1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * Definitions for the 'struct sk_buff' memory handlers. 4 * 5 * Authors: 6 * Alan Cox, <[email protected]> 7 * Florian La Roche, <[email protected]> 8 */ 9 10 #ifndef _LINUX_SKBUFF_H 11 #define _LINUX_SKBUFF_H 12 13 #include <linux/kernel.h> 14 #include <linux/compiler.h> 15 #include <linux/time.h> 16 #include <linux/bug.h> 17 #include <linux/bvec.h> 18 #include <linux/cache.h> 19 #include <linux/rbtree.h> 20 #include <linux/socket.h> 21 #include <linux/refcount.h> 22 23 #include <linux/atomic.h> 24 #include <asm/types.h> 25 #include <linux/spinlock.h> 26 #include <net/checksum.h> 27 #include <linux/rcupdate.h> 28 #include <linux/dma-mapping.h> 29 #include <linux/netdev_features.h> 30 #include <net/flow_dissector.h> 31 #include <linux/in6.h> 32 #include <linux/if_packet.h> 33 #include <linux/llist.h> 34 #include <net/flow.h> 35 #include <net/page_pool.h> 36 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 37 #include <linux/netfilter/nf_conntrack_common.h> 38 #endif 39 #include <net/net_debug.h> 40 #include <net/dropreason.h> 41 42 /** 43 * DOC: skb checksums 44 * 45 * The interface for checksum offload between the stack and networking drivers 46 * is as follows... 47 * 48 * IP checksum related features 49 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 50 * 51 * Drivers advertise checksum offload capabilities in the features of a device. 52 * From the stack's point of view these are capabilities offered by the driver. 53 * A driver typically only advertises features that it is capable of offloading 54 * to its device. 55 * 56 * .. flat-table:: Checksum related device features 57 * :widths: 1 10 58 * 59 * * - %NETIF_F_HW_CSUM 60 * - The driver (or its device) is able to compute one 61 * IP (one's complement) checksum for any combination 62 * of protocols or protocol layering. The checksum is 63 * computed and set in a packet per the CHECKSUM_PARTIAL 64 * interface (see below). 65 * 66 * * - %NETIF_F_IP_CSUM 67 * - Driver (device) is only able to checksum plain 68 * TCP or UDP packets over IPv4. These are specifically 69 * unencapsulated packets of the form IPv4|TCP or 70 * IPv4|UDP where the Protocol field in the IPv4 header 71 * is TCP or UDP. The IPv4 header may contain IP options. 72 * This feature cannot be set in features for a device 73 * with NETIF_F_HW_CSUM also set. This feature is being 74 * DEPRECATED (see below). 75 * 76 * * - %NETIF_F_IPV6_CSUM 77 * - Driver (device) is only able to checksum plain 78 * TCP or UDP packets over IPv6. These are specifically 79 * unencapsulated packets of the form IPv6|TCP or 80 * IPv6|UDP where the Next Header field in the IPv6 81 * header is either TCP or UDP. IPv6 extension headers 82 * are not supported with this feature. This feature 83 * cannot be set in features for a device with 84 * NETIF_F_HW_CSUM also set. This feature is being 85 * DEPRECATED (see below). 86 * 87 * * - %NETIF_F_RXCSUM 88 * - Driver (device) performs receive checksum offload. 89 * This flag is only used to disable the RX checksum 90 * feature for a device. The stack will accept receive 91 * checksum indication in packets received on a device 92 * regardless of whether NETIF_F_RXCSUM is set. 93 * 94 * Checksumming of received packets by device 95 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 96 * 97 * Indication of checksum verification is set in &sk_buff.ip_summed. 98 * Possible values are: 99 * 100 * - %CHECKSUM_NONE 101 * 102 * Device did not checksum this packet e.g. due to lack of capabilities. 103 * The packet contains full (though not verified) checksum in packet but 104 * not in skb->csum. Thus, skb->csum is undefined in this case. 105 * 106 * - %CHECKSUM_UNNECESSARY 107 * 108 * The hardware you're dealing with doesn't calculate the full checksum 109 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums 110 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY 111 * if their checksums are okay. &sk_buff.csum is still undefined in this case 112 * though. A driver or device must never modify the checksum field in the 113 * packet even if checksum is verified. 114 * 115 * %CHECKSUM_UNNECESSARY is applicable to following protocols: 116 * 117 * - TCP: IPv6 and IPv4. 118 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 119 * zero UDP checksum for either IPv4 or IPv6, the networking stack 120 * may perform further validation in this case. 121 * - GRE: only if the checksum is present in the header. 122 * - SCTP: indicates the CRC in SCTP header has been validated. 123 * - FCOE: indicates the CRC in FC frame has been validated. 124 * 125 * &sk_buff.csum_level indicates the number of consecutive checksums found in 126 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY. 127 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 128 * and a device is able to verify the checksums for UDP (possibly zero), 129 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to 130 * two. If the device were only able to verify the UDP checksum and not 131 * GRE, either because it doesn't support GRE checksum or because GRE 132 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 133 * not considered in this case). 134 * 135 * - %CHECKSUM_COMPLETE 136 * 137 * This is the most generic way. The device supplied checksum of the _whole_ 138 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the 139 * hardware doesn't need to parse L3/L4 headers to implement this. 140 * 141 * Notes: 142 * 143 * - Even if device supports only some protocols, but is able to produce 144 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 145 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 146 * 147 * - %CHECKSUM_PARTIAL 148 * 149 * A checksum is set up to be offloaded to a device as described in the 150 * output description for CHECKSUM_PARTIAL. This may occur on a packet 151 * received directly from another Linux OS, e.g., a virtualized Linux kernel 152 * on the same host, or it may be set in the input path in GRO or remote 153 * checksum offload. For the purposes of checksum verification, the checksum 154 * referred to by skb->csum_start + skb->csum_offset and any preceding 155 * checksums in the packet are considered verified. Any checksums in the 156 * packet that are after the checksum being offloaded are not considered to 157 * be verified. 158 * 159 * Checksumming on transmit for non-GSO 160 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 161 * 162 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet. 163 * Values are: 164 * 165 * - %CHECKSUM_PARTIAL 166 * 167 * The driver is required to checksum the packet as seen by hard_start_xmit() 168 * from &sk_buff.csum_start up to the end, and to record/write the checksum at 169 * offset &sk_buff.csum_start + &sk_buff.csum_offset. 170 * A driver may verify that the 171 * csum_start and csum_offset values are valid values given the length and 172 * offset of the packet, but it should not attempt to validate that the 173 * checksum refers to a legitimate transport layer checksum -- it is the 174 * purview of the stack to validate that csum_start and csum_offset are set 175 * correctly. 176 * 177 * When the stack requests checksum offload for a packet, the driver MUST 178 * ensure that the checksum is set correctly. A driver can either offload the 179 * checksum calculation to the device, or call skb_checksum_help (in the case 180 * that the device does not support offload for a particular checksum). 181 * 182 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of 183 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate 184 * checksum offload capability. 185 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based 186 * on network device checksumming capabilities: if a packet does not match 187 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of 188 * &sk_buff.csum_not_inet, see :ref:`crc`) 189 * is called to resolve the checksum. 190 * 191 * - %CHECKSUM_NONE 192 * 193 * The skb was already checksummed by the protocol, or a checksum is not 194 * required. 195 * 196 * - %CHECKSUM_UNNECESSARY 197 * 198 * This has the same meaning as CHECKSUM_NONE for checksum offload on 199 * output. 200 * 201 * - %CHECKSUM_COMPLETE 202 * 203 * Not used in checksum output. If a driver observes a packet with this value 204 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set. 205 * 206 * .. _crc: 207 * 208 * Non-IP checksum (CRC) offloads 209 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 210 * 211 * .. flat-table:: 212 * :widths: 1 10 213 * 214 * * - %NETIF_F_SCTP_CRC 215 * - This feature indicates that a device is capable of 216 * offloading the SCTP CRC in a packet. To perform this offload the stack 217 * will set csum_start and csum_offset accordingly, set ip_summed to 218 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication 219 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c. 220 * A driver that supports both IP checksum offload and SCTP CRC32c offload 221 * must verify which offload is configured for a packet by testing the 222 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to 223 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 224 * 225 * * - %NETIF_F_FCOE_CRC 226 * - This feature indicates that a device is capable of offloading the FCOE 227 * CRC in a packet. To perform this offload the stack will set ip_summed 228 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset 229 * accordingly. Note that there is no indication in the skbuff that the 230 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports 231 * both IP checksum offload and FCOE CRC offload must verify which offload 232 * is configured for a packet, presumably by inspecting packet headers. 233 * 234 * Checksumming on output with GSO 235 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 236 * 237 * In the case of a GSO packet (skb_is_gso() is true), checksum offload 238 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 239 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as 240 * part of the GSO operation is implied. If a checksum is being offloaded 241 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and 242 * csum_offset are set to refer to the outermost checksum being offloaded 243 * (two offloaded checksums are possible with UDP encapsulation). 244 */ 245 246 /* Don't change this without changing skb_csum_unnecessary! */ 247 #define CHECKSUM_NONE 0 248 #define CHECKSUM_UNNECESSARY 1 249 #define CHECKSUM_COMPLETE 2 250 #define CHECKSUM_PARTIAL 3 251 252 /* Maximum value in skb->csum_level */ 253 #define SKB_MAX_CSUM_LEVEL 3 254 255 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 256 #define SKB_WITH_OVERHEAD(X) \ 257 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 258 #define SKB_MAX_ORDER(X, ORDER) \ 259 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 260 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 261 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 262 263 /* return minimum truesize of one skb containing X bytes of data */ 264 #define SKB_TRUESIZE(X) ((X) + \ 265 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 266 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 267 268 struct ahash_request; 269 struct net_device; 270 struct scatterlist; 271 struct pipe_inode_info; 272 struct iov_iter; 273 struct napi_struct; 274 struct bpf_prog; 275 union bpf_attr; 276 struct skb_ext; 277 struct ts_config; 278 279 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 280 struct nf_bridge_info { 281 enum { 282 BRNF_PROTO_UNCHANGED, 283 BRNF_PROTO_8021Q, 284 BRNF_PROTO_PPPOE 285 } orig_proto:8; 286 u8 pkt_otherhost:1; 287 u8 in_prerouting:1; 288 u8 bridged_dnat:1; 289 __u16 frag_max_size; 290 struct net_device *physindev; 291 292 /* always valid & non-NULL from FORWARD on, for physdev match */ 293 struct net_device *physoutdev; 294 union { 295 /* prerouting: detect dnat in orig/reply direction */ 296 __be32 ipv4_daddr; 297 struct in6_addr ipv6_daddr; 298 299 /* after prerouting + nat detected: store original source 300 * mac since neigh resolution overwrites it, only used while 301 * skb is out in neigh layer. 302 */ 303 char neigh_header[8]; 304 }; 305 }; 306 #endif 307 308 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 309 /* Chain in tc_skb_ext will be used to share the tc chain with 310 * ovs recirc_id. It will be set to the current chain by tc 311 * and read by ovs to recirc_id. 312 */ 313 struct tc_skb_ext { 314 __u32 chain; 315 __u16 mru; 316 __u16 zone; 317 u8 post_ct:1; 318 u8 post_ct_snat:1; 319 u8 post_ct_dnat:1; 320 }; 321 #endif 322 323 struct sk_buff_head { 324 /* These two members must be first to match sk_buff. */ 325 struct_group_tagged(sk_buff_list, list, 326 struct sk_buff *next; 327 struct sk_buff *prev; 328 ); 329 330 __u32 qlen; 331 spinlock_t lock; 332 }; 333 334 struct sk_buff; 335 336 /* To allow 64K frame to be packed as single skb without frag_list we 337 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 338 * buffers which do not start on a page boundary. 339 * 340 * Since GRO uses frags we allocate at least 16 regardless of page 341 * size. 342 */ 343 #if (65536/PAGE_SIZE + 1) < 16 344 #define MAX_SKB_FRAGS 16UL 345 #else 346 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 347 #endif 348 extern int sysctl_max_skb_frags; 349 350 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 351 * segment using its current segmentation instead. 352 */ 353 #define GSO_BY_FRAGS 0xFFFF 354 355 typedef struct bio_vec skb_frag_t; 356 357 /** 358 * skb_frag_size() - Returns the size of a skb fragment 359 * @frag: skb fragment 360 */ 361 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 362 { 363 return frag->bv_len; 364 } 365 366 /** 367 * skb_frag_size_set() - Sets the size of a skb fragment 368 * @frag: skb fragment 369 * @size: size of fragment 370 */ 371 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 372 { 373 frag->bv_len = size; 374 } 375 376 /** 377 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 378 * @frag: skb fragment 379 * @delta: value to add 380 */ 381 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 382 { 383 frag->bv_len += delta; 384 } 385 386 /** 387 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 388 * @frag: skb fragment 389 * @delta: value to subtract 390 */ 391 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 392 { 393 frag->bv_len -= delta; 394 } 395 396 /** 397 * skb_frag_must_loop - Test if %p is a high memory page 398 * @p: fragment's page 399 */ 400 static inline bool skb_frag_must_loop(struct page *p) 401 { 402 #if defined(CONFIG_HIGHMEM) 403 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 404 return true; 405 #endif 406 return false; 407 } 408 409 /** 410 * skb_frag_foreach_page - loop over pages in a fragment 411 * 412 * @f: skb frag to operate on 413 * @f_off: offset from start of f->bv_page 414 * @f_len: length from f_off to loop over 415 * @p: (temp var) current page 416 * @p_off: (temp var) offset from start of current page, 417 * non-zero only on first page. 418 * @p_len: (temp var) length in current page, 419 * < PAGE_SIZE only on first and last page. 420 * @copied: (temp var) length so far, excluding current p_len. 421 * 422 * A fragment can hold a compound page, in which case per-page 423 * operations, notably kmap_atomic, must be called for each 424 * regular page. 425 */ 426 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 427 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 428 p_off = (f_off) & (PAGE_SIZE - 1), \ 429 p_len = skb_frag_must_loop(p) ? \ 430 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 431 copied = 0; \ 432 copied < f_len; \ 433 copied += p_len, p++, p_off = 0, \ 434 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 435 436 #define HAVE_HW_TIME_STAMP 437 438 /** 439 * struct skb_shared_hwtstamps - hardware time stamps 440 * @hwtstamp: hardware time stamp transformed into duration 441 * since arbitrary point in time 442 * @netdev_data: address/cookie of network device driver used as 443 * reference to actual hardware time stamp 444 * 445 * Software time stamps generated by ktime_get_real() are stored in 446 * skb->tstamp. 447 * 448 * hwtstamps can only be compared against other hwtstamps from 449 * the same device. 450 * 451 * This structure is attached to packets as part of the 452 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 453 */ 454 struct skb_shared_hwtstamps { 455 union { 456 ktime_t hwtstamp; 457 void *netdev_data; 458 }; 459 }; 460 461 /* Definitions for tx_flags in struct skb_shared_info */ 462 enum { 463 /* generate hardware time stamp */ 464 SKBTX_HW_TSTAMP = 1 << 0, 465 466 /* generate software time stamp when queueing packet to NIC */ 467 SKBTX_SW_TSTAMP = 1 << 1, 468 469 /* device driver is going to provide hardware time stamp */ 470 SKBTX_IN_PROGRESS = 1 << 2, 471 472 /* generate hardware time stamp based on cycles if supported */ 473 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3, 474 475 /* generate wifi status information (where possible) */ 476 SKBTX_WIFI_STATUS = 1 << 4, 477 478 /* determine hardware time stamp based on time or cycles */ 479 SKBTX_HW_TSTAMP_NETDEV = 1 << 5, 480 481 /* generate software time stamp when entering packet scheduling */ 482 SKBTX_SCHED_TSTAMP = 1 << 6, 483 }; 484 485 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 486 SKBTX_SCHED_TSTAMP) 487 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \ 488 SKBTX_HW_TSTAMP_USE_CYCLES | \ 489 SKBTX_ANY_SW_TSTAMP) 490 491 /* Definitions for flags in struct skb_shared_info */ 492 enum { 493 /* use zcopy routines */ 494 SKBFL_ZEROCOPY_ENABLE = BIT(0), 495 496 /* This indicates at least one fragment might be overwritten 497 * (as in vmsplice(), sendfile() ...) 498 * If we need to compute a TX checksum, we'll need to copy 499 * all frags to avoid possible bad checksum 500 */ 501 SKBFL_SHARED_FRAG = BIT(1), 502 503 /* segment contains only zerocopy data and should not be 504 * charged to the kernel memory. 505 */ 506 SKBFL_PURE_ZEROCOPY = BIT(2), 507 508 SKBFL_DONT_ORPHAN = BIT(3), 509 510 /* page references are managed by the ubuf_info, so it's safe to 511 * use frags only up until ubuf_info is released 512 */ 513 SKBFL_MANAGED_FRAG_REFS = BIT(4), 514 }; 515 516 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG) 517 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \ 518 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS) 519 520 /* 521 * The callback notifies userspace to release buffers when skb DMA is done in 522 * lower device, the skb last reference should be 0 when calling this. 523 * The zerocopy_success argument is true if zero copy transmit occurred, 524 * false on data copy or out of memory error caused by data copy attempt. 525 * The ctx field is used to track device context. 526 * The desc field is used to track userspace buffer index. 