1 /* 2 * Definitions for the 'struct sk_buff' memory handlers. 3 * 4 * Authors: 5 * Alan Cox, <gw4pts@gw4pts.ampr.org> 6 * Florian La Roche, <rzsfl@rz.uni-sb.de> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 */ 13 14 #ifndef _LINUX_SKBUFF_H 15 #define _LINUX_SKBUFF_H 16 17 #include <linux/kernel.h> 18 #include <linux/kmemcheck.h> 19 #include <linux/compiler.h> 20 #include <linux/time.h> 21 #include <linux/bug.h> 22 #include <linux/cache.h> 23 #include <linux/rbtree.h> 24 #include <linux/socket.h> 25 26 #include <linux/atomic.h> 27 #include <asm/types.h> 28 #include <linux/spinlock.h> 29 #include <linux/net.h> 30 #include <linux/textsearch.h> 31 #include <net/checksum.h> 32 #include <linux/rcupdate.h> 33 #include <linux/hrtimer.h> 34 #include <linux/dma-mapping.h> 35 #include <linux/netdev_features.h> 36 #include <linux/sched.h> 37 #include <net/flow_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 }; 386 387 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 388 SKBTX_SCHED_TSTAMP) 389 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 390 391 /* 392 * The callback notifies userspace to release buffers when skb DMA is done in 393 * lower device, the skb last reference should be 0 when calling this. 394 * The zerocopy_success argument is true if zero copy transmit occurred, 395 * false on data copy or out of memory error caused by data copy attempt. 396 * The ctx field is used to track device context. 397 * The desc field is used to track userspace buffer index. 398 */ 399 struct ubuf_info { 400 void (*callback)(struct ubuf_info *, bool zerocopy_success); 401 void *ctx; 402 unsigned long desc; 403 }; 404 405 /* This data is invariant across clones and lives at 406 * the end of the header data, ie. at skb->end. 407 */ 408 struct skb_shared_info { 409 unsigned char nr_frags; 410 __u8 tx_flags; 411 unsigned short gso_size; 412 /* Warning: this field is not always filled in (UFO)! */ 413 unsigned short gso_segs; 414 unsigned short gso_type; 415 struct sk_buff *frag_list; 416 struct skb_shared_hwtstamps hwtstamps; 417 u32 tskey; 418 __be32 ip6_frag_id; 419 420 /* 421 * Warning : all fields before dataref are cleared in __alloc_skb() 422 */ 423 atomic_t dataref; 424 425 /* Intermediate layers must ensure that destructor_arg 426 * remains valid until skb destructor */ 427 void * destructor_arg; 428 429 /* must be last field, see pskb_expand_head() */ 430 skb_frag_t frags[MAX_SKB_FRAGS]; 431 }; 432 433 /* We divide dataref into two halves. The higher 16 bits hold references 434 * to the payload part of skb->data. The lower 16 bits hold references to 435 * the entire skb->data. A clone of a headerless skb holds the length of 436 * the header in skb->hdr_len. 437 * 438 * All users must obey the rule that the skb->data reference count must be 439 * greater than or equal to the payload reference count. 440 * 441 * Holding a reference to the payload part means that the user does not 442 * care about modifications to the header part of skb->data. 443 */ 444 #define SKB_DATAREF_SHIFT 16 445 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 446 447 448 enum { 449 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 450 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 451 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 452 }; 453 454 enum { 455 SKB_GSO_TCPV4 = 1 << 0, 456 SKB_GSO_UDP = 1 << 1, 457 458 /* This indicates the skb is from an untrusted source. */ 459 SKB_GSO_DODGY = 1 << 2, 460 461 /* This indicates the tcp segment has CWR set. */ 462 SKB_GSO_TCP_ECN = 1 << 3, 463 464 SKB_GSO_TCP_FIXEDID = 1 << 4, 465 466 SKB_GSO_TCPV6 = 1 << 5, 467 468 SKB_GSO_FCOE = 1 << 6, 469 470 SKB_GSO_GRE = 1 << 7, 471 472 SKB_GSO_GRE_CSUM = 1 << 8, 473 474 SKB_GSO_IPXIP4 = 1 << 9, 475 476 SKB_GSO_IPXIP6 = 1 << 10, 477 478 SKB_GSO_UDP_TUNNEL = 1 << 11, 479 480 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12, 481 482 SKB_GSO_PARTIAL = 1 << 13, 483 484 SKB_GSO_TUNNEL_REMCSUM = 1 << 14, 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_hash_symmetric(struct sk_buff *skb); 1066 u32 skb_get_poff(const struct sk_buff *skb); 1067 u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1068 const struct flow_keys *keys, int hlen); 1069 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1070 void *data, int hlen_proto); 1071 1072 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1073 int thoff, u8 ip_proto) 1074 { 1075 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1076 } 1077 1078 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1079 const struct flow_dissector_key *key, 1080 unsigned int key_count); 1081 1082 bool __skb_flow_dissect(const struct sk_buff *skb, 1083 struct flow_dissector *flow_dissector, 1084 void *target_container, 1085 void *data, __be16 proto, int nhoff, int hlen, 1086 unsigned int flags); 1087 1088 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1089 struct flow_dissector *flow_dissector, 1090 void *target_container, unsigned int flags) 1091 { 1092 return __skb_flow_dissect(skb, flow_dissector, target_container, 1093 NULL, 0, 0, 0, flags); 1094 } 1095 1096 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1097 struct flow_keys *flow, 1098 unsigned int flags) 1099 { 1100 memset(flow, 0, sizeof(*flow)); 1101 return __skb_flow_dissect(skb, &flow_keys_dissector, flow, 1102 NULL, 0, 0, 0, flags); 1103 } 1104 1105 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow, 1106 void *data, __be16 proto, 1107 int nhoff, int hlen, 1108 unsigned int flags) 1109 { 1110 memset(flow, 0, sizeof(*flow)); 1111 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow, 1112 data, proto, nhoff, hlen, flags); 1113 } 1114 1115 static inline __u32 skb_get_hash(struct sk_buff *skb) 1116 { 1117 if (!skb->l4_hash && !skb->sw_hash) 1118 __skb_get_hash(skb); 1119 1120 return skb->hash; 1121 } 1122 1123 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6); 1124 1125 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1126 { 1127 if (!skb->l4_hash && !skb->sw_hash) { 1128 struct flow_keys keys; 1129 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1130 1131 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1132 } 1133 1134 return skb->hash; 1135 } 1136 1137 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl); 1138 1139 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4) 1140 { 1141 if (!skb->l4_hash && !skb->sw_hash) { 1142 struct flow_keys keys; 1143 __u32 hash = __get_hash_from_flowi4(fl4, &keys); 1144 1145 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1146 } 1147 1148 return skb->hash; 1149 } 1150 1151 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1152 1153 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1154 { 1155 return skb->hash; 1156 } 1157 1158 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1159 { 1160 to->hash = from->hash; 1161 to->sw_hash = from->sw_hash; 1162 to->l4_hash = from->l4_hash; 1163 }; 1164 1165 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1166 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1167 { 1168 return skb->head + skb->end; 1169 } 1170 1171 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1172 { 1173 return skb->end; 1174 } 1175 #else 1176 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1177 { 1178 return skb->end; 1179 } 1180 1181 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1182 { 1183 return skb->end - skb->head; 1184 } 1185 #endif 1186 1187 /* Internal */ 1188 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1189 1190 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1191 { 1192 return &skb_shinfo(skb)->hwtstamps; 1193 } 1194 1195 /** 1196 * skb_queue_empty - check if a queue is empty 1197 * @list: queue head 1198 * 1199 * Returns true if the queue is empty, false otherwise. 1200 */ 1201 static inline int skb_queue_empty(const struct sk_buff_head *list) 1202 { 1203 return list->next == (const struct sk_buff *) list; 1204 } 1205 1206 /** 1207 * skb_queue_is_last - check if skb is the last entry in the queue 1208 * @list: queue head 1209 * @skb: buffer 1210 * 1211 * Returns true if @skb is the last buffer on the list. 1212 */ 1213 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1214 const struct sk_buff *skb) 1215 { 1216 return skb->next == (const struct sk_buff *) list; 1217 } 1218 1219 /** 1220 * skb_queue_is_first - check if skb is the first entry in the queue 1221 * @list: queue head 1222 * @skb: buffer 1223 * 1224 * Returns true if @skb is the first buffer on the list. 1225 */ 1226 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1227 const struct sk_buff *skb) 1228 { 1229 return skb->prev == (const struct sk_buff *) list; 1230 } 1231 1232 /** 1233 * skb_queue_next - return the next packet in the queue 1234 * @list: queue head 1235 * @skb: current buffer 1236 * 1237 * Return the next packet in @list after @skb. It is only valid to 1238 * call this if skb_queue_is_last() evaluates to false. 1239 */ 1240 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1241 const struct sk_buff *skb) 1242 { 1243 /* This BUG_ON may seem severe, but if we just return then we 1244 * are going to dereference garbage. 1245 */ 1246 BUG_ON(skb_queue_is_last(list, skb)); 1247 return skb->next; 1248 } 1249 1250 /** 1251 * skb_queue_prev - return the prev packet in the queue 1252 * @list: queue head 1253 * @skb: current buffer 1254 * 1255 * Return the prev packet in @list before @skb. It is only valid to 1256 * call this if skb_queue_is_first() evaluates to false. 1257 */ 1258 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1259 const struct sk_buff *skb) 1260 { 1261 /* This BUG_ON may seem severe, but if we just return then we 1262 * are going to dereference garbage. 1263 */ 1264 BUG_ON(skb_queue_is_first(list, skb)); 1265 return skb->prev; 1266 } 1267 1268 /** 1269 * skb_get - reference buffer 1270 * @skb: buffer to reference 1271 * 1272 * Makes another reference to a socket buffer and returns a pointer 1273 * to the buffer. 1274 */ 1275 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1276 { 1277 atomic_inc(&skb->users); 1278 return skb; 1279 } 1280 1281 /* 1282 * If users == 1, we are the only owner and are can avoid redundant 1283 * atomic change. 1284 */ 1285 1286 /** 1287 * skb_cloned - is the buffer a clone 1288 * @skb: buffer to check 1289 * 1290 * Returns true if the buffer was generated with skb_clone() and is 1291 * one of multiple shared copies of the buffer. Cloned buffers are 1292 * shared data so must not be written to under normal circumstances. 1293 */ 1294 static inline int skb_cloned(const struct sk_buff *skb) 1295 { 1296 return skb->cloned && 1297 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1298 } 1299 1300 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1301 { 1302 might_sleep_if(gfpflags_allow_blocking(pri)); 1303 1304 if (skb_cloned(skb)) 1305 return pskb_expand_head(skb, 0, 0, pri); 1306 1307 return 0; 1308 } 1309 1310 /** 1311 * skb_header_cloned - is the header a clone 1312 * @skb: buffer to check 1313 * 1314 * Returns true if modifying the header part of the buffer requires 1315 * the data to be copied. 