2 * Definitions for the 'struct sk_buff' memory handlers.
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
17 #include <linux/kernel.h>
18 #include <linux/kmemcheck.h>
19 #include <linux/compiler.h>
20 #include <linux/time.h>
21 #include <linux/bug.h>
22 #include <linux/cache.h>
23 #include <linux/rbtree.h>
24 #include <linux/socket.h>
26 #include <linux/atomic.h>
27 #include <asm/types.h>
28 #include <linux/spinlock.h>
29 #include <linux/net.h>
30 #include <linux/textsearch.h>
31 #include <net/checksum.h>
32 #include <linux/rcupdate.h>
33 #include <linux/hrtimer.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/netdev_features.h>
36 #include <linux/sched.h>
37 #include <net/flow_dissector.h>
38 #include <linux/splice.h>
39 #include <linux/in6.h>
42 /* A. Checksumming of received packets by device.
46 * Device failed to checksum this packet e.g. due to lack of capabilities.
47 * The packet contains full (though not verified) checksum in packet but
48 * not in skb->csum. Thus, skb->csum is undefined in this case.
50 * CHECKSUM_UNNECESSARY:
52 * The hardware you're dealing with doesn't calculate the full checksum
53 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
54 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
55 * if their checksums are okay. skb->csum is still undefined in this case
56 * though. It is a bad option, but, unfortunately, nowadays most vendors do
57 * this. Apparently with the secret goal to sell you new devices, when you
58 * will add new protocol to your host, f.e. IPv6 8)
60 * CHECKSUM_UNNECESSARY is applicable to following protocols:
62 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
63 * zero UDP checksum for either IPv4 or IPv6, the networking stack
64 * may perform further validation in this case.
65 * GRE: only if the checksum is present in the header.
66 * SCTP: indicates the CRC in SCTP header has been validated.
68 * skb->csum_level indicates the number of consecutive checksums found in
69 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
70 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
71 * and a device is able to verify the checksums for UDP (possibly zero),
72 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
73 * two. If the device were only able to verify the UDP checksum and not
74 * GRE, either because it doesn't support GRE checksum of because GRE
75 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
76 * not considered in this case).
80 * This is the most generic way. The device supplied checksum of the _whole_
81 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
82 * hardware doesn't need to parse L3/L4 headers to implement this.
84 * Note: Even if device supports only some protocols, but is able to produce
85 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
89 * A checksum is set up to be offloaded to a device as described in the
90 * output description for CHECKSUM_PARTIAL. This may occur on a packet
91 * received directly from another Linux OS, e.g., a virtualized Linux kernel
92 * on the same host, or it may be set in the input path in GRO or remote
93 * checksum offload. For the purposes of checksum verification, the checksum
94 * referred to by skb->csum_start + skb->csum_offset and any preceding
95 * checksums in the packet are considered verified. Any checksums in the
96 * packet that are after the checksum being offloaded are not considered to
99 * B. Checksumming on output.
103 * The skb was already checksummed by the protocol, or a checksum is not
108 * The device is required to checksum the packet as seen by hard_start_xmit()
109 * from skb->csum_start up to the end, and to record/write the checksum at
110 * offset skb->csum_start + skb->csum_offset.
112 * The device must show its capabilities in dev->features, set up at device
113 * setup time, e.g. netdev_features.h:
115 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything.
116 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over
117 * IPv4. Sigh. Vendors like this way for an unknown reason.
118 * Though, see comment above about CHECKSUM_UNNECESSARY. 8)
119 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead.
120 * NETIF_F_... - Well, you get the picture.
122 * CHECKSUM_UNNECESSARY:
124 * Normally, the device will do per protocol specific checksumming. Protocol
125 * implementations that do not want the NIC to perform the checksum
126 * calculation should use this flag in their outgoing skbs.
128 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC
129 * offload. Correspondingly, the FCoE protocol driver
130 * stack should use CHECKSUM_UNNECESSARY.
132 * Any questions? No questions, good. --ANK
135 /* Don't change this without changing skb_csum_unnecessary! */
136 #define CHECKSUM_NONE 0
137 #define CHECKSUM_UNNECESSARY 1
138 #define CHECKSUM_COMPLETE 2
139 #define CHECKSUM_PARTIAL 3
141 /* Maximum value in skb->csum_level */
142 #define SKB_MAX_CSUM_LEVEL 3
144 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
145 #define SKB_WITH_OVERHEAD(X) \
146 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
147 #define SKB_MAX_ORDER(X, ORDER) \
148 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
149 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
150 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
152 /* return minimum truesize of one skb containing X bytes of data */
153 #define SKB_TRUESIZE(X) ((X) + \
154 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
155 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
159 struct pipe_inode_info;
163 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
164 struct nf_conntrack {
169 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
170 struct nf_bridge_info {
173 BRNF_PROTO_UNCHANGED,
181 struct net_device *physindev;
183 /* always valid & non-NULL from FORWARD on, for physdev match */
184 struct net_device *physoutdev;
186 /* prerouting: detect dnat in orig/reply direction */
188 struct in6_addr ipv6_daddr;
190 /* after prerouting + nat detected: store original source
191 * mac since neigh resolution overwrites it, only used while
192 * skb is out in neigh layer.
194 char neigh_header[8];
199 struct sk_buff_head {
200 /* These two members must be first. */
201 struct sk_buff *next;
202 struct sk_buff *prev;
206 raw_spinlock_t raw_lock;
211 /* To allow 64K frame to be packed as single skb without frag_list we
212 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
213 * buffers which do not start on a page boundary.
215 * Since GRO uses frags we allocate at least 16 regardless of page
218 #if (65536/PAGE_SIZE + 1) < 16
219 #define MAX_SKB_FRAGS 16UL
221 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
223 extern int sysctl_max_skb_frags;
225 typedef struct skb_frag_struct skb_frag_t;
227 struct skb_frag_struct {
231 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
240 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
245 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
250 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
255 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
260 #define HAVE_HW_TIME_STAMP
263 * struct skb_shared_hwtstamps - hardware time stamps
264 * @hwtstamp: hardware time stamp transformed into duration
265 * since arbitrary point in time
267 * Software time stamps generated by ktime_get_real() are stored in
270 * hwtstamps can only be compared against other hwtstamps from
273 * This structure is attached to packets as part of the
274 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
276 struct skb_shared_hwtstamps {
280 /* Definitions for tx_flags in struct skb_shared_info */
282 /* generate hardware time stamp */
283 SKBTX_HW_TSTAMP = 1 << 0,
285 /* generate software time stamp when queueing packet to NIC */
286 SKBTX_SW_TSTAMP = 1 << 1,
288 /* device driver is going to provide hardware time stamp */
289 SKBTX_IN_PROGRESS = 1 << 2,
291 /* device driver supports TX zero-copy buffers */
292 SKBTX_DEV_ZEROCOPY = 1 << 3,
294 /* generate wifi status information (where possible) */
295 SKBTX_WIFI_STATUS = 1 << 4,
297 /* This indicates at least one fragment might be overwritten
298 * (as in vmsplice(), sendfile() ...)
299 * If we need to compute a TX checksum, we'll need to copy
300 * all frags to avoid possible bad checksum
302 SKBTX_SHARED_FRAG = 1 << 5,
304 /* generate software time stamp when entering packet scheduling */
305 SKBTX_SCHED_TSTAMP = 1 << 6,
307 /* generate software timestamp on peer data acknowledgment */
308 SKBTX_ACK_TSTAMP = 1 << 7,
311 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
312 SKBTX_SCHED_TSTAMP | \
314 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
317 * The callback notifies userspace to release buffers when skb DMA is done in
318 * lower device, the skb last reference should be 0 when calling this.
319 * The zerocopy_success argument is true if zero copy transmit occurred,
320 * false on data copy or out of memory error caused by data copy attempt.
321 * The ctx field is used to track device context.
322 * The desc field is used to track userspace buffer index.
325 void (*callback)(struct ubuf_info *, bool zerocopy_success);
330 /* This data is invariant across clones and lives at
331 * the end of the header data, ie. at skb->end.
333 struct skb_shared_info {
334 unsigned char nr_frags;
336 unsigned short gso_size;
337 /* Warning: this field is not always filled in (UFO)! */
338 unsigned short gso_segs;
339 unsigned short gso_type;
340 struct sk_buff *frag_list;
341 struct skb_shared_hwtstamps hwtstamps;
346 * Warning : all fields before dataref are cleared in __alloc_skb()
350 /* Intermediate layers must ensure that destructor_arg
351 * remains valid until skb destructor */
352 void * destructor_arg;
354 /* must be last field, see pskb_expand_head() */
355 skb_frag_t frags[MAX_SKB_FRAGS];
358 /* We divide dataref into two halves. The higher 16 bits hold references
359 * to the payload part of skb->data. The lower 16 bits hold references to
360 * the entire skb->data. A clone of a headerless skb holds the length of
361 * the header in skb->hdr_len.
363 * All users must obey the rule that the skb->data reference count must be
364 * greater than or equal to the payload reference count.
366 * Holding a reference to the payload part means that the user does not
367 * care about modifications to the header part of skb->data.
