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 /* A. Checksumming of received packets by device.
46 * Device failed to checksum this packet e.g. due to lack of capabilities.
47 * The packet contains full (though not verified) checksum in packet but
48 * not in skb->csum. Thus, skb->csum is undefined in this case.
50 * CHECKSUM_UNNECESSARY:
52 * The hardware you're dealing with doesn't calculate the full checksum
53 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
54 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
55 * if their checksums are okay. skb->csum is still undefined in this case
56 * though. It is a bad option, but, unfortunately, nowadays most vendors do
57 * this. Apparently with the secret goal to sell you new devices, when you
58 * will add new protocol to your host, f.e. IPv6 8)
60 * CHECKSUM_UNNECESSARY is applicable to following protocols:
62 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
63 * zero UDP checksum for either IPv4 or IPv6, the networking stack
64 * may perform further validation in this case.
65 * GRE: only if the checksum is present in the header.
66 * SCTP: indicates the CRC in SCTP header has been validated.
68 * skb->csum_level indicates the number of consecutive checksums found in
69 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
70 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
71 * and a device is able to verify the checksums for UDP (possibly zero),
72 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
73 * two. If the device were only able to verify the UDP checksum and not
74 * GRE, either because it doesn't support GRE checksum of because GRE
75 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
76 * not considered in this case).
80 * This is the most generic way. The device supplied checksum of the _whole_
81 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
82 * hardware doesn't need to parse L3/L4 headers to implement this.
84 * Note: Even if device supports only some protocols, but is able to produce
85 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
89 * A checksum is set up to be offloaded to a device as described in the
90 * output description for CHECKSUM_PARTIAL. This may occur on a packet
91 * received directly from another Linux OS, e.g., a virtualized Linux kernel
92 * on the same host, or it may be set in the input path in GRO or remote
93 * checksum offload. For the purposes of checksum verification, the checksum
94 * referred to by skb->csum_start + skb->csum_offset and any preceding
95 * checksums in the packet are considered verified. Any checksums in the
96 * packet that are after the checksum being offloaded are not considered to
99 * B. Checksumming on output.
103 * The skb was already checksummed by the protocol, or a checksum is not
108 * The device is required to checksum the packet as seen by hard_start_xmit()
109 * from skb->csum_start up to the end, and to record/write the checksum at
110 * offset skb->csum_start + skb->csum_offset.
112 * The device must show its capabilities in dev->features, set up at device
113 * setup time, e.g. netdev_features.h:
115 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything.
116 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over
117 * IPv4. Sigh. Vendors like this way for an unknown reason.
118 * Though, see comment above about CHECKSUM_UNNECESSARY. 8)
119 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead.
120 * NETIF_F_... - Well, you get the picture.
122 * CHECKSUM_UNNECESSARY:
124 * Normally, the device will do per protocol specific checksumming. Protocol
125 * implementations that do not want the NIC to perform the checksum
126 * calculation should use this flag in their outgoing skbs.
128 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC
129 * offload. Correspondingly, the FCoE protocol driver
130 * stack should use CHECKSUM_UNNECESSARY.
132 * Any questions? No questions, good. --ANK
135 /* Don't change this without changing skb_csum_unnecessary! */
136 #define CHECKSUM_NONE 0
137 #define CHECKSUM_UNNECESSARY 1
138 #define CHECKSUM_COMPLETE 2
139 #define CHECKSUM_PARTIAL 3
141 /* Maximum value in skb->csum_level */
142 #define SKB_MAX_CSUM_LEVEL 3
144 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
145 #define SKB_WITH_OVERHEAD(X) \
146 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
147 #define SKB_MAX_ORDER(X, ORDER) \
148 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
149 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
150 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
152 /* return minimum truesize of one skb containing X bytes of data */
153 #define SKB_TRUESIZE(X) ((X) + \
154 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
155 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
159 struct pipe_inode_info;
163 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
164 struct nf_conntrack {
169 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
170 struct nf_bridge_info {
173 BRNF_PROTO_UNCHANGED,
181 struct net_device *physindev;
183 /* prerouting: detect dnat in orig/reply direction */
185 struct in6_addr ipv6_daddr;
187 /* after prerouting + nat detected: store original source
188 * mac since neigh resolution overwrites it, only used while
189 * skb is out in neigh layer.
191 char neigh_header[8];
193 /* always valid & non-NULL from FORWARD on, for physdev match */
194 struct net_device *physoutdev;
199 struct sk_buff_head {
200 /* These two members must be first. */
201 struct sk_buff *next;
202 struct sk_buff *prev;
210 /* To allow 64K frame to be packed as single skb without frag_list we
211 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
212 * buffers which do not start on a page boundary.
214 * Since GRO uses frags we allocate at least 16 regardless of page
217 #if (65536/PAGE_SIZE + 1) < 16
218 #define MAX_SKB_FRAGS 16UL
220 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
223 typedef struct skb_frag_struct skb_frag_t;
225 struct skb_frag_struct {
229 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
238 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
243 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
248 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
253 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
258 #define HAVE_HW_TIME_STAMP
261 * struct skb_shared_hwtstamps - hardware time stamps
262 * @hwtstamp: hardware time stamp transformed into duration
263 * since arbitrary point in time
265 * Software time stamps generated by ktime_get_real() are stored in
268 * hwtstamps can only be compared against other hwtstamps from
271 * This structure is attached to packets as part of the
272 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
274 struct skb_shared_hwtstamps {
278 /* Definitions for tx_flags in struct skb_shared_info */
280 /* generate hardware time stamp */
281 SKBTX_HW_TSTAMP = 1 << 0,
283 /* generate software time stamp when queueing packet to NIC */
284 SKBTX_SW_TSTAMP = 1 << 1,
286 /* device driver is going to provide hardware time stamp */
287 SKBTX_IN_PROGRESS = 1 << 2,
289 /* device driver supports TX zero-copy buffers */
290 SKBTX_DEV_ZEROCOPY = 1 << 3,
292 /* generate wifi status information (where possible) */
293 SKBTX_WIFI_STATUS = 1 << 4,
295 /* This indicates at least one fragment might be overwritten
296 * (as in vmsplice(), sendfile() ...)
297 * If we need to compute a TX checksum, we'll need to copy
298 * all frags to avoid possible bad checksum
300 SKBTX_SHARED_FRAG = 1 << 5,
302 /* generate software time stamp when entering packet scheduling */
303 SKBTX_SCHED_TSTAMP = 1 << 6,
305 /* generate software timestamp on peer data acknowledgment */
306 SKBTX_ACK_TSTAMP = 1 << 7,
309 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
310 SKBTX_SCHED_TSTAMP | \
312 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
315 * The callback notifies userspace to release buffers when skb DMA is done in
316 * lower device, the skb last reference should be 0 when calling this.
317 * The zerocopy_success argument is true if zero copy transmit occurred,
318 * false on data copy or out of memory error caused by data copy attempt.
319 * The ctx field is used to track device context.
320 * The desc field is used to track userspace buffer index.
323 void (*callback)(struct ubuf_info *, bool zerocopy_success);
328 /* This data is invariant across clones and lives at
329 * the end of the header data, ie. at skb->end.
331 struct skb_shared_info {
332 unsigned char nr_frags;
334 unsigned short gso_size;
335 /* Warning: this field is not always filled in (UFO)! */
336 unsigned short gso_segs;
337 unsigned short gso_type;
338 struct sk_buff *frag_list;
339 struct skb_shared_hwtstamps hwtstamps;
344 * Warning : all fields before dataref are cleared in __alloc_skb()
348 /* Intermediate layers must ensure that destructor_arg
349 * remains valid until skb destructor */
350 void * destructor_arg;
352 /* must be last field, see pskb_expand_head() */
353 skb_frag_t frags[MAX_SKB_FRAGS];
356 /* We divide dataref into two halves. The higher 16 bits hold references
357 * to the payload part of skb->data. The lower 16 bits hold references to
358 * the entire skb->data. A clone of a headerless skb holds the length of
359 * the header in skb->hdr_len.
361 * All users must obey the rule that the skb->data reference count must be
362 * greater than or equal to the payload reference count.
364 * Holding a reference to the payload part means that the user does not
365 * care about modifications to the header part of skb->data.
