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>
40 #include <linux/if_packet.h>
43 /* The interface for checksum offload between the stack and networking drivers
46 * A. IP checksum related features
48 * Drivers advertise checksum offload capabilities in the features of a device.
49 * From the stack's point of view these are capabilities offered by the driver,
50 * a driver typically only advertises features that it is capable of offloading
53 * The checksum related features are:
55 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
56 * IP (one's complement) checksum for any combination
57 * of protocols or protocol layering. The checksum is
58 * computed and set in a packet per the CHECKSUM_PARTIAL
59 * interface (see below).
61 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
62 * TCP or UDP packets over IPv4. These are specifically
63 * unencapsulated packets of the form IPv4|TCP or
64 * IPv4|UDP where the Protocol field in the IPv4 header
65 * is TCP or UDP. The IPv4 header may contain IP options
66 * This feature cannot be set in features for a device
67 * with NETIF_F_HW_CSUM also set. This feature is being
68 * DEPRECATED (see below).
70 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
71 * TCP or UDP packets over IPv6. These are specifically
72 * unencapsulated packets of the form IPv6|TCP or
73 * IPv4|UDP where the Next Header field in the IPv6
74 * header is either TCP or UDP. IPv6 extension headers
75 * are not supported with this feature. This feature
76 * cannot be set in features for a device with
77 * NETIF_F_HW_CSUM also set. This feature is being
78 * DEPRECATED (see below).
80 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
81 * This flag is used only used to disable the RX checksum
82 * feature for a device. The stack will accept receive
83 * checksum indication in packets received on a device
84 * regardless of whether NETIF_F_RXCSUM is set.
86 * B. Checksumming of received packets by device. Indication of checksum
87 * verification is in set skb->ip_summed. Possible values are:
91 * Device did not checksum this packet e.g. due to lack of capabilities.
92 * The packet contains full (though not verified) checksum in packet but
93 * not in skb->csum. Thus, skb->csum is undefined in this case.
95 * CHECKSUM_UNNECESSARY:
97 * The hardware you're dealing with doesn't calculate the full checksum
98 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
99 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
100 * if their checksums are okay. skb->csum is still undefined in this case
101 * though. A driver or device must never modify the checksum field in the
102 * packet even if checksum is verified.
104 * CHECKSUM_UNNECESSARY is applicable to following protocols:
105 * TCP: IPv6 and IPv4.
106 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
107 * zero UDP checksum for either IPv4 or IPv6, the networking stack
108 * may perform further validation in this case.
109 * GRE: only if the checksum is present in the header.
110 * SCTP: indicates the CRC in SCTP header has been validated.
112 * skb->csum_level indicates the number of consecutive checksums found in
113 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
114 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
115 * and a device is able to verify the checksums for UDP (possibly zero),
116 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
117 * two. If the device were only able to verify the UDP checksum and not
118 * GRE, either because it doesn't support GRE checksum of because GRE
119 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
120 * not considered in this case).
124 * This is the most generic way. The device supplied checksum of the _whole_
125 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
126 * hardware doesn't need to parse L3/L4 headers to implement this.
128 * Note: Even if device supports only some protocols, but is able to produce
129 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
133 * A checksum is set up to be offloaded to a device as described in the
134 * output description for CHECKSUM_PARTIAL. This may occur on a packet
135 * received directly from another Linux OS, e.g., a virtualized Linux kernel
136 * on the same host, or it may be set in the input path in GRO or remote
137 * checksum offload. For the purposes of checksum verification, the checksum
138 * referred to by skb->csum_start + skb->csum_offset and any preceding
139 * checksums in the packet are considered verified. Any checksums in the
140 * packet that are after the checksum being offloaded are not considered to
143 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
144 * in the skb->ip_summed for a packet. Values are:
148 * The driver is required to checksum the packet as seen by hard_start_xmit()
149 * from skb->csum_start up to the end, and to record/write the checksum at
150 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
151 * csum_start and csum_offset values are valid values given the length and
152 * offset of the packet, however they should not attempt to validate that the
153 * checksum refers to a legitimate transport layer checksum-- it is the
154 * purview of the stack to validate that csum_start and csum_offset are set
157 * When the stack requests checksum offload for a packet, the driver MUST
158 * ensure that the checksum is set correctly. A driver can either offload the
159 * checksum calculation to the device, or call skb_checksum_help (in the case
160 * that the device does not support offload for a particular checksum).
162 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
163 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
164 * checksum offload capability. If a device has limited checksum capabilities
165 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as
166 * described above) a helper function can be called to resolve
167 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper
168 * function takes a spec argument that describes the protocol layer that is
169 * supported for checksum offload and can be called for each packet. If a
170 * packet does not match the specification for offload, skb_checksum_help
171 * is called to resolve the checksum.
175 * The skb was already checksummed by the protocol, or a checksum is not
178 * CHECKSUM_UNNECESSARY:
180 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
184 * Not used in checksum output. If a driver observes a packet with this value
185 * set in skbuff, if should treat as CHECKSUM_NONE being set.
187 * D. Non-IP checksum (CRC) offloads
189 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
190 * offloading the SCTP CRC in a packet. To perform this offload the stack
191 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
192 * accordingly. Note the there is no indication in the skbuff that the
193 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports
194 * both IP checksum offload and SCTP CRC offload must verify which offload
195 * is configured for a packet presumably by inspecting packet headers.
197 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
198 * offloading the FCOE CRC in a packet. To perform this offload the stack
199 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
200 * accordingly. Note the there is no indication in the skbuff that the
201 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
202 * both IP checksum offload and FCOE CRC offload must verify which offload
203 * is configured for a packet presumably by inspecting packet headers.
205 * E. Checksumming on output with GSO.
207 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
208 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
209 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
210 * part of the GSO operation is implied. If a checksum is being offloaded
211 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
212 * are set to refer to the outermost checksum being offload (two offloaded
213 * checksums are possible with UDP encapsulation).
216 /* Don't change this without changing skb_csum_unnecessary! */
217 #define CHECKSUM_NONE 0
218 #define CHECKSUM_UNNECESSARY 1
219 #define CHECKSUM_COMPLETE 2
220 #define CHECKSUM_PARTIAL 3
222 /* Maximum value in skb->csum_level */
223 #define SKB_MAX_CSUM_LEVEL 3
225 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
226 #define SKB_WITH_OVERHEAD(X) \
227 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
228 #define SKB_MAX_ORDER(X, ORDER) \
229 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
230 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
231 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
233 /* return minimum truesize of one skb containing X bytes of data */
234 #define SKB_TRUESIZE(X) ((X) + \
235 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
236 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
240 struct pipe_inode_info;
244 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
245 struct nf_conntrack {
250 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
251 struct nf_bridge_info {
254 BRNF_PROTO_UNCHANGED,
262 struct net_device *physindev;
264 /* always valid & non-NULL from FORWARD on, for physdev match */
265 struct net_device *physoutdev;
267 /* prerouting: detect dnat in orig/reply direction */
269 struct in6_addr ipv6_daddr;
271 /* after prerouting + nat detected: store original source
272 * mac since neigh resolution overwrites it, only used while
273 * skb is out in neigh layer.
275 char neigh_header[8];
280 struct sk_buff_head {
281 /* These two members must be first. */
282 struct sk_buff *next;
283 struct sk_buff *prev;
291 /* To allow 64K frame to be packed as single skb without frag_list we
292 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
293 * buffers which do not start on a page boundary.
295 * Since GRO uses frags we allocate at least 16 regardless of page
298 #if (65536/PAGE_SIZE + 1) < 16
299 #define MAX_SKB_FRAGS 16UL
301 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
303 extern int sysctl_max_skb_frags;
305 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
306 * segment using its current segmentation instead.
308 #define GSO_BY_FRAGS 0xFFFF
310 typedef struct skb_frag_struct skb_frag_t;
312 struct skb_frag_struct {
316 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
325 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
330 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
335 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
340 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
345 #define HAVE_HW_TIME_STAMP
348 * struct skb_shared_hwtstamps - hardware time stamps
349 * @hwtstamp: hardware time stamp transformed into duration
350 * since arbitrary point in time
352 * Software time stamps generated by ktime_get_real() are stored in
355 * hwtstamps can only be compared against other hwtstamps from
358 * This structure is attached to packets as part of the
359 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
361 struct skb_shared_hwtstamps {
365 /* Definitions for tx_flags in struct skb_shared_info */
367 /* generate hardware time stamp */
368 SKBTX_HW_TSTAMP = 1 << 0,
370 /* generate software time stamp when queueing packet to NIC */
371 SKBTX_SW_TSTAMP = 1 << 1,
373 /* device driver is going to provide hardware time stamp */
374 SKBTX_IN_PROGRESS = 1 << 2,
376 /* device driver supports TX zero-copy buffers */
377 SKBTX_DEV_ZEROCOPY = 1 << 3,
379 /* generate wifi status information (where possible) */
380 SKBTX_WIFI_STATUS = 1 << 4,
382 /* This indicates at least one fragment might be overwritten
383 * (as in vmsplice(), sendfile() ...)
384 * If we need to compute a TX checksum, we'll need to copy
385 * all frags to avoid possible bad checksum
387 SKBTX_SHARED_FRAG = 1 << 5,
389 /* generate software time stamp when entering packet scheduling */
390 SKBTX_SCHED_TSTAMP = 1 << 6,
393 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
395 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
398 * The callback notifies userspace to release buffers when skb DMA is done in
399 * lower device, the skb last reference should be 0 when calling this.
400 * The zerocopy_success argument is true if zero copy transmit occurred,
401 * false on data copy or out of memory error caused by data copy attempt.
402 * The ctx field is used to track device context.
403 * The desc field is used to track userspace buffer index.
406 void (*callback)(struct ubuf_info *, bool zerocopy_success);
411 /* This data is invariant across clones and lives at
412 * the end of the header data, ie. at skb->end.
