2 * Definitions for the 'struct sk_buff' memory handlers.
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
17 #include <linux/kernel.h>
18 #include <linux/kmemcheck.h>
19 #include <linux/compiler.h>
20 #include <linux/time.h>
21 #include <linux/bug.h>
22 #include <linux/cache.h>
24 #include <linux/atomic.h>
25 #include <asm/types.h>
26 #include <linux/spinlock.h>
27 #include <linux/net.h>
28 #include <linux/textsearch.h>
29 #include <net/checksum.h>
30 #include <linux/rcupdate.h>
31 #include <linux/hrtimer.h>
32 #include <linux/dma-mapping.h>
33 #include <linux/netdev_features.h>
34 #include <linux/sched.h>
35 #include <net/flow_keys.h>
37 /* A. Checksumming of received packets by device.
41 * Device failed to checksum this packet e.g. due to lack of capabilities.
42 * The packet contains full (though not verified) checksum in packet but
43 * not in skb->csum. Thus, skb->csum is undefined in this case.
45 * CHECKSUM_UNNECESSARY:
47 * The hardware you're dealing with doesn't calculate the full checksum
48 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
49 * for specific protocols e.g. TCP/UDP/SCTP, then, for such packets it will
50 * set CHECKSUM_UNNECESSARY if their checksums are okay. skb->csum is still
51 * undefined in this case though. It is a bad option, but, unfortunately,
52 * nowadays most vendors do this. Apparently with the secret goal to sell
53 * you new devices, when you will add new protocol to your host, f.e. IPv6 8)
57 * This is the most generic way. The device supplied checksum of the _whole_
58 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
59 * hardware doesn't need to parse L3/L4 headers to implement this.
61 * Note: Even if device supports only some protocols, but is able to produce
62 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
66 * This is identical to the case for output below. This may occur on a packet
67 * received directly from another Linux OS, e.g., a virtualized Linux kernel
68 * on the same host. The packet can be treated in the same way as
69 * CHECKSUM_UNNECESSARY, except that on output (i.e., forwarding) the
70 * checksum must be filled in by the OS or the hardware.
72 * B. Checksumming on output.
76 * The skb was already checksummed by the protocol, or a checksum is not
81 * The device is required to checksum the packet as seen by hard_start_xmit()
82 * from skb->csum_start up to the end, and to record/write the checksum at
83 * offset skb->csum_start + skb->csum_offset.
85 * The device must show its capabilities in dev->features, set up at device
86 * setup time, e.g. netdev_features.h:
88 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything.
89 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over
90 * IPv4. Sigh. Vendors like this way for an unknown reason.
91 * Though, see comment above about CHECKSUM_UNNECESSARY. 8)
92 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead.
93 * NETIF_F_... - Well, you get the picture.
95 * CHECKSUM_UNNECESSARY:
97 * Normally, the device will do per protocol specific checksumming. Protocol
98 * implementations that do not want the NIC to perform the checksum
99 * calculation should use this flag in their outgoing skbs.
101 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC
102 * offload. Correspondingly, the FCoE protocol driver
103 * stack should use CHECKSUM_UNNECESSARY.
105 * Any questions? No questions, good. --ANK
108 /* Don't change this without changing skb_csum_unnecessary! */
109 #define CHECKSUM_NONE 0
110 #define CHECKSUM_UNNECESSARY 1
111 #define CHECKSUM_COMPLETE 2
112 #define CHECKSUM_PARTIAL 3
114 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
115 #define SKB_WITH_OVERHEAD(X) \
116 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
117 #define SKB_MAX_ORDER(X, ORDER) \
118 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
119 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
120 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
122 /* return minimum truesize of one skb containing X bytes of data */
123 #define SKB_TRUESIZE(X) ((X) + \
124 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
125 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
129 struct pipe_inode_info;
131 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
132 struct nf_conntrack {
137 #ifdef CONFIG_BRIDGE_NETFILTER
138 struct nf_bridge_info {
141 struct net_device *physindev;
142 struct net_device *physoutdev;
143 unsigned long data[32 / sizeof(unsigned long)];
147 struct sk_buff_head {
148 /* These two members must be first. */
149 struct sk_buff *next;
150 struct sk_buff *prev;
158 /* To allow 64K frame to be packed as single skb without frag_list we
159 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
160 * buffers which do not start on a page boundary.
162 * Since GRO uses frags we allocate at least 16 regardless of page
165 #if (65536/PAGE_SIZE + 1) < 16
166 #define MAX_SKB_FRAGS 16UL
168 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
171 typedef struct skb_frag_struct skb_frag_t;
173 struct skb_frag_struct {
177 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
186 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
191 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
196 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
201 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
206 #define HAVE_HW_TIME_STAMP
209 * struct skb_shared_hwtstamps - hardware time stamps
210 * @hwtstamp: hardware time stamp transformed into duration
211 * since arbitrary point in time
213 * Software time stamps generated by ktime_get_real() are stored in
216 * hwtstamps can only be compared against other hwtstamps from
219 * This structure is attached to packets as part of the
220 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
222 struct skb_shared_hwtstamps {
226 /* Definitions for tx_flags in struct skb_shared_info */
228 /* generate hardware time stamp */
229 SKBTX_HW_TSTAMP = 1 << 0,
231 /* generate software time stamp when queueing packet to NIC */
232 SKBTX_SW_TSTAMP = 1 << 1,
234 /* device driver is going to provide hardware time stamp */
235 SKBTX_IN_PROGRESS = 1 << 2,
237 /* device driver supports TX zero-copy buffers */
238 SKBTX_DEV_ZEROCOPY = 1 << 3,
240 /* generate wifi status information (where possible) */
241 SKBTX_WIFI_STATUS = 1 << 4,
243 /* This indicates at least one fragment might be overwritten
244 * (as in vmsplice(), sendfile() ...)
245 * If we need to compute a TX checksum, we'll need to copy
246 * all frags to avoid possible bad checksum
248 SKBTX_SHARED_FRAG = 1 << 5,
250 /* generate software time stamp when entering packet scheduling */
251 SKBTX_SCHED_TSTAMP = 1 << 6,
253 /* generate software timestamp on peer data acknowledgment */
254 SKBTX_ACK_TSTAMP = 1 << 7,
257 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
258 SKBTX_SCHED_TSTAMP | \
260 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
263 * The callback notifies userspace to release buffers when skb DMA is done in
264 * lower device, the skb last reference should be 0 when calling this.
265 * The zerocopy_success argument is true if zero copy transmit occurred,
266 * false on data copy or out of memory error caused by data copy attempt.
267 * The ctx field is used to track device context.
268 * The desc field is used to track userspace buffer index.
271 void (*callback)(struct ubuf_info *, bool zerocopy_success);
276 /* This data is invariant across clones and lives at
277 * the end of the header data, ie. at skb->end.
279 struct skb_shared_info {
280 unsigned char nr_frags;
282 unsigned short gso_size;
283 /* Warning: this field is not always filled in (UFO)! */
284 unsigned short gso_segs;
285 unsigned short gso_type;
286 struct sk_buff *frag_list;
287 struct skb_shared_hwtstamps hwtstamps;
292 * Warning : all fields before dataref are cleared in __alloc_skb()
296 /* Intermediate layers must ensure that destructor_arg
297 * remains valid until skb destructor */
298 void * destructor_arg;
300 /* must be last field, see pskb_expand_head() */
301 skb_frag_t frags[MAX_SKB_FRAGS];
304 /* We divide dataref into two halves. The higher 16 bits hold references
305 * to the payload part of skb->data. The lower 16 bits hold references to
306 * the entire skb->data. A clone of a headerless skb holds the length of
307 * the header in skb->hdr_len.
309 * All users must obey the rule that the skb->data reference count must be
310 * greater than or equal to the payload reference count.
312 * Holding a reference to the payload part means that the user does not
313 * care about modifications to the header part of skb->data.
