2 * Definitions for the 'struct sk_buff' memory handlers.
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
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 <net/flow_dissector.h>
38 #include <linux/splice.h>
39 #include <linux/in6.h>
42 /* The interface for checksum offload between the stack and networking drivers
45 * A. IP checksum related features
47 * Drivers advertise checksum offload capabilities in the features of a device.
48 * From the stack's point of view these are capabilities offered by the driver,
49 * a driver typically only advertises features that it is capable of offloading
52 * The checksum related features are:
54 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
55 * IP (one's complement) checksum for any combination
56 * of protocols or protocol layering. The checksum is
57 * computed and set in a packet per the CHECKSUM_PARTIAL
58 * interface (see below).
60 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
61 * TCP or UDP packets over IPv4. These are specifically
62 * unencapsulated packets of the form IPv4|TCP or
63 * IPv4|UDP where the Protocol field in the IPv4 header
64 * is TCP or UDP. The IPv4 header may contain IP options
65 * This feature cannot be set in features for a device
66 * with NETIF_F_HW_CSUM also set. This feature is being
67 * DEPRECATED (see below).
69 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
70 * TCP or UDP packets over IPv6. These are specifically
71 * unencapsulated packets of the form IPv6|TCP or
72 * IPv4|UDP where the Next Header field in the IPv6
73 * header is either TCP or UDP. IPv6 extension headers
74 * are not supported with this feature. This feature
75 * cannot be set in features for a device with
76 * NETIF_F_HW_CSUM also set. This feature is being
77 * DEPRECATED (see below).
79 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
80 * This flag is used only used to disable the RX checksum
81 * feature for a device. The stack will accept receive
82 * checksum indication in packets received on a device
83 * regardless of whether NETIF_F_RXCSUM is set.
85 * B. Checksumming of received packets by device. Indication of checksum
86 * verification is in set skb->ip_summed. Possible values are:
90 * Device did not checksum this packet e.g. due to lack of capabilities.
91 * The packet contains full (though not verified) checksum in packet but
92 * not in skb->csum. Thus, skb->csum is undefined in this case.
94 * CHECKSUM_UNNECESSARY:
96 * The hardware you're dealing with doesn't calculate the full checksum
97 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
98 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
99 * if their checksums are okay. skb->csum is still undefined in this case
100 * though. A driver or device must never modify the checksum field in the
101 * packet even if checksum is verified.
103 * CHECKSUM_UNNECESSARY is applicable to following protocols:
104 * TCP: IPv6 and IPv4.
105 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
106 * zero UDP checksum for either IPv4 or IPv6, the networking stack
107 * may perform further validation in this case.
108 * GRE: only if the checksum is present in the header.
109 * SCTP: indicates the CRC in SCTP header has been validated.
111 * skb->csum_level indicates the number of consecutive checksums found in
112 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
113 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
114 * and a device is able to verify the checksums for UDP (possibly zero),
115 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
116 * two. If the device were only able to verify the UDP checksum and not
117 * GRE, either because it doesn't support GRE checksum of because GRE
118 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
119 * not considered in this case).
123 * This is the most generic way. The device supplied checksum of the _whole_
124 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
125 * hardware doesn't need to parse L3/L4 headers to implement this.
127 * Note: Even if device supports only some protocols, but is able to produce
128 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
132 * A checksum is set up to be offloaded to a device as described in the
133 * output description for CHECKSUM_PARTIAL. This may occur on a packet
134 * received directly from another Linux OS, e.g., a virtualized Linux kernel
135 * on the same host, or it may be set in the input path in GRO or remote
136 * checksum offload. For the purposes of checksum verification, the checksum
137 * referred to by skb->csum_start + skb->csum_offset and any preceding
138 * checksums in the packet are considered verified. Any checksums in the
139 * packet that are after the checksum being offloaded are not considered to
142 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
143 * in the skb->ip_summed for a packet. Values are:
147 * The driver is required to checksum the packet as seen by hard_start_xmit()
148 * from skb->csum_start up to the end, and to record/write the checksum at
149 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
150 * csum_start and csum_offset values are valid values given the length and
151 * offset of the packet, however they should not attempt to validate that the
152 * checksum refers to a legitimate transport layer checksum-- it is the
153 * purview of the stack to validate that csum_start and csum_offset are set
156 * When the stack requests checksum offload for a packet, the driver MUST
157 * ensure that the checksum is set correctly. A driver can either offload the
158 * checksum calculation to the device, or call skb_checksum_help (in the case
159 * that the device does not support offload for a particular checksum).
161 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
162 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
163 * checksum offload capability. If a device has limited checksum capabilities
164 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as
165 * described above) a helper function can be called to resolve
166 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper
167 * function takes a spec argument that describes the protocol layer that is
168 * supported for checksum offload and can be called for each packet. If a
169 * packet does not match the specification for offload, skb_checksum_help
170 * is called to resolve the checksum.
174 * The skb was already checksummed by the protocol, or a checksum is not
177 * CHECKSUM_UNNECESSARY:
179 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
183 * Not used in checksum output. If a driver observes a packet with this value
184 * set in skbuff, if should treat as CHECKSUM_NONE being set.
186 * D. Non-IP checksum (CRC) offloads
188 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
189 * offloading the SCTP CRC in a packet. To perform this offload the stack
190 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
191 * accordingly. Note the there is no indication in the skbuff that the
192 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports
193 * both IP checksum offload and SCTP CRC offload must verify which offload
194 * is configured for a packet presumably by inspecting packet headers.
196 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
197 * offloading the FCOE CRC in a packet. To perform this offload the stack
198 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
199 * accordingly. Note the there is no indication in the skbuff that the
200 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
201 * both IP checksum offload and FCOE CRC offload must verify which offload
202 * is configured for a packet presumably by inspecting packet headers.
204 * E. Checksumming on output with GSO.
206 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
207 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
208 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
209 * part of the GSO operation is implied. If a checksum is being offloaded
210 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
211 * are set to refer to the outermost checksum being offload (two offloaded
212 * checksums are possible with UDP encapsulation).
215 /* Don't change this without changing skb_csum_unnecessary! */
216 #define CHECKSUM_NONE 0
217 #define CHECKSUM_UNNECESSARY 1
218 #define CHECKSUM_COMPLETE 2
219 #define CHECKSUM_PARTIAL 3
221 /* Maximum value in skb->csum_level */
222 #define SKB_MAX_CSUM_LEVEL 3
224 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
225 #define SKB_WITH_OVERHEAD(X) \
226 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
227 #define SKB_MAX_ORDER(X, ORDER) \
228 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
229 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
230 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
232 /* return minimum truesize of one skb containing X bytes of data */
233 #define SKB_TRUESIZE(X) ((X) + \
234 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
235 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
239 struct pipe_inode_info;
243 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
244 struct nf_conntrack {
249 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
250 struct nf_bridge_info {
253 BRNF_PROTO_UNCHANGED,
261 struct net_device *physindev;
263 /* always valid & non-NULL from FORWARD on, for physdev match */
264 struct net_device *physoutdev;
266 /* prerouting: detect dnat in orig/reply direction */
268 struct in6_addr ipv6_daddr;
270 /* after prerouting + nat detected: store original source
271 * mac since neigh resolution overwrites it, only used while
272 * skb is out in neigh layer.
274 char neigh_header[8];
279 struct sk_buff_head {
280 /* These two members must be first. */
281 struct sk_buff *next;
282 struct sk_buff *prev;
290 /* To allow 64K frame to be packed as single skb without frag_list we
291 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
292 * buffers which do not start on a page boundary.
294 * Since GRO uses frags we allocate at least 16 regardless of page
297 #if (65536/PAGE_SIZE + 1) < 16
298 #define MAX_SKB_FRAGS 16UL
300 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
302 extern int sysctl_max_skb_frags;
304 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
305 * segment using its current segmentation instead.
307 #define GSO_BY_FRAGS 0xFFFF
309 typedef struct skb_frag_struct skb_frag_t;
311 struct skb_frag_struct {
315 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
324 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
329 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
334 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
339 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
344 #define HAVE_HW_TIME_STAMP
347 * struct skb_shared_hwtstamps - hardware time stamps
348 * @hwtstamp: hardware time stamp transformed into duration
349 * since arbitrary point in time
351 * Software time stamps generated by ktime_get_real() are stored in
354 * hwtstamps can only be compared against other hwtstamps from
357 * This structure is attached to packets as part of the
358 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
360 struct skb_shared_hwtstamps {
364 /* Definitions for tx_flags in struct skb_shared_info */
366 /* generate hardware time stamp */
367 SKBTX_HW_TSTAMP = 1 << 0,
369 /* generate software time stamp when queueing packet to NIC */
370 SKBTX_SW_TSTAMP = 1 << 1,
372 /* device driver is going to provide hardware time stamp */
373 SKBTX_IN_PROGRESS = 1 << 2,
375 /* device driver supports TX zero-copy buffers */
376 SKBTX_DEV_ZEROCOPY = 1 << 3,
378 /* generate wifi status information (where possible) */
379 SKBTX_WIFI_STATUS = 1 << 4,
381 /* This indicates at least one fragment might be overwritten
382 * (as in vmsplice(), sendfile() ...)
383 * If we need to compute a TX checksum, we'll need to copy
384 * all frags to avoid possible bad checksum
386 SKBTX_SHARED_FRAG = 1 << 5,
388 /* generate software time stamp when entering packet scheduling */
389 SKBTX_SCHED_TSTAMP = 1 << 6,
392 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
394 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
397 * The callback notifies userspace to release buffers when skb DMA is done in
398 * lower device, the skb last reference should be 0 when calling this.
399 * The zerocopy_success argument is true if zero copy transmit occurred,
400 * false on data copy or out of memory error caused by data copy attempt.
401 * The ctx field is used to track device context.
402 * The desc field is used to track userspace buffer index.
405 void (*callback)(struct ubuf_info *, bool zerocopy_success);
410 /* This data is invariant across clones and lives at
411 * the end of the header data, ie. at skb->end.