1316 */ 1317 static inline int skb_header_cloned(const struct sk_buff *skb) 1318 { 1319 int dataref; 1320 1321 if (!skb->cloned) 1322 return 0; 1323 1324 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1325 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1326 return dataref != 1; 1327 } 1328 1329 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1330 { 1331 might_sleep_if(gfpflags_allow_blocking(pri)); 1332 1333 if (skb_header_cloned(skb)) 1334 return pskb_expand_head(skb, 0, 0, pri); 1335 1336 return 0; 1337 } 1338 1339 /** 1340 * skb_header_release - release reference to header 1341 * @skb: buffer to operate on 1342 * 1343 * Drop a reference to the header part of the buffer. This is done 1344 * by acquiring a payload reference. You must not read from the header 1345 * part of skb->data after this. 1346 * Note : Check if you can use __skb_header_release() instead. 1347 */ 1348 static inline void skb_header_release(struct sk_buff *skb) 1349 { 1350 BUG_ON(skb->nohdr); 1351 skb->nohdr = 1; 1352 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1353 } 1354 1355 /** 1356 * __skb_header_release - release reference to header 1357 * @skb: buffer to operate on 1358 * 1359 * Variant of skb_header_release() assuming skb is private to caller. 1360 * We can avoid one atomic operation. 1361 */ 1362 static inline void __skb_header_release(struct sk_buff *skb) 1363 { 1364 skb->nohdr = 1; 1365 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1366 } 1367 1368 1369 /** 1370 * skb_shared - is the buffer shared 1371 * @skb: buffer to check 1372 * 1373 * Returns true if more than one person has a reference to this 1374 * buffer. 1375 */ 1376 static inline int skb_shared(const struct sk_buff *skb) 1377 { 1378 return atomic_read(&skb->users) != 1; 1379 } 1380 1381 /** 1382 * skb_share_check - check if buffer is shared and if so clone it 1383 * @skb: buffer to check 1384 * @pri: priority for memory allocation 1385 * 1386 * If the buffer is shared the buffer is cloned and the old copy 1387 * drops a reference. A new clone with a single reference is returned. 1388 * If the buffer is not shared the original buffer is returned. When 1389 * being called from interrupt status or with spinlocks held pri must 1390 * be GFP_ATOMIC. 1391 * 1392 * NULL is returned on a memory allocation failure. 1393 */ 1394 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1395 { 1396 might_sleep_if(gfpflags_allow_blocking(pri)); 1397 if (skb_shared(skb)) { 1398 struct sk_buff *nskb = skb_clone(skb, pri); 1399 1400 if (likely(nskb)) 1401 consume_skb(skb); 1402 else 1403 kfree_skb(skb); 1404 skb = nskb; 1405 } 1406 return skb; 1407 } 1408 1409 /* 1410 * Copy shared buffers into a new sk_buff. We effectively do COW on 1411 * packets to handle cases where we have a local reader and forward 1412 * and a couple of other messy ones. The normal one is tcpdumping 1413 * a packet thats being forwarded. 1414 */ 1415 1416 /** 1417 * skb_unshare - make a copy of a shared buffer 1418 * @skb: buffer to check 1419 * @pri: priority for memory allocation 1420 * 1421 * If the socket buffer is a clone then this function creates a new 1422 * copy of the data, drops a reference count on the old copy and returns 1423 * the new copy with the reference count at 1. If the buffer is not a clone 1424 * the original buffer is returned. When called with a spinlock held or 1425 * from interrupt state @pri must be %GFP_ATOMIC 1426 * 1427 * %NULL is returned on a memory allocation failure. 1428 */ 1429 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1430 gfp_t pri) 1431 { 1432 might_sleep_if(gfpflags_allow_blocking(pri)); 1433 if (skb_cloned(skb)) { 1434 struct sk_buff *nskb = skb_copy(skb, pri); 1435 1436 /* Free our shared copy */ 1437 if (likely(nskb)) 1438 consume_skb(skb); 1439 else 1440 kfree_skb(skb); 1441 skb = nskb; 1442 } 1443 return skb; 1444 } 1445 1446 /** 1447 * skb_peek - peek at the head of an &sk_buff_head 1448 * @list_: list to peek at 1449 * 1450 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1451 * be careful with this one. A peek leaves the buffer on the 1452 * list and someone else may run off with it. You must hold 1453 * the appropriate locks or have a private queue to do this. 1454 * 1455 * Returns %NULL for an empty list or a pointer to the head element. 1456 * The reference count is not incremented and the reference is therefore 1457 * volatile. Use with caution. 1458 */ 1459 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1460 { 1461 struct sk_buff *skb = list_->next; 1462 1463 if (skb == (struct sk_buff *)list_) 1464 skb = NULL; 1465 return skb; 1466 } 1467 1468 /** 1469 * skb_peek_next - peek skb following the given one from a queue 1470 * @skb: skb to start from 1471 * @list_: list to peek at 1472 * 1473 * Returns %NULL when the end of the list is met or a pointer to the 1474 * next element. The reference count is not incremented and the 1475 * reference is therefore volatile. Use with caution. 1476 */ 1477 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1478 const struct sk_buff_head *list_) 1479 { 1480 struct sk_buff *next = skb->next; 1481 1482 if (next == (struct sk_buff *)list_) 1483 next = NULL; 1484 return next; 1485 } 1486 1487 /** 1488 * skb_peek_tail - peek at the tail of an &sk_buff_head 1489 * @list_: list to peek at 1490 * 1491 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1492 * be careful with this one. A peek leaves the buffer on the 1493 * list and someone else may run off with it. You must hold 1494 * the appropriate locks or have a private queue to do this. 1495 * 1496 * Returns %NULL for an empty list or a pointer to the tail element. 1497 * The reference count is not incremented and the reference is therefore 1498 * volatile. Use with caution. 1499 */ 1500 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1501 { 1502 struct sk_buff *skb = list_->prev; 1503 1504 if (skb == (struct sk_buff *)list_) 1505 skb = NULL; 1506 return skb; 1507 1508 } 1509 1510 /** 1511 * skb_queue_len - get queue length 1512 * @list_: list to measure 1513 * 1514 * Return the length of an &sk_buff queue. 1515 */ 1516 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1517 { 1518 return list_->qlen; 1519 } 1520 1521 /** 1522 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1523 * @list: queue to initialize 1524 * 1525 * This initializes only the list and queue length aspects of 1526 * an sk_buff_head object. This allows to initialize the list 1527 * aspects of an sk_buff_head without reinitializing things like 1528 * the spinlock. It can also be used for on-stack sk_buff_head 1529 * objects where the spinlock is known to not be used. 1530 */ 1531 static inline void __skb_queue_head_init(struct sk_buff_head *list) 1532 { 1533 list->prev = list->next = (struct sk_buff *)list; 1534 list->qlen = 0; 1535 } 1536 1537 /* 1538 * This function creates a split out lock class for each invocation; 1539 * this is needed for now since a whole lot of users of the skb-queue 1540 * infrastructure in drivers have different locking usage (in hardirq) 1541 * than the networking core (in softirq only). In the long run either the 1542 * network layer or drivers should need annotation to consolidate the 1543 * main types of usage into 3 classes. 1544 */ 1545 static inline void skb_queue_head_init(struct sk_buff_head *list) 1546 { 1547 spin_lock_init(&list->lock); 1548 __skb_queue_head_init(list); 1549 } 1550 1551 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1552 struct lock_class_key *class) 1553 { 1554 skb_queue_head_init(list); 1555 lockdep_set_class(&list->lock, class); 1556 } 1557 1558 /* 1559 * Insert an sk_buff on a list. 1560 * 1561 * The "__skb_xxxx()" functions are the non-atomic ones that 1562 * can only be called with interrupts disabled. 1563 */ 1564 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1565 struct sk_buff_head *list); 1566 static inline void __skb_insert(struct sk_buff *newsk, 1567 struct sk_buff *prev, struct sk_buff *next, 1568 struct sk_buff_head *list) 1569 { 1570 newsk->next = next; 1571 newsk->prev = prev; 1572 next->prev = prev->next = newsk; 1573 list->qlen++; 1574 } 1575 1576 static inline void __skb_queue_splice(const struct sk_buff_head *list, 1577 struct sk_buff *prev, 1578 struct sk_buff *next) 1579 { 1580 struct sk_buff *first = list->next; 1581 struct sk_buff *last = list->prev; 1582 1583 first->prev = prev; 1584 prev->next = first; 1585 1586 last->next = next; 1587 next->prev = last; 1588 } 1589 1590 /** 1591 * skb_queue_splice - join two skb lists, this is designed for stacks 1592 * @list: the new list to add 1593 * @head: the place to add it in the first list 1594 */ 1595 static inline void skb_queue_splice(const struct sk_buff_head *list, 1596 struct sk_buff_head *head) 1597 { 1598 if (!skb_queue_empty(list)) { 1599 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1600 head->qlen += list->qlen; 1601 } 1602 } 1603 1604 /** 1605 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1606 * @list: the new list to add 1607 * @head: the place to add it in the first list 1608 * 1609 * The list at @list is reinitialised 1610 */ 1611 static inline void skb_queue_splice_init(struct sk_buff_head *list, 1612 struct sk_buff_head *head) 1613 { 1614 if (!skb_queue_empty(list)) { 1615 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1616 head->qlen += list->qlen; 1617 __skb_queue_head_init(list); 1618 } 1619 } 1620 1621 /** 1622 * skb_queue_splice_tail - join two skb lists, each list being a queue 1623 * @list: the new list to add 1624 * @head: the place to add it in the first list 1625 */ 1626 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1627 struct sk_buff_head *head) 1628 { 1629 if (!skb_queue_empty(list)) { 1630 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1631 head->qlen += list->qlen; 1632 } 1633 } 1634 1635 /** 1636 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1637 * @list: the new list to add 1638 * @head: the place to add it in the first list 1639 * 1640 * Each of the lists is a queue. 