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>
24 #include <linux/atomic.h>
25 #include <asm/types.h>
26 #include <linux/spinlock.h>
27 #include <linux/net.h>
28 #include <linux/textsearch.h>
29 #include <net/checksum.h>
30 #include <linux/rcupdate.h>
31 #include <linux/hrtimer.h>
32 #include <linux/dma-mapping.h>
33 #include <linux/netdev_features.h>
34 #include <linux/sched.h>
35 #include <net/flow_keys.h>
37 /* A. Checksumming of received packets by device.
41 * Device failed to checksum this packet e.g. due to lack of capabilities.
42 * The packet contains full (though not verified) checksum in packet but
43 * not in skb->csum. Thus, skb->csum is undefined in this case.
45 * CHECKSUM_UNNECESSARY:
47 * The hardware you're dealing with doesn't calculate the full checksum
48 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
49 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
50 * if their checksums are okay. skb->csum is still undefined in this case
51 * though. It is a bad option, but, unfortunately, nowadays most vendors do
52 * this. Apparently with the secret goal to sell you new devices, when you
53 * will add new protocol to your host, f.e. IPv6 8)
55 * CHECKSUM_UNNECESSARY is applicable to following protocols:
57 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
58 * zero UDP checksum for either IPv4 or IPv6, the networking stack
59 * may perform further validation in this case.
60 * GRE: only if the checksum is present in the header.
61 * SCTP: indicates the CRC in SCTP header has been validated.
63 * skb->csum_level indicates the number of consecutive checksums found in
64 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
65 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
66 * and a device is able to verify the checksums for UDP (possibly zero),
67 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
68 * two. If the device were only able to verify the UDP checksum and not
69 * GRE, either because it doesn't support GRE checksum of because GRE
70 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
71 * not considered in this case).
75 * This is the most generic way. The device supplied checksum of the _whole_
76 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
77 * hardware doesn't need to parse L3/L4 headers to implement this.
79 * Note: Even if device supports only some protocols, but is able to produce
80 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
84 * This is identical to the case for output below. This may occur on a packet
85 * received directly from another Linux OS, e.g., a virtualized Linux kernel
86 * on the same host. The packet can be treated in the same way as
87 * CHECKSUM_UNNECESSARY, except that on output (i.e., forwarding) the
88 * checksum must be filled in by the OS or the hardware.
90 * B. Checksumming on output.
94 * The skb was already checksummed by the protocol, or a checksum is not
99 * The device is required to checksum the packet as seen by hard_start_xmit()
100 * from skb->csum_start up to the end, and to record/write the checksum at
101 * offset skb->csum_start + skb->csum_offset.
103 * The device must show its capabilities in dev->features, set up at device
104 * setup time, e.g. netdev_features.h:
106 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything.
107 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over
108 * IPv4. Sigh. Vendors like this way for an unknown reason.
109 * Though, see comment above about CHECKSUM_UNNECESSARY. 8)
110 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead.
111 * NETIF_F_... - Well, you get the picture.
113 * CHECKSUM_UNNECESSARY:
115 * Normally, the device will do per protocol specific checksumming. Protocol
116 * implementations that do not want the NIC to perform the checksum
117 * calculation should use this flag in their outgoing skbs.
119 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC
120 * offload. Correspondingly, the FCoE protocol driver
121 * stack should use CHECKSUM_UNNECESSARY.
123 * Any questions? No questions, good. --ANK
126 /* Don't change this without changing skb_csum_unnecessary! */
127 #define CHECKSUM_NONE 0
128 #define CHECKSUM_UNNECESSARY 1
129 #define CHECKSUM_COMPLETE 2
130 #define CHECKSUM_PARTIAL 3
132 /* Maximum value in skb->csum_level */
133 #define SKB_MAX_CSUM_LEVEL 3
135 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
136 #define SKB_WITH_OVERHEAD(X) \
137 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
138 #define SKB_MAX_ORDER(X, ORDER) \
139 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
140 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
141 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
143 /* return minimum truesize of one skb containing X bytes of data */
144 #define SKB_TRUESIZE(X) ((X) + \
145 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
146 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
150 struct pipe_inode_info;
152 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
153 struct nf_conntrack {
158 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
159 struct nf_bridge_info {
162 struct net_device *physindev;
163 struct net_device *physoutdev;
164 unsigned long data[32 / sizeof(unsigned long)];
168 struct sk_buff_head {
169 /* These two members must be first. */
170 struct sk_buff *next;
171 struct sk_buff *prev;
179 /* To allow 64K frame to be packed as single skb without frag_list we
180 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
181 * buffers which do not start on a page boundary.
183 * Since GRO uses frags we allocate at least 16 regardless of page
186 #if (65536/PAGE_SIZE + 1) < 16
187 #define MAX_SKB_FRAGS 16UL
189 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
192 typedef struct skb_frag_struct skb_frag_t;
194 struct skb_frag_struct {
198 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
207 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
212 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
217 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
222 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
227 #define HAVE_HW_TIME_STAMP
230 * struct skb_shared_hwtstamps - hardware time stamps
231 * @hwtstamp: hardware time stamp transformed into duration
232 * since arbitrary point in time
234 * Software time stamps generated by ktime_get_real() are stored in
237 * hwtstamps can only be compared against other hwtstamps from
240 * This structure is attached to packets as part of the
241 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
243 struct skb_shared_hwtstamps {
247 /* Definitions for tx_flags in struct skb_shared_info */
249 /* generate hardware time stamp */
250 SKBTX_HW_TSTAMP = 1 << 0,
252 /* generate software time stamp when queueing packet to NIC */
253 SKBTX_SW_TSTAMP = 1 << 1,
255 /* device driver is going to provide hardware time stamp */
256 SKBTX_IN_PROGRESS = 1 << 2,
258 /* device driver supports TX zero-copy buffers */
259 SKBTX_DEV_ZEROCOPY = 1 << 3,
261 /* generate wifi status information (where possible) */
262 SKBTX_WIFI_STATUS = 1 << 4,
264 /* This indicates at least one fragment might be overwritten
265 * (as in vmsplice(), sendfile() ...)
266 * If we need to compute a TX checksum, we'll need to copy
267 * all frags to avoid possible bad checksum
269 SKBTX_SHARED_FRAG = 1 << 5,
271 /* generate software time stamp when entering packet scheduling */
272 SKBTX_SCHED_TSTAMP = 1 << 6,
274 /* generate software timestamp on peer data acknowledgment */
275 SKBTX_ACK_TSTAMP = 1 << 7,
278 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
279 SKBTX_SCHED_TSTAMP | \
281 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
284 * The callback notifies userspace to release buffers when skb DMA is done in
285 * lower device, the skb last reference should be 0 when calling this.
286 * The zerocopy_success argument is true if zero copy transmit occurred,
287 * false on data copy or out of memory error caused by data copy attempt.
288 * The ctx field is used to track device context.
289 * The desc field is used to track userspace buffer index.
292 void (*callback)(struct ubuf_info *, bool zerocopy_success);
297 /* This data is invariant across clones and lives at
298 * the end of the header data, ie. at skb->end.
300 struct skb_shared_info {
301 unsigned char nr_frags;
303 unsigned short gso_size;
304 /* Warning: this field is not always filled in (UFO)! */
305 unsigned short gso_segs;
306 unsigned short gso_type;
307 struct sk_buff *frag_list;
308 struct skb_shared_hwtstamps hwtstamps;
313 * Warning : all fields before dataref are cleared in __alloc_skb()
317 /* Intermediate layers must ensure that destructor_arg
318 * remains valid until skb destructor */
319 void * destructor_arg;
321 /* must be last field, see pskb_expand_head() */
322 skb_frag_t frags[MAX_SKB_FRAGS];
325 /* We divide dataref into two halves. The higher 16 bits hold references
326 * to the payload part of skb->data. The lower 16 bits hold references to
327 * the entire skb->data. A clone of a headerless skb holds the length of
328 * the header in skb->hdr_len.
330 * All users must obey the rule that the skb->data reference count must be
331 * greater than or equal to the payload reference count.
333 * Holding a reference to the payload part means that the user does not
334 * care about modifications to the header part of skb->data.
