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 * @wifi_acked_valid: wifi_acked was set
459 * @wifi_acked: whether frame was acked on wifi or not
460 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
461 * @dma_cookie: a cookie to one of several possible DMA operations
462 * done by skb DMA functions
463 * @napi_id: id of the NAPI struct this skb came from
464 * @secmark: security marking
465 * @mark: Generic packet mark
466 * @dropcount: total number of sk_receive_queue overflows
467 * @vlan_proto: vlan encapsulation protocol
468 * @vlan_tci: vlan tag control information
469 * @inner_protocol: Protocol (encapsulation)
470 * @inner_transport_header: Inner transport layer header (encapsulation)
471 * @inner_network_header: Network layer header (encapsulation)
472 * @inner_mac_header: Link layer header (encapsulation)
473 * @transport_header: Transport layer header
474 * @network_header: Network layer header
475 * @mac_header: Link layer header
476 * @tail: Tail pointer
478 * @head: Head of buffer
479 * @data: Data head pointer
480 * @truesize: Buffer size
481 * @users: User count - see {datagram,tcp}.c
485 /* These two members must be first. */
486 struct sk_buff *next;
487 struct sk_buff *prev;
491 struct skb_mstamp skb_mstamp;
495 struct net_device *dev;
498 * This is the control buffer. It is free to use for every
499 * layer. Please put your private variables there. If you
500 * want to keep them across layers you have to do a skb_clone()
501 * first. This is owned by whoever has the skb queued ATM.
503 char cb[48] __aligned(8);
505 unsigned long _skb_refdst;
521 kmemcheck_bitfield_begin(flags1);
532 kmemcheck_bitfield_end(flags1);
535 void (*destructor)(struct sk_buff *skb);
536 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
537 struct nf_conntrack *nfct;
539 #ifdef CONFIG_BRIDGE_NETFILTER
540 struct nf_bridge_info *nf_bridge;
550 #ifdef CONFIG_NET_SCHED
551 __u16 tc_index; /* traffic control index */
552 #ifdef CONFIG_NET_CLS_ACT
553 __u16 tc_verd; /* traffic control verdict */
558 kmemcheck_bitfield_begin(flags2);
559 #ifdef CONFIG_IPV6_NDISC_NODETYPE
560 __u8 ndisc_nodetype:2;
565 __u8 wifi_acked_valid:1;
569 /* Encapsulation protocol and NIC drivers should use
570 * this flag to indicate to each other if the skb contains
571 * encapsulated packet or not and maybe use the inner packet
574 __u8 encapsulation:1;
575 __u8 encap_hdr_csum:1;
577 __u8 csum_complete_sw:1;
578 /* 3/5 bit hole (depending on ndisc_nodetype presence) */
579 kmemcheck_bitfield_end(flags2);
581 #if defined CONFIG_NET_DMA || defined CONFIG_NET_RX_BUSY_POLL
583 unsigned int napi_id;
584 dma_cookie_t dma_cookie;
587 #ifdef CONFIG_NETWORK_SECMARK
593 __u32 reserved_tailroom;
596 __be16 inner_protocol;
597 __u16 inner_transport_header;
598 __u16 inner_network_header;
599 __u16 inner_mac_header;
600 __u16 transport_header;
601 __u16 network_header;
603 /* These elements must be at the end, see alloc_skb() for details. */
608 unsigned int truesize;
614 * Handling routines are only of interest to the kernel
616 #include <linux/slab.h>
619 #define SKB_ALLOC_FCLONE 0x01
620 #define SKB_ALLOC_RX 0x02
622 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
623 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
625 return unlikely(skb->pfmemalloc);
629 * skb might have a dst pointer attached, refcounted or not.
630 * _skb_refdst low order bit is set if refcount was _not_ taken
632 #define SKB_DST_NOREF 1UL
633 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
636 * skb_dst - returns skb dst_entry
639 * Returns skb dst_entry, regardless of reference taken or not.
641 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
643 /* If refdst was not refcounted, check we still are in a
644 * rcu_read_lock section
646 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
647 !rcu_read_lock_held() &&
648 !rcu_read_lock_bh_held());
649 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
653 * skb_dst_set - sets skb dst
657 * Sets skb dst, assuming a reference was taken on dst and should
658 * be released by skb_dst_drop()
660 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
662 skb->_skb_refdst = (unsigned long)dst;
665 void __skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst,
669 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
673 * Sets skb dst, assuming a reference was not taken on dst.
674 * If dst entry is cached, we do not take reference and dst_release
675 * will be avoided by refdst_drop. If dst entry is not cached, we take
676 * reference, so that last dst_release can destroy the dst immediately.
678 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
680 __skb_dst_set_noref(skb, dst, false);
684 * skb_dst_set_noref_force - sets skb dst, without taking reference
688 * Sets skb dst, assuming a reference was not taken on dst.
689 * No reference is taken and no dst_release will be called. While for
690 * cached dsts deferred reclaim is a basic feature, for entries that are
691 * not cached it is caller's job to guarantee that last dst_release for
692 * provided dst happens when nobody uses it, eg. after a RCU grace period.
694 static inline void skb_dst_set_noref_force(struct sk_buff *skb,
695 struct dst_entry *dst)
697 __skb_dst_set_noref(skb, dst, true);
701 * skb_dst_is_noref - Test if skb dst isn't refcounted
704 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
706 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
709 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
711 return (struct rtable *)skb_dst(skb);
714 void kfree_skb(struct sk_buff *skb);
715 void kfree_skb_list(struct sk_buff *segs);
716 void skb_tx_error(struct sk_buff *skb);
717 void consume_skb(struct sk_buff *skb);
718 void __kfree_skb(struct sk_buff *skb);
719 extern struct kmem_cache *skbuff_head_cache;
721 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
722 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
723 bool *fragstolen, int *delta_truesize);
725 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
727 struct sk_buff *build_skb(void *data, unsigned int frag_size);
728 static inline struct sk_buff *alloc_skb(unsigned int size,
731 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
734 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
737 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
740 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
741 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
743 return __alloc_skb_head(priority, -1);
746 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
747 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
748 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
749 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
750 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
751 gfp_t gfp_mask, bool fclone);
752 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
755 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
758 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
759 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
760 unsigned int headroom);
761 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
762 int newtailroom, gfp_t priority);
763 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
764 int offset, int len);
765 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
767 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
768 int skb_pad(struct sk_buff *skb, int pad);
769 #define dev_kfree_skb(a) consume_skb(a)
771 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
772 int getfrag(void *from, char *to, int offset,
773 int len, int odd, struct sk_buff *skb),
774 void *from, int length);
776 struct skb_seq_state {
780 __u32 stepped_offset;
781 struct sk_buff *root_skb;
782 struct sk_buff *cur_skb;
786 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
787 unsigned int to, struct skb_seq_state *st);
788 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
789 struct skb_seq_state *st);
790 void skb_abort_seq_read(struct skb_seq_state *st);
792 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
793 unsigned int to, struct ts_config *config,
794 struct ts_state *state);
797 * Packet hash types specify the type of hash in skb_set_hash.