414 struct skb_shared_info {
415 unsigned char nr_frags;
417 unsigned short gso_size;
418 /* Warning: this field is not always filled in (UFO)! */
419 unsigned short gso_segs;
420 unsigned short gso_type;
421 struct sk_buff *frag_list;
422 struct skb_shared_hwtstamps hwtstamps;
427 * Warning : all fields before dataref are cleared in __alloc_skb()
431 /* Intermediate layers must ensure that destructor_arg
432 * remains valid until skb destructor */
433 void * destructor_arg;
435 /* must be last field, see pskb_expand_head() */
436 skb_frag_t frags[MAX_SKB_FRAGS];
439 /* We divide dataref into two halves. The higher 16 bits hold references
440 * to the payload part of skb->data. The lower 16 bits hold references to
441 * the entire skb->data. A clone of a headerless skb holds the length of
442 * the header in skb->hdr_len.
444 * All users must obey the rule that the skb->data reference count must be
445 * greater than or equal to the payload reference count.
447 * Holding a reference to the payload part means that the user does not
448 * care about modifications to the header part of skb->data.
450 #define SKB_DATAREF_SHIFT 16
451 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
455 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
456 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
457 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
461 SKB_GSO_TCPV4 = 1 << 0,
462 SKB_GSO_UDP = 1 << 1,
464 /* This indicates the skb is from an untrusted source. */
465 SKB_GSO_DODGY = 1 << 2,
467 /* This indicates the tcp segment has CWR set. */
468 SKB_GSO_TCP_ECN = 1 << 3,
470 SKB_GSO_TCP_FIXEDID = 1 << 4,
472 SKB_GSO_TCPV6 = 1 << 5,
474 SKB_GSO_FCOE = 1 << 6,
476 SKB_GSO_GRE = 1 << 7,
478 SKB_GSO_GRE_CSUM = 1 << 8,
480 SKB_GSO_IPXIP4 = 1 << 9,
482 SKB_GSO_IPXIP6 = 1 << 10,
484 SKB_GSO_UDP_TUNNEL = 1 << 11,
486 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
488 SKB_GSO_PARTIAL = 1 << 13,
490 SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
492 SKB_GSO_SCTP = 1 << 15,
495 #if BITS_PER_LONG > 32
496 #define NET_SKBUFF_DATA_USES_OFFSET 1
499 #ifdef NET_SKBUFF_DATA_USES_OFFSET
500 typedef unsigned int sk_buff_data_t;
502 typedef unsigned char *sk_buff_data_t;
506 * struct skb_mstamp - multi resolution time stamps
507 * @stamp_us: timestamp in us resolution
508 * @stamp_jiffies: timestamp in jiffies
521 * skb_mstamp_get - get current timestamp
522 * @cl: place to store timestamps
524 static inline void skb_mstamp_get(struct skb_mstamp *cl)
526 u64 val = local_clock();
528 do_div(val, NSEC_PER_USEC);
529 cl->stamp_us = (u32)val;
530 cl->stamp_jiffies = (u32)jiffies;
534 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
535 * @t1: pointer to newest sample
536 * @t0: pointer to oldest sample
538 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
539 const struct skb_mstamp *t0)
541 s32 delta_us = t1->stamp_us - t0->stamp_us;
542 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
544 /* If delta_us is negative, this might be because interval is too big,
545 * or local_clock() drift is too big : fallback using jiffies.
548 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
550 delta_us = jiffies_to_usecs(delta_jiffies);
555 static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
556 const struct skb_mstamp *t0)
558 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
561 diff = t1->stamp_us - t0->stamp_us;
566 * struct sk_buff - socket buffer
567 * @next: Next buffer in list
568 * @prev: Previous buffer in list
569 * @tstamp: Time we arrived/left
570 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
571 * @sk: Socket we are owned by
572 * @dev: Device we arrived on/are leaving by
573 * @cb: Control buffer. Free for use by every layer. Put private vars here
574 * @_skb_refdst: destination entry (with norefcount bit)
575 * @sp: the security path, used for xfrm
576 * @len: Length of actual data
577 * @data_len: Data length
578 * @mac_len: Length of link layer header
579 * @hdr_len: writable header length of cloned skb
580 * @csum: Checksum (must include start/offset pair)
581 * @csum_start: Offset from skb->head where checksumming should start
582 * @csum_offset: Offset from csum_start where checksum should be stored
583 * @priority: Packet queueing priority
584 * @ignore_df: allow local fragmentation
585 * @cloned: Head may be cloned (check refcnt to be sure)
586 * @ip_summed: Driver fed us an IP checksum
587 * @nohdr: Payload reference only, must not modify header
588 * @nfctinfo: Relationship of this skb to the connection
589 * @pkt_type: Packet class
590 * @fclone: skbuff clone status
591 * @ipvs_property: skbuff is owned by ipvs
592 * @peeked: this packet has been seen already, so stats have been
593 * done for it, don't do them again
594 * @nf_trace: netfilter packet trace flag
595 * @protocol: Packet protocol from driver
596 * @destructor: Destruct function
597 * @nfct: Associated connection, if any
598 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
599 * @skb_iif: ifindex of device we arrived on
600 * @tc_index: Traffic control index
601 * @tc_verd: traffic control verdict
602 * @hash: the packet hash
603 * @queue_mapping: Queue mapping for multiqueue devices
604 * @xmit_more: More SKBs are pending for this queue
605 * @ndisc_nodetype: router type (from link layer)
606 * @ooo_okay: allow the mapping of a socket to a queue to be changed
607 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
609 * @sw_hash: indicates hash was computed in software stack
610 * @wifi_acked_valid: wifi_acked was set
611 * @wifi_acked: whether frame was acked on wifi or not
612 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
613 * @napi_id: id of the NAPI struct this skb came from
614 * @secmark: security marking
615 * @mark: Generic packet mark
616 * @vlan_proto: vlan encapsulation protocol
617 * @vlan_tci: vlan tag control information
618 * @inner_protocol: Protocol (encapsulation)
619 * @inner_transport_header: Inner transport layer header (encapsulation)
620 * @inner_network_header: Network layer header (encapsulation)
621 * @inner_mac_header: Link layer header (encapsulation)
622 * @transport_header: Transport layer header
623 * @network_header: Network layer header
624 * @mac_header: Link layer header
625 * @tail: Tail pointer
627 * @head: Head of buffer
628 * @data: Data head pointer
629 * @truesize: Buffer size
630 * @users: User count - see {datagram,tcp}.c
636 /* These two members must be first. */
637 struct sk_buff *next;
638 struct sk_buff *prev;
642 struct skb_mstamp skb_mstamp;
645 struct rb_node rbnode; /* used in netem & tcp stack */
648 struct net_device *dev;
651 * This is the control buffer. It is free to use for every
652 * layer. Please put your private variables there. If you
653 * want to keep them across layers you have to do a skb_clone()
654 * first. This is owned by whoever has the skb queued ATM.
656 char cb[48] __aligned(8);
658 unsigned long _skb_refdst;
659 void (*destructor)(struct sk_buff *skb);
663 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
664 struct nf_conntrack *nfct;
666 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
667 struct nf_bridge_info *nf_bridge;
674 /* Following fields are _not_ copied in __copy_skb_header()
675 * Note that queue_mapping is here mostly to fill a hole.
677 kmemcheck_bitfield_begin(flags1);
680 /* if you move cloned around you also must adapt those constants */
681 #ifdef __BIG_ENDIAN_BITFIELD
682 #define CLONED_MASK (1 << 7)
684 #define CLONED_MASK 1
686 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
688 __u8 __cloned_offset[0];
695 __unused:1; /* one bit hole */
696 kmemcheck_bitfield_end(flags1);
698 /* fields enclosed in headers_start/headers_end are copied
699 * using a single memcpy() in __copy_skb_header()
702 __u32 headers_start[0];
705 /* if you move pkt_type around you also must adapt those constants */
706 #ifdef __BIG_ENDIAN_BITFIELD
707 #define PKT_TYPE_MAX (7 << 5)
709 #define PKT_TYPE_MAX 7
711 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
713 __u8 __pkt_type_offset[0];
724 __u8 wifi_acked_valid:1;
728 /* Indicates the inner headers are valid in the skbuff. */
729 __u8 encapsulation:1;
730 __u8 encap_hdr_csum:1;
732 __u8 csum_complete_sw:1;
736 #ifdef CONFIG_IPV6_NDISC_NODETYPE
737 __u8 ndisc_nodetype:2;
739 __u8 ipvs_property:1;
740 __u8 inner_protocol_type:1;
741 __u8 remcsum_offload:1;
742 #ifdef CONFIG_NET_SWITCHDEV
743 __u8 offload_fwd_mark:1;
745 /* 2, 4 or 5 bit hole */
747 #ifdef CONFIG_NET_SCHED
748 __u16 tc_index; /* traffic control index */
749 #ifdef CONFIG_NET_CLS_ACT
750 __u16 tc_verd; /* traffic control verdict */
766 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
768 unsigned int napi_id;
769 unsigned int sender_cpu;
772 #ifdef CONFIG_NETWORK_SECMARK
778 __u32 reserved_tailroom;
782 __be16 inner_protocol;
786 __u16 inner_transport_header;
787 __u16 inner_network_header;
788 __u16 inner_mac_header;
791 __u16 transport_header;
792 __u16 network_header;
796 __u32 headers_end[0];
799 /* These elements must be at the end, see alloc_skb() for details. */
804 unsigned int truesize;
810 * Handling routines are only of interest to the kernel
812 #include <linux/slab.h>
815 #define SKB_ALLOC_FCLONE 0x01
816 #define SKB_ALLOC_RX 0x02
817 #define SKB_ALLOC_NAPI 0x04
819 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
820 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
822 return unlikely(skb->pfmemalloc);
826 * skb might have a dst pointer attached, refcounted or not.
827 * _skb_refdst low order bit is set if refcount was _not_ taken
829 #define SKB_DST_NOREF 1UL
830 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
833 * skb_dst - returns skb dst_entry
836 * Returns skb dst_entry, regardless of reference taken or not.
838 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
840 /* If refdst was not refcounted, check we still are in a
841 * rcu_read_lock section
843 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
844 !rcu_read_lock_held() &&
845 !rcu_read_lock_bh_held());
846 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
850 * skb_dst_set - sets skb dst
854 * Sets skb dst, assuming a reference was taken on dst and should
855 * be released by skb_dst_drop()
857 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
859 skb->_skb_refdst = (unsigned long)dst;
863 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
867 * Sets skb dst, assuming a reference was not taken on dst.
868 * If dst entry is cached, we do not take reference and dst_release
869 * will be avoided by refdst_drop. If dst entry is not cached, we take
870 * reference, so that last dst_release can destroy the dst immediately.
