2 * Definitions for the 'struct sk_buff' memory handlers.
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
17 #include <linux/kernel.h>
18 #include <linux/kmemcheck.h>
19 #include <linux/compiler.h>
20 #include <linux/time.h>
21 #include <linux/bug.h>
22 #include <linux/cache.h>
23 #include <linux/rbtree.h>
24 #include <linux/socket.h>
26 #include <linux/atomic.h>
27 #include <asm/types.h>
28 #include <linux/spinlock.h>
29 #include <linux/net.h>
30 #include <linux/textsearch.h>
31 #include <net/checksum.h>
32 #include <linux/rcupdate.h>
33 #include <linux/hrtimer.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/netdev_features.h>
36 #include <linux/sched.h>
37 #include <net/flow_dissector.h>
38 #include <linux/splice.h>
39 #include <linux/in6.h>
42 /* The interface for checksum offload between the stack and networking drivers
45 * A. IP checksum related features
47 * Drivers advertise checksum offload capabilities in the features of a device.
48 * From the stack's point of view these are capabilities offered by the driver,
49 * a driver typically only advertises features that it is capable of offloading
52 * The checksum related features are:
54 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
55 * IP (one's complement) checksum for any combination
56 * of protocols or protocol layering. The checksum is
57 * computed and set in a packet per the CHECKSUM_PARTIAL
58 * interface (see below).
60 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
61 * TCP or UDP packets over IPv4. These are specifically
62 * unencapsulated packets of the form IPv4|TCP or
63 * IPv4|UDP where the Protocol field in the IPv4 header
64 * is TCP or UDP. The IPv4 header may contain IP options
65 * This feature cannot be set in features for a device
66 * with NETIF_F_HW_CSUM also set. This feature is being
67 * DEPRECATED (see below).
69 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
70 * TCP or UDP packets over IPv6. These are specifically
71 * unencapsulated packets of the form IPv6|TCP or
72 * IPv4|UDP where the Next Header field in the IPv6
73 * header is either TCP or UDP. IPv6 extension headers
74 * are not supported with this feature. This feature
75 * cannot be set in features for a device with
76 * NETIF_F_HW_CSUM also set. This feature is being
77 * DEPRECATED (see below).
79 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
80 * This flag is used only used to disable the RX checksum
81 * feature for a device. The stack will accept receive
82 * checksum indication in packets received on a device
83 * regardless of whether NETIF_F_RXCSUM is set.
85 * B. Checksumming of received packets by device. Indication of checksum
86 * verification is in set skb->ip_summed. Possible values are:
90 * Device did not checksum this packet e.g. due to lack of capabilities.
91 * The packet contains full (though not verified) checksum in packet but
92 * not in skb->csum. Thus, skb->csum is undefined in this case.
94 * CHECKSUM_UNNECESSARY:
96 * The hardware you're dealing with doesn't calculate the full checksum
97 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
98 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
99 * if their checksums are okay. skb->csum is still undefined in this case
100 * though. A driver or device must never modify the checksum field in the
101 * packet even if checksum is verified.
103 * CHECKSUM_UNNECESSARY is applicable to following protocols:
104 * TCP: IPv6 and IPv4.
105 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
106 * zero UDP checksum for either IPv4 or IPv6, the networking stack
107 * may perform further validation in this case.
108 * GRE: only if the checksum is present in the header.
109 * SCTP: indicates the CRC in SCTP header has been validated.
111 * skb->csum_level indicates the number of consecutive checksums found in
112 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
113 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
114 * and a device is able to verify the checksums for UDP (possibly zero),
115 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
116 * two. If the device were only able to verify the UDP checksum and not
117 * GRE, either because it doesn't support GRE checksum of because GRE
118 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
119 * not considered in this case).
123 * This is the most generic way. The device supplied checksum of the _whole_
124 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
125 * hardware doesn't need to parse L3/L4 headers to implement this.
127 * Note: Even if device supports only some protocols, but is able to produce
128 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
132 * A checksum is set up to be offloaded to a device as described in the
133 * output description for CHECKSUM_PARTIAL. This may occur on a packet
134 * received directly from another Linux OS, e.g., a virtualized Linux kernel
135 * on the same host, or it may be set in the input path in GRO or remote
136 * checksum offload. For the purposes of checksum verification, the checksum
137 * referred to by skb->csum_start + skb->csum_offset and any preceding
138 * checksums in the packet are considered verified. Any checksums in the
139 * packet that are after the checksum being offloaded are not considered to
142 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
143 * in the skb->ip_summed for a packet. Values are:
147 * The driver is required to checksum the packet as seen by hard_start_xmit()
148 * from skb->csum_start up to the end, and to record/write the checksum at
149 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
150 * csum_start and csum_offset values are valid values given the length and
151 * offset of the packet, however they should not attempt to validate that the
152 * checksum refers to a legitimate transport layer checksum-- it is the
153 * purview of the stack to validate that csum_start and csum_offset are set
156 * When the stack requests checksum offload for a packet, the driver MUST
157 * ensure that the checksum is set correctly. A driver can either offload the
158 * checksum calculation to the device, or call skb_checksum_help (in the case
159 * that the device does not support offload for a particular checksum).
161 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
162 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
163 * checksum offload capability. If a device has limited checksum capabilities
164 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as
165 * described above) a helper function can be called to resolve
166 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper
167 * function takes a spec argument that describes the protocol layer that is
168 * supported for checksum offload and can be called for each packet. If a
169 * packet does not match the specification for offload, skb_checksum_help
170 * is called to resolve the checksum.
174 * The skb was already checksummed by the protocol, or a checksum is not
177 * CHECKSUM_UNNECESSARY:
179 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
183 * Not used in checksum output. If a driver observes a packet with this value
184 * set in skbuff, if should treat as CHECKSUM_NONE being set.
186 * D. Non-IP checksum (CRC) offloads
188 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
189 * offloading the SCTP CRC in a packet. To perform this offload the stack
190 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
191 * accordingly. Note the there is no indication in the skbuff that the
192 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports
193 * both IP checksum offload and SCTP CRC offload must verify which offload
194 * is configured for a packet presumably by inspecting packet headers.
196 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
197 * offloading the FCOE CRC in a packet. To perform this offload the stack
198 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
199 * accordingly. Note the there is no indication in the skbuff that the
200 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
201 * both IP checksum offload and FCOE CRC offload must verify which offload
202 * is configured for a packet presumably by inspecting packet headers.
204 * E. Checksumming on output with GSO.
206 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
207 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
208 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
209 * part of the GSO operation is implied. If a checksum is being offloaded
210 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
211 * are set to refer to the outermost checksum being offload (two offloaded
212 * checksums are possible with UDP encapsulation).
215 /* Don't change this without changing skb_csum_unnecessary! */
216 #define CHECKSUM_NONE 0
217 #define CHECKSUM_UNNECESSARY 1
218 #define CHECKSUM_COMPLETE 2
219 #define CHECKSUM_PARTIAL 3
221 /* Maximum value in skb->csum_level */
222 #define SKB_MAX_CSUM_LEVEL 3
224 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
225 #define SKB_WITH_OVERHEAD(X) \
226 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
227 #define SKB_MAX_ORDER(X, ORDER) \
228 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
229 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
230 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
232 /* return minimum truesize of one skb containing X bytes of data */
233 #define SKB_TRUESIZE(X) ((X) + \
234 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
235 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
239 struct pipe_inode_info;
243 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
244 struct nf_conntrack {
249 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
250 struct nf_bridge_info {
253 BRNF_PROTO_UNCHANGED,
261 struct net_device *physindev;
263 /* always valid & non-NULL from FORWARD on, for physdev match */
264 struct net_device *physoutdev;
266 /* prerouting: detect dnat in orig/reply direction */
268 struct in6_addr ipv6_daddr;
270 /* after prerouting + nat detected: store original source
271 * mac since neigh resolution overwrites it, only used while
272 * skb is out in neigh layer.
274 char neigh_header[8];
279 struct sk_buff_head {
280 /* These two members must be first. */
281 struct sk_buff *next;
282 struct sk_buff *prev;
290 /* To allow 64K frame to be packed as single skb without frag_list we
291 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
292 * buffers which do not start on a page boundary.
294 * Since GRO uses frags we allocate at least 16 regardless of page
297 #if (65536/PAGE_SIZE + 1) < 16
298 #define MAX_SKB_FRAGS 16UL
300 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
302 extern int sysctl_max_skb_frags;
304 typedef struct skb_frag_struct skb_frag_t;
306 struct skb_frag_struct {
310 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
319 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
324 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
329 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
334 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
339 #define HAVE_HW_TIME_STAMP
342 * struct skb_shared_hwtstamps - hardware time stamps
343 * @hwtstamp: hardware time stamp transformed into duration
344 * since arbitrary point in time
346 * Software time stamps generated by ktime_get_real() are stored in
349 * hwtstamps can only be compared against other hwtstamps from
352 * This structure is attached to packets as part of the
353 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
355 struct skb_shared_hwtstamps {
359 /* Definitions for tx_flags in struct skb_shared_info */
361 /* generate hardware time stamp */
362 SKBTX_HW_TSTAMP = 1 << 0,
364 /* generate software time stamp when queueing packet to NIC */
365 SKBTX_SW_TSTAMP = 1 << 1,
367 /* device driver is going to provide hardware time stamp */
368 SKBTX_IN_PROGRESS = 1 << 2,
370 /* device driver supports TX zero-copy buffers */
371 SKBTX_DEV_ZEROCOPY = 1 << 3,
373 /* generate wifi status information (where possible) */
374 SKBTX_WIFI_STATUS = 1 << 4,
376 /* This indicates at least one fragment might be overwritten
377 * (as in vmsplice(), sendfile() ...)
378 * If we need to compute a TX checksum, we'll need to copy
379 * all frags to avoid possible bad checksum
381 SKBTX_SHARED_FRAG = 1 << 5,
383 /* generate software time stamp when entering packet scheduling */
384 SKBTX_SCHED_TSTAMP = 1 << 6,
386 /* generate software timestamp on peer data acknowledgment */
387 SKBTX_ACK_TSTAMP = 1 << 7,
390 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
391 SKBTX_SCHED_TSTAMP | \
393 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
396 * The callback notifies userspace to release buffers when skb DMA is done in
397 * lower device, the skb last reference should be 0 when calling this.