369 #define SKB_DATAREF_SHIFT 16
370 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
374 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
375 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
376 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
380 SKB_GSO_TCPV4 = 1 << 0,
381 SKB_GSO_UDP = 1 << 1,
383 /* This indicates the skb is from an untrusted source. */
384 SKB_GSO_DODGY = 1 << 2,
386 /* This indicates the tcp segment has CWR set. */
387 SKB_GSO_TCP_ECN = 1 << 3,
389 SKB_GSO_TCPV6 = 1 << 4,
391 SKB_GSO_FCOE = 1 << 5,
393 SKB_GSO_GRE = 1 << 6,
395 SKB_GSO_GRE_CSUM = 1 << 7,
397 SKB_GSO_IPIP = 1 << 8,
399 SKB_GSO_SIT = 1 << 9,
401 SKB_GSO_UDP_TUNNEL = 1 << 10,
403 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
405 SKB_GSO_TUNNEL_REMCSUM = 1 << 12,
408 #if BITS_PER_LONG > 32
409 #define NET_SKBUFF_DATA_USES_OFFSET 1
412 #ifdef NET_SKBUFF_DATA_USES_OFFSET
413 typedef unsigned int sk_buff_data_t;
415 typedef unsigned char *sk_buff_data_t;
419 * struct skb_mstamp - multi resolution time stamps
420 * @stamp_us: timestamp in us resolution
421 * @stamp_jiffies: timestamp in jiffies
434 * skb_mstamp_get - get current timestamp
435 * @cl: place to store timestamps
437 static inline void skb_mstamp_get(struct skb_mstamp *cl)
439 u64 val = local_clock();
441 do_div(val, NSEC_PER_USEC);
442 cl->stamp_us = (u32)val;
443 cl->stamp_jiffies = (u32)jiffies;
447 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
448 * @t1: pointer to newest sample
449 * @t0: pointer to oldest sample
451 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
452 const struct skb_mstamp *t0)
454 s32 delta_us = t1->stamp_us - t0->stamp_us;
455 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
457 /* If delta_us is negative, this might be because interval is too big,
458 * or local_clock() drift is too big : fallback using jiffies.
461 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
463 delta_us = jiffies_to_usecs(delta_jiffies);
468 static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
469 const struct skb_mstamp *t0)
471 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
474 diff = t1->stamp_us - t0->stamp_us;
479 * struct sk_buff - socket buffer
480 * @next: Next buffer in list
481 * @prev: Previous buffer in list
482 * @tstamp: Time we arrived/left
483 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
484 * @sk: Socket we are owned by
485 * @dev: Device we arrived on/are leaving by
486 * @cb: Control buffer. Free for use by every layer. Put private vars here
487 * @_skb_refdst: destination entry (with norefcount bit)
488 * @sp: the security path, used for xfrm
489 * @len: Length of actual data
490 * @data_len: Data length
491 * @mac_len: Length of link layer header
492 * @hdr_len: writable header length of cloned skb
493 * @csum: Checksum (must include start/offset pair)
494 * @csum_start: Offset from skb->head where checksumming should start
495 * @csum_offset: Offset from csum_start where checksum should be stored
496 * @priority: Packet queueing priority
497 * @ignore_df: allow local fragmentation
498 * @cloned: Head may be cloned (check refcnt to be sure)
499 * @ip_summed: Driver fed us an IP checksum
500 * @nohdr: Payload reference only, must not modify header
501 * @nfctinfo: Relationship of this skb to the connection
502 * @pkt_type: Packet class
503 * @fclone: skbuff clone status
504 * @ipvs_property: skbuff is owned by ipvs
505 * @peeked: this packet has been seen already, so stats have been
506 * done for it, don't do them again
507 * @nf_trace: netfilter packet trace flag
508 * @protocol: Packet protocol from driver
509 * @destructor: Destruct function
510 * @nfct: Associated connection, if any
511 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
512 * @skb_iif: ifindex of device we arrived on
513 * @tc_index: Traffic control index
514 * @tc_verd: traffic control verdict
515 * @hash: the packet hash
516 * @queue_mapping: Queue mapping for multiqueue devices
517 * @xmit_more: More SKBs are pending for this queue
518 * @ndisc_nodetype: router type (from link layer)
519 * @ooo_okay: allow the mapping of a socket to a queue to be changed
520 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
522 * @sw_hash: indicates hash was computed in software stack
523 * @wifi_acked_valid: wifi_acked was set
524 * @wifi_acked: whether frame was acked on wifi or not
525 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
526 * @napi_id: id of the NAPI struct this skb came from
527 * @secmark: security marking
528 * @offload_fwd_mark: fwding offload mark
529 * @mark: Generic packet mark
530 * @vlan_proto: vlan encapsulation protocol
531 * @vlan_tci: vlan tag control information
532 * @inner_protocol: Protocol (encapsulation)
533 * @inner_transport_header: Inner transport layer header (encapsulation)
534 * @inner_network_header: Network layer header (encapsulation)
535 * @inner_mac_header: Link layer header (encapsulation)
536 * @transport_header: Transport layer header
537 * @network_header: Network layer header
538 * @mac_header: Link layer header
539 * @tail: Tail pointer
541 * @head: Head of buffer
542 * @data: Data head pointer
543 * @truesize: Buffer size
544 * @users: User count - see {datagram,tcp}.c
550 /* These two members must be first. */
551 struct sk_buff *next;
552 struct sk_buff *prev;
556 struct skb_mstamp skb_mstamp;
559 struct rb_node rbnode; /* used in netem & tcp stack */
562 struct net_device *dev;
565 * This is the control buffer. It is free to use for every
566 * layer. Please put your private variables there. If you
567 * want to keep them across layers you have to do a skb_clone()
568 * first. This is owned by whoever has the skb queued ATM.
570 char cb[48] __aligned(8);
572 unsigned long _skb_refdst;
573 void (*destructor)(struct sk_buff *skb);
577 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
578 struct nf_conntrack *nfct;
580 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
581 struct nf_bridge_info *nf_bridge;
588 /* Following fields are _not_ copied in __copy_skb_header()
589 * Note that queue_mapping is here mostly to fill a hole.
591 kmemcheck_bitfield_begin(flags1);
600 kmemcheck_bitfield_end(flags1);
602 /* fields enclosed in headers_start/headers_end are copied
603 * using a single memcpy() in __copy_skb_header()
606 __u32 headers_start[0];
609 /* if you move pkt_type around you also must adapt those constants */
610 #ifdef __BIG_ENDIAN_BITFIELD
611 #define PKT_TYPE_MAX (7 << 5)
613 #define PKT_TYPE_MAX 7
615 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
617 __u8 __pkt_type_offset[0];
628 __u8 wifi_acked_valid:1;
632 /* Indicates the inner headers are valid in the skbuff. */
633 __u8 encapsulation:1;
634 __u8 encap_hdr_csum:1;
636 __u8 csum_complete_sw:1;
640 #ifdef CONFIG_IPV6_NDISC_NODETYPE
641 __u8 ndisc_nodetype:2;
643 __u8 ipvs_property:1;
644 __u8 inner_protocol_type:1;
645 __u8 remcsum_offload:1;
646 /* 3 or 5 bit hole */
648 #ifdef CONFIG_NET_SCHED
649 __u16 tc_index; /* traffic control index */
650 #ifdef CONFIG_NET_CLS_ACT
651 __u16 tc_verd; /* traffic control verdict */
667 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
669 unsigned int napi_id;
670 unsigned int sender_cpu;
674 #ifdef CONFIG_NETWORK_SECMARK
677 #ifdef CONFIG_NET_SWITCHDEV
678 __u32 offload_fwd_mark;
684 __u32 reserved_tailroom;
688 __be16 inner_protocol;
692 __u16 inner_transport_header;
693 __u16 inner_network_header;
694 __u16 inner_mac_header;
697 __u16 transport_header;
698 __u16 network_header;
702 __u32 headers_end[0];
705 /* These elements must be at the end, see alloc_skb() for details. */
710 unsigned int truesize;
716 * Handling routines are only of interest to the kernel
718 #include <linux/slab.h>
721 #define SKB_ALLOC_FCLONE 0x01
722 #define SKB_ALLOC_RX 0x02
723 #define SKB_ALLOC_NAPI 0x04
725 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
726 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
728 return unlikely(skb->pfmemalloc);
732 * skb might have a dst pointer attached, refcounted or not.
733 * _skb_refdst low order bit is set if refcount was _not_ taken
735 #define SKB_DST_NOREF 1UL
736 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
739 * skb_dst - returns skb dst_entry
742 * Returns skb dst_entry, regardless of reference taken or not.
744 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
746 /* If refdst was not refcounted, check we still are in a
747 * rcu_read_lock section
749 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
750 !rcu_read_lock_held() &&
751 !rcu_read_lock_bh_held());
752 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
756 * skb_dst_set - sets skb dst
760 * Sets skb dst, assuming a reference was taken on dst and should
761 * be released by skb_dst_drop()
763 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
765 skb->_skb_refdst = (unsigned long)dst;
769 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
773 * Sets skb dst, assuming a reference was not taken on dst.
774 * If dst entry is cached, we do not take reference and dst_release
775 * will be avoided by refdst_drop. If dst entry is not cached, we take
776 * reference, so that last dst_release can destroy the dst immediately.
778 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
780 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
781 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
785 * skb_dst_is_noref - Test if skb dst isn't refcounted
788 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
790 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
793 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
795 return (struct rtable *)skb_dst(skb);
798 void kfree_skb(struct sk_buff *skb);
799 void kfree_skb_list(struct sk_buff *segs);
800 void skb_tx_error(struct sk_buff *skb);
801 void consume_skb(struct sk_buff *skb);
802 void __kfree_skb(struct sk_buff *skb);
803 extern struct kmem_cache *skbuff_head_cache;
805 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
806 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
807 bool *fragstolen, int *delta_truesize);
809 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
811 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
812 struct sk_buff *build_skb(void *data, unsigned int frag_size);
813 static inline struct sk_buff *alloc_skb(unsigned int size,
816 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
819 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
820 unsigned long data_len,
825 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
826 struct sk_buff_fclones {
835 * skb_fclone_busy - check if fclone is busy
838 * Returns true is skb is a fast clone, and its clone is not freed.
839 * Some drivers call skb_orphan() in their ndo_start_xmit(),
840 * so we also check that this didnt happen.
