2 * Definitions for the 'struct sk_buff' memory handlers.
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) ALIGN(X, SMP_CACHE_BYTES)
116 #define SKB_WITH_OVERHEAD(X) \
117 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
118 #define SKB_MAX_ORDER(X, ORDER) \
119 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
120 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
121 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
123 /* return minimum truesize of one skb containing X bytes of data */
124 #define SKB_TRUESIZE(X) ((X) + \
125 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
126 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
130 struct pipe_inode_info;
132 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
133 struct nf_conntrack {
138 #ifdef CONFIG_BRIDGE_NETFILTER
139 struct nf_bridge_info {
142 struct net_device *physindev;
143 struct net_device *physoutdev;
144 unsigned long data[32 / sizeof(unsigned long)];
148 struct sk_buff_head {
149 /* These two members must be first. */
150 struct sk_buff *next;
151 struct sk_buff *prev;
159 /* To allow 64K frame to be packed as single skb without frag_list we
160 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
161 * buffers which do not start on a page boundary.
163 * Since GRO uses frags we allocate at least 16 regardless of page
166 #if (65536/PAGE_SIZE + 1) < 16
167 #define MAX_SKB_FRAGS 16UL
169 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
172 typedef struct skb_frag_struct skb_frag_t;
174 struct skb_frag_struct {
178 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
187 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
192 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
197 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
202 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
207 #define HAVE_HW_TIME_STAMP
210 * struct skb_shared_hwtstamps - hardware time stamps
211 * @hwtstamp: hardware time stamp transformed into duration
212 * since arbitrary point in time
214 * Software time stamps generated by ktime_get_real() are stored in
217 * hwtstamps can only be compared against other hwtstamps from
220 * This structure is attached to packets as part of the
221 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
223 struct skb_shared_hwtstamps {
227 /* Definitions for tx_flags in struct skb_shared_info */
229 /* generate hardware time stamp */
230 SKBTX_HW_TSTAMP = 1 << 0,
232 /* generate software time stamp when queueing packet to NIC */
233 SKBTX_SW_TSTAMP = 1 << 1,
235 /* device driver is going to provide hardware time stamp */
236 SKBTX_IN_PROGRESS = 1 << 2,
238 /* device driver supports TX zero-copy buffers */
239 SKBTX_DEV_ZEROCOPY = 1 << 3,
241 /* generate wifi status information (where possible) */
242 SKBTX_WIFI_STATUS = 1 << 4,
244 /* This indicates at least one fragment might be overwritten
245 * (as in vmsplice(), sendfile() ...)
246 * If we need to compute a TX checksum, we'll need to copy
247 * all frags to avoid possible bad checksum
249 SKBTX_SHARED_FRAG = 1 << 5,
251 /* generate software time stamp when entering packet scheduling */
252 SKBTX_SCHED_TSTAMP = 1 << 6,
254 /* generate software timestamp on peer data acknowledgment */
255 SKBTX_ACK_TSTAMP = 1 << 7,
258 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
259 SKBTX_SCHED_TSTAMP | \
261 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
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;
293 * Warning : all fields before dataref are cleared in __alloc_skb()
297 /* Intermediate layers must ensure that destructor_arg
298 * remains valid until skb destructor */
299 void * destructor_arg;
301 /* must be last field, see pskb_expand_head() */
302 skb_frag_t frags[MAX_SKB_FRAGS];
305 /* We divide dataref into two halves. The higher 16 bits hold references
306 * to the payload part of skb->data. The lower 16 bits hold references to
307 * the entire skb->data. A clone of a headerless skb holds the length of
308 * the header in skb->hdr_len.
310 * All users must obey the rule that the skb->data reference count must be
311 * greater than or equal to the payload reference count.
313 * Holding a reference to the payload part means that the user does not
314 * care about modifications to the header part of skb->data.
316 #define SKB_DATAREF_SHIFT 16
317 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
321 SKB_FCLONE_UNAVAILABLE,
327 SKB_GSO_TCPV4 = 1 << 0,
328 SKB_GSO_UDP = 1 << 1,
330 /* This indicates the skb is from an untrusted source. */
331 SKB_GSO_DODGY = 1 << 2,
333 /* This indicates the tcp segment has CWR set. */
334 SKB_GSO_TCP_ECN = 1 << 3,
336 SKB_GSO_TCPV6 = 1 << 4,
338 SKB_GSO_FCOE = 1 << 5,
340 SKB_GSO_GRE = 1 << 6,
342 SKB_GSO_GRE_CSUM = 1 << 7,
344 SKB_GSO_IPIP = 1 << 8,
346 SKB_GSO_SIT = 1 << 9,
348 SKB_GSO_UDP_TUNNEL = 1 << 10,
350 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
352 SKB_GSO_MPLS = 1 << 12,
356 #if BITS_PER_LONG > 32
357 #define NET_SKBUFF_DATA_USES_OFFSET 1
360 #ifdef NET_SKBUFF_DATA_USES_OFFSET
361 typedef unsigned int sk_buff_data_t;
363 typedef unsigned char *sk_buff_data_t;
367 * struct skb_mstamp - multi resolution time stamps
368 * @stamp_us: timestamp in us resolution
369 * @stamp_jiffies: timestamp in jiffies
382 * skb_mstamp_get - get current timestamp
383 * @cl: place to store timestamps
385 static inline void skb_mstamp_get(struct skb_mstamp *cl)
387 u64 val = local_clock();
389 do_div(val, NSEC_PER_USEC);
390 cl->stamp_us = (u32)val;
391 cl->stamp_jiffies = (u32)jiffies;
395 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
396 * @t1: pointer to newest sample
397 * @t0: pointer to oldest sample
399 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
400 const struct skb_mstamp *t0)
402 s32 delta_us = t1->stamp_us - t0->stamp_us;
403 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
405 /* If delta_us is negative, this might be because interval is too big,
406 * or local_clock() drift is too big : fallback using jiffies.
409 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
411 delta_us = jiffies_to_usecs(delta_jiffies);
418 * struct sk_buff - socket buffer
419 * @next: Next buffer in list
420 * @prev: Previous buffer in list
421 * @tstamp: Time we arrived/left
422 * @sk: Socket we are owned by
423 * @dev: Device we arrived on/are leaving by
424 * @cb: Control buffer. Free for use by every layer. Put private vars here
425 * @_skb_refdst: destination entry (with norefcount bit)
426 * @sp: the security path, used for xfrm
427 * @len: Length of actual data
428 * @data_len: Data length
429 * @mac_len: Length of link layer header
430 * @hdr_len: writable header length of cloned skb
431 * @csum: Checksum (must include start/offset pair)
432 * @csum_start: Offset from skb->head where checksumming should start
433 * @csum_offset: Offset from csum_start where checksum should be stored
434 * @priority: Packet queueing priority
435 * @ignore_df: allow local fragmentation
436 * @cloned: Head may be cloned (check refcnt to be sure)
437 * @ip_summed: Driver fed us an IP checksum
438 * @nohdr: Payload reference only, must not modify header
439 * @nfctinfo: Relationship of this skb to the connection
440 * @pkt_type: Packet class
441 * @fclone: skbuff clone status
442 * @ipvs_property: skbuff is owned by ipvs
443 * @peeked: this packet has been seen already, so stats have been
444 * done for it, don't do them again
445 * @nf_trace: netfilter packet trace flag
446 * @protocol: Packet protocol from driver
447 * @destructor: Destruct function
448 * @nfct: Associated connection, if any
449 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
450 * @skb_iif: ifindex of device we arrived on
451 * @tc_index: Traffic control index
452 * @tc_verd: traffic control verdict
453 * @hash: the packet hash
454 * @queue_mapping: Queue mapping for multiqueue devices
455 * @ndisc_nodetype: router type (from link layer)
456 * @ooo_okay: allow the mapping of a socket to a queue to be changed
457 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
459 * @sw_hash: indicates hash was computed in software stack
460 * @wifi_acked_valid: wifi_acked was set
461 * @wifi_acked: whether frame was acked on wifi or not
462 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
463 * @dma_cookie: a cookie to one of several possible DMA operations
464 * done by skb DMA functions
465 * @napi_id: id of the NAPI struct this skb came from
466 * @secmark: security marking
467 * @mark: Generic packet mark
468 * @dropcount: total number of sk_receive_queue overflows
469 * @vlan_proto: vlan encapsulation protocol
470 * @vlan_tci: vlan tag control information
471 * @inner_protocol: Protocol (encapsulation)
472 * @inner_transport_header: Inner transport layer header (encapsulation)
473 * @inner_network_header: Network layer header (encapsulation)
474 * @inner_mac_header: Link layer header (encapsulation)
475 * @transport_header: Transport layer header
476 * @network_header: Network layer header
477 * @mac_header: Link layer header
478 * @tail: Tail pointer
480 * @head: Head of buffer
481 * @data: Data head pointer
482 * @truesize: Buffer size
483 * @users: User count - see {datagram,tcp}.c
487 /* These two members must be first. */
488 struct sk_buff *next;
489 struct sk_buff *prev;
493 struct skb_mstamp skb_mstamp;
497 struct net_device *dev;
500 * This is the control buffer. It is free to use for every
501 * layer. Please put your private variables there. If you
502 * want to keep them across layers you have to do a skb_clone()
503 * first. This is owned by whoever has the skb queued ATM.