799 * Hash types refer to the protocol layer addresses which are used to
800 * construct a packet's hash. The hashes are used to differentiate or identify
801 * flows of the protocol layer for the hash type. Hash types are either
802 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
804 * Properties of hashes:
806 * 1) Two packets in different flows have different hash values
807 * 2) Two packets in the same flow should have the same hash value
809 * A hash at a higher layer is considered to be more specific. A driver should
810 * set the most specific hash possible.
812 * A driver cannot indicate a more specific hash than the layer at which a hash
813 * was computed. For instance an L3 hash cannot be set as an L4 hash.
815 * A driver may indicate a hash level which is less specific than the
816 * actual layer the hash was computed on. For instance, a hash computed
817 * at L4 may be considered an L3 hash. This should only be done if the
818 * driver can't unambiguously determine that the HW computed the hash at
819 * the higher layer. Note that the "should" in the second property above
822 enum pkt_hash_types {
823 PKT_HASH_TYPE_NONE, /* Undefined type */
824 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
825 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
826 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
830 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
832 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)
845 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
850 static inline void skb_clear_hash(struct sk_buff *skb)
856 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
862 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
864 to->hash = from->hash;
865 to->l4_hash = from->l4_hash;
868 #ifdef NET_SKBUFF_DATA_USES_OFFSET
869 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
871 return skb->head + skb->end;
874 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
879 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
884 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
886 return skb->end - skb->head;
891 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
893 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
895 return &skb_shinfo(skb)->hwtstamps;
899 * skb_queue_empty - check if a queue is empty
902 * Returns true if the queue is empty, false otherwise.
904 static inline int skb_queue_empty(const struct sk_buff_head *list)
906 return list->next == (const struct sk_buff *) list;
910 * skb_queue_is_last - check if skb is the last entry in the queue
914 * Returns true if @skb is the last buffer on the list.
916 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
917 const struct sk_buff *skb)
919 return skb->next == (const struct sk_buff *) list;
923 * skb_queue_is_first - check if skb is the first entry in the queue
927 * Returns true if @skb is the first buffer on the list.
929 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
930 const struct sk_buff *skb)
932 return skb->prev == (const struct sk_buff *) list;
936 * skb_queue_next - return the next packet in the queue
938 * @skb: current buffer
940 * Return the next packet in @list after @skb. It is only valid to
941 * call this if skb_queue_is_last() evaluates to false.
943 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
944 const struct sk_buff *skb)
946 /* This BUG_ON may seem severe, but if we just return then we
947 * are going to dereference garbage.
949 BUG_ON(skb_queue_is_last(list, skb));
954 * skb_queue_prev - return the prev packet in the queue
956 * @skb: current buffer
958 * Return the prev packet in @list before @skb. It is only valid to
959 * call this if skb_queue_is_first() evaluates to false.
961 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
962 const struct sk_buff *skb)
964 /* This BUG_ON may seem severe, but if we just return then we
965 * are going to dereference garbage.
967 BUG_ON(skb_queue_is_first(list, skb));
972 * skb_get - reference buffer
973 * @skb: buffer to reference
975 * Makes another reference to a socket buffer and returns a pointer
978 static inline struct sk_buff *skb_get(struct sk_buff *skb)
980 atomic_inc(&skb->users);
985 * If users == 1, we are the only owner and are can avoid redundant
990 * skb_cloned - is the buffer a clone
991 * @skb: buffer to check
993 * Returns true if the buffer was generated with skb_clone() and is
994 * one of multiple shared copies of the buffer. Cloned buffers are
995 * shared data so must not be written to under normal circumstances.
997 static inline int skb_cloned(const struct sk_buff *skb)
999 return skb->cloned &&
1000 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1003 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1005 might_sleep_if(pri & __GFP_WAIT);
1007 if (skb_cloned(skb))
1008 return pskb_expand_head(skb, 0, 0, pri);
1014 * skb_header_cloned - is the header a clone
1015 * @skb: buffer to check
1017 * Returns true if modifying the header part of the buffer requires
1018 * the data to be copied.
1020 static inline int skb_header_cloned(const struct sk_buff *skb)
1027 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1028 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1029 return dataref != 1;
1033 * skb_header_release - release reference to header
1034 * @skb: buffer to operate on
1036 * Drop a reference to the header part of the buffer. This is done
1037 * by acquiring a payload reference. You must not read from the header
1038 * part of skb->data after this.
1040 static inline void skb_header_release(struct sk_buff *skb)
1044 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1048 * skb_shared - is the buffer shared
1049 * @skb: buffer to check
1051 * Returns true if more than one person has a reference to this
1054 static inline int skb_shared(const struct sk_buff *skb)
1056 return atomic_read(&skb->users) != 1;
1060 * skb_share_check - check if buffer is shared and if so clone it
1061 * @skb: buffer to check
1062 * @pri: priority for memory allocation
1064 * If the buffer is shared the buffer is cloned and the old copy
1065 * drops a reference. A new clone with a single reference is returned.
1066 * If the buffer is not shared the original buffer is returned. When
1067 * being called from interrupt status or with spinlocks held pri must
1070 * NULL is returned on a memory allocation failure.
1072 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1074 might_sleep_if(pri & __GFP_WAIT);
1075 if (skb_shared(skb)) {
1076 struct sk_buff *nskb = skb_clone(skb, pri);
1088 * Copy shared buffers into a new sk_buff. We effectively do COW on
1089 * packets to handle cases where we have a local reader and forward
1090 * and a couple of other messy ones. The normal one is tcpdumping
1091 * a packet thats being forwarded.
1095 * skb_unshare - make a copy of a shared buffer
1096 * @skb: buffer to check
1097 * @pri: priority for memory allocation
1099 * If the socket buffer is a clone then this function creates a new
1100 * copy of the data, drops a reference count on the old copy and returns
1101 * the new copy with the reference count at 1. If the buffer is not a clone
1102 * the original buffer is returned. When called with a spinlock held or
1103 * from interrupt state @pri must be %GFP_ATOMIC
1105 * %NULL is returned on a memory allocation failure.
1107 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1110 might_sleep_if(pri & __GFP_WAIT);
1111 if (skb_cloned(skb)) {
1112 struct sk_buff *nskb = skb_copy(skb, pri);
1113 kfree_skb(skb); /* Free our shared copy */
1120 * skb_peek - peek at the head of an &sk_buff_head
1121 * @list_: list to peek at
1123 * Peek an &sk_buff. Unlike most other operations you _MUST_
1124 * be careful with this one. A peek leaves the buffer on the
1125 * list and someone else may run off with it. You must hold
1126 * the appropriate locks or have a private queue to do this.
1128 * Returns %NULL for an empty list or a pointer to the head element.
1129 * The reference count is not incremented and the reference is therefore
1130 * volatile. Use with caution.
1132 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1134 struct sk_buff *skb = list_->next;
1136 if (skb == (struct sk_buff *)list_)
1142 * skb_peek_next - peek skb following the given one from a queue
1143 * @skb: skb to start from
1144 * @list_: list to peek at
1146 * Returns %NULL when the end of the list is met or a pointer to the
1147 * next element. The reference count is not incremented and the
1148 * reference is therefore volatile. Use with caution.
