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/dmaengine.h>
32 #include <linux/hrtimer.h>
33 #include <linux/dma-mapping.h>
34 #include <linux/netdev_features.h>
35 #include <linux/sched.h>
36 #include <net/flow_keys.h>
38 /* A. Checksumming of received packets by device.
42 * Device failed to checksum this packet e.g. due to lack of capabilities.
43 * The packet contains full (though not verified) checksum in packet but
44 * not in skb->csum. Thus, skb->csum is undefined in this case.
46 * CHECKSUM_UNNECESSARY:
48 * The hardware you're dealing with doesn't calculate the full checksum
49 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
50 * for specific protocols e.g. TCP/UDP/SCTP, then, for such packets it will
51 * set CHECKSUM_UNNECESSARY if their checksums are okay. skb->csum is still
52 * undefined in this case though. It is a bad option, but, unfortunately,
53 * nowadays most vendors do this. Apparently with the secret goal to sell
54 * you new devices, when you will add new protocol to your host, f.e. IPv6 8)
58 * This is the most generic way. The device supplied checksum of the _whole_
59 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
60 * hardware doesn't need to parse L3/L4 headers to implement this.
62 * Note: Even if device supports only some protocols, but is able to produce
63 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
67 * This is identical to the case for output below. This may occur on a packet
68 * received directly from another Linux OS, e.g., a virtualized Linux kernel
69 * on the same host. The packet can be treated in the same way as
70 * CHECKSUM_UNNECESSARY, except that on output (i.e., forwarding) the
71 * checksum must be filled in by the OS or the hardware.
73 * B. Checksumming on output.
77 * The skb was already checksummed by the protocol, or a checksum is not
82 * The device is required to checksum the packet as seen by hard_start_xmit()
83 * from skb->csum_start up to the end, and to record/write the checksum at
84 * offset skb->csum_start + skb->csum_offset.
86 * The device must show its capabilities in dev->features, set up at device
87 * setup time, e.g. netdev_features.h:
89 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything.
90 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over
91 * IPv4. Sigh. Vendors like this way for an unknown reason.
92 * Though, see comment above about CHECKSUM_UNNECESSARY. 8)
93 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead.
94 * NETIF_F_... - Well, you get the picture.
96 * CHECKSUM_UNNECESSARY:
98 * Normally, the device will do per protocol specific checksumming. Protocol
99 * implementations that do not want the NIC to perform the checksum
100 * calculation should use this flag in their outgoing skbs.
102 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC
103 * offload. Correspondingly, the FCoE protocol driver
104 * stack should use CHECKSUM_UNNECESSARY.
106 * Any questions? No questions, good. --ANK
109 /* Don't change this without changing skb_csum_unnecessary! */
110 #define CHECKSUM_NONE 0
111 #define CHECKSUM_UNNECESSARY 1
112 #define CHECKSUM_COMPLETE 2
113 #define CHECKSUM_PARTIAL 3
115 #define SKB_DATA_ALIGN(X) (((X) + (SMP_CACHE_BYTES - 1)) & \
116 ~(SMP_CACHE_BYTES - 1))
117 #define SKB_WITH_OVERHEAD(X) \
118 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
119 #define SKB_MAX_ORDER(X, ORDER) \
120 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
121 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
122 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
124 /* return minimum truesize of one skb containing X bytes of data */
125 #define SKB_TRUESIZE(X) ((X) + \
126 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
127 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
131 struct pipe_inode_info;
133 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
134 struct nf_conntrack {
139 #ifdef CONFIG_BRIDGE_NETFILTER
140 struct nf_bridge_info {
143 struct net_device *physindev;
144 struct net_device *physoutdev;
145 unsigned long data[32 / sizeof(unsigned long)];
149 struct sk_buff_head {
150 /* These two members must be first. */
151 struct sk_buff *next;
152 struct sk_buff *prev;
160 /* To allow 64K frame to be packed as single skb without frag_list we
161 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
162 * buffers which do not start on a page boundary.
164 * Since GRO uses frags we allocate at least 16 regardless of page
167 #if (65536/PAGE_SIZE + 1) < 16
168 #define MAX_SKB_FRAGS 16UL
170 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
173 typedef struct skb_frag_struct skb_frag_t;
175 struct skb_frag_struct {
179 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
188 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
193 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
198 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
203 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
208 #define HAVE_HW_TIME_STAMP
211 * struct skb_shared_hwtstamps - hardware time stamps
212 * @hwtstamp: hardware time stamp transformed into duration
213 * since arbitrary point in time
214 * @syststamp: hwtstamp transformed to system time base
216 * Software time stamps generated by ktime_get_real() are stored in
217 * skb->tstamp. The relation between the different kinds of time
218 * stamps is as follows:
220 * syststamp and tstamp can be compared against each other in
221 * arbitrary combinations. The accuracy of a
222 * syststamp/tstamp/"syststamp from other device" comparison is
223 * limited by the accuracy of the transformation into system time
224 * base. This depends on the device driver and its underlying
227 * hwtstamps can only be compared against other hwtstamps from
230 * This structure is attached to packets as part of the
231 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
233 struct skb_shared_hwtstamps {
238 /* Definitions for tx_flags in struct skb_shared_info */
240 /* generate hardware time stamp */
241 SKBTX_HW_TSTAMP = 1 << 0,
243 /* generate software time stamp */
244 SKBTX_SW_TSTAMP = 1 << 1,
246 /* device driver is going to provide hardware time stamp */
247 SKBTX_IN_PROGRESS = 1 << 2,
249 /* device driver supports TX zero-copy buffers */
250 SKBTX_DEV_ZEROCOPY = 1 << 3,
252 /* generate wifi status information (where possible) */
253 SKBTX_WIFI_STATUS = 1 << 4,
255 /* This indicates at least one fragment might be overwritten
256 * (as in vmsplice(), sendfile() ...)
257 * If we need to compute a TX checksum, we'll need to copy
258 * all frags to avoid possible bad checksum
260 SKBTX_SHARED_FRAG = 1 << 5,
264 * The callback notifies userspace to release buffers when skb DMA is done in
265 * lower device, the skb last reference should be 0 when calling this.
266 * The zerocopy_success argument is true if zero copy transmit occurred,
267 * false on data copy or out of memory error caused by data copy attempt.
268 * The ctx field is used to track device context.
269 * The desc field is used to track userspace buffer index.
272 void (*callback)(struct ubuf_info *, bool zerocopy_success);
277 /* This data is invariant across clones and lives at
278 * the end of the header data, ie. at skb->end.
280 struct skb_shared_info {
281 unsigned char nr_frags;
283 unsigned short gso_size;
284 /* Warning: this field is not always filled in (UFO)! */
285 unsigned short gso_segs;
286 unsigned short gso_type;
287 struct sk_buff *frag_list;
288 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 #if defined CONFIG_NET_DMA || defined CONFIG_NET_RX_BUSY_POLL
585 unsigned int napi_id;
586 dma_cookie_t dma_cookie;
589 #ifdef CONFIG_NETWORK_SECMARK
595 __u32 reserved_tailroom;
598 __be16 inner_protocol;
599 __u16 inner_transport_header;
600 __u16 inner_network_header;
601 __u16 inner_mac_header;
602 __u16 transport_header;
603 __u16 network_header;
605 /* These elements must be at the end, see alloc_skb() for details. */
610 unsigned int truesize;
616 * Handling routines are only of interest to the kernel
618 #include <linux/slab.h>
621 #define SKB_ALLOC_FCLONE 0x01
622 #define SKB_ALLOC_RX 0x02
624 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
625 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
627 return unlikely(skb->pfmemalloc);
631 * skb might have a dst pointer attached, refcounted or not.