505 char cb[48] __aligned(8);
507 unsigned long _skb_refdst;
523 kmemcheck_bitfield_begin(flags1);
534 kmemcheck_bitfield_end(flags1);
537 void (*destructor)(struct sk_buff *skb);
538 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
539 struct nf_conntrack *nfct;
541 #ifdef CONFIG_BRIDGE_NETFILTER
542 struct nf_bridge_info *nf_bridge;
552 #ifdef CONFIG_NET_SCHED
553 __u16 tc_index; /* traffic control index */
554 #ifdef CONFIG_NET_CLS_ACT
555 __u16 tc_verd; /* traffic control verdict */
560 kmemcheck_bitfield_begin(flags2);
561 #ifdef CONFIG_IPV6_NDISC_NODETYPE
562 __u8 ndisc_nodetype:2;
568 __u8 wifi_acked_valid:1;
572 /* Encapsulation protocol and NIC drivers should use
573 * this flag to indicate to each other if the skb contains
574 * encapsulated packet or not and maybe use the inner packet
577 __u8 encapsulation:1;
578 __u8 encap_hdr_csum:1;
580 __u8 csum_complete_sw:1;
581 /* 2/4 bit hole (depending on ndisc_nodetype presence) */
582 kmemcheck_bitfield_end(flags2);
584 #if defined CONFIG_NET_DMA || defined CONFIG_NET_RX_BUSY_POLL
586 unsigned int napi_id;
587 dma_cookie_t dma_cookie;
590 #ifdef CONFIG_NETWORK_SECMARK
596 __u32 reserved_tailroom;
599 __be16 inner_protocol;
600 __u16 inner_transport_header;
601 __u16 inner_network_header;
602 __u16 inner_mac_header;
603 __u16 transport_header;
604 __u16 network_header;
606 /* These elements must be at the end, see alloc_skb() for details. */
611 unsigned int truesize;
617 * Handling routines are only of interest to the kernel
619 #include <linux/slab.h>
622 #define SKB_ALLOC_FCLONE 0x01
623 #define SKB_ALLOC_RX 0x02
625 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
626 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
628 return unlikely(skb->pfmemalloc);
632 * skb might have a dst pointer attached, refcounted or not.
633 * _skb_refdst low order bit is set if refcount was _not_ taken
635 #define SKB_DST_NOREF 1UL
636 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
639 * skb_dst - returns skb dst_entry
642 * Returns skb dst_entry, regardless of reference taken or not.
644 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
646 /* If refdst was not refcounted, check we still are in a
647 * rcu_read_lock section
649 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
650 !rcu_read_lock_held() &&
651 !rcu_read_lock_bh_held());
652 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
656 * skb_dst_set - sets skb dst
660 * Sets skb dst, assuming a reference was taken on dst and should
661 * be released by skb_dst_drop()
663 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
665 skb->_skb_refdst = (unsigned long)dst;
668 void __skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst,
672 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
676 * Sets skb dst, assuming a reference was not taken on dst.
677 * If dst entry is cached, we do not take reference and dst_release
678 * will be avoided by refdst_drop. If dst entry is not cached, we take
679 * reference, so that last dst_release can destroy the dst immediately.
681 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
683 __skb_dst_set_noref(skb, dst, false);
687 * skb_dst_set_noref_force - sets skb dst, without taking reference
691 * Sets skb dst, assuming a reference was not taken on dst.
692 * No reference is taken and no dst_release will be called. While for
693 * cached dsts deferred reclaim is a basic feature, for entries that are
694 * not cached it is caller's job to guarantee that last dst_release for
695 * provided dst happens when nobody uses it, eg. after a RCU grace period.
697 static inline void skb_dst_set_noref_force(struct sk_buff *skb,
698 struct dst_entry *dst)
700 __skb_dst_set_noref(skb, dst, true);
704 * skb_dst_is_noref - Test if skb dst isn't refcounted
707 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
709 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
712 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
714 return (struct rtable *)skb_dst(skb);
717 void kfree_skb(struct sk_buff *skb);
718 void kfree_skb_list(struct sk_buff *segs);
719 void skb_tx_error(struct sk_buff *skb);
720 void consume_skb(struct sk_buff *skb);
721 void __kfree_skb(struct sk_buff *skb);
722 extern struct kmem_cache *skbuff_head_cache;
724 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
725 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
726 bool *fragstolen, int *delta_truesize);
728 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
730 struct sk_buff *build_skb(void *data, unsigned int frag_size);
731 static inline struct sk_buff *alloc_skb(unsigned int size,
734 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
737 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
740 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
743 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
744 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
746 return __alloc_skb_head(priority, -1);
749 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
750 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
751 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
752 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
753 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
754 gfp_t gfp_mask, bool fclone);
755 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
758 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
761 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
762 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
763 unsigned int headroom);
764 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
765 int newtailroom, gfp_t priority);
766 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
767 int offset, int len);
768 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
770 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
771 int skb_pad(struct sk_buff *skb, int pad);
772 #define dev_kfree_skb(a) consume_skb(a)
774 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
775 int getfrag(void *from, char *to, int offset,
776 int len, int odd, struct sk_buff *skb),
777 void *from, int length);
779 struct skb_seq_state {
783 __u32 stepped_offset;
784 struct sk_buff *root_skb;
785 struct sk_buff *cur_skb;
789 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
790 unsigned int to, struct skb_seq_state *st);
791 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
792 struct skb_seq_state *st);
793 void skb_abort_seq_read(struct skb_seq_state *st);
795 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
796 unsigned int to, struct ts_config *config,
797 struct ts_state *state);
800 * Packet hash types specify the type of hash in skb_set_hash.
802 * Hash types refer to the protocol layer addresses which are used to
803 * construct a packet's hash. The hashes are used to differentiate or identify
804 * flows of the protocol layer for the hash type. Hash types are either
805 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
807 * Properties of hashes:
809 * 1) Two packets in different flows have different hash values
810 * 2) Two packets in the same flow should have the same hash value
812 * A hash at a higher layer is considered to be more specific. A driver should
813 * set the most specific hash possible.
815 * A driver cannot indicate a more specific hash than the layer at which a hash
816 * was computed. For instance an L3 hash cannot be set as an L4 hash.
818 * A driver may indicate a hash level which is less specific than the
819 * actual layer the hash was computed on. For instance, a hash computed
820 * at L4 may be considered an L3 hash. This should only be done if the
821 * driver can't unambiguously determine that the HW computed the hash at
822 * the higher layer. Note that the "should" in the second property above
825 enum pkt_hash_types {
826 PKT_HASH_TYPE_NONE, /* Undefined type */
827 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
828 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
829 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
833 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
835 skb->l4_hash = (type == PKT_HASH_TYPE_L4);
840 void __skb_get_hash(struct sk_buff *skb);
841 static inline __u32 skb_get_hash(struct sk_buff *skb)
843 if (!skb->l4_hash && !skb->sw_hash)
849 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
854 static inline void skb_clear_hash(struct sk_buff *skb)
861 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
867 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
869 to->hash = from->hash;
870 to->sw_hash = from->sw_hash;
871 to->l4_hash = from->l4_hash;
874 #ifdef NET_SKBUFF_DATA_USES_OFFSET
875 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
877 return skb->head + skb->end;
880 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
885 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
890 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
892 return skb->end - skb->head;
897 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
899 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
901 return &skb_shinfo(skb)->hwtstamps;
905 * skb_queue_empty - check if a queue is empty
908 * Returns true if the queue is empty, false otherwise.