315 #define SKB_DATAREF_SHIFT 16
316 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
320 SKB_FCLONE_UNAVAILABLE,
326 SKB_GSO_TCPV4 = 1 << 0,
327 SKB_GSO_UDP = 1 << 1,
329 /* This indicates the skb is from an untrusted source. */
330 SKB_GSO_DODGY = 1 << 2,
332 /* This indicates the tcp segment has CWR set. */
333 SKB_GSO_TCP_ECN = 1 << 3,
335 SKB_GSO_TCPV6 = 1 << 4,
337 SKB_GSO_FCOE = 1 << 5,
339 SKB_GSO_GRE = 1 << 6,
341 SKB_GSO_GRE_CSUM = 1 << 7,
343 SKB_GSO_IPIP = 1 << 8,
345 SKB_GSO_SIT = 1 << 9,
347 SKB_GSO_UDP_TUNNEL = 1 << 10,
349 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
351 SKB_GSO_MPLS = 1 << 12,
355 #if BITS_PER_LONG > 32
356 #define NET_SKBUFF_DATA_USES_OFFSET 1
359 #ifdef NET_SKBUFF_DATA_USES_OFFSET
360 typedef unsigned int sk_buff_data_t;
362 typedef unsigned char *sk_buff_data_t;
366 * struct skb_mstamp - multi resolution time stamps
367 * @stamp_us: timestamp in us resolution
368 * @stamp_jiffies: timestamp in jiffies
381 * skb_mstamp_get - get current timestamp
382 * @cl: place to store timestamps
384 static inline void skb_mstamp_get(struct skb_mstamp *cl)
386 u64 val = local_clock();
388 do_div(val, NSEC_PER_USEC);
389 cl->stamp_us = (u32)val;
390 cl->stamp_jiffies = (u32)jiffies;
394 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
395 * @t1: pointer to newest sample
396 * @t0: pointer to oldest sample
398 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
399 const struct skb_mstamp *t0)
401 s32 delta_us = t1->stamp_us - t0->stamp_us;
402 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
404 /* If delta_us is negative, this might be because interval is too big,
405 * or local_clock() drift is too big : fallback using jiffies.
408 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
410 delta_us = jiffies_to_usecs(delta_jiffies);
417 * struct sk_buff - socket buffer
418 * @next: Next buffer in list
419 * @prev: Previous buffer in list
420 * @tstamp: Time we arrived/left
421 * @sk: Socket we are owned by
422 * @dev: Device we arrived on/are leaving by
423 * @cb: Control buffer. Free for use by every layer. Put private vars here
424 * @_skb_refdst: destination entry (with norefcount bit)
425 * @sp: the security path, used for xfrm
426 * @len: Length of actual data
427 * @data_len: Data length
428 * @mac_len: Length of link layer header
429 * @hdr_len: writable header length of cloned skb
430 * @csum: Checksum (must include start/offset pair)
431 * @csum_start: Offset from skb->head where checksumming should start
432 * @csum_offset: Offset from csum_start where checksum should be stored
433 * @priority: Packet queueing priority
434 * @ignore_df: allow local fragmentation
435 * @cloned: Head may be cloned (check refcnt to be sure)
436 * @ip_summed: Driver fed us an IP checksum
437 * @nohdr: Payload reference only, must not modify header
438 * @nfctinfo: Relationship of this skb to the connection
439 * @pkt_type: Packet class
440 * @fclone: skbuff clone status
441 * @ipvs_property: skbuff is owned by ipvs
442 * @peeked: this packet has been seen already, so stats have been
443 * done for it, don't do them again
444 * @nf_trace: netfilter packet trace flag
445 * @protocol: Packet protocol from driver
446 * @destructor: Destruct function
447 * @nfct: Associated connection, if any
448 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
449 * @skb_iif: ifindex of device we arrived on
450 * @tc_index: Traffic control index
451 * @tc_verd: traffic control verdict
452 * @hash: the packet hash
453 * @queue_mapping: Queue mapping for multiqueue devices
454 * @ndisc_nodetype: router type (from link layer)
455 * @ooo_okay: allow the mapping of a socket to a queue to be changed
456 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
458 * @sw_hash: indicates hash was computed in software stack
459 * @wifi_acked_valid: wifi_acked was set
460 * @wifi_acked: whether frame was acked on wifi or not
461 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
462 * @dma_cookie: a cookie to one of several possible DMA operations
463 * done by skb DMA functions
464 * @napi_id: id of the NAPI struct this skb came from
465 * @secmark: security marking
466 * @mark: Generic packet mark
467 * @dropcount: total number of sk_receive_queue overflows
468 * @vlan_proto: vlan encapsulation protocol
469 * @vlan_tci: vlan tag control information
470 * @inner_protocol: Protocol (encapsulation)
471 * @inner_transport_header: Inner transport layer header (encapsulation)
472 * @inner_network_header: Network layer header (encapsulation)
473 * @inner_mac_header: Link layer header (encapsulation)
474 * @transport_header: Transport layer header
475 * @network_header: Network layer header
476 * @mac_header: Link layer header
477 * @tail: Tail pointer
479 * @head: Head of buffer
480 * @data: Data head pointer
481 * @truesize: Buffer size
482 * @users: User count - see {datagram,tcp}.c
486 /* These two members must be first. */
487 struct sk_buff *next;
488 struct sk_buff *prev;
492 struct skb_mstamp skb_mstamp;
496 struct net_device *dev;
499 * This is the control buffer. It is free to use for every
500 * layer. Please put your private variables there. If you
501 * want to keep them across layers you have to do a skb_clone()
502 * first. This is owned by whoever has the skb queued ATM.
504 char cb[48] __aligned(8);
506 unsigned long _skb_refdst;
522 kmemcheck_bitfield_begin(flags1);
533 kmemcheck_bitfield_end(flags1);
536 void (*destructor)(struct sk_buff *skb);
537 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
538 struct nf_conntrack *nfct;
540 #ifdef CONFIG_BRIDGE_NETFILTER
541 struct nf_bridge_info *nf_bridge;
551 #ifdef CONFIG_NET_SCHED
552 __u16 tc_index; /* traffic control index */
553 #ifdef CONFIG_NET_CLS_ACT
554 __u16 tc_verd; /* traffic control verdict */
559 kmemcheck_bitfield_begin(flags2);
560 #ifdef CONFIG_IPV6_NDISC_NODETYPE
561 __u8 ndisc_nodetype:2;
567 __u8 wifi_acked_valid:1;
571 /* Encapsulation protocol and NIC drivers should use
572 * this flag to indicate to each other if the skb contains
573 * encapsulated packet or not and maybe use the inner packet
576 __u8 encapsulation:1;
577 __u8 encap_hdr_csum:1;
579 __u8 csum_complete_sw:1;
580 /* 2/4 bit hole (depending on ndisc_nodetype presence) */
581 kmemcheck_bitfield_end(flags2);
583 #ifdef CONFIG_NET_RX_BUSY_POLL
584 unsigned int napi_id;
586 #ifdef CONFIG_NETWORK_SECMARK
592 __u32 reserved_tailroom;
595 __be16 inner_protocol;
596 __u16 inner_transport_header;
597 __u16 inner_network_header;
598 __u16 inner_mac_header;
599 __u16 transport_header;
600 __u16 network_header;
602 /* These elements must be at the end, see alloc_skb() for details. */
607 unsigned int truesize;
613 * Handling routines are only of interest to the kernel
615 #include <linux/slab.h>
618 #define SKB_ALLOC_FCLONE 0x01
619 #define SKB_ALLOC_RX 0x02
621 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
622 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
624 return unlikely(skb->pfmemalloc);
628 * skb might have a dst pointer attached, refcounted or not.
629 * _skb_refdst low order bit is set if refcount was _not_ taken
631 #define SKB_DST_NOREF 1UL
632 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
635 * skb_dst - returns skb dst_entry
638 * Returns skb dst_entry, regardless of reference taken or not.
640 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
642 /* If refdst was not refcounted, check we still are in a
643 * rcu_read_lock section
645 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
646 !rcu_read_lock_held() &&
647 !rcu_read_lock_bh_held());
648 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
652 * skb_dst_set - sets skb dst
656 * Sets skb dst, assuming a reference was taken on dst and should
657 * be released by skb_dst_drop()
659 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
661 skb->_skb_refdst = (unsigned long)dst;
664 void __skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst,
668 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
672 * Sets skb dst, assuming a reference was not taken on dst.
673 * If dst entry is cached, we do not take reference and dst_release
674 * will be avoided by refdst_drop. If dst entry is not cached, we take
675 * reference, so that last dst_release can destroy the dst immediately.
677 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
679 __skb_dst_set_noref(skb, dst, false);
683 * skb_dst_set_noref_force - sets skb dst, without taking reference
687 * Sets skb dst, assuming a reference was not taken on dst.
688 * No reference is taken and no dst_release will be called. While for
689 * cached dsts deferred reclaim is a basic feature, for entries that are
690 * not cached it is caller's job to guarantee that last dst_release for
691 * provided dst happens when nobody uses it, eg. after a RCU grace period.
