1 /* 2 * Definitions for the 'struct sk_buff' memory handlers. 3 * 4 * Authors: 5 * Alan Cox, <[email protected]> 6 * Florian La Roche, <[email protected]> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 */ 13 14 #ifndef _LINUX_SKBUFF_H 15 #define _LINUX_SKBUFF_H 16 17 #include <linux/kernel.h> 18 #include <linux/kmemcheck.h> 19 #include <linux/compiler.h> 20 #include <linux/time.h> 21 #include <linux/bug.h> 22 #include <linux/cache.h> 23 #include <linux/rbtree.h> 24 #include <linux/socket.h> 25 26 #include <linux/atomic.h> 27 #include <asm/types.h> 28 #include <linux/spinlock.h> 29 #include <linux/net.h> 30 #include <linux/textsearch.h> 31 #include <net/checksum.h> 32 #include <linux/rcupdate.h> 33 #include <linux/hrtimer.h> 34 #include <linux/dma-mapping.h> 35 #include <linux/netdev_features.h> 36 #include <linux/sched.h> 37 #include <net/flow_dissector.h> 38 #include <linux/splice.h> 39 #include <linux/in6.h> 40 #include <net/flow.h> 41 42 /* The interface for checksum offload between the stack and networking drivers 43 * is as follows... 44 * 45 * A. IP checksum related features 46 * 47 * Drivers advertise checksum offload capabilities in the features of a device. 48 * From the stack's point of view these are capabilities offered by the driver, 49 * a driver typically only advertises features that it is capable of offloading 50 * to its device. 51 * 52 * The checksum related features are: 53 * 54 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one 55 * IP (one's complement) checksum for any combination 56 * of protocols or protocol layering. The checksum is 57 * computed and set in a packet per the CHECKSUM_PARTIAL 58 * interface (see below). 59 * 60 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain 61 * TCP or UDP packets over IPv4. These are specifically 62 * unencapsulated packets of the form IPv4|TCP or 63 * IPv4|UDP where the Protocol field in the IPv4 header 64 * is TCP or UDP. The IPv4 header may contain IP options 65 * This feature cannot be set in features for a device 66 * with NETIF_F_HW_CSUM also set. This feature is being 67 * DEPRECATED (see below). 68 * 69 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain 70 * TCP or UDP packets over IPv6. These are specifically 71 * unencapsulated packets of the form IPv6|TCP or 72 * IPv4|UDP where the Next Header field in the IPv6 73 * header is either TCP or UDP. IPv6 extension headers 74 * are not supported with this feature. This feature 75 * cannot be set in features for a device with 76 * NETIF_F_HW_CSUM also set. This feature is being 77 * DEPRECATED (see below). 78 * 79 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload. 80 * This flag is used only used to disable the RX checksum 81 * feature for a device. The stack will accept receive 82 * checksum indication in packets received on a device 83 * regardless of whether NETIF_F_RXCSUM is set. 84 * 85 * B. Checksumming of received packets by device. Indication of checksum 86 * verification is in set skb->ip_summed. Possible values are: 87 * 88 * CHECKSUM_NONE: 89 * 90 * Device did not checksum this packet e.g. due to lack of capabilities. 91 * The packet contains full (though not verified) checksum in packet but 92 * not in skb->csum. Thus, skb->csum is undefined in this case. 93 * 94 * CHECKSUM_UNNECESSARY: 95 * 96 * The hardware you're dealing with doesn't calculate the full checksum 97 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 98 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY 99 * if their checksums are okay. skb->csum is still undefined in this case 100 * though. A driver or device must never modify the checksum field in the 101 * packet even if checksum is verified. 102 * 103 * CHECKSUM_UNNECESSARY is applicable to following protocols: 104 * TCP: IPv6 and IPv4. 105 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 106 * zero UDP checksum for either IPv4 or IPv6, the networking stack 107 * may perform further validation in this case. 108 * GRE: only if the checksum is present in the header. 109 * SCTP: indicates the CRC in SCTP header has been validated. 110 * 111 * skb->csum_level indicates the number of consecutive checksums found in 112 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY. 113 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 114 * and a device is able to verify the checksums for UDP (possibly zero), 115 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to 116 * two. If the device were only able to verify the UDP checksum and not 117 * GRE, either because it doesn't support GRE checksum of because GRE 118 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 119 * not considered in this case). 120 * 121 * CHECKSUM_COMPLETE: 122 * 123 * This is the most generic way. The device supplied checksum of the _whole_ 124 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the 125 * hardware doesn't need to parse L3/L4 headers to implement this. 126 * 127 * Note: Even if device supports only some protocols, but is able to produce 128 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 129 * 130 * CHECKSUM_PARTIAL: 131 * 132 * A checksum is set up to be offloaded to a device as described in the 133 * output description for CHECKSUM_PARTIAL. This may occur on a packet 134 * received directly from another Linux OS, e.g., a virtualized Linux kernel 135 * on the same host, or it may be set in the input path in GRO or remote 136 * checksum offload. For the purposes of checksum verification, the checksum 137 * referred to by skb->csum_start + skb->csum_offset and any preceding 138 * checksums in the packet are considered verified. Any checksums in the 139 * packet that are after the checksum being offloaded are not considered to 140 * be verified. 141 * 142 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload 143 * in the skb->ip_summed for a packet. Values are: 144 * 145 * CHECKSUM_PARTIAL: 146 * 147 * The driver is required to checksum the packet as seen by hard_start_xmit() 148 * from skb->csum_start up to the end, and to record/write the checksum at 149 * offset skb->csum_start + skb->csum_offset. A driver may verify that the 150 * csum_start and csum_offset values are valid values given the length and 151 * offset of the packet, however they should not attempt to validate that the 152 * checksum refers to a legitimate transport layer checksum-- it is the 153 * purview of the stack to validate that csum_start and csum_offset are set 154 * correctly. 155 * 156 * When the stack requests checksum offload for a packet, the driver MUST 157 * ensure that the checksum is set correctly. A driver can either offload the 158 * checksum calculation to the device, or call skb_checksum_help (in the case 159 * that the device does not support offload for a particular checksum). 160 * 161 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of 162 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate 163 * checksum offload capability. If a device has limited checksum capabilities 164 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as 165 * described above) a helper function can be called to resolve 166 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper 167 * function takes a spec argument that describes the protocol layer that is 168 * supported for checksum offload and can be called for each packet. If a 169 * packet does not match the specification for offload, skb_checksum_help 170 * is called to resolve the checksum. 171 * 172 * CHECKSUM_NONE: 173 * 174 * The skb was already checksummed by the protocol, or a checksum is not 175 * required. 176 * 177 * CHECKSUM_UNNECESSARY: 178 * 179 * This has the same meaning on as CHECKSUM_NONE for checksum offload on 180 * output. 181 * 182 * CHECKSUM_COMPLETE: 183 * Not used in checksum output. If a driver observes a packet with this value 184 * set in skbuff, if should treat as CHECKSUM_NONE being set. 185 * 186 * D. Non-IP checksum (CRC) offloads 187 * 188 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of 189 * offloading the SCTP CRC in a packet. To perform this offload the stack 190 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 191 * accordingly. Note the there is no indication in the skbuff that the 192 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports 193 * both IP checksum offload and SCTP CRC offload must verify which offload 194 * is configured for a packet presumably by inspecting packet headers. 195 * 196 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of 197 * offloading the FCOE CRC in a packet. To perform this offload the stack 198 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 199 * accordingly. Note the there is no indication in the skbuff that the 200 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports 201 * both IP checksum offload and FCOE CRC offload must verify which offload 202 * is configured for a packet presumably by inspecting packet headers. 203 * 204 * E. Checksumming on output with GSO. 205 * 206 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload 207 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 208 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as 209 * part of the GSO operation is implied. If a checksum is being offloaded 210 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset 211 * are set to refer to the outermost checksum being offload (two offloaded 212 * checksums are possible with UDP encapsulation). 213 */ 214 215 /* Don't change this without changing skb_csum_unnecessary! */ 216 #define CHECKSUM_NONE 0 217 #define CHECKSUM_UNNECESSARY 1 218 #define CHECKSUM_COMPLETE 2 219 #define CHECKSUM_PARTIAL 3 220 221 /* Maximum value in skb->csum_level */ 222 #define SKB_MAX_CSUM_LEVEL 3 223 224 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 225 #define SKB_WITH_OVERHEAD(X) \ 226 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 227 #define SKB_MAX_ORDER(X, ORDER) \ 228 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 229 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 230 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 231 232 /* return minimum truesize of one skb containing X bytes of data */ 233 #define SKB_TRUESIZE(X) ((X) + \ 234 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 235 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 236 237 struct net_device; 238 struct scatterlist; 239 struct pipe_inode_info; 240 struct iov_iter; 241 struct napi_struct; 242 243 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 244 struct nf_conntrack { 245 atomic_t use; 246 }; 247 #endif 248 249 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 250 struct nf_bridge_info { 251 atomic_t use; 252 enum { 253 BRNF_PROTO_UNCHANGED, 254 BRNF_PROTO_8021Q, 255 BRNF_PROTO_PPPOE 256 } orig_proto:8; 257 u8 pkt_otherhost:1; 258 u8 in_prerouting:1; 259 u8 bridged_dnat:1; 260 __u16 frag_max_size; 261 struct net_device *physindev; 262 263 /* always valid & non-NULL from FORWARD on, for physdev match */ 264 struct net_device *physoutdev; 265 union { 266 /* prerouting: detect dnat in orig/reply direction */ 267 __be32 ipv4_daddr; 268 struct in6_addr ipv6_daddr; 269 270 /* after prerouting + nat detected: store original source 271 * mac since neigh resolution overwrites it, only used while 272 * skb is out in neigh layer. 273 */ 274 char neigh_header[8]; 275 }; 276 }; 277 #endif 278 279 struct sk_buff_head { 280 /* These two members must be first. */ 281 struct sk_buff *next; 282 struct sk_buff *prev; 283 284 __u32 qlen; 285 spinlock_t lock; 286 }; 287 288 struct sk_buff; 289 290 /* To allow 64K frame to be packed as single skb without frag_list we 291 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 292 * buffers which do not start on a page boundary. 293 * 294 * Since GRO uses frags we allocate at least 16 regardless of page 295 * size. 296 */ 297 #if (65536/PAGE_SIZE + 1) < 16 298 #define MAX_SKB_FRAGS 16UL 299 #else 300 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 301 #endif 302 303 typedef struct skb_frag_struct skb_frag_t; 304 305 struct skb_frag_struct { 306 struct { 307 struct page *p; 308 } page; 309 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 310 __u32 page_offset; 311 __u32 size; 312 #else 313 __u16 page_offset; 314 __u16 size; 315 #endif 316 }; 317 318 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 319 { 320 return frag->size; 321 } 322 323 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 324 { 325 frag->size = size; 326 } 327 328 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 329 { 330 frag->size += delta; 331 } 332 333 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 334 { 335 frag->size -= delta; 336 } 337 338 #define HAVE_HW_TIME_STAMP 339 340 /** 341 * struct skb_shared_hwtstamps - hardware time stamps 342 * @hwtstamp: hardware time stamp transformed into duration 343 * since arbitrary point in time 344 * 345 * Software time stamps generated by ktime_get_real() are stored in 346 * skb->tstamp. 347 * 348 * hwtstamps can only be compared against other hwtstamps from 349 * the same device. 350 * 351 * This structure is attached to packets as part of the 352 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 353 */ 354 struct skb_shared_hwtstamps { 355 ktime_t hwtstamp; 356 }; 357 358 /* Definitions for tx_flags in struct skb_shared_info */ 359 enum { 360 /* generate hardware time stamp */ 361 SKBTX_HW_TSTAMP = 1 << 0, 362 363 /* generate software time stamp when queueing packet to NIC */ 364 SKBTX_SW_TSTAMP = 1 << 1, 365 366 /* device driver is going to provide hardware time stamp */ 367 SKBTX_IN_PROGRESS = 1 << 2, 368 369 /* device driver supports TX zero-copy buffers */ 370 SKBTX_DEV_ZEROCOPY = 1 << 3, 371 372 /* generate wifi status information (where possible) */ 373 SKBTX_WIFI_STATUS = 1 << 4, 374 375 /* This indicates at least one fragment might be overwritten 376 * (as in vmsplice(), sendfile() ...) 377 * If we need to compute a TX checksum, we'll need to copy 378 * all frags to avoid possible bad checksum 379 */ 380 SKBTX_SHARED_FRAG = 1 << 5, 381 382 /* generate software time stamp when entering packet scheduling */ 383 SKBTX_SCHED_TSTAMP = 1 << 6, 384 385 /* generate software timestamp on peer data acknowledgment */ 386 SKBTX_ACK_TSTAMP = 1 << 7, 387 }; 388 389 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 390 SKBTX_SCHED_TSTAMP | \ 391 SKBTX_ACK_TSTAMP) 392 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 393 394 /* 395 * The callback notifies userspace to release buffers when skb DMA is done in 396 * lower device, the skb last reference should be 0 when calling this. 397 * The zerocopy_success argument is true if zero copy transmit occurred, 398 * false on data copy or out of memory error caused by data copy attempt. 