336 #define SKB_DATAREF_SHIFT 16
337 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
341 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
342 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
343 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
344 SKB_FCLONE_FREE, /* this companion fclone skb is available */
348 SKB_GSO_TCPV4 = 1 << 0,
349 SKB_GSO_UDP = 1 << 1,
351 /* This indicates the skb is from an untrusted source. */
352 SKB_GSO_DODGY = 1 << 2,
354 /* This indicates the tcp segment has CWR set. */
355 SKB_GSO_TCP_ECN = 1 << 3,
357 SKB_GSO_TCPV6 = 1 << 4,
359 SKB_GSO_FCOE = 1 << 5,
361 SKB_GSO_GRE = 1 << 6,
363 SKB_GSO_GRE_CSUM = 1 << 7,
365 SKB_GSO_IPIP = 1 << 8,
367 SKB_GSO_SIT = 1 << 9,
369 SKB_GSO_UDP_TUNNEL = 1 << 10,
371 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
373 SKB_GSO_MPLS = 1 << 12,
377 #if BITS_PER_LONG > 32
378 #define NET_SKBUFF_DATA_USES_OFFSET 1
381 #ifdef NET_SKBUFF_DATA_USES_OFFSET
382 typedef unsigned int sk_buff_data_t;
384 typedef unsigned char *sk_buff_data_t;
388 * struct skb_mstamp - multi resolution time stamps
389 * @stamp_us: timestamp in us resolution
390 * @stamp_jiffies: timestamp in jiffies
403 * skb_mstamp_get - get current timestamp
404 * @cl: place to store timestamps
406 static inline void skb_mstamp_get(struct skb_mstamp *cl)
408 u64 val = local_clock();
410 do_div(val, NSEC_PER_USEC);
411 cl->stamp_us = (u32)val;
412 cl->stamp_jiffies = (u32)jiffies;
416 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
417 * @t1: pointer to newest sample
418 * @t0: pointer to oldest sample
420 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
421 const struct skb_mstamp *t0)
423 s32 delta_us = t1->stamp_us - t0->stamp_us;
424 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
426 /* If delta_us is negative, this might be because interval is too big,
427 * or local_clock() drift is too big : fallback using jiffies.
430 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
432 delta_us = jiffies_to_usecs(delta_jiffies);
439 * struct sk_buff - socket buffer
440 * @next: Next buffer in list
441 * @prev: Previous buffer in list
442 * @tstamp: Time we arrived/left
443 * @sk: Socket we are owned by
444 * @dev: Device we arrived on/are leaving by
445 * @cb: Control buffer. Free for use by every layer. Put private vars here
446 * @_skb_refdst: destination entry (with norefcount bit)
447 * @sp: the security path, used for xfrm
448 * @len: Length of actual data
449 * @data_len: Data length
450 * @mac_len: Length of link layer header
451 * @hdr_len: writable header length of cloned skb
452 * @csum: Checksum (must include start/offset pair)
453 * @csum_start: Offset from skb->head where checksumming should start
454 * @csum_offset: Offset from csum_start where checksum should be stored
455 * @priority: Packet queueing priority
456 * @ignore_df: allow local fragmentation
457 * @cloned: Head may be cloned (check refcnt to be sure)
458 * @ip_summed: Driver fed us an IP checksum
459 * @nohdr: Payload reference only, must not modify header
460 * @nfctinfo: Relationship of this skb to the connection
461 * @pkt_type: Packet class
462 * @fclone: skbuff clone status
463 * @ipvs_property: skbuff is owned by ipvs
464 * @peeked: this packet has been seen already, so stats have been
465 * done for it, don't do them again
466 * @nf_trace: netfilter packet trace flag
467 * @protocol: Packet protocol from driver
468 * @destructor: Destruct function
469 * @nfct: Associated connection, if any
470 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
471 * @skb_iif: ifindex of device we arrived on
472 * @tc_index: Traffic control index
473 * @tc_verd: traffic control verdict
474 * @hash: the packet hash
475 * @queue_mapping: Queue mapping for multiqueue devices
476 * @xmit_more: More SKBs are pending for this queue
477 * @ndisc_nodetype: router type (from link layer)
478 * @ooo_okay: allow the mapping of a socket to a queue to be changed
479 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
481 * @sw_hash: indicates hash was computed in software stack
482 * @wifi_acked_valid: wifi_acked was set
483 * @wifi_acked: whether frame was acked on wifi or not
484 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
485 * @napi_id: id of the NAPI struct this skb came from
486 * @secmark: security marking
487 * @mark: Generic packet mark
488 * @dropcount: total number of sk_receive_queue overflows
489 * @vlan_proto: vlan encapsulation protocol
490 * @vlan_tci: vlan tag control information
491 * @inner_protocol: Protocol (encapsulation)
492 * @inner_transport_header: Inner transport layer header (encapsulation)
493 * @inner_network_header: Network layer header (encapsulation)
494 * @inner_mac_header: Link layer header (encapsulation)
495 * @transport_header: Transport layer header
496 * @network_header: Network layer header
497 * @mac_header: Link layer header
498 * @tail: Tail pointer
500 * @head: Head of buffer
501 * @data: Data head pointer
502 * @truesize: Buffer size
503 * @users: User count - see {datagram,tcp}.c
507 /* These two members must be first. */
508 struct sk_buff *next;
509 struct sk_buff *prev;
513 struct skb_mstamp skb_mstamp;
517 struct net_device *dev;
520 * This is the control buffer. It is free to use for every
521 * layer. Please put your private variables there. If you
522 * want to keep them across layers you have to do a skb_clone()
523 * first. This is owned by whoever has the skb queued ATM.
525 char cb[48] __aligned(8);
527 unsigned long _skb_refdst;
528 void (*destructor)(struct sk_buff *skb);
532 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
533 struct nf_conntrack *nfct;
535 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
536 struct nf_bridge_info *nf_bridge;
543 /* Following fields are _not_ copied in __copy_skb_header()
544 * Note that queue_mapping is here mostly to fill a hole.
546 kmemcheck_bitfield_begin(flags1);
555 kmemcheck_bitfield_end(flags1);
557 /* fields enclosed in headers_start/headers_end are copied
558 * using a single memcpy() in __copy_skb_header()
561 __u32 headers_start[0];
564 /* if you move pkt_type around you also must adapt those constants */
565 #ifdef __BIG_ENDIAN_BITFIELD
566 #define PKT_TYPE_MAX (7 << 5)
568 #define PKT_TYPE_MAX 7
570 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
572 __u8 __pkt_type_offset[0];
583 __u8 wifi_acked_valid:1;
587 /* Indicates the inner headers are valid in the skbuff. */
588 __u8 encapsulation:1;
589 __u8 encap_hdr_csum:1;
591 __u8 csum_complete_sw:1;
595 #ifdef CONFIG_IPV6_NDISC_NODETYPE
596 __u8 ndisc_nodetype:2;
598 __u8 ipvs_property:1;
599 __u8 inner_protocol_type:1;
600 /* 4 or 6 bit hole */
602 #ifdef CONFIG_NET_SCHED
603 __u16 tc_index; /* traffic control index */
604 #ifdef CONFIG_NET_CLS_ACT
605 __u16 tc_verd; /* traffic control verdict */
621 #ifdef CONFIG_NET_RX_BUSY_POLL
622 unsigned int napi_id;
624 #ifdef CONFIG_NETWORK_SECMARK
630 __u32 reserved_tailroom;
634 __be16 inner_protocol;
638 __u16 inner_transport_header;
639 __u16 inner_network_header;
640 __u16 inner_mac_header;
643 __u16 transport_header;
644 __u16 network_header;
648 __u32 headers_end[0];
651 /* These elements must be at the end, see alloc_skb() for details. */
656 unsigned int truesize;
662 * Handling routines are only of interest to the kernel
664 #include <linux/slab.h>
667 #define SKB_ALLOC_FCLONE 0x01
668 #define SKB_ALLOC_RX 0x02
670 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
671 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
673 return unlikely(skb->pfmemalloc);
677 * skb might have a dst pointer attached, refcounted or not.
678 * _skb_refdst low order bit is set if refcount was _not_ taken
680 #define SKB_DST_NOREF 1UL
681 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
684 * skb_dst - returns skb dst_entry
687 * Returns skb dst_entry, regardless of reference taken or not.
689 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
691 /* If refdst was not refcounted, check we still are in a
692 * rcu_read_lock section
694 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
695 !rcu_read_lock_held() &&
696 !rcu_read_lock_bh_held());
697 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
701 * skb_dst_set - sets skb dst
705 * Sets skb dst, assuming a reference was taken on dst and should
706 * be released by skb_dst_drop()
708 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
710 skb->_skb_refdst = (unsigned long)dst;
713 void __skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst,
717 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
721 * Sets skb dst, assuming a reference was not taken on dst.
722 * If dst entry is cached, we do not take reference and dst_release
723 * will be avoided by refdst_drop. If dst entry is not cached, we take
724 * reference, so that last dst_release can destroy the dst immediately.
726 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
728 __skb_dst_set_noref(skb, dst, false);
732 * skb_dst_set_noref_force - sets skb dst, without taking reference
736 * Sets skb dst, assuming a reference was not taken on dst.
737 * No reference is taken and no dst_release will be called. While for
738 * cached dsts deferred reclaim is a basic feature, for entries that are
739 * not cached it is caller's job to guarantee that last dst_release for
740 * provided dst happens when nobody uses it, eg. after a RCU grace period.