693 static inline void skb_dst_set_noref_force(struct sk_buff *skb,
694 struct dst_entry *dst)
696 __skb_dst_set_noref(skb, dst, true);
700 * skb_dst_is_noref - Test if skb dst isn't refcounted
703 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
705 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
708 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
710 return (struct rtable *)skb_dst(skb);
713 void kfree_skb(struct sk_buff *skb);
714 void kfree_skb_list(struct sk_buff *segs);
715 void skb_tx_error(struct sk_buff *skb);
716 void consume_skb(struct sk_buff *skb);
717 void __kfree_skb(struct sk_buff *skb);
718 extern struct kmem_cache *skbuff_head_cache;
720 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
721 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
722 bool *fragstolen, int *delta_truesize);
724 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
726 struct sk_buff *build_skb(void *data, unsigned int frag_size);
727 static inline struct sk_buff *alloc_skb(unsigned int size,
730 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
733 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
736 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
739 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
740 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
742 return __alloc_skb_head(priority, -1);
745 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
746 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
747 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
748 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
749 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
750 gfp_t gfp_mask, bool fclone);
751 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
754 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
757 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
758 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
759 unsigned int headroom);
760 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
761 int newtailroom, gfp_t priority);
762 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
763 int offset, int len);
764 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
766 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
767 int skb_pad(struct sk_buff *skb, int pad);
768 #define dev_kfree_skb(a) consume_skb(a)
770 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
771 int getfrag(void *from, char *to, int offset,
772 int len, int odd, struct sk_buff *skb),
773 void *from, int length);
775 struct skb_seq_state {
779 __u32 stepped_offset;
780 struct sk_buff *root_skb;
781 struct sk_buff *cur_skb;
785 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
786 unsigned int to, struct skb_seq_state *st);
787 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
788 struct skb_seq_state *st);
789 void skb_abort_seq_read(struct skb_seq_state *st);
791 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
792 unsigned int to, struct ts_config *config,
793 struct ts_state *state);
796 * Packet hash types specify the type of hash in skb_set_hash.
798 * Hash types refer to the protocol layer addresses which are used to
799 * construct a packet's hash. The hashes are used to differentiate or identify
800 * flows of the protocol layer for the hash type. Hash types are either
801 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
803 * Properties of hashes:
805 * 1) Two packets in different flows have different hash values
806 * 2) Two packets in the same flow should have the same hash value
808 * A hash at a higher layer is considered to be more specific. A driver should
809 * set the most specific hash possible.
811 * A driver cannot indicate a more specific hash than the layer at which a hash
812 * was computed. For instance an L3 hash cannot be set as an L4 hash.
814 * A driver may indicate a hash level which is less specific than the
815 * actual layer the hash was computed on. For instance, a hash computed
816 * at L4 may be considered an L3 hash. This should only be done if the
817 * driver can't unambiguously determine that the HW computed the hash at
818 * the higher layer. Note that the "should" in the second property above
821 enum pkt_hash_types {
822 PKT_HASH_TYPE_NONE, /* Undefined type */
823 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
824 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
825 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
829 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
831 skb->l4_hash = (type == PKT_HASH_TYPE_L4);
836 void __skb_get_hash(struct sk_buff *skb);
837 static inline __u32 skb_get_hash(struct sk_buff *skb)
839 if (!skb->l4_hash && !skb->sw_hash)
845 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
850 static inline void skb_clear_hash(struct sk_buff *skb)
857 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
863 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
865 to->hash = from->hash;
866 to->sw_hash = from->sw_hash;
867 to->l4_hash = from->l4_hash;
870 #ifdef NET_SKBUFF_DATA_USES_OFFSET
871 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
873 return skb->head + skb->end;
876 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
881 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
886 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
888 return skb->end - skb->head;
893 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
895 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
897 return &skb_shinfo(skb)->hwtstamps;
901 * skb_queue_empty - check if a queue is empty
904 * Returns true if the queue is empty, false otherwise.
906 static inline int skb_queue_empty(const struct sk_buff_head *list)
908 return list->next == (const struct sk_buff *) list;
912 * skb_queue_is_last - check if skb is the last entry in the queue
916 * Returns true if @skb is the last buffer on the list.
918 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
919 const struct sk_buff *skb)
921 return skb->next == (const struct sk_buff *) list;
925 * skb_queue_is_first - check if skb is the first entry in the queue
929 * Returns true if @skb is the first buffer on the list.
931 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
932 const struct sk_buff *skb)
934 return skb->prev == (const struct sk_buff *) list;
938 * skb_queue_next - return the next packet in the queue
940 * @skb: current buffer
942 * Return the next packet in @list after @skb. It is only valid to
943 * call this if skb_queue_is_last() evaluates to false.
945 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
946 const struct sk_buff *skb)
948 /* This BUG_ON may seem severe, but if we just return then we
949 * are going to dereference garbage.
951 BUG_ON(skb_queue_is_last(list, skb));
956 * skb_queue_prev - return the prev packet in the queue
958 * @skb: current buffer
960 * Return the prev packet in @list before @skb. It is only valid to
961 * call this if skb_queue_is_first() evaluates to false.
963 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
964 const struct sk_buff *skb)
966 /* This BUG_ON may seem severe, but if we just return then we
967 * are going to dereference garbage.
969 BUG_ON(skb_queue_is_first(list, skb));
974 * skb_get - reference buffer
975 * @skb: buffer to reference
977 * Makes another reference to a socket buffer and returns a pointer
980 static inline struct sk_buff *skb_get(struct sk_buff *skb)
982 atomic_inc(&skb->users);
987 * If users == 1, we are the only owner and are can avoid redundant
992 * skb_cloned - is the buffer a clone
993 * @skb: buffer to check
995 * Returns true if the buffer was generated with skb_clone() and is
996 * one of multiple shared copies of the buffer. Cloned buffers are
997 * shared data so must not be written to under normal circumstances.
999 static inline int skb_cloned(const struct sk_buff *skb)
1001 return skb->cloned &&
1002 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1005 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1007 might_sleep_if(pri & __GFP_WAIT);
1009 if (skb_cloned(skb))
1010 return pskb_expand_head(skb, 0, 0, pri);
1016 * skb_header_cloned - is the header a clone
1017 * @skb: buffer to check
1019 * Returns true if modifying the header part of the buffer requires
1020 * the data to be copied.
1022 static inline int skb_header_cloned(const struct sk_buff *skb)
1029 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1030 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1031 return dataref != 1;
1035 * skb_header_release - release reference to header
1036 * @skb: buffer to operate on
1038 * Drop a reference to the header part of the buffer. This is done
1039 * by acquiring a payload reference. You must not read from the header
1040 * part of skb->data after this.
1042 static inline void skb_header_release(struct sk_buff *skb)
1046 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1050 * skb_shared - is the buffer shared
1051 * @skb: buffer to check
1053 * Returns true if more than one person has a reference to this
1056 static inline int skb_shared(const struct sk_buff *skb)
1058 return atomic_read(&skb->users) != 1;
1062 * skb_share_check - check if buffer is shared and if so clone it
1063 * @skb: buffer to check
1064 * @pri: priority for memory allocation
1066 * If the buffer is shared the buffer is cloned and the old copy
1067 * drops a reference. A new clone with a single reference is returned.
1068 * If the buffer is not shared the original buffer is returned. When
1069 * being called from interrupt status or with spinlocks held pri must
1072 * NULL is returned on a memory allocation failure.
1074 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1076 might_sleep_if(pri & __GFP_WAIT);
1077 if (skb_shared(skb)) {
1078 struct sk_buff *nskb = skb_clone(skb, pri);
1090 * Copy shared buffers into a new sk_buff. We effectively do COW on
1091 * packets to handle cases where we have a local reader and forward
1092 * and a couple of other messy ones. The normal one is tcpdumping
1093 * a packet thats being forwarded.
1097 * skb_unshare - make a copy of a shared buffer
1098 * @skb: buffer to check
1099 * @pri: priority for memory allocation
1101 * If the socket buffer is a clone then this function creates a new
1102 * copy of the data, drops a reference count on the old copy and returns
1103 * the new copy with the reference count at 1. If the buffer is not a clone
1104 * the original buffer is returned. When called with a spinlock held or
1105 * from interrupt state @pri must be %GFP_ATOMIC
1107 * %NULL is returned on a memory allocation failure.
1109 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1112 might_sleep_if(pri & __GFP_WAIT);
1113 if (skb_cloned(skb)) {
1114 struct sk_buff *nskb = skb_copy(skb, pri);
1115 kfree_skb(skb); /* Free our shared copy */
1122 * skb_peek - peek at the head of an &sk_buff_head
1123 * @list_: list to peek at
1125 * Peek an &sk_buff. Unlike most other operations you _MUST_
1126 * be careful with this one. A peek leaves the buffer on the
1127 * list and someone else may run off with it. You must hold
1128 * the appropriate locks or have a private queue to do this.
1130 * Returns %NULL for an empty list or a pointer to the head element.
1131 * The reference count is not incremented and the reference is therefore
1132 * volatile. Use with caution.
