2 * Definitions for the 'struct sk_buff' memory handlers.
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
17 #include <linux/kernel.h>
18 #include <linux/kmemcheck.h>
19 #include <linux/compiler.h>
20 #include <linux/time.h>
21 #include <linux/bug.h>
22 #include <linux/cache.h>
23 #include <linux/rbtree.h>
24 #include <linux/socket.h>
26 #include <linux/atomic.h>
27 #include <asm/types.h>
28 #include <linux/spinlock.h>
29 #include <linux/net.h>
30 #include <linux/textsearch.h>
31 #include <net/checksum.h>
32 #include <linux/rcupdate.h>
33 #include <linux/hrtimer.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/netdev_features.h>
36 #include <linux/sched.h>
37 #include <linux/sched/clock.h>
38 #include <net/flow_dissector.h>
39 #include <linux/splice.h>
40 #include <linux/in6.h>
41 #include <linux/if_packet.h>
44 /* The interface for checksum offload between the stack and networking drivers
47 * A. IP checksum related features
49 * Drivers advertise checksum offload capabilities in the features of a device.
50 * From the stack's point of view these are capabilities offered by the driver,
51 * a driver typically only advertises features that it is capable of offloading
54 * The checksum related features are:
56 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
57 * IP (one's complement) checksum for any combination
58 * of protocols or protocol layering. The checksum is
59 * computed and set in a packet per the CHECKSUM_PARTIAL
60 * interface (see below).
62 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
63 * TCP or UDP packets over IPv4. These are specifically
64 * unencapsulated packets of the form IPv4|TCP or
65 * IPv4|UDP where the Protocol field in the IPv4 header
66 * is TCP or UDP. The IPv4 header may contain IP options
67 * This feature cannot be set in features for a device
68 * with NETIF_F_HW_CSUM also set. This feature is being
69 * DEPRECATED (see below).
71 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
72 * TCP or UDP packets over IPv6. These are specifically
73 * unencapsulated packets of the form IPv6|TCP or
74 * IPv4|UDP where the Next Header field in the IPv6
75 * header is either TCP or UDP. IPv6 extension headers
76 * are not supported with this feature. This feature
77 * cannot be set in features for a device with
78 * NETIF_F_HW_CSUM also set. This feature is being
79 * DEPRECATED (see below).
81 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
82 * This flag is used only used to disable the RX checksum
83 * feature for a device. The stack will accept receive
84 * checksum indication in packets received on a device
85 * regardless of whether NETIF_F_RXCSUM is set.
87 * B. Checksumming of received packets by device. Indication of checksum
88 * verification is in set skb->ip_summed. Possible values are:
92 * Device did not checksum this packet e.g. due to lack of capabilities.
93 * The packet contains full (though not verified) checksum in packet but
94 * not in skb->csum. Thus, skb->csum is undefined in this case.
96 * CHECKSUM_UNNECESSARY:
98 * The hardware you're dealing with doesn't calculate the full checksum
99 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
100 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
101 * if their checksums are okay. skb->csum is still undefined in this case
102 * though. A driver or device must never modify the checksum field in the
103 * packet even if checksum is verified.
105 * CHECKSUM_UNNECESSARY is applicable to following protocols:
106 * TCP: IPv6 and IPv4.
107 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
108 * zero UDP checksum for either IPv4 or IPv6, the networking stack
109 * may perform further validation in this case.
110 * GRE: only if the checksum is present in the header.
111 * SCTP: indicates the CRC in SCTP header has been validated.
112 * FCOE: indicates the CRC in FC frame has been validated.
114 * skb->csum_level indicates the number of consecutive checksums found in
115 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
116 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
117 * and a device is able to verify the checksums for UDP (possibly zero),
118 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
119 * two. If the device were only able to verify the UDP checksum and not
120 * GRE, either because it doesn't support GRE checksum of because GRE
121 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
122 * not considered in this case).
126 * This is the most generic way. The device supplied checksum of the _whole_
127 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
128 * hardware doesn't need to parse L3/L4 headers to implement this.
131 * - Even if device supports only some protocols, but is able to produce
132 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
133 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
137 * A checksum is set up to be offloaded to a device as described in the
138 * output description for CHECKSUM_PARTIAL. This may occur on a packet
139 * received directly from another Linux OS, e.g., a virtualized Linux kernel
140 * on the same host, or it may be set in the input path in GRO or remote
141 * checksum offload. For the purposes of checksum verification, the checksum
142 * referred to by skb->csum_start + skb->csum_offset and any preceding
143 * checksums in the packet are considered verified. Any checksums in the
144 * packet that are after the checksum being offloaded are not considered to
147 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
148 * in the skb->ip_summed for a packet. Values are:
152 * The driver is required to checksum the packet as seen by hard_start_xmit()
153 * from skb->csum_start up to the end, and to record/write the checksum at
154 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
155 * csum_start and csum_offset values are valid values given the length and
156 * offset of the packet, however they should not attempt to validate that the
157 * checksum refers to a legitimate transport layer checksum-- it is the
158 * purview of the stack to validate that csum_start and csum_offset are set
161 * When the stack requests checksum offload for a packet, the driver MUST
162 * ensure that the checksum is set correctly. A driver can either offload the
163 * checksum calculation to the device, or call skb_checksum_help (in the case
164 * that the device does not support offload for a particular checksum).
166 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
167 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
168 * checksum offload capability.
169 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
170 * on network device checksumming capabilities: if a packet does not match
171 * them, skb_checksum_help or skb_crc32c_help (depending on the value of
172 * csum_not_inet, see item D.) is called to resolve the checksum.
176 * The skb was already checksummed by the protocol, or a checksum is not
179 * CHECKSUM_UNNECESSARY:
181 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
185 * Not used in checksum output. If a driver observes a packet with this value
186 * set in skbuff, if should treat as CHECKSUM_NONE being set.
188 * D. Non-IP checksum (CRC) offloads
190 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
191 * offloading the SCTP CRC in a packet. To perform this offload the stack
192 * will set set csum_start and csum_offset accordingly, set ip_summed to
193 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
194 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
195 * A driver that supports both IP checksum offload and SCTP CRC32c offload
196 * must verify which offload is configured for a packet by testing the
197 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
198 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
200 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
201 * offloading the FCOE CRC in a packet. To perform this offload the stack
202 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
203 * accordingly. Note the there is no indication in the skbuff that the
204 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
205 * both IP checksum offload and FCOE CRC offload must verify which offload
206 * is configured for a packet presumably by inspecting packet headers.
208 * E. Checksumming on output with GSO.
210 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
211 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
212 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
213 * part of the GSO operation is implied. If a checksum is being offloaded
214 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
215 * are set to refer to the outermost checksum being offload (two offloaded
216 * checksums are possible with UDP encapsulation).
219 /* Don't change this without changing skb_csum_unnecessary! */
220 #define CHECKSUM_NONE 0
221 #define CHECKSUM_UNNECESSARY 1
222 #define CHECKSUM_COMPLETE 2
223 #define CHECKSUM_PARTIAL 3
225 /* Maximum value in skb->csum_level */
226 #define SKB_MAX_CSUM_LEVEL 3
228 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
229 #define SKB_WITH_OVERHEAD(X) \
230 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
231 #define SKB_MAX_ORDER(X, ORDER) \
232 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
233 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
234 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
236 /* return minimum truesize of one skb containing X bytes of data */
237 #define SKB_TRUESIZE(X) ((X) + \
238 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
239 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
243 struct pipe_inode_info;
247 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
248 struct nf_conntrack {
253 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
254 struct nf_bridge_info {
257 BRNF_PROTO_UNCHANGED,
265 struct net_device *physindev;
267 /* always valid & non-NULL from FORWARD on, for physdev match */
268 struct net_device *physoutdev;
270 /* prerouting: detect dnat in orig/reply direction */
272 struct in6_addr ipv6_daddr;
274 /* after prerouting + nat detected: store original source
275 * mac since neigh resolution overwrites it, only used while
276 * skb is out in neigh layer.
278 char neigh_header[8];
283 struct sk_buff_head {
284 /* These two members must be first. */
285 struct sk_buff *next;
286 struct sk_buff *prev;
294 /* To allow 64K frame to be packed as single skb without frag_list we
295 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
296 * buffers which do not start on a page boundary.
298 * Since GRO uses frags we allocate at least 16 regardless of page
301 #if (65536/PAGE_SIZE + 1) < 16
302 #define MAX_SKB_FRAGS 16UL
304 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
306 extern int sysctl_max_skb_frags;
308 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
309 * segment using its current segmentation instead.
311 #define GSO_BY_FRAGS 0xFFFF
313 typedef struct skb_frag_struct skb_frag_t;
315 struct skb_frag_struct {
319 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
328 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
333 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
338 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
343 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
348 #define HAVE_HW_TIME_STAMP
351 * struct skb_shared_hwtstamps - hardware time stamps
352 * @hwtstamp: hardware time stamp transformed into duration
353 * since arbitrary point in time
355 * Software time stamps generated by ktime_get_real() are stored in
358 * hwtstamps can only be compared against other hwtstamps from
361 * This structure is attached to packets as part of the
362 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
364 struct skb_shared_hwtstamps {
368 /* Definitions for tx_flags in struct skb_shared_info */
370 /* generate hardware time stamp */
371 SKBTX_HW_TSTAMP = 1 << 0,
373 /* generate software time stamp when queueing packet to NIC */
374 SKBTX_SW_TSTAMP = 1 << 1,
376 /* device driver is going to provide hardware time stamp */
377 SKBTX_IN_PROGRESS = 1 << 2,
379 /* device driver supports TX zero-copy buffers */
380 SKBTX_DEV_ZEROCOPY = 1 << 3,
382 /* generate wifi status information (where possible) */
383 SKBTX_WIFI_STATUS = 1 << 4,
385 /* This indicates at least one fragment might be overwritten
386 * (as in vmsplice(), sendfile() ...)
387 * If we need to compute a TX checksum, we'll need to copy
388 * all frags to avoid possible bad checksum
390 SKBTX_SHARED_FRAG = 1 << 5,
392 /* generate software time stamp when entering packet scheduling */
393 SKBTX_SCHED_TSTAMP = 1 << 6,
396 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
398 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
401 * The callback notifies userspace to release buffers when skb DMA is done in
402 * lower device, the skb last reference should be 0 when calling this.
403 * The zerocopy_success argument is true if zero copy transmit occurred,
404 * false on data copy or out of memory error caused by data copy attempt.
405 * The ctx field is used to track device context.
