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 */
940 static inline void skb_clear_hash(struct sk_buff
*skb
)
947 static inline void skb_clear_hash_if_not_l4(struct sk_buff
*skb
)
954 __skb_set_hash(struct sk_buff
*skb
, __u32 hash
, bool is_sw
, bool is_l4
)
956 skb
->l4_hash
= is_l4
;
957 skb
->sw_hash
= is_sw
;
962 skb_set_hash(struct sk_buff
*skb
, __u32 hash
, enum pkt_hash_types type
)
964 /* Used by drivers to set hash from HW */
965 __skb_set_hash(skb
, hash
, false, type
== PKT_HASH_TYPE_L4
);
969 __skb_set_sw_hash(struct sk_buff
*skb
, __u32 hash
, bool is_l4
)
971 __skb_set_hash(skb
, hash
, true, is_l4
);
974 void __skb_get_hash(struct sk_buff
*skb
);
975 u32
skb_get_poff(const struct sk_buff
*skb
);
976 u32
__skb_get_poff(const struct sk_buff
*skb
, void *data
,
977 const struct flow_keys
*keys
, int hlen
);
978 __be32
__skb_flow_get_ports(const struct sk_buff
*skb
, int thoff
, u8 ip_proto
,
979 void *data
, int hlen_proto
);
981 static inline __be32
skb_flow_get_ports(const struct sk_buff
*skb
,
982 int thoff
, u8 ip_proto
)
984 return __skb_flow_get_ports(skb
, thoff
, ip_proto
, NULL
, 0);
987 void skb_flow_dissector_init(struct flow_dissector
*flow_dissector
,
988 const struct flow_dissector_key
*key
,
989 unsigned int key_count
);
991 bool __skb_flow_dissect(const struct sk_buff
*skb
,
992 struct flow_dissector
*flow_dissector
,
993 void *target_container
,
994 void *data
, __be16 proto
, int nhoff
, int hlen
,
997 static inline bool skb_flow_dissect(const struct sk_buff
*skb
,
998 struct flow_dissector
*flow_dissector
,
999 void *target_container
, unsigned int flags
)
1001 return __skb_flow_dissect(skb
, flow_dissector
, target_container
,
1002 NULL
, 0, 0, 0, flags
);
1005 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff
*skb
,
1006 struct flow_keys
*flow
,
1009 memset(flow
, 0, sizeof(*flow
));
1010 return __skb_flow_dissect(skb
, &flow_keys_dissector
, flow
,
1011 NULL
, 0, 0, 0, flags
);
1014 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys
*flow
,
1015 void *data
, __be16 proto
,
1016 int nhoff
, int hlen
,
1019 memset(flow
, 0, sizeof(*flow
));
1020 return __skb_flow_dissect(NULL
, &flow_keys_buf_dissector
, flow
,
1021 data
, proto
, nhoff
, hlen
, flags
);
1024 static inline __u32
skb_get_hash(struct sk_buff
*skb
)
1026 if (!skb
->l4_hash
&& !skb
->sw_hash
)
1027 __skb_get_hash(skb
);
1032 __u32
__skb_get_hash_flowi6(struct sk_buff
*skb
, const struct flowi6
*fl6
);
1034 static inline __u32
skb_get_hash_flowi6(struct sk_buff
*skb
, const struct flowi6
*fl6
)
1036 if (!skb
->l4_hash
&& !skb
->sw_hash
) {
1037 struct flow_keys keys
;
1038 __u32 hash
= __get_hash_from_flowi6(fl6
, &keys
);
1040 __skb_set_sw_hash(skb
, hash
, flow_keys_have_l4(&keys
));
1046 __u32
__skb_get_hash_flowi4(struct sk_buff
*skb
, const struct flowi4
*fl
);
1048 static inline __u32
skb_get_hash_flowi4(struct sk_buff
*skb
, const struct flowi4
*fl4
)
1050 if (!skb
->l4_hash
&& !skb
->sw_hash
) {
1051 struct flow_keys keys
;
1052 __u32 hash
= __get_hash_from_flowi4(fl4
, &keys
);
1054 __skb_set_sw_hash(skb
, hash
, flow_keys_have_l4(&keys
));
1060 __u32
skb_get_hash_perturb(const struct sk_buff
*skb
, u32 perturb
);
1062 static inline __u32
skb_get_hash_raw(const struct sk_buff
*skb
)
1067 static inline void skb_copy_hash(struct sk_buff
*to
, const struct sk_buff
*from
)
1069 to
->hash
= from
->hash
;
1070 to
->sw_hash
= from
->sw_hash
;
1071 to
->l4_hash
= from
->l4_hash
;
1074 static inline void skb_sender_cpu_clear(struct sk_buff
*skb
)
1077 skb
->sender_cpu
= 0;
1081 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1082 static inline unsigned char *skb_end_pointer(const struct sk_buff
*skb
)
1084 return skb
->head
+ skb
->end
;
1087 static inline unsigned int skb_end_offset(const struct sk_buff
*skb
)
1092 static inline unsigned char *skb_end_pointer(const struct sk_buff
*skb
)
1097 static inline unsigned int skb_end_offset(const struct sk_buff
*skb
)
1099 return skb
->end
- skb
->head
;
1104 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1106 static inline struct skb_shared_hwtstamps
*skb_hwtstamps(struct sk_buff
*skb
)
1108 return &skb_shinfo(skb
)->hwtstamps
;
1112 * skb_queue_empty - check if a queue is empty
1115 * Returns true if the queue is empty, false otherwise.
1117 static inline int skb_queue_empty(const struct sk_buff_head
*list
)
1119 return list
->next
== (const struct sk_buff
*) list
;
1123 * skb_queue_is_last - check if skb is the last entry in the queue
1127 * Returns true if @skb is the last buffer on the list.
1129 static inline bool skb_queue_is_last(const struct sk_buff_head
*list
,
1130 const struct sk_buff
*skb
)
1132 return skb
->next
== (const struct sk_buff
*) list
;
1136 * skb_queue_is_first - check if skb is the first entry in the queue
1140 * Returns true if @skb is the first buffer on the list.
1142 static inline bool skb_queue_is_first(const struct sk_buff_head
*list
,
1143 const struct sk_buff
*skb
)
1145 return skb
->prev
== (const struct sk_buff
*) list
;
1149 * skb_queue_next - return the next packet in the queue
1151 * @skb: current buffer
1153 * Return the next packet in @list after @skb. It is only valid to
1154 * call this if skb_queue_is_last() evaluates to false.
1156 static inline struct sk_buff
*skb_queue_next(const struct sk_buff_head
*list
,
1157 const struct sk_buff
*skb
)
1159 /* This BUG_ON may seem severe, but if we just return then we
1160 * are going to dereference garbage.
1162 BUG_ON(skb_queue_is_last(list
, skb
));
1167 * skb_queue_prev - return the prev packet in the queue
1169 * @skb: current buffer
1171 * Return the prev packet in @list before @skb. It is only valid to
1172 * call this if skb_queue_is_first() evaluates to false.
1174 static inline struct sk_buff
*skb_queue_prev(const struct sk_buff_head
*list
,
1175 const struct sk_buff
*skb
)
1177 /* This BUG_ON may seem severe, but if we just return then we
1178 * are going to dereference garbage.
1180 BUG_ON(skb_queue_is_first(list
, skb
));
1185 * skb_get - reference buffer
1186 * @skb: buffer to reference
1188 * Makes another reference to a socket buffer and returns a pointer
1191 static inline struct sk_buff
*skb_get(struct sk_buff
*skb
)
1193 atomic_inc(&skb
->users
);
1198 * If users == 1, we are the only owner and are can avoid redundant
1203 * skb_cloned - is the buffer a clone
1204 * @skb: buffer to check
1206 * Returns true if the buffer was generated with skb_clone() and is
1207 * one of multiple shared copies of the buffer. Cloned buffers are
1208 * shared data so must not be written to under normal circumstances.
1210 static inline int skb_cloned(const struct sk_buff
*skb
)
1212 return skb
->cloned
&&
1213 (atomic_read(&skb_shinfo(skb
)->dataref
) & SKB_DATAREF_MASK
) != 1;
1216 static inline int skb_unclone(struct sk_buff
*skb
, gfp_t pri
)
1218 might_sleep_if(pri
& __GFP_WAIT
);
1220 if (skb_cloned(skb
))
1221 return pskb_expand_head(skb
, 0, 0, pri
);
1227 * skb_header_cloned - is the header a clone
1228 * @skb: buffer to check
1230 * Returns true if modifying the header part of the buffer requires
1231 * the data to be copied.
1233 static inline int skb_header_cloned(const struct sk_buff
*skb
)
1240 dataref
= atomic_read(&skb_shinfo(skb
)->dataref
);
1241 dataref
= (dataref
& SKB_DATAREF_MASK
) - (dataref
>> SKB_DATAREF_SHIFT
);
1242 return dataref
!= 1;
1246 * skb_header_release - release reference to header
1247 * @skb: buffer to operate on
1249 * Drop a reference to the header part of the buffer. This is done
1250 * by acquiring a payload reference. You must not read from the header
1251 * part of skb->data after this.
1252 * Note : Check if you can use __skb_header_release() instead.
1254 static inline void skb_header_release(struct sk_buff
*skb
)
1258 atomic_add(1 << SKB_DATAREF_SHIFT
, &skb_shinfo(skb
)->dataref
);
1262 * __skb_header_release - release reference to header
1263 * @skb: buffer to operate on
1265 * Variant of skb_header_release() assuming skb is private to caller.
1266 * We can avoid one atomic operation.