842 static inline bool skb_fclone_busy(const struct sock *sk,
843 const struct sk_buff *skb)
845 const struct sk_buff_fclones *fclones;
847 fclones = container_of(skb, struct sk_buff_fclones, skb1);
849 return skb->fclone == SKB_FCLONE_ORIG &&
850 atomic_read(&fclones->fclone_ref) > 1 &&
851 fclones->skb2.sk == sk;
854 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
857 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
860 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
861 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
863 return __alloc_skb_head(priority, -1);
866 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
867 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
868 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
869 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
870 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
871 gfp_t gfp_mask, bool fclone);
872 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
875 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
878 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
879 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
880 unsigned int headroom);
881 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
882 int newtailroom, gfp_t priority);
883 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
884 int offset, int len);
885 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
887 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
888 int skb_pad(struct sk_buff *skb, int pad);
889 #define dev_kfree_skb(a) consume_skb(a)
891 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
892 int getfrag(void *from, char *to, int offset,
893 int len, int odd, struct sk_buff *skb),
894 void *from, int length);
896 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
897 int offset, size_t size);
899 struct skb_seq_state {
903 __u32 stepped_offset;
904 struct sk_buff *root_skb;
905 struct sk_buff *cur_skb;
909 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
910 unsigned int to, struct skb_seq_state *st);
911 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
912 struct skb_seq_state *st);
913 void skb_abort_seq_read(struct skb_seq_state *st);
915 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
916 unsigned int to, struct ts_config *config);
919 * Packet hash types specify the type of hash in skb_set_hash.
921 * Hash types refer to the protocol layer addresses which are used to
922 * construct a packet's hash. The hashes are used to differentiate or identify
923 * flows of the protocol layer for the hash type. Hash types are either
924 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
926 * Properties of hashes:
928 * 1) Two packets in different flows have different hash values
929 * 2) Two packets in the same flow should have the same hash value
931 * A hash at a higher layer is considered to be more specific. A driver should
932 * set the most specific hash possible.
934 * A driver cannot indicate a more specific hash than the layer at which a hash
935 * was computed. For instance an L3 hash cannot be set as an L4 hash.
937 * A driver may indicate a hash level which is less specific than the
938 * actual layer the hash was computed on. For instance, a hash computed
939 * at L4 may be considered an L3 hash. This should only be done if the
940 * driver can't unambiguously determine that the HW computed the hash at
941 * the higher layer. Note that the "should" in the second property above
944 enum pkt_hash_types {
945 PKT_HASH_TYPE_NONE, /* Undefined type */
946 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
947 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
948 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
951 static inline void skb_clear_hash(struct sk_buff *skb)
958 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
965 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
967 skb->l4_hash = is_l4;
968 skb->sw_hash = is_sw;
973 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
975 /* Used by drivers to set hash from HW */
976 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
980 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
982 __skb_set_hash(skb, hash, true, is_l4);
985 void __skb_get_hash(struct sk_buff *skb);
986 u32 __skb_get_hash_symmetric(struct sk_buff *skb);
987 u32 skb_get_poff(const struct sk_buff *skb);
988 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
989 const struct flow_keys *keys, int hlen);
990 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
991 void *data, int hlen_proto);
993 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
994 int thoff, u8 ip_proto)
996 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
999 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1000 const struct flow_dissector_key *key,
1001 unsigned int key_count);
1003 bool __skb_flow_dissect(const struct sk_buff *skb,
1004 struct flow_dissector *flow_dissector,
1005 void *target_container,
1006 void *data, __be16 proto, int nhoff, int hlen,
1007 unsigned int flags);
1009 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1010 struct flow_dissector *flow_dissector,
1011 void *target_container, unsigned int flags)
1013 return __skb_flow_dissect(skb, flow_dissector, target_container,
1014 NULL, 0, 0, 0, flags);
1017 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1018 struct flow_keys *flow,
1021 memset(flow, 0, sizeof(*flow));
1022 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1023 NULL, 0, 0, 0, flags);
1026 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1027 void *data, __be16 proto,
1028 int nhoff, int hlen,
1031 memset(flow, 0, sizeof(*flow));
1032 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1033 data, proto, nhoff, hlen, flags);
1036 static inline __u32 skb_get_hash(struct sk_buff *skb)
1038 if (!skb->l4_hash && !skb->sw_hash)
1039 __skb_get_hash(skb);
1044 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1046 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1048 if (!skb->l4_hash && !skb->sw_hash) {
1049 struct flow_keys keys;
1050 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1052 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1058 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1060 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1062 if (!skb->l4_hash && !skb->sw_hash) {
1063 struct flow_keys keys;
1064 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
1066 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1072 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1074 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1079 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1081 to->hash = from->hash;
1082 to->sw_hash = from->sw_hash;
1083 to->l4_hash = from->l4_hash;
1086 static inline void skb_sender_cpu_clear(struct sk_buff *skb)
1089 skb->sender_cpu = 0;
1093 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1094 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1096 return skb->head + skb->end;
1099 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1104 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1109 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1111 return skb->end - skb->head;
1116 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1118 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1120 return &skb_shinfo(skb)->hwtstamps;
1124 * skb_queue_empty - check if a queue is empty
1127 * Returns true if the queue is empty, false otherwise.
1129 static inline int skb_queue_empty(const struct sk_buff_head *list)
1131 return list->next == (const struct sk_buff *) list;
1135 * skb_queue_is_last - check if skb is the last entry in the queue
1139 * Returns true if @skb is the last buffer on the list.
1141 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1142 const struct sk_buff *skb)
1144 return skb->next == (const struct sk_buff *) list;
1148 * skb_queue_is_first - check if skb is the first entry in the queue
1152 * Returns true if @skb is the first buffer on the list.
1154 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1155 const struct sk_buff *skb)
1157 return skb->prev == (const struct sk_buff *) list;
1161 * skb_queue_next - return the next packet in the queue
1163 * @skb: current buffer
1165 * Return the next packet in @list after @skb. It is only valid to
1166 * call this if skb_queue_is_last() evaluates to false.
1168 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1169 const struct sk_buff *skb)
1171 /* This BUG_ON may seem severe, but if we just return then we
1172 * are going to dereference garbage.
1174 BUG_ON(skb_queue_is_last(list, skb));
1179 * skb_queue_prev - return the prev packet in the queue
1181 * @skb: current buffer
1183 * Return the prev packet in @list before @skb. It is only valid to
1184 * call this if skb_queue_is_first() evaluates to false.
1186 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1187 const struct sk_buff *skb)
1189 /* This BUG_ON may seem severe, but if we just return then we
1190 * are going to dereference garbage.
1192 BUG_ON(skb_queue_is_first(list, skb));
1197 * skb_get - reference buffer
1198 * @skb: buffer to reference
1200 * Makes another reference to a socket buffer and returns a pointer
1203 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1205 atomic_inc(&skb->users);
1210 * If users == 1, we are the only owner and are can avoid redundant
1215 * skb_cloned - is the buffer a clone
1216 * @skb: buffer to check
1218 * Returns true if the buffer was generated with skb_clone() and is
1219 * one of multiple shared copies of the buffer. Cloned buffers are
1220 * shared data so must not be written to under normal circumstances.
1222 static inline int skb_cloned(const struct sk_buff *skb)
1224 return skb->cloned &&
1225 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1228 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1230 might_sleep_if(gfpflags_allow_blocking(pri));
1232 if (skb_cloned(skb))
1233 return pskb_expand_head(skb, 0, 0, pri);
1239 * skb_header_cloned - is the header a clone
1240 * @skb: buffer to check
1242 * Returns true if modifying the header part of the buffer requires
1243 * the data to be copied.
1245 static inline int skb_header_cloned(const struct sk_buff *skb)
1252 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1253 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1254 return dataref != 1;
1258 * skb_header_release - release reference to header
1259 * @skb: buffer to operate on
1261 * Drop a reference to the header part of the buffer. This is done
1262 * by acquiring a payload reference. You must not read from the header
1263 * part of skb->data after this.
1264 * Note : Check if you can use __skb_header_release() instead.
1266 static inline void skb_header_release(struct sk_buff *skb)
1270 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1274 * __skb_header_release - release reference to header
1275 * @skb: buffer to operate on
1277 * Variant of skb_header_release() assuming skb is private to caller.
1278 * We can avoid one atomic operation.
1280 static inline void __skb_header_release(struct sk_buff *skb)
1283 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1288 * skb_shared - is the buffer shared
1289 * @skb: buffer to check
1291 * Returns true if more than one person has a reference to this
1294 static inline int skb_shared(const struct sk_buff *skb)
1296 return atomic_read(&skb->users) != 1;
1300 * skb_share_check - check if buffer is shared and if so clone it
1301 * @skb: buffer to check
1302 * @pri: priority for memory allocation
1304 * If the buffer is shared the buffer is cloned and the old copy
1305 * drops a reference. A new clone with a single reference is returned.
1306 * If the buffer is not shared the original buffer is returned. When
1307 * being called from interrupt status or with spinlocks held pri must
1310 * NULL is returned on a memory allocation failure.
1312 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1314 might_sleep_if(gfpflags_allow_blocking(pri));
1315 if (skb_shared(skb)) {
1316 struct sk_buff *nskb = skb_clone(skb, pri);
1328 * Copy shared buffers into a new sk_buff. We effectively do COW on
1329 * packets to handle cases where we have a local reader and forward
1330 * and a couple of other messy ones. The normal one is tcpdumping
1331 * a packet thats being forwarded.