406 * The desc field is used to track userspace buffer index.
409 void (*callback)(struct ubuf_info *, bool zerocopy_success);
414 /* This data is invariant across clones and lives at
415 * the end of the header data, ie. at skb->end.
417 struct skb_shared_info {
418 unsigned short _unused;
419 unsigned char nr_frags;
421 unsigned short gso_size;
422 /* Warning: this field is not always filled in (UFO)! */
423 unsigned short gso_segs;
424 struct sk_buff *frag_list;
425 struct skb_shared_hwtstamps hwtstamps;
426 unsigned int gso_type;
431 * Warning : all fields before dataref are cleared in __alloc_skb()
435 /* Intermediate layers must ensure that destructor_arg
436 * remains valid until skb destructor */
437 void * destructor_arg;
439 /* must be last field, see pskb_expand_head() */
440 skb_frag_t frags[MAX_SKB_FRAGS];
443 /* We divide dataref into two halves. The higher 16 bits hold references
444 * to the payload part of skb->data. The lower 16 bits hold references to
445 * the entire skb->data. A clone of a headerless skb holds the length of
446 * the header in skb->hdr_len.
448 * All users must obey the rule that the skb->data reference count must be
449 * greater than or equal to the payload reference count.
451 * Holding a reference to the payload part means that the user does not
452 * care about modifications to the header part of skb->data.
454 #define SKB_DATAREF_SHIFT 16
455 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
459 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
460 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
461 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
465 SKB_GSO_TCPV4 = 1 << 0,
466 SKB_GSO_UDP = 1 << 1,
468 /* This indicates the skb is from an untrusted source. */
469 SKB_GSO_DODGY = 1 << 2,
471 /* This indicates the tcp segment has CWR set. */
472 SKB_GSO_TCP_ECN = 1 << 3,
474 SKB_GSO_TCP_FIXEDID = 1 << 4,
476 SKB_GSO_TCPV6 = 1 << 5,
478 SKB_GSO_FCOE = 1 << 6,
480 SKB_GSO_GRE = 1 << 7,
482 SKB_GSO_GRE_CSUM = 1 << 8,
484 SKB_GSO_IPXIP4 = 1 << 9,
486 SKB_GSO_IPXIP6 = 1 << 10,
488 SKB_GSO_UDP_TUNNEL = 1 << 11,
490 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
492 SKB_GSO_PARTIAL = 1 << 13,
494 SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
496 SKB_GSO_SCTP = 1 << 15,
498 SKB_GSO_ESP = 1 << 16,
501 #if BITS_PER_LONG > 32
502 #define NET_SKBUFF_DATA_USES_OFFSET 1
505 #ifdef NET_SKBUFF_DATA_USES_OFFSET
506 typedef unsigned int sk_buff_data_t;
508 typedef unsigned char *sk_buff_data_t;
512 * struct sk_buff - socket buffer
513 * @next: Next buffer in list
514 * @prev: Previous buffer in list
515 * @tstamp: Time we arrived/left
516 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
517 * @sk: Socket we are owned by
518 * @dev: Device we arrived on/are leaving by
519 * @cb: Control buffer. Free for use by every layer. Put private vars here
520 * @_skb_refdst: destination entry (with norefcount bit)
521 * @sp: the security path, used for xfrm
522 * @len: Length of actual data
523 * @data_len: Data length
524 * @mac_len: Length of link layer header
525 * @hdr_len: writable header length of cloned skb
526 * @csum: Checksum (must include start/offset pair)
527 * @csum_start: Offset from skb->head where checksumming should start
528 * @csum_offset: Offset from csum_start where checksum should be stored
529 * @priority: Packet queueing priority
530 * @ignore_df: allow local fragmentation
531 * @cloned: Head may be cloned (check refcnt to be sure)
532 * @ip_summed: Driver fed us an IP checksum
533 * @nohdr: Payload reference only, must not modify header
534 * @pkt_type: Packet class
535 * @fclone: skbuff clone status
536 * @ipvs_property: skbuff is owned by ipvs
537 * @tc_skip_classify: do not classify packet. set by IFB device
538 * @tc_at_ingress: used within tc_classify to distinguish in/egress
539 * @tc_redirected: packet was redirected by a tc action
540 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
541 * @peeked: this packet has been seen already, so stats have been
542 * done for it, don't do them again
543 * @nf_trace: netfilter packet trace flag
544 * @protocol: Packet protocol from driver
545 * @destructor: Destruct function
546 * @_nfct: Associated connection, if any (with nfctinfo bits)
547 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
548 * @skb_iif: ifindex of device we arrived on
549 * @tc_index: Traffic control index
550 * @hash: the packet hash
551 * @queue_mapping: Queue mapping for multiqueue devices
552 * @xmit_more: More SKBs are pending for this queue
553 * @ndisc_nodetype: router type (from link layer)
554 * @ooo_okay: allow the mapping of a socket to a queue to be changed
555 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
557 * @sw_hash: indicates hash was computed in software stack
558 * @wifi_acked_valid: wifi_acked was set
559 * @wifi_acked: whether frame was acked on wifi or not
560 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
561 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
562 * @dst_pending_confirm: need to confirm neighbour
563 * @napi_id: id of the NAPI struct this skb came from
564 * @secmark: security marking
565 * @mark: Generic packet mark
566 * @vlan_proto: vlan encapsulation protocol
567 * @vlan_tci: vlan tag control information
568 * @inner_protocol: Protocol (encapsulation)
569 * @inner_transport_header: Inner transport layer header (encapsulation)
570 * @inner_network_header: Network layer header (encapsulation)
571 * @inner_mac_header: Link layer header (encapsulation)
572 * @transport_header: Transport layer header
573 * @network_header: Network layer header
574 * @mac_header: Link layer header
575 * @tail: Tail pointer
577 * @head: Head of buffer
578 * @data: Data head pointer
579 * @truesize: Buffer size
580 * @users: User count - see {datagram,tcp}.c
586 /* These two members must be first. */
587 struct sk_buff *next;
588 struct sk_buff *prev;
595 struct rb_node rbnode; /* used in netem & tcp stack */
600 struct net_device *dev;
601 /* Some protocols might use this space to store information,
602 * while device pointer would be NULL.
603 * UDP receive path is one user.
605 unsigned long dev_scratch;
608 * This is the control buffer. It is free to use for every
609 * layer. Please put your private variables there. If you
610 * want to keep them across layers you have to do a skb_clone()
611 * first. This is owned by whoever has the skb queued ATM.
613 char cb[48] __aligned(8);
615 unsigned long _skb_refdst;
616 void (*destructor)(struct sk_buff *skb);
620 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
623 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
624 struct nf_bridge_info *nf_bridge;
631 /* Following fields are _not_ copied in __copy_skb_header()
632 * Note that queue_mapping is here mostly to fill a hole.
634 kmemcheck_bitfield_begin(flags1);
637 /* if you move cloned around you also must adapt those constants */
638 #ifdef __BIG_ENDIAN_BITFIELD
639 #define CLONED_MASK (1 << 7)
641 #define CLONED_MASK 1
643 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
645 __u8 __cloned_offset[0];
652 __unused:1; /* one bit hole */
653 kmemcheck_bitfield_end(flags1);
655 /* fields enclosed in headers_start/headers_end are copied
656 * using a single memcpy() in __copy_skb_header()
659 __u32 headers_start[0];
662 /* if you move pkt_type around you also must adapt those constants */
663 #ifdef __BIG_ENDIAN_BITFIELD
664 #define PKT_TYPE_MAX (7 << 5)
666 #define PKT_TYPE_MAX 7
668 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
670 __u8 __pkt_type_offset[0];
680 __u8 wifi_acked_valid:1;
684 /* Indicates the inner headers are valid in the skbuff. */
685 __u8 encapsulation:1;
686 __u8 encap_hdr_csum:1;
688 __u8 csum_complete_sw:1;
690 __u8 csum_not_inet:1;
692 __u8 dst_pending_confirm:1;
693 #ifdef CONFIG_IPV6_NDISC_NODETYPE
694 __u8 ndisc_nodetype:2;
696 __u8 ipvs_property:1;
697 __u8 inner_protocol_type:1;
698 __u8 remcsum_offload:1;
699 #ifdef CONFIG_NET_SWITCHDEV
700 __u8 offload_fwd_mark:1;
702 #ifdef CONFIG_NET_CLS_ACT
703 __u8 tc_skip_classify:1;
704 __u8 tc_at_ingress:1;
705 __u8 tc_redirected:1;
706 __u8 tc_from_ingress:1;
709 #ifdef CONFIG_NET_SCHED
710 __u16 tc_index; /* traffic control index */
725 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
727 unsigned int napi_id;
728 unsigned int sender_cpu;
731 #ifdef CONFIG_NETWORK_SECMARK
737 __u32 reserved_tailroom;
741 __be16 inner_protocol;
745 __u16 inner_transport_header;
746 __u16 inner_network_header;
747 __u16 inner_mac_header;
750 __u16 transport_header;
751 __u16 network_header;
755 __u32 headers_end[0];
758 /* These elements must be at the end, see alloc_skb() for details. */
763 unsigned int truesize;
769 * Handling routines are only of interest to the kernel
771 #include <linux/slab.h>
774 #define SKB_ALLOC_FCLONE 0x01
775 #define SKB_ALLOC_RX 0x02
776 #define SKB_ALLOC_NAPI 0x04
778 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
779 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
781 return unlikely(skb->pfmemalloc);
785 * skb might have a dst pointer attached, refcounted or not.
786 * _skb_refdst low order bit is set if refcount was _not_ taken
788 #define SKB_DST_NOREF 1UL
789 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
791 #define SKB_NFCT_PTRMASK ~(7UL)
793 * skb_dst - returns skb dst_entry
796 * Returns skb dst_entry, regardless of reference taken or not.
798 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
800 /* If refdst was not refcounted, check we still are in a
801 * rcu_read_lock section
803 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
804 !rcu_read_lock_held() &&
805 !rcu_read_lock_bh_held());
806 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
810 * skb_dst_set - sets skb dst
814 * Sets skb dst, assuming a reference was taken on dst and should
815 * be released by skb_dst_drop()
817 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
819 skb->_skb_refdst = (unsigned long)dst;
823 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
827 * Sets skb dst, assuming a reference was not taken on dst.
828 * If dst entry is cached, we do not take reference and dst_release
829 * will be avoided by refdst_drop. If dst entry is not cached, we take
830 * reference, so that last dst_release can destroy the dst immediately.