1150 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1151 const struct sk_buff_head *list_)
1153 struct sk_buff *next = skb->next;
1155 if (next == (struct sk_buff *)list_)
1161 * skb_peek_tail - peek at the tail of an &sk_buff_head
1162 * @list_: list to peek at
1164 * Peek an &sk_buff. Unlike most other operations you _MUST_
1165 * be careful with this one. A peek leaves the buffer on the
1166 * list and someone else may run off with it. You must hold
1167 * the appropriate locks or have a private queue to do this.
1169 * Returns %NULL for an empty list or a pointer to the tail element.
1170 * The reference count is not incremented and the reference is therefore
1171 * volatile. Use with caution.
1173 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1175 struct sk_buff *skb = list_->prev;
1177 if (skb == (struct sk_buff *)list_)
1184 * skb_queue_len - get queue length
1185 * @list_: list to measure
1187 * Return the length of an &sk_buff queue.
1189 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1195 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1196 * @list: queue to initialize
1198 * This initializes only the list and queue length aspects of
1199 * an sk_buff_head object. This allows to initialize the list
1200 * aspects of an sk_buff_head without reinitializing things like
1201 * the spinlock. It can also be used for on-stack sk_buff_head
1202 * objects where the spinlock is known to not be used.
1204 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1206 list->prev = list->next = (struct sk_buff *)list;
1211 * This function creates a split out lock class for each invocation;
1212 * this is needed for now since a whole lot of users of the skb-queue
1213 * infrastructure in drivers have different locking usage (in hardirq)
1214 * than the networking core (in softirq only). In the long run either the
1215 * network layer or drivers should need annotation to consolidate the
1216 * main types of usage into 3 classes.
1218 static inline void skb_queue_head_init(struct sk_buff_head *list)
1220 spin_lock_init(&list->lock);
1221 __skb_queue_head_init(list);
1224 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1225 struct lock_class_key *class)
1227 skb_queue_head_init(list);
1228 lockdep_set_class(&list->lock, class);
1232 * Insert an sk_buff on a list.
1234 * The "__skb_xxxx()" functions are the non-atomic ones that
1235 * can only be called with interrupts disabled.
1237 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1238 struct sk_buff_head *list);
1239 static inline void __skb_insert(struct sk_buff *newsk,
1240 struct sk_buff *prev, struct sk_buff *next,
1241 struct sk_buff_head *list)
1245 next->prev = prev->next = newsk;
1249 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1250 struct sk_buff *prev,
1251 struct sk_buff *next)
1253 struct sk_buff *first = list->next;
1254 struct sk_buff *last = list->prev;
1264 * skb_queue_splice - join two skb lists, this is designed for stacks
1265 * @list: the new list to add
1266 * @head: the place to add it in the first list
1268 static inline void skb_queue_splice(const struct sk_buff_head *list,
1269 struct sk_buff_head *head)
1271 if (!skb_queue_empty(list)) {
1272 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1273 head->qlen += list->qlen;
1278 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1279 * @list: the new list to add
1280 * @head: the place to add it in the first list
1282 * The list at @list is reinitialised
1284 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1285 struct sk_buff_head *head)
1287 if (!skb_queue_empty(list)) {
1288 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1289 head->qlen += list->qlen;
1290 __skb_queue_head_init(list);
1295 * skb_queue_splice_tail - join two skb lists, each list being a queue
1296 * @list: the new list to add
1297 * @head: the place to add it in the first list
1299 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1300 struct sk_buff_head *head)
1302 if (!skb_queue_empty(list)) {
1303 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1304 head->qlen += list->qlen;
1309 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1310 * @list: the new list to add
1311 * @head: the place to add it in the first list
1313 * Each of the lists is a queue.
1314 * The list at @list is reinitialised
1316 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1317 struct sk_buff_head *head)
1319 if (!skb_queue_empty(list)) {
1320 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1321 head->qlen += list->qlen;
1322 __skb_queue_head_init(list);
1327 * __skb_queue_after - queue a buffer at the list head
1328 * @list: list to use
1329 * @prev: place after this buffer
1330 * @newsk: buffer to queue
1332 * Queue a buffer int the middle of a list. This function takes no locks
1333 * and you must therefore hold required locks before calling it.
1335 * A buffer cannot be placed on two lists at the same time.
1337 static inline void __skb_queue_after(struct sk_buff_head *list,
1338 struct sk_buff *prev,
1339 struct sk_buff *newsk)
1341 __skb_insert(newsk, prev, prev->next, list);
1344 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1345 struct sk_buff_head *list);
1347 static inline void __skb_queue_before(struct sk_buff_head *list,
1348 struct sk_buff *next,
1349 struct sk_buff *newsk)
1351 __skb_insert(newsk, next->prev, next, list);
1355 * __skb_queue_head - queue a buffer at the list head
1356 * @list: list to use
1357 * @newsk: buffer to queue
1359 * Queue a buffer at the start of a list. This function takes no locks
1360 * and you must therefore hold required locks before calling it.
1362 * A buffer cannot be placed on two lists at the same time.
1364 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1365 static inline void __skb_queue_head(struct sk_buff_head *list,
1366 struct sk_buff *newsk)
1368 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1372 * __skb_queue_tail - queue a buffer at the list tail
1373 * @list: list to use
1374 * @newsk: buffer to queue
1376 * Queue a buffer at the end of a list. This function takes no locks
1377 * and you must therefore hold required locks before calling it.
1379 * A buffer cannot be placed on two lists at the same time.
1381 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1382 static inline void __skb_queue_tail(struct sk_buff_head *list,
1383 struct sk_buff *newsk)
1385 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1389 * remove sk_buff from list. _Must_ be called atomically, and with
1392 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1393 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1395 struct sk_buff *next, *prev;
1400 skb->next = skb->prev = NULL;
1406 * __skb_dequeue - remove from the head of the queue
1407 * @list: list to dequeue from
1409 * Remove the head of the list. This function does not take any locks
1410 * so must be used with appropriate locks held only. The head item is
1411 * returned or %NULL if the list is empty.
1413 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1414 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1416 struct sk_buff *skb = skb_peek(list);
1418 __skb_unlink(skb, list);
1423 * __skb_dequeue_tail - remove from the tail of the queue
1424 * @list: list to dequeue from
1426 * Remove the tail of the list. This function does not take any locks
1427 * so must be used with appropriate locks held only. The tail item is
1428 * returned or %NULL if the list is empty.
1430 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1431 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1433 struct sk_buff *skb = skb_peek_tail(list);
1435 __skb_unlink(skb, list);
1440 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1442 return skb->data_len;
1445 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1447 return skb->len - skb->data_len;
1450 static inline int skb_pagelen(const struct sk_buff *skb)
1454 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1455 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1456 return len + skb_headlen(skb);
1460 * __skb_fill_page_desc - initialise a paged fragment in an skb
1461 * @skb: buffer containing fragment to be initialised
1462 * @i: paged fragment index to initialise
1463 * @page: the page to use for this fragment
1464 * @off: the offset to the data with @page
1465 * @size: the length of the data
1467 * Initialises the @i'th fragment of @skb to point to &size bytes at
1468 * offset @off within @page.