742 static inline void skb_dst_set_noref_force(struct sk_buff *skb,
743 struct dst_entry *dst)
745 __skb_dst_set_noref(skb, dst, true);
749 * skb_dst_is_noref - Test if skb dst isn't refcounted
752 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
754 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
757 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
759 return (struct rtable *)skb_dst(skb);
762 void kfree_skb(struct sk_buff *skb);
763 void kfree_skb_list(struct sk_buff *segs);
764 void skb_tx_error(struct sk_buff *skb);
765 void consume_skb(struct sk_buff *skb);
766 void __kfree_skb(struct sk_buff *skb);
767 extern struct kmem_cache *skbuff_head_cache;
769 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
770 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
771 bool *fragstolen, int *delta_truesize);
773 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
775 struct sk_buff *build_skb(void *data, unsigned int frag_size);
776 static inline struct sk_buff *alloc_skb(unsigned int size,
779 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
782 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
783 unsigned long data_len,
788 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
789 struct sk_buff_fclones {
798 * skb_fclone_busy - check if fclone is busy
801 * Returns true is skb is a fast clone, and its clone is not freed.
802 * Some drivers call skb_orphan() in their ndo_start_xmit(),
803 * so we also check that this didnt happen.
805 static inline bool skb_fclone_busy(const struct sock *sk,
806 const struct sk_buff *skb)
808 const struct sk_buff_fclones *fclones;
810 fclones = container_of(skb, struct sk_buff_fclones, skb1);
812 return skb->fclone == SKB_FCLONE_ORIG &&
813 fclones->skb2.fclone == SKB_FCLONE_CLONE &&
814 fclones->skb2.sk == sk;
817 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
820 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
823 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
824 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
826 return __alloc_skb_head(priority, -1);
829 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
830 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
831 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
832 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
833 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
834 gfp_t gfp_mask, bool fclone);
835 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
838 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
841 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
842 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
843 unsigned int headroom);
844 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
845 int newtailroom, gfp_t priority);
846 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
847 int offset, int len);
848 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
850 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
851 int skb_pad(struct sk_buff *skb, int pad);
852 #define dev_kfree_skb(a) consume_skb(a)
854 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
855 int getfrag(void *from, char *to, int offset,
856 int len, int odd, struct sk_buff *skb),
857 void *from, int length);
859 struct skb_seq_state {
863 __u32 stepped_offset;
864 struct sk_buff *root_skb;
865 struct sk_buff *cur_skb;
869 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
870 unsigned int to, struct skb_seq_state *st);
871 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
872 struct skb_seq_state *st);
873 void skb_abort_seq_read(struct skb_seq_state *st);
875 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
876 unsigned int to, struct ts_config *config,
877 struct ts_state *state);
880 * Packet hash types specify the type of hash in skb_set_hash.
882 * Hash types refer to the protocol layer addresses which are used to
883 * construct a packet's hash. The hashes are used to differentiate or identify
884 * flows of the protocol layer for the hash type. Hash types are either
885 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
887 * Properties of hashes:
889 * 1) Two packets in different flows have different hash values
890 * 2) Two packets in the same flow should have the same hash value
892 * A hash at a higher layer is considered to be more specific. A driver should
893 * set the most specific hash possible.
895 * A driver cannot indicate a more specific hash than the layer at which a hash
896 * was computed. For instance an L3 hash cannot be set as an L4 hash.
898 * A driver may indicate a hash level which is less specific than the
899 * actual layer the hash was computed on. For instance, a hash computed
900 * at L4 may be considered an L3 hash. This should only be done if the
901 * driver can't unambiguously determine that the HW computed the hash at
902 * the higher layer. Note that the "should" in the second property above
905 enum pkt_hash_types {
906 PKT_HASH_TYPE_NONE, /* Undefined type */
907 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
908 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
909 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
913 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
915 skb->l4_hash = (type == PKT_HASH_TYPE_L4);
920 void __skb_get_hash(struct sk_buff *skb);
921 static inline __u32 skb_get_hash(struct sk_buff *skb)
923 if (!skb->l4_hash && !skb->sw_hash)
929 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
934 static inline void skb_clear_hash(struct sk_buff *skb)
941 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
947 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
949 to->hash = from->hash;
950 to->sw_hash = from->sw_hash;
951 to->l4_hash = from->l4_hash;
954 #ifdef NET_SKBUFF_DATA_USES_OFFSET
955 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
957 return skb->head + skb->end;
960 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
965 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
970 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
972 return skb->end - skb->head;
977 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
979 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
981 return &skb_shinfo(skb)->hwtstamps;
985 * skb_queue_empty - check if a queue is empty
988 * Returns true if the queue is empty, false otherwise.
990 static inline int skb_queue_empty(const struct sk_buff_head *list)
992 return list->next == (const struct sk_buff *) list;
996 * skb_queue_is_last - check if skb is the last entry in the queue
1000 * Returns true if @skb is the last buffer on the list.
1002 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1003 const struct sk_buff *skb)
1005 return skb->next == (const struct sk_buff *) list;
1009 * skb_queue_is_first - check if skb is the first entry in the queue
1013 * Returns true if @skb is the first buffer on the list.
1015 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1016 const struct sk_buff *skb)
1018 return skb->prev == (const struct sk_buff *) list;
1022 * skb_queue_next - return the next packet in the queue
1024 * @skb: current buffer
1026 * Return the next packet in @list after @skb. It is only valid to
1027 * call this if skb_queue_is_last() evaluates to false.
1029 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1030 const struct sk_buff *skb)
1032 /* This BUG_ON may seem severe, but if we just return then we
1033 * are going to dereference garbage.
1035 BUG_ON(skb_queue_is_last(list, skb));
1040 * skb_queue_prev - return the prev packet in the queue
1042 * @skb: current buffer
1044 * Return the prev packet in @list before @skb. It is only valid to
1045 * call this if skb_queue_is_first() evaluates to false.
1047 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1048 const struct sk_buff *skb)
1050 /* This BUG_ON may seem severe, but if we just return then we
1051 * are going to dereference garbage.
1053 BUG_ON(skb_queue_is_first(list, skb));
1058 * skb_get - reference buffer
1059 * @skb: buffer to reference
1061 * Makes another reference to a socket buffer and returns a pointer
1064 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1066 atomic_inc(&skb->users);
1071 * If users == 1, we are the only owner and are can avoid redundant
1076 * skb_cloned - is the buffer a clone
1077 * @skb: buffer to check
1079 * Returns true if the buffer was generated with skb_clone() and is
1080 * one of multiple shared copies of the buffer. Cloned buffers are
1081 * shared data so must not be written to under normal circumstances.
1083 static inline int skb_cloned(const struct sk_buff *skb)
1085 return skb->cloned &&
1086 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1089 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1091 might_sleep_if(pri & __GFP_WAIT);
1093 if (skb_cloned(skb))
1094 return pskb_expand_head(skb, 0, 0, pri);
1100 * skb_header_cloned - is the header a clone
1101 * @skb: buffer to check
1103 * Returns true if modifying the header part of the buffer requires
1104 * the data to be copied.
1106 static inline int skb_header_cloned(const struct sk_buff *skb)
1113 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1114 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1115 return dataref != 1;
1119 * skb_header_release - release reference to header
1120 * @skb: buffer to operate on
1122 * Drop a reference to the header part of the buffer. This is done
1123 * by acquiring a payload reference. You must not read from the header
1124 * part of skb->data after this.
1125 * Note : Check if you can use __skb_header_release() instead.
1127 static inline void skb_header_release(struct sk_buff *skb)
1131 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1135 * __skb_header_release - release reference to header
1136 * @skb: buffer to operate on
1138 * Variant of skb_header_release() assuming skb is private to caller.
1139 * We can avoid one atomic operation.
1141 static inline void __skb_header_release(struct sk_buff *skb)
1144 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1149 * skb_shared - is the buffer shared
1150 * @skb: buffer to check
1152 * Returns true if more than one person has a reference to this
1155 static inline int skb_shared(const struct sk_buff *skb)
1157 return atomic_read(&skb->users) != 1;
1161 * skb_share_check - check if buffer is shared and if so clone it
1162 * @skb: buffer to check
1163 * @pri: priority for memory allocation
1165 * If the buffer is shared the buffer is cloned and the old copy
1166 * drops a reference. A new clone with a single reference is returned.
1167 * If the buffer is not shared the original buffer is returned. When
1168 * being called from interrupt status or with spinlocks held pri must
1171 * NULL is returned on a memory allocation failure.
1173 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1175 might_sleep_if(pri & __GFP_WAIT);
1176 if (skb_shared(skb)) {
1177 struct sk_buff *nskb = skb_clone(skb, pri);
1189 * Copy shared buffers into a new sk_buff. We effectively do COW on
1190 * packets to handle cases where we have a local reader and forward
1191 * and a couple of other messy ones. The normal one is tcpdumping
1192 * a packet thats being forwarded.
1196 * skb_unshare - make a copy of a shared buffer
1197 * @skb: buffer to check
1198 * @pri: priority for memory allocation
1200 * If the socket buffer is a clone then this function creates a new
1201 * copy of the data, drops a reference count on the old copy and returns
1202 * the new copy with the reference count at 1. If the buffer is not a clone
1203 * the original buffer is returned. When called with a spinlock held or
1204 * from interrupt state @pri must be %GFP_ATOMIC
1206 * %NULL is returned on a memory allocation failure.