1268 static inline void __skb_header_release(struct sk_buff
*skb
)
1271 atomic_set(&skb_shinfo(skb
)->dataref
, 1 + (1 << SKB_DATAREF_SHIFT
));
1276 * skb_shared - is the buffer shared
1277 * @skb: buffer to check
1279 * Returns true if more than one person has a reference to this
1282 static inline int skb_shared(const struct sk_buff
*skb
)
1284 return atomic_read(&skb
->users
) != 1;
1288 * skb_share_check - check if buffer is shared and if so clone it
1289 * @skb: buffer to check
1290 * @pri: priority for memory allocation
1292 * If the buffer is shared the buffer is cloned and the old copy
1293 * drops a reference. A new clone with a single reference is returned.
1294 * If the buffer is not shared the original buffer is returned. When
1295 * being called from interrupt status or with spinlocks held pri must
1298 * NULL is returned on a memory allocation failure.
1300 static inline struct sk_buff
*skb_share_check(struct sk_buff
*skb
, gfp_t pri
)
1302 might_sleep_if(pri
& __GFP_WAIT
);
1303 if (skb_shared(skb
)) {
1304 struct sk_buff
*nskb
= skb_clone(skb
, pri
);
1316 * Copy shared buffers into a new sk_buff. We effectively do COW on
1317 * packets to handle cases where we have a local reader and forward
1318 * and a couple of other messy ones. The normal one is tcpdumping
1319 * a packet thats being forwarded.
1323 * skb_unshare - make a copy of a shared buffer
1324 * @skb: buffer to check
1325 * @pri: priority for memory allocation
1327 * If the socket buffer is a clone then this function creates a new
1328 * copy of the data, drops a reference count on the old copy and returns
1329 * the new copy with the reference count at 1. If the buffer is not a clone
1330 * the original buffer is returned. When called with a spinlock held or
1331 * from interrupt state @pri must be %GFP_ATOMIC
1333 * %NULL is returned on a memory allocation failure.
1335 static inline struct sk_buff
*skb_unshare(struct sk_buff
*skb
,
1338 might_sleep_if(pri
& __GFP_WAIT
);
1339 if (skb_cloned(skb
)) {
1340 struct sk_buff
*nskb
= skb_copy(skb
, pri
);
1342 /* Free our shared copy */
1353 * skb_peek - peek at the head of an &sk_buff_head
1354 * @list_: list to peek at
1356 * Peek an &sk_buff. Unlike most other operations you _MUST_
1357 * be careful with this one. A peek leaves the buffer on the
1358 * list and someone else may run off with it. You must hold
1359 * the appropriate locks or have a private queue to do this.
1361 * Returns %NULL for an empty list or a pointer to the head element.
1362 * The reference count is not incremented and the reference is therefore
1363 * volatile. Use with caution.
1365 static inline struct sk_buff
*skb_peek(const struct sk_buff_head
*list_
)
1367 struct sk_buff
*skb
= list_
->next
;
1369 if (skb
== (struct sk_buff
*)list_
)
1375 * skb_peek_next - peek skb following the given one from a queue
1376 * @skb: skb to start from
1377 * @list_: list to peek at
1379 * Returns %NULL when the end of the list is met or a pointer to the
1380 * next element. The reference count is not incremented and the
1381 * reference is therefore volatile. Use with caution.
1383 static inline struct sk_buff
*skb_peek_next(struct sk_buff
*skb
,
1384 const struct sk_buff_head
*list_
)
1386 struct sk_buff
*next
= skb
->next
;
1388 if (next
== (struct sk_buff
*)list_
)
1394 * skb_peek_tail - peek at the tail of an &sk_buff_head
1395 * @list_: list to peek at
1397 * Peek an &sk_buff. Unlike most other operations you _MUST_
1398 * be careful with this one. A peek leaves the buffer on the
1399 * list and someone else may run off with it. You must hold
1400 * the appropriate locks or have a private queue to do this.
1402 * Returns %NULL for an empty list or a pointer to the tail element.
1403 * The reference count is not incremented and the reference is therefore
1404 * volatile. Use with caution.
1406 static inline struct sk_buff
*skb_peek_tail(const struct sk_buff_head
*list_
)
1408 struct sk_buff
*skb
= list_
->prev
;
1410 if (skb
== (struct sk_buff
*)list_
)
1417 * skb_queue_len - get queue length
1418 * @list_: list to measure
1420 * Return the length of an &sk_buff queue.
1422 static inline __u32
skb_queue_len(const struct sk_buff_head
*list_
)
1428 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1429 * @list: queue to initialize
1431 * This initializes only the list and queue length aspects of
1432 * an sk_buff_head object. This allows to initialize the list
1433 * aspects of an sk_buff_head without reinitializing things like
1434 * the spinlock. It can also be used for on-stack sk_buff_head
1435 * objects where the spinlock is known to not be used.
1437 static inline void __skb_queue_head_init(struct sk_buff_head
*list
)
1439 list
->prev
= list
->next
= (struct sk_buff
*)list
;
1444 * This function creates a split out lock class for each invocation;
1445 * this is needed for now since a whole lot of users of the skb-queue
1446 * infrastructure in drivers have different locking usage (in hardirq)
1447 * than the networking core (in softirq only). In the long run either the
1448 * network layer or drivers should need annotation to consolidate the
1449 * main types of usage into 3 classes.
1451 static inline void skb_queue_head_init(struct sk_buff_head
*list
)
1453 spin_lock_init(&list
->lock
);
1454 __skb_queue_head_init(list
);
1457 static inline void skb_queue_head_init_class(struct sk_buff_head
*list
,
1458 struct lock_class_key
*class)
1460 skb_queue_head_init(list
);
1461 lockdep_set_class(&list
->lock
, class);
1465 * Insert an sk_buff on a list.
1467 * The "__skb_xxxx()" functions are the non-atomic ones that
1468 * can only be called with interrupts disabled.
1470 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
,
1471 struct sk_buff_head
*list
);
1472 static inline void __skb_insert(struct sk_buff
*newsk
,
1473 struct sk_buff
*prev
, struct sk_buff
*next
,
1474 struct sk_buff_head
*list
)
1478 next
->prev
= prev
->next
= newsk
;
1482 static inline void __skb_queue_splice(const struct sk_buff_head
*list
,
1483 struct sk_buff
*prev
,
1484 struct sk_buff
*next
)
1486 struct sk_buff
*first
= list
->next
;
1487 struct sk_buff
*last
= list
->prev
;
1497 * skb_queue_splice - join two skb lists, this is designed for stacks
1498 * @list: the new list to add
1499 * @head: the place to add it in the first list
1501 static inline void skb_queue_splice(const struct sk_buff_head
*list
,
1502 struct sk_buff_head
*head
)
1504 if (!skb_queue_empty(list
)) {
1505 __skb_queue_splice(list
, (struct sk_buff
*) head
, head
->next
);
1506 head
->qlen
+= list
->qlen
;
1511 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1512 * @list: the new list to add
1513 * @head: the place to add it in the first list
1515 * The list at @list is reinitialised
1517 static inline void skb_queue_splice_init(struct sk_buff_head
*list
,
1518 struct sk_buff_head
*head
)
1520 if (!skb_queue_empty(list
)) {
1521 __skb_queue_splice(list
, (struct sk_buff
*) head
, head
->next
);
1522 head
->qlen
+= list
->qlen
;
1523 __skb_queue_head_init(list
);
1528 * skb_queue_splice_tail - join two skb lists, each list being a queue
1529 * @list: the new list to add
1530 * @head: the place to add it in the first list
1532 static inline void skb_queue_splice_tail(const struct sk_buff_head
*list
,
1533 struct sk_buff_head
*head
)
1535 if (!skb_queue_empty(list
)) {
1536 __skb_queue_splice(list
, head
->prev
, (struct sk_buff
*) head
);
1537 head
->qlen
+= list
->qlen
;
1542 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1543 * @list: the new list to add
1544 * @head: the place to add it in the first list
1546 * Each of the lists is a queue.
1547 * The list at @list is reinitialised
1549 static inline void skb_queue_splice_tail_init(struct sk_buff_head
*list
,
1550 struct sk_buff_head
*head
)
1552 if (!skb_queue_empty(list
)) {
1553 __skb_queue_splice(list
, head
->prev
, (struct sk_buff
*) head
);
1554 head
->qlen
+= list
->qlen
;
1555 __skb_queue_head_init(list
);
1560 * __skb_queue_after - queue a buffer at the list head
1561 * @list: list to use
1562 * @prev: place after this buffer
1563 * @newsk: buffer to queue
1565 * Queue a buffer int the middle of a list. This function takes no locks
1566 * and you must therefore hold required locks before calling it.
1568 * A buffer cannot be placed on two lists at the same time.
1570 static inline void __skb_queue_after(struct sk_buff_head
*list
,
1571 struct sk_buff
*prev
,
1572 struct sk_buff
*newsk
)
1574 __skb_insert(newsk
, prev
, prev
->next
, list
);
1577 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
,
1578 struct sk_buff_head
*list
);
1580 static inline void __skb_queue_before(struct sk_buff_head
*list
,
1581 struct sk_buff
*next
,
1582 struct sk_buff
*newsk
)
1584 __skb_insert(newsk
, next
->prev
, next
, list
);
1588 * __skb_queue_head - queue a buffer at the list head
1589 * @list: list to use
1590 * @newsk: buffer to queue
1592 * Queue a buffer at the start of a list. This function takes no locks
1593 * and you must therefore hold required locks before calling it.
1595 * A buffer cannot be placed on two lists at the same time.
1597 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
);
1598 static inline void __skb_queue_head(struct sk_buff_head
*list
,
1599 struct sk_buff
*newsk
)
1601 __skb_queue_after(list
, (struct sk_buff
*)list
, newsk
);
1605 * __skb_queue_tail - queue a buffer at the list tail
1606 * @list: list to use
1607 * @newsk: buffer to queue
1609 * Queue a buffer at the end of a list. This function takes no locks
1610 * and you must therefore hold required locks before calling it.
1612 * A buffer cannot be placed on two lists at the same time.