367 #define SKB_DATAREF_SHIFT 16
368 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
372 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
373 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
374 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
378 SKB_GSO_TCPV4 = 1 << 0,
379 SKB_GSO_UDP = 1 << 1,
381 /* This indicates the skb is from an untrusted source. */
382 SKB_GSO_DODGY = 1 << 2,
384 /* This indicates the tcp segment has CWR set. */
385 SKB_GSO_TCP_ECN = 1 << 3,
387 SKB_GSO_TCPV6 = 1 << 4,
389 SKB_GSO_FCOE = 1 << 5,
391 SKB_GSO_GRE = 1 << 6,
393 SKB_GSO_GRE_CSUM = 1 << 7,
395 SKB_GSO_IPIP = 1 << 8,
397 SKB_GSO_SIT = 1 << 9,
399 SKB_GSO_UDP_TUNNEL = 1 << 10,
401 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
403 SKB_GSO_TUNNEL_REMCSUM = 1 << 12,
406 #if BITS_PER_LONG > 32
407 #define NET_SKBUFF_DATA_USES_OFFSET 1
410 #ifdef NET_SKBUFF_DATA_USES_OFFSET
411 typedef unsigned int sk_buff_data_t;
413 typedef unsigned char *sk_buff_data_t;
417 * struct skb_mstamp - multi resolution time stamps
418 * @stamp_us: timestamp in us resolution
419 * @stamp_jiffies: timestamp in jiffies
432 * skb_mstamp_get - get current timestamp
433 * @cl: place to store timestamps
435 static inline void skb_mstamp_get(struct skb_mstamp *cl)
437 u64 val = local_clock();
439 do_div(val, NSEC_PER_USEC);
440 cl->stamp_us = (u32)val;
441 cl->stamp_jiffies = (u32)jiffies;
445 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
446 * @t1: pointer to newest sample
447 * @t0: pointer to oldest sample
449 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
450 const struct skb_mstamp *t0)
452 s32 delta_us = t1->stamp_us - t0->stamp_us;
453 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
455 /* If delta_us is negative, this might be because interval is too big,
456 * or local_clock() drift is too big : fallback using jiffies.
459 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
461 delta_us = jiffies_to_usecs(delta_jiffies);
468 * struct sk_buff - socket buffer
469 * @next: Next buffer in list
470 * @prev: Previous buffer in list
471 * @tstamp: Time we arrived/left
472 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
473 * @sk: Socket we are owned by
474 * @dev: Device we arrived on/are leaving by
475 * @cb: Control buffer. Free for use by every layer. Put private vars here
476 * @_skb_refdst: destination entry (with norefcount bit)
477 * @sp: the security path, used for xfrm
478 * @len: Length of actual data
479 * @data_len: Data length
480 * @mac_len: Length of link layer header
481 * @hdr_len: writable header length of cloned skb
482 * @csum: Checksum (must include start/offset pair)
483 * @csum_start: Offset from skb->head where checksumming should start
484 * @csum_offset: Offset from csum_start where checksum should be stored
485 * @priority: Packet queueing priority
486 * @ignore_df: allow local fragmentation
487 * @cloned: Head may be cloned (check refcnt to be sure)
488 * @ip_summed: Driver fed us an IP checksum
489 * @nohdr: Payload reference only, must not modify header
490 * @nfctinfo: Relationship of this skb to the connection
491 * @pkt_type: Packet class
492 * @fclone: skbuff clone status
493 * @ipvs_property: skbuff is owned by ipvs
494 * @peeked: this packet has been seen already, so stats have been
495 * done for it, don't do them again
496 * @nf_trace: netfilter packet trace flag
497 * @protocol: Packet protocol from driver
498 * @destructor: Destruct function
499 * @nfct: Associated connection, if any
500 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
501 * @skb_iif: ifindex of device we arrived on
502 * @tc_index: Traffic control index
503 * @tc_verd: traffic control verdict
504 * @hash: the packet hash
505 * @queue_mapping: Queue mapping for multiqueue devices
506 * @xmit_more: More SKBs are pending for this queue
507 * @ndisc_nodetype: router type (from link layer)
508 * @ooo_okay: allow the mapping of a socket to a queue to be changed
509 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
511 * @sw_hash: indicates hash was computed in software stack
512 * @wifi_acked_valid: wifi_acked was set
513 * @wifi_acked: whether frame was acked on wifi or not
514 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
515 * @napi_id: id of the NAPI struct this skb came from
516 * @secmark: security marking
517 * @offload_fwd_mark: fwding offload mark
518 * @mark: Generic packet mark
519 * @vlan_proto: vlan encapsulation protocol
520 * @vlan_tci: vlan tag control information
521 * @inner_protocol: Protocol (encapsulation)
522 * @inner_transport_header: Inner transport layer header (encapsulation)
523 * @inner_network_header: Network layer header (encapsulation)
524 * @inner_mac_header: Link layer header (encapsulation)
525 * @transport_header: Transport layer header
526 * @network_header: Network layer header
527 * @mac_header: Link layer header
528 * @tail: Tail pointer
530 * @head: Head of buffer
531 * @data: Data head pointer
532 * @truesize: Buffer size
533 * @users: User count - see {datagram,tcp}.c
539 /* These two members must be first. */
540 struct sk_buff *next;
541 struct sk_buff *prev;
545 struct skb_mstamp skb_mstamp;
548 struct rb_node rbnode; /* used in netem & tcp stack */
551 struct net_device *dev;
554 * This is the control buffer. It is free to use for every
555 * layer. Please put your private variables there. If you
556 * want to keep them across layers you have to do a skb_clone()
557 * first. This is owned by whoever has the skb queued ATM.
559 char cb[48] __aligned(8);
561 unsigned long _skb_refdst;
562 void (*destructor)(struct sk_buff *skb);
566 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
567 struct nf_conntrack *nfct;
569 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
570 struct nf_bridge_info *nf_bridge;
577 /* Following fields are _not_ copied in __copy_skb_header()
578 * Note that queue_mapping is here mostly to fill a hole.
580 kmemcheck_bitfield_begin(flags1);
589 kmemcheck_bitfield_end(flags1);
591 /* fields enclosed in headers_start/headers_end are copied
592 * using a single memcpy() in __copy_skb_header()
595 __u32 headers_start[0];
598 /* if you move pkt_type around you also must adapt those constants */
599 #ifdef __BIG_ENDIAN_BITFIELD
600 #define PKT_TYPE_MAX (7 << 5)
602 #define PKT_TYPE_MAX 7
604 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
606 __u8 __pkt_type_offset[0];
617 __u8 wifi_acked_valid:1;
621 /* Indicates the inner headers are valid in the skbuff. */
622 __u8 encapsulation:1;
623 __u8 encap_hdr_csum:1;
625 __u8 csum_complete_sw:1;
629 #ifdef CONFIG_IPV6_NDISC_NODETYPE
630 __u8 ndisc_nodetype:2;
632 __u8 ipvs_property:1;
633 __u8 inner_protocol_type:1;
634 __u8 remcsum_offload:1;
635 /* 3 or 5 bit hole */
637 #ifdef CONFIG_NET_SCHED
638 __u16 tc_index; /* traffic control index */
639 #ifdef CONFIG_NET_CLS_ACT
640 __u16 tc_verd; /* traffic control verdict */
656 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
658 unsigned int napi_id;
659 unsigned int sender_cpu;
663 #ifdef CONFIG_NETWORK_SECMARK
666 #ifdef CONFIG_NET_SWITCHDEV
667 __u32 offload_fwd_mark;
673 __u32 reserved_tailroom;
677 __be16 inner_protocol;
681 __u16 inner_transport_header;
682 __u16 inner_network_header;
683 __u16 inner_mac_header;
686 __u16 transport_header;
687 __u16 network_header;
691 __u32 headers_end[0];
694 /* These elements must be at the end, see alloc_skb() for details. */
699 unsigned int truesize;
705 * Handling routines are only of interest to the kernel
707 #include <linux/slab.h>
710 #define SKB_ALLOC_FCLONE 0x01
711 #define SKB_ALLOC_RX 0x02
712 #define SKB_ALLOC_NAPI 0x04
714 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
715 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
717 return unlikely(skb->pfmemalloc);
721 * skb might have a dst pointer attached, refcounted or not.
722 * _skb_refdst low order bit is set if refcount was _not_ taken
724 #define SKB_DST_NOREF 1UL
725 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
728 * skb_dst - returns skb dst_entry
731 * Returns skb dst_entry, regardless of reference taken or not.
733 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
735 /* If refdst was not refcounted, check we still are in a
736 * rcu_read_lock section
738 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
739 !rcu_read_lock_held() &&
740 !rcu_read_lock_bh_held());
741 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
745 * skb_dst_set - sets skb dst
749 * Sets skb dst, assuming a reference was taken on dst and should
750 * be released by skb_dst_drop()
752 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
754 skb->_skb_refdst = (unsigned long)dst;
758 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
762 * Sets skb dst, assuming a reference was not taken on dst.
763 * If dst entry is cached, we do not take reference and dst_release
764 * will be avoided by refdst_drop. If dst entry is not cached, we take
765 * reference, so that last dst_release can destroy the dst immediately.
767 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
769 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
770 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
774 * skb_dst_is_noref - Test if skb dst isn't refcounted
777 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
779 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
782 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
784 return (struct rtable *)skb_dst(skb);
787 void kfree_skb(struct sk_buff *skb);
788 void kfree_skb_list(struct sk_buff *segs);
789 void skb_tx_error(struct sk_buff *skb);
790 void consume_skb(struct sk_buff *skb);
791 void __kfree_skb(struct sk_buff *skb);
792 extern struct kmem_cache *skbuff_head_cache;
794 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
795 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
796 bool *fragstolen, int *delta_truesize);
798 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
800 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
801 struct sk_buff *build_skb(void *data, unsigned int frag_size);
802 static inline struct sk_buff *alloc_skb(unsigned int size,
805 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
808 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
809 unsigned long data_len,
814 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
815 struct sk_buff_fclones {
824 * skb_fclone_busy - check if fclone is busy
827 * Returns true is skb is a fast clone, and its clone is not freed.
828 * Some drivers call skb_orphan() in their ndo_start_xmit(),
829 * so we also check that this didnt happen.