1641 * The list at @list is reinitialised 1642 */ 1643 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1644 struct sk_buff_head *head) 1645 { 1646 if (!skb_queue_empty(list)) { 1647 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1648 head->qlen += list->qlen; 1649 __skb_queue_head_init(list); 1650 } 1651 } 1652 1653 /** 1654 * __skb_queue_after - queue a buffer at the list head 1655 * @list: list to use 1656 * @prev: place after this buffer 1657 * @newsk: buffer to queue 1658 * 1659 * Queue a buffer int the middle of a list. This function takes no locks 1660 * and you must therefore hold required locks before calling it. 1661 * 1662 * A buffer cannot be placed on two lists at the same time. 1663 */ 1664 static inline void __skb_queue_after(struct sk_buff_head *list, 1665 struct sk_buff *prev, 1666 struct sk_buff *newsk) 1667 { 1668 __skb_insert(newsk, prev, prev->next, list); 1669 } 1670 1671 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1672 struct sk_buff_head *list); 1673 1674 static inline void __skb_queue_before(struct sk_buff_head *list, 1675 struct sk_buff *next, 1676 struct sk_buff *newsk) 1677 { 1678 __skb_insert(newsk, next->prev, next, list); 1679 } 1680 1681 /** 1682 * __skb_queue_head - queue a buffer at the list head 1683 * @list: list to use 1684 * @newsk: buffer to queue 1685 * 1686 * Queue a buffer at the start of a list. This function takes no locks 1687 * and you must therefore hold required locks before calling it. 1688 * 1689 * A buffer cannot be placed on two lists at the same time. 1690 */ 1691 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1692 static inline void __skb_queue_head(struct sk_buff_head *list, 1693 struct sk_buff *newsk) 1694 { 1695 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1696 } 1697 1698 /** 1699 * __skb_queue_tail - queue a buffer at the list tail 1700 * @list: list to use 1701 * @newsk: buffer to queue 1702 * 1703 * Queue a buffer at the end of a list. This function takes no locks 1704 * and you must therefore hold required locks before calling it. 1705 * 1706 * A buffer cannot be placed on two lists at the same time. 1707 */ 1708 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1709 static inline void __skb_queue_tail(struct sk_buff_head *list, 1710 struct sk_buff *newsk) 1711 { 1712 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1713 } 1714 1715 /* 1716 * remove sk_buff from list. _Must_ be called atomically, and with 1717 * the list known.. 1718 */ 1719 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1720 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1721 { 1722 struct sk_buff *next, *prev; 1723 1724 list->qlen--; 1725 next = skb->next; 1726 prev = skb->prev; 1727 skb->next = skb->prev = NULL; 1728 next->prev = prev; 1729 prev->next = next; 1730 } 1731 1732 /** 1733 * __skb_dequeue - remove from the head of the queue 1734 * @list: list to dequeue from 1735 * 1736 * Remove the head of the list. This function does not take any locks 1737 * so must be used with appropriate locks held only. The head item is 1738 * returned or %NULL if the list is empty. 1739 */ 1740 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1741 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1742 { 1743 struct sk_buff *skb = skb_peek(list); 1744 if (skb) 1745 __skb_unlink(skb, list); 1746 return skb; 1747 } 1748 1749 /** 1750 * __skb_dequeue_tail - remove from the tail of the queue 1751 * @list: list to dequeue from 1752 * 1753 * Remove the tail of the list. This function does not take any locks 1754 * so must be used with appropriate locks held only. The tail item is 1755 * returned or %NULL if the list is empty. 1756 */ 1757 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1758 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1759 { 1760 struct sk_buff *skb = skb_peek_tail(list); 1761 if (skb) 1762 __skb_unlink(skb, list); 1763 return skb; 1764 } 1765 1766 1767 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1768 { 1769 return skb->data_len; 1770 } 1771 1772 static inline unsigned int skb_headlen(const struct sk_buff *skb) 1773 { 1774 return skb->len - skb->data_len; 1775 } 1776 1777 static inline int skb_pagelen(const struct sk_buff *skb) 1778 { 1779 int i, len = 0; 1780 1781 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--) 1782 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1783 return len + skb_headlen(skb); 1784 } 1785 1786 /** 1787 * __skb_fill_page_desc - initialise a paged fragment in an skb 1788 * @skb: buffer containing fragment to be initialised 1789 * @i: paged fragment index to initialise 1790 * @page: the page to use for this fragment 1791 * @off: the offset to the data with @page 1792 * @size: the length of the data 1793 * 1794 * Initialises the @i'th fragment of @skb to point to &size bytes at 1795 * offset @off within @page. 1796 * 1797 * Does not take any additional reference on the fragment. 1798 */ 1799 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1800 struct page *page, int off, int size) 1801 { 1802 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1803 1804 /* 1805 * Propagate page pfmemalloc to the skb if we can. The problem is 1806 * that not all callers have unique ownership of the page but rely 1807 * on page_is_pfmemalloc doing the right thing(tm). 1808 */ 1809 frag->page.p = page; 1810 frag->page_offset = off; 1811 skb_frag_size_set(frag, size); 1812 1813 page = compound_head(page); 1814 if (page_is_pfmemalloc(page)) 1815 skb->pfmemalloc = true; 1816 } 1817 1818 /** 1819 * skb_fill_page_desc - initialise a paged fragment in an skb 1820 * @skb: buffer containing fragment to be initialised 1821 * @i: paged fragment index to initialise 1822 * @page: the page to use for this fragment 1823 * @off: the offset to the data with @page 1824 * @size: the length of the data 1825 * 1826 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1827 * @skb to point to @size bytes at offset @off within @page. In 1828 * addition updates @skb such that @i is the last fragment. 1829 * 1830 * Does not take any additional reference on the fragment. 1831 */ 1832 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1833 struct page *page, int off, int size) 1834 { 1835 __skb_fill_page_desc(skb, i, page, off, size); 1836 skb_shinfo(skb)->nr_frags = i + 1; 1837 } 1838 1839 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1840 int size, unsigned int truesize); 1841 1842 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1843 unsigned int truesize); 1844 1845 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1846 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1847 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1848 1849 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1850 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1851 { 1852 return skb->head + skb->tail; 1853 } 1854 1855 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1856 { 1857 skb->tail = skb->data - skb->head; 1858 } 1859 1860 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1861 { 1862 skb_reset_tail_pointer(skb); 1863 skb->tail += offset; 1864 } 1865 1866 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 1867 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1868 { 1869 return skb->tail; 1870 } 1871 1872 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1873 { 1874 skb->tail = skb->data; 1875 } 1876 1877 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1878 { 1879 skb->tail = skb->data + offset; 1880 } 1881 1882 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1883 1884 /* 1885 * Add data to an sk_buff 1886 */ 1887 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1888 unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1889 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1890 { 1891 unsigned char *tmp = skb_tail_pointer(skb); 1892 SKB_LINEAR_ASSERT(skb); 1893 skb->tail += len; 1894 skb->len += len; 1895 return tmp; 1896 } 1897 1898 unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1899 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1900 { 1901 skb->data -= len; 1902 skb->len += len; 1903 return skb->data; 1904 } 1905 1906 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1907 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1908 { 1909 skb->len -= len; 1910 BUG_ON(skb->len < skb->data_len); 1911 return skb->data += len; 1912 } 1913 1914 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1915 { 1916 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1917 } 1918 1919 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1920 1921 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1922 { 1923 if (len > skb_headlen(skb) && 1924 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1925 return NULL; 1926 skb->len -= len; 1927 return skb->data += len; 1928 } 1929 1930 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1931 { 1932 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1933 } 1934 1935 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1936 { 1937 if (likely(len <= skb_headlen(skb))) 1938 return 1; 1939 if (unlikely(len > skb->len)) 1940 return 0; 1941 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1942 } 1943 1944 /** 1945 * skb_headroom - bytes at buffer head 1946 * @skb: buffer to check 1947 * 1948 * Return the number of bytes of free space at the head of an &sk_buff. 1949 */ 1950 static inline unsigned int skb_headroom(const struct sk_buff *skb) 1951 { 1952 return skb->data - skb->head; 1953 } 1954 1955 /** 1956 * skb_tailroom - bytes at buffer end 1957 * @skb: buffer to check 1958 * 1959 * Return the number of bytes of free space at the tail of an sk_buff 1960 */ 1961 static inline int skb_tailroom(const struct sk_buff *skb) 1962 { 1963 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 1964 } 1965 1966 /** 1967 * skb_availroom - bytes at buffer end 1968 * @skb: buffer to check 1969 * 1970 * Return the number of bytes of free space at the tail of an sk_buff 1971 * allocated by sk_stream_alloc() 1972 */ 1973 static inline int skb_availroom(const struct sk_buff *skb) 1974 { 1975 if (skb_is_nonlinear(skb)) 1976 return 0; 1977 1978 return skb->end - skb->tail - skb->reserved_tailroom; 1979 } 1980 1981 /** 1982 * skb_reserve - adjust headroom 1983 * @skb: buffer to alter 1984 * @len: bytes to move 1985 * 1986 * Increase the headroom of an empty &sk_buff by reducing the tail 1987 * room. This is only allowed for an empty buffer. 1988 */ 1989 static inline void skb_reserve(struct sk_buff *skb, int len) 1990 { 1991 skb->data += len; 1992 skb->tail += len; 1993 } 1994 1995 /** 1996 * skb_tailroom_reserve - adjust reserved_tailroom 1997 * @skb: buffer to alter 1998 * @mtu: maximum amount of headlen permitted 1999 * @needed_tailroom: minimum amount of reserved_tailroom 2000 * 2001 * Set reserved_tailroom so that headlen can be as large as possible but 2002 * not larger than mtu and tailroom cannot be smaller than 2003 * needed_tailroom. 2004 * The required headroom should already have been reserved before using 2005 * this function. 