1614 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
);
1615 static inline void __skb_queue_tail(struct sk_buff_head
*list
,
1616 struct sk_buff
*newsk
)
1618 __skb_queue_before(list
, (struct sk_buff
*)list
, newsk
);
1622 * remove sk_buff from list. _Must_ be called atomically, and with
1625 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
);
1626 static inline void __skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
1628 struct sk_buff
*next
, *prev
;
1633 skb
->next
= skb
->prev
= NULL
;
1639 * __skb_dequeue - remove from the head of the queue
1640 * @list: list to dequeue from
1642 * Remove the head of the list. This function does not take any locks
1643 * so must be used with appropriate locks held only. The head item is
1644 * returned or %NULL if the list is empty.
1646 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
);
1647 static inline struct sk_buff
*__skb_dequeue(struct sk_buff_head
*list
)
1649 struct sk_buff
*skb
= skb_peek(list
);
1651 __skb_unlink(skb
, list
);
1656 * __skb_dequeue_tail - remove from the tail of the queue
1657 * @list: list to dequeue from
1659 * Remove the tail of the list. This function does not take any locks
1660 * so must be used with appropriate locks held only. The tail item is
1661 * returned or %NULL if the list is empty.
1663 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
);
1664 static inline struct sk_buff
*__skb_dequeue_tail(struct sk_buff_head
*list
)
1666 struct sk_buff
*skb
= skb_peek_tail(list
);
1668 __skb_unlink(skb
, list
);
1673 static inline bool skb_is_nonlinear(const struct sk_buff
*skb
)
1675 return skb
->data_len
;
1678 static inline unsigned int skb_headlen(const struct sk_buff
*skb
)
1680 return skb
->len
- skb
->data_len
;
1683 static inline int skb_pagelen(const struct sk_buff
*skb
)
1687 for (i
= (int)skb_shinfo(skb
)->nr_frags
- 1; i
>= 0; i
--)
1688 len
+= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1689 return len
+ skb_headlen(skb
);
1693 * __skb_fill_page_desc - initialise a paged fragment in an skb
1694 * @skb: buffer containing fragment to be initialised
1695 * @i: paged fragment index to initialise
1696 * @page: the page to use for this fragment
1697 * @off: the offset to the data with @page
1698 * @size: the length of the data
1700 * Initialises the @i'th fragment of @skb to point to &size bytes at
1701 * offset @off within @page.
1703 * Does not take any additional reference on the fragment.
1705 static inline void __skb_fill_page_desc(struct sk_buff
*skb
, int i
,
1706 struct page
*page
, int off
, int size
)
1708 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1711 * Propagate page pfmemalloc to the skb if we can. The problem is
1712 * that not all callers have unique ownership of the page but rely
1713 * on page_is_pfmemalloc doing the right thing(tm).
1715 frag
->page
.p
= page
;
1716 frag
->page_offset
= off
;
1717 skb_frag_size_set(frag
, size
);
1719 page
= compound_head(page
);
1720 if (page_is_pfmemalloc(page
))
1721 skb
->pfmemalloc
= true;
1725 * skb_fill_page_desc - initialise a paged fragment in an skb
1726 * @skb: buffer containing fragment to be initialised
1727 * @i: paged fragment index to initialise
1728 * @page: the page to use for this fragment
1729 * @off: the offset to the data with @page
1730 * @size: the length of the data
1732 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1733 * @skb to point to @size bytes at offset @off within @page. In
1734 * addition updates @skb such that @i is the last fragment.
1736 * Does not take any additional reference on the fragment.
1738 static inline void skb_fill_page_desc(struct sk_buff
*skb
, int i
,
1739 struct page
*page
, int off
, int size
)
1741 __skb_fill_page_desc(skb
, i
, page
, off
, size
);
1742 skb_shinfo(skb
)->nr_frags
= i
+ 1;
1745 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
1746 int size
, unsigned int truesize
);
1748 void skb_coalesce_rx_frag(struct sk_buff
*skb
, int i
, int size
,
1749 unsigned int truesize
);
1751 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1752 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1753 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1755 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1756 static inline unsigned char *skb_tail_pointer(const struct sk_buff
*skb
)
1758 return skb
->head
+ skb
->tail
;
1761 static inline void skb_reset_tail_pointer(struct sk_buff
*skb
)
1763 skb
->tail
= skb
->data
- skb
->head
;
1766 static inline void skb_set_tail_pointer(struct sk_buff
*skb
, const int offset
)
1768 skb_reset_tail_pointer(skb
);
1769 skb
->tail
+= offset
;
1772 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1773 static inline unsigned char *skb_tail_pointer(const struct sk_buff
*skb
)
1778 static inline void skb_reset_tail_pointer(struct sk_buff
*skb
)
1780 skb
->tail
= skb
->data
;
1783 static inline void skb_set_tail_pointer(struct sk_buff
*skb
, const int offset
)
1785 skb
->tail
= skb
->data
+ offset
;
1788 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1791 * Add data to an sk_buff
1793 unsigned char *pskb_put(struct sk_buff
*skb
, struct sk_buff
*tail
, int len
);
1794 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
);
1795 static inline unsigned char *__skb_put(struct sk_buff
*skb
, unsigned int len
)
1797 unsigned char *tmp
= skb_tail_pointer(skb
);
1798 SKB_LINEAR_ASSERT(skb
);
1804 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
);
1805 static inline unsigned char *__skb_push(struct sk_buff
*skb
, unsigned int len
)
1812 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
);
1813 static inline unsigned char *__skb_pull(struct sk_buff
*skb
, unsigned int len
)
1816 BUG_ON(skb
->len
< skb
->data_len
);
1817 return skb
->data
+= len
;
1820 static inline unsigned char *skb_pull_inline(struct sk_buff
*skb
, unsigned int len
)
1822 return unlikely(len
> skb
->len
) ? NULL
: __skb_pull(skb
, len
);
1825 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
);
1827 static inline unsigned char *__pskb_pull(struct sk_buff
*skb
, unsigned int len
)
1829 if (len
> skb_headlen(skb
) &&
1830 !__pskb_pull_tail(skb
, len
- skb_headlen(skb
)))
1833 return skb
->data
+= len
;
1836 static inline unsigned char *pskb_pull(struct sk_buff
*skb
, unsigned int len
)
1838 return unlikely(len
> skb
->len
) ? NULL
: __pskb_pull(skb
, len
);
1841 static inline int pskb_may_pull(struct sk_buff
*skb
, unsigned int len
)
1843 if (likely(len
<= skb_headlen(skb
)))
1845 if (unlikely(len
> skb
->len
))
1847 return __pskb_pull_tail(skb
, len
- skb_headlen(skb
)) != NULL
;
1851 * skb_headroom - bytes at buffer head
1852 * @skb: buffer to check
1854 * Return the number of bytes of free space at the head of an &sk_buff.
1856 static inline unsigned int skb_headroom(const struct sk_buff
*skb
)
1858 return skb
->data
- skb
->head
;
1862 * skb_tailroom - bytes at buffer end
1863 * @skb: buffer to check
1865 * Return the number of bytes of free space at the tail of an sk_buff
1867 static inline int skb_tailroom(const struct sk_buff
*skb
)
1869 return skb_is_nonlinear(skb
) ? 0 : skb
->end
- skb
->tail
;
1873 * skb_availroom - bytes at buffer end
1874 * @skb: buffer to check
1876 * Return the number of bytes of free space at the tail of an sk_buff
1877 * allocated by sk_stream_alloc()
1879 static inline int skb_availroom(const struct sk_buff
*skb
)
1881 if (skb_is_nonlinear(skb
))
1884 return skb
->end
- skb
->tail
- skb
->reserved_tailroom
;
1888 * skb_reserve - adjust headroom
1889 * @skb: buffer to alter
1890 * @len: bytes to move
1892 * Increase the headroom of an empty &sk_buff by reducing the tail
1893 * room. This is only allowed for an empty buffer.
