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 /* always valid & non-NULL from FORWARD on, for physdev match */
184 struct net_device *physoutdev;
186 /* prerouting: detect dnat in orig/reply direction */
188 struct in6_addr ipv6_daddr;
190 /* after prerouting + nat detected: store original source
191 * mac since neigh resolution overwrites it, only used while
192 * skb is out in neigh layer.
194 char neigh_header[8];
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)
222 extern int sysctl_max_skb_frags;
224 typedef struct skb_frag_struct skb_frag_t;
226 struct skb_frag_struct {
230 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
239 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
244 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
249 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
254 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
259 #define HAVE_HW_TIME_STAMP
262 * struct skb_shared_hwtstamps - hardware time stamps
263 * @hwtstamp: hardware time stamp transformed into duration
264 * since arbitrary point in time
266 * Software time stamps generated by ktime_get_real() are stored in
269 * hwtstamps can only be compared against other hwtstamps from
272 * This structure is attached to packets as part of the
273 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
275 struct skb_shared_hwtstamps {
279 /* Definitions for tx_flags in struct skb_shared_info */
281 /* generate hardware time stamp */
282 SKBTX_HW_TSTAMP = 1 << 0,
284 /* generate software time stamp when queueing packet to NIC */
285 SKBTX_SW_TSTAMP = 1 << 1,
287 /* device driver is going to provide hardware time stamp */
288 SKBTX_IN_PROGRESS = 1 << 2,
290 /* device driver supports TX zero-copy buffers */
291 SKBTX_DEV_ZEROCOPY = 1 << 3,
293 /* generate wifi status information (where possible) */
294 SKBTX_WIFI_STATUS = 1 << 4,
296 /* This indicates at least one fragment might be overwritten
297 * (as in vmsplice(), sendfile() ...)
298 * If we need to compute a TX checksum, we'll need to copy
299 * all frags to avoid possible bad checksum
301 SKBTX_SHARED_FRAG = 1 << 5,
303 /* generate software time stamp when entering packet scheduling */
304 SKBTX_SCHED_TSTAMP = 1 << 6,
306 /* generate software timestamp on peer data acknowledgment */
307 SKBTX_ACK_TSTAMP = 1 << 7,
310 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
311 SKBTX_SCHED_TSTAMP | \
313 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
316 * The callback notifies userspace to release buffers when skb DMA is done in
317 * lower device, the skb last reference should be 0 when calling this.
318 * The zerocopy_success argument is true if zero copy transmit occurred,
319 * false on data copy or out of memory error caused by data copy attempt.
320 * The ctx field is used to track device context.
321 * The desc field is used to track userspace buffer index.
324 void (*callback)(struct ubuf_info *, bool zerocopy_success);
329 /* This data is invariant across clones and lives at
330 * the end of the header data, ie. at skb->end.
332 struct skb_shared_info {
333 unsigned char nr_frags;
335 unsigned short gso_size;
336 /* Warning: this field is not always filled in (UFO)! */
337 unsigned short gso_segs;
338 unsigned short gso_type;
339 struct sk_buff *frag_list;
340 struct skb_shared_hwtstamps hwtstamps;
345 * Warning : all fields before dataref are cleared in __alloc_skb()
349 /* Intermediate layers must ensure that destructor_arg
350 * remains valid until skb destructor */
351 void * destructor_arg;
353 /* must be last field, see pskb_expand_head() */
354 skb_frag_t frags[MAX_SKB_FRAGS];
357 /* We divide dataref into two halves. The higher 16 bits hold references
358 * to the payload part of skb->data. The lower 16 bits hold references to
359 * the entire skb->data. A clone of a headerless skb holds the length of
360 * the header in skb->hdr_len.
362 * All users must obey the rule that the skb->data reference count must be
363 * greater than or equal to the payload reference count.
365 * Holding a reference to the payload part means that the user does not
366 * care about modifications to the header part of skb->data.
368 #define SKB_DATAREF_SHIFT 16
369 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
373 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
374 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
375 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
379 SKB_GSO_TCPV4 = 1 << 0,
380 SKB_GSO_UDP = 1 << 1,
382 /* This indicates the skb is from an untrusted source. */
383 SKB_GSO_DODGY = 1 << 2,
385 /* This indicates the tcp segment has CWR set. */
386 SKB_GSO_TCP_ECN = 1 << 3,
388 SKB_GSO_TCPV6 = 1 << 4,
390 SKB_GSO_FCOE = 1 << 5,
392 SKB_GSO_GRE = 1 << 6,
394 SKB_GSO_GRE_CSUM = 1 << 7,
396 SKB_GSO_IPIP = 1 << 8,
398 SKB_GSO_SIT = 1 << 9,
400 SKB_GSO_UDP_TUNNEL = 1 << 10,
402 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
404 SKB_GSO_TUNNEL_REMCSUM = 1 << 12,
407 #if BITS_PER_LONG > 32
408 #define NET_SKBUFF_DATA_USES_OFFSET 1
411 #ifdef NET_SKBUFF_DATA_USES_OFFSET
412 typedef unsigned int sk_buff_data_t;
414 typedef unsigned char *sk_buff_data_t;
418 * struct skb_mstamp - multi resolution time stamps
419 * @stamp_us: timestamp in us resolution
420 * @stamp_jiffies: timestamp in jiffies
433 * skb_mstamp_get - get current timestamp
434 * @cl: place to store timestamps
436 static inline void skb_mstamp_get(struct skb_mstamp *cl)
438 u64 val = local_clock();
440 do_div(val, NSEC_PER_USEC);
441 cl->stamp_us = (u32)val;
442 cl->stamp_jiffies = (u32)jiffies;
446 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
447 * @t1: pointer to newest sample
448 * @t0: pointer to oldest sample
450 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
451 const struct skb_mstamp *t0)
453 s32 delta_us = t1->stamp_us - t0->stamp_us;
454 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
456 /* If delta_us is negative, this might be because interval is too big,
457 * or local_clock() drift is too big : fallback using jiffies.
460 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
462 delta_us = jiffies_to_usecs(delta_jiffies);
467 static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
468 const struct skb_mstamp *t0)
470 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
473 diff = t1->stamp_us - t0->stamp_us;
478 * struct sk_buff - socket buffer
479 * @next: Next buffer in list
480 * @prev: Previous buffer in list
481 * @tstamp: Time we arrived/left
482 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
483 * @sk: Socket we are owned by
484 * @dev: Device we arrived on/are leaving by
485 * @cb: Control buffer. Free for use by every layer. Put private vars here
486 * @_skb_refdst: destination entry (with norefcount bit)
487 * @sp: the security path, used for xfrm
488 * @len: Length of actual data
489 * @data_len: Data length
490 * @mac_len: Length of link layer header
491 * @hdr_len: writable header length of cloned skb
492 * @csum: Checksum (must include start/offset pair)
493 * @csum_start: Offset from skb->head where checksumming should start
494 * @csum_offset: Offset from csum_start where checksum should be stored
495 * @priority: Packet queueing priority
496 * @ignore_df: allow local fragmentation
497 * @cloned: Head may be cloned (check refcnt to be sure)
498 * @ip_summed: Driver fed us an IP checksum
499 * @nohdr: Payload reference only, must not modify header
500 * @nfctinfo: Relationship of this skb to the connection
501 * @pkt_type: Packet class
502 * @fclone: skbuff clone status
503 * @ipvs_property: skbuff is owned by ipvs
504 * @peeked: this packet has been seen already, so stats have been
505 * done for it, don't do them again
506 * @nf_trace: netfilter packet trace flag
507 * @protocol: Packet protocol from driver
508 * @destructor: Destruct function
509 * @nfct: Associated connection, if any
510 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
511 * @skb_iif: ifindex of device we arrived on
512 * @tc_index: Traffic control index
513 * @tc_verd: traffic control verdict
514 * @hash: the packet hash
515 * @queue_mapping: Queue mapping for multiqueue devices
516 * @xmit_more: More SKBs are pending for this queue
517 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
518 * @ndisc_nodetype: router type (from link layer)
519 * @ooo_okay: allow the mapping of a socket to a queue to be changed
520 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
522 * @sw_hash: indicates hash was computed in software stack
523 * @wifi_acked_valid: wifi_acked was set
524 * @wifi_acked: whether frame was acked on wifi or not
525 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
526 * @napi_id: id of the NAPI struct this skb came from
527 * @secmark: security marking
528 * @offload_fwd_mark: fwding offload mark
529 * @mark: Generic packet mark
530 * @vlan_proto: vlan encapsulation protocol
531 * @vlan_tci: vlan tag control information
532 * @inner_protocol: Protocol (encapsulation)
533 * @inner_transport_header: Inner transport layer header (encapsulation)
534 * @inner_network_header: Network layer header (encapsulation)
535 * @inner_mac_header: Link layer header (encapsulation)
536 * @transport_header: Transport layer header
537 * @network_header: Network layer header
538 * @mac_header: Link layer header
539 * @tail: Tail pointer
541 * @head: Head of buffer
542 * @data: Data head pointer
543 * @truesize: Buffer size
544 * @users: User count - see {datagram,tcp}.c
550 /* These two members must be first. */
551 struct sk_buff *next;
552 struct sk_buff *prev;
556 struct skb_mstamp skb_mstamp;
559 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
564 int ip_defrag_offset;
567 struct net_device *dev;
570 * This is the control buffer. It is free to use for every
571 * layer. Please put your private variables there. If you
572 * want to keep them across layers you have to do a skb_clone()
573 * first. This is owned by whoever has the skb queued ATM.