632 * _skb_refdst low order bit is set if refcount was _not_ taken
634 #define SKB_DST_NOREF 1UL
635 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
638 * skb_dst - returns skb dst_entry
641 * Returns skb dst_entry, regardless of reference taken or not.
643 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
645 /* If refdst was not refcounted, check we still are in a
646 * rcu_read_lock section
648 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
649 !rcu_read_lock_held() &&
650 !rcu_read_lock_bh_held());
651 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
655 * skb_dst_set - sets skb dst
659 * Sets skb dst, assuming a reference was taken on dst and should
660 * be released by skb_dst_drop()
662 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
664 skb->_skb_refdst = (unsigned long)dst;
667 void __skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst,
671 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
675 * Sets skb dst, assuming a reference was not taken on dst.
676 * If dst entry is cached, we do not take reference and dst_release
677 * will be avoided by refdst_drop. If dst entry is not cached, we take
678 * reference, so that last dst_release can destroy the dst immediately.
680 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
682 __skb_dst_set_noref(skb, dst, false);
686 * skb_dst_set_noref_force - sets skb dst, without taking reference
690 * Sets skb dst, assuming a reference was not taken on dst.
691 * No reference is taken and no dst_release will be called. While for
692 * cached dsts deferred reclaim is a basic feature, for entries that are
693 * not cached it is caller's job to guarantee that last dst_release for
694 * provided dst happens when nobody uses it, eg. after a RCU grace period.
696 static inline void skb_dst_set_noref_force(struct sk_buff *skb,
697 struct dst_entry *dst)
699 __skb_dst_set_noref(skb, dst, true);
703 * skb_dst_is_noref - Test if skb dst isn't refcounted
706 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
708 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
711 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
713 return (struct rtable *)skb_dst(skb);
716 void kfree_skb(struct sk_buff *skb);
717 void kfree_skb_list(struct sk_buff *segs);
718 void skb_tx_error(struct sk_buff *skb);
719 void consume_skb(struct sk_buff *skb);
720 void __kfree_skb(struct sk_buff *skb);
721 extern struct kmem_cache *skbuff_head_cache;
723 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
724 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
725 bool *fragstolen, int *delta_truesize);
727 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
729 struct sk_buff *build_skb(void *data, unsigned int frag_size);
730 static inline struct sk_buff *alloc_skb(unsigned int size,
733 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
736 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
739 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
742 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
743 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
745 return __alloc_skb_head(priority, -1);
748 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
749 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
750 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
751 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
752 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
753 gfp_t gfp_mask, bool fclone);
754 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
757 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
760 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
761 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
762 unsigned int headroom);
763 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
764 int newtailroom, gfp_t priority);
765 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
766 int offset, int len);
767 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
769 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
770 int skb_pad(struct sk_buff *skb, int pad);
771 #define dev_kfree_skb(a) consume_skb(a)
773 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
774 int getfrag(void *from, char *to, int offset,
775 int len, int odd, struct sk_buff *skb),
776 void *from, int length);
778 struct skb_seq_state {
782 __u32 stepped_offset;
783 struct sk_buff *root_skb;
784 struct sk_buff *cur_skb;
788 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
789 unsigned int to, struct skb_seq_state *st);
790 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
791 struct skb_seq_state *st);
792 void skb_abort_seq_read(struct skb_seq_state *st);
794 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
795 unsigned int to, struct ts_config *config,
796 struct ts_state *state);
799 * Packet hash types specify the type of hash in skb_set_hash.
801 * Hash types refer to the protocol layer addresses which are used to
802 * construct a packet's hash. The hashes are used to differentiate or identify
803 * flows of the protocol layer for the hash type. Hash types are either
804 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
806 * Properties of hashes:
808 * 1) Two packets in different flows have different hash values
809 * 2) Two packets in the same flow should have the same hash value
811 * A hash at a higher layer is considered to be more specific. A driver should
812 * set the most specific hash possible.
814 * A driver cannot indicate a more specific hash than the layer at which a hash
815 * was computed. For instance an L3 hash cannot be set as an L4 hash.
817 * A driver may indicate a hash level which is less specific than the
818 * actual layer the hash was computed on. For instance, a hash computed
819 * at L4 may be considered an L3 hash. This should only be done if the
820 * driver can't unambiguously determine that the HW computed the hash at
821 * the higher layer. Note that the "should" in the second property above
824 enum pkt_hash_types {
825 PKT_HASH_TYPE_NONE, /* Undefined type */
826 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
827 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
828 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
832 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
834 skb->l4_hash = (type == PKT_HASH_TYPE_L4);
839 void __skb_get_hash(struct sk_buff *skb);
840 static inline __u32 skb_get_hash(struct sk_buff *skb)
842 if (!skb->l4_hash && !skb->sw_hash)
848 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
853 static inline void skb_clear_hash(struct sk_buff *skb)
860 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
866 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
868 to->hash = from->hash;
869 to->sw_hash = from->sw_hash;
870 to->l4_hash = from->l4_hash;
873 #ifdef NET_SKBUFF_DATA_USES_OFFSET
874 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
876 return skb->head + skb->end;
879 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
884 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
889 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
891 return skb->end - skb->head;
896 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
898 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
900 return &skb_shinfo(skb)->hwtstamps;
904 * skb_queue_empty - check if a queue is empty
907 * Returns true if the queue is empty, false otherwise.
909 static inline int skb_queue_empty(const struct sk_buff_head *list)
911 return list->next == (const struct sk_buff *) list;
915 * skb_queue_is_last - check if skb is the last entry in the queue
919 * Returns true if @skb is the last buffer on the list.
921 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
922 const struct sk_buff *skb)
924 return skb->next == (const struct sk_buff *) list;
928 * skb_queue_is_first - check if skb is the first entry in the queue
932 * Returns true if @skb is the first buffer on the list.
934 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
935 const struct sk_buff *skb)
937 return skb->prev == (const struct sk_buff *) list;
941 * skb_queue_next - return the next packet in the queue
943 * @skb: current buffer
945 * Return the next packet in @list after @skb. It is only valid to
946 * call this if skb_queue_is_last() evaluates to false.
948 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
949 const struct sk_buff *skb)
951 /* This BUG_ON may seem severe, but if we just return then we
952 * are going to dereference garbage.
954 BUG_ON(skb_queue_is_last(list, skb));
959 * skb_queue_prev - return the prev packet in the queue
961 * @skb: current buffer
963 * Return the prev packet in @list before @skb. It is only valid to
964 * call this if skb_queue_is_first() evaluates to false.
966 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
967 const struct sk_buff *skb)
969 /* This BUG_ON may seem severe, but if we just return then we
970 * are going to dereference garbage.
972 BUG_ON(skb_queue_is_first(list, skb));
977 * skb_get - reference buffer
978 * @skb: buffer to reference
980 * Makes another reference to a socket buffer and returns a pointer
983 static inline struct sk_buff *skb_get(struct sk_buff *skb)
985 atomic_inc(&skb->users);
990 * If users == 1, we are the only owner and are can avoid redundant
995 * skb_cloned - is the buffer a clone
996 * @skb: buffer to check
998 * Returns true if the buffer was generated with skb_clone() and is
999 * one of multiple shared copies of the buffer. Cloned buffers are
1000 * shared data so must not be written to under normal circumstances.
1002 static inline int skb_cloned(const struct sk_buff *skb)
1004 return skb->cloned &&
1005 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1008 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1010 might_sleep_if(pri & __GFP_WAIT);
1012 if (skb_cloned(skb))
1013 return pskb_expand_head(skb, 0, 0, pri);
1019 * skb_header_cloned - is the header a clone
1020 * @skb: buffer to check
1022 * Returns true if modifying the header part of the buffer requires
1023 * the data to be copied.