832 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
834 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
835 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
839 * skb_dst_is_noref - Test if skb dst isn't refcounted
842 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
844 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
847 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
849 return (struct rtable *)skb_dst(skb);
852 /* For mangling skb->pkt_type from user space side from applications
853 * such as nft, tc, etc, we only allow a conservative subset of
854 * possible pkt_types to be set.
856 static inline bool skb_pkt_type_ok(u32 ptype)
858 return ptype <= PACKET_OTHERHOST;
861 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
863 #ifdef CONFIG_NET_RX_BUSY_POLL
870 /* decrement the reference count and return true if we can free the skb */
871 static inline bool skb_unref(struct sk_buff *skb)
875 if (likely(refcount_read(&skb->users) == 1))
877 else if (likely(!refcount_dec_and_test(&skb->users)))
883 void skb_release_head_state(struct sk_buff *skb);
884 void kfree_skb(struct sk_buff *skb);
885 void kfree_skb_list(struct sk_buff *segs);
886 void skb_tx_error(struct sk_buff *skb);
887 void consume_skb(struct sk_buff *skb);
888 void consume_stateless_skb(struct sk_buff *skb);
889 void __kfree_skb(struct sk_buff *skb);
890 extern struct kmem_cache *skbuff_head_cache;
892 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
893 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
894 bool *fragstolen, int *delta_truesize);
896 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
898 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
899 struct sk_buff *build_skb(void *data, unsigned int frag_size);
900 static inline struct sk_buff *alloc_skb(unsigned int size,
903 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
906 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
907 unsigned long data_len,
912 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
913 struct sk_buff_fclones {
918 refcount_t fclone_ref;
922 * skb_fclone_busy - check if fclone is busy
926 * Returns true if skb is a fast clone, and its clone is not freed.
927 * Some drivers call skb_orphan() in their ndo_start_xmit(),
928 * so we also check that this didnt happen.
930 static inline bool skb_fclone_busy(const struct sock *sk,
931 const struct sk_buff *skb)
933 const struct sk_buff_fclones *fclones;
935 fclones = container_of(skb, struct sk_buff_fclones, skb1);
937 return skb->fclone == SKB_FCLONE_ORIG &&
938 refcount_read(&fclones->fclone_ref) > 1 &&
939 fclones->skb2.sk == sk;
942 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
945 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
948 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
949 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
951 return __alloc_skb_head(priority, -1);
954 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
955 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
956 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
957 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
958 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
959 gfp_t gfp_mask, bool fclone);
960 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
963 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
966 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
967 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
968 unsigned int headroom);
969 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
970 int newtailroom, gfp_t priority);
971 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
972 int offset, int len);
973 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
974 int offset, int len);
975 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
976 int skb_pad(struct sk_buff *skb, int pad);
977 #define dev_kfree_skb(a) consume_skb(a)
979 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
980 int getfrag(void *from, char *to, int offset,
981 int len, int odd, struct sk_buff *skb),
982 void *from, int length);
984 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
985 int offset, size_t size);
987 struct skb_seq_state {
991 __u32 stepped_offset;
992 struct sk_buff *root_skb;
993 struct sk_buff *cur_skb;
997 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
998 unsigned int to, struct skb_seq_state *st);
999 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1000 struct skb_seq_state *st);
1001 void skb_abort_seq_read(struct skb_seq_state *st);
1003 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1004 unsigned int to, struct ts_config *config);
1007 * Packet hash types specify the type of hash in skb_set_hash.
1009 * Hash types refer to the protocol layer addresses which are used to
1010 * construct a packet's hash. The hashes are used to differentiate or identify
1011 * flows of the protocol layer for the hash type. Hash types are either
1012 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1014 * Properties of hashes:
1016 * 1) Two packets in different flows have different hash values
1017 * 2) Two packets in the same flow should have the same hash value
1019 * A hash at a higher layer is considered to be more specific. A driver should
1020 * set the most specific hash possible.
1022 * A driver cannot indicate a more specific hash than the layer at which a hash
1023 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1025 * A driver may indicate a hash level which is less specific than the
1026 * actual layer the hash was computed on. For instance, a hash computed
1027 * at L4 may be considered an L3 hash. This should only be done if the
1028 * driver can't unambiguously determine that the HW computed the hash at
1029 * the higher layer. Note that the "should" in the second property above
1032 enum pkt_hash_types {
1033 PKT_HASH_TYPE_NONE, /* Undefined type */
1034 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1035 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1036 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1039 static inline void skb_clear_hash(struct sk_buff *skb)
1046 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1049 skb_clear_hash(skb);
1053 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1055 skb->l4_hash = is_l4;
1056 skb->sw_hash = is_sw;
1061 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1063 /* Used by drivers to set hash from HW */
1064 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1068 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1070 __skb_set_hash(skb, hash, true, is_l4);
1073 void __skb_get_hash(struct sk_buff *skb);
1074 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1075 u32 skb_get_poff(const struct sk_buff *skb);
1076 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1077 const struct flow_keys *keys, int hlen);
1078 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1079 void *data, int hlen_proto);
1081 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1082 int thoff, u8 ip_proto)
1084 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1087 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1088 const struct flow_dissector_key *key,
1089 unsigned int key_count);
1091 bool __skb_flow_dissect(const struct sk_buff *skb,
1092 struct flow_dissector *flow_dissector,
1093 void *target_container,
1094 void *data, __be16 proto, int nhoff, int hlen,
1095 unsigned int flags);
1097 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1098 struct flow_dissector *flow_dissector,
1099 void *target_container, unsigned int flags)
1101 return __skb_flow_dissect(skb, flow_dissector, target_container,
1102 NULL, 0, 0, 0, flags);
1105 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1106 struct flow_keys *flow,
1109 memset(flow, 0, sizeof(*flow));
1110 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1111 NULL, 0, 0, 0, flags);
1114 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1115 void *data, __be16 proto,
1116 int nhoff, int hlen,
1119 memset(flow, 0, sizeof(*flow));
1120 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1121 data, proto, nhoff, hlen, flags);
1124 static inline __u32 skb_get_hash(struct sk_buff *skb)
1126 if (!skb->l4_hash && !skb->sw_hash)
1127 __skb_get_hash(skb);
1132 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1134 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1136 if (!skb->l4_hash && !skb->sw_hash) {
1137 struct flow_keys keys;
1138 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1140 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1146 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1148 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1150 if (!skb->l4_hash && !skb->sw_hash) {
1151 struct flow_keys keys;
1152 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
1154 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1160 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1162 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1167 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1169 to->hash = from->hash;
1170 to->sw_hash = from->sw_hash;
1171 to->l4_hash = from->l4_hash;
1174 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1175 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1177 return skb->head + skb->end;
1180 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1185 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1190 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1192 return skb->end - skb->head;
1197 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1199 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1201 return &skb_shinfo(skb)->hwtstamps;
1205 * skb_queue_empty - check if a queue is empty
1208 * Returns true if the queue is empty, false otherwise.
1210 static inline int skb_queue_empty(const struct sk_buff_head *list)
1212 return list->next == (const struct sk_buff *) list;
1216 * skb_queue_is_last - check if skb is the last entry in the queue
1220 * Returns true if @skb is the last buffer on the list.
1222 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1223 const struct sk_buff *skb)
1225 return skb->next == (const struct sk_buff *) list;
1229 * skb_queue_is_first - check if skb is the first entry in the queue
1233 * Returns true if @skb is the first buffer on the list.
1235 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1236 const struct sk_buff *skb)
1238 return skb->prev == (const struct sk_buff *) list;
1242 * skb_queue_next - return the next packet in the queue
1244 * @skb: current buffer
1246 * Return the next packet in @list after @skb. It is only valid to
1247 * call this if skb_queue_is_last() evaluates to false.
1249 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1250 const struct sk_buff *skb)
1252 /* This BUG_ON may seem severe, but if we just return then we
1253 * are going to dereference garbage.
1255 BUG_ON(skb_queue_is_last(list, skb));
1260 * skb_queue_prev - return the prev packet in the queue
1262 * @skb: current buffer
1264 * Return the prev packet in @list before @skb. It is only valid to
1265 * call this if skb_queue_is_first() evaluates to false.
1267 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1268 const struct sk_buff *skb)
1270 /* This BUG_ON may seem severe, but if we just return then we
1271 * are going to dereference garbage.
1273 BUG_ON(skb_queue_is_first(list, skb));
1278 * skb_get - reference buffer
1279 * @skb: buffer to reference
1281 * Makes another reference to a socket buffer and returns a pointer
1284 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1286 refcount_inc(&skb->users);
1291 * If users == 1, we are the only owner and are can avoid redundant
1296 * skb_cloned - is the buffer a clone
1297 * @skb: buffer to check
1299 * Returns true if the buffer was generated with skb_clone() and is
1300 * one of multiple shared copies of the buffer. Cloned buffers are
1301 * shared data so must not be written to under normal circumstances.
1303 static inline int skb_cloned(const struct sk_buff *skb)
1305 return skb->cloned &&
1306 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1309 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1311 might_sleep_if(gfpflags_allow_blocking(pri));
1313 if (skb_cloned(skb))
1314 return pskb_expand_head(skb, 0, 0, pri);
1320 * skb_header_cloned - is the header a clone
1321 * @skb: buffer to check
1323 * Returns true if modifying the header part of the buffer requires
1324 * the data to be copied.
1326 static inline int skb_header_cloned(const struct sk_buff *skb)
1333 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1334 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1335 return dataref != 1;
1338 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1340 might_sleep_if(gfpflags_allow_blocking(pri));
1342 if (skb_header_cloned(skb))
1343 return pskb_expand_head(skb, 0, 0, pri);
1349 * skb_header_release - release reference to header
1350 * @skb: buffer to operate on
1352 * Drop a reference to the header part of the buffer. This is done
1353 * by acquiring a payload reference. You must not read from the header
1354 * part of skb->data after this.
1355 * Note : Check if you can use __skb_header_release() instead.
1357 static inline void skb_header_release(struct sk_buff *skb)
1361 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1365 * __skb_header_release - release reference to header
1366 * @skb: buffer to operate on
1368 * Variant of skb_header_release() assuming skb is private to caller.
1369 * We can avoid one atomic operation.