575 char cb[48] __aligned(8);
577 unsigned long _skb_refdst;
578 void (*destructor)(struct sk_buff *skb);
582 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
583 struct nf_conntrack *nfct;
585 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
586 struct nf_bridge_info *nf_bridge;
593 /* Following fields are _not_ copied in __copy_skb_header()
594 * Note that queue_mapping is here mostly to fill a hole.
596 kmemcheck_bitfield_begin(flags1);
605 kmemcheck_bitfield_end(flags1);
607 /* fields enclosed in headers_start/headers_end are copied
608 * using a single memcpy() in __copy_skb_header()
611 __u32 headers_start[0];
614 /* if you move pkt_type around you also must adapt those constants */
615 #ifdef __BIG_ENDIAN_BITFIELD
616 #define PKT_TYPE_MAX (7 << 5)
618 #define PKT_TYPE_MAX 7
620 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
622 __u8 __pkt_type_offset[0];
632 __u8 wifi_acked_valid:1;
636 /* Indicates the inner headers are valid in the skbuff. */
637 __u8 encapsulation:1;
638 __u8 encap_hdr_csum:1;
640 __u8 csum_complete_sw:1;
643 #ifdef CONFIG_IPV6_NDISC_NODETYPE
644 __u8 ndisc_nodetype:2;
646 __u8 ipvs_property:1;
648 __u8 inner_protocol_type:1;
649 __u8 remcsum_offload:1;
650 /* 3 or 5 bit hole */
652 #ifdef CONFIG_NET_SCHED
653 __u16 tc_index; /* traffic control index */
654 #ifdef CONFIG_NET_CLS_ACT
655 __u16 tc_verd; /* traffic control verdict */
671 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
673 unsigned int napi_id;
674 unsigned int sender_cpu;
678 #ifdef CONFIG_NETWORK_SECMARK
681 #ifdef CONFIG_NET_SWITCHDEV
682 __u32 offload_fwd_mark;
688 __u32 reserved_tailroom;
692 __be16 inner_protocol;
696 __u16 inner_transport_header;
697 __u16 inner_network_header;
698 __u16 inner_mac_header;
701 __u16 transport_header;
702 __u16 network_header;
706 __u32 headers_end[0];
709 /* These elements must be at the end, see alloc_skb() for details. */
714 unsigned int truesize;
720 * Handling routines are only of interest to the kernel
722 #include <linux/slab.h>
725 #define SKB_ALLOC_FCLONE 0x01
726 #define SKB_ALLOC_RX 0x02
727 #define SKB_ALLOC_NAPI 0x04
729 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
730 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
732 return unlikely(skb->pfmemalloc);
736 * skb might have a dst pointer attached, refcounted or not.
737 * _skb_refdst low order bit is set if refcount was _not_ taken
739 #define SKB_DST_NOREF 1UL
740 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
743 * skb_dst - returns skb dst_entry
746 * Returns skb dst_entry, regardless of reference taken or not.
748 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
750 /* If refdst was not refcounted, check we still are in a
751 * rcu_read_lock section
753 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
754 !rcu_read_lock_held() &&
755 !rcu_read_lock_bh_held());
756 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
760 * skb_dst_set - sets skb dst
764 * Sets skb dst, assuming a reference was taken on dst and should
765 * be released by skb_dst_drop()
767 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
769 skb->_skb_refdst = (unsigned long)dst;
773 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
777 * Sets skb dst, assuming a reference was not taken on dst.
778 * If dst entry is cached, we do not take reference and dst_release
779 * will be avoided by refdst_drop. If dst entry is not cached, we take
780 * reference, so that last dst_release can destroy the dst immediately.
782 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
784 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
785 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
789 * skb_dst_is_noref - Test if skb dst isn't refcounted
792 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
794 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
797 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
799 return (struct rtable *)skb_dst(skb);
802 void kfree_skb(struct sk_buff *skb);
803 void kfree_skb_list(struct sk_buff *segs);
804 void skb_tx_error(struct sk_buff *skb);
805 void consume_skb(struct sk_buff *skb);
806 void __kfree_skb(struct sk_buff *skb);
807 extern struct kmem_cache *skbuff_head_cache;
809 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
810 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
811 bool *fragstolen, int *delta_truesize);
813 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
815 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
816 struct sk_buff *build_skb(void *data, unsigned int frag_size);
817 static inline struct sk_buff *alloc_skb(unsigned int size,
820 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
823 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
824 unsigned long data_len,
829 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
830 struct sk_buff_fclones {
839 * skb_fclone_busy - check if fclone is busy
842 * Returns true is skb is a fast clone, and its clone is not freed.
843 * Some drivers call skb_orphan() in their ndo_start_xmit(),
844 * so we also check that this didnt happen.
846 static inline bool skb_fclone_busy(const struct sock *sk,
847 const struct sk_buff *skb)
849 const struct sk_buff_fclones *fclones;
851 fclones = container_of(skb, struct sk_buff_fclones, skb1);
853 return skb->fclone == SKB_FCLONE_ORIG &&
854 atomic_read(&fclones->fclone_ref) > 1 &&
855 fclones->skb2.sk == sk;
858 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
861 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
864 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
865 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
867 return __alloc_skb_head(priority, -1);
870 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
871 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
872 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
873 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
874 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
875 gfp_t gfp_mask, bool fclone);
876 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
879 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
882 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
883 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
884 unsigned int headroom);
885 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
886 int newtailroom, gfp_t priority);
887 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
888 int offset, int len);
889 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
890 int offset, int len);
891 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
892 int skb_pad(struct sk_buff *skb, int pad);
893 #define dev_kfree_skb(a) consume_skb(a)
895 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
896 int getfrag(void *from, char *to, int offset,
897 int len, int odd, struct sk_buff *skb),
898 void *from, int length);
900 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
901 int offset, size_t size);
903 struct skb_seq_state {
907 __u32 stepped_offset;
908 struct sk_buff *root_skb;
909 struct sk_buff *cur_skb;
913 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
914 unsigned int to, struct skb_seq_state *st);
915 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
916 struct skb_seq_state *st);
917 void skb_abort_seq_read(struct skb_seq_state *st);
919 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
920 unsigned int to, struct ts_config *config);
923 * Packet hash types specify the type of hash in skb_set_hash.
925 * Hash types refer to the protocol layer addresses which are used to
926 * construct a packet's hash. The hashes are used to differentiate or identify
927 * flows of the protocol layer for the hash type. Hash types are either
928 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
930 * Properties of hashes:
932 * 1) Two packets in different flows have different hash values
933 * 2) Two packets in the same flow should have the same hash value
935 * A hash at a higher layer is considered to be more specific. A driver should
936 * set the most specific hash possible.
938 * A driver cannot indicate a more specific hash than the layer at which a hash
939 * was computed. For instance an L3 hash cannot be set as an L4 hash.
941 * A driver may indicate a hash level which is less specific than the
942 * actual layer the hash was computed on. For instance, a hash computed
943 * at L4 may be considered an L3 hash. This should only be done if the
944 * driver can't unambiguously determine that the HW computed the hash at
945 * the higher layer. Note that the "should" in the second property above
948 enum pkt_hash_types {
949 PKT_HASH_TYPE_NONE, /* Undefined type */
950 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
951 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
952 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
955 static inline void skb_clear_hash(struct sk_buff *skb)
962 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
969 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
971 skb->l4_hash = is_l4;
972 skb->sw_hash = is_sw;
977 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
979 /* Used by drivers to set hash from HW */
980 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
984 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
986 __skb_set_hash(skb, hash, true, is_l4);
989 void __skb_get_hash(struct sk_buff *skb);
990 u32 __skb_get_hash_symmetric(struct sk_buff *skb);
991 u32 skb_get_poff(const struct sk_buff *skb);
992 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
993 const struct flow_keys *keys, int hlen);
994 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
995 void *data, int hlen_proto);
997 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
998 int thoff, u8 ip_proto)
1000 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1003 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1004 const struct flow_dissector_key *key,
1005 unsigned int key_count);
1007 bool __skb_flow_dissect(const struct sk_buff *skb,
1008 struct flow_dissector *flow_dissector,
1009 void *target_container,
1010 void *data, __be16 proto, int nhoff, int hlen,
1011 unsigned int flags);
1013 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1014 struct flow_dissector *flow_dissector,
1015 void *target_container, unsigned int flags)
1017 return __skb_flow_dissect(skb, flow_dissector, target_container,
1018 NULL, 0, 0, 0, flags);
1021 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1022 struct flow_keys *flow,
1025 memset(flow, 0, sizeof(*flow));
1026 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1027 NULL, 0, 0, 0, flags);
1030 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1031 void *data, __be16 proto,
1032 int nhoff, int hlen,
1035 memset(flow, 0, sizeof(*flow));
1036 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1037 data, proto, nhoff, hlen, flags);
1040 static inline __u32 skb_get_hash(struct sk_buff *skb)
1042 if (!skb->l4_hash && !skb->sw_hash)
1043 __skb_get_hash(skb);
1048 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1050 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1052 if (!skb->l4_hash && !skb->sw_hash) {
1053 struct flow_keys keys;
1054 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1056 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1062 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1064 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1066 if (!skb->l4_hash && !skb->sw_hash) {
1067 struct flow_keys keys;
1068 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
1070 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1076 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1077 const siphash_key_t *perturb);
1079 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1084 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1086 to->hash = from->hash;
1087 to->sw_hash = from->sw_hash;
1088 to->l4_hash = from->l4_hash;
1091 static inline void skb_sender_cpu_clear(struct sk_buff *skb)
1095 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1096 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1098 return skb->head + skb->end;
1101 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1106 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1111 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1113 return skb->end - skb->head;
1118 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1120 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1122 return &skb_shinfo(skb)->hwtstamps;
1126 * skb_queue_empty - check if a queue is empty
1129 * Returns true if the queue is empty, false otherwise.