1470 * Does not take any additional reference on the fragment.
1472 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1473 struct page *page, int off, int size)
1475 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1478 * Propagate page->pfmemalloc to the skb if we can. The problem is
1479 * that not all callers have unique ownership of the page. If
1480 * pfmemalloc is set, we check the mapping as a mapping implies
1481 * page->index is set (index and pfmemalloc share space).
1482 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1483 * do not lose pfmemalloc information as the pages would not be
1484 * allocated using __GFP_MEMALLOC.
1486 frag->page.p = page;
1487 frag->page_offset = off;
1488 skb_frag_size_set(frag, size);
1490 page = compound_head(page);
1491 if (page->pfmemalloc && !page->mapping)
1492 skb->pfmemalloc = true;
1496 * skb_fill_page_desc - initialise a paged fragment in an skb
1497 * @skb: buffer containing fragment to be initialised
1498 * @i: paged fragment index to initialise
1499 * @page: the page to use for this fragment
1500 * @off: the offset to the data with @page
1501 * @size: the length of the data
1503 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1504 * @skb to point to @size bytes at offset @off within @page. In
1505 * addition updates @skb such that @i is the last fragment.
1507 * Does not take any additional reference on the fragment.
1509 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1510 struct page *page, int off, int size)
1512 __skb_fill_page_desc(skb, i, page, off, size);
1513 skb_shinfo(skb)->nr_frags = i + 1;
1516 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1517 int size, unsigned int truesize);
1519 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1520 unsigned int truesize);
1522 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1523 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1524 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1526 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1527 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1529 return skb->head + skb->tail;
1532 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1534 skb->tail = skb->data - skb->head;
1537 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1539 skb_reset_tail_pointer(skb);
1540 skb->tail += offset;
1543 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1544 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1549 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1551 skb->tail = skb->data;
1554 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1556 skb->tail = skb->data + offset;
1559 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1562 * Add data to an sk_buff
1564 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1565 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1566 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1568 unsigned char *tmp = skb_tail_pointer(skb);
1569 SKB_LINEAR_ASSERT(skb);
1575 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1576 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1583 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1584 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1587 BUG_ON(skb->len < skb->data_len);
1588 return skb->data += len;
1591 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1593 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1596 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1598 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1600 if (len > skb_headlen(skb) &&
1601 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1604 return skb->data += len;
1607 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1609 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1612 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1614 if (likely(len <= skb_headlen(skb)))
1616 if (unlikely(len > skb->len))
1618 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1622 * skb_headroom - bytes at buffer head
1623 * @skb: buffer to check
1625 * Return the number of bytes of free space at the head of an &sk_buff.
1627 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1629 return skb->data - skb->head;
1633 * skb_tailroom - bytes at buffer end
1634 * @skb: buffer to check
1636 * Return the number of bytes of free space at the tail of an sk_buff
1638 static inline int skb_tailroom(const struct sk_buff *skb)
1640 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1644 * skb_availroom - bytes at buffer end
1645 * @skb: buffer to check
1647 * Return the number of bytes of free space at the tail of an sk_buff
1648 * allocated by sk_stream_alloc()
1650 static inline int skb_availroom(const struct sk_buff *skb)
1652 if (skb_is_nonlinear(skb))
1655 return skb->end - skb->tail - skb->reserved_tailroom;
1659 * skb_reserve - adjust headroom
1660 * @skb: buffer to alter
1661 * @len: bytes to move
1663 * Increase the headroom of an empty &sk_buff by reducing the tail
1664 * room. This is only allowed for an empty buffer.
1666 static inline void skb_reserve(struct sk_buff *skb, int len)
1672 static inline void skb_reset_inner_headers(struct sk_buff *skb)
1674 skb->inner_mac_header = skb->mac_header;
1675 skb->inner_network_header = skb->network_header;
1676 skb->inner_transport_header = skb->transport_header;
1679 static inline void skb_reset_mac_len(struct sk_buff *skb)
1681 skb->mac_len = skb->network_header - skb->mac_header;
1684 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1687 return skb->head + skb->inner_transport_header;
1690 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1692 skb->inner_transport_header = skb->data - skb->head;
1695 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1698 skb_reset_inner_transport_header(skb);
1699 skb->inner_transport_header += offset;
1702 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1704 return skb->head + skb->inner_network_header;
1707 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1709 skb->inner_network_header = skb->data - skb->head;
1712 static inline void skb_set_inner_network_header(struct sk_buff *skb,
1715 skb_reset_inner_network_header(skb);
1716 skb->inner_network_header += offset;
1719 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1721 return skb->head + skb->inner_mac_header;
1724 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1726 skb->inner_mac_header = skb->data - skb->head;
1729 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1732 skb_reset_inner_mac_header(skb);
1733 skb->inner_mac_header += offset;
1735 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1737 return skb->transport_header != (typeof(skb->transport_header))~0U;
1740 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1742 return skb->head + skb->transport_header;
1745 static inline void skb_reset_transport_header(struct sk_buff *skb)
1747 skb->transport_header = skb->data - skb->head;
1750 static inline void skb_set_transport_header(struct sk_buff *skb,
1753 skb_reset_transport_header(skb);
1754 skb->transport_header += offset;
1757 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1759 return skb->head + skb->network_header;
1762 static inline void skb_reset_network_header(struct sk_buff *skb)
1764 skb->network_header = skb->data - skb->head;
1767 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1769 skb_reset_network_header(skb);
1770 skb->network_header += offset;
1773 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1775 return skb->head + skb->mac_header;
1778 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1780 return skb->mac_header != (typeof(skb->mac_header))~0U;
1783 static inline void skb_reset_mac_header(struct sk_buff *skb)
1785 skb->mac_header = skb->data - skb->head;
1788 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1790 skb_reset_mac_header(skb);
1791 skb->mac_header += offset;
1794 static inline void skb_pop_mac_header(struct sk_buff *skb)
1796 skb->mac_header = skb->network_header;
1799 static inline void skb_probe_transport_header(struct sk_buff *skb,
1800 const int offset_hint)
1802 struct flow_keys keys;
1804 if (skb_transport_header_was_set(skb))
1806 else if (skb_flow_dissect(skb, &keys))
1807 skb_set_transport_header(skb, keys.thoff);
1809 skb_set_transport_header(skb, offset_hint);
1812 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1814 if (skb_mac_header_was_set(skb)) {
1815 const unsigned char *old_mac = skb_mac_header(skb);
1817 skb_set_mac_header(skb, -skb->mac_len);
1818 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
1822 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
1824 return skb->csum_start - skb_headroom(skb);
1827 static inline int skb_transport_offset(const struct sk_buff *skb)
1829 return skb_transport_header(skb) - skb->data;
1832 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1834 return skb->transport_header - skb->network_header;
1837 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
1839 return skb->inner_transport_header - skb->inner_network_header;
1842 static inline int skb_network_offset(const struct sk_buff *skb)
1844 return skb_network_header(skb) - skb->data;
1847 static inline int skb_inner_network_offset(const struct sk_buff *skb)
1849 return skb_inner_network_header(skb) - skb->data;
1852 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1854 return pskb_may_pull(skb, skb_network_offset(skb) + len);
1857 static inline void skb_pop_rcv_encapsulation(struct sk_buff *skb)
1859 /* Only continue with checksum unnecessary if device indicated
1860 * it is valid across encapsulation (skb->encapsulation was set).