413 struct skb_shared_info {
414 unsigned char nr_frags;
416 unsigned short gso_size;
417 /* Warning: this field is not always filled in (UFO)! */
418 unsigned short gso_segs;
419 unsigned short gso_type;
420 struct sk_buff *frag_list;
421 struct skb_shared_hwtstamps hwtstamps;
426 * Warning : all fields before dataref are cleared in __alloc_skb()
430 /* Intermediate layers must ensure that destructor_arg
431 * remains valid until skb destructor */
432 void * destructor_arg;
434 /* must be last field, see pskb_expand_head() */
435 skb_frag_t frags[MAX_SKB_FRAGS];
438 /* We divide dataref into two halves. The higher 16 bits hold references
439 * to the payload part of skb->data. The lower 16 bits hold references to
440 * the entire skb->data. A clone of a headerless skb holds the length of
441 * the header in skb->hdr_len.
443 * All users must obey the rule that the skb->data reference count must be
444 * greater than or equal to the payload reference count.
446 * Holding a reference to the payload part means that the user does not
447 * care about modifications to the header part of skb->data.
449 #define SKB_DATAREF_SHIFT 16
450 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
454 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
455 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
456 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
460 SKB_GSO_TCPV4 = 1 << 0,
461 SKB_GSO_UDP = 1 << 1,
463 /* This indicates the skb is from an untrusted source. */
464 SKB_GSO_DODGY = 1 << 2,
466 /* This indicates the tcp segment has CWR set. */
467 SKB_GSO_TCP_ECN = 1 << 3,
469 SKB_GSO_TCP_FIXEDID = 1 << 4,
471 SKB_GSO_TCPV6 = 1 << 5,
473 SKB_GSO_FCOE = 1 << 6,
475 SKB_GSO_GRE = 1 << 7,
477 SKB_GSO_GRE_CSUM = 1 << 8,
479 SKB_GSO_IPXIP4 = 1 << 9,
481 SKB_GSO_IPXIP6 = 1 << 10,
483 SKB_GSO_UDP_TUNNEL = 1 << 11,
485 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
487 SKB_GSO_PARTIAL = 1 << 13,
489 SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
492 #if BITS_PER_LONG > 32
493 #define NET_SKBUFF_DATA_USES_OFFSET 1
496 #ifdef NET_SKBUFF_DATA_USES_OFFSET
497 typedef unsigned int sk_buff_data_t;
499 typedef unsigned char *sk_buff_data_t;
503 * struct skb_mstamp - multi resolution time stamps
504 * @stamp_us: timestamp in us resolution
505 * @stamp_jiffies: timestamp in jiffies
518 * skb_mstamp_get - get current timestamp
519 * @cl: place to store timestamps
521 static inline void skb_mstamp_get(struct skb_mstamp *cl)
523 u64 val = local_clock();
525 do_div(val, NSEC_PER_USEC);
526 cl->stamp_us = (u32)val;
527 cl->stamp_jiffies = (u32)jiffies;
531 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
532 * @t1: pointer to newest sample
533 * @t0: pointer to oldest sample
535 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
536 const struct skb_mstamp *t0)
538 s32 delta_us = t1->stamp_us - t0->stamp_us;
539 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
541 /* If delta_us is negative, this might be because interval is too big,
542 * or local_clock() drift is too big : fallback using jiffies.
545 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
547 delta_us = jiffies_to_usecs(delta_jiffies);
552 static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
553 const struct skb_mstamp *t0)
555 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
558 diff = t1->stamp_us - t0->stamp_us;
563 * struct sk_buff - socket buffer
564 * @next: Next buffer in list
565 * @prev: Previous buffer in list
566 * @tstamp: Time we arrived/left
567 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
568 * @sk: Socket we are owned by
569 * @dev: Device we arrived on/are leaving by
570 * @cb: Control buffer. Free for use by every layer. Put private vars here
571 * @_skb_refdst: destination entry (with norefcount bit)
572 * @sp: the security path, used for xfrm
573 * @len: Length of actual data
574 * @data_len: Data length
575 * @mac_len: Length of link layer header
576 * @hdr_len: writable header length of cloned skb
577 * @csum: Checksum (must include start/offset pair)
578 * @csum_start: Offset from skb->head where checksumming should start
579 * @csum_offset: Offset from csum_start where checksum should be stored
580 * @priority: Packet queueing priority
581 * @ignore_df: allow local fragmentation
582 * @cloned: Head may be cloned (check refcnt to be sure)
583 * @ip_summed: Driver fed us an IP checksum
584 * @nohdr: Payload reference only, must not modify header
585 * @nfctinfo: Relationship of this skb to the connection
586 * @pkt_type: Packet class
587 * @fclone: skbuff clone status
588 * @ipvs_property: skbuff is owned by ipvs
589 * @peeked: this packet has been seen already, so stats have been
590 * done for it, don't do them again
591 * @nf_trace: netfilter packet trace flag
592 * @protocol: Packet protocol from driver
593 * @destructor: Destruct function
594 * @nfct: Associated connection, if any
595 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
596 * @skb_iif: ifindex of device we arrived on
597 * @tc_index: Traffic control index
598 * @tc_verd: traffic control verdict
599 * @hash: the packet hash
600 * @queue_mapping: Queue mapping for multiqueue devices
601 * @xmit_more: More SKBs are pending for this queue
602 * @ndisc_nodetype: router type (from link layer)
603 * @ooo_okay: allow the mapping of a socket to a queue to be changed
604 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
606 * @sw_hash: indicates hash was computed in software stack
607 * @wifi_acked_valid: wifi_acked was set
608 * @wifi_acked: whether frame was acked on wifi or not
609 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
610 * @napi_id: id of the NAPI struct this skb came from
611 * @secmark: security marking
612 * @offload_fwd_mark: fwding offload mark
613 * @mark: Generic packet mark
614 * @vlan_proto: vlan encapsulation protocol
615 * @vlan_tci: vlan tag control information
616 * @inner_protocol: Protocol (encapsulation)
617 * @inner_transport_header: Inner transport layer header (encapsulation)
618 * @inner_network_header: Network layer header (encapsulation)
619 * @inner_mac_header: Link layer header (encapsulation)
620 * @transport_header: Transport layer header
621 * @network_header: Network layer header
622 * @mac_header: Link layer header
623 * @tail: Tail pointer
625 * @head: Head of buffer
626 * @data: Data head pointer
627 * @truesize: Buffer size
628 * @users: User count - see {datagram,tcp}.c
634 /* These two members must be first. */
635 struct sk_buff *next;
636 struct sk_buff *prev;
640 struct skb_mstamp skb_mstamp;
643 struct rb_node rbnode; /* used in netem & tcp stack */
646 struct net_device *dev;
649 * This is the control buffer. It is free to use for every
650 * layer. Please put your private variables there. If you
651 * want to keep them across layers you have to do a skb_clone()
652 * first. This is owned by whoever has the skb queued ATM.
654 char cb[48] __aligned(8);
656 unsigned long _skb_refdst;
657 void (*destructor)(struct sk_buff *skb);
661 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
662 struct nf_conntrack *nfct;
664 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
665 struct nf_bridge_info *nf_bridge;
672 /* Following fields are _not_ copied in __copy_skb_header()
673 * Note that queue_mapping is here mostly to fill a hole.
675 kmemcheck_bitfield_begin(flags1);
684 kmemcheck_bitfield_end(flags1);
686 /* fields enclosed in headers_start/headers_end are copied
687 * using a single memcpy() in __copy_skb_header()
690 __u32 headers_start[0];
693 /* if you move pkt_type around you also must adapt those constants */
694 #ifdef __BIG_ENDIAN_BITFIELD
695 #define PKT_TYPE_MAX (7 << 5)
697 #define PKT_TYPE_MAX 7
699 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
701 __u8 __pkt_type_offset[0];
712 __u8 wifi_acked_valid:1;
716 /* Indicates the inner headers are valid in the skbuff. */
717 __u8 encapsulation:1;
718 __u8 encap_hdr_csum:1;
720 __u8 csum_complete_sw:1;
724 #ifdef CONFIG_IPV6_NDISC_NODETYPE
725 __u8 ndisc_nodetype:2;
727 __u8 ipvs_property:1;
728 __u8 inner_protocol_type:1;
729 __u8 remcsum_offload:1;
730 /* 3 or 5 bit hole */
732 #ifdef CONFIG_NET_SCHED
733 __u16 tc_index; /* traffic control index */
734 #ifdef CONFIG_NET_CLS_ACT
735 __u16 tc_verd; /* traffic control verdict */
751 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
753 unsigned int napi_id;
754 unsigned int sender_cpu;
758 #ifdef CONFIG_NETWORK_SECMARK
761 #ifdef CONFIG_NET_SWITCHDEV
762 __u32 offload_fwd_mark;
768 __u32 reserved_tailroom;
772 __be16 inner_protocol;
776 __u16 inner_transport_header;
777 __u16 inner_network_header;
778 __u16 inner_mac_header;
781 __u16 transport_header;
782 __u16 network_header;
786 __u32 headers_end[0];
789 /* These elements must be at the end, see alloc_skb() for details. */
794 unsigned int truesize;
800 * Handling routines are only of interest to the kernel
802 #include <linux/slab.h>
805 #define SKB_ALLOC_FCLONE 0x01
806 #define SKB_ALLOC_RX 0x02
807 #define SKB_ALLOC_NAPI 0x04
809 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
810 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
812 return unlikely(skb->pfmemalloc);
816 * skb might have a dst pointer attached, refcounted or not.
817 * _skb_refdst low order bit is set if refcount was _not_ taken
819 #define SKB_DST_NOREF 1UL
820 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
823 * skb_dst - returns skb dst_entry
826 * Returns skb dst_entry, regardless of reference taken or not.
828 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
830 /* If refdst was not refcounted, check we still are in a
831 * rcu_read_lock section
833 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
834 !rcu_read_lock_held() &&
835 !rcu_read_lock_bh_held());
836 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
840 * skb_dst_set - sets skb dst
844 * Sets skb dst, assuming a reference was taken on dst and should
845 * be released by skb_dst_drop()
847 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
849 skb->_skb_refdst = (unsigned long)dst;
853 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
857 * Sets skb dst, assuming a reference was not taken on dst.