1131 static inline int skb_queue_empty(const struct sk_buff_head *list)
1133 return list->next == (const struct sk_buff *) list;
1137 * skb_queue_is_last - check if skb is the last entry in the queue
1141 * Returns true if @skb is the last buffer on the list.
1143 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1144 const struct sk_buff *skb)
1146 return skb->next == (const struct sk_buff *) list;
1150 * skb_queue_is_first - check if skb is the first entry in the queue
1154 * Returns true if @skb is the first buffer on the list.
1156 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1157 const struct sk_buff *skb)
1159 return skb->prev == (const struct sk_buff *) list;
1163 * skb_queue_next - return the next packet in the queue
1165 * @skb: current buffer
1167 * Return the next packet in @list after @skb. It is only valid to
1168 * call this if skb_queue_is_last() evaluates to false.
1170 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1171 const struct sk_buff *skb)
1173 /* This BUG_ON may seem severe, but if we just return then we
1174 * are going to dereference garbage.
1176 BUG_ON(skb_queue_is_last(list, skb));
1181 * skb_queue_prev - return the prev packet in the queue
1183 * @skb: current buffer
1185 * Return the prev packet in @list before @skb. It is only valid to
1186 * call this if skb_queue_is_first() evaluates to false.
1188 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1189 const struct sk_buff *skb)
1191 /* This BUG_ON may seem severe, but if we just return then we
1192 * are going to dereference garbage.
1194 BUG_ON(skb_queue_is_first(list, skb));
1199 * skb_get - reference buffer
1200 * @skb: buffer to reference
1202 * Makes another reference to a socket buffer and returns a pointer
1205 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1207 atomic_inc(&skb->users);
1212 * If users == 1, we are the only owner and are can avoid redundant
1217 * skb_cloned - is the buffer a clone
1218 * @skb: buffer to check
1220 * Returns true if the buffer was generated with skb_clone() and is
1221 * one of multiple shared copies of the buffer. Cloned buffers are
1222 * shared data so must not be written to under normal circumstances.
1224 static inline int skb_cloned(const struct sk_buff *skb)
1226 return skb->cloned &&
1227 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1230 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1232 might_sleep_if(gfpflags_allow_blocking(pri));
1234 if (skb_cloned(skb))
1235 return pskb_expand_head(skb, 0, 0, pri);
1241 * skb_header_cloned - is the header a clone
1242 * @skb: buffer to check
1244 * Returns true if modifying the header part of the buffer requires
1245 * the data to be copied.
1247 static inline int skb_header_cloned(const struct sk_buff *skb)
1254 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1255 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1256 return dataref != 1;
1260 * skb_header_release - release reference to header
1261 * @skb: buffer to operate on
1263 * Drop a reference to the header part of the buffer. This is done
1264 * by acquiring a payload reference. You must not read from the header
1265 * part of skb->data after this.
1266 * Note : Check if you can use __skb_header_release() instead.
1268 static inline void skb_header_release(struct sk_buff *skb)
1272 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1276 * __skb_header_release - release reference to header
1277 * @skb: buffer to operate on
1279 * Variant of skb_header_release() assuming skb is private to caller.
1280 * We can avoid one atomic operation.
1282 static inline void __skb_header_release(struct sk_buff *skb)
1285 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1290 * skb_shared - is the buffer shared
1291 * @skb: buffer to check
1293 * Returns true if more than one person has a reference to this
1296 static inline int skb_shared(const struct sk_buff *skb)
1298 return atomic_read(&skb->users) != 1;
1302 * skb_share_check - check if buffer is shared and if so clone it
1303 * @skb: buffer to check
1304 * @pri: priority for memory allocation
1306 * If the buffer is shared the buffer is cloned and the old copy
1307 * drops a reference. A new clone with a single reference is returned.
1308 * If the buffer is not shared the original buffer is returned. When
1309 * being called from interrupt status or with spinlocks held pri must
1312 * NULL is returned on a memory allocation failure.
1314 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1316 might_sleep_if(gfpflags_allow_blocking(pri));
1317 if (skb_shared(skb)) {
1318 struct sk_buff *nskb = skb_clone(skb, pri);
1330 * Copy shared buffers into a new sk_buff. We effectively do COW on
1331 * packets to handle cases where we have a local reader and forward
1332 * and a couple of other messy ones. The normal one is tcpdumping
1333 * a packet thats being forwarded.
1337 * skb_unshare - make a copy of a shared buffer
1338 * @skb: buffer to check
1339 * @pri: priority for memory allocation
1341 * If the socket buffer is a clone then this function creates a new
1342 * copy of the data, drops a reference count on the old copy and returns
1343 * the new copy with the reference count at 1. If the buffer is not a clone
1344 * the original buffer is returned. When called with a spinlock held or
1345 * from interrupt state @pri must be %GFP_ATOMIC
1347 * %NULL is returned on a memory allocation failure.
1349 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1352 might_sleep_if(gfpflags_allow_blocking(pri));
1353 if (skb_cloned(skb)) {
1354 struct sk_buff *nskb = skb_copy(skb, pri);
1356 /* Free our shared copy */
1367 * skb_peek - peek at the head of an &sk_buff_head
1368 * @list_: list to peek at
1370 * Peek an &sk_buff. Unlike most other operations you _MUST_
1371 * be careful with this one. A peek leaves the buffer on the
1372 * list and someone else may run off with it. You must hold
1373 * the appropriate locks or have a private queue to do this.
1375 * Returns %NULL for an empty list or a pointer to the head element.
1376 * The reference count is not incremented and the reference is therefore
1377 * volatile. Use with caution.
1379 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1381 struct sk_buff *skb = list_->next;
1383 if (skb == (struct sk_buff *)list_)
1389 * skb_peek_next - peek skb following the given one from a queue
1390 * @skb: skb to start from
1391 * @list_: list to peek at
1393 * Returns %NULL when the end of the list is met or a pointer to the
1394 * next element. The reference count is not incremented and the
1395 * reference is therefore volatile. Use with caution.
1397 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1398 const struct sk_buff_head *list_)
1400 struct sk_buff *next = skb->next;
1402 if (next == (struct sk_buff *)list_)
1408 * skb_peek_tail - peek at the tail of an &sk_buff_head
1409 * @list_: list to peek at
1411 * Peek an &sk_buff. Unlike most other operations you _MUST_
1412 * be careful with this one. A peek leaves the buffer on the
1413 * list and someone else may run off with it. You must hold
1414 * the appropriate locks or have a private queue to do this.
1416 * Returns %NULL for an empty list or a pointer to the tail element.
1417 * The reference count is not incremented and the reference is therefore
1418 * volatile. Use with caution.
1420 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1422 struct sk_buff *skb = list_->prev;
1424 if (skb == (struct sk_buff *)list_)
1431 * skb_queue_len - get queue length
1432 * @list_: list to measure
1434 * Return the length of an &sk_buff queue.
1436 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1442 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1443 * @list: queue to initialize
1445 * This initializes only the list and queue length aspects of
1446 * an sk_buff_head object. This allows to initialize the list
1447 * aspects of an sk_buff_head without reinitializing things like
1448 * the spinlock. It can also be used for on-stack sk_buff_head
1449 * objects where the spinlock is known to not be used.
1451 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1453 list->prev = list->next = (struct sk_buff *)list;
1458 * This function creates a split out lock class for each invocation;
1459 * this is needed for now since a whole lot of users of the skb-queue
1460 * infrastructure in drivers have different locking usage (in hardirq)
1461 * than the networking core (in softirq only). In the long run either the
1462 * network layer or drivers should need annotation to consolidate the
1463 * main types of usage into 3 classes.
1465 static inline void skb_queue_head_init(struct sk_buff_head *list)
1467 spin_lock_init(&list->lock);
1468 __skb_queue_head_init(list);
1471 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1472 struct lock_class_key *class)
1474 skb_queue_head_init(list);
1475 lockdep_set_class(&list->lock, class);
1479 * Insert an sk_buff on a list.
1481 * The "__skb_xxxx()" functions are the non-atomic ones that
1482 * can only be called with interrupts disabled.