831 static inline bool skb_fclone_busy(const struct sock *sk,
832 const struct sk_buff *skb)
834 const struct sk_buff_fclones *fclones;
836 fclones = container_of(skb, struct sk_buff_fclones, skb1);
838 return skb->fclone == SKB_FCLONE_ORIG &&
839 atomic_read(&fclones->fclone_ref) > 1 &&
840 fclones->skb2.sk == sk;
843 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
846 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
849 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
850 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
852 return __alloc_skb_head(priority, -1);
855 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
856 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
857 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
858 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
859 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
860 gfp_t gfp_mask, bool fclone);
861 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
864 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
867 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
868 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
869 unsigned int headroom);
870 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
871 int newtailroom, gfp_t priority);
872 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
873 int offset, int len);
874 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
876 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
877 int skb_pad(struct sk_buff *skb, int pad);
878 #define dev_kfree_skb(a) consume_skb(a)
880 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
881 int getfrag(void *from, char *to, int offset,
882 int len, int odd, struct sk_buff *skb),
883 void *from, int length);
885 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
886 int offset, size_t size);
888 struct skb_seq_state {
892 __u32 stepped_offset;
893 struct sk_buff *root_skb;
894 struct sk_buff *cur_skb;
898 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
899 unsigned int to, struct skb_seq_state *st);
900 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
901 struct skb_seq_state *st);
902 void skb_abort_seq_read(struct skb_seq_state *st);
904 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
905 unsigned int to, struct ts_config *config);
908 * Packet hash types specify the type of hash in skb_set_hash.
910 * Hash types refer to the protocol layer addresses which are used to
911 * construct a packet's hash. The hashes are used to differentiate or identify
912 * flows of the protocol layer for the hash type. Hash types are either
913 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
915 * Properties of hashes:
917 * 1) Two packets in different flows have different hash values
918 * 2) Two packets in the same flow should have the same hash value
920 * A hash at a higher layer is considered to be more specific. A driver should
921 * set the most specific hash possible.
923 * A driver cannot indicate a more specific hash than the layer at which a hash
924 * was computed. For instance an L3 hash cannot be set as an L4 hash.
926 * A driver may indicate a hash level which is less specific than the
927 * actual layer the hash was computed on. For instance, a hash computed
928 * at L4 may be considered an L3 hash. This should only be done if the
929 * driver can't unambiguously determine that the HW computed the hash at
930 * the higher layer. Note that the "should" in the second property above
933 enum pkt_hash_types {
934 PKT_HASH_TYPE_NONE, /* Undefined type */
935 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
936 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
937 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
941 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
943 skb->l4_hash = (type == PKT_HASH_TYPE_L4);
948 static inline __u32 skb_get_hash(struct sk_buff *skb)
950 if (!skb->l4_hash && !skb->sw_hash)
956 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, struct flowi6 *fl6);
958 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, struct flowi6 *fl6)
960 if (!skb->l4_hash && !skb->sw_hash)
961 __skb_get_hash_flowi6(skb, fl6);
966 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, struct flowi4 *fl);
968 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, struct flowi4 *fl4)
970 if (!skb->l4_hash && !skb->sw_hash)
971 __skb_get_hash_flowi4(skb, fl4);
976 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
978 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
983 static inline void skb_clear_hash(struct sk_buff *skb)
990 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
996 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
998 to->hash = from->hash;
999 to->sw_hash = from->sw_hash;
1000 to->l4_hash = from->l4_hash;
1003 static inline void skb_sender_cpu_clear(struct sk_buff *skb)
1006 skb->sender_cpu = 0;
1010 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1011 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1013 return skb->head + skb->end;
1016 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1021 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1026 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1028 return skb->end - skb->head;
1033 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1035 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1037 return &skb_shinfo(skb)->hwtstamps;
1041 * skb_queue_empty - check if a queue is empty
1044 * Returns true if the queue is empty, false otherwise.
1046 static inline int skb_queue_empty(const struct sk_buff_head *list)
1048 return list->next == (const struct sk_buff *) list;
1052 * skb_queue_is_last - check if skb is the last entry in the queue
1056 * Returns true if @skb is the last buffer on the list.
1058 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1059 const struct sk_buff *skb)
1061 return skb->next == (const struct sk_buff *) list;
1065 * skb_queue_is_first - check if skb is the first entry in the queue
1069 * Returns true if @skb is the first buffer on the list.
1071 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1072 const struct sk_buff *skb)
1074 return skb->prev == (const struct sk_buff *) list;
1078 * skb_queue_next - return the next packet in the queue
1080 * @skb: current buffer
1082 * Return the next packet in @list after @skb. It is only valid to
1083 * call this if skb_queue_is_last() evaluates to false.
1085 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1086 const struct sk_buff *skb)
1088 /* This BUG_ON may seem severe, but if we just return then we
1089 * are going to dereference garbage.
1091 BUG_ON(skb_queue_is_last(list, skb));
1096 * skb_queue_prev - return the prev packet in the queue
1098 * @skb: current buffer
1100 * Return the prev packet in @list before @skb. It is only valid to
1101 * call this if skb_queue_is_first() evaluates to false.
1103 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1104 const struct sk_buff *skb)
1106 /* This BUG_ON may seem severe, but if we just return then we
1107 * are going to dereference garbage.
1109 BUG_ON(skb_queue_is_first(list, skb));
1114 * skb_get - reference buffer
1115 * @skb: buffer to reference
1117 * Makes another reference to a socket buffer and returns a pointer
1120 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1122 atomic_inc(&skb->users);
1127 * If users == 1, we are the only owner and are can avoid redundant
1132 * skb_cloned - is the buffer a clone
1133 * @skb: buffer to check
1135 * Returns true if the buffer was generated with skb_clone() and is
1136 * one of multiple shared copies of the buffer. Cloned buffers are
1137 * shared data so must not be written to under normal circumstances.
1139 static inline int skb_cloned(const struct sk_buff *skb)
1141 return skb->cloned &&
1142 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1145 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1147 might_sleep_if(pri & __GFP_WAIT);
1149 if (skb_cloned(skb))
1150 return pskb_expand_head(skb, 0, 0, pri);
1156 * skb_header_cloned - is the header a clone
1157 * @skb: buffer to check
1159 * Returns true if modifying the header part of the buffer requires
1160 * the data to be copied.
1162 static inline int skb_header_cloned(const struct sk_buff *skb)
1169 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1170 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1171 return dataref != 1;
1175 * skb_header_release - release reference to header
1176 * @skb: buffer to operate on
1178 * Drop a reference to the header part of the buffer. This is done
1179 * by acquiring a payload reference. You must not read from the header
1180 * part of skb->data after this.
1181 * Note : Check if you can use __skb_header_release() instead.
1183 static inline void skb_header_release(struct sk_buff *skb)
1187 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1191 * __skb_header_release - release reference to header
1192 * @skb: buffer to operate on
1194 * Variant of skb_header_release() assuming skb is private to caller.
1195 * We can avoid one atomic operation.
1197 static inline void __skb_header_release(struct sk_buff *skb)
1200 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1205 * skb_shared - is the buffer shared
1206 * @skb: buffer to check
1208 * Returns true if more than one person has a reference to this
1211 static inline int skb_shared(const struct sk_buff *skb)
1213 return atomic_read(&skb->users) != 1;
1217 * skb_share_check - check if buffer is shared and if so clone it
1218 * @skb: buffer to check
1219 * @pri: priority for memory allocation
1221 * If the buffer is shared the buffer is cloned and the old copy
1222 * drops a reference. A new clone with a single reference is returned.
1223 * If the buffer is not shared the original buffer is returned. When
1224 * being called from interrupt status or with spinlocks held pri must
1227 * NULL is returned on a memory allocation failure.
1229 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1231 might_sleep_if(pri & __GFP_WAIT);
1232 if (skb_shared(skb)) {
1233 struct sk_buff *nskb = skb_clone(skb, pri);
1245 * Copy shared buffers into a new sk_buff. We effectively do COW on
1246 * packets to handle cases where we have a local reader and forward
1247 * and a couple of other messy ones. The normal one is tcpdumping
1248 * a packet thats being forwarded.
1252 * skb_unshare - make a copy of a shared buffer
1253 * @skb: buffer to check
1254 * @pri: priority for memory allocation
1256 * If the socket buffer is a clone then this function creates a new
1257 * copy of the data, drops a reference count on the old copy and returns
1258 * the new copy with the reference count at 1. If the buffer is not a clone
1259 * the original buffer is returned. When called with a spinlock held or
1260 * from interrupt state @pri must be %GFP_ATOMIC
1262 * %NULL is returned on a memory allocation failure.
1264 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1267 might_sleep_if(pri & __GFP_WAIT);
1268 if (skb_cloned(skb)) {
1269 struct sk_buff *nskb = skb_copy(skb, pri);
1271 /* Free our shared copy */
1282 * skb_peek - peek at the head of an &sk_buff_head
1283 * @list_: list to peek at
1285 * Peek an &sk_buff. Unlike most other operations you _MUST_
1286 * be careful with this one. A peek leaves the buffer on the
1287 * list and someone else may run off with it. You must hold
1288 * the appropriate locks or have a private queue to do this.