858 * If dst entry is cached, we do not take reference and dst_release
859 * will be avoided by refdst_drop. If dst entry is not cached, we take
860 * reference, so that last dst_release can destroy the dst immediately.
862 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
864 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
865 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
869 * skb_dst_is_noref - Test if skb dst isn't refcounted
872 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
874 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
877 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
879 return (struct rtable *)skb_dst(skb);
882 void kfree_skb(struct sk_buff *skb);
883 void kfree_skb_list(struct sk_buff *segs);
884 void skb_tx_error(struct sk_buff *skb);
885 void consume_skb(struct sk_buff *skb);
886 void __kfree_skb(struct sk_buff *skb);
887 extern struct kmem_cache *skbuff_head_cache;
889 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
890 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
891 bool *fragstolen, int *delta_truesize);
893 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
895 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
896 struct sk_buff *build_skb(void *data, unsigned int frag_size);
897 static inline struct sk_buff *alloc_skb(unsigned int size,
900 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
903 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
904 unsigned long data_len,
909 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
910 struct sk_buff_fclones {
919 * skb_fclone_busy - check if fclone is busy
922 * Returns true if skb is a fast clone, and its clone is not freed.
923 * Some drivers call skb_orphan() in their ndo_start_xmit(),
924 * so we also check that this didnt happen.
926 static inline bool skb_fclone_busy(const struct sock *sk,
927 const struct sk_buff *skb)
929 const struct sk_buff_fclones *fclones;
931 fclones = container_of(skb, struct sk_buff_fclones, skb1);
933 return skb->fclone == SKB_FCLONE_ORIG &&
934 atomic_read(&fclones->fclone_ref) > 1 &&
935 fclones->skb2.sk == sk;
938 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
941 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
944 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
945 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
947 return __alloc_skb_head(priority, -1);
950 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
951 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
952 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
953 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
954 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
955 gfp_t gfp_mask, bool fclone);
956 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
959 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
962 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
963 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
964 unsigned int headroom);
965 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
966 int newtailroom, gfp_t priority);
967 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
968 int offset, int len);
969 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
971 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
972 int skb_pad(struct sk_buff *skb, int pad);
973 #define dev_kfree_skb(a) consume_skb(a)
975 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
976 int getfrag(void *from, char *to, int offset,
977 int len, int odd, struct sk_buff *skb),
978 void *from, int length);
980 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
981 int offset, size_t size);
983 struct skb_seq_state {
987 __u32 stepped_offset;
988 struct sk_buff *root_skb;
989 struct sk_buff *cur_skb;
993 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
994 unsigned int to, struct skb_seq_state *st);
995 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
996 struct skb_seq_state *st);
997 void skb_abort_seq_read(struct skb_seq_state *st);
999 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1000 unsigned int to, struct ts_config *config);
1003 * Packet hash types specify the type of hash in skb_set_hash.
1005 * Hash types refer to the protocol layer addresses which are used to
1006 * construct a packet's hash. The hashes are used to differentiate or identify
1007 * flows of the protocol layer for the hash type. Hash types are either
1008 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1010 * Properties of hashes:
1012 * 1) Two packets in different flows have different hash values
1013 * 2) Two packets in the same flow should have the same hash value
1015 * A hash at a higher layer is considered to be more specific. A driver should
1016 * set the most specific hash possible.
1018 * A driver cannot indicate a more specific hash than the layer at which a hash
1019 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1021 * A driver may indicate a hash level which is less specific than the
1022 * actual layer the hash was computed on. For instance, a hash computed
1023 * at L4 may be considered an L3 hash. This should only be done if the
1024 * driver can't unambiguously determine that the HW computed the hash at
1025 * the higher layer. Note that the "should" in the second property above
1028 enum pkt_hash_types {
1029 PKT_HASH_TYPE_NONE, /* Undefined type */
1030 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1031 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1032 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1035 static inline void skb_clear_hash(struct sk_buff *skb)
1042 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1045 skb_clear_hash(skb);
1049 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1051 skb->l4_hash = is_l4;
1052 skb->sw_hash = is_sw;
1057 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1059 /* Used by drivers to set hash from HW */
1060 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1064 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1066 __skb_set_hash(skb, hash, true, is_l4);
1069 void __skb_get_hash(struct sk_buff *skb);
1070 u32 skb_get_poff(const struct sk_buff *skb);
1071 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1072 const struct flow_keys *keys, int hlen);
1073 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1074 void *data, int hlen_proto);
1076 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1077 int thoff, u8 ip_proto)
1079 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1082 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1083 const struct flow_dissector_key *key,
1084 unsigned int key_count);
1086 bool __skb_flow_dissect(const struct sk_buff *skb,
1087 struct flow_dissector *flow_dissector,
1088 void *target_container,
1089 void *data, __be16 proto, int nhoff, int hlen,
1090 unsigned int flags);
1092 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1093 struct flow_dissector *flow_dissector,
1094 void *target_container, unsigned int flags)
1096 return __skb_flow_dissect(skb, flow_dissector, target_container,
1097 NULL, 0, 0, 0, flags);
1100 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1101 struct flow_keys *flow,
1104 memset(flow, 0, sizeof(*flow));
1105 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1106 NULL, 0, 0, 0, flags);
1109 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1110 void *data, __be16 proto,
1111 int nhoff, int hlen,
1114 memset(flow, 0, sizeof(*flow));
1115 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1116 data, proto, nhoff, hlen, flags);
1119 static inline __u32 skb_get_hash(struct sk_buff *skb)
1121 if (!skb->l4_hash && !skb->sw_hash)
1122 __skb_get_hash(skb);
1127 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1129 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1131 if (!skb->l4_hash && !skb->sw_hash) {
1132 struct flow_keys keys;
1133 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1135 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1141 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1143 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1145 if (!skb->l4_hash && !skb->sw_hash) {
1146 struct flow_keys keys;
1147 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
1149 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1155 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1157 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1162 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1164 to->hash = from->hash;
1165 to->sw_hash = from->sw_hash;
1166 to->l4_hash = from->l4_hash;
1169 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1170 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1172 return skb->head + skb->end;
1175 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1180 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1185 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1187 return skb->end - skb->head;
1192 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1194 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1196 return &skb_shinfo(skb)->hwtstamps;
1200 * skb_queue_empty - check if a queue is empty
1203 * Returns true if the queue is empty, false otherwise.
1205 static inline int skb_queue_empty(const struct sk_buff_head *list)
1207 return list->next == (const struct sk_buff *) list;
1211 * skb_queue_is_last - check if skb is the last entry in the queue
1215 * Returns true if @skb is the last buffer on the list.
1217 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1218 const struct sk_buff *skb)
1220 return skb->next == (const struct sk_buff *) list;
1224 * skb_queue_is_first - check if skb is the first entry in the queue
1228 * Returns true if @skb is the first buffer on the list.
1230 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1231 const struct sk_buff *skb)
1233 return skb->prev == (const struct sk_buff *) list;
1237 * skb_queue_next - return the next packet in the queue
1239 * @skb: current buffer
1241 * Return the next packet in @list after @skb. It is only valid to
1242 * call this if skb_queue_is_last() evaluates to false.
1244 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1245 const struct sk_buff *skb)
1247 /* This BUG_ON may seem severe, but if we just return then we
1248 * are going to dereference garbage.
1250 BUG_ON(skb_queue_is_last(list, skb));
1255 * skb_queue_prev - return the prev packet in the queue
1257 * @skb: current buffer
1259 * Return the prev packet in @list before @skb. It is only valid to
1260 * call this if skb_queue_is_first() evaluates to false.
1262 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1263 const struct sk_buff *skb)
1265 /* This BUG_ON may seem severe, but if we just return then we
1266 * are going to dereference garbage.
1268 BUG_ON(skb_queue_is_first(list, skb));
1273 * skb_get - reference buffer
1274 * @skb: buffer to reference
1276 * Makes another reference to a socket buffer and returns a pointer
1279 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1281 atomic_inc(&skb->users);
1286 * If users == 1, we are the only owner and are can avoid redundant
1291 * skb_cloned - is the buffer a clone
1292 * @skb: buffer to check
1294 * Returns true if the buffer was generated with skb_clone() and is
1295 * one of multiple shared copies of the buffer. Cloned buffers are
1296 * shared data so must not be written to under normal circumstances.
1298 static inline int skb_cloned(const struct sk_buff *skb)
1300 return skb->cloned &&
1301 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1304 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1306 might_sleep_if(gfpflags_allow_blocking(pri));
1308 if (skb_cloned(skb))
1309 return pskb_expand_head(skb, 0, 0, pri);
1315 * skb_header_cloned - is the header a clone
1316 * @skb: buffer to check
1318 * Returns true if modifying the header part of the buffer requires
1319 * the data to be copied.
1321 static inline int skb_header_cloned(const struct sk_buff *skb)
1328 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1329 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1330 return dataref != 1;
1333 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1335 might_sleep_if(gfpflags_allow_blocking(pri));
1337 if (skb_header_cloned(skb))
1338 return pskb_expand_head(skb, 0, 0, pri);
1344 * skb_header_release - release reference to header
1345 * @skb: buffer to operate on
1347 * Drop a reference to the header part of the buffer. This is done
1348 * by acquiring a payload reference. You must not read from the header
1349 * part of skb->data after this.
1350 * Note : Check if you can use __skb_header_release() instead.
1352 static inline void skb_header_release(struct sk_buff *skb)
1356 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1360 * __skb_header_release - release reference to header
1361 * @skb: buffer to operate on
1363 * Variant of skb_header_release() assuming skb is private to caller.
1364 * We can avoid one atomic operation.
1366 static inline void __skb_header_release(struct sk_buff *skb)
1369 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1374 * skb_shared - is the buffer shared
1375 * @skb: buffer to check
1377 * Returns true if more than one person has a reference to this
1380 static inline int skb_shared(const struct sk_buff *skb)
1382 return atomic_read(&skb->users) != 1;
1386 * skb_share_check - check if buffer is shared and if so clone it
1387 * @skb: buffer to check
1388 * @pri: priority for memory allocation
1390 * If the buffer is shared the buffer is cloned and the old copy
1391 * drops a reference. A new clone with a single reference is returned.