398 * The zerocopy_success argument is true if zero copy transmit occurred,
399 * false on data copy or out of memory error caused by data copy attempt.
400 * The ctx field is used to track device context.
401 * The desc field is used to track userspace buffer index.
404 void (*callback)(struct ubuf_info *, bool zerocopy_success);
409 /* This data is invariant across clones and lives at
410 * the end of the header data, ie. at skb->end.
412 struct skb_shared_info {
413 unsigned char nr_frags;
415 unsigned short gso_size;
416 /* Warning: this field is not always filled in (UFO)! */
417 unsigned short gso_segs;
418 unsigned short gso_type;
419 struct sk_buff *frag_list;
420 struct skb_shared_hwtstamps hwtstamps;
425 * Warning : all fields before dataref are cleared in __alloc_skb()
429 /* Intermediate layers must ensure that destructor_arg
430 * remains valid until skb destructor */
431 void * destructor_arg;
433 /* must be last field, see pskb_expand_head() */
434 skb_frag_t frags[MAX_SKB_FRAGS];
437 /* We divide dataref into two halves. The higher 16 bits hold references
438 * to the payload part of skb->data. The lower 16 bits hold references to
439 * the entire skb->data. A clone of a headerless skb holds the length of
440 * the header in skb->hdr_len.
442 * All users must obey the rule that the skb->data reference count must be
443 * greater than or equal to the payload reference count.
445 * Holding a reference to the payload part means that the user does not
446 * care about modifications to the header part of skb->data.
448 #define SKB_DATAREF_SHIFT 16
449 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
453 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
454 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
455 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
459 SKB_GSO_TCPV4 = 1 << 0,
460 SKB_GSO_UDP = 1 << 1,
462 /* This indicates the skb is from an untrusted source. */
463 SKB_GSO_DODGY = 1 << 2,
465 /* This indicates the tcp segment has CWR set. */
466 SKB_GSO_TCP_ECN = 1 << 3,
468 SKB_GSO_TCP_FIXEDID = 1 << 4,
470 SKB_GSO_TCPV6 = 1 << 5,
472 SKB_GSO_FCOE = 1 << 6,
474 SKB_GSO_GRE = 1 << 7,
476 SKB_GSO_GRE_CSUM = 1 << 8,
478 SKB_GSO_IPIP = 1 << 9,
480 SKB_GSO_SIT = 1 << 10,
482 SKB_GSO_UDP_TUNNEL = 1 << 11,
484 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
486 SKB_GSO_PARTIAL = 1 << 13,
488 SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
491 #if BITS_PER_LONG > 32
492 #define NET_SKBUFF_DATA_USES_OFFSET 1
495 #ifdef NET_SKBUFF_DATA_USES_OFFSET
496 typedef unsigned int sk_buff_data_t;
498 typedef unsigned char *sk_buff_data_t;
502 * struct skb_mstamp - multi resolution time stamps
503 * @stamp_us: timestamp in us resolution
504 * @stamp_jiffies: timestamp in jiffies
517 * skb_mstamp_get - get current timestamp
518 * @cl: place to store timestamps
520 static inline void skb_mstamp_get(struct skb_mstamp *cl)
522 u64 val = local_clock();
524 do_div(val, NSEC_PER_USEC);
525 cl->stamp_us = (u32)val;
526 cl->stamp_jiffies = (u32)jiffies;
530 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
531 * @t1: pointer to newest sample
532 * @t0: pointer to oldest sample
534 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
535 const struct skb_mstamp *t0)
537 s32 delta_us = t1->stamp_us - t0->stamp_us;
538 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
540 /* If delta_us is negative, this might be because interval is too big,
541 * or local_clock() drift is too big : fallback using jiffies.
544 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
546 delta_us = jiffies_to_usecs(delta_jiffies);
551 static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
552 const struct skb_mstamp *t0)
554 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
557 diff = t1->stamp_us - t0->stamp_us;
562 * struct sk_buff - socket buffer
563 * @next: Next buffer in list
564 * @prev: Previous buffer in list
565 * @tstamp: Time we arrived/left
566 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
567 * @sk: Socket we are owned by
568 * @dev: Device we arrived on/are leaving by
569 * @cb: Control buffer. Free for use by every layer. Put private vars here
570 * @_skb_refdst: destination entry (with norefcount bit)
571 * @sp: the security path, used for xfrm
572 * @len: Length of actual data
573 * @data_len: Data length
574 * @mac_len: Length of link layer header
575 * @hdr_len: writable header length of cloned skb
576 * @csum: Checksum (must include start/offset pair)
577 * @csum_start: Offset from skb->head where checksumming should start
578 * @csum_offset: Offset from csum_start where checksum should be stored
579 * @priority: Packet queueing priority
580 * @ignore_df: allow local fragmentation
581 * @cloned: Head may be cloned (check refcnt to be sure)
582 * @ip_summed: Driver fed us an IP checksum
583 * @nohdr: Payload reference only, must not modify header
584 * @nfctinfo: Relationship of this skb to the connection
585 * @pkt_type: Packet class
586 * @fclone: skbuff clone status
587 * @ipvs_property: skbuff is owned by ipvs
588 * @peeked: this packet has been seen already, so stats have been
589 * done for it, don't do them again
590 * @nf_trace: netfilter packet trace flag
591 * @protocol: Packet protocol from driver
592 * @destructor: Destruct function
593 * @nfct: Associated connection, if any
594 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
595 * @skb_iif: ifindex of device we arrived on
596 * @tc_index: Traffic control index
597 * @tc_verd: traffic control verdict
598 * @hash: the packet hash
599 * @queue_mapping: Queue mapping for multiqueue devices
600 * @xmit_more: More SKBs are pending for this queue
601 * @ndisc_nodetype: router type (from link layer)
602 * @ooo_okay: allow the mapping of a socket to a queue to be changed
603 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
605 * @sw_hash: indicates hash was computed in software stack
606 * @wifi_acked_valid: wifi_acked was set
607 * @wifi_acked: whether frame was acked on wifi or not
608 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
609 * @napi_id: id of the NAPI struct this skb came from
610 * @secmark: security marking
611 * @offload_fwd_mark: fwding offload mark
612 * @mark: Generic packet mark
613 * @vlan_proto: vlan encapsulation protocol
614 * @vlan_tci: vlan tag control information
615 * @inner_protocol: Protocol (encapsulation)
616 * @inner_transport_header: Inner transport layer header (encapsulation)
617 * @inner_network_header: Network layer header (encapsulation)
618 * @inner_mac_header: Link layer header (encapsulation)
619 * @transport_header: Transport layer header
620 * @network_header: Network layer header
621 * @mac_header: Link layer header
622 * @tail: Tail pointer
624 * @head: Head of buffer
625 * @data: Data head pointer
626 * @truesize: Buffer size
627 * @users: User count - see {datagram,tcp}.c
633 /* These two members must be first. */
634 struct sk_buff *next;
635 struct sk_buff *prev;
639 struct skb_mstamp skb_mstamp;
642 struct rb_node rbnode; /* used in netem & tcp stack */
645 struct net_device *dev;
648 * This is the control buffer. It is free to use for every
649 * layer. Please put your private variables there. If you
650 * want to keep them across layers you have to do a skb_clone()
651 * first. This is owned by whoever has the skb queued ATM.
653 char cb[48] __aligned(8);
655 unsigned long _skb_refdst;
656 void (*destructor)(struct sk_buff *skb);
660 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
661 struct nf_conntrack *nfct;
663 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
664 struct nf_bridge_info *nf_bridge;
671 /* Following fields are _not_ copied in __copy_skb_header()
672 * Note that queue_mapping is here mostly to fill a hole.
674 kmemcheck_bitfield_begin(flags1);
683 kmemcheck_bitfield_end(flags1);
685 /* fields enclosed in headers_start/headers_end are copied
686 * using a single memcpy() in __copy_skb_header()
689 __u32 headers_start[0];
692 /* if you move pkt_type around you also must adapt those constants */
693 #ifdef __BIG_ENDIAN_BITFIELD
694 #define PKT_TYPE_MAX (7 << 5)
696 #define PKT_TYPE_MAX 7
698 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
700 __u8 __pkt_type_offset[0];
711 __u8 wifi_acked_valid:1;
715 /* Indicates the inner headers are valid in the skbuff. */
716 __u8 encapsulation:1;
717 __u8 encap_hdr_csum:1;
719 __u8 csum_complete_sw:1;
723 #ifdef CONFIG_IPV6_NDISC_NODETYPE
724 __u8 ndisc_nodetype:2;
726 __u8 ipvs_property:1;
727 __u8 inner_protocol_type:1;
728 __u8 remcsum_offload:1;
729 /* 3 or 5 bit hole */
731 #ifdef CONFIG_NET_SCHED
732 __u16 tc_index; /* traffic control index */
733 #ifdef CONFIG_NET_CLS_ACT
734 __u16 tc_verd; /* traffic control verdict */
750 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
752 unsigned int napi_id;
753 unsigned int sender_cpu;
757 #ifdef CONFIG_NETWORK_SECMARK
760 #ifdef CONFIG_NET_SWITCHDEV
761 __u32 offload_fwd_mark;
767 __u32 reserved_tailroom;
771 __be16 inner_protocol;
775 __u16 inner_transport_header;
776 __u16 inner_network_header;
777 __u16 inner_mac_header;
780 __u16 transport_header;
781 __u16 network_header;
785 __u32 headers_end[0];
788 /* These elements must be at the end, see alloc_skb() for details. */
793 unsigned int truesize;
799 * Handling routines are only of interest to the kernel
801 #include <linux/slab.h>
804 #define SKB_ALLOC_FCLONE 0x01
805 #define SKB_ALLOC_RX 0x02
806 #define SKB_ALLOC_NAPI 0x04
808 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
809 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
811 return unlikely(skb->pfmemalloc);
815 * skb might have a dst pointer attached, refcounted or not.