1025 static inline int skb_header_cloned(const struct sk_buff *skb)
1032 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1033 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1034 return dataref != 1;
1038 * skb_header_release - release reference to header
1039 * @skb: buffer to operate on
1041 * Drop a reference to the header part of the buffer. This is done
1042 * by acquiring a payload reference. You must not read from the header
1043 * part of skb->data after this.
1045 static inline void skb_header_release(struct sk_buff *skb)
1049 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1053 * skb_shared - is the buffer shared
1054 * @skb: buffer to check
1056 * Returns true if more than one person has a reference to this
1059 static inline int skb_shared(const struct sk_buff *skb)
1061 return atomic_read(&skb->users) != 1;
1065 * skb_share_check - check if buffer is shared and if so clone it
1066 * @skb: buffer to check
1067 * @pri: priority for memory allocation
1069 * If the buffer is shared the buffer is cloned and the old copy
1070 * drops a reference. A new clone with a single reference is returned.
1071 * If the buffer is not shared the original buffer is returned. When
1072 * being called from interrupt status or with spinlocks held pri must
1075 * NULL is returned on a memory allocation failure.
1077 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1079 might_sleep_if(pri & __GFP_WAIT);
1080 if (skb_shared(skb)) {
1081 struct sk_buff *nskb = skb_clone(skb, pri);
1093 * Copy shared buffers into a new sk_buff. We effectively do COW on
1094 * packets to handle cases where we have a local reader and forward
1095 * and a couple of other messy ones. The normal one is tcpdumping
1096 * a packet thats being forwarded.
1100 * skb_unshare - make a copy of a shared buffer
1101 * @skb: buffer to check
1102 * @pri: priority for memory allocation
1104 * If the socket buffer is a clone then this function creates a new
1105 * copy of the data, drops a reference count on the old copy and returns
1106 * the new copy with the reference count at 1. If the buffer is not a clone
1107 * the original buffer is returned. When called with a spinlock held or
1108 * from interrupt state @pri must be %GFP_ATOMIC
1110 * %NULL is returned on a memory allocation failure.
1112 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1115 might_sleep_if(pri & __GFP_WAIT);
1116 if (skb_cloned(skb)) {
1117 struct sk_buff *nskb = skb_copy(skb, pri);
1118 kfree_skb(skb); /* Free our shared copy */
1125 * skb_peek - peek at the head of an &sk_buff_head
1126 * @list_: list to peek at
1128 * Peek an &sk_buff. Unlike most other operations you _MUST_
1129 * be careful with this one. A peek leaves the buffer on the
1130 * list and someone else may run off with it. You must hold
1131 * the appropriate locks or have a private queue to do this.
1133 * Returns %NULL for an empty list or a pointer to the head element.
1134 * The reference count is not incremented and the reference is therefore
1135 * volatile. Use with caution.
1137 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1139 struct sk_buff *skb = list_->next;
1141 if (skb == (struct sk_buff *)list_)
1147 * skb_peek_next - peek skb following the given one from a queue
1148 * @skb: skb to start from
1149 * @list_: list to peek at
1151 * Returns %NULL when the end of the list is met or a pointer to the
1152 * next element. The reference count is not incremented and the
1153 * reference is therefore volatile. Use with caution.
1155 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1156 const struct sk_buff_head *list_)
1158 struct sk_buff *next = skb->next;
1160 if (next == (struct sk_buff *)list_)
1166 * skb_peek_tail - peek at the tail of an &sk_buff_head
1167 * @list_: list to peek at
1169 * Peek an &sk_buff. Unlike most other operations you _MUST_
1170 * be careful with this one. A peek leaves the buffer on the
1171 * list and someone else may run off with it. You must hold
1172 * the appropriate locks or have a private queue to do this.
1174 * Returns %NULL for an empty list or a pointer to the tail element.
1175 * The reference count is not incremented and the reference is therefore
1176 * volatile. Use with caution.
1178 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1180 struct sk_buff *skb = list_->prev;
1182 if (skb == (struct sk_buff *)list_)
1189 * skb_queue_len - get queue length
1190 * @list_: list to measure
1192 * Return the length of an &sk_buff queue.
1194 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1200 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1201 * @list: queue to initialize
1203 * This initializes only the list and queue length aspects of
1204 * an sk_buff_head object. This allows to initialize the list
1205 * aspects of an sk_buff_head without reinitializing things like
1206 * the spinlock. It can also be used for on-stack sk_buff_head
1207 * objects where the spinlock is known to not be used.
1209 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1211 list->prev = list->next = (struct sk_buff *)list;
1216 * This function creates a split out lock class for each invocation;
1217 * this is needed for now since a whole lot of users of the skb-queue
1218 * infrastructure in drivers have different locking usage (in hardirq)
1219 * than the networking core (in softirq only). In the long run either the
1220 * network layer or drivers should need annotation to consolidate the
1221 * main types of usage into 3 classes.
1223 static inline void skb_queue_head_init(struct sk_buff_head *list)
1225 spin_lock_init(&list->lock);
1226 __skb_queue_head_init(list);
1229 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1230 struct lock_class_key *class)
1232 skb_queue_head_init(list);
1233 lockdep_set_class(&list->lock, class);
1237 * Insert an sk_buff on a list.
1239 * The "__skb_xxxx()" functions are the non-atomic ones that
1240 * can only be called with interrupts disabled.
1242 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1243 struct sk_buff_head *list);
1244 static inline void __skb_insert(struct sk_buff *newsk,
1245 struct sk_buff *prev, struct sk_buff *next,
1246 struct sk_buff_head *list)
1250 next->prev = prev->next = newsk;
1254 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1255 struct sk_buff *prev,
1256 struct sk_buff *next)
1258 struct sk_buff *first = list->next;
1259 struct sk_buff *last = list->prev;
1269 * skb_queue_splice - join two skb lists, this is designed for stacks
1270 * @list: the new list to add
1271 * @head: the place to add it in the first list
1273 static inline void skb_queue_splice(const struct sk_buff_head *list,
1274 struct sk_buff_head *head)
1276 if (!skb_queue_empty(list)) {
1277 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1278 head->qlen += list->qlen;
1283 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1284 * @list: the new list to add
1285 * @head: the place to add it in the first list
1287 * The list at @list is reinitialised
1289 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1290 struct sk_buff_head *head)
1292 if (!skb_queue_empty(list)) {
1293 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1294 head->qlen += list->qlen;
1295 __skb_queue_head_init(list);
1300 * skb_queue_splice_tail - join two skb lists, each list being a queue
1301 * @list: the new list to add
1302 * @head: the place to add it in the first list
1304 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1305 struct sk_buff_head *head)
1307 if (!skb_queue_empty(list)) {
1308 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1309 head->qlen += list->qlen;
1314 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1315 * @list: the new list to add
1316 * @head: the place to add it in the first list
1318 * Each of the lists is a queue.
1319 * The list at @list is reinitialised
1321 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1322 struct sk_buff_head *head)
1324 if (!skb_queue_empty(list)) {
1325 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1326 head->qlen += list->qlen;
1327 __skb_queue_head_init(list);
1332 * __skb_queue_after - queue a buffer at the list head
1333 * @list: list to use
1334 * @prev: place after this buffer
1335 * @newsk: buffer to queue
1337 * Queue a buffer int the middle of a list. This function takes no locks
1338 * and you must therefore hold required locks before calling it.
1340 * A buffer cannot be placed on two lists at the same time.
