 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 3677 unsigned int transport_len, 3678 __sum16(*skb_chkf)(struct sk_buff *skb)); 3679 3680 /** 3681 * skb_head_is_locked - Determine if the skb->head is locked down 3682 * @skb: skb to check 3683 * 3684 * The head on skbs build around a head frag can be removed if they are 3685 * not cloned. This function returns true if the skb head is locked down 3686 * due to either being allocated via kmalloc, or by being a clone with 3687 * multiple references to the head. 3688 */ 3689 static inline bool skb_head_is_locked(const struct sk_buff *skb) 3690 { 3691 return !skb->head_frag || skb_cloned(skb); 3692 } 3693 3694 /** 3695 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3696 * 3697 * @skb: GSO skb 3698 * 3699 * skb_gso_network_seglen is used to determine the real size of the 3700 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3701 * 3702 * The MAC/L2 header is not accounted for. 3703 */ 3704 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 3705 { 3706 unsigned int hdr_len = skb_transport_header(skb) - 3707 skb_network_header(skb); 3708 return hdr_len + skb_gso_transport_seglen(skb); 3709 } 3710 3711 #endif /* __KERNEL__ */ 3712 #endif /* _LINUX_SKBUFF_H */ 3713