1392 * If the buffer is not shared the original buffer is returned. When
1393 * being called from interrupt status or with spinlocks held pri must
1396 * NULL is returned on a memory allocation failure.
1398 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1400 might_sleep_if(gfpflags_allow_blocking(pri));
1401 if (skb_shared(skb)) {
1402 struct sk_buff *nskb = skb_clone(skb, pri);
1414 * Copy shared buffers into a new sk_buff. We effectively do COW on
1415 * packets to handle cases where we have a local reader and forward
1416 * and a couple of other messy ones. The normal one is tcpdumping
1417 * a packet thats being forwarded.
1421 * skb_unshare - make a copy of a shared buffer
1422 * @skb: buffer to check
1423 * @pri: priority for memory allocation
1425 * If the socket buffer is a clone then this function creates a new
1426 * copy of the data, drops a reference count on the old copy and returns
1427 * the new copy with the reference count at 1. If the buffer is not a clone
1428 * the original buffer is returned. When called with a spinlock held or
1429 * from interrupt state @pri must be %GFP_ATOMIC
1431 * %NULL is returned on a memory allocation failure.
1433 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1436 might_sleep_if(gfpflags_allow_blocking(pri));
1437 if (skb_cloned(skb)) {
1438 struct sk_buff *nskb = skb_copy(skb, pri);
1440 /* Free our shared copy */
1451 * skb_peek - peek at the head of an &sk_buff_head
1452 * @list_: list to peek at
1454 * Peek an &sk_buff. Unlike most other operations you _MUST_
1455 * be careful with this one. A peek leaves the buffer on the
1456 * list and someone else may run off with it. You must hold
1457 * the appropriate locks or have a private queue to do this.
1459 * Returns %NULL for an empty list or a pointer to the head element.
1460 * The reference count is not incremented and the reference is therefore
1461 * volatile. Use with caution.
1463 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1465 struct sk_buff *skb = list_->next;
1467 if (skb == (struct sk_buff *)list_)
1473 * skb_peek_next - peek skb following the given one from a queue
1474 * @skb: skb to start from
1475 * @list_: list to peek at
1477 * Returns %NULL when the end of the list is met or a pointer to the
1478 * next element. The reference count is not incremented and the
1479 * reference is therefore volatile. Use with caution.
1481 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1482 const struct sk_buff_head *list_)
1484 struct sk_buff *next = skb->next;
1486 if (next == (struct sk_buff *)list_)
1492 * skb_peek_tail - peek at the tail of an &sk_buff_head
1493 * @list_: list to peek at
1495 * Peek an &sk_buff. Unlike most other operations you _MUST_
1496 * be careful with this one. A peek leaves the buffer on the
1497 * list and someone else may run off with it. You must hold
1498 * the appropriate locks or have a private queue to do this.
1500 * Returns %NULL for an empty list or a pointer to the tail element.
1501 * The reference count is not incremented and the reference is therefore
1502 * volatile. Use with caution.
1504 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1506 struct sk_buff *skb = list_->prev;
1508 if (skb == (struct sk_buff *)list_)
1515 * skb_queue_len - get queue length
1516 * @list_: list to measure
1518 * Return the length of an &sk_buff queue.
1520 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1526 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1527 * @list: queue to initialize
1529 * This initializes only the list and queue length aspects of
1530 * an sk_buff_head object. This allows to initialize the list
1531 * aspects of an sk_buff_head without reinitializing things like
1532 * the spinlock. It can also be used for on-stack sk_buff_head
1533 * objects where the spinlock is known to not be used.
1535 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1537 list->prev = list->next = (struct sk_buff *)list;
1542 * This function creates a split out lock class for each invocation;
1543 * this is needed for now since a whole lot of users of the skb-queue
1544 * infrastructure in drivers have different locking usage (in hardirq)
1545 * than the networking core (in softirq only). In the long run either the
1546 * network layer or drivers should need annotation to consolidate the
1547 * main types of usage into 3 classes.
1549 static inline void skb_queue_head_init(struct sk_buff_head *list)
1551 spin_lock_init(&list->lock);
1552 __skb_queue_head_init(list);
1555 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1556 struct lock_class_key *class)
1558 skb_queue_head_init(list);
1559 lockdep_set_class(&list->lock, class);
1563 * Insert an sk_buff on a list.
1565 * The "__skb_xxxx()" functions are the non-atomic ones that
1566 * can only be called with interrupts disabled.
1568 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1569 struct sk_buff_head *list);
1570 static inline void __skb_insert(struct sk_buff *newsk,
1571 struct sk_buff *prev, struct sk_buff *next,
1572 struct sk_buff_head *list)
1576 next->prev = prev->next = newsk;
1580 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1581 struct sk_buff *prev,
1582 struct sk_buff *next)
1584 struct sk_buff *first = list->next;
1585 struct sk_buff *last = list->prev;
1595 * skb_queue_splice - join two skb lists, this is designed for stacks
1596 * @list: the new list to add
1597 * @head: the place to add it in the first list
1599 static inline void skb_queue_splice(const struct sk_buff_head *list,
1600 struct sk_buff_head *head)
1602 if (!skb_queue_empty(list)) {
1603 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1604 head->qlen += list->qlen;
1609 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1610 * @list: the new list to add
1611 * @head: the place to add it in the first list
1613 * The list at @list is reinitialised
1615 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1616 struct sk_buff_head *head)
1618 if (!skb_queue_empty(list)) {
1619 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1620 head->qlen += list->qlen;
1621 __skb_queue_head_init(list);
1626 * skb_queue_splice_tail - join two skb lists, each list being a queue
1627 * @list: the new list to add
1628 * @head: the place to add it in the first list
1630 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1631 struct sk_buff_head *head)
1633 if (!skb_queue_empty(list)) {
1634 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1635 head->qlen += list->qlen;
1640 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1641 * @list: the new list to add
1642 * @head: the place to add it in the first list
1644 * Each of the lists is a queue.
1645 * The list at @list is reinitialised
1647 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1648 struct sk_buff_head *head)
1650 if (!skb_queue_empty(list)) {
1651 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1652 head->qlen += list->qlen;
1653 __skb_queue_head_init(list);
1658 * __skb_queue_after - queue a buffer at the list head
1659 * @list: list to use
1660 * @prev: place after this buffer
1661 * @newsk: buffer to queue
1663 * Queue a buffer int the middle of a list. This function takes no locks
1664 * and you must therefore hold required locks before calling it.
1666 * A buffer cannot be placed on two lists at the same time.
1668 static inline void __skb_queue_after(struct sk_buff_head *list,
1669 struct sk_buff *prev,
1670 struct sk_buff *newsk)
1672 __skb_insert(newsk, prev, prev->next, list);
1675 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1676 struct sk_buff_head *list);
1678 static inline void __skb_queue_before(struct sk_buff_head *list,
1679 struct sk_buff *next,
1680 struct sk_buff *newsk)
1682 __skb_insert(newsk, next->prev, next, list);
1686 * __skb_queue_head - queue a buffer at the list head
1687 * @list: list to use
1688 * @newsk: buffer to queue
1690 * Queue a buffer at the start of a list. This function takes no locks
1691 * and you must therefore hold required locks before calling it.
1693 * A buffer cannot be placed on two lists at the same time.
1695 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1696 static inline void __skb_queue_head(struct sk_buff_head *list,
1697 struct sk_buff *newsk)
1699 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1703 * __skb_queue_tail - queue a buffer at the list tail
1704 * @list: list to use
1705 * @newsk: buffer to queue
1707 * Queue a buffer at the end of a list. This function takes no locks
1708 * and you must therefore hold required locks before calling it.
1710 * A buffer cannot be placed on two lists at the same time.
1712 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1713 static inline void __skb_queue_tail(struct sk_buff_head *list,
1714 struct sk_buff *newsk)
1716 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1720 * remove sk_buff from list. _Must_ be called atomically, and with
1723 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1724 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1726 struct sk_buff *next, *prev;
1731 skb->next = skb->prev = NULL;
1737 * __skb_dequeue - remove from the head of the queue
1738 * @list: list to dequeue from
1740 * Remove the head of the list. This function does not take any locks
1741 * so must be used with appropriate locks held only. The head item is
1742 * returned or %NULL if the list is empty.
1744 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1745 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1747 struct sk_buff *skb = skb_peek(list);
1749 __skb_unlink(skb, list);
1754 * __skb_dequeue_tail - remove from the tail of the queue
1755 * @list: list to dequeue from
1757 * Remove the tail of the list. This function does not take any locks
1758 * so must be used with appropriate locks held only. The tail item is
1759 * returned or %NULL if the list is empty.
1761 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1762 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1764 struct sk_buff *skb = skb_peek_tail(list);
1766 __skb_unlink(skb, list);
1771 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1773 return skb->data_len;
1776 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1778 return skb->len - skb->data_len;
1781 static inline int skb_pagelen(const struct sk_buff *skb)
1785 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1786 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1787 return len + skb_headlen(skb);
1791 * __skb_fill_page_desc - initialise a paged fragment in an skb
1792 * @skb: buffer containing fragment to be initialised
1793 * @i: paged fragment index to initialise
1794 * @page: the page to use for this fragment
1795 * @off: the offset to the data with @page
1796 * @size: the length of the data
1798 * Initialises the @i'th fragment of @skb to point to &size bytes at
1799 * offset @off within @page.
1801 * Does not take any additional reference on the fragment.
1803 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1804 struct page *page, int off, int size)
1806 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1809 * Propagate page pfmemalloc to the skb if we can. The problem is
1810 * that not all callers have unique ownership of the page but rely
1811 * on page_is_pfmemalloc doing the right thing(tm).