1895 static inline void skb_reserve(struct sk_buff
*skb
, int len
)
1901 #define ENCAP_TYPE_ETHER 0
1902 #define ENCAP_TYPE_IPPROTO 1
1904 static inline void skb_set_inner_protocol(struct sk_buff
*skb
,
1907 skb
->inner_protocol
= protocol
;
1908 skb
->inner_protocol_type
= ENCAP_TYPE_ETHER
;
1911 static inline void skb_set_inner_ipproto(struct sk_buff
*skb
,
1914 skb
->inner_ipproto
= ipproto
;
1915 skb
->inner_protocol_type
= ENCAP_TYPE_IPPROTO
;
1918 static inline void skb_reset_inner_headers(struct sk_buff
*skb
)
1920 skb
->inner_mac_header
= skb
->mac_header
;
1921 skb
->inner_network_header
= skb
->network_header
;
1922 skb
->inner_transport_header
= skb
->transport_header
;
1925 static inline void skb_reset_mac_len(struct sk_buff
*skb
)
1927 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1930 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1933 return skb
->head
+ skb
->inner_transport_header
;
1936 static inline void skb_reset_inner_transport_header(struct sk_buff
*skb
)
1938 skb
->inner_transport_header
= skb
->data
- skb
->head
;
1941 static inline void skb_set_inner_transport_header(struct sk_buff
*skb
,
1944 skb_reset_inner_transport_header(skb
);
1945 skb
->inner_transport_header
+= offset
;
1948 static inline unsigned char *skb_inner_network_header(const struct sk_buff
*skb
)
1950 return skb
->head
+ skb
->inner_network_header
;
1953 static inline void skb_reset_inner_network_header(struct sk_buff
*skb
)
1955 skb
->inner_network_header
= skb
->data
- skb
->head
;
1958 static inline void skb_set_inner_network_header(struct sk_buff
*skb
,
1961 skb_reset_inner_network_header(skb
);
1962 skb
->inner_network_header
+= offset
;
1965 static inline unsigned char *skb_inner_mac_header(const struct sk_buff
*skb
)
1967 return skb
->head
+ skb
->inner_mac_header
;
1970 static inline void skb_reset_inner_mac_header(struct sk_buff
*skb
)
1972 skb
->inner_mac_header
= skb
->data
- skb
->head
;
1975 static inline void skb_set_inner_mac_header(struct sk_buff
*skb
,
1978 skb_reset_inner_mac_header(skb
);
1979 skb
->inner_mac_header
+= offset
;
1981 static inline bool skb_transport_header_was_set(const struct sk_buff
*skb
)
1983 return skb
->transport_header
!= (typeof(skb
->transport_header
))~0U;
1986 static inline unsigned char *skb_transport_header(const struct sk_buff
*skb
)
1988 return skb
->head
+ skb
->transport_header
;
1991 static inline void skb_reset_transport_header(struct sk_buff
*skb
)
1993 skb
->transport_header
= skb
->data
- skb
->head
;
1996 static inline void skb_set_transport_header(struct sk_buff
*skb
,
1999 skb_reset_transport_header(skb
);
2000 skb
->transport_header
+= offset
;
2003 static inline unsigned char *skb_network_header(const struct sk_buff
*skb
)
2005 return skb
->head
+ skb
->network_header
;
2008 static inline void skb_reset_network_header(struct sk_buff
*skb
)
2010 skb
->network_header
= skb
->data
- skb
->head
;
2013 static inline void skb_set_network_header(struct sk_buff
*skb
, const int offset
)
2015 skb_reset_network_header(skb
);
2016 skb
->network_header
+= offset
;
2019 static inline unsigned char *skb_mac_header(const struct sk_buff
*skb
)
2021 return skb
->head
+ skb
->mac_header
;
2024 static inline int skb_mac_header_was_set(const struct sk_buff
*skb
)
2026 return skb
->mac_header
!= (typeof(skb
->mac_header
))~0U;
2029 static inline void skb_reset_mac_header(struct sk_buff
*skb
)
2031 skb
->mac_header
= skb
->data
- skb
->head
;
2034 static inline void skb_set_mac_header(struct sk_buff
*skb
, const int offset
)
2036 skb_reset_mac_header(skb
);
2037 skb
->mac_header
+= offset
;
2040 static inline void skb_pop_mac_header(struct sk_buff
*skb
)
2042 skb
->mac_header
= skb
->network_header
;
2045 static inline void skb_probe_transport_header(struct sk_buff
*skb
,
2046 const int offset_hint
)
2048 struct flow_keys keys
;
2050 if (skb_transport_header_was_set(skb
))
2052 else if (skb_flow_dissect_flow_keys(skb
, &keys
, 0))
2053 skb_set_transport_header(skb
, keys
.control
.thoff
);
2055 skb_set_transport_header(skb
, offset_hint
);
2058 static inline void skb_mac_header_rebuild(struct sk_buff
*skb
)
2060 if (skb_mac_header_was_set(skb
)) {
2061 const unsigned char *old_mac
= skb_mac_header(skb
);
2063 skb_set_mac_header(skb
, -skb
->mac_len
);
2064 memmove(skb_mac_header(skb
), old_mac
, skb
->mac_len
);
2068 static inline int skb_checksum_start_offset(const struct sk_buff
*skb
)
2070 return skb
->csum_start
- skb_headroom(skb
);
2073 static inline int skb_transport_offset(const struct sk_buff
*skb
)
2075 return skb_transport_header(skb
) - skb
->data
;
2078 static inline u32
skb_network_header_len(const struct sk_buff
*skb
)
2080 return skb
->transport_header
- skb
->network_header
;
2083 static inline u32
skb_inner_network_header_len(const struct sk_buff
*skb
)
2085 return skb
->inner_transport_header
- skb
->inner_network_header
;
2088 static inline int skb_network_offset(const struct sk_buff
*skb
)
2090 return skb_network_header(skb
) - skb
->data
;
2093 static inline int skb_inner_network_offset(const struct sk_buff
*skb
)
2095 return skb_inner_network_header(skb
) - skb
->data
;
2098 static inline int pskb_network_may_pull(struct sk_buff
*skb
, unsigned int len
)
2100 return pskb_may_pull(skb
, skb_network_offset(skb
) + len
);
2104 * CPUs often take a performance hit when accessing unaligned memory
2105 * locations. The actual performance hit varies, it can be small if the
2106 * hardware handles it or large if we have to take an exception and fix it
2109 * Since an ethernet header is 14 bytes network drivers often end up with
2110 * the IP header at an unaligned offset. The IP header can be aligned by
2111 * shifting the start of the packet by 2 bytes. Drivers should do this
2114 * skb_reserve(skb, NET_IP_ALIGN);
2116 * The downside to this alignment of the IP header is that the DMA is now
2117 * unaligned. On some architectures the cost of an unaligned DMA is high
2118 * and this cost outweighs the gains made by aligning the IP header.
2120 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2123 #ifndef NET_IP_ALIGN
2124 #define NET_IP_ALIGN 2
2128 * The networking layer reserves some headroom in skb data (via
2129 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2130 * the header has to grow. In the default case, if the header has to grow
2131 * 32 bytes or less we avoid the reallocation.
2133 * Unfortunately this headroom changes the DMA alignment of the resulting
2134 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2135 * on some architectures. An architecture can override this value,
2136 * perhaps setting it to a cacheline in size (since that will maintain
2137 * cacheline alignment of the DMA). It must be a power of 2.
2139 * Various parts of the networking layer expect at least 32 bytes of
2140 * headroom, you should not reduce this.
2142 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2143 * to reduce average number of cache lines per packet.
2144 * get_rps_cpus() for example only access one 64 bytes aligned block :
2145 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2148 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2151 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
);
2153 static inline void __skb_trim(struct sk_buff
*skb
, unsigned int len
)
2155 if (unlikely(skb_is_nonlinear(skb
))) {
2160 skb_set_tail_pointer(skb
, len
);
2163 void skb_trim(struct sk_buff
*skb
, unsigned int len
);
2165 static inline int __pskb_trim(struct sk_buff
*skb
, unsigned int len
)
2168 return ___pskb_trim(skb
, len
);
2169 __skb_trim(skb
, len
);
2173 static inline int pskb_trim(struct sk_buff
*skb
, unsigned int len
)
2175 return (len
< skb
->len
) ? __pskb_trim(skb
, len
) : 0;
2179 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2180 * @skb: buffer to alter
2183 * This is identical to pskb_trim except that the caller knows that
2184 * the skb is not cloned so we should never get an error due to out-
2187 static inline void pskb_trim_unique(struct sk_buff
*skb
, unsigned int len
)
2189 int err
= pskb_trim(skb
, len
);
2194 * skb_orphan - orphan a buffer
2195 * @skb: buffer to orphan
2197 * If a buffer currently has an owner then we call the owner's
2198 * destructor function and make the @skb unowned. The buffer continues
2199 * to exist but is no longer charged to its former owner.
2201 static inline void skb_orphan(struct sk_buff
*skb
)
2203 if (skb
->destructor
) {
2204 skb
->destructor(skb
);
2205 skb
->destructor
= NULL
;
2213 * skb_orphan_frags - orphan the frags contained in a buffer
2214 * @skb: buffer to orphan frags from
2215 * @gfp_mask: allocation mask for replacement pages
2217 * For each frag in the SKB which needs a destructor (i.e. has an
2218 * owner) create a copy of that frag and release the original
2219 * page by calling the destructor.
2221 static inline int skb_orphan_frags(struct sk_buff
*skb
, gfp_t gfp_mask
)
2223 if (likely(!(skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
)))
2225 return skb_copy_ubufs(skb
, gfp_mask
);
2229 * __skb_queue_purge - empty a list
2230 * @list: list to empty
2232 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2233 * the list and one reference dropped. This function does not take the
2234 * list lock and the caller must hold the relevant locks to use it.
2236 void skb_queue_purge(struct sk_buff_head
*list
);
2237 static inline void __skb_queue_purge(struct sk_buff_head
*list
)
2239 struct sk_buff
*skb
;
2240 while ((skb
= __skb_dequeue(list
)) != NULL
)
2244 void *netdev_alloc_frag(unsigned int fragsz
);
2246 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
, unsigned int length
,
2250 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2251 * @dev: network device to receive on
2252 * @length: length to allocate
2254 * Allocate a new &sk_buff and assign it a usage count of one. The
2255 * buffer has unspecified headroom built in. Users should allocate
2256 * the headroom they think they need without accounting for the
2257 * built in space. The built in space is used for optimisations.
2259 * %NULL is returned if there is no free memory. Although this function
2260 * allocates memory it can be called from an interrupt.
2262 static inline struct sk_buff
*netdev_alloc_skb(struct net_device
*dev
,
2263 unsigned int length
)
2265 return __netdev_alloc_skb(dev
, length
, GFP_ATOMIC
);
2268 /* legacy helper around __netdev_alloc_skb() */
2269 static inline struct sk_buff
*__dev_alloc_skb(unsigned int length
,
2272 return __netdev_alloc_skb(NULL
, length
, gfp_mask
);
2275 /* legacy helper around netdev_alloc_skb() */
2276 static inline struct sk_buff
*dev_alloc_skb(unsigned int length
)
2278 return netdev_alloc_skb(NULL
, length
);
2282 static inline struct sk_buff
*__netdev_alloc_skb_ip_align(struct net_device
*dev
,
2283 unsigned int length
, gfp_t gfp
)
2285 struct sk_buff
*skb
= __netdev_alloc_skb(dev
, length
+ NET_IP_ALIGN
, gfp
);
2287 if (NET_IP_ALIGN
&& skb
)
2288 skb_reserve(skb
, NET_IP_ALIGN
);
2292 static inline struct sk_buff
*netdev_alloc_skb_ip_align(struct net_device
*dev
,
2293 unsigned int length
)
2295 return __netdev_alloc_skb_ip_align(dev
, length
, GFP_ATOMIC
);
2298 static inline void skb_free_frag(void *addr
)
2300 __free_page_frag(addr
);
2303 void *napi_alloc_frag(unsigned int fragsz
);
2304 struct sk_buff
*__napi_alloc_skb(struct napi_struct
*napi
,
2305 unsigned int length
, gfp_t gfp_mask
);
2306 static inline struct sk_buff
*napi_alloc_skb(struct napi_struct
*napi
,
2307 unsigned int length
)
2309 return __napi_alloc_skb(napi
, length
, GFP_ATOMIC
);
2313 * __dev_alloc_pages - allocate page for network Rx
2314 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2315 * @order: size of the allocation
2317 * Allocate a new page.