1134 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1136 struct sk_buff *skb = list_->next;
1138 if (skb == (struct sk_buff *)list_)
1144 * skb_peek_next - peek skb following the given one from a queue
1145 * @skb: skb to start from
1146 * @list_: list to peek at
1148 * Returns %NULL when the end of the list is met or a pointer to the
1149 * next element. The reference count is not incremented and the
1150 * reference is therefore volatile. Use with caution.
1152 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1153 const struct sk_buff_head *list_)
1155 struct sk_buff *next = skb->next;
1157 if (next == (struct sk_buff *)list_)
1163 * skb_peek_tail - peek at the tail of an &sk_buff_head
1164 * @list_: list to peek at
1166 * Peek an &sk_buff. Unlike most other operations you _MUST_
1167 * be careful with this one. A peek leaves the buffer on the
1168 * list and someone else may run off with it. You must hold
1169 * the appropriate locks or have a private queue to do this.
1171 * Returns %NULL for an empty list or a pointer to the tail element.
1172 * The reference count is not incremented and the reference is therefore
1173 * volatile. Use with caution.
1175 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1177 struct sk_buff *skb = list_->prev;
1179 if (skb == (struct sk_buff *)list_)
1186 * skb_queue_len - get queue length
1187 * @list_: list to measure
1189 * Return the length of an &sk_buff queue.
1191 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1197 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1198 * @list: queue to initialize
1200 * This initializes only the list and queue length aspects of
1201 * an sk_buff_head object. This allows to initialize the list
1202 * aspects of an sk_buff_head without reinitializing things like
1203 * the spinlock. It can also be used for on-stack sk_buff_head
1204 * objects where the spinlock is known to not be used.
1206 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1208 list->prev = list->next = (struct sk_buff *)list;
1213 * This function creates a split out lock class for each invocation;
1214 * this is needed for now since a whole lot of users of the skb-queue
1215 * infrastructure in drivers have different locking usage (in hardirq)
1216 * than the networking core (in softirq only). In the long run either the
1217 * network layer or drivers should need annotation to consolidate the
1218 * main types of usage into 3 classes.
1220 static inline void skb_queue_head_init(struct sk_buff_head *list)
1222 spin_lock_init(&list->lock);
1223 __skb_queue_head_init(list);
1226 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1227 struct lock_class_key *class)
1229 skb_queue_head_init(list);
1230 lockdep_set_class(&list->lock, class);
1234 * Insert an sk_buff on a list.
1236 * The "__skb_xxxx()" functions are the non-atomic ones that
1237 * can only be called with interrupts disabled.
1239 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1240 struct sk_buff_head *list);
1241 static inline void __skb_insert(struct sk_buff *newsk,
1242 struct sk_buff *prev, struct sk_buff *next,
1243 struct sk_buff_head *list)
1247 next->prev = prev->next = newsk;
1251 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1252 struct sk_buff *prev,
1253 struct sk_buff *next)
1255 struct sk_buff *first = list->next;
1256 struct sk_buff *last = list->prev;
1266 * skb_queue_splice - join two skb lists, this is designed for stacks
1267 * @list: the new list to add
1268 * @head: the place to add it in the first list
1270 static inline void skb_queue_splice(const struct sk_buff_head *list,
1271 struct sk_buff_head *head)
1273 if (!skb_queue_empty(list)) {
1274 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1275 head->qlen += list->qlen;
1280 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1281 * @list: the new list to add
1282 * @head: the place to add it in the first list
1284 * The list at @list is reinitialised
1286 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1287 struct sk_buff_head *head)
1289 if (!skb_queue_empty(list)) {
1290 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1291 head->qlen += list->qlen;
1292 __skb_queue_head_init(list);
1297 * skb_queue_splice_tail - join two skb lists, each list being a queue
1298 * @list: the new list to add
1299 * @head: the place to add it in the first list
1301 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1302 struct sk_buff_head *head)
1304 if (!skb_queue_empty(list)) {
1305 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1306 head->qlen += list->qlen;
1311 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1312 * @list: the new list to add
1313 * @head: the place to add it in the first list
1315 * Each of the lists is a queue.
1316 * The list at @list is reinitialised
1318 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1319 struct sk_buff_head *head)
1321 if (!skb_queue_empty(list)) {
1322 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1323 head->qlen += list->qlen;
1324 __skb_queue_head_init(list);
1329 * __skb_queue_after - queue a buffer at the list head
1330 * @list: list to use
1331 * @prev: place after this buffer
1332 * @newsk: buffer to queue
1334 * Queue a buffer int the middle of a list. This function takes no locks
1335 * and you must therefore hold required locks before calling it.
1337 * A buffer cannot be placed on two lists at the same time.
1339 static inline void __skb_queue_after(struct sk_buff_head *list,
1340 struct sk_buff *prev,
1341 struct sk_buff *newsk)
1343 __skb_insert(newsk, prev, prev->next, list);
1346 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1347 struct sk_buff_head *list);
1349 static inline void __skb_queue_before(struct sk_buff_head *list,
1350 struct sk_buff *next,
1351 struct sk_buff *newsk)
1353 __skb_insert(newsk, next->prev, next, list);
1357 * __skb_queue_head - queue a buffer at the list head
1358 * @list: list to use
1359 * @newsk: buffer to queue
1361 * Queue a buffer at the start of a list. This function takes no locks
1362 * and you must therefore hold required locks before calling it.
1364 * A buffer cannot be placed on two lists at the same time.
1366 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1367 static inline void __skb_queue_head(struct sk_buff_head *list,
1368 struct sk_buff *newsk)
1370 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1374 * __skb_queue_tail - queue a buffer at the list tail
1375 * @list: list to use
1376 * @newsk: buffer to queue
1378 * Queue a buffer at the end of a list. This function takes no locks
1379 * and you must therefore hold required locks before calling it.
1381 * A buffer cannot be placed on two lists at the same time.
1383 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1384 static inline void __skb_queue_tail(struct sk_buff_head *list,
1385 struct sk_buff *newsk)
1387 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1391 * remove sk_buff from list. _Must_ be called atomically, and with
1394 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1395 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1397 struct sk_buff *next, *prev;
1402 skb->next = skb->prev = NULL;
1408 * __skb_dequeue - remove from the head of the queue
1409 * @list: list to dequeue from
1411 * Remove the head of the list. This function does not take any locks
1412 * so must be used with appropriate locks held only. The head item is
1413 * returned or %NULL if the list is empty.
1415 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1416 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1418 struct sk_buff *skb = skb_peek(list);
1420 __skb_unlink(skb, list);
1425 * __skb_dequeue_tail - remove from the tail of the queue
1426 * @list: list to dequeue from
1428 * Remove the tail of the list. This function does not take any locks
1429 * so must be used with appropriate locks held only. The tail item is
1430 * returned or %NULL if the list is empty.
1432 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1433 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1435 struct sk_buff *skb = skb_peek_tail(list);
1437 __skb_unlink(skb, list);
1442 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1444 return skb->data_len;
1447 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1449 return skb->len - skb->data_len;
1452 static inline int skb_pagelen(const struct sk_buff *skb)
1456 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1457 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1458 return len + skb_headlen(skb);
1462 * __skb_fill_page_desc - initialise a paged fragment in an skb
1463 * @skb: buffer containing fragment to be initialised
1464 * @i: paged fragment index to initialise
1465 * @page: the page to use for this fragment
1466 * @off: the offset to the data with @page
1467 * @size: the length of the data
1469 * Initialises the @i'th fragment of @skb to point to &size bytes at
1470 * offset @off within @page.
1472 * Does not take any additional reference on the fragment.
1474 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1475 struct page *page, int off, int size)
1477 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1480 * Propagate page->pfmemalloc to the skb if we can. The problem is
1481 * that not all callers have unique ownership of the page. If
1482 * pfmemalloc is set, we check the mapping as a mapping implies
1483 * page->index is set (index and pfmemalloc share space).
1484 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1485 * do not lose pfmemalloc information as the pages would not be
1486 * allocated using __GFP_MEMALLOC.
1488 frag->page.p = page;
1489 frag->page_offset = off;
1490 skb_frag_size_set(frag, size);
1492 page = compound_head(page);
1493 if (page->pfmemalloc && !page->mapping)
1494 skb->pfmemalloc = true;
1498 * skb_fill_page_desc - initialise a paged fragment in an skb
1499 * @skb: buffer containing fragment to be initialised
1500 * @i: paged fragment index to initialise
1501 * @page: the page to use for this fragment
1502 * @off: the offset to the data with @page
1503 * @size: the length of the data
1505 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1506 * @skb to point to @size bytes at offset @off within @page. In
1507 * addition updates @skb such that @i is the last fragment.