1813 frag->page.p = page;
1814 frag->page_offset = off;
1815 skb_frag_size_set(frag, size);
1817 page = compound_head(page);
1818 if (page_is_pfmemalloc(page))
1819 skb->pfmemalloc = true;
1823 * skb_fill_page_desc - initialise a paged fragment in an skb
1824 * @skb: buffer containing fragment to be initialised
1825 * @i: paged fragment index to initialise
1826 * @page: the page to use for this fragment
1827 * @off: the offset to the data with @page
1828 * @size: the length of the data
1830 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1831 * @skb to point to @size bytes at offset @off within @page. In
1832 * addition updates @skb such that @i is the last fragment.
1834 * Does not take any additional reference on the fragment.
1836 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1837 struct page *page, int off, int size)
1839 __skb_fill_page_desc(skb, i, page, off, size);
1840 skb_shinfo(skb)->nr_frags = i + 1;
1843 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1844 int size, unsigned int truesize);
1846 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1847 unsigned int truesize);
1849 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1850 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1851 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1853 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1854 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1856 return skb->head + skb->tail;
1859 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1861 skb->tail = skb->data - skb->head;
1864 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1866 skb_reset_tail_pointer(skb);
1867 skb->tail += offset;
1870 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1871 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1876 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1878 skb->tail = skb->data;
1881 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1883 skb->tail = skb->data + offset;
1886 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1889 * Add data to an sk_buff
1891 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1892 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1893 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1895 unsigned char *tmp = skb_tail_pointer(skb);
1896 SKB_LINEAR_ASSERT(skb);
1902 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1903 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1910 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1911 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1914 BUG_ON(skb->len < skb->data_len);
1915 return skb->data += len;
1918 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1920 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1923 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1925 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1927 if (len > skb_headlen(skb) &&
1928 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1931 return skb->data += len;
1934 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1936 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1939 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1941 if (likely(len <= skb_headlen(skb)))
1943 if (unlikely(len > skb->len))
1945 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1949 * skb_headroom - bytes at buffer head
1950 * @skb: buffer to check
1952 * Return the number of bytes of free space at the head of an &sk_buff.
1954 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1956 return skb->data - skb->head;
1960 * skb_tailroom - bytes at buffer end
1961 * @skb: buffer to check
1963 * Return the number of bytes of free space at the tail of an sk_buff
1965 static inline int skb_tailroom(const struct sk_buff *skb)
1967 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1971 * skb_availroom - bytes at buffer end
1972 * @skb: buffer to check
1974 * Return the number of bytes of free space at the tail of an sk_buff
1975 * allocated by sk_stream_alloc()
1977 static inline int skb_availroom(const struct sk_buff *skb)
1979 if (skb_is_nonlinear(skb))
1982 return skb->end - skb->tail - skb->reserved_tailroom;
1986 * skb_reserve - adjust headroom
1987 * @skb: buffer to alter
1988 * @len: bytes to move
1990 * Increase the headroom of an empty &sk_buff by reducing the tail
1991 * room. This is only allowed for an empty buffer.
1993 static inline void skb_reserve(struct sk_buff *skb, int len)
2000 * skb_tailroom_reserve - adjust reserved_tailroom
2001 * @skb: buffer to alter
2002 * @mtu: maximum amount of headlen permitted
2003 * @needed_tailroom: minimum amount of reserved_tailroom
2005 * Set reserved_tailroom so that headlen can be as large as possible but
2006 * not larger than mtu and tailroom cannot be smaller than
2008 * The required headroom should already have been reserved before using
2011 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2012 unsigned int needed_tailroom)
2014 SKB_LINEAR_ASSERT(skb);
2015 if (mtu < skb_tailroom(skb) - needed_tailroom)
2016 /* use at most mtu */
2017 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2019 /* use up to all available space */
2020 skb->reserved_tailroom = needed_tailroom;
2023 #define ENCAP_TYPE_ETHER 0
2024 #define ENCAP_TYPE_IPPROTO 1
2026 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2029 skb->inner_protocol = protocol;
2030 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2033 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2036 skb->inner_ipproto = ipproto;
2037 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2040 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2042 skb->inner_mac_header = skb->mac_header;
2043 skb->inner_network_header = skb->network_header;
2044 skb->inner_transport_header = skb->transport_header;
2047 static inline void skb_reset_mac_len(struct sk_buff *skb)
2049 skb->mac_len = skb->network_header - skb->mac_header;
2052 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2055 return skb->head + skb->inner_transport_header;
2058 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2060 return skb_inner_transport_header(skb) - skb->data;
2063 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2065 skb->inner_transport_header = skb->data - skb->head;
2068 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2071 skb_reset_inner_transport_header(skb);
2072 skb->inner_transport_header += offset;
2075 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2077 return skb->head + skb->inner_network_header;
2080 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2082 skb->inner_network_header = skb->data - skb->head;
2085 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2088 skb_reset_inner_network_header(skb);
2089 skb->inner_network_header += offset;
2092 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2094 return skb->head + skb->inner_mac_header;
2097 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2099 skb->inner_mac_header = skb->data - skb->head;
2102 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2105 skb_reset_inner_mac_header(skb);
2106 skb->inner_mac_header += offset;
2108 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2110 return skb->transport_header != (typeof(skb->transport_header))~0U;
2113 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2115 return skb->head + skb->transport_header;
2118 static inline void skb_reset_transport_header(struct sk_buff *skb)
2120 skb->transport_header = skb->data - skb->head;
2123 static inline void skb_set_transport_header(struct sk_buff *skb,
2126 skb_reset_transport_header(skb);
2127 skb->transport_header += offset;
2130 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2132 return skb->head + skb->network_header;
2135 static inline void skb_reset_network_header(struct sk_buff *skb)
2137 skb->network_header = skb->data - skb->head;
2140 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2142 skb_reset_network_header(skb);
2143 skb->network_header += offset;
2146 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2148 return skb->head + skb->mac_header;
2151 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2153 return skb->mac_header != (typeof(skb->mac_header))~0U;
2156 static inline void skb_reset_mac_header(struct sk_buff *skb)
2158 skb->mac_header = skb->data - skb->head;
2161 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2163 skb_reset_mac_header(skb);
2164 skb->mac_header += offset;
2167 static inline void skb_pop_mac_header(struct sk_buff *skb)
2169 skb->mac_header = skb->network_header;
2172 static inline void skb_probe_transport_header(struct sk_buff *skb,
2173 const int offset_hint)
2175 struct flow_keys keys;
2177 if (skb_transport_header_was_set(skb))
2179 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2180 skb_set_transport_header(skb, keys.control.thoff);
2182 skb_set_transport_header(skb, offset_hint);
2185 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2187 if (skb_mac_header_was_set(skb)) {
2188 const unsigned char *old_mac = skb_mac_header(skb);
2190 skb_set_mac_header(skb, -skb->mac_len);
2191 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2195 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2197 return skb->csum_start - skb_headroom(skb);
2200 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2202 return skb->head + skb->csum_start;
2205 static inline int skb_transport_offset(const struct sk_buff *skb)
2207 return skb_transport_header(skb) - skb->data;
2210 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2212 return skb->transport_header - skb->network_header;
2215 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2217 return skb->inner_transport_header - skb->inner_network_header;
2220 static inline int skb_network_offset(const struct sk_buff *skb)
2222 return skb_network_header(skb) - skb->data;
2225 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2227 return skb_inner_network_header(skb) - skb->data;
2230 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2232 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2236 * CPUs often take a performance hit when accessing unaligned memory
2237 * locations. The actual performance hit varies, it can be small if the
2238 * hardware handles it or large if we have to take an exception and fix it
2241 * Since an ethernet header is 14 bytes network drivers often end up with
2242 * the IP header at an unaligned offset. The IP header can be aligned by
2243 * shifting the start of the packet by 2 bytes. Drivers should do this
2246 * skb_reserve(skb, NET_IP_ALIGN);
2248 * The downside to this alignment of the IP header is that the DMA is now
2249 * unaligned. On some architectures the cost of an unaligned DMA is high
2250 * and this cost outweighs the gains made by aligning the IP header.
2252 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2255 #ifndef NET_IP_ALIGN
2256 #define NET_IP_ALIGN 2
2260 * The networking layer reserves some headroom in skb data (via
2261 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2262 * the header has to grow. In the default case, if the header has to grow
2263 * 32 bytes or less we avoid the reallocation.
2265 * Unfortunately this headroom changes the DMA alignment of the resulting
2266 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2267 * on some architectures. An architecture can override this value,
2268 * perhaps setting it to a cacheline in size (since that will maintain
2269 * cacheline alignment of the DMA). It must be a power of 2.
2271 * Various parts of the networking layer expect at least 32 bytes of
2272 * headroom, you should not reduce this.
2274 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2275 * to reduce average number of cache lines per packet.
2276 * get_rps_cpus() for example only access one 64 bytes aligned block :
2277 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2280 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2283 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2285 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2287 if (unlikely(skb_is_nonlinear(skb))) {
2292 skb_set_tail_pointer(skb, len);
2295 void skb_trim(struct sk_buff *skb, unsigned int len);
2297 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2300 return ___pskb_trim(skb, len);
2301 __skb_trim(skb, len);
2305 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2307 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2311 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2312 * @skb: buffer to alter
2315 * This is identical to pskb_trim except that the caller knows that
2316 * the skb is not cloned so we should never get an error due to out-
2319 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2321 int err = pskb_trim(skb, len);
2326 * skb_orphan - orphan a buffer
2327 * @skb: buffer to orphan
2329 * If a buffer currently has an owner then we call the owner's
2330 * destructor function and make the @skb unowned. The buffer continues
2331 * to exist but is no longer charged to its former owner.
2333 static inline void skb_orphan(struct sk_buff *skb)
2335 if (skb->destructor) {
2336 skb->destructor(skb);
2337 skb->destructor = NULL;
2345 * skb_orphan_frags - orphan the frags contained in a buffer
2346 * @skb: buffer to orphan frags from
2347 * @gfp_mask: allocation mask for replacement pages
2349 * For each frag in the SKB which needs a destructor (i.e. has an
2350 * owner) create a copy of that frag and release the original
2351 * page by calling the destructor.