399 * The ctx field is used to track device context. 400 * The desc field is used to track userspace buffer index. 401 */ 402 struct ubuf_info { 403 void (*callback)(struct ubuf_info *, bool zerocopy_success); 404 void *ctx; 405 unsigned long desc; 406 }; 407 408 /* This data is invariant across clones and lives at 409 * the end of the header data, ie. at skb->end. 410 */ 411 struct skb_shared_info { 412 unsigned char nr_frags; 413 __u8 tx_flags; 414 unsigned short gso_size; 415 /* Warning: this field is not always filled in (UFO)! */ 416 unsigned short gso_segs; 417 unsigned short gso_type; 418 struct sk_buff *frag_list; 419 struct skb_shared_hwtstamps hwtstamps; 420 u32 tskey; 421 __be32 ip6_frag_id; 422 423 /* 424 * Warning : all fields before dataref are cleared in __alloc_skb() 425 */ 426 atomic_t dataref; 427 428 /* Intermediate layers must ensure that destructor_arg 429 * remains valid until skb destructor */ 430 void * destructor_arg; 431 432 /* must be last field, see pskb_expand_head() */ 433 skb_frag_t frags[MAX_SKB_FRAGS]; 434 }; 435 436 /* We divide dataref into two halves. The higher 16 bits hold references 437 * to the payload part of skb->data. The lower 16 bits hold references to 438 * the entire skb->data. A clone of a headerless skb holds the length of 439 * the header in skb->hdr_len. 440 * 441 * All users must obey the rule that the skb->data reference count must be 442 * greater than or equal to the payload reference count. 443 * 444 * Holding a reference to the payload part means that the user does not 445 * care about modifications to the header part of skb->data. 446 */ 447 #define SKB_DATAREF_SHIFT 16 448 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 449 450 451 enum { 452 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 453 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 454 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 455 }; 456 457 enum { 458 SKB_GSO_TCPV4 = 1 << 0, 459 SKB_GSO_UDP = 1 << 1, 460 461 /* This indicates the skb is from an untrusted source. */ 462 SKB_GSO_DODGY = 1 << 2, 463 464 /* This indicates the tcp segment has CWR set. */ 465 SKB_GSO_TCP_ECN = 1 << 3, 466 467 SKB_GSO_TCPV6 = 1 << 4, 468 469 SKB_GSO_FCOE = 1 << 5, 470 471 SKB_GSO_GRE = 1 << 6, 472 473 SKB_GSO_GRE_CSUM = 1 << 7, 474 475 SKB_GSO_IPIP = 1 << 8, 476 477 SKB_GSO_SIT = 1 << 9, 478 479 SKB_GSO_UDP_TUNNEL = 1 << 10, 480 481 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 482 483 SKB_GSO_TUNNEL_REMCSUM = 1 << 12, 484 }; 485 486 #if BITS_PER_LONG > 32 487 #define NET_SKBUFF_DATA_USES_OFFSET 1 488 #endif 489 490 #ifdef NET_SKBUFF_DATA_USES_OFFSET 491 typedef unsigned int sk_buff_data_t; 492 #else 493 typedef unsigned char *sk_buff_data_t; 494 #endif 495 496 /** 497 * struct skb_mstamp - multi resolution time stamps 498 * @stamp_us: timestamp in us resolution 499 * @stamp_jiffies: timestamp in jiffies 500 */ 501 struct skb_mstamp { 502 union { 503 u64 v64; 504 struct { 505 u32 stamp_us; 506 u32 stamp_jiffies; 507 }; 508 }; 509 }; 510 511 /** 512 * skb_mstamp_get - get current timestamp 513 * @cl: place to store timestamps 514 */ 515 static inline void skb_mstamp_get(struct skb_mstamp *cl) 516 { 517 u64 val = local_clock(); 518 519 do_div(val, NSEC_PER_USEC); 520 cl->stamp_us = (u32)val; 521 cl->stamp_jiffies = (u32)jiffies; 522 } 523 524 /** 525 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp 526 * @t1: pointer to newest sample 527 * @t0: pointer to oldest sample 528 */ 529 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1, 530 const struct skb_mstamp *t0) 531 { 532 s32 delta_us = t1->stamp_us - t0->stamp_us; 533 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies; 534 535 /* If delta_us is negative, this might be because interval is too big, 536 * or local_clock() drift is too big : fallback using jiffies. 537 */ 538 if (delta_us <= 0 || 539 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ))) 540 541 delta_us = jiffies_to_usecs(delta_jiffies); 542 543 return delta_us; 544 } 545 546 static inline bool skb_mstamp_after(const struct skb_mstamp *t1, 547 const struct skb_mstamp *t0) 548 { 549 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies; 550 551 if (!diff) 552 diff = t1->stamp_us - t0->stamp_us; 553 return diff > 0; 554 } 555 556 /** 557 * struct sk_buff - socket buffer 558 * @next: Next buffer in list 559 * @prev: Previous buffer in list 560 * @tstamp: Time we arrived/left 561 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 562 * @sk: Socket we are owned by 563 * @dev: Device we arrived on/are leaving by 564 * @cb: Control buffer. Free for use by every layer. Put private vars here 565 * @_skb_refdst: destination entry (with norefcount bit) 566 * @sp: the security path, used for xfrm 567 * @len: Length of actual data 568 * @data_len: Data length 569 * @mac_len: Length of link layer header 570 * @hdr_len: writable header length of cloned skb 571 * @csum: Checksum (must include start/offset pair) 572 * @csum_start: Offset from skb->head where checksumming should start 573 * @csum_offset: Offset from csum_start where checksum should be stored 574 * @priority: Packet queueing priority 575 * @ignore_df: allow local fragmentation 576 * @cloned: Head may be cloned (check refcnt to be sure) 577 * @ip_summed: Driver fed us an IP checksum 578 * @nohdr: Payload reference only, must not modify header 579 * @nfctinfo: Relationship of this skb to the connection 580 * @pkt_type: Packet class 581 * @fclone: skbuff clone status 582 * @ipvs_property: skbuff is owned by ipvs 583 * @peeked: this packet has been seen already, so stats have been 584 * done for it, don't do them again 585 * @nf_trace: netfilter packet trace flag 586 * @protocol: Packet protocol from driver 587 * @destructor: Destruct function 588 * @nfct: Associated connection, if any 589 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 590 * @skb_iif: ifindex of device we arrived on 591 * @tc_index: Traffic control index 592 * @tc_verd: traffic control verdict 593 * @hash: the packet hash 594 * @queue_mapping: Queue mapping for multiqueue devices 595 * @xmit_more: More SKBs are pending for this queue 596 * @ndisc_nodetype: router type (from link layer) 597 * @ooo_okay: allow the mapping of a socket to a queue to be changed 598 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 599 * ports. 600 * @sw_hash: indicates hash was computed in software stack 601 * @wifi_acked_valid: wifi_acked was set 602 * @wifi_acked: whether frame was acked on wifi or not 603 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 604 * @napi_id: id of the NAPI struct this skb came from 605 * @secmark: security marking 606 * @offload_fwd_mark: fwding offload mark 607 * @mark: Generic packet mark 608 * @vlan_proto: vlan encapsulation protocol 609 * @vlan_tci: vlan tag control information 610 * @inner_protocol: Protocol (encapsulation) 611 * @inner_transport_header: Inner transport layer header (encapsulation) 612 * @inner_network_header: Network layer header (encapsulation) 613 * @inner_mac_header: Link layer header (encapsulation) 614 * @transport_header: Transport layer header 615 * @network_header: Network layer header 616 * @mac_header: Link layer header 617 * @tail: Tail pointer 618 * @end: End pointer 619 * @head: Head of buffer 620 * @data: Data head pointer 621 * @truesize: Buffer size 622 * @users: User count - see {datagram,tcp}.c 623 */ 624 625 struct sk_buff { 626 union { 627 struct { 628 /* These two members must be first. */ 629 struct sk_buff *next; 630 struct sk_buff *prev; 631 632 union { 633 ktime_t tstamp; 634 struct skb_mstamp skb_mstamp; 635 }; 636 }; 637 struct rb_node rbnode; /* used in netem & tcp stack */ 638 }; 639 struct sock *sk; 640 struct net_device *dev; 641 642 /* 643 * This is the control buffer. It is free to use for every 644 * layer. Please put your private variables there. If you 645 * want to keep them across layers you have to do a skb_clone() 646 * first. This is owned by whoever has the skb queued ATM. 647 */ 648 char cb[48] __aligned(8); 649 650 unsigned long _skb_refdst; 651 void (*destructor)(struct sk_buff *skb); 652 #ifdef CONFIG_XFRM 653 struct sec_path *sp; 654 #endif 655 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 656 struct nf_conntrack *nfct; 657 #endif 658 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 659 struct nf_bridge_info *nf_bridge; 660 #endif 661 unsigned int len, 662 data_len; 663 __u16 mac_len, 664 hdr_len; 665 666 /* Following fields are _not_ copied in __copy_skb_header() 667 * Note that queue_mapping is here mostly to fill a hole. 668 */ 669 kmemcheck_bitfield_begin(flags1); 670 __u16 queue_mapping; 671 __u8 cloned:1, 672 nohdr:1, 673 fclone:2, 674 peeked:1, 675 head_frag:1, 676 xmit_more:1; 677 /* one bit hole */ 678 kmemcheck_bitfield_end(flags1); 679 680 /* fields enclosed in headers_start/headers_end are copied 681 * using a single memcpy() in __copy_skb_header() 682 */ 683 /* private: */ 684 __u32 headers_start[0]; 685 /* public: */ 686 687 /* if you move pkt_type around you also must adapt those constants */ 688 #ifdef __BIG_ENDIAN_BITFIELD 689 #define PKT_TYPE_MAX (7 << 5) 690 #else 691 #define PKT_TYPE_MAX 7 692 #endif 693 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 694 695 __u8 __pkt_type_offset[0]; 696 __u8 pkt_type:3; 697 __u8 pfmemalloc:1; 698 __u8 ignore_df:1; 699 __u8 nfctinfo:3; 700 701 __u8 nf_trace:1; 702 __u8 ip_summed:2; 703 __u8 ooo_okay:1; 704 __u8 l4_hash:1; 705 __u8 sw_hash:1; 706 __u8 wifi_acked_valid:1; 707 __u8 wifi_acked:1; 708 709 __u8 no_fcs:1; 710 /* Indicates the inner headers are valid in the skbuff. */ 711 __u8 encapsulation:1; 712 __u8 encap_hdr_csum:1; 713 __u8 csum_valid:1; 714 __u8 csum_complete_sw:1; 715 __u8 csum_level:2; 716 __u8 csum_bad:1; 717 718 #ifdef CONFIG_IPV6_NDISC_NODETYPE 719 __u8 ndisc_nodetype:2; 720 #endif 721 __u8 ipvs_property:1; 722 __u8 inner_protocol_type:1; 723 __u8 remcsum_offload:1; 724 /* 3 or 5 bit hole */ 725 726 #ifdef CONFIG_NET_SCHED 727 __u16 tc_index; /* traffic control index */ 728 #ifdef CONFIG_NET_CLS_ACT 729 __u16 tc_verd; /* traffic control verdict */ 730 #endif 731 #endif 732 733 union { 734 __wsum csum; 735 struct { 736 __u16 csum_start; 737 __u16 csum_offset; 738 }; 739 }; 740 __u32 priority; 741 int skb_iif; 742 __u32 hash; 743 __be16 vlan_proto; 744 __u16 vlan_tci; 745 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 746 union { 747 unsigned int napi_id; 748 unsigned int sender_cpu; 749 }; 750 #endif 751 union { 752 #ifdef CONFIG_NETWORK_SECMARK 753 __u32 secmark; 754 #endif 755 #ifdef CONFIG_NET_SWITCHDEV 756 __u32 offload_fwd_mark; 757 #endif 758 }; 759 760 union { 761 __u32 mark; 762 __u32 reserved_tailroom; 763 }; 764 765 union { 766 __be16 inner_protocol; 767 __u8 inner_ipproto; 768 }; 769 770 __u16 inner_transport_header; 771 __u16 inner_network_header; 772 __u16 inner_mac_header; 773 774 __be16 protocol; 775 __u16 transport_header; 776 __u16 network_header; 777 __u16 mac_header; 778 779 /* private: */ 780 __u32 headers_end[0]; 781 /* public: */ 782 783 /* These elements must be at the end, see alloc_skb() for details. */ 784 sk_buff_data_t tail; 785 sk_buff_data_t end; 786 unsigned char *head, 787 *data; 788 unsigned int truesize; 789 atomic_t users; 790 }; 791 792 #ifdef __KERNEL__ 793 /* 794 * Handling routines are only of interest to the kernel 795 */ 796 #include <linux/slab.h> 797 798 799 #define SKB_ALLOC_FCLONE 0x01 800 #define SKB_ALLOC_RX 0x02 801 #define SKB_ALLOC_NAPI 0x04 802 803 /* Returns true if the skb was allocated from PFMEMALLOC reserves */ 804 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 805 { 806 return unlikely(skb->pfmemalloc); 807 } 808 809 /* 810 * skb might have a dst pointer attached, refcounted or not. 811 * _skb_refdst low order bit is set if refcount was _not_ taken 812 */ 813 #define SKB_DST_NOREF 1UL 814 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 815 816 /** 817 * skb_dst - returns skb dst_entry 818 * @skb: buffer 819 * 820 * Returns skb dst_entry, regardless of reference taken or not. 821 */ 822 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 823 { 824 /* If refdst was not refcounted, check we still are in a 825 * rcu_read_lock section 826 */ 827 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 828 !rcu_read_lock_held() && 829 !rcu_read_lock_bh_held()); 830 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 831 } 832 833 /** 834 * skb_dst_set - sets skb dst 835 * @skb: buffer 836 * @dst: dst entry 837 * 838 * Sets skb dst, assuming a reference was taken on dst and should 839 * be released by skb_dst_drop() 840 */ 841 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 842 { 843 skb->_skb_refdst = (unsigned long)dst; 844 } 845 846 /** 847 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 848 * @skb: buffer 849 * @dst: dst entry 850 * 851 * Sets skb dst, assuming a reference was not taken on dst. 852 * If dst entry is cached, we do not take reference and dst_release 853 * will be avoided by refdst_drop. If dst entry is not cached, we take 854 * reference, so that last dst_release can destroy the dst immediately. 855 */ 856 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 857 { 858 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 859 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 860 } 861 862 /** 863 * skb_dst_is_noref - Test if skb dst isn't refcounted 864 * @skb: buffer 865 */ 866 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 867 { 868 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 869 } 870 871 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 872 { 873 return (struct rtable *)skb_dst(skb); 874 } 875 876 void kfree_skb(struct sk_buff *skb); 877 void kfree_skb_list(struct sk_buff *segs); 878 void skb_tx_error(struct sk_buff *skb); 879 void consume_skb(struct sk_buff *skb); 880 void __kfree_skb(struct sk_buff *skb); 881 extern struct kmem_cache *skbuff_head_cache; 882 883 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 884 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 885 bool *fragstolen, int *delta_truesize); 886 887 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 888 int node); 889 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 890 struct sk_buff *build_skb(void *data, unsigned int frag_size); 891 static inline struct sk_buff *alloc_skb(unsigned int size, 892 gfp_t priority) 893 { 894 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 895 } 896 897 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 898 unsigned long data_len, 899 int max_page_order, 900 int *errcode, 901 gfp_t gfp_mask); 902 903 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 904 struct sk_buff_fclones { 905 struct sk_buff skb1; 906 907 struct sk_buff skb2; 908 909 atomic_t fclone_ref; 910 }; 911 912 /** 913 * skb_fclone_busy - check if fclone is busy 914 * @skb: buffer 915 * 916 * Returns true if skb is a fast clone, and its clone is not freed. 