1290 * Returns %NULL for an empty list or a pointer to the head element.
1291 * The reference count is not incremented and the reference is therefore
1292 * volatile. Use with caution.
1294 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1296 struct sk_buff *skb = list_->next;
1298 if (skb == (struct sk_buff *)list_)
1304 * skb_peek_next - peek skb following the given one from a queue
1305 * @skb: skb to start from
1306 * @list_: list to peek at
1308 * Returns %NULL when the end of the list is met or a pointer to the
1309 * next element. The reference count is not incremented and the
1310 * reference is therefore volatile. Use with caution.
1312 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1313 const struct sk_buff_head *list_)
1315 struct sk_buff *next = skb->next;
1317 if (next == (struct sk_buff *)list_)
1323 * skb_peek_tail - peek at the tail of an &sk_buff_head
1324 * @list_: list to peek at
1326 * Peek an &sk_buff. Unlike most other operations you _MUST_
1327 * be careful with this one. A peek leaves the buffer on the
1328 * list and someone else may run off with it. You must hold
1329 * the appropriate locks or have a private queue to do this.
1331 * Returns %NULL for an empty list or a pointer to the tail element.
1332 * The reference count is not incremented and the reference is therefore
1333 * volatile. Use with caution.
1335 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1337 struct sk_buff *skb = list_->prev;
1339 if (skb == (struct sk_buff *)list_)
1346 * skb_queue_len - get queue length
1347 * @list_: list to measure
1349 * Return the length of an &sk_buff queue.
1351 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1357 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1358 * @list: queue to initialize
1360 * This initializes only the list and queue length aspects of
1361 * an sk_buff_head object. This allows to initialize the list
1362 * aspects of an sk_buff_head without reinitializing things like
1363 * the spinlock. It can also be used for on-stack sk_buff_head
1364 * objects where the spinlock is known to not be used.
1366 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1368 list->prev = list->next = (struct sk_buff *)list;
1373 * This function creates a split out lock class for each invocation;
1374 * this is needed for now since a whole lot of users of the skb-queue
1375 * infrastructure in drivers have different locking usage (in hardirq)
1376 * than the networking core (in softirq only). In the long run either the
1377 * network layer or drivers should need annotation to consolidate the
1378 * main types of usage into 3 classes.
1380 static inline void skb_queue_head_init(struct sk_buff_head *list)
1382 spin_lock_init(&list->lock);
1383 __skb_queue_head_init(list);
1386 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1387 struct lock_class_key *class)
1389 skb_queue_head_init(list);
1390 lockdep_set_class(&list->lock, class);
1394 * Insert an sk_buff on a list.
1396 * The "__skb_xxxx()" functions are the non-atomic ones that
1397 * can only be called with interrupts disabled.
1399 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1400 struct sk_buff_head *list);
1401 static inline void __skb_insert(struct sk_buff *newsk,
1402 struct sk_buff *prev, struct sk_buff *next,
1403 struct sk_buff_head *list)
1407 next->prev = prev->next = newsk;
1411 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1412 struct sk_buff *prev,
1413 struct sk_buff *next)
1415 struct sk_buff *first = list->next;
1416 struct sk_buff *last = list->prev;
1426 * skb_queue_splice - join two skb lists, this is designed for stacks
1427 * @list: the new list to add
1428 * @head: the place to add it in the first list
1430 static inline void skb_queue_splice(const struct sk_buff_head *list,
1431 struct sk_buff_head *head)
1433 if (!skb_queue_empty(list)) {
1434 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1435 head->qlen += list->qlen;
1440 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1441 * @list: the new list to add
1442 * @head: the place to add it in the first list
1444 * The list at @list is reinitialised
1446 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1447 struct sk_buff_head *head)
1449 if (!skb_queue_empty(list)) {
1450 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1451 head->qlen += list->qlen;
1452 __skb_queue_head_init(list);
1457 * skb_queue_splice_tail - join two skb lists, each list being a queue
1458 * @list: the new list to add
1459 * @head: the place to add it in the first list
1461 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1462 struct sk_buff_head *head)
1464 if (!skb_queue_empty(list)) {
1465 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1466 head->qlen += list->qlen;
1471 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1472 * @list: the new list to add
1473 * @head: the place to add it in the first list
1475 * Each of the lists is a queue.
1476 * The list at @list is reinitialised
1478 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1479 struct sk_buff_head *head)
1481 if (!skb_queue_empty(list)) {
1482 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1483 head->qlen += list->qlen;
1484 __skb_queue_head_init(list);
1489 * __skb_queue_after - queue a buffer at the list head
1490 * @list: list to use
1491 * @prev: place after this buffer
1492 * @newsk: buffer to queue
1494 * Queue a buffer int the middle of a list. This function takes no locks
1495 * and you must therefore hold required locks before calling it.
1497 * A buffer cannot be placed on two lists at the same time.
1499 static inline void __skb_queue_after(struct sk_buff_head *list,
1500 struct sk_buff *prev,
1501 struct sk_buff *newsk)
1503 __skb_insert(newsk, prev, prev->next, list);
1506 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1507 struct sk_buff_head *list);
1509 static inline void __skb_queue_before(struct sk_buff_head *list,
1510 struct sk_buff *next,
1511 struct sk_buff *newsk)
1513 __skb_insert(newsk, next->prev, next, list);
1517 * __skb_queue_head - queue a buffer at the list head
1518 * @list: list to use
1519 * @newsk: buffer to queue
1521 * Queue a buffer at the start of a list. This function takes no locks
1522 * and you must therefore hold required locks before calling it.
1524 * A buffer cannot be placed on two lists at the same time.
1526 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1527 static inline void __skb_queue_head(struct sk_buff_head *list,
1528 struct sk_buff *newsk)
1530 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1534 * __skb_queue_tail - queue a buffer at the list tail
1535 * @list: list to use
1536 * @newsk: buffer to queue
1538 * Queue a buffer at the end of a list. This function takes no locks
1539 * and you must therefore hold required locks before calling it.
1541 * A buffer cannot be placed on two lists at the same time.
1543 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1544 static inline void __skb_queue_tail(struct sk_buff_head *list,
1545 struct sk_buff *newsk)
1547 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1551 * remove sk_buff from list. _Must_ be called atomically, and with
1554 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1555 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1557 struct sk_buff *next, *prev;
1562 skb->next = skb->prev = NULL;
1568 * __skb_dequeue - remove from the head of the queue
1569 * @list: list to dequeue from
1571 * Remove the head of the list. This function does not take any locks
1572 * so must be used with appropriate locks held only. The head item is
1573 * returned or %NULL if the list is empty.
1575 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1576 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1578 struct sk_buff *skb = skb_peek(list);
1580 __skb_unlink(skb, list);
1585 * __skb_dequeue_tail - remove from the tail of the queue
1586 * @list: list to dequeue from
1588 * Remove the tail of the list. This function does not take any locks
1589 * so must be used with appropriate locks held only. The tail item is
1590 * returned or %NULL if the list is empty.
1592 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1593 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1595 struct sk_buff *skb = skb_peek_tail(list);
1597 __skb_unlink(skb, list);
1602 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1604 return skb->data_len;
1607 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1609 return skb->len - skb->data_len;
1612 static inline int skb_pagelen(const struct sk_buff *skb)
1616 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1617 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1618 return len + skb_headlen(skb);
1622 * __skb_fill_page_desc - initialise a paged fragment in an skb
1623 * @skb: buffer containing fragment to be initialised
1624 * @i: paged fragment index to initialise
1625 * @page: the page to use for this fragment
1626 * @off: the offset to the data with @page
1627 * @size: the length of the data
1629 * Initialises the @i'th fragment of @skb to point to &size bytes at
1630 * offset @off within @page.
1632 * Does not take any additional reference on the fragment.
1634 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1635 struct page *page, int off, int size)
1637 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1640 * Propagate page->pfmemalloc to the skb if we can. The problem is
1641 * that not all callers have unique ownership of the page. If
1642 * pfmemalloc is set, we check the mapping as a mapping implies
1643 * page->index is set (index and pfmemalloc share space).
1644 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1645 * do not lose pfmemalloc information as the pages would not be
1646 * allocated using __GFP_MEMALLOC.
1648 frag->page.p = page;
1649 frag->page_offset = off;
1650 skb_frag_size_set(frag, size);
1652 page = compound_head(page);
1653 if (page->pfmemalloc && !page->mapping)
1654 skb->pfmemalloc = true;
1658 * skb_fill_page_desc - initialise a paged fragment in an skb
1659 * @skb: buffer containing fragment to be initialised
1660 * @i: paged fragment index to initialise
1661 * @page: the page to use for this fragment
1662 * @off: the offset to the data with @page
1663 * @size: the length of the data
1665 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1666 * @skb to point to @size bytes at offset @off within @page. In
1667 * addition updates @skb such that @i is the last fragment.
1669 * Does not take any additional reference on the fragment.