527 */ 528 struct ubuf_info { 529 void (*callback)(struct sk_buff *, struct ubuf_info *, 530 bool zerocopy_success); 531 refcount_t refcnt; 532 u8 flags; 533 }; 534 535 struct ubuf_info_msgzc { 536 struct ubuf_info ubuf; 537 538 union { 539 struct { 540 unsigned long desc; 541 void *ctx; 542 }; 543 struct { 544 u32 id; 545 u16 len; 546 u16 zerocopy:1; 547 u32 bytelen; 548 }; 549 }; 550 551 struct mmpin { 552 struct user_struct *user; 553 unsigned int num_pg; 554 } mmp; 555 }; 556 557 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 558 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \ 559 ubuf) 560 561 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 562 void mm_unaccount_pinned_pages(struct mmpin *mmp); 563 564 /* This data is invariant across clones and lives at 565 * the end of the header data, ie. at skb->end. 566 */ 567 struct skb_shared_info { 568 __u8 flags; 569 __u8 meta_len; 570 __u8 nr_frags; 571 __u8 tx_flags; 572 unsigned short gso_size; 573 /* Warning: this field is not always filled in (UFO)! */ 574 unsigned short gso_segs; 575 struct sk_buff *frag_list; 576 struct skb_shared_hwtstamps hwtstamps; 577 unsigned int gso_type; 578 u32 tskey; 579 580 /* 581 * Warning : all fields before dataref are cleared in __alloc_skb() 582 */ 583 atomic_t dataref; 584 unsigned int xdp_frags_size; 585 586 /* Intermediate layers must ensure that destructor_arg 587 * remains valid until skb destructor */ 588 void * destructor_arg; 589 590 /* must be last field, see pskb_expand_head() */ 591 skb_frag_t frags[MAX_SKB_FRAGS]; 592 }; 593 594 /** 595 * DOC: dataref and headerless skbs 596 * 597 * Transport layers send out clones of payload skbs they hold for 598 * retransmissions. To allow lower layers of the stack to prepend their headers 599 * we split &skb_shared_info.dataref into two halves. 600 * The lower 16 bits count the overall number of references. 601 * The higher 16 bits indicate how many of the references are payload-only. 602 * skb_header_cloned() checks if skb is allowed to add / write the headers. 603 * 604 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr 605 * (via __skb_header_release()). Any clone created from marked skb will get 606 * &sk_buff.hdr_len populated with the available headroom. 607 * If there's the only clone in existence it's able to modify the headroom 608 * at will. The sequence of calls inside the transport layer is:: 609 * 610 * <alloc skb> 611 * skb_reserve() 612 * __skb_header_release() 613 * skb_clone() 614 * // send the clone down the stack 615 * 616 * This is not a very generic construct and it depends on the transport layers 617 * doing the right thing. In practice there's usually only one payload-only skb. 618 * Having multiple payload-only skbs with different lengths of hdr_len is not 619 * possible. The payload-only skbs should never leave their owner. 620 */ 621 #define SKB_DATAREF_SHIFT 16 622 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 623 624 625 enum { 626 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 627 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 628 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 629 }; 630 631 enum { 632 SKB_GSO_TCPV4 = 1 << 0, 633 634 /* This indicates the skb is from an untrusted source. */ 635 SKB_GSO_DODGY = 1 << 1, 636 637 /* This indicates the tcp segment has CWR set. */ 638 SKB_GSO_TCP_ECN = 1 << 2, 639 640 SKB_GSO_TCP_FIXEDID = 1 << 3, 641 642 SKB_GSO_TCPV6 = 1 << 4, 643 644 SKB_GSO_FCOE = 1 << 5, 645 646 SKB_GSO_GRE = 1 << 6, 647 648 SKB_GSO_GRE_CSUM = 1 << 7, 649 650 SKB_GSO_IPXIP4 = 1 << 8, 651 652 SKB_GSO_IPXIP6 = 1 << 9, 653 654 SKB_GSO_UDP_TUNNEL = 1 << 10, 655 656 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 657 658 SKB_GSO_PARTIAL = 1 << 12, 659 660 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 661 662 SKB_GSO_SCTP = 1 << 14, 663 664 SKB_GSO_ESP = 1 << 15, 665 666 SKB_GSO_UDP = 1 << 16, 667 668 SKB_GSO_UDP_L4 = 1 << 17, 669 670 SKB_GSO_FRAGLIST = 1 << 18, 671 }; 672 673 #if BITS_PER_LONG > 32 674 #define NET_SKBUFF_DATA_USES_OFFSET 1 675 #endif 676 677 #ifdef NET_SKBUFF_DATA_USES_OFFSET 678 typedef unsigned int sk_buff_data_t; 679 #else 680 typedef unsigned char *sk_buff_data_t; 681 #endif 682 683 /** 684 * DOC: Basic sk_buff geometry 685 * 686 * struct sk_buff itself is a metadata structure and does not hold any packet 687 * data. All the data is held in associated buffers. 688 * 689 * &sk_buff.head points to the main "head" buffer. The head buffer is divided 690 * into two parts: 691 * 692 * - data buffer, containing headers and sometimes payload; 693 * this is the part of the skb operated on by the common helpers 694 * such as skb_put() or skb_pull(); 695 * - shared info (struct skb_shared_info) which holds an array of pointers 696 * to read-only data in the (page, offset, length) format. 697 * 698 * Optionally &skb_shared_info.frag_list may point to another skb. 699 * 700 * Basic diagram may look like this:: 701 * 702 * --------------- 703 * | sk_buff | 704 * --------------- 705 * ,--------------------------- + head 706 * / ,----------------- + data 707 * / / ,----------- + tail 708 * | | | , + end 709 * | | | | 710 * v v v v 711 * ----------------------------------------------- 712 * | headroom | data | tailroom | skb_shared_info | 713 * ----------------------------------------------- 714 * + [page frag] 715 * + [page frag] 716 * + [page frag] 717 * + [page frag] --------- 718 * + frag_list --> | sk_buff | 719 * --------- 720 * 721 */ 722 723 /** 724 * struct sk_buff - socket buffer 725 * @next: Next buffer in list 726 * @prev: Previous buffer in list 727 * @tstamp: Time we arrived/left 728 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 729 * for retransmit timer 730 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 731 * @list: queue head 732 * @ll_node: anchor in an llist (eg socket defer_list) 733 * @sk: Socket we are owned by 734 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in 735 * fragmentation management 736 * @dev: Device we arrived on/are leaving by 737 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 738 * @cb: Control buffer. Free for use by every layer. Put private vars here 739 * @_skb_refdst: destination entry (with norefcount bit) 740 * @sp: the security path, used for xfrm 741 * @len: Length of actual data 742 * @data_len: Data length 743 * @mac_len: Length of link layer header 744 * @hdr_len: writable header length of cloned skb 745 * @csum: Checksum (must include start/offset pair) 746 * @csum_start: Offset from skb->head where checksumming should start 747 * @csum_offset: Offset from csum_start where checksum should be stored 748 * @priority: Packet queueing priority 749 * @ignore_df: allow local fragmentation 750 * @cloned: Head may be cloned (check refcnt to be sure) 751 * @ip_summed: Driver fed us an IP checksum 752 * @nohdr: Payload reference only, must not modify header 753 * @pkt_type: Packet class 754 * @fclone: skbuff clone status 755 * @ipvs_property: skbuff is owned by ipvs 756 * @inner_protocol_type: whether the inner protocol is 757 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 758 * @remcsum_offload: remote checksum offload is enabled 759 * @offload_fwd_mark: Packet was L2-forwarded in hardware 760 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 761 * @tc_skip_classify: do not classify packet. set by IFB device 762 * @tc_at_ingress: used within tc_classify to distinguish in/egress 763 * @redirected: packet was redirected by packet classifier 764 * @from_ingress: packet was redirected from the ingress path 765 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 766 * @peeked: this packet has been seen already, so stats have been 767 * done for it, don't do them again 768 * @nf_trace: netfilter packet trace flag 769 * @protocol: Packet protocol from driver 770 * @destructor: Destruct function 771 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 772 * @_sk_redir: socket redirection information for skmsg 773 * @_nfct: Associated connection, if any (with nfctinfo bits) 774 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 775 * @skb_iif: ifindex of device we arrived on 776 * @tc_index: Traffic control index 777 * @hash: the packet hash 778 * @queue_mapping: Queue mapping for multiqueue devices 779 * @head_frag: skb was allocated from page fragments, 780 * not allocated by kmalloc() or vmalloc(). 781 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 782 * @pp_recycle: mark the packet for recycling instead of freeing (implies 783 * page_pool support on driver) 784 * @active_extensions: active extensions (skb_ext_id types) 785 * @ndisc_nodetype: router type (from link layer) 786 * @ooo_okay: allow the mapping of a socket to a queue to be changed 787 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 788 * ports. 789 * @sw_hash: indicates hash was computed in software stack 790 * @wifi_acked_valid: wifi_acked was set 791 * @wifi_acked: whether frame was acked on wifi or not 792 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 793 * @encapsulation: indicates the inner headers in the skbuff are valid 794 * @encap_hdr_csum: software checksum is needed 795 * @csum_valid: checksum is already valid 796 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 797 * @csum_complete_sw: checksum was completed by software 798 * @csum_level: indicates the number of consecutive checksums found in 799 * the packet minus one that have been verified as 800 * CHECKSUM_UNNECESSARY (max 3) 801 * @scm_io_uring: SKB holds io_uring registered files 802 * @dst_pending_confirm: need to confirm neighbour 803 * @decrypted: Decrypted SKB 804 * @slow_gro: state present at GRO time, slower prepare step required 805 * @mono_delivery_time: When set, skb->tstamp has the 806 * delivery_time in mono clock base (i.e. EDT). Otherwise, the 807 * skb->tstamp has the (rcv) timestamp at ingress and 808 * delivery_time at egress. 809 * @napi_id: id of the NAPI struct this skb came from 810 * @sender_cpu: (aka @napi_id) source CPU in XPS 811 * @alloc_cpu: CPU which did the skb allocation. 812 * @secmark: security marking 813 * @mark: Generic packet mark 814 * @reserved_tailroom: (aka @mark) number of bytes of free space available 815 * at the tail of an sk_buff 816 * @vlan_all: vlan fields (proto & tci) 817 * @vlan_proto: vlan encapsulation protocol 818 * @vlan_tci: vlan tag control information 819 * @inner_protocol: Protocol (encapsulation) 820 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 821 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 822 * @inner_transport_header: Inner transport layer header (encapsulation) 823 * @inner_network_header: Network layer header (encapsulation) 824 * @inner_mac_header: Link layer header (encapsulation) 825 * @transport_header: Transport layer header 826 * @network_header: Network layer header 827 * @mac_header: Link layer header 828 * @kcov_handle: KCOV remote handle for remote coverage collection 829 * @tail: Tail pointer 830 * @end: End pointer 831 * @head: Head of buffer 832 * @data: Data head pointer 833 * @truesize: Buffer size 834 * @users: User count - see {datagram,tcp}.c 835 * @extensions: allocated extensions, valid if active_extensions is nonzero 836 */ 837 838 struct sk_buff { 839 union { 840 struct { 841 /* These two members must be first to match sk_buff_head. */ 842 struct sk_buff *next; 843 struct sk_buff *prev; 844 845 union { 846 struct net_device *dev; 847 /* Some protocols might use this space to store information, 848 * while device pointer would be NULL. 849 * UDP receive path is one user. 850 */ 851 unsigned long dev_scratch; 852 }; 853 }; 854 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 855 struct list_head list; 856 struct llist_node ll_node; 857 }; 858 859 union { 860 struct sock *sk; 861 int ip_defrag_offset; 862 }; 863 864 union { 865 ktime_t tstamp; 866 u64 skb_mstamp_ns; /* earliest departure time */ 867 }; 868 /* 869 * This is the control buffer. It is free to use for every 870 * layer. Please put your private variables there. If you 871 * want to keep them across layers you have to do a skb_clone() 872 * first. This is owned by whoever has the skb queued ATM. 873 */ 874 char cb[48] __aligned(8); 875 876 union { 877 struct { 878 unsigned long _skb_refdst; 879 void (*destructor)(struct sk_buff *skb); 880 }; 881 struct list_head tcp_tsorted_anchor; 882 #ifdef CONFIG_NET_SOCK_MSG 883 unsigned long _sk_redir; 884 #endif 885 }; 886 887 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 888 unsigned long _nfct; 889 #endif 890 unsigned int len, 891 data_len; 892 __u16 mac_len, 893 hdr_len; 894 895 /* Following fields are _not_ copied in __copy_skb_header() 896 * Note that queue_mapping is here mostly to fill a hole. 897 */ 898 __u16 queue_mapping; 899 900 /* if you move cloned around you also must adapt those constants */ 901 #ifdef __BIG_ENDIAN_BITFIELD 902 #define CLONED_MASK (1 << 7) 903 #else 904 #define CLONED_MASK 1 905 #endif 906 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 907 908 /* private: */ 909 __u8 __cloned_offset[0]; 910 /* public: */ 911 __u8 cloned:1, 912 nohdr:1, 913 fclone:2, 914 peeked:1, 915 head_frag:1, 916 pfmemalloc:1, 917 pp_recycle:1; /* page_pool recycle indicator */ 918 #ifdef CONFIG_SKB_EXTENSIONS 919 __u8 active_extensions; 920 #endif 921 922 /* Fields enclosed in headers group are copied 923 * using a single memcpy() in __copy_skb_header() 924 */ 925 struct_group(headers, 926 927 /* private: */ 928 __u8 __pkt_type_offset[0]; 929 /* public: */ 930 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 931 __u8 ignore_df:1; 932 __u8 nf_trace:1; 933 __u8 ip_summed:2; 934 __u8 ooo_okay:1; 935 936 __u8 l4_hash:1; 937 __u8 sw_hash:1; 938 __u8 wifi_acked_valid:1; 939 __u8 wifi_acked:1; 940 __u8 no_fcs:1; 941 /* Indicates the inner headers are valid in the skbuff. */ 942 __u8 encapsulation:1; 943 __u8 encap_hdr_csum:1; 944 __u8 csum_valid:1; 945 946 /* private: */ 947 __u8 __pkt_vlan_present_offset[0]; 948 /* public: */ 949 __u8 remcsum_offload:1; 950 __u8 csum_complete_sw:1; 951 __u8 csum_level:2; 952 __u8 dst_pending_confirm:1; 953 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */ 954 #ifdef CONFIG_NET_CLS_ACT 955 __u8 tc_skip_classify:1; 956 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 957 #endif 958 #ifdef CONFIG_IPV6_NDISC_NODETYPE 959 __u8 ndisc_nodetype:2; 960 #endif 961 962 __u8 ipvs_property:1; 963 __u8 inner_protocol_type:1; 964 #ifdef CONFIG_NET_SWITCHDEV 965 __u8 offload_fwd_mark:1; 966 __u8 offload_l3_fwd_mark:1; 967 #endif 968 __u8 redirected:1; 969 #ifdef CONFIG_NET_REDIRECT 970 __u8 from_ingress:1; 971 #endif 972 #ifdef CONFIG_NETFILTER_SKIP_EGRESS 973 __u8 nf_skip_egress:1; 974 #endif 975 #ifdef CONFIG_TLS_DEVICE 976 __u8 decrypted:1; 977 #endif 978 __u8 slow_gro:1; 979 __u8 csum_not_inet:1; 980 __u8 scm_io_uring:1; 981 982 #ifdef CONFIG_NET_SCHED 983 __u16 tc_index; /* traffic control index */ 984 #endif 985 986 union { 987 __wsum csum; 988 struct { 989 __u16 csum_start; 990 __u16 csum_offset; 991 }; 992 }; 993 __u32 priority; 994 int skb_iif; 995 __u32 hash; 996 union { 997 u32 vlan_all; 998 struct { 999 __be16 vlan_proto; 1000 __u16 vlan_tci; 1001 }; 1002 }; 1003 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1004 union { 1005 unsigned int napi_id; 1006 unsigned int sender_cpu; 1007 }; 1008 #endif 1009 u16 alloc_cpu; 1010 #ifdef CONFIG_NETWORK_SECMARK 1011 __u32 secmark; 1012 #endif 1013 1014 union { 1015 __u32 mark; 1016 __u32 reserved_tailroom; 1017 }; 1018 1019 union { 1020 __be16 inner_protocol; 1021 __u8 inner_ipproto; 1022 }; 1023 1024 __u16 inner_transport_header; 1025 __u16 inner_network_header; 1026 __u16 inner_mac_header; 1027 1028 __be16 protocol; 1029 __u16 transport_header; 1030 __u16 network_header; 1031 __u16 mac_header; 1032 1033 #ifdef CONFIG_KCOV 1034 u64 kcov_handle; 1035 #endif 1036 1037 ); /* end headers group */ 1038 1039 /* These elements must be at the end, see alloc_skb() for details. */ 1040 sk_buff_data_t tail; 1041 sk_buff_data_t end; 1042 unsigned char *head, 1043 *data; 1044 unsigned int truesize; 1045 refcount_t users; 1046 1047 #ifdef CONFIG_SKB_EXTENSIONS 1048 /* only useable after checking ->active_extensions != 0 */ 1049 struct skb_ext *extensions; 1050 #endif 1051 }; 1052 1053 /* if you move pkt_type around you also must adapt those constants */ 1054 #ifdef __BIG_ENDIAN_BITFIELD 1055 #define PKT_TYPE_MAX (7 << 5) 1056 #else 1057 #define PKT_TYPE_MAX 7 1058 #endif 1059 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1060 1061 /* if you move tc_at_ingress or mono_delivery_time 1062 * around, you also must adapt these constants. 1063 */ 1064 #ifdef __BIG_ENDIAN_BITFIELD 1065 #define TC_AT_INGRESS_MASK (1 << 0) 1066 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 2) 1067 #else 1068 #define TC_AT_INGRESS_MASK (1 << 7) 1069 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 5) 1070 #endif 1071 #define PKT_VLAN_PRESENT_OFFSET offsetof(struct sk_buff, __pkt_vlan_present_offset) 1072 1073 #ifdef __KERNEL__ 1074 /* 1075 * Handling routines are only of interest to the kernel 1076 */ 1077 1078 #define SKB_ALLOC_FCLONE 0x01 1079 #define SKB_ALLOC_RX 0x02 1080 #define SKB_ALLOC_NAPI 0x04 1081 1082 /** 1083 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1084 * @skb: buffer 1085 */ 1086 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1087 { 1088 return unlikely(skb->pfmemalloc); 1089 } 1090 1091 /* 1092 * skb might have a dst pointer attached, refcounted or not. 1093 * _skb_refdst low order bit is set if refcount was _not_ taken 1094 */ 1095 #define SKB_DST_NOREF 1UL 1096 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1097 1098 /** 1099 * skb_dst - returns skb dst_entry 1100 * @skb: buffer 1101 * 1102 * Returns skb dst_entry, regardless of reference taken or not. 1103 */ 1104 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1105 { 1106 /* If refdst was not refcounted, check we still are in a 1107 * rcu_read_lock section 1108 */ 1109 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1110 !rcu_read_lock_held() && 1111 !rcu_read_lock_bh_held()); 1112 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1113 } 1114 1115 /** 1116 * skb_dst_set - sets skb dst 1117 * @skb: buffer 1118 * @dst: dst entry 1119 * 1120 * Sets skb dst, assuming a reference was taken on dst and should 1121 * be released by skb_dst_drop() 1122 */ 1123 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1124 { 1125 skb->slow_gro |= !!dst; 1126 skb->_skb_refdst = (unsigned long)dst; 1127 } 1128 1129 /** 1130 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1131 * @skb: buffer 1132 * @dst: dst entry 1133 * 1134 * Sets skb dst, assuming a reference was not taken on dst. 1135 * If dst entry is cached, we do not take reference and dst_release 1136 * will be avoided by refdst_drop. If dst entry is not cached, we take 1137 * reference, so that last dst_release can destroy the dst immediately. 