1509 * Does not take any additional reference on the fragment.
1511 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1512 struct page *page, int off, int size)
1514 __skb_fill_page_desc(skb, i, page, off, size);
1515 skb_shinfo(skb)->nr_frags = i + 1;
1518 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1519 int size, unsigned int truesize);
1521 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1522 unsigned int truesize);
1524 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1525 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1526 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1528 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1529 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1531 return skb->head + skb->tail;
1534 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1536 skb->tail = skb->data - skb->head;
1539 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1541 skb_reset_tail_pointer(skb);
1542 skb->tail += offset;
1545 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1546 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1551 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1553 skb->tail = skb->data;
1556 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1558 skb->tail = skb->data + offset;
1561 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1564 * Add data to an sk_buff
1566 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1567 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1568 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1570 unsigned char *tmp = skb_tail_pointer(skb);
1571 SKB_LINEAR_ASSERT(skb);
1577 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1578 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1585 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1586 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1589 BUG_ON(skb->len < skb->data_len);
1590 return skb->data += len;
1593 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1595 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1598 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1600 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1602 if (len > skb_headlen(skb) &&
1603 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1606 return skb->data += len;
1609 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1611 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1614 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1616 if (likely(len <= skb_headlen(skb)))
1618 if (unlikely(len > skb->len))
1620 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1624 * skb_headroom - bytes at buffer head
1625 * @skb: buffer to check
1627 * Return the number of bytes of free space at the head of an &sk_buff.
1629 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1631 return skb->data - skb->head;
1635 * skb_tailroom - bytes at buffer end
1636 * @skb: buffer to check
1638 * Return the number of bytes of free space at the tail of an sk_buff
1640 static inline int skb_tailroom(const struct sk_buff *skb)
1642 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1646 * skb_availroom - bytes at buffer end
1647 * @skb: buffer to check
1649 * Return the number of bytes of free space at the tail of an sk_buff
1650 * allocated by sk_stream_alloc()
1652 static inline int skb_availroom(const struct sk_buff *skb)
1654 if (skb_is_nonlinear(skb))
1657 return skb->end - skb->tail - skb->reserved_tailroom;
1661 * skb_reserve - adjust headroom
1662 * @skb: buffer to alter
1663 * @len: bytes to move
1665 * Increase the headroom of an empty &sk_buff by reducing the tail
1666 * room. This is only allowed for an empty buffer.
1668 static inline void skb_reserve(struct sk_buff *skb, int len)
1674 static inline void skb_reset_inner_headers(struct sk_buff *skb)
1676 skb->inner_mac_header = skb->mac_header;
1677 skb->inner_network_header = skb->network_header;
1678 skb->inner_transport_header = skb->transport_header;
1681 static inline void skb_reset_mac_len(struct sk_buff *skb)
1683 skb->mac_len = skb->network_header - skb->mac_header;
1686 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1689 return skb->head + skb->inner_transport_header;
1692 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1694 skb->inner_transport_header = skb->data - skb->head;
1697 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1700 skb_reset_inner_transport_header(skb);
1701 skb->inner_transport_header += offset;
1704 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1706 return skb->head + skb->inner_network_header;
1709 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1711 skb->inner_network_header = skb->data - skb->head;
1714 static inline void skb_set_inner_network_header(struct sk_buff *skb,
1717 skb_reset_inner_network_header(skb);
1718 skb->inner_network_header += offset;
1721 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1723 return skb->head + skb->inner_mac_header;
1726 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1728 skb->inner_mac_header = skb->data - skb->head;
1731 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1734 skb_reset_inner_mac_header(skb);
1735 skb->inner_mac_header += offset;
1737 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1739 return skb->transport_header != (typeof(skb->transport_header))~0U;
1742 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1744 return skb->head + skb->transport_header;
1747 static inline void skb_reset_transport_header(struct sk_buff *skb)
1749 skb->transport_header = skb->data - skb->head;
1752 static inline void skb_set_transport_header(struct sk_buff *skb,
1755 skb_reset_transport_header(skb);
1756 skb->transport_header += offset;
1759 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1761 return skb->head + skb->network_header;
1764 static inline void skb_reset_network_header(struct sk_buff *skb)
1766 skb->network_header = skb->data - skb->head;
1769 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1771 skb_reset_network_header(skb);
1772 skb->network_header += offset;
1775 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1777 return skb->head + skb->mac_header;
1780 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1782 return skb->mac_header != (typeof(skb->mac_header))~0U;
1785 static inline void skb_reset_mac_header(struct sk_buff *skb)
1787 skb->mac_header = skb->data - skb->head;
1790 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1792 skb_reset_mac_header(skb);
1793 skb->mac_header += offset;
1796 static inline void skb_pop_mac_header(struct sk_buff *skb)
1798 skb->mac_header = skb->network_header;
1801 static inline void skb_probe_transport_header(struct sk_buff *skb,
1802 const int offset_hint)
1804 struct flow_keys keys;
1806 if (skb_transport_header_was_set(skb))
1808 else if (skb_flow_dissect(skb, &keys))
1809 skb_set_transport_header(skb, keys.thoff);
1811 skb_set_transport_header(skb, offset_hint);
1814 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1816 if (skb_mac_header_was_set(skb)) {
1817 const unsigned char *old_mac = skb_mac_header(skb);
1819 skb_set_mac_header(skb, -skb->mac_len);
1820 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
1824 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
1826 return skb->csum_start - skb_headroom(skb);
1829 static inline int skb_transport_offset(const struct sk_buff *skb)
1831 return skb_transport_header(skb) - skb->data;
1834 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1836 return skb->transport_header - skb->network_header;
1839 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
1841 return skb->inner_transport_header - skb->inner_network_header;
1844 static inline int skb_network_offset(const struct sk_buff *skb)
1846 return skb_network_header(skb) - skb->data;
1849 static inline int skb_inner_network_offset(const struct sk_buff *skb)
1851 return skb_inner_network_header(skb) - skb->data;
1854 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1856 return pskb_may_pull(skb, skb_network_offset(skb) + len);
1859 static inline void skb_pop_rcv_encapsulation(struct sk_buff *skb)
1861 /* Only continue with checksum unnecessary if device indicated
1862 * it is valid across encapsulation (skb->encapsulation was set).
1864 if (skb->ip_summed == CHECKSUM_UNNECESSARY && !skb->encapsulation)
1865 skb->ip_summed = CHECKSUM_NONE;
1867 skb->encapsulation = 0;
1868 skb->csum_valid = 0;
1872 * CPUs often take a performance hit when accessing unaligned memory
1873 * locations. The actual performance hit varies, it can be small if the
1874 * hardware handles it or large if we have to take an exception and fix it
1877 * Since an ethernet header is 14 bytes network drivers often end up with
1878 * the IP header at an unaligned offset. The IP header can be aligned by
1879 * shifting the start of the packet by 2 bytes. Drivers should do this
1882 * skb_reserve(skb, NET_IP_ALIGN);
1884 * The downside to this alignment of the IP header is that the DMA is now
1885 * unaligned. On some architectures the cost of an unaligned DMA is high
1886 * and this cost outweighs the gains made by aligning the IP header.
1888 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1891 #ifndef NET_IP_ALIGN
1892 #define NET_IP_ALIGN 2
1896 * The networking layer reserves some headroom in skb data (via
1897 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
1898 * the header has to grow. In the default case, if the header has to grow
1899 * 32 bytes or less we avoid the reallocation.
1901 * Unfortunately this headroom changes the DMA alignment of the resulting
1902 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
1903 * on some architectures. An architecture can override this value,
1904 * perhaps setting it to a cacheline in size (since that will maintain
1905 * cacheline alignment of the DMA). It must be a power of 2.
1907 * Various parts of the networking layer expect at least 32 bytes of
1908 * headroom, you should not reduce this.
1910 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
1911 * to reduce average number of cache lines per packet.
1912 * get_rps_cpus() for example only access one 64 bytes aligned block :
1913 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
1916 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
1919 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
1921 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
1923 if (unlikely(skb_is_nonlinear(skb))) {
1928 skb_set_tail_pointer(skb, len);
1931 void skb_trim(struct sk_buff *skb, unsigned int len);
1933 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
1936 return ___pskb_trim(skb, len);
1937 __skb_trim(skb, len);
1941 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
1943 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
1947 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
1948 * @skb: buffer to alter
1951 * This is identical to pskb_trim except that the caller knows that
1952 * the skb is not cloned so we should never get an error due to out-
1955 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
1957 int err = pskb_trim(skb, len);
1962 * skb_orphan - orphan a buffer
1963 * @skb: buffer to orphan
1965 * If a buffer currently has an owner then we call the owner's
1966 * destructor function and make the @skb unowned. The buffer continues
1967 * to exist but is no longer charged to its former owner.