1484 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1485 struct sk_buff_head *list);
1486 static inline void __skb_insert(struct sk_buff *newsk,
1487 struct sk_buff *prev, struct sk_buff *next,
1488 struct sk_buff_head *list)
1492 next->prev = prev->next = newsk;
1496 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1497 struct sk_buff *prev,
1498 struct sk_buff *next)
1500 struct sk_buff *first = list->next;
1501 struct sk_buff *last = list->prev;
1511 * skb_queue_splice - join two skb lists, this is designed for stacks
1512 * @list: the new list to add
1513 * @head: the place to add it in the first list
1515 static inline void skb_queue_splice(const struct sk_buff_head *list,
1516 struct sk_buff_head *head)
1518 if (!skb_queue_empty(list)) {
1519 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1520 head->qlen += list->qlen;
1525 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1526 * @list: the new list to add
1527 * @head: the place to add it in the first list
1529 * The list at @list is reinitialised
1531 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1532 struct sk_buff_head *head)
1534 if (!skb_queue_empty(list)) {
1535 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1536 head->qlen += list->qlen;
1537 __skb_queue_head_init(list);
1542 * skb_queue_splice_tail - join two skb lists, each list being a queue
1543 * @list: the new list to add
1544 * @head: the place to add it in the first list
1546 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1547 struct sk_buff_head *head)
1549 if (!skb_queue_empty(list)) {
1550 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1551 head->qlen += list->qlen;
1556 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1557 * @list: the new list to add
1558 * @head: the place to add it in the first list
1560 * Each of the lists is a queue.
1561 * The list at @list is reinitialised
1563 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1564 struct sk_buff_head *head)
1566 if (!skb_queue_empty(list)) {
1567 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1568 head->qlen += list->qlen;
1569 __skb_queue_head_init(list);
1574 * __skb_queue_after - queue a buffer at the list head
1575 * @list: list to use
1576 * @prev: place after this buffer
1577 * @newsk: buffer to queue
1579 * Queue a buffer int the middle of a list. This function takes no locks
1580 * and you must therefore hold required locks before calling it.
1582 * A buffer cannot be placed on two lists at the same time.
1584 static inline void __skb_queue_after(struct sk_buff_head *list,
1585 struct sk_buff *prev,
1586 struct sk_buff *newsk)
1588 __skb_insert(newsk, prev, prev->next, list);
1591 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1592 struct sk_buff_head *list);
1594 static inline void __skb_queue_before(struct sk_buff_head *list,
1595 struct sk_buff *next,
1596 struct sk_buff *newsk)
1598 __skb_insert(newsk, next->prev, next, list);
1602 * __skb_queue_head - queue a buffer at the list head
1603 * @list: list to use
1604 * @newsk: buffer to queue
1606 * Queue a buffer at the start of a list. This function takes no locks
1607 * and you must therefore hold required locks before calling it.
1609 * A buffer cannot be placed on two lists at the same time.
1611 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1612 static inline void __skb_queue_head(struct sk_buff_head *list,
1613 struct sk_buff *newsk)
1615 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1619 * __skb_queue_tail - queue a buffer at the list tail
1620 * @list: list to use
1621 * @newsk: buffer to queue
1623 * Queue a buffer at the end of a list. This function takes no locks
1624 * and you must therefore hold required locks before calling it.
1626 * A buffer cannot be placed on two lists at the same time.
1628 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1629 static inline void __skb_queue_tail(struct sk_buff_head *list,
1630 struct sk_buff *newsk)
1632 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1636 * remove sk_buff from list. _Must_ be called atomically, and with
1639 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1640 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1642 struct sk_buff *next, *prev;
1647 skb->next = skb->prev = NULL;
1653 * __skb_dequeue - remove from the head of the queue
1654 * @list: list to dequeue from
1656 * Remove the head of the list. This function does not take any locks
1657 * so must be used with appropriate locks held only. The head item is
1658 * returned or %NULL if the list is empty.
1660 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1661 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1663 struct sk_buff *skb = skb_peek(list);
1665 __skb_unlink(skb, list);
1670 * __skb_dequeue_tail - remove from the tail of the queue
1671 * @list: list to dequeue from
1673 * Remove the tail of the list. This function does not take any locks
1674 * so must be used with appropriate locks held only. The tail item is
1675 * returned or %NULL if the list is empty.
1677 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1678 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1680 struct sk_buff *skb = skb_peek_tail(list);
1682 __skb_unlink(skb, list);
1687 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1689 return skb->data_len;
1692 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1694 return skb->len - skb->data_len;
1697 static inline int skb_pagelen(const struct sk_buff *skb)
1701 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1702 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1703 return len + skb_headlen(skb);
1707 * __skb_fill_page_desc - initialise a paged fragment in an skb
1708 * @skb: buffer containing fragment to be initialised
1709 * @i: paged fragment index to initialise
1710 * @page: the page to use for this fragment
1711 * @off: the offset to the data with @page
1712 * @size: the length of the data
1714 * Initialises the @i'th fragment of @skb to point to &size bytes at
1715 * offset @off within @page.
1717 * Does not take any additional reference on the fragment.
1719 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1720 struct page *page, int off, int size)
1722 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1725 * Propagate page pfmemalloc to the skb if we can. The problem is
1726 * that not all callers have unique ownership of the page but rely
1727 * on page_is_pfmemalloc doing the right thing(tm).
1729 frag->page.p = page;
1730 frag->page_offset = off;
1731 skb_frag_size_set(frag, size);
1733 page = compound_head(page);
1734 if (page_is_pfmemalloc(page))
1735 skb->pfmemalloc = true;
1739 * skb_fill_page_desc - initialise a paged fragment in an skb
1740 * @skb: buffer containing fragment to be initialised
1741 * @i: paged fragment index to initialise
1742 * @page: the page to use for this fragment
1743 * @off: the offset to the data with @page
1744 * @size: the length of the data
1746 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1747 * @skb to point to @size bytes at offset @off within @page. In
1748 * addition updates @skb such that @i is the last fragment.
1750 * Does not take any additional reference on the fragment.
1752 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1753 struct page *page, int off, int size)
1755 __skb_fill_page_desc(skb, i, page, off, size);
1756 skb_shinfo(skb)->nr_frags = i + 1;
1759 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1760 int size, unsigned int truesize);
1762 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1763 unsigned int truesize);
1765 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1766 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1767 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1769 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1770 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1772 return skb->head + skb->tail;
1775 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1777 skb->tail = skb->data - skb->head;
1780 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1782 skb_reset_tail_pointer(skb);
1783 skb->tail += offset;
1786 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1787 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1792 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1794 skb->tail = skb->data;
1797 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1799 skb->tail = skb->data + offset;
1802 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1805 * Add data to an sk_buff
1807 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1808 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1809 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1811 unsigned char *tmp = skb_tail_pointer(skb);
1812 SKB_LINEAR_ASSERT(skb);
1818 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1819 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1826 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1827 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1830 BUG_ON(skb->len < skb->data_len);
1831 return skb->data += len;
1834 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1836 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1839 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1841 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1843 if (len > skb_headlen(skb) &&
1844 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1847 return skb->data += len;
1850 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1852 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1855 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1857 if (likely(len <= skb_headlen(skb)))
1859 if (unlikely(len > skb->len))
1861 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1865 * skb_headroom - bytes at buffer head
1866 * @skb: buffer to check
1868 * Return the number of bytes of free space at the head of an &sk_buff.
1870 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1872 return skb->data - skb->head;
1876 * skb_tailroom - bytes at buffer end
1877 * @skb: buffer to check
1879 * Return the number of bytes of free space at the tail of an sk_buff
1881 static inline int skb_tailroom(const struct sk_buff *skb)
1883 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1887 * skb_availroom - bytes at buffer end
1888 * @skb: buffer to check
1890 * Return the number of bytes of free space at the tail of an sk_buff
1891 * allocated by sk_stream_alloc()
1893 static inline int skb_availroom(const struct sk_buff *skb)
1895 if (skb_is_nonlinear(skb))
1898 return skb->end - skb->tail - skb->reserved_tailroom;
1902 * skb_reserve - adjust headroom
1903 * @skb: buffer to alter
1904 * @len: bytes to move
1906 * Increase the headroom of an empty &sk_buff by reducing the tail
1907 * room. This is only allowed for an empty buffer.