1862 if (skb->ip_summed == CHECKSUM_UNNECESSARY && !skb->encapsulation)
1863 skb->ip_summed = CHECKSUM_NONE;
1865 skb->encapsulation = 0;
1866 skb->csum_valid = 0;
1870 * CPUs often take a performance hit when accessing unaligned memory
1871 * locations. The actual performance hit varies, it can be small if the
1872 * hardware handles it or large if we have to take an exception and fix it
1875 * Since an ethernet header is 14 bytes network drivers often end up with
1876 * the IP header at an unaligned offset. The IP header can be aligned by
1877 * shifting the start of the packet by 2 bytes. Drivers should do this
1880 * skb_reserve(skb, NET_IP_ALIGN);
1882 * The downside to this alignment of the IP header is that the DMA is now
1883 * unaligned. On some architectures the cost of an unaligned DMA is high
1884 * and this cost outweighs the gains made by aligning the IP header.
1886 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1889 #ifndef NET_IP_ALIGN
1890 #define NET_IP_ALIGN 2
1894 * The networking layer reserves some headroom in skb data (via
1895 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
1896 * the header has to grow. In the default case, if the header has to grow
1897 * 32 bytes or less we avoid the reallocation.
1899 * Unfortunately this headroom changes the DMA alignment of the resulting
1900 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
1901 * on some architectures. An architecture can override this value,
1902 * perhaps setting it to a cacheline in size (since that will maintain
1903 * cacheline alignment of the DMA). It must be a power of 2.
1905 * Various parts of the networking layer expect at least 32 bytes of
1906 * headroom, you should not reduce this.
1908 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
1909 * to reduce average number of cache lines per packet.
1910 * get_rps_cpus() for example only access one 64 bytes aligned block :
1911 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
1914 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
1917 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
1919 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
1921 if (unlikely(skb_is_nonlinear(skb))) {
1926 skb_set_tail_pointer(skb, len);
1929 void skb_trim(struct sk_buff *skb, unsigned int len);
1931 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
1934 return ___pskb_trim(skb, len);
1935 __skb_trim(skb, len);
1939 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
1941 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
1945 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
1946 * @skb: buffer to alter
1949 * This is identical to pskb_trim except that the caller knows that
1950 * the skb is not cloned so we should never get an error due to out-
1953 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
1955 int err = pskb_trim(skb, len);
1960 * skb_orphan - orphan a buffer
1961 * @skb: buffer to orphan
1963 * If a buffer currently has an owner then we call the owner's
1964 * destructor function and make the @skb unowned. The buffer continues
1965 * to exist but is no longer charged to its former owner.
1967 static inline void skb_orphan(struct sk_buff *skb)
1969 if (skb->destructor) {
1970 skb->destructor(skb);
1971 skb->destructor = NULL;
1979 * skb_orphan_frags - orphan the frags contained in a buffer
1980 * @skb: buffer to orphan frags from
1981 * @gfp_mask: allocation mask for replacement pages
1983 * For each frag in the SKB which needs a destructor (i.e. has an
1984 * owner) create a copy of that frag and release the original
1985 * page by calling the destructor.
1987 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
1989 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
1991 return skb_copy_ubufs(skb, gfp_mask);
1995 * __skb_queue_purge - empty a list
1996 * @list: list to empty
1998 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1999 * the list and one reference dropped. This function does not take the
2000 * list lock and the caller must hold the relevant locks to use it.
2002 void skb_queue_purge(struct sk_buff_head *list);
2003 static inline void __skb_queue_purge(struct sk_buff_head *list)
2005 struct sk_buff *skb;
2006 while ((skb = __skb_dequeue(list)) != NULL)
2010 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
2011 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
2012 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
2014 void *netdev_alloc_frag(unsigned int fragsz);
2016 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2020 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2021 * @dev: network device to receive on
2022 * @length: length to allocate
2024 * Allocate a new &sk_buff and assign it a usage count of one. The
2025 * buffer has unspecified headroom built in. Users should allocate
2026 * the headroom they think they need without accounting for the
2027 * built in space. The built in space is used for optimisations.
2029 * %NULL is returned if there is no free memory. Although this function
2030 * allocates memory it can be called from an interrupt.
2032 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2033 unsigned int length)
2035 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2038 /* legacy helper around __netdev_alloc_skb() */
2039 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2042 return __netdev_alloc_skb(NULL, length, gfp_mask);
2045 /* legacy helper around netdev_alloc_skb() */
2046 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2048 return netdev_alloc_skb(NULL, length);
2052 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2053 unsigned int length, gfp_t gfp)
2055 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2057 if (NET_IP_ALIGN && skb)
2058 skb_reserve(skb, NET_IP_ALIGN);
2062 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2063 unsigned int length)
2065 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2069 * __skb_alloc_pages - allocate pages for ps-rx on a skb and preserve pfmemalloc data
2070 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2071 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2072 * @order: size of the allocation
2074 * Allocate a new page.
2076 * %NULL is returned if there is no free memory.
2078 static inline struct page *__skb_alloc_pages(gfp_t gfp_mask,
2079 struct sk_buff *skb,
2084 gfp_mask |= __GFP_COLD;
2086 if (!(gfp_mask & __GFP_NOMEMALLOC))
2087 gfp_mask |= __GFP_MEMALLOC;
2089 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2090 if (skb && page && page->pfmemalloc)
2091 skb->pfmemalloc = true;
2097 * __skb_alloc_page - allocate a page for ps-rx for a given skb and preserve pfmemalloc data
2098 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2099 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2101 * Allocate a new page.
2103 * %NULL is returned if there is no free memory.
2105 static inline struct page *__skb_alloc_page(gfp_t gfp_mask,
2106 struct sk_buff *skb)
2108 return __skb_alloc_pages(gfp_mask, skb, 0);
2112 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2113 * @page: The page that was allocated from skb_alloc_page
2114 * @skb: The skb that may need pfmemalloc set
2116 static inline void skb_propagate_pfmemalloc(struct page *page,
2117 struct sk_buff *skb)
2119 if (page && page->pfmemalloc)
2120 skb->pfmemalloc = true;
2124 * skb_frag_page - retrieve the page referred to by a paged fragment
2125 * @frag: the paged fragment
2127 * Returns the &struct page associated with @frag.
2129 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2131 return frag->page.p;
2135 * __skb_frag_ref - take an addition reference on a paged fragment.
2136 * @frag: the paged fragment
2138 * Takes an additional reference on the paged fragment @frag.
2140 static inline void __skb_frag_ref(skb_frag_t *frag)
2142 get_page(skb_frag_page(frag));
2146 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2148 * @f: the fragment offset.
2150 * Takes an additional reference on the @f'th paged fragment of @skb.
2152 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2154 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2158 * __skb_frag_unref - release a reference on a paged fragment.
2159 * @frag: the paged fragment
2161 * Releases a reference on the paged fragment @frag.