872 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
874 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
875 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
879 * skb_dst_is_noref - Test if skb dst isn't refcounted
882 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
884 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
887 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
889 return (struct rtable *)skb_dst(skb);
892 /* For mangling skb->pkt_type from user space side from applications
893 * such as nft, tc, etc, we only allow a conservative subset of
894 * possible pkt_types to be set.
896 static inline bool skb_pkt_type_ok(u32 ptype)
898 return ptype <= PACKET_OTHERHOST;
901 void kfree_skb(struct sk_buff *skb);
902 void kfree_skb_list(struct sk_buff *segs);
903 void skb_tx_error(struct sk_buff *skb);
904 void consume_skb(struct sk_buff *skb);
905 void __kfree_skb(struct sk_buff *skb);
906 extern struct kmem_cache *skbuff_head_cache;
908 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
909 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
910 bool *fragstolen, int *delta_truesize);
912 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
914 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
915 struct sk_buff *build_skb(void *data, unsigned int frag_size);
916 static inline struct sk_buff *alloc_skb(unsigned int size,
919 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
922 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
923 unsigned long data_len,
928 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
929 struct sk_buff_fclones {
938 * skb_fclone_busy - check if fclone is busy
942 * Returns true if skb is a fast clone, and its clone is not freed.
943 * Some drivers call skb_orphan() in their ndo_start_xmit(),
944 * so we also check that this didnt happen.
946 static inline bool skb_fclone_busy(const struct sock *sk,
947 const struct sk_buff *skb)
949 const struct sk_buff_fclones *fclones;
951 fclones = container_of(skb, struct sk_buff_fclones, skb1);
953 return skb->fclone == SKB_FCLONE_ORIG &&
954 atomic_read(&fclones->fclone_ref) > 1 &&
955 fclones->skb2.sk == sk;
958 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
961 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
964 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
965 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
967 return __alloc_skb_head(priority, -1);
970 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
971 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
972 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
973 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
974 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
975 gfp_t gfp_mask, bool fclone);
976 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
979 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
982 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
983 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
984 unsigned int headroom);
985 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
986 int newtailroom, gfp_t priority);
987 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
988 int offset, int len);
989 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
991 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
992 int skb_pad(struct sk_buff *skb, int pad);
993 #define dev_kfree_skb(a) consume_skb(a)
995 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
996 int getfrag(void *from, char *to, int offset,
997 int len, int odd, struct sk_buff *skb),
998 void *from, int length);
1000 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1001 int offset, size_t size);
1003 struct skb_seq_state {
1007 __u32 stepped_offset;
1008 struct sk_buff *root_skb;
1009 struct sk_buff *cur_skb;
1013 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1014 unsigned int to, struct skb_seq_state *st);
1015 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1016 struct skb_seq_state *st);
1017 void skb_abort_seq_read(struct skb_seq_state *st);
1019 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1020 unsigned int to, struct ts_config *config);
1023 * Packet hash types specify the type of hash in skb_set_hash.
1025 * Hash types refer to the protocol layer addresses which are used to
1026 * construct a packet's hash. The hashes are used to differentiate or identify
1027 * flows of the protocol layer for the hash type. Hash types are either
1028 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1030 * Properties of hashes:
1032 * 1) Two packets in different flows have different hash values
1033 * 2) Two packets in the same flow should have the same hash value
1035 * A hash at a higher layer is considered to be more specific. A driver should
1036 * set the most specific hash possible.
1038 * A driver cannot indicate a more specific hash than the layer at which a hash
1039 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1041 * A driver may indicate a hash level which is less specific than the
1042 * actual layer the hash was computed on. For instance, a hash computed
1043 * at L4 may be considered an L3 hash. This should only be done if the
1044 * driver can't unambiguously determine that the HW computed the hash at
1045 * the higher layer. Note that the "should" in the second property above
1048 enum pkt_hash_types {
1049 PKT_HASH_TYPE_NONE, /* Undefined type */
1050 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1051 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1052 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1055 static inline void skb_clear_hash(struct sk_buff *skb)
1062 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1065 skb_clear_hash(skb);
1069 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1071 skb->l4_hash = is_l4;
1072 skb->sw_hash = is_sw;
1077 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1079 /* Used by drivers to set hash from HW */
1080 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1084 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1086 __skb_set_hash(skb, hash, true, is_l4);
1089 void __skb_get_hash(struct sk_buff *skb);
1090 u32 __skb_get_hash_symmetric(struct sk_buff *skb);
1091 u32 skb_get_poff(const struct sk_buff *skb);
1092 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1093 const struct flow_keys *keys, int hlen);
1094 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1095 void *data, int hlen_proto);
1097 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1098 int thoff, u8 ip_proto)
1100 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1103 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1104 const struct flow_dissector_key *key,
1105 unsigned int key_count);
1107 bool __skb_flow_dissect(const struct sk_buff *skb,
1108 struct flow_dissector *flow_dissector,
1109 void *target_container,
1110 void *data, __be16 proto, int nhoff, int hlen,
1111 unsigned int flags);
1113 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1114 struct flow_dissector *flow_dissector,
1115 void *target_container, unsigned int flags)
1117 return __skb_flow_dissect(skb, flow_dissector, target_container,
1118 NULL, 0, 0, 0, flags);
1121 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1122 struct flow_keys *flow,
1125 memset(flow, 0, sizeof(*flow));
1126 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1127 NULL, 0, 0, 0, flags);
1130 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1131 void *data, __be16 proto,
1132 int nhoff, int hlen,
1135 memset(flow, 0, sizeof(*flow));
1136 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1137 data, proto, nhoff, hlen, flags);
1140 static inline __u32 skb_get_hash(struct sk_buff *skb)
1142 if (!skb->l4_hash && !skb->sw_hash)
1143 __skb_get_hash(skb);
1148 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1150 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1152 if (!skb->l4_hash && !skb->sw_hash) {
1153 struct flow_keys keys;
1154 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1156 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1162 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1164 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1166 if (!skb->l4_hash && !skb->sw_hash) {
1167 struct flow_keys keys;
1168 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
1170 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1176 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1178 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1183 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1185 to->hash = from->hash;
1186 to->sw_hash = from->sw_hash;
1187 to->l4_hash = from->l4_hash;
1190 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1191 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1193 return skb->head + skb->end;
1196 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1201 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1206 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1208 return skb->end - skb->head;
1213 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1215 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1217 return &skb_shinfo(skb)->hwtstamps;
1221 * skb_queue_empty - check if a queue is empty
1224 * Returns true if the queue is empty, false otherwise.
1226 static inline int skb_queue_empty(const struct sk_buff_head *list)
1228 return list->next == (const struct sk_buff *) list;
1232 * skb_queue_is_last - check if skb is the last entry in the queue
1236 * Returns true if @skb is the last buffer on the list.
1238 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1239 const struct sk_buff *skb)
1241 return skb->next == (const struct sk_buff *) list;
1245 * skb_queue_is_first - check if skb is the first entry in the queue
1249 * Returns true if @skb is the first buffer on the list.
1251 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1252 const struct sk_buff *skb)
1254 return skb->prev == (const struct sk_buff *) list;
1258 * skb_queue_next - return the next packet in the queue
1260 * @skb: current buffer
1262 * Return the next packet in @list after @skb. It is only valid to
1263 * call this if skb_queue_is_last() evaluates to false.
1265 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1266 const struct sk_buff *skb)
1268 /* This BUG_ON may seem severe, but if we just return then we
1269 * are going to dereference garbage.
1271 BUG_ON(skb_queue_is_last(list, skb));
1276 * skb_queue_prev - return the prev packet in the queue
1278 * @skb: current buffer
1280 * Return the prev packet in @list before @skb. It is only valid to
1281 * call this if skb_queue_is_first() evaluates to false.
1283 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1284 const struct sk_buff *skb)
1286 /* This BUG_ON may seem severe, but if we just return then we
1287 * are going to dereference garbage.
1289 BUG_ON(skb_queue_is_first(list, skb));
1294 * skb_get - reference buffer
1295 * @skb: buffer to reference
1297 * Makes another reference to a socket buffer and returns a pointer
1300 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1302 atomic_inc(&skb->users);
1307 * If users == 1, we are the only owner and are can avoid redundant
1312 * skb_cloned - is the buffer a clone
1313 * @skb: buffer to check
1315 * Returns true if the buffer was generated with skb_clone() and is
1316 * one of multiple shared copies of the buffer. Cloned buffers are
1317 * shared data so must not be written to under normal circumstances.
1319 static inline int skb_cloned(const struct sk_buff *skb)
1321 return skb->cloned &&
1322 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1325 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1327 might_sleep_if(gfpflags_allow_blocking(pri));
1329 if (skb_cloned(skb))
1330 return pskb_expand_head(skb, 0, 0, pri);
1336 * skb_header_cloned - is the header a clone
1337 * @skb: buffer to check
1339 * Returns true if modifying the header part of the buffer requires
1340 * the data to be copied.
1342 static inline int skb_header_cloned(const struct sk_buff *skb)
1349 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1350 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1351 return dataref != 1;
1354 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1356 might_sleep_if(gfpflags_allow_blocking(pri));
1358 if (skb_header_cloned(skb))
1359 return pskb_expand_head(skb, 0, 0, pri);
1365 * skb_header_release - release reference to header
1366 * @skb: buffer to operate on
1368 * Drop a reference to the header part of the buffer. This is done
1369 * by acquiring a payload reference. You must not read from the header
1370 * part of skb->data after this.
1371 * Note : Check if you can use __skb_header_release() instead.
1373 static inline void skb_header_release(struct sk_buff *skb)
1377 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1381 * __skb_header_release - release reference to header
1382 * @skb: buffer to operate on
1384 * Variant of skb_header_release() assuming skb is private to caller.
1385 * We can avoid one atomic operation.
1387 static inline void __skb_header_release(struct sk_buff *skb)
1390 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1395 * skb_shared - is the buffer shared
1396 * @skb: buffer to check
1398 * Returns true if more than one person has a reference to this
1401 static inline int skb_shared(const struct sk_buff *skb)
1403 return atomic_read(&skb->users) != 1;
1407 * skb_share_check - check if buffer is shared and if so clone it
1408 * @skb: buffer to check
1409 * @pri: priority for memory allocation
1411 * If the buffer is shared the buffer is cloned and the old copy
1412 * drops a reference. A new clone with a single reference is returned.