2006 */ 2007 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2008 unsigned int needed_tailroom) 2009 { 2010 SKB_LINEAR_ASSERT(skb); 2011 if (mtu < skb_tailroom(skb) - needed_tailroom) 2012 /* use at most mtu */ 2013 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2014 else 2015 /* use up to all available space */ 2016 skb->reserved_tailroom = needed_tailroom; 2017 } 2018 2019 #define ENCAP_TYPE_ETHER 0 2020 #define ENCAP_TYPE_IPPROTO 1 2021 2022 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2023 __be16 protocol) 2024 { 2025 skb->inner_protocol = protocol; 2026 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2027 } 2028 2029 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2030 __u8 ipproto) 2031 { 2032 skb->inner_ipproto = ipproto; 2033 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2034 } 2035 2036 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2037 { 2038 skb->inner_mac_header = skb->mac_header; 2039 skb->inner_network_header = skb->network_header; 2040 skb->inner_transport_header = skb->transport_header; 2041 } 2042 2043 static inline void skb_reset_mac_len(struct sk_buff *skb) 2044 { 2045 skb->mac_len = skb->network_header - skb->mac_header; 2046 } 2047 2048 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2049 *skb) 2050 { 2051 return skb->head + skb->inner_transport_header; 2052 } 2053 2054 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2055 { 2056 return skb_inner_transport_header(skb) - skb->data; 2057 } 2058 2059 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2060 { 2061 skb->inner_transport_header = skb->data - skb->head; 2062 } 2063 2064 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2065 const int offset) 2066 { 2067 skb_reset_inner_transport_header(skb); 2068 skb->inner_transport_header += offset; 2069 } 2070 2071 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2072 { 2073 return skb->head + skb->inner_network_header; 2074 } 2075 2076 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2077 { 2078 skb->inner_network_header = skb->data - skb->head; 2079 } 2080 2081 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2082 const int offset) 2083 { 2084 skb_reset_inner_network_header(skb); 2085 skb->inner_network_header += offset; 2086 } 2087 2088 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2089 { 2090 return skb->head + skb->inner_mac_header; 2091 } 2092 2093 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2094 { 2095 skb->inner_mac_header = skb->data - skb->head; 2096 } 2097 2098 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2099 const int offset) 2100 { 2101 skb_reset_inner_mac_header(skb); 2102 skb->inner_mac_header += offset; 2103 } 2104 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2105 { 2106 return skb->transport_header != (typeof(skb->transport_header))~0U; 2107 } 2108 2109 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2110 { 2111 return skb->head + skb->transport_header; 2112 } 2113 2114 static inline void skb_reset_transport_header(struct sk_buff *skb) 2115 { 2116 skb->transport_header = skb->data - skb->head; 2117 } 2118 2119 static inline void skb_set_transport_header(struct sk_buff *skb, 2120 const int offset) 2121 { 2122 skb_reset_transport_header(skb); 2123 skb->transport_header += offset; 2124 } 2125 2126 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2127 { 2128 return skb->head + skb->network_header; 2129 } 2130 2131 static inline void skb_reset_network_header(struct sk_buff *skb) 2132 { 2133 skb->network_header = skb->data - skb->head; 2134 } 2135 2136 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2137 { 2138 skb_reset_network_header(skb); 2139 skb->network_header += offset; 2140 } 2141 2142 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2143 { 2144 return skb->head + skb->mac_header; 2145 } 2146 2147 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2148 { 2149 return skb->mac_header != (typeof(skb->mac_header))~0U; 2150 } 2151 2152 static inline void skb_reset_mac_header(struct sk_buff *skb) 2153 { 2154 skb->mac_header = skb->data - skb->head; 2155 } 2156 2157 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2158 { 2159 skb_reset_mac_header(skb); 2160 skb->mac_header += offset; 2161 } 2162 2163 static inline void skb_pop_mac_header(struct sk_buff *skb) 2164 { 2165 skb->mac_header = skb->network_header; 2166 } 2167 2168 static inline void skb_probe_transport_header(struct sk_buff *skb, 2169 const int offset_hint) 2170 { 2171 struct flow_keys keys; 2172 2173 if (skb_transport_header_was_set(skb)) 2174 return; 2175 else if (skb_flow_dissect_flow_keys(skb, &keys, 0)) 2176 skb_set_transport_header(skb, keys.control.thoff); 2177 else 2178 skb_set_transport_header(skb, offset_hint); 2179 } 2180 2181 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2182 { 2183 if (skb_mac_header_was_set(skb)) { 2184 const unsigned char *old_mac = skb_mac_header(skb); 2185 2186 skb_set_mac_header(skb, -skb->mac_len); 2187 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2188 } 2189 } 2190 2191 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2192 { 2193 return skb->csum_start - skb_headroom(skb); 2194 } 2195 2196 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2197 { 2198 return skb->head + skb->csum_start; 2199 } 2200 2201 static inline int skb_transport_offset(const struct sk_buff *skb) 2202 { 2203 return skb_transport_header(skb) - skb->data; 2204 } 2205 2206 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2207 { 2208 return skb->transport_header - skb->network_header; 2209 } 2210 2211 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2212 { 2213 return skb->inner_transport_header - skb->inner_network_header; 2214 } 2215 2216 static inline int skb_network_offset(const struct sk_buff *skb) 2217 { 2218 return skb_network_header(skb) - skb->data; 2219 } 2220 2221 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2222 { 2223 return skb_inner_network_header(skb) - skb->data; 2224 } 2225 2226 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2227 { 2228 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2229 } 2230 2231 /* 2232 * CPUs often take a performance hit when accessing unaligned memory 2233 * locations. The actual performance hit varies, it can be small if the 2234 * hardware handles it or large if we have to take an exception and fix it 2235 * in software. 2236 * 2237 * Since an ethernet header is 14 bytes network drivers often end up with 2238 * the IP header at an unaligned offset. The IP header can be aligned by 2239 * shifting the start of the packet by 2 bytes. Drivers should do this 2240 * with: 2241 * 2242 * skb_reserve(skb, NET_IP_ALIGN); 2243 * 2244 * The downside to this alignment of the IP header is that the DMA is now 2245 * unaligned. On some architectures the cost of an unaligned DMA is high 2246 * and this cost outweighs the gains made by aligning the IP header. 2247 * 2248 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2249 * to be overridden. 2250 */ 2251 #ifndef NET_IP_ALIGN 2252 #define NET_IP_ALIGN 2 2253 #endif 2254 2255 /* 2256 * The networking layer reserves some headroom in skb data (via 2257 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2258 * the header has to grow. In the default case, if the header has to grow 2259 * 32 bytes or less we avoid the reallocation. 2260 * 2261 * Unfortunately this headroom changes the DMA alignment of the resulting 2262 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2263 * on some architectures. An architecture can override this value, 2264 * perhaps setting it to a cacheline in size (since that will maintain 2265 * cacheline alignment of the DMA). It must be a power of 2. 2266 * 2267 * Various parts of the networking layer expect at least 32 bytes of 2268 * headroom, you should not reduce this. 2269 * 2270 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2271 * to reduce average number of cache lines per packet. 2272 * get_rps_cpus() for example only access one 64 bytes aligned block : 2273 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2274 */ 2275 #ifndef NET_SKB_PAD 2276 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2277 #endif 2278 2279 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2280 2281 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2282 { 2283 if (unlikely(skb_is_nonlinear(skb))) { 2284 WARN_ON(1); 2285 return; 2286 } 2287 skb->len = len; 2288 skb_set_tail_pointer(skb, len); 2289 } 2290 2291 void skb_trim(struct sk_buff *skb, unsigned int len); 2292 2293 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2294 { 2295 if (skb->data_len) 2296 return ___pskb_trim(skb, len); 2297 __skb_trim(skb, len); 2298 return 0; 2299 } 2300 2301 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2302 { 2303 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2304 } 2305 2306 /** 2307 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2308 * @skb: buffer to alter 2309 * @len: new length 2310 * 2311 * This is identical to pskb_trim except that the caller knows that 2312 * the skb is not cloned so we should never get an error due to out- 2313 * of-memory. 2314 */ 2315 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2316 { 2317 int err = pskb_trim(skb, len); 2318 BUG_ON(err); 2319 } 2320 2321 /** 2322 * skb_orphan - orphan a buffer 2323 * @skb: buffer to orphan 2324 * 2325 * If a buffer currently has an owner then we call the owner's 2326 * destructor function and make the @skb unowned. The buffer continues 2327 * to exist but is no longer charged to its former owner. 2328 */ 2329 static inline void skb_orphan(struct sk_buff *skb) 2330 { 2331 if (skb->destructor) { 2332 skb->destructor(skb); 2333 skb->destructor = NULL; 2334 skb->sk = NULL; 2335 } else { 2336 BUG_ON(skb->sk); 2337 } 2338 } 2339 2340 /** 2341 * skb_orphan_frags - orphan the frags contained in a buffer 2342 * @skb: buffer to orphan frags from 2343 * @gfp_mask: allocation mask for replacement pages 2344 * 2345 * For each frag in the SKB which needs a destructor (i.e. has an 2346 * owner) create a copy of that frag and release the original 2347 * page by calling the destructor. 2348 */ 2349 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2350 { 2351 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 2352 return 0; 2353 return skb_copy_ubufs(skb, gfp_mask); 2354 } 2355 2356 /** 2357 * __skb_queue_purge - empty a list 2358 * @list: list to empty 2359 * 2360 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2361 * the list and one reference dropped. This function does not take the 2362 * list lock and the caller must hold the relevant locks to use it. 2363 */ 2364 void skb_queue_purge(struct sk_buff_head *list); 2365 static inline void __skb_queue_purge(struct sk_buff_head *list) 2366 { 2367 struct sk_buff *skb; 2368 while ((skb = __skb_dequeue(list)) != NULL) 2369 kfree_skb(skb); 2370 } 2371 2372 void *netdev_alloc_frag(unsigned int fragsz); 2373 2374 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2375 gfp_t gfp_mask); 2376 2377 /** 2378 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2379 * @dev: network device to receive on 2380 * @length: length to allocate 2381 * 2382 * Allocate a new &sk_buff and assign it a usage count of one. The 2383 * buffer has unspecified headroom built in. Users should allocate 2384 * the headroom they think they need without accounting for the 2385 * built in space. The built in space is used for optimisations. 2386 * 2387 * %NULL is returned if there is no free memory. Although this function 2388 * allocates memory it can be called from an interrupt. 