1671 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1672 struct page *page, int off, int size)
1674 __skb_fill_page_desc(skb, i, page, off, size);
1675 skb_shinfo(skb)->nr_frags = i + 1;
1678 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1679 int size, unsigned int truesize);
1681 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1682 unsigned int truesize);
1684 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1685 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1686 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1688 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1689 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1691 return skb->head + skb->tail;
1694 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1696 skb->tail = skb->data - skb->head;
1699 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1701 skb_reset_tail_pointer(skb);
1702 skb->tail += offset;
1705 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1706 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1711 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1713 skb->tail = skb->data;
1716 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1718 skb->tail = skb->data + offset;
1721 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1724 * Add data to an sk_buff
1726 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1727 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1728 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1730 unsigned char *tmp = skb_tail_pointer(skb);
1731 SKB_LINEAR_ASSERT(skb);
1737 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1738 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1745 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1746 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1749 BUG_ON(skb->len < skb->data_len);
1750 return skb->data += len;
1753 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1755 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1758 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1760 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1762 if (len > skb_headlen(skb) &&
1763 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1766 return skb->data += len;
1769 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1771 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1774 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1776 if (likely(len <= skb_headlen(skb)))
1778 if (unlikely(len > skb->len))
1780 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1784 * skb_headroom - bytes at buffer head
1785 * @skb: buffer to check
1787 * Return the number of bytes of free space at the head of an &sk_buff.
1789 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1791 return skb->data - skb->head;
1795 * skb_tailroom - bytes at buffer end
1796 * @skb: buffer to check
1798 * Return the number of bytes of free space at the tail of an sk_buff
1800 static inline int skb_tailroom(const struct sk_buff *skb)
1802 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1806 * skb_availroom - bytes at buffer end
1807 * @skb: buffer to check
1809 * Return the number of bytes of free space at the tail of an sk_buff
1810 * allocated by sk_stream_alloc()
1812 static inline int skb_availroom(const struct sk_buff *skb)
1814 if (skb_is_nonlinear(skb))
1817 return skb->end - skb->tail - skb->reserved_tailroom;
1821 * skb_reserve - adjust headroom
1822 * @skb: buffer to alter
1823 * @len: bytes to move
1825 * Increase the headroom of an empty &sk_buff by reducing the tail
1826 * room. This is only allowed for an empty buffer.
1828 static inline void skb_reserve(struct sk_buff *skb, int len)
1834 #define ENCAP_TYPE_ETHER 0
1835 #define ENCAP_TYPE_IPPROTO 1
1837 static inline void skb_set_inner_protocol(struct sk_buff *skb,
1840 skb->inner_protocol = protocol;
1841 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
1844 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
1847 skb->inner_ipproto = ipproto;
1848 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
1851 static inline void skb_reset_inner_headers(struct sk_buff *skb)
1853 skb->inner_mac_header = skb->mac_header;
1854 skb->inner_network_header = skb->network_header;
1855 skb->inner_transport_header = skb->transport_header;
1858 static inline void skb_reset_mac_len(struct sk_buff *skb)
1860 skb->mac_len = skb->network_header - skb->mac_header;
1863 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1866 return skb->head + skb->inner_transport_header;
1869 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1871 skb->inner_transport_header = skb->data - skb->head;
1874 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1877 skb_reset_inner_transport_header(skb);
1878 skb->inner_transport_header += offset;
1881 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1883 return skb->head + skb->inner_network_header;
1886 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1888 skb->inner_network_header = skb->data - skb->head;
1891 static inline void skb_set_inner_network_header(struct sk_buff *skb,
1894 skb_reset_inner_network_header(skb);
1895 skb->inner_network_header += offset;
1898 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1900 return skb->head + skb->inner_mac_header;
1903 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1905 skb->inner_mac_header = skb->data - skb->head;
1908 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1911 skb_reset_inner_mac_header(skb);
1912 skb->inner_mac_header += offset;
1914 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1916 return skb->transport_header != (typeof(skb->transport_header))~0U;
1919 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1921 return skb->head + skb->transport_header;
1924 static inline void skb_reset_transport_header(struct sk_buff *skb)
1926 skb->transport_header = skb->data - skb->head;
1929 static inline void skb_set_transport_header(struct sk_buff *skb,
1932 skb_reset_transport_header(skb);
1933 skb->transport_header += offset;
1936 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1938 return skb->head + skb->network_header;
1941 static inline void skb_reset_network_header(struct sk_buff *skb)
1943 skb->network_header = skb->data - skb->head;
1946 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1948 skb_reset_network_header(skb);
1949 skb->network_header += offset;
1952 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1954 return skb->head + skb->mac_header;
1957 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1959 return skb->mac_header != (typeof(skb->mac_header))~0U;
1962 static inline void skb_reset_mac_header(struct sk_buff *skb)
1964 skb->mac_header = skb->data - skb->head;
1967 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1969 skb_reset_mac_header(skb);
1970 skb->mac_header += offset;
1973 static inline void skb_pop_mac_header(struct sk_buff *skb)
1975 skb->mac_header = skb->network_header;
1978 static inline void skb_probe_transport_header(struct sk_buff *skb,
1979 const int offset_hint)
1981 struct flow_keys keys;
1983 if (skb_transport_header_was_set(skb))
1985 else if (skb_flow_dissect_flow_keys(skb, &keys))
1986 skb_set_transport_header(skb, keys.control.thoff);
1988 skb_set_transport_header(skb, offset_hint);
1991 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1993 if (skb_mac_header_was_set(skb)) {
1994 const unsigned char *old_mac = skb_mac_header(skb);
1996 skb_set_mac_header(skb, -skb->mac_len);
1997 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2001 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2003 return skb->csum_start - skb_headroom(skb);
2006 static inline int skb_transport_offset(const struct sk_buff *skb)
2008 return skb_transport_header(skb) - skb->data;
2011 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2013 return skb->transport_header - skb->network_header;
2016 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2018 return skb->inner_transport_header - skb->inner_network_header;
2021 static inline int skb_network_offset(const struct sk_buff *skb)
2023 return skb_network_header(skb) - skb->data;
2026 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2028 return skb_inner_network_header(skb) - skb->data;
2031 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2033 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2037 * CPUs often take a performance hit when accessing unaligned memory
2038 * locations. The actual performance hit varies, it can be small if the
2039 * hardware handles it or large if we have to take an exception and fix it
2042 * Since an ethernet header is 14 bytes network drivers often end up with
2043 * the IP header at an unaligned offset. The IP header can be aligned by
2044 * shifting the start of the packet by 2 bytes. Drivers should do this
2047 * skb_reserve(skb, NET_IP_ALIGN);
2049 * The downside to this alignment of the IP header is that the DMA is now
2050 * unaligned. On some architectures the cost of an unaligned DMA is high
2051 * and this cost outweighs the gains made by aligning the IP header.
2053 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2056 #ifndef NET_IP_ALIGN
2057 #define NET_IP_ALIGN 2
2061 * The networking layer reserves some headroom in skb data (via
2062 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2063 * the header has to grow. In the default case, if the header has to grow
2064 * 32 bytes or less we avoid the reallocation.
2066 * Unfortunately this headroom changes the DMA alignment of the resulting
2067 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2068 * on some architectures. An architecture can override this value,
2069 * perhaps setting it to a cacheline in size (since that will maintain
2070 * cacheline alignment of the DMA). It must be a power of 2.
2072 * Various parts of the networking layer expect at least 32 bytes of
2073 * headroom, you should not reduce this.
2075 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2076 * to reduce average number of cache lines per packet.
2077 * get_rps_cpus() for example only access one 64 bytes aligned block :
2078 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2081 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2084 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2086 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2088 if (unlikely(skb_is_nonlinear(skb))) {
2093 skb_set_tail_pointer(skb, len);
2096 void skb_trim(struct sk_buff *skb, unsigned int len);
2098 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2101 return ___pskb_trim(skb, len);
2102 __skb_trim(skb, len);
2106 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2108 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2112 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2113 * @skb: buffer to alter
2116 * This is identical to pskb_trim except that the caller knows that
2117 * the skb is not cloned so we should never get an error due to out-
2120 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2122 int err = pskb_trim(skb, len);
2127 * skb_orphan - orphan a buffer
2128 * @skb: buffer to orphan
2130 * If a buffer currently has an owner then we call the owner's
2131 * destructor function and make the @skb unowned. The buffer continues
2132 * to exist but is no longer charged to its former owner.
2134 static inline void skb_orphan(struct sk_buff *skb)
2136 if (skb->destructor) {
2137 skb->destructor(skb);
2138 skb->destructor = NULL;
2146 * skb_orphan_frags - orphan the frags contained in a buffer
2147 * @skb: buffer to orphan frags from
2148 * @gfp_mask: allocation mask for replacement pages
2150 * For each frag in the SKB which needs a destructor (i.e. has an
2151 * owner) create a copy of that frag and release the original
2152 * page by calling the destructor.
2154 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2156 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2158 return skb_copy_ubufs(skb, gfp_mask);
2162 * __skb_queue_purge - empty a list
2163 * @list: list to empty
2165 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2166 * the list and one reference dropped. This function does not take the
2167 * list lock and the caller must hold the relevant locks to use it.
2169 void skb_queue_purge(struct sk_buff_head *list);
2170 static inline void __skb_queue_purge(struct sk_buff_head *list)
2172 struct sk_buff *skb;
2173 while ((skb = __skb_dequeue(list)) != NULL)
2177 void *netdev_alloc_frag(unsigned int fragsz);
2179 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2183 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2184 * @dev: network device to receive on
2185 * @length: length to allocate
2187 * Allocate a new &sk_buff and assign it a usage count of one. The
2188 * buffer has unspecified headroom built in. Users should allocate
2189 * the headroom they think they need without accounting for the
2190 * built in space. The built in space is used for optimisations.
