2 * Definitions for the 'struct sk_buff' memory handlers.
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
17 #include <linux/kernel.h>
18 #include <linux/compiler.h>
19 #include <linux/time.h>
20 #include <linux/bug.h>
21 #include <linux/cache.h>
22 #include <linux/rbtree.h>
23 #include <linux/socket.h>
24 #include <linux/refcount.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 <linux/sched/clock.h>
38 #include <net/flow_dissector.h>
39 #include <linux/splice.h>
40 #include <linux/in6.h>
41 #include <linux/if_packet.h>
44 /* The interface for checksum offload between the stack and networking drivers
47 * A. IP checksum related features
49 * Drivers advertise checksum offload capabilities in the features of a device.
50 * From the stack's point of view these are capabilities offered by the driver,
51 * a driver typically only advertises features that it is capable of offloading
54 * The checksum related features are:
56 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
57 * IP (one's complement) checksum for any combination
58 * of protocols or protocol layering. The checksum is
59 * computed and set in a packet per the CHECKSUM_PARTIAL
60 * interface (see below).
62 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
63 * TCP or UDP packets over IPv4. These are specifically
64 * unencapsulated packets of the form IPv4|TCP or
65 * IPv4|UDP where the Protocol field in the IPv4 header
66 * is TCP or UDP. The IPv4 header may contain IP options
67 * This feature cannot be set in features for a device
68 * with NETIF_F_HW_CSUM also set. This feature is being
69 * DEPRECATED (see below).
71 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
72 * TCP or UDP packets over IPv6. These are specifically
73 * unencapsulated packets of the form IPv6|TCP or
74 * IPv4|UDP where the Next Header field in the IPv6
75 * header is either TCP or UDP. IPv6 extension headers
76 * are not supported with this feature. This feature
77 * cannot be set in features for a device with
78 * NETIF_F_HW_CSUM also set. This feature is being
79 * DEPRECATED (see below).
81 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
82 * This flag is used only used to disable the RX checksum
83 * feature for a device. The stack will accept receive
84 * checksum indication in packets received on a device
85 * regardless of whether NETIF_F_RXCSUM is set.
87 * B. Checksumming of received packets by device. Indication of checksum
88 * verification is in set skb->ip_summed. Possible values are:
92 * Device did not checksum this packet e.g. due to lack of capabilities.
93 * The packet contains full (though not verified) checksum in packet but
94 * not in skb->csum. Thus, skb->csum is undefined in this case.
96 * CHECKSUM_UNNECESSARY:
98 * The hardware you're dealing with doesn't calculate the full checksum
99 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
100 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
101 * if their checksums are okay. skb->csum is still undefined in this case
102 * though. A driver or device must never modify the checksum field in the
103 * packet even if checksum is verified.
105 * CHECKSUM_UNNECESSARY is applicable to following protocols:
106 * TCP: IPv6 and IPv4.
107 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
108 * zero UDP checksum for either IPv4 or IPv6, the networking stack
109 * may perform further validation