910 static inline int skb_queue_empty(const struct sk_buff_head *list)
912 return list->next == (const struct sk_buff *) list;
916 * skb_queue_is_last - check if skb is the last entry in the queue
920 * Returns true if @skb is the last buffer on the list.
922 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
923 const struct sk_buff *skb)
925 return skb->next == (const struct sk_buff *) list;
929 * skb_queue_is_first - check if skb is the first entry in the queue
933 * Returns true if @skb is the first buffer on the list.
935 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
936 const struct sk_buff *skb)
938 return skb->prev == (const struct sk_buff *) list;
942 * skb_queue_next - return the next packet in the queue
944 * @skb: current buffer
946 * Return the next packet in @list after @skb. It is only valid to
947 * call this if skb_queue_is_last() evaluates to false.
949 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
950 const struct sk_buff *skb)
952 /* This BUG_ON may seem severe, but if we just return then we
953 * are going to dereference garbage.
955 BUG_ON(skb_queue_is_last(list, skb));
960 * skb_queue_prev - return the prev packet in the queue
962 * @skb: current buffer
964 * Return the prev packet in @list before @skb. It is only valid to
965 * call this if skb_queue_is_first() evaluates to false.
967 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
968 const struct sk_buff *skb)
970 /* This BUG_ON may seem severe, but if we just return then we
971 * are going to dereference garbage.
973 BUG_ON(skb_queue_is_first(list, skb));
978 * skb_get - reference buffer
979 * @skb: buffer to reference
981 * Makes another reference to a socket buffer and returns a pointer
984 static inline struct sk_buff *skb_get(struct sk_buff *skb)
986 atomic_inc(&skb->users);
991 * If users == 1, we are the only owner and are can avoid redundant
996 * skb_cloned - is the buffer a clone
997 * @skb: buffer to check
999 * Returns true if the buffer was generated with skb_clone() and is
1000 * one of multiple shared copies of the buffer. Cloned buffers are
1001 * shared data so must not be written to under normal circumstances.
1003 static inline int skb_cloned(const struct sk_buff *skb)
1005 return skb->cloned &&
1006 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1009 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1011 might_sleep_if(pri & __GFP_WAIT);
1013 if (skb_cloned(skb))
1014 return pskb_expand_head(skb, 0, 0, pri);
1020 * skb_header_cloned - is the header a clone
1021 * @skb: buffer to check
1023 * Returns true if modifying the header part of the buffer requires
1024 * the data to be copied.
1026 static inline int skb_header_cloned(const struct sk_buff *skb)
1033 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1034 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1035 return dataref != 1;
1039 * skb_header_release - release reference to header
1040 * @skb: buffer to operate on
1042 * Drop a reference to the header part of the buffer. This is done
1043 * by acquiring a payload reference. You must not read from the header
1044 * part of skb->data after this.
1046 static inline void skb_header_release(struct sk_buff *skb)
1050 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1054 * skb_shared - is the buffer shared
1055 * @skb: buffer to check
1057 * Returns true if more than one person has a reference to this
1060 static inline int skb_shared(const struct sk_buff *skb)
1062 return atomic_read(&skb->users) != 1;
1066 * skb_share_check - check if buffer is shared and if so clone it
1067 * @skb: buffer to check
1068 * @pri: priority for memory allocation
1070 * If the buffer is shared the buffer is cloned and the old copy
1071 * drops a reference. A new clone with a single reference is returned.
1072 * If the buffer is not shared the original buffer is returned. When
1073 * being called from interrupt status or with spinlocks held pri must
1076 * NULL is returned on a memory allocation failure.
1078 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1080 might_sleep_if(pri & __GFP_WAIT);
1081 if (skb_shared(skb)) {
1082 struct sk_buff *nskb = skb_clone(skb, pri);
1094 * Copy shared buffers into a new sk_buff. We effectively do COW on
1095 * packets to handle cases where we have a local reader and forward
1096 * and a couple of other messy ones. The normal one is tcpdumping
1097 * a packet thats being forwarded.
1101 * skb_unshare - make a copy of a shared buffer
1102 * @skb: buffer to check
1103 * @pri: priority for memory allocation
1105 * If the socket buffer is a clone then this function creates a new
1106 * copy of the data, drops a reference count on the old copy and returns
1107 * the new copy with the reference count at 1. If the buffer is not a clone
1108 * the original buffer is returned. When called with a spinlock held or
1109 * from interrupt state @pri must be %GFP_ATOMIC
1111 * %NULL is returned on a memory allocation failure.
1113 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1116 might_sleep_if(pri & __GFP_WAIT);
1117 if (skb_cloned(skb)) {
1118 struct sk_buff *nskb = skb_copy(skb, pri);
1119 kfree_skb(skb); /* Free our shared copy */
1126 * skb_peek - peek at the head of an &sk_buff_head
1127 * @list_: list to peek at
1129 * Peek an &sk_buff. Unlike most other operations you _MUST_
1130 * be careful with this one. A peek leaves the buffer on the
1131 * list and someone else may run off with it. You must hold
1132 * the appropriate locks or have a private queue to do this.
1134 * Returns %NULL for an empty list or a pointer to the head element.
1135 * The reference count is not incremented and the reference is therefore
1136 * volatile. Use with caution.
1138 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1140 struct sk_buff *skb = list_->next;
1142 if (skb == (struct sk_buff *)list_)
1148 * skb_peek_next - peek skb following the given one from a queue
1149 * @skb: skb to start from
1150 * @list_: list to peek at
1152 * Returns %NULL when the end of the list is met or a pointer to the
1153 * next element. The reference count is not incremented and the
1154 * reference is therefore volatile. Use with caution.
1156 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1157 const struct sk_buff_head *list_)
1159 struct sk_buff *next = skb->next;
1161 if (next == (struct sk_buff *)list_)
1167 * skb_peek_tail - peek at the tail of an &sk_buff_head
1168 * @list_: list to peek at
1170 * Peek an &sk_buff. Unlike most other operations you _MUST_
1171 * be careful with this one. A peek leaves the buffer on the
1172 * list and someone else may run off with it. You must hold
1173 * the appropriate locks or have a private queue to do this.
1175 * Returns %NULL for an empty list or a pointer to the tail element.
1176 * The reference count is not incremented and the reference is therefore
1177 * volatile. Use with caution.
1179 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1181 struct sk_buff *skb = list_->prev;
1183 if (skb == (struct sk_buff *)list_)
1190 * skb_queue_len - get queue length
1191 * @list_: list to measure
1193 * Return the length of an &sk_buff queue.
1195 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1201 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1202 * @list: queue to initialize
1204 * This initializes only the list and queue length aspects of
1205 * an sk_buff_head object. This allows to initialize the list
1206 * aspects of an sk_buff_head without reinitializing things like
1207 * the spinlock. It can also be used for on-stack sk_buff_head
1208 * objects where the spinlock is known to not be used.
1210 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1212 list->prev = list->next = (struct sk_buff *)list;
1217 * This function creates a split out lock class for each invocation;
1218 * this is needed for now since a whole lot of users of the skb-queue
1219 * infrastructure in drivers have different locking usage (in hardirq)
1220 * than the networking core (in softirq only). In the long run either the
1221 * network layer or drivers should need annotation to consolidate the
1222 * main types of usage into 3 classes.
1224 static inline void skb_queue_head_init(struct sk_buff_head *list)
1226 spin_lock_init(&list->lock);
1227 __skb_queue_head_init(list);
1230 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1231 struct lock_class_key *class)
1233 skb_queue_head_init(list);
1234 lockdep_set_class(&list->lock, class);
1238 * Insert an sk_buff on a list.
1240 * The "__skb_xxxx()" functions are the non-atomic ones that
1241 * can only be called with interrupts disabled.