816 * _skb_refdst low order bit is set if refcount was _not_ taken
818 #define SKB_DST_NOREF 1UL
819 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
822 * skb_dst - returns skb dst_entry
825 * Returns skb dst_entry, regardless of reference taken or not.
827 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
829 /* If refdst was not refcounted, check we still are in a
830 * rcu_read_lock section
832 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
833 !rcu_read_lock_held() &&
834 !rcu_read_lock_bh_held());
835 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
839 * skb_dst_set - sets skb dst
843 * Sets skb dst, assuming a reference was taken on dst and should
844 * be released by skb_dst_drop()
846 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
848 skb->_skb_refdst = (unsigned long)dst;
852 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
856 * Sets skb dst, assuming a reference was not taken on dst.
857 * If dst entry is cached, we do not take reference and dst_release
858 * will be avoided by refdst_drop. If dst entry is not cached, we take
859 * reference, so that last dst_release can destroy the dst immediately.
861 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
863 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
864 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
868 * skb_dst_is_noref - Test if skb dst isn't refcounted
871 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
873 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
876 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
878 return (struct rtable *)skb_dst(skb);
881 void kfree_skb(struct sk_buff *skb);
882 void kfree_skb_list(struct sk_buff *segs);
883 void skb_tx_error(struct sk_buff *skb);
884 void consume_skb(struct sk_buff *skb);
885 void __kfree_skb(struct sk_buff *skb);
886 extern struct kmem_cache *skbuff_head_cache;
888 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
889 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
890 bool *fragstolen, int *delta_truesize);
892 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
894 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
895 struct sk_buff *build_skb(void *data, unsigned int frag_size);
896 static inline struct sk_buff *alloc_skb(unsigned int size,
899 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
902 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
903 unsigned long data_len,
908 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
909 struct sk_buff_fclones {
918 * skb_fclone_busy - check if fclone is busy
921 * Returns true if skb is a fast clone, and its clone is not freed.
922 * Some drivers call skb_orphan() in their ndo_start_xmit(),
923 * so we also check that this didnt happen.
925 static inline bool skb_fclone_busy(const struct sock *sk,
926 const struct sk_buff *skb)
928 const struct sk_buff_fclones *fclones;
930 fclones = container_of(skb, struct sk_buff_fclones, skb1);
932 return skb->fclone == SKB_FCLONE_ORIG &&
933 atomic_read(&fclones->fclone_ref) > 1 &&
934 fclones->skb2.sk == sk;
937 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
940 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
943 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
944 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
946 return __alloc_skb_head(priority, -1);
949 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
950 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
951 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
952 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
953 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
954 gfp_t gfp_mask, bool fclone);
955 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
958 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
961 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
962 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
963 unsigned int headroom);
964 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
965 int newtailroom, gfp_t priority);
966 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
967 int offset, int len);
968 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
970 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
971 int skb_pad(struct sk_buff *skb, int pad);
972 #define dev_kfree_skb(a) consume_skb(a)
974 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
975 int getfrag(void *from, char *to, int offset,
976 int len, int odd, struct sk_buff *skb),
977 void *from, int length);
979 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
980 int offset, size_t size);
982 struct skb_seq_state {
986 __u32 stepped_offset;
987 struct sk_buff *root_skb;
988 struct sk_buff *cur_skb;
992 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
993 unsigned int to, struct skb_seq_state *st);
994 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
995 struct skb_seq_state *st);
996 void skb_abort_seq_read(struct skb_seq_state *st);
998 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
999 unsigned int to, struct ts_config *config);
1002 * Packet hash types specify the type of hash in skb_set_hash.
1004 * Hash types refer to the protocol layer addresses which are used to
1005 * construct a packet's hash. The hashes are used to differentiate or identify
1006 * flows of the protocol layer for the hash type. Hash types are either
1007 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1009 * Properties of hashes:
1011 * 1) Two packets in different flows have different hash values
1012 * 2) Two packets in the same flow should have the same hash value
1014 * A hash at a higher layer is considered to be more specific. A driver should
1015 * set the most specific hash possible.
1017 * A driver cannot indicate a more specific hash than the layer at which a hash
1018 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1020 * A driver may indicate a hash level which is less specific than the
1021 * actual layer the hash was computed on. For instance, a hash computed
1022 * at L4 may be considered an L3 hash. This should only be done if the
1023 * driver can't unambiguously determine that the HW computed the hash at
1024 * the higher layer. Note that the "should" in the second property above
1027 enum pkt_hash_types {
1028 PKT_HASH_TYPE_NONE, /* Undefined type */
1029 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1030 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1031 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1034 static inline void skb_clear_hash(struct sk_buff *skb)
1041 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1044 skb_clear_hash(skb);
1048 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1050 skb->l4_hash = is_l4;
1051 skb->sw_hash = is_sw;
1056 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1058 /* Used by drivers to set hash from HW */
1059 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1063 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1065 __skb_set_hash(skb, hash, true, is_l4);
1068 void __skb_get_hash(struct sk_buff *skb);
1069 u32 skb_get_poff(const struct sk_buff *skb);
1070 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1071 const struct flow_keys *keys, int hlen);
1072 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1073 void *data, int hlen_proto);
1075 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1076 int thoff, u8 ip_proto)
1078 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1081 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1082 const struct flow_dissector_key *key,
1083 unsigned int key_count);
1085 bool __skb_flow_dissect(const struct sk_buff *skb,
1086 struct flow_dissector *flow_dissector,
1087 void *target_container,
1088 void *data, __be16 proto, int nhoff, int hlen,
1089 unsigned int flags);
1091 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1092 struct flow_dissector *flow_dissector,
1093 void *target_container, unsigned int flags)
1095 return __skb_flow_dissect(skb, flow_dissector, target_container,
1096 NULL, 0, 0, 0, flags);
1099 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1100 struct flow_keys *flow,
1103 memset(flow, 0, sizeof(*flow));
1104 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1105 NULL, 0, 0, 0, flags);
1108 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1109 void *data, __be16 proto,
1110 int nhoff, int hlen,
1113 memset(flow, 0, sizeof(*flow));
1114 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1115 data, proto, nhoff, hlen, flags);
1118 static inline __u32 skb_get_hash(struct sk_buff *skb)
1120 if (!skb->l4_hash && !skb->sw_hash)
1121 __skb_get_hash(skb);
1126 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1128 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1130 if (!skb->l4_hash && !skb->sw_hash) {
1131 struct flow_keys keys;
1132 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1134 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1140 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1142 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1144 if (!skb->l4_hash && !skb->sw_hash) {
1145 struct flow_keys keys;
1146 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
1148 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1154 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1156 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1161 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1163 to->hash = from->hash;
1164 to->sw_hash = from->sw_hash;
1165 to->l4_hash = from->l4_hash;
1168 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1169 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1171 return skb->head + skb->end;
1174 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1179 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1184 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1186 return skb->end - skb->head;
1191 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1193 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1195 return &skb_shinfo(skb)->hwtstamps;
1199 * skb_queue_empty - check if a queue is empty
1202 * Returns true if the queue is empty, false otherwise.
1204 static inline int skb_queue_empty(const struct sk_buff_head *list)
1206 return list->next == (const struct sk_buff *) list;
1210 * skb_queue_is_last - check if skb is the last entry in the queue
1214 * Returns true if @skb is the last buffer on the list.
1216 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1217 const struct sk_buff *skb)
1219 return skb->next == (const struct sk_buff *) list;
1223 * skb_queue_is_first - check if skb is the first entry in the queue
1227 * Returns true if @skb is the first buffer on the list.
1229 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1230 const struct sk_buff *skb)
1232 return skb->prev == (const struct sk_buff *) list;
1236 * skb_queue_next - return the next packet in the queue
1238 * @skb: current buffer
1240 * Return the next packet in @list after @skb. It is only valid to
1241 * call this if skb_queue_is_last() evaluates to false.
1243 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1244 const struct sk_buff *skb)
1246 /* This BUG_ON may seem severe, but if we just return then we
1247 * are going to dereference garbage.
1249 BUG_ON(skb_queue_is_last(list, skb));
1254 * skb_queue_prev - return the prev packet in the queue
1256 * @skb: current buffer
1258 * Return the prev packet in @list before @skb. It is only valid to
1259 * call this if skb_queue_is_first() evaluates to false.
1261 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1262 const struct sk_buff *skb)
1264 /* This BUG_ON may seem severe, but if we just return then we
1265 * are going to dereference garbage.
1267 BUG_ON(skb_queue_is_first(list, skb));
1272 * skb_get - reference buffer
1273 * @skb: buffer to reference
1275 * Makes another reference to a socket buffer and returns a pointer
1278 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1280 atomic_inc(&skb->users);
1285 * If users == 1, we are the only owner and are can avoid redundant
1290 * skb_cloned - is the buffer a clone
1291 * @skb: buffer to check
1293 * Returns true if the buffer was generated with skb_clone() and is
1294 * one of multiple shared copies of the buffer. Cloned buffers are
1295 * shared data so must not be written to under normal circumstances.
1297 static inline int skb_cloned(const struct sk_buff *skb)
1299 return skb->cloned &&
1300 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1303 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1305 might_sleep_if(gfpflags_allow_blocking(pri));
1307 if (skb_cloned(skb))
1308 return pskb_expand_head(skb, 0, 0, pri);
1314 * skb_header_cloned - is the header a clone
1315 * @skb: buffer to check
1317 * Returns true if modifying the header part of the buffer requires
1318 * the data to be copied.
1320 static inline int skb_header_cloned(const struct sk_buff *skb)
1327 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1328 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1329 return dataref != 1;
1333 * skb_header_release - release reference to header
1334 * @skb: buffer to operate on
1336 * Drop a reference to the header part of the buffer. This is done
1337 * by acquiring a payload reference. You must not read from the header
1338 * part of skb->data after this.
1339 * Note : Check if you can use __skb_header_release() instead.
1341 static inline void skb_header_release(struct sk_buff *skb)
1345 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1349 * __skb_header_release - release reference to header
1350 * @skb: buffer to operate on
1352 * Variant of skb_header_release() assuming skb is private to caller.
1353 * We can avoid one atomic operation.