1138 */ 1139 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1140 { 1141 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1142 skb->slow_gro |= !!dst; 1143 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1144 } 1145 1146 /** 1147 * skb_dst_is_noref - Test if skb dst isn't refcounted 1148 * @skb: buffer 1149 */ 1150 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1151 { 1152 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1153 } 1154 1155 /** 1156 * skb_rtable - Returns the skb &rtable 1157 * @skb: buffer 1158 */ 1159 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 1160 { 1161 return (struct rtable *)skb_dst(skb); 1162 } 1163 1164 /* For mangling skb->pkt_type from user space side from applications 1165 * such as nft, tc, etc, we only allow a conservative subset of 1166 * possible pkt_types to be set. 1167 */ 1168 static inline bool skb_pkt_type_ok(u32 ptype) 1169 { 1170 return ptype <= PACKET_OTHERHOST; 1171 } 1172 1173 /** 1174 * skb_napi_id - Returns the skb's NAPI id 1175 * @skb: buffer 1176 */ 1177 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1178 { 1179 #ifdef CONFIG_NET_RX_BUSY_POLL 1180 return skb->napi_id; 1181 #else 1182 return 0; 1183 #endif 1184 } 1185 1186 /** 1187 * skb_unref - decrement the skb's reference count 1188 * @skb: buffer 1189 * 1190 * Returns true if we can free the skb. 1191 */ 1192 static inline bool skb_unref(struct sk_buff *skb) 1193 { 1194 if (unlikely(!skb)) 1195 return false; 1196 if (likely(refcount_read(&skb->users) == 1)) 1197 smp_rmb(); 1198 else if (likely(!refcount_dec_and_test(&skb->users))) 1199 return false; 1200 1201 return true; 1202 } 1203 1204 void __fix_address 1205 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason); 1206 1207 /** 1208 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1209 * @skb: buffer to free 1210 */ 1211 static inline void kfree_skb(struct sk_buff *skb) 1212 { 1213 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1214 } 1215 1216 void skb_release_head_state(struct sk_buff *skb); 1217 void kfree_skb_list_reason(struct sk_buff *segs, 1218 enum skb_drop_reason reason); 1219 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1220 void skb_tx_error(struct sk_buff *skb); 1221 1222 static inline void kfree_skb_list(struct sk_buff *segs) 1223 { 1224 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1225 } 1226 1227 #ifdef CONFIG_TRACEPOINTS 1228 void consume_skb(struct sk_buff *skb); 1229 #else 1230 static inline void consume_skb(struct sk_buff *skb) 1231 { 1232 return kfree_skb(skb); 1233 } 1234 #endif 1235 1236 void __consume_stateless_skb(struct sk_buff *skb); 1237 void __kfree_skb(struct sk_buff *skb); 1238 extern struct kmem_cache *skbuff_head_cache; 1239 1240 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1241 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1242 bool *fragstolen, int *delta_truesize); 1243 1244 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1245 int node); 1246 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1247 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1248 struct sk_buff *build_skb_around(struct sk_buff *skb, 1249 void *data, unsigned int frag_size); 1250 void skb_attempt_defer_free(struct sk_buff *skb); 1251 1252 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1253 struct sk_buff *slab_build_skb(void *data); 1254 1255 /** 1256 * alloc_skb - allocate a network buffer 1257 * @size: size to allocate 1258 * @priority: allocation mask 1259 * 1260 * This function is a convenient wrapper around __alloc_skb(). 1261 */ 1262 static inline struct sk_buff *alloc_skb(unsigned int size, 1263 gfp_t priority) 1264 { 1265 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1266 } 1267 1268 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1269 unsigned long data_len, 1270 int max_page_order, 1271 int *errcode, 1272 gfp_t gfp_mask); 1273 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1274 1275 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1276 struct sk_buff_fclones { 1277 struct sk_buff skb1; 1278 1279 struct sk_buff skb2; 1280 1281 refcount_t fclone_ref; 1282 }; 1283 1284 /** 1285 * skb_fclone_busy - check if fclone is busy 1286 * @sk: socket 1287 * @skb: buffer 1288 * 1289 * Returns true if skb is a fast clone, and its clone is not freed. 1290 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1291 * so we also check that this didnt happen. 1292 */ 1293 static inline bool skb_fclone_busy(const struct sock *sk, 1294 const struct sk_buff *skb) 1295 { 1296 const struct sk_buff_fclones *fclones; 1297 1298 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1299 1300 return skb->fclone == SKB_FCLONE_ORIG && 1301 refcount_read(&fclones->fclone_ref) > 1 && 1302 READ_ONCE(fclones->skb2.sk) == sk; 1303 } 1304 1305 /** 1306 * alloc_skb_fclone - allocate a network buffer from fclone cache 1307 * @size: size to allocate 1308 * @priority: allocation mask 1309 * 1310 * This function is a convenient wrapper around __alloc_skb(). 1311 */ 1312 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1313 gfp_t priority) 1314 { 1315 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1316 } 1317 1318 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1319 void skb_headers_offset_update(struct sk_buff *skb, int off); 1320 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1321 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1322 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1323 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1324 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1325 gfp_t gfp_mask, bool fclone); 1326 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1327 gfp_t gfp_mask) 1328 { 1329 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1330 } 1331 1332 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1333 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1334 unsigned int headroom); 1335 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1336 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1337 int newtailroom, gfp_t priority); 1338 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1339 int offset, int len); 1340 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1341 int offset, int len); 1342 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1343 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1344 1345 /** 1346 * skb_pad - zero pad the tail of an skb 1347 * @skb: buffer to pad 1348 * @pad: space to pad 1349 * 1350 * Ensure that a buffer is followed by a padding area that is zero 1351 * filled. Used by network drivers which may DMA or transfer data 1352 * beyond the buffer end onto the wire. 1353 * 1354 * May return error in out of memory cases. The skb is freed on error. 1355 */ 1356 static inline int skb_pad(struct sk_buff *skb, int pad) 1357 { 1358 return __skb_pad(skb, pad, true); 1359 } 1360 #define dev_kfree_skb(a) consume_skb(a) 1361 1362 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1363 int offset, size_t size); 1364 1365 struct skb_seq_state { 1366 __u32 lower_offset; 1367 __u32 upper_offset; 1368 __u32 frag_idx; 1369 __u32 stepped_offset; 1370 struct sk_buff *root_skb; 1371 struct sk_buff *cur_skb; 1372 __u8 *frag_data; 1373 __u32 frag_off; 1374 }; 1375 1376 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1377 unsigned int to, struct skb_seq_state *st); 1378 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1379 struct skb_seq_state *st); 1380 void skb_abort_seq_read(struct skb_seq_state *st); 1381 1382 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1383 unsigned int to, struct ts_config *config); 1384 1385 /* 1386 * Packet hash types specify the type of hash in skb_set_hash. 1387 * 1388 * Hash types refer to the protocol layer addresses which are used to 1389 * construct a packet's hash. The hashes are used to differentiate or identify 1390 * flows of the protocol layer for the hash type. Hash types are either 1391 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1392 * 1393 * Properties of hashes: 1394 * 1395 * 1) Two packets in different flows have different hash values 1396 * 2) Two packets in the same flow should have the same hash value 1397 * 1398 * A hash at a higher layer is considered to be more specific. A driver should 1399 * set the most specific hash possible. 1400 * 1401 * A driver cannot indicate a more specific hash than the layer at which a hash 1402 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1403 * 1404 * A driver may indicate a hash level which is less specific than the 1405 * actual layer the hash was computed on. For instance, a hash computed 1406 * at L4 may be considered an L3 hash. This should only be done if the 1407 * driver can't unambiguously determine that the HW computed the hash at 1408 * the higher layer. Note that the "should" in the second property above 1409 * permits this. 1410 */ 1411 enum pkt_hash_types { 1412 PKT_HASH_TYPE_NONE, /* Undefined type */ 1413 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1414 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1415 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1416 }; 1417 1418 static inline void skb_clear_hash(struct sk_buff *skb) 1419 { 1420 skb->hash = 0; 1421 skb->sw_hash = 0; 1422 skb->l4_hash = 0; 1423 } 1424 1425 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1426 { 1427 if (!skb->l4_hash) 1428 skb_clear_hash(skb); 1429 } 1430 1431 static inline void 1432 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1433 { 1434 skb->l4_hash = is_l4; 1435 skb->sw_hash = is_sw; 1436 skb->hash = hash; 1437 } 1438 1439 static inline void 1440 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1441 { 1442 /* Used by drivers to set hash from HW */ 1443 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1444 } 1445 1446 static inline void 1447 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1448 { 1449 __skb_set_hash(skb, hash, true, is_l4); 1450 } 1451 1452 void __skb_get_hash(struct sk_buff *skb); 1453 u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1454 u32 skb_get_poff(const struct sk_buff *skb); 1455 u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1456 const struct flow_keys_basic *keys, int hlen); 1457 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1458 const void *data, int hlen_proto); 1459 1460 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1461 int thoff, u8 ip_proto) 1462 { 1463 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1464 } 1465 1466 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1467 const struct flow_dissector_key *key, 1468 unsigned int key_count); 1469 1470 struct bpf_flow_dissector; 1471 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1472 __be16 proto, int nhoff, int hlen, unsigned int flags); 1473 1474 bool __skb_flow_dissect(const struct net *net, 1475 const struct sk_buff *skb, 1476 struct flow_dissector *flow_dissector, 1477 void *target_container, const void *data, 1478 __be16 proto, int nhoff, int hlen, unsigned int flags); 1479 1480 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1481 struct flow_dissector *flow_dissector, 1482 void *target_container, unsigned int flags) 1483 { 1484 return __skb_flow_dissect(NULL, skb, flow_dissector, 1485 target_container, NULL, 0, 0, 0, flags); 1486 } 1487 1488 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1489 struct flow_keys *flow, 1490 unsigned int flags) 1491 { 1492 memset(flow, 0, sizeof(*flow)); 1493 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1494 flow, NULL, 0, 0, 0, flags); 1495 } 1496 1497 static inline bool 1498 skb_flow_dissect_flow_keys_basic(const struct net *net, 1499 const struct sk_buff *skb, 1500 struct flow_keys_basic *flow, 1501 const void *data, __be16 proto, 1502 int nhoff, int hlen, unsigned int flags) 1503 { 1504 memset(flow, 0, sizeof(*flow)); 1505 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1506 data, proto, nhoff, hlen, flags); 1507 } 1508 1509 void skb_flow_dissect_meta(const struct sk_buff *skb, 1510 struct flow_dissector *flow_dissector, 1511 void *target_container); 1512 1513 /* Gets a skb connection tracking info, ctinfo map should be a 1514 * map of mapsize to translate enum ip_conntrack_info states 1515 * to user states. 1516 */ 1517 void 1518 skb_flow_dissect_ct(const struct sk_buff *skb, 1519 struct flow_dissector *flow_dissector, 1520 void *target_container, 1521 u16 *ctinfo_map, size_t mapsize, 1522 bool post_ct, u16 zone); 1523 void 1524 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1525 struct flow_dissector *flow_dissector, 1526 void *target_container); 1527 1528 void skb_flow_dissect_hash(const struct sk_buff *skb, 1529 struct flow_dissector *flow_dissector, 1530 void *target_container); 1531 1532 static inline __u32 skb_get_hash(struct sk_buff *skb) 1533 { 1534 if (!skb->l4_hash && !skb->sw_hash) 1535 __skb_get_hash(skb); 1536 1537 return skb->hash; 1538 } 1539 1540 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1541 { 1542 if (!skb->l4_hash && !skb->sw_hash) { 1543 struct flow_keys keys; 1544 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1545 1546 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1547 } 1548 1549 return skb->hash; 1550 } 1551 1552 __u32 skb_get_hash_perturb(const struct sk_buff *skb, 1553 const siphash_key_t *perturb); 1554 1555 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1556 { 1557 return skb->hash; 1558 } 1559 1560 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1561 { 1562 to->hash = from->hash; 1563 to->sw_hash = from->sw_hash; 1564 to->l4_hash = from->l4_hash; 1565 }; 1566 1567 static inline void skb_copy_decrypted(struct sk_buff *to, 1568 const struct sk_buff *from) 1569 { 1570 #ifdef CONFIG_TLS_DEVICE 1571 to->decrypted = from->decrypted; 1572 #endif 1573 } 1574 1575 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1576 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1577 { 1578 return skb->head + skb->end; 1579 } 1580 1581 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1582 { 1583 return skb->end; 1584 } 1585 1586 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1587 { 1588 skb->end = offset; 1589 } 1590 #else 1591 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1592 { 1593 return skb->end; 1594 } 1595 1596 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1597 { 1598 return skb->end - skb->head; 1599 } 1600 1601 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1602 { 1603 skb->end = skb->head + offset; 1604 } 1605 #endif 1606 1607 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1608 struct ubuf_info *uarg); 1609 1610 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1611 1612 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg, 1613 bool success); 1614 1615 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, 1616 struct sk_buff *skb, struct iov_iter *from, 1617 size_t length); 1618 1619 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1620 struct msghdr *msg, int len) 1621 { 1622 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len); 1623 } 1624 1625 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1626 struct msghdr *msg, int len, 1627 struct ubuf_info *uarg); 1628 1629 /* Internal */ 1630 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1631 1632 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1633 { 1634 return &skb_shinfo(skb)->hwtstamps; 1635 } 1636 1637 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1638 { 1639 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1640 1641 return is_zcopy ? skb_uarg(skb) : NULL; 1642 } 1643 1644 static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1645 { 1646 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1647 } 1648 1649 static inline bool skb_zcopy_managed(const struct sk_buff *skb) 1650 { 1651 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS; 1652 } 1653 1654 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1655 const struct sk_buff *skb2) 1656 { 1657 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1658 } 1659 1660 static inline void net_zcopy_get(struct ubuf_info *uarg) 1661 { 1662 refcount_inc(&uarg->refcnt); 1663 } 1664 1665 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1666 { 1667 skb_shinfo(skb)->destructor_arg = uarg; 1668 skb_shinfo(skb)->flags |= uarg->flags; 1669 } 1670 1671 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1672 bool *have_ref) 1673 { 1674 if (skb && uarg && !skb_zcopy(skb)) { 1675 if (unlikely(have_ref && *have_ref)) 1676 *have_ref = false; 1677 else 1678 net_zcopy_get(uarg); 1679 skb_zcopy_init(skb, uarg); 1680 } 1681 } 1682 1683 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1684 { 1685 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1686 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1687 } 1688 1689 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1690 { 1691 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1692 } 1693 1694 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1695 { 1696 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1697 } 1698 1699 static inline void net_zcopy_put(struct ubuf_info *uarg) 1700 { 1701 if (uarg) 1702 uarg->callback(NULL, uarg, true); 1703 } 1704 1705 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1706 { 1707 if (uarg) { 1708 if (uarg->callback == msg_zerocopy_callback) 1709 msg_zerocopy_put_abort(uarg, have_uref); 1710 else if (have_uref) 1711 net_zcopy_put(uarg); 1712 } 1713 } 1714 1715 /* Release a reference on a zerocopy structure */ 1716 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1717 { 1718 struct ubuf_info *uarg = skb_zcopy(skb); 1719 1720 if (uarg) { 1721 if (!skb_zcopy_is_nouarg(skb)) 1722 uarg->callback(skb, uarg, zerocopy_success); 1723 1724 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1725 } 1726 } 1727 1728 void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1729 1730 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1731 { 1732 if (unlikely(skb_zcopy_managed(skb))) 1733 __skb_zcopy_downgrade_managed(skb); 1734 } 1735 1736 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1737 { 1738 skb->next = NULL; 1739 } 1740 1741 static inline void skb_poison_list(struct sk_buff *skb) 1742 { 1743 #ifdef CONFIG_DEBUG_NET 1744 skb->next = SKB_LIST_POISON_NEXT; 1745 #endif 1746 } 1747 1748 /* Iterate through singly-linked GSO fragments of an skb. */ 1749 #define skb_list_walk_safe(first, skb, next_skb) \ 1750 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1751 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1752 1753 static inline void skb_list_del_init(struct sk_buff *skb) 1754 { 1755 __list_del_entry(&skb->list); 1756 skb_mark_not_on_list(skb); 1757 } 1758 1759 /** 1760 * skb_queue_empty - check if a queue is empty 1761 * @list: queue head 1762 * 1763 * Returns true if the queue is empty, false otherwise. 1764 */ 1765 static inline int skb_queue_empty(const struct sk_buff_head *list) 1766 { 1767 return list->next == (const struct sk_buff *) list; 1768 } 1769 1770 /** 1771 * skb_queue_empty_lockless - check if a queue is empty 1772 * @list: queue head 1773 * 1774 * Returns true if the queue is empty, false otherwise. 1775 * This variant can be used in lockless contexts. 1776 */ 1777 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1778 { 1779 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1780 } 1781 1782 1783 /** 1784 * skb_queue_is_last - check if skb is the last entry in the queue 1785 * @list: queue head 1786 * @skb: buffer 1787 * 1788 * Returns true if @skb is the last buffer on the list. 1789 */ 1790 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1791 const struct sk_buff *skb) 1792 { 1793 return skb->next == (const struct sk_buff *) list; 1794 } 1795 1796 /** 1797 * skb_queue_is_first - check if skb is the first entry in the queue 1798 * @list: queue head 1799 * @skb: buffer 1800 * 1801 * Returns true if @skb is the first buffer on the list. 1802 */ 1803 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1804 const struct sk_buff *skb) 1805 { 1806 return skb->prev == (const struct sk_buff *) list; 1807 } 1808 1809 /** 1810 * skb_queue_next - return the next packet in the queue 1811 * @list: queue head 1812 * @skb: current buffer 1813 * 1814 * Return the next packet in @list after @skb. It is only valid to 1815 * call this if skb_queue_is_last() evaluates to false. 1816 */ 1817 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1818 const struct sk_buff *skb) 1819 { 1820 /* This BUG_ON may seem severe, but if we just return then we 1821 * are going to dereference garbage. 