2192 * %NULL is returned if there is no free memory. Although this function
2193 * allocates memory it can be called from an interrupt.
2195 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2196 unsigned int length)
2198 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2201 /* legacy helper around __netdev_alloc_skb() */
2202 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2205 return __netdev_alloc_skb(NULL, length, gfp_mask);
2208 /* legacy helper around netdev_alloc_skb() */
2209 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2211 return netdev_alloc_skb(NULL, length);
2215 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2216 unsigned int length, gfp_t gfp)
2218 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2220 if (NET_IP_ALIGN && skb)
2221 skb_reserve(skb, NET_IP_ALIGN);
2225 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2226 unsigned int length)
2228 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2231 static inline void skb_free_frag(void *addr)
2233 __free_page_frag(addr);
2236 void *napi_alloc_frag(unsigned int fragsz);
2237 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2238 unsigned int length, gfp_t gfp_mask);
2239 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2240 unsigned int length)
2242 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2246 * __dev_alloc_pages - allocate page for network Rx
2247 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2248 * @order: size of the allocation
2250 * Allocate a new page.
2252 * %NULL is returned if there is no free memory.
2254 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2257 /* This piece of code contains several assumptions.
2258 * 1. This is for device Rx, therefor a cold page is preferred.
2259 * 2. The expectation is the user wants a compound page.
2260 * 3. If requesting a order 0 page it will not be compound
2261 * due to the check to see if order has a value in prep_new_page
2262 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2263 * code in gfp_to_alloc_flags that should be enforcing this.
2265 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2267 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2270 static inline struct page *dev_alloc_pages(unsigned int order)
2272 return __dev_alloc_pages(GFP_ATOMIC, order);
2276 * __dev_alloc_page - allocate a page for network Rx
2277 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2279 * Allocate a new page.
2281 * %NULL is returned if there is no free memory.
2283 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2285 return __dev_alloc_pages(gfp_mask, 0);
2288 static inline struct page *dev_alloc_page(void)
2290 return __dev_alloc_page(GFP_ATOMIC);
2294 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2295 * @page: The page that was allocated from skb_alloc_page
2296 * @skb: The skb that may need pfmemalloc set
2298 static inline void skb_propagate_pfmemalloc(struct page *page,
2299 struct sk_buff *skb)
2301 if (page && page->pfmemalloc)
2302 skb->pfmemalloc = true;
2306 * skb_frag_page - retrieve the page referred to by a paged fragment
2307 * @frag: the paged fragment
2309 * Returns the &struct page associated with @frag.
2311 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2313 return frag->page.p;
2317 * __skb_frag_ref - take an addition reference on a paged fragment.
2318 * @frag: the paged fragment
2320 * Takes an additional reference on the paged fragment @frag.
2322 static inline void __skb_frag_ref(skb_frag_t *frag)
2324 get_page(skb_frag_page(frag));
2328 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2330 * @f: the fragment offset.
2332 * Takes an additional reference on the @f'th paged fragment of @skb.
2334 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2336 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2340 * __skb_frag_unref - release a reference on a paged fragment.
2341 * @frag: the paged fragment
2343 * Releases a reference on the paged fragment @frag.
2345 static inline void __skb_frag_unref(skb_frag_t *frag)
2347 put_page(skb_frag_page(frag));
2351 * skb_frag_unref - release a reference on a paged fragment of an skb.
2353 * @f: the fragment offset
2355 * Releases a reference on the @f'th paged fragment of @skb.
2357 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2359 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2363 * skb_frag_address - gets the address of the data contained in a paged fragment
2364 * @frag: the paged fragment buffer
2366 * Returns the address of the data within @frag. The page must already
2369 static inline void *skb_frag_address(const skb_frag_t *frag)
2371 return page_address(skb_frag_page(frag)) + frag->page_offset;
2375 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2376 * @frag: the paged fragment buffer
2378 * Returns the address of the data within @frag. Checks that the page
2379 * is mapped and returns %NULL otherwise.
2381 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2383 void *ptr = page_address(skb_frag_page(frag));
2387 return ptr + frag->page_offset;
2391 * __skb_frag_set_page - sets the page contained in a paged fragment
2392 * @frag: the paged fragment
2393 * @page: the page to set
2395 * Sets the fragment @frag to contain @page.
2397 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2399 frag->page.p = page;
2403 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2405 * @f: the fragment offset
2406 * @page: the page to set
2408 * Sets the @f'th fragment of @skb to contain @page.
2410 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2413 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2416 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2419 * skb_frag_dma_map - maps a paged fragment via the DMA API
2420 * @dev: the device to map the fragment to
2421 * @frag: the paged fragment to map
2422 * @offset: the offset within the fragment (starting at the
2423 * fragment's own offset)
2424 * @size: the number of bytes to map
2425 * @dir: the direction of the mapping (%PCI_DMA_*)
2427 * Maps the page associated with @frag to @device.
2429 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2430 const skb_frag_t *frag,
2431 size_t offset, size_t size,
2432 enum dma_data_direction dir)
2434 return dma_map_page(dev, skb_frag_page(frag),
2435 frag->page_offset + offset, size, dir);
2438 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2441 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2445 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2448 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2453 * skb_clone_writable - is the header of a clone writable
2454 * @skb: buffer to check
2455 * @len: length up to which to write
2457 * Returns true if modifying the header part of the cloned buffer
2458 * does not requires the data to be copied.
2460 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2462 return !skb_header_cloned(skb) &&
2463 skb_headroom(skb) + len <= skb->hdr_len;
2466 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2471 if (headroom > skb_headroom(skb))
2472 delta = headroom - skb_headroom(skb);
2474 if (delta || cloned)
2475 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2481 * skb_cow - copy header of skb when it is required
2482 * @skb: buffer to cow
2483 * @headroom: needed headroom
2485 * If the skb passed lacks sufficient headroom or its data part
2486 * is shared, data is reallocated. If reallocation fails, an error
2487 * is returned and original skb is not changed.
2489 * The result is skb with writable area skb->head...skb->tail
2490 * and at least @headroom of space at head.
2492 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2494 return __skb_cow(skb, headroom, skb_cloned(skb));
2498 * skb_cow_head - skb_cow but only making the head writable
2499 * @skb: buffer to cow
2500 * @headroom: needed headroom
2502 * This function is identical to skb_cow except that we replace the
2503 * skb_cloned check by skb_header_cloned. It should be used when
2504 * you only need to push on some header and do not need to modify
2507 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2509 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2513 * skb_padto - pad an skbuff up to a minimal size
2514 * @skb: buffer to pad
2515 * @len: minimal length
2517 * Pads up a buffer to ensure the trailing bytes exist and are
2518 * blanked. If the buffer already contains sufficient data it
2519 * is untouched. Otherwise it is extended. Returns zero on
2520 * success. The skb is freed on error.
2522 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2524 unsigned int size = skb->len;
2525 if (likely(size >= len))
2527 return skb_pad(skb, len - size);
2531 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2532 * @skb: buffer to pad
2533 * @len: minimal length
2535 * Pads up a buffer to ensure the trailing bytes exist and are
2536 * blanked. If the buffer already contains sufficient data it
2537 * is untouched. Otherwise it is extended. Returns zero on
2538 * success. The skb is freed on error.
2540 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2542 unsigned int size = skb->len;
2544 if (unlikely(size < len)) {
2546 if (skb_pad(skb, len))
2548 __skb_put(skb, len);
2553 static inline int skb_add_data(struct sk_buff *skb,
2554 struct iov_iter *from, int copy)
2556 const int off = skb->len;
2558 if (skb->ip_summed == CHECKSUM_NONE) {
2560 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2561 &csum, from) == copy) {
2562 skb->csum = csum_block_add(skb->csum, csum, off);
2565 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
2568 __skb_trim(skb, off);
2572 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2573 const struct page *page, int off)
2576 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2578 return page == skb_frag_page(frag) &&
2579 off == frag->page_offset + skb_frag_size(frag);
2584 static inline int __skb_linearize(struct sk_buff *skb)
2586 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2590 * skb_linearize - convert paged skb to linear one
2591 * @skb: buffer to linarize
2593 * If there is no free memory -ENOMEM is returned, otherwise zero
2594 * is returned and the old skb data released.
2596 static inline int skb_linearize(struct sk_buff *skb)
2598 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2602 * skb_has_shared_frag - can any frag be overwritten
2603 * @skb: buffer to test
2605 * Return true if the skb has at least one frag that might be modified
2606 * by an external entity (as in vmsplice()/sendfile())
2608 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2610 return skb_is_nonlinear(skb) &&
2611 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2615 * skb_linearize_cow - make sure skb is linear and writable
2616 * @skb: buffer to process
2618 * If there is no free memory -ENOMEM is returned, otherwise zero
2619 * is returned and the old skb data released.