1371 static inline void __skb_header_release(struct sk_buff *skb)
1374 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1379 * skb_shared - is the buffer shared
1380 * @skb: buffer to check
1382 * Returns true if more than one person has a reference to this
1385 static inline int skb_shared(const struct sk_buff *skb)
1387 return refcount_read(&skb->users) != 1;
1391 * skb_share_check - check if buffer is shared and if so clone it
1392 * @skb: buffer to check
1393 * @pri: priority for memory allocation
1395 * If the buffer is shared the buffer is cloned and the old copy
1396 * drops a reference. A new clone with a single reference is returned.
1397 * If the buffer is not shared the original buffer is returned. When
1398 * being called from interrupt status or with spinlocks held pri must
1401 * NULL is returned on a memory allocation failure.
1403 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1405 might_sleep_if(gfpflags_allow_blocking(pri));
1406 if (skb_shared(skb)) {
1407 struct sk_buff *nskb = skb_clone(skb, pri);
1419 * Copy shared buffers into a new sk_buff. We effectively do COW on
1420 * packets to handle cases where we have a local reader and forward
1421 * and a couple of other messy ones. The normal one is tcpdumping
1422 * a packet thats being forwarded.
1426 * skb_unshare - make a copy of a shared buffer
1427 * @skb: buffer to check
1428 * @pri: priority for memory allocation
1430 * If the socket buffer is a clone then this function creates a new
1431 * copy of the data, drops a reference count on the old copy and returns
1432 * the new copy with the reference count at 1. If the buffer is not a clone
1433 * the original buffer is returned. When called with a spinlock held or
1434 * from interrupt state @pri must be %GFP_ATOMIC
1436 * %NULL is returned on a memory allocation failure.
1438 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1441 might_sleep_if(gfpflags_allow_blocking(pri));
1442 if (skb_cloned(skb)) {
1443 struct sk_buff *nskb = skb_copy(skb, pri);
1445 /* Free our shared copy */
1456 * skb_peek - peek at the head of an &sk_buff_head
1457 * @list_: list to peek at
1459 * Peek an &sk_buff. Unlike most other operations you _MUST_
1460 * be careful with this one. A peek leaves the buffer on the
1461 * list and someone else may run off with it. You must hold
1462 * the appropriate locks or have a private queue to do this.
1464 * Returns %NULL for an empty list or a pointer to the head element.
1465 * The reference count is not incremented and the reference is therefore
1466 * volatile. Use with caution.
1468 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1470 struct sk_buff *skb = list_->next;
1472 if (skb == (struct sk_buff *)list_)
1478 * skb_peek_next - peek skb following the given one from a queue
1479 * @skb: skb to start from
1480 * @list_: list to peek at
1482 * Returns %NULL when the end of the list is met or a pointer to the
1483 * next element. The reference count is not incremented and the
1484 * reference is therefore volatile. Use with caution.
1486 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1487 const struct sk_buff_head *list_)
1489 struct sk_buff *next = skb->next;
1491 if (next == (struct sk_buff *)list_)
1497 * skb_peek_tail - peek at the tail of an &sk_buff_head
1498 * @list_: list to peek at
1500 * Peek an &sk_buff. Unlike most other operations you _MUST_
1501 * be careful with this one. A peek leaves the buffer on the
1502 * list and someone else may run off with it. You must hold
1503 * the appropriate locks or have a private queue to do this.
1505 * Returns %NULL for an empty list or a pointer to the tail element.
1506 * The reference count is not incremented and the reference is therefore
1507 * volatile. Use with caution.
1509 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1511 struct sk_buff *skb = list_->prev;
1513 if (skb == (struct sk_buff *)list_)
1520 * skb_queue_len - get queue length
1521 * @list_: list to measure
1523 * Return the length of an &sk_buff queue.
1525 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1531 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1532 * @list: queue to initialize
1534 * This initializes only the list and queue length aspects of
1535 * an sk_buff_head object. This allows to initialize the list
1536 * aspects of an sk_buff_head without reinitializing things like
1537 * the spinlock. It can also be used for on-stack sk_buff_head
1538 * objects where the spinlock is known to not be used.
1540 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1542 list->prev = list->next = (struct sk_buff *)list;
1547 * This function creates a split out lock class for each invocation;
1548 * this is needed for now since a whole lot of users of the skb-queue
1549 * infrastructure in drivers have different locking usage (in hardirq)
1550 * than the networking core (in softirq only). In the long run either the
1551 * network layer or drivers should need annotation to consolidate the
1552 * main types of usage into 3 classes.
1554 static inline void skb_queue_head_init(struct sk_buff_head *list)
1556 spin_lock_init(&list->lock);
1557 __skb_queue_head_init(list);
1560 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1561 struct lock_class_key *class)
1563 skb_queue_head_init(list);
1564 lockdep_set_class(&list->lock, class);
1568 * Insert an sk_buff on a list.
1570 * The "__skb_xxxx()" functions are the non-atomic ones that
1571 * can only be called with interrupts disabled.
1573 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1574 struct sk_buff_head *list);
1575 static inline void __skb_insert(struct sk_buff *newsk,
1576 struct sk_buff *prev, struct sk_buff *next,
1577 struct sk_buff_head *list)
1581 next->prev = prev->next = newsk;
1585 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1586 struct sk_buff *prev,
1587 struct sk_buff *next)
1589 struct sk_buff *first = list->next;
1590 struct sk_buff *last = list->prev;
1600 * skb_queue_splice - join two skb lists, this is designed for stacks
1601 * @list: the new list to add
1602 * @head: the place to add it in the first list
1604 static inline void skb_queue_splice(const struct sk_buff_head *list,
1605 struct sk_buff_head *head)
1607 if (!skb_queue_empty(list)) {
1608 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1609 head->qlen += list->qlen;
1614 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1615 * @list: the new list to add
1616 * @head: the place to add it in the first list
1618 * The list at @list is reinitialised
1620 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1621 struct sk_buff_head *head)
1623 if (!skb_queue_empty(list)) {
1624 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1625 head->qlen += list->qlen;
1626 __skb_queue_head_init(list);
1631 * skb_queue_splice_tail - join two skb lists, each list being a queue
1632 * @list: the new list to add
1633 * @head: the place to add it in the first list
1635 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1636 struct sk_buff_head *head)
1638 if (!skb_queue_empty(list)) {
1639 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1640 head->qlen += list->qlen;
1645 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1646 * @list: the new list to add
1647 * @head: the place to add it in the first list
1649 * Each of the lists is a queue.
1650 * The list at @list is reinitialised
1652 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1653 struct sk_buff_head *head)
1655 if (!skb_queue_empty(list)) {
1656 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1657 head->qlen += list->qlen;
1658 __skb_queue_head_init(list);
1663 * __skb_queue_after - queue a buffer at the list head
1664 * @list: list to use
1665 * @prev: place after this buffer
1666 * @newsk: buffer to queue
1668 * Queue a buffer int the middle of a list. This function takes no locks
1669 * and you must therefore hold required locks before calling it.
1671 * A buffer cannot be placed on two lists at the same time.
1673 static inline void __skb_queue_after(struct sk_buff_head *list,
1674 struct sk_buff *prev,
1675 struct sk_buff *newsk)
1677 __skb_insert(newsk, prev, prev->next, list);
1680 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1681 struct sk_buff_head *list);
1683 static inline void __skb_queue_before(struct sk_buff_head *list,
1684 struct sk_buff *next,
1685 struct sk_buff *newsk)
1687 __skb_insert(newsk, next->prev, next, list);
1691 * __skb_queue_head - queue a buffer at the list head
1692 * @list: list to use
1693 * @newsk: buffer to queue
1695 * Queue a buffer at the start of a list. This function takes no locks
1696 * and you must therefore hold required locks before calling it.
1698 * A buffer cannot be placed on two lists at the same time.
1700 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1701 static inline void __skb_queue_head(struct sk_buff_head *list,
1702 struct sk_buff *newsk)
1704 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1708 * __skb_queue_tail - queue a buffer at the list tail
1709 * @list: list to use
1710 * @newsk: buffer to queue
1712 * Queue a buffer at the end of a list. This function takes no locks
1713 * and you must therefore hold required locks before calling it.
1715 * A buffer cannot be placed on two lists at the same time.
1717 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1718 static inline void __skb_queue_tail(struct sk_buff_head *list,
1719 struct sk_buff *newsk)
1721 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1725 * remove sk_buff from list. _Must_ be called atomically, and with
1728 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1729 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1731 struct sk_buff *next, *prev;
1736 skb->next = skb->prev = NULL;
1742 * __skb_dequeue - remove from the head of the queue
1743 * @list: list to dequeue from
1745 * Remove the head of the list. This function does not take any locks
1746 * so must be used with appropriate locks held only. The head item is
1747 * returned or %NULL if the list is empty.
1749 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1750 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1752 struct sk_buff *skb = skb_peek(list);
1754 __skb_unlink(skb, list);
1759 * __skb_dequeue_tail - remove from the tail of the queue
1760 * @list: list to dequeue from
1762 * Remove the tail of the list. This function does not take any locks
1763 * so must be used with appropriate locks held only. The tail item is
1764 * returned or %NULL if the list is empty.
1766 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1767 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1769 struct sk_buff *skb = skb_peek_tail(list);
1771 __skb_unlink(skb, list);
1776 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1778 return skb->data_len;
1781 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1783 return skb->len - skb->data_len;
1786 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1788 unsigned int i, len = 0;
1790 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1791 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1792 return len + skb_headlen(skb);
1796 * __skb_fill_page_desc - initialise a paged fragment in an skb
1797 * @skb: buffer containing fragment to be initialised
1798 * @i: paged fragment index to initialise
1799 * @page: the page to use for this fragment
1800 * @off: the offset to the data with @page
1801 * @size: the length of the data
1803 * Initialises the @i'th fragment of @skb to point to &size bytes at
1804 * offset @off within @page.
1806 * Does not take any additional reference on the fragment.
1808 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1809 struct page *page, int off, int size)
1811 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1814 * Propagate page pfmemalloc to the skb if we can. The problem is
1815 * that not all callers have unique ownership of the page but rely
1816 * on page_is_pfmemalloc doing the right thing(tm).
1818 frag->page.p = page;
1819 frag->page_offset = off;
1820 skb_frag_size_set(frag, size);
1822 page = compound_head(page);
1823 if (page_is_pfmemalloc(page))
1824 skb->pfmemalloc = true;
1828 * skb_fill_page_desc - initialise a paged fragment in an skb
1829 * @skb: buffer containing fragment to be initialised
1830 * @i: paged fragment index to initialise
1831 * @page: the page to use for this fragment
1832 * @off: the offset to the data with @page
1833 * @size: the length of the data
1835 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1836 * @skb to point to @size bytes at offset @off within @page. In
1837 * addition updates @skb such that @i is the last fragment.