1335 * skb_unshare - make a copy of a shared buffer
1336 * @skb: buffer to check
1337 * @pri: priority for memory allocation
1339 * If the socket buffer is a clone then this function creates a new
1340 * copy of the data, drops a reference count on the old copy and returns
1341 * the new copy with the reference count at 1. If the buffer is not a clone
1342 * the original buffer is returned. When called with a spinlock held or
1343 * from interrupt state @pri must be %GFP_ATOMIC
1345 * %NULL is returned on a memory allocation failure.
1347 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1350 might_sleep_if(gfpflags_allow_blocking(pri));
1351 if (skb_cloned(skb)) {
1352 struct sk_buff *nskb = skb_copy(skb, pri);
1354 /* Free our shared copy */
1365 * skb_peek - peek at the head of an &sk_buff_head
1366 * @list_: list to peek at
1368 * Peek an &sk_buff. Unlike most other operations you _MUST_
1369 * be careful with this one. A peek leaves the buffer on the
1370 * list and someone else may run off with it. You must hold
1371 * the appropriate locks or have a private queue to do this.
1373 * Returns %NULL for an empty list or a pointer to the head element.
1374 * The reference count is not incremented and the reference is therefore
1375 * volatile. Use with caution.
1377 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1379 struct sk_buff *skb = list_->next;
1381 if (skb == (struct sk_buff *)list_)
1387 * skb_peek_next - peek skb following the given one from a queue
1388 * @skb: skb to start from
1389 * @list_: list to peek at
1391 * Returns %NULL when the end of the list is met or a pointer to the
1392 * next element. The reference count is not incremented and the
1393 * reference is therefore volatile. Use with caution.
1395 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1396 const struct sk_buff_head *list_)
1398 struct sk_buff *next = skb->next;
1400 if (next == (struct sk_buff *)list_)
1406 * skb_peek_tail - peek at the tail of an &sk_buff_head
1407 * @list_: list to peek at
1409 * Peek an &sk_buff. Unlike most other operations you _MUST_
1410 * be careful with this one. A peek leaves the buffer on the
1411 * list and someone else may run off with it. You must hold
1412 * the appropriate locks or have a private queue to do this.
1414 * Returns %NULL for an empty list or a pointer to the tail element.
1415 * The reference count is not incremented and the reference is therefore
1416 * volatile. Use with caution.
1418 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1420 struct sk_buff *skb = list_->prev;
1422 if (skb == (struct sk_buff *)list_)
1429 * skb_queue_len - get queue length
1430 * @list_: list to measure
1432 * Return the length of an &sk_buff queue.
1434 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1440 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1441 * @list: queue to initialize
1443 * This initializes only the list and queue length aspects of
1444 * an sk_buff_head object. This allows to initialize the list
1445 * aspects of an sk_buff_head without reinitializing things like
1446 * the spinlock. It can also be used for on-stack sk_buff_head
1447 * objects where the spinlock is known to not be used.
1449 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1451 list->prev = list->next = (struct sk_buff *)list;
1456 * This function creates a split out lock class for each invocation;
1457 * this is needed for now since a whole lot of users of the skb-queue
1458 * infrastructure in drivers have different locking usage (in hardirq)
1459 * than the networking core (in softirq only). In the long run either the
1460 * network layer or drivers should need annotation to consolidate the
1461 * main types of usage into 3 classes.
1463 static inline void skb_queue_head_init(struct sk_buff_head *list)
1465 spin_lock_init(&list->lock);
1466 __skb_queue_head_init(list);
1469 static inline void skb_queue_head_init_raw(struct sk_buff_head *list)
1471 raw_spin_lock_init(&list->raw_lock);
1472 __skb_queue_head_init(list);
1475 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1476 struct lock_class_key *class)
1478 skb_queue_head_init(list);
1479 lockdep_set_class(&list->lock, class);
1483 * Insert an sk_buff on a list.
1485 * The "__skb_xxxx()" functions are the non-atomic ones that
1486 * can only be called with interrupts disabled.
1488 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1489 struct sk_buff_head *list);
1490 static inline void __skb_insert(struct sk_buff *newsk,
1491 struct sk_buff *prev, struct sk_buff *next,
1492 struct sk_buff_head *list)
1496 next->prev = prev->next = newsk;
1500 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1501 struct sk_buff *prev,
1502 struct sk_buff *next)
1504 struct sk_buff *first = list->next;
1505 struct sk_buff *last = list->prev;
1515 * skb_queue_splice - join two skb lists, this is designed for stacks
1516 * @list: the new list to add
1517 * @head: the place to add it in the first list
1519 static inline void skb_queue_splice(const struct sk_buff_head *list,
1520 struct sk_buff_head *head)
1522 if (!skb_queue_empty(list)) {
1523 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1524 head->qlen += list->qlen;
1529 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1530 * @list: the new list to add
1531 * @head: the place to add it in the first list
1533 * The list at @list is reinitialised
1535 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1536 struct sk_buff_head *head)
1538 if (!skb_queue_empty(list)) {
1539 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1540 head->qlen += list->qlen;
1541 __skb_queue_head_init(list);
1546 * skb_queue_splice_tail - join two skb lists, each list being a queue
1547 * @list: the new list to add
1548 * @head: the place to add it in the first list
1550 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1551 struct sk_buff_head *head)
1553 if (!skb_queue_empty(list)) {
1554 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1555 head->qlen += list->qlen;
1560 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1561 * @list: the new list to add
1562 * @head: the place to add it in the first list
1564 * Each of the lists is a queue.
1565 * The list at @list is reinitialised
1567 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1568 struct sk_buff_head *head)
1570 if (!skb_queue_empty(list)) {
1571 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1572 head->qlen += list->qlen;
1573 __skb_queue_head_init(list);
1578 * __skb_queue_after - queue a buffer at the list head
1579 * @list: list to use
1580 * @prev: place after this buffer
1581 * @newsk: buffer to queue
1583 * Queue a buffer int the middle of a list. This function takes no locks
1584 * and you must therefore hold required locks before calling it.
1586 * A buffer cannot be placed on two lists at the same time.
1588 static inline void __skb_queue_after(struct sk_buff_head *list,
1589 struct sk_buff *prev,
1590 struct sk_buff *newsk)
1592 __skb_insert(newsk, prev, prev->next, list);
1595 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1596 struct sk_buff_head *list);
1598 static inline void __skb_queue_before(struct sk_buff_head *list,
1599 struct sk_buff *next,
1600 struct sk_buff *newsk)
1602 __skb_insert(newsk, next->prev, next, list);
1606 * __skb_queue_head - queue a buffer at the list head
1607 * @list: list to use
1608 * @newsk: buffer to queue
1610 * Queue a buffer at the start of a list. This function takes no locks
1611 * and you must therefore hold required locks before calling it.
1613 * A buffer cannot be placed on two lists at the same time.
1615 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1616 static inline void __skb_queue_head(struct sk_buff_head *list,
1617 struct sk_buff *newsk)
1619 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1623 * __skb_queue_tail - queue a buffer at the list tail
1624 * @list: list to use
1625 * @newsk: buffer to queue
1627 * Queue a buffer at the end of a list. This function takes no locks
1628 * and you must therefore hold required locks before calling it.
1630 * A buffer cannot be placed on two lists at the same time.
1632 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1633 static inline void __skb_queue_tail(struct sk_buff_head *list,
1634 struct sk_buff *newsk)
1636 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1640 * remove sk_buff from list. _Must_ be called atomically, and with
1643 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1644 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1646 struct sk_buff *next, *prev;
1651 skb->next = skb->prev = NULL;
1657 * __skb_dequeue - remove from the head of the queue
1658 * @list: list to dequeue from
1660 * Remove the head of the list. This function does not take any locks
1661 * so must be used with appropriate locks held only. The head item is
1662 * returned or %NULL if the list is empty.
1664 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1665 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1667 struct sk_buff *skb = skb_peek(list);
1669 __skb_unlink(skb, list);
1674 * __skb_dequeue_tail - remove from the tail of the queue
1675 * @list: list to dequeue from
1677 * Remove the tail of the list. This function does not take any locks
1678 * so must be used with appropriate locks held only. The tail item is
1679 * returned or %NULL if the list is empty.
1681 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1682 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1684 struct sk_buff *skb = skb_peek_tail(list);
1686 __skb_unlink(skb, list);
1691 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1693 return skb->data_len;
1696 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1698 return skb->len - skb->data_len;
1701 static inline int skb_pagelen(const struct sk_buff *skb)
1705 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1706 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1707 return len + skb_headlen(skb);
1711 * __skb_fill_page_desc - initialise a paged fragment in an skb
1712 * @skb: buffer containing fragment to be initialised
1713 * @i: paged fragment index to initialise
1714 * @page: the page to use for this fragment
1715 * @off: the offset to the data with @page
1716 * @size: the length of the data
1718 * Initialises the @i'th fragment of @skb to point to &size bytes at
1719 * offset @off within @page.
1721 * Does not take any additional reference on the fragment.
1723 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1724 struct page *page, int off, int size)
1726 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1729 * Propagate page pfmemalloc to the skb if we can. The problem is
1730 * that not all callers have unique ownership of the page but rely
1731 * on page_is_pfmemalloc doing the right thing(tm).
1733 frag->page.p = page;
1734 frag->page_offset = off;
1735 skb_frag_size_set(frag, size);
1737 page = compound_head(page);
1738 if (page_is_pfmemalloc(page))
1739 skb->pfmemalloc = true;
1743 * skb_fill_page_desc - initialise a paged fragment in an skb
1744 * @skb: buffer containing fragment to be initialised
1745 * @i: paged fragment index to initialise
1746 * @page: the page to use for this fragment
1747 * @off: the offset to the data with @page
1748 * @size: the length of the data
1750 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1751 * @skb to point to @size bytes at offset @off within @page. In
1752 * addition updates @skb such that @i is the last fragment.
1754 * Does not take any additional reference on the fragment.