1909 static inline void skb_reserve(struct sk_buff *skb, int len)
1916 * skb_tailroom_reserve - adjust reserved_tailroom
1917 * @skb: buffer to alter
1918 * @mtu: maximum amount of headlen permitted
1919 * @needed_tailroom: minimum amount of reserved_tailroom
1921 * Set reserved_tailroom so that headlen can be as large as possible but
1922 * not larger than mtu and tailroom cannot be smaller than
1924 * The required headroom should already have been reserved before using
1927 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
1928 unsigned int needed_tailroom)
1930 SKB_LINEAR_ASSERT(skb);
1931 if (mtu < skb_tailroom(skb) - needed_tailroom)
1932 /* use at most mtu */
1933 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
1935 /* use up to all available space */
1936 skb->reserved_tailroom = needed_tailroom;
1939 #define ENCAP_TYPE_ETHER 0
1940 #define ENCAP_TYPE_IPPROTO 1
1942 static inline void skb_set_inner_protocol(struct sk_buff *skb,
1945 skb->inner_protocol = protocol;
1946 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
1949 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
1952 skb->inner_ipproto = ipproto;
1953 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
1956 static inline void skb_reset_inner_headers(struct sk_buff *skb)
1958 skb->inner_mac_header = skb->mac_header;
1959 skb->inner_network_header = skb->network_header;
1960 skb->inner_transport_header = skb->transport_header;
1963 static inline void skb_reset_mac_len(struct sk_buff *skb)
1965 skb->mac_len = skb->network_header - skb->mac_header;
1968 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1971 return skb->head + skb->inner_transport_header;
1974 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1976 skb->inner_transport_header = skb->data - skb->head;
1979 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1982 skb_reset_inner_transport_header(skb);
1983 skb->inner_transport_header += offset;
1986 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1988 return skb->head + skb->inner_network_header;
1991 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1993 skb->inner_network_header = skb->data - skb->head;
1996 static inline void skb_set_inner_network_header(struct sk_buff *skb,
1999 skb_reset_inner_network_header(skb);
2000 skb->inner_network_header += offset;
2003 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2005 return skb->head + skb->inner_mac_header;
2008 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2010 skb->inner_mac_header = skb->data - skb->head;
2013 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2016 skb_reset_inner_mac_header(skb);
2017 skb->inner_mac_header += offset;
2019 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2021 return skb->transport_header != (typeof(skb->transport_header))~0U;
2024 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2026 return skb->head + skb->transport_header;
2029 static inline void skb_reset_transport_header(struct sk_buff *skb)
2031 skb->transport_header = skb->data - skb->head;
2034 static inline void skb_set_transport_header(struct sk_buff *skb,
2037 skb_reset_transport_header(skb);
2038 skb->transport_header += offset;
2041 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2043 return skb->head + skb->network_header;
2046 static inline void skb_reset_network_header(struct sk_buff *skb)
2048 skb->network_header = skb->data - skb->head;
2051 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2053 skb_reset_network_header(skb);
2054 skb->network_header += offset;
2057 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2059 return skb->head + skb->mac_header;
2062 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2064 return skb->mac_header != (typeof(skb->mac_header))~0U;
2067 static inline void skb_reset_mac_header(struct sk_buff *skb)
2069 skb->mac_header = skb->data - skb->head;
2072 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2074 skb_reset_mac_header(skb);
2075 skb->mac_header += offset;
2078 static inline void skb_pop_mac_header(struct sk_buff *skb)
2080 skb->mac_header = skb->network_header;
2083 static inline void skb_probe_transport_header(struct sk_buff *skb,
2084 const int offset_hint)
2086 struct flow_keys keys;
2088 if (skb_transport_header_was_set(skb))
2090 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2091 skb_set_transport_header(skb, keys.control.thoff);
2093 skb_set_transport_header(skb, offset_hint);
2096 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2098 if (skb_mac_header_was_set(skb)) {
2099 const unsigned char *old_mac = skb_mac_header(skb);
2101 skb_set_mac_header(skb, -skb->mac_len);
2102 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2106 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2108 return skb->csum_start - skb_headroom(skb);
2111 static inline int skb_transport_offset(const struct sk_buff *skb)
2113 return skb_transport_header(skb) - skb->data;
2116 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2118 return skb->transport_header - skb->network_header;
2121 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2123 return skb->inner_transport_header - skb->inner_network_header;
2126 static inline int skb_network_offset(const struct sk_buff *skb)
2128 return skb_network_header(skb) - skb->data;
2131 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2133 return skb_inner_network_header(skb) - skb->data;
2136 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2138 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2142 * CPUs often take a performance hit when accessing unaligned memory
2143 * locations. The actual performance hit varies, it can be small if the
2144 * hardware handles it or large if we have to take an exception and fix it
2147 * Since an ethernet header is 14 bytes network drivers often end up with
2148 * the IP header at an unaligned offset. The IP header can be aligned by
2149 * shifting the start of the packet by 2 bytes. Drivers should do this
2152 * skb_reserve(skb, NET_IP_ALIGN);
2154 * The downside to this alignment of the IP header is that the DMA is now
2155 * unaligned. On some architectures the cost of an unaligned DMA is high
2156 * and this cost outweighs the gains made by aligning the IP header.
2158 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2161 #ifndef NET_IP_ALIGN
2162 #define NET_IP_ALIGN 2
2166 * The networking layer reserves some headroom in skb data (via
2167 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2168 * the header has to grow. In the default case, if the header has to grow
2169 * 32 bytes or less we avoid the reallocation.
2171 * Unfortunately this headroom changes the DMA alignment of the resulting
2172 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2173 * on some architectures. An architecture can override this value,
2174 * perhaps setting it to a cacheline in size (since that will maintain
2175 * cacheline alignment of the DMA). It must be a power of 2.
2177 * Various parts of the networking layer expect at least 32 bytes of
2178 * headroom, you should not reduce this.
2180 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2181 * to reduce average number of cache lines per packet.
2182 * get_rps_cpus() for example only access one 64 bytes aligned block :
2183 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2186 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2189 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2191 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2193 if (unlikely(skb_is_nonlinear(skb))) {
2198 skb_set_tail_pointer(skb, len);
2201 void skb_trim(struct sk_buff *skb, unsigned int len);
2203 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2206 return ___pskb_trim(skb, len);
2207 __skb_trim(skb, len);
2211 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2213 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2217 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2218 * @skb: buffer to alter
2221 * This is identical to pskb_trim except that the caller knows that
2222 * the skb is not cloned so we should never get an error due to out-
2225 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2227 int err = pskb_trim(skb, len);
2232 * skb_orphan - orphan a buffer
2233 * @skb: buffer to orphan
2235 * If a buffer currently has an owner then we call the owner's
2236 * destructor function and make the @skb unowned. The buffer continues
2237 * to exist but is no longer charged to its former owner.
2239 static inline void skb_orphan(struct sk_buff *skb)
2241 if (skb->destructor) {
2242 skb->destructor(skb);
2243 skb->destructor = NULL;
2251 * skb_orphan_frags - orphan the frags contained in a buffer
2252 * @skb: buffer to orphan frags from
2253 * @gfp_mask: allocation mask for replacement pages
2255 * For each frag in the SKB which needs a destructor (i.e. has an
2256 * owner) create a copy of that frag and release the original
2257 * page by calling the destructor.
2259 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2261 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2263 return skb_copy_ubufs(skb, gfp_mask);
2267 * __skb_queue_purge - empty a list
2268 * @list: list to empty
2270 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2271 * the list and one reference dropped. This function does not take the
2272 * list lock and the caller must hold the relevant locks to use it.
2274 void skb_queue_purge(struct sk_buff_head *list);
2275 static inline void __skb_queue_purge(struct sk_buff_head *list)
2277 struct sk_buff *skb;
2278 while ((skb = __skb_dequeue(list)) != NULL)
2282 unsigned int skb_rbtree_purge(struct rb_root *root);
2284 void *netdev_alloc_frag(unsigned int fragsz);
2286 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2290 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2291 * @dev: network device to receive on
2292 * @length: length to allocate
2294 * Allocate a new &sk_buff and assign it a usage count of one. The
2295 * buffer has unspecified headroom built in. Users should allocate
2296 * the headroom they think they need without accounting for the
2297 * built in space. The built in space is used for optimisations.
2299 * %NULL is returned if there is no free memory. Although this function
2300 * allocates memory it can be called from an interrupt.
2302 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2303 unsigned int length)
2305 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2308 /* legacy helper around __netdev_alloc_skb() */
2309 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2312 return __netdev_alloc_skb(NULL, length, gfp_mask);
2315 /* legacy helper around netdev_alloc_skb() */
2316 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2318 return netdev_alloc_skb(NULL, length);
2322 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2323 unsigned int length, gfp_t gfp)
2325 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2327 if (NET_IP_ALIGN && skb)
2328 skb_reserve(skb, NET_IP_ALIGN);
2332 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2333 unsigned int length)
2335 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2338 static inline void skb_free_frag(void *addr)
2340 __free_page_frag(addr);
2343 void *napi_alloc_frag(unsigned int fragsz);
2344 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2345 unsigned int length, gfp_t gfp_mask);
2346 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2347 unsigned int length)
2349 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2353 * __dev_alloc_pages - allocate page for network Rx
2354 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2355 * @order: size of the allocation
2357 * Allocate a new page.
2359 * %NULL is returned if there is no free memory.
2361 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2364 /* This piece of code contains several assumptions.
2365 * 1. This is for device Rx, therefor a cold page is preferred.
2366 * 2. The expectation is the user wants a compound page.
2367 * 3. If requesting a order 0 page it will not be compound
2368 * due to the check to see if order has a value in prep_new_page
2369 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2370 * code in gfp_to_alloc_flags that should be enforcing this.
2372 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2374 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2377 static inline struct page *dev_alloc_pages(unsigned int order)
2379 return __dev_alloc_pages(GFP_ATOMIC, order);
2383 * __dev_alloc_page - allocate a page for network Rx
2384 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2386 * Allocate a new page.
2388 * %NULL is returned if there is no free memory.