1969 static inline void skb_orphan(struct sk_buff *skb)
1971 if (skb->destructor) {
1972 skb->destructor(skb);
1973 skb->destructor = NULL;
1981 * skb_orphan_frags - orphan the frags contained in a buffer
1982 * @skb: buffer to orphan frags from
1983 * @gfp_mask: allocation mask for replacement pages
1985 * For each frag in the SKB which needs a destructor (i.e. has an
1986 * owner) create a copy of that frag and release the original
1987 * page by calling the destructor.
1989 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
1991 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
1993 return skb_copy_ubufs(skb, gfp_mask);
1997 * __skb_queue_purge - empty a list
1998 * @list: list to empty
2000 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2001 * the list and one reference dropped. This function does not take the
2002 * list lock and the caller must hold the relevant locks to use it.
2004 void skb_queue_purge(struct sk_buff_head *list);
2005 static inline void __skb_queue_purge(struct sk_buff_head *list)
2007 struct sk_buff *skb;
2008 while ((skb = __skb_dequeue(list)) != NULL)
2012 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
2013 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
2014 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
2016 void *netdev_alloc_frag(unsigned int fragsz);
2018 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2022 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2023 * @dev: network device to receive on
2024 * @length: length to allocate
2026 * Allocate a new &sk_buff and assign it a usage count of one. The
2027 * buffer has unspecified headroom built in. Users should allocate
2028 * the headroom they think they need without accounting for the
2029 * built in space. The built in space is used for optimisations.
2031 * %NULL is returned if there is no free memory. Although this function
2032 * allocates memory it can be called from an interrupt.
2034 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2035 unsigned int length)
2037 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2040 /* legacy helper around __netdev_alloc_skb() */
2041 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2044 return __netdev_alloc_skb(NULL, length, gfp_mask);
2047 /* legacy helper around netdev_alloc_skb() */
2048 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2050 return netdev_alloc_skb(NULL, length);
2054 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2055 unsigned int length, gfp_t gfp)
2057 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2059 if (NET_IP_ALIGN && skb)
2060 skb_reserve(skb, NET_IP_ALIGN);
2064 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2065 unsigned int length)
2067 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2071 * __skb_alloc_pages - allocate pages for ps-rx on a skb and preserve pfmemalloc data
2072 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2073 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2074 * @order: size of the allocation
2076 * Allocate a new page.
2078 * %NULL is returned if there is no free memory.
2080 static inline struct page *__skb_alloc_pages(gfp_t gfp_mask,
2081 struct sk_buff *skb,
2086 gfp_mask |= __GFP_COLD;
2088 if (!(gfp_mask & __GFP_NOMEMALLOC))
2089 gfp_mask |= __GFP_MEMALLOC;
2091 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2092 if (skb && page && page->pfmemalloc)
2093 skb->pfmemalloc = true;
2099 * __skb_alloc_page - allocate a page for ps-rx for a given skb and preserve pfmemalloc data
2100 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2101 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2103 * Allocate a new page.
2105 * %NULL is returned if there is no free memory.
2107 static inline struct page *__skb_alloc_page(gfp_t gfp_mask,
2108 struct sk_buff *skb)
2110 return __skb_alloc_pages(gfp_mask, skb, 0);
2114 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2115 * @page: The page that was allocated from skb_alloc_page
2116 * @skb: The skb that may need pfmemalloc set
2118 static inline void skb_propagate_pfmemalloc(struct page *page,
2119 struct sk_buff *skb)
2121 if (page && page->pfmemalloc)
2122 skb->pfmemalloc = true;
2126 * skb_frag_page - retrieve the page referred to by a paged fragment
2127 * @frag: the paged fragment
2129 * Returns the &struct page associated with @frag.
2131 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2133 return frag->page.p;
2137 * __skb_frag_ref - take an addition reference on a paged fragment.
2138 * @frag: the paged fragment
2140 * Takes an additional reference on the paged fragment @frag.
2142 static inline void __skb_frag_ref(skb_frag_t *frag)
2144 get_page(skb_frag_page(frag));
2148 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2150 * @f: the fragment offset.
2152 * Takes an additional reference on the @f'th paged fragment of @skb.
2154 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2156 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2160 * __skb_frag_unref - release a reference on a paged fragment.
2161 * @frag: the paged fragment
2163 * Releases a reference on the paged fragment @frag.
2165 static inline void __skb_frag_unref(skb_frag_t *frag)
2167 put_page(skb_frag_page(frag));
2171 * skb_frag_unref - release a reference on a paged fragment of an skb.
2173 * @f: the fragment offset
2175 * Releases a reference on the @f'th paged fragment of @skb.
2177 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2179 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2183 * skb_frag_address - gets the address of the data contained in a paged fragment
2184 * @frag: the paged fragment buffer
2186 * Returns the address of the data within @frag. The page must already
2189 static inline void *skb_frag_address(const skb_frag_t *frag)
2191 return page_address(skb_frag_page(frag)) + frag->page_offset;
2195 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2196 * @frag: the paged fragment buffer
2198 * Returns the address of the data within @frag. Checks that the page
2199 * is mapped and returns %NULL otherwise.
2201 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2203 void *ptr = page_address(skb_frag_page(frag));
2207 return ptr + frag->page_offset;
2211 * __skb_frag_set_page - sets the page contained in a paged fragment
2212 * @frag: the paged fragment
2213 * @page: the page to set
2215 * Sets the fragment @frag to contain @page.
2217 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2219 frag->page.p = page;
2223 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2225 * @f: the fragment offset
2226 * @page: the page to set
2228 * Sets the @f'th fragment of @skb to contain @page.
2230 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2233 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2236 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2239 * skb_frag_dma_map - maps a paged fragment via the DMA API
2240 * @dev: the device to map the fragment to
2241 * @frag: the paged fragment to map
2242 * @offset: the offset within the fragment (starting at the
2243 * fragment's own offset)
2244 * @size: the number of bytes to map
2245 * @dir: the direction of the mapping (%PCI_DMA_*)
2247 * Maps the page associated with @frag to @device.
2249 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2250 const skb_frag_t *frag,
2251 size_t offset, size_t size,
2252 enum dma_data_direction dir)
2254 return dma_map_page(dev, skb_frag_page(frag),
2255 frag->page_offset + offset, size, dir);
2258 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2261 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2265 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2268 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2273 * skb_clone_writable - is the header of a clone writable
2274 * @skb: buffer to check
2275 * @len: length up to which to write
2277 * Returns true if modifying the header part of the cloned buffer
2278 * does not requires the data to be copied.
2280 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2282 return !skb_header_cloned(skb) &&
2283 skb_headroom(skb) + len <= skb->hdr_len;
2286 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2291 if (headroom > skb_headroom(skb))
2292 delta = headroom - skb_headroom(skb);
2294 if (delta || cloned)
2295 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2301 * skb_cow - copy header of skb when it is required
2302 * @skb: buffer to cow
2303 * @headroom: needed headroom
2305 * If the skb passed lacks sufficient headroom or its data part
2306 * is shared, data is reallocated. If reallocation fails, an error
2307 * is returned and original skb is not changed.
2309 * The result is skb with writable area skb->head...skb->tail
2310 * and at least @headroom of space at head.
2312 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2314 return __skb_cow(skb, headroom, skb_cloned(skb));
2318 * skb_cow_head - skb_cow but only making the head writable
2319 * @skb: buffer to cow
2320 * @headroom: needed headroom
2322 * This function is identical to skb_cow except that we replace the
2323 * skb_cloned check by skb_header_cloned. It should be used when
2324 * you only need to push on some header and do not need to modify
2327 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2329 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2333 * skb_padto - pad an skbuff up to a minimal size
2334 * @skb: buffer to pad
2335 * @len: minimal length
2337 * Pads up a buffer to ensure the trailing bytes exist and are
2338 * blanked. If the buffer already contains sufficient data it
2339 * is untouched. Otherwise it is extended. Returns zero on
2340 * success. The skb is freed on error.
2343 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2345 unsigned int size = skb->len;
2346 if (likely(size >= len))
2348 return skb_pad(skb, len - size);
2351 static inline int skb_add_data(struct sk_buff *skb,
2352 char __user *from, int copy)
2354 const int off = skb->len;
2356 if (skb->ip_summed == CHECKSUM_NONE) {
2358 __wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
2361 skb->csum = csum_block_add(skb->csum, csum, off);
2364 } else if (!copy_from_user(skb_put(skb, copy), from, copy))
2367 __skb_trim(skb, off);
2371 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2372 const struct page *page, int off)
2375 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2377 return page == skb_frag_page(frag) &&
2378 off == frag->page_offset + skb_frag_size(frag);
2383 static inline int __skb_linearize(struct sk_buff *skb)
2385 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2389 * skb_linearize - convert paged skb to linear one
2390 * @skb: buffer to linarize
2392 * If there is no free memory -ENOMEM is returned, otherwise zero
2393 * is returned and the old skb data released.