1839 * Does not take any additional reference on the fragment.
1841 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1842 struct page *page, int off, int size)
1844 __skb_fill_page_desc(skb, i, page, off, size);
1845 skb_shinfo(skb)->nr_frags = i + 1;
1848 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1849 int size, unsigned int truesize);
1851 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1852 unsigned int truesize);
1854 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1855 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1856 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1858 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1859 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1861 return skb->head + skb->tail;
1864 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1866 skb->tail = skb->data - skb->head;
1869 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1871 skb_reset_tail_pointer(skb);
1872 skb->tail += offset;
1875 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1876 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1881 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1883 skb->tail = skb->data;
1886 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1888 skb->tail = skb->data + offset;
1891 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1894 * Add data to an sk_buff
1896 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1897 void *skb_put(struct sk_buff *skb, unsigned int len);
1898 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
1900 void *tmp = skb_tail_pointer(skb);
1901 SKB_LINEAR_ASSERT(skb);
1907 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
1909 void *tmp = __skb_put(skb, len);
1911 memset(tmp, 0, len);
1915 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
1918 void *tmp = __skb_put(skb, len);
1920 memcpy(tmp, data, len);
1924 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
1926 *(u8 *)__skb_put(skb, 1) = val;
1929 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
1931 void *tmp = skb_put(skb, len);
1933 memset(tmp, 0, len);
1938 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
1941 void *tmp = skb_put(skb, len);
1943 memcpy(tmp, data, len);
1948 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
1950 *(u8 *)skb_put(skb, 1) = val;
1953 void *skb_push(struct sk_buff *skb, unsigned int len);
1954 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
1961 void *skb_pull(struct sk_buff *skb, unsigned int len);
1962 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
1965 BUG_ON(skb->len < skb->data_len);
1966 return skb->data += len;
1969 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1971 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1974 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
1976 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
1978 if (len > skb_headlen(skb) &&
1979 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1982 return skb->data += len;
1985 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
1987 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1990 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1992 if (likely(len <= skb_headlen(skb)))
1994 if (unlikely(len > skb->len))
1996 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1999 void skb_condense(struct sk_buff *skb);
2002 * skb_headroom - bytes at buffer head
2003 * @skb: buffer to check
2005 * Return the number of bytes of free space at the head of an &sk_buff.
2007 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2009 return skb->data - skb->head;
2013 * skb_tailroom - bytes at buffer end
2014 * @skb: buffer to check
2016 * Return the number of bytes of free space at the tail of an sk_buff
2018 static inline int skb_tailroom(const struct sk_buff *skb)
2020 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2024 * skb_availroom - bytes at buffer end
2025 * @skb: buffer to check
2027 * Return the number of bytes of free space at the tail of an sk_buff
2028 * allocated by sk_stream_alloc()
2030 static inline int skb_availroom(const struct sk_buff *skb)
2032 if (skb_is_nonlinear(skb))
2035 return skb->end - skb->tail - skb->reserved_tailroom;
2039 * skb_reserve - adjust headroom
2040 * @skb: buffer to alter
2041 * @len: bytes to move
2043 * Increase the headroom of an empty &sk_buff by reducing the tail
2044 * room. This is only allowed for an empty buffer.
2046 static inline void skb_reserve(struct sk_buff *skb, int len)
2053 * skb_tailroom_reserve - adjust reserved_tailroom
2054 * @skb: buffer to alter
2055 * @mtu: maximum amount of headlen permitted
2056 * @needed_tailroom: minimum amount of reserved_tailroom
2058 * Set reserved_tailroom so that headlen can be as large as possible but
2059 * not larger than mtu and tailroom cannot be smaller than
2061 * The required headroom should already have been reserved before using
2064 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2065 unsigned int needed_tailroom)
2067 SKB_LINEAR_ASSERT(skb);
2068 if (mtu < skb_tailroom(skb) - needed_tailroom)
2069 /* use at most mtu */
2070 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2072 /* use up to all available space */
2073 skb->reserved_tailroom = needed_tailroom;
2076 #define ENCAP_TYPE_ETHER 0
2077 #define ENCAP_TYPE_IPPROTO 1
2079 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2082 skb->inner_protocol = protocol;
2083 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2086 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2089 skb->inner_ipproto = ipproto;
2090 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2093 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2095 skb->inner_mac_header = skb->mac_header;
2096 skb->inner_network_header = skb->network_header;
2097 skb->inner_transport_header = skb->transport_header;
2100 static inline void skb_reset_mac_len(struct sk_buff *skb)
2102 skb->mac_len = skb->network_header - skb->mac_header;
2105 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2108 return skb->head + skb->inner_transport_header;
2111 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2113 return skb_inner_transport_header(skb) - skb->data;
2116 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2118 skb->inner_transport_header = skb->data - skb->head;
2121 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2124 skb_reset_inner_transport_header(skb);
2125 skb->inner_transport_header += offset;
2128 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2130 return skb->head + skb->inner_network_header;
2133 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2135 skb->inner_network_header = skb->data - skb->head;
2138 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2141 skb_reset_inner_network_header(skb);
2142 skb->inner_network_header += offset;
2145 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2147 return skb->head + skb->inner_mac_header;
2150 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2152 skb->inner_mac_header = skb->data - skb->head;
2155 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2158 skb_reset_inner_mac_header(skb);
2159 skb->inner_mac_header += offset;
2161 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2163 return skb->transport_header != (typeof(skb->transport_header))~0U;
2166 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2168 return skb->head + skb->transport_header;
2171 static inline void skb_reset_transport_header(struct sk_buff *skb)
2173 skb->transport_header = skb->data - skb->head;
2176 static inline void skb_set_transport_header(struct sk_buff *skb,
2179 skb_reset_transport_header(skb);
2180 skb->transport_header += offset;
2183 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2185 return skb->head + skb->network_header;
2188 static inline void skb_reset_network_header(struct sk_buff *skb)
2190 skb->network_header = skb->data - skb->head;
2193 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2195 skb_reset_network_header(skb);
2196 skb->network_header += offset;
2199 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2201 return skb->head + skb->mac_header;
2204 static inline int skb_mac_offset(const struct sk_buff *skb)
2206 return skb_mac_header(skb) - skb->data;
2209 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2211 return skb->network_header - skb->mac_header;
2214 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2216 return skb->mac_header != (typeof(skb->mac_header))~0U;
2219 static inline void skb_reset_mac_header(struct sk_buff *skb)
2221 skb->mac_header = skb->data - skb->head;
2224 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2226 skb_reset_mac_header(skb);
2227 skb->mac_header += offset;
2230 static inline void skb_pop_mac_header(struct sk_buff *skb)
2232 skb->mac_header = skb->network_header;
2235 static inline void skb_probe_transport_header(struct sk_buff *skb,
2236 const int offset_hint)
2238 struct flow_keys keys;
2240 if (skb_transport_header_was_set(skb))
2242 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2243 skb_set_transport_header(skb, keys.control.thoff);
2245 skb_set_transport_header(skb, offset_hint);
2248 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2250 if (skb_mac_header_was_set(skb)) {
2251 const unsigned char *old_mac = skb_mac_header(skb);
2253 skb_set_mac_header(skb, -skb->mac_len);
2254 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2258 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2260 return skb->csum_start - skb_headroom(skb);
2263 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2265 return skb->head + skb->csum_start;
2268 static inline int skb_transport_offset(const struct sk_buff *skb)
2270 return skb_transport_header(skb) - skb->data;
2273 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2275 return skb->transport_header - skb->network_header;
2278 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2280 return skb->inner_transport_header - skb->inner_network_header;
2283 static inline int skb_network_offset(const struct sk_buff *skb)
2285 return skb_network_header(skb) - skb->data;
2288 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2290 return skb_inner_network_header(skb) - skb->data;
2293 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2295 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2299 * CPUs often take a performance hit when accessing unaligned memory
2300 * locations. The actual performance hit varies, it can be small if the
2301 * hardware handles it or large if we have to take an exception and fix it
2304 * Since an ethernet header is 14 bytes network drivers often end up with
2305 * the IP header at an unaligned offset. The IP header can be aligned by
2306 * shifting the start of the packet by 2 bytes. Drivers should do this
2309 * skb_reserve(skb, NET_IP_ALIGN);
2311 * The downside to this alignment of the IP header is that the DMA is now
2312 * unaligned. On some architectures the cost of an unaligned DMA is high
2313 * and this cost outweighs the gains made by aligning the IP header.
2315 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2318 #ifndef NET_IP_ALIGN
2319 #define NET_IP_ALIGN 2
2323 * The networking layer reserves some headroom in skb data (via
2324 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2325 * the header has to grow. In the default case, if the header has to grow
2326 * 32 bytes or less we avoid the reallocation.
2328 * Unfortunately this headroom changes the DMA alignment of the resulting
2329 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2330 * on some architectures. An architecture can override this value,
2331 * perhaps setting it to a cacheline in size (since that will maintain
2332 * cacheline alignment of the DMA). It must be a power of 2.
2334 * Various parts of the networking layer expect at least 32 bytes of
2335 * headroom, you should not reduce this.
2337 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2338 * to reduce average number of cache lines per packet.
2339 * get_rps_cpus() for example only access one 64 bytes aligned block :
2340 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2343 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2346 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2348 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2350 if (unlikely(skb_is_nonlinear(skb))) {
2355 skb_set_tail_pointer(skb, len);
2358 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2360 __skb_set_length(skb, len);
2363 void skb_trim(struct sk_buff *skb, unsigned int len);
2365 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2368 return ___pskb_trim(skb, len);
2369 __skb_trim(skb, len);
2373 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2375 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2379 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2380 * @skb: buffer to alter
2383 * This is identical to pskb_trim except that the caller knows that
2384 * the skb is not cloned so we should never get an error due to out-
2387 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2389 int err = pskb_trim(skb, len);
2393 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2395 unsigned int diff = len - skb->len;
2397 if (skb_tailroom(skb) < diff) {
2398 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2403 __skb_set_length(skb, len);
2408 * skb_orphan - orphan a buffer
2409 * @skb: buffer to orphan
2411 * If a buffer currently has an owner then we call the owner's
2412 * destructor function and make the @skb unowned. The buffer continues
2413 * to exist but is no longer charged to its former owner.