2390 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2392 return __dev_alloc_pages(gfp_mask, 0);
2395 static inline struct page *dev_alloc_page(void)
2397 return __dev_alloc_page(GFP_ATOMIC);
2401 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2402 * @page: The page that was allocated from skb_alloc_page
2403 * @skb: The skb that may need pfmemalloc set
2405 static inline void skb_propagate_pfmemalloc(struct page *page,
2406 struct sk_buff *skb)
2408 if (page_is_pfmemalloc(page))
2409 skb->pfmemalloc = true;
2413 * skb_frag_page - retrieve the page referred to by a paged fragment
2414 * @frag: the paged fragment
2416 * Returns the &struct page associated with @frag.
2418 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2420 return frag->page.p;
2424 * __skb_frag_ref - take an addition reference on a paged fragment.
2425 * @frag: the paged fragment
2427 * Takes an additional reference on the paged fragment @frag.
2429 static inline void __skb_frag_ref(skb_frag_t *frag)
2431 get_page(skb_frag_page(frag));
2435 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2437 * @f: the fragment offset.
2439 * Takes an additional reference on the @f'th paged fragment of @skb.
2441 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2443 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2447 * __skb_frag_unref - release a reference on a paged fragment.
2448 * @frag: the paged fragment
2450 * Releases a reference on the paged fragment @frag.
2452 static inline void __skb_frag_unref(skb_frag_t *frag)
2454 put_page(skb_frag_page(frag));
2458 * skb_frag_unref - release a reference on a paged fragment of an skb.
2460 * @f: the fragment offset
2462 * Releases a reference on the @f'th paged fragment of @skb.
2464 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2466 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2470 * skb_frag_address - gets the address of the data contained in a paged fragment
2471 * @frag: the paged fragment buffer
2473 * Returns the address of the data within @frag. The page must already
2476 static inline void *skb_frag_address(const skb_frag_t *frag)
2478 return page_address(skb_frag_page(frag)) + frag->page_offset;
2482 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2483 * @frag: the paged fragment buffer
2485 * Returns the address of the data within @frag. Checks that the page
2486 * is mapped and returns %NULL otherwise.
2488 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2490 void *ptr = page_address(skb_frag_page(frag));
2494 return ptr + frag->page_offset;
2498 * __skb_frag_set_page - sets the page contained in a paged fragment
2499 * @frag: the paged fragment
2500 * @page: the page to set
2502 * Sets the fragment @frag to contain @page.
2504 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2506 frag->page.p = page;
2510 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2512 * @f: the fragment offset
2513 * @page: the page to set
2515 * Sets the @f'th fragment of @skb to contain @page.
2517 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2520 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2523 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2526 * skb_frag_dma_map - maps a paged fragment via the DMA API
2527 * @dev: the device to map the fragment to
2528 * @frag: the paged fragment to map
2529 * @offset: the offset within the fragment (starting at the
2530 * fragment's own offset)
2531 * @size: the number of bytes to map
2532 * @dir: the direction of the mapping (%PCI_DMA_*)
2534 * Maps the page associated with @frag to @device.
2536 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2537 const skb_frag_t *frag,
2538 size_t offset, size_t size,
2539 enum dma_data_direction dir)
2541 return dma_map_page(dev, skb_frag_page(frag),
2542 frag->page_offset + offset, size, dir);
2545 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2548 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2552 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2555 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2560 * skb_clone_writable - is the header of a clone writable
2561 * @skb: buffer to check
2562 * @len: length up to which to write
2564 * Returns true if modifying the header part of the cloned buffer
2565 * does not requires the data to be copied.
2567 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2569 return !skb_header_cloned(skb) &&
2570 skb_headroom(skb) + len <= skb->hdr_len;
2573 static inline int skb_try_make_writable(struct sk_buff *skb,
2574 unsigned int write_len)
2576 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2577 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2580 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2585 if (headroom > skb_headroom(skb))
2586 delta = headroom - skb_headroom(skb);
2588 if (delta || cloned)
2589 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2595 * skb_cow - copy header of skb when it is required
2596 * @skb: buffer to cow
2597 * @headroom: needed headroom
2599 * If the skb passed lacks sufficient headroom or its data part
2600 * is shared, data is reallocated. If reallocation fails, an error
2601 * is returned and original skb is not changed.
2603 * The result is skb with writable area skb->head...skb->tail
2604 * and at least @headroom of space at head.
2606 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2608 return __skb_cow(skb, headroom, skb_cloned(skb));
2612 * skb_cow_head - skb_cow but only making the head writable
2613 * @skb: buffer to cow
2614 * @headroom: needed headroom
2616 * This function is identical to skb_cow except that we replace the
2617 * skb_cloned check by skb_header_cloned. It should be used when
2618 * you only need to push on some header and do not need to modify
2621 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2623 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2627 * skb_padto - pad an skbuff up to a minimal size
2628 * @skb: buffer to pad
2629 * @len: minimal length
2631 * Pads up a buffer to ensure the trailing bytes exist and are
2632 * blanked. If the buffer already contains sufficient data it
2633 * is untouched. Otherwise it is extended. Returns zero on
2634 * success. The skb is freed on error.
2636 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2638 unsigned int size = skb->len;
2639 if (likely(size >= len))
2641 return skb_pad(skb, len - size);
2645 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2646 * @skb: buffer to pad
2647 * @len: minimal length
2649 * Pads up a buffer to ensure the trailing bytes exist and are
2650 * blanked. If the buffer already contains sufficient data it
2651 * is untouched. Otherwise it is extended. Returns zero on
2652 * success. The skb is freed on error.
2654 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2656 unsigned int size = skb->len;
2658 if (unlikely(size < len)) {
2660 if (skb_pad(skb, len))
2662 __skb_put(skb, len);
2667 static inline int skb_add_data(struct sk_buff *skb,
2668 struct iov_iter *from, int copy)
2670 const int off = skb->len;
2672 if (skb->ip_summed == CHECKSUM_NONE) {
2674 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2675 &csum, from) == copy) {
2676 skb->csum = csum_block_add(skb->csum, csum, off);
2679 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
2682 __skb_trim(skb, off);
2686 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2687 const struct page *page, int off)
2690 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2692 return page == skb_frag_page(frag) &&
2693 off == frag->page_offset + skb_frag_size(frag);
2698 static inline int __skb_linearize(struct sk_buff *skb)
2700 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2704 * skb_linearize - convert paged skb to linear one
2705 * @skb: buffer to linarize
2707 * If there is no free memory -ENOMEM is returned, otherwise zero
2708 * is returned and the old skb data released.
2710 static inline int skb_linearize(struct sk_buff *skb)
2712 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2716 * skb_has_shared_frag - can any frag be overwritten
2717 * @skb: buffer to test
2719 * Return true if the skb has at least one frag that might be modified
2720 * by an external entity (as in vmsplice()/sendfile())
2722 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2724 return skb_is_nonlinear(skb) &&
2725 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2729 * skb_linearize_cow - make sure skb is linear and writable
2730 * @skb: buffer to process
2732 * If there is no free memory -ENOMEM is returned, otherwise zero
2733 * is returned and the old skb data released.
2735 static inline int skb_linearize_cow(struct sk_buff *skb)
2737 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2738 __skb_linearize(skb) : 0;
2742 * skb_postpull_rcsum - update checksum for received skb after pull
2743 * @skb: buffer to update
2744 * @start: start of data before pull
2745 * @len: length of data pulled
2747 * After doing a pull on a received packet, you need to call this to
2748 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2749 * CHECKSUM_NONE so that it can be recomputed from scratch.
2752 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2753 const void *start, unsigned int len)
2755 if (skb->ip_summed == CHECKSUM_COMPLETE)
2756 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2757 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2758 skb_checksum_start_offset(skb) < 0)
2759 skb->ip_summed = CHECKSUM_NONE;
2762 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2764 static inline void skb_postpush_rcsum(struct sk_buff *skb,
2765 const void *start, unsigned int len)
2767 /* For performing the reverse operation to skb_postpull_rcsum(),
2768 * we can instead of ...
2770 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0));
2772 * ... just use this equivalent version here to save a few
2773 * instructions. Feeding csum of 0 in csum_partial() and later
2774 * on adding skb->csum is equivalent to feed skb->csum in the
2777 if (skb->ip_summed == CHECKSUM_COMPLETE)
2778 skb->csum = csum_partial(start, len, skb->csum);
2782 * skb_push_rcsum - push skb and update receive checksum
2783 * @skb: buffer to update
2784 * @len: length of data pulled
2786 * This function performs an skb_push on the packet and updates
2787 * the CHECKSUM_COMPLETE checksum. It should be used on
2788 * receive path processing instead of skb_push unless you know
2789 * that the checksum difference is zero (e.g., a valid IP header)
2790 * or you are setting ip_summed to CHECKSUM_NONE.
2792 static inline unsigned char *skb_push_rcsum(struct sk_buff *skb,
2796 skb_postpush_rcsum(skb, skb->data, len);
2800 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
2802 * pskb_trim_rcsum - trim received skb and update checksum
2803 * @skb: buffer to trim
2806 * This is exactly the same as pskb_trim except that it ensures the
2807 * checksum of received packets are still valid after the operation.