1342 static inline void __skb_queue_after(struct sk_buff_head *list,
1343 struct sk_buff *prev,
1344 struct sk_buff *newsk)
1346 __skb_insert(newsk, prev, prev->next, list);
1349 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1350 struct sk_buff_head *list);
1352 static inline void __skb_queue_before(struct sk_buff_head *list,
1353 struct sk_buff *next,
1354 struct sk_buff *newsk)
1356 __skb_insert(newsk, next->prev, next, list);
1360 * __skb_queue_head - queue a buffer at the list head
1361 * @list: list to use
1362 * @newsk: buffer to queue
1364 * Queue a buffer at the start of a list. This function takes no locks
1365 * and you must therefore hold required locks before calling it.
1367 * A buffer cannot be placed on two lists at the same time.
1369 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1370 static inline void __skb_queue_head(struct sk_buff_head *list,
1371 struct sk_buff *newsk)
1373 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1377 * __skb_queue_tail - queue a buffer at the list tail
1378 * @list: list to use
1379 * @newsk: buffer to queue
1381 * Queue a buffer at the end of a list. This function takes no locks
1382 * and you must therefore hold required locks before calling it.
1384 * A buffer cannot be placed on two lists at the same time.
1386 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1387 static inline void __skb_queue_tail(struct sk_buff_head *list,
1388 struct sk_buff *newsk)
1390 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1394 * remove sk_buff from list. _Must_ be called atomically, and with
1397 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1398 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1400 struct sk_buff *next, *prev;
1405 skb->next = skb->prev = NULL;
1411 * __skb_dequeue - remove from the head of the queue
1412 * @list: list to dequeue from
1414 * Remove the head of the list. This function does not take any locks
1415 * so must be used with appropriate locks held only. The head item is
1416 * returned or %NULL if the list is empty.
1418 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1419 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1421 struct sk_buff *skb = skb_peek(list);
1423 __skb_unlink(skb, list);
1428 * __skb_dequeue_tail - remove from the tail of the queue
1429 * @list: list to dequeue from
1431 * Remove the tail of the list. This function does not take any locks
1432 * so must be used with appropriate locks held only. The tail item is
1433 * returned or %NULL if the list is empty.
1435 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1436 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1438 struct sk_buff *skb = skb_peek_tail(list);
1440 __skb_unlink(skb, list);
1445 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1447 return skb->data_len;
1450 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1452 return skb->len - skb->data_len;
1455 static inline int skb_pagelen(const struct sk_buff *skb)
1459 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1460 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1461 return len + skb_headlen(skb);
1465 * __skb_fill_page_desc - initialise a paged fragment in an skb
1466 * @skb: buffer containing fragment to be initialised
1467 * @i: paged fragment index to initialise
1468 * @page: the page to use for this fragment
1469 * @off: the offset to the data with @page
1470 * @size: the length of the data
1472 * Initialises the @i'th fragment of @skb to point to &size bytes at
1473 * offset @off within @page.
1475 * Does not take any additional reference on the fragment.
1477 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1478 struct page *page, int off, int size)
1480 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1483 * Propagate page->pfmemalloc to the skb if we can. The problem is
1484 * that not all callers have unique ownership of the page. If
1485 * pfmemalloc is set, we check the mapping as a mapping implies
1486 * page->index is set (index and pfmemalloc share space).
1487 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1488 * do not lose pfmemalloc information as the pages would not be
1489 * allocated using __GFP_MEMALLOC.
1491 frag->page.p = page;
1492 frag->page_offset = off;
1493 skb_frag_size_set(frag, size);
1495 page = compound_head(page);
1496 if (page->pfmemalloc && !page->mapping)
1497 skb->pfmemalloc = true;
1501 * skb_fill_page_desc - initialise a paged fragment in an skb
1502 * @skb: buffer containing fragment to be initialised
1503 * @i: paged fragment index to initialise
1504 * @page: the page to use for this fragment
1505 * @off: the offset to the data with @page
1506 * @size: the length of the data
1508 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1509 * @skb to point to @size bytes at offset @off within @page. In
1510 * addition updates @skb such that @i is the last fragment.
1512 * Does not take any additional reference on the fragment.
1514 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1515 struct page *page, int off, int size)
1517 __skb_fill_page_desc(skb, i, page, off, size);
1518 skb_shinfo(skb)->nr_frags = i + 1;
1521 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1522 int size, unsigned int truesize);
1524 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1525 unsigned int truesize);
1527 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1528 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1529 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1531 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1532 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1534 return skb->head + skb->tail;
1537 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1539 skb->tail = skb->data - skb->head;
1542 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1544 skb_reset_tail_pointer(skb);
1545 skb->tail += offset;
1548 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1549 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1554 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1556 skb->tail = skb->data;
1559 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1561 skb->tail = skb->data + offset;
1564 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1567 * Add data to an sk_buff
1569 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1570 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1571 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1573 unsigned char *tmp = skb_tail_pointer(skb);
1574 SKB_LINEAR_ASSERT(skb);
1580 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1581 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1588 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1589 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1592 BUG_ON(skb->len < skb->data_len);
1593 return skb->data += len;
1596 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1598 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1601 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1603 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1605 if (len > skb_headlen(skb) &&
1606 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1609 return skb->data += len;
1612 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1614 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1617 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1619 if (likely(len <= skb_headlen(skb)))
1621 if (unlikely(len > skb->len))
1623 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1627 * skb_headroom - bytes at buffer head
1628 * @skb: buffer to check
1630 * Return the number of bytes of free space at the head of an &sk_buff.
1632 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1634 return skb->data - skb->head;
1638 * skb_tailroom - bytes at buffer end
1639 * @skb: buffer to check
1641 * Return the number of bytes of free space at the tail of an sk_buff
1643 static inline int skb_tailroom(const struct sk_buff *skb)
1645 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1649 * skb_availroom - bytes at buffer end
1650 * @skb: buffer to check
1652 * Return the number of bytes of free space at the tail of an sk_buff
1653 * allocated by sk_stream_alloc()
1655 static inline int skb_availroom(const struct sk_buff *skb)
1657 if (skb_is_nonlinear(skb))
1660 return skb->end - skb->tail - skb->reserved_tailroom;
1664 * skb_reserve - adjust headroom
1665 * @skb: buffer to alter
1666 * @len: bytes to move
1668 * Increase the headroom of an empty &sk_buff by reducing the tail
1669 * room. This is only allowed for an empty buffer.