2415 static inline void skb_orphan(struct sk_buff *skb)
2417 if (skb->destructor) {
2418 skb->destructor(skb);
2419 skb->destructor = NULL;
2427 * skb_orphan_frags - orphan the frags contained in a buffer
2428 * @skb: buffer to orphan frags from
2429 * @gfp_mask: allocation mask for replacement pages
2431 * For each frag in the SKB which needs a destructor (i.e. has an
2432 * owner) create a copy of that frag and release the original
2433 * page by calling the destructor.
2435 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2437 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2439 return skb_copy_ubufs(skb, gfp_mask);
2443 * __skb_queue_purge - empty a list
2444 * @list: list to empty
2446 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2447 * the list and one reference dropped. This function does not take the
2448 * list lock and the caller must hold the relevant locks to use it.
2450 void skb_queue_purge(struct sk_buff_head *list);
2451 static inline void __skb_queue_purge(struct sk_buff_head *list)
2453 struct sk_buff *skb;
2454 while ((skb = __skb_dequeue(list)) != NULL)
2458 void skb_rbtree_purge(struct rb_root *root);
2460 void *netdev_alloc_frag(unsigned int fragsz);
2462 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2466 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2467 * @dev: network device to receive on
2468 * @length: length to allocate
2470 * Allocate a new &sk_buff and assign it a usage count of one. The
2471 * buffer has unspecified headroom built in. Users should allocate
2472 * the headroom they think they need without accounting for the
2473 * built in space. The built in space is used for optimisations.
2475 * %NULL is returned if there is no free memory. Although this function
2476 * allocates memory it can be called from an interrupt.
2478 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2479 unsigned int length)
2481 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2484 /* legacy helper around __netdev_alloc_skb() */
2485 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2488 return __netdev_alloc_skb(NULL, length, gfp_mask);
2491 /* legacy helper around netdev_alloc_skb() */
2492 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2494 return netdev_alloc_skb(NULL, length);
2498 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2499 unsigned int length, gfp_t gfp)
2501 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2503 if (NET_IP_ALIGN && skb)
2504 skb_reserve(skb, NET_IP_ALIGN);
2508 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2509 unsigned int length)
2511 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2514 static inline void skb_free_frag(void *addr)
2516 page_frag_free(addr);
2519 void *napi_alloc_frag(unsigned int fragsz);
2520 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2521 unsigned int length, gfp_t gfp_mask);
2522 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2523 unsigned int length)
2525 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2527 void napi_consume_skb(struct sk_buff *skb, int budget);
2529 void __kfree_skb_flush(void);
2530 void __kfree_skb_defer(struct sk_buff *skb);
2533 * __dev_alloc_pages - allocate page for network Rx
2534 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2535 * @order: size of the allocation
2537 * Allocate a new page.
2539 * %NULL is returned if there is no free memory.
2541 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2544 /* This piece of code contains several assumptions.
2545 * 1. This is for device Rx, therefor a cold page is preferred.
2546 * 2. The expectation is the user wants a compound page.
2547 * 3. If requesting a order 0 page it will not be compound
2548 * due to the check to see if order has a value in prep_new_page
2549 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2550 * code in gfp_to_alloc_flags that should be enforcing this.
2552 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2554 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2557 static inline struct page *dev_alloc_pages(unsigned int order)
2559 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2563 * __dev_alloc_page - allocate a page for network Rx
2564 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2566 * Allocate a new page.
2568 * %NULL is returned if there is no free memory.
2570 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2572 return __dev_alloc_pages(gfp_mask, 0);
2575 static inline struct page *dev_alloc_page(void)
2577 return dev_alloc_pages(0);
2581 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2582 * @page: The page that was allocated from skb_alloc_page
2583 * @skb: The skb that may need pfmemalloc set
2585 static inline void skb_propagate_pfmemalloc(struct page *page,
2586 struct sk_buff *skb)
2588 if (page_is_pfmemalloc(page))
2589 skb->pfmemalloc = true;
2593 * skb_frag_page - retrieve the page referred to by a paged fragment
2594 * @frag: the paged fragment
2596 * Returns the &struct page associated with @frag.
2598 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2600 return frag->page.p;
2604 * __skb_frag_ref - take an addition reference on a paged fragment.
2605 * @frag: the paged fragment
2607 * Takes an additional reference on the paged fragment @frag.
2609 static inline void __skb_frag_ref(skb_frag_t *frag)
2611 get_page(skb_frag_page(frag));
2615 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2617 * @f: the fragment offset.
2619 * Takes an additional reference on the @f'th paged fragment of @skb.
2621 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2623 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2627 * __skb_frag_unref - release a reference on a paged fragment.
2628 * @frag: the paged fragment
2630 * Releases a reference on the paged fragment @frag.
2632 static inline void __skb_frag_unref(skb_frag_t *frag)
2634 put_page(skb_frag_page(frag));
2638 * skb_frag_unref - release a reference on a paged fragment of an skb.
2640 * @f: the fragment offset
2642 * Releases a reference on the @f'th paged fragment of @skb.
2644 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2646 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2650 * skb_frag_address - gets the address of the data contained in a paged fragment
2651 * @frag: the paged fragment buffer
2653 * Returns the address of the data within @frag. The page must already
2656 static inline void *skb_frag_address(const skb_frag_t *frag)
2658 return page_address(skb_frag_page(frag)) + frag->page_offset;
2662 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2663 * @frag: the paged fragment buffer
2665 * Returns the address of the data within @frag. Checks that the page
2666 * is mapped and returns %NULL otherwise.
2668 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2670 void *ptr = page_address(skb_frag_page(frag));
2674 return ptr + frag->page_offset;
2678 * __skb_frag_set_page - sets the page contained in a paged fragment
2679 * @frag: the paged fragment
2680 * @page: the page to set
2682 * Sets the fragment @frag to contain @page.
2684 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2686 frag->page.p = page;
2690 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2692 * @f: the fragment offset
2693 * @page: the page to set
2695 * Sets the @f'th fragment of @skb to contain @page.
2697 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2700 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2703 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2706 * skb_frag_dma_map - maps a paged fragment via the DMA API
2707 * @dev: the device to map the fragment to
2708 * @frag: the paged fragment to map
2709 * @offset: the offset within the fragment (starting at the
2710 * fragment's own offset)
2711 * @size: the number of bytes to map
2712 * @dir: the direction of the mapping (``PCI_DMA_*``)
2714 * Maps the page associated with @frag to @device.
2716 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2717 const skb_frag_t *frag,
2718 size_t offset, size_t size,
2719 enum dma_data_direction dir)
2721 return dma_map_page(dev, skb_frag_page(frag),
2722 frag->page_offset + offset, size, dir);
2725 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2728 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2732 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2735 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2740 * skb_clone_writable - is the header of a clone writable
2741 * @skb: buffer to check
2742 * @len: length up to which to write
2744 * Returns true if modifying the header part of the cloned buffer
2745 * does not requires the data to be copied.
2747 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2749 return !skb_header_cloned(skb) &&
2750 skb_headroom(skb) + len <= skb->hdr_len;
2753 static inline int skb_try_make_writable(struct sk_buff *skb,
2754 unsigned int write_len)
2756 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2757 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2760 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2765 if (headroom > skb_headroom(skb))
2766 delta = headroom - skb_headroom(skb);
2768 if (delta || cloned)
2769 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2775 * skb_cow - copy header of skb when it is required
2776 * @skb: buffer to cow
2777 * @headroom: needed headroom
2779 * If the skb passed lacks sufficient headroom or its data part
2780 * is shared, data is reallocated. If reallocation fails, an error
2781 * is returned and original skb is not changed.
2783 * The result is skb with writable area skb->head...skb->tail
2784 * and at least @headroom of space at head.
2786 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2788 return __skb_cow(skb, headroom, skb_cloned(skb));
2792 * skb_cow_head - skb_cow but only making the head writable
2793 * @skb: buffer to cow
2794 * @headroom: needed headroom
2796 * This function is identical to skb_cow except that we replace the
2797 * skb_cloned check by skb_header_cloned. It should be used when
2798 * you only need to push on some header and do not need to modify
2801 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2803 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2807 * skb_padto - pad an skbuff up to a minimal size
2808 * @skb: buffer to pad
2809 * @len: minimal length
2811 * Pads up a buffer to ensure the trailing bytes exist and are
2812 * blanked. If the buffer already contains sufficient data it
2813 * is untouched. Otherwise it is extended. Returns zero on
2814 * success. The skb is freed on error.
2816 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2818 unsigned int size = skb->len;
2819 if (likely(size >= len))
2821 return skb_pad(skb, len - size);
2825 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2826 * @skb: buffer to pad
2827 * @len: minimal length
2829 * Pads up a buffer to ensure the trailing bytes exist and are
2830 * blanked. If the buffer already contains sufficient data it
2831 * is untouched. Otherwise it is extended. Returns zero on
2832 * success. The skb is freed on error.
2834 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2836 unsigned int size = skb->len;
2838 if (unlikely(size < len)) {
2840 if (skb_pad(skb, len))
2842 __skb_put(skb, len);
2847 static inline int skb_add_data(struct sk_buff *skb,
2848 struct iov_iter *from, int copy)
2850 const int off = skb->len;
2852 if (skb->ip_summed == CHECKSUM_NONE) {
2854 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2856 skb->csum = csum_block_add(skb->csum, csum, off);
2859 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
2862 __skb_trim(skb, off);
2866 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2867 const struct page *page, int off)
2870 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2872 return page == skb_frag_page(frag) &&
2873 off == frag->page_offset + skb_frag_size(frag);
2878 static inline int __skb_linearize(struct sk_buff *skb)
2880 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2884 * skb_linearize - convert paged skb to linear one
2885 * @skb: buffer to linarize
2887 * If there is no free memory -ENOMEM is returned, otherwise zero
2888 * is returned and the old skb data released.
2890 static inline int skb_linearize(struct sk_buff *skb)
2892 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2896 * skb_has_shared_frag - can any frag be overwritten
2897 * @skb: buffer to test
2899 * Return true if the skb has at least one frag that might be modified
2900 * by an external entity (as in vmsplice()/sendfile())
2902 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2904 return skb_is_nonlinear(skb) &&
2905 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2909 * skb_linearize_cow - make sure skb is linear and writable
2910 * @skb: buffer to process
2912 * If there is no free memory -ENOMEM is returned, otherwise zero
2913 * is returned and the old skb data released.