1756 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1757 struct page *page, int off, int size)
1759 __skb_fill_page_desc(skb, i, page, off, size);
1760 skb_shinfo(skb)->nr_frags = i + 1;
1763 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1764 int size, unsigned int truesize);
1766 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1767 unsigned int truesize);
1769 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1770 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1771 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1773 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1774 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1776 return skb->head + skb->tail;
1779 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1781 skb->tail = skb->data - skb->head;
1784 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1786 skb_reset_tail_pointer(skb);
1787 skb->tail += offset;
1790 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1791 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1796 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1798 skb->tail = skb->data;
1801 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1803 skb->tail = skb->data + offset;
1806 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1809 * Add data to an sk_buff
1811 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1812 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1813 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1815 unsigned char *tmp = skb_tail_pointer(skb);
1816 SKB_LINEAR_ASSERT(skb);
1822 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1823 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1830 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1831 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1834 BUG_ON(skb->len < skb->data_len);
1835 return skb->data += len;
1838 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1840 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1843 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1845 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1847 if (len > skb_headlen(skb) &&
1848 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1851 return skb->data += len;
1854 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1856 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1859 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1861 if (likely(len <= skb_headlen(skb)))
1863 if (unlikely(len > skb->len))
1865 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1869 * skb_headroom - bytes at buffer head
1870 * @skb: buffer to check
1872 * Return the number of bytes of free space at the head of an &sk_buff.
1874 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1876 return skb->data - skb->head;
1880 * skb_tailroom - bytes at buffer end
1881 * @skb: buffer to check
1883 * Return the number of bytes of free space at the tail of an sk_buff
1885 static inline int skb_tailroom(const struct sk_buff *skb)
1887 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1891 * skb_availroom - bytes at buffer end
1892 * @skb: buffer to check
1894 * Return the number of bytes of free space at the tail of an sk_buff
1895 * allocated by sk_stream_alloc()
1897 static inline int skb_availroom(const struct sk_buff *skb)
1899 if (skb_is_nonlinear(skb))
1902 return skb->end - skb->tail - skb->reserved_tailroom;
1906 * skb_reserve - adjust headroom
1907 * @skb: buffer to alter
1908 * @len: bytes to move
1910 * Increase the headroom of an empty &sk_buff by reducing the tail
1911 * room. This is only allowed for an empty buffer.
1913 static inline void skb_reserve(struct sk_buff *skb, int len)
1920 * skb_tailroom_reserve - adjust reserved_tailroom
1921 * @skb: buffer to alter
1922 * @mtu: maximum amount of headlen permitted
1923 * @needed_tailroom: minimum amount of reserved_tailroom
1925 * Set reserved_tailroom so that headlen can be as large as possible but
1926 * not larger than mtu and tailroom cannot be smaller than
1928 * The required headroom should already have been reserved before using
1931 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
1932 unsigned int needed_tailroom)
1934 SKB_LINEAR_ASSERT(skb);
1935 if (mtu < skb_tailroom(skb) - needed_tailroom)
1936 /* use at most mtu */
1937 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
1939 /* use up to all available space */
1940 skb->reserved_tailroom = needed_tailroom;
1943 #define ENCAP_TYPE_ETHER 0
1944 #define ENCAP_TYPE_IPPROTO 1
1946 static inline void skb_set_inner_protocol(struct sk_buff *skb,
1949 skb->inner_protocol = protocol;
1950 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
1953 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
1956 skb->inner_ipproto = ipproto;
1957 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
1960 static inline void skb_reset_inner_headers(struct sk_buff *skb)
1962 skb->inner_mac_header = skb->mac_header;
1963 skb->inner_network_header = skb->network_header;
1964 skb->inner_transport_header = skb->transport_header;
1967 static inline void skb_reset_mac_len(struct sk_buff *skb)
1969 skb->mac_len = skb->network_header - skb->mac_header;
1972 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1975 return skb->head + skb->inner_transport_header;
1978 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1980 skb->inner_transport_header = skb->data - skb->head;
1983 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1986 skb_reset_inner_transport_header(skb);
1987 skb->inner_transport_header += offset;
1990 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1992 return skb->head + skb->inner_network_header;
1995 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1997 skb->inner_network_header = skb->data - skb->head;
2000 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2003 skb_reset_inner_network_header(skb);
2004 skb->inner_network_header += offset;
2007 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2009 return skb->head + skb->inner_mac_header;
2012 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2014 skb->inner_mac_header = skb->data - skb->head;
2017 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2020 skb_reset_inner_mac_header(skb);
2021 skb->inner_mac_header += offset;
2023 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2025 return skb->transport_header != (typeof(skb->transport_header))~0U;
2028 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2030 return skb->head + skb->transport_header;
2033 static inline void skb_reset_transport_header(struct sk_buff *skb)
2035 skb->transport_header = skb->data - skb->head;
2038 static inline void skb_set_transport_header(struct sk_buff *skb,
2041 skb_reset_transport_header(skb);
2042 skb->transport_header += offset;
2045 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2047 return skb->head + skb->network_header;
2050 static inline void skb_reset_network_header(struct sk_buff *skb)
2052 skb->network_header = skb->data - skb->head;
2055 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2057 skb_reset_network_header(skb);
2058 skb->network_header += offset;
2061 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2063 return skb->head + skb->mac_header;
2066 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2068 return skb->mac_header != (typeof(skb->mac_header))~0U;
2071 static inline void skb_reset_mac_header(struct sk_buff *skb)
2073 skb->mac_header = skb->data - skb->head;
2076 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2078 skb_reset_mac_header(skb);
2079 skb->mac_header += offset;
2082 static inline void skb_pop_mac_header(struct sk_buff *skb)
2084 skb->mac_header = skb->network_header;
2087 static inline void skb_probe_transport_header(struct sk_buff *skb,
2088 const int offset_hint)
2090 struct flow_keys keys;
2092 if (skb_transport_header_was_set(skb))
2094 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2095 skb_set_transport_header(skb, keys.control.thoff);
2097 skb_set_transport_header(skb, offset_hint);
2100 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2102 if (skb_mac_header_was_set(skb)) {
2103 const unsigned char *old_mac = skb_mac_header(skb);
2105 skb_set_mac_header(skb, -skb->mac_len);
2106 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2110 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2112 return skb->csum_start - skb_headroom(skb);
2115 static inline int skb_transport_offset(const struct sk_buff *skb)
2117 return skb_transport_header(skb) - skb->data;
2120 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2122 return skb->transport_header - skb->network_header;
2125 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2127 return skb->inner_transport_header - skb->inner_network_header;
2130 static inline int skb_network_offset(const struct sk_buff *skb)
2132 return skb_network_header(skb) - skb->data;
2135 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2137 return skb_inner_network_header(skb) - skb->data;
2140 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2142 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2146 * CPUs often take a performance hit when accessing unaligned memory
2147 * locations. The actual performance hit varies, it can be small if the
2148 * hardware handles it or large if we have to take an exception and fix it
2151 * Since an ethernet header is 14 bytes network drivers often end up with
2152 * the IP header at an unaligned offset. The IP header can be aligned by
2153 * shifting the start of the packet by 2 bytes. Drivers should do this
2156 * skb_reserve(skb, NET_IP_ALIGN);
2158 * The downside to this alignment of the IP header is that the DMA is now
2159 * unaligned. On some architectures the cost of an unaligned DMA is high
2160 * and this cost outweighs the gains made by aligning the IP header.
2162 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2165 #ifndef NET_IP_ALIGN
2166 #define NET_IP_ALIGN 2
2170 * The networking layer reserves some headroom in skb data (via
2171 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2172 * the header has to grow. In the default case, if the header has to grow
2173 * 32 bytes or less we avoid the reallocation.
2175 * Unfortunately this headroom changes the DMA alignment of the resulting
2176 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2177 * on some architectures. An architecture can override this value,
2178 * perhaps setting it to a cacheline in size (since that will maintain
2179 * cacheline alignment of the DMA). It must be a power of 2.
2181 * Various parts of the networking layer expect at least 32 bytes of
2182 * headroom, you should not reduce this.
2184 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2185 * to reduce average number of cache lines per packet.
2186 * get_rps_cpus() for example only access one 64 bytes aligned block :
2187 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2190 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2193 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2195 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2197 if (unlikely(skb_is_nonlinear(skb))) {
2202 skb_set_tail_pointer(skb, len);
2205 void skb_trim(struct sk_buff *skb, unsigned int len);
2207 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2210 return ___pskb_trim(skb, len);
2211 __skb_trim(skb, len);
2215 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2217 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2221 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2222 * @skb: buffer to alter
2225 * This is identical to pskb_trim except that the caller knows that
2226 * the skb is not cloned so we should never get an error due to out-
2229 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2231 int err = pskb_trim(skb, len);
2236 * skb_orphan - orphan a buffer
2237 * @skb: buffer to orphan
2239 * If a buffer currently has an owner then we call the owner's
2240 * destructor function and make the @skb unowned. The buffer continues
2241 * to exist but is no longer charged to its former owner.
2243 static inline void skb_orphan(struct sk_buff *skb)
2245 if (skb->destructor) {
2246 skb->destructor(skb);
2247 skb->destructor = NULL;
2255 * skb_orphan_frags - orphan the frags contained in a buffer
2256 * @skb: buffer to orphan frags from
2257 * @gfp_mask: allocation mask for replacement pages
2259 * For each frag in the SKB which needs a destructor (i.e. has an
2260 * owner) create a copy of that frag and release the original
2261 * page by calling the destructor.
2263 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2265 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2267 return skb_copy_ubufs(skb, gfp_mask);
2271 * __skb_queue_purge - empty a list
2272 * @list: list to empty
2274 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2275 * the list and one reference dropped. This function does not take the
2276 * list lock and the caller must hold the relevant locks to use it.