2353 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2355 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2357 return skb_copy_ubufs(skb, gfp_mask);
2361 * __skb_queue_purge - empty a list
2362 * @list: list to empty
2364 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2365 * the list and one reference dropped. This function does not take the
2366 * list lock and the caller must hold the relevant locks to use it.
2368 void skb_queue_purge(struct sk_buff_head *list);
2369 static inline void __skb_queue_purge(struct sk_buff_head *list)
2371 struct sk_buff *skb;
2372 while ((skb = __skb_dequeue(list)) != NULL)
2376 void *netdev_alloc_frag(unsigned int fragsz);
2378 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2382 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2383 * @dev: network device to receive on
2384 * @length: length to allocate
2386 * Allocate a new &sk_buff and assign it a usage count of one. The
2387 * buffer has unspecified headroom built in. Users should allocate
2388 * the headroom they think they need without accounting for the
2389 * built in space. The built in space is used for optimisations.
2391 * %NULL is returned if there is no free memory. Although this function
2392 * allocates memory it can be called from an interrupt.
2394 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2395 unsigned int length)
2397 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2400 /* legacy helper around __netdev_alloc_skb() */
2401 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2404 return __netdev_alloc_skb(NULL, length, gfp_mask);
2407 /* legacy helper around netdev_alloc_skb() */
2408 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2410 return netdev_alloc_skb(NULL, length);
2414 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2415 unsigned int length, gfp_t gfp)
2417 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2419 if (NET_IP_ALIGN && skb)
2420 skb_reserve(skb, NET_IP_ALIGN);
2424 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2425 unsigned int length)
2427 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2430 static inline void skb_free_frag(void *addr)
2432 __free_page_frag(addr);
2435 void *napi_alloc_frag(unsigned int fragsz);
2436 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2437 unsigned int length, gfp_t gfp_mask);
2438 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2439 unsigned int length)
2441 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2443 void napi_consume_skb(struct sk_buff *skb, int budget);
2445 void __kfree_skb_flush(void);
2446 void __kfree_skb_defer(struct sk_buff *skb);
2449 * __dev_alloc_pages - allocate page for network Rx
2450 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2451 * @order: size of the allocation
2453 * Allocate a new page.
2455 * %NULL is returned if there is no free memory.
2457 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2460 /* This piece of code contains several assumptions.
2461 * 1. This is for device Rx, therefor a cold page is preferred.
2462 * 2. The expectation is the user wants a compound page.
2463 * 3. If requesting a order 0 page it will not be compound
2464 * due to the check to see if order has a value in prep_new_page
2465 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2466 * code in gfp_to_alloc_flags that should be enforcing this.
2468 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2470 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2473 static inline struct page *dev_alloc_pages(unsigned int order)
2475 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2479 * __dev_alloc_page - allocate a page for network Rx
2480 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2482 * Allocate a new page.
2484 * %NULL is returned if there is no free memory.
2486 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2488 return __dev_alloc_pages(gfp_mask, 0);
2491 static inline struct page *dev_alloc_page(void)
2493 return dev_alloc_pages(0);
2497 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2498 * @page: The page that was allocated from skb_alloc_page
2499 * @skb: The skb that may need pfmemalloc set
2501 static inline void skb_propagate_pfmemalloc(struct page *page,
2502 struct sk_buff *skb)
2504 if (page_is_pfmemalloc(page))
2505 skb->pfmemalloc = true;
2509 * skb_frag_page - retrieve the page referred to by a paged fragment
2510 * @frag: the paged fragment
2512 * Returns the &struct page associated with @frag.
2514 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2516 return frag->page.p;
2520 * __skb_frag_ref - take an addition reference on a paged fragment.
2521 * @frag: the paged fragment
2523 * Takes an additional reference on the paged fragment @frag.
2525 static inline void __skb_frag_ref(skb_frag_t *frag)
2527 get_page(skb_frag_page(frag));
2531 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2533 * @f: the fragment offset.
2535 * Takes an additional reference on the @f'th paged fragment of @skb.
2537 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2539 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2543 * __skb_frag_unref - release a reference on a paged fragment.
2544 * @frag: the paged fragment
2546 * Releases a reference on the paged fragment @frag.
2548 static inline void __skb_frag_unref(skb_frag_t *frag)
2550 put_page(skb_frag_page(frag));
2554 * skb_frag_unref - release a reference on a paged fragment of an skb.
2556 * @f: the fragment offset
2558 * Releases a reference on the @f'th paged fragment of @skb.
2560 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2562 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2566 * skb_frag_address - gets the address of the data contained in a paged fragment
2567 * @frag: the paged fragment buffer
2569 * Returns the address of the data within @frag. The page must already
2572 static inline void *skb_frag_address(const skb_frag_t *frag)
2574 return page_address(skb_frag_page(frag)) + frag->page_offset;
2578 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2579 * @frag: the paged fragment buffer
2581 * Returns the address of the data within @frag. Checks that the page
2582 * is mapped and returns %NULL otherwise.
2584 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2586 void *ptr = page_address(skb_frag_page(frag));
2590 return ptr + frag->page_offset;
2594 * __skb_frag_set_page - sets the page contained in a paged fragment
2595 * @frag: the paged fragment
2596 * @page: the page to set
2598 * Sets the fragment @frag to contain @page.
2600 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2602 frag->page.p = page;
2606 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2608 * @f: the fragment offset
2609 * @page: the page to set
2611 * Sets the @f'th fragment of @skb to contain @page.
2613 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2616 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2619 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2622 * skb_frag_dma_map - maps a paged fragment via the DMA API
2623 * @dev: the device to map the fragment to
2624 * @frag: the paged fragment to map
2625 * @offset: the offset within the fragment (starting at the
2626 * fragment's own offset)
2627 * @size: the number of bytes to map
2628 * @dir: the direction of the mapping (%PCI_DMA_*)
2630 * Maps the page associated with @frag to @device.
2632 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2633 const skb_frag_t *frag,
2634 size_t offset, size_t size,
2635 enum dma_data_direction dir)
2637 return dma_map_page(dev, skb_frag_page(frag),
2638 frag->page_offset + offset, size, dir);
2641 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2644 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2648 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2651 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2656 * skb_clone_writable - is the header of a clone writable
2657 * @skb: buffer to check
2658 * @len: length up to which to write
2660 * Returns true if modifying the header part of the cloned buffer
2661 * does not requires the data to be copied.
2663 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2665 return !skb_header_cloned(skb) &&
2666 skb_headroom(skb) + len <= skb->hdr_len;
2669 static inline int skb_try_make_writable(struct sk_buff *skb,
2670 unsigned int write_len)
2672 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2673 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2676 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2681 if (headroom > skb_headroom(skb))
2682 delta = headroom - skb_headroom(skb);
2684 if (delta || cloned)
2685 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2691 * skb_cow - copy header of skb when it is required
2692 * @skb: buffer to cow
2693 * @headroom: needed headroom
2695 * If the skb passed lacks sufficient headroom or its data part
2696 * is shared, data is reallocated. If reallocation fails, an error
2697 * is returned and original skb is not changed.
2699 * The result is skb with writable area skb->head...skb->tail
2700 * and at least @headroom of space at head.
2702 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2704 return __skb_cow(skb, headroom, skb_cloned(skb));
2708 * skb_cow_head - skb_cow but only making the head writable
2709 * @skb: buffer to cow
2710 * @headroom: needed headroom
2712 * This function is identical to skb_cow except that we replace the
2713 * skb_cloned check by skb_header_cloned. It should be used when
2714 * you only need to push on some header and do not need to modify
2717 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2719 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2723 * skb_padto - pad an skbuff up to a minimal size
2724 * @skb: buffer to pad
2725 * @len: minimal length
2727 * Pads up a buffer to ensure the trailing bytes exist and are
2728 * blanked. If the buffer already contains sufficient data it
2729 * is untouched. Otherwise it is extended. Returns zero on
2730 * success. The skb is freed on error.
2732 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2734 unsigned int size = skb->len;
2735 if (likely(size >= len))
2737 return skb_pad(skb, len - size);
2741 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2742 * @skb: buffer to pad
2743 * @len: minimal length
2745 * Pads up a buffer to ensure the trailing bytes exist and are
2746 * blanked. If the buffer already contains sufficient data it
2747 * is untouched. Otherwise it is extended. Returns zero on
2748 * success. The skb is freed on error.
2750 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2752 unsigned int size = skb->len;
2754 if (unlikely(size < len)) {
2756 if (skb_pad(skb, len))
2758 __skb_put(skb, len);
2763 static inline int skb_add_data(struct sk_buff *skb,
2764 struct iov_iter *from, int copy)
2766 const int off = skb->len;
2768 if (skb->ip_summed == CHECKSUM_NONE) {
2770 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2771 &csum, from) == copy) {
2772 skb->csum = csum_block_add(skb->csum, csum, off);
2775 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
2778 __skb_trim(skb, off);
2782 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2783 const struct page *page, int off)
2786 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2788 return page == skb_frag_page(frag) &&
2789 off == frag->page_offset + skb_frag_size(frag);
2794 static inline int __skb_linearize(struct sk_buff *skb)
2796 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2800 * skb_linearize - convert paged skb to linear one
2801 * @skb: buffer to linarize
2803 * If there is no free memory -ENOMEM is returned, otherwise zero
2804 * is returned and the old skb data released.
2806 static inline int skb_linearize(struct sk_buff *skb)
2808 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2812 * skb_has_shared_frag - can any frag be overwritten
2813 * @skb: buffer to test
2815 * Return true if the skb has at least one frag that might be modified
2816 * by an external entity (as in vmsplice()/sendfile())
2818 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2820 return skb_is_nonlinear(skb) &&
2821 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2825 * skb_linearize_cow - make sure skb is linear and writable
2826 * @skb: buffer to process
2828 * If there is no free memory -ENOMEM is returned, otherwise zero
2829 * is returned and the old skb data released.
2831 static inline int skb_linearize_cow(struct sk_buff *skb)
2833 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2834 __skb_linearize(skb) : 0;
2838 * skb_postpull_rcsum - update checksum for received skb after pull
2839 * @skb: buffer to update
2840 * @start: start of data before pull
2841 * @len: length of data pulled
2843 * After doing a pull on a received packet, you need to call this to
2844 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2845 * CHECKSUM_NONE so that it can be recomputed from scratch.