1822 */ 1823 BUG_ON(skb_queue_is_last(list, skb)); 1824 return skb->next; 1825 } 1826 1827 /** 1828 * skb_queue_prev - return the prev packet in the queue 1829 * @list: queue head 1830 * @skb: current buffer 1831 * 1832 * Return the prev packet in @list before @skb. It is only valid to 1833 * call this if skb_queue_is_first() evaluates to false. 1834 */ 1835 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1836 const struct sk_buff *skb) 1837 { 1838 /* This BUG_ON may seem severe, but if we just return then we 1839 * are going to dereference garbage. 1840 */ 1841 BUG_ON(skb_queue_is_first(list, skb)); 1842 return skb->prev; 1843 } 1844 1845 /** 1846 * skb_get - reference buffer 1847 * @skb: buffer to reference 1848 * 1849 * Makes another reference to a socket buffer and returns a pointer 1850 * to the buffer. 1851 */ 1852 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1853 { 1854 refcount_inc(&skb->users); 1855 return skb; 1856 } 1857 1858 /* 1859 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1860 */ 1861 1862 /** 1863 * skb_cloned - is the buffer a clone 1864 * @skb: buffer to check 1865 * 1866 * Returns true if the buffer was generated with skb_clone() and is 1867 * one of multiple shared copies of the buffer. Cloned buffers are 1868 * shared data so must not be written to under normal circumstances. 1869 */ 1870 static inline int skb_cloned(const struct sk_buff *skb) 1871 { 1872 return skb->cloned && 1873 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1874 } 1875 1876 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1877 { 1878 might_sleep_if(gfpflags_allow_blocking(pri)); 1879 1880 if (skb_cloned(skb)) 1881 return pskb_expand_head(skb, 0, 0, pri); 1882 1883 return 0; 1884 } 1885 1886 /* This variant of skb_unclone() makes sure skb->truesize 1887 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 1888 * 1889 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 1890 * when various debugging features are in place. 1891 */ 1892 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 1893 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 1894 { 1895 might_sleep_if(gfpflags_allow_blocking(pri)); 1896 1897 if (skb_cloned(skb)) 1898 return __skb_unclone_keeptruesize(skb, pri); 1899 return 0; 1900 } 1901 1902 /** 1903 * skb_header_cloned - is the header a clone 1904 * @skb: buffer to check 1905 * 1906 * Returns true if modifying the header part of the buffer requires 1907 * the data to be copied. 1908 */ 1909 static inline int skb_header_cloned(const struct sk_buff *skb) 1910 { 1911 int dataref; 1912 1913 if (!skb->cloned) 1914 return 0; 1915 1916 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1917 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1918 return dataref != 1; 1919 } 1920 1921 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1922 { 1923 might_sleep_if(gfpflags_allow_blocking(pri)); 1924 1925 if (skb_header_cloned(skb)) 1926 return pskb_expand_head(skb, 0, 0, pri); 1927 1928 return 0; 1929 } 1930 1931 /** 1932 * __skb_header_release() - allow clones to use the headroom 1933 * @skb: buffer to operate on 1934 * 1935 * See "DOC: dataref and headerless skbs". 1936 */ 1937 static inline void __skb_header_release(struct sk_buff *skb) 1938 { 1939 skb->nohdr = 1; 1940 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1941 } 1942 1943 1944 /** 1945 * skb_shared - is the buffer shared 1946 * @skb: buffer to check 1947 * 1948 * Returns true if more than one person has a reference to this 1949 * buffer. 1950 */ 1951 static inline int skb_shared(const struct sk_buff *skb) 1952 { 1953 return refcount_read(&skb->users) != 1; 1954 } 1955 1956 /** 1957 * skb_share_check - check if buffer is shared and if so clone it 1958 * @skb: buffer to check 1959 * @pri: priority for memory allocation 1960 * 1961 * If the buffer is shared the buffer is cloned and the old copy 1962 * drops a reference. A new clone with a single reference is returned. 1963 * If the buffer is not shared the original buffer is returned. When 1964 * being called from interrupt status or with spinlocks held pri must 1965 * be GFP_ATOMIC. 1966 * 1967 * NULL is returned on a memory allocation failure. 1968 */ 1969 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1970 { 1971 might_sleep_if(gfpflags_allow_blocking(pri)); 1972 if (skb_shared(skb)) { 1973 struct sk_buff *nskb = skb_clone(skb, pri); 1974 1975 if (likely(nskb)) 1976 consume_skb(skb); 1977 else 1978 kfree_skb(skb); 1979 skb = nskb; 1980 } 1981 return skb; 1982 } 1983 1984 /* 1985 * Copy shared buffers into a new sk_buff. We effectively do COW on 1986 * packets to handle cases where we have a local reader and forward 1987 * and a couple of other messy ones. The normal one is tcpdumping 1988 * a packet thats being forwarded. 1989 */ 1990 1991 /** 1992 * skb_unshare - make a copy of a shared buffer 1993 * @skb: buffer to check 1994 * @pri: priority for memory allocation 1995 * 1996 * If the socket buffer is a clone then this function creates a new 1997 * copy of the data, drops a reference count on the old copy and returns 1998 * the new copy with the reference count at 1. If the buffer is not a clone 1999 * the original buffer is returned. When called with a spinlock held or 2000 * from interrupt state @pri must be %GFP_ATOMIC 2001 * 2002 * %NULL is returned on a memory allocation failure. 2003 */ 2004 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2005 gfp_t pri) 2006 { 2007 might_sleep_if(gfpflags_allow_blocking(pri)); 2008 if (skb_cloned(skb)) { 2009 struct sk_buff *nskb = skb_copy(skb, pri); 2010 2011 /* Free our shared copy */ 2012 if (likely(nskb)) 2013 consume_skb(skb); 2014 else 2015 kfree_skb(skb); 2016 skb = nskb; 2017 } 2018 return skb; 2019 } 2020 2021 /** 2022 * skb_peek - peek at the head of an &sk_buff_head 2023 * @list_: list to peek at 2024 * 2025 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2026 * be careful with this one. A peek leaves the buffer on the 2027 * list and someone else may run off with it. You must hold 2028 * the appropriate locks or have a private queue to do this. 2029 * 2030 * Returns %NULL for an empty list or a pointer to the head element. 2031 * The reference count is not incremented and the reference is therefore 2032 * volatile. Use with caution. 2033 */ 2034 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2035 { 2036 struct sk_buff *skb = list_->next; 2037 2038 if (skb == (struct sk_buff *)list_) 2039 skb = NULL; 2040 return skb; 2041 } 2042 2043 /** 2044 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2045 * @list_: list to peek at 2046 * 2047 * Like skb_peek(), but the caller knows that the list is not empty. 2048 */ 2049 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2050 { 2051 return list_->next; 2052 } 2053 2054 /** 2055 * skb_peek_next - peek skb following the given one from a queue 2056 * @skb: skb to start from 2057 * @list_: list to peek at 2058 * 2059 * Returns %NULL when the end of the list is met or a pointer to the 2060 * next element. The reference count is not incremented and the 2061 * reference is therefore volatile. Use with caution. 2062 */ 2063 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2064 const struct sk_buff_head *list_) 2065 { 2066 struct sk_buff *next = skb->next; 2067 2068 if (next == (struct sk_buff *)list_) 2069 next = NULL; 2070 return next; 2071 } 2072 2073 /** 2074 * skb_peek_tail - peek at the tail of an &sk_buff_head 2075 * @list_: list to peek at 2076 * 2077 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2078 * be careful with this one. A peek leaves the buffer on the 2079 * list and someone else may run off with it. You must hold 2080 * the appropriate locks or have a private queue to do this. 2081 * 2082 * Returns %NULL for an empty list or a pointer to the tail element. 2083 * The reference count is not incremented and the reference is therefore 2084 * volatile. Use with caution. 2085 */ 2086 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2087 { 2088 struct sk_buff *skb = READ_ONCE(list_->prev); 2089 2090 if (skb == (struct sk_buff *)list_) 2091 skb = NULL; 2092 return skb; 2093 2094 } 2095 2096 /** 2097 * skb_queue_len - get queue length 2098 * @list_: list to measure 2099 * 2100 * Return the length of an &sk_buff queue. 2101 */ 2102 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2103 { 2104 return list_->qlen; 2105 } 2106 2107 /** 2108 * skb_queue_len_lockless - get queue length 2109 * @list_: list to measure 2110 * 2111 * Return the length of an &sk_buff queue. 2112 * This variant can be used in lockless contexts. 2113 */ 2114 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2115 { 2116 return READ_ONCE(list_->qlen); 2117 } 2118 2119 /** 2120 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2121 * @list: queue to initialize 2122 * 2123 * This initializes only the list and queue length aspects of 2124 * an sk_buff_head object. This allows to initialize the list 2125 * aspects of an sk_buff_head without reinitializing things like 2126 * the spinlock. It can also be used for on-stack sk_buff_head 2127 * objects where the spinlock is known to not be used. 2128 */ 2129 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2130 { 2131 list->prev = list->next = (struct sk_buff *)list; 2132 list->qlen = 0; 2133 } 2134 2135 /* 2136 * This function creates a split out lock class for each invocation; 2137 * this is needed for now since a whole lot of users of the skb-queue 2138 * infrastructure in drivers have different locking usage (in hardirq) 2139 * than the networking core (in softirq only). In the long run either the 2140 * network layer or drivers should need annotation to consolidate the 2141 * main types of usage into 3 classes. 2142 */ 2143 static inline void skb_queue_head_init(struct sk_buff_head *list) 2144 { 2145 spin_lock_init(&list->lock); 2146 __skb_queue_head_init(list); 2147 } 2148 2149 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2150 struct lock_class_key *class) 2151 { 2152 skb_queue_head_init(list); 2153 lockdep_set_class(&list->lock, class); 2154 } 2155 2156 /* 2157 * Insert an sk_buff on a list. 2158 * 2159 * The "__skb_xxxx()" functions are the non-atomic ones that 2160 * can only be called with interrupts disabled. 2161 */ 2162 static inline void __skb_insert(struct sk_buff *newsk, 2163 struct sk_buff *prev, struct sk_buff *next, 2164 struct sk_buff_head *list) 2165 { 2166 /* See skb_queue_empty_lockless() and skb_peek_tail() 2167 * for the opposite READ_ONCE() 2168 */ 2169 WRITE_ONCE(newsk->next, next); 2170 WRITE_ONCE(newsk->prev, prev); 2171 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2172 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2173 WRITE_ONCE(list->qlen, list->qlen + 1); 2174 } 2175 2176 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2177 struct sk_buff *prev, 2178 struct sk_buff *next) 2179 { 2180 struct sk_buff *first = list->next; 2181 struct sk_buff *last = list->prev; 2182 2183 WRITE_ONCE(first->prev, prev); 2184 WRITE_ONCE(prev->next, first); 2185 2186 WRITE_ONCE(last->next, next); 2187 WRITE_ONCE(next->prev, last); 2188 } 2189 2190 /** 2191 * skb_queue_splice - join two skb lists, this is designed for stacks 2192 * @list: the new list to add 2193 * @head: the place to add it in the first list 2194 */ 2195 static inline void skb_queue_splice(const struct sk_buff_head *list, 2196 struct sk_buff_head *head) 2197 { 2198 if (!skb_queue_empty(list)) { 2199 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2200 head->qlen += list->qlen; 2201 } 2202 } 2203 2204 /** 2205 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2206 * @list: the new list to add 2207 * @head: the place to add it in the first list 2208 * 2209 * The list at @list is reinitialised 2210 */ 2211 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2212 struct sk_buff_head *head) 2213 { 2214 if (!skb_queue_empty(list)) { 2215 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2216 head->qlen += list->qlen; 2217 __skb_queue_head_init(list); 2218 } 2219 } 2220 2221 /** 2222 * skb_queue_splice_tail - join two skb lists, each list being a queue 2223 * @list: the new list to add 2224 * @head: the place to add it in the first list 2225 */ 2226 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2227 struct sk_buff_head *head) 2228 { 2229 if (!skb_queue_empty(list)) { 2230 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2231 head->qlen += list->qlen; 2232 } 2233 } 2234 2235 /** 2236 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2237 * @list: the new list to add 2238 * @head: the place to add it in the first list 2239 * 2240 * Each of the lists is a queue. 2241 * The list at @list is reinitialised 2242 */ 2243 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2244 struct sk_buff_head *head) 2245 { 2246 if (!skb_queue_empty(list)) { 2247 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2248 head->qlen += list->qlen; 2249 __skb_queue_head_init(list); 2250 } 2251 } 2252 2253 /** 2254 * __skb_queue_after - queue a buffer at the list head 2255 * @list: list to use 2256 * @prev: place after this buffer 2257 * @newsk: buffer to queue 2258 * 2259 * Queue a buffer int the middle of a list. This function takes no locks 2260 * and you must therefore hold required locks before calling it. 2261 * 2262 * A buffer cannot be placed on two lists at the same time. 2263 */ 2264 static inline void __skb_queue_after(struct sk_buff_head *list, 2265 struct sk_buff *prev, 2266 struct sk_buff *newsk) 2267 { 2268 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2269 } 2270 2271 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2272 struct sk_buff_head *list); 2273 2274 static inline void __skb_queue_before(struct sk_buff_head *list, 2275 struct sk_buff *next, 2276 struct sk_buff *newsk) 2277 { 2278 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2279 } 2280 2281 /** 2282 * __skb_queue_head - queue a buffer at the list head 2283 * @list: list to use 2284 * @newsk: buffer to queue 2285 * 2286 * Queue a buffer at the start of a list. This function takes no locks 2287 * and you must therefore hold required locks before calling it. 2288 * 2289 * A buffer cannot be placed on two lists at the same time. 2290 */ 2291 static inline void __skb_queue_head(struct sk_buff_head *list, 2292 struct sk_buff *newsk) 2293 { 2294 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2295 } 2296 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2297 2298 /** 2299 * __skb_queue_tail - queue a buffer at the list tail 2300 * @list: list to use 2301 * @newsk: buffer to queue 2302 * 2303 * Queue a buffer at the end of a list. This function takes no locks 2304 * and you must therefore hold required locks before calling it. 2305 * 2306 * A buffer cannot be placed on two lists at the same time. 2307 */ 2308 static inline void __skb_queue_tail(struct sk_buff_head *list, 2309 struct sk_buff *newsk) 2310 { 2311 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2312 } 2313 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2314 2315 /* 2316 * remove sk_buff from list. _Must_ be called atomically, and with 2317 * the list known.. 2318 */ 2319 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2320 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2321 { 2322 struct sk_buff *next, *prev; 2323 2324 WRITE_ONCE(list->qlen, list->qlen - 1); 2325 next = skb->next; 2326 prev = skb->prev; 2327 skb->next = skb->prev = NULL; 2328 WRITE_ONCE(next->prev, prev); 2329 WRITE_ONCE(prev->next, next); 2330 } 2331 2332 /** 2333 * __skb_dequeue - remove from the head of the queue 2334 * @list: list to dequeue from 2335 * 2336 * Remove the head of the list. This function does not take any locks 2337 * so must be used with appropriate locks held only. The head item is 2338 * returned or %NULL if the list is empty. 2339 */ 2340 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2341 { 2342 struct sk_buff *skb = skb_peek(list); 2343 if (skb) 2344 __skb_unlink(skb, list); 2345 return skb; 2346 } 2347 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2348 2349 /** 2350 * __skb_dequeue_tail - remove from the tail of the queue 2351 * @list: list to dequeue from 2352 * 2353 * Remove the tail of the list. This function does not take any locks 2354 * so must be used with appropriate locks held only. The tail item is 2355 * returned or %NULL if the list is empty. 2356 */ 2357 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2358 { 2359 struct sk_buff *skb = skb_peek_tail(list); 2360 if (skb) 2361 __skb_unlink(skb, list); 2362 return skb; 2363 } 2364 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2365 2366 2367 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2368 { 2369 return skb->data_len; 2370 } 2371 2372 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2373 { 2374 return skb->len - skb->data_len; 2375 } 2376 2377 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2378 { 2379 unsigned int i, len = 0; 2380 2381 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2382 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2383 return len; 2384 } 2385 2386 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2387 { 2388 return skb_headlen(skb) + __skb_pagelen(skb); 2389 } 2390 2391 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2392 int i, struct page *page, 2393 int off, int size) 2394 { 2395 skb_frag_t *frag = &shinfo->frags[i]; 2396 2397 /* 2398 * Propagate page pfmemalloc to the skb if we can. The problem is 2399 * that not all callers have unique ownership of the page but rely 2400 * on page_is_pfmemalloc doing the right thing(tm). 2401 */ 2402 frag->bv_page = page; 2403 frag->bv_offset = off; 2404 skb_frag_size_set(frag, size); 2405 } 2406 2407 /** 2408 * skb_len_add - adds a number to len fields of skb 2409 * @skb: buffer to add len to 2410 * @delta: number of bytes to add 2411 */ 2412 static inline void skb_len_add(struct sk_buff *skb, int delta) 2413 { 2414 skb->len += delta; 2415 skb->data_len += delta; 2416 skb->truesize += delta; 2417 } 2418 2419 /** 2420 * __skb_fill_page_desc - initialise a paged fragment in an skb 2421 * @skb: buffer containing fragment to be initialised 2422 * @i: paged fragment index to initialise 2423 * @page: the page to use for this fragment 2424 * @off: the offset to the data with @page 2425 * @size: the length of the data 2426 * 2427 * Initialises the @i'th fragment of @skb to point to &size bytes at 2428 * offset @off within @page. 2429 * 2430 * Does not take any additional reference on the fragment. 2431 */ 2432 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2433 struct page *page, int off, int size) 2434 { 2435 __skb_fill_page_desc_noacc(skb_shinfo(skb), i, page, off, size); 2436 page = compound_head(page); 2437 if (page_is_pfmemalloc(page)) 2438 skb->pfmemalloc = true; 2439 } 2440 2441 /** 2442 * skb_fill_page_desc - initialise a paged fragment in an skb 2443 * @skb: buffer containing fragment to be initialised 2444 * @i: paged fragment index to initialise 2445 * @page: the page to use for this fragment 2446 * @off: the offset to the data with @page 2447 * @size: the length of the data 2448 * 2449 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2450 * @skb to point to @size bytes at offset @off within @page. In 2451 * addition updates @skb such that @i is the last fragment. 2452 * 2453 * Does not take any additional reference on the fragment. 