1413 * If the buffer is not shared the original buffer is returned. When
1414 * being called from interrupt status or with spinlocks held pri must
1417 * NULL is returned on a memory allocation failure.
1419 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1421 might_sleep_if(gfpflags_allow_blocking(pri));
1422 if (skb_shared(skb)) {
1423 struct sk_buff *nskb = skb_clone(skb, pri);
1435 * Copy shared buffers into a new sk_buff. We effectively do COW on
1436 * packets to handle cases where we have a local reader and forward
1437 * and a couple of other messy ones. The normal one is tcpdumping
1438 * a packet thats being forwarded.
1442 * skb_unshare - make a copy of a shared buffer
1443 * @skb: buffer to check
1444 * @pri: priority for memory allocation
1446 * If the socket buffer is a clone then this function creates a new
1447 * copy of the data, drops a reference count on the old copy and returns
1448 * the new copy with the reference count at 1. If the buffer is not a clone
1449 * the original buffer is returned. When called with a spinlock held or
1450 * from interrupt state @pri must be %GFP_ATOMIC
1452 * %NULL is returned on a memory allocation failure.
1454 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1457 might_sleep_if(gfpflags_allow_blocking(pri));
1458 if (skb_cloned(skb)) {
1459 struct sk_buff *nskb = skb_copy(skb, pri);
1461 /* Free our shared copy */
1472 * skb_peek - peek at the head of an &sk_buff_head
1473 * @list_: list to peek at
1475 * Peek an &sk_buff. Unlike most other operations you _MUST_
1476 * be careful with this one. A peek leaves the buffer on the
1477 * list and someone else may run off with it. You must hold
1478 * the appropriate locks or have a private queue to do this.
1480 * Returns %NULL for an empty list or a pointer to the head element.
1481 * The reference count is not incremented and the reference is therefore
1482 * volatile. Use with caution.
1484 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1486 struct sk_buff *skb = list_->next;
1488 if (skb == (struct sk_buff *)list_)
1494 * skb_peek_next - peek skb following the given one from a queue
1495 * @skb: skb to start from
1496 * @list_: list to peek at
1498 * Returns %NULL when the end of the list is met or a pointer to the
1499 * next element. The reference count is not incremented and the
1500 * reference is therefore volatile. Use with caution.
1502 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1503 const struct sk_buff_head *list_)
1505 struct sk_buff *next = skb->next;
1507 if (next == (struct sk_buff *)list_)
1513 * skb_peek_tail - peek at the tail of an &sk_buff_head
1514 * @list_: list to peek at
1516 * Peek an &sk_buff. Unlike most other operations you _MUST_
1517 * be careful with this one. A peek leaves the buffer on the
1518 * list and someone else may run off with it. You must hold
1519 * the appropriate locks or have a private queue to do this.
1521 * Returns %NULL for an empty list or a pointer to the tail element.
1522 * The reference count is not incremented and the reference is therefore
1523 * volatile. Use with caution.
1525 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1527 struct sk_buff *skb = list_->prev;
1529 if (skb == (struct sk_buff *)list_)
1536 * skb_queue_len - get queue length
1537 * @list_: list to measure
1539 * Return the length of an &sk_buff queue.
1541 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1547 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1548 * @list: queue to initialize
1550 * This initializes only the list and queue length aspects of
1551 * an sk_buff_head object. This allows to initialize the list
1552 * aspects of an sk_buff_head without reinitializing things like
1553 * the spinlock. It can also be used for on-stack sk_buff_head
1554 * objects where the spinlock is known to not be used.
1556 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1558 list->prev = list->next = (struct sk_buff *)list;
1563 * This function creates a split out lock class for each invocation;
1564 * this is needed for now since a whole lot of users of the skb-queue
1565 * infrastructure in drivers have different locking usage (in hardirq)
1566 * than the networking core (in softirq only). In the long run either the
1567 * network layer or drivers should need annotation to consolidate the
1568 * main types of usage into 3 classes.
1570 static inline void skb_queue_head_init(struct sk_buff_head *list)
1572 spin_lock_init(&list->lock);
1573 __skb_queue_head_init(list);
1576 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1577 struct lock_class_key *class)
1579 skb_queue_head_init(list);
1580 lockdep_set_class(&list->lock, class);
1584 * Insert an sk_buff on a list.
1586 * The "__skb_xxxx()" functions are the non-atomic ones that
1587 * can only be called with interrupts disabled.
1589 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1590 struct sk_buff_head *list);
1591 static inline void __skb_insert(struct sk_buff *newsk,
1592 struct sk_buff *prev, struct sk_buff *next,
1593 struct sk_buff_head *list)
1597 next->prev = prev->next = newsk;
1601 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1602 struct sk_buff *prev,
1603 struct sk_buff *next)
1605 struct sk_buff *first = list->next;
1606 struct sk_buff *last = list->prev;
1616 * skb_queue_splice - join two skb lists, this is designed for stacks
1617 * @list: the new list to add
1618 * @head: the place to add it in the first list
1620 static inline void skb_queue_splice(const struct sk_buff_head *list,
1621 struct sk_buff_head *head)
1623 if (!skb_queue_empty(list)) {
1624 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1625 head->qlen += list->qlen;
1630 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1631 * @list: the new list to add
1632 * @head: the place to add it in the first list
1634 * The list at @list is reinitialised
1636 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1637 struct sk_buff_head *head)
1639 if (!skb_queue_empty(list)) {
1640 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1641 head->qlen += list->qlen;
1642 __skb_queue_head_init(list);
1647 * skb_queue_splice_tail - join two skb lists, each list being a queue
1648 * @list: the new list to add
1649 * @head: the place to add it in the first list
1651 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1652 struct sk_buff_head *head)
1654 if (!skb_queue_empty(list)) {
1655 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1656 head->qlen += list->qlen;
1661 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1662 * @list: the new list to add
1663 * @head: the place to add it in the first list
1665 * Each of the lists is a queue.
1666 * The list at @list is reinitialised
1668 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1669 struct sk_buff_head *head)
1671 if (!skb_queue_empty(list)) {
1672 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1673 head->qlen += list->qlen;
1674 __skb_queue_head_init(list);
1679 * __skb_queue_after - queue a buffer at the list head
1680 * @list: list to use
1681 * @prev: place after this buffer
1682 * @newsk: buffer to queue
1684 * Queue a buffer int the middle of a list. This function takes no locks
1685 * and you must therefore hold required locks before calling it.
1687 * A buffer cannot be placed on two lists at the same time.
1689 static inline void __skb_queue_after(struct sk_buff_head *list,
1690 struct sk_buff *prev,
1691 struct sk_buff *newsk)
1693 __skb_insert(newsk, prev, prev->next, list);
1696 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1697 struct sk_buff_head *list);
1699 static inline void __skb_queue_before(struct sk_buff_head *list,
1700 struct sk_buff *next,
1701 struct sk_buff *newsk)
1703 __skb_insert(newsk, next->prev, next, list);
1707 * __skb_queue_head - queue a buffer at the list head
1708 * @list: list to use
1709 * @newsk: buffer to queue
1711 * Queue a buffer at the start of a list. This function takes no locks
1712 * and you must therefore hold required locks before calling it.
1714 * A buffer cannot be placed on two lists at the same time.
1716 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1717 static inline void __skb_queue_head(struct sk_buff_head *list,
1718 struct sk_buff *newsk)
1720 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1724 * __skb_queue_tail - queue a buffer at the list tail
1725 * @list: list to use
1726 * @newsk: buffer to queue
1728 * Queue a buffer at the end of a list. This function takes no locks
1729 * and you must therefore hold required locks before calling it.
1731 * A buffer cannot be placed on two lists at the same time.
1733 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1734 static inline void __skb_queue_tail(struct sk_buff_head *list,
1735 struct sk_buff *newsk)
1737 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1741 * remove sk_buff from list. _Must_ be called atomically, and with
1744 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1745 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1747 struct sk_buff *next, *prev;
1752 skb->next = skb->prev = NULL;
1758 * __skb_dequeue - remove from the head of the queue
1759 * @list: list to dequeue from
1761 * Remove the head of the list. This function does not take any locks
1762 * so must be used with appropriate locks held only. The head item is
1763 * returned or %NULL if the list is empty.
1765 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1766 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1768 struct sk_buff *skb = skb_peek(list);
1770 __skb_unlink(skb, list);
1775 * __skb_dequeue_tail - remove from the tail of the queue
1776 * @list: list to dequeue from
1778 * Remove the tail of the list. This function does not take any locks
1779 * so must be used with appropriate locks held only. The tail item is
1780 * returned or %NULL if the list is empty.
1782 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1783 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1785 struct sk_buff *skb = skb_peek_tail(list);
1787 __skb_unlink(skb, list);
1792 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1794 return skb->data_len;
1797 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1799 return skb->len - skb->data_len;
1802 static inline int skb_pagelen(const struct sk_buff *skb)
1806 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1807 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1808 return len + skb_headlen(skb);
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 * Initialises the @i'th fragment of @skb to point to &size bytes at
1820 * offset @off within @page.
1822 * Does not take any additional reference on the fragment.
1824 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1825 struct page *page, int off, int size)
1827 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1830 * Propagate page pfmemalloc to the skb if we can. The problem is
1831 * that not all callers have unique ownership of the page but rely
1832 * on page_is_pfmemalloc doing the right thing(tm).
1834 frag->page.p = page;
1835 frag->page_offset = off;
1836 skb_frag_size_set(frag, size);
1838 page = compound_head(page);
1839 if (page_is_pfmemalloc(page))
1840 skb->pfmemalloc = true;
1844 * skb_fill_page_desc - initialise a paged fragment in an skb
1845 * @skb: buffer containing fragment to be initialised
1846 * @i: paged fragment index to initialise
1847 * @page: the page to use for this fragment
1848 * @off: the offset to the data with @page
1849 * @size: the length of the data
1851 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1852 * @skb to point to @size bytes at offset @off within @page. In
1853 * addition updates @skb such that @i is the last fragment.
1855 * Does not take any additional reference on the fragment.