2848 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2849 const void *start, unsigned int len)
2851 if (skb->ip_summed == CHECKSUM_COMPLETE)
2852 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2853 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2854 skb_checksum_start_offset(skb) < 0)
2855 skb->ip_summed = CHECKSUM_NONE;
2858 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2860 static inline void skb_postpush_rcsum(struct sk_buff *skb,
2861 const void *start, unsigned int len)
2863 /* For performing the reverse operation to skb_postpull_rcsum(),
2864 * we can instead of ...
2866 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0));
2868 * ... just use this equivalent version here to save a few
2869 * instructions. Feeding csum of 0 in csum_partial() and later
2870 * on adding skb->csum is equivalent to feed skb->csum in the
2873 if (skb->ip_summed == CHECKSUM_COMPLETE)
2874 skb->csum = csum_partial(start, len, skb->csum);
2878 * pskb_trim_rcsum - trim received skb and update checksum
2879 * @skb: buffer to trim
2882 * This is exactly the same as pskb_trim except that it ensures the
2883 * checksum of received packets are still valid after the operation.
2886 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2888 if (likely(len >= skb->len))
2890 if (skb->ip_summed == CHECKSUM_COMPLETE)
2891 skb->ip_summed = CHECKSUM_NONE;
2892 return __pskb_trim(skb, len);
2895 #define skb_queue_walk(queue, skb) \
2896 for (skb = (queue)->next; \
2897 skb != (struct sk_buff *)(queue); \
2900 #define skb_queue_walk_safe(queue, skb, tmp) \
2901 for (skb = (queue)->next, tmp = skb->next; \
2902 skb != (struct sk_buff *)(queue); \
2903 skb = tmp, tmp = skb->next)
2905 #define skb_queue_walk_from(queue, skb) \
2906 for (; skb != (struct sk_buff *)(queue); \
2909 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2910 for (tmp = skb->next; \
2911 skb != (struct sk_buff *)(queue); \
2912 skb = tmp, tmp = skb->next)
2914 #define skb_queue_reverse_walk(queue, skb) \
2915 for (skb = (queue)->prev; \
2916 skb != (struct sk_buff *)(queue); \
2919 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2920 for (skb = (queue)->prev, tmp = skb->prev; \
2921 skb != (struct sk_buff *)(queue); \
2922 skb = tmp, tmp = skb->prev)
2924 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2925 for (tmp = skb->prev; \
2926 skb != (struct sk_buff *)(queue); \
2927 skb = tmp, tmp = skb->prev)
2929 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2931 return skb_shinfo(skb)->frag_list != NULL;
2934 static inline void skb_frag_list_init(struct sk_buff *skb)
2936 skb_shinfo(skb)->frag_list = NULL;
2939 #define skb_walk_frags(skb, iter) \
2940 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2943 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
2944 const struct sk_buff *skb);
2945 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
2946 int *peeked, int *off, int *err,
2947 struct sk_buff **last);
2948 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2949 int *peeked, int *off, int *err);
2950 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2952 unsigned int datagram_poll(struct file *file, struct socket *sock,
2953 struct poll_table_struct *wait);
2954 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2955 struct iov_iter *to, int size);
2956 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2957 struct msghdr *msg, int size)
2959 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
2961 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2962 struct msghdr *msg);
2963 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2964 struct iov_iter *from, int len);
2965 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
2966 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2967 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
2968 static inline void skb_free_datagram_locked(struct sock *sk,
2969 struct sk_buff *skb)
2971 __skb_free_datagram_locked(sk, skb, 0);
2973 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2974 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2975 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2976 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2977 int len, __wsum csum);
2978 ssize_t skb_socket_splice(struct sock *sk,
2979 struct pipe_inode_info *pipe,
2980 struct splice_pipe_desc *spd);
2981 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2982 struct pipe_inode_info *pipe, unsigned int len,
2984 ssize_t (*splice_cb)(struct sock *,
2985 struct pipe_inode_info *,
2986 struct splice_pipe_desc *));
2987 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2988 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2989 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2991 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2992 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2993 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2994 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2995 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2996 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
2997 int skb_ensure_writable(struct sk_buff *skb, int write_len);
2998 int skb_vlan_pop(struct sk_buff *skb);
2999 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3000 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3003 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3005 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3008 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3010 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3013 struct skb_checksum_ops {
3014 __wsum (*update)(const void *mem, int len, __wsum wsum);
3015 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3018 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3019 __wsum csum, const struct skb_checksum_ops *ops);
3020 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3023 static inline void * __must_check
3024 __skb_header_pointer(const struct sk_buff *skb, int offset,
3025 int len, void *data, int hlen, void *buffer)
3027 if (hlen - offset >= len)
3028 return data + offset;
3031 skb_copy_bits(skb, offset, buffer, len) < 0)
3037 static inline void * __must_check
3038 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3040 return __skb_header_pointer(skb, offset, len, skb->data,
3041 skb_headlen(skb), buffer);
3045 * skb_needs_linearize - check if we need to linearize a given skb
3046 * depending on the given device features.
3047 * @skb: socket buffer to check
3048 * @features: net device features
3050 * Returns true if either:
3051 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3052 * 2. skb is fragmented and the device does not support SG.
3054 static inline bool skb_needs_linearize(struct sk_buff *skb,
3055 netdev_features_t features)
3057 return skb_is_nonlinear(skb) &&
3058 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3059 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3062 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3064 const unsigned int len)
3066 memcpy(to, skb->data, len);
3069 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3070 const int offset, void *to,
3071 const unsigned int len)
3073 memcpy(to, skb->data + offset, len);
3076 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3078 const unsigned int len)
3080 memcpy(skb->data, from, len);
3083 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3086 const unsigned int len)
3088 memcpy(skb->data + offset, from, len);
3091 void skb_init(void);
3093 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3099 * skb_get_timestamp - get timestamp from a skb
3100 * @skb: skb to get stamp from
3101 * @stamp: pointer to struct timeval to store stamp in
3103 * Timestamps are stored in the skb as offsets to a base timestamp.
3104 * This function converts the offset back to a struct timeval and stores
3107 static inline void skb_get_timestamp(const struct sk_buff *skb,
3108 struct timeval *stamp)
3110 *stamp = ktime_to_timeval(skb->tstamp);
3113 static inline void skb_get_timestampns(const struct sk_buff *skb,
3114 struct timespec *stamp)
3116 *stamp = ktime_to_timespec(skb->tstamp);
3119 static inline void __net_timestamp(struct sk_buff *skb)
3121 skb->tstamp = ktime_get_real();
3124 static inline ktime_t net_timedelta(ktime_t t)
3126 return ktime_sub(ktime_get_real(), t);
3129 static inline ktime_t net_invalid_timestamp(void)
3131 return ktime_set(0, 0);
3134 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3136 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3138 void skb_clone_tx_timestamp(struct sk_buff *skb);
3139 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3141 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3143 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3147 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3152 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3155 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3157 * PHY drivers may accept clones of transmitted packets for
3158 * timestamping via their phy_driver.txtstamp method. These drivers
3159 * must call this function to return the skb back to the stack with a
3162 * @skb: clone of the the original outgoing packet
3163 * @hwtstamps: hardware time stamps
3166 void skb_complete_tx_timestamp(struct sk_buff *skb,
3167 struct skb_shared_hwtstamps *hwtstamps);
3169 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3170 struct skb_shared_hwtstamps *hwtstamps,
3171 struct sock *sk, int tstype);
3174 * skb_tstamp_tx - queue clone of skb with send time stamps
3175 * @orig_skb: the original outgoing packet
3176 * @hwtstamps: hardware time stamps, may be NULL if not available
3178 * If the skb has a socket associated, then this function clones the
3179 * skb (thus sharing the actual data and optional structures), stores
3180 * the optional hardware time stamping information (if non NULL) or
3181 * generates a software time stamp (otherwise), then queues the clone
3182 * to the error queue of the socket. Errors are silently ignored.
3184 void skb_tstamp_tx(struct sk_buff *orig_skb,
3185 struct skb_shared_hwtstamps *hwtstamps);
3187 static inline void sw_tx_timestamp(struct sk_buff *skb)
3189 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3190 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
3191 skb_tstamp_tx(skb, NULL);
3195 * skb_tx_timestamp() - Driver hook for transmit timestamping
3197 * Ethernet MAC Drivers should call this function in their hard_xmit()
3198 * function immediately before giving the sk_buff to the MAC hardware.
3200 * Specifically, one should make absolutely sure that this function is
3201 * called before TX completion of this packet can trigger. Otherwise
3202 * the packet could potentially already be freed.
3204 * @skb: A socket buffer.
3206 static inline void skb_tx_timestamp(struct sk_buff *skb)
3208 skb_clone_tx_timestamp(skb);
3209 sw_tx_timestamp(skb);
3213 * skb_complete_wifi_ack - deliver skb with wifi status
3215 * @skb: the original outgoing packet
3216 * @acked: ack status
3219 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3221 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3222 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3224 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3226 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3228 (skb->ip_summed == CHECKSUM_PARTIAL &&
3229 skb_checksum_start_offset(skb) >= 0));
3233 * skb_checksum_complete - Calculate checksum of an entire packet
3234 * @skb: packet to process
3236 * This function calculates the checksum over the entire packet plus
3237 * the value of skb->csum. The latter can be used to supply the
3238 * checksum of a pseudo header as used by TCP/UDP. It returns the
3241 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3242 * this function can be used to verify that checksum on received
3243 * packets. In that case the function should return zero if the
3244 * checksum is correct. In particular, this function will return zero
3245 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3246 * hardware has already verified the correctness of the checksum.
3248 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3250 return skb_csum_unnecessary(skb) ?