1208 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1211 might_sleep_if(pri & __GFP_WAIT);
1212 if (skb_cloned(skb)) {
1213 struct sk_buff *nskb = skb_copy(skb, pri);
1215 /* Free our shared copy */
1226 * skb_peek - peek at the head of an &sk_buff_head
1227 * @list_: list to peek at
1229 * Peek an &sk_buff. Unlike most other operations you _MUST_
1230 * be careful with this one. A peek leaves the buffer on the
1231 * list and someone else may run off with it. You must hold
1232 * the appropriate locks or have a private queue to do this.
1234 * Returns %NULL for an empty list or a pointer to the head element.
1235 * The reference count is not incremented and the reference is therefore
1236 * volatile. Use with caution.
1238 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1240 struct sk_buff *skb = list_->next;
1242 if (skb == (struct sk_buff *)list_)
1248 * skb_peek_next - peek skb following the given one from a queue
1249 * @skb: skb to start from
1250 * @list_: list to peek at
1252 * Returns %NULL when the end of the list is met or a pointer to the
1253 * next element. The reference count is not incremented and the
1254 * reference is therefore volatile. Use with caution.
1256 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1257 const struct sk_buff_head *list_)
1259 struct sk_buff *next = skb->next;
1261 if (next == (struct sk_buff *)list_)
1267 * skb_peek_tail - peek at the tail of an &sk_buff_head
1268 * @list_: list to peek at
1270 * Peek an &sk_buff. Unlike most other operations you _MUST_
1271 * be careful with this one. A peek leaves the buffer on the
1272 * list and someone else may run off with it. You must hold
1273 * the appropriate locks or have a private queue to do this.
1275 * Returns %NULL for an empty list or a pointer to the tail element.
1276 * The reference count is not incremented and the reference is therefore
1277 * volatile. Use with caution.
1279 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1281 struct sk_buff *skb = list_->prev;
1283 if (skb == (struct sk_buff *)list_)
1290 * skb_queue_len - get queue length
1291 * @list_: list to measure
1293 * Return the length of an &sk_buff queue.
1295 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1301 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1302 * @list: queue to initialize
1304 * This initializes only the list and queue length aspects of
1305 * an sk_buff_head object. This allows to initialize the list
1306 * aspects of an sk_buff_head without reinitializing things like
1307 * the spinlock. It can also be used for on-stack sk_buff_head
1308 * objects where the spinlock is known to not be used.
1310 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1312 list->prev = list->next = (struct sk_buff *)list;
1317 * This function creates a split out lock class for each invocation;
1318 * this is needed for now since a whole lot of users of the skb-queue
1319 * infrastructure in drivers have different locking usage (in hardirq)
1320 * than the networking core (in softirq only). In the long run either the
1321 * network layer or drivers should need annotation to consolidate the
1322 * main types of usage into 3 classes.
1324 static inline void skb_queue_head_init(struct sk_buff_head *list)
1326 spin_lock_init(&list->lock);
1327 __skb_queue_head_init(list);
1330 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1331 struct lock_class_key *class)
1333 skb_queue_head_init(list);
1334 lockdep_set_class(&list->lock, class);
1338 * Insert an sk_buff on a list.
1340 * The "__skb_xxxx()" functions are the non-atomic ones that
1341 * can only be called with interrupts disabled.
1343 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1344 struct sk_buff_head *list);
1345 static inline void __skb_insert(struct sk_buff *newsk,
1346 struct sk_buff *prev, struct sk_buff *next,
1347 struct sk_buff_head *list)
1351 next->prev = prev->next = newsk;
1355 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1356 struct sk_buff *prev,
1357 struct sk_buff *next)
1359 struct sk_buff *first = list->next;
1360 struct sk_buff *last = list->prev;
1370 * skb_queue_splice - join two skb lists, this is designed for stacks
1371 * @list: the new list to add
1372 * @head: the place to add it in the first list
1374 static inline void skb_queue_splice(const struct sk_buff_head *list,
1375 struct sk_buff_head *head)
1377 if (!skb_queue_empty(list)) {
1378 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1379 head->qlen += list->qlen;
1384 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1385 * @list: the new list to add
1386 * @head: the place to add it in the first list
1388 * The list at @list is reinitialised
1390 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1391 struct sk_buff_head *head)
1393 if (!skb_queue_empty(list)) {
1394 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1395 head->qlen += list->qlen;
1396 __skb_queue_head_init(list);
1401 * skb_queue_splice_tail - join two skb lists, each list being a queue
1402 * @list: the new list to add
1403 * @head: the place to add it in the first list
1405 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1406 struct sk_buff_head *head)
1408 if (!skb_queue_empty(list)) {
1409 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1410 head->qlen += list->qlen;
1415 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1416 * @list: the new list to add
1417 * @head: the place to add it in the first list
1419 * Each of the lists is a queue.
1420 * The list at @list is reinitialised
1422 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1423 struct sk_buff_head *head)
1425 if (!skb_queue_empty(list)) {
1426 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1427 head->qlen += list->qlen;
1428 __skb_queue_head_init(list);
1433 * __skb_queue_after - queue a buffer at the list head
1434 * @list: list to use
1435 * @prev: place after this buffer
1436 * @newsk: buffer to queue
1438 * Queue a buffer int the middle of a list. This function takes no locks
1439 * and you must therefore hold required locks before calling it.
1441 * A buffer cannot be placed on two lists at the same time.
1443 static inline void __skb_queue_after(struct sk_buff_head *list,
1444 struct sk_buff *prev,
1445 struct sk_buff *newsk)
1447 __skb_insert(newsk, prev, prev->next, list);
1450 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1451 struct sk_buff_head *list);
1453 static inline void __skb_queue_before(struct sk_buff_head *list,
1454 struct sk_buff *next,
1455 struct sk_buff *newsk)
1457 __skb_insert(newsk, next->prev, next, list);
1461 * __skb_queue_head - queue a buffer at the list head
1462 * @list: list to use
1463 * @newsk: buffer to queue
1465 * Queue a buffer at the start of a list. This function takes no locks
1466 * and you must therefore hold required locks before calling it.
1468 * A buffer cannot be placed on two lists at the same time.
1470 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1471 static inline void __skb_queue_head(struct sk_buff_head *list,
1472 struct sk_buff *newsk)
1474 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1478 * __skb_queue_tail - queue a buffer at the list tail
1479 * @list: list to use
1480 * @newsk: buffer to queue
1482 * Queue a buffer at the end of a list. This function takes no locks
1483 * and you must therefore hold required locks before calling it.
1485 * A buffer cannot be placed on two lists at the same time.
1487 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1488 static inline void __skb_queue_tail(struct sk_buff_head *list,
1489 struct sk_buff *newsk)
1491 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1495 * remove sk_buff from list. _Must_ be called atomically, and with
1498 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1499 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1501 struct sk_buff *next, *prev;
1506 skb->next = skb->prev = NULL;
1512 * __skb_dequeue - remove from the head of the queue
1513 * @list: list to dequeue from
1515 * Remove the head of the list. This function does not take any locks
1516 * so must be used with appropriate locks held only. The head item is
1517 * returned or %NULL if the list is empty.
1519 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1520 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1522 struct sk_buff *skb = skb_peek(list);
1524 __skb_unlink(skb, list);
1529 * __skb_dequeue_tail - remove from the tail of the queue
1530 * @list: list to dequeue from
1532 * Remove the tail of the list. This function does not take any locks
1533 * so must be used with appropriate locks held only. The tail item is
1534 * returned or %NULL if the list is empty.
1536 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1537 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1539 struct sk_buff *skb = skb_peek_tail(list);
1541 __skb_unlink(skb, list);
1546 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1548 return skb->data_len;
1551 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1553 return skb->len - skb->data_len;
1556 static inline int skb_pagelen(const struct sk_buff *skb)
1560 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1561 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1562 return len + skb_headlen(skb);
1566 * __skb_fill_page_desc - initialise a paged fragment in an skb
1567 * @skb: buffer containing fragment to be initialised
1568 * @i: paged fragment index to initialise
1569 * @page: the page to use for this fragment
1570 * @off: the offset to the data with @page
1571 * @size: the length of the data
1573 * Initialises the @i'th fragment of @skb to point to &size bytes at
1574 * offset @off within @page.
1576 * Does not take any additional reference on the fragment.
1578 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1579 struct page *page, int off, int size)
1581 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1584 * Propagate page->pfmemalloc to the skb if we can. The problem is
1585 * that not all callers have unique ownership of the page. If
1586 * pfmemalloc is set, we check the mapping as a mapping implies
1587 * page->index is set (index and pfmemalloc share space).
1588 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1589 * do not lose pfmemalloc information as the pages would not be
1590 * allocated using __GFP_MEMALLOC.