1671 static inline void skb_reserve(struct sk_buff *skb, int len)
1677 static inline void skb_reset_inner_headers(struct sk_buff *skb)
1679 skb->inner_mac_header = skb->mac_header;
1680 skb->inner_network_header = skb->network_header;
1681 skb->inner_transport_header = skb->transport_header;
1684 static inline void skb_reset_mac_len(struct sk_buff *skb)
1686 skb->mac_len = skb->network_header - skb->mac_header;
1689 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1692 return skb->head + skb->inner_transport_header;
1695 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1697 skb->inner_transport_header = skb->data - skb->head;
1700 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1703 skb_reset_inner_transport_header(skb);
1704 skb->inner_transport_header += offset;
1707 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1709 return skb->head + skb->inner_network_header;
1712 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1714 skb->inner_network_header = skb->data - skb->head;
1717 static inline void skb_set_inner_network_header(struct sk_buff *skb,
1720 skb_reset_inner_network_header(skb);
1721 skb->inner_network_header += offset;
1724 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1726 return skb->head + skb->inner_mac_header;
1729 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1731 skb->inner_mac_header = skb->data - skb->head;
1734 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1737 skb_reset_inner_mac_header(skb);
1738 skb->inner_mac_header += offset;
1740 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1742 return skb->transport_header != (typeof(skb->transport_header))~0U;
1745 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1747 return skb->head + skb->transport_header;
1750 static inline void skb_reset_transport_header(struct sk_buff *skb)
1752 skb->transport_header = skb->data - skb->head;
1755 static inline void skb_set_transport_header(struct sk_buff *skb,
1758 skb_reset_transport_header(skb);
1759 skb->transport_header += offset;
1762 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1764 return skb->head + skb->network_header;
1767 static inline void skb_reset_network_header(struct sk_buff *skb)
1769 skb->network_header = skb->data - skb->head;
1772 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1774 skb_reset_network_header(skb);
1775 skb->network_header += offset;
1778 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1780 return skb->head + skb->mac_header;
1783 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1785 return skb->mac_header != (typeof(skb->mac_header))~0U;
1788 static inline void skb_reset_mac_header(struct sk_buff *skb)
1790 skb->mac_header = skb->data - skb->head;
1793 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1795 skb_reset_mac_header(skb);
1796 skb->mac_header += offset;
1799 static inline void skb_pop_mac_header(struct sk_buff *skb)
1801 skb->mac_header = skb->network_header;
1804 static inline void skb_probe_transport_header(struct sk_buff *skb,
1805 const int offset_hint)
1807 struct flow_keys keys;
1809 if (skb_transport_header_was_set(skb))
1811 else if (skb_flow_dissect(skb, &keys))
1812 skb_set_transport_header(skb, keys.thoff);
1814 skb_set_transport_header(skb, offset_hint);
1817 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1819 if (skb_mac_header_was_set(skb)) {
1820 const unsigned char *old_mac = skb_mac_header(skb);
1822 skb_set_mac_header(skb, -skb->mac_len);
1823 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
1827 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
1829 return skb->csum_start - skb_headroom(skb);
1832 static inline int skb_transport_offset(const struct sk_buff *skb)
1834 return skb_transport_header(skb) - skb->data;
1837 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1839 return skb->transport_header - skb->network_header;
1842 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
1844 return skb->inner_transport_header - skb->inner_network_header;
1847 static inline int skb_network_offset(const struct sk_buff *skb)
1849 return skb_network_header(skb) - skb->data;
1852 static inline int skb_inner_network_offset(const struct sk_buff *skb)
1854 return skb_inner_network_header(skb) - skb->data;
1857 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1859 return pskb_may_pull(skb, skb_network_offset(skb) + len);
1862 static inline void skb_pop_rcv_encapsulation(struct sk_buff *skb)
1864 /* Only continue with checksum unnecessary if device indicated
1865 * it is valid across encapsulation (skb->encapsulation was set).
1867 if (skb->ip_summed == CHECKSUM_UNNECESSARY && !skb->encapsulation)
1868 skb->ip_summed = CHECKSUM_NONE;
1870 skb->encapsulation = 0;
1871 skb->csum_valid = 0;
1875 * CPUs often take a performance hit when accessing unaligned memory
1876 * locations. The actual performance hit varies, it can be small if the
1877 * hardware handles it or large if we have to take an exception and fix it
1880 * Since an ethernet header is 14 bytes network drivers often end up with
1881 * the IP header at an unaligned offset. The IP header can be aligned by
1882 * shifting the start of the packet by 2 bytes. Drivers should do this
1885 * skb_reserve(skb, NET_IP_ALIGN);
1887 * The downside to this alignment of the IP header is that the DMA is now
1888 * unaligned. On some architectures the cost of an unaligned DMA is high
1889 * and this cost outweighs the gains made by aligning the IP header.
1891 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1894 #ifndef NET_IP_ALIGN
1895 #define NET_IP_ALIGN 2
1899 * The networking layer reserves some headroom in skb data (via
1900 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
1901 * the header has to grow. In the default case, if the header has to grow
1902 * 32 bytes or less we avoid the reallocation.
1904 * Unfortunately this headroom changes the DMA alignment of the resulting
1905 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
1906 * on some architectures. An architecture can override this value,
1907 * perhaps setting it to a cacheline in size (since that will maintain
1908 * cacheline alignment of the DMA). It must be a power of 2.
1910 * Various parts of the networking layer expect at least 32 bytes of
1911 * headroom, you should not reduce this.
1913 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
1914 * to reduce average number of cache lines per packet.
1915 * get_rps_cpus() for example only access one 64 bytes aligned block :
1916 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
1919 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
1922 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
1924 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
1926 if (unlikely(skb_is_nonlinear(skb))) {
1931 skb_set_tail_pointer(skb, len);
1934 void skb_trim(struct sk_buff *skb, unsigned int len);
1936 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
1939 return ___pskb_trim(skb, len);
1940 __skb_trim(skb, len);
1944 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
1946 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
1950 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
1951 * @skb: buffer to alter
1954 * This is identical to pskb_trim except that the caller knows that
1955 * the skb is not cloned so we should never get an error due to out-
1958 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
1960 int err = pskb_trim(skb, len);
1965 * skb_orphan - orphan a buffer
1966 * @skb: buffer to orphan
1968 * If a buffer currently has an owner then we call the owner's
1969 * destructor function and make the @skb unowned. The buffer continues
1970 * to exist but is no longer charged to its former owner.
1972 static inline void skb_orphan(struct sk_buff *skb)
1974 if (skb->destructor) {
1975 skb->destructor(skb);
1976 skb->destructor = NULL;
1984 * skb_orphan_frags - orphan the frags contained in a buffer
1985 * @skb: buffer to orphan frags from
1986 * @gfp_mask: allocation mask for replacement pages
1988 * For each frag in the SKB which needs a destructor (i.e. has an
1989 * owner) create a copy of that frag and release the original
1990 * page by calling the destructor.
1992 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
1994 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
1996 return skb_copy_ubufs(skb, gfp_mask);
2000 * __skb_queue_purge - empty a list
2001 * @list: list to empty
2003 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2004 * the list and one reference dropped. This function does not take the
2005 * list lock and the caller must hold the relevant locks to use it.
2007 void skb_queue_purge(struct sk_buff_head *list);
2008 static inline void __skb_queue_purge(struct sk_buff_head *list)
2010 struct sk_buff *skb;
2011 while ((skb = __skb_dequeue(list)) != NULL)
2015 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
2016 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
2017 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
2019 void *netdev_alloc_frag(unsigned int fragsz);
2021 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2025 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2026 * @dev: network device to receive on
2027 * @length: length to allocate
2029 * Allocate a new &sk_buff and assign it a usage count of one. The
2030 * buffer has unspecified headroom built in. Users should allocate
2031 * the headroom they think they need without accounting for the
2032 * built in space. The built in space is used for optimisations.
2034 * %NULL is returned if there is no free memory. Although this function
2035 * allocates memory it can be called from an interrupt.
2037 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2038 unsigned int length)
2040 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2043 /* legacy helper around __netdev_alloc_skb() */
2044 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2047 return __netdev_alloc_skb(NULL, length, gfp_mask);
2050 /* legacy helper around netdev_alloc_skb() */
2051 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2053 return netdev_alloc_skb(NULL, length);
2057 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2058 unsigned int length, gfp_t gfp)
2060 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2062 if (NET_IP_ALIGN && skb)
2063 skb_reserve(skb, NET_IP_ALIGN);
2067 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2068 unsigned int length)
2070 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2074 * __skb_alloc_pages - allocate pages for ps-rx on a skb and preserve pfmemalloc data
2075 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2076 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2077 * @order: size of the allocation
2079 * Allocate a new page.
2081 * %NULL is returned if there is no free memory.
2083 static inline struct page *__skb_alloc_pages(gfp_t gfp_mask,
2084 struct sk_buff *skb,
2089 gfp_mask |= __GFP_COLD;
2091 if (!(gfp_mask & __GFP_NOMEMALLOC))
2092 gfp_mask |= __GFP_MEMALLOC;
2094 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2095 if (skb && page && page->pfmemalloc)
2096 skb->pfmemalloc = true;
2102 * __skb_alloc_page - allocate a page for ps-rx for a given skb and preserve pfmemalloc data
2103 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2104 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2106 * Allocate a new page.