2278 void skb_queue_purge(struct sk_buff_head *list);
2279 static inline void __skb_queue_purge(struct sk_buff_head *list)
2281 struct sk_buff *skb;
2282 while ((skb = __skb_dequeue(list)) != NULL)
2286 void *netdev_alloc_frag(unsigned int fragsz);
2288 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2292 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2293 * @dev: network device to receive on
2294 * @length: length to allocate
2296 * Allocate a new &sk_buff and assign it a usage count of one. The
2297 * buffer has unspecified headroom built in. Users should allocate
2298 * the headroom they think they need without accounting for the
2299 * built in space. The built in space is used for optimisations.
2301 * %NULL is returned if there is no free memory. Although this function
2302 * allocates memory it can be called from an interrupt.
2304 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2305 unsigned int length)
2307 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2310 /* legacy helper around __netdev_alloc_skb() */
2311 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2314 return __netdev_alloc_skb(NULL, length, gfp_mask);
2317 /* legacy helper around netdev_alloc_skb() */
2318 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2320 return netdev_alloc_skb(NULL, length);
2324 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2325 unsigned int length, gfp_t gfp)
2327 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2329 if (NET_IP_ALIGN && skb)
2330 skb_reserve(skb, NET_IP_ALIGN);
2334 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2335 unsigned int length)
2337 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2340 static inline void skb_free_frag(void *addr)
2342 __free_page_frag(addr);
2345 void *napi_alloc_frag(unsigned int fragsz);
2346 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2347 unsigned int length, gfp_t gfp_mask);
2348 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2349 unsigned int length)
2351 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2355 * __dev_alloc_pages - allocate page for network Rx
2356 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2357 * @order: size of the allocation
2359 * Allocate a new page.
2361 * %NULL is returned if there is no free memory.
2363 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2366 /* This piece of code contains several assumptions.
2367 * 1. This is for device Rx, therefor a cold page is preferred.
2368 * 2. The expectation is the user wants a compound page.
2369 * 3. If requesting a order 0 page it will not be compound
2370 * due to the check to see if order has a value in prep_new_page
2371 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2372 * code in gfp_to_alloc_flags that should be enforcing this.
2374 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2376 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2379 static inline struct page *dev_alloc_pages(unsigned int order)
2381 return __dev_alloc_pages(GFP_ATOMIC, order);
2385 * __dev_alloc_page - allocate a page for network Rx
2386 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2388 * Allocate a new page.
2390 * %NULL is returned if there is no free memory.
2392 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2394 return __dev_alloc_pages(gfp_mask, 0);
2397 static inline struct page *dev_alloc_page(void)
2399 return __dev_alloc_page(GFP_ATOMIC);
2403 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2404 * @page: The page that was allocated from skb_alloc_page
2405 * @skb: The skb that may need pfmemalloc set
2407 static inline void skb_propagate_pfmemalloc(struct page *page,
2408 struct sk_buff *skb)
2410 if (page_is_pfmemalloc(page))
2411 skb->pfmemalloc = true;
2415 * skb_frag_page - retrieve the page referred to by a paged fragment
2416 * @frag: the paged fragment
2418 * Returns the &struct page associated with @frag.
2420 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2422 return frag->page.p;
2426 * __skb_frag_ref - take an addition reference on a paged fragment.
2427 * @frag: the paged fragment
2429 * Takes an additional reference on the paged fragment @frag.
2431 static inline void __skb_frag_ref(skb_frag_t *frag)
2433 get_page(skb_frag_page(frag));
2437 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2439 * @f: the fragment offset.
2441 * Takes an additional reference on the @f'th paged fragment of @skb.
2443 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2445 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2449 * __skb_frag_unref - release a reference on a paged fragment.
2450 * @frag: the paged fragment
2452 * Releases a reference on the paged fragment @frag.
2454 static inline void __skb_frag_unref(skb_frag_t *frag)
2456 put_page(skb_frag_page(frag));
2460 * skb_frag_unref - release a reference on a paged fragment of an skb.
2462 * @f: the fragment offset
2464 * Releases a reference on the @f'th paged fragment of @skb.
2466 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2468 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2472 * skb_frag_address - gets the address of the data contained in a paged fragment
2473 * @frag: the paged fragment buffer
2475 * Returns the address of the data within @frag. The page must already
2478 static inline void *skb_frag_address(const skb_frag_t *frag)
2480 return page_address(skb_frag_page(frag)) + frag->page_offset;
2484 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2485 * @frag: the paged fragment buffer
2487 * Returns the address of the data within @frag. Checks that the page
2488 * is mapped and returns %NULL otherwise.
2490 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2492 void *ptr = page_address(skb_frag_page(frag));
2496 return ptr + frag->page_offset;
2500 * __skb_frag_set_page - sets the page contained in a paged fragment
2501 * @frag: the paged fragment
2502 * @page: the page to set
2504 * Sets the fragment @frag to contain @page.
2506 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2508 frag->page.p = page;
2512 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2514 * @f: the fragment offset
2515 * @page: the page to set
2517 * Sets the @f'th fragment of @skb to contain @page.
2519 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2522 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2525 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2528 * skb_frag_dma_map - maps a paged fragment via the DMA API
2529 * @dev: the device to map the fragment to
2530 * @frag: the paged fragment to map
2531 * @offset: the offset within the fragment (starting at the
2532 * fragment's own offset)
2533 * @size: the number of bytes to map
2534 * @dir: the direction of the mapping (%PCI_DMA_*)
2536 * Maps the page associated with @frag to @device.
2538 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2539 const skb_frag_t *frag,
2540 size_t offset, size_t size,
2541 enum dma_data_direction dir)
2543 return dma_map_page(dev, skb_frag_page(frag),
2544 frag->page_offset + offset, size, dir);
2547 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2550 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2554 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2557 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2562 * skb_clone_writable - is the header of a clone writable
2563 * @skb: buffer to check
2564 * @len: length up to which to write
2566 * Returns true if modifying the header part of the cloned buffer
2567 * does not requires the data to be copied.
2569 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2571 return !skb_header_cloned(skb) &&
2572 skb_headroom(skb) + len <= skb->hdr_len;
2575 static inline int skb_try_make_writable(struct sk_buff *skb,
2576 unsigned int write_len)
2578 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2579 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2582 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2587 if (headroom > skb_headroom(skb))
2588 delta = headroom - skb_headroom(skb);
2590 if (delta || cloned)
2591 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2597 * skb_cow - copy header of skb when it is required
2598 * @skb: buffer to cow
2599 * @headroom: needed headroom
2601 * If the skb passed lacks sufficient headroom or its data part
2602 * is shared, data is reallocated. If reallocation fails, an error
2603 * is returned and original skb is not changed.
2605 * The result is skb with writable area skb->head...skb->tail
2606 * and at least @headroom of space at head.
2608 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2610 return __skb_cow(skb, headroom, skb_cloned(skb));
2614 * skb_cow_head - skb_cow but only making the head writable
2615 * @skb: buffer to cow
2616 * @headroom: needed headroom
2618 * This function is identical to skb_cow except that we replace the
2619 * skb_cloned check by skb_header_cloned. It should be used when
2620 * you only need to push on some header and do not need to modify
2623 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2625 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2629 * skb_padto - pad an skbuff up to a minimal size
2630 * @skb: buffer to pad
2631 * @len: minimal length
2633 * Pads up a buffer to ensure the trailing bytes exist and are
2634 * blanked. If the buffer already contains sufficient data it
2635 * is untouched. Otherwise it is extended. Returns zero on
2636 * success. The skb is freed on error.
2638 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2640 unsigned int size = skb->len;
2641 if (likely(size >= len))
2643 return skb_pad(skb, len - size);
2647 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2648 * @skb: buffer to pad
2649 * @len: minimal length
2651 * Pads up a buffer to ensure the trailing bytes exist and are
2652 * blanked. If the buffer already contains sufficient data it
2653 * is untouched. Otherwise it is extended. Returns zero on
2654 * success. The skb is freed on error.
2656 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2658 unsigned int size = skb->len;
2660 if (unlikely(size < len)) {
2662 if (skb_pad(skb, len))
2664 __skb_put(skb, len);
2669 static inline int skb_add_data(struct sk_buff *skb,
2670 struct iov_iter *from, int copy)
2672 const int off = skb->len;
2674 if (skb->ip_summed == CHECKSUM_NONE) {
2676 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2677 &csum, from) == copy) {
2678 skb->csum = csum_block_add(skb->csum, csum, off);
2681 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
2684 __skb_trim(skb, off);
2688 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2689 const struct page *page, int off)
2692 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2694 return page == skb_frag_page(frag) &&
2695 off == frag->page_offset + skb_frag_size(frag);
2700 static inline int __skb_linearize(struct sk_buff *skb)
2702 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2706 * skb_linearize - convert paged skb to linear one
2707 * @skb: buffer to linarize
2709 * If there is no free memory -ENOMEM is returned, otherwise zero
2710 * is returned and the old skb data released.
2712 static inline int skb_linearize(struct sk_buff *skb)
2714 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2718 * skb_has_shared_frag - can any frag be overwritten
2719 * @skb: buffer to test
2721 * Return true if the skb has at least one frag that might be modified
2722 * by an external entity (as in vmsplice()/sendfile())
2724 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2726 return skb_is_nonlinear(skb) &&
2727 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2731 * skb_linearize_cow - make sure skb is linear and writable
2732 * @skb: buffer to process
2734 * If there is no free memory -ENOMEM is returned, otherwise zero
2735 * is returned and the old skb data released.