2454 */ 2455 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2456 struct page *page, int off, int size) 2457 { 2458 __skb_fill_page_desc(skb, i, page, off, size); 2459 skb_shinfo(skb)->nr_frags = i + 1; 2460 } 2461 2462 /** 2463 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2464 * @skb: buffer containing fragment to be initialised 2465 * @i: paged fragment index to initialise 2466 * @page: the page to use for this fragment 2467 * @off: the offset to the data with @page 2468 * @size: the length of the data 2469 * 2470 * Variant of skb_fill_page_desc() which does not deal with 2471 * pfmemalloc, if page is not owned by us. 2472 */ 2473 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2474 struct page *page, int off, 2475 int size) 2476 { 2477 struct skb_shared_info *shinfo = skb_shinfo(skb); 2478 2479 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2480 shinfo->nr_frags = i + 1; 2481 } 2482 2483 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 2484 int size, unsigned int truesize); 2485 2486 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2487 unsigned int truesize); 2488 2489 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2490 2491 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2492 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2493 { 2494 return skb->head + skb->tail; 2495 } 2496 2497 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2498 { 2499 skb->tail = skb->data - skb->head; 2500 } 2501 2502 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2503 { 2504 skb_reset_tail_pointer(skb); 2505 skb->tail += offset; 2506 } 2507 2508 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2509 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2510 { 2511 return skb->tail; 2512 } 2513 2514 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2515 { 2516 skb->tail = skb->data; 2517 } 2518 2519 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2520 { 2521 skb->tail = skb->data + offset; 2522 } 2523 2524 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2525 2526 static inline void skb_assert_len(struct sk_buff *skb) 2527 { 2528 #ifdef CONFIG_DEBUG_NET 2529 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2530 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2531 #endif /* CONFIG_DEBUG_NET */ 2532 } 2533 2534 /* 2535 * Add data to an sk_buff 2536 */ 2537 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2538 void *skb_put(struct sk_buff *skb, unsigned int len); 2539 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2540 { 2541 void *tmp = skb_tail_pointer(skb); 2542 SKB_LINEAR_ASSERT(skb); 2543 skb->tail += len; 2544 skb->len += len; 2545 return tmp; 2546 } 2547 2548 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2549 { 2550 void *tmp = __skb_put(skb, len); 2551 2552 memset(tmp, 0, len); 2553 return tmp; 2554 } 2555 2556 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2557 unsigned int len) 2558 { 2559 void *tmp = __skb_put(skb, len); 2560 2561 memcpy(tmp, data, len); 2562 return tmp; 2563 } 2564 2565 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2566 { 2567 *(u8 *)__skb_put(skb, 1) = val; 2568 } 2569 2570 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2571 { 2572 void *tmp = skb_put(skb, len); 2573 2574 memset(tmp, 0, len); 2575 2576 return tmp; 2577 } 2578 2579 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2580 unsigned int len) 2581 { 2582 void *tmp = skb_put(skb, len); 2583 2584 memcpy(tmp, data, len); 2585 2586 return tmp; 2587 } 2588 2589 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2590 { 2591 *(u8 *)skb_put(skb, 1) = val; 2592 } 2593 2594 void *skb_push(struct sk_buff *skb, unsigned int len); 2595 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2596 { 2597 skb->data -= len; 2598 skb->len += len; 2599 return skb->data; 2600 } 2601 2602 void *skb_pull(struct sk_buff *skb, unsigned int len); 2603 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2604 { 2605 skb->len -= len; 2606 if (unlikely(skb->len < skb->data_len)) { 2607 #if defined(CONFIG_DEBUG_NET) 2608 skb->len += len; 2609 pr_err("__skb_pull(len=%u)\n", len); 2610 skb_dump(KERN_ERR, skb, false); 2611 #endif 2612 BUG(); 2613 } 2614 return skb->data += len; 2615 } 2616 2617 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2618 { 2619 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2620 } 2621 2622 void *skb_pull_data(struct sk_buff *skb, size_t len); 2623 2624 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2625 2626 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2627 { 2628 if (likely(len <= skb_headlen(skb))) 2629 return true; 2630 if (unlikely(len > skb->len)) 2631 return false; 2632 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 2633 } 2634 2635 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2636 { 2637 if (!pskb_may_pull(skb, len)) 2638 return NULL; 2639 2640 skb->len -= len; 2641 return skb->data += len; 2642 } 2643 2644 void skb_condense(struct sk_buff *skb); 2645 2646 /** 2647 * skb_headroom - bytes at buffer head 2648 * @skb: buffer to check 2649 * 2650 * Return the number of bytes of free space at the head of an &sk_buff. 2651 */ 2652 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2653 { 2654 return skb->data - skb->head; 2655 } 2656 2657 /** 2658 * skb_tailroom - bytes at buffer end 2659 * @skb: buffer to check 2660 * 2661 * Return the number of bytes of free space at the tail of an sk_buff 2662 */ 2663 static inline int skb_tailroom(const struct sk_buff *skb) 2664 { 2665 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2666 } 2667 2668 /** 2669 * skb_availroom - bytes at buffer end 2670 * @skb: buffer to check 2671 * 2672 * Return the number of bytes of free space at the tail of an sk_buff 2673 * allocated by sk_stream_alloc() 2674 */ 2675 static inline int skb_availroom(const struct sk_buff *skb) 2676 { 2677 if (skb_is_nonlinear(skb)) 2678 return 0; 2679 2680 return skb->end - skb->tail - skb->reserved_tailroom; 2681 } 2682 2683 /** 2684 * skb_reserve - adjust headroom 2685 * @skb: buffer to alter 2686 * @len: bytes to move 2687 * 2688 * Increase the headroom of an empty &sk_buff by reducing the tail 2689 * room. This is only allowed for an empty buffer. 2690 */ 2691 static inline void skb_reserve(struct sk_buff *skb, int len) 2692 { 2693 skb->data += len; 2694 skb->tail += len; 2695 } 2696 2697 /** 2698 * skb_tailroom_reserve - adjust reserved_tailroom 2699 * @skb: buffer to alter 2700 * @mtu: maximum amount of headlen permitted 2701 * @needed_tailroom: minimum amount of reserved_tailroom 2702 * 2703 * Set reserved_tailroom so that headlen can be as large as possible but 2704 * not larger than mtu and tailroom cannot be smaller than 2705 * needed_tailroom. 2706 * The required headroom should already have been reserved before using 2707 * this function. 2708 */ 2709 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2710 unsigned int needed_tailroom) 2711 { 2712 SKB_LINEAR_ASSERT(skb); 2713 if (mtu < skb_tailroom(skb) - needed_tailroom) 2714 /* use at most mtu */ 2715 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2716 else 2717 /* use up to all available space */ 2718 skb->reserved_tailroom = needed_tailroom; 2719 } 2720 2721 #define ENCAP_TYPE_ETHER 0 2722 #define ENCAP_TYPE_IPPROTO 1 2723 2724 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2725 __be16 protocol) 2726 { 2727 skb->inner_protocol = protocol; 2728 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2729 } 2730 2731 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2732 __u8 ipproto) 2733 { 2734 skb->inner_ipproto = ipproto; 2735 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2736 } 2737 2738 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2739 { 2740 skb->inner_mac_header = skb->mac_header; 2741 skb->inner_network_header = skb->network_header; 2742 skb->inner_transport_header = skb->transport_header; 2743 } 2744 2745 static inline void skb_reset_mac_len(struct sk_buff *skb) 2746 { 2747 skb->mac_len = skb->network_header - skb->mac_header; 2748 } 2749 2750 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2751 *skb) 2752 { 2753 return skb->head + skb->inner_transport_header; 2754 } 2755 2756 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2757 { 2758 return skb_inner_transport_header(skb) - skb->data; 2759 } 2760 2761 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2762 { 2763 skb->inner_transport_header = skb->data - skb->head; 2764 } 2765 2766 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2767 const int offset) 2768 { 2769 skb_reset_inner_transport_header(skb); 2770 skb->inner_transport_header += offset; 2771 } 2772 2773 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2774 { 2775 return skb->head + skb->inner_network_header; 2776 } 2777 2778 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2779 { 2780 skb->inner_network_header = skb->data - skb->head; 2781 } 2782 2783 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2784 const int offset) 2785 { 2786 skb_reset_inner_network_header(skb); 2787 skb->inner_network_header += offset; 2788 } 2789 2790 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2791 { 2792 return skb->head + skb->inner_mac_header; 2793 } 2794 2795 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2796 { 2797 skb->inner_mac_header = skb->data - skb->head; 2798 } 2799 2800 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2801 const int offset) 2802 { 2803 skb_reset_inner_mac_header(skb); 2804 skb->inner_mac_header += offset; 2805 } 2806 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2807 { 2808 return skb->transport_header != (typeof(skb->transport_header))~0U; 2809 } 2810 2811 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2812 { 2813 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 2814 return skb->head + skb->transport_header; 2815 } 2816 2817 static inline void skb_reset_transport_header(struct sk_buff *skb) 2818 { 2819 skb->transport_header = skb->data - skb->head; 2820 } 2821 2822 static inline void skb_set_transport_header(struct sk_buff *skb, 2823 const int offset) 2824 { 2825 skb_reset_transport_header(skb); 2826 skb->transport_header += offset; 2827 } 2828 2829 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2830 { 2831 return skb->head + skb->network_header; 2832 } 2833 2834 static inline void skb_reset_network_header(struct sk_buff *skb) 2835 { 2836 skb->network_header = skb->data - skb->head; 2837 } 2838 2839 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2840 { 2841 skb_reset_network_header(skb); 2842 skb->network_header += offset; 2843 } 2844 2845 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2846 { 2847 return skb->mac_header != (typeof(skb->mac_header))~0U; 2848 } 2849 2850 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2851 { 2852 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 2853 return skb->head + skb->mac_header; 2854 } 2855 2856 static inline int skb_mac_offset(const struct sk_buff *skb) 2857 { 2858 return skb_mac_header(skb) - skb->data; 2859 } 2860 2861 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2862 { 2863 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 2864 return skb->network_header - skb->mac_header; 2865 } 2866 2867 static inline void skb_unset_mac_header(struct sk_buff *skb) 2868 { 2869 skb->mac_header = (typeof(skb->mac_header))~0U; 2870 } 2871 2872 static inline void skb_reset_mac_header(struct sk_buff *skb) 2873 { 2874 skb->mac_header = skb->data - skb->head; 2875 } 2876 2877 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2878 { 2879 skb_reset_mac_header(skb); 2880 skb->mac_header += offset; 2881 } 2882 2883 static inline void skb_pop_mac_header(struct sk_buff *skb) 2884 { 2885 skb->mac_header = skb->network_header; 2886 } 2887 2888 static inline void skb_probe_transport_header(struct sk_buff *skb) 2889 { 2890 struct flow_keys_basic keys; 2891 2892 if (skb_transport_header_was_set(skb)) 2893 return; 2894 2895 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 2896 NULL, 0, 0, 0, 0)) 2897 skb_set_transport_header(skb, keys.control.thoff); 2898 } 2899 2900 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2901 { 2902 if (skb_mac_header_was_set(skb)) { 2903 const unsigned char *old_mac = skb_mac_header(skb); 2904 2905 skb_set_mac_header(skb, -skb->mac_len); 2906 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2907 } 2908 } 2909 2910 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2911 { 2912 return skb->csum_start - skb_headroom(skb); 2913 } 2914 2915 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2916 { 2917 return skb->head + skb->csum_start; 2918 } 2919 2920 static inline int skb_transport_offset(const struct sk_buff *skb) 2921 { 2922 return skb_transport_header(skb) - skb->data; 2923 } 2924 2925 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2926 { 2927 return skb->transport_header - skb->network_header; 2928 } 2929 2930 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2931 { 2932 return skb->inner_transport_header - skb->inner_network_header; 2933 } 2934 2935 static inline int skb_network_offset(const struct sk_buff *skb) 2936 { 2937 return skb_network_header(skb) - skb->data; 2938 } 2939 2940 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2941 { 2942 return skb_inner_network_header(skb) - skb->data; 2943 } 2944 2945 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2946 { 2947 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2948 } 2949 2950 /* 2951 * CPUs often take a performance hit when accessing unaligned memory 2952 * locations. The actual performance hit varies, it can be small if the 2953 * hardware handles it or large if we have to take an exception and fix it 2954 * in software. 2955 * 2956 * Since an ethernet header is 14 bytes network drivers often end up with 2957 * the IP header at an unaligned offset. The IP header can be aligned by 2958 * shifting the start of the packet by 2 bytes. Drivers should do this 2959 * with: 2960 * 2961 * skb_reserve(skb, NET_IP_ALIGN); 2962 * 2963 * The downside to this alignment of the IP header is that the DMA is now 2964 * unaligned. On some architectures the cost of an unaligned DMA is high 2965 * and this cost outweighs the gains made by aligning the IP header. 2966 * 2967 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2968 * to be overridden. 2969 */ 2970 #ifndef NET_IP_ALIGN 2971 #define NET_IP_ALIGN 2 2972 #endif 2973 2974 /* 2975 * The networking layer reserves some headroom in skb data (via 2976 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2977 * the header has to grow. In the default case, if the header has to grow 2978 * 32 bytes or less we avoid the reallocation. 2979 * 2980 * Unfortunately this headroom changes the DMA alignment of the resulting 2981 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2982 * on some architectures. An architecture can override this value, 2983 * perhaps setting it to a cacheline in size (since that will maintain 2984 * cacheline alignment of the DMA). It must be a power of 2. 2985 * 2986 * Various parts of the networking layer expect at least 32 bytes of 2987 * headroom, you should not reduce this. 2988 * 2989 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2990 * to reduce average number of cache lines per packet. 2991 * get_rps_cpu() for example only access one 64 bytes aligned block : 2992 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2993 */ 2994 #ifndef NET_SKB_PAD 2995 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2996 #endif 2997 2998 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2999 3000 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3001 { 3002 if (WARN_ON(skb_is_nonlinear(skb))) 3003 return; 3004 skb->len = len; 3005 skb_set_tail_pointer(skb, len); 3006 } 3007 3008 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3009 { 3010 __skb_set_length(skb, len); 3011 } 3012 3013 void skb_trim(struct sk_buff *skb, unsigned int len); 3014 3015 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3016 { 3017 if (skb->data_len) 3018 return ___pskb_trim(skb, len); 3019 __skb_trim(skb, len); 3020 return 0; 3021 } 3022 3023 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3024 { 3025 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3026 } 3027 3028 /** 3029 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3030 * @skb: buffer to alter 3031 * @len: new length 3032 * 3033 * This is identical to pskb_trim except that the caller knows that 3034 * the skb is not cloned so we should never get an error due to out- 3035 * of-memory. 3036 */ 3037 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3038 { 3039 int err = pskb_trim(skb, len); 3040 BUG_ON(err); 3041 } 3042 3043 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3044 { 3045 unsigned int diff = len - skb->len; 3046 3047 if (skb_tailroom(skb) < diff) { 3048 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3049 GFP_ATOMIC); 3050 if (ret) 3051 return ret; 3052 } 3053 __skb_set_length(skb, len); 3054 return 0; 3055 } 3056 3057 /** 3058 * skb_orphan - orphan a buffer 3059 * @skb: buffer to orphan 3060 * 3061 * If a buffer currently has an owner then we call the owner's 3062 * destructor function and make the @skb unowned. The buffer continues 3063 * to exist but is no longer charged to its former owner. 3064 */ 3065 static inline void skb_orphan(struct sk_buff *skb) 3066 { 3067 if (skb->destructor) { 3068 skb->destructor(skb); 3069 skb->destructor = NULL; 3070 skb->sk = NULL; 3071 } else { 3072 BUG_ON(skb->sk); 3073 } 3074 } 3075 3076 /** 3077 * skb_orphan_frags - orphan the frags contained in a buffer 3078 * @skb: buffer to orphan frags from 3079 * @gfp_mask: allocation mask for replacement pages 3080 * 3081 * For each frag in the SKB which needs a destructor (i.e. has an 3082 * owner) create a copy of that frag and release the original 3083 * page by calling the destructor. 3084 */ 3085 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3086 { 3087 if (likely(!skb_zcopy(skb))) 3088 return 0; 3089 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3090 return 0; 3091 return skb_copy_ubufs(skb, gfp_mask); 3092 } 3093 3094 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3095 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3096 { 3097 if (likely(!skb_zcopy(skb))) 3098 return 0; 3099 return skb_copy_ubufs(skb, gfp_mask); 3100 } 3101 3102 /** 3103 * __skb_queue_purge - empty a list 3104 * @list: list to empty 3105 * 3106 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3107 * the list and one reference dropped. This function does not take the 3108 * list lock and the caller must hold the relevant locks to use it. 3109 */ 3110 static inline void __skb_queue_purge(struct sk_buff_head *list) 3111 { 3112 struct sk_buff *skb; 3113 while ((skb = __skb_dequeue(list)) != NULL) 3114 kfree_skb(skb); 3115 } 3116 void skb_queue_purge(struct sk_buff_head *list); 3117 3118 unsigned int skb_rbtree_purge(struct rb_root *root); 3119 3120 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3121 3122 /** 3123 * netdev_alloc_frag - allocate a page fragment 3124 * @fragsz: fragment size 3125 * 3126 * Allocates a frag from a page for receive buffer. 3127 * Uses GFP_ATOMIC allocations. 3128 */ 3129 static inline void *netdev_alloc_frag(unsigned int fragsz) 3130 { 3131 return __netdev_alloc_frag_align(fragsz, ~0u); 3132 } 3133 3134 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3135 unsigned int align) 3136 { 3137 WARN_ON_ONCE(!is_power_of_2(align)); 3138 return __netdev_alloc_frag_align(fragsz, -align); 3139 } 3140 3141 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3142 gfp_t gfp_mask); 3143 3144 /** 3145 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3146 * @dev: network device to receive on 3147 * @length: length to allocate 3148 * 3149 * Allocate a new &sk_buff and assign it a usage count of one. The 3150 * buffer has unspecified headroom built in. Users should allocate 3151 * the headroom they think they need without accounting for the 3152 * built in space. The built in space is used for optimisations. 