1857 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1858 struct page *page, int off, int size)
1860 __skb_fill_page_desc(skb, i, page, off, size);
1861 skb_shinfo(skb)->nr_frags = i + 1;
1864 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1865 int size, unsigned int truesize);
1867 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1868 unsigned int truesize);
1870 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1871 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1872 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1874 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1875 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1877 return skb->head + skb->tail;
1880 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1882 skb->tail = skb->data - skb->head;
1885 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1887 skb_reset_tail_pointer(skb);
1888 skb->tail += offset;
1891 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1892 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1897 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1899 skb->tail = skb->data;
1902 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1904 skb->tail = skb->data + offset;
1907 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1910 * Add data to an sk_buff
1912 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1913 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1914 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1916 unsigned char *tmp = skb_tail_pointer(skb);
1917 SKB_LINEAR_ASSERT(skb);
1923 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1924 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1931 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1932 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1935 BUG_ON(skb->len < skb->data_len);
1936 return skb->data += len;
1939 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1941 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1944 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1946 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1948 if (len > skb_headlen(skb) &&
1949 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1952 return skb->data += len;
1955 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1957 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1960 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1962 if (likely(len <= skb_headlen(skb)))
1964 if (unlikely(len > skb->len))
1966 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1970 * skb_headroom - bytes at buffer head
1971 * @skb: buffer to check
1973 * Return the number of bytes of free space at the head of an &sk_buff.
1975 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1977 return skb->data - skb->head;
1981 * skb_tailroom - bytes at buffer end
1982 * @skb: buffer to check
1984 * Return the number of bytes of free space at the tail of an sk_buff
1986 static inline int skb_tailroom(const struct sk_buff *skb)
1988 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1992 * skb_availroom - bytes at buffer end
1993 * @skb: buffer to check
1995 * Return the number of bytes of free space at the tail of an sk_buff
1996 * allocated by sk_stream_alloc()
1998 static inline int skb_availroom(const struct sk_buff *skb)
2000 if (skb_is_nonlinear(skb))
2003 return skb->end - skb->tail - skb->reserved_tailroom;
2007 * skb_reserve - adjust headroom
2008 * @skb: buffer to alter
2009 * @len: bytes to move
2011 * Increase the headroom of an empty &sk_buff by reducing the tail
2012 * room. This is only allowed for an empty buffer.
2014 static inline void skb_reserve(struct sk_buff *skb, int len)
2021 * skb_tailroom_reserve - adjust reserved_tailroom
2022 * @skb: buffer to alter
2023 * @mtu: maximum amount of headlen permitted
2024 * @needed_tailroom: minimum amount of reserved_tailroom
2026 * Set reserved_tailroom so that headlen can be as large as possible but
2027 * not larger than mtu and tailroom cannot be smaller than
2029 * The required headroom should already have been reserved before using
2032 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2033 unsigned int needed_tailroom)
2035 SKB_LINEAR_ASSERT(skb);
2036 if (mtu < skb_tailroom(skb) - needed_tailroom)
2037 /* use at most mtu */
2038 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2040 /* use up to all available space */
2041 skb->reserved_tailroom = needed_tailroom;
2044 #define ENCAP_TYPE_ETHER 0
2045 #define ENCAP_TYPE_IPPROTO 1
2047 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2050 skb->inner_protocol = protocol;
2051 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2054 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2057 skb->inner_ipproto = ipproto;
2058 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2061 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2063 skb->inner_mac_header = skb->mac_header;
2064 skb->inner_network_header = skb->network_header;
2065 skb->inner_transport_header = skb->transport_header;
2068 static inline void skb_reset_mac_len(struct sk_buff *skb)
2070 skb->mac_len = skb->network_header - skb->mac_header;
2073 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2076 return skb->head + skb->inner_transport_header;
2079 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2081 return skb_inner_transport_header(skb) - skb->data;
2084 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2086 skb->inner_transport_header = skb->data - skb->head;
2089 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2092 skb_reset_inner_transport_header(skb);
2093 skb->inner_transport_header += offset;
2096 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2098 return skb->head + skb->inner_network_header;
2101 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2103 skb->inner_network_header = skb->data - skb->head;
2106 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2109 skb_reset_inner_network_header(skb);
2110 skb->inner_network_header += offset;
2113 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2115 return skb->head + skb->inner_mac_header;
2118 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2120 skb->inner_mac_header = skb->data - skb->head;
2123 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2126 skb_reset_inner_mac_header(skb);
2127 skb->inner_mac_header += offset;
2129 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2131 return skb->transport_header != (typeof(skb->transport_header))~0U;
2134 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2136 return skb->head + skb->transport_header;
2139 static inline void skb_reset_transport_header(struct sk_buff *skb)
2141 skb->transport_header = skb->data - skb->head;
2144 static inline void skb_set_transport_header(struct sk_buff *skb,
2147 skb_reset_transport_header(skb);
2148 skb->transport_header += offset;
2151 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2153 return skb->head + skb->network_header;
2156 static inline void skb_reset_network_header(struct sk_buff *skb)
2158 skb->network_header = skb->data - skb->head;
2161 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2163 skb_reset_network_header(skb);
2164 skb->network_header += offset;
2167 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2169 return skb->head + skb->mac_header;
2172 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2174 return skb->mac_header != (typeof(skb->mac_header))~0U;
2177 static inline void skb_reset_mac_header(struct sk_buff *skb)
2179 skb->mac_header = skb->data - skb->head;
2182 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2184 skb_reset_mac_header(skb);
2185 skb->mac_header += offset;
2188 static inline void skb_pop_mac_header(struct sk_buff *skb)
2190 skb->mac_header = skb->network_header;
2193 static inline void skb_probe_transport_header(struct sk_buff *skb,
2194 const int offset_hint)
2196 struct flow_keys keys;
2198 if (skb_transport_header_was_set(skb))
2200 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2201 skb_set_transport_header(skb, keys.control.thoff);
2203 skb_set_transport_header(skb, offset_hint);
2206 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2208 if (skb_mac_header_was_set(skb)) {
2209 const unsigned char *old_mac = skb_mac_header(skb);
2211 skb_set_mac_header(skb, -skb->mac_len);
2212 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2216 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2218 return skb->csum_start - skb_headroom(skb);
2221 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2223 return skb->head + skb->csum_start;
2226 static inline int skb_transport_offset(const struct sk_buff *skb)
2228 return skb_transport_header(skb) - skb->data;
2231 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2233 return skb->transport_header - skb->network_header;
2236 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2238 return skb->inner_transport_header - skb->inner_network_header;
2241 static inline int skb_network_offset(const struct sk_buff *skb)
2243 return skb_network_header(skb) - skb->data;
2246 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2248 return skb_inner_network_header(skb) - skb->data;
2251 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2253 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2257 * CPUs often take a performance hit when accessing unaligned memory
2258 * locations. The actual performance hit varies, it can be small if the
2259 * hardware handles it or large if we have to take an exception and fix it
2262 * Since an ethernet header is 14 bytes network drivers often end up with
2263 * the IP header at an unaligned offset. The IP header can be aligned by
2264 * shifting the start of the packet by 2 bytes. Drivers should do this
2267 * skb_reserve(skb, NET_IP_ALIGN);
2269 * The downside to this alignment of the IP header is that the DMA is now
2270 * unaligned. On some architectures the cost of an unaligned DMA is high
2271 * and this cost outweighs the gains made by aligning the IP header.
2273 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2276 #ifndef NET_IP_ALIGN
2277 #define NET_IP_ALIGN 2
2281 * The networking layer reserves some headroom in skb data (via
2282 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2283 * the header has to grow. In the default case, if the header has to grow
2284 * 32 bytes or less we avoid the reallocation.
2286 * Unfortunately this headroom changes the DMA alignment of the resulting
2287 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2288 * on some architectures. An architecture can override this value,
2289 * perhaps setting it to a cacheline in size (since that will maintain
2290 * cacheline alignment of the DMA). It must be a power of 2.
2292 * Various parts of the networking layer expect at least 32 bytes of
2293 * headroom, you should not reduce this.
2295 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2296 * to reduce average number of cache lines per packet.
2297 * get_rps_cpus() for example only access one 64 bytes aligned block :
2298 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2301 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2304 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2306 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2308 if (unlikely(skb_is_nonlinear(skb))) {
2313 skb_set_tail_pointer(skb, len);
2316 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2318 __skb_set_length(skb, len);
2321 void skb_trim(struct sk_buff *skb, unsigned int len);
2323 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2326 return ___pskb_trim(skb, len);
2327 __skb_trim(skb, len);
2331 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2333 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2337 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2338 * @skb: buffer to alter
2341 * This is identical to pskb_trim except that the caller knows that
2342 * the skb is not cloned so we should never get an error due to out-
2345 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2347 int err = pskb_trim(skb, len);
2351 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2353 unsigned int diff = len - skb->len;
2355 if (skb_tailroom(skb) < diff) {
2356 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2361 __skb_set_length(skb, len);
2366 * skb_orphan - orphan a buffer
2367 * @skb: buffer to orphan
2369 * If a buffer currently has an owner then we call the owner's
2370 * destructor function and make the @skb unowned. The buffer continues
2371 * to exist but is no longer charged to its former owner.
2373 static inline void skb_orphan(struct sk_buff *skb)
2375 if (skb->destructor) {
2376 skb->destructor(skb);
2377 skb->destructor = NULL;
2385 * skb_orphan_frags - orphan the frags contained in a buffer
2386 * @skb: buffer to orphan frags from
2387 * @gfp_mask: allocation mask for replacement pages
2389 * For each frag in the SKB which needs a destructor (i.e. has an
2390 * owner) create a copy of that frag and release the original
2391 * page by calling the destructor.
2393 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2395 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2397 return skb_copy_ubufs(skb, gfp_mask);
2401 * __skb_queue_purge - empty a list
2402 * @list: list to empty
2404 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2405 * the list and one reference dropped. This function does not take the
2406 * list lock and the caller must hold the relevant locks to use it.
2408 void skb_queue_purge(struct sk_buff_head *list);
2409 static inline void __skb_queue_purge(struct sk_buff_head *list)
2411 struct sk_buff *skb;
2412 while ((skb = __skb_dequeue(list)) != NULL)
2416 void skb_rbtree_purge(struct rb_root *root);
2418 void *netdev_alloc_frag(unsigned int fragsz);
2420 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2424 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2425 * @dev: network device to receive on
2426 * @length: length to allocate
2428 * Allocate a new &sk_buff and assign it a usage count of one. The
2429 * buffer has unspecified headroom built in. Users should allocate
2430 * the headroom they think they need without accounting for the
2431 * built in space. The built in space is used for optimisations.