2915 static inline int skb_linearize_cow(struct sk_buff *skb)
2917 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2918 __skb_linearize(skb) : 0;
2921 static __always_inline void
2922 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2925 if (skb->ip_summed == CHECKSUM_COMPLETE)
2926 skb->csum = csum_block_sub(skb->csum,
2927 csum_partial(start, len, 0), off);
2928 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2929 skb_checksum_start_offset(skb) < 0)
2930 skb->ip_summed = CHECKSUM_NONE;
2934 * skb_postpull_rcsum - update checksum for received skb after pull
2935 * @skb: buffer to update
2936 * @start: start of data before pull
2937 * @len: length of data pulled
2939 * After doing a pull on a received packet, you need to call this to
2940 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2941 * CHECKSUM_NONE so that it can be recomputed from scratch.
2943 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2944 const void *start, unsigned int len)
2946 __skb_postpull_rcsum(skb, start, len, 0);
2949 static __always_inline void
2950 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2953 if (skb->ip_summed == CHECKSUM_COMPLETE)
2954 skb->csum = csum_block_add(skb->csum,
2955 csum_partial(start, len, 0), off);
2959 * skb_postpush_rcsum - update checksum for received skb after push
2960 * @skb: buffer to update
2961 * @start: start of data after push
2962 * @len: length of data pushed
2964 * After doing a push on a received packet, you need to call this to
2965 * update the CHECKSUM_COMPLETE checksum.
2967 static inline void skb_postpush_rcsum(struct sk_buff *skb,
2968 const void *start, unsigned int len)
2970 __skb_postpush_rcsum(skb, start, len, 0);
2973 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2976 * skb_push_rcsum - push skb and update receive checksum
2977 * @skb: buffer to update
2978 * @len: length of data pulled
2980 * This function performs an skb_push on the packet and updates
2981 * the CHECKSUM_COMPLETE checksum. It should be used on
2982 * receive path processing instead of skb_push unless you know
2983 * that the checksum difference is zero (e.g., a valid IP header)
2984 * or you are setting ip_summed to CHECKSUM_NONE.
2986 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
2989 skb_postpush_rcsum(skb, skb->data, len);
2994 * pskb_trim_rcsum - trim received skb and update checksum
2995 * @skb: buffer to trim
2998 * This is exactly the same as pskb_trim except that it ensures the
2999 * checksum of received packets are still valid after the operation.
3002 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3004 if (likely(len >= skb->len))
3006 if (skb->ip_summed == CHECKSUM_COMPLETE)
3007 skb->ip_summed = CHECKSUM_NONE;
3008 return __pskb_trim(skb, len);
3011 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3013 if (skb->ip_summed == CHECKSUM_COMPLETE)
3014 skb->ip_summed = CHECKSUM_NONE;
3015 __skb_trim(skb, len);
3019 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3021 if (skb->ip_summed == CHECKSUM_COMPLETE)
3022 skb->ip_summed = CHECKSUM_NONE;
3023 return __skb_grow(skb, len);
3026 #define skb_queue_walk(queue, skb) \
3027 for (skb = (queue)->next; \
3028 skb != (struct sk_buff *)(queue); \
3031 #define skb_queue_walk_safe(queue, skb, tmp) \
3032 for (skb = (queue)->next, tmp = skb->next; \
3033 skb != (struct sk_buff *)(queue); \
3034 skb = tmp, tmp = skb->next)
3036 #define skb_queue_walk_from(queue, skb) \
3037 for (; skb != (struct sk_buff *)(queue); \
3040 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3041 for (tmp = skb->next; \
3042 skb != (struct sk_buff *)(queue); \
3043 skb = tmp, tmp = skb->next)
3045 #define skb_queue_reverse_walk(queue, skb) \
3046 for (skb = (queue)->prev; \
3047 skb != (struct sk_buff *)(queue); \
3050 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3051 for (skb = (queue)->prev, tmp = skb->prev; \
3052 skb != (struct sk_buff *)(queue); \
3053 skb = tmp, tmp = skb->prev)
3055 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3056 for (tmp = skb->prev; \
3057 skb != (struct sk_buff *)(queue); \
3058 skb = tmp, tmp = skb->prev)
3060 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3062 return skb_shinfo(skb)->frag_list != NULL;
3065 static inline void skb_frag_list_init(struct sk_buff *skb)
3067 skb_shinfo(skb)->frag_list = NULL;
3070 #define skb_walk_frags(skb, iter) \
3071 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3074 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3075 const struct sk_buff *skb);
3076 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3077 struct sk_buff_head *queue,
3079 void (*destructor)(struct sock *sk,
3080 struct sk_buff *skb),
3081 int *peeked, int *off, int *err,
3082 struct sk_buff **last);
3083 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3084 void (*destructor)(struct sock *sk,
3085 struct sk_buff *skb),
3086 int *peeked, int *off, int *err,
3087 struct sk_buff **last);
3088 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3089 void (*destructor)(struct sock *sk,
3090 struct sk_buff *skb),
3091 int *peeked, int *off, int *err);
3092 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3094 unsigned int datagram_poll(struct file *file, struct socket *sock,
3095 struct poll_table_struct *wait);
3096 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3097 struct iov_iter *to, int size);
3098 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3099 struct msghdr *msg, int size)
3101 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3103 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3104 struct msghdr *msg);
3105 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3106 struct iov_iter *from, int len);
3107 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3108 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3109 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3110 static inline void skb_free_datagram_locked(struct sock *sk,
3111 struct sk_buff *skb)
3113 __skb_free_datagram_locked(sk, skb, 0);
3115 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3116 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3117 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3118 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3119 int len, __wsum csum);
3120 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3121 struct pipe_inode_info *pipe, unsigned int len,
3122 unsigned int flags);
3123 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3124 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3125 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3127 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3128 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3129 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3130 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3131 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3132 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3133 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3134 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3135 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3136 int skb_vlan_pop(struct sk_buff *skb);
3137 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3138 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3141 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3143 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3146 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3148 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3151 struct skb_checksum_ops {
3152 __wsum (*update)(const void *mem, int len, __wsum wsum);
3153 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3156 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3158 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3159 __wsum csum, const struct skb_checksum_ops *ops);
3160 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3163 static inline void * __must_check
3164 __skb_header_pointer(const struct sk_buff *skb, int offset,
3165 int len, void *data, int hlen, void *buffer)
3167 if (hlen - offset >= len)
3168 return data + offset;
3171 skb_copy_bits(skb, offset, buffer, len) < 0)
3177 static inline void * __must_check
3178 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3180 return __skb_header_pointer(skb, offset, len, skb->data,
3181 skb_headlen(skb), buffer);
3185 * skb_needs_linearize - check if we need to linearize a given skb
3186 * depending on the given device features.
3187 * @skb: socket buffer to check
3188 * @features: net device features
3190 * Returns true if either:
3191 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3192 * 2. skb is fragmented and the device does not support SG.
3194 static inline bool skb_needs_linearize(struct sk_buff *skb,
3195 netdev_features_t features)
3197 return skb_is_nonlinear(skb) &&
3198 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3199 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3202 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3204 const unsigned int len)
3206 memcpy(to, skb->data, len);
3209 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3210 const int offset, void *to,
3211 const unsigned int len)
3213 memcpy(to, skb->data + offset, len);
3216 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3218 const unsigned int len)
3220 memcpy(skb->data, from, len);
3223 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3226 const unsigned int len)
3228 memcpy(skb->data + offset, from, len);
3231 void skb_init(void);
3233 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3239 * skb_get_timestamp - get timestamp from a skb
3240 * @skb: skb to get stamp from
3241 * @stamp: pointer to struct timeval to store stamp in
3243 * Timestamps are stored in the skb as offsets to a base timestamp.
3244 * This function converts the offset back to a struct timeval and stores
3247 static inline void skb_get_timestamp(const struct sk_buff *skb,
3248 struct timeval *stamp)
3250 *stamp = ktime_to_timeval(skb->tstamp);
3253 static inline void skb_get_timestampns(const struct sk_buff *skb,
3254 struct timespec *stamp)
3256 *stamp = ktime_to_timespec(skb->tstamp);
3259 static inline void __net_timestamp(struct sk_buff *skb)
3261 skb->tstamp = ktime_get_real();
3264 static inline ktime_t net_timedelta(ktime_t t)
3266 return ktime_sub(ktime_get_real(), t);
3269 static inline ktime_t net_invalid_timestamp(void)
3274 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3276 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3278 void skb_clone_tx_timestamp(struct sk_buff *skb);
3279 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3281 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3283 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3287 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3292 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3295 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3297 * PHY drivers may accept clones of transmitted packets for
3298 * timestamping via their phy_driver.txtstamp method. These drivers
3299 * must call this function to return the skb back to the stack with a
3302 * @skb: clone of the the original outgoing packet
3303 * @hwtstamps: hardware time stamps
3306 void skb_complete_tx_timestamp(struct sk_buff *skb,
3307 struct skb_shared_hwtstamps *hwtstamps);
3309 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3310 struct skb_shared_hwtstamps *hwtstamps,
3311 struct sock *sk, int tstype);
3314 * skb_tstamp_tx - queue clone of skb with send time stamps
3315 * @orig_skb: the original outgoing packet
3316 * @hwtstamps: hardware time stamps, may be NULL if not available
3318 * If the skb has a socket associated, then this function clones the
3319 * skb (thus sharing the actual data and optional structures), stores
3320 * the optional hardware time stamping information (if non NULL) or
3321 * generates a software time stamp (otherwise), then queues the clone
3322 * to the error queue of the socket. Errors are silently ignored.
3324 void skb_tstamp_tx(struct sk_buff *orig_skb,
3325 struct skb_shared_hwtstamps *hwtstamps);
3328 * skb_tx_timestamp() - Driver hook for transmit timestamping
3330 * Ethernet MAC Drivers should call this function in their hard_xmit()
3331 * function immediately before giving the sk_buff to the MAC hardware.
3333 * Specifically, one should make absolutely sure that this function is
3334 * called before TX completion of this packet can trigger. Otherwise
3335 * the packet could potentially already be freed.
3337 * @skb: A socket buffer.
3339 static inline void skb_tx_timestamp(struct sk_buff *skb)
3341 skb_clone_tx_timestamp(skb);
3342 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3343 skb_tstamp_tx(skb, NULL);
3347 * skb_complete_wifi_ack - deliver skb with wifi status
3349 * @skb: the original outgoing packet
3350 * @acked: ack status
3353 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3355 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3356 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3358 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3360 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3362 (skb->ip_summed == CHECKSUM_PARTIAL &&
3363 skb_checksum_start_offset(skb) >= 0));
3367 * skb_checksum_complete - Calculate checksum of an entire packet
3368 * @skb: packet to process
3370 * This function calculates the checksum over the entire packet plus
3371 * the value of skb->csum. The latter can be used to supply the
3372 * checksum of a pseudo header as used by TCP/UDP. It returns the
3375 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3376 * this function can be used to verify that checksum on received
3377 * packets. In that case the function should return zero if the
3378 * checksum is correct. In particular, this function will return zero
3379 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3380 * hardware has already verified the correctness of the checksum.