3153 * 3154 * %NULL is returned if there is no free memory. Although this function 3155 * allocates memory it can be called from an interrupt. 3156 */ 3157 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3158 unsigned int length) 3159 { 3160 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3161 } 3162 3163 /* legacy helper around __netdev_alloc_skb() */ 3164 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3165 gfp_t gfp_mask) 3166 { 3167 return __netdev_alloc_skb(NULL, length, gfp_mask); 3168 } 3169 3170 /* legacy helper around netdev_alloc_skb() */ 3171 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3172 { 3173 return netdev_alloc_skb(NULL, length); 3174 } 3175 3176 3177 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3178 unsigned int length, gfp_t gfp) 3179 { 3180 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3181 3182 if (NET_IP_ALIGN && skb) 3183 skb_reserve(skb, NET_IP_ALIGN); 3184 return skb; 3185 } 3186 3187 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3188 unsigned int length) 3189 { 3190 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3191 } 3192 3193 static inline void skb_free_frag(void *addr) 3194 { 3195 page_frag_free(addr); 3196 } 3197 3198 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3199 3200 static inline void *napi_alloc_frag(unsigned int fragsz) 3201 { 3202 return __napi_alloc_frag_align(fragsz, ~0u); 3203 } 3204 3205 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3206 unsigned int align) 3207 { 3208 WARN_ON_ONCE(!is_power_of_2(align)); 3209 return __napi_alloc_frag_align(fragsz, -align); 3210 } 3211 3212 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 3213 unsigned int length, gfp_t gfp_mask); 3214 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 3215 unsigned int length) 3216 { 3217 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 3218 } 3219 void napi_consume_skb(struct sk_buff *skb, int budget); 3220 3221 void napi_skb_free_stolen_head(struct sk_buff *skb); 3222 void __kfree_skb_defer(struct sk_buff *skb); 3223 3224 /** 3225 * __dev_alloc_pages - allocate page for network Rx 3226 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3227 * @order: size of the allocation 3228 * 3229 * Allocate a new page. 3230 * 3231 * %NULL is returned if there is no free memory. 3232 */ 3233 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 3234 unsigned int order) 3235 { 3236 /* This piece of code contains several assumptions. 3237 * 1. This is for device Rx, therefor a cold page is preferred. 3238 * 2. The expectation is the user wants a compound page. 3239 * 3. If requesting a order 0 page it will not be compound 3240 * due to the check to see if order has a value in prep_new_page 3241 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3242 * code in gfp_to_alloc_flags that should be enforcing this. 3243 */ 3244 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3245 3246 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 3247 } 3248 3249 static inline struct page *dev_alloc_pages(unsigned int order) 3250 { 3251 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 3252 } 3253 3254 /** 3255 * __dev_alloc_page - allocate a page for network Rx 3256 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3257 * 3258 * Allocate a new page. 3259 * 3260 * %NULL is returned if there is no free memory. 3261 */ 3262 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 3263 { 3264 return __dev_alloc_pages(gfp_mask, 0); 3265 } 3266 3267 static inline struct page *dev_alloc_page(void) 3268 { 3269 return dev_alloc_pages(0); 3270 } 3271 3272 /** 3273 * dev_page_is_reusable - check whether a page can be reused for network Rx 3274 * @page: the page to test 3275 * 3276 * A page shouldn't be considered for reusing/recycling if it was allocated 3277 * under memory pressure or at a distant memory node. 3278 * 3279 * Returns false if this page should be returned to page allocator, true 3280 * otherwise. 3281 */ 3282 static inline bool dev_page_is_reusable(const struct page *page) 3283 { 3284 return likely(page_to_nid(page) == numa_mem_id() && 3285 !page_is_pfmemalloc(page)); 3286 } 3287 3288 /** 3289 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3290 * @page: The page that was allocated from skb_alloc_page 3291 * @skb: The skb that may need pfmemalloc set 3292 */ 3293 static inline void skb_propagate_pfmemalloc(const struct page *page, 3294 struct sk_buff *skb) 3295 { 3296 if (page_is_pfmemalloc(page)) 3297 skb->pfmemalloc = true; 3298 } 3299 3300 /** 3301 * skb_frag_off() - Returns the offset of a skb fragment 3302 * @frag: the paged fragment 3303 */ 3304 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3305 { 3306 return frag->bv_offset; 3307 } 3308 3309 /** 3310 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3311 * @frag: skb fragment 3312 * @delta: value to add 3313 */ 3314 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3315 { 3316 frag->bv_offset += delta; 3317 } 3318 3319 /** 3320 * skb_frag_off_set() - Sets the offset of a skb fragment 3321 * @frag: skb fragment 3322 * @offset: offset of fragment 3323 */ 3324 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3325 { 3326 frag->bv_offset = offset; 3327 } 3328 3329 /** 3330 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3331 * @fragto: skb fragment where offset is set 3332 * @fragfrom: skb fragment offset is copied from 3333 */ 3334 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3335 const skb_frag_t *fragfrom) 3336 { 3337 fragto->bv_offset = fragfrom->bv_offset; 3338 } 3339 3340 /** 3341 * skb_frag_page - retrieve the page referred to by a paged fragment 3342 * @frag: the paged fragment 3343 * 3344 * Returns the &struct page associated with @frag. 3345 */ 3346 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3347 { 3348 return frag->bv_page; 3349 } 3350 3351 /** 3352 * __skb_frag_ref - take an addition reference on a paged fragment. 3353 * @frag: the paged fragment 3354 * 3355 * Takes an additional reference on the paged fragment @frag. 3356 */ 3357 static inline void __skb_frag_ref(skb_frag_t *frag) 3358 { 3359 get_page(skb_frag_page(frag)); 3360 } 3361 3362 /** 3363 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 3364 * @skb: the buffer 3365 * @f: the fragment offset. 3366 * 3367 * Takes an additional reference on the @f'th paged fragment of @skb. 3368 */ 3369 static inline void skb_frag_ref(struct sk_buff *skb, int f) 3370 { 3371 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 3372 } 3373 3374 /** 3375 * __skb_frag_unref - release a reference on a paged fragment. 3376 * @frag: the paged fragment 3377 * @recycle: recycle the page if allocated via page_pool 3378 * 3379 * Releases a reference on the paged fragment @frag 3380 * or recycles the page via the page_pool API. 3381 */ 3382 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle) 3383 { 3384 struct page *page = skb_frag_page(frag); 3385 3386 #ifdef CONFIG_PAGE_POOL 3387 if (recycle && page_pool_return_skb_page(page)) 3388 return; 3389 #endif 3390 put_page(page); 3391 } 3392 3393 /** 3394 * skb_frag_unref - release a reference on a paged fragment of an skb. 3395 * @skb: the buffer 3396 * @f: the fragment offset 3397 * 3398 * Releases a reference on the @f'th paged fragment of @skb. 3399 */ 3400 static inline void skb_frag_unref(struct sk_buff *skb, int f) 3401 { 3402 struct skb_shared_info *shinfo = skb_shinfo(skb); 3403 3404 if (!skb_zcopy_managed(skb)) 3405 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle); 3406 } 3407 3408 /** 3409 * skb_frag_address - gets the address of the data contained in a paged fragment 3410 * @frag: the paged fragment buffer 3411 * 3412 * Returns the address of the data within @frag. The page must already 3413 * be mapped. 3414 */ 3415 static inline void *skb_frag_address(const skb_frag_t *frag) 3416 { 3417 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3418 } 3419 3420 /** 3421 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3422 * @frag: the paged fragment buffer 3423 * 3424 * Returns the address of the data within @frag. Checks that the page 3425 * is mapped and returns %NULL otherwise. 3426 */ 3427 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3428 { 3429 void *ptr = page_address(skb_frag_page(frag)); 3430 if (unlikely(!ptr)) 3431 return NULL; 3432 3433 return ptr + skb_frag_off(frag); 3434 } 3435 3436 /** 3437 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3438 * @fragto: skb fragment where page is set 3439 * @fragfrom: skb fragment page is copied from 3440 */ 3441 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3442 const skb_frag_t *fragfrom) 3443 { 3444 fragto->bv_page = fragfrom->bv_page; 3445 } 3446 3447 /** 3448 * __skb_frag_set_page - sets the page contained in a paged fragment 3449 * @frag: the paged fragment 3450 * @page: the page to set 3451 * 3452 * Sets the fragment @frag to contain @page. 3453 */ 3454 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 3455 { 3456 frag->bv_page = page; 3457 } 3458 3459 /** 3460 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 3461 * @skb: the buffer 3462 * @f: the fragment offset 3463 * @page: the page to set 3464 * 3465 * Sets the @f'th fragment of @skb to contain @page. 3466 */ 3467 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 3468 struct page *page) 3469 { 3470 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 3471 } 3472 3473 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3474 3475 /** 3476 * skb_frag_dma_map - maps a paged fragment via the DMA API 3477 * @dev: the device to map the fragment to 3478 * @frag: the paged fragment to map 3479 * @offset: the offset within the fragment (starting at the 3480 * fragment's own offset) 3481 * @size: the number of bytes to map 3482 * @dir: the direction of the mapping (``PCI_DMA_*``) 3483 * 3484 * Maps the page associated with @frag to @device. 3485 */ 3486 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 3487 const skb_frag_t *frag, 3488 size_t offset, size_t size, 3489 enum dma_data_direction dir) 3490 { 3491 return dma_map_page(dev, skb_frag_page(frag), 3492 skb_frag_off(frag) + offset, size, dir); 3493 } 3494 3495 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3496 gfp_t gfp_mask) 3497 { 3498 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3499 } 3500 3501 3502 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3503 gfp_t gfp_mask) 3504 { 3505 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3506 } 3507 3508 3509 /** 3510 * skb_clone_writable - is the header of a clone writable 3511 * @skb: buffer to check 3512 * @len: length up to which to write 3513 * 3514 * Returns true if modifying the header part of the cloned buffer 3515 * does not requires the data to be copied. 3516 */ 3517 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3518 { 3519 return !skb_header_cloned(skb) && 3520 skb_headroom(skb) + len <= skb->hdr_len; 3521 } 3522 3523 static inline int skb_try_make_writable(struct sk_buff *skb, 3524 unsigned int write_len) 3525 { 3526 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3527 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3528 } 3529 3530 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3531 int cloned) 3532 { 3533 int delta = 0; 3534 3535 if (headroom > skb_headroom(skb)) 3536 delta = headroom - skb_headroom(skb); 3537 3538 if (delta || cloned) 3539 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3540 GFP_ATOMIC); 3541 return 0; 3542 } 3543 3544 /** 3545 * skb_cow - copy header of skb when it is required 3546 * @skb: buffer to cow 3547 * @headroom: needed headroom 3548 * 3549 * If the skb passed lacks sufficient headroom or its data part 3550 * is shared, data is reallocated. If reallocation fails, an error 3551 * is returned and original skb is not changed. 3552 * 3553 * The result is skb with writable area skb->head...skb->tail 3554 * and at least @headroom of space at head. 3555 */ 3556 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3557 { 3558 return __skb_cow(skb, headroom, skb_cloned(skb)); 3559 } 3560 3561 /** 3562 * skb_cow_head - skb_cow but only making the head writable 3563 * @skb: buffer to cow 3564 * @headroom: needed headroom 3565 * 3566 * This function is identical to skb_cow except that we replace the 3567 * skb_cloned check by skb_header_cloned. It should be used when 3568 * you only need to push on some header and do not need to modify 3569 * the data. 3570 */ 3571 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3572 { 3573 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3574 } 3575 3576 /** 3577 * skb_padto - pad an skbuff up to a minimal size 3578 * @skb: buffer to pad 3579 * @len: minimal length 3580 * 3581 * Pads up a buffer to ensure the trailing bytes exist and are 3582 * blanked. If the buffer already contains sufficient data it 3583 * is untouched. Otherwise it is extended. Returns zero on 3584 * success. The skb is freed on error. 3585 */ 3586 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3587 { 3588 unsigned int size = skb->len; 3589 if (likely(size >= len)) 3590 return 0; 3591 return skb_pad(skb, len - size); 3592 } 3593 3594 /** 3595 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3596 * @skb: buffer to pad 3597 * @len: minimal length 3598 * @free_on_error: free buffer on error 3599 * 3600 * Pads up a buffer to ensure the trailing bytes exist and are 3601 * blanked. If the buffer already contains sufficient data it 3602 * is untouched. Otherwise it is extended. Returns zero on 3603 * success. The skb is freed on error if @free_on_error is true. 3604 */ 3605 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3606 unsigned int len, 3607 bool free_on_error) 3608 { 3609 unsigned int size = skb->len; 3610 3611 if (unlikely(size < len)) { 3612 len -= size; 3613 if (__skb_pad(skb, len, free_on_error)) 3614 return -ENOMEM; 3615 __skb_put(skb, len); 3616 } 3617 return 0; 3618 } 3619 3620 /** 3621 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3622 * @skb: buffer to pad 3623 * @len: minimal length 3624 * 3625 * Pads up a buffer to ensure the trailing bytes exist and are 3626 * blanked. If the buffer already contains sufficient data it 3627 * is untouched. Otherwise it is extended. Returns zero on 3628 * success. The skb is freed on error. 3629 */ 3630 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3631 { 3632 return __skb_put_padto(skb, len, true); 3633 } 3634 3635 static inline int skb_add_data(struct sk_buff *skb, 3636 struct iov_iter *from, int copy) 3637 { 3638 const int off = skb->len; 3639 3640 if (skb->ip_summed == CHECKSUM_NONE) { 3641 __wsum csum = 0; 3642 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 3643 &csum, from)) { 3644 skb->csum = csum_block_add(skb->csum, csum, off); 3645 return 0; 3646 } 3647 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3648 return 0; 3649 3650 __skb_trim(skb, off); 3651 return -EFAULT; 3652 } 3653 3654 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3655 const struct page *page, int off) 3656 { 3657 if (skb_zcopy(skb)) 3658 return false; 3659 if (i) { 3660 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3661 3662 return page == skb_frag_page(frag) && 3663 off == skb_frag_off(frag) + skb_frag_size(frag); 3664 } 3665 return false; 3666 } 3667 3668 static inline int __skb_linearize(struct sk_buff *skb) 3669 { 3670 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3671 } 3672 3673 /** 3674 * skb_linearize - convert paged skb to linear one 3675 * @skb: buffer to linarize 3676 * 3677 * If there is no free memory -ENOMEM is returned, otherwise zero 3678 * is returned and the old skb data released. 3679 */ 3680 static inline int skb_linearize(struct sk_buff *skb) 3681 { 3682 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3683 } 3684 3685 /** 3686 * skb_has_shared_frag - can any frag be overwritten 3687 * @skb: buffer to test 3688 * 3689 * Return true if the skb has at least one frag that might be modified 3690 * by an external entity (as in vmsplice()/sendfile()) 3691 */ 3692 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3693 { 3694 return skb_is_nonlinear(skb) && 3695 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 3696 } 3697 3698 /** 3699 * skb_linearize_cow - make sure skb is linear and writable 3700 * @skb: buffer to process 3701 * 3702 * If there is no free memory -ENOMEM is returned, otherwise zero 3703 * is returned and the old skb data released. 3704 */ 3705 static inline int skb_linearize_cow(struct sk_buff *skb) 3706 { 3707 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3708 __skb_linearize(skb) : 0; 3709 } 3710 3711 static __always_inline void 3712 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3713 unsigned int off) 3714 { 3715 if (skb->ip_summed == CHECKSUM_COMPLETE) 3716 skb->csum = csum_block_sub(skb->csum, 3717 csum_partial(start, len, 0), off); 3718 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3719 skb_checksum_start_offset(skb) < 0) 3720 skb->ip_summed = CHECKSUM_NONE; 3721 } 3722 3723 /** 3724 * skb_postpull_rcsum - update checksum for received skb after pull 3725 * @skb: buffer to update 3726 * @start: start of data before pull 3727 * @len: length of data pulled 3728 * 3729 * After doing a pull on a received packet, you need to call this to 3730 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3731 * CHECKSUM_NONE so that it can be recomputed from scratch. 3732 */ 3733 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3734 const void *start, unsigned int len) 3735 { 3736 if (skb->ip_summed == CHECKSUM_COMPLETE) 3737 skb->csum = wsum_negate(csum_partial(start, len, 3738 wsum_negate(skb->csum))); 3739 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3740 skb_checksum_start_offset(skb) < 0) 3741 skb->ip_summed = CHECKSUM_NONE; 3742 } 3743 3744 static __always_inline void 3745 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3746 unsigned int off) 3747 { 3748 if (skb->ip_summed == CHECKSUM_COMPLETE) 3749 skb->csum = csum_block_add(skb->csum, 3750 csum_partial(start, len, 0), off); 3751 } 3752 3753 /** 3754 * skb_postpush_rcsum - update checksum for received skb after push 3755 * @skb: buffer to update 3756 * @start: start of data after push 3757 * @len: length of data pushed 3758 * 3759 * After doing a push on a received packet, you need to call this to 3760 * update the CHECKSUM_COMPLETE checksum. 3761 */ 3762 static inline void skb_postpush_rcsum(struct sk_buff *skb, 3763 const void *start, unsigned int len) 3764 { 3765 __skb_postpush_rcsum(skb, start, len, 0); 3766 } 3767 3768 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3769 3770 /** 3771 * skb_push_rcsum - push skb and update receive checksum 3772 * @skb: buffer to update 3773 * @len: length of data pulled 3774 * 3775 * This function performs an skb_push on the packet and updates 3776 * the CHECKSUM_COMPLETE checksum. It should be used on 3777 * receive path processing instead of skb_push unless you know 3778 * that the checksum difference is zero (e.g., a valid IP header) 3779 * or you are setting ip_summed to CHECKSUM_NONE. 3780 */ 3781 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3782 { 3783 skb_push(skb, len); 3784 skb_postpush_rcsum(skb, skb->data, len); 3785 return skb->data; 3786 } 3787 3788 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 3789 /** 3790 * pskb_trim_rcsum - trim received skb and update checksum 3791 * @skb: buffer to trim 3792 * @len: new length 3793 * 3794 * This is exactly the same as pskb_trim except that it ensures the 3795 * checksum of received packets are still valid after the operation. 