2433 * %NULL is returned if there is no free memory. Although this function
2434 * allocates memory it can be called from an interrupt.
2436 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2437 unsigned int length)
2439 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2442 /* legacy helper around __netdev_alloc_skb() */
2443 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2446 return __netdev_alloc_skb(NULL, length, gfp_mask);
2449 /* legacy helper around netdev_alloc_skb() */
2450 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2452 return netdev_alloc_skb(NULL, length);
2456 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2457 unsigned int length, gfp_t gfp)
2459 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2461 if (NET_IP_ALIGN && skb)
2462 skb_reserve(skb, NET_IP_ALIGN);
2466 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2467 unsigned int length)
2469 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2472 static inline void skb_free_frag(void *addr)
2474 __free_page_frag(addr);
2477 void *napi_alloc_frag(unsigned int fragsz);
2478 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2479 unsigned int length, gfp_t gfp_mask);
2480 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2481 unsigned int length)
2483 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2485 void napi_consume_skb(struct sk_buff *skb, int budget);
2487 void __kfree_skb_flush(void);
2488 void __kfree_skb_defer(struct sk_buff *skb);
2491 * __dev_alloc_pages - allocate page for network Rx
2492 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2493 * @order: size of the allocation
2495 * Allocate a new page.
2497 * %NULL is returned if there is no free memory.
2499 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2502 /* This piece of code contains several assumptions.
2503 * 1. This is for device Rx, therefor a cold page is preferred.
2504 * 2. The expectation is the user wants a compound page.
2505 * 3. If requesting a order 0 page it will not be compound
2506 * due to the check to see if order has a value in prep_new_page
2507 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2508 * code in gfp_to_alloc_flags that should be enforcing this.
2510 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2512 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2515 static inline struct page *dev_alloc_pages(unsigned int order)
2517 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2521 * __dev_alloc_page - allocate a page for network Rx
2522 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2524 * Allocate a new page.
2526 * %NULL is returned if there is no free memory.
2528 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2530 return __dev_alloc_pages(gfp_mask, 0);
2533 static inline struct page *dev_alloc_page(void)
2535 return dev_alloc_pages(0);
2539 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2540 * @page: The page that was allocated from skb_alloc_page
2541 * @skb: The skb that may need pfmemalloc set
2543 static inline void skb_propagate_pfmemalloc(struct page *page,
2544 struct sk_buff *skb)
2546 if (page_is_pfmemalloc(page))
2547 skb->pfmemalloc = true;
2551 * skb_frag_page - retrieve the page referred to by a paged fragment
2552 * @frag: the paged fragment
2554 * Returns the &struct page associated with @frag.
2556 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2558 return frag->page.p;
2562 * __skb_frag_ref - take an addition reference on a paged fragment.
2563 * @frag: the paged fragment
2565 * Takes an additional reference on the paged fragment @frag.
2567 static inline void __skb_frag_ref(skb_frag_t *frag)
2569 get_page(skb_frag_page(frag));
2573 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2575 * @f: the fragment offset.
2577 * Takes an additional reference on the @f'th paged fragment of @skb.
2579 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2581 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2585 * __skb_frag_unref - release a reference on a paged fragment.
2586 * @frag: the paged fragment
2588 * Releases a reference on the paged fragment @frag.
2590 static inline void __skb_frag_unref(skb_frag_t *frag)
2592 put_page(skb_frag_page(frag));
2596 * skb_frag_unref - release a reference on a paged fragment of an skb.
2598 * @f: the fragment offset
2600 * Releases a reference on the @f'th paged fragment of @skb.
2602 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2604 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2608 * skb_frag_address - gets the address of the data contained in a paged fragment
2609 * @frag: the paged fragment buffer
2611 * Returns the address of the data within @frag. The page must already
2614 static inline void *skb_frag_address(const skb_frag_t *frag)
2616 return page_address(skb_frag_page(frag)) + frag->page_offset;
2620 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2621 * @frag: the paged fragment buffer
2623 * Returns the address of the data within @frag. Checks that the page
2624 * is mapped and returns %NULL otherwise.
2626 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2628 void *ptr = page_address(skb_frag_page(frag));
2632 return ptr + frag->page_offset;
2636 * __skb_frag_set_page - sets the page contained in a paged fragment
2637 * @frag: the paged fragment
2638 * @page: the page to set
2640 * Sets the fragment @frag to contain @page.
2642 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2644 frag->page.p = page;
2648 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2650 * @f: the fragment offset
2651 * @page: the page to set
2653 * Sets the @f'th fragment of @skb to contain @page.
2655 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2658 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2661 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2664 * skb_frag_dma_map - maps a paged fragment via the DMA API
2665 * @dev: the device to map the fragment to
2666 * @frag: the paged fragment to map
2667 * @offset: the offset within the fragment (starting at the
2668 * fragment's own offset)
2669 * @size: the number of bytes to map
2670 * @dir: the direction of the mapping (%PCI_DMA_*)
2672 * Maps the page associated with @frag to @device.
2674 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2675 const skb_frag_t *frag,
2676 size_t offset, size_t size,
2677 enum dma_data_direction dir)
2679 return dma_map_page(dev, skb_frag_page(frag),
2680 frag->page_offset + offset, size, dir);
2683 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2686 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2690 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2693 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2698 * skb_clone_writable - is the header of a clone writable
2699 * @skb: buffer to check
2700 * @len: length up to which to write
2702 * Returns true if modifying the header part of the cloned buffer
2703 * does not requires the data to be copied.
2705 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2707 return !skb_header_cloned(skb) &&
2708 skb_headroom(skb) + len <= skb->hdr_len;
2711 static inline int skb_try_make_writable(struct sk_buff *skb,
2712 unsigned int write_len)
2714 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2715 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2718 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2723 if (headroom > skb_headroom(skb))
2724 delta = headroom - skb_headroom(skb);
2726 if (delta || cloned)
2727 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2733 * skb_cow - copy header of skb when it is required
2734 * @skb: buffer to cow
2735 * @headroom: needed headroom
2737 * If the skb passed lacks sufficient headroom or its data part
2738 * is shared, data is reallocated. If reallocation fails, an error
2739 * is returned and original skb is not changed.
2741 * The result is skb with writable area skb->head...skb->tail
2742 * and at least @headroom of space at head.
2744 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2746 return __skb_cow(skb, headroom, skb_cloned(skb));
2750 * skb_cow_head - skb_cow but only making the head writable
2751 * @skb: buffer to cow
2752 * @headroom: needed headroom
2754 * This function is identical to skb_cow except that we replace the
2755 * skb_cloned check by skb_header_cloned. It should be used when
2756 * you only need to push on some header and do not need to modify
2759 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2761 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2765 * skb_padto - pad an skbuff up to a minimal size
2766 * @skb: buffer to pad
2767 * @len: minimal length
2769 * Pads up a buffer to ensure the trailing bytes exist and are
2770 * blanked. If the buffer already contains sufficient data it
2771 * is untouched. Otherwise it is extended. Returns zero on
2772 * success. The skb is freed on error.
2774 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2776 unsigned int size = skb->len;
2777 if (likely(size >= len))
2779 return skb_pad(skb, len - size);
2783 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2784 * @skb: buffer to pad
2785 * @len: minimal length
2787 * Pads up a buffer to ensure the trailing bytes exist and are
2788 * blanked. If the buffer already contains sufficient data it
2789 * is untouched. Otherwise it is extended. Returns zero on
2790 * success. The skb is freed on error.
2792 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2794 unsigned int size = skb->len;
2796 if (unlikely(size < len)) {
2798 if (skb_pad(skb, len))
2800 __skb_put(skb, len);
2805 static inline int skb_add_data(struct sk_buff *skb,
2806 struct iov_iter *from, int copy)
2808 const int off = skb->len;
2810 if (skb->ip_summed == CHECKSUM_NONE) {
2812 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2813 &csum, from) == copy) {
2814 skb->csum = csum_block_add(skb->csum, csum, off);
2817 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
2820 __skb_trim(skb, off);
2824 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2825 const struct page *page, int off)
2828 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2830 return page == skb_frag_page(frag) &&
2831 off == frag->page_offset + skb_frag_size(frag);
2836 static inline int __skb_linearize(struct sk_buff *skb)
2838 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2842 * skb_linearize - convert paged skb to linear one
2843 * @skb: buffer to linarize
2845 * If there is no free memory -ENOMEM is returned, otherwise zero
2846 * is returned and the old skb data released.
2848 static inline int skb_linearize(struct sk_buff *skb)
2850 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2854 * skb_has_shared_frag - can any frag be overwritten
2855 * @skb: buffer to test
2857 * Return true if the skb has at least one frag that might be modified
2858 * by an external entity (as in vmsplice()/sendfile())
2860 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2862 return skb_is_nonlinear(skb) &&
2863 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2867 * skb_linearize_cow - make sure skb is linear and writable
2868 * @skb: buffer to process
2870 * If there is no free memory -ENOMEM is returned, otherwise zero
2871 * is returned and the old skb data released.
2873 static inline int skb_linearize_cow(struct sk_buff *skb)
2875 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2876 __skb_linearize(skb) : 0;
2879 static __always_inline void
2880 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2883 if (skb->ip_summed == CHECKSUM_COMPLETE)
2884 skb->csum = csum_block_sub(skb->csum,
2885 csum_partial(start, len, 0), off);
2886 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2887 skb_checksum_start_offset(skb) < 0)
2888 skb->ip_summed = CHECKSUM_NONE;
2892 * skb_postpull_rcsum - update checksum for received skb after pull
2893 * @skb: buffer to update
2894 * @start: start of data before pull
2895 * @len: length of data pulled
2897 * After doing a pull on a received packet, you need to call this to
2898 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2899 * CHECKSUM_NONE so that it can be recomputed from scratch.