1355 static inline void __skb_header_release(struct sk_buff *skb)
1358 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1363 * skb_shared - is the buffer shared
1364 * @skb: buffer to check
1366 * Returns true if more than one person has a reference to this
1369 static inline int skb_shared(const struct sk_buff *skb)
1371 return atomic_read(&skb->users) != 1;
1375 * skb_share_check - check if buffer is shared and if so clone it
1376 * @skb: buffer to check
1377 * @pri: priority for memory allocation
1379 * If the buffer is shared the buffer is cloned and the old copy
1380 * drops a reference. A new clone with a single reference is returned.
1381 * If the buffer is not shared the original buffer is returned. When
1382 * being called from interrupt status or with spinlocks held pri must
1385 * NULL is returned on a memory allocation failure.
1387 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1389 might_sleep_if(gfpflags_allow_blocking(pri));
1390 if (skb_shared(skb)) {
1391 struct sk_buff *nskb = skb_clone(skb, pri);
1403 * Copy shared buffers into a new sk_buff. We effectively do COW on
1404 * packets to handle cases where we have a local reader and forward
1405 * and a couple of other messy ones. The normal one is tcpdumping
1406 * a packet thats being forwarded.
1410 * skb_unshare - make a copy of a shared buffer
1411 * @skb: buffer to check
1412 * @pri: priority for memory allocation
1414 * If the socket buffer is a clone then this function creates a new
1415 * copy of the data, drops a reference count on the old copy and returns
1416 * the new copy with the reference count at 1. If the buffer is not a clone
1417 * the original buffer is returned. When called with a spinlock held or
1418 * from interrupt state @pri must be %GFP_ATOMIC
1420 * %NULL is returned on a memory allocation failure.
1422 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1425 might_sleep_if(gfpflags_allow_blocking(pri));
1426 if (skb_cloned(skb)) {
1427 struct sk_buff *nskb = skb_copy(skb, pri);
1429 /* Free our shared copy */
1440 * skb_peek - peek at the head of an &sk_buff_head
1441 * @list_: list to peek at
1443 * Peek an &sk_buff. Unlike most other operations you _MUST_
1444 * be careful with this one. A peek leaves the buffer on the
1445 * list and someone else may run off with it. You must hold
1446 * the appropriate locks or have a private queue to do this.
1448 * Returns %NULL for an empty list or a pointer to the head element.
1449 * The reference count is not incremented and the reference is therefore
1450 * volatile. Use with caution.
1452 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1454 struct sk_buff *skb = list_->next;
1456 if (skb == (struct sk_buff *)list_)
1462 * skb_peek_next - peek skb following the given one from a queue
1463 * @skb: skb to start from
1464 * @list_: list to peek at
1466 * Returns %NULL when the end of the list is met or a pointer to the
1467 * next element. The reference count is not incremented and the
1468 * reference is therefore volatile. Use with caution.
1470 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1471 const struct sk_buff_head *list_)
1473 struct sk_buff *next = skb->next;
1475 if (next == (struct sk_buff *)list_)
1481 * skb_peek_tail - peek at the tail of an &sk_buff_head
1482 * @list_: list to peek at
1484 * Peek an &sk_buff. Unlike most other operations you _MUST_
1485 * be careful with this one. A peek leaves the buffer on the
1486 * list and someone else may run off with it. You must hold
1487 * the appropriate locks or have a private queue to do this.
1489 * Returns %NULL for an empty list or a pointer to the tail element.
1490 * The reference count is not incremented and the reference is therefore
1491 * volatile. Use with caution.
1493 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1495 struct sk_buff *skb = list_->prev;
1497 if (skb == (struct sk_buff *)list_)
1504 * skb_queue_len - get queue length
1505 * @list_: list to measure
1507 * Return the length of an &sk_buff queue.
1509 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1515 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1516 * @list: queue to initialize
1518 * This initializes only the list and queue length aspects of
1519 * an sk_buff_head object. This allows to initialize the list
1520 * aspects of an sk_buff_head without reinitializing things like
1521 * the spinlock. It can also be used for on-stack sk_buff_head
1522 * objects where the spinlock is known to not be used.
1524 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1526 list->prev = list->next = (struct sk_buff *)list;
1531 * This function creates a split out lock class for each invocation;
1532 * this is needed for now since a whole lot of users of the skb-queue
1533 * infrastructure in drivers have different locking usage (in hardirq)
1534 * than the networking core (in softirq only). In the long run either the
1535 * network layer or drivers should need annotation to consolidate the
1536 * main types of usage into 3 classes.
1538 static inline void skb_queue_head_init(struct sk_buff_head *list)
1540 spin_lock_init(&list->lock);
1541 __skb_queue_head_init(list);
1544 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1545 struct lock_class_key *class)
1547 skb_queue_head_init(list);
1548 lockdep_set_class(&list->lock, class);
1552 * Insert an sk_buff on a list.
1554 * The "__skb_xxxx()" functions are the non-atomic ones that
1555 * can only be called with interrupts disabled.
1557 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1558 struct sk_buff_head *list);
1559 static inline void __skb_insert(struct sk_buff *newsk,
1560 struct sk_buff *prev, struct sk_buff *next,
1561 struct sk_buff_head *list)
1565 next->prev = prev->next = newsk;
1569 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1570 struct sk_buff *prev,
1571 struct sk_buff *next)
1573 struct sk_buff *first = list->next;
1574 struct sk_buff *last = list->prev;
1584 * skb_queue_splice - join two skb lists, this is designed for stacks
1585 * @list: the new list to add
1586 * @head: the place to add it in the first list
1588 static inline void skb_queue_splice(const struct sk_buff_head *list,
1589 struct sk_buff_head *head)
1591 if (!skb_queue_empty(list)) {
1592 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1593 head->qlen += list->qlen;
1598 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1599 * @list: the new list to add
1600 * @head: the place to add it in the first list
1602 * The list at @list is reinitialised
1604 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1605 struct sk_buff_head *head)
1607 if (!skb_queue_empty(list)) {
1608 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1609 head->qlen += list->qlen;
1610 __skb_queue_head_init(list);
1615 * skb_queue_splice_tail - join two skb lists, each list being a queue
1616 * @list: the new list to add
1617 * @head: the place to add it in the first list
1619 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1620 struct sk_buff_head *head)
1622 if (!skb_queue_empty(list)) {
1623 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1624 head->qlen += list->qlen;
1629 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1630 * @list: the new list to add
1631 * @head: the place to add it in the first list
1633 * Each of the lists is a queue.
1634 * The list at @list is reinitialised
1636 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1637 struct sk_buff_head *head)
1639 if (!skb_queue_empty(list)) {
1640 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1641 head->qlen += list->qlen;
1642 __skb_queue_head_init(list);
1647 * __skb_queue_after - queue a buffer at the list head
1648 * @list: list to use
1649 * @prev: place after this buffer
1650 * @newsk: buffer to queue
1652 * Queue a buffer int the middle of a list. This function takes no locks
1653 * and you must therefore hold required locks before calling it.
1655 * A buffer cannot be placed on two lists at the same time.
1657 static inline void __skb_queue_after(struct sk_buff_head *list,
1658 struct sk_buff *prev,
1659 struct sk_buff *newsk)
1661 __skb_insert(newsk, prev, prev->next, list);
1664 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1665 struct sk_buff_head *list);
1667 static inline void __skb_queue_before(struct sk_buff_head *list,
1668 struct sk_buff *next,
1669 struct sk_buff *newsk)
1671 __skb_insert(newsk, next->prev, next, list);
1675 * __skb_queue_head - queue a buffer at the list head
1676 * @list: list to use
1677 * @newsk: buffer to queue
1679 * Queue a buffer at the start of a list. This function takes no locks
1680 * and you must therefore hold required locks before calling it.
1682 * A buffer cannot be placed on two lists at the same time.
1684 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1685 static inline void __skb_queue_head(struct sk_buff_head *list,
1686 struct sk_buff *newsk)
1688 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1692 * __skb_queue_tail - queue a buffer at the list tail
1693 * @list: list to use
1694 * @newsk: buffer to queue
1696 * Queue a buffer at the end of a list. This function takes no locks
1697 * and you must therefore hold required locks before calling it.
1699 * A buffer cannot be placed on two lists at the same time.
1701 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1702 static inline void __skb_queue_tail(struct sk_buff_head *list,
1703 struct sk_buff *newsk)
1705 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1709 * remove sk_buff from list. _Must_ be called atomically, and with
1712 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1713 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1715 struct sk_buff *next, *prev;
1720 skb->next = skb->prev = NULL;
1726 * __skb_dequeue - remove from the head of the queue
1727 * @list: list to dequeue from
1729 * Remove the head of the list. This function does not take any locks
1730 * so must be used with appropriate locks held only. The head item is
1731 * returned or %NULL if the list is empty.
1733 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1734 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1736 struct sk_buff *skb = skb_peek(list);
1738 __skb_unlink(skb, list);
1743 * __skb_dequeue_tail - remove from the tail of the queue
1744 * @list: list to dequeue from
1746 * Remove the tail of the list. This function does not take any locks
1747 * so must be used with appropriate locks held only. The tail item is
1748 * returned or %NULL if the list is empty.
1750 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1751 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1753 struct sk_buff *skb = skb_peek_tail(list);
1755 __skb_unlink(skb, list);
1760 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1762 return skb->data_len;
1765 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1767 return skb->len - skb->data_len;
1770 static inline int skb_pagelen(const struct sk_buff *skb)
1774 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1775 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1776 return len + skb_headlen(skb);
1780 * __skb_fill_page_desc - initialise a paged fragment in an skb
1781 * @skb: buffer containing fragment to be initialised
1782 * @i: paged fragment index to initialise
1783 * @page: the page to use for this fragment
1784 * @off: the offset to the data with @page
1785 * @size: the length of the data
1787 * Initialises the @i'th fragment of @skb to point to &size bytes at
1788 * offset @off within @page.
1790 * Does not take any additional reference on the fragment.
1792 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1793 struct page *page, int off, int size)
1795 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1798 * Propagate page pfmemalloc to the skb if we can. The problem is
1799 * that not all callers have unique ownership of the page but rely
1800 * on page_is_pfmemalloc doing the right thing(tm).