2108 * %NULL is returned if there is no free memory.
2110 static inline struct page *__skb_alloc_page(gfp_t gfp_mask,
2111 struct sk_buff *skb)
2113 return __skb_alloc_pages(gfp_mask, skb, 0);
2117 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2118 * @page: The page that was allocated from skb_alloc_page
2119 * @skb: The skb that may need pfmemalloc set
2121 static inline void skb_propagate_pfmemalloc(struct page *page,
2122 struct sk_buff *skb)
2124 if (page && page->pfmemalloc)
2125 skb->pfmemalloc = true;
2129 * skb_frag_page - retrieve the page referred to by a paged fragment
2130 * @frag: the paged fragment
2132 * Returns the &struct page associated with @frag.
2134 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2136 return frag->page.p;
2140 * __skb_frag_ref - take an addition reference on a paged fragment.
2141 * @frag: the paged fragment
2143 * Takes an additional reference on the paged fragment @frag.
2145 static inline void __skb_frag_ref(skb_frag_t *frag)
2147 get_page(skb_frag_page(frag));
2151 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2153 * @f: the fragment offset.
2155 * Takes an additional reference on the @f'th paged fragment of @skb.
2157 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2159 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2163 * __skb_frag_unref - release a reference on a paged fragment.
2164 * @frag: the paged fragment
2166 * Releases a reference on the paged fragment @frag.
2168 static inline void __skb_frag_unref(skb_frag_t *frag)
2170 put_page(skb_frag_page(frag));
2174 * skb_frag_unref - release a reference on a paged fragment of an skb.
2176 * @f: the fragment offset
2178 * Releases a reference on the @f'th paged fragment of @skb.
2180 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2182 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2186 * skb_frag_address - gets the address of the data contained in a paged fragment
2187 * @frag: the paged fragment buffer
2189 * Returns the address of the data within @frag. The page must already
2192 static inline void *skb_frag_address(const skb_frag_t *frag)
2194 return page_address(skb_frag_page(frag)) + frag->page_offset;
2198 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2199 * @frag: the paged fragment buffer
2201 * Returns the address of the data within @frag. Checks that the page
2202 * is mapped and returns %NULL otherwise.
2204 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2206 void *ptr = page_address(skb_frag_page(frag));
2210 return ptr + frag->page_offset;
2214 * __skb_frag_set_page - sets the page contained in a paged fragment
2215 * @frag: the paged fragment
2216 * @page: the page to set
2218 * Sets the fragment @frag to contain @page.
2220 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2222 frag->page.p = page;
2226 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2228 * @f: the fragment offset
2229 * @page: the page to set
2231 * Sets the @f'th fragment of @skb to contain @page.
2233 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2236 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2239 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2242 * skb_frag_dma_map - maps a paged fragment via the DMA API
2243 * @dev: the device to map the fragment to
2244 * @frag: the paged fragment to map
2245 * @offset: the offset within the fragment (starting at the
2246 * fragment's own offset)
2247 * @size: the number of bytes to map
2248 * @dir: the direction of the mapping (%PCI_DMA_*)
2250 * Maps the page associated with @frag to @device.
2252 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2253 const skb_frag_t *frag,
2254 size_t offset, size_t size,
2255 enum dma_data_direction dir)
2257 return dma_map_page(dev, skb_frag_page(frag),
2258 frag->page_offset + offset, size, dir);
2261 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2264 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2268 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2271 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2276 * skb_clone_writable - is the header of a clone writable
2277 * @skb: buffer to check
2278 * @len: length up to which to write
2280 * Returns true if modifying the header part of the cloned buffer
2281 * does not requires the data to be copied.
2283 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2285 return !skb_header_cloned(skb) &&
2286 skb_headroom(skb) + len <= skb->hdr_len;
2289 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2294 if (headroom > skb_headroom(skb))
2295 delta = headroom - skb_headroom(skb);
2297 if (delta || cloned)
2298 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2304 * skb_cow - copy header of skb when it is required
2305 * @skb: buffer to cow
2306 * @headroom: needed headroom
2308 * If the skb passed lacks sufficient headroom or its data part
2309 * is shared, data is reallocated. If reallocation fails, an error
2310 * is returned and original skb is not changed.
2312 * The result is skb with writable area skb->head...skb->tail
2313 * and at least @headroom of space at head.
2315 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2317 return __skb_cow(skb, headroom, skb_cloned(skb));
2321 * skb_cow_head - skb_cow but only making the head writable
2322 * @skb: buffer to cow
2323 * @headroom: needed headroom
2325 * This function is identical to skb_cow except that we replace the
2326 * skb_cloned check by skb_header_cloned. It should be used when
2327 * you only need to push on some header and do not need to modify
2330 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2332 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2336 * skb_padto - pad an skbuff up to a minimal size
2337 * @skb: buffer to pad
2338 * @len: minimal length
2340 * Pads up a buffer to ensure the trailing bytes exist and are
2341 * blanked. If the buffer already contains sufficient data it
2342 * is untouched. Otherwise it is extended. Returns zero on
2343 * success. The skb is freed on error.
2346 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2348 unsigned int size = skb->len;
2349 if (likely(size >= len))
2351 return skb_pad(skb, len - size);
2354 static inline int skb_add_data(struct sk_buff *skb,
2355 char __user *from, int copy)
2357 const int off = skb->len;
2359 if (skb->ip_summed == CHECKSUM_NONE) {
2361 __wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
2364 skb->csum = csum_block_add(skb->csum, csum, off);
2367 } else if (!copy_from_user(skb_put(skb, copy), from, copy))
2370 __skb_trim(skb, off);
2374 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2375 const struct page *page, int off)
2378 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2380 return page == skb_frag_page(frag) &&
2381 off == frag->page_offset + skb_frag_size(frag);
2386 static inline int __skb_linearize(struct sk_buff *skb)
2388 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2392 * skb_linearize - convert paged skb to linear one
2393 * @skb: buffer to linarize
2395 * If there is no free memory -ENOMEM is returned, otherwise zero
2396 * is returned and the old skb data released.
2398 static inline int skb_linearize(struct sk_buff *skb)
2400 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2404 * skb_has_shared_frag - can any frag be overwritten
2405 * @skb: buffer to test
2407 * Return true if the skb has at least one frag that might be modified
2408 * by an external entity (as in vmsplice()/sendfile())
2410 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2412 return skb_is_nonlinear(skb) &&
2413 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2417 * skb_linearize_cow - make sure skb is linear and writable
2418 * @skb: buffer to process
2420 * If there is no free memory -ENOMEM is returned, otherwise zero
2421 * is returned and the old skb data released.
2423 static inline int skb_linearize_cow(struct sk_buff *skb)
2425 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2426 __skb_linearize(skb) : 0;
2430 * skb_postpull_rcsum - update checksum for received skb after pull
2431 * @skb: buffer to update
2432 * @start: start of data before pull
2433 * @len: length of data pulled
2435 * After doing a pull on a received packet, you need to call this to
2436 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2437 * CHECKSUM_NONE so that it can be recomputed from scratch.
2440 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2441 const void *start, unsigned int len)
2443 if (skb->ip_summed == CHECKSUM_COMPLETE)
2444 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2447 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2450 * pskb_trim_rcsum - trim received skb and update checksum
2451 * @skb: buffer to trim
2454 * This is exactly the same as pskb_trim except that it ensures the
2455 * checksum of received packets are still valid after the operation.