2163 static inline void __skb_frag_unref(skb_frag_t *frag)
2165 put_page(skb_frag_page(frag));
2169 * skb_frag_unref - release a reference on a paged fragment of an skb.
2171 * @f: the fragment offset
2173 * Releases a reference on the @f'th paged fragment of @skb.
2175 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2177 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2181 * skb_frag_address - gets the address of the data contained in a paged fragment
2182 * @frag: the paged fragment buffer
2184 * Returns the address of the data within @frag. The page must already
2187 static inline void *skb_frag_address(const skb_frag_t *frag)
2189 return page_address(skb_frag_page(frag)) + frag->page_offset;
2193 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2194 * @frag: the paged fragment buffer
2196 * Returns the address of the data within @frag. Checks that the page
2197 * is mapped and returns %NULL otherwise.
2199 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2201 void *ptr = page_address(skb_frag_page(frag));
2205 return ptr + frag->page_offset;
2209 * __skb_frag_set_page - sets the page contained in a paged fragment
2210 * @frag: the paged fragment
2211 * @page: the page to set
2213 * Sets the fragment @frag to contain @page.
2215 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2217 frag->page.p = page;
2221 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2223 * @f: the fragment offset
2224 * @page: the page to set
2226 * Sets the @f'th fragment of @skb to contain @page.
2228 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2231 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2234 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2237 * skb_frag_dma_map - maps a paged fragment via the DMA API
2238 * @dev: the device to map the fragment to
2239 * @frag: the paged fragment to map
2240 * @offset: the offset within the fragment (starting at the
2241 * fragment's own offset)
2242 * @size: the number of bytes to map
2243 * @dir: the direction of the mapping (%PCI_DMA_*)
2245 * Maps the page associated with @frag to @device.
2247 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2248 const skb_frag_t *frag,
2249 size_t offset, size_t size,
2250 enum dma_data_direction dir)
2252 return dma_map_page(dev, skb_frag_page(frag),
2253 frag->page_offset + offset, size, dir);
2256 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2259 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2263 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2266 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2271 * skb_clone_writable - is the header of a clone writable
2272 * @skb: buffer to check
2273 * @len: length up to which to write
2275 * Returns true if modifying the header part of the cloned buffer
2276 * does not requires the data to be copied.
2278 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2280 return !skb_header_cloned(skb) &&
2281 skb_headroom(skb) + len <= skb->hdr_len;
2284 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2289 if (headroom > skb_headroom(skb))
2290 delta = headroom - skb_headroom(skb);
2292 if (delta || cloned)
2293 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2299 * skb_cow - copy header of skb when it is required
2300 * @skb: buffer to cow
2301 * @headroom: needed headroom
2303 * If the skb passed lacks sufficient headroom or its data part
2304 * is shared, data is reallocated. If reallocation fails, an error
2305 * is returned and original skb is not changed.
2307 * The result is skb with writable area skb->head...skb->tail
2308 * and at least @headroom of space at head.
2310 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2312 return __skb_cow(skb, headroom, skb_cloned(skb));
2316 * skb_cow_head - skb_cow but only making the head writable
2317 * @skb: buffer to cow
2318 * @headroom: needed headroom
2320 * This function is identical to skb_cow except that we replace the
2321 * skb_cloned check by skb_header_cloned. It should be used when
2322 * you only need to push on some header and do not need to modify
2325 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2327 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2331 * skb_padto - pad an skbuff up to a minimal size
2332 * @skb: buffer to pad
2333 * @len: minimal length
2335 * Pads up a buffer to ensure the trailing bytes exist and are
2336 * blanked. If the buffer already contains sufficient data it
2337 * is untouched. Otherwise it is extended. Returns zero on
2338 * success. The skb is freed on error.
2341 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2343 unsigned int size = skb->len;
2344 if (likely(size >= len))
2346 return skb_pad(skb, len - size);
2349 static inline int skb_add_data(struct sk_buff *skb,
2350 char __user *from, int copy)
2352 const int off = skb->len;
2354 if (skb->ip_summed == CHECKSUM_NONE) {
2356 __wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
2359 skb->csum = csum_block_add(skb->csum, csum, off);
2362 } else if (!copy_from_user(skb_put(skb, copy), from, copy))
2365 __skb_trim(skb, off);
2369 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2370 const struct page *page, int off)
2373 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2375 return page == skb_frag_page(frag) &&
2376 off == frag->page_offset + skb_frag_size(frag);
2381 static inline int __skb_linearize(struct sk_buff *skb)
2383 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2387 * skb_linearize - convert paged skb to linear one
2388 * @skb: buffer to linarize
2390 * If there is no free memory -ENOMEM is returned, otherwise zero
2391 * is returned and the old skb data released.
2393 static inline int skb_linearize(struct sk_buff *skb)
2395 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2399 * skb_has_shared_frag - can any frag be overwritten
2400 * @skb: buffer to test
2402 * Return true if the skb has at least one frag that might be modified
2403 * by an external entity (as in vmsplice()/sendfile())
2405 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2407 return skb_is_nonlinear(skb) &&
2408 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2412 * skb_linearize_cow - make sure skb is linear and writable
2413 * @skb: buffer to process
2415 * If there is no free memory -ENOMEM is returned, otherwise zero
2416 * is returned and the old skb data released.
2418 static inline int skb_linearize_cow(struct sk_buff *skb)
2420 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2421 __skb_linearize(skb) : 0;
2425 * skb_postpull_rcsum - update checksum for received skb after pull
2426 * @skb: buffer to update
2427 * @start: start of data before pull
2428 * @len: length of data pulled
2430 * After doing a pull on a received packet, you need to call this to
2431 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2432 * CHECKSUM_NONE so that it can be recomputed from scratch.
2435 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2436 const void *start, unsigned int len)
2438 if (skb->ip_summed == CHECKSUM_COMPLETE)
2439 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2442 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2445 * pskb_trim_rcsum - trim received skb and update checksum
2446 * @skb: buffer to trim
2449 * This is exactly the same as pskb_trim except that it ensures the
2450 * checksum of received packets are still valid after the operation.