3251 0 : __skb_checksum_complete(skb);
3254 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3256 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3257 if (skb->csum_level == 0)
3258 skb->ip_summed = CHECKSUM_NONE;
3264 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3266 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3267 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3269 } else if (skb->ip_summed == CHECKSUM_NONE) {
3270 skb->ip_summed = CHECKSUM_UNNECESSARY;
3271 skb->csum_level = 0;
3275 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3277 /* Mark current checksum as bad (typically called from GRO
3278 * path). In the case that ip_summed is CHECKSUM_NONE
3279 * this must be the first checksum encountered in the packet.
3280 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3281 * checksum after the last one validated. For UDP, a zero
3282 * checksum can not be marked as bad.
3285 if (skb->ip_summed == CHECKSUM_NONE ||
3286 skb->ip_summed == CHECKSUM_UNNECESSARY)
3290 /* Check if we need to perform checksum complete validation.
3292 * Returns true if checksum complete is needed, false otherwise
3293 * (either checksum is unnecessary or zero checksum is allowed).
3295 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3299 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3300 skb->csum_valid = 1;
3301 __skb_decr_checksum_unnecessary(skb);
3308 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3311 #define CHECKSUM_BREAK 76
3313 /* Unset checksum-complete
3315 * Unset checksum complete can be done when packet is being modified
3316 * (uncompressed for instance) and checksum-complete value is
3319 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3321 if (skb->ip_summed == CHECKSUM_COMPLETE)
3322 skb->ip_summed = CHECKSUM_NONE;
3325 /* Validate (init) checksum based on checksum complete.
3328 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3329 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3330 * checksum is stored in skb->csum for use in __skb_checksum_complete
3331 * non-zero: value of invalid checksum
3334 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3338 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3339 if (!csum_fold(csum_add(psum, skb->csum))) {
3340 skb->csum_valid = 1;
3343 } else if (skb->csum_bad) {
3344 /* ip_summed == CHECKSUM_NONE in this case */
3345 return (__force __sum16)1;
3350 if (complete || skb->len <= CHECKSUM_BREAK) {
3353 csum = __skb_checksum_complete(skb);
3354 skb->csum_valid = !csum;
3361 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3366 /* Perform checksum validate (init). Note that this is a macro since we only
3367 * want to calculate the pseudo header which is an input function if necessary.
3368 * First we try to validate without any computation (checksum unnecessary) and
3369 * then calculate based on checksum complete calling the function to compute
3373 * 0: checksum is validated or try to in skb_checksum_complete
3374 * non-zero: value of invalid checksum
3376 #define __skb_checksum_validate(skb, proto, complete, \
3377 zero_okay, check, compute_pseudo) \
3379 __sum16 __ret = 0; \
3380 skb->csum_valid = 0; \
3381 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3382 __ret = __skb_checksum_validate_complete(skb, \
3383 complete, compute_pseudo(skb, proto)); \
3387 #define skb_checksum_init(skb, proto, compute_pseudo) \
3388 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3390 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3391 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3393 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3394 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3396 #define skb_checksum_validate_zero_check(skb, proto, check, \
3398 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3400 #define skb_checksum_simple_validate(skb) \
3401 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3403 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3405 return (skb->ip_summed == CHECKSUM_NONE &&
3406 skb->csum_valid && !skb->csum_bad);
3409 static inline void __skb_checksum_convert(struct sk_buff *skb,
3410 __sum16 check, __wsum pseudo)
3412 skb->csum = ~pseudo;
3413 skb->ip_summed = CHECKSUM_COMPLETE;
3416 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3418 if (__skb_checksum_convert_check(skb)) \
3419 __skb_checksum_convert(skb, check, \
3420 compute_pseudo(skb, proto)); \
3423 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3424 u16 start, u16 offset)
3426 skb->ip_summed = CHECKSUM_PARTIAL;
3427 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3428 skb->csum_offset = offset - start;
3431 /* Update skbuf and packet to reflect the remote checksum offload operation.
3432 * When called, ptr indicates the starting point for skb->csum when
3433 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3434 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3436 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3437 int start, int offset, bool nopartial)
3442 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3446 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3447 __skb_checksum_complete(skb);
3448 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3451 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3453 /* Adjust skb->csum since we changed the packet */
3454 skb->csum = csum_add(skb->csum, delta);
3457 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3458 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3459 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3461 if (nfct && atomic_dec_and_test(&nfct->use))
3462 nf_conntrack_destroy(nfct);
3464 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3467 atomic_inc(&nfct->use);
3470 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3471 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3473 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3476 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3479 atomic_inc(&nf_bridge->use);
3481 #endif /* CONFIG_BRIDGE_NETFILTER */
3482 static inline void nf_reset(struct sk_buff *skb)
3484 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3485 nf_conntrack_put(skb->nfct);
3488 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3489 nf_bridge_put(skb->nf_bridge);
3490 skb->nf_bridge = NULL;
3494 static inline void nf_reset_trace(struct sk_buff *skb)
3496 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3501 /* Note: This doesn't put any conntrack and bridge info in dst. */
3502 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3505 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3506 dst->nfct = src->nfct;
3507 nf_conntrack_get(src->nfct);
3509 dst->nfctinfo = src->nfctinfo;
3511 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3512 dst->nf_bridge = src->nf_bridge;
3513 nf_bridge_get(src->nf_bridge);
3515 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3517 dst->nf_trace = src->nf_trace;
3521 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3523 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3524 nf_conntrack_put(dst->nfct);
3526 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3527 nf_bridge_put(dst->nf_bridge);
3529 __nf_copy(dst, src, true);
3532 #ifdef CONFIG_NETWORK_SECMARK
3533 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3535 to->secmark = from->secmark;
3538 static inline void skb_init_secmark(struct sk_buff *skb)
3543 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3546 static inline void skb_init_secmark(struct sk_buff *skb)
3550 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3552 return !skb->destructor &&
3553 #if IS_ENABLED(CONFIG_XFRM)
3556 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3559 !skb->_skb_refdst &&
3560 !skb_has_frag_list(skb);
3563 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3565 skb->queue_mapping = queue_mapping;
3568 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3570 return skb->queue_mapping;
3573 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3575 to->queue_mapping = from->queue_mapping;
3578 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3580 skb->queue_mapping = rx_queue + 1;
3583 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3585 return skb->queue_mapping - 1;
3588 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3590 return skb->queue_mapping != 0;
3593 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3602 /* Keeps track of mac header offset relative to skb->head.
3603 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3604 * For non-tunnel skb it points to skb_mac_header() and for
3605 * tunnel skb it points to outer mac header.
3606 * Keeps track of level of encapsulation of network headers.
3617 #define SKB_SGO_CB_OFFSET 32
3618 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3620 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3622 return (skb_mac_header(inner_skb) - inner_skb->head) -
3623 SKB_GSO_CB(inner_skb)->mac_offset;
3626 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3628 int new_headroom, headroom;
3631 headroom = skb_headroom(skb);
3632 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3636 new_headroom = skb_headroom(skb);
3637 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3641 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3643 /* Do not update partial checksums if remote checksum is enabled. */
3644 if (skb->remcsum_offload)
3647 SKB_GSO_CB(skb)->csum = res;
3648 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3651 /* Compute the checksum for a gso segment. First compute the checksum value
3652 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3653 * then add in skb->csum (checksum from csum_start to end of packet).
3654 * skb->csum and csum_start are then updated to reflect the checksum of the
3655 * resultant packet starting from the transport header-- the resultant checksum
3656 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3659 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3661 unsigned char *csum_start = skb_transport_header(skb);
3662 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3663 __wsum partial = SKB_GSO_CB(skb)->csum;
3665 SKB_GSO_CB(skb)->csum = res;
3666 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
3668 return csum_fold(csum_partial(csum_start, plen, partial));
3671 static inline bool skb_is_gso(const struct sk_buff *skb)
3673 return skb_shinfo(skb)->gso_size;
3676 /* Note: Should be called only if skb_is_gso(skb) is true */
3677 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3679 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3682 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3684 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3686 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3687 * wanted then gso_type will be set. */
3688 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3690 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3691 unlikely(shinfo->gso_type == 0)) {
3692 __skb_warn_lro_forwarding(skb);
3698 static inline void skb_forward_csum(struct sk_buff *skb)
3700 /* Unfortunately we don't support this one. Any brave souls? */
3701 if (skb->ip_summed == CHECKSUM_COMPLETE)
3702 skb->ip_summed = CHECKSUM_NONE;
3706 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3707 * @skb: skb to check
3709 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3710 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3711 * use this helper, to document places where we make this assertion.
3713 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3716 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3720 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3722 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3723 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3724 unsigned int transport_len,
3725 __sum16(*skb_chkf)(struct sk_buff *skb));
3728 * skb_head_is_locked - Determine if the skb->head is locked down
3729 * @skb: skb to check
3731 * The head on skbs build around a head frag can be removed if they are
3732 * not cloned. This function returns true if the skb head is locked down
3733 * due to either being allocated via kmalloc, or by being a clone with
3734 * multiple references to the head.
3736 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3738 return !skb->head_frag || skb_cloned(skb);
3742 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3746 * skb_gso_network_seglen is used to determine the real size of the
3747 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3749 * The MAC/L2 header is not accounted for.
3751 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3753 unsigned int hdr_len = skb_transport_header(skb) -
3754 skb_network_header(skb);
3755 return hdr_len + skb_gso_transport_seglen(skb);
3758 /* Local Checksum Offload.
3759 * Compute outer checksum based on the assumption that the
3760 * inner checksum will be offloaded later.
3761 * See Documentation/networking/checksum-offloads.txt for
3762 * explanation of how this works.
3763 * Fill in outer checksum adjustment (e.g. with sum of outer
3764 * pseudo-header) before calling.
3765 * Also ensure that inner checksum is in linear data area.
3767 static inline __wsum lco_csum(struct sk_buff *skb)
3769 unsigned char *csum_start = skb_checksum_start(skb);
3770 unsigned char *l4_hdr = skb_transport_header(skb);
3773 /* Start with complement of inner checksum adjustment */
3774 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3777 /* Add in checksum of our headers (incl. outer checksum
3778 * adjustment filled in by caller) and return result.
3780 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
3783 #endif /* __KERNEL__ */
3784 #endif /* _LINUX_SKBUFF_H */