2458 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2460 if (likely(len >= skb->len))
2462 if (skb->ip_summed == CHECKSUM_COMPLETE)
2463 skb->ip_summed = CHECKSUM_NONE;
2464 return __pskb_trim(skb, len);
2467 #define skb_queue_walk(queue, skb) \
2468 for (skb = (queue)->next; \
2469 skb != (struct sk_buff *)(queue); \
2472 #define skb_queue_walk_safe(queue, skb, tmp) \
2473 for (skb = (queue)->next, tmp = skb->next; \
2474 skb != (struct sk_buff *)(queue); \
2475 skb = tmp, tmp = skb->next)
2477 #define skb_queue_walk_from(queue, skb) \
2478 for (; skb != (struct sk_buff *)(queue); \
2481 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2482 for (tmp = skb->next; \
2483 skb != (struct sk_buff *)(queue); \
2484 skb = tmp, tmp = skb->next)
2486 #define skb_queue_reverse_walk(queue, skb) \
2487 for (skb = (queue)->prev; \
2488 skb != (struct sk_buff *)(queue); \
2491 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2492 for (skb = (queue)->prev, tmp = skb->prev; \
2493 skb != (struct sk_buff *)(queue); \
2494 skb = tmp, tmp = skb->prev)
2496 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2497 for (tmp = skb->prev; \
2498 skb != (struct sk_buff *)(queue); \
2499 skb = tmp, tmp = skb->prev)
2501 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2503 return skb_shinfo(skb)->frag_list != NULL;
2506 static inline void skb_frag_list_init(struct sk_buff *skb)
2508 skb_shinfo(skb)->frag_list = NULL;
2511 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
2513 frag->next = skb_shinfo(skb)->frag_list;
2514 skb_shinfo(skb)->frag_list = frag;
2517 #define skb_walk_frags(skb, iter) \
2518 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2520 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2521 int *peeked, int *off, int *err);
2522 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2524 unsigned int datagram_poll(struct file *file, struct socket *sock,
2525 struct poll_table_struct *wait);
2526 int skb_copy_datagram_iovec(const struct sk_buff *from, int offset,
2527 struct iovec *to, int size);
2528 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, int hlen,
2530 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
2531 const struct iovec *from, int from_offset,
2533 int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *frm,
2534 int offset, size_t count);
2535 int skb_copy_datagram_const_iovec(const struct sk_buff *from, int offset,
2536 const struct iovec *to, int to_offset,
2538 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2539 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2540 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2541 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2542 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2543 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2544 int len, __wsum csum);
2545 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
2546 struct pipe_inode_info *pipe, unsigned int len,
2547 unsigned int flags);
2548 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2549 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2550 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2552 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2553 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2554 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2555 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2556 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2558 struct skb_checksum_ops {
2559 __wsum (*update)(const void *mem, int len, __wsum wsum);
2560 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2563 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2564 __wsum csum, const struct skb_checksum_ops *ops);
2565 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2568 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
2569 int len, void *buffer)
2571 int hlen = skb_headlen(skb);
2573 if (hlen - offset >= len)
2574 return skb->data + offset;
2576 if (skb_copy_bits(skb, offset, buffer, len) < 0)
2583 * skb_needs_linearize - check if we need to linearize a given skb
2584 * depending on the given device features.
2585 * @skb: socket buffer to check
2586 * @features: net device features
2588 * Returns true if either:
2589 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2590 * 2. skb is fragmented and the device does not support SG.
2592 static inline bool skb_needs_linearize(struct sk_buff *skb,
2593 netdev_features_t features)
2595 return skb_is_nonlinear(skb) &&
2596 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2597 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2600 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2602 const unsigned int len)
2604 memcpy(to, skb->data, len);
2607 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2608 const int offset, void *to,
2609 const unsigned int len)
2611 memcpy(to, skb->data + offset, len);
2614 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2616 const unsigned int len)
2618 memcpy(skb->data, from, len);
2621 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2624 const unsigned int len)
2626 memcpy(skb->data + offset, from, len);
2629 void skb_init(void);
2631 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2637 * skb_get_timestamp - get timestamp from a skb
2638 * @skb: skb to get stamp from
2639 * @stamp: pointer to struct timeval to store stamp in
2641 * Timestamps are stored in the skb as offsets to a base timestamp.
2642 * This function converts the offset back to a struct timeval and stores
2645 static inline void skb_get_timestamp(const struct sk_buff *skb,
2646 struct timeval *stamp)
2648 *stamp = ktime_to_timeval(skb->tstamp);
2651 static inline void skb_get_timestampns(const struct sk_buff *skb,
2652 struct timespec *stamp)
2654 *stamp = ktime_to_timespec(skb->tstamp);
2657 static inline void __net_timestamp(struct sk_buff *skb)
2659 skb->tstamp = ktime_get_real();
2662 static inline ktime_t net_timedelta(ktime_t t)
2664 return ktime_sub(ktime_get_real(), t);
2667 static inline ktime_t net_invalid_timestamp(void)
2669 return ktime_set(0, 0);
2672 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2674 void skb_clone_tx_timestamp(struct sk_buff *skb);
2675 bool skb_defer_rx_timestamp(struct sk_buff *skb);
2677 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2679 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2683 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2688 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2691 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2693 * PHY drivers may accept clones of transmitted packets for
2694 * timestamping via their phy_driver.txtstamp method. These drivers
2695 * must call this function to return the skb back to the stack, with
2696 * or without a timestamp.
2698 * @skb: clone of the the original outgoing packet
2699 * @hwtstamps: hardware time stamps, may be NULL if not available
2702 void skb_complete_tx_timestamp(struct sk_buff *skb,
2703 struct skb_shared_hwtstamps *hwtstamps);
2706 * skb_tstamp_tx - queue clone of skb with send time stamps
2707 * @orig_skb: the original outgoing packet
2708 * @hwtstamps: hardware time stamps, may be NULL if not available
2710 * If the skb has a socket associated, then this function clones the
2711 * skb (thus sharing the actual data and optional structures), stores
2712 * the optional hardware time stamping information (if non NULL) or
2713 * generates a software time stamp (otherwise), then queues the clone
2714 * to the error queue of the socket. Errors are silently ignored.
2716 void skb_tstamp_tx(struct sk_buff *orig_skb,
2717 struct skb_shared_hwtstamps *hwtstamps);
2719 static inline void sw_tx_timestamp(struct sk_buff *skb)
2721 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2722 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
2723 skb_tstamp_tx(skb, NULL);
2727 * skb_tx_timestamp() - Driver hook for transmit timestamping
2729 * Ethernet MAC Drivers should call this function in their hard_xmit()
2730 * function immediately before giving the sk_buff to the MAC hardware.
2732 * Specifically, one should make absolutely sure that this function is
2733 * called before TX completion of this packet can trigger. Otherwise
2734 * the packet could potentially already be freed.
2736 * @skb: A socket buffer.
2738 static inline void skb_tx_timestamp(struct sk_buff *skb)
2740 skb_clone_tx_timestamp(skb);
2741 sw_tx_timestamp(skb);
2745 * skb_complete_wifi_ack - deliver skb with wifi status
2747 * @skb: the original outgoing packet
2748 * @acked: ack status
2751 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2753 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2754 __sum16 __skb_checksum_complete(struct sk_buff *skb);
2756 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2758 return ((skb->ip_summed & CHECKSUM_UNNECESSARY) || skb->csum_valid);
2762 * skb_checksum_complete - Calculate checksum of an entire packet
2763 * @skb: packet to process
2765 * This function calculates the checksum over the entire packet plus
2766 * the value of skb->csum. The latter can be used to supply the
2767 * checksum of a pseudo header as used by TCP/UDP. It returns the
2770 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2771 * this function can be used to verify that checksum on received
2772 * packets. In that case the function should return zero if the
2773 * checksum is correct. In particular, this function will return zero
2774 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2775 * hardware has already verified the correctness of the checksum.