2389 */ 2390 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2391 unsigned int length) 2392 { 2393 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2394 } 2395 2396 /* legacy helper around __netdev_alloc_skb() */ 2397 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2398 gfp_t gfp_mask) 2399 { 2400 return __netdev_alloc_skb(NULL, length, gfp_mask); 2401 } 2402 2403 /* legacy helper around netdev_alloc_skb() */ 2404 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2405 { 2406 return netdev_alloc_skb(NULL, length); 2407 } 2408 2409 2410 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2411 unsigned int length, gfp_t gfp) 2412 { 2413 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2414 2415 if (NET_IP_ALIGN && skb) 2416 skb_reserve(skb, NET_IP_ALIGN); 2417 return skb; 2418 } 2419 2420 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2421 unsigned int length) 2422 { 2423 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2424 } 2425 2426 static inline void skb_free_frag(void *addr) 2427 { 2428 __free_page_frag(addr); 2429 } 2430 2431 void *napi_alloc_frag(unsigned int fragsz); 2432 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2433 unsigned int length, gfp_t gfp_mask); 2434 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2435 unsigned int length) 2436 { 2437 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2438 } 2439 void napi_consume_skb(struct sk_buff *skb, int budget); 2440 2441 void __kfree_skb_flush(void); 2442 void __kfree_skb_defer(struct sk_buff *skb); 2443 2444 /** 2445 * __dev_alloc_pages - allocate page for network Rx 2446 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2447 * @order: size of the allocation 2448 * 2449 * Allocate a new page. 2450 * 2451 * %NULL is returned if there is no free memory. 2452 */ 2453 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2454 unsigned int order) 2455 { 2456 /* This piece of code contains several assumptions. 2457 * 1. This is for device Rx, therefor a cold page is preferred. 2458 * 2. The expectation is the user wants a compound page. 2459 * 3. If requesting a order 0 page it will not be compound 2460 * due to the check to see if order has a value in prep_new_page 2461 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2462 * code in gfp_to_alloc_flags that should be enforcing this. 2463 */ 2464 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2465 2466 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2467 } 2468 2469 static inline struct page *dev_alloc_pages(unsigned int order) 2470 { 2471 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 2472 } 2473 2474 /** 2475 * __dev_alloc_page - allocate a page for network Rx 2476 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2477 * 2478 * Allocate a new page. 2479 * 2480 * %NULL is returned if there is no free memory. 2481 */ 2482 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2483 { 2484 return __dev_alloc_pages(gfp_mask, 0); 2485 } 2486 2487 static inline struct page *dev_alloc_page(void) 2488 { 2489 return dev_alloc_pages(0); 2490 } 2491 2492 /** 2493 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2494 * @page: The page that was allocated from skb_alloc_page 2495 * @skb: The skb that may need pfmemalloc set 2496 */ 2497 static inline void skb_propagate_pfmemalloc(struct page *page, 2498 struct sk_buff *skb) 2499 { 2500 if (page_is_pfmemalloc(page)) 2501 skb->pfmemalloc = true; 2502 } 2503 2504 /** 2505 * skb_frag_page - retrieve the page referred to by a paged fragment 2506 * @frag: the paged fragment 2507 * 2508 * Returns the &struct page associated with @frag. 2509 */ 2510 static inline struct page *skb_frag_page(const skb_frag_t *frag) 2511 { 2512 return frag->page.p; 2513 } 2514 2515 /** 2516 * __skb_frag_ref - take an addition reference on a paged fragment. 2517 * @frag: the paged fragment 2518 * 2519 * Takes an additional reference on the paged fragment @frag. 2520 */ 2521 static inline void __skb_frag_ref(skb_frag_t *frag) 2522 { 2523 get_page(skb_frag_page(frag)); 2524 } 2525 2526 /** 2527 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2528 * @skb: the buffer 2529 * @f: the fragment offset. 2530 * 2531 * Takes an additional reference on the @f'th paged fragment of @skb. 2532 */ 2533 static inline void skb_frag_ref(struct sk_buff *skb, int f) 2534 { 2535 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2536 } 2537 2538 /** 2539 * __skb_frag_unref - release a reference on a paged fragment. 2540 * @frag: the paged fragment 2541 * 2542 * Releases a reference on the paged fragment @frag. 2543 */ 2544 static inline void __skb_frag_unref(skb_frag_t *frag) 2545 { 2546 put_page(skb_frag_page(frag)); 2547 } 2548 2549 /** 2550 * skb_frag_unref - release a reference on a paged fragment of an skb. 2551 * @skb: the buffer 2552 * @f: the fragment offset 2553 * 2554 * Releases a reference on the @f'th paged fragment of @skb. 2555 */ 2556 static inline void skb_frag_unref(struct sk_buff *skb, int f) 2557 { 2558 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2559 } 2560 2561 /** 2562 * skb_frag_address - gets the address of the data contained in a paged fragment 2563 * @frag: the paged fragment buffer 2564 * 2565 * Returns the address of the data within @frag. The page must already 2566 * be mapped. 2567 */ 2568 static inline void *skb_frag_address(const skb_frag_t *frag) 2569 { 2570 return page_address(skb_frag_page(frag)) + frag->page_offset; 2571 } 2572 2573 /** 2574 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2575 * @frag: the paged fragment buffer 2576 * 2577 * Returns the address of the data within @frag. Checks that the page 2578 * is mapped and returns %NULL otherwise. 2579 */ 2580 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2581 { 2582 void *ptr = page_address(skb_frag_page(frag)); 2583 if (unlikely(!ptr)) 2584 return NULL; 2585 2586 return ptr + frag->page_offset; 2587 } 2588 2589 /** 2590 * __skb_frag_set_page - sets the page contained in a paged fragment 2591 * @frag: the paged fragment 2592 * @page: the page to set 2593 * 2594 * Sets the fragment @frag to contain @page. 2595 */ 2596 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2597 { 2598 frag->page.p = page; 2599 } 2600 2601 /** 2602 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2603 * @skb: the buffer 2604 * @f: the fragment offset 2605 * @page: the page to set 2606 * 2607 * Sets the @f'th fragment of @skb to contain @page. 2608 */ 2609 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2610 struct page *page) 2611 { 2612 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2613 } 2614 2615 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2616 2617 /** 2618 * skb_frag_dma_map - maps a paged fragment via the DMA API 2619 * @dev: the device to map the fragment to 2620 * @frag: the paged fragment to map 2621 * @offset: the offset within the fragment (starting at the 2622 * fragment's own offset) 2623 * @size: the number of bytes to map 2624 * @dir: the direction of the mapping (%PCI_DMA_*) 2625 * 2626 * Maps the page associated with @frag to @device. 2627 */ 2628 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2629 const skb_frag_t *frag, 2630 size_t offset, size_t size, 2631 enum dma_data_direction dir) 2632 { 2633 return dma_map_page(dev, skb_frag_page(frag), 2634 frag->page_offset + offset, size, dir); 2635 } 2636 2637 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2638 gfp_t gfp_mask) 2639 { 2640 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2641 } 2642 2643 2644 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2645 gfp_t gfp_mask) 2646 { 2647 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2648 } 2649 2650 2651 /** 2652 * skb_clone_writable - is the header of a clone writable 2653 * @skb: buffer to check 2654 * @len: length up to which to write 2655 * 2656 * Returns true if modifying the header part of the cloned buffer 2657 * does not requires the data to be copied. 2658 */ 2659 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2660 { 2661 return !skb_header_cloned(skb) && 2662 skb_headroom(skb) + len <= skb->hdr_len; 2663 } 2664 2665 static inline int skb_try_make_writable(struct sk_buff *skb, 2666 unsigned int write_len) 2667 { 2668 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 2669 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2670 } 2671 2672 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2673 int cloned) 2674 { 2675 int delta = 0; 2676 2677 if (headroom > skb_headroom(skb)) 2678 delta = headroom - skb_headroom(skb); 2679 2680 if (delta || cloned) 2681 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2682 GFP_ATOMIC); 2683 return 0; 2684 } 2685 2686 /** 2687 * skb_cow - copy header of skb when it is required 2688 * @skb: buffer to cow 2689 * @headroom: needed headroom 2690 * 2691 * If the skb passed lacks sufficient headroom or its data part 2692 * is shared, data is reallocated. If reallocation fails, an error 2693 * is returned and original skb is not changed. 2694 * 2695 * The result is skb with writable area skb->head...skb->tail 2696 * and at least @headroom of space at head. 2697 */ 2698 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2699 { 2700 return __skb_cow(skb, headroom, skb_cloned(skb)); 2701 } 2702 2703 /** 2704 * skb_cow_head - skb_cow but only making the head writable 2705 * @skb: buffer to cow 2706 * @headroom: needed headroom 2707 * 2708 * This function is identical to skb_cow except that we replace the 2709 * skb_cloned check by skb_header_cloned. It should be used when 2710 * you only need to push on some header and do not need to modify 2711 * the data. 2712 */ 2713 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2714 { 2715 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2716 } 2717 2718 /** 2719 * skb_padto - pad an skbuff up to a minimal size 2720 * @skb: buffer to pad 2721 * @len: minimal length 2722 * 2723 * Pads up a buffer to ensure the trailing bytes exist and are 2724 * blanked. If the buffer already contains sufficient data it 2725 * is untouched. Otherwise it is extended. Returns zero on 2726 * success. The skb is freed on error. 2727 */ 2728 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2729 { 2730 unsigned int size = skb->len; 2731 if (likely(size >= len)) 2732 return 0; 2733 return skb_pad(skb, len - size); 2734 } 2735 2736 /** 2737 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2738 * @skb: buffer to pad 2739 * @len: minimal length 2740 * 2741 * Pads up a buffer to ensure the trailing bytes exist and are 2742 * blanked. If the buffer already contains sufficient data it 2743 * is untouched. Otherwise it is extended. Returns zero on 2744 * success. The skb is freed on error. 