3796 * It can change skb pointers. 3797 */ 3798 3799 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3800 { 3801 if (likely(len >= skb->len)) 3802 return 0; 3803 return pskb_trim_rcsum_slow(skb, len); 3804 } 3805 3806 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3807 { 3808 if (skb->ip_summed == CHECKSUM_COMPLETE) 3809 skb->ip_summed = CHECKSUM_NONE; 3810 __skb_trim(skb, len); 3811 return 0; 3812 } 3813 3814 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3815 { 3816 if (skb->ip_summed == CHECKSUM_COMPLETE) 3817 skb->ip_summed = CHECKSUM_NONE; 3818 return __skb_grow(skb, len); 3819 } 3820 3821 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3822 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 3823 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 3824 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3825 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3826 3827 #define skb_queue_walk(queue, skb) \ 3828 for (skb = (queue)->next; \ 3829 skb != (struct sk_buff *)(queue); \ 3830 skb = skb->next) 3831 3832 #define skb_queue_walk_safe(queue, skb, tmp) \ 3833 for (skb = (queue)->next, tmp = skb->next; \ 3834 skb != (struct sk_buff *)(queue); \ 3835 skb = tmp, tmp = skb->next) 3836 3837 #define skb_queue_walk_from(queue, skb) \ 3838 for (; skb != (struct sk_buff *)(queue); \ 3839 skb = skb->next) 3840 3841 #define skb_rbtree_walk(skb, root) \ 3842 for (skb = skb_rb_first(root); skb != NULL; \ 3843 skb = skb_rb_next(skb)) 3844 3845 #define skb_rbtree_walk_from(skb) \ 3846 for (; skb != NULL; \ 3847 skb = skb_rb_next(skb)) 3848 3849 #define skb_rbtree_walk_from_safe(skb, tmp) \ 3850 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3851 skb = tmp) 3852 3853 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 3854 for (tmp = skb->next; \ 3855 skb != (struct sk_buff *)(queue); \ 3856 skb = tmp, tmp = skb->next) 3857 3858 #define skb_queue_reverse_walk(queue, skb) \ 3859 for (skb = (queue)->prev; \ 3860 skb != (struct sk_buff *)(queue); \ 3861 skb = skb->prev) 3862 3863 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3864 for (skb = (queue)->prev, tmp = skb->prev; \ 3865 skb != (struct sk_buff *)(queue); \ 3866 skb = tmp, tmp = skb->prev) 3867 3868 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3869 for (tmp = skb->prev; \ 3870 skb != (struct sk_buff *)(queue); \ 3871 skb = tmp, tmp = skb->prev) 3872 3873 static inline bool skb_has_frag_list(const struct sk_buff *skb) 3874 { 3875 return skb_shinfo(skb)->frag_list != NULL; 3876 } 3877 3878 static inline void skb_frag_list_init(struct sk_buff *skb) 3879 { 3880 skb_shinfo(skb)->frag_list = NULL; 3881 } 3882 3883 #define skb_walk_frags(skb, iter) \ 3884 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3885 3886 3887 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 3888 int *err, long *timeo_p, 3889 const struct sk_buff *skb); 3890 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3891 struct sk_buff_head *queue, 3892 unsigned int flags, 3893 int *off, int *err, 3894 struct sk_buff **last); 3895 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 3896 struct sk_buff_head *queue, 3897 unsigned int flags, int *off, int *err, 3898 struct sk_buff **last); 3899 struct sk_buff *__skb_recv_datagram(struct sock *sk, 3900 struct sk_buff_head *sk_queue, 3901 unsigned int flags, int *off, int *err); 3902 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 3903 __poll_t datagram_poll(struct file *file, struct socket *sock, 3904 struct poll_table_struct *wait); 3905 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3906 struct iov_iter *to, int size); 3907 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3908 struct msghdr *msg, int size) 3909 { 3910 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3911 } 3912 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3913 struct msghdr *msg); 3914 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 3915 struct iov_iter *to, int len, 3916 struct ahash_request *hash); 3917 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3918 struct iov_iter *from, int len); 3919 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3920 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3921 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3922 static inline void skb_free_datagram_locked(struct sock *sk, 3923 struct sk_buff *skb) 3924 { 3925 __skb_free_datagram_locked(sk, skb, 0); 3926 } 3927 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3928 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3929 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3930 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3931 int len); 3932 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3933 struct pipe_inode_info *pipe, unsigned int len, 3934 unsigned int flags); 3935 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 3936 int len); 3937 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 3938 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3939 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3940 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3941 int len, int hlen); 3942 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3943 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3944 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3945 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu); 3946 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len); 3947 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3948 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 3949 unsigned int offset); 3950 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3951 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 3952 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3953 int skb_vlan_pop(struct sk_buff *skb); 3954 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3955 int skb_eth_pop(struct sk_buff *skb); 3956 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 3957 const unsigned char *src); 3958 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 3959 int mac_len, bool ethernet); 3960 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 3961 bool ethernet); 3962 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 3963 int skb_mpls_dec_ttl(struct sk_buff *skb); 3964 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3965 gfp_t gfp); 3966 3967 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3968 { 3969 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 3970 } 3971 3972 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3973 { 3974 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3975 } 3976 3977 struct skb_checksum_ops { 3978 __wsum (*update)(const void *mem, int len, __wsum wsum); 3979 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3980 }; 3981 3982 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 3983 3984 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3985 __wsum csum, const struct skb_checksum_ops *ops); 3986 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3987 __wsum csum); 3988 3989 static inline void * __must_check 3990 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 3991 const void *data, int hlen, void *buffer) 3992 { 3993 if (likely(hlen - offset >= len)) 3994 return (void *)data + offset; 3995 3996 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 3997 return NULL; 3998 3999 return buffer; 4000 } 4001 4002 static inline void * __must_check 4003 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4004 { 4005 return __skb_header_pointer(skb, offset, len, skb->data, 4006 skb_headlen(skb), buffer); 4007 } 4008 4009 /** 4010 * skb_needs_linearize - check if we need to linearize a given skb 4011 * depending on the given device features. 4012 * @skb: socket buffer to check 4013 * @features: net device features 4014 * 4015 * Returns true if either: 4016 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4017 * 2. skb is fragmented and the device does not support SG. 4018 */ 4019 static inline bool skb_needs_linearize(struct sk_buff *skb, 4020 netdev_features_t features) 4021 { 4022 return skb_is_nonlinear(skb) && 4023 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4024 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4025 } 4026 4027 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4028 void *to, 4029 const unsigned int len) 4030 { 4031 memcpy(to, skb->data, len); 4032 } 4033 4034 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4035 const int offset, void *to, 4036 const unsigned int len) 4037 { 4038 memcpy(to, skb->data + offset, len); 4039 } 4040 4041 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4042 const void *from, 4043 const unsigned int len) 4044 { 4045 memcpy(skb->data, from, len); 4046 } 4047 4048 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4049 const int offset, 4050 const void *from, 4051 const unsigned int len) 4052 { 4053 memcpy(skb->data + offset, from, len); 4054 } 4055 4056 void skb_init(void); 4057 4058 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4059 { 4060 return skb->tstamp; 4061 } 4062 4063 /** 4064 * skb_get_timestamp - get timestamp from a skb 4065 * @skb: skb to get stamp from 4066 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4067 * 4068 * Timestamps are stored in the skb as offsets to a base timestamp. 4069 * This function converts the offset back to a struct timeval and stores 4070 * it in stamp. 4071 */ 4072 static inline void skb_get_timestamp(const struct sk_buff *skb, 4073 struct __kernel_old_timeval *stamp) 4074 { 4075 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4076 } 4077 4078 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4079 struct __kernel_sock_timeval *stamp) 4080 { 4081 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4082 4083 stamp->tv_sec = ts.tv_sec; 4084 stamp->tv_usec = ts.tv_nsec / 1000; 4085 } 4086 4087 static inline void skb_get_timestampns(const struct sk_buff *skb, 4088 struct __kernel_old_timespec *stamp) 4089 { 4090 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4091 4092 stamp->tv_sec = ts.tv_sec; 4093 stamp->tv_nsec = ts.tv_nsec; 4094 } 4095 4096 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4097 struct __kernel_timespec *stamp) 4098 { 4099 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4100 4101 stamp->tv_sec = ts.tv_sec; 4102 stamp->tv_nsec = ts.tv_nsec; 4103 } 4104 4105 static inline void __net_timestamp(struct sk_buff *skb) 4106 { 4107 skb->tstamp = ktime_get_real(); 4108 skb->mono_delivery_time = 0; 4109 } 4110 4111 static inline ktime_t net_timedelta(ktime_t t) 4112 { 4113 return ktime_sub(ktime_get_real(), t); 4114 } 4115 4116 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4117 bool mono) 4118 { 4119 skb->tstamp = kt; 4120 skb->mono_delivery_time = kt && mono; 4121 } 4122 4123 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4124 4125 /* It is used in the ingress path to clear the delivery_time. 4126 * If needed, set the skb->tstamp to the (rcv) timestamp. 4127 */ 4128 static inline void skb_clear_delivery_time(struct sk_buff *skb) 4129 { 4130 if (skb->mono_delivery_time) { 4131 skb->mono_delivery_time = 0; 4132 if (static_branch_unlikely(&netstamp_needed_key)) 4133 skb->tstamp = ktime_get_real(); 4134 else 4135 skb->tstamp = 0; 4136 } 4137 } 4138 4139 static inline void skb_clear_tstamp(struct sk_buff *skb) 4140 { 4141 if (skb->mono_delivery_time) 4142 return; 4143 4144 skb->tstamp = 0; 4145 } 4146 4147 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4148 { 4149 if (skb->mono_delivery_time) 4150 return 0; 4151 4152 return skb->tstamp; 4153 } 4154 4155 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4156 { 4157 if (!skb->mono_delivery_time && skb->tstamp) 4158 return skb->tstamp; 4159 4160 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4161 return ktime_get_real(); 4162 4163 return 0; 4164 } 4165 4166 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4167 { 4168 return skb_shinfo(skb)->meta_len; 4169 } 4170 4171 static inline void *skb_metadata_end(const struct sk_buff *skb) 4172 { 4173 return skb_mac_header(skb); 4174 } 4175 4176 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4177 const struct sk_buff *skb_b, 4178 u8 meta_len) 4179 { 4180 const void *a = skb_metadata_end(skb_a); 4181 const void *b = skb_metadata_end(skb_b); 4182 /* Using more efficient varaiant than plain call to memcmp(). */ 4183 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64 4184 u64 diffs = 0; 4185 4186 switch (meta_len) { 4187 #define __it(x, op) (x -= sizeof(u##op)) 4188 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4189 case 32: diffs |= __it_diff(a, b, 64); 4190 fallthrough; 4191 case 24: diffs |= __it_diff(a, b, 64); 4192 fallthrough; 4193 case 16: diffs |= __it_diff(a, b, 64); 4194 fallthrough; 4195 case 8: diffs |= __it_diff(a, b, 64); 4196 break; 4197 case 28: diffs |= __it_diff(a, b, 64); 4198 fallthrough; 4199 case 20: diffs |= __it_diff(a, b, 64); 4200 fallthrough; 4201 case 12: diffs |= __it_diff(a, b, 64); 4202 fallthrough; 4203 case 4: diffs |= __it_diff(a, b, 32); 4204 break; 4205 } 4206 return diffs; 4207 #else 4208 return memcmp(a - meta_len, b - meta_len, meta_len); 4209 #endif 4210 } 4211 4212 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4213 const struct sk_buff *skb_b) 4214 { 4215 u8 len_a = skb_metadata_len(skb_a); 4216 u8 len_b = skb_metadata_len(skb_b); 4217 4218 if (!(len_a | len_b)) 4219 return false; 4220 4221 return len_a != len_b ? 4222 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4223 } 4224 4225 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4226 { 4227 skb_shinfo(skb)->meta_len = meta_len; 4228 } 4229 4230 static inline void skb_metadata_clear(struct sk_buff *skb) 4231 { 4232 skb_metadata_set(skb, 0); 4233 } 4234 4235 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4236 4237 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4238 4239 void skb_clone_tx_timestamp(struct sk_buff *skb); 4240 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4241 4242 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4243 4244 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4245 { 4246 } 4247 4248 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4249 { 4250 return false; 4251 } 4252 4253 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4254 4255 /** 4256 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4257 * 4258 * PHY drivers may accept clones of transmitted packets for 4259 * timestamping via their phy_driver.txtstamp method. These drivers 4260 * must call this function to return the skb back to the stack with a 4261 * timestamp. 4262 * 4263 * @skb: clone of the original outgoing packet 4264 * @hwtstamps: hardware time stamps 4265 * 4266 */ 4267 void skb_complete_tx_timestamp(struct sk_buff *skb, 4268 struct skb_shared_hwtstamps *hwtstamps); 4269 4270 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4271 struct skb_shared_hwtstamps *hwtstamps, 4272 struct sock *sk, int tstype); 4273 4274 /** 4275 * skb_tstamp_tx - queue clone of skb with send time stamps 4276 * @orig_skb: the original outgoing packet 4277 * @hwtstamps: hardware time stamps, may be NULL if not available 4278 * 4279 * If the skb has a socket associated, then this function clones the 4280 * skb (thus sharing the actual data and optional structures), stores 4281 * the optional hardware time stamping information (if non NULL) or 4282 * generates a software time stamp (otherwise), then queues the clone 4283 * to the error queue of the socket. Errors are silently ignored. 4284 */ 4285 void skb_tstamp_tx(struct sk_buff *orig_skb, 4286 struct skb_shared_hwtstamps *hwtstamps); 4287 4288 /** 4289 * skb_tx_timestamp() - Driver hook for transmit timestamping 4290 * 4291 * Ethernet MAC Drivers should call this function in their hard_xmit() 4292 * function immediately before giving the sk_buff to the MAC hardware. 4293 * 4294 * Specifically, one should make absolutely sure that this function is 4295 * called before TX completion of this packet can trigger. Otherwise 4296 * the packet could potentially already be freed. 4297 * 4298 * @skb: A socket buffer. 4299 */ 4300 static inline void skb_tx_timestamp(struct sk_buff *skb) 4301 { 4302 skb_clone_tx_timestamp(skb); 4303 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 4304 skb_tstamp_tx(skb, NULL); 4305 } 4306 4307 /** 4308 * skb_complete_wifi_ack - deliver skb with wifi status 4309 * 4310 * @skb: the original outgoing packet 4311 * @acked: ack status 4312 * 4313 */ 4314 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4315 4316 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4317 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4318 4319 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4320 { 4321 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4322 skb->csum_valid || 4323 (skb->ip_summed == CHECKSUM_PARTIAL && 4324 skb_checksum_start_offset(skb) >= 0)); 4325 } 4326 4327 /** 4328 * skb_checksum_complete - Calculate checksum of an entire packet 4329 * @skb: packet to process 4330 * 4331 * This function calculates the checksum over the entire packet plus 4332 * the value of skb->csum. The latter can be used to supply the 4333 * checksum of a pseudo header as used by TCP/UDP. It returns the 4334 * checksum. 4335 * 4336 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4337 * this function can be used to verify that checksum on received 4338 * packets. In that case the function should return zero if the 4339 * checksum is correct. In particular, this function will return zero 4340 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4341 * hardware has already verified the correctness of the checksum. 4342 */ 4343 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4344 { 4345 return skb_csum_unnecessary(skb) ? 4346 0 : __skb_checksum_complete(skb); 4347 } 4348 4349 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4350 { 4351 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4352 if (skb->csum_level == 0) 4353 skb->ip_summed = CHECKSUM_NONE; 4354 else 4355 skb->csum_level--; 4356 } 4357 } 4358 4359 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4360 { 4361 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4362 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4363 skb->csum_level++; 4364 } else if (skb->ip_summed == CHECKSUM_NONE) { 4365 skb->ip_summed = CHECKSUM_UNNECESSARY; 4366 skb->csum_level = 0; 4367 } 4368 } 4369 4370 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4371 { 4372 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4373 skb->ip_summed = CHECKSUM_NONE; 4374 skb->csum_level = 0; 4375 } 4376 } 4377 4378 /* Check if we need to perform checksum complete validation. 4379 * 4380 * Returns true if checksum complete is needed, false otherwise 4381 * (either checksum is unnecessary or zero checksum is allowed). 