917 * Some drivers call skb_orphan() in their ndo_start_xmit(), 918 * so we also check that this didnt happen. 919 */ 920 static inline bool skb_fclone_busy(const struct sock *sk, 921 const struct sk_buff *skb) 922 { 923 const struct sk_buff_fclones *fclones; 924 925 fclones = container_of(skb, struct sk_buff_fclones, skb1); 926 927 return skb->fclone == SKB_FCLONE_ORIG && 928 atomic_read(&fclones->fclone_ref) > 1 && 929 fclones->skb2.sk == sk; 930 } 931 932 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 933 gfp_t priority) 934 { 935 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 936 } 937 938 struct sk_buff *__alloc_skb_head(gfp_t priority, int node); 939 static inline struct sk_buff *alloc_skb_head(gfp_t priority) 940 { 941 return __alloc_skb_head(priority, -1); 942 } 943 944 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 945 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 946 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 947 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 948 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 949 gfp_t gfp_mask, bool fclone); 950 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 951 gfp_t gfp_mask) 952 { 953 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 954 } 955 956 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 957 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 958 unsigned int headroom); 959 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 960 int newtailroom, gfp_t priority); 961 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 962 int offset, int len); 963 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, 964 int len); 965 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 966 int skb_pad(struct sk_buff *skb, int pad); 967 #define dev_kfree_skb(a) consume_skb(a) 968 969 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 970 int getfrag(void *from, char *to, int offset, 971 int len, int odd, struct sk_buff *skb), 972 void *from, int length); 973 974 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 975 int offset, size_t size); 976 977 struct skb_seq_state { 978 __u32 lower_offset; 979 __u32 upper_offset; 980 __u32 frag_idx; 981 __u32 stepped_offset; 982 struct sk_buff *root_skb; 983 struct sk_buff *cur_skb; 984 __u8 *frag_data; 985 }; 986 987 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 988 unsigned int to, struct skb_seq_state *st); 989 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 990 struct skb_seq_state *st); 991 void skb_abort_seq_read(struct skb_seq_state *st); 992 993 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 994 unsigned int to, struct ts_config *config); 995 996 /* 997 * Packet hash types specify the type of hash in skb_set_hash. 998 * 999 * Hash types refer to the protocol layer addresses which are used to 1000 * construct a packet's hash. The hashes are used to differentiate or identify 1001 * flows of the protocol layer for the hash type. Hash types are either 1002 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1003 * 1004 * Properties of hashes: 1005 * 1006 * 1) Two packets in different flows have different hash values 1007 * 2) Two packets in the same flow should have the same hash value 1008 * 1009 * A hash at a higher layer is considered to be more specific. A driver should 1010 * set the most specific hash possible. 1011 * 1012 * A driver cannot indicate a more specific hash than the layer at which a hash 1013 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1014 * 1015 * A driver may indicate a hash level which is less specific than the 1016 * actual layer the hash was computed on. For instance, a hash computed 1017 * at L4 may be considered an L3 hash. This should only be done if the 1018 * driver can't unambiguously determine that the HW computed the hash at 1019 * the higher layer. Note that the "should" in the second property above 1020 * permits this. 1021 */ 1022 enum pkt_hash_types { 1023 PKT_HASH_TYPE_NONE, /* Undefined type */ 1024 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1025 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1026 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1027 }; 1028 1029 static inline void skb_clear_hash(struct sk_buff *skb) 1030 { 1031 skb->hash = 0; 1032 skb->sw_hash = 0; 1033 skb->l4_hash = 0; 1034 } 1035 1036 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1037 { 1038 if (!skb->l4_hash) 1039 skb_clear_hash(skb); 1040 } 1041 1042 static inline void 1043 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1044 { 1045 skb->l4_hash = is_l4; 1046 skb->sw_hash = is_sw; 1047 skb->hash = hash; 1048 } 1049 1050 static inline void 1051 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1052 { 1053 /* Used by drivers to set hash from HW */ 1054 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1055 } 1056 1057 static inline void 1058 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1059 { 1060 __skb_set_hash(skb, hash, true, is_l4); 1061 } 1062 1063 void __skb_get_hash(struct sk_buff *skb); 1064 u32 skb_get_poff(const struct sk_buff *skb); 1065 u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1066 const struct flow_keys *keys, int hlen); 1067 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1068 void *data, int hlen_proto); 1069 1070 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1071 int thoff, u8 ip_proto) 1072 { 1073 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1074 } 1075 1076 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1077 const struct flow_dissector_key *key, 1078 unsigned int key_count); 1079 1080 bool __skb_flow_dissect(const struct sk_buff *skb, 1081 struct flow_dissector *flow_dissector, 1082 void *target_container, 1083 void *data, __be16 proto, int nhoff, int hlen, 1084 unsigned int flags); 1085 1086 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1087 struct flow_dissector *flow_dissector, 1088 void *target_container, unsigned int flags) 1089 { 1090 return __skb_flow_dissect(skb, flow_dissector, target_container, 1091 NULL, 0, 0, 0, flags); 1092 } 1093 1094 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1095 struct flow_keys *flow, 1096 unsigned int flags) 1097 { 1098 memset(flow, 0, sizeof(*flow)); 1099 return __skb_flow_dissect(skb, &flow_keys_dissector, flow, 1100 NULL, 0, 0, 0, flags); 1101 } 1102 1103 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow, 1104 void *data, __be16 proto, 1105 int nhoff, int hlen, 1106 unsigned int flags) 1107 { 1108 memset(flow, 0, sizeof(*flow)); 1109 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow, 1110 data, proto, nhoff, hlen, flags); 1111 } 1112 1113 static inline __u32 skb_get_hash(struct sk_buff *skb) 1114 { 1115 if (!skb->l4_hash && !skb->sw_hash) 1116 __skb_get_hash(skb); 1117 1118 return skb->hash; 1119 } 1120 1121 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6); 1122 1123 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1124 { 1125 if (!skb->l4_hash && !skb->sw_hash) { 1126 struct flow_keys keys; 1127 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1128 1129 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1130 } 1131 1132 return skb->hash; 1133 } 1134 1135 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl); 1136 1137 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4) 1138 { 1139 if (!skb->l4_hash && !skb->sw_hash) { 1140 struct flow_keys keys; 1141 __u32 hash = __get_hash_from_flowi4(fl4, &keys); 1142 1143 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1144 } 1145 1146 return skb->hash; 1147 } 1148 1149 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1150 1151 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1152 { 1153 return skb->hash; 1154 } 1155 1156 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1157 { 1158 to->hash = from->hash; 1159 to->sw_hash = from->sw_hash; 1160 to->l4_hash = from->l4_hash; 1161 }; 1162 1163 static inline void skb_sender_cpu_clear(struct sk_buff *skb) 1164 { 1165 } 1166 1167 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1168 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1169 { 1170 return skb->head + skb->end; 1171 } 1172 1173 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1174 { 1175 return skb->end; 1176 } 1177 #else 1178 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1179 { 1180 return skb->end; 1181 } 1182 1183 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1184 { 1185 return skb->end - skb->head; 1186 } 1187 #endif 1188 1189 /* Internal */ 1190 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1191 1192 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1193 { 1194 return &skb_shinfo(skb)->hwtstamps; 1195 } 1196 1197 /** 1198 * skb_queue_empty - check if a queue is empty 1199 * @list: queue head 1200 * 1201 * Returns true if the queue is empty, false otherwise. 1202 */ 1203 static inline int skb_queue_empty(const struct sk_buff_head *list) 1204 { 1205 return list->next == (const struct sk_buff *) list; 1206 } 1207 1208 /** 1209 * skb_queue_is_last - check if skb is the last entry in the queue 1210 * @list: queue head 1211 * @skb: buffer 1212 * 1213 * Returns true if @skb is the last buffer on the list. 1214 */ 1215 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1216 const struct sk_buff *skb) 1217 { 1218 return skb->next == (const struct sk_buff *) list; 1219 } 1220 1221 /** 1222 * skb_queue_is_first - check if skb is the first entry in the queue 1223 * @list: queue head 1224 * @skb: buffer 1225 * 1226 * Returns true if @skb is the first buffer on the list. 1227 */ 1228 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1229 const struct sk_buff *skb) 1230 { 1231 return skb->prev == (const struct sk_buff *) list; 1232 } 1233 1234 /** 1235 * skb_queue_next - return the next packet in the queue 1236 * @list: queue head 1237 * @skb: current buffer 1238 * 1239 * Return the next packet in @list after @skb. It is only valid to 1240 * call this if skb_queue_is_last() evaluates to false. 1241 */ 1242 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1243 const struct sk_buff *skb) 1244 { 1245 /* This BUG_ON may seem severe, but if we just return then we 1246 * are going to dereference garbage. 1247 */ 1248 BUG_ON(skb_queue_is_last(list, skb)); 1249 return skb->next; 1250 } 1251 1252 /** 1253 * skb_queue_prev - return the prev packet in the queue 1254 * @list: queue head 1255 * @skb: current buffer 1256 * 1257 * Return the prev packet in @list before @skb. It is only valid to 1258 * call this if skb_queue_is_first() evaluates to false. 1259 */ 1260 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1261 const struct sk_buff *skb) 1262 { 1263 /* This BUG_ON may seem severe, but if we just return then we 1264 * are going to dereference garbage. 1265 */ 1266 BUG_ON(skb_queue_is_first(list, skb)); 1267 return skb->prev; 1268 } 1269 1270 /** 1271 * skb_get - reference buffer 1272 * @skb: buffer to reference 1273 * 1274 * Makes another reference to a socket buffer and returns a pointer 1275 * to the buffer. 1276 */ 1277 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1278 { 1279 atomic_inc(&skb->users); 1280 return skb; 1281 } 1282 1283 /* 1284 * If users == 1, we are the only owner and are can avoid redundant 1285 * atomic change. 1286 */ 1287 1288 /** 1289 * skb_cloned - is the buffer a clone 1290 * @skb: buffer to check 1291 * 1292 * Returns true if the buffer was generated with skb_clone() and is 1293 * one of multiple shared copies of the buffer. Cloned buffers are 1294 * shared data so must not be written to under normal circumstances. 1295 */ 1296 static inline int skb_cloned(const struct sk_buff *skb) 1297 { 1298 return skb->cloned && 1299 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1300 } 1301 1302 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1303 { 1304 might_sleep_if(gfpflags_allow_blocking(pri)); 1305 1306 if (skb_cloned(skb)) 1307 return pskb_expand_head(skb, 0, 0, pri); 1308 1309 return 0; 1310 } 1311 1312 /** 1313 * skb_header_cloned - is the header a clone 1314 * @skb: buffer to check 1315 * 1316 * Returns true if modifying the header part of the buffer requires 1317 * the data to be copied. 1318 */ 1319 static inline int skb_header_cloned(const struct sk_buff *skb) 1320 { 1321 int dataref; 1322 1323 if (!skb->cloned) 1324 return 0; 1325 1326 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1327 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1328 return dataref != 1; 1329 } 1330 1331 /** 1332 * skb_header_release - release reference to header 1333 * @skb: buffer to operate on 1334 * 1335 * Drop a reference to the header part of the buffer. This is done 1336 * by acquiring a payload reference. You must not read from the header 1337 * part of skb->data after this. 1338 * Note : Check if you can use __skb_header_release() instead. 1339 */ 1340 static inline void skb_header_release(struct sk_buff *skb) 1341 { 1342 BUG_ON(skb->nohdr); 1343 skb->nohdr = 1; 1344 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1345 } 1346 1347 /** 1348 * __skb_header_release - release reference to header 1349 * @skb: buffer to operate on 1350 * 1351 * Variant of skb_header_release() assuming skb is private to caller. 1352 * We can avoid one atomic operation. 1353 */ 1354 static inline void __skb_header_release(struct sk_buff *skb) 1355 { 1356 skb->nohdr = 1; 1357 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1358 } 1359 1360 1361 /** 1362 * skb_shared - is the buffer shared 1363 * @skb: buffer to check 1364 * 1365 * Returns true if more than one person has a reference to this 1366 * buffer. 1367 */ 1368 static inline int skb_shared(const struct sk_buff *skb) 1369 { 1370 return atomic_read(&skb->users) != 1; 1371 } 1372 1373 /** 1374 * skb_share_check - check if buffer is shared and if so clone it 1375 * @skb: buffer to check 1376 * @pri: priority for memory allocation 1377 * 1378 * If the buffer is shared the buffer is cloned and the old copy 1379 * drops a reference. A new clone with a single reference is returned. 