2737 static inline int skb_linearize_cow(struct sk_buff *skb)
2739 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2740 __skb_linearize(skb) : 0;
2744 * skb_postpull_rcsum - update checksum for received skb after pull
2745 * @skb: buffer to update
2746 * @start: start of data before pull
2747 * @len: length of data pulled
2749 * After doing a pull on a received packet, you need to call this to
2750 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2751 * CHECKSUM_NONE so that it can be recomputed from scratch.
2754 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2755 const void *start, unsigned int len)
2757 if (skb->ip_summed == CHECKSUM_COMPLETE)
2758 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2759 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2760 skb_checksum_start_offset(skb) < 0)
2761 skb->ip_summed = CHECKSUM_NONE;
2764 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2766 static inline void skb_postpush_rcsum(struct sk_buff *skb,
2767 const void *start, unsigned int len)
2769 /* For performing the reverse operation to skb_postpull_rcsum(),
2770 * we can instead of ...
2772 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0));
2774 * ... just use this equivalent version here to save a few
2775 * instructions. Feeding csum of 0 in csum_partial() and later
2776 * on adding skb->csum is equivalent to feed skb->csum in the
2779 if (skb->ip_summed == CHECKSUM_COMPLETE)
2780 skb->csum = csum_partial(start, len, skb->csum);
2784 * skb_push_rcsum - push skb and update receive checksum
2785 * @skb: buffer to update
2786 * @len: length of data pulled
2788 * This function performs an skb_push on the packet and updates
2789 * the CHECKSUM_COMPLETE checksum. It should be used on
2790 * receive path processing instead of skb_push unless you know
2791 * that the checksum difference is zero (e.g., a valid IP header)
2792 * or you are setting ip_summed to CHECKSUM_NONE.
2794 static inline unsigned char *skb_push_rcsum(struct sk_buff *skb,
2798 skb_postpush_rcsum(skb, skb->data, len);
2803 * pskb_trim_rcsum - trim received skb and update checksum
2804 * @skb: buffer to trim
2807 * This is exactly the same as pskb_trim except that it ensures the
2808 * checksum of received packets are still valid after the operation.
2811 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2813 if (likely(len >= skb->len))
2815 if (skb->ip_summed == CHECKSUM_COMPLETE)
2816 skb->ip_summed = CHECKSUM_NONE;
2817 return __pskb_trim(skb, len);
2820 #define skb_queue_walk(queue, skb) \
2821 for (skb = (queue)->next; \
2822 skb != (struct sk_buff *)(queue); \
2825 #define skb_queue_walk_safe(queue, skb, tmp) \
2826 for (skb = (queue)->next, tmp = skb->next; \
2827 skb != (struct sk_buff *)(queue); \
2828 skb = tmp, tmp = skb->next)
2830 #define skb_queue_walk_from(queue, skb) \
2831 for (; skb != (struct sk_buff *)(queue); \
2834 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2835 for (tmp = skb->next; \
2836 skb != (struct sk_buff *)(queue); \
2837 skb = tmp, tmp = skb->next)
2839 #define skb_queue_reverse_walk(queue, skb) \
2840 for (skb = (queue)->prev; \
2841 skb != (struct sk_buff *)(queue); \
2844 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2845 for (skb = (queue)->prev, tmp = skb->prev; \
2846 skb != (struct sk_buff *)(queue); \
2847 skb = tmp, tmp = skb->prev)
2849 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2850 for (tmp = skb->prev; \
2851 skb != (struct sk_buff *)(queue); \
2852 skb = tmp, tmp = skb->prev)
2854 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2856 return skb_shinfo(skb)->frag_list != NULL;
2859 static inline void skb_frag_list_init(struct sk_buff *skb)
2861 skb_shinfo(skb)->frag_list = NULL;
2864 #define skb_walk_frags(skb, iter) \
2865 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2867 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2868 int *peeked, int *off, int *err);
2869 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2871 unsigned int datagram_poll(struct file *file, struct socket *sock,
2872 struct poll_table_struct *wait);
2873 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2874 struct iov_iter *to, int size);
2875 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2876 struct msghdr *msg, int size)
2878 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
2880 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2881 struct msghdr *msg);
2882 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2883 struct iov_iter *from, int len);
2884 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
2885 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2886 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2887 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2888 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2889 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2890 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2891 int len, __wsum csum);
2892 ssize_t skb_socket_splice(struct sock *sk,
2893 struct pipe_inode_info *pipe,
2894 struct splice_pipe_desc *spd);
2895 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2896 struct pipe_inode_info *pipe, unsigned int len,
2898 ssize_t (*splice_cb)(struct sock *,
2899 struct pipe_inode_info *,
2900 struct splice_pipe_desc *));
2901 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2902 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2903 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2905 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2906 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2907 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2908 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2909 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2910 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
2911 int skb_ensure_writable(struct sk_buff *skb, int write_len);
2912 int skb_vlan_pop(struct sk_buff *skb);
2913 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
2915 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2917 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
2920 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
2922 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
2925 struct skb_checksum_ops {
2926 __wsum (*update)(const void *mem, int len, __wsum wsum);
2927 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2930 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2931 __wsum csum, const struct skb_checksum_ops *ops);
2932 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2935 static inline void * __must_check
2936 __skb_header_pointer(const struct sk_buff *skb, int offset,
2937 int len, void *data, int hlen, void *buffer)
2939 if (hlen - offset >= len)
2940 return data + offset;
2943 skb_copy_bits(skb, offset, buffer, len) < 0)
2949 static inline void * __must_check
2950 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
2952 return __skb_header_pointer(skb, offset, len, skb->data,
2953 skb_headlen(skb), buffer);
2957 * skb_needs_linearize - check if we need to linearize a given skb
2958 * depending on the given device features.
2959 * @skb: socket buffer to check
2960 * @features: net device features
2962 * Returns true if either:
2963 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2964 * 2. skb is fragmented and the device does not support SG.
2966 static inline bool skb_needs_linearize(struct sk_buff *skb,
2967 netdev_features_t features)
2969 return skb_is_nonlinear(skb) &&
2970 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2971 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2974 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2976 const unsigned int len)
2978 memcpy(to, skb->data, len);
2981 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2982 const int offset, void *to,
2983 const unsigned int len)
2985 memcpy(to, skb->data + offset, len);
2988 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2990 const unsigned int len)
2992 memcpy(skb->data, from, len);
2995 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2998 const unsigned int len)
3000 memcpy(skb->data + offset, from, len);
3003 void skb_init(void);
3005 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3011 * skb_get_timestamp - get timestamp from a skb
3012 * @skb: skb to get stamp from
3013 * @stamp: pointer to struct timeval to store stamp in
3015 * Timestamps are stored in the skb as offsets to a base timestamp.
3016 * This function converts the offset back to a struct timeval and stores
3019 static inline void skb_get_timestamp(const struct sk_buff *skb,
3020 struct timeval *stamp)
3022 *stamp = ktime_to_timeval(skb->tstamp);
3025 static inline void skb_get_timestampns(const struct sk_buff *skb,
3026 struct timespec *stamp)
3028 *stamp = ktime_to_timespec(skb->tstamp);
3031 static inline void __net_timestamp(struct sk_buff *skb)
3033 skb->tstamp = ktime_get_real();
3036 static inline ktime_t net_timedelta(ktime_t t)
3038 return ktime_sub(ktime_get_real(), t);
3041 static inline ktime_t net_invalid_timestamp(void)
3043 return ktime_set(0, 0);
3046 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3048 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3050 void skb_clone_tx_timestamp(struct sk_buff *skb);
3051 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3053 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3055 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3059 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3064 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3067 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3069 * PHY drivers may accept clones of transmitted packets for
3070 * timestamping via their phy_driver.txtstamp method. These drivers
3071 * must call this function to return the skb back to the stack with a
3074 * @skb: clone of the the original outgoing packet
3075 * @hwtstamps: hardware time stamps
3078 void skb_complete_tx_timestamp(struct sk_buff *skb,
3079 struct skb_shared_hwtstamps *hwtstamps);
3081 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3082 struct skb_shared_hwtstamps *hwtstamps,
3083 struct sock *sk, int tstype);
3086 * skb_tstamp_tx - queue clone of skb with send time stamps
3087 * @orig_skb: the original outgoing packet
3088 * @hwtstamps: hardware time stamps, may be NULL if not available
3090 * If the skb has a socket associated, then this function clones the
3091 * skb (thus sharing the actual data and optional structures), stores
3092 * the optional hardware time stamping information (if non NULL) or
3093 * generates a software time stamp (otherwise), then queues the clone
3094 * to the error queue of the socket. Errors are silently ignored.
3096 void skb_tstamp_tx(struct sk_buff *orig_skb,
3097 struct skb_shared_hwtstamps *hwtstamps);
3099 static inline void sw_tx_timestamp(struct sk_buff *skb)
3101 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3102 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
3103 skb_tstamp_tx(skb, NULL);
3107 * skb_tx_timestamp() - Driver hook for transmit timestamping
3109 * Ethernet MAC Drivers should call this function in their hard_xmit()
3110 * function immediately before giving the sk_buff to the MAC hardware.
3112 * Specifically, one should make absolutely sure that this function is
3113 * called before TX completion of this packet can trigger. Otherwise
3114 * the packet could potentially already be freed.
3116 * @skb: A socket buffer.
3118 static inline void skb_tx_timestamp(struct sk_buff *skb)
3120 skb_clone_tx_timestamp(skb);
3121 sw_tx_timestamp(skb);
3125 * skb_complete_wifi_ack - deliver skb with wifi status
3127 * @skb: the original outgoing packet
3128 * @acked: ack status
3131 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3133 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3134 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3136 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3138 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3140 (skb->ip_summed == CHECKSUM_PARTIAL &&
3141 skb_checksum_start_offset(skb) >= 0));
3145 * skb_checksum_complete - Calculate checksum of an entire packet
3146 * @skb: packet to process
3148 * This function calculates the checksum over the entire packet plus
3149 * the value of skb->csum. The latter can be used to supply the
3150 * checksum of a pseudo header as used by TCP/UDP. It returns the
3153 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3154 * this function can be used to verify that checksum on received
3155 * packets. In that case the function should return zero if the
3156 * checksum is correct. In particular, this function will return zero
3157 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3158 * hardware has already verified the correctness of the checksum.