2319 * %NULL is returned if there is no free memory.
2321 static inline struct page
*__dev_alloc_pages(gfp_t gfp_mask
,
2324 /* This piece of code contains several assumptions.
2325 * 1. This is for device Rx, therefor a cold page is preferred.
2326 * 2. The expectation is the user wants a compound page.
2327 * 3. If requesting a order 0 page it will not be compound
2328 * due to the check to see if order has a value in prep_new_page
2329 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2330 * code in gfp_to_alloc_flags that should be enforcing this.
2332 gfp_mask
|= __GFP_COLD
| __GFP_COMP
| __GFP_MEMALLOC
;
2334 return alloc_pages_node(NUMA_NO_NODE
, gfp_mask
, order
);
2337 static inline struct page
*dev_alloc_pages(unsigned int order
)
2339 return __dev_alloc_pages(GFP_ATOMIC
, order
);
2343 * __dev_alloc_page - allocate a page for network Rx
2344 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2346 * Allocate a new page.
2348 * %NULL is returned if there is no free memory.
2350 static inline struct page
*__dev_alloc_page(gfp_t gfp_mask
)
2352 return __dev_alloc_pages(gfp_mask
, 0);
2355 static inline struct page
*dev_alloc_page(void)
2357 return __dev_alloc_page(GFP_ATOMIC
);
2361 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2362 * @page: The page that was allocated from skb_alloc_page
2363 * @skb: The skb that may need pfmemalloc set
2365 static inline void skb_propagate_pfmemalloc(struct page
*page
,
2366 struct sk_buff
*skb
)
2368 if (page_is_pfmemalloc(page
))
2369 skb
->pfmemalloc
= true;
2373 * skb_frag_page - retrieve the page referred to by a paged fragment
2374 * @frag: the paged fragment
2376 * Returns the &struct page associated with @frag.
2378 static inline struct page
*skb_frag_page(const skb_frag_t
*frag
)
2380 return frag
->page
.p
;
2384 * __skb_frag_ref - take an addition reference on a paged fragment.
2385 * @frag: the paged fragment
2387 * Takes an additional reference on the paged fragment @frag.
2389 static inline void __skb_frag_ref(skb_frag_t
*frag
)
2391 get_page(skb_frag_page(frag
));
2395 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2397 * @f: the fragment offset.
2399 * Takes an additional reference on the @f'th paged fragment of @skb.
2401 static inline void skb_frag_ref(struct sk_buff
*skb
, int f
)
2403 __skb_frag_ref(&skb_shinfo(skb
)->frags
[f
]);
2407 * __skb_frag_unref - release a reference on a paged fragment.
2408 * @frag: the paged fragment
2410 * Releases a reference on the paged fragment @frag.
2412 static inline void __skb_frag_unref(skb_frag_t
*frag
)
2414 put_page(skb_frag_page(frag
));
2418 * skb_frag_unref - release a reference on a paged fragment of an skb.
2420 * @f: the fragment offset
2422 * Releases a reference on the @f'th paged fragment of @skb.
2424 static inline void skb_frag_unref(struct sk_buff
*skb
, int f
)
2426 __skb_frag_unref(&skb_shinfo(skb
)->frags
[f
]);
2430 * skb_frag_address - gets the address of the data contained in a paged fragment
2431 * @frag: the paged fragment buffer
2433 * Returns the address of the data within @frag. The page must already
2436 static inline void *skb_frag_address(const skb_frag_t
*frag
)
2438 return page_address(skb_frag_page(frag
)) + frag
->page_offset
;
2442 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2443 * @frag: the paged fragment buffer
2445 * Returns the address of the data within @frag. Checks that the page
2446 * is mapped and returns %NULL otherwise.
2448 static inline void *skb_frag_address_safe(const skb_frag_t
*frag
)
2450 void *ptr
= page_address(skb_frag_page(frag
));
2454 return ptr
+ frag
->page_offset
;
2458 * __skb_frag_set_page - sets the page contained in a paged fragment
2459 * @frag: the paged fragment
2460 * @page: the page to set
2462 * Sets the fragment @frag to contain @page.
2464 static inline void __skb_frag_set_page(skb_frag_t
*frag
, struct page
*page
)
2466 frag
->page
.p
= page
;
2470 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2472 * @f: the fragment offset
2473 * @page: the page to set
2475 * Sets the @f'th fragment of @skb to contain @page.
2477 static inline void skb_frag_set_page(struct sk_buff
*skb
, int f
,
2480 __skb_frag_set_page(&skb_shinfo(skb
)->frags
[f
], page
);
2483 bool skb_page_frag_refill(unsigned int sz
, struct page_frag
*pfrag
, gfp_t prio
);
2486 * skb_frag_dma_map - maps a paged fragment via the DMA API
2487 * @dev: the device to map the fragment to
2488 * @frag: the paged fragment to map
2489 * @offset: the offset within the fragment (starting at the
2490 * fragment's own offset)
2491 * @size: the number of bytes to map
2492 * @dir: the direction of the mapping (%PCI_DMA_*)
2494 * Maps the page associated with @frag to @device.
2496 static inline dma_addr_t
skb_frag_dma_map(struct device
*dev
,
2497 const skb_frag_t
*frag
,
2498 size_t offset
, size_t size
,
2499 enum dma_data_direction dir
)
2501 return dma_map_page(dev
, skb_frag_page(frag
),
2502 frag
->page_offset
+ offset
, size
, dir
);
2505 static inline struct sk_buff
*pskb_copy(struct sk_buff
*skb
,
2508 return __pskb_copy(skb
, skb_headroom(skb
), gfp_mask
);
2512 static inline struct sk_buff
*pskb_copy_for_clone(struct sk_buff
*skb
,
2515 return __pskb_copy_fclone(skb
, skb_headroom(skb
), gfp_mask
, true);
2520 * skb_clone_writable - is the header of a clone writable
2521 * @skb: buffer to check
2522 * @len: length up to which to write
2524 * Returns true if modifying the header part of the cloned buffer
2525 * does not requires the data to be copied.
2527 static inline int skb_clone_writable(const struct sk_buff
*skb
, unsigned int len
)
2529 return !skb_header_cloned(skb
) &&
2530 skb_headroom(skb
) + len
<= skb
->hdr_len
;
2533 static inline int __skb_cow(struct sk_buff
*skb
, unsigned int headroom
,
2538 if (headroom
> skb_headroom(skb
))
2539 delta
= headroom
- skb_headroom(skb
);
2541 if (delta
|| cloned
)
2542 return pskb_expand_head(skb
, ALIGN(delta
, NET_SKB_PAD
), 0,
2548 * skb_cow - copy header of skb when it is required
2549 * @skb: buffer to cow
2550 * @headroom: needed headroom
2552 * If the skb passed lacks sufficient headroom or its data part
2553 * is shared, data is reallocated. If reallocation fails, an error
2554 * is returned and original skb is not changed.
2556 * The result is skb with writable area skb->head...skb->tail
2557 * and at least @headroom of space at head.
2559 static inline int skb_cow(struct sk_buff
*skb
, unsigned int headroom
)
2561 return __skb_cow(skb
, headroom
, skb_cloned(skb
));
2565 * skb_cow_head - skb_cow but only making the head writable
2566 * @skb: buffer to cow
2567 * @headroom: needed headroom
2569 * This function is identical to skb_cow except that we replace the
2570 * skb_cloned check by skb_header_cloned. It should be used when
2571 * you only need to push on some header and do not need to modify
2574 static inline int skb_cow_head(struct sk_buff
*skb
, unsigned int headroom
)
2576 return __skb_cow(skb
, headroom
, skb_header_cloned(skb
));
2580 * skb_padto - pad an skbuff up to a minimal size
2581 * @skb: buffer to pad
2582 * @len: minimal length
2584 * Pads up a buffer to ensure the trailing bytes exist and are
2585 * blanked. If the buffer already contains sufficient data it
2586 * is untouched. Otherwise it is extended. Returns zero on
2587 * success. The skb is freed on error.
2589 static inline int skb_padto(struct sk_buff
*skb
, unsigned int len
)
2591 unsigned int size
= skb
->len
;
2592 if (likely(size
>= len
))
2594 return skb_pad(skb
, len
- size
);
2598 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2599 * @skb: buffer to pad
2600 * @len: minimal length
2602 * Pads up a buffer to ensure the trailing bytes exist and are
2603 * blanked. If the buffer already contains sufficient data it
2604 * is untouched. Otherwise it is extended. Returns zero on
2605 * success. The skb is freed on error.