1380 * If the buffer is not shared the original buffer is returned. When 1381 * being called from interrupt status or with spinlocks held pri must 1382 * be GFP_ATOMIC. 1383 * 1384 * NULL is returned on a memory allocation failure. 1385 */ 1386 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1387 { 1388 might_sleep_if(gfpflags_allow_blocking(pri)); 1389 if (skb_shared(skb)) { 1390 struct sk_buff *nskb = skb_clone(skb, pri); 1391 1392 if (likely(nskb)) 1393 consume_skb(skb); 1394 else 1395 kfree_skb(skb); 1396 skb = nskb; 1397 } 1398 return skb; 1399 } 1400 1401 /* 1402 * Copy shared buffers into a new sk_buff. We effectively do COW on 1403 * packets to handle cases where we have a local reader and forward 1404 * and a couple of other messy ones. The normal one is tcpdumping 1405 * a packet thats being forwarded. 1406 */ 1407 1408 /** 1409 * skb_unshare - make a copy of a shared buffer 1410 * @skb: buffer to check 1411 * @pri: priority for memory allocation 1412 * 1413 * If the socket buffer is a clone then this function creates a new 1414 * copy of the data, drops a reference count on the old copy and returns 1415 * the new copy with the reference count at 1. If the buffer is not a clone 1416 * the original buffer is returned. When called with a spinlock held or 1417 * from interrupt state @pri must be %GFP_ATOMIC 1418 * 1419 * %NULL is returned on a memory allocation failure. 1420 */ 1421 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1422 gfp_t pri) 1423 { 1424 might_sleep_if(gfpflags_allow_blocking(pri)); 1425 if (skb_cloned(skb)) { 1426 struct sk_buff *nskb = skb_copy(skb, pri); 1427 1428 /* Free our shared copy */ 1429 if (likely(nskb)) 1430 consume_skb(skb); 1431 else 1432 kfree_skb(skb); 1433 skb = nskb; 1434 } 1435 return skb; 1436 } 1437 1438 /** 1439 * skb_peek - peek at the head of an &sk_buff_head 1440 * @list_: list to peek at 1441 * 1442 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1443 * be careful with this one. A peek leaves the buffer on the 1444 * list and someone else may run off with it. You must hold 1445 * the appropriate locks or have a private queue to do this. 1446 * 1447 * Returns %NULL for an empty list or a pointer to the head element. 1448 * The reference count is not incremented and the reference is therefore 1449 * volatile. Use with caution. 1450 */ 1451 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1452 { 1453 struct sk_buff *skb = list_->next; 1454 1455 if (skb == (struct sk_buff *)list_) 1456 skb = NULL; 1457 return skb; 1458 } 1459 1460 /** 1461 * skb_peek_next - peek skb following the given one from a queue 1462 * @skb: skb to start from 1463 * @list_: list to peek at 1464 * 1465 * Returns %NULL when the end of the list is met or a pointer to the 1466 * next element. The reference count is not incremented and the 1467 * reference is therefore volatile. Use with caution. 1468 */ 1469 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1470 const struct sk_buff_head *list_) 1471 { 1472 struct sk_buff *next = skb->next; 1473 1474 if (next == (struct sk_buff *)list_) 1475 next = NULL; 1476 return next; 1477 } 1478 1479 /** 1480 * skb_peek_tail - peek at the tail of an &sk_buff_head 1481 * @list_: list to peek at 1482 * 1483 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1484 * be careful with this one. A peek leaves the buffer on the 1485 * list and someone else may run off with it. You must hold 1486 * the appropriate locks or have a private queue to do this. 1487 * 1488 * Returns %NULL for an empty list or a pointer to the tail element. 1489 * The reference count is not incremented and the reference is therefore 1490 * volatile. Use with caution. 1491 */ 1492 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1493 { 1494 struct sk_buff *skb = list_->prev; 1495 1496 if (skb == (struct sk_buff *)list_) 1497 skb = NULL; 1498 return skb; 1499 1500 } 1501 1502 /** 1503 * skb_queue_len - get queue length 1504 * @list_: list to measure 1505 * 1506 * Return the length of an &sk_buff queue. 1507 */ 1508 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1509 { 1510 return list_->qlen; 1511 } 1512 1513 /** 1514 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1515 * @list: queue to initialize 1516 * 1517 * This initializes only the list and queue length aspects of 1518 * an sk_buff_head object. This allows to initialize the list 1519 * aspects of an sk_buff_head without reinitializing things like 1520 * the spinlock. It can also be used for on-stack sk_buff_head 1521 * objects where the spinlock is known to not be used. 1522 */ 1523 static inline void __skb_queue_head_init(struct sk_buff_head *list) 1524 { 1525 list->prev = list->next = (struct sk_buff *)list; 1526 list->qlen = 0; 1527 } 1528 1529 /* 1530 * This function creates a split out lock class for each invocation; 1531 * this is needed for now since a whole lot of users of the skb-queue 1532 * infrastructure in drivers have different locking usage (in hardirq) 1533 * than the networking core (in softirq only). In the long run either the 1534 * network layer or drivers should need annotation to consolidate the 1535 * main types of usage into 3 classes. 1536 */ 1537 static inline void skb_queue_head_init(struct sk_buff_head *list) 1538 { 1539 spin_lock_init(&list->lock); 1540 __skb_queue_head_init(list); 1541 } 1542 1543 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1544 struct lock_class_key *class) 1545 { 1546 skb_queue_head_init(list); 1547 lockdep_set_class(&list->lock, class); 1548 } 1549 1550 /* 1551 * Insert an sk_buff on a list. 1552 * 1553 * The "__skb_xxxx()" functions are the non-atomic ones that 1554 * can only be called with interrupts disabled. 1555 */ 1556 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1557 struct sk_buff_head *list); 1558 static inline void __skb_insert(struct sk_buff *newsk, 1559 struct sk_buff *prev, struct sk_buff *next, 1560 struct sk_buff_head *list) 1561 { 1562 newsk->next = next; 1563 newsk->prev = prev; 1564 next->prev = prev->next = newsk; 1565 list->qlen++; 1566 } 1567 1568 static inline void __skb_queue_splice(const struct sk_buff_head *list, 1569 struct sk_buff *prev, 1570 struct sk_buff *next) 1571 { 1572 struct sk_buff *first = list->next; 1573 struct sk_buff *last = list->prev; 1574 1575 first->prev = prev; 1576 prev->next = first; 1577 1578 last->next = next; 1579 next->prev = last; 1580 } 1581 1582 /** 1583 * skb_queue_splice - join two skb lists, this is designed for stacks 1584 * @list: the new list to add 1585 * @head: the place to add it in the first list 1586 */ 1587 static inline void skb_queue_splice(const struct sk_buff_head *list, 1588 struct sk_buff_head *head) 1589 { 1590 if (!skb_queue_empty(list)) { 1591 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1592 head->qlen += list->qlen; 1593 } 1594 } 1595 1596 /** 1597 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1598 * @list: the new list to add 1599 * @head: the place to add it in the first list 1600 * 1601 * The list at @list is reinitialised 1602 */ 1603 static inline void skb_queue_splice_init(struct sk_buff_head *list, 1604 struct sk_buff_head *head) 1605 { 1606 if (!skb_queue_empty(list)) { 1607 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1608 head->qlen += list->qlen; 1609 __skb_queue_head_init(list); 1610 } 1611 } 1612 1613 /** 1614 * skb_queue_splice_tail - join two skb lists, each list being a queue 1615 * @list: the new list to add 1616 * @head: the place to add it in the first list 1617 */ 1618 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1619 struct sk_buff_head *head) 1620 { 1621 if (!skb_queue_empty(list)) { 1622 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1623 head->qlen += list->qlen; 1624 } 1625 } 1626 1627 /** 1628 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1629 * @list: the new list to add 1630 * @head: the place to add it in the first list 1631 * 1632 * Each of the lists is a queue. 1633 * The list at @list is reinitialised 1634 */ 1635 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1636 struct sk_buff_head *head) 1637 { 1638 if (!skb_queue_empty(list)) { 1639 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1640 head->qlen += list->qlen; 1641 __skb_queue_head_init(list); 1642 } 1643 } 1644 1645 /** 1646 * __skb_queue_after - queue a buffer at the list head 1647 * @list: list to use 1648 * @prev: place after this buffer 1649 * @newsk: buffer to queue 1650 * 1651 * Queue a buffer int the middle of a list. This function takes no locks 1652 * and you must therefore hold required locks before calling it. 1653 * 1654 * A buffer cannot be placed on two lists at the same time. 1655 */ 1656 static inline void __skb_queue_after(struct sk_buff_head *list, 1657 struct sk_buff *prev, 1658 struct sk_buff *newsk) 1659 { 1660 __skb_insert(newsk, prev, prev->next, list); 1661 } 1662 1663 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1664 struct sk_buff_head *list); 1665 1666 static inline void __skb_queue_before(struct sk_buff_head *list, 1667 struct sk_buff *next, 1668 struct sk_buff *newsk) 1669 { 1670 __skb_insert(newsk, next->prev, next, list); 1671 } 1672 1673 /** 1674 * __skb_queue_head - queue a buffer at the list head 1675 * @list: list to use 1676 * @newsk: buffer to queue 1677 * 1678 * Queue a buffer at the start of a list. This function takes no locks 1679 * and you must therefore hold required locks before calling it. 1680 * 1681 * A buffer cannot be placed on two lists at the same time. 1682 */ 1683 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1684 static inline void __skb_queue_head(struct sk_buff_head *list, 1685 struct sk_buff *newsk) 1686 { 1687 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1688 } 1689 1690 /** 1691 * __skb_queue_tail - queue a buffer at the list tail 1692 * @list: list to use 1693 * @newsk: buffer to queue 1694 * 1695 * Queue a buffer at the end of a list. This function takes no locks 1696 * and you must therefore hold required locks before calling it. 1697 * 1698 * A buffer cannot be placed on two lists at the same time. 1699 */ 1700 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1701 static inline void __skb_queue_tail(struct sk_buff_head *list, 1702 struct sk_buff *newsk) 1703 { 1704 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1705 } 1706 1707 /* 1708 * remove sk_buff from list. _Must_ be called atomically, and with 1709 * the list known.. 1710 */ 1711 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1712 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1713 { 1714 struct sk_buff *next, *prev; 1715 1716 list->qlen--; 1717 next = skb->next; 1718 prev = skb->prev; 1719 skb->next = skb->prev = NULL; 1720 next->prev = prev; 1721 prev->next = next; 1722 } 1723 1724 /** 1725 * __skb_dequeue - remove from the head of the queue 1726 * @list: list to dequeue from 1727 * 1728 * Remove the head of the list. This function does not take any locks 1729 * so must be used with appropriate locks held only. The head item is 1730 * returned or %NULL if the list is empty. 1731 */ 1732 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1733 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1734 { 1735 struct sk_buff *skb = skb_peek(list); 1736 if (skb) 1737 __skb_unlink(skb, list); 1738 return skb; 1739 } 1740 1741 /** 1742 * __skb_dequeue_tail - remove from the tail of the queue 1743 * @list: list to dequeue from 1744 * 1745 * Remove the tail of the list. This function does not take any locks 1746 * so must be used with appropriate locks held only. The tail item is 1747 * returned or %NULL if the list is empty. 1748 */ 1749 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1750 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1751 { 1752 struct sk_buff *skb = skb_peek_tail(list); 1753 if (skb) 1754 __skb_unlink(skb, list); 1755 return skb; 1756 } 1757 1758 1759 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1760 { 1761 return skb->data_len; 1762 } 1763 1764 static inline unsigned int skb_headlen(const struct sk_buff *skb) 1765 { 1766 return skb->len - skb->data_len; 1767 } 1768 1769 static inline int skb_pagelen(const struct sk_buff *skb) 1770 { 1771 int i, len = 0; 1772 1773 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--) 1774 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1775 return len + skb_headlen(skb); 1776 } 1777 1778 /** 1779 * __skb_fill_page_desc - initialise a paged fragment in an skb 1780 * @skb: buffer containing fragment to be initialised 1781 * @i: paged fragment index to initialise 1782 * @page: the page to use for this fragment 1783 * @off: the offset to the data with @page 1784 * @size: the length of the data 1785 * 1786 * Initialises the @i'th fragment of @skb to point to &size bytes at 1787 * offset @off within @page. 1788 * 1789 * Does not take any additional reference on the fragment. 1790 */ 1791 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1792 struct page *page, int off, int size) 1793 { 1794 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1795 1796 /* 1797 * Propagate page pfmemalloc to the skb if we can. The problem is 1798 * that not all callers have unique ownership of the page but rely 1799 * on page_is_pfmemalloc doing the right thing(tm). 1800 */ 1801 frag->page.p = page; 1802 frag->page_offset = off; 1803 skb_frag_size_set(frag, size); 1804 1805 page = compound_head(page); 1806 if (page_is_pfmemalloc(page)) 1807 skb->pfmemalloc = true; 1808 } 1809 1810 /** 1811 * skb_fill_page_desc - initialise a paged fragment in an skb 1812 * @skb: buffer containing fragment to be initialised 1813 * @i: paged fragment index to initialise 1814 * @page: the page to use for this fragment 1815 * @off: the offset to the data with @page 1816 * @size: the length of the data 1817 * 1818 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1819 * @skb to point to @size bytes at offset @off within @page. In 1820 * addition updates @skb such that @i is the last fragment. 1821 * 1822 * Does not take any additional reference on the fragment. 