1243 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1244 struct sk_buff_head *list);
1245 static inline void __skb_insert(struct sk_buff *newsk,
1246 struct sk_buff *prev, struct sk_buff *next,
1247 struct sk_buff_head *list)
1251 next->prev = prev->next = newsk;
1255 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1256 struct sk_buff *prev,
1257 struct sk_buff *next)
1259 struct sk_buff *first = list->next;
1260 struct sk_buff *last = list->prev;
1270 * skb_queue_splice - join two skb lists, this is designed for stacks
1271 * @list: the new list to add
1272 * @head: the place to add it in the first list
1274 static inline void skb_queue_splice(const struct sk_buff_head *list,
1275 struct sk_buff_head *head)
1277 if (!skb_queue_empty(list)) {
1278 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1279 head->qlen += list->qlen;
1284 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1285 * @list: the new list to add
1286 * @head: the place to add it in the first list
1288 * The list at @list is reinitialised
1290 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1291 struct sk_buff_head *head)
1293 if (!skb_queue_empty(list)) {
1294 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1295 head->qlen += list->qlen;
1296 __skb_queue_head_init(list);
1301 * skb_queue_splice_tail - join two skb lists, each list being a queue
1302 * @list: the new list to add
1303 * @head: the place to add it in the first list
1305 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1306 struct sk_buff_head *head)
1308 if (!skb_queue_empty(list)) {
1309 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1310 head->qlen += list->qlen;
1315 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1316 * @list: the new list to add
1317 * @head: the place to add it in the first list
1319 * Each of the lists is a queue.
1320 * The list at @list is reinitialised
1322 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1323 struct sk_buff_head *head)
1325 if (!skb_queue_empty(list)) {
1326 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1327 head->qlen += list->qlen;
1328 __skb_queue_head_init(list);
1333 * __skb_queue_after - queue a buffer at the list head
1334 * @list: list to use
1335 * @prev: place after this buffer
1336 * @newsk: buffer to queue
1338 * Queue a buffer int the middle of a list. This function takes no locks
1339 * and you must therefore hold required locks before calling it.
1341 * A buffer cannot be placed on two lists at the same time.
1343 static inline void __skb_queue_after(struct sk_buff_head *list,
1344 struct sk_buff *prev,
1345 struct sk_buff *newsk)
1347 __skb_insert(newsk, prev, prev->next, list);
1350 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1351 struct sk_buff_head *list);
1353 static inline void __skb_queue_before(struct sk_buff_head *list,
1354 struct sk_buff *next,
1355 struct sk_buff *newsk)
1357 __skb_insert(newsk, next->prev, next, list);
1361 * __skb_queue_head - queue a buffer at the list head
1362 * @list: list to use
1363 * @newsk: buffer to queue
1365 * Queue a buffer at the start of a list. This function takes no locks
1366 * and you must therefore hold required locks before calling it.
1368 * A buffer cannot be placed on two lists at the same time.
1370 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1371 static inline void __skb_queue_head(struct sk_buff_head *list,
1372 struct sk_buff *newsk)
1374 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1378 * __skb_queue_tail - queue a buffer at the list tail
1379 * @list: list to use
1380 * @newsk: buffer to queue
1382 * Queue a buffer at the end of a list. This function takes no locks
1383 * and you must therefore hold required locks before calling it.
1385 * A buffer cannot be placed on two lists at the same time.
1387 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1388 static inline void __skb_queue_tail(struct sk_buff_head *list,
1389 struct sk_buff *newsk)
1391 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1395 * remove sk_buff from list. _Must_ be called atomically, and with
1398 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1399 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1401 struct sk_buff *next, *prev;
1406 skb->next = skb->prev = NULL;
1412 * __skb_dequeue - remove from the head of the queue
1413 * @list: list to dequeue from
1415 * Remove the head of the list. This function does not take any locks
1416 * so must be used with appropriate locks held only. The head item is
1417 * returned or %NULL if the list is empty.
1419 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1420 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1422 struct sk_buff *skb = skb_peek(list);
1424 __skb_unlink(skb, list);
1429 * __skb_dequeue_tail - remove from the tail of the queue
1430 * @list: list to dequeue from
1432 * Remove the tail of the list. This function does not take any locks
1433 * so must be used with appropriate locks held only. The tail item is
1434 * returned or %NULL if the list is empty.
1436 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1437 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1439 struct sk_buff *skb = skb_peek_tail(list);
1441 __skb_unlink(skb, list);
1446 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1448 return skb->data_len;
1451 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1453 return skb->len - skb->data_len;
1456 static inline int skb_pagelen(const struct sk_buff *skb)
1460 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1461 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1462 return len + skb_headlen(skb);
1466 * __skb_fill_page_desc - initialise a paged fragment in an skb
1467 * @skb: buffer containing fragment to be initialised
1468 * @i: paged fragment index to initialise
1469 * @page: the page to use for this fragment
1470 * @off: the offset to the data with @page
1471 * @size: the length of the data
1473 * Initialises the @i'th fragment of @skb to point to &size bytes at
1474 * offset @off within @page.
1476 * Does not take any additional reference on the fragment.
1478 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1479 struct page *page, int off, int size)
1481 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1484 * Propagate page->pfmemalloc to the skb if we can. The problem is
1485 * that not all callers have unique ownership of the page. If
1486 * pfmemalloc is set, we check the mapping as a mapping implies
1487 * page->index is set (index and pfmemalloc share space).
1488 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1489 * do not lose pfmemalloc information as the pages would not be
1490 * allocated using __GFP_MEMALLOC.
1492 frag->page.p = page;
1493 frag->page_offset = off;
1494 skb_frag_size_set(frag, size);
1496 page = compound_head(page);
1497 if (page->pfmemalloc && !page->mapping)
1498 skb->pfmemalloc = true;
1502 * skb_fill_page_desc - initialise a paged fragment in an skb
1503 * @skb: buffer containing fragment to be initialised
1504 * @i: paged fragment index to initialise
1505 * @page: the page to use for this fragment
1506 * @off: the offset to the data with @page
1507 * @size: the length of the data
1509 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1510 * @skb to point to @size bytes at offset @off within @page. In
1511 * addition updates @skb such that @i is the last fragment.
1513 * Does not take any additional reference on the fragment.
1515 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1516 struct page *page, int off, int size)
1518 __skb_fill_page_desc(skb, i, page, off, size);
1519 skb_shinfo(skb)->nr_frags = i + 1;
1522 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1523 int size, unsigned int truesize);
1525 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1526 unsigned int truesize);
1528 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1529 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1530 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1532 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1533 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1535 return skb->head + skb->tail;
1538 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1540 skb->tail = skb->data - skb->head;
1543 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1545 skb_reset_tail_pointer(skb);
1546 skb->tail += offset;
1549 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1550 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1555 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1557 skb->tail = skb->data;
1560 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1562 skb->tail = skb->data + offset;
1565 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1568 * Add data to an sk_buff
1570 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1571 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1572 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1574 unsigned char *tmp = skb_tail_pointer(skb);
1575 SKB_LINEAR_ASSERT(skb);
1581 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1582 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1589 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1590 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1593 BUG_ON(skb->len < skb->data_len);
1594 return skb->data += len;
1597 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1599 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1602 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1604 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1606 if (len > skb_headlen(skb) &&
1607 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1610 return skb->data += len;
1613 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1615 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1618 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1620 if (likely(len <= skb_headlen(skb)))
1622 if (unlikely(len > skb->len))
1624 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1628 * skb_headroom - bytes at buffer head
1629 * @skb: buffer to check
1631 * Return the number of bytes of free space at the head of an &sk_buff.
1633 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1635 return skb->data - skb->head;
1639 * skb_tailroom - bytes at buffer end
1640 * @skb: buffer to check
1642 * Return the number of bytes of free space at the tail of an sk_buff
1644 static inline int skb_tailroom(const struct sk_buff *skb)
1646 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1650 * skb_availroom - bytes at buffer end
1651 * @skb: buffer to check
1653 * Return the number of bytes of free space at the tail of an sk_buff
1654 * allocated by sk_stream_alloc()
1656 static inline int skb_availroom(const struct sk_buff *skb)
1658 if (skb_is_nonlinear(skb))
1661 return skb->end - skb->tail - skb->reserved_tailroom;
1665 * skb_reserve - adjust headroom
1666 * @skb: buffer to alter
1667 * @len: bytes to move
1669 * Increase the headroom of an empty &sk_buff by reducing the tail
1670 * room. This is only allowed for an empty buffer.