1592 frag->page.p = page;
1593 frag->page_offset = off;
1594 skb_frag_size_set(frag, size);
1596 page = compound_head(page);
1597 if (page->pfmemalloc && !page->mapping)
1598 skb->pfmemalloc = true;
1602 * skb_fill_page_desc - initialise a paged fragment in an skb
1603 * @skb: buffer containing fragment to be initialised
1604 * @i: paged fragment index to initialise
1605 * @page: the page to use for this fragment
1606 * @off: the offset to the data with @page
1607 * @size: the length of the data
1609 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1610 * @skb to point to @size bytes at offset @off within @page. In
1611 * addition updates @skb such that @i is the last fragment.
1613 * Does not take any additional reference on the fragment.
1615 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1616 struct page *page, int off, int size)
1618 __skb_fill_page_desc(skb, i, page, off, size);
1619 skb_shinfo(skb)->nr_frags = i + 1;
1622 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1623 int size, unsigned int truesize);
1625 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1626 unsigned int truesize);
1628 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1629 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1630 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1632 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1633 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1635 return skb->head + skb->tail;
1638 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1640 skb->tail = skb->data - skb->head;
1643 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1645 skb_reset_tail_pointer(skb);
1646 skb->tail += offset;
1649 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1650 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1655 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1657 skb->tail = skb->data;
1660 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1662 skb->tail = skb->data + offset;
1665 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1668 * Add data to an sk_buff
1670 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1671 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1672 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1674 unsigned char *tmp = skb_tail_pointer(skb);
1675 SKB_LINEAR_ASSERT(skb);
1681 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1682 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1689 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1690 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1693 BUG_ON(skb->len < skb->data_len);
1694 return skb->data += len;
1697 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1699 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1702 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1704 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1706 if (len > skb_headlen(skb) &&
1707 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1710 return skb->data += len;
1713 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1715 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1718 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1720 if (likely(len <= skb_headlen(skb)))
1722 if (unlikely(len > skb->len))
1724 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1728 * skb_headroom - bytes at buffer head
1729 * @skb: buffer to check
1731 * Return the number of bytes of free space at the head of an &sk_buff.
1733 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1735 return skb->data - skb->head;
1739 * skb_tailroom - bytes at buffer end
1740 * @skb: buffer to check
1742 * Return the number of bytes of free space at the tail of an sk_buff
1744 static inline int skb_tailroom(const struct sk_buff *skb)
1746 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1750 * skb_availroom - bytes at buffer end
1751 * @skb: buffer to check
1753 * Return the number of bytes of free space at the tail of an sk_buff
1754 * allocated by sk_stream_alloc()
1756 static inline int skb_availroom(const struct sk_buff *skb)
1758 if (skb_is_nonlinear(skb))
1761 return skb->end - skb->tail - skb->reserved_tailroom;
1765 * skb_reserve - adjust headroom
1766 * @skb: buffer to alter
1767 * @len: bytes to move
1769 * Increase the headroom of an empty &sk_buff by reducing the tail
1770 * room. This is only allowed for an empty buffer.
1772 static inline void skb_reserve(struct sk_buff *skb, int len)
1778 #define ENCAP_TYPE_ETHER 0
1779 #define ENCAP_TYPE_IPPROTO 1
1781 static inline void skb_set_inner_protocol(struct sk_buff *skb,
1784 skb->inner_protocol = protocol;
1785 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
1788 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
1791 skb->inner_ipproto = ipproto;
1792 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
1795 static inline void skb_reset_inner_headers(struct sk_buff *skb)
1797 skb->inner_mac_header = skb->mac_header;
1798 skb->inner_network_header = skb->network_header;
1799 skb->inner_transport_header = skb->transport_header;
1802 static inline void skb_reset_mac_len(struct sk_buff *skb)
1804 skb->mac_len = skb->network_header - skb->mac_header;
1807 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1810 return skb->head + skb->inner_transport_header;
1813 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1815 skb->inner_transport_header = skb->data - skb->head;
1818 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1821 skb_reset_inner_transport_header(skb);
1822 skb->inner_transport_header += offset;
1825 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1827 return skb->head + skb->inner_network_header;
1830 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1832 skb->inner_network_header = skb->data - skb->head;
1835 static inline void skb_set_inner_network_header(struct sk_buff *skb,
1838 skb_reset_inner_network_header(skb);
1839 skb->inner_network_header += offset;
1842 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1844 return skb->head + skb->inner_mac_header;
1847 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1849 skb->inner_mac_header = skb->data - skb->head;
1852 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1855 skb_reset_inner_mac_header(skb);
1856 skb->inner_mac_header += offset;
1858 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1860 return skb->transport_header != (typeof(skb->transport_header))~0U;
1863 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1865 return skb->head + skb->transport_header;
1868 static inline void skb_reset_transport_header(struct sk_buff *skb)
1870 skb->transport_header = skb->data - skb->head;
1873 static inline void skb_set_transport_header(struct sk_buff *skb,
1876 skb_reset_transport_header(skb);
1877 skb->transport_header += offset;
1880 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1882 return skb->head + skb->network_header;
1885 static inline void skb_reset_network_header(struct sk_buff *skb)
1887 skb->network_header = skb->data - skb->head;
1890 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1892 skb_reset_network_header(skb);
1893 skb->network_header += offset;
1896 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1898 return skb->head + skb->mac_header;
1901 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1903 return skb->mac_header != (typeof(skb->mac_header))~0U;
1906 static inline void skb_reset_mac_header(struct sk_buff *skb)
1908 skb->mac_header = skb->data - skb->head;
1911 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1913 skb_reset_mac_header(skb);
1914 skb->mac_header += offset;
1917 static inline void skb_pop_mac_header(struct sk_buff *skb)
1919 skb->mac_header = skb->network_header;
1922 static inline void skb_probe_transport_header(struct sk_buff *skb,
1923 const int offset_hint)
1925 struct flow_keys keys;
1927 if (skb_transport_header_was_set(skb))
1929 else if (skb_flow_dissect(skb, &keys))
1930 skb_set_transport_header(skb, keys.thoff);
1932 skb_set_transport_header(skb, offset_hint);
1935 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1937 if (skb_mac_header_was_set(skb)) {
1938 const unsigned char *old_mac = skb_mac_header(skb);
1940 skb_set_mac_header(skb, -skb->mac_len);
1941 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
1945 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
1947 return skb->csum_start - skb_headroom(skb);
1950 static inline int skb_transport_offset(const struct sk_buff *skb)
1952 return skb_transport_header(skb) - skb->data;
1955 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1957 return skb->transport_header - skb->network_header;
1960 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
1962 return skb->inner_transport_header - skb->inner_network_header;
1965 static inline int skb_network_offset(const struct sk_buff *skb)
1967 return skb_network_header(skb) - skb->data;
1970 static inline int skb_inner_network_offset(const struct sk_buff *skb)
1972 return skb_inner_network_header(skb) - skb->data;
1975 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1977 return pskb_may_pull(skb, skb_network_offset(skb) + len);
1981 * CPUs often take a performance hit when accessing unaligned memory
1982 * locations. The actual performance hit varies, it can be small if the
1983 * hardware handles it or large if we have to take an exception and fix it
1986 * Since an ethernet header is 14 bytes network drivers often end up with
1987 * the IP header at an unaligned offset. The IP header can be aligned by
1988 * shifting the start of the packet by 2 bytes. Drivers should do this
1991 * skb_reserve(skb, NET_IP_ALIGN);
1993 * The downside to this alignment of the IP header is that the DMA is now
1994 * unaligned. On some architectures the cost of an unaligned DMA is high
1995 * and this cost outweighs the gains made by aligning the IP header.
1997 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2000 #ifndef NET_IP_ALIGN
2001 #define NET_IP_ALIGN 2
2005 * The networking layer reserves some headroom in skb data (via
2006 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2007 * the header has to grow. In the default case, if the header has to grow
2008 * 32 bytes or less we avoid the reallocation.
2010 * Unfortunately this headroom changes the DMA alignment of the resulting
2011 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2012 * on some architectures. An architecture can override this value,
2013 * perhaps setting it to a cacheline in size (since that will maintain
2014 * cacheline alignment of the DMA). It must be a power of 2.
2016 * Various parts of the networking layer expect at least 32 bytes of
2017 * headroom, you should not reduce this.
2019 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2020 * to reduce average number of cache lines per packet.
2021 * get_rps_cpus() for example only access one 64 bytes aligned block :
2022 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2025 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2028 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2030 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2032 if (unlikely(skb_is_nonlinear(skb))) {
2037 skb_set_tail_pointer(skb, len);
2040 void skb_trim(struct sk_buff *skb, unsigned int len);
2042 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2045 return ___pskb_trim(skb, len);
2046 __skb_trim(skb, len);
2050 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2052 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2056 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2057 * @skb: buffer to alter
2060 * This is identical to pskb_trim except that the caller knows that
2061 * the skb is not cloned so we should never get an error due to out-
2064 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2066 int err = pskb_trim(skb, len);
2071 * skb_orphan - orphan a buffer
2072 * @skb: buffer to orphan
2074 * If a buffer currently has an owner then we call the owner's
2075 * destructor function and make the @skb unowned. The buffer continues
2076 * to exist but is no longer charged to its former owner.