1802 frag->page.p = page;
1803 frag->page_offset = off;
1804 skb_frag_size_set(frag, size);
1806 page = compound_head(page);
1807 if (page_is_pfmemalloc(page))
1808 skb->pfmemalloc = true;
1812 * skb_fill_page_desc - initialise a paged fragment in an skb
1813 * @skb: buffer containing fragment to be initialised
1814 * @i: paged fragment index to initialise
1815 * @page: the page to use for this fragment
1816 * @off: the offset to the data with @page
1817 * @size: the length of the data
1819 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1820 * @skb to point to @size bytes at offset @off within @page. In
1821 * addition updates @skb such that @i is the last fragment.
1823 * Does not take any additional reference on the fragment.
1825 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1826 struct page *page, int off, int size)
1828 __skb_fill_page_desc(skb, i, page, off, size);
1829 skb_shinfo(skb)->nr_frags = i + 1;
1832 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1833 int size, unsigned int truesize);
1835 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1836 unsigned int truesize);
1838 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1839 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1840 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1842 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1843 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1845 return skb->head + skb->tail;
1848 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1850 skb->tail = skb->data - skb->head;
1853 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1855 skb_reset_tail_pointer(skb);
1856 skb->tail += offset;
1859 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1860 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1865 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1867 skb->tail = skb->data;
1870 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1872 skb->tail = skb->data + offset;
1875 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1878 * Add data to an sk_buff
1880 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1881 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1882 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1884 unsigned char *tmp = skb_tail_pointer(skb);
1885 SKB_LINEAR_ASSERT(skb);
1891 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1892 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1899 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1900 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1903 BUG_ON(skb->len < skb->data_len);
1904 return skb->data += len;
1907 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1909 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1912 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1914 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1916 if (len > skb_headlen(skb) &&
1917 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1920 return skb->data += len;
1923 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1925 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1928 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1930 if (likely(len <= skb_headlen(skb)))
1932 if (unlikely(len > skb->len))
1934 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1938 * skb_headroom - bytes at buffer head
1939 * @skb: buffer to check
1941 * Return the number of bytes of free space at the head of an &sk_buff.
1943 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1945 return skb->data - skb->head;
1949 * skb_tailroom - bytes at buffer end
1950 * @skb: buffer to check
1952 * Return the number of bytes of free space at the tail of an sk_buff
1954 static inline int skb_tailroom(const struct sk_buff *skb)
1956 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1960 * skb_availroom - bytes at buffer end
1961 * @skb: buffer to check
1963 * Return the number of bytes of free space at the tail of an sk_buff
1964 * allocated by sk_stream_alloc()
1966 static inline int skb_availroom(const struct sk_buff *skb)
1968 if (skb_is_nonlinear(skb))
1971 return skb->end - skb->tail - skb->reserved_tailroom;
1975 * skb_reserve - adjust headroom
1976 * @skb: buffer to alter
1977 * @len: bytes to move
1979 * Increase the headroom of an empty &sk_buff by reducing the tail
1980 * room. This is only allowed for an empty buffer.
1982 static inline void skb_reserve(struct sk_buff *skb, int len)
1989 * skb_tailroom_reserve - adjust reserved_tailroom
1990 * @skb: buffer to alter
1991 * @mtu: maximum amount of headlen permitted
1992 * @needed_tailroom: minimum amount of reserved_tailroom
1994 * Set reserved_tailroom so that headlen can be as large as possible but
1995 * not larger than mtu and tailroom cannot be smaller than
1997 * The required headroom should already have been reserved before using
2000 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2001 unsigned int needed_tailroom)
2003 SKB_LINEAR_ASSERT(skb);
2004 if (mtu < skb_tailroom(skb) - needed_tailroom)
2005 /* use at most mtu */
2006 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2008 /* use up to all available space */
2009 skb->reserved_tailroom = needed_tailroom;
2012 #define ENCAP_TYPE_ETHER 0
2013 #define ENCAP_TYPE_IPPROTO 1
2015 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2018 skb->inner_protocol = protocol;
2019 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2022 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2025 skb->inner_ipproto = ipproto;
2026 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2029 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2031 skb->inner_mac_header = skb->mac_header;
2032 skb->inner_network_header = skb->network_header;
2033 skb->inner_transport_header = skb->transport_header;
2036 static inline void skb_reset_mac_len(struct sk_buff *skb)
2038 skb->mac_len = skb->network_header - skb->mac_header;
2041 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2044 return skb->head + skb->inner_transport_header;
2047 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2049 return skb_inner_transport_header(skb) - skb->data;
2052 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2054 skb->inner_transport_header = skb->data - skb->head;
2057 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2060 skb_reset_inner_transport_header(skb);
2061 skb->inner_transport_header += offset;
2064 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2066 return skb->head + skb->inner_network_header;
2069 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2071 skb->inner_network_header = skb->data - skb->head;
2074 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2077 skb_reset_inner_network_header(skb);
2078 skb->inner_network_header += offset;
2081 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2083 return skb->head + skb->inner_mac_header;
2086 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2088 skb->inner_mac_header = skb->data - skb->head;
2091 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2094 skb_reset_inner_mac_header(skb);
2095 skb->inner_mac_header += offset;
2097 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2099 return skb->transport_header != (typeof(skb->transport_header))~0U;
2102 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2104 return skb->head + skb->transport_header;
2107 static inline void skb_reset_transport_header(struct sk_buff *skb)
2109 skb->transport_header = skb->data - skb->head;
2112 static inline void skb_set_transport_header(struct sk_buff *skb,
2115 skb_reset_transport_header(skb);
2116 skb->transport_header += offset;
2119 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2121 return skb->head + skb->network_header;
2124 static inline void skb_reset_network_header(struct sk_buff *skb)
2126 skb->network_header = skb->data - skb->head;
2129 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2131 skb_reset_network_header(skb);
2132 skb->network_header += offset;
2135 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2137 return skb->head + skb->mac_header;
2140 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2142 return skb->mac_header != (typeof(skb->mac_header))~0U;
2145 static inline void skb_reset_mac_header(struct sk_buff *skb)
2147 skb->mac_header = skb->data - skb->head;
2150 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2152 skb_reset_mac_header(skb);
2153 skb->mac_header += offset;
2156 static inline void skb_pop_mac_header(struct sk_buff *skb)
2158 skb->mac_header = skb->network_header;
2161 static inline void skb_probe_transport_header(struct sk_buff *skb,
2162 const int offset_hint)
2164 struct flow_keys keys;
2166 if (skb_transport_header_was_set(skb))
2168 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2169 skb_set_transport_header(skb, keys.control.thoff);
2171 skb_set_transport_header(skb, offset_hint);
2174 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2176 if (skb_mac_header_was_set(skb)) {
2177 const unsigned char *old_mac = skb_mac_header(skb);
2179 skb_set_mac_header(skb, -skb->mac_len);
2180 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2184 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2186 return skb->csum_start - skb_headroom(skb);
2189 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2191 return skb->head + skb->csum_start;
2194 static inline int skb_transport_offset(const struct sk_buff *skb)
2196 return skb_transport_header(skb) - skb->data;
2199 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2201 return skb->transport_header - skb->network_header;
2204 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2206 return skb->inner_transport_header - skb->inner_network_header;
2209 static inline int skb_network_offset(const struct sk_buff *skb)
2211 return skb_network_header(skb) - skb->data;
2214 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2216 return skb_inner_network_header(skb) - skb->data;
2219 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2221 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2225 * CPUs often take a performance hit when accessing unaligned memory
2226 * locations. The actual performance hit varies, it can be small if the
2227 * hardware handles it or large if we have to take an exception and fix it
2230 * Since an ethernet header is 14 bytes network drivers often end up with
2231 * the IP header at an unaligned offset. The IP header can be aligned by
2232 * shifting the start of the packet by 2 bytes. Drivers should do this
2235 * skb_reserve(skb, NET_IP_ALIGN);
2237 * The downside to this alignment of the IP header is that the DMA is now
2238 * unaligned. On some architectures the cost of an unaligned DMA is high
2239 * and this cost outweighs the gains made by aligning the IP header.
2241 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2244 #ifndef NET_IP_ALIGN
2245 #define NET_IP_ALIGN 2
2249 * The networking layer reserves some headroom in skb data (via
2250 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2251 * the header has to grow. In the default case, if the header has to grow
2252 * 32 bytes or less we avoid the reallocation.
2254 * Unfortunately this headroom changes the DMA alignment of the resulting
2255 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2256 * on some architectures. An architecture can override this value,
2257 * perhaps setting it to a cacheline in size (since that will maintain
2258 * cacheline alignment of the DMA). It must be a power of 2.
2260 * Various parts of the networking layer expect at least 32 bytes of
2261 * headroom, you should not reduce this.
2263 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2264 * to reduce average number of cache lines per packet.
2265 * get_rps_cpus() for example only access one 64 bytes aligned block :
2266 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2269 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2272 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2274 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2276 if (unlikely(skb_is_nonlinear(skb))) {
2281 skb_set_tail_pointer(skb, len);
2284 void skb_trim(struct sk_buff *skb, unsigned int len);
2286 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2289 return ___pskb_trim(skb, len);
2290 __skb_trim(skb, len);
2294 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2296 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2300 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2301 * @skb: buffer to alter
2304 * This is identical to pskb_trim except that the caller knows that
2305 * the skb is not cloned so we should never get an error due to out-
2308 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2310 int err = pskb_trim(skb, len);
2315 * skb_orphan - orphan a buffer
2316 * @skb: buffer to orphan
2318 * If a buffer currently has an owner then we call the owner's
2319 * destructor function and make the @skb unowned. The buffer continues
2320 * to exist but is no longer charged to its former owner.
2322 static inline void skb_orphan(struct sk_buff *skb)
2324 if (skb->destructor) {
2325 skb->destructor(skb);
2326 skb->destructor = NULL;
2334 * skb_orphan_frags - orphan the frags contained in a buffer
2335 * @skb: buffer to orphan frags from
2336 * @gfp_mask: allocation mask for replacement pages
2338 * For each frag in the SKB which needs a destructor (i.e. has an
2339 * owner) create a copy of that frag and release the original
2340 * page by calling the destructor.