2745 */ 2746 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2747 { 2748 unsigned int size = skb->len; 2749 2750 if (unlikely(size < len)) { 2751 len -= size; 2752 if (skb_pad(skb, len)) 2753 return -ENOMEM; 2754 __skb_put(skb, len); 2755 } 2756 return 0; 2757 } 2758 2759 static inline int skb_add_data(struct sk_buff *skb, 2760 struct iov_iter *from, int copy) 2761 { 2762 const int off = skb->len; 2763 2764 if (skb->ip_summed == CHECKSUM_NONE) { 2765 __wsum csum = 0; 2766 if (csum_and_copy_from_iter(skb_put(skb, copy), copy, 2767 &csum, from) == copy) { 2768 skb->csum = csum_block_add(skb->csum, csum, off); 2769 return 0; 2770 } 2771 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy) 2772 return 0; 2773 2774 __skb_trim(skb, off); 2775 return -EFAULT; 2776 } 2777 2778 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2779 const struct page *page, int off) 2780 { 2781 if (i) { 2782 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2783 2784 return page == skb_frag_page(frag) && 2785 off == frag->page_offset + skb_frag_size(frag); 2786 } 2787 return false; 2788 } 2789 2790 static inline int __skb_linearize(struct sk_buff *skb) 2791 { 2792 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2793 } 2794 2795 /** 2796 * skb_linearize - convert paged skb to linear one 2797 * @skb: buffer to linarize 2798 * 2799 * If there is no free memory -ENOMEM is returned, otherwise zero 2800 * is returned and the old skb data released. 2801 */ 2802 static inline int skb_linearize(struct sk_buff *skb) 2803 { 2804 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2805 } 2806 2807 /** 2808 * skb_has_shared_frag - can any frag be overwritten 2809 * @skb: buffer to test 2810 * 2811 * Return true if the skb has at least one frag that might be modified 2812 * by an external entity (as in vmsplice()/sendfile()) 2813 */ 2814 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 2815 { 2816 return skb_is_nonlinear(skb) && 2817 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2818 } 2819 2820 /** 2821 * skb_linearize_cow - make sure skb is linear and writable 2822 * @skb: buffer to process 2823 * 2824 * If there is no free memory -ENOMEM is returned, otherwise zero 2825 * is returned and the old skb data released. 2826 */ 2827 static inline int skb_linearize_cow(struct sk_buff *skb) 2828 { 2829 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2830 __skb_linearize(skb) : 0; 2831 } 2832 2833 /** 2834 * skb_postpull_rcsum - update checksum for received skb after pull 2835 * @skb: buffer to update 2836 * @start: start of data before pull 2837 * @len: length of data pulled 2838 * 2839 * After doing a pull on a received packet, you need to call this to 2840 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2841 * CHECKSUM_NONE so that it can be recomputed from scratch. 2842 */ 2843 2844 static inline void skb_postpull_rcsum(struct sk_buff *skb, 2845 const void *start, unsigned int len) 2846 { 2847 if (skb->ip_summed == CHECKSUM_COMPLETE) 2848 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0)); 2849 else if (skb->ip_summed == CHECKSUM_PARTIAL && 2850 skb_checksum_start_offset(skb) < 0) 2851 skb->ip_summed = CHECKSUM_NONE; 2852 } 2853 2854 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2855 2856 static inline void skb_postpush_rcsum(struct sk_buff *skb, 2857 const void *start, unsigned int len) 2858 { 2859 /* For performing the reverse operation to skb_postpull_rcsum(), 2860 * we can instead of ... 2861 * 2862 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0)); 2863 * 2864 * ... just use this equivalent version here to save a few 2865 * instructions. Feeding csum of 0 in csum_partial() and later 2866 * on adding skb->csum is equivalent to feed skb->csum in the 2867 * first place. 2868 */ 2869 if (skb->ip_summed == CHECKSUM_COMPLETE) 2870 skb->csum = csum_partial(start, len, skb->csum); 2871 } 2872 2873 /** 2874 * skb_push_rcsum - push skb and update receive checksum 2875 * @skb: buffer to update 2876 * @len: length of data pulled 2877 * 2878 * This function performs an skb_push on the packet and updates 2879 * the CHECKSUM_COMPLETE checksum. It should be used on 2880 * receive path processing instead of skb_push unless you know 2881 * that the checksum difference is zero (e.g., a valid IP header) 2882 * or you are setting ip_summed to CHECKSUM_NONE. 2883 */ 2884 static inline unsigned char *skb_push_rcsum(struct sk_buff *skb, 2885 unsigned int len) 2886 { 2887 skb_push(skb, len); 2888 skb_postpush_rcsum(skb, skb->data, len); 2889 return skb->data; 2890 } 2891 2892 /** 2893 * pskb_trim_rcsum - trim received skb and update checksum 2894 * @skb: buffer to trim 2895 * @len: new length 2896 * 2897 * This is exactly the same as pskb_trim except that it ensures the 2898 * checksum of received packets are still valid after the operation. 2899 */ 2900 2901 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2902 { 2903 if (likely(len >= skb->len)) 2904 return 0; 2905 if (skb->ip_summed == CHECKSUM_COMPLETE) 2906 skb->ip_summed = CHECKSUM_NONE; 2907 return __pskb_trim(skb, len); 2908 } 2909 2910 #define skb_queue_walk(queue, skb) \ 2911 for (skb = (queue)->next; \ 2912 skb != (struct sk_buff *)(queue); \ 2913 skb = skb->next) 2914 2915 #define skb_queue_walk_safe(queue, skb, tmp) \ 2916 for (skb = (queue)->next, tmp = skb->next; \ 2917 skb != (struct sk_buff *)(queue); \ 2918 skb = tmp, tmp = skb->next) 2919 2920 #define skb_queue_walk_from(queue, skb) \ 2921 for (; skb != (struct sk_buff *)(queue); \ 2922 skb = skb->next) 2923 2924 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 2925 for (tmp = skb->next; \ 2926 skb != (struct sk_buff *)(queue); \ 2927 skb = tmp, tmp = skb->next) 2928 2929 #define skb_queue_reverse_walk(queue, skb) \ 2930 for (skb = (queue)->prev; \ 2931 skb != (struct sk_buff *)(queue); \ 2932 skb = skb->prev) 2933 2934 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 2935 for (skb = (queue)->prev, tmp = skb->prev; \ 2936 skb != (struct sk_buff *)(queue); \ 2937 skb = tmp, tmp = skb->prev) 2938 2939 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 2940 for (tmp = skb->prev; \ 2941 skb != (struct sk_buff *)(queue); \ 2942 skb = tmp, tmp = skb->prev) 2943 2944 static inline bool skb_has_frag_list(const struct sk_buff *skb) 2945 { 2946 return skb_shinfo(skb)->frag_list != NULL; 2947 } 2948 2949 static inline void skb_frag_list_init(struct sk_buff *skb) 2950 { 2951 skb_shinfo(skb)->frag_list = NULL; 2952 } 2953 2954 #define skb_walk_frags(skb, iter) \ 2955 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 2956 2957 2958 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, 2959 const struct sk_buff *skb); 2960 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags, 2961 int *peeked, int *off, int *err, 2962 struct sk_buff **last); 2963 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 2964 int *peeked, int *off, int *err); 2965 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 2966 int *err); 2967 unsigned int datagram_poll(struct file *file, struct socket *sock, 2968 struct poll_table_struct *wait); 2969 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 2970 struct iov_iter *to, int size); 2971 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 2972 struct msghdr *msg, int size) 2973 { 2974 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 2975 } 2976 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 2977 struct msghdr *msg); 2978 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 2979 struct iov_iter *from, int len); 2980 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 2981 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 2982 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 2983 static inline void skb_free_datagram_locked(struct sock *sk, 2984 struct sk_buff *skb) 2985 { 2986 __skb_free_datagram_locked(sk, skb, 0); 2987 } 2988 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 2989 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 2990 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 2991 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 2992 int len, __wsum csum); 2993 ssize_t skb_socket_splice(struct sock *sk, 2994 struct pipe_inode_info *pipe, 2995 struct splice_pipe_desc *spd); 2996 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 2997 struct pipe_inode_info *pipe, unsigned int len, 2998 unsigned int flags, 2999 ssize_t (*splice_cb)(struct sock *, 3000 struct pipe_inode_info *, 3001 struct splice_pipe_desc *)); 3002 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3003 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3004 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3005 int len, int hlen); 3006 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3007 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3008 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3009 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 3010 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3011 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3012 int skb_ensure_writable(struct sk_buff *skb, int write_len); 3013 int skb_vlan_pop(struct sk_buff *skb); 3014 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3015 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3016 gfp_t gfp); 3017 3018 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3019 { 3020 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3021 } 3022 3023 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3024 { 3025 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3026 } 3027 3028 struct skb_checksum_ops { 3029 __wsum (*update)(const void *mem, int len, __wsum wsum); 3030 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3031 }; 3032 3033 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3034 __wsum csum, const struct skb_checksum_ops *ops); 3035 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3036 __wsum csum); 3037 3038 static inline void * __must_check 3039 __skb_header_pointer(const struct sk_buff *skb, int offset, 3040 int len, void *data, int hlen, void *buffer) 3041 { 3042 if (hlen - offset >= len) 3043 return data + offset; 3044 3045 if (!skb || 3046 skb_copy_bits(skb, offset, buffer, len) < 0) 3047 return NULL; 3048 3049 return buffer; 3050 } 3051 3052 static inline void * __must_check 3053 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 3054 { 3055 return __skb_header_pointer(skb, offset, len, skb->data, 3056 skb_headlen(skb), buffer); 3057 } 3058 3059 /** 3060 * skb_needs_linearize - check if we need to linearize a given skb 3061 * depending on the given device features. 3062 * @skb: socket buffer to check 3063 * @features: net device features 3064 * 3065 * Returns true if either: 3066 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3067 * 2. skb is fragmented and the device does not support SG. 3068 */ 3069 static inline bool skb_needs_linearize(struct sk_buff *skb, 3070 netdev_features_t features) 3071 { 3072 return skb_is_nonlinear(skb) && 3073 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3074 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3075 } 3076 3077 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3078 void *to, 3079 const unsigned int len) 3080 { 3081 memcpy(to, skb->data, len); 3082 } 3083 3084 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3085 const int offset, void *to, 3086 const unsigned int len) 3087 { 3088 memcpy(to, skb->data + offset, len); 3089 } 3090 3091 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3092 const void *from, 3093 const unsigned int len) 3094 { 3095 memcpy(skb->data, from, len); 3096 } 3097 3098 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3099 const int offset, 3100 const void *from, 3101 const unsigned int len) 3102 { 3103 memcpy(skb->data + offset, from, len); 3104 } 3105 3106 void skb_init(void); 3107 3108 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3109 { 3110 return skb->tstamp; 3111 } 3112 3113 /** 3114 * skb_get_timestamp - get timestamp from a skb 3115 * @skb: skb to get stamp from 3116 * @stamp: pointer to struct timeval to store stamp in 3117 * 3118 * Timestamps are stored in the skb as offsets to a base timestamp. 