2078 static inline void skb_orphan(struct sk_buff *skb)
2080 if (skb->destructor) {
2081 skb->destructor(skb);
2082 skb->destructor = NULL;
2090 * skb_orphan_frags - orphan the frags contained in a buffer
2091 * @skb: buffer to orphan frags from
2092 * @gfp_mask: allocation mask for replacement pages
2094 * For each frag in the SKB which needs a destructor (i.e. has an
2095 * owner) create a copy of that frag and release the original
2096 * page by calling the destructor.
2098 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2100 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2102 return skb_copy_ubufs(skb, gfp_mask);
2106 * __skb_queue_purge - empty a list
2107 * @list: list to empty
2109 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2110 * the list and one reference dropped. This function does not take the
2111 * list lock and the caller must hold the relevant locks to use it.
2113 void skb_queue_purge(struct sk_buff_head *list);
2114 static inline void __skb_queue_purge(struct sk_buff_head *list)
2116 struct sk_buff *skb;
2117 while ((skb = __skb_dequeue(list)) != NULL)
2121 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
2122 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
2123 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
2125 void *netdev_alloc_frag(unsigned int fragsz);
2127 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2131 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2132 * @dev: network device to receive on
2133 * @length: length to allocate
2135 * Allocate a new &sk_buff and assign it a usage count of one. The
2136 * buffer has unspecified headroom built in. Users should allocate
2137 * the headroom they think they need without accounting for the
2138 * built in space. The built in space is used for optimisations.
2140 * %NULL is returned if there is no free memory. Although this function
2141 * allocates memory it can be called from an interrupt.
2143 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2144 unsigned int length)
2146 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2149 /* legacy helper around __netdev_alloc_skb() */
2150 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2153 return __netdev_alloc_skb(NULL, length, gfp_mask);
2156 /* legacy helper around netdev_alloc_skb() */
2157 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2159 return netdev_alloc_skb(NULL, length);
2163 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2164 unsigned int length, gfp_t gfp)
2166 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2168 if (NET_IP_ALIGN && skb)
2169 skb_reserve(skb, NET_IP_ALIGN);
2173 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2174 unsigned int length)
2176 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2180 * __skb_alloc_pages - allocate pages for ps-rx on a skb and preserve pfmemalloc data
2181 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2182 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2183 * @order: size of the allocation
2185 * Allocate a new page.
2187 * %NULL is returned if there is no free memory.
2189 static inline struct page *__skb_alloc_pages(gfp_t gfp_mask,
2190 struct sk_buff *skb,
2195 gfp_mask |= __GFP_COLD;
2197 if (!(gfp_mask & __GFP_NOMEMALLOC))
2198 gfp_mask |= __GFP_MEMALLOC;
2200 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2201 if (skb && page && page->pfmemalloc)
2202 skb->pfmemalloc = true;
2208 * __skb_alloc_page - allocate a page for ps-rx for a given skb and preserve pfmemalloc data
2209 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2210 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2212 * Allocate a new page.
2214 * %NULL is returned if there is no free memory.
2216 static inline struct page *__skb_alloc_page(gfp_t gfp_mask,
2217 struct sk_buff *skb)
2219 return __skb_alloc_pages(gfp_mask, skb, 0);
2223 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2224 * @page: The page that was allocated from skb_alloc_page
2225 * @skb: The skb that may need pfmemalloc set
2227 static inline void skb_propagate_pfmemalloc(struct page *page,
2228 struct sk_buff *skb)
2230 if (page && page->pfmemalloc)
2231 skb->pfmemalloc = true;
2235 * skb_frag_page - retrieve the page referred to by a paged fragment
2236 * @frag: the paged fragment
2238 * Returns the &struct page associated with @frag.
2240 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2242 return frag->page.p;
2246 * __skb_frag_ref - take an addition reference on a paged fragment.
2247 * @frag: the paged fragment
2249 * Takes an additional reference on the paged fragment @frag.
2251 static inline void __skb_frag_ref(skb_frag_t *frag)
2253 get_page(skb_frag_page(frag));
2257 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2259 * @f: the fragment offset.
2261 * Takes an additional reference on the @f'th paged fragment of @skb.
2263 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2265 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2269 * __skb_frag_unref - release a reference on a paged fragment.
2270 * @frag: the paged fragment
2272 * Releases a reference on the paged fragment @frag.
2274 static inline void __skb_frag_unref(skb_frag_t *frag)
2276 put_page(skb_frag_page(frag));
2280 * skb_frag_unref - release a reference on a paged fragment of an skb.
2282 * @f: the fragment offset
2284 * Releases a reference on the @f'th paged fragment of @skb.
2286 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2288 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2292 * skb_frag_address - gets the address of the data contained in a paged fragment
2293 * @frag: the paged fragment buffer
2295 * Returns the address of the data within @frag. The page must already
2298 static inline void *skb_frag_address(const skb_frag_t *frag)
2300 return page_address(skb_frag_page(frag)) + frag->page_offset;
2304 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2305 * @frag: the paged fragment buffer
2307 * Returns the address of the data within @frag. Checks that the page
2308 * is mapped and returns %NULL otherwise.
2310 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2312 void *ptr = page_address(skb_frag_page(frag));
2316 return ptr + frag->page_offset;
2320 * __skb_frag_set_page - sets the page contained in a paged fragment
2321 * @frag: the paged fragment
2322 * @page: the page to set
2324 * Sets the fragment @frag to contain @page.
2326 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2328 frag->page.p = page;
2332 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2334 * @f: the fragment offset
2335 * @page: the page to set
2337 * Sets the @f'th fragment of @skb to contain @page.
2339 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2342 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2345 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2348 * skb_frag_dma_map - maps a paged fragment via the DMA API
2349 * @dev: the device to map the fragment to
2350 * @frag: the paged fragment to map
2351 * @offset: the offset within the fragment (starting at the
2352 * fragment's own offset)
2353 * @size: the number of bytes to map
2354 * @dir: the direction of the mapping (%PCI_DMA_*)
2356 * Maps the page associated with @frag to @device.
2358 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2359 const skb_frag_t *frag,
2360 size_t offset, size_t size,
2361 enum dma_data_direction dir)
2363 return dma_map_page(dev, skb_frag_page(frag),
2364 frag->page_offset + offset, size, dir);
2367 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2370 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2374 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2377 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2382 * skb_clone_writable - is the header of a clone writable
2383 * @skb: buffer to check
2384 * @len: length up to which to write
2386 * Returns true if modifying the header part of the cloned buffer
2387 * does not requires the data to be copied.
2389 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2391 return !skb_header_cloned(skb) &&
2392 skb_headroom(skb) + len <= skb->hdr_len;
2395 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2400 if (headroom > skb_headroom(skb))
2401 delta = headroom - skb_headroom(skb);
2403 if (delta || cloned)
2404 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2410 * skb_cow - copy header of skb when it is required
2411 * @skb: buffer to cow
2412 * @headroom: needed headroom
2414 * If the skb passed lacks sufficient headroom or its data part
2415 * is shared, data is reallocated. If reallocation fails, an error
2416 * is returned and original skb is not changed.
2418 * The result is skb with writable area skb->head...skb->tail
2419 * and at least @headroom of space at head.
2421 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2423 return __skb_cow(skb, headroom, skb_cloned(skb));
2427 * skb_cow_head - skb_cow but only making the head writable
2428 * @skb: buffer to cow
2429 * @headroom: needed headroom
2431 * This function is identical to skb_cow except that we replace the
2432 * skb_cloned check by skb_header_cloned. It should be used when
2433 * you only need to push on some header and do not need to modify
2436 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2438 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2442 * skb_padto - pad an skbuff up to a minimal size
2443 * @skb: buffer to pad
2444 * @len: minimal length
2446 * Pads up a buffer to ensure the trailing bytes exist and are
2447 * blanked. If the buffer already contains sufficient data it
2448 * is untouched. Otherwise it is extended. Returns zero on
2449 * success. The skb is freed on error.
2452 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2454 unsigned int size = skb->len;
2455 if (likely(size >= len))
2457 return skb_pad(skb, len - size);
2460 static inline int skb_add_data(struct sk_buff *skb,
2461 char __user *from, int copy)
2463 const int off = skb->len;
2465 if (skb->ip_summed == CHECKSUM_NONE) {
2467 __wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
2470 skb->csum = csum_block_add(skb->csum, csum, off);
2473 } else if (!copy_from_user(skb_put(skb, copy), from, copy))
2476 __skb_trim(skb, off);
2480 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2481 const struct page *page, int off)
2484 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2486 return page == skb_frag_page(frag) &&
2487 off == frag->page_offset + skb_frag_size(frag);
2492 static inline int __skb_linearize(struct sk_buff *skb)
2494 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2498 * skb_linearize - convert paged skb to linear one
2499 * @skb: buffer to linarize
2501 * If there is no free memory -ENOMEM is returned, otherwise zero
2502 * is returned and the old skb data released.