2342 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2344 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2346 return skb_copy_ubufs(skb, gfp_mask);
2350 * __skb_queue_purge - empty a list
2351 * @list: list to empty
2353 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2354 * the list and one reference dropped. This function does not take the
2355 * list lock and the caller must hold the relevant locks to use it.
2357 void skb_queue_purge(struct sk_buff_head *list);
2358 static inline void __skb_queue_purge(struct sk_buff_head *list)
2360 struct sk_buff *skb;
2361 while ((skb = __skb_dequeue(list)) != NULL)
2365 void *netdev_alloc_frag(unsigned int fragsz);
2367 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2371 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2372 * @dev: network device to receive on
2373 * @length: length to allocate
2375 * Allocate a new &sk_buff and assign it a usage count of one. The
2376 * buffer has unspecified headroom built in. Users should allocate
2377 * the headroom they think they need without accounting for the
2378 * built in space. The built in space is used for optimisations.
2380 * %NULL is returned if there is no free memory. Although this function
2381 * allocates memory it can be called from an interrupt.
2383 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2384 unsigned int length)
2386 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2389 /* legacy helper around __netdev_alloc_skb() */
2390 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2393 return __netdev_alloc_skb(NULL, length, gfp_mask);
2396 /* legacy helper around netdev_alloc_skb() */
2397 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2399 return netdev_alloc_skb(NULL, length);
2403 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2404 unsigned int length, gfp_t gfp)
2406 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2408 if (NET_IP_ALIGN && skb)
2409 skb_reserve(skb, NET_IP_ALIGN);
2413 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2414 unsigned int length)
2416 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2419 static inline void skb_free_frag(void *addr)
2421 __free_page_frag(addr);
2424 void *napi_alloc_frag(unsigned int fragsz);
2425 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2426 unsigned int length, gfp_t gfp_mask);
2427 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2428 unsigned int length)
2430 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2432 void napi_consume_skb(struct sk_buff *skb, int budget);
2434 void __kfree_skb_flush(void);
2435 void __kfree_skb_defer(struct sk_buff *skb);
2438 * __dev_alloc_pages - allocate page for network Rx
2439 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2440 * @order: size of the allocation
2442 * Allocate a new page.
2444 * %NULL is returned if there is no free memory.
2446 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2449 /* This piece of code contains several assumptions.
2450 * 1. This is for device Rx, therefor a cold page is preferred.
2451 * 2. The expectation is the user wants a compound page.
2452 * 3. If requesting a order 0 page it will not be compound
2453 * due to the check to see if order has a value in prep_new_page
2454 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2455 * code in gfp_to_alloc_flags that should be enforcing this.
2457 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2459 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2462 static inline struct page *dev_alloc_pages(unsigned int order)
2464 return __dev_alloc_pages(GFP_ATOMIC, order);
2468 * __dev_alloc_page - allocate a page for network Rx
2469 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2471 * Allocate a new page.
2473 * %NULL is returned if there is no free memory.
2475 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2477 return __dev_alloc_pages(gfp_mask, 0);
2480 static inline struct page *dev_alloc_page(void)
2482 return __dev_alloc_page(GFP_ATOMIC);
2486 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2487 * @page: The page that was allocated from skb_alloc_page
2488 * @skb: The skb that may need pfmemalloc set
2490 static inline void skb_propagate_pfmemalloc(struct page *page,
2491 struct sk_buff *skb)
2493 if (page_is_pfmemalloc(page))
2494 skb->pfmemalloc = true;
2498 * skb_frag_page - retrieve the page referred to by a paged fragment
2499 * @frag: the paged fragment
2501 * Returns the &struct page associated with @frag.
2503 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2505 return frag->page.p;
2509 * __skb_frag_ref - take an addition reference on a paged fragment.
2510 * @frag: the paged fragment
2512 * Takes an additional reference on the paged fragment @frag.
2514 static inline void __skb_frag_ref(skb_frag_t *frag)
2516 get_page(skb_frag_page(frag));
2520 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2522 * @f: the fragment offset.
2524 * Takes an additional reference on the @f'th paged fragment of @skb.
2526 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2528 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2532 * __skb_frag_unref - release a reference on a paged fragment.
2533 * @frag: the paged fragment
2535 * Releases a reference on the paged fragment @frag.
2537 static inline void __skb_frag_unref(skb_frag_t *frag)
2539 put_page(skb_frag_page(frag));
2543 * skb_frag_unref - release a reference on a paged fragment of an skb.
2545 * @f: the fragment offset
2547 * Releases a reference on the @f'th paged fragment of @skb.
2549 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2551 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2555 * skb_frag_address - gets the address of the data contained in a paged fragment
2556 * @frag: the paged fragment buffer
2558 * Returns the address of the data within @frag. The page must already
2561 static inline void *skb_frag_address(const skb_frag_t *frag)
2563 return page_address(skb_frag_page(frag)) + frag->page_offset;
2567 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2568 * @frag: the paged fragment buffer
2570 * Returns the address of the data within @frag. Checks that the page
2571 * is mapped and returns %NULL otherwise.
2573 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2575 void *ptr = page_address(skb_frag_page(frag));
2579 return ptr + frag->page_offset;
2583 * __skb_frag_set_page - sets the page contained in a paged fragment
2584 * @frag: the paged fragment
2585 * @page: the page to set
2587 * Sets the fragment @frag to contain @page.
2589 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2591 frag->page.p = page;
2595 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2597 * @f: the fragment offset
2598 * @page: the page to set
2600 * Sets the @f'th fragment of @skb to contain @page.
2602 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2605 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2608 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2611 * skb_frag_dma_map - maps a paged fragment via the DMA API
2612 * @dev: the device to map the fragment to
2613 * @frag: the paged fragment to map
2614 * @offset: the offset within the fragment (starting at the
2615 * fragment's own offset)
2616 * @size: the number of bytes to map
2617 * @dir: the direction of the mapping (%PCI_DMA_*)
2619 * Maps the page associated with @frag to @device.
2621 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2622 const skb_frag_t *frag,
2623 size_t offset, size_t size,
2624 enum dma_data_direction dir)
2626 return dma_map_page(dev, skb_frag_page(frag),
2627 frag->page_offset + offset, size, dir);
2630 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2633 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2637 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2640 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2645 * skb_clone_writable - is the header of a clone writable
2646 * @skb: buffer to check
2647 * @len: length up to which to write
2649 * Returns true if modifying the header part of the cloned buffer
2650 * does not requires the data to be copied.
2652 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2654 return !skb_header_cloned(skb) &&
2655 skb_headroom(skb) + len <= skb->hdr_len;
2658 static inline int skb_try_make_writable(struct sk_buff *skb,
2659 unsigned int write_len)
2661 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2662 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2665 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2670 if (headroom > skb_headroom(skb))
2671 delta = headroom - skb_headroom(skb);
2673 if (delta || cloned)
2674 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2680 * skb_cow - copy header of skb when it is required
2681 * @skb: buffer to cow
2682 * @headroom: needed headroom
2684 * If the skb passed lacks sufficient headroom or its data part
2685 * is shared, data is reallocated. If reallocation fails, an error
2686 * is returned and original skb is not changed.
2688 * The result is skb with writable area skb->head...skb->tail
2689 * and at least @headroom of space at head.
2691 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2693 return __skb_cow(skb, headroom, skb_cloned(skb));
2697 * skb_cow_head - skb_cow but only making the head writable
2698 * @skb: buffer to cow
2699 * @headroom: needed headroom
2701 * This function is identical to skb_cow except that we replace the
2702 * skb_cloned check by skb_header_cloned. It should be used when
2703 * you only need to push on some header and do not need to modify
2706 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2708 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2712 * skb_padto - pad an skbuff up to a minimal size
2713 * @skb: buffer to pad
2714 * @len: minimal length
2716 * Pads up a buffer to ensure the trailing bytes exist and are
2717 * blanked. If the buffer already contains sufficient data it
2718 * is untouched. Otherwise it is extended. Returns zero on
2719 * success. The skb is freed on error.
2721 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2723 unsigned int size = skb->len;
2724 if (likely(size >= len))
2726 return skb_pad(skb, len - size);
2730 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2731 * @skb: buffer to pad
2732 * @len: minimal length
2734 * Pads up a buffer to ensure the trailing bytes exist and are
2735 * blanked. If the buffer already contains sufficient data it
2736 * is untouched. Otherwise it is extended. Returns zero on
2737 * success. The skb is freed on error.
2739 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2741 unsigned int size = skb->len;
2743 if (unlikely(size < len)) {
2745 if (skb_pad(skb, len))
2747 __skb_put(skb, len);
2752 static inline int skb_add_data(struct sk_buff *skb,
2753 struct iov_iter *from, int copy)
2755 const int off = skb->len;
2757 if (skb->ip_summed == CHECKSUM_NONE) {
2759 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2760 &csum, from) == copy) {
2761 skb->csum = csum_block_add(skb->csum, csum, off);
2764 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
2767 __skb_trim(skb, off);
2771 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2772 const struct page *page, int off)
2775 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2777 return page == skb_frag_page(frag) &&
2778 off == frag->page_offset + skb_frag_size(frag);
2783 static inline int __skb_linearize(struct sk_buff *skb)
2785 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2789 * skb_linearize - convert paged skb to linear one
2790 * @skb: buffer to linarize
2792 * If there is no free memory -ENOMEM is returned, otherwise zero
2793 * is returned and the old skb data released.
2795 static inline int skb_linearize(struct sk_buff *skb)
2797 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2801 * skb_has_shared_frag - can any frag be overwritten
2802 * @skb: buffer to test
2804 * Return true if the skb has at least one frag that might be modified
2805 * by an external entity (as in vmsplice()/sendfile())
2807 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2809 return skb_is_nonlinear(skb) &&
2810 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2814 * skb_linearize_cow - make sure skb is linear and writable
2815 * @skb: buffer to process
2817 * If there is no free memory -ENOMEM is returned, otherwise zero
2818 * is returned and the old skb data released.