2453 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2455 if (likely(len >= skb->len))
2457 if (skb->ip_summed == CHECKSUM_COMPLETE)
2458 skb->ip_summed = CHECKSUM_NONE;
2459 return __pskb_trim(skb, len);
2462 #define skb_queue_walk(queue, skb) \
2463 for (skb = (queue)->next; \
2464 skb != (struct sk_buff *)(queue); \
2467 #define skb_queue_walk_safe(queue, skb, tmp) \
2468 for (skb = (queue)->next, tmp = skb->next; \
2469 skb != (struct sk_buff *)(queue); \
2470 skb = tmp, tmp = skb->next)
2472 #define skb_queue_walk_from(queue, skb) \
2473 for (; skb != (struct sk_buff *)(queue); \
2476 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2477 for (tmp = skb->next; \
2478 skb != (struct sk_buff *)(queue); \
2479 skb = tmp, tmp = skb->next)
2481 #define skb_queue_reverse_walk(queue, skb) \
2482 for (skb = (queue)->prev; \
2483 skb != (struct sk_buff *)(queue); \
2486 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2487 for (skb = (queue)->prev, tmp = skb->prev; \
2488 skb != (struct sk_buff *)(queue); \
2489 skb = tmp, tmp = skb->prev)
2491 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2492 for (tmp = skb->prev; \
2493 skb != (struct sk_buff *)(queue); \
2494 skb = tmp, tmp = skb->prev)
2496 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2498 return skb_shinfo(skb)->frag_list != NULL;
2501 static inline void skb_frag_list_init(struct sk_buff *skb)
2503 skb_shinfo(skb)->frag_list = NULL;
2506 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
2508 frag->next = skb_shinfo(skb)->frag_list;
2509 skb_shinfo(skb)->frag_list = frag;
2512 #define skb_walk_frags(skb, iter) \
2513 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2515 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2516 int *peeked, int *off, int *err);
2517 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2519 unsigned int datagram_poll(struct file *file, struct socket *sock,
2520 struct poll_table_struct *wait);
2521 int skb_copy_datagram_iovec(const struct sk_buff *from, int offset,
2522 struct iovec *to, int size);
2523 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, int hlen,
2525 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
2526 const struct iovec *from, int from_offset,
2528 int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *frm,
2529 int offset, size_t count);
2530 int skb_copy_datagram_const_iovec(const struct sk_buff *from, int offset,
2531 const struct iovec *to, int to_offset,
2533 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2534 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2535 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2536 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2537 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2538 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2539 int len, __wsum csum);
2540 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
2541 struct pipe_inode_info *pipe, unsigned int len,
2542 unsigned int flags);
2543 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2544 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2545 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2547 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2548 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2549 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2550 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2551 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2553 struct skb_checksum_ops {
2554 __wsum (*update)(const void *mem, int len, __wsum wsum);
2555 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2558 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2559 __wsum csum, const struct skb_checksum_ops *ops);
2560 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2563 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
2564 int len, void *buffer)
2566 int hlen = skb_headlen(skb);
2568 if (hlen - offset >= len)
2569 return skb->data + offset;
2571 if (skb_copy_bits(skb, offset, buffer, len) < 0)
2578 * skb_needs_linearize - check if we need to linearize a given skb
2579 * depending on the given device features.
2580 * @skb: socket buffer to check
2581 * @features: net device features
2583 * Returns true if either:
2584 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2585 * 2. skb is fragmented and the device does not support SG.
2587 static inline bool skb_needs_linearize(struct sk_buff *skb,
2588 netdev_features_t features)
2590 return skb_is_nonlinear(skb) &&
2591 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2592 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2595 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2597 const unsigned int len)
2599 memcpy(to, skb->data, len);
2602 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2603 const int offset, void *to,
2604 const unsigned int len)
2606 memcpy(to, skb->data + offset, len);
2609 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2611 const unsigned int len)
2613 memcpy(skb->data, from, len);
2616 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2619 const unsigned int len)
2621 memcpy(skb->data + offset, from, len);
2624 void skb_init(void);
2626 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2632 * skb_get_timestamp - get timestamp from a skb
2633 * @skb: skb to get stamp from
2634 * @stamp: pointer to struct timeval to store stamp in
2636 * Timestamps are stored in the skb as offsets to a base timestamp.
2637 * This function converts the offset back to a struct timeval and stores
2640 static inline void skb_get_timestamp(const struct sk_buff *skb,
2641 struct timeval *stamp)
2643 *stamp = ktime_to_timeval(skb->tstamp);
2646 static inline void skb_get_timestampns(const struct sk_buff *skb,
2647 struct timespec *stamp)
2649 *stamp = ktime_to_timespec(skb->tstamp);
2652 static inline void __net_timestamp(struct sk_buff *skb)
2654 skb->tstamp = ktime_get_real();
2657 static inline ktime_t net_timedelta(ktime_t t)
2659 return ktime_sub(ktime_get_real(), t);
2662 static inline ktime_t net_invalid_timestamp(void)
2664 return ktime_set(0, 0);
2667 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2669 void skb_clone_tx_timestamp(struct sk_buff *skb);
2670 bool skb_defer_rx_timestamp(struct sk_buff *skb);
2672 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2674 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2678 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2683 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2686 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2688 * PHY drivers may accept clones of transmitted packets for
2689 * timestamping via their phy_driver.txtstamp method. These drivers
2690 * must call this function to return the skb back to the stack, with
2691 * or without a timestamp.
2693 * @skb: clone of the the original outgoing packet
2694 * @hwtstamps: hardware time stamps, may be NULL if not available
2697 void skb_complete_tx_timestamp(struct sk_buff *skb,
2698 struct skb_shared_hwtstamps *hwtstamps);
2701 * skb_tstamp_tx - queue clone of skb with send time stamps
2702 * @orig_skb: the original outgoing packet
2703 * @hwtstamps: hardware time stamps, may be NULL if not available
2705 * If the skb has a socket associated, then this function clones the
2706 * skb (thus sharing the actual data and optional structures), stores
2707 * the optional hardware time stamping information (if non NULL) or
2708 * generates a software time stamp (otherwise), then queues the clone
2709 * to the error queue of the socket. Errors are silently ignored.
2711 void skb_tstamp_tx(struct sk_buff *orig_skb,
2712 struct skb_shared_hwtstamps *hwtstamps);
2714 static inline void sw_tx_timestamp(struct sk_buff *skb)
2716 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2717 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
2718 skb_tstamp_tx(skb, NULL);
2722 * skb_tx_timestamp() - Driver hook for transmit timestamping
2724 * Ethernet MAC Drivers should call this function in their hard_xmit()
2725 * function immediately before giving the sk_buff to the MAC hardware.
2727 * Specifically, one should make absolutely sure that this function is
2728 * called before TX completion of this packet can trigger. Otherwise
2729 * the packet could potentially already be freed.
2731 * @skb: A socket buffer.
2733 static inline void skb_tx_timestamp(struct sk_buff *skb)
2735 skb_clone_tx_timestamp(skb);
2736 sw_tx_timestamp(skb);
2740 * skb_complete_wifi_ack - deliver skb with wifi status
2742 * @skb: the original outgoing packet
2743 * @acked: ack status
2746 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2748 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2749 __sum16 __skb_checksum_complete(struct sk_buff *skb);
2751 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2753 return ((skb->ip_summed & CHECKSUM_UNNECESSARY) || skb->csum_valid);
2757 * skb_checksum_complete - Calculate checksum of an entire packet
2758 * @skb: packet to process
2760 * This function calculates the checksum over the entire packet plus
2761 * the value of skb->csum. The latter can be used to supply the
2762 * checksum of a pseudo header as used by TCP/UDP. It returns the
2765 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2766 * this function can be used to verify that checksum on received
2767 * packets. In that case the function should return zero if the
2768 * checksum is correct. In particular, this function will return zero
2769 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2770 * hardware has already verified the correctness of the checksum.
2772 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
2774 return skb_csum_unnecessary(skb) ?
2775 0 : __skb_checksum_complete(skb);
2778 /* Check if we need to perform checksum complete validation.
2780 * Returns true if checksum complete is needed, false otherwise
2781 * (either checksum is unnecessary or zero checksum is allowed).