1823 */ 1824 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1825 struct page *page, int off, int size) 1826 { 1827 __skb_fill_page_desc(skb, i, page, off, size); 1828 skb_shinfo(skb)->nr_frags = i + 1; 1829 } 1830 1831 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1832 int size, unsigned int truesize); 1833 1834 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1835 unsigned int truesize); 1836 1837 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1838 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1839 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1840 1841 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1842 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1843 { 1844 return skb->head + skb->tail; 1845 } 1846 1847 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1848 { 1849 skb->tail = skb->data - skb->head; 1850 } 1851 1852 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1853 { 1854 skb_reset_tail_pointer(skb); 1855 skb->tail += offset; 1856 } 1857 1858 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 1859 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1860 { 1861 return skb->tail; 1862 } 1863 1864 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1865 { 1866 skb->tail = skb->data; 1867 } 1868 1869 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1870 { 1871 skb->tail = skb->data + offset; 1872 } 1873 1874 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1875 1876 /* 1877 * Add data to an sk_buff 1878 */ 1879 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1880 unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1881 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1882 { 1883 unsigned char *tmp = skb_tail_pointer(skb); 1884 SKB_LINEAR_ASSERT(skb); 1885 skb->tail += len; 1886 skb->len += len; 1887 return tmp; 1888 } 1889 1890 unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1891 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1892 { 1893 skb->data -= len; 1894 skb->len += len; 1895 return skb->data; 1896 } 1897 1898 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1899 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1900 { 1901 skb->len -= len; 1902 BUG_ON(skb->len < skb->data_len); 1903 return skb->data += len; 1904 } 1905 1906 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1907 { 1908 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1909 } 1910 1911 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1912 1913 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1914 { 1915 if (len > skb_headlen(skb) && 1916 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1917 return NULL; 1918 skb->len -= len; 1919 return skb->data += len; 1920 } 1921 1922 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1923 { 1924 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1925 } 1926 1927 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1928 { 1929 if (likely(len <= skb_headlen(skb))) 1930 return 1; 1931 if (unlikely(len > skb->len)) 1932 return 0; 1933 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1934 } 1935 1936 /** 1937 * skb_headroom - bytes at buffer head 1938 * @skb: buffer to check 1939 * 1940 * Return the number of bytes of free space at the head of an &sk_buff. 1941 */ 1942 static inline unsigned int skb_headroom(const struct sk_buff *skb) 1943 { 1944 return skb->data - skb->head; 1945 } 1946 1947 /** 1948 * skb_tailroom - bytes at buffer end 1949 * @skb: buffer to check 1950 * 1951 * Return the number of bytes of free space at the tail of an sk_buff 1952 */ 1953 static inline int skb_tailroom(const struct sk_buff *skb) 1954 { 1955 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 1956 } 1957 1958 /** 1959 * skb_availroom - bytes at buffer end 1960 * @skb: buffer to check 1961 * 1962 * Return the number of bytes of free space at the tail of an sk_buff 1963 * allocated by sk_stream_alloc() 1964 */ 1965 static inline int skb_availroom(const struct sk_buff *skb) 1966 { 1967 if (skb_is_nonlinear(skb)) 1968 return 0; 1969 1970 return skb->end - skb->tail - skb->reserved_tailroom; 1971 } 1972 1973 /** 1974 * skb_reserve - adjust headroom 1975 * @skb: buffer to alter 1976 * @len: bytes to move 1977 * 1978 * Increase the headroom of an empty &sk_buff by reducing the tail 1979 * room. This is only allowed for an empty buffer. 1980 */ 1981 static inline void skb_reserve(struct sk_buff *skb, int len) 1982 { 1983 skb->data += len; 1984 skb->tail += len; 1985 } 1986 1987 #define ENCAP_TYPE_ETHER 0 1988 #define ENCAP_TYPE_IPPROTO 1 1989 1990 static inline void skb_set_inner_protocol(struct sk_buff *skb, 1991 __be16 protocol) 1992 { 1993 skb->inner_protocol = protocol; 1994 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 1995 } 1996 1997 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 1998 __u8 ipproto) 1999 { 2000 skb->inner_ipproto = ipproto; 2001 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2002 } 2003 2004 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2005 { 2006 skb->inner_mac_header = skb->mac_header; 2007 skb->inner_network_header = skb->network_header; 2008 skb->inner_transport_header = skb->transport_header; 2009 } 2010 2011 static inline void skb_reset_mac_len(struct sk_buff *skb) 2012 { 2013 skb->mac_len = skb->network_header - skb->mac_header; 2014 } 2015 2016 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2017 *skb) 2018 { 2019 return skb->head + skb->inner_transport_header; 2020 } 2021 2022 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2023 { 2024 return skb_inner_transport_header(skb) - skb->data; 2025 } 2026 2027 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2028 { 2029 skb->inner_transport_header = skb->data - skb->head; 2030 } 2031 2032 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2033 const int offset) 2034 { 2035 skb_reset_inner_transport_header(skb); 2036 skb->inner_transport_header += offset; 2037 } 2038 2039 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2040 { 2041 return skb->head + skb->inner_network_header; 2042 } 2043 2044 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2045 { 2046 skb->inner_network_header = skb->data - skb->head; 2047 } 2048 2049 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2050 const int offset) 2051 { 2052 skb_reset_inner_network_header(skb); 2053 skb->inner_network_header += offset; 2054 } 2055 2056 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2057 { 2058 return skb->head + skb->inner_mac_header; 2059 } 2060 2061 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2062 { 2063 skb->inner_mac_header = skb->data - skb->head; 2064 } 2065 2066 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2067 const int offset) 2068 { 2069 skb_reset_inner_mac_header(skb); 2070 skb->inner_mac_header += offset; 2071 } 2072 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2073 { 2074 return skb->transport_header != (typeof(skb->transport_header))~0U; 2075 } 2076 2077 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2078 { 2079 return skb->head + skb->transport_header; 2080 } 2081 2082 static inline void skb_reset_transport_header(struct sk_buff *skb) 2083 { 2084 skb->transport_header = skb->data - skb->head; 2085 } 2086 2087 static inline void skb_set_transport_header(struct sk_buff *skb, 2088 const int offset) 2089 { 2090 skb_reset_transport_header(skb); 2091 skb->transport_header += offset; 2092 } 2093 2094 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2095 { 2096 return skb->head + skb->network_header; 2097 } 2098 2099 static inline void skb_reset_network_header(struct sk_buff *skb) 2100 { 2101 skb->network_header = skb->data - skb->head; 2102 } 2103 2104 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2105 { 2106 skb_reset_network_header(skb); 2107 skb->network_header += offset; 2108 } 2109 2110 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2111 { 2112 return skb->head + skb->mac_header; 2113 } 2114 2115 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2116 { 2117 return skb->mac_header != (typeof(skb->mac_header))~0U; 2118 } 2119 2120 static inline void skb_reset_mac_header(struct sk_buff *skb) 2121 { 2122 skb->mac_header = skb->data - skb->head; 2123 } 2124 2125 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2126 { 2127 skb_reset_mac_header(skb); 2128 skb->mac_header += offset; 2129 } 2130 2131 static inline void skb_pop_mac_header(struct sk_buff *skb) 2132 { 2133 skb->mac_header = skb->network_header; 2134 } 2135 2136 static inline void skb_probe_transport_header(struct sk_buff *skb, 2137 const int offset_hint) 2138 { 2139 struct flow_keys keys; 2140 2141 if (skb_transport_header_was_set(skb)) 2142 return; 2143 else if (skb_flow_dissect_flow_keys(skb, &keys, 0)) 2144 skb_set_transport_header(skb, keys.control.thoff); 2145 else 2146 skb_set_transport_header(skb, offset_hint); 2147 } 2148 2149 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2150 { 2151 if (skb_mac_header_was_set(skb)) { 2152 const unsigned char *old_mac = skb_mac_header(skb); 2153 2154 skb_set_mac_header(skb, -skb->mac_len); 2155 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2156 } 2157 } 2158 2159 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2160 { 2161 return skb->csum_start - skb_headroom(skb); 2162 } 2163 2164 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2165 { 2166 return skb->head + skb->csum_start; 2167 } 2168 2169 static inline int skb_transport_offset(const struct sk_buff *skb) 2170 { 2171 return skb_transport_header(skb) - skb->data; 2172 } 2173 2174 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2175 { 2176 return skb->transport_header - skb->network_header; 2177 } 2178 2179 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2180 { 2181 return skb->inner_transport_header - skb->inner_network_header; 2182 } 2183 2184 static inline int skb_network_offset(const struct sk_buff *skb) 2185 { 2186 return skb_network_header(skb) - skb->data; 2187 } 2188 2189 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2190 { 2191 return skb_inner_network_header(skb) - skb->data; 2192 } 2193 2194 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2195 { 2196 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2197 } 2198 2199 /* 2200 * CPUs often take a performance hit when accessing unaligned memory 2201 * locations. The actual performance hit varies, it can be small if the 2202 * hardware handles it or large if we have to take an exception and fix it 2203 * in software. 2204 * 2205 * Since an ethernet header is 14 bytes network drivers often end up with 2206 * the IP header at an unaligned offset. The IP header can be aligned by 2207 * shifting the start of the packet by 2 bytes. Drivers should do this 2208 * with: 2209 * 2210 * skb_reserve(skb, NET_IP_ALIGN); 2211 * 2212 * The downside to this alignment of the IP header is that the DMA is now 2213 * unaligned. On some architectures the cost of an unaligned DMA is high 2214 * and this cost outweighs the gains made by aligning the IP header. 2215 * 2216 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2217 * to be overridden. 2218 */ 2219 #ifndef NET_IP_ALIGN 2220 #define NET_IP_ALIGN 2 2221 #endif 2222 2223 /* 2224 * The networking layer reserves some headroom in skb data (via 2225 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2226 * the header has to grow. In the default case, if the header has to grow 2227 * 32 bytes or less we avoid the reallocation. 2228 * 2229 * Unfortunately this headroom changes the DMA alignment of the resulting 2230 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2231 * on some architectures. An architecture can override this value, 2232 * perhaps setting it to a cacheline in size (since that will maintain 2233 * cacheline alignment of the DMA). It must be a power of 2. 2234 * 2235 * Various parts of the networking layer expect at least 32 bytes of 2236 * headroom, you should not reduce this. 2237 * 2238 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2239 * to reduce average number of cache lines per packet. 2240 * get_rps_cpus() for example only access one 64 bytes aligned block : 2241 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2242 */ 2243 #ifndef NET_SKB_PAD 2244 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2245 #endif 2246 2247 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2248 2249 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2250 { 2251 if (unlikely(skb_is_nonlinear(skb))) { 2252 WARN_ON(1); 2253 return; 2254 } 2255 skb->len = len; 2256 skb_set_tail_pointer(skb, len); 2257 } 2258 2259 void skb_trim(struct sk_buff *skb, unsigned int len); 2260 2261 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2262 { 2263 if (skb->data_len) 2264 return ___pskb_trim(skb, len); 2265 __skb_trim(skb, len); 2266 return 0; 2267 } 2268 2269 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2270 { 2271 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2272 } 2273 2274 /** 2275 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2276 * @skb: buffer to alter 2277 * @len: new length 2278 * 2279 * This is identical to pskb_trim except that the caller knows that 2280 * the skb is not cloned so we should never get an error due to out- 2281 * of-memory. 2282 */ 2283 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2284 { 2285 int err = pskb_trim(skb, len); 2286 BUG_ON(err); 2287 } 2288 2289 /** 2290 * skb_orphan - orphan a buffer 2291 * @skb: buffer to orphan 2292 * 2293 * If a buffer currently has an owner then we call the owner's 2294 * destructor function and make the @skb unowned. The buffer continues 2295 * to exist but is no longer charged to its former owner. 2296 */ 2297 static inline void skb_orphan(struct sk_buff *skb) 2298 { 2299 if (skb->destructor) { 2300 skb->destructor(skb); 2301 skb->destructor = NULL; 2302 skb->sk = NULL; 2303 } else { 2304 BUG_ON(skb->sk); 2305 } 2306 } 2307 2308 /** 2309 * skb_orphan_frags - orphan the frags contained in a buffer 2310 * @skb: buffer to orphan frags from 2311 * @gfp_mask: allocation mask for replacement pages 2312 * 2313 * For each frag in the SKB which needs a destructor (i.e. has an 2314 * owner) create a copy of that frag and release the original 2315 * page by calling the destructor. 2316 */ 2317 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2318 { 2319 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 2320 return 0; 2321 return skb_copy_ubufs(skb, gfp_mask); 2322 } 2323 2324 /** 2325 * __skb_queue_purge - empty a list 2326 * @list: list to empty 2327 * 2328 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2329 * the list and one reference dropped. This function does not take the 2330 * list lock and the caller must hold the relevant locks to use it. 2331 */ 2332 void skb_queue_purge(struct sk_buff_head *list); 2333 static inline void __skb_queue_purge(struct sk_buff_head *list) 2334 { 2335 struct sk_buff *skb; 2336 while ((skb = __skb_dequeue(list)) != NULL) 2337 kfree_skb(skb); 2338 } 2339 2340 void *netdev_alloc_frag(unsigned int fragsz); 2341 2342 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2343 gfp_t gfp_mask); 2344 2345 /** 2346 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2347 * @dev: network device to receive on 2348 * @length: length to allocate 2349 * 2350 * Allocate a new &sk_buff and assign it a usage count of one. The 2351 * buffer has unspecified headroom built in. Users should allocate 2352 * the headroom they think they need without accounting for the 2353 * built in space. The built in space is used for optimisations. 