2395 static inline int skb_linearize(struct sk_buff *skb)
2397 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2401 * skb_has_shared_frag - can any frag be overwritten
2402 * @skb: buffer to test
2404 * Return true if the skb has at least one frag that might be modified
2405 * by an external entity (as in vmsplice()/sendfile())
2407 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2409 return skb_is_nonlinear(skb) &&
2410 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2414 * skb_linearize_cow - make sure skb is linear and writable
2415 * @skb: buffer to process
2417 * If there is no free memory -ENOMEM is returned, otherwise zero
2418 * is returned and the old skb data released.
2420 static inline int skb_linearize_cow(struct sk_buff *skb)
2422 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2423 __skb_linearize(skb) : 0;
2427 * skb_postpull_rcsum - update checksum for received skb after pull
2428 * @skb: buffer to update
2429 * @start: start of data before pull
2430 * @len: length of data pulled
2432 * After doing a pull on a received packet, you need to call this to
2433 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2434 * CHECKSUM_NONE so that it can be recomputed from scratch.
2437 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2438 const void *start, unsigned int len)
2440 if (skb->ip_summed == CHECKSUM_COMPLETE)
2441 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2444 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2447 * pskb_trim_rcsum - trim received skb and update checksum
2448 * @skb: buffer to trim
2451 * This is exactly the same as pskb_trim except that it ensures the
2452 * checksum of received packets are still valid after the operation.
2455 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2457 if (likely(len >= skb->len))
2459 if (skb->ip_summed == CHECKSUM_COMPLETE)
2460 skb->ip_summed = CHECKSUM_NONE;
2461 return __pskb_trim(skb, len);
2464 #define skb_queue_walk(queue, skb) \
2465 for (skb = (queue)->next; \
2466 skb != (struct sk_buff *)(queue); \
2469 #define skb_queue_walk_safe(queue, skb, tmp) \
2470 for (skb = (queue)->next, tmp = skb->next; \
2471 skb != (struct sk_buff *)(queue); \
2472 skb = tmp, tmp = skb->next)
2474 #define skb_queue_walk_from(queue, skb) \
2475 for (; skb != (struct sk_buff *)(queue); \
2478 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2479 for (tmp = skb->next; \
2480 skb != (struct sk_buff *)(queue); \
2481 skb = tmp, tmp = skb->next)
2483 #define skb_queue_reverse_walk(queue, skb) \
2484 for (skb = (queue)->prev; \
2485 skb != (struct sk_buff *)(queue); \
2488 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2489 for (skb = (queue)->prev, tmp = skb->prev; \
2490 skb != (struct sk_buff *)(queue); \
2491 skb = tmp, tmp = skb->prev)
2493 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2494 for (tmp = skb->prev; \
2495 skb != (struct sk_buff *)(queue); \
2496 skb = tmp, tmp = skb->prev)
2498 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2500 return skb_shinfo(skb)->frag_list != NULL;
2503 static inline void skb_frag_list_init(struct sk_buff *skb)
2505 skb_shinfo(skb)->frag_list = NULL;
2508 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
2510 frag->next = skb_shinfo(skb)->frag_list;
2511 skb_shinfo(skb)->frag_list = frag;
2514 #define skb_walk_frags(skb, iter) \
2515 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2517 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2518 int *peeked, int *off, int *err);
2519 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2521 unsigned int datagram_poll(struct file *file, struct socket *sock,
2522 struct poll_table_struct *wait);
2523 int skb_copy_datagram_iovec(const struct sk_buff *from, int offset,
2524 struct iovec *to, int size);
2525 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, int hlen,
2527 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
2528 const struct iovec *from, int from_offset,
2530 int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *frm,
2531 int offset, size_t count);
2532 int skb_copy_datagram_const_iovec(const struct sk_buff *from, int offset,
2533 const struct iovec *to, int to_offset,
2535 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2536 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2537 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2538 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2539 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2540 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2541 int len, __wsum csum);
2542 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
2543 struct pipe_inode_info *pipe, unsigned int len,
2544 unsigned int flags);
2545 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2546 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2547 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2549 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2550 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2551 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2552 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2553 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2554 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
2556 struct skb_checksum_ops {
2557 __wsum (*update)(const void *mem, int len, __wsum wsum);
2558 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2561 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2562 __wsum csum, const struct skb_checksum_ops *ops);
2563 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2566 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
2567 int len, void *buffer)
2569 int hlen = skb_headlen(skb);
2571 if (hlen - offset >= len)
2572 return skb->data + offset;
2574 if (skb_copy_bits(skb, offset, buffer, len) < 0)
2581 * skb_needs_linearize - check if we need to linearize a given skb
2582 * depending on the given device features.
2583 * @skb: socket buffer to check
2584 * @features: net device features
2586 * Returns true if either:
2587 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2588 * 2. skb is fragmented and the device does not support SG.
2590 static inline bool skb_needs_linearize(struct sk_buff *skb,
2591 netdev_features_t features)
2593 return skb_is_nonlinear(skb) &&
2594 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2595 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2598 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2600 const unsigned int len)
2602 memcpy(to, skb->data, len);
2605 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2606 const int offset, void *to,
2607 const unsigned int len)
2609 memcpy(to, skb->data + offset, len);
2612 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2614 const unsigned int len)
2616 memcpy(skb->data, from, len);
2619 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2622 const unsigned int len)
2624 memcpy(skb->data + offset, from, len);
2627 void skb_init(void);
2629 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2635 * skb_get_timestamp - get timestamp from a skb
2636 * @skb: skb to get stamp from
2637 * @stamp: pointer to struct timeval to store stamp in
2639 * Timestamps are stored in the skb as offsets to a base timestamp.
2640 * This function converts the offset back to a struct timeval and stores
2643 static inline void skb_get_timestamp(const struct sk_buff *skb,
2644 struct timeval *stamp)
2646 *stamp = ktime_to_timeval(skb->tstamp);
2649 static inline void skb_get_timestampns(const struct sk_buff *skb,
2650 struct timespec *stamp)
2652 *stamp = ktime_to_timespec(skb->tstamp);
2655 static inline void __net_timestamp(struct sk_buff *skb)
2657 skb->tstamp = ktime_get_real();
2660 static inline ktime_t net_timedelta(ktime_t t)
2662 return ktime_sub(ktime_get_real(), t);
2665 static inline ktime_t net_invalid_timestamp(void)
2667 return ktime_set(0, 0);
2670 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2672 void skb_clone_tx_timestamp(struct sk_buff *skb);
2673 bool skb_defer_rx_timestamp(struct sk_buff *skb);
2675 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2677 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2681 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2686 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2689 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2691 * PHY drivers may accept clones of transmitted packets for
2692 * timestamping via their phy_driver.txtstamp method. These drivers
2693 * must call this function to return the skb back to the stack, with
2694 * or without a timestamp.
2696 * @skb: clone of the the original outgoing packet
2697 * @hwtstamps: hardware time stamps, may be NULL if not available
2700 void skb_complete_tx_timestamp(struct sk_buff *skb,
2701 struct skb_shared_hwtstamps *hwtstamps);
2703 void __skb_tstamp_tx(struct sk_buff *orig_skb,
2704 struct skb_shared_hwtstamps *hwtstamps,
2705 struct sock *sk, int tstype);
2708 * skb_tstamp_tx - queue clone of skb with send time stamps
2709 * @orig_skb: the original outgoing packet
2710 * @hwtstamps: hardware time stamps, may be NULL if not available
2712 * If the skb has a socket associated, then this function clones the
2713 * skb (thus sharing the actual data and optional structures), stores
2714 * the optional hardware time stamping information (if non NULL) or
2715 * generates a software time stamp (otherwise), then queues the clone
2716 * to the error queue of the socket. Errors are silently ignored.
2718 void skb_tstamp_tx(struct sk_buff *orig_skb,
2719 struct skb_shared_hwtstamps *hwtstamps);
2721 static inline void sw_tx_timestamp(struct sk_buff *skb)
2723 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2724 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
2725 skb_tstamp_tx(skb, NULL);
2729 * skb_tx_timestamp() - Driver hook for transmit timestamping
2731 * Ethernet MAC Drivers should call this function in their hard_xmit()
2732 * function immediately before giving the sk_buff to the MAC hardware.