3119 * This function converts the offset back to a struct timeval and stores 3120 * it in stamp. 3121 */ 3122 static inline void skb_get_timestamp(const struct sk_buff *skb, 3123 struct timeval *stamp) 3124 { 3125 *stamp = ktime_to_timeval(skb->tstamp); 3126 } 3127 3128 static inline void skb_get_timestampns(const struct sk_buff *skb, 3129 struct timespec *stamp) 3130 { 3131 *stamp = ktime_to_timespec(skb->tstamp); 3132 } 3133 3134 static inline void __net_timestamp(struct sk_buff *skb) 3135 { 3136 skb->tstamp = ktime_get_real(); 3137 } 3138 3139 static inline ktime_t net_timedelta(ktime_t t) 3140 { 3141 return ktime_sub(ktime_get_real(), t); 3142 } 3143 3144 static inline ktime_t net_invalid_timestamp(void) 3145 { 3146 return ktime_set(0, 0); 3147 } 3148 3149 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3150 3151 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3152 3153 void skb_clone_tx_timestamp(struct sk_buff *skb); 3154 bool skb_defer_rx_timestamp(struct sk_buff *skb); 3155 3156 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3157 3158 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3159 { 3160 } 3161 3162 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3163 { 3164 return false; 3165 } 3166 3167 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3168 3169 /** 3170 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3171 * 3172 * PHY drivers may accept clones of transmitted packets for 3173 * timestamping via their phy_driver.txtstamp method. These drivers 3174 * must call this function to return the skb back to the stack with a 3175 * timestamp. 3176 * 3177 * @skb: clone of the the original outgoing packet 3178 * @hwtstamps: hardware time stamps 3179 * 3180 */ 3181 void skb_complete_tx_timestamp(struct sk_buff *skb, 3182 struct skb_shared_hwtstamps *hwtstamps); 3183 3184 void __skb_tstamp_tx(struct sk_buff *orig_skb, 3185 struct skb_shared_hwtstamps *hwtstamps, 3186 struct sock *sk, int tstype); 3187 3188 /** 3189 * skb_tstamp_tx - queue clone of skb with send time stamps 3190 * @orig_skb: the original outgoing packet 3191 * @hwtstamps: hardware time stamps, may be NULL if not available 3192 * 3193 * If the skb has a socket associated, then this function clones the 3194 * skb (thus sharing the actual data and optional structures), stores 3195 * the optional hardware time stamping information (if non NULL) or 3196 * generates a software time stamp (otherwise), then queues the clone 3197 * to the error queue of the socket. Errors are silently ignored. 3198 */ 3199 void skb_tstamp_tx(struct sk_buff *orig_skb, 3200 struct skb_shared_hwtstamps *hwtstamps); 3201 3202 static inline void sw_tx_timestamp(struct sk_buff *skb) 3203 { 3204 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 3205 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 3206 skb_tstamp_tx(skb, NULL); 3207 } 3208 3209 /** 3210 * skb_tx_timestamp() - Driver hook for transmit timestamping 3211 * 3212 * Ethernet MAC Drivers should call this function in their hard_xmit() 3213 * function immediately before giving the sk_buff to the MAC hardware. 3214 * 3215 * Specifically, one should make absolutely sure that this function is 3216 * called before TX completion of this packet can trigger. Otherwise 3217 * the packet could potentially already be freed. 3218 * 3219 * @skb: A socket buffer. 3220 */ 3221 static inline void skb_tx_timestamp(struct sk_buff *skb) 3222 { 3223 skb_clone_tx_timestamp(skb); 3224 sw_tx_timestamp(skb); 3225 } 3226 3227 /** 3228 * skb_complete_wifi_ack - deliver skb with wifi status 3229 * 3230 * @skb: the original outgoing packet 3231 * @acked: ack status 3232 * 3233 */ 3234 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3235 3236 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3237 __sum16 __skb_checksum_complete(struct sk_buff *skb); 3238 3239 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3240 { 3241 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3242 skb->csum_valid || 3243 (skb->ip_summed == CHECKSUM_PARTIAL && 3244 skb_checksum_start_offset(skb) >= 0)); 3245 } 3246 3247 /** 3248 * skb_checksum_complete - Calculate checksum of an entire packet 3249 * @skb: packet to process 3250 * 3251 * This function calculates the checksum over the entire packet plus 3252 * the value of skb->csum. The latter can be used to supply the 3253 * checksum of a pseudo header as used by TCP/UDP. It returns the 3254 * checksum. 3255 * 3256 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3257 * this function can be used to verify that checksum on received 3258 * packets. In that case the function should return zero if the 3259 * checksum is correct. In particular, this function will return zero 3260 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3261 * hardware has already verified the correctness of the checksum. 3262 */ 3263 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3264 { 3265 return skb_csum_unnecessary(skb) ? 3266 0 : __skb_checksum_complete(skb); 3267 } 3268 3269 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3270 { 3271 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3272 if (skb->csum_level == 0) 3273 skb->ip_summed = CHECKSUM_NONE; 3274 else 3275 skb->csum_level--; 3276 } 3277 } 3278 3279 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3280 { 3281 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3282 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3283 skb->csum_level++; 3284 } else if (skb->ip_summed == CHECKSUM_NONE) { 3285 skb->ip_summed = CHECKSUM_UNNECESSARY; 3286 skb->csum_level = 0; 3287 } 3288 } 3289 3290 static inline void __skb_mark_checksum_bad(struct sk_buff *skb) 3291 { 3292 /* Mark current checksum as bad (typically called from GRO 3293 * path). In the case that ip_summed is CHECKSUM_NONE 3294 * this must be the first checksum encountered in the packet. 3295 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first 3296 * checksum after the last one validated. For UDP, a zero 3297 * checksum can not be marked as bad. 3298 */ 3299 3300 if (skb->ip_summed == CHECKSUM_NONE || 3301 skb->ip_summed == CHECKSUM_UNNECESSARY) 3302 skb->csum_bad = 1; 3303 } 3304 3305 /* Check if we need to perform checksum complete validation. 3306 * 3307 * Returns true if checksum complete is needed, false otherwise 3308 * (either checksum is unnecessary or zero checksum is allowed). 3309 */ 3310 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3311 bool zero_okay, 3312 __sum16 check) 3313 { 3314 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3315 skb->csum_valid = 1; 3316 __skb_decr_checksum_unnecessary(skb); 3317 return false; 3318 } 3319 3320 return true; 3321 } 3322 3323 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3324 * in checksum_init. 3325 */ 3326 #define CHECKSUM_BREAK 76 3327 3328 /* Unset checksum-complete 3329 * 3330 * Unset checksum complete can be done when packet is being modified 3331 * (uncompressed for instance) and checksum-complete value is 3332 * invalidated. 3333 */ 3334 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3335 { 3336 if (skb->ip_summed == CHECKSUM_COMPLETE) 3337 skb->ip_summed = CHECKSUM_NONE; 3338 } 3339 3340 /* Validate (init) checksum based on checksum complete. 3341 * 3342 * Return values: 3343 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3344 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3345 * checksum is stored in skb->csum for use in __skb_checksum_complete 3346 * non-zero: value of invalid checksum 3347 * 3348 */ 3349 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3350 bool complete, 3351 __wsum psum) 3352 { 3353 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3354 if (!csum_fold(csum_add(psum, skb->csum))) { 3355 skb->csum_valid = 1; 3356 return 0; 3357 } 3358 } else if (skb->csum_bad) { 3359 /* ip_summed == CHECKSUM_NONE in this case */ 3360 return (__force __sum16)1; 3361 } 3362 3363 skb->csum = psum; 3364 3365 if (complete || skb->len <= CHECKSUM_BREAK) { 3366 __sum16 csum; 3367 3368 csum = __skb_checksum_complete(skb); 3369 skb->csum_valid = !csum; 3370 return csum; 3371 } 3372 3373 return 0; 3374 } 3375 3376 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3377 { 3378 return 0; 3379 } 3380 3381 /* Perform checksum validate (init). Note that this is a macro since we only 3382 * want to calculate the pseudo header which is an input function if necessary. 3383 * First we try to validate without any computation (checksum unnecessary) and 3384 * then calculate based on checksum complete calling the function to compute 3385 * pseudo header. 3386 * 3387 * Return values: 3388 * 0: checksum is validated or try to in skb_checksum_complete 3389 * non-zero: value of invalid checksum 3390 */ 3391 #define __skb_checksum_validate(skb, proto, complete, \ 3392 zero_okay, check, compute_pseudo) \ 3393 ({ \ 3394 __sum16 __ret = 0; \ 3395 skb->csum_valid = 0; \ 3396 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3397 __ret = __skb_checksum_validate_complete(skb, \ 3398 complete, compute_pseudo(skb, proto)); \ 3399 __ret; \ 3400 }) 3401 3402 #define skb_checksum_init(skb, proto, compute_pseudo) \ 3403 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3404 3405 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3406 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3407 3408 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 3409 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3410 3411 #define skb_checksum_validate_zero_check(skb, proto, check, \ 3412 compute_pseudo) \ 3413 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3414 3415 #define skb_checksum_simple_validate(skb) \ 3416 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3417 3418 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3419 { 3420 return (skb->ip_summed == CHECKSUM_NONE && 3421 skb->csum_valid && !skb->csum_bad); 3422 } 3423 3424 static inline void __skb_checksum_convert(struct sk_buff *skb, 3425 __sum16 check, __wsum pseudo) 3426 { 3427 skb->csum = ~pseudo; 3428 skb->ip_summed = CHECKSUM_COMPLETE; 3429 } 3430 3431 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3432 do { \ 3433 if (__skb_checksum_convert_check(skb)) \ 3434 __skb_checksum_convert(skb, check, \ 3435 compute_pseudo(skb, proto)); \ 3436 } while (0) 3437 3438 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3439 u16 start, u16 offset) 3440 { 3441 skb->ip_summed = CHECKSUM_PARTIAL; 3442 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3443 skb->csum_offset = offset - start; 3444 } 3445 3446 /* Update skbuf and packet to reflect the remote checksum offload operation. 