2820 static inline int skb_linearize_cow(struct sk_buff *skb)
2822 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2823 __skb_linearize(skb) : 0;
2827 * skb_postpull_rcsum - update checksum for received skb after pull
2828 * @skb: buffer to update
2829 * @start: start of data before pull
2830 * @len: length of data pulled
2832 * After doing a pull on a received packet, you need to call this to
2833 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2834 * CHECKSUM_NONE so that it can be recomputed from scratch.
2837 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2838 const void *start, unsigned int len)
2840 if (skb->ip_summed == CHECKSUM_COMPLETE)
2841 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2842 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2843 skb_checksum_start_offset(skb) < 0)
2844 skb->ip_summed = CHECKSUM_NONE;
2847 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2849 static inline void skb_postpush_rcsum(struct sk_buff *skb,
2850 const void *start, unsigned int len)
2852 /* For performing the reverse operation to skb_postpull_rcsum(),
2853 * we can instead of ...
2855 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0));
2857 * ... just use this equivalent version here to save a few
2858 * instructions. Feeding csum of 0 in csum_partial() and later
2859 * on adding skb->csum is equivalent to feed skb->csum in the
2862 if (skb->ip_summed == CHECKSUM_COMPLETE)
2863 skb->csum = csum_partial(start, len, skb->csum);
2867 * pskb_trim_rcsum - trim received skb and update checksum
2868 * @skb: buffer to trim
2871 * This is exactly the same as pskb_trim except that it ensures the
2872 * checksum of received packets are still valid after the operation.
2875 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2877 if (likely(len >= skb->len))
2879 if (skb->ip_summed == CHECKSUM_COMPLETE)
2880 skb->ip_summed = CHECKSUM_NONE;
2881 return __pskb_trim(skb, len);
2884 #define skb_queue_walk(queue, skb) \
2885 for (skb = (queue)->next; \
2886 skb != (struct sk_buff *)(queue); \
2889 #define skb_queue_walk_safe(queue, skb, tmp) \
2890 for (skb = (queue)->next, tmp = skb->next; \
2891 skb != (struct sk_buff *)(queue); \
2892 skb = tmp, tmp = skb->next)
2894 #define skb_queue_walk_from(queue, skb) \
2895 for (; skb != (struct sk_buff *)(queue); \
2898 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2899 for (tmp = skb->next; \
2900 skb != (struct sk_buff *)(queue); \
2901 skb = tmp, tmp = skb->next)
2903 #define skb_queue_reverse_walk(queue, skb) \
2904 for (skb = (queue)->prev; \
2905 skb != (struct sk_buff *)(queue); \
2908 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2909 for (skb = (queue)->prev, tmp = skb->prev; \
2910 skb != (struct sk_buff *)(queue); \
2911 skb = tmp, tmp = skb->prev)
2913 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2914 for (tmp = skb->prev; \
2915 skb != (struct sk_buff *)(queue); \
2916 skb = tmp, tmp = skb->prev)
2918 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2920 return skb_shinfo(skb)->frag_list != NULL;
2923 static inline void skb_frag_list_init(struct sk_buff *skb)
2925 skb_shinfo(skb)->frag_list = NULL;
2928 #define skb_walk_frags(skb, iter) \
2929 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2932 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
2933 const struct sk_buff *skb);
2934 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
2935 int *peeked, int *off, int *err,
2936 struct sk_buff **last);
2937 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2938 int *peeked, int *off, int *err);
2939 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2941 unsigned int datagram_poll(struct file *file, struct socket *sock,
2942 struct poll_table_struct *wait);
2943 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2944 struct iov_iter *to, int size);
2945 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2946 struct msghdr *msg, int size)
2948 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
2950 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2951 struct msghdr *msg);
2952 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2953 struct iov_iter *from, int len);
2954 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
2955 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2956 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
2957 static inline void skb_free_datagram_locked(struct sock *sk,
2958 struct sk_buff *skb)
2960 __skb_free_datagram_locked(sk, skb, 0);
2962 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2963 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2964 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2965 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2966 int len, __wsum csum);
2967 ssize_t skb_socket_splice(struct sock *sk,
2968 struct pipe_inode_info *pipe,
2969 struct splice_pipe_desc *spd);
2970 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2971 struct pipe_inode_info *pipe, unsigned int len,
2973 ssize_t (*splice_cb)(struct sock *,
2974 struct pipe_inode_info *,
2975 struct splice_pipe_desc *));
2976 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2977 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2978 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2980 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2981 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2982 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2983 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2984 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2985 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
2986 int skb_ensure_writable(struct sk_buff *skb, int write_len);
2987 int skb_vlan_pop(struct sk_buff *skb);
2988 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
2990 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2992 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
2995 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
2997 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3000 struct skb_checksum_ops {
3001 __wsum (*update)(const void *mem, int len, __wsum wsum);
3002 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3005 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3006 __wsum csum, const struct skb_checksum_ops *ops);
3007 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3010 static inline void * __must_check
3011 __skb_header_pointer(const struct sk_buff *skb, int offset,
3012 int len, void *data, int hlen, void *buffer)
3014 if (hlen - offset >= len)
3015 return data + offset;
3018 skb_copy_bits(skb, offset, buffer, len) < 0)
3024 static inline void * __must_check
3025 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3027 return __skb_header_pointer(skb, offset, len, skb->data,
3028 skb_headlen(skb), buffer);
3032 * skb_needs_linearize - check if we need to linearize a given skb
3033 * depending on the given device features.
3034 * @skb: socket buffer to check
3035 * @features: net device features
3037 * Returns true if either:
3038 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3039 * 2. skb is fragmented and the device does not support SG.
3041 static inline bool skb_needs_linearize(struct sk_buff *skb,
3042 netdev_features_t features)
3044 return skb_is_nonlinear(skb) &&
3045 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3046 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3049 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3051 const unsigned int len)
3053 memcpy(to, skb->data, len);
3056 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3057 const int offset, void *to,
3058 const unsigned int len)
3060 memcpy(to, skb->data + offset, len);
3063 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3065 const unsigned int len)
3067 memcpy(skb->data, from, len);
3070 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3073 const unsigned int len)
3075 memcpy(skb->data + offset, from, len);
3078 void skb_init(void);
3080 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3086 * skb_get_timestamp - get timestamp from a skb
3087 * @skb: skb to get stamp from
3088 * @stamp: pointer to struct timeval to store stamp in
3090 * Timestamps are stored in the skb as offsets to a base timestamp.
3091 * This function converts the offset back to a struct timeval and stores
3094 static inline void skb_get_timestamp(const struct sk_buff *skb,
3095 struct timeval *stamp)
3097 *stamp = ktime_to_timeval(skb->tstamp);
3100 static inline void skb_get_timestampns(const struct sk_buff *skb,
3101 struct timespec *stamp)
3103 *stamp = ktime_to_timespec(skb->tstamp);
3106 static inline void __net_timestamp(struct sk_buff *skb)
3108 skb->tstamp = ktime_get_real();
3111 static inline ktime_t net_timedelta(ktime_t t)
3113 return ktime_sub(ktime_get_real(), t);
3116 static inline ktime_t net_invalid_timestamp(void)
3118 return ktime_set(0, 0);
3121 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3123 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3125 void skb_clone_tx_timestamp(struct sk_buff *skb);
3126 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3128 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3130 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3134 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3139 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3142 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3144 * PHY drivers may accept clones of transmitted packets for
3145 * timestamping via their phy_driver.txtstamp method. These drivers
3146 * must call this function to return the skb back to the stack with a
3149 * @skb: clone of the the original outgoing packet
3150 * @hwtstamps: hardware time stamps
3153 void skb_complete_tx_timestamp(struct sk_buff *skb,
3154 struct skb_shared_hwtstamps *hwtstamps);
3156 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3157 struct skb_shared_hwtstamps *hwtstamps,
3158 struct sock *sk, int tstype);
3161 * skb_tstamp_tx - queue clone of skb with send time stamps
3162 * @orig_skb: the original outgoing packet
3163 * @hwtstamps: hardware time stamps, may be NULL if not available
3165 * If the skb has a socket associated, then this function clones the
3166 * skb (thus sharing the actual data and optional structures), stores
3167 * the optional hardware time stamping information (if non NULL) or
3168 * generates a software time stamp (otherwise), then queues the clone
3169 * to the error queue of the socket. Errors are silently ignored.
3171 void skb_tstamp_tx(struct sk_buff *orig_skb,
3172 struct skb_shared_hwtstamps *hwtstamps);
3174 static inline void sw_tx_timestamp(struct sk_buff *skb)
3176 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3177 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
3178 skb_tstamp_tx(skb, NULL);
3182 * skb_tx_timestamp() - Driver hook for transmit timestamping
3184 * Ethernet MAC Drivers should call this function in their hard_xmit()
3185 * function immediately before giving the sk_buff to the MAC hardware.
3187 * Specifically, one should make absolutely sure that this function is
3188 * called before TX completion of this packet can trigger. Otherwise
3189 * the packet could potentially already be freed.
3191 * @skb: A socket buffer.
3193 static inline void skb_tx_timestamp(struct sk_buff *skb)
3195 skb_clone_tx_timestamp(skb);
3196 sw_tx_timestamp(skb);
3200 * skb_complete_wifi_ack - deliver skb with wifi status
3202 * @skb: the original outgoing packet
3203 * @acked: ack status
3206 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3208 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3209 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3211 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3213 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3215 (skb->ip_summed == CHECKSUM_PARTIAL &&
3216 skb_checksum_start_offset(skb) >= 0));
3220 * skb_checksum_complete - Calculate checksum of an entire packet
3221 * @skb: packet to process
3223 * This function calculates the checksum over the entire packet plus
3224 * the value of skb->csum. The latter can be used to supply the
3225 * checksum of a pseudo header as used by TCP/UDP. It returns the
3228 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3229 * this function can be used to verify that checksum on received
3230 * packets. In that case the function should return zero if the
3231 * checksum is correct. In particular, this function will return zero
3232 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3233 * hardware has already verified the correctness of the checksum.
3235 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3237 return skb_csum_unnecessary(skb) ?