2734 * Specifically, one should make absolutely sure that this function is
2735 * called before TX completion of this packet can trigger. Otherwise
2736 * the packet could potentially already be freed.
2738 * @skb: A socket buffer.
2740 static inline void skb_tx_timestamp(struct sk_buff *skb)
2742 skb_clone_tx_timestamp(skb);
2743 sw_tx_timestamp(skb);
2747 * skb_complete_wifi_ack - deliver skb with wifi status
2749 * @skb: the original outgoing packet
2750 * @acked: ack status
2753 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2755 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2756 __sum16 __skb_checksum_complete(struct sk_buff *skb);
2758 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2760 return ((skb->ip_summed & CHECKSUM_UNNECESSARY) || skb->csum_valid);
2764 * skb_checksum_complete - Calculate checksum of an entire packet
2765 * @skb: packet to process
2767 * This function calculates the checksum over the entire packet plus
2768 * the value of skb->csum. The latter can be used to supply the
2769 * checksum of a pseudo header as used by TCP/UDP. It returns the
2772 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2773 * this function can be used to verify that checksum on received
2774 * packets. In that case the function should return zero if the
2775 * checksum is correct. In particular, this function will return zero
2776 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2777 * hardware has already verified the correctness of the checksum.
2779 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
2781 return skb_csum_unnecessary(skb) ?
2782 0 : __skb_checksum_complete(skb);
2785 /* Check if we need to perform checksum complete validation.
2787 * Returns true if checksum complete is needed, false otherwise
2788 * (either checksum is unnecessary or zero checksum is allowed).
2790 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
2794 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
2795 skb->csum_valid = 1;
2802 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
2805 #define CHECKSUM_BREAK 76
2807 /* Validate (init) checksum based on checksum complete.
2810 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
2811 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
2812 * checksum is stored in skb->csum for use in __skb_checksum_complete
2813 * non-zero: value of invalid checksum
2816 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
2820 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2821 if (!csum_fold(csum_add(psum, skb->csum))) {
2822 skb->csum_valid = 1;
2829 if (complete || skb->len <= CHECKSUM_BREAK) {
2832 csum = __skb_checksum_complete(skb);
2833 skb->csum_valid = !csum;
2840 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
2845 /* Perform checksum validate (init). Note that this is a macro since we only
2846 * want to calculate the pseudo header which is an input function if necessary.
2847 * First we try to validate without any computation (checksum unnecessary) and
2848 * then calculate based on checksum complete calling the function to compute
2852 * 0: checksum is validated or try to in skb_checksum_complete
2853 * non-zero: value of invalid checksum
2855 #define __skb_checksum_validate(skb, proto, complete, \
2856 zero_okay, check, compute_pseudo) \
2858 __sum16 __ret = 0; \
2859 skb->csum_valid = 0; \
2860 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
2861 __ret = __skb_checksum_validate_complete(skb, \
2862 complete, compute_pseudo(skb, proto)); \
2866 #define skb_checksum_init(skb, proto, compute_pseudo) \
2867 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
2869 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
2870 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
2872 #define skb_checksum_validate(skb, proto, compute_pseudo) \
2873 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
2875 #define skb_checksum_validate_zero_check(skb, proto, check, \
2877 __skb_checksum_validate_(skb, proto, true, true, check, compute_pseudo)
2879 #define skb_checksum_simple_validate(skb) \
2880 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
2882 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2883 void nf_conntrack_destroy(struct nf_conntrack *nfct);
2884 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
2886 if (nfct && atomic_dec_and_test(&nfct->use))
2887 nf_conntrack_destroy(nfct);
2889 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
2892 atomic_inc(&nfct->use);
2895 #ifdef CONFIG_BRIDGE_NETFILTER
2896 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
2898 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
2901 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
2904 atomic_inc(&nf_bridge->use);
2906 #endif /* CONFIG_BRIDGE_NETFILTER */
2907 static inline void nf_reset(struct sk_buff *skb)
2909 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2910 nf_conntrack_put(skb->nfct);
2913 #ifdef CONFIG_BRIDGE_NETFILTER
2914 nf_bridge_put(skb->nf_bridge);
2915 skb->nf_bridge = NULL;
2919 static inline void nf_reset_trace(struct sk_buff *skb)
2921 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
2926 /* Note: This doesn't put any conntrack and bridge info in dst. */
2927 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2929 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2930 dst->nfct = src->nfct;
2931 nf_conntrack_get(src->nfct);
2932 dst->nfctinfo = src->nfctinfo;
2934 #ifdef CONFIG_BRIDGE_NETFILTER
2935 dst->nf_bridge = src->nf_bridge;
2936 nf_bridge_get(src->nf_bridge);
2938 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
2939 dst->nf_trace = src->nf_trace;
2943 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2945 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2946 nf_conntrack_put(dst->nfct);
2948 #ifdef CONFIG_BRIDGE_NETFILTER
2949 nf_bridge_put(dst->nf_bridge);
2951 __nf_copy(dst, src);
2954 #ifdef CONFIG_NETWORK_SECMARK
2955 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2957 to->secmark = from->secmark;
2960 static inline void skb_init_secmark(struct sk_buff *skb)
2965 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2968 static inline void skb_init_secmark(struct sk_buff *skb)
2972 static inline bool skb_irq_freeable(const struct sk_buff *skb)
2974 return !skb->destructor &&
2975 #if IS_ENABLED(CONFIG_XFRM)
2978 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
2981 !skb->_skb_refdst &&
2982 !skb_has_frag_list(skb);
2985 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
2987 skb->queue_mapping = queue_mapping;
2990 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
2992 return skb->queue_mapping;
2995 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
2997 to->queue_mapping = from->queue_mapping;
3000 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3002 skb->queue_mapping = rx_queue + 1;
3005 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3007 return skb->queue_mapping - 1;
3010 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3012 return skb->queue_mapping != 0;
3015 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
3016 unsigned int num_tx_queues);
3018 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3027 /* Keeps track of mac header offset relative to skb->head.
3028 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3029 * For non-tunnel skb it points to skb_mac_header() and for
3030 * tunnel skb it points to outer mac header.
3031 * Keeps track of level of encapsulation of network headers.
3038 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3040 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3042 return (skb_mac_header(inner_skb) - inner_skb->head) -
3043 SKB_GSO_CB(inner_skb)->mac_offset;
3046 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3048 int new_headroom, headroom;
3051 headroom = skb_headroom(skb);
3052 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3056 new_headroom = skb_headroom(skb);
3057 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3061 /* Compute the checksum for a gso segment. First compute the checksum value
3062 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3063 * then add in skb->csum (checksum from csum_start to end of packet).
3064 * skb->csum and csum_start are then updated to reflect the checksum of the
3065 * resultant packet starting from the transport header-- the resultant checksum
3066 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3069 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3071 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3072 skb_transport_offset(skb);
3075 csum = csum_fold(csum_partial(skb_transport_header(skb),
3078 SKB_GSO_CB(skb)->csum_start -= plen;
3083 static inline bool skb_is_gso(const struct sk_buff *skb)
3085 return skb_shinfo(skb)->gso_size;
3088 /* Note: Should be called only if skb_is_gso(skb) is true */
3089 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3091 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3094 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3096 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3098 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3099 * wanted then gso_type will be set. */
3100 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3102 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3103 unlikely(shinfo->gso_type == 0)) {
3104 __skb_warn_lro_forwarding(skb);
3110 static inline void skb_forward_csum(struct sk_buff *skb)
3112 /* Unfortunately we don't support this one. Any brave souls? */
3113 if (skb->ip_summed == CHECKSUM_COMPLETE)
3114 skb->ip_summed = CHECKSUM_NONE;
3118 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3119 * @skb: skb to check
3121 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3122 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3123 * use this helper, to document places where we make this assertion.
3125 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3128 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3132 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3134 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3136 u32 __skb_get_poff(const struct sk_buff *skb);
3139 * skb_head_is_locked - Determine if the skb->head is locked down
3140 * @skb: skb to check
3142 * The head on skbs build around a head frag can be removed if they are
3143 * not cloned. This function returns true if the skb head is locked down
3144 * due to either being allocated via kmalloc, or by being a clone with
3145 * multiple references to the head.
3147 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3149 return !skb->head_frag || skb_cloned(skb);
3153 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3157 * skb_gso_network_seglen is used to determine the real size of the
3158 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3160 * The MAC/L2 header is not accounted for.
3162 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3164 unsigned int hdr_len = skb_transport_header(skb) -
3165 skb_network_header(skb);
3166 return hdr_len + skb_gso_transport_seglen(skb);
3168 #endif /* __KERNEL__ */
3169 #endif /* _LINUX_SKBUFF_H */