1672 static inline void skb_reserve(struct sk_buff *skb, int len)
1678 static inline void skb_reset_inner_headers(struct sk_buff *skb)
1680 skb->inner_mac_header = skb->mac_header;
1681 skb->inner_network_header = skb->network_header;
1682 skb->inner_transport_header = skb->transport_header;
1685 static inline void skb_reset_mac_len(struct sk_buff *skb)
1687 skb->mac_len = skb->network_header - skb->mac_header;
1690 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1693 return skb->head + skb->inner_transport_header;
1696 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1698 skb->inner_transport_header = skb->data - skb->head;
1701 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1704 skb_reset_inner_transport_header(skb);
1705 skb->inner_transport_header += offset;
1708 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1710 return skb->head + skb->inner_network_header;
1713 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1715 skb->inner_network_header = skb->data - skb->head;
1718 static inline void skb_set_inner_network_header(struct sk_buff *skb,
1721 skb_reset_inner_network_header(skb);
1722 skb->inner_network_header += offset;
1725 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1727 return skb->head + skb->inner_mac_header;
1730 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1732 skb->inner_mac_header = skb->data - skb->head;
1735 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1738 skb_reset_inner_mac_header(skb);
1739 skb->inner_mac_header += offset;
1741 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1743 return skb->transport_header != (typeof(skb->transport_header))~0U;
1746 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1748 return skb->head + skb->transport_header;
1751 static inline void skb_reset_transport_header(struct sk_buff *skb)
1753 skb->transport_header = skb->data - skb->head;
1756 static inline void skb_set_transport_header(struct sk_buff *skb,
1759 skb_reset_transport_header(skb);
1760 skb->transport_header += offset;
1763 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1765 return skb->head + skb->network_header;
1768 static inline void skb_reset_network_header(struct sk_buff *skb)
1770 skb->network_header = skb->data - skb->head;
1773 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1775 skb_reset_network_header(skb);
1776 skb->network_header += offset;
1779 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1781 return skb->head + skb->mac_header;
1784 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1786 return skb->mac_header != (typeof(skb->mac_header))~0U;
1789 static inline void skb_reset_mac_header(struct sk_buff *skb)
1791 skb->mac_header = skb->data - skb->head;
1794 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1796 skb_reset_mac_header(skb);
1797 skb->mac_header += offset;
1800 static inline void skb_pop_mac_header(struct sk_buff *skb)
1802 skb->mac_header = skb->network_header;
1805 static inline void skb_probe_transport_header(struct sk_buff *skb,
1806 const int offset_hint)
1808 struct flow_keys keys;
1810 if (skb_transport_header_was_set(skb))
1812 else if (skb_flow_dissect(skb, &keys))
1813 skb_set_transport_header(skb, keys.thoff);
1815 skb_set_transport_header(skb, offset_hint);
1818 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1820 if (skb_mac_header_was_set(skb)) {
1821 const unsigned char *old_mac = skb_mac_header(skb);
1823 skb_set_mac_header(skb, -skb->mac_len);
1824 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
1828 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
1830 return skb->csum_start - skb_headroom(skb);
1833 static inline int skb_transport_offset(const struct sk_buff *skb)
1835 return skb_transport_header(skb) - skb->data;
1838 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1840 return skb->transport_header - skb->network_header;
1843 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
1845 return skb->inner_transport_header - skb->inner_network_header;
1848 static inline int skb_network_offset(const struct sk_buff *skb)
1850 return skb_network_header(skb) - skb->data;
1853 static inline int skb_inner_network_offset(const struct sk_buff *skb)
1855 return skb_inner_network_header(skb) - skb->data;
1858 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1860 return pskb_may_pull(skb, skb_network_offset(skb) + len);
1863 static inline void skb_pop_rcv_encapsulation(struct sk_buff *skb)
1865 /* Only continue with checksum unnecessary if device indicated
1866 * it is valid across encapsulation (skb->encapsulation was set).
1868 if (skb->ip_summed == CHECKSUM_UNNECESSARY && !skb->encapsulation)
1869 skb->ip_summed = CHECKSUM_NONE;
1871 skb->encapsulation = 0;
1872 skb->csum_valid = 0;
1876 * CPUs often take a performance hit when accessing unaligned memory
1877 * locations. The actual performance hit varies, it can be small if the
1878 * hardware handles it or large if we have to take an exception and fix it
1881 * Since an ethernet header is 14 bytes network drivers often end up with
1882 * the IP header at an unaligned offset. The IP header can be aligned by
1883 * shifting the start of the packet by 2 bytes. Drivers should do this
1886 * skb_reserve(skb, NET_IP_ALIGN);
1888 * The downside to this alignment of the IP header is that the DMA is now
1889 * unaligned. On some architectures the cost of an unaligned DMA is high
1890 * and this cost outweighs the gains made by aligning the IP header.
1892 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1895 #ifndef NET_IP_ALIGN
1896 #define NET_IP_ALIGN 2
1900 * The networking layer reserves some headroom in skb data (via
1901 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
1902 * the header has to grow. In the default case, if the header has to grow
1903 * 32 bytes or less we avoid the reallocation.
1905 * Unfortunately this headroom changes the DMA alignment of the resulting
1906 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
1907 * on some architectures. An architecture can override this value,
1908 * perhaps setting it to a cacheline in size (since that will maintain
1909 * cacheline alignment of the DMA). It must be a power of 2.
1911 * Various parts of the networking layer expect at least 32 bytes of
1912 * headroom, you should not reduce this.
1914 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
1915 * to reduce average number of cache lines per packet.
1916 * get_rps_cpus() for example only access one 64 bytes aligned block :
1917 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
1920 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
1923 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
1925 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
1927 if (unlikely(skb_is_nonlinear(skb))) {
1932 skb_set_tail_pointer(skb, len);
1935 void skb_trim(struct sk_buff *skb, unsigned int len);
1937 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
1940 return ___pskb_trim(skb, len);
1941 __skb_trim(skb, len);
1945 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
1947 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
1951 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
1952 * @skb: buffer to alter
1955 * This is identical to pskb_trim except that the caller knows that
1956 * the skb is not cloned so we should never get an error due to out-
1959 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
1961 int err = pskb_trim(skb, len);
1966 * skb_orphan - orphan a buffer
1967 * @skb: buffer to orphan
1969 * If a buffer currently has an owner then we call the owner's
1970 * destructor function and make the @skb unowned. The buffer continues
1971 * to exist but is no longer charged to its former owner.
1973 static inline void skb_orphan(struct sk_buff *skb)
1975 if (skb->destructor) {
1976 skb->destructor(skb);
1977 skb->destructor = NULL;
1985 * skb_orphan_frags - orphan the frags contained in a buffer
1986 * @skb: buffer to orphan frags from
1987 * @gfp_mask: allocation mask for replacement pages
1989 * For each frag in the SKB which needs a destructor (i.e. has an
1990 * owner) create a copy of that frag and release the original
1991 * page by calling the destructor.
1993 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
1995 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
1997 return skb_copy_ubufs(skb, gfp_mask);
2001 * __skb_queue_purge - empty a list
2002 * @list: list to empty
2004 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2005 * the list and one reference dropped. This function does not take the
2006 * list lock and the caller must hold the relevant locks to use it.
2008 void skb_queue_purge(struct sk_buff_head *list);
2009 static inline void __skb_queue_purge(struct sk_buff_head *list)
2011 struct sk_buff *skb;
2012 while ((skb = __skb_dequeue(list)) != NULL)
2016 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
2017 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
2018 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
2020 void *netdev_alloc_frag(unsigned int fragsz);
2022 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2026 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2027 * @dev: network device to receive on
2028 * @length: length to allocate
2030 * Allocate a new &sk_buff and assign it a usage count of one. The
2031 * buffer has unspecified headroom built in. Users should allocate
2032 * the headroom they think they need without accounting for the
2033 * built in space. The built in space is used for optimisations.