2504 static inline int skb_linearize(struct sk_buff *skb)
2506 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2510 * skb_has_shared_frag - can any frag be overwritten
2511 * @skb: buffer to test
2513 * Return true if the skb has at least one frag that might be modified
2514 * by an external entity (as in vmsplice()/sendfile())
2516 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2518 return skb_is_nonlinear(skb) &&
2519 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2523 * skb_linearize_cow - make sure skb is linear and writable
2524 * @skb: buffer to process
2526 * If there is no free memory -ENOMEM is returned, otherwise zero
2527 * is returned and the old skb data released.
2529 static inline int skb_linearize_cow(struct sk_buff *skb)
2531 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2532 __skb_linearize(skb) : 0;
2536 * skb_postpull_rcsum - update checksum for received skb after pull
2537 * @skb: buffer to update
2538 * @start: start of data before pull
2539 * @len: length of data pulled
2541 * After doing a pull on a received packet, you need to call this to
2542 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2543 * CHECKSUM_NONE so that it can be recomputed from scratch.
2546 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2547 const void *start, unsigned int len)
2549 if (skb->ip_summed == CHECKSUM_COMPLETE)
2550 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2553 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2556 * pskb_trim_rcsum - trim received skb and update checksum
2557 * @skb: buffer to trim
2560 * This is exactly the same as pskb_trim except that it ensures the
2561 * checksum of received packets are still valid after the operation.
2564 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2566 if (likely(len >= skb->len))
2568 if (skb->ip_summed == CHECKSUM_COMPLETE)
2569 skb->ip_summed = CHECKSUM_NONE;
2570 return __pskb_trim(skb, len);
2573 #define skb_queue_walk(queue, skb) \
2574 for (skb = (queue)->next; \
2575 skb != (struct sk_buff *)(queue); \
2578 #define skb_queue_walk_safe(queue, skb, tmp) \
2579 for (skb = (queue)->next, tmp = skb->next; \
2580 skb != (struct sk_buff *)(queue); \
2581 skb = tmp, tmp = skb->next)
2583 #define skb_queue_walk_from(queue, skb) \
2584 for (; skb != (struct sk_buff *)(queue); \
2587 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2588 for (tmp = skb->next; \
2589 skb != (struct sk_buff *)(queue); \
2590 skb = tmp, tmp = skb->next)
2592 #define skb_queue_reverse_walk(queue, skb) \
2593 for (skb = (queue)->prev; \
2594 skb != (struct sk_buff *)(queue); \
2597 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2598 for (skb = (queue)->prev, tmp = skb->prev; \
2599 skb != (struct sk_buff *)(queue); \
2600 skb = tmp, tmp = skb->prev)
2602 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2603 for (tmp = skb->prev; \
2604 skb != (struct sk_buff *)(queue); \
2605 skb = tmp, tmp = skb->prev)
2607 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2609 return skb_shinfo(skb)->frag_list != NULL;
2612 static inline void skb_frag_list_init(struct sk_buff *skb)
2614 skb_shinfo(skb)->frag_list = NULL;
2617 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
2619 frag->next = skb_shinfo(skb)->frag_list;
2620 skb_shinfo(skb)->frag_list = frag;
2623 #define skb_walk_frags(skb, iter) \
2624 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2626 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2627 int *peeked, int *off, int *err);
2628 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2630 unsigned int datagram_poll(struct file *file, struct socket *sock,
2631 struct poll_table_struct *wait);
2632 int skb_copy_datagram_iovec(const struct sk_buff *from, int offset,
2633 struct iovec *to, int size);
2634 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, int hlen,
2636 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
2637 const struct iovec *from, int from_offset,
2639 int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *frm,
2640 int offset, size_t count);
2641 int skb_copy_datagram_const_iovec(const struct sk_buff *from, int offset,
2642 const struct iovec *to, int to_offset,
2644 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2645 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2646 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2647 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2648 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2649 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2650 int len, __wsum csum);
2651 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
2652 struct pipe_inode_info *pipe, unsigned int len,
2653 unsigned int flags);
2654 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2655 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2656 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2658 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2659 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2660 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2661 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2662 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2663 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
2665 struct skb_checksum_ops {
2666 __wsum (*update)(const void *mem, int len, __wsum wsum);
2667 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2670 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2671 __wsum csum, const struct skb_checksum_ops *ops);
2672 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2675 static inline void *__skb_header_pointer(const struct sk_buff *skb, int offset,
2676 int len, void *data, int hlen, void *buffer)
2678 if (hlen - offset >= len)
2679 return data + offset;
2682 skb_copy_bits(skb, offset, buffer, len) < 0)
2688 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
2689 int len, void *buffer)
2691 return __skb_header_pointer(skb, offset, len, skb->data,
2692 skb_headlen(skb), buffer);
2696 * skb_needs_linearize - check if we need to linearize a given skb
2697 * depending on the given device features.
2698 * @skb: socket buffer to check
2699 * @features: net device features
2701 * Returns true if either:
2702 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2703 * 2. skb is fragmented and the device does not support SG.
2705 static inline bool skb_needs_linearize(struct sk_buff *skb,
2706 netdev_features_t features)
2708 return skb_is_nonlinear(skb) &&
2709 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2710 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2713 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2715 const unsigned int len)
2717 memcpy(to, skb->data, len);
2720 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2721 const int offset, void *to,
2722 const unsigned int len)
2724 memcpy(to, skb->data + offset, len);
2727 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2729 const unsigned int len)
2731 memcpy(skb->data, from, len);
2734 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2737 const unsigned int len)
2739 memcpy(skb->data + offset, from, len);
2742 void skb_init(void);
2744 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2750 * skb_get_timestamp - get timestamp from a skb
2751 * @skb: skb to get stamp from
2752 * @stamp: pointer to struct timeval to store stamp in
2754 * Timestamps are stored in the skb as offsets to a base timestamp.
2755 * This function converts the offset back to a struct timeval and stores
2758 static inline void skb_get_timestamp(const struct sk_buff *skb,
2759 struct timeval *stamp)
2761 *stamp = ktime_to_timeval(skb->tstamp);
2764 static inline void skb_get_timestampns(const struct sk_buff *skb,
2765 struct timespec *stamp)
2767 *stamp = ktime_to_timespec(skb->tstamp);
2770 static inline void __net_timestamp(struct sk_buff *skb)
2772 skb->tstamp = ktime_get_real();
2775 static inline ktime_t net_timedelta(ktime_t t)
2777 return ktime_sub(ktime_get_real(), t);
2780 static inline ktime_t net_invalid_timestamp(void)
2782 return ktime_set(0, 0);
2785 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
2787 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2789 void skb_clone_tx_timestamp(struct sk_buff *skb);
2790 bool skb_defer_rx_timestamp(struct sk_buff *skb);
2792 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2794 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2798 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2803 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2806 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2808 * PHY drivers may accept clones of transmitted packets for
2809 * timestamping via their phy_driver.txtstamp method. These drivers
2810 * must call this function to return the skb back to the stack, with
2811 * or without a timestamp.
2813 * @skb: clone of the the original outgoing packet
2814 * @hwtstamps: hardware time stamps, may be NULL if not available
2817 void skb_complete_tx_timestamp(struct sk_buff *skb,
2818 struct skb_shared_hwtstamps *hwtstamps);
2820 void __skb_tstamp_tx(struct sk_buff *orig_skb,
2821 struct skb_shared_hwtstamps *hwtstamps,
2822 struct sock *sk, int tstype);
2825 * skb_tstamp_tx - queue clone of skb with send time stamps
2826 * @orig_skb: the original outgoing packet
2827 * @hwtstamps: hardware time stamps, may be NULL if not available
2829 * If the skb has a socket associated, then this function clones the
2830 * skb (thus sharing the actual data and optional structures), stores
2831 * the optional hardware time stamping information (if non NULL) or
2832 * generates a software time stamp (otherwise), then queues the clone
2833 * to the error queue of the socket. Errors are silently ignored.
2835 void skb_tstamp_tx(struct sk_buff *orig_skb,
2836 struct skb_shared_hwtstamps *hwtstamps);
2838 static inline void sw_tx_timestamp(struct sk_buff *skb)
2840 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2841 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
2842 skb_tstamp_tx(skb, NULL);
2846 * skb_tx_timestamp() - Driver hook for transmit timestamping
2848 * Ethernet MAC Drivers should call this function in their hard_xmit()
2849 * function immediately before giving the sk_buff to the MAC hardware.
2851 * Specifically, one should make absolutely sure that this function is
2852 * called before TX completion of this packet can trigger. Otherwise
2853 * the packet could potentially already be freed.
2855 * @skb: A socket buffer.
2857 static inline void skb_tx_timestamp(struct sk_buff *skb)
2859 skb_clone_tx_timestamp(skb);
2860 sw_tx_timestamp(skb);
2864 * skb_complete_wifi_ack - deliver skb with wifi status
2866 * @skb: the original outgoing packet
2867 * @acked: ack status
2870 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2872 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2873 __sum16 __skb_checksum_complete(struct sk_buff *skb);
2875 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2877 return ((skb->ip_summed & CHECKSUM_UNNECESSARY) || skb->csum_valid);
2881 * skb_checksum_complete - Calculate checksum of an entire packet
2882 * @skb: packet to process
2884 * This function calculates the checksum over the entire packet plus
2885 * the value of skb->csum. The latter can be used to supply the
2886 * checksum of a pseudo header as used by TCP/UDP. It returns the
2889 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2890 * this function can be used to verify that checksum on received
2891 * packets. In that case the function should return zero if the
2892 * checksum is correct. In particular, this function will return zero
2893 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2894 * hardware has already verified the correctness of the checksum.