4382 */ 4383 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4384 bool zero_okay, 4385 __sum16 check) 4386 { 4387 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4388 skb->csum_valid = 1; 4389 __skb_decr_checksum_unnecessary(skb); 4390 return false; 4391 } 4392 4393 return true; 4394 } 4395 4396 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4397 * in checksum_init. 4398 */ 4399 #define CHECKSUM_BREAK 76 4400 4401 /* Unset checksum-complete 4402 * 4403 * Unset checksum complete can be done when packet is being modified 4404 * (uncompressed for instance) and checksum-complete value is 4405 * invalidated. 4406 */ 4407 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4408 { 4409 if (skb->ip_summed == CHECKSUM_COMPLETE) 4410 skb->ip_summed = CHECKSUM_NONE; 4411 } 4412 4413 /* Validate (init) checksum based on checksum complete. 4414 * 4415 * Return values: 4416 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4417 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4418 * checksum is stored in skb->csum for use in __skb_checksum_complete 4419 * non-zero: value of invalid checksum 4420 * 4421 */ 4422 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4423 bool complete, 4424 __wsum psum) 4425 { 4426 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4427 if (!csum_fold(csum_add(psum, skb->csum))) { 4428 skb->csum_valid = 1; 4429 return 0; 4430 } 4431 } 4432 4433 skb->csum = psum; 4434 4435 if (complete || skb->len <= CHECKSUM_BREAK) { 4436 __sum16 csum; 4437 4438 csum = __skb_checksum_complete(skb); 4439 skb->csum_valid = !csum; 4440 return csum; 4441 } 4442 4443 return 0; 4444 } 4445 4446 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4447 { 4448 return 0; 4449 } 4450 4451 /* Perform checksum validate (init). Note that this is a macro since we only 4452 * want to calculate the pseudo header which is an input function if necessary. 4453 * First we try to validate without any computation (checksum unnecessary) and 4454 * then calculate based on checksum complete calling the function to compute 4455 * pseudo header. 4456 * 4457 * Return values: 4458 * 0: checksum is validated or try to in skb_checksum_complete 4459 * non-zero: value of invalid checksum 4460 */ 4461 #define __skb_checksum_validate(skb, proto, complete, \ 4462 zero_okay, check, compute_pseudo) \ 4463 ({ \ 4464 __sum16 __ret = 0; \ 4465 skb->csum_valid = 0; \ 4466 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4467 __ret = __skb_checksum_validate_complete(skb, \ 4468 complete, compute_pseudo(skb, proto)); \ 4469 __ret; \ 4470 }) 4471 4472 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4473 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4474 4475 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4476 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4477 4478 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4479 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4480 4481 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4482 compute_pseudo) \ 4483 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4484 4485 #define skb_checksum_simple_validate(skb) \ 4486 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4487 4488 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4489 { 4490 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4491 } 4492 4493 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4494 { 4495 skb->csum = ~pseudo; 4496 skb->ip_summed = CHECKSUM_COMPLETE; 4497 } 4498 4499 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4500 do { \ 4501 if (__skb_checksum_convert_check(skb)) \ 4502 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4503 } while (0) 4504 4505 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4506 u16 start, u16 offset) 4507 { 4508 skb->ip_summed = CHECKSUM_PARTIAL; 4509 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4510 skb->csum_offset = offset - start; 4511 } 4512 4513 /* Update skbuf and packet to reflect the remote checksum offload operation. 4514 * When called, ptr indicates the starting point for skb->csum when 4515 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4516 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4517 */ 4518 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4519 int start, int offset, bool nopartial) 4520 { 4521 __wsum delta; 4522 4523 if (!nopartial) { 4524 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4525 return; 4526 } 4527 4528 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4529 __skb_checksum_complete(skb); 4530 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4531 } 4532 4533 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4534 4535 /* Adjust skb->csum since we changed the packet */ 4536 skb->csum = csum_add(skb->csum, delta); 4537 } 4538 4539 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4540 { 4541 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4542 return (void *)(skb->_nfct & NFCT_PTRMASK); 4543 #else 4544 return NULL; 4545 #endif 4546 } 4547 4548 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4549 { 4550 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4551 return skb->_nfct; 4552 #else 4553 return 0UL; 4554 #endif 4555 } 4556 4557 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4558 { 4559 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4560 skb->slow_gro |= !!nfct; 4561 skb->_nfct = nfct; 4562 #endif 4563 } 4564 4565 #ifdef CONFIG_SKB_EXTENSIONS 4566 enum skb_ext_id { 4567 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4568 SKB_EXT_BRIDGE_NF, 4569 #endif 4570 #ifdef CONFIG_XFRM 4571 SKB_EXT_SEC_PATH, 4572 #endif 4573 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4574 TC_SKB_EXT, 4575 #endif 4576 #if IS_ENABLED(CONFIG_MPTCP) 4577 SKB_EXT_MPTCP, 4578 #endif 4579 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 4580 SKB_EXT_MCTP, 4581 #endif 4582 SKB_EXT_NUM, /* must be last */ 4583 }; 4584 4585 /** 4586 * struct skb_ext - sk_buff extensions 4587 * @refcnt: 1 on allocation, deallocated on 0 4588 * @offset: offset to add to @data to obtain extension address 4589 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4590 * @data: start of extension data, variable sized 4591 * 4592 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4593 * to use 'u8' types while allowing up to 2kb worth of extension data. 4594 */ 4595 struct skb_ext { 4596 refcount_t refcnt; 4597 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4598 u8 chunks; /* same */ 4599 char data[] __aligned(8); 4600 }; 4601 4602 struct skb_ext *__skb_ext_alloc(gfp_t flags); 4603 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 4604 struct skb_ext *ext); 4605 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4606 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4607 void __skb_ext_put(struct skb_ext *ext); 4608 4609 static inline void skb_ext_put(struct sk_buff *skb) 4610 { 4611 if (skb->active_extensions) 4612 __skb_ext_put(skb->extensions); 4613 } 4614 4615 static inline void __skb_ext_copy(struct sk_buff *dst, 4616 const struct sk_buff *src) 4617 { 4618 dst->active_extensions = src->active_extensions; 4619 4620 if (src->active_extensions) { 4621 struct skb_ext *ext = src->extensions; 4622 4623 refcount_inc(&ext->refcnt); 4624 dst->extensions = ext; 4625 } 4626 } 4627 4628 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4629 { 4630 skb_ext_put(dst); 4631 __skb_ext_copy(dst, src); 4632 } 4633 4634 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4635 { 4636 return !!ext->offset[i]; 4637 } 4638 4639 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4640 { 4641 return skb->active_extensions & (1 << id); 4642 } 4643 4644 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4645 { 4646 if (skb_ext_exist(skb, id)) 4647 __skb_ext_del(skb, id); 4648 } 4649 4650 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4651 { 4652 if (skb_ext_exist(skb, id)) { 4653 struct skb_ext *ext = skb->extensions; 4654 4655 return (void *)ext + (ext->offset[id] << 3); 4656 } 4657 4658 return NULL; 4659 } 4660 4661 static inline void skb_ext_reset(struct sk_buff *skb) 4662 { 4663 if (unlikely(skb->active_extensions)) { 4664 __skb_ext_put(skb->extensions); 4665 skb->active_extensions = 0; 4666 } 4667 } 4668 4669 static inline bool skb_has_extensions(struct sk_buff *skb) 4670 { 4671 return unlikely(skb->active_extensions); 4672 } 4673 #else 4674 static inline void skb_ext_put(struct sk_buff *skb) {} 4675 static inline void skb_ext_reset(struct sk_buff *skb) {} 4676 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4677 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4678 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4679 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 4680 #endif /* CONFIG_SKB_EXTENSIONS */ 4681 4682 static inline void nf_reset_ct(struct sk_buff *skb) 4683 { 4684 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4685 nf_conntrack_put(skb_nfct(skb)); 4686 skb->_nfct = 0; 4687 #endif 4688 } 4689 4690 static inline void nf_reset_trace(struct sk_buff *skb) 4691 { 4692 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4693 skb->nf_trace = 0; 4694 #endif 4695 } 4696 4697 static inline void ipvs_reset(struct sk_buff *skb) 4698 { 4699 #if IS_ENABLED(CONFIG_IP_VS) 4700 skb->ipvs_property = 0; 4701 #endif 4702 } 4703 4704 /* Note: This doesn't put any conntrack info in dst. */ 4705 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4706 bool copy) 4707 { 4708 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4709 dst->_nfct = src->_nfct; 4710 nf_conntrack_get(skb_nfct(src)); 4711 #endif 4712 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4713 if (copy) 4714 dst->nf_trace = src->nf_trace; 4715 #endif 4716 } 4717 4718 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 4719 { 4720 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4721 nf_conntrack_put(skb_nfct(dst)); 4722 #endif 4723 dst->slow_gro = src->slow_gro; 4724 __nf_copy(dst, src, true); 4725 } 4726 4727 #ifdef CONFIG_NETWORK_SECMARK 4728 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4729 { 4730 to->secmark = from->secmark; 4731 } 4732 4733 static inline void skb_init_secmark(struct sk_buff *skb) 4734 { 4735 skb->secmark = 0; 4736 } 4737 #else 4738 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4739 { } 4740 4741 static inline void skb_init_secmark(struct sk_buff *skb) 4742 { } 4743 #endif 4744 4745 static inline int secpath_exists(const struct sk_buff *skb) 4746 { 4747 #ifdef CONFIG_XFRM 4748 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 4749 #else 4750 return 0; 4751 #endif 4752 } 4753 4754 static inline bool skb_irq_freeable(const struct sk_buff *skb) 4755 { 4756 return !skb->destructor && 4757 !secpath_exists(skb) && 4758 !skb_nfct(skb) && 4759 !skb->_skb_refdst && 4760 !skb_has_frag_list(skb); 4761 } 4762 4763 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 4764 { 4765 skb->queue_mapping = queue_mapping; 4766 } 4767 4768 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 4769 { 4770 return skb->queue_mapping; 4771 } 4772 4773 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 4774 { 4775 to->queue_mapping = from->queue_mapping; 4776 } 4777 4778 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 4779 { 4780 skb->queue_mapping = rx_queue + 1; 4781 } 4782 4783 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 4784 { 4785 return skb->queue_mapping - 1; 4786 } 4787 4788 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 4789 { 4790 return skb->queue_mapping != 0; 4791 } 4792 4793 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 4794 { 4795 skb->dst_pending_confirm = val; 4796 } 4797 4798 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 4799 { 4800 return skb->dst_pending_confirm != 0; 4801 } 4802 4803 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 4804 { 4805 #ifdef CONFIG_XFRM 4806 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 4807 #else 4808 return NULL; 4809 #endif 4810 } 4811 4812 /* Keeps track of mac header offset relative to skb->head. 4813 * It is useful for TSO of Tunneling protocol. e.g. GRE. 4814 * For non-tunnel skb it points to skb_mac_header() and for 4815 * tunnel skb it points to outer mac header. 4816 * Keeps track of level of encapsulation of network headers. 4817 */ 4818 struct skb_gso_cb { 4819 union { 4820 int mac_offset; 4821 int data_offset; 4822 }; 4823 int encap_level; 4824 __wsum csum; 4825 __u16 csum_start; 4826 }; 4827 #define SKB_GSO_CB_OFFSET 32 4828 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET)) 4829 4830 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 4831 { 4832 return (skb_mac_header(inner_skb) - inner_skb->head) - 4833 SKB_GSO_CB(inner_skb)->mac_offset; 4834 } 4835 4836 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 4837 { 4838 int new_headroom, headroom; 4839 int ret; 4840 4841 headroom = skb_headroom(skb); 4842 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 4843 if (ret) 4844 return ret; 4845 4846 new_headroom = skb_headroom(skb); 4847 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 4848 return 0; 4849 } 4850 4851 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 4852 { 4853 /* Do not update partial checksums if remote checksum is enabled. */ 4854 if (skb->remcsum_offload) 4855 return; 4856 4857 SKB_GSO_CB(skb)->csum = res; 4858 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 4859 } 4860 4861 /* Compute the checksum for a gso segment. First compute the checksum value 4862 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 4863 * then add in skb->csum (checksum from csum_start to end of packet). 4864 * skb->csum and csum_start are then updated to reflect the checksum of the 4865 * resultant packet starting from the transport header-- the resultant checksum 4866 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 4867 * header. 4868 */ 4869 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 4870 { 4871 unsigned char *csum_start = skb_transport_header(skb); 4872 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 4873 __wsum partial = SKB_GSO_CB(skb)->csum; 4874 4875 SKB_GSO_CB(skb)->csum = res; 4876 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 4877 4878 return csum_fold(csum_partial(csum_start, plen, partial)); 4879 } 4880 4881 static inline bool skb_is_gso(const struct sk_buff *skb) 4882 { 4883 return skb_shinfo(skb)->gso_size; 4884 } 4885 4886 /* Note: Should be called only if skb_is_gso(skb) is true */ 4887 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4888 { 4889 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4890 } 4891 4892 /* Note: Should be called only if skb_is_gso(skb) is true */ 4893 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 4894 { 4895 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 4896 } 4897 4898 /* Note: Should be called only if skb_is_gso(skb) is true */ 4899 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 4900 { 4901 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 4902 } 4903 4904 static inline void skb_gso_reset(struct sk_buff *skb) 4905 { 4906 skb_shinfo(skb)->gso_size = 0; 4907 skb_shinfo(skb)->gso_segs = 0; 4908 skb_shinfo(skb)->gso_type = 0; 4909 } 4910 4911 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 4912 u16 increment) 4913 { 4914 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4915 return; 4916 shinfo->gso_size += increment; 4917 } 4918 4919 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 4920 u16 decrement) 4921 { 4922 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4923 return; 4924 shinfo->gso_size -= decrement; 4925 } 4926 4927 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 4928 4929 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 4930 { 4931 /* LRO sets gso_size but not gso_type, whereas if GSO is really 4932 * wanted then gso_type will be set. */ 4933 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4934 4935 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 4936 unlikely(shinfo->gso_type == 0)) { 4937 __skb_warn_lro_forwarding(skb); 4938 return true; 4939 } 4940 return false; 4941 } 4942 4943 static inline void skb_forward_csum(struct sk_buff *skb) 4944 { 4945 /* Unfortunately we don't support this one. Any brave souls? */ 4946 if (skb->ip_summed == CHECKSUM_COMPLETE) 4947 skb->ip_summed = CHECKSUM_NONE; 4948 } 4949 4950 /** 4951 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 4952 * @skb: skb to check 4953 * 4954 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 4955 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 4956 * use this helper, to document places where we make this assertion. 4957 */ 4958 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 4959 { 4960 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 4961 } 4962 4963 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 4964 4965 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 4966 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4967 unsigned int transport_len, 4968 __sum16(*skb_chkf)(struct sk_buff *skb)); 4969 4970 /** 4971 * skb_head_is_locked - Determine if the skb->head is locked down 4972 * @skb: skb to check 4973 * 4974 * The head on skbs build around a head frag can be removed if they are 4975 * not cloned. This function returns true if the skb head is locked down 4976 * due to either being allocated via kmalloc, or by being a clone with 4977 * multiple references to the head. 4978 */ 4979 static inline bool skb_head_is_locked(const struct sk_buff *skb) 4980 { 4981 return !skb->head_frag || skb_cloned(skb); 4982 } 4983 4984 /* Local Checksum Offload. 4985 * Compute outer checksum based on the assumption that the 4986 * inner checksum will be offloaded later. 4987 * See Documentation/networking/checksum-offloads.rst for 4988 * explanation of how this works. 4989 * Fill in outer checksum adjustment (e.g. with sum of outer 4990 * pseudo-header) before calling. 4991 * Also ensure that inner checksum is in linear data area. 4992 */ 4993 static inline __wsum lco_csum(struct sk_buff *skb) 4994 { 4995 unsigned char *csum_start = skb_checksum_start(skb); 4996 unsigned char *l4_hdr = skb_transport_header(skb); 4997 __wsum partial; 4998 4999 /* Start with complement of inner checksum adjustment */ 5000 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 5001 skb->csum_offset)); 5002 5003 /* Add in checksum of our headers (incl. outer checksum 5004 * adjustment filled in by caller) and return result. 5005 */ 5006 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5007 } 5008 5009 static inline bool skb_is_redirected(const struct sk_buff *skb) 5010 { 5011 return skb->redirected; 5012 } 5013 5014 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5015 { 5016 skb->redirected = 1; 5017 #ifdef CONFIG_NET_REDIRECT 5018 skb->from_ingress = from_ingress; 5019 if (skb->from_ingress) 5020 skb_clear_tstamp(skb); 5021 #endif 5022 } 5023 5024 static inline void skb_reset_redirect(struct sk_buff *skb) 5025 { 5026 skb->redirected = 0; 5027 } 5028 5029 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5030 { 5031 return skb->csum_not_inet; 5032 } 5033 5034 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5035 const u64 kcov_handle) 5036 { 5037 #ifdef CONFIG_KCOV 5038 skb->kcov_handle = kcov_handle; 5039 #endif 5040 } 5041 5042 static inline u64 skb_get_kcov_handle(struct sk_buff *skb) 5043 { 5044 #ifdef CONFIG_KCOV 5045 return skb->kcov_handle; 5046 #else 5047 return 0; 5048 #endif 5049 } 5050 5051 #ifdef CONFIG_PAGE_POOL 5052 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5053 { 5054 skb->pp_recycle = 1; 5055 } 5056 #endif 5057 5058 #endif /* __KERNEL__ */ 5059 #endif /* _LINUX_SKBUFF_H */ 5060