2035 * %NULL is returned if there is no free memory. Although this function
2036 * allocates memory it can be called from an interrupt.
2038 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2039 unsigned int length)
2041 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2044 /* legacy helper around __netdev_alloc_skb() */
2045 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2048 return __netdev_alloc_skb(NULL, length, gfp_mask);
2051 /* legacy helper around netdev_alloc_skb() */
2052 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2054 return netdev_alloc_skb(NULL, length);
2058 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2059 unsigned int length, gfp_t gfp)
2061 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2063 if (NET_IP_ALIGN && skb)
2064 skb_reserve(skb, NET_IP_ALIGN);
2068 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2069 unsigned int length)
2071 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2075 * __skb_alloc_pages - allocate pages for ps-rx on a skb and preserve pfmemalloc data
2076 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2077 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2078 * @order: size of the allocation
2080 * Allocate a new page.
2082 * %NULL is returned if there is no free memory.
2084 static inline struct page *__skb_alloc_pages(gfp_t gfp_mask,
2085 struct sk_buff *skb,
2090 gfp_mask |= __GFP_COLD;
2092 if (!(gfp_mask & __GFP_NOMEMALLOC))
2093 gfp_mask |= __GFP_MEMALLOC;
2095 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2096 if (skb && page && page->pfmemalloc)
2097 skb->pfmemalloc = true;
2103 * __skb_alloc_page - allocate a page for ps-rx for a given skb and preserve pfmemalloc data
2104 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2105 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2107 * Allocate a new page.
2109 * %NULL is returned if there is no free memory.
2111 static inline struct page *__skb_alloc_page(gfp_t gfp_mask,
2112 struct sk_buff *skb)
2114 return __skb_alloc_pages(gfp_mask, skb, 0);
2118 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2119 * @page: The page that was allocated from skb_alloc_page
2120 * @skb: The skb that may need pfmemalloc set
2122 static inline void skb_propagate_pfmemalloc(struct page *page,
2123 struct sk_buff *skb)
2125 if (page && page->pfmemalloc)
2126 skb->pfmemalloc = true;
2130 * skb_frag_page - retrieve the page referred to by a paged fragment
2131 * @frag: the paged fragment
2133 * Returns the &struct page associated with @frag.
2135 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2137 return frag->page.p;
2141 * __skb_frag_ref - take an addition reference on a paged fragment.
2142 * @frag: the paged fragment
2144 * Takes an additional reference on the paged fragment @frag.
2146 static inline void __skb_frag_ref(skb_frag_t *frag)
2148 get_page(skb_frag_page(frag));
2152 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2154 * @f: the fragment offset.
2156 * Takes an additional reference on the @f'th paged fragment of @skb.
2158 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2160 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2164 * __skb_frag_unref - release a reference on a paged fragment.
2165 * @frag: the paged fragment
2167 * Releases a reference on the paged fragment @frag.
2169 static inline void __skb_frag_unref(skb_frag_t *frag)
2171 put_page(skb_frag_page(frag));
2175 * skb_frag_unref - release a reference on a paged fragment of an skb.
2177 * @f: the fragment offset
2179 * Releases a reference on the @f'th paged fragment of @skb.
2181 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2183 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2187 * skb_frag_address - gets the address of the data contained in a paged fragment
2188 * @frag: the paged fragment buffer
2190 * Returns the address of the data within @frag. The page must already
2193 static inline void *skb_frag_address(const skb_frag_t *frag)
2195 return page_address(skb_frag_page(frag)) + frag->page_offset;
2199 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2200 * @frag: the paged fragment buffer
2202 * Returns the address of the data within @frag. Checks that the page
2203 * is mapped and returns %NULL otherwise.
2205 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2207 void *ptr = page_address(skb_frag_page(frag));
2211 return ptr + frag->page_offset;
2215 * __skb_frag_set_page - sets the page contained in a paged fragment
2216 * @frag: the paged fragment
2217 * @page: the page to set
2219 * Sets the fragment @frag to contain @page.
2221 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2223 frag->page.p = page;
2227 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2229 * @f: the fragment offset
2230 * @page: the page to set
2232 * Sets the @f'th fragment of @skb to contain @page.
2234 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2237 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2240 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2243 * skb_frag_dma_map - maps a paged fragment via the DMA API
2244 * @dev: the device to map the fragment to
2245 * @frag: the paged fragment to map
2246 * @offset: the offset within the fragment (starting at the
2247 * fragment's own offset)
2248 * @size: the number of bytes to map
2249 * @dir: the direction of the mapping (%PCI_DMA_*)
2251 * Maps the page associated with @frag to @device.
2253 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2254 const skb_frag_t *frag,
2255 size_t offset, size_t size,
2256 enum dma_data_direction dir)
2258 return dma_map_page(dev, skb_frag_page(frag),
2259 frag->page_offset + offset, size, dir);
2262 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2265 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2269 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2272 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2277 * skb_clone_writable - is the header of a clone writable
2278 * @skb: buffer to check
2279 * @len: length up to which to write
2281 * Returns true if modifying the header part of the cloned buffer
2282 * does not requires the data to be copied.
2284 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2286 return !skb_header_cloned(skb) &&
2287 skb_headroom(skb) + len <= skb->hdr_len;
2290 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2295 if (headroom > skb_headroom(skb))
2296 delta = headroom - skb_headroom(skb);
2298 if (delta || cloned)
2299 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2305 * skb_cow - copy header of skb when it is required
2306 * @skb: buffer to cow
2307 * @headroom: needed headroom
2309 * If the skb passed lacks sufficient headroom or its data part
2310 * is shared, data is reallocated. If reallocation fails, an error
2311 * is returned and original skb is not changed.
2313 * The result is skb with writable area skb->head...skb->tail
2314 * and at least @headroom of space at head.
2316 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2318 return __skb_cow(skb, headroom, skb_cloned(skb));
2322 * skb_cow_head - skb_cow but only making the head writable
2323 * @skb: buffer to cow
2324 * @headroom: needed headroom
2326 * This function is identical to skb_cow except that we replace the
2327 * skb_cloned check by skb_header_cloned. It should be used when
2328 * you only need to push on some header and do not need to modify
2331 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2333 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2337 * skb_padto - pad an skbuff up to a minimal size
2338 * @skb: buffer to pad
2339 * @len: minimal length
2341 * Pads up a buffer to ensure the trailing bytes exist and are
2342 * blanked. If the buffer already contains sufficient data it
2343 * is untouched. Otherwise it is extended. Returns zero on
2344 * success. The skb is freed on error.
2347 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2349 unsigned int size = skb->len;
2350 if (likely(size >= len))
2352 return skb_pad(skb, len - size);
2355 static inline int skb_add_data(struct sk_buff *skb,
2356 char __user *from, int copy)
2358 const int off = skb->len;
2360 if (skb->ip_summed == CHECKSUM_NONE) {
2362 __wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
2365 skb->csum = csum_block_add(skb->csum, csum, off);
2368 } else if (!copy_from_user(skb_put(skb, copy), from, copy))
2371 __skb_trim(skb, off);
2375 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2376 const struct page *page, int off)
2379 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2381 return page == skb_frag_page(frag) &&
2382 off == frag->page_offset + skb_frag_size(frag);
2387 static inline int __skb_linearize(struct sk_buff *skb)
2389 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2393 * skb_linearize - convert paged skb to linear one
2394 * @skb: buffer to linarize
2396 * If there is no free memory -ENOMEM is returned, otherwise zero
2397 * is returned and the old skb data released.
2399 static inline int skb_linearize(struct sk_buff *skb)
2401 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2405 * skb_has_shared_frag - can any frag be overwritten
2406 * @skb: buffer to test
2408 * Return true if the skb has at least one frag that might be modified
2409 * by an external entity (as in vmsplice()/sendfile())
2411 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2413 return skb_is_nonlinear(skb) &&
2414 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2418 * skb_linearize_cow - make sure skb is linear and writable
2419 * @skb: buffer to process
2421 * If there is no free memory -ENOMEM is returned, otherwise zero
2422 * is returned and the old skb data released.