2354 * 2355 * %NULL is returned if there is no free memory. Although this function 2356 * allocates memory it can be called from an interrupt. 2357 */ 2358 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2359 unsigned int length) 2360 { 2361 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2362 } 2363 2364 /* legacy helper around __netdev_alloc_skb() */ 2365 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2366 gfp_t gfp_mask) 2367 { 2368 return __netdev_alloc_skb(NULL, length, gfp_mask); 2369 } 2370 2371 /* legacy helper around netdev_alloc_skb() */ 2372 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2373 { 2374 return netdev_alloc_skb(NULL, length); 2375 } 2376 2377 2378 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2379 unsigned int length, gfp_t gfp) 2380 { 2381 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2382 2383 if (NET_IP_ALIGN && skb) 2384 skb_reserve(skb, NET_IP_ALIGN); 2385 return skb; 2386 } 2387 2388 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2389 unsigned int length) 2390 { 2391 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2392 } 2393 2394 static inline void skb_free_frag(void *addr) 2395 { 2396 __free_page_frag(addr); 2397 } 2398 2399 void *napi_alloc_frag(unsigned int fragsz); 2400 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2401 unsigned int length, gfp_t gfp_mask); 2402 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2403 unsigned int length) 2404 { 2405 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2406 } 2407 void napi_consume_skb(struct sk_buff *skb, int budget); 2408 2409 void __kfree_skb_flush(void); 2410 void __kfree_skb_defer(struct sk_buff *skb); 2411 2412 /** 2413 * __dev_alloc_pages - allocate page for network Rx 2414 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2415 * @order: size of the allocation 2416 * 2417 * Allocate a new page. 2418 * 2419 * %NULL is returned if there is no free memory. 2420 */ 2421 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2422 unsigned int order) 2423 { 2424 /* This piece of code contains several assumptions. 2425 * 1. This is for device Rx, therefor a cold page is preferred. 2426 * 2. The expectation is the user wants a compound page. 2427 * 3. If requesting a order 0 page it will not be compound 2428 * due to the check to see if order has a value in prep_new_page 2429 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2430 * code in gfp_to_alloc_flags that should be enforcing this. 2431 */ 2432 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2433 2434 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2435 } 2436 2437 static inline struct page *dev_alloc_pages(unsigned int order) 2438 { 2439 return __dev_alloc_pages(GFP_ATOMIC, order); 2440 } 2441 2442 /** 2443 * __dev_alloc_page - allocate a page for network Rx 2444 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2445 * 2446 * Allocate a new page. 2447 * 2448 * %NULL is returned if there is no free memory. 2449 */ 2450 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2451 { 2452 return __dev_alloc_pages(gfp_mask, 0); 2453 } 2454 2455 static inline struct page *dev_alloc_page(void) 2456 { 2457 return __dev_alloc_page(GFP_ATOMIC); 2458 } 2459 2460 /** 2461 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2462 * @page: The page that was allocated from skb_alloc_page 2463 * @skb: The skb that may need pfmemalloc set 2464 */ 2465 static inline void skb_propagate_pfmemalloc(struct page *page, 2466 struct sk_buff *skb) 2467 { 2468 if (page_is_pfmemalloc(page)) 2469 skb->pfmemalloc = true; 2470 } 2471 2472 /** 2473 * skb_frag_page - retrieve the page referred to by a paged fragment 2474 * @frag: the paged fragment 2475 * 2476 * Returns the &struct page associated with @frag. 2477 */ 2478 static inline struct page *skb_frag_page(const skb_frag_t *frag) 2479 { 2480 return frag->page.p; 2481 } 2482 2483 /** 2484 * __skb_frag_ref - take an addition reference on a paged fragment. 2485 * @frag: the paged fragment 2486 * 2487 * Takes an additional reference on the paged fragment @frag. 2488 */ 2489 static inline void __skb_frag_ref(skb_frag_t *frag) 2490 { 2491 get_page(skb_frag_page(frag)); 2492 } 2493 2494 /** 2495 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2496 * @skb: the buffer 2497 * @f: the fragment offset. 2498 * 2499 * Takes an additional reference on the @f'th paged fragment of @skb. 2500 */ 2501 static inline void skb_frag_ref(struct sk_buff *skb, int f) 2502 { 2503 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2504 } 2505 2506 /** 2507 * __skb_frag_unref - release a reference on a paged fragment. 2508 * @frag: the paged fragment 2509 * 2510 * Releases a reference on the paged fragment @frag. 2511 */ 2512 static inline void __skb_frag_unref(skb_frag_t *frag) 2513 { 2514 put_page(skb_frag_page(frag)); 2515 } 2516 2517 /** 2518 * skb_frag_unref - release a reference on a paged fragment of an skb. 2519 * @skb: the buffer 2520 * @f: the fragment offset 2521 * 2522 * Releases a reference on the @f'th paged fragment of @skb. 2523 */ 2524 static inline void skb_frag_unref(struct sk_buff *skb, int f) 2525 { 2526 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2527 } 2528 2529 /** 2530 * skb_frag_address - gets the address of the data contained in a paged fragment 2531 * @frag: the paged fragment buffer 2532 * 2533 * Returns the address of the data within @frag. The page must already 2534 * be mapped. 2535 */ 2536 static inline void *skb_frag_address(const skb_frag_t *frag) 2537 { 2538 return page_address(skb_frag_page(frag)) + frag->page_offset; 2539 } 2540 2541 /** 2542 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2543 * @frag: the paged fragment buffer 2544 * 2545 * Returns the address of the data within @frag. Checks that the page 2546 * is mapped and returns %NULL otherwise. 2547 */ 2548 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2549 { 2550 void *ptr = page_address(skb_frag_page(frag)); 2551 if (unlikely(!ptr)) 2552 return NULL; 2553 2554 return ptr + frag->page_offset; 2555 } 2556 2557 /** 2558 * __skb_frag_set_page - sets the page contained in a paged fragment 2559 * @frag: the paged fragment 2560 * @page: the page to set 2561 * 2562 * Sets the fragment @frag to contain @page. 2563 */ 2564 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2565 { 2566 frag->page.p = page; 2567 } 2568 2569 /** 2570 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2571 * @skb: the buffer 2572 * @f: the fragment offset 2573 * @page: the page to set 2574 * 2575 * Sets the @f'th fragment of @skb to contain @page. 2576 */ 2577 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2578 struct page *page) 2579 { 2580 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2581 } 2582 2583 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2584 2585 /** 2586 * skb_frag_dma_map - maps a paged fragment via the DMA API 2587 * @dev: the device to map the fragment to 2588 * @frag: the paged fragment to map 2589 * @offset: the offset within the fragment (starting at the 2590 * fragment's own offset) 2591 * @size: the number of bytes to map 2592 * @dir: the direction of the mapping (%PCI_DMA_*) 2593 * 2594 * Maps the page associated with @frag to @device. 2595 */ 2596 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2597 const skb_frag_t *frag, 2598 size_t offset, size_t size, 2599 enum dma_data_direction dir) 2600 { 2601 return dma_map_page(dev, skb_frag_page(frag), 2602 frag->page_offset + offset, size, dir); 2603 } 2604 2605 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2606 gfp_t gfp_mask) 2607 { 2608 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2609 } 2610 2611 2612 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2613 gfp_t gfp_mask) 2614 { 2615 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2616 } 2617 2618 2619 /** 2620 * skb_clone_writable - is the header of a clone writable 2621 * @skb: buffer to check 2622 * @len: length up to which to write 2623 * 2624 * Returns true if modifying the header part of the cloned buffer 2625 * does not requires the data to be copied. 2626 */ 2627 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2628 { 2629 return !skb_header_cloned(skb) && 2630 skb_headroom(skb) + len <= skb->hdr_len; 2631 } 2632 2633 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2634 int cloned) 2635 { 2636 int delta = 0; 2637 2638 if (headroom > skb_headroom(skb)) 2639 delta = headroom - skb_headroom(skb); 2640 2641 if (delta || cloned) 2642 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2643 GFP_ATOMIC); 2644 return 0; 2645 } 2646 2647 /** 2648 * skb_cow - copy header of skb when it is required 2649 * @skb: buffer to cow 2650 * @headroom: needed headroom 2651 * 2652 * If the skb passed lacks sufficient headroom or its data part 2653 * is shared, data is reallocated. If reallocation fails, an error 2654 * is returned and original skb is not changed. 2655 * 2656 * The result is skb with writable area skb->head...skb->tail 2657 * and at least @headroom of space at head. 2658 */ 2659 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2660 { 2661 return __skb_cow(skb, headroom, skb_cloned(skb)); 2662 } 2663 2664 /** 2665 * skb_cow_head - skb_cow but only making the head writable 2666 * @skb: buffer to cow 2667 * @headroom: needed headroom 2668 * 2669 * This function is identical to skb_cow except that we replace the 2670 * skb_cloned check by skb_header_cloned. It should be used when 2671 * you only need to push on some header and do not need to modify 2672 * the data. 2673 */ 2674 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2675 { 2676 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2677 } 2678 2679 /** 2680 * skb_padto - pad an skbuff up to a minimal size 2681 * @skb: buffer to pad 2682 * @len: minimal length 2683 * 2684 * Pads up a buffer to ensure the trailing bytes exist and are 2685 * blanked. If the buffer already contains sufficient data it 2686 * is untouched. Otherwise it is extended. Returns zero on 2687 * success. The skb is freed on error. 2688 */ 2689 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2690 { 2691 unsigned int size = skb->len; 2692 if (likely(size >= len)) 2693 return 0; 2694 return skb_pad(skb, len - size); 2695 } 2696 2697 /** 2698 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2699 * @skb: buffer to pad 2700 * @len: minimal length 2701 * 2702 * Pads up a buffer to ensure the trailing bytes exist and are 2703 * blanked. If the buffer already contains sufficient data it 2704 * is untouched. Otherwise it is extended. Returns zero on 2705 * success. The skb is freed on error. 2706 */ 2707 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2708 { 2709 unsigned int size = skb->len; 2710 2711 if (unlikely(size < len)) { 2712 len -= size; 2713 if (skb_pad(skb, len)) 2714 return -ENOMEM; 2715 __skb_put(skb, len); 2716 } 2717 return 0; 2718 } 2719 2720 static inline int skb_add_data(struct sk_buff *skb, 2721 struct iov_iter *from, int copy) 2722 { 2723 const int off = skb->len; 2724 2725 if (skb->ip_summed == CHECKSUM_NONE) { 2726 __wsum csum = 0; 2727 if (csum_and_copy_from_iter(skb_put(skb, copy), copy, 2728 &csum, from) == copy) { 2729 skb->csum = csum_block_add(skb->csum, csum, off); 2730 return 0; 2731 } 2732 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy) 2733 return 0; 2734 2735 __skb_trim(skb, off); 2736 return -EFAULT; 2737 } 2738 2739 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2740 const struct page *page, int off) 2741 { 2742 if (i) { 2743 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2744 2745 return page == skb_frag_page(frag) && 2746 off == frag->page_offset + skb_frag_size(frag); 2747 } 2748 return false; 2749 } 2750 2751 static inline int __skb_linearize(struct sk_buff *skb) 2752 { 2753 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2754 } 2755 2756 /** 2757 * skb_linearize - convert paged skb to linear one 2758 * @skb: buffer to linarize 2759 * 2760 * If there is no free memory -ENOMEM is returned, otherwise zero 2761 * is returned and the old skb data released. 2762 */ 2763 static inline int skb_linearize(struct sk_buff *skb) 2764 { 2765 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2766 } 2767 2768 /** 2769 * skb_has_shared_frag - can any frag be overwritten 2770 * @skb: buffer to test 2771 * 2772 * Return true if the skb has at least one frag that might be modified 2773 * by an external entity (as in vmsplice()/sendfile()) 2774 */ 2775 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 2776 { 2777 return skb_is_nonlinear(skb) && 2778 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2779 } 2780 2781 /** 2782 * skb_linearize_cow - make sure skb is linear and writable 2783 * @skb: buffer to process 2784 * 2785 * If there is no free memory -ENOMEM is returned, otherwise zero 2786 * is returned and the old skb data released. 2787 */ 2788 static inline int skb_linearize_cow(struct sk_buff *skb) 2789 { 2790 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2791 __skb_linearize(skb) : 0; 2792 } 2793 2794 /** 2795 * skb_postpull_rcsum - update checksum for received skb after pull 2796 * @skb: buffer to update 2797 * @start: start of data before pull 2798 * @len: length of data pulled 2799 * 2800 * After doing a pull on a received packet, you need to call this to 2801 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2802 * CHECKSUM_NONE so that it can be recomputed from scratch. 2803 */ 2804 2805 static inline void skb_postpull_rcsum(struct sk_buff *skb, 2806 const void *start, unsigned int len) 2807 { 2808 if (skb->ip_summed == CHECKSUM_COMPLETE) 2809 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0)); 2810 else if (skb->ip_summed == CHECKSUM_PARTIAL && 2811 skb_checksum_start_offset(skb) < 0) 2812 skb->ip_summed = CHECKSUM_NONE; 2813 } 2814 2815 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2816 2817 static inline void skb_postpush_rcsum(struct sk_buff *skb, 2818 const void *start, unsigned int len) 2819 { 2820 /* For performing the reverse operation to skb_postpull_rcsum(), 2821 * we can instead of ... 