2901 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2902 const void *start, unsigned int len)
2904 __skb_postpull_rcsum(skb, start, len, 0);
2907 static __always_inline void
2908 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2911 if (skb->ip_summed == CHECKSUM_COMPLETE)
2912 skb->csum = csum_block_add(skb->csum,
2913 csum_partial(start, len, 0), off);
2917 * skb_postpush_rcsum - update checksum for received skb after push
2918 * @skb: buffer to update
2919 * @start: start of data after push
2920 * @len: length of data pushed
2922 * After doing a push on a received packet, you need to call this to
2923 * update the CHECKSUM_COMPLETE checksum.
2925 static inline void skb_postpush_rcsum(struct sk_buff *skb,
2926 const void *start, unsigned int len)
2928 __skb_postpush_rcsum(skb, start, len, 0);
2931 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2934 * skb_push_rcsum - push skb and update receive checksum
2935 * @skb: buffer to update
2936 * @len: length of data pulled
2938 * This function performs an skb_push on the packet and updates
2939 * the CHECKSUM_COMPLETE checksum. It should be used on
2940 * receive path processing instead of skb_push unless you know
2941 * that the checksum difference is zero (e.g., a valid IP header)
2942 * or you are setting ip_summed to CHECKSUM_NONE.
2944 static inline unsigned char *skb_push_rcsum(struct sk_buff *skb,
2948 skb_postpush_rcsum(skb, skb->data, len);
2953 * pskb_trim_rcsum - trim received skb and update checksum
2954 * @skb: buffer to trim
2957 * This is exactly the same as pskb_trim except that it ensures the
2958 * checksum of received packets are still valid after the operation.
2961 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2963 if (likely(len >= skb->len))
2965 if (skb->ip_summed == CHECKSUM_COMPLETE)
2966 skb->ip_summed = CHECKSUM_NONE;
2967 return __pskb_trim(skb, len);
2970 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2972 if (skb->ip_summed == CHECKSUM_COMPLETE)
2973 skb->ip_summed = CHECKSUM_NONE;
2974 __skb_trim(skb, len);
2978 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
2980 if (skb->ip_summed == CHECKSUM_COMPLETE)
2981 skb->ip_summed = CHECKSUM_NONE;
2982 return __skb_grow(skb, len);
2985 #define skb_queue_walk(queue, skb) \
2986 for (skb = (queue)->next; \
2987 skb != (struct sk_buff *)(queue); \
2990 #define skb_queue_walk_safe(queue, skb, tmp) \
2991 for (skb = (queue)->next, tmp = skb->next; \
2992 skb != (struct sk_buff *)(queue); \
2993 skb = tmp, tmp = skb->next)
2995 #define skb_queue_walk_from(queue, skb) \
2996 for (; skb != (struct sk_buff *)(queue); \
2999 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3000 for (tmp = skb->next; \
3001 skb != (struct sk_buff *)(queue); \
3002 skb = tmp, tmp = skb->next)
3004 #define skb_queue_reverse_walk(queue, skb) \
3005 for (skb = (queue)->prev; \
3006 skb != (struct sk_buff *)(queue); \
3009 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3010 for (skb = (queue)->prev, tmp = skb->prev; \
3011 skb != (struct sk_buff *)(queue); \
3012 skb = tmp, tmp = skb->prev)
3014 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3015 for (tmp = skb->prev; \
3016 skb != (struct sk_buff *)(queue); \
3017 skb = tmp, tmp = skb->prev)
3019 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3021 return skb_shinfo(skb)->frag_list != NULL;
3024 static inline void skb_frag_list_init(struct sk_buff *skb)
3026 skb_shinfo(skb)->frag_list = NULL;
3029 #define skb_walk_frags(skb, iter) \
3030 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3033 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3034 const struct sk_buff *skb);
3035 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3036 int *peeked, int *off, int *err,
3037 struct sk_buff **last);
3038 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3039 int *peeked, int *off, int *err);
3040 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3042 unsigned int datagram_poll(struct file *file, struct socket *sock,
3043 struct poll_table_struct *wait);
3044 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3045 struct iov_iter *to, int size);
3046 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3047 struct msghdr *msg, int size)
3049 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3051 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3052 struct msghdr *msg);
3053 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3054 struct iov_iter *from, int len);
3055 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3056 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3057 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3058 static inline void skb_free_datagram_locked(struct sock *sk,
3059 struct sk_buff *skb)
3061 __skb_free_datagram_locked(sk, skb, 0);
3063 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3064 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3065 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3066 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3067 int len, __wsum csum);
3068 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3069 struct pipe_inode_info *pipe, unsigned int len,
3070 unsigned int flags);
3071 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3072 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3073 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3075 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3076 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3077 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3078 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3079 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3080 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3081 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3082 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3083 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3084 int skb_vlan_pop(struct sk_buff *skb);
3085 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3086 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3089 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3091 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3094 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3096 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3099 struct skb_checksum_ops {
3100 __wsum (*update)(const void *mem, int len, __wsum wsum);
3101 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3104 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3105 __wsum csum, const struct skb_checksum_ops *ops);
3106 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3109 static inline void * __must_check
3110 __skb_header_pointer(const struct sk_buff *skb, int offset,
3111 int len, void *data, int hlen, void *buffer)
3113 if (hlen - offset >= len)
3114 return data + offset;
3117 skb_copy_bits(skb, offset, buffer, len) < 0)
3123 static inline void * __must_check
3124 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3126 return __skb_header_pointer(skb, offset, len, skb->data,
3127 skb_headlen(skb), buffer);
3131 * skb_needs_linearize - check if we need to linearize a given skb
3132 * depending on the given device features.
3133 * @skb: socket buffer to check
3134 * @features: net device features
3136 * Returns true if either:
3137 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3138 * 2. skb is fragmented and the device does not support SG.
3140 static inline bool skb_needs_linearize(struct sk_buff *skb,
3141 netdev_features_t features)
3143 return skb_is_nonlinear(skb) &&
3144 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3145 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3148 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3150 const unsigned int len)
3152 memcpy(to, skb->data, len);
3155 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3156 const int offset, void *to,
3157 const unsigned int len)
3159 memcpy(to, skb->data + offset, len);
3162 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3164 const unsigned int len)
3166 memcpy(skb->data, from, len);
3169 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3172 const unsigned int len)
3174 memcpy(skb->data + offset, from, len);
3177 void skb_init(void);
3179 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3185 * skb_get_timestamp - get timestamp from a skb
3186 * @skb: skb to get stamp from
3187 * @stamp: pointer to struct timeval to store stamp in
3189 * Timestamps are stored in the skb as offsets to a base timestamp.
3190 * This function converts the offset back to a struct timeval and stores
3193 static inline void skb_get_timestamp(const struct sk_buff *skb,
3194 struct timeval *stamp)
3196 *stamp = ktime_to_timeval(skb->tstamp);
3199 static inline void skb_get_timestampns(const struct sk_buff *skb,
3200 struct timespec *stamp)
3202 *stamp = ktime_to_timespec(skb->tstamp);
3205 static inline void __net_timestamp(struct sk_buff *skb)
3207 skb->tstamp = ktime_get_real();
3210 static inline ktime_t net_timedelta(ktime_t t)
3212 return ktime_sub(ktime_get_real(), t);
3215 static inline ktime_t net_invalid_timestamp(void)
3217 return ktime_set(0, 0);
3220 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3222 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3224 void skb_clone_tx_timestamp(struct sk_buff *skb);
3225 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3227 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3229 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3233 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3238 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3241 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3243 * PHY drivers may accept clones of transmitted packets for
3244 * timestamping via their phy_driver.txtstamp method. These drivers
3245 * must call this function to return the skb back to the stack with a
3248 * @skb: clone of the the original outgoing packet
3249 * @hwtstamps: hardware time stamps
3252 void skb_complete_tx_timestamp(struct sk_buff *skb,
3253 struct skb_shared_hwtstamps *hwtstamps);
3255 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3256 struct skb_shared_hwtstamps *hwtstamps,
3257 struct sock *sk, int tstype);
3260 * skb_tstamp_tx - queue clone of skb with send time stamps
3261 * @orig_skb: the original outgoing packet
3262 * @hwtstamps: hardware time stamps, may be NULL if not available
3264 * If the skb has a socket associated, then this function clones the
3265 * skb (thus sharing the actual data and optional structures), stores
3266 * the optional hardware time stamping information (if non NULL) or
3267 * generates a software time stamp (otherwise), then queues the clone
3268 * to the error queue of the socket. Errors are silently ignored.
3270 void skb_tstamp_tx(struct sk_buff *orig_skb,
3271 struct skb_shared_hwtstamps *hwtstamps);
3273 static inline void sw_tx_timestamp(struct sk_buff *skb)
3275 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3276 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
3277 skb_tstamp_tx(skb, NULL);
3281 * skb_tx_timestamp() - Driver hook for transmit timestamping
3283 * Ethernet MAC Drivers should call this function in their hard_xmit()
3284 * function immediately before giving the sk_buff to the MAC hardware.
3286 * Specifically, one should make absolutely sure that this function is
3287 * called before TX completion of this packet can trigger. Otherwise
3288 * the packet could potentially already be freed.
3290 * @skb: A socket buffer.
3292 static inline void skb_tx_timestamp(struct sk_buff *skb)
3294 skb_clone_tx_timestamp(skb);
3295 sw_tx_timestamp(skb);
3299 * skb_complete_wifi_ack - deliver skb with wifi status
3301 * @skb: the original outgoing packet
3302 * @acked: ack status
3305 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3307 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3308 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3310 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3312 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3314 (skb->ip_summed == CHECKSUM_PARTIAL &&
3315 skb_checksum_start_offset(skb) >= 0));
3319 * skb_checksum_complete - Calculate checksum of an entire packet
3320 * @skb: packet to process
3322 * This function calculates the checksum over the entire packet plus
3323 * the value of skb->csum. The latter can be used to supply the
3324 * checksum of a pseudo header as used by TCP/UDP. It returns the
3327 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3328 * this function can be used to verify that checksum on received
3329 * packets. In that case the function should return zero if the
3330 * checksum is correct. In particular, this function will return zero
3331 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3332 * hardware has already verified the correctness of the checksum.
3334 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3336 return skb_csum_unnecessary(skb) ?