2424 static inline int skb_linearize_cow(struct sk_buff *skb)
2426 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2427 __skb_linearize(skb) : 0;
2431 * skb_postpull_rcsum - update checksum for received skb after pull
2432 * @skb: buffer to update
2433 * @start: start of data before pull
2434 * @len: length of data pulled
2436 * After doing a pull on a received packet, you need to call this to
2437 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2438 * CHECKSUM_NONE so that it can be recomputed from scratch.
2441 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2442 const void *start, unsigned int len)
2444 if (skb->ip_summed == CHECKSUM_COMPLETE)
2445 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2448 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2451 * pskb_trim_rcsum - trim received skb and update checksum
2452 * @skb: buffer to trim
2455 * This is exactly the same as pskb_trim except that it ensures the
2456 * checksum of received packets are still valid after the operation.
2459 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2461 if (likely(len >= skb->len))
2463 if (skb->ip_summed == CHECKSUM_COMPLETE)
2464 skb->ip_summed = CHECKSUM_NONE;
2465 return __pskb_trim(skb, len);
2468 #define skb_queue_walk(queue, skb) \
2469 for (skb = (queue)->next; \
2470 skb != (struct sk_buff *)(queue); \
2473 #define skb_queue_walk_safe(queue, skb, tmp) \
2474 for (skb = (queue)->next, tmp = skb->next; \
2475 skb != (struct sk_buff *)(queue); \
2476 skb = tmp, tmp = skb->next)
2478 #define skb_queue_walk_from(queue, skb) \
2479 for (; skb != (struct sk_buff *)(queue); \
2482 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2483 for (tmp = skb->next; \
2484 skb != (struct sk_buff *)(queue); \
2485 skb = tmp, tmp = skb->next)
2487 #define skb_queue_reverse_walk(queue, skb) \
2488 for (skb = (queue)->prev; \
2489 skb != (struct sk_buff *)(queue); \
2492 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2493 for (skb = (queue)->prev, tmp = skb->prev; \
2494 skb != (struct sk_buff *)(queue); \
2495 skb = tmp, tmp = skb->prev)
2497 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2498 for (tmp = skb->prev; \
2499 skb != (struct sk_buff *)(queue); \
2500 skb = tmp, tmp = skb->prev)
2502 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2504 return skb_shinfo(skb)->frag_list != NULL;
2507 static inline void skb_frag_list_init(struct sk_buff *skb)
2509 skb_shinfo(skb)->frag_list = NULL;
2512 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
2514 frag->next = skb_shinfo(skb)->frag_list;
2515 skb_shinfo(skb)->frag_list = frag;
2518 #define skb_walk_frags(skb, iter) \
2519 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2521 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2522 int *peeked, int *off, int *err);
2523 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2525 unsigned int datagram_poll(struct file *file, struct socket *sock,
2526 struct poll_table_struct *wait);
2527 int skb_copy_datagram_iovec(const struct sk_buff *from, int offset,
2528 struct iovec *to, int size);
2529 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, int hlen,
2531 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
2532 const struct iovec *from, int from_offset,
2534 int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *frm,
2535 int offset, size_t count);
2536 int skb_copy_datagram_const_iovec(const struct sk_buff *from, int offset,
2537 const struct iovec *to, int to_offset,
2539 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2540 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2541 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2542 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2543 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2544 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2545 int len, __wsum csum);
2546 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
2547 struct pipe_inode_info *pipe, unsigned int len,
2548 unsigned int flags);
2549 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2550 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2551 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2553 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2554 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2555 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2556 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2557 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2558 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
2560 struct skb_checksum_ops {
2561 __wsum (*update)(const void *mem, int len, __wsum wsum);
2562 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2565 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2566 __wsum csum, const struct skb_checksum_ops *ops);
2567 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2570 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
2571 int len, void *buffer)
2573 int hlen = skb_headlen(skb);
2575 if (hlen - offset >= len)
2576 return skb->data + offset;
2578 if (skb_copy_bits(skb, offset, buffer, len) < 0)
2585 * skb_needs_linearize - check if we need to linearize a given skb
2586 * depending on the given device features.
2587 * @skb: socket buffer to check
2588 * @features: net device features
2590 * Returns true if either:
2591 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2592 * 2. skb is fragmented and the device does not support SG.
2594 static inline bool skb_needs_linearize(struct sk_buff *skb,
2595 netdev_features_t features)
2597 return skb_is_nonlinear(skb) &&
2598 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2599 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2602 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2604 const unsigned int len)
2606 memcpy(to, skb->data, len);
2609 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2610 const int offset, void *to,
2611 const unsigned int len)
2613 memcpy(to, skb->data + offset, len);
2616 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2618 const unsigned int len)
2620 memcpy(skb->data, from, len);
2623 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2626 const unsigned int len)
2628 memcpy(skb->data + offset, from, len);
2631 void skb_init(void);
2633 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2639 * skb_get_timestamp - get timestamp from a skb
2640 * @skb: skb to get stamp from
2641 * @stamp: pointer to struct timeval to store stamp in
2643 * Timestamps are stored in the skb as offsets to a base timestamp.
2644 * This function converts the offset back to a struct timeval and stores
2647 static inline void skb_get_timestamp(const struct sk_buff *skb,
2648 struct timeval *stamp)
2650 *stamp = ktime_to_timeval(skb->tstamp);
2653 static inline void skb_get_timestampns(const struct sk_buff *skb,
2654 struct timespec *stamp)
2656 *stamp = ktime_to_timespec(skb->tstamp);
2659 static inline void __net_timestamp(struct sk_buff *skb)
2661 skb->tstamp = ktime_get_real();
2664 static inline ktime_t net_timedelta(ktime_t t)
2666 return ktime_sub(ktime_get_real(), t);
2669 static inline ktime_t net_invalid_timestamp(void)
2671 return ktime_set(0, 0);
2674 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2676 void skb_clone_tx_timestamp(struct sk_buff *skb);
2677 bool skb_defer_rx_timestamp(struct sk_buff *skb);
2679 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2681 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2685 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2690 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2693 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2695 * PHY drivers may accept clones of transmitted packets for
2696 * timestamping via their phy_driver.txtstamp method. These drivers
2697 * must call this function to return the skb back to the stack, with
2698 * or without a timestamp.
2700 * @skb: clone of the the original outgoing packet
2701 * @hwtstamps: hardware time stamps, may be NULL if not available
2704 void skb_complete_tx_timestamp(struct sk_buff *skb,
2705 struct skb_shared_hwtstamps *hwtstamps);
2707 void __skb_tstamp_tx(struct sk_buff *orig_skb,
2708 struct skb_shared_hwtstamps *hwtstamps,
2709 struct sock *sk, int tstype);
2712 * skb_tstamp_tx - queue clone of skb with send time stamps
2713 * @orig_skb: the original outgoing packet
2714 * @hwtstamps: hardware time stamps, may be NULL if not available
2716 * If the skb has a socket associated, then this function clones the
2717 * skb (thus sharing the actual data and optional structures), stores
2718 * the optional hardware time stamping information (if non NULL) or
2719 * generates a software time stamp (otherwise), then queues the clone
2720 * to the error queue of the socket. Errors are silently ignored.
2722 void skb_tstamp_tx(struct sk_buff *orig_skb,
2723 struct skb_shared_hwtstamps *hwtstamps);
2725 static inline void sw_tx_timestamp(struct sk_buff *skb)
2727 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2728 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
2729 skb_tstamp_tx(skb, NULL);
2733 * skb_tx_timestamp() - Driver hook for transmit timestamping
2735 * Ethernet MAC Drivers should call this function in their hard_xmit()
2736 * function immediately before giving the sk_buff to the MAC hardware.
2738 * Specifically, one should make absolutely sure that this function is
2739 * called before TX completion of this packet can trigger. Otherwise
2740 * the packet could potentially already be freed.
2742 * @skb: A socket buffer.