2822 * 2823 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0)); 2824 * 2825 * ... just use this equivalent version here to save a few 2826 * instructions. Feeding csum of 0 in csum_partial() and later 2827 * on adding skb->csum is equivalent to feed skb->csum in the 2828 * first place. 2829 */ 2830 if (skb->ip_summed == CHECKSUM_COMPLETE) 2831 skb->csum = csum_partial(start, len, skb->csum); 2832 } 2833 2834 /** 2835 * pskb_trim_rcsum - trim received skb and update checksum 2836 * @skb: buffer to trim 2837 * @len: new length 2838 * 2839 * This is exactly the same as pskb_trim except that it ensures the 2840 * checksum of received packets are still valid after the operation. 2841 */ 2842 2843 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2844 { 2845 if (likely(len >= skb->len)) 2846 return 0; 2847 if (skb->ip_summed == CHECKSUM_COMPLETE) 2848 skb->ip_summed = CHECKSUM_NONE; 2849 return __pskb_trim(skb, len); 2850 } 2851 2852 #define skb_queue_walk(queue, skb) \ 2853 for (skb = (queue)->next; \ 2854 skb != (struct sk_buff *)(queue); \ 2855 skb = skb->next) 2856 2857 #define skb_queue_walk_safe(queue, skb, tmp) \ 2858 for (skb = (queue)->next, tmp = skb->next; \ 2859 skb != (struct sk_buff *)(queue); \ 2860 skb = tmp, tmp = skb->next) 2861 2862 #define skb_queue_walk_from(queue, skb) \ 2863 for (; skb != (struct sk_buff *)(queue); \ 2864 skb = skb->next) 2865 2866 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 2867 for (tmp = skb->next; \ 2868 skb != (struct sk_buff *)(queue); \ 2869 skb = tmp, tmp = skb->next) 2870 2871 #define skb_queue_reverse_walk(queue, skb) \ 2872 for (skb = (queue)->prev; \ 2873 skb != (struct sk_buff *)(queue); \ 2874 skb = skb->prev) 2875 2876 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 2877 for (skb = (queue)->prev, tmp = skb->prev; \ 2878 skb != (struct sk_buff *)(queue); \ 2879 skb = tmp, tmp = skb->prev) 2880 2881 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 2882 for (tmp = skb->prev; \ 2883 skb != (struct sk_buff *)(queue); \ 2884 skb = tmp, tmp = skb->prev) 2885 2886 static inline bool skb_has_frag_list(const struct sk_buff *skb) 2887 { 2888 return skb_shinfo(skb)->frag_list != NULL; 2889 } 2890 2891 static inline void skb_frag_list_init(struct sk_buff *skb) 2892 { 2893 skb_shinfo(skb)->frag_list = NULL; 2894 } 2895 2896 #define skb_walk_frags(skb, iter) \ 2897 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 2898 2899 2900 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, 2901 const struct sk_buff *skb); 2902 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags, 2903 int *peeked, int *off, int *err, 2904 struct sk_buff **last); 2905 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 2906 int *peeked, int *off, int *err); 2907 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 2908 int *err); 2909 unsigned int datagram_poll(struct file *file, struct socket *sock, 2910 struct poll_table_struct *wait); 2911 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 2912 struct iov_iter *to, int size); 2913 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 2914 struct msghdr *msg, int size) 2915 { 2916 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 2917 } 2918 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 2919 struct msghdr *msg); 2920 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 2921 struct iov_iter *from, int len); 2922 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 2923 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 2924 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb); 2925 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 2926 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 2927 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 2928 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 2929 int len, __wsum csum); 2930 ssize_t skb_socket_splice(struct sock *sk, 2931 struct pipe_inode_info *pipe, 2932 struct splice_pipe_desc *spd); 2933 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 2934 struct pipe_inode_info *pipe, unsigned int len, 2935 unsigned int flags, 2936 ssize_t (*splice_cb)(struct sock *, 2937 struct pipe_inode_info *, 2938 struct splice_pipe_desc *)); 2939 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 2940 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 2941 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 2942 int len, int hlen); 2943 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 2944 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 2945 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 2946 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 2947 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 2948 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 2949 int skb_ensure_writable(struct sk_buff *skb, int write_len); 2950 int skb_vlan_pop(struct sk_buff *skb); 2951 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 2952 2953 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 2954 { 2955 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2956 } 2957 2958 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 2959 { 2960 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2961 } 2962 2963 struct skb_checksum_ops { 2964 __wsum (*update)(const void *mem, int len, __wsum wsum); 2965 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 2966 }; 2967 2968 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 2969 __wsum csum, const struct skb_checksum_ops *ops); 2970 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 2971 __wsum csum); 2972 2973 static inline void * __must_check 2974 __skb_header_pointer(const struct sk_buff *skb, int offset, 2975 int len, void *data, int hlen, void *buffer) 2976 { 2977 if (hlen - offset >= len) 2978 return data + offset; 2979 2980 if (!skb || 2981 skb_copy_bits(skb, offset, buffer, len) < 0) 2982 return NULL; 2983 2984 return buffer; 2985 } 2986 2987 static inline void * __must_check 2988 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 2989 { 2990 return __skb_header_pointer(skb, offset, len, skb->data, 2991 skb_headlen(skb), buffer); 2992 } 2993 2994 /** 2995 * skb_needs_linearize - check if we need to linearize a given skb 2996 * depending on the given device features. 2997 * @skb: socket buffer to check 2998 * @features: net device features 2999 * 3000 * Returns true if either: 3001 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3002 * 2. skb is fragmented and the device does not support SG. 3003 */ 3004 static inline bool skb_needs_linearize(struct sk_buff *skb, 3005 netdev_features_t features) 3006 { 3007 return skb_is_nonlinear(skb) && 3008 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3009 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3010 } 3011 3012 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3013 void *to, 3014 const unsigned int len) 3015 { 3016 memcpy(to, skb->data, len); 3017 } 3018 3019 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3020 const int offset, void *to, 3021 const unsigned int len) 3022 { 3023 memcpy(to, skb->data + offset, len); 3024 } 3025 3026 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3027 const void *from, 3028 const unsigned int len) 3029 { 3030 memcpy(skb->data, from, len); 3031 } 3032 3033 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3034 const int offset, 3035 const void *from, 3036 const unsigned int len) 3037 { 3038 memcpy(skb->data + offset, from, len); 3039 } 3040 3041 void skb_init(void); 3042 3043 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3044 { 3045 return skb->tstamp; 3046 } 3047 3048 /** 3049 * skb_get_timestamp - get timestamp from a skb 3050 * @skb: skb to get stamp from 3051 * @stamp: pointer to struct timeval to store stamp in 3052 * 3053 * Timestamps are stored in the skb as offsets to a base timestamp. 3054 * This function converts the offset back to a struct timeval and stores 3055 * it in stamp. 3056 */ 3057 static inline void skb_get_timestamp(const struct sk_buff *skb, 3058 struct timeval *stamp) 3059 { 3060 *stamp = ktime_to_timeval(skb->tstamp); 3061 } 3062 3063 static inline void skb_get_timestampns(const struct sk_buff *skb, 3064 struct timespec *stamp) 3065 { 3066 *stamp = ktime_to_timespec(skb->tstamp); 3067 } 3068 3069 static inline void __net_timestamp(struct sk_buff *skb) 3070 { 3071 skb->tstamp = ktime_get_real(); 3072 } 3073 3074 static inline ktime_t net_timedelta(ktime_t t) 3075 { 3076 return ktime_sub(ktime_get_real(), t); 3077 } 3078 3079 static inline ktime_t net_invalid_timestamp(void) 3080 { 3081 return ktime_set(0, 0); 3082 } 3083 3084 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3085 3086 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3087 3088 void skb_clone_tx_timestamp(struct sk_buff *skb); 3089 bool skb_defer_rx_timestamp(struct sk_buff *skb); 3090 3091 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3092 3093 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3094 { 3095 } 3096 3097 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3098 { 3099 return false; 3100 } 3101 3102 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3103 3104 /** 3105 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3106 * 3107 * PHY drivers may accept clones of transmitted packets for 3108 * timestamping via their phy_driver.txtstamp method. These drivers 3109 * must call this function to return the skb back to the stack with a 3110 * timestamp. 3111 * 3112 * @skb: clone of the the original outgoing packet 3113 * @hwtstamps: hardware time stamps 3114 * 3115 */ 3116 void skb_complete_tx_timestamp(struct sk_buff *skb, 3117 struct skb_shared_hwtstamps *hwtstamps); 3118 3119 void __skb_tstamp_tx(struct sk_buff *orig_skb, 3120 struct skb_shared_hwtstamps *hwtstamps, 3121 struct sock *sk, int tstype); 3122 3123 /** 3124 * skb_tstamp_tx - queue clone of skb with send time stamps 3125 * @orig_skb: the original outgoing packet 3126 * @hwtstamps: hardware time stamps, may be NULL if not available 3127 * 3128 * If the skb has a socket associated, then this function clones the 3129 * skb (thus sharing the actual data and optional structures), stores 3130 * the optional hardware time stamping information (if non NULL) or 3131 * generates a software time stamp (otherwise), then queues the clone 3132 * to the error queue of the socket. Errors are silently ignored. 3133 */ 3134 void skb_tstamp_tx(struct sk_buff *orig_skb, 3135 struct skb_shared_hwtstamps *hwtstamps); 3136 3137 static inline void sw_tx_timestamp(struct sk_buff *skb) 3138 { 3139 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 3140 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 3141 skb_tstamp_tx(skb, NULL); 3142 } 3143 3144 /** 3145 * skb_tx_timestamp() - Driver hook for transmit timestamping 3146 * 3147 * Ethernet MAC Drivers should call this function in their hard_xmit() 3148 * function immediately before giving the sk_buff to the MAC hardware. 3149 * 3150 * Specifically, one should make absolutely sure that this function is 3151 * called before TX completion of this packet can trigger. Otherwise 3152 * the packet could potentially already be freed. 3153 * 3154 * @skb: A socket buffer. 3155 */ 3156 static inline void skb_tx_timestamp(struct sk_buff *skb) 3157 { 3158 skb_clone_tx_timestamp(skb); 3159 sw_tx_timestamp(skb); 3160 } 3161 3162 /** 3163 * skb_complete_wifi_ack - deliver skb with wifi status 3164 * 3165 * @skb: the original outgoing packet 3166 * @acked: ack status 3167 * 3168 */ 3169 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3170 3171 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3172 __sum16 __skb_checksum_complete(struct sk_buff *skb); 3173 3174 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3175 { 3176 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3177 skb->csum_valid || 3178 (skb->ip_summed == CHECKSUM_PARTIAL && 3179 skb_checksum_start_offset(skb) >= 0)); 3180 } 3181 3182 /** 3183 * skb_checksum_complete - Calculate checksum of an entire packet 3184 * @skb: packet to process 3185 * 3186 * This function calculates the checksum over the entire packet plus 3187 * the value of skb->csum. The latter can be used to supply the 3188 * checksum of a pseudo header as used by TCP/UDP. It returns the 3189 * checksum. 3190 * 3191 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3192 * this function can be used to verify that checksum on received 3193 * packets. In that case the function should return zero if the 3194 * checksum is correct. In particular, this function will return zero 3195 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3196 * hardware has already verified the correctness of the checksum. 3197 */ 3198 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3199 { 3200 return skb_csum_unnecessary(skb) ? 3201 0 : __skb_checksum_complete(skb); 3202 } 3203 3204 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3205 { 3206 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3207 if (skb->csum_level == 0) 3208 skb->ip_summed = CHECKSUM_NONE; 3209 else 3210 skb->csum_level--; 3211 } 3212 } 3213 3214 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3215 { 3216 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3217 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3218 skb->csum_level++; 3219 } else if (skb->ip_summed == CHECKSUM_NONE) { 3220 skb->ip_summed = CHECKSUM_UNNECESSARY; 3221 skb->csum_level = 0; 3222 } 3223 } 3224 3225 static inline void __skb_mark_checksum_bad(struct sk_buff *skb) 3226 { 3227 /* Mark current checksum as bad (typically called from GRO 3228 * path). In the case that ip_summed is CHECKSUM_NONE 3229 * this must be the first checksum encountered in the packet. 3230 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first 3231 * checksum after the last one validated. For UDP, a zero 3232 * checksum can not be marked as bad. 3233 */ 3234 3235 if (skb->ip_summed == CHECKSUM_NONE || 3236 skb->ip_summed == CHECKSUM_UNNECESSARY) 3237 skb->csum_bad = 1; 3238 } 3239 3240 /* Check if we need to perform checksum complete validation. 3241 * 3242 * Returns true if checksum complete is needed, false otherwise 3243 * (either checksum is unnecessary or zero checksum is allowed). 3244 */ 3245 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3246 bool zero_okay, 3247 __sum16 check) 3248 { 3249 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3250 skb->csum_valid = 1; 3251 __skb_decr_checksum_unnecessary(skb); 3252 return false; 3253 } 3254 3255 return true; 3256 } 3257 3258 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3259 * in checksum_init. 3260 */ 3261 #define CHECKSUM_BREAK 76 3262 3263 /* Unset checksum-complete 3264 * 3265 * Unset checksum complete can be done when packet is being modified 3266 * (uncompressed for instance) and checksum-complete value is 3267 * invalidated. 