3447 * When called, ptr indicates the starting point for skb->csum when 3448 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3449 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3450 */ 3451 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3452 int start, int offset, bool nopartial) 3453 { 3454 __wsum delta; 3455 3456 if (!nopartial) { 3457 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3458 return; 3459 } 3460 3461 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3462 __skb_checksum_complete(skb); 3463 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3464 } 3465 3466 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3467 3468 /* Adjust skb->csum since we changed the packet */ 3469 skb->csum = csum_add(skb->csum, delta); 3470 } 3471 3472 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3473 void nf_conntrack_destroy(struct nf_conntrack *nfct); 3474 static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3475 { 3476 if (nfct && atomic_dec_and_test(&nfct->use)) 3477 nf_conntrack_destroy(nfct); 3478 } 3479 static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3480 { 3481 if (nfct) 3482 atomic_inc(&nfct->use); 3483 } 3484 #endif 3485 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3486 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3487 { 3488 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 3489 kfree(nf_bridge); 3490 } 3491 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3492 { 3493 if (nf_bridge) 3494 atomic_inc(&nf_bridge->use); 3495 } 3496 #endif /* CONFIG_BRIDGE_NETFILTER */ 3497 static inline void nf_reset(struct sk_buff *skb) 3498 { 3499 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3500 nf_conntrack_put(skb->nfct); 3501 skb->nfct = NULL; 3502 #endif 3503 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3504 nf_bridge_put(skb->nf_bridge); 3505 skb->nf_bridge = NULL; 3506 #endif 3507 } 3508 3509 static inline void nf_reset_trace(struct sk_buff *skb) 3510 { 3511 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3512 skb->nf_trace = 0; 3513 #endif 3514 } 3515 3516 /* Note: This doesn't put any conntrack and bridge info in dst. */ 3517 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3518 bool copy) 3519 { 3520 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3521 dst->nfct = src->nfct; 3522 nf_conntrack_get(src->nfct); 3523 if (copy) 3524 dst->nfctinfo = src->nfctinfo; 3525 #endif 3526 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3527 dst->nf_bridge = src->nf_bridge; 3528 nf_bridge_get(src->nf_bridge); 3529 #endif 3530 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3531 if (copy) 3532 dst->nf_trace = src->nf_trace; 3533 #endif 3534 } 3535 3536 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3537 { 3538 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3539 nf_conntrack_put(dst->nfct); 3540 #endif 3541 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3542 nf_bridge_put(dst->nf_bridge); 3543 #endif 3544 __nf_copy(dst, src, true); 3545 } 3546 3547 #ifdef CONFIG_NETWORK_SECMARK 3548 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3549 { 3550 to->secmark = from->secmark; 3551 } 3552 3553 static inline void skb_init_secmark(struct sk_buff *skb) 3554 { 3555 skb->secmark = 0; 3556 } 3557 #else 3558 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3559 { } 3560 3561 static inline void skb_init_secmark(struct sk_buff *skb) 3562 { } 3563 #endif 3564 3565 static inline bool skb_irq_freeable(const struct sk_buff *skb) 3566 { 3567 return !skb->destructor && 3568 #if IS_ENABLED(CONFIG_XFRM) 3569 !skb->sp && 3570 #endif 3571 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 3572 !skb->nfct && 3573 #endif 3574 !skb->_skb_refdst && 3575 !skb_has_frag_list(skb); 3576 } 3577 3578 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3579 { 3580 skb->queue_mapping = queue_mapping; 3581 } 3582 3583 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3584 { 3585 return skb->queue_mapping; 3586 } 3587 3588 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3589 { 3590 to->queue_mapping = from->queue_mapping; 3591 } 3592 3593 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3594 { 3595 skb->queue_mapping = rx_queue + 1; 3596 } 3597 3598 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3599 { 3600 return skb->queue_mapping - 1; 3601 } 3602 3603 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3604 { 3605 return skb->queue_mapping != 0; 3606 } 3607 3608 static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3609 { 3610 #ifdef CONFIG_XFRM 3611 return skb->sp; 3612 #else 3613 return NULL; 3614 #endif 3615 } 3616 3617 /* Keeps track of mac header offset relative to skb->head. 3618 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3619 * For non-tunnel skb it points to skb_mac_header() and for 3620 * tunnel skb it points to outer mac header. 3621 * Keeps track of level of encapsulation of network headers. 3622 */ 3623 struct skb_gso_cb { 3624 union { 3625 int mac_offset; 3626 int data_offset; 3627 }; 3628 int encap_level; 3629 __wsum csum; 3630 __u16 csum_start; 3631 }; 3632 #define SKB_SGO_CB_OFFSET 32 3633 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 3634 3635 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3636 { 3637 return (skb_mac_header(inner_skb) - inner_skb->head) - 3638 SKB_GSO_CB(inner_skb)->mac_offset; 3639 } 3640 3641 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3642 { 3643 int new_headroom, headroom; 3644 int ret; 3645 3646 headroom = skb_headroom(skb); 3647 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3648 if (ret) 3649 return ret; 3650 3651 new_headroom = skb_headroom(skb); 3652 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3653 return 0; 3654 } 3655 3656 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 3657 { 3658 /* Do not update partial checksums if remote checksum is enabled. */ 3659 if (skb->remcsum_offload) 3660 return; 3661 3662 SKB_GSO_CB(skb)->csum = res; 3663 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 3664 } 3665 3666 /* Compute the checksum for a gso segment. First compute the checksum value 3667 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3668 * then add in skb->csum (checksum from csum_start to end of packet). 3669 * skb->csum and csum_start are then updated to reflect the checksum of the 3670 * resultant packet starting from the transport header-- the resultant checksum 3671 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3672 * header. 3673 */ 3674 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3675 { 3676 unsigned char *csum_start = skb_transport_header(skb); 3677 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 3678 __wsum partial = SKB_GSO_CB(skb)->csum; 3679 3680 SKB_GSO_CB(skb)->csum = res; 3681 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 3682 3683 return csum_fold(csum_partial(csum_start, plen, partial)); 3684 } 3685 3686 static inline bool skb_is_gso(const struct sk_buff *skb) 3687 { 3688 return skb_shinfo(skb)->gso_size; 3689 } 3690 3691 /* Note: Should be called only if skb_is_gso(skb) is true */ 3692 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3693 { 3694 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3695 } 3696 3697 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3698 3699 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3700 { 3701 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3702 * wanted then gso_type will be set. */ 3703 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3704 3705 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3706 unlikely(shinfo->gso_type == 0)) { 3707 __skb_warn_lro_forwarding(skb); 3708 return true; 3709 } 3710 return false; 3711 } 3712 3713 static inline void skb_forward_csum(struct sk_buff *skb) 3714 { 3715 /* Unfortunately we don't support this one. Any brave souls? */ 3716 if (skb->ip_summed == CHECKSUM_COMPLETE) 3717 skb->ip_summed = CHECKSUM_NONE; 3718 } 3719 3720 /** 3721 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3722 * @skb: skb to check 3723 * 3724 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3725 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3726 * use this helper, to document places where we make this assertion. 3727 */ 3728 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3729 { 3730 #ifdef DEBUG 3731 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3732 #endif 3733 } 3734 3735 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3736 3737 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3738 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 3739 unsigned int transport_len, 3740 __sum16(*skb_chkf)(struct sk_buff *skb)); 3741 3742 /** 3743 * skb_head_is_locked - Determine if the skb->head is locked down 3744 * @skb: skb to check 3745 * 3746 * The head on skbs build around a head frag can be removed if they are 3747 * not cloned. This function returns true if the skb head is locked down 3748 * due to either being allocated via kmalloc, or by being a clone with 3749 * multiple references to the head. 3750 */ 3751 static inline bool skb_head_is_locked(const struct sk_buff *skb) 3752 { 3753 return !skb->head_frag || skb_cloned(skb); 3754 } 3755 3756 /** 3757 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3758 * 3759 * @skb: GSO skb 3760 * 3761 * skb_gso_network_seglen is used to determine the real size of the 3762 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3763 * 3764 * The MAC/L2 header is not accounted for. 3765 */ 3766 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 3767 { 3768 unsigned int hdr_len = skb_transport_header(skb) - 3769 skb_network_header(skb); 3770 return hdr_len + skb_gso_transport_seglen(skb); 3771 } 3772 3773 /* Local Checksum Offload. 3774 * Compute outer checksum based on the assumption that the 3775 * inner checksum will be offloaded later. 3776 * See Documentation/networking/checksum-offloads.txt for 3777 * explanation of how this works. 3778 * Fill in outer checksum adjustment (e.g. with sum of outer 3779 * pseudo-header) before calling. 3780 * Also ensure that inner checksum is in linear data area. 3781 */ 3782 static inline __wsum lco_csum(struct sk_buff *skb) 3783 { 3784 unsigned char *csum_start = skb_checksum_start(skb); 3785 unsigned char *l4_hdr = skb_transport_header(skb); 3786 __wsum partial; 3787 3788 /* Start with complement of inner checksum adjustment */ 3789 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 3790 skb->csum_offset)); 3791 3792 /* Add in checksum of our headers (incl. outer checksum 3793 * adjustment filled in by caller) and return result. 3794 */ 3795 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 3796 } 3797 3798 #endif /* __KERNEL__ */ 3799 #endif /* _LINUX_SKBUFF_H */ 3800
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.