3382 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3384 return skb_csum_unnecessary(skb) ?
3385 0 : __skb_checksum_complete(skb);
3388 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3390 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3391 if (skb->csum_level == 0)
3392 skb->ip_summed = CHECKSUM_NONE;
3398 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3400 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3401 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3403 } else if (skb->ip_summed == CHECKSUM_NONE) {
3404 skb->ip_summed = CHECKSUM_UNNECESSARY;
3405 skb->csum_level = 0;
3409 /* Check if we need to perform checksum complete validation.
3411 * Returns true if checksum complete is needed, false otherwise
3412 * (either checksum is unnecessary or zero checksum is allowed).
3414 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3418 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3419 skb->csum_valid = 1;
3420 __skb_decr_checksum_unnecessary(skb);
3427 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3430 #define CHECKSUM_BREAK 76
3432 /* Unset checksum-complete
3434 * Unset checksum complete can be done when packet is being modified
3435 * (uncompressed for instance) and checksum-complete value is
3438 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3440 if (skb->ip_summed == CHECKSUM_COMPLETE)
3441 skb->ip_summed = CHECKSUM_NONE;
3444 /* Validate (init) checksum based on checksum complete.
3447 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3448 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3449 * checksum is stored in skb->csum for use in __skb_checksum_complete
3450 * non-zero: value of invalid checksum
3453 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3457 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3458 if (!csum_fold(csum_add(psum, skb->csum))) {
3459 skb->csum_valid = 1;
3466 if (complete || skb->len <= CHECKSUM_BREAK) {
3469 csum = __skb_checksum_complete(skb);
3470 skb->csum_valid = !csum;
3477 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3482 /* Perform checksum validate (init). Note that this is a macro since we only
3483 * want to calculate the pseudo header which is an input function if necessary.
3484 * First we try to validate without any computation (checksum unnecessary) and
3485 * then calculate based on checksum complete calling the function to compute
3489 * 0: checksum is validated or try to in skb_checksum_complete
3490 * non-zero: value of invalid checksum
3492 #define __skb_checksum_validate(skb, proto, complete, \
3493 zero_okay, check, compute_pseudo) \
3495 __sum16 __ret = 0; \
3496 skb->csum_valid = 0; \
3497 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3498 __ret = __skb_checksum_validate_complete(skb, \
3499 complete, compute_pseudo(skb, proto)); \
3503 #define skb_checksum_init(skb, proto, compute_pseudo) \
3504 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3506 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3507 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3509 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3510 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3512 #define skb_checksum_validate_zero_check(skb, proto, check, \
3514 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3516 #define skb_checksum_simple_validate(skb) \
3517 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3519 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3521 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3524 static inline void __skb_checksum_convert(struct sk_buff *skb,
3525 __sum16 check, __wsum pseudo)
3527 skb->csum = ~pseudo;
3528 skb->ip_summed = CHECKSUM_COMPLETE;
3531 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3533 if (__skb_checksum_convert_check(skb)) \
3534 __skb_checksum_convert(skb, check, \
3535 compute_pseudo(skb, proto)); \
3538 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3539 u16 start, u16 offset)
3541 skb->ip_summed = CHECKSUM_PARTIAL;
3542 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3543 skb->csum_offset = offset - start;
3546 /* Update skbuf and packet to reflect the remote checksum offload operation.
3547 * When called, ptr indicates the starting point for skb->csum when
3548 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3549 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3551 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3552 int start, int offset, bool nopartial)
3557 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3561 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3562 __skb_checksum_complete(skb);
3563 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3566 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3568 /* Adjust skb->csum since we changed the packet */
3569 skb->csum = csum_add(skb->csum, delta);
3572 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3574 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3575 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3581 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3582 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3583 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3585 if (nfct && atomic_dec_and_test(&nfct->use))
3586 nf_conntrack_destroy(nfct);
3588 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3591 atomic_inc(&nfct->use);
3594 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3595 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3597 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use))
3600 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3603 refcount_inc(&nf_bridge->use);
3605 #endif /* CONFIG_BRIDGE_NETFILTER */
3606 static inline void nf_reset(struct sk_buff *skb)
3608 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3609 nf_conntrack_put(skb_nfct(skb));
3612 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3613 nf_bridge_put(skb->nf_bridge);
3614 skb->nf_bridge = NULL;
3618 static inline void nf_reset_trace(struct sk_buff *skb)
3620 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3625 /* Note: This doesn't put any conntrack and bridge info in dst. */
3626 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3629 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3630 dst->_nfct = src->_nfct;
3631 nf_conntrack_get(skb_nfct(src));
3633 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3634 dst->nf_bridge = src->nf_bridge;
3635 nf_bridge_get(src->nf_bridge);
3637 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3639 dst->nf_trace = src->nf_trace;
3643 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3645 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3646 nf_conntrack_put(skb_nfct(dst));
3648 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3649 nf_bridge_put(dst->nf_bridge);
3651 __nf_copy(dst, src, true);
3654 #ifdef CONFIG_NETWORK_SECMARK
3655 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3657 to->secmark = from->secmark;
3660 static inline void skb_init_secmark(struct sk_buff *skb)
3665 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3668 static inline void skb_init_secmark(struct sk_buff *skb)
3672 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3674 return !skb->destructor &&
3675 #if IS_ENABLED(CONFIG_XFRM)
3679 !skb->_skb_refdst &&
3680 !skb_has_frag_list(skb);
3683 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3685 skb->queue_mapping = queue_mapping;
3688 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3690 return skb->queue_mapping;
3693 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3695 to->queue_mapping = from->queue_mapping;
3698 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3700 skb->queue_mapping = rx_queue + 1;
3703 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3705 return skb->queue_mapping - 1;
3708 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3710 return skb->queue_mapping != 0;
3713 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3715 skb->dst_pending_confirm = val;
3718 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3720 return skb->dst_pending_confirm != 0;
3723 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3732 /* Keeps track of mac header offset relative to skb->head.
3733 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3734 * For non-tunnel skb it points to skb_mac_header() and for
3735 * tunnel skb it points to outer mac header.
3736 * Keeps track of level of encapsulation of network headers.
3747 #define SKB_SGO_CB_OFFSET 32
3748 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3750 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3752 return (skb_mac_header(inner_skb) - inner_skb->head) -
3753 SKB_GSO_CB(inner_skb)->mac_offset;
3756 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3758 int new_headroom, headroom;
3761 headroom = skb_headroom(skb);
3762 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3766 new_headroom = skb_headroom(skb);
3767 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3771 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3773 /* Do not update partial checksums if remote checksum is enabled. */
3774 if (skb->remcsum_offload)
3777 SKB_GSO_CB(skb)->csum = res;
3778 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3781 /* Compute the checksum for a gso segment. First compute the checksum value
3782 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3783 * then add in skb->csum (checksum from csum_start to end of packet).
3784 * skb->csum and csum_start are then updated to reflect the checksum of the
3785 * resultant packet starting from the transport header-- the resultant checksum
3786 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3789 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3791 unsigned char *csum_start = skb_transport_header(skb);
3792 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3793 __wsum partial = SKB_GSO_CB(skb)->csum;
3795 SKB_GSO_CB(skb)->csum = res;
3796 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
3798 return csum_fold(csum_partial(csum_start, plen, partial));
3801 static inline bool skb_is_gso(const struct sk_buff *skb)
3803 return skb_shinfo(skb)->gso_size;
3806 /* Note: Should be called only if skb_is_gso(skb) is true */
3807 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3809 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3812 static inline void skb_gso_reset(struct sk_buff *skb)
3814 skb_shinfo(skb)->gso_size = 0;
3815 skb_shinfo(skb)->gso_segs = 0;
3816 skb_shinfo(skb)->gso_type = 0;
3819 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3821 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3823 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3824 * wanted then gso_type will be set. */
3825 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3827 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3828 unlikely(shinfo->gso_type == 0)) {
3829 __skb_warn_lro_forwarding(skb);
3835 static inline void skb_forward_csum(struct sk_buff *skb)
3837 /* Unfortunately we don't support this one. Any brave souls? */
3838 if (skb->ip_summed == CHECKSUM_COMPLETE)
3839 skb->ip_summed = CHECKSUM_NONE;
3843 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3844 * @skb: skb to check
3846 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3847 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3848 * use this helper, to document places where we make this assertion.
3850 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3853 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3857 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3859 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3860 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3861 unsigned int transport_len,
3862 __sum16(*skb_chkf)(struct sk_buff *skb));
3865 * skb_head_is_locked - Determine if the skb->head is locked down
3866 * @skb: skb to check
3868 * The head on skbs build around a head frag can be removed if they are
3869 * not cloned. This function returns true if the skb head is locked down
3870 * due to either being allocated via kmalloc, or by being a clone with
3871 * multiple references to the head.
3873 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3875 return !skb->head_frag || skb_cloned(skb);
3879 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3883 * skb_gso_network_seglen is used to determine the real size of the
3884 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3886 * The MAC/L2 header is not accounted for.
3888 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3890 unsigned int hdr_len = skb_transport_header(skb) -
3891 skb_network_header(skb);
3892 return hdr_len + skb_gso_transport_seglen(skb);
3895 /* Local Checksum Offload.
3896 * Compute outer checksum based on the assumption that the
3897 * inner checksum will be offloaded later.
3898 * See Documentation/networking/checksum-offloads.txt for
3899 * explanation of how this works.
3900 * Fill in outer checksum adjustment (e.g. with sum of outer
3901 * pseudo-header) before calling.
3902 * Also ensure that inner checksum is in linear data area.
3904 static inline __wsum lco_csum(struct sk_buff *skb)
3906 unsigned char *csum_start = skb_checksum_start(skb);
3907 unsigned char *l4_hdr = skb_transport_header(skb);
3910 /* Start with complement of inner checksum adjustment */
3911 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3914 /* Add in checksum of our headers (incl. outer checksum
3915 * adjustment filled in by caller) and return result.
3917 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
3920 #endif /* __KERNEL__ */
3921 #endif /* _LINUX_SKBUFF_H */