2783 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
2787 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
2788 skb->csum_valid = 1;
2795 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
2798 #define CHECKSUM_BREAK 76
2800 /* Validate (init) checksum based on checksum complete.
2803 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
2804 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
2805 * checksum is stored in skb->csum for use in __skb_checksum_complete
2806 * non-zero: value of invalid checksum
2809 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
2813 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2814 if (!csum_fold(csum_add(psum, skb->csum))) {
2815 skb->csum_valid = 1;
2822 if (complete || skb->len <= CHECKSUM_BREAK) {
2825 csum = __skb_checksum_complete(skb);
2826 skb->csum_valid = !csum;
2833 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
2838 /* Perform checksum validate (init). Note that this is a macro since we only
2839 * want to calculate the pseudo header which is an input function if necessary.
2840 * First we try to validate without any computation (checksum unnecessary) and
2841 * then calculate based on checksum complete calling the function to compute
2845 * 0: checksum is validated or try to in skb_checksum_complete
2846 * non-zero: value of invalid checksum
2848 #define __skb_checksum_validate(skb, proto, complete, \
2849 zero_okay, check, compute_pseudo) \
2851 __sum16 __ret = 0; \
2852 skb->csum_valid = 0; \
2853 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
2854 __ret = __skb_checksum_validate_complete(skb, \
2855 complete, compute_pseudo(skb, proto)); \
2859 #define skb_checksum_init(skb, proto, compute_pseudo) \
2860 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
2862 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
2863 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
2865 #define skb_checksum_validate(skb, proto, compute_pseudo) \
2866 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
2868 #define skb_checksum_validate_zero_check(skb, proto, check, \
2870 __skb_checksum_validate_(skb, proto, true, true, check, compute_pseudo)
2872 #define skb_checksum_simple_validate(skb) \
2873 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
2875 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2876 void nf_conntrack_destroy(struct nf_conntrack *nfct);
2877 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
2879 if (nfct && atomic_dec_and_test(&nfct->use))
2880 nf_conntrack_destroy(nfct);
2882 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
2885 atomic_inc(&nfct->use);
2888 #ifdef CONFIG_BRIDGE_NETFILTER
2889 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
2891 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
2894 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
2897 atomic_inc(&nf_bridge->use);
2899 #endif /* CONFIG_BRIDGE_NETFILTER */
2900 static inline void nf_reset(struct sk_buff *skb)
2902 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2903 nf_conntrack_put(skb->nfct);
2906 #ifdef CONFIG_BRIDGE_NETFILTER
2907 nf_bridge_put(skb->nf_bridge);
2908 skb->nf_bridge = NULL;
2912 static inline void nf_reset_trace(struct sk_buff *skb)
2914 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
2919 /* Note: This doesn't put any conntrack and bridge info in dst. */
2920 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2922 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2923 dst->nfct = src->nfct;
2924 nf_conntrack_get(src->nfct);
2925 dst->nfctinfo = src->nfctinfo;
2927 #ifdef CONFIG_BRIDGE_NETFILTER
2928 dst->nf_bridge = src->nf_bridge;
2929 nf_bridge_get(src->nf_bridge);
2931 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
2932 dst->nf_trace = src->nf_trace;
2936 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2938 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2939 nf_conntrack_put(dst->nfct);
2941 #ifdef CONFIG_BRIDGE_NETFILTER
2942 nf_bridge_put(dst->nf_bridge);
2944 __nf_copy(dst, src);
2947 #ifdef CONFIG_NETWORK_SECMARK
2948 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2950 to->secmark = from->secmark;
2953 static inline void skb_init_secmark(struct sk_buff *skb)
2958 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2961 static inline void skb_init_secmark(struct sk_buff *skb)
2965 static inline bool skb_irq_freeable(const struct sk_buff *skb)
2967 return !skb->destructor &&
2968 #if IS_ENABLED(CONFIG_XFRM)
2971 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
2974 !skb->_skb_refdst &&
2975 !skb_has_frag_list(skb);
2978 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
2980 skb->queue_mapping = queue_mapping;
2983 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
2985 return skb->queue_mapping;
2988 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
2990 to->queue_mapping = from->queue_mapping;
2993 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
2995 skb->queue_mapping = rx_queue + 1;
2998 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3000 return skb->queue_mapping - 1;
3003 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3005 return skb->queue_mapping != 0;
3008 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
3009 unsigned int num_tx_queues);
3011 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3020 /* Keeps track of mac header offset relative to skb->head.
3021 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3022 * For non-tunnel skb it points to skb_mac_header() and for
3023 * tunnel skb it points to outer mac header.
3024 * Keeps track of level of encapsulation of network headers.
3031 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3033 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3035 return (skb_mac_header(inner_skb) - inner_skb->head) -
3036 SKB_GSO_CB(inner_skb)->mac_offset;
3039 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3041 int new_headroom, headroom;
3044 headroom = skb_headroom(skb);
3045 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3049 new_headroom = skb_headroom(skb);
3050 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3054 /* Compute the checksum for a gso segment. First compute the checksum value
3055 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3056 * then add in skb->csum (checksum from csum_start to end of packet).
3057 * skb->csum and csum_start are then updated to reflect the checksum of the
3058 * resultant packet starting from the transport header-- the resultant checksum
3059 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3062 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3064 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3065 skb_transport_offset(skb);
3068 csum = csum_fold(csum_partial(skb_transport_header(skb),
3071 SKB_GSO_CB(skb)->csum_start -= plen;
3076 static inline bool skb_is_gso(const struct sk_buff *skb)
3078 return skb_shinfo(skb)->gso_size;
3081 /* Note: Should be called only if skb_is_gso(skb) is true */
3082 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3084 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3087 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3089 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3091 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3092 * wanted then gso_type will be set. */
3093 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3095 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3096 unlikely(shinfo->gso_type == 0)) {
3097 __skb_warn_lro_forwarding(skb);
3103 static inline void skb_forward_csum(struct sk_buff *skb)
3105 /* Unfortunately we don't support this one. Any brave souls? */
3106 if (skb->ip_summed == CHECKSUM_COMPLETE)
3107 skb->ip_summed = CHECKSUM_NONE;
3111 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3112 * @skb: skb to check
3114 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3115 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3116 * use this helper, to document places where we make this assertion.
3118 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3121 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3125 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3127 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3129 u32 __skb_get_poff(const struct sk_buff *skb);
3132 * skb_head_is_locked - Determine if the skb->head is locked down
3133 * @skb: skb to check
3135 * The head on skbs build around a head frag can be removed if they are
3136 * not cloned. This function returns true if the skb head is locked down
3137 * due to either being allocated via kmalloc, or by being a clone with
3138 * multiple references to the head.
3140 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3142 return !skb->head_frag || skb_cloned(skb);
3146 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3150 * skb_gso_network_seglen is used to determine the real size of the
3151 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3153 * The MAC/L2 header is not accounted for.
3155 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3157 unsigned int hdr_len = skb_transport_header(skb) -
3158 skb_network_header(skb);
3159 return hdr_len + skb_gso_transport_seglen(skb);
3161 #endif /* __KERNEL__ */
3162 #endif /* _LINUX_SKBUFF_H */