2808 * It can change skb pointers.
2811 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2813 if (likely(len >= skb->len))
2815 return pskb_trim_rcsum_slow(skb, len);
2818 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
2819 #define skb_rb_first(root) rb_to_skb(rb_first(root))
2820 #define skb_rb_last(root) rb_to_skb(rb_last(root))
2821 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
2822 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
2824 #define skb_queue_walk(queue, skb) \
2825 for (skb = (queue)->next; \
2826 skb != (struct sk_buff *)(queue); \
2829 #define skb_queue_walk_safe(queue, skb, tmp) \
2830 for (skb = (queue)->next, tmp = skb->next; \
2831 skb != (struct sk_buff *)(queue); \
2832 skb = tmp, tmp = skb->next)
2834 #define skb_queue_walk_from(queue, skb) \
2835 for (; skb != (struct sk_buff *)(queue); \
2838 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2839 for (tmp = skb->next; \
2840 skb != (struct sk_buff *)(queue); \
2841 skb = tmp, tmp = skb->next)
2843 #define skb_queue_reverse_walk(queue, skb) \
2844 for (skb = (queue)->prev; \
2845 skb != (struct sk_buff *)(queue); \
2848 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2849 for (skb = (queue)->prev, tmp = skb->prev; \
2850 skb != (struct sk_buff *)(queue); \
2851 skb = tmp, tmp = skb->prev)
2853 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2854 for (tmp = skb->prev; \
2855 skb != (struct sk_buff *)(queue); \
2856 skb = tmp, tmp = skb->prev)
2858 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2860 return skb_shinfo(skb)->frag_list != NULL;
2863 static inline void skb_frag_list_init(struct sk_buff *skb)
2865 skb_shinfo(skb)->frag_list = NULL;
2868 #define skb_walk_frags(skb, iter) \
2869 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2871 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2872 int *peeked, int *off, int *err);
2873 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2875 unsigned int datagram_poll(struct file *file, struct socket *sock,
2876 struct poll_table_struct *wait);
2877 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2878 struct iov_iter *to, int size);
2879 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2880 struct msghdr *msg, int size)
2882 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
2884 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2885 struct msghdr *msg);
2886 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2887 struct iov_iter *from, int len);
2888 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
2889 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2890 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2891 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2892 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2893 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2894 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2895 int len, __wsum csum);
2896 ssize_t skb_socket_splice(struct sock *sk,
2897 struct pipe_inode_info *pipe,
2898 struct splice_pipe_desc *spd);
2899 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2900 struct pipe_inode_info *pipe, unsigned int len,
2902 ssize_t (*splice_cb)(struct sock *,
2903 struct pipe_inode_info *,
2904 struct splice_pipe_desc *));
2905 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2906 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2907 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2909 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2910 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2911 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2912 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2913 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2914 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
2915 int skb_ensure_writable(struct sk_buff *skb, int write_len);
2916 int skb_vlan_pop(struct sk_buff *skb);
2917 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
2919 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2921 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
2924 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
2926 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
2929 struct skb_checksum_ops {
2930 __wsum (*update)(const void *mem, int len, __wsum wsum);
2931 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2934 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2935 __wsum csum, const struct skb_checksum_ops *ops);
2936 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2939 static inline void * __must_check
2940 __skb_header_pointer(const struct sk_buff *skb, int offset,
2941 int len, void *data, int hlen, void *buffer)
2943 if (hlen - offset >= len)
2944 return data + offset;
2947 skb_copy_bits(skb, offset, buffer, len) < 0)
2953 static inline void * __must_check
2954 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
2956 return __skb_header_pointer(skb, offset, len, skb->data,
2957 skb_headlen(skb), buffer);
2961 * skb_needs_linearize - check if we need to linearize a given skb
2962 * depending on the given device features.
2963 * @skb: socket buffer to check
2964 * @features: net device features
2966 * Returns true if either:
2967 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2968 * 2. skb is fragmented and the device does not support SG.
2970 static inline bool skb_needs_linearize(struct sk_buff *skb,
2971 netdev_features_t features)
2973 return skb_is_nonlinear(skb) &&
2974 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2975 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2978 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2980 const unsigned int len)
2982 memcpy(to, skb->data, len);
2985 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2986 const int offset, void *to,
2987 const unsigned int len)
2989 memcpy(to, skb->data + offset, len);
2992 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2994 const unsigned int len)
2996 memcpy(skb->data, from, len);
2999 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3002 const unsigned int len)
3004 memcpy(skb->data + offset, from, len);
3007 void skb_init(void);
3009 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3015 * skb_get_timestamp - get timestamp from a skb
3016 * @skb: skb to get stamp from
3017 * @stamp: pointer to struct timeval to store stamp in
3019 * Timestamps are stored in the skb as offsets to a base timestamp.
3020 * This function converts the offset back to a struct timeval and stores
3023 static inline void skb_get_timestamp(const struct sk_buff *skb,
3024 struct timeval *stamp)
3026 *stamp = ktime_to_timeval(skb->tstamp);
3029 static inline void skb_get_timestampns(const struct sk_buff *skb,
3030 struct timespec *stamp)
3032 *stamp = ktime_to_timespec(skb->tstamp);
3035 static inline void __net_timestamp(struct sk_buff *skb)
3037 skb->tstamp = ktime_get_real();
3040 static inline ktime_t net_timedelta(ktime_t t)
3042 return ktime_sub(ktime_get_real(), t);
3045 static inline ktime_t net_invalid_timestamp(void)
3047 return ktime_set(0, 0);
3050 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3052 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3054 void skb_clone_tx_timestamp(struct sk_buff *skb);
3055 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3057 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3059 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3063 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3068 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3071 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3073 * PHY drivers may accept clones of transmitted packets for
3074 * timestamping via their phy_driver.txtstamp method. These drivers
3075 * must call this function to return the skb back to the stack with a
3078 * @skb: clone of the the original outgoing packet
3079 * @hwtstamps: hardware time stamps
3082 void skb_complete_tx_timestamp(struct sk_buff *skb,
3083 struct skb_shared_hwtstamps *hwtstamps);
3085 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3086 struct skb_shared_hwtstamps *hwtstamps,
3087 struct sock *sk, int tstype);
3090 * skb_tstamp_tx - queue clone of skb with send time stamps
3091 * @orig_skb: the original outgoing packet
3092 * @hwtstamps: hardware time stamps, may be NULL if not available
3094 * If the skb has a socket associated, then this function clones the
3095 * skb (thus sharing the actual data and optional structures), stores
3096 * the optional hardware time stamping information (if non NULL) or
3097 * generates a software time stamp (otherwise), then queues the clone
3098 * to the error queue of the socket. Errors are silently ignored.
3100 void skb_tstamp_tx(struct sk_buff *orig_skb,
3101 struct skb_shared_hwtstamps *hwtstamps);
3103 static inline void sw_tx_timestamp(struct sk_buff *skb)
3105 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3106 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
3107 skb_tstamp_tx(skb, NULL);
3111 * skb_tx_timestamp() - Driver hook for transmit timestamping
3113 * Ethernet MAC Drivers should call this function in their hard_xmit()
3114 * function immediately before giving the sk_buff to the MAC hardware.
3116 * Specifically, one should make absolutely sure that this function is
3117 * called before TX completion of this packet can trigger. Otherwise
3118 * the packet could potentially already be freed.
3120 * @skb: A socket buffer.
3122 static inline void skb_tx_timestamp(struct sk_buff *skb)
3124 skb_clone_tx_timestamp(skb);
3125 sw_tx_timestamp(skb);
3129 * skb_complete_wifi_ack - deliver skb with wifi status
3131 * @skb: the original outgoing packet
3132 * @acked: ack status
3135 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3137 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3138 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3140 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3142 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3144 (skb->ip_summed == CHECKSUM_PARTIAL &&
3145 skb_checksum_start_offset(skb) >= 0));
3149 * skb_checksum_complete - Calculate checksum of an entire packet
3150 * @skb: packet to process
3152 * This function calculates the checksum over the entire packet plus
3153 * the value of skb->csum. The latter can be used to supply the
3154 * checksum of a pseudo header as used by TCP/UDP. It returns the
3157 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3158 * this function can be used to verify that checksum on received
3159 * packets. In that case the function should return zero if the
3160 * checksum is correct. In particular, this function will return zero
3161 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3162 * hardware has already verified the correctness of the checksum.
3164 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3166 return skb_csum_unnecessary(skb) ?
3167 0 : __skb_checksum_complete(skb);
3170 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3172 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3173 if (skb->csum_level == 0)
3174 skb->ip_summed = CHECKSUM_NONE;
3180 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3182 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3183 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3185 } else if (skb->ip_summed == CHECKSUM_NONE) {
3186 skb->ip_summed = CHECKSUM_UNNECESSARY;
3187 skb->csum_level = 0;
3191 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3193 /* Mark current checksum as bad (typically called from GRO
3194 * path). In the case that ip_summed is CHECKSUM_NONE
3195 * this must be the first checksum encountered in the packet.
3196 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3197 * checksum after the last one validated. For UDP, a zero
3198 * checksum can not be marked as bad.