2896 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
2898 return skb_csum_unnecessary(skb) ?
2899 0 : __skb_checksum_complete(skb);
2902 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
2904 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2905 if (skb->csum_level == 0)
2906 skb->ip_summed = CHECKSUM_NONE;
2912 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
2914 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2915 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
2917 } else if (skb->ip_summed == CHECKSUM_NONE) {
2918 skb->ip_summed = CHECKSUM_UNNECESSARY;
2919 skb->csum_level = 0;
2923 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
2925 /* Mark current checksum as bad (typically called from GRO
2926 * path). In the case that ip_summed is CHECKSUM_NONE
2927 * this must be the first checksum encountered in the packet.
2928 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
2929 * checksum after the last one validated. For UDP, a zero
2930 * checksum can not be marked as bad.
2933 if (skb->ip_summed == CHECKSUM_NONE ||
2934 skb->ip_summed == CHECKSUM_UNNECESSARY)
2938 /* Check if we need to perform checksum complete validation.
2940 * Returns true if checksum complete is needed, false otherwise
2941 * (either checksum is unnecessary or zero checksum is allowed).
2943 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
2947 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
2948 skb->csum_valid = 1;
2949 __skb_decr_checksum_unnecessary(skb);
2956 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
2959 #define CHECKSUM_BREAK 76
2961 /* Validate (init) checksum based on checksum complete.
2964 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
2965 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
2966 * checksum is stored in skb->csum for use in __skb_checksum_complete
2967 * non-zero: value of invalid checksum
2970 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
2974 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2975 if (!csum_fold(csum_add(psum, skb->csum))) {
2976 skb->csum_valid = 1;
2979 } else if (skb->csum_bad) {
2980 /* ip_summed == CHECKSUM_NONE in this case */
2986 if (complete || skb->len <= CHECKSUM_BREAK) {
2989 csum = __skb_checksum_complete(skb);
2990 skb->csum_valid = !csum;
2997 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3002 /* Perform checksum validate (init). Note that this is a macro since we only
3003 * want to calculate the pseudo header which is an input function if necessary.
3004 * First we try to validate without any computation (checksum unnecessary) and
3005 * then calculate based on checksum complete calling the function to compute
3009 * 0: checksum is validated or try to in skb_checksum_complete
3010 * non-zero: value of invalid checksum
3012 #define __skb_checksum_validate(skb, proto, complete, \
3013 zero_okay, check, compute_pseudo) \
3015 __sum16 __ret = 0; \
3016 skb->csum_valid = 0; \
3017 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3018 __ret = __skb_checksum_validate_complete(skb, \
3019 complete, compute_pseudo(skb, proto)); \
3023 #define skb_checksum_init(skb, proto, compute_pseudo) \
3024 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3026 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3027 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3029 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3030 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3032 #define skb_checksum_validate_zero_check(skb, proto, check, \
3034 __skb_checksum_validate_(skb, proto, true, true, check, compute_pseudo)
3036 #define skb_checksum_simple_validate(skb) \
3037 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3039 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3041 return (skb->ip_summed == CHECKSUM_NONE &&
3042 skb->csum_valid && !skb->csum_bad);
3045 static inline void __skb_checksum_convert(struct sk_buff *skb,
3046 __sum16 check, __wsum pseudo)
3048 skb->csum = ~pseudo;
3049 skb->ip_summed = CHECKSUM_COMPLETE;
3052 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3054 if (__skb_checksum_convert_check(skb)) \
3055 __skb_checksum_convert(skb, check, \
3056 compute_pseudo(skb, proto)); \
3059 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3060 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3061 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3063 if (nfct && atomic_dec_and_test(&nfct->use))
3064 nf_conntrack_destroy(nfct);
3066 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3069 atomic_inc(&nfct->use);
3072 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3073 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3075 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3078 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3081 atomic_inc(&nf_bridge->use);
3083 #endif /* CONFIG_BRIDGE_NETFILTER */
3084 static inline void nf_reset(struct sk_buff *skb)
3086 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3087 nf_conntrack_put(skb->nfct);
3090 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3091 nf_bridge_put(skb->nf_bridge);
3092 skb->nf_bridge = NULL;
3096 static inline void nf_reset_trace(struct sk_buff *skb)
3098 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3103 /* Note: This doesn't put any conntrack and bridge info in dst. */
3104 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3107 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3108 dst->nfct = src->nfct;
3109 nf_conntrack_get(src->nfct);
3111 dst->nfctinfo = src->nfctinfo;
3113 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3114 dst->nf_bridge = src->nf_bridge;
3115 nf_bridge_get(src->nf_bridge);
3117 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3119 dst->nf_trace = src->nf_trace;
3123 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3125 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3126 nf_conntrack_put(dst->nfct);
3128 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3129 nf_bridge_put(dst->nf_bridge);
3131 __nf_copy(dst, src, true);
3134 #ifdef CONFIG_NETWORK_SECMARK
3135 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3137 to->secmark = from->secmark;
3140 static inline void skb_init_secmark(struct sk_buff *skb)
3145 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3148 static inline void skb_init_secmark(struct sk_buff *skb)
3152 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3154 return !skb->destructor &&
3155 #if IS_ENABLED(CONFIG_XFRM)
3158 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3161 !skb->_skb_refdst &&
3162 !skb_has_frag_list(skb);
3165 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3167 skb->queue_mapping = queue_mapping;
3170 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3172 return skb->queue_mapping;
3175 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3177 to->queue_mapping = from->queue_mapping;
3180 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3182 skb->queue_mapping = rx_queue + 1;
3185 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3187 return skb->queue_mapping - 1;
3190 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3192 return skb->queue_mapping != 0;
3195 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
3196 unsigned int num_tx_queues);
3198 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3207 /* Keeps track of mac header offset relative to skb->head.
3208 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3209 * For non-tunnel skb it points to skb_mac_header() and for
3210 * tunnel skb it points to outer mac header.
3211 * Keeps track of level of encapsulation of network headers.
3218 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3220 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3222 return (skb_mac_header(inner_skb) - inner_skb->head) -
3223 SKB_GSO_CB(inner_skb)->mac_offset;
3226 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3228 int new_headroom, headroom;
3231 headroom = skb_headroom(skb);
3232 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3236 new_headroom = skb_headroom(skb);
3237 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3241 /* Compute the checksum for a gso segment. First compute the checksum value
3242 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3243 * then add in skb->csum (checksum from csum_start to end of packet).
3244 * skb->csum and csum_start are then updated to reflect the checksum of the
3245 * resultant packet starting from the transport header-- the resultant checksum
3246 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3249 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3251 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3252 skb_transport_offset(skb);
3255 csum = csum_fold(csum_partial(skb_transport_header(skb),
3258 SKB_GSO_CB(skb)->csum_start -= plen;
3263 static inline bool skb_is_gso(const struct sk_buff *skb)
3265 return skb_shinfo(skb)->gso_size;
3268 /* Note: Should be called only if skb_is_gso(skb) is true */
3269 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3271 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3274 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3276 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3278 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3279 * wanted then gso_type will be set. */
3280 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3282 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3283 unlikely(shinfo->gso_type == 0)) {
3284 __skb_warn_lro_forwarding(skb);
3290 static inline void skb_forward_csum(struct sk_buff *skb)
3292 /* Unfortunately we don't support this one. Any brave souls? */
3293 if (skb->ip_summed == CHECKSUM_COMPLETE)
3294 skb->ip_summed = CHECKSUM_NONE;
3298 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3299 * @skb: skb to check
3301 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3302 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3303 * use this helper, to document places where we make this assertion.
3305 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3308 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3312 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3314 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3316 u32 skb_get_poff(const struct sk_buff *skb);
3317 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
3318 const struct flow_keys *keys, int hlen);
3321 * skb_head_is_locked - Determine if the skb->head is locked down
3322 * @skb: skb to check
3324 * The head on skbs build around a head frag can be removed if they are
3325 * not cloned. This function returns true if the skb head is locked down
3326 * due to either being allocated via kmalloc, or by being a clone with
3327 * multiple references to the head.
3329 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3331 return !skb->head_frag || skb_cloned(skb);
3335 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3339 * skb_gso_network_seglen is used to determine the real size of the
3340 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3342 * The MAC/L2 header is not accounted for.
3344 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3346 unsigned int hdr_len = skb_transport_header(skb) -
3347 skb_network_header(skb);
3348 return hdr_len + skb_gso_transport_seglen(skb);
3350 #endif /* __KERNEL__ */
3351 #endif /* _LINUX_SKBUFF_H */