2744 static inline void skb_tx_timestamp(struct sk_buff *skb)
2746 skb_clone_tx_timestamp(skb);
2747 sw_tx_timestamp(skb);
2751 * skb_complete_wifi_ack - deliver skb with wifi status
2753 * @skb: the original outgoing packet
2754 * @acked: ack status
2757 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2759 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2760 __sum16 __skb_checksum_complete(struct sk_buff *skb);
2762 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2764 return ((skb->ip_summed & CHECKSUM_UNNECESSARY) || skb->csum_valid);
2768 * skb_checksum_complete - Calculate checksum of an entire packet
2769 * @skb: packet to process
2771 * This function calculates the checksum over the entire packet plus
2772 * the value of skb->csum. The latter can be used to supply the
2773 * checksum of a pseudo header as used by TCP/UDP. It returns the
2776 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2777 * this function can be used to verify that checksum on received
2778 * packets. In that case the function should return zero if the
2779 * checksum is correct. In particular, this function will return zero
2780 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2781 * hardware has already verified the correctness of the checksum.
2783 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
2785 return skb_csum_unnecessary(skb) ?
2786 0 : __skb_checksum_complete(skb);
2789 /* Check if we need to perform checksum complete validation.
2791 * Returns true if checksum complete is needed, false otherwise
2792 * (either checksum is unnecessary or zero checksum is allowed).
2794 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
2798 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
2799 skb->csum_valid = 1;
2806 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
2809 #define CHECKSUM_BREAK 76
2811 /* Validate (init) checksum based on checksum complete.
2814 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
2815 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
2816 * checksum is stored in skb->csum for use in __skb_checksum_complete
2817 * non-zero: value of invalid checksum
2820 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
2824 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2825 if (!csum_fold(csum_add(psum, skb->csum))) {
2826 skb->csum_valid = 1;
2833 if (complete || skb->len <= CHECKSUM_BREAK) {
2836 csum = __skb_checksum_complete(skb);
2837 skb->csum_valid = !csum;
2844 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
2849 /* Perform checksum validate (init). Note that this is a macro since we only
2850 * want to calculate the pseudo header which is an input function if necessary.
2851 * First we try to validate without any computation (checksum unnecessary) and
2852 * then calculate based on checksum complete calling the function to compute
2856 * 0: checksum is validated or try to in skb_checksum_complete
2857 * non-zero: value of invalid checksum
2859 #define __skb_checksum_validate(skb, proto, complete, \
2860 zero_okay, check, compute_pseudo) \
2862 __sum16 __ret = 0; \
2863 skb->csum_valid = 0; \
2864 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
2865 __ret = __skb_checksum_validate_complete(skb, \
2866 complete, compute_pseudo(skb, proto)); \
2870 #define skb_checksum_init(skb, proto, compute_pseudo) \
2871 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
2873 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
2874 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
2876 #define skb_checksum_validate(skb, proto, compute_pseudo) \
2877 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
2879 #define skb_checksum_validate_zero_check(skb, proto, check, \
2881 __skb_checksum_validate_(skb, proto, true, true, check, compute_pseudo)
2883 #define skb_checksum_simple_validate(skb) \
2884 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
2886 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2887 void nf_conntrack_destroy(struct nf_conntrack *nfct);
2888 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
2890 if (nfct && atomic_dec_and_test(&nfct->use))
2891 nf_conntrack_destroy(nfct);
2893 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
2896 atomic_inc(&nfct->use);
2899 #ifdef CONFIG_BRIDGE_NETFILTER
2900 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
2902 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
2905 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
2908 atomic_inc(&nf_bridge->use);
2910 #endif /* CONFIG_BRIDGE_NETFILTER */
2911 static inline void nf_reset(struct sk_buff *skb)
2913 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2914 nf_conntrack_put(skb->nfct);
2917 #ifdef CONFIG_BRIDGE_NETFILTER
2918 nf_bridge_put(skb->nf_bridge);
2919 skb->nf_bridge = NULL;
2923 static inline void nf_reset_trace(struct sk_buff *skb)
2925 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
2930 /* Note: This doesn't put any conntrack and bridge info in dst. */
2931 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2933 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2934 dst->nfct = src->nfct;
2935 nf_conntrack_get(src->nfct);
2936 dst->nfctinfo = src->nfctinfo;
2938 #ifdef CONFIG_BRIDGE_NETFILTER
2939 dst->nf_bridge = src->nf_bridge;
2940 nf_bridge_get(src->nf_bridge);
2942 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
2943 dst->nf_trace = src->nf_trace;
2947 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2949 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2950 nf_conntrack_put(dst->nfct);
2952 #ifdef CONFIG_BRIDGE_NETFILTER
2953 nf_bridge_put(dst->nf_bridge);
2955 __nf_copy(dst, src);
2958 #ifdef CONFIG_NETWORK_SECMARK
2959 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2961 to->secmark = from->secmark;
2964 static inline void skb_init_secmark(struct sk_buff *skb)
2969 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2972 static inline void skb_init_secmark(struct sk_buff *skb)
2976 static inline bool skb_irq_freeable(const struct sk_buff *skb)
2978 return !skb->destructor &&
2979 #if IS_ENABLED(CONFIG_XFRM)
2982 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
2985 !skb->_skb_refdst &&
2986 !skb_has_frag_list(skb);
2989 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
2991 skb->queue_mapping = queue_mapping;
2994 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
2996 return skb->queue_mapping;
2999 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3001 to->queue_mapping = from->queue_mapping;
3004 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3006 skb->queue_mapping = rx_queue + 1;
3009 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3011 return skb->queue_mapping - 1;
3014 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3016 return skb->queue_mapping != 0;
3019 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
3020 unsigned int num_tx_queues);
3022 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3031 /* Keeps track of mac header offset relative to skb->head.
3032 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3033 * For non-tunnel skb it points to skb_mac_header() and for
3034 * tunnel skb it points to outer mac header.
3035 * Keeps track of level of encapsulation of network headers.
3042 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3044 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3046 return (skb_mac_header(inner_skb) - inner_skb->head) -
3047 SKB_GSO_CB(inner_skb)->mac_offset;
3050 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3052 int new_headroom, headroom;
3055 headroom = skb_headroom(skb);
3056 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3060 new_headroom = skb_headroom(skb);
3061 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3065 /* Compute the checksum for a gso segment. First compute the checksum value
3066 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3067 * then add in skb->csum (checksum from csum_start to end of packet).
3068 * skb->csum and csum_start are then updated to reflect the checksum of the
3069 * resultant packet starting from the transport header-- the resultant checksum
3070 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3073 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3075 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3076 skb_transport_offset(skb);
3079 csum = csum_fold(csum_partial(skb_transport_header(skb),
3082 SKB_GSO_CB(skb)->csum_start -= plen;
3087 static inline bool skb_is_gso(const struct sk_buff *skb)
3089 return skb_shinfo(skb)->gso_size;
3092 /* Note: Should be called only if skb_is_gso(skb) is true */
3093 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3095 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3098 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3100 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3102 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3103 * wanted then gso_type will be set. */
3104 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3106 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3107 unlikely(shinfo->gso_type == 0)) {
3108 __skb_warn_lro_forwarding(skb);
3114 static inline void skb_forward_csum(struct sk_buff *skb)
3116 /* Unfortunately we don't support this one. Any brave souls? */
3117 if (skb->ip_summed == CHECKSUM_COMPLETE)
3118 skb->ip_summed = CHECKSUM_NONE;
3122 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3123 * @skb: skb to check
3125 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3126 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3127 * use this helper, to document places where we make this assertion.
3129 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3132 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3136 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3138 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3140 u32 __skb_get_poff(const struct sk_buff *skb);
3143 * skb_head_is_locked - Determine if the skb->head is locked down
3144 * @skb: skb to check
3146 * The head on skbs build around a head frag can be removed if they are
3147 * not cloned. This function returns true if the skb head is locked down
3148 * due to either being allocated via kmalloc, or by being a clone with
3149 * multiple references to the head.
3151 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3153 return !skb->head_frag || skb_cloned(skb);
3157 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3161 * skb_gso_network_seglen is used to determine the real size of the
3162 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3164 * The MAC/L2 header is not accounted for.
3166 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3168 unsigned int hdr_len = skb_transport_header(skb) -
3169 skb_network_header(skb);
3170 return hdr_len + skb_gso_transport_seglen(skb);
3172 #endif /* __KERNEL__ */
3173 #endif /* _LINUX_SKBUFF_H */