3238 0 : __skb_checksum_complete(skb);
3241 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3243 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3244 if (skb->csum_level == 0)
3245 skb->ip_summed = CHECKSUM_NONE;
3251 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3253 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3254 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3256 } else if (skb->ip_summed == CHECKSUM_NONE) {
3257 skb->ip_summed = CHECKSUM_UNNECESSARY;
3258 skb->csum_level = 0;
3262 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3264 /* Mark current checksum as bad (typically called from GRO
3265 * path). In the case that ip_summed is CHECKSUM_NONE
3266 * this must be the first checksum encountered in the packet.
3267 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3268 * checksum after the last one validated. For UDP, a zero
3269 * checksum can not be marked as bad.
3272 if (skb->ip_summed == CHECKSUM_NONE ||
3273 skb->ip_summed == CHECKSUM_UNNECESSARY)
3277 /* Check if we need to perform checksum complete validation.
3279 * Returns true if checksum complete is needed, false otherwise
3280 * (either checksum is unnecessary or zero checksum is allowed).
3282 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3286 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3287 skb->csum_valid = 1;
3288 __skb_decr_checksum_unnecessary(skb);
3295 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3298 #define CHECKSUM_BREAK 76
3300 /* Unset checksum-complete
3302 * Unset checksum complete can be done when packet is being modified
3303 * (uncompressed for instance) and checksum-complete value is
3306 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3308 if (skb->ip_summed == CHECKSUM_COMPLETE)
3309 skb->ip_summed = CHECKSUM_NONE;
3312 /* Validate (init) checksum based on checksum complete.
3315 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3316 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3317 * checksum is stored in skb->csum for use in __skb_checksum_complete
3318 * non-zero: value of invalid checksum
3321 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3325 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3326 if (!csum_fold(csum_add(psum, skb->csum))) {
3327 skb->csum_valid = 1;
3330 } else if (skb->csum_bad) {
3331 /* ip_summed == CHECKSUM_NONE in this case */
3332 return (__force __sum16)1;
3337 if (complete || skb->len <= CHECKSUM_BREAK) {
3340 csum = __skb_checksum_complete(skb);
3341 skb->csum_valid = !csum;
3348 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3353 /* Perform checksum validate (init). Note that this is a macro since we only
3354 * want to calculate the pseudo header which is an input function if necessary.
3355 * First we try to validate without any computation (checksum unnecessary) and
3356 * then calculate based on checksum complete calling the function to compute
3360 * 0: checksum is validated or try to in skb_checksum_complete
3361 * non-zero: value of invalid checksum
3363 #define __skb_checksum_validate(skb, proto, complete, \
3364 zero_okay, check, compute_pseudo) \
3366 __sum16 __ret = 0; \
3367 skb->csum_valid = 0; \
3368 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3369 __ret = __skb_checksum_validate_complete(skb, \
3370 complete, compute_pseudo(skb, proto)); \
3374 #define skb_checksum_init(skb, proto, compute_pseudo) \
3375 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3377 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3378 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3380 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3381 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3383 #define skb_checksum_validate_zero_check(skb, proto, check, \
3385 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3387 #define skb_checksum_simple_validate(skb) \
3388 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3390 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3392 return (skb->ip_summed == CHECKSUM_NONE &&
3393 skb->csum_valid && !skb->csum_bad);
3396 static inline void __skb_checksum_convert(struct sk_buff *skb,
3397 __sum16 check, __wsum pseudo)
3399 skb->csum = ~pseudo;
3400 skb->ip_summed = CHECKSUM_COMPLETE;
3403 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3405 if (__skb_checksum_convert_check(skb)) \
3406 __skb_checksum_convert(skb, check, \
3407 compute_pseudo(skb, proto)); \
3410 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3411 u16 start, u16 offset)
3413 skb->ip_summed = CHECKSUM_PARTIAL;
3414 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3415 skb->csum_offset = offset - start;
3418 /* Update skbuf and packet to reflect the remote checksum offload operation.
3419 * When called, ptr indicates the starting point for skb->csum when
3420 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3421 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3423 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3424 int start, int offset, bool nopartial)
3429 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3433 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3434 __skb_checksum_complete(skb);
3435 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3438 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3440 /* Adjust skb->csum since we changed the packet */
3441 skb->csum = csum_add(skb->csum, delta);
3444 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3445 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3446 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3448 if (nfct && atomic_dec_and_test(&nfct->use))
3449 nf_conntrack_destroy(nfct);
3451 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3454 atomic_inc(&nfct->use);
3457 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3458 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3460 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3463 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3466 atomic_inc(&nf_bridge->use);
3468 #endif /* CONFIG_BRIDGE_NETFILTER */
3469 static inline void nf_reset(struct sk_buff *skb)
3471 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3472 nf_conntrack_put(skb->nfct);
3475 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3476 nf_bridge_put(skb->nf_bridge);
3477 skb->nf_bridge = NULL;
3481 static inline void nf_reset_trace(struct sk_buff *skb)
3483 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3488 /* Note: This doesn't put any conntrack and bridge info in dst. */
3489 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3492 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3493 dst->nfct = src->nfct;
3494 nf_conntrack_get(src->nfct);
3496 dst->nfctinfo = src->nfctinfo;
3498 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3499 dst->nf_bridge = src->nf_bridge;
3500 nf_bridge_get(src->nf_bridge);
3502 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3504 dst->nf_trace = src->nf_trace;
3508 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3510 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3511 nf_conntrack_put(dst->nfct);
3513 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3514 nf_bridge_put(dst->nf_bridge);
3516 __nf_copy(dst, src, true);
3519 #ifdef CONFIG_NETWORK_SECMARK
3520 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3522 to->secmark = from->secmark;
3525 static inline void skb_init_secmark(struct sk_buff *skb)
3530 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3533 static inline void skb_init_secmark(struct sk_buff *skb)
3537 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3539 return !skb->destructor &&
3540 #if IS_ENABLED(CONFIG_XFRM)
3543 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3546 !skb->_skb_refdst &&
3547 !skb_has_frag_list(skb);
3550 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3552 skb->queue_mapping = queue_mapping;
3555 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3557 return skb->queue_mapping;
3560 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3562 to->queue_mapping = from->queue_mapping;
3565 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3567 skb->queue_mapping = rx_queue + 1;
3570 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3572 return skb->queue_mapping - 1;
3575 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3577 return skb->queue_mapping != 0;
3580 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3589 /* Keeps track of mac header offset relative to skb->head.
3590 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3591 * For non-tunnel skb it points to skb_mac_header() and for
3592 * tunnel skb it points to outer mac header.
3593 * Keeps track of level of encapsulation of network headers.
3604 #define SKB_SGO_CB_OFFSET 32
3605 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3607 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3609 return (skb_mac_header(inner_skb) - inner_skb->head) -
3610 SKB_GSO_CB(inner_skb)->mac_offset;
3613 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3615 int new_headroom, headroom;
3618 headroom = skb_headroom(skb);
3619 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3623 new_headroom = skb_headroom(skb);
3624 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3628 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3630 /* Do not update partial checksums if remote checksum is enabled. */
3631 if (skb->remcsum_offload)
3634 SKB_GSO_CB(skb)->csum = res;
3635 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3638 /* Compute the checksum for a gso segment. First compute the checksum value
3639 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3640 * then add in skb->csum (checksum from csum_start to end of packet).
3641 * skb->csum and csum_start are then updated to reflect the checksum of the
3642 * resultant packet starting from the transport header-- the resultant checksum
3643 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3646 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3648 unsigned char *csum_start = skb_transport_header(skb);
3649 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3650 __wsum partial = SKB_GSO_CB(skb)->csum;
3652 SKB_GSO_CB(skb)->csum = res;
3653 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
3655 return csum_fold(csum_partial(csum_start, plen, partial));
3658 static inline bool skb_is_gso(const struct sk_buff *skb)
3660 return skb_shinfo(skb)->gso_size;
3663 /* Note: Should be called only if skb_is_gso(skb) is true */
3664 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3666 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3669 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3671 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3673 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3674 * wanted then gso_type will be set. */
3675 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3677 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3678 unlikely(shinfo->gso_type == 0)) {
3679 __skb_warn_lro_forwarding(skb);
3685 static inline void skb_forward_csum(struct sk_buff *skb)
3687 /* Unfortunately we don't support this one. Any brave souls? */
3688 if (skb->ip_summed == CHECKSUM_COMPLETE)
3689 skb->ip_summed = CHECKSUM_NONE;
3693 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3694 * @skb: skb to check
3696 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3697 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3698 * use this helper, to document places where we make this assertion.
3700 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3703 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3707 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3709 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3710 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3711 unsigned int transport_len,
3712 __sum16(*skb_chkf)(struct sk_buff *skb));
3715 * skb_head_is_locked - Determine if the skb->head is locked down
3716 * @skb: skb to check
3718 * The head on skbs build around a head frag can be removed if they are
3719 * not cloned. This function returns true if the skb head is locked down
3720 * due to either being allocated via kmalloc, or by being a clone with
3721 * multiple references to the head.
3723 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3725 return !skb->head_frag || skb_cloned(skb);
3729 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3733 * skb_gso_network_seglen is used to determine the real size of the
3734 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3736 * The MAC/L2 header is not accounted for.
3738 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3740 unsigned int hdr_len = skb_transport_header(skb) -
3741 skb_network_header(skb);
3742 return hdr_len + skb_gso_transport_seglen(skb);
3745 /* Local Checksum Offload.
3746 * Compute outer checksum based on the assumption that the
3747 * inner checksum will be offloaded later.
3748 * See Documentation/networking/checksum-offloads.txt for
3749 * explanation of how this works.
3750 * Fill in outer checksum adjustment (e.g. with sum of outer
3751 * pseudo-header) before calling.
3752 * Also ensure that inner checksum is in linear data area.
3754 static inline __wsum lco_csum(struct sk_buff *skb)
3756 unsigned char *csum_start = skb_checksum_start(skb);
3757 unsigned char *l4_hdr = skb_transport_header(skb);
3760 /* Start with complement of inner checksum adjustment */
3761 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3764 /* Add in checksum of our headers (incl. outer checksum
3765 * adjustment filled in by caller) and return result.
3767 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
3770 #endif /* __KERNEL__ */
3771 #endif /* _LINUX_SKBUFF_H */