2621 static inline int skb_linearize_cow(struct sk_buff *skb)
2623 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2624 __skb_linearize(skb) : 0;
2628 * skb_postpull_rcsum - update checksum for received skb after pull
2629 * @skb: buffer to update
2630 * @start: start of data before pull
2631 * @len: length of data pulled
2633 * After doing a pull on a received packet, you need to call this to
2634 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2635 * CHECKSUM_NONE so that it can be recomputed from scratch.
2638 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2639 const void *start, unsigned int len)
2641 if (skb->ip_summed == CHECKSUM_COMPLETE)
2642 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2645 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2648 * pskb_trim_rcsum - trim received skb and update checksum
2649 * @skb: buffer to trim
2652 * This is exactly the same as pskb_trim except that it ensures the
2653 * checksum of received packets are still valid after the operation.
2656 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2658 if (likely(len >= skb->len))
2660 if (skb->ip_summed == CHECKSUM_COMPLETE)
2661 skb->ip_summed = CHECKSUM_NONE;
2662 return __pskb_trim(skb, len);
2665 #define skb_queue_walk(queue, skb) \
2666 for (skb = (queue)->next; \
2667 skb != (struct sk_buff *)(queue); \
2670 #define skb_queue_walk_safe(queue, skb, tmp) \
2671 for (skb = (queue)->next, tmp = skb->next; \
2672 skb != (struct sk_buff *)(queue); \
2673 skb = tmp, tmp = skb->next)
2675 #define skb_queue_walk_from(queue, skb) \
2676 for (; skb != (struct sk_buff *)(queue); \
2679 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2680 for (tmp = skb->next; \
2681 skb != (struct sk_buff *)(queue); \
2682 skb = tmp, tmp = skb->next)
2684 #define skb_queue_reverse_walk(queue, skb) \
2685 for (skb = (queue)->prev; \
2686 skb != (struct sk_buff *)(queue); \
2689 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2690 for (skb = (queue)->prev, tmp = skb->prev; \
2691 skb != (struct sk_buff *)(queue); \
2692 skb = tmp, tmp = skb->prev)
2694 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2695 for (tmp = skb->prev; \
2696 skb != (struct sk_buff *)(queue); \
2697 skb = tmp, tmp = skb->prev)
2699 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2701 return skb_shinfo(skb)->frag_list != NULL;
2704 static inline void skb_frag_list_init(struct sk_buff *skb)
2706 skb_shinfo(skb)->frag_list = NULL;
2709 #define skb_walk_frags(skb, iter) \
2710 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2712 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2713 int *peeked, int *off, int *err);
2714 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2716 unsigned int datagram_poll(struct file *file, struct socket *sock,
2717 struct poll_table_struct *wait);
2718 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2719 struct iov_iter *to, int size);
2720 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2721 struct msghdr *msg, int size)
2723 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
2725 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2726 struct msghdr *msg);
2727 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2728 struct iov_iter *from, int len);
2729 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
2730 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2731 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2732 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2733 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2734 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2735 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2736 int len, __wsum csum);
2737 ssize_t skb_socket_splice(struct sock *sk,
2738 struct pipe_inode_info *pipe,
2739 struct splice_pipe_desc *spd);
2740 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2741 struct pipe_inode_info *pipe, unsigned int len,
2743 ssize_t (*splice_cb)(struct sock *,
2744 struct pipe_inode_info *,
2745 struct splice_pipe_desc *));
2746 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2747 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2748 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2750 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2751 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2752 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2753 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2754 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2755 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
2756 int skb_ensure_writable(struct sk_buff *skb, int write_len);
2757 int skb_vlan_pop(struct sk_buff *skb);
2758 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
2760 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2762 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
2765 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
2767 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
2770 struct skb_checksum_ops {
2771 __wsum (*update)(const void *mem, int len, __wsum wsum);
2772 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2775 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2776 __wsum csum, const struct skb_checksum_ops *ops);
2777 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2780 static inline void * __must_check
2781 __skb_header_pointer(const struct sk_buff *skb, int offset,
2782 int len, void *data, int hlen, void *buffer)
2784 if (hlen - offset >= len)
2785 return data + offset;
2788 skb_copy_bits(skb, offset, buffer, len) < 0)
2794 static inline void * __must_check
2795 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
2797 return __skb_header_pointer(skb, offset, len, skb->data,
2798 skb_headlen(skb), buffer);
2802 * skb_needs_linearize - check if we need to linearize a given skb
2803 * depending on the given device features.
2804 * @skb: socket buffer to check
2805 * @features: net device features
2807 * Returns true if either:
2808 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2809 * 2. skb is fragmented and the device does not support SG.
2811 static inline bool skb_needs_linearize(struct sk_buff *skb,
2812 netdev_features_t features)
2814 return skb_is_nonlinear(skb) &&
2815 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2816 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2819 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2821 const unsigned int len)
2823 memcpy(to, skb->data, len);
2826 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2827 const int offset, void *to,
2828 const unsigned int len)
2830 memcpy(to, skb->data + offset, len);
2833 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2835 const unsigned int len)
2837 memcpy(skb->data, from, len);
2840 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2843 const unsigned int len)
2845 memcpy(skb->data + offset, from, len);
2848 void skb_init(void);
2850 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2856 * skb_get_timestamp - get timestamp from a skb
2857 * @skb: skb to get stamp from
2858 * @stamp: pointer to struct timeval to store stamp in
2860 * Timestamps are stored in the skb as offsets to a base timestamp.
2861 * This function converts the offset back to a struct timeval and stores
2864 static inline void skb_get_timestamp(const struct sk_buff *skb,
2865 struct timeval *stamp)
2867 *stamp = ktime_to_timeval(skb->tstamp);
2870 static inline void skb_get_timestampns(const struct sk_buff *skb,
2871 struct timespec *stamp)
2873 *stamp = ktime_to_timespec(skb->tstamp);
2876 static inline void __net_timestamp(struct sk_buff *skb)
2878 skb->tstamp = ktime_get_real();
2881 static inline ktime_t net_timedelta(ktime_t t)
2883 return ktime_sub(ktime_get_real(), t);
2886 static inline ktime_t net_invalid_timestamp(void)
2888 return ktime_set(0, 0);
2891 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
2893 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2895 void skb_clone_tx_timestamp(struct sk_buff *skb);
2896 bool skb_defer_rx_timestamp(struct sk_buff *skb);
2898 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2900 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2904 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2909 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2912 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2914 * PHY drivers may accept clones of transmitted packets for
2915 * timestamping via their phy_driver.txtstamp method. These drivers
2916 * must call this function to return the skb back to the stack with a
2919 * @skb: clone of the the original outgoing packet
2920 * @hwtstamps: hardware time stamps
2923 void skb_complete_tx_timestamp(struct sk_buff *skb,
2924 struct skb_shared_hwtstamps *hwtstamps);
2926 void __skb_tstamp_tx(struct sk_buff *orig_skb,
2927 struct skb_shared_hwtstamps *hwtstamps,
2928 struct sock *sk, int tstype);
2931 * skb_tstamp_tx - queue clone of skb with send time stamps
2932 * @orig_skb: the original outgoing packet
2933 * @hwtstamps: hardware time stamps, may be NULL if not available
2935 * If the skb has a socket associated, then this function clones the
2936 * skb (thus sharing the actual data and optional structures), stores
2937 * the optional hardware time stamping information (if non NULL) or
2938 * generates a software time stamp (otherwise), then queues the clone
2939 * to the error queue of the socket. Errors are silently ignored.
2941 void skb_tstamp_tx(struct sk_buff *orig_skb,
2942 struct skb_shared_hwtstamps *hwtstamps);
2944 static inline void sw_tx_timestamp(struct sk_buff *skb)
2946 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2947 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
2948 skb_tstamp_tx(skb, NULL);
2952 * skb_tx_timestamp() - Driver hook for transmit timestamping
2954 * Ethernet MAC Drivers should call this function in their hard_xmit()
2955 * function immediately before giving the sk_buff to the MAC hardware.
2957 * Specifically, one should make absolutely sure that this function is
2958 * called before TX completion of this packet can trigger. Otherwise
2959 * the packet could potentially already be freed.
2961 * @skb: A socket buffer.
2963 static inline void skb_tx_timestamp(struct sk_buff *skb)
2965 skb_clone_tx_timestamp(skb);
2966 sw_tx_timestamp(skb);
2970 * skb_complete_wifi_ack - deliver skb with wifi status
2972 * @skb: the original outgoing packet
2973 * @acked: ack status
2976 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2978 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2979 __sum16 __skb_checksum_complete(struct sk_buff *skb);
2981 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2983 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
2985 (skb->ip_summed == CHECKSUM_PARTIAL &&
2986 skb_checksum_start_offset(skb) >= 0));
2990 * skb_checksum_complete - Calculate checksum of an entire packet
2991 * @skb: packet to process
2993 * This function calculates the checksum over the entire packet plus
2994 * the value of skb->csum. The latter can be used to supply the
2995 * checksum of a pseudo header as used by TCP/UDP. It returns the
2998 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2999 * this function can be used to verify that checksum on received
3000 * packets. In that case the function should return zero if the
3001 * checksum is correct. In particular, this function will return zero
3002 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3003 * hardware has already verified the correctness of the checksum.
3005 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3007 return skb_csum_unnecessary(skb) ?