3160 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3162 return skb_csum_unnecessary(skb) ?
3163 0 : __skb_checksum_complete(skb);
3166 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3168 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3169 if (skb->csum_level == 0)
3170 skb->ip_summed = CHECKSUM_NONE;
3176 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3178 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3179 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3181 } else if (skb->ip_summed == CHECKSUM_NONE) {
3182 skb->ip_summed = CHECKSUM_UNNECESSARY;
3183 skb->csum_level = 0;
3187 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3189 /* Mark current checksum as bad (typically called from GRO
3190 * path). In the case that ip_summed is CHECKSUM_NONE
3191 * this must be the first checksum encountered in the packet.
3192 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3193 * checksum after the last one validated. For UDP, a zero
3194 * checksum can not be marked as bad.
3197 if (skb->ip_summed == CHECKSUM_NONE ||
3198 skb->ip_summed == CHECKSUM_UNNECESSARY)
3202 /* Check if we need to perform checksum complete validation.
3204 * Returns true if checksum complete is needed, false otherwise
3205 * (either checksum is unnecessary or zero checksum is allowed).
3207 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3211 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3212 skb->csum_valid = 1;
3213 __skb_decr_checksum_unnecessary(skb);
3220 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3223 #define CHECKSUM_BREAK 76
3225 /* Unset checksum-complete
3227 * Unset checksum complete can be done when packet is being modified
3228 * (uncompressed for instance) and checksum-complete value is
3231 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3233 if (skb->ip_summed == CHECKSUM_COMPLETE)
3234 skb->ip_summed = CHECKSUM_NONE;
3237 /* Validate (init) checksum based on checksum complete.
3240 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3241 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3242 * checksum is stored in skb->csum for use in __skb_checksum_complete
3243 * non-zero: value of invalid checksum
3246 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3250 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3251 if (!csum_fold(csum_add(psum, skb->csum))) {
3252 skb->csum_valid = 1;
3255 } else if (skb->csum_bad) {
3256 /* ip_summed == CHECKSUM_NONE in this case */
3257 return (__force __sum16)1;
3262 if (complete || skb->len <= CHECKSUM_BREAK) {
3265 csum = __skb_checksum_complete(skb);
3266 skb->csum_valid = !csum;
3273 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3278 /* Perform checksum validate (init). Note that this is a macro since we only
3279 * want to calculate the pseudo header which is an input function if necessary.
3280 * First we try to validate without any computation (checksum unnecessary) and
3281 * then calculate based on checksum complete calling the function to compute
3285 * 0: checksum is validated or try to in skb_checksum_complete
3286 * non-zero: value of invalid checksum
3288 #define __skb_checksum_validate(skb, proto, complete, \
3289 zero_okay, check, compute_pseudo) \
3291 __sum16 __ret = 0; \
3292 skb->csum_valid = 0; \
3293 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3294 __ret = __skb_checksum_validate_complete(skb, \
3295 complete, compute_pseudo(skb, proto)); \
3299 #define skb_checksum_init(skb, proto, compute_pseudo) \
3300 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3302 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3303 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3305 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3306 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3308 #define skb_checksum_validate_zero_check(skb, proto, check, \
3310 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3312 #define skb_checksum_simple_validate(skb) \
3313 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3315 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3317 return (skb->ip_summed == CHECKSUM_NONE &&
3318 skb->csum_valid && !skb->csum_bad);
3321 static inline void __skb_checksum_convert(struct sk_buff *skb,
3322 __sum16 check, __wsum pseudo)
3324 skb->csum = ~pseudo;
3325 skb->ip_summed = CHECKSUM_COMPLETE;
3328 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3330 if (__skb_checksum_convert_check(skb)) \
3331 __skb_checksum_convert(skb, check, \
3332 compute_pseudo(skb, proto)); \
3335 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3336 u16 start, u16 offset)
3338 skb->ip_summed = CHECKSUM_PARTIAL;
3339 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3340 skb->csum_offset = offset - start;
3343 /* Update skbuf and packet to reflect the remote checksum offload operation.
3344 * When called, ptr indicates the starting point for skb->csum when
3345 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3346 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3348 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3349 int start, int offset, bool nopartial)
3354 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3358 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3359 __skb_checksum_complete(skb);
3360 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3363 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3365 /* Adjust skb->csum since we changed the packet */
3366 skb->csum = csum_add(skb->csum, delta);
3369 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3370 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3371 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3373 if (nfct && atomic_dec_and_test(&nfct->use))
3374 nf_conntrack_destroy(nfct);
3376 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3379 atomic_inc(&nfct->use);
3382 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3383 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3385 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3388 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3391 atomic_inc(&nf_bridge->use);
3393 #endif /* CONFIG_BRIDGE_NETFILTER */
3394 static inline void nf_reset(struct sk_buff *skb)
3396 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3397 nf_conntrack_put(skb->nfct);
3400 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3401 nf_bridge_put(skb->nf_bridge);
3402 skb->nf_bridge = NULL;
3406 static inline void nf_reset_trace(struct sk_buff *skb)
3408 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3413 /* Note: This doesn't put any conntrack and bridge info in dst. */
3414 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3417 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3418 dst->nfct = src->nfct;
3419 nf_conntrack_get(src->nfct);
3421 dst->nfctinfo = src->nfctinfo;
3423 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3424 dst->nf_bridge = src->nf_bridge;
3425 nf_bridge_get(src->nf_bridge);
3427 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3429 dst->nf_trace = src->nf_trace;
3433 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3435 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3436 nf_conntrack_put(dst->nfct);
3438 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3439 nf_bridge_put(dst->nf_bridge);
3441 __nf_copy(dst, src, true);
3444 #ifdef CONFIG_NETWORK_SECMARK
3445 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3447 to->secmark = from->secmark;
3450 static inline void skb_init_secmark(struct sk_buff *skb)
3455 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3458 static inline void skb_init_secmark(struct sk_buff *skb)
3462 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3464 return !skb->destructor &&
3465 #if IS_ENABLED(CONFIG_XFRM)
3468 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3471 !skb->_skb_refdst &&
3472 !skb_has_frag_list(skb);
3475 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3477 skb->queue_mapping = queue_mapping;
3480 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3482 return skb->queue_mapping;
3485 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3487 to->queue_mapping = from->queue_mapping;
3490 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3492 skb->queue_mapping = rx_queue + 1;
3495 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3497 return skb->queue_mapping - 1;
3500 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3502 return skb->queue_mapping != 0;
3505 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3514 /* Keeps track of mac header offset relative to skb->head.
3515 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3516 * For non-tunnel skb it points to skb_mac_header() and for
3517 * tunnel skb it points to outer mac header.
3518 * Keeps track of level of encapsulation of network headers.
3525 #define SKB_SGO_CB_OFFSET 32
3526 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3528 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3530 return (skb_mac_header(inner_skb) - inner_skb->head) -
3531 SKB_GSO_CB(inner_skb)->mac_offset;
3534 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3536 int new_headroom, headroom;
3539 headroom = skb_headroom(skb);
3540 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3544 new_headroom = skb_headroom(skb);
3545 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3549 /* Compute the checksum for a gso segment. First compute the checksum value
3550 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3551 * then add in skb->csum (checksum from csum_start to end of packet).
3552 * skb->csum and csum_start are then updated to reflect the checksum of the
3553 * resultant packet starting from the transport header-- the resultant checksum
3554 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3557 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3559 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3560 skb_transport_offset(skb);
3563 partial = csum_partial(skb_transport_header(skb), plen, skb->csum);
3565 SKB_GSO_CB(skb)->csum_start -= plen;
3567 return csum_fold(partial);
3570 static inline bool skb_is_gso(const struct sk_buff *skb)
3572 return skb_shinfo(skb)->gso_size;
3575 /* Note: Should be called only if skb_is_gso(skb) is true */
3576 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3578 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3581 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3583 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3585 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3586 * wanted then gso_type will be set. */
3587 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3589 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3590 unlikely(shinfo->gso_type == 0)) {
3591 __skb_warn_lro_forwarding(skb);
3597 static inline void skb_forward_csum(struct sk_buff *skb)
3599 /* Unfortunately we don't support this one. Any brave souls? */
3600 if (skb->ip_summed == CHECKSUM_COMPLETE)
3601 skb->ip_summed = CHECKSUM_NONE;
3605 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3606 * @skb: skb to check
3608 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3609 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3610 * use this helper, to document places where we make this assertion.
3612 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3615 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3619 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3621 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3622 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3623 unsigned int transport_len,
3624 __sum16(*skb_chkf)(struct sk_buff *skb));
3627 * skb_head_is_locked - Determine if the skb->head is locked down
3628 * @skb: skb to check
3630 * The head on skbs build around a head frag can be removed if they are
3631 * not cloned. This function returns true if the skb head is locked down
3632 * due to either being allocated via kmalloc, or by being a clone with
3633 * multiple references to the head.
3635 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3637 return !skb->head_frag || skb_cloned(skb);
3641 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3645 * skb_gso_network_seglen is used to determine the real size of the
3646 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3648 * The MAC/L2 header is not accounted for.
3650 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3652 unsigned int hdr_len = skb_transport_header(skb) -
3653 skb_network_header(skb);
3654 return hdr_len + skb_gso_transport_seglen(skb);
3657 #endif /* __KERNEL__ */
3658 #endif /* _LINUX_SKBUFF_H */