3337 0 : __skb_checksum_complete(skb);
3340 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3342 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3343 if (skb->csum_level == 0)
3344 skb->ip_summed = CHECKSUM_NONE;
3350 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3352 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3353 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3355 } else if (skb->ip_summed == CHECKSUM_NONE) {
3356 skb->ip_summed = CHECKSUM_UNNECESSARY;
3357 skb->csum_level = 0;
3361 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3363 /* Mark current checksum as bad (typically called from GRO
3364 * path). In the case that ip_summed is CHECKSUM_NONE
3365 * this must be the first checksum encountered in the packet.
3366 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3367 * checksum after the last one validated. For UDP, a zero
3368 * checksum can not be marked as bad.
3371 if (skb->ip_summed == CHECKSUM_NONE ||
3372 skb->ip_summed == CHECKSUM_UNNECESSARY)
3376 /* Check if we need to perform checksum complete validation.
3378 * Returns true if checksum complete is needed, false otherwise
3379 * (either checksum is unnecessary or zero checksum is allowed).
3381 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3385 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3386 skb->csum_valid = 1;
3387 __skb_decr_checksum_unnecessary(skb);
3394 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3397 #define CHECKSUM_BREAK 76
3399 /* Unset checksum-complete
3401 * Unset checksum complete can be done when packet is being modified
3402 * (uncompressed for instance) and checksum-complete value is
3405 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3407 if (skb->ip_summed == CHECKSUM_COMPLETE)
3408 skb->ip_summed = CHECKSUM_NONE;
3411 /* Validate (init) checksum based on checksum complete.
3414 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3415 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3416 * checksum is stored in skb->csum for use in __skb_checksum_complete
3417 * non-zero: value of invalid checksum
3420 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3424 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3425 if (!csum_fold(csum_add(psum, skb->csum))) {
3426 skb->csum_valid = 1;
3429 } else if (skb->csum_bad) {
3430 /* ip_summed == CHECKSUM_NONE in this case */
3431 return (__force __sum16)1;
3436 if (complete || skb->len <= CHECKSUM_BREAK) {
3439 csum = __skb_checksum_complete(skb);
3440 skb->csum_valid = !csum;
3447 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3452 /* Perform checksum validate (init). Note that this is a macro since we only
3453 * want to calculate the pseudo header which is an input function if necessary.
3454 * First we try to validate without any computation (checksum unnecessary) and
3455 * then calculate based on checksum complete calling the function to compute
3459 * 0: checksum is validated or try to in skb_checksum_complete
3460 * non-zero: value of invalid checksum
3462 #define __skb_checksum_validate(skb, proto, complete, \
3463 zero_okay, check, compute_pseudo) \
3465 __sum16 __ret = 0; \
3466 skb->csum_valid = 0; \
3467 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3468 __ret = __skb_checksum_validate_complete(skb, \
3469 complete, compute_pseudo(skb, proto)); \
3473 #define skb_checksum_init(skb, proto, compute_pseudo) \
3474 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3476 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3477 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3479 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3480 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3482 #define skb_checksum_validate_zero_check(skb, proto, check, \
3484 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3486 #define skb_checksum_simple_validate(skb) \
3487 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3489 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3491 return (skb->ip_summed == CHECKSUM_NONE &&
3492 skb->csum_valid && !skb->csum_bad);
3495 static inline void __skb_checksum_convert(struct sk_buff *skb,
3496 __sum16 check, __wsum pseudo)
3498 skb->csum = ~pseudo;
3499 skb->ip_summed = CHECKSUM_COMPLETE;
3502 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3504 if (__skb_checksum_convert_check(skb)) \
3505 __skb_checksum_convert(skb, check, \
3506 compute_pseudo(skb, proto)); \
3509 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3510 u16 start, u16 offset)
3512 skb->ip_summed = CHECKSUM_PARTIAL;
3513 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3514 skb->csum_offset = offset - start;
3517 /* Update skbuf and packet to reflect the remote checksum offload operation.
3518 * When called, ptr indicates the starting point for skb->csum when
3519 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3520 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3522 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3523 int start, int offset, bool nopartial)
3528 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3532 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3533 __skb_checksum_complete(skb);
3534 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3537 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3539 /* Adjust skb->csum since we changed the packet */
3540 skb->csum = csum_add(skb->csum, delta);
3543 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3544 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3545 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3547 if (nfct && atomic_dec_and_test(&nfct->use))
3548 nf_conntrack_destroy(nfct);
3550 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3553 atomic_inc(&nfct->use);
3556 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3557 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3559 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3562 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3565 atomic_inc(&nf_bridge->use);
3567 #endif /* CONFIG_BRIDGE_NETFILTER */
3568 static inline void nf_reset(struct sk_buff *skb)
3570 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3571 nf_conntrack_put(skb->nfct);
3574 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3575 nf_bridge_put(skb->nf_bridge);
3576 skb->nf_bridge = NULL;
3580 static inline void nf_reset_trace(struct sk_buff *skb)
3582 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3587 /* Note: This doesn't put any conntrack and bridge info in dst. */
3588 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3591 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3592 dst->nfct = src->nfct;
3593 nf_conntrack_get(src->nfct);
3595 dst->nfctinfo = src->nfctinfo;
3597 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3598 dst->nf_bridge = src->nf_bridge;
3599 nf_bridge_get(src->nf_bridge);
3601 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3603 dst->nf_trace = src->nf_trace;
3607 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3609 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3610 nf_conntrack_put(dst->nfct);
3612 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3613 nf_bridge_put(dst->nf_bridge);
3615 __nf_copy(dst, src, true);
3618 #ifdef CONFIG_NETWORK_SECMARK
3619 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3621 to->secmark = from->secmark;
3624 static inline void skb_init_secmark(struct sk_buff *skb)
3629 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3632 static inline void skb_init_secmark(struct sk_buff *skb)
3636 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3638 return !skb->destructor &&
3639 #if IS_ENABLED(CONFIG_XFRM)
3642 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3645 !skb->_skb_refdst &&
3646 !skb_has_frag_list(skb);
3649 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3651 skb->queue_mapping = queue_mapping;
3654 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3656 return skb->queue_mapping;
3659 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3661 to->queue_mapping = from->queue_mapping;
3664 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3666 skb->queue_mapping = rx_queue + 1;
3669 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3671 return skb->queue_mapping - 1;
3674 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3676 return skb->queue_mapping != 0;
3679 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3688 /* Keeps track of mac header offset relative to skb->head.
3689 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3690 * For non-tunnel skb it points to skb_mac_header() and for
3691 * tunnel skb it points to outer mac header.
3692 * Keeps track of level of encapsulation of network headers.
3703 #define SKB_SGO_CB_OFFSET 32
3704 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3706 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3708 return (skb_mac_header(inner_skb) - inner_skb->head) -
3709 SKB_GSO_CB(inner_skb)->mac_offset;
3712 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3714 int new_headroom, headroom;
3717 headroom = skb_headroom(skb);
3718 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3722 new_headroom = skb_headroom(skb);
3723 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3727 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3729 /* Do not update partial checksums if remote checksum is enabled. */
3730 if (skb->remcsum_offload)
3733 SKB_GSO_CB(skb)->csum = res;
3734 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3737 /* Compute the checksum for a gso segment. First compute the checksum value
3738 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3739 * then add in skb->csum (checksum from csum_start to end of packet).
3740 * skb->csum and csum_start are then updated to reflect the checksum of the
3741 * resultant packet starting from the transport header-- the resultant checksum
3742 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3745 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3747 unsigned char *csum_start = skb_transport_header(skb);
3748 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3749 __wsum partial = SKB_GSO_CB(skb)->csum;
3751 SKB_GSO_CB(skb)->csum = res;
3752 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
3754 return csum_fold(csum_partial(csum_start, plen, partial));
3757 static inline bool skb_is_gso(const struct sk_buff *skb)
3759 return skb_shinfo(skb)->gso_size;
3762 /* Note: Should be called only if skb_is_gso(skb) is true */
3763 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3765 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3768 static inline void skb_gso_reset(struct sk_buff *skb)
3770 skb_shinfo(skb)->gso_size = 0;
3771 skb_shinfo(skb)->gso_segs = 0;
3772 skb_shinfo(skb)->gso_type = 0;
3775 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3777 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3779 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3780 * wanted then gso_type will be set. */
3781 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3783 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3784 unlikely(shinfo->gso_type == 0)) {
3785 __skb_warn_lro_forwarding(skb);
3791 static inline void skb_forward_csum(struct sk_buff *skb)
3793 /* Unfortunately we don't support this one. Any brave souls? */
3794 if (skb->ip_summed == CHECKSUM_COMPLETE)
3795 skb->ip_summed = CHECKSUM_NONE;
3799 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3800 * @skb: skb to check
3802 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3803 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3804 * use this helper, to document places where we make this assertion.
3806 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3809 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3813 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3815 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3816 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3817 unsigned int transport_len,
3818 __sum16(*skb_chkf)(struct sk_buff *skb));
3821 * skb_head_is_locked - Determine if the skb->head is locked down
3822 * @skb: skb to check
3824 * The head on skbs build around a head frag can be removed if they are
3825 * not cloned. This function returns true if the skb head is locked down
3826 * due to either being allocated via kmalloc, or by being a clone with
3827 * multiple references to the head.
3829 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3831 return !skb->head_frag || skb_cloned(skb);
3835 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3839 * skb_gso_network_seglen is used to determine the real size of the
3840 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3842 * The MAC/L2 header is not accounted for.
3844 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3846 unsigned int hdr_len = skb_transport_header(skb) -
3847 skb_network_header(skb);
3848 return hdr_len + skb_gso_transport_seglen(skb);
3851 /* Local Checksum Offload.
3852 * Compute outer checksum based on the assumption that the
3853 * inner checksum will be offloaded later.
3854 * See Documentation/networking/checksum-offloads.txt for
3855 * explanation of how this works.
3856 * Fill in outer checksum adjustment (e.g. with sum of outer
3857 * pseudo-header) before calling.
3858 * Also ensure that inner checksum is in linear data area.
3860 static inline __wsum lco_csum(struct sk_buff *skb)
3862 unsigned char *csum_start = skb_checksum_start(skb);
3863 unsigned char *l4_hdr = skb_transport_header(skb);
3866 /* Start with complement of inner checksum adjustment */
3867 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3870 /* Add in checksum of our headers (incl. outer checksum
3871 * adjustment filled in by caller) and return result.
3873 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
3876 #endif /* __KERNEL__ */
3877 #endif /* _LINUX_SKBUFF_H */