3008 0 : __skb_checksum_complete(skb);
3011 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3013 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3014 if (skb->csum_level == 0)
3015 skb->ip_summed = CHECKSUM_NONE;
3021 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3023 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3024 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3026 } else if (skb->ip_summed == CHECKSUM_NONE) {
3027 skb->ip_summed = CHECKSUM_UNNECESSARY;
3028 skb->csum_level = 0;
3032 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3034 /* Mark current checksum as bad (typically called from GRO
3035 * path). In the case that ip_summed is CHECKSUM_NONE
3036 * this must be the first checksum encountered in the packet.
3037 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3038 * checksum after the last one validated. For UDP, a zero
3039 * checksum can not be marked as bad.
3042 if (skb->ip_summed == CHECKSUM_NONE ||
3043 skb->ip_summed == CHECKSUM_UNNECESSARY)
3047 /* Check if we need to perform checksum complete validation.
3049 * Returns true if checksum complete is needed, false otherwise
3050 * (either checksum is unnecessary or zero checksum is allowed).
3052 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3056 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3057 skb->csum_valid = 1;
3058 __skb_decr_checksum_unnecessary(skb);
3065 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3068 #define CHECKSUM_BREAK 76
3070 /* Unset checksum-complete
3072 * Unset checksum complete can be done when packet is being modified
3073 * (uncompressed for instance) and checksum-complete value is
3076 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3078 if (skb->ip_summed == CHECKSUM_COMPLETE)
3079 skb->ip_summed = CHECKSUM_NONE;
3082 /* Validate (init) checksum based on checksum complete.
3085 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3086 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3087 * checksum is stored in skb->csum for use in __skb_checksum_complete
3088 * non-zero: value of invalid checksum
3091 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3095 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3096 if (!csum_fold(csum_add(psum, skb->csum))) {
3097 skb->csum_valid = 1;
3100 } else if (skb->csum_bad) {
3101 /* ip_summed == CHECKSUM_NONE in this case */
3102 return (__force __sum16)1;
3107 if (complete || skb->len <= CHECKSUM_BREAK) {
3110 csum = __skb_checksum_complete(skb);
3111 skb->csum_valid = !csum;
3118 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3123 /* Perform checksum validate (init). Note that this is a macro since we only
3124 * want to calculate the pseudo header which is an input function if necessary.
3125 * First we try to validate without any computation (checksum unnecessary) and
3126 * then calculate based on checksum complete calling the function to compute
3130 * 0: checksum is validated or try to in skb_checksum_complete
3131 * non-zero: value of invalid checksum
3133 #define __skb_checksum_validate(skb, proto, complete, \
3134 zero_okay, check, compute_pseudo) \
3136 __sum16 __ret = 0; \
3137 skb->csum_valid = 0; \
3138 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3139 __ret = __skb_checksum_validate_complete(skb, \
3140 complete, compute_pseudo(skb, proto)); \
3144 #define skb_checksum_init(skb, proto, compute_pseudo) \
3145 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3147 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3148 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3150 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3151 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3153 #define skb_checksum_validate_zero_check(skb, proto, check, \
3155 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3157 #define skb_checksum_simple_validate(skb) \
3158 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3160 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3162 return (skb->ip_summed == CHECKSUM_NONE &&
3163 skb->csum_valid && !skb->csum_bad);
3166 static inline void __skb_checksum_convert(struct sk_buff *skb,
3167 __sum16 check, __wsum pseudo)
3169 skb->csum = ~pseudo;
3170 skb->ip_summed = CHECKSUM_COMPLETE;
3173 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3175 if (__skb_checksum_convert_check(skb)) \
3176 __skb_checksum_convert(skb, check, \
3177 compute_pseudo(skb, proto)); \
3180 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3181 u16 start, u16 offset)
3183 skb->ip_summed = CHECKSUM_PARTIAL;
3184 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3185 skb->csum_offset = offset - start;
3188 /* Update skbuf and packet to reflect the remote checksum offload operation.
3189 * When called, ptr indicates the starting point for skb->csum when
3190 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3191 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3193 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3194 int start, int offset, bool nopartial)
3199 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3203 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3204 __skb_checksum_complete(skb);
3205 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3208 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3210 /* Adjust skb->csum since we changed the packet */
3211 skb->csum = csum_add(skb->csum, delta);
3214 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3215 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3216 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3218 if (nfct && atomic_dec_and_test(&nfct->use))
3219 nf_conntrack_destroy(nfct);
3221 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3224 atomic_inc(&nfct->use);
3227 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3228 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3230 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3233 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3236 atomic_inc(&nf_bridge->use);
3238 #endif /* CONFIG_BRIDGE_NETFILTER */
3239 static inline void nf_reset(struct sk_buff *skb)
3241 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3242 nf_conntrack_put(skb->nfct);
3245 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3246 nf_bridge_put(skb->nf_bridge);
3247 skb->nf_bridge = NULL;
3251 static inline void nf_reset_trace(struct sk_buff *skb)
3253 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3258 /* Note: This doesn't put any conntrack and bridge info in dst. */
3259 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3262 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3263 dst->nfct = src->nfct;
3264 nf_conntrack_get(src->nfct);
3266 dst->nfctinfo = src->nfctinfo;
3268 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3269 dst->nf_bridge = src->nf_bridge;
3270 nf_bridge_get(src->nf_bridge);
3272 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3274 dst->nf_trace = src->nf_trace;
3278 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3280 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3281 nf_conntrack_put(dst->nfct);
3283 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3284 nf_bridge_put(dst->nf_bridge);
3286 __nf_copy(dst, src, true);
3289 #ifdef CONFIG_NETWORK_SECMARK
3290 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3292 to->secmark = from->secmark;
3295 static inline void skb_init_secmark(struct sk_buff *skb)
3300 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3303 static inline void skb_init_secmark(struct sk_buff *skb)
3307 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3309 return !skb->destructor &&
3310 #if IS_ENABLED(CONFIG_XFRM)
3313 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3316 !skb->_skb_refdst &&
3317 !skb_has_frag_list(skb);
3320 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3322 skb->queue_mapping = queue_mapping;
3325 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3327 return skb->queue_mapping;
3330 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3332 to->queue_mapping = from->queue_mapping;
3335 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3337 skb->queue_mapping = rx_queue + 1;
3340 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3342 return skb->queue_mapping - 1;
3345 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3347 return skb->queue_mapping != 0;
3350 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3359 /* Keeps track of mac header offset relative to skb->head.
3360 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3361 * For non-tunnel skb it points to skb_mac_header() and for
3362 * tunnel skb it points to outer mac header.
3363 * Keeps track of level of encapsulation of network headers.
3370 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3372 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3374 return (skb_mac_header(inner_skb) - inner_skb->head) -
3375 SKB_GSO_CB(inner_skb)->mac_offset;
3378 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3380 int new_headroom, headroom;
3383 headroom = skb_headroom(skb);
3384 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3388 new_headroom = skb_headroom(skb);
3389 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3393 /* Compute the checksum for a gso segment. First compute the checksum value
3394 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3395 * then add in skb->csum (checksum from csum_start to end of packet).
3396 * skb->csum and csum_start are then updated to reflect the checksum of the
3397 * resultant packet starting from the transport header-- the resultant checksum
3398 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3401 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3403 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3404 skb_transport_offset(skb);
3407 partial = csum_partial(skb_transport_header(skb), plen, skb->csum);
3409 SKB_GSO_CB(skb)->csum_start -= plen;
3411 return csum_fold(partial);
3414 static inline bool skb_is_gso(const struct sk_buff *skb)
3416 return skb_shinfo(skb)->gso_size;
3419 /* Note: Should be called only if skb_is_gso(skb) is true */
3420 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3422 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3425 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3427 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3429 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3430 * wanted then gso_type will be set. */
3431 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3433 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3434 unlikely(shinfo->gso_type == 0)) {
3435 __skb_warn_lro_forwarding(skb);
3441 static inline void skb_forward_csum(struct sk_buff *skb)
3443 /* Unfortunately we don't support this one. Any brave souls? */
3444 if (skb->ip_summed == CHECKSUM_COMPLETE)
3445 skb->ip_summed = CHECKSUM_NONE;
3449 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3450 * @skb: skb to check
3452 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3453 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3454 * use this helper, to document places where we make this assertion.
3456 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3459 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3463 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3465 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3466 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3467 unsigned int transport_len,
3468 __sum16(*skb_chkf)(struct sk_buff *skb));
3471 * skb_head_is_locked - Determine if the skb->head is locked down
3472 * @skb: skb to check
3474 * The head on skbs build around a head frag can be removed if they are
3475 * not cloned. This function returns true if the skb head is locked down
3476 * due to either being allocated via kmalloc, or by being a clone with
3477 * multiple references to the head.
3479 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3481 return !skb->head_frag || skb_cloned(skb);
3485 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3489 * skb_gso_network_seglen is used to determine the real size of the
3490 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3492 * The MAC/L2 header is not accounted for.
3494 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3496 unsigned int hdr_len = skb_transport_header(skb) -
3497 skb_network_header(skb);
3498 return hdr_len + skb_gso_transport_seglen(skb);
3501 #endif /* __KERNEL__ */
3502 #endif /* _LINUX_SKBUFF_H */