2607 static inline int skb_put_padto(struct sk_buff
*skb
, unsigned int len
)
2609 unsigned int size
= skb
->len
;
2611 if (unlikely(size
< len
)) {
2613 if (skb_pad(skb
, len
))
2615 __skb_put(skb
, len
);
2620 static inline int skb_add_data(struct sk_buff
*skb
,
2621 struct iov_iter
*from
, int copy
)
2623 const int off
= skb
->len
;
2625 if (skb
->ip_summed
== CHECKSUM_NONE
) {
2627 if (csum_and_copy_from_iter(skb_put(skb
, copy
), copy
,
2628 &csum
, from
) == copy
) {
2629 skb
->csum
= csum_block_add(skb
->csum
, csum
, off
);
2632 } else if (copy_from_iter(skb_put(skb
, copy
), copy
, from
) == copy
)
2635 __skb_trim(skb
, off
);
2639 static inline bool skb_can_coalesce(struct sk_buff
*skb
, int i
,
2640 const struct page
*page
, int off
)
2643 const struct skb_frag_struct
*frag
= &skb_shinfo(skb
)->frags
[i
- 1];
2645 return page
== skb_frag_page(frag
) &&
2646 off
== frag
->page_offset
+ skb_frag_size(frag
);
2651 static inline int __skb_linearize(struct sk_buff
*skb
)
2653 return __pskb_pull_tail(skb
, skb
->data_len
) ? 0 : -ENOMEM
;
2657 * skb_linearize - convert paged skb to linear one
2658 * @skb: buffer to linarize
2660 * If there is no free memory -ENOMEM is returned, otherwise zero
2661 * is returned and the old skb data released.
2663 static inline int skb_linearize(struct sk_buff
*skb
)
2665 return skb_is_nonlinear(skb
) ? __skb_linearize(skb
) : 0;
2669 * skb_has_shared_frag - can any frag be overwritten
2670 * @skb: buffer to test
2672 * Return true if the skb has at least one frag that might be modified
2673 * by an external entity (as in vmsplice()/sendfile())
2675 static inline bool skb_has_shared_frag(const struct sk_buff
*skb
)
2677 return skb_is_nonlinear(skb
) &&
2678 skb_shinfo(skb
)->tx_flags
& SKBTX_SHARED_FRAG
;
2682 * skb_linearize_cow - make sure skb is linear and writable
2683 * @skb: buffer to process
2685 * If there is no free memory -ENOMEM is returned, otherwise zero
2686 * is returned and the old skb data released.
2688 static inline int skb_linearize_cow(struct sk_buff
*skb
)
2690 return skb_is_nonlinear(skb
) || skb_cloned(skb
) ?
2691 __skb_linearize(skb
) : 0;
2695 * skb_postpull_rcsum - update checksum for received skb after pull
2696 * @skb: buffer to update
2697 * @start: start of data before pull
2698 * @len: length of data pulled
2700 * After doing a pull on a received packet, you need to call this to
2701 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2702 * CHECKSUM_NONE so that it can be recomputed from scratch.
2705 static inline void skb_postpull_rcsum(struct sk_buff
*skb
,
2706 const void *start
, unsigned int len
)
2708 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2709 skb
->csum
= csum_sub(skb
->csum
, csum_partial(start
, len
, 0));
2712 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
);
2715 * pskb_trim_rcsum - trim received skb and update checksum
2716 * @skb: buffer to trim
2719 * This is exactly the same as pskb_trim except that it ensures the
2720 * checksum of received packets are still valid after the operation.
2723 static inline int pskb_trim_rcsum(struct sk_buff
*skb
, unsigned int len
)
2725 if (likely(len
>= skb
->len
))
2727 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
2728 skb
->ip_summed
= CHECKSUM_NONE
;
2729 return __pskb_trim(skb
, len
);
2732 #define skb_queue_walk(queue, skb) \
2733 for (skb = (queue)->next; \
2734 skb != (struct sk_buff *)(queue); \
2737 #define skb_queue_walk_safe(queue, skb, tmp) \
2738 for (skb = (queue)->next, tmp = skb->next; \
2739 skb != (struct sk_buff *)(queue); \
2740 skb = tmp, tmp = skb->next)
2742 #define skb_queue_walk_from(queue, skb) \
2743 for (; skb != (struct sk_buff *)(queue); \
2746 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2747 for (tmp = skb->next; \
2748 skb != (struct sk_buff *)(queue); \
2749 skb = tmp, tmp = skb->next)
2751 #define skb_queue_reverse_walk(queue, skb) \
2752 for (skb = (queue)->prev; \
2753 skb != (struct sk_buff *)(queue); \
2756 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2757 for (skb = (queue)->prev, tmp = skb->prev; \
2758 skb != (struct sk_buff *)(queue); \
2759 skb = tmp, tmp = skb->prev)
2761 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2762 for (tmp = skb->prev; \
2763 skb != (struct sk_buff *)(queue); \
2764 skb = tmp, tmp = skb->prev)
2766 static inline bool skb_has_frag_list(const struct sk_buff
*skb
)
2768 return skb_shinfo(skb
)->frag_list
!= NULL
;
2771 static inline void skb_frag_list_init(struct sk_buff
*skb
)
2773 skb_shinfo(skb
)->frag_list
= NULL
;
2776 #define skb_walk_frags(skb, iter) \
2777 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2779 struct sk_buff
*__skb_recv_datagram(struct sock
*sk
, unsigned flags
,
2780 int *peeked
, int *off
, int *err
);
2781 struct sk_buff
*skb_recv_datagram(struct sock
*sk
, unsigned flags
, int noblock
,
2783 unsigned int datagram_poll(struct file
*file
, struct socket
*sock
,
2784 struct poll_table_struct
*wait
);
2785 int skb_copy_datagram_iter(const struct sk_buff
*from
, int offset
,
2786 struct iov_iter
*to
, int size
);
2787 static inline int skb_copy_datagram_msg(const struct sk_buff
*from
, int offset
,
2788 struct msghdr
*msg
, int size
)
2790 return skb_copy_datagram_iter(from
, offset
, &msg
->msg_iter
, size
);
2792 int skb_copy_and_csum_datagram_msg(struct sk_buff
*skb
, int hlen
,
2793 struct msghdr
*msg
);
2794 int skb_copy_datagram_from_iter(struct sk_buff
*skb
, int offset
,
2795 struct iov_iter
*from
, int len
);
2796 int zerocopy_sg_from_iter(struct sk_buff
*skb
, struct iov_iter
*frm
);
2797 void skb_free_datagram(struct sock
*sk
, struct sk_buff
*skb
);
2798 void skb_free_datagram_locked(struct sock
*sk
, struct sk_buff
*skb
);
2799 int skb_kill_datagram(struct sock
*sk
, struct sk_buff
*skb
, unsigned int flags
);
2800 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
);
2801 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
);
2802 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
, u8
*to
,
2803 int len
, __wsum csum
);
2804 ssize_t
skb_socket_splice(struct sock
*sk
,
2805 struct pipe_inode_info
*pipe
,
2806 struct splice_pipe_desc
*spd
);
2807 int skb_splice_bits(struct sk_buff
*skb
, struct sock
*sk
, unsigned int offset
,
2808 struct pipe_inode_info
*pipe
, unsigned int len
,
2810 ssize_t (*splice_cb
)(struct sock
*,
2811 struct pipe_inode_info
*,
2812 struct splice_pipe_desc
*));
2813 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
);
2814 unsigned int skb_zerocopy_headlen(const struct sk_buff
*from
);
2815 int skb_zerocopy(struct sk_buff
*to
, struct sk_buff
*from
,
2817 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
);
2818 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
);
2819 void skb_scrub_packet(struct sk_buff
*skb
, bool xnet
);
2820 unsigned int skb_gso_transport_seglen(const struct sk_buff
*skb
);
2821 struct sk_buff
*skb_segment(struct sk_buff
*skb
, netdev_features_t features
);
2822 struct sk_buff
*skb_vlan_untag(struct sk_buff
*skb
);
2823 int skb_ensure_writable(struct sk_buff
*skb
, int write_len
);
2824 int skb_vlan_pop(struct sk_buff
*skb
);
2825 int skb_vlan_push(struct sk_buff
*skb
, __be16 vlan_proto
, u16 vlan_tci
);
2827 static inline int memcpy_from_msg(void *data
, struct msghdr
*msg
, int len
)
2829 return copy_from_iter(data
, len
, &msg
->msg_iter
) == len
? 0 : -EFAULT
;
2832 static inline int memcpy_to_msg(struct msghdr
*msg
, void *data
, int len
)
2834 return copy_to_iter(data
, len
, &msg
->msg_iter
) == len
? 0 : -EFAULT
;
2837 struct skb_checksum_ops
{
2838 __wsum (*update
)(const void *mem
, int len
, __wsum wsum
);
2839 __wsum (*combine
)(__wsum csum
, __wsum csum2
, int offset
, int len
);
2842 __wsum
__skb_checksum(const struct sk_buff
*skb
, int offset
, int len
,
2843 __wsum csum
, const struct skb_checksum_ops
*ops
);
2844 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
, int len
,
2847 static inline void * __must_check
2848 __skb_header_pointer(const struct sk_buff
*skb
, int offset
,
2849 int len
, void *data
, int hlen
, void *buffer
)
2851 if (hlen
- offset
>= len
)
2852 return data
+ offset
;
2855 skb_copy_bits(skb
, offset
, buffer
, len
) < 0)
2861 static inline void * __must_check
2862 skb_header_pointer(const struct sk_buff
*skb
, int offset
, int len
, void *buffer
)
2864 return __skb_header_pointer(skb
, offset
, len
, skb
->data
,
2865 skb_headlen(skb
), buffer
);
2869 * skb_needs_linearize - check if we need to linearize a given skb
2870 * depending on the given device features.
2871 * @skb: socket buffer to check
2872 * @features: net device features
2874 * Returns true if either:
2875 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2876 * 2. skb is fragmented and the device does not support SG.