3201 if (skb->ip_summed == CHECKSUM_NONE ||
3202 skb->ip_summed == CHECKSUM_UNNECESSARY)
3206 /* Check if we need to perform checksum complete validation.
3208 * Returns true if checksum complete is needed, false otherwise
3209 * (either checksum is unnecessary or zero checksum is allowed).
3211 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3215 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3216 skb->csum_valid = 1;
3217 __skb_decr_checksum_unnecessary(skb);
3224 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3227 #define CHECKSUM_BREAK 76
3229 /* Unset checksum-complete
3231 * Unset checksum complete can be done when packet is being modified
3232 * (uncompressed for instance) and checksum-complete value is
3235 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3237 if (skb->ip_summed == CHECKSUM_COMPLETE)
3238 skb->ip_summed = CHECKSUM_NONE;
3241 /* Validate (init) checksum based on checksum complete.
3244 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3245 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3246 * checksum is stored in skb->csum for use in __skb_checksum_complete
3247 * non-zero: value of invalid checksum
3250 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3254 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3255 if (!csum_fold(csum_add(psum, skb->csum))) {
3256 skb->csum_valid = 1;
3259 } else if (skb->csum_bad) {
3260 /* ip_summed == CHECKSUM_NONE in this case */
3261 return (__force __sum16)1;
3266 if (complete || skb->len <= CHECKSUM_BREAK) {
3269 csum = __skb_checksum_complete(skb);
3270 skb->csum_valid = !csum;
3277 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3282 /* Perform checksum validate (init). Note that this is a macro since we only
3283 * want to calculate the pseudo header which is an input function if necessary.
3284 * First we try to validate without any computation (checksum unnecessary) and
3285 * then calculate based on checksum complete calling the function to compute
3289 * 0: checksum is validated or try to in skb_checksum_complete
3290 * non-zero: value of invalid checksum
3292 #define __skb_checksum_validate(skb, proto, complete, \
3293 zero_okay, check, compute_pseudo) \
3295 __sum16 __ret = 0; \
3296 skb->csum_valid = 0; \
3297 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3298 __ret = __skb_checksum_validate_complete(skb, \
3299 complete, compute_pseudo(skb, proto)); \
3303 #define skb_checksum_init(skb, proto, compute_pseudo) \
3304 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3306 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3307 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3309 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3310 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3312 #define skb_checksum_validate_zero_check(skb, proto, check, \
3314 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3316 #define skb_checksum_simple_validate(skb) \
3317 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3319 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3321 return (skb->ip_summed == CHECKSUM_NONE &&
3322 skb->csum_valid && !skb->csum_bad);
3325 static inline void __skb_checksum_convert(struct sk_buff *skb,
3326 __sum16 check, __wsum pseudo)
3328 skb->csum = ~pseudo;
3329 skb->ip_summed = CHECKSUM_COMPLETE;
3332 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3334 if (__skb_checksum_convert_check(skb)) \
3335 __skb_checksum_convert(skb, check, \
3336 compute_pseudo(skb, proto)); \
3339 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3340 u16 start, u16 offset)
3342 skb->ip_summed = CHECKSUM_PARTIAL;
3343 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3344 skb->csum_offset = offset - start;
3347 /* Update skbuf and packet to reflect the remote checksum offload operation.
3348 * When called, ptr indicates the starting point for skb->csum when
3349 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3350 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3352 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3353 int start, int offset, bool nopartial)
3358 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3362 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3363 __skb_checksum_complete(skb);
3364 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3367 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3369 /* Adjust skb->csum since we changed the packet */
3370 skb->csum = csum_add(skb->csum, delta);
3373 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3374 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3375 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3377 if (nfct && atomic_dec_and_test(&nfct->use))
3378 nf_conntrack_destroy(nfct);
3380 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3383 atomic_inc(&nfct->use);
3386 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3387 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3389 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3392 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3395 atomic_inc(&nf_bridge->use);
3397 #endif /* CONFIG_BRIDGE_NETFILTER */
3398 static inline void nf_reset(struct sk_buff *skb)
3400 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3401 nf_conntrack_put(skb->nfct);
3404 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3405 nf_bridge_put(skb->nf_bridge);
3406 skb->nf_bridge = NULL;
3410 static inline void nf_reset_trace(struct sk_buff *skb)
3412 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3417 static inline void ipvs_reset(struct sk_buff *skb)
3419 #if IS_ENABLED(CONFIG_IP_VS)
3420 skb->ipvs_property = 0;
3424 /* Note: This doesn't put any conntrack and bridge info in dst. */
3425 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3428 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3429 dst->nfct = src->nfct;
3430 nf_conntrack_get(src->nfct);
3432 dst->nfctinfo = src->nfctinfo;
3434 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3435 dst->nf_bridge = src->nf_bridge;
3436 nf_bridge_get(src->nf_bridge);
3438 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3440 dst->nf_trace = src->nf_trace;
3444 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3446 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3447 nf_conntrack_put(dst->nfct);
3449 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3450 nf_bridge_put(dst->nf_bridge);
3452 __nf_copy(dst, src, true);
3455 #ifdef CONFIG_NETWORK_SECMARK
3456 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3458 to->secmark = from->secmark;
3461 static inline void skb_init_secmark(struct sk_buff *skb)
3466 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3469 static inline void skb_init_secmark(struct sk_buff *skb)
3473 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3475 return !skb->destructor &&
3476 #if IS_ENABLED(CONFIG_XFRM)
3479 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3482 !skb->_skb_refdst &&
3483 !skb_has_frag_list(skb);
3486 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3488 skb->queue_mapping = queue_mapping;
3491 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3493 return skb->queue_mapping;
3496 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3498 to->queue_mapping = from->queue_mapping;
3501 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3503 skb->queue_mapping = rx_queue + 1;
3506 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3508 return skb->queue_mapping - 1;
3511 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3513 return skb->queue_mapping != 0;
3516 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3525 /* Keeps track of mac header offset relative to skb->head.
3526 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3527 * For non-tunnel skb it points to skb_mac_header() and for
3528 * tunnel skb it points to outer mac header.
3529 * Keeps track of level of encapsulation of network headers.
3536 #define SKB_SGO_CB_OFFSET 32
3537 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3539 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3541 return (skb_mac_header(inner_skb) - inner_skb->head) -
3542 SKB_GSO_CB(inner_skb)->mac_offset;
3545 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3547 int new_headroom, headroom;
3550 headroom = skb_headroom(skb);
3551 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3555 new_headroom = skb_headroom(skb);
3556 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3560 /* Compute the checksum for a gso segment. First compute the checksum value
3561 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3562 * then add in skb->csum (checksum from csum_start to end of packet).
3563 * skb->csum and csum_start are then updated to reflect the checksum of the
3564 * resultant packet starting from the transport header-- the resultant checksum
3565 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3568 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3570 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3571 skb_transport_offset(skb);
3574 partial = csum_partial(skb_transport_header(skb), plen, skb->csum);
3576 SKB_GSO_CB(skb)->csum_start -= plen;
3578 return csum_fold(partial);
3581 static inline bool skb_is_gso(const struct sk_buff *skb)
3583 return skb_shinfo(skb)->gso_size;
3586 /* Note: Should be called only if skb_is_gso(skb) is true */
3587 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3589 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3592 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3594 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3596 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3597 * wanted then gso_type will be set. */
3598 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3600 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3601 unlikely(shinfo->gso_type == 0)) {
3602 __skb_warn_lro_forwarding(skb);
3608 static inline void skb_forward_csum(struct sk_buff *skb)
3610 /* Unfortunately we don't support this one. Any brave souls? */
3611 if (skb->ip_summed == CHECKSUM_COMPLETE)
3612 skb->ip_summed = CHECKSUM_NONE;
3616 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3617 * @skb: skb to check
3619 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3620 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3621 * use this helper, to document places where we make this assertion.
3623 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3626 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3630 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3632 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3633 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3634 unsigned int transport_len,
3635 __sum16(*skb_chkf)(struct sk_buff *skb));
3638 * skb_head_is_locked - Determine if the skb->head is locked down
3639 * @skb: skb to check
3641 * The head on skbs build around a head frag can be removed if they are
3642 * not cloned. This function returns true if the skb head is locked down
3643 * due to either being allocated via kmalloc, or by being a clone with
3644 * multiple references to the head.
3646 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3648 return !skb->head_frag || skb_cloned(skb);
3652 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3656 * skb_gso_network_seglen is used to determine the real size of the
3657 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3659 * The MAC/L2 header is not accounted for.
3661 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3663 unsigned int hdr_len = skb_transport_header(skb) -
3664 skb_network_header(skb);
3665 return hdr_len + skb_gso_transport_seglen(skb);
3669 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
3673 * skb_gso_mac_seglen is used to determine the real size of the
3674 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
3675 * headers (TCP/UDP).
3677 static inline unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
3679 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3680 return hdr_len + skb_gso_transport_seglen(skb);
3684 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
3687 * @len: length to validate against
3689 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
3690 * length once split, including L2, L3 and L4 headers and the payload.
3693 skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
3695 return skb_gso_mac_seglen(skb) <= len;
3698 #endif /* __KERNEL__ */
3699 #endif /* _LINUX_SKBUFF_H */