3268 */ 3269 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3270 { 3271 if (skb->ip_summed == CHECKSUM_COMPLETE) 3272 skb->ip_summed = CHECKSUM_NONE; 3273 } 3274 3275 /* Validate (init) checksum based on checksum complete. 3276 * 3277 * Return values: 3278 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3279 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3280 * checksum is stored in skb->csum for use in __skb_checksum_complete 3281 * non-zero: value of invalid checksum 3282 * 3283 */ 3284 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3285 bool complete, 3286 __wsum psum) 3287 { 3288 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3289 if (!csum_fold(csum_add(psum, skb->csum))) { 3290 skb->csum_valid = 1; 3291 return 0; 3292 } 3293 } else if (skb->csum_bad) { 3294 /* ip_summed == CHECKSUM_NONE in this case */ 3295 return (__force __sum16)1; 3296 } 3297 3298 skb->csum = psum; 3299 3300 if (complete || skb->len <= CHECKSUM_BREAK) { 3301 __sum16 csum; 3302 3303 csum = __skb_checksum_complete(skb); 3304 skb->csum_valid = !csum; 3305 return csum; 3306 } 3307 3308 return 0; 3309 } 3310 3311 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3312 { 3313 return 0; 3314 } 3315 3316 /* Perform checksum validate (init). Note that this is a macro since we only 3317 * want to calculate the pseudo header which is an input function if necessary. 3318 * First we try to validate without any computation (checksum unnecessary) and 3319 * then calculate based on checksum complete calling the function to compute 3320 * pseudo header. 3321 * 3322 * Return values: 3323 * 0: checksum is validated or try to in skb_checksum_complete 3324 * non-zero: value of invalid checksum 3325 */ 3326 #define __skb_checksum_validate(skb, proto, complete, \ 3327 zero_okay, check, compute_pseudo) \ 3328 ({ \ 3329 __sum16 __ret = 0; \ 3330 skb->csum_valid = 0; \ 3331 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3332 __ret = __skb_checksum_validate_complete(skb, \ 3333 complete, compute_pseudo(skb, proto)); \ 3334 __ret; \ 3335 }) 3336 3337 #define skb_checksum_init(skb, proto, compute_pseudo) \ 3338 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3339 3340 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3341 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3342 3343 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 3344 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3345 3346 #define skb_checksum_validate_zero_check(skb, proto, check, \ 3347 compute_pseudo) \ 3348 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3349 3350 #define skb_checksum_simple_validate(skb) \ 3351 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3352 3353 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3354 { 3355 return (skb->ip_summed == CHECKSUM_NONE && 3356 skb->csum_valid && !skb->csum_bad); 3357 } 3358 3359 static inline void __skb_checksum_convert(struct sk_buff *skb, 3360 __sum16 check, __wsum pseudo) 3361 { 3362 skb->csum = ~pseudo; 3363 skb->ip_summed = CHECKSUM_COMPLETE; 3364 } 3365 3366 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3367 do { \ 3368 if (__skb_checksum_convert_check(skb)) \ 3369 __skb_checksum_convert(skb, check, \ 3370 compute_pseudo(skb, proto)); \ 3371 } while (0) 3372 3373 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3374 u16 start, u16 offset) 3375 { 3376 skb->ip_summed = CHECKSUM_PARTIAL; 3377 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3378 skb->csum_offset = offset - start; 3379 } 3380 3381 /* Update skbuf and packet to reflect the remote checksum offload operation. 3382 * When called, ptr indicates the starting point for skb->csum when 3383 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3384 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3385 */ 3386 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3387 int start, int offset, bool nopartial) 3388 { 3389 __wsum delta; 3390 3391 if (!nopartial) { 3392 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3393 return; 3394 } 3395 3396 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3397 __skb_checksum_complete(skb); 3398 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3399 } 3400 3401 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3402 3403 /* Adjust skb->csum since we changed the packet */ 3404 skb->csum = csum_add(skb->csum, delta); 3405 } 3406 3407 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3408 void nf_conntrack_destroy(struct nf_conntrack *nfct); 3409 static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3410 { 3411 if (nfct && atomic_dec_and_test(&nfct->use)) 3412 nf_conntrack_destroy(nfct); 3413 } 3414 static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3415 { 3416 if (nfct) 3417 atomic_inc(&nfct->use); 3418 } 3419 #endif 3420 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3421 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3422 { 3423 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 3424 kfree(nf_bridge); 3425 } 3426 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3427 { 3428 if (nf_bridge) 3429 atomic_inc(&nf_bridge->use); 3430 } 3431 #endif /* CONFIG_BRIDGE_NETFILTER */ 3432 static inline void nf_reset(struct sk_buff *skb) 3433 { 3434 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3435 nf_conntrack_put(skb->nfct); 3436 skb->nfct = NULL; 3437 #endif 3438 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3439 nf_bridge_put(skb->nf_bridge); 3440 skb->nf_bridge = NULL; 3441 #endif 3442 } 3443 3444 static inline void nf_reset_trace(struct sk_buff *skb) 3445 { 3446 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3447 skb->nf_trace = 0; 3448 #endif 3449 } 3450 3451 /* Note: This doesn't put any conntrack and bridge info in dst. */ 3452 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3453 bool copy) 3454 { 3455 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3456 dst->nfct = src->nfct; 3457 nf_conntrack_get(src->nfct); 3458 if (copy) 3459 dst->nfctinfo = src->nfctinfo; 3460 #endif 3461 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3462 dst->nf_bridge = src->nf_bridge; 3463 nf_bridge_get(src->nf_bridge); 3464 #endif 3465 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3466 if (copy) 3467 dst->nf_trace = src->nf_trace; 3468 #endif 3469 } 3470 3471 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3472 { 3473 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3474 nf_conntrack_put(dst->nfct); 3475 #endif 3476 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3477 nf_bridge_put(dst->nf_bridge); 3478 #endif 3479 __nf_copy(dst, src, true); 3480 } 3481 3482 #ifdef CONFIG_NETWORK_SECMARK 3483 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3484 { 3485 to->secmark = from->secmark; 3486 } 3487 3488 static inline void skb_init_secmark(struct sk_buff *skb) 3489 { 3490 skb->secmark = 0; 3491 } 3492 #else 3493 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3494 { } 3495 3496 static inline void skb_init_secmark(struct sk_buff *skb) 3497 { } 3498 #endif 3499 3500 static inline bool skb_irq_freeable(const struct sk_buff *skb) 3501 { 3502 return !skb->destructor && 3503 #if IS_ENABLED(CONFIG_XFRM) 3504 !skb->sp && 3505 #endif 3506 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 3507 !skb->nfct && 3508 #endif 3509 !skb->_skb_refdst && 3510 !skb_has_frag_list(skb); 3511 } 3512 3513 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3514 { 3515 skb->queue_mapping = queue_mapping; 3516 } 3517 3518 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3519 { 3520 return skb->queue_mapping; 3521 } 3522 3523 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3524 { 3525 to->queue_mapping = from->queue_mapping; 3526 } 3527 3528 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3529 { 3530 skb->queue_mapping = rx_queue + 1; 3531 } 3532 3533 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3534 { 3535 return skb->queue_mapping - 1; 3536 } 3537 3538 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3539 { 3540 return skb->queue_mapping != 0; 3541 } 3542 3543 static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3544 { 3545 #ifdef CONFIG_XFRM 3546 return skb->sp; 3547 #else 3548 return NULL; 3549 #endif 3550 } 3551 3552 /* Keeps track of mac header offset relative to skb->head. 3553 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3554 * For non-tunnel skb it points to skb_mac_header() and for 3555 * tunnel skb it points to outer mac header. 3556 * Keeps track of level of encapsulation of network headers. 3557 */ 3558 struct skb_gso_cb { 3559 int mac_offset; 3560 int encap_level; 3561 __wsum csum; 3562 __u16 csum_start; 3563 }; 3564 #define SKB_SGO_CB_OFFSET 32 3565 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 3566 3567 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3568 { 3569 return (skb_mac_header(inner_skb) - inner_skb->head) - 3570 SKB_GSO_CB(inner_skb)->mac_offset; 3571 } 3572 3573 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3574 { 3575 int new_headroom, headroom; 3576 int ret; 3577 3578 headroom = skb_headroom(skb); 3579 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3580 if (ret) 3581 return ret; 3582 3583 new_headroom = skb_headroom(skb); 3584 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3585 return 0; 3586 } 3587 3588 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 3589 { 3590 /* Do not update partial checksums if remote checksum is enabled. */ 3591 if (skb->remcsum_offload) 3592 return; 3593 3594 SKB_GSO_CB(skb)->csum = res; 3595 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 3596 } 3597 3598 /* Compute the checksum for a gso segment. First compute the checksum value 3599 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3600 * then add in skb->csum (checksum from csum_start to end of packet). 3601 * skb->csum and csum_start are then updated to reflect the checksum of the 3602 * resultant packet starting from the transport header-- the resultant checksum 3603 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3604 * header. 3605 */ 3606 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3607 { 3608 unsigned char *csum_start = skb_transport_header(skb); 3609 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 3610 __wsum partial = SKB_GSO_CB(skb)->csum; 3611 3612 SKB_GSO_CB(skb)->csum = res; 3613 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 3614 3615 return csum_fold(csum_partial(csum_start, plen, partial)); 3616 } 3617 3618 static inline bool skb_is_gso(const struct sk_buff *skb) 3619 { 3620 return skb_shinfo(skb)->gso_size; 3621 } 3622 3623 /* Note: Should be called only if skb_is_gso(skb) is true */ 3624 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3625 { 3626 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3627 } 3628 3629 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3630 3631 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3632 { 3633 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3634 * wanted then gso_type will be set. */ 3635 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3636 3637 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3638 unlikely(shinfo->gso_type == 0)) { 3639 __skb_warn_lro_forwarding(skb); 3640 return true; 3641 } 3642 return false; 3643 } 3644 3645 static inline void skb_forward_csum(struct sk_buff *skb) 3646 { 3647 /* Unfortunately we don't support this one. Any brave souls? */ 3648 if (skb->ip_summed == CHECKSUM_COMPLETE) 3649 skb->ip_summed = CHECKSUM_NONE; 3650 } 3651 3652 /** 3653 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3654 * @skb: skb to check 3655 * 3656 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3657 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3658 * use this helper, to document places where we make this assertion. 3659 */ 3660 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3661 { 3662 #ifdef DEBUG 3663 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3664 #endif 3665 } 3666 3667 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3668 3669 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3670 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 3671 unsigned int transport_len, 3672 __sum16(*skb_chkf)(struct sk_buff *skb)); 3673 3674 /** 3675 * skb_head_is_locked - Determine if the skb->head is locked down 3676 * @skb: skb to check 3677 * 3678 * The head on skbs build around a head frag can be removed if they are 3679 * not cloned. This function returns true if the skb head is locked down 3680 * due to either being allocated via kmalloc, or by being a clone with 3681 * multiple references to the head. 3682 */ 3683 static inline bool skb_head_is_locked(const struct sk_buff *skb) 3684 { 3685 return !skb->head_frag || skb_cloned(skb); 3686 } 3687 3688 /** 3689 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3690 * 3691 * @skb: GSO skb 3692 * 3693 * skb_gso_network_seglen is used to determine the real size of the 3694 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3695 * 3696 * The MAC/L2 header is not accounted for. 3697 */ 3698 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 3699 { 3700 unsigned int hdr_len = skb_transport_header(skb) - 3701 skb_network_header(skb); 3702 return hdr_len + skb_gso_transport_seglen(skb); 3703 } 3704 3705 /* Local Checksum Offload. 3706 * Compute outer checksum based on the assumption that the 3707 * inner checksum will be offloaded later. 3708 * See Documentation/networking/checksum-offloads.txt for 3709 * explanation of how this works. 3710 * Fill in outer checksum adjustment (e.g. with sum of outer 3711 * pseudo-header) before calling. 3712 * Also ensure that inner checksum is in linear data area. 3713 */ 3714 static inline __wsum lco_csum(struct sk_buff *skb) 3715 { 3716 unsigned char *csum_start = skb_checksum_start(skb); 3717 unsigned char *l4_hdr = skb_transport_header(skb); 3718 __wsum partial; 3719 3720 /* Start with complement of inner checksum adjustment */ 3721 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 3722 skb->csum_offset)); 3723 3724 /* Add in checksum of our headers (incl. outer checksum 3725 * adjustment filled in by caller) and return result. 3726 */ 3727 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 3728 } 3729 3730 #endif /* __KERNEL__ */ 3731 #endif /* _LINUX_SKBUFF_H */ 3732