2777 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
2779 return skb_csum_unnecessary(skb) ?
2780 0 : __skb_checksum_complete(skb);
2783 /* Check if we need to perform checksum complete validation.
2785 * Returns true if checksum complete is needed, false otherwise
2786 * (either checksum is unnecessary or zero checksum is allowed).
2788 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
2792 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
2793 skb->csum_valid = 1;
2800 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
2803 #define CHECKSUM_BREAK 76
2805 /* Validate (init) checksum based on checksum complete.
2808 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
2809 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
2810 * checksum is stored in skb->csum for use in __skb_checksum_complete
2811 * non-zero: value of invalid checksum
2814 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
2818 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2819 if (!csum_fold(csum_add(psum, skb->csum))) {
2820 skb->csum_valid = 1;
2827 if (complete || skb->len <= CHECKSUM_BREAK) {
2830 csum = __skb_checksum_complete(skb);
2831 skb->csum_valid = !csum;
2838 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
2843 /* Perform checksum validate (init). Note that this is a macro since we only
2844 * want to calculate the pseudo header which is an input function if necessary.
2845 * First we try to validate without any computation (checksum unnecessary) and
2846 * then calculate based on checksum complete calling the function to compute
2850 * 0: checksum is validated or try to in skb_checksum_complete
2851 * non-zero: value of invalid checksum
2853 #define __skb_checksum_validate(skb, proto, complete, \
2854 zero_okay, check, compute_pseudo) \
2856 __sum16 __ret = 0; \
2857 skb->csum_valid = 0; \
2858 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
2859 __ret = __skb_checksum_validate_complete(skb, \
2860 complete, compute_pseudo(skb, proto)); \
2864 #define skb_checksum_init(skb, proto, compute_pseudo) \
2865 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
2867 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
2868 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
2870 #define skb_checksum_validate(skb, proto, compute_pseudo) \
2871 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
2873 #define skb_checksum_validate_zero_check(skb, proto, check, \
2875 __skb_checksum_validate_(skb, proto, true, true, check, compute_pseudo)
2877 #define skb_checksum_simple_validate(skb) \
2878 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
2880 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2881 void nf_conntrack_destroy(struct nf_conntrack *nfct);
2882 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
2884 if (nfct && atomic_dec_and_test(&nfct->use))
2885 nf_conntrack_destroy(nfct);
2887 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
2890 atomic_inc(&nfct->use);
2893 #ifdef CONFIG_BRIDGE_NETFILTER
2894 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
2896 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
2899 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
2902 atomic_inc(&nf_bridge->use);
2904 #endif /* CONFIG_BRIDGE_NETFILTER */
2905 static inline void nf_reset(struct sk_buff *skb)
2907 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2908 nf_conntrack_put(skb->nfct);
2911 #ifdef CONFIG_BRIDGE_NETFILTER
2912 nf_bridge_put(skb->nf_bridge);
2913 skb->nf_bridge = NULL;
2917 static inline void nf_reset_trace(struct sk_buff *skb)
2919 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
2924 /* Note: This doesn't put any conntrack and bridge info in dst. */
2925 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2927 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2928 dst->nfct = src->nfct;
2929 nf_conntrack_get(src->nfct);
2930 dst->nfctinfo = src->nfctinfo;
2932 #ifdef CONFIG_BRIDGE_NETFILTER
2933 dst->nf_bridge = src->nf_bridge;
2934 nf_bridge_get(src->nf_bridge);
2936 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
2937 dst->nf_trace = src->nf_trace;
2941 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2943 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2944 nf_conntrack_put(dst->nfct);
2946 #ifdef CONFIG_BRIDGE_NETFILTER
2947 nf_bridge_put(dst->nf_bridge);
2949 __nf_copy(dst, src);
2952 #ifdef CONFIG_NETWORK_SECMARK
2953 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2955 to->secmark = from->secmark;
2958 static inline void skb_init_secmark(struct sk_buff *skb)
2963 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2966 static inline void skb_init_secmark(struct sk_buff *skb)
2970 static inline bool skb_irq_freeable(const struct sk_buff *skb)
2972 return !skb->destructor &&
2973 #if IS_ENABLED(CONFIG_XFRM)
2976 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
2979 !skb->_skb_refdst &&
2980 !skb_has_frag_list(skb);
2983 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
2985 skb->queue_mapping = queue_mapping;
2988 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
2990 return skb->queue_mapping;
2993 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
2995 to->queue_mapping = from->queue_mapping;
2998 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3000 skb->queue_mapping = rx_queue + 1;
3003 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3005 return skb->queue_mapping - 1;
3008 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3010 return skb->queue_mapping != 0;
3013 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
3014 unsigned int num_tx_queues);
3016 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3025 /* Keeps track of mac header offset relative to skb->head.
3026 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3027 * For non-tunnel skb it points to skb_mac_header() and for
3028 * tunnel skb it points to outer mac header.
3029 * Keeps track of level of encapsulation of network headers.
3036 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3038 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3040 return (skb_mac_header(inner_skb) - inner_skb->head) -
3041 SKB_GSO_CB(inner_skb)->mac_offset;
3044 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3046 int new_headroom, headroom;
3049 headroom = skb_headroom(skb);
3050 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3054 new_headroom = skb_headroom(skb);
3055 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3059 /* Compute the checksum for a gso segment. First compute the checksum value
3060 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3061 * then add in skb->csum (checksum from csum_start to end of packet).
3062 * skb->csum and csum_start are then updated to reflect the checksum of the
3063 * resultant packet starting from the transport header-- the resultant checksum
3064 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3067 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3069 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3070 skb_transport_offset(skb);
3073 csum = csum_fold(csum_partial(skb_transport_header(skb),
3076 SKB_GSO_CB(skb)->csum_start -= plen;
3081 static inline bool skb_is_gso(const struct sk_buff *skb)
3083 return skb_shinfo(skb)->gso_size;
3086 /* Note: Should be called only if skb_is_gso(skb) is true */
3087 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3089 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3092 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3094 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3096 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3097 * wanted then gso_type will be set. */
3098 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3100 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3101 unlikely(shinfo->gso_type == 0)) {
3102 __skb_warn_lro_forwarding(skb);
3108 static inline void skb_forward_csum(struct sk_buff *skb)
3110 /* Unfortunately we don't support this one. Any brave souls? */
3111 if (skb->ip_summed == CHECKSUM_COMPLETE)
3112 skb->ip_summed = CHECKSUM_NONE;
3116 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3117 * @skb: skb to check
3119 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3120 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3121 * use this helper, to document places where we make this assertion.
3123 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3126 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3130 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3132 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3134 u32 __skb_get_poff(const struct sk_buff *skb);
3137 * skb_head_is_locked - Determine if the skb->head is locked down
3138 * @skb: skb to check
3140 * The head on skbs build around a head frag can be removed if they are
3141 * not cloned. This function returns true if the skb head is locked down
3142 * due to either being allocated via kmalloc, or by being a clone with
3143 * multiple references to the head.
3145 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3147 return !skb->head_frag || skb_cloned(skb);
3151 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3155 * skb_gso_network_seglen is used to determine the real size of the
3156 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3158 * The MAC/L2 header is not accounted for.
3160 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3162 unsigned int hdr_len = skb_transport_header(skb) -
3163 skb_network_header(skb);
3164 return hdr_len + skb_gso_transport_seglen(skb);
3166 #endif /* __KERNEL__ */
3167 #endif /* _LINUX_SKBUFF_H */