2878 static inline bool skb_needs_linearize(struct sk_buff
*skb
,
2879 netdev_features_t features
)
2881 return skb_is_nonlinear(skb
) &&
2882 ((skb_has_frag_list(skb
) && !(features
& NETIF_F_FRAGLIST
)) ||
2883 (skb_shinfo(skb
)->nr_frags
&& !(features
& NETIF_F_SG
)));
2886 static inline void skb_copy_from_linear_data(const struct sk_buff
*skb
,
2888 const unsigned int len
)
2890 memcpy(to
, skb
->data
, len
);
2893 static inline void skb_copy_from_linear_data_offset(const struct sk_buff
*skb
,
2894 const int offset
, void *to
,
2895 const unsigned int len
)
2897 memcpy(to
, skb
->data
+ offset
, len
);
2900 static inline void skb_copy_to_linear_data(struct sk_buff
*skb
,
2902 const unsigned int len
)
2904 memcpy(skb
->data
, from
, len
);
2907 static inline void skb_copy_to_linear_data_offset(struct sk_buff
*skb
,
2910 const unsigned int len
)
2912 memcpy(skb
->data
+ offset
, from
, len
);
2915 void skb_init(void);
2917 static inline ktime_t
skb_get_ktime(const struct sk_buff
*skb
)
2923 * skb_get_timestamp - get timestamp from a skb
2924 * @skb: skb to get stamp from
2925 * @stamp: pointer to struct timeval to store stamp in
2927 * Timestamps are stored in the skb as offsets to a base timestamp.
2928 * This function converts the offset back to a struct timeval and stores
2931 static inline void skb_get_timestamp(const struct sk_buff
*skb
,
2932 struct timeval
*stamp
)
2934 *stamp
= ktime_to_timeval(skb
->tstamp
);
2937 static inline void skb_get_timestampns(const struct sk_buff
*skb
,
2938 struct timespec
*stamp
)
2940 *stamp
= ktime_to_timespec(skb
->tstamp
);
2943 static inline void __net_timestamp(struct sk_buff
*skb
)
2945 skb
->tstamp
= ktime_get_real();
2948 static inline ktime_t
net_timedelta(ktime_t t
)
2950 return ktime_sub(ktime_get_real(), t
);
2953 static inline ktime_t
net_invalid_timestamp(void)
2955 return ktime_set(0, 0);
2958 struct sk_buff
*skb_clone_sk(struct sk_buff
*skb
);
2960 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2962 void skb_clone_tx_timestamp(struct sk_buff
*skb
);
2963 bool skb_defer_rx_timestamp(struct sk_buff
*skb
);
2965 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2967 static inline void skb_clone_tx_timestamp(struct sk_buff
*skb
)
2971 static inline bool skb_defer_rx_timestamp(struct sk_buff
*skb
)
2976 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2979 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2981 * PHY drivers may accept clones of transmitted packets for
2982 * timestamping via their phy_driver.txtstamp method. These drivers
2983 * must call this function to return the skb back to the stack with a
2986 * @skb: clone of the the original outgoing packet
2987 * @hwtstamps: hardware time stamps
2990 void skb_complete_tx_timestamp(struct sk_buff
*skb
,
2991 struct skb_shared_hwtstamps
*hwtstamps
);
2993 void __skb_tstamp_tx(struct sk_buff
*orig_skb
,
2994 struct skb_shared_hwtstamps
*hwtstamps
,
2995 struct sock
*sk
, int tstype
);
2998 * skb_tstamp_tx - queue clone of skb with send time stamps
2999 * @orig_skb: the original outgoing packet
3000 * @hwtstamps: hardware time stamps, may be NULL if not available
3002 * If the skb has a socket associated, then this function clones the
3003 * skb (thus sharing the actual data and optional structures), stores
3004 * the optional hardware time stamping information (if non NULL) or
3005 * generates a software time stamp (otherwise), then queues the clone
3006 * to the error queue of the socket. Errors are silently ignored.
3008 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3009 struct skb_shared_hwtstamps
*hwtstamps
);
3011 static inline void sw_tx_timestamp(struct sk_buff
*skb
)
3013 if (skb_shinfo(skb
)->tx_flags
& SKBTX_SW_TSTAMP
&&
3014 !(skb_shinfo(skb
)->tx_flags
& SKBTX_IN_PROGRESS
))
3015 skb_tstamp_tx(skb
, NULL
);
3019 * skb_tx_timestamp() - Driver hook for transmit timestamping
3021 * Ethernet MAC Drivers should call this function in their hard_xmit()
3022 * function immediately before giving the sk_buff to the MAC hardware.
3024 * Specifically, one should make absolutely sure that this function is
3025 * called before TX completion of this packet can trigger. Otherwise
3026 * the packet could potentially already be freed.
3028 * @skb: A socket buffer.
3030 static inline void skb_tx_timestamp(struct sk_buff
*skb
)
3032 skb_clone_tx_timestamp(skb
);
3033 sw_tx_timestamp(skb
);
3037 * skb_complete_wifi_ack - deliver skb with wifi status
3039 * @skb: the original outgoing packet
3040 * @acked: ack status
3043 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
);
3045 __sum16
__skb_checksum_complete_head(struct sk_buff
*skb
, int len
);
3046 __sum16
__skb_checksum_complete(struct sk_buff
*skb
);
3048 static inline int skb_csum_unnecessary(const struct sk_buff
*skb
)
3050 return ((skb
->ip_summed
== CHECKSUM_UNNECESSARY
) ||
3052 (skb
->ip_summed
== CHECKSUM_PARTIAL
&&
3053 skb_checksum_start_offset(skb
) >= 0));
3057 * skb_checksum_complete - Calculate checksum of an entire packet
3058 * @skb: packet to process
3060 * This function calculates the checksum over the entire packet plus
3061 * the value of skb->csum. The latter can be used to supply the
3062 * checksum of a pseudo header as used by TCP/UDP. It returns the
3065 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3066 * this function can be used to verify that checksum on received
3067 * packets. In that case the function should return zero if the
3068 * checksum is correct. In particular, this function will return zero
3069 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3070 * hardware has already verified the correctness of the checksum.
3072 static inline __sum16
skb_checksum_complete(struct sk_buff
*skb
)
3074 return skb_csum_unnecessary(skb
) ?
3075 0 : __skb_checksum_complete(skb
);
3078 static inline void __skb_decr_checksum_unnecessary(struct sk_buff
*skb
)
3080 if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
3081 if (skb
->csum_level
== 0)
3082 skb
->ip_summed
= CHECKSUM_NONE
;
3088 static inline void __skb_incr_checksum_unnecessary(struct sk_buff
*skb
)
3090 if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
3091 if (skb
->csum_level
< SKB_MAX_CSUM_LEVEL
)
3093 } else if (skb
->ip_summed
== CHECKSUM_NONE
) {
3094 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
3095 skb
->csum_level
= 0;
3099 static inline void __skb_mark_checksum_bad(struct sk_buff
*skb
)
3101 /* Mark current checksum as bad (typically called from GRO
3102 * path). In the case that ip_summed is CHECKSUM_NONE
3103 * this must be the first checksum encountered in the packet.
3104 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3105 * checksum after the last one validated. For UDP, a zero
3106 * checksum can not be marked as bad.
3109 if (skb
->ip_summed
== CHECKSUM_NONE
||
3110 skb
->ip_summed
== CHECKSUM_UNNECESSARY
)
3114 /* Check if we need to perform checksum complete validation.
3116 * Returns true if checksum complete is needed, false otherwise
3117 * (either checksum is unnecessary or zero checksum is allowed).
3119 static inline bool __skb_checksum_validate_needed(struct sk_buff
*skb
,
3123 if (skb_csum_unnecessary(skb
) || (zero_okay
&& !check
)) {
3124 skb
->csum_valid
= 1;
3125 __skb_decr_checksum_unnecessary(skb
);
3132 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3135 #define CHECKSUM_BREAK 76
3137 /* Unset checksum-complete
3139 * Unset checksum complete can be done when packet is being modified
3140 * (uncompressed for instance) and checksum-complete value is
3143 static inline void skb_checksum_complete_unset(struct sk_buff
*skb
)
3145 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
3146 skb
->ip_summed
= CHECKSUM_NONE
;
3149 /* Validate (init) checksum based on checksum complete.
3152 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3153 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3154 * checksum is stored in skb->csum for use in __skb_checksum_complete
3155 * non-zero: value of invalid checksum
3158 static inline __sum16
__skb_checksum_validate_complete(struct sk_buff
*skb
,
3162 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
3163 if (!csum_fold(csum_add(psum
, skb
->csum
))) {
3164 skb
->csum_valid
= 1;
3167 } else if (skb
->csum_bad
) {
3168 /* ip_summed == CHECKSUM_NONE in this case */
3169 return (__force __sum16
)1;
3174 if (complete
|| skb
->len
<= CHECKSUM_BREAK
) {
3177 csum
= __skb_checksum_complete(skb
);
3178 skb
->csum_valid
= !csum
;
3185 static inline __wsum
null_compute_pseudo(struct sk_buff
*skb
, int proto
)
3190 /* Perform checksum validate (init). Note that this is a macro since we only
3191 * want to calculate the pseudo header which is an input function if necessary.
3192 * First we try to validate without any computation (checksum unnecessary) and
3193 * then calculate based on checksum complete calling the function to compute
3197 * 0: checksum is validated or try to in skb_checksum_complete
3198 * non-zero: value of invalid checksum
3200 #define __skb_checksum_validate(skb, proto, complete, \
3201 zero_okay, check, compute_pseudo) \
3203 __sum16 __ret = 0; \
3204 skb->csum_valid = 0; \
3205 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3206 __ret = __skb_checksum_validate_complete(skb, \
3207 complete, compute_pseudo(skb, proto)); \
3211 #define skb_checksum_init(skb, proto, compute_pseudo) \
3212 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3214 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3215 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3217 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3218 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3220 #define skb_checksum_validate_zero_check(skb, proto, check, \
3222 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3224 #define skb_checksum_simple_validate(skb) \
3225 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3227 static inline bool __skb_checksum_convert_check(struct sk_buff
*skb
)
3229 return (skb
->ip_summed
== CHECKSUM_NONE
&&
3230 skb
->csum_valid
&& !skb
->csum_bad
);
3233 static inline void __skb_checksum_convert(struct sk_buff
*skb
,
3234 __sum16 check
, __wsum pseudo
)
3236 skb
->csum
= ~pseudo
;
3237 skb
->ip_summed
= CHECKSUM_COMPLETE
;
3240 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3242 if (__skb_checksum_convert_check(skb)) \
3243 __skb_checksum_convert(skb, check, \
3244 compute_pseudo(skb, proto)); \
3247 static inline void skb_remcsum_adjust_partial(struct sk_buff
*skb
, void *ptr
,
3248 u16 start
, u16 offset
)
3250 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3251 skb
->csum_start
= ((unsigned char *)ptr
+ start
) - skb
->head
;
3252 skb
->csum_offset
= offset
- start
;
3255 /* Update skbuf and packet to reflect the remote checksum offload operation.
3256 * When called, ptr indicates the starting point for skb->csum when
3257 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3258 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3260 static inline void skb_remcsum_process(struct sk_buff
*skb
, void *ptr
,
3261 int start
, int offset
, bool nopartial
)
3266 skb_remcsum_adjust_partial(skb
, ptr
, start
, offset
);
3270 if (unlikely(skb
->ip_summed
!= CHECKSUM_COMPLETE
)) {
3271 __skb_checksum_complete(skb
);
3272 skb_postpull_rcsum(skb
, skb
->data
, ptr
- (void *)skb
->data
);
3275 delta
= remcsum_adjust(ptr
, skb
->csum
, start
, offset
);
3277 /* Adjust skb->csum since we changed the packet */
3278 skb
->csum
= csum_add(skb
->csum
, delta
);
3281 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3282 void nf_conntrack_destroy(struct nf_conntrack
*nfct
);
3283 static inline void nf_conntrack_put(struct nf_conntrack
*nfct
)
3285 if (nfct
&& atomic_dec_and_test(&nfct
->use
))
3286 nf_conntrack_destroy(nfct
);
3288 static inline void nf_conntrack_get(struct nf_conntrack
*nfct
)
3291 atomic_inc(&nfct
->use
);
3294 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3295 static inline void nf_bridge_put(struct nf_bridge_info
*nf_bridge
)
3297 if (nf_bridge
&& atomic_dec_and_test(&nf_bridge
->use
))
3300 static inline void nf_bridge_get(struct nf_bridge_info
*nf_bridge
)
3303 atomic_inc(&nf_bridge
->use
);
3305 #endif /* CONFIG_BRIDGE_NETFILTER */
3306 static inline void nf_reset(struct sk_buff
*skb
)
3308 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3309 nf_conntrack_put(skb
->nfct
);
3312 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3313 nf_bridge_put(skb
->nf_bridge
);
3314 skb
->nf_bridge
= NULL
;
3318 static inline void nf_reset_trace(struct sk_buff
*skb
)
3320 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3325 /* Note: This doesn't put any conntrack and bridge info in dst. */
3326 static inline void __nf_copy(struct sk_buff
*dst
, const struct sk_buff
*src
,
3329 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3330 dst
->nfct
= src
->nfct
;
3331 nf_conntrack_get(src
->nfct
);
3333 dst
->nfctinfo
= src
->nfctinfo
;
3335 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3336 dst
->nf_bridge
= src
->nf_bridge
;
3337 nf_bridge_get(src
->nf_bridge
);
3339 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3341 dst
->nf_trace
= src
->nf_trace
;
3345 static inline void nf_copy(struct sk_buff
*dst
, const struct sk_buff
*src
)
3347 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3348 nf_conntrack_put(dst
->nfct
);
3350 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3351 nf_bridge_put(dst
->nf_bridge
);
3353 __nf_copy(dst
, src
, true);
3356 #ifdef CONFIG_NETWORK_SECMARK
3357 static inline void skb_copy_secmark(struct sk_buff
*to
, const struct sk_buff
*from
)
3359 to
->secmark
= from
->secmark
;
3362 static inline void skb_init_secmark(struct sk_buff
*skb
)
3367 static inline void skb_copy_secmark(struct sk_buff
*to
, const struct sk_buff
*from
)
3370 static inline void skb_init_secmark(struct sk_buff
*skb
)
3374 static inline bool skb_irq_freeable(const struct sk_buff
*skb
)
3376 return !skb
->destructor
&&
3377 #if IS_ENABLED(CONFIG_XFRM)
3380 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3383 !skb
->_skb_refdst
&&
3384 !skb_has_frag_list(skb
);
3387 static inline void skb_set_queue_mapping(struct sk_buff
*skb
, u16 queue_mapping
)
3389 skb
->queue_mapping
= queue_mapping
;
3392 static inline u16
skb_get_queue_mapping(const struct sk_buff
*skb
)
3394 return skb
->queue_mapping
;
3397 static inline void skb_copy_queue_mapping(struct sk_buff
*to
, const struct sk_buff
*from
)
3399 to
->queue_mapping
= from
->queue_mapping
;
3402 static inline void skb_record_rx_queue(struct sk_buff
*skb
, u16 rx_queue
)
3404 skb
->queue_mapping
= rx_queue
+ 1;
3407 static inline u16
skb_get_rx_queue(const struct sk_buff
*skb
)
3409 return skb
->queue_mapping
- 1;
3412 static inline bool skb_rx_queue_recorded(const struct sk_buff
*skb
)
3414 return skb
->queue_mapping
!= 0;
3417 static inline struct sec_path
*skb_sec_path(struct sk_buff
*skb
)
3426 /* Keeps track of mac header offset relative to skb->head.
3427 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3428 * For non-tunnel skb it points to skb_mac_header() and for
3429 * tunnel skb it points to outer mac header.
3430 * Keeps track of level of encapsulation of network headers.
3437 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3439 static inline int skb_tnl_header_len(const struct sk_buff
*inner_skb
)
3441 return (skb_mac_header(inner_skb
) - inner_skb
->head
) -
3442 SKB_GSO_CB(inner_skb
)->mac_offset
;
3445 static inline int gso_pskb_expand_head(struct sk_buff
*skb
, int extra
)
3447 int new_headroom
, headroom
;
3450 headroom
= skb_headroom(skb
);
3451 ret
= pskb_expand_head(skb
, extra
, 0, GFP_ATOMIC
);
3455 new_headroom
= skb_headroom(skb
);
3456 SKB_GSO_CB(skb
)->mac_offset
+= (new_headroom
- headroom
);
3460 /* Compute the checksum for a gso segment. First compute the checksum value
3461 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3462 * then add in skb->csum (checksum from csum_start to end of packet).
3463 * skb->csum and csum_start are then updated to reflect the checksum of the
3464 * resultant packet starting from the transport header-- the resultant checksum
3465 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3468 static inline __sum16
gso_make_checksum(struct sk_buff
*skb
, __wsum res
)
3470 int plen
= SKB_GSO_CB(skb
)->csum_start
- skb_headroom(skb
) -
3471 skb_transport_offset(skb
);
3474 partial
= csum_partial(skb_transport_header(skb
), plen
, skb
->csum
);
3476 SKB_GSO_CB(skb
)->csum_start
-= plen
;
3478 return csum_fold(partial
);
3481 static inline bool skb_is_gso(const struct sk_buff
*skb
)
3483 return skb_shinfo(skb
)->gso_size
;
3486 /* Note: Should be called only if skb_is_gso(skb) is true */
3487 static inline bool skb_is_gso_v6(const struct sk_buff
*skb
)
3489 return skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
;
3492 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
);
3494 static inline bool skb_warn_if_lro(const struct sk_buff
*skb
)
3496 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3497 * wanted then gso_type will be set. */
3498 const struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
3500 if (skb_is_nonlinear(skb
) && shinfo
->gso_size
!= 0 &&
3501 unlikely(shinfo
->gso_type
== 0)) {
3502 __skb_warn_lro_forwarding(skb
);
3508 static inline void skb_forward_csum(struct sk_buff
*skb
)
3510 /* Unfortunately we don't support this one. Any brave souls? */
3511 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
3512 skb
->ip_summed
= CHECKSUM_NONE
;
3516 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3517 * @skb: skb to check
3519 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3520 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3521 * use this helper, to document places where we make this assertion.
3523 static inline void skb_checksum_none_assert(const struct sk_buff
*skb
)
3526 BUG_ON(skb
->ip_summed
!= CHECKSUM_NONE
);
3530 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
);
3532 int skb_checksum_setup(struct sk_buff
*skb
, bool recalculate
);
3533 struct sk_buff
*skb_checksum_trimmed(struct sk_buff
*skb
,
3534 unsigned int transport_len
,
3535 __sum16(*skb_chkf
)(struct sk_buff
*skb
));
3538 * skb_head_is_locked - Determine if the skb->head is locked down
3539 * @skb: skb to check
3541 * The head on skbs build around a head frag can be removed if they are
3542 * not cloned. This function returns true if the skb head is locked down
3543 * due to either being allocated via kmalloc, or by being a clone with
3544 * multiple references to the head.
3546 static inline bool skb_head_is_locked(const struct sk_buff
*skb
)
3548 return !skb
->head_frag
|| skb_cloned(skb
);
3552 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3556 * skb_gso_network_seglen is used to determine the real size of the
3557 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3559 * The MAC/L2 header is not accounted for.
3561 static inline unsigned int skb_gso_network_seglen(const struct sk_buff
*skb
)
3563 unsigned int hdr_len
= skb_transport_header(skb
) -
3564 skb_network_header(skb
);
3565 return hdr_len
+ skb_gso_transport_seglen(skb
);
3568 #endif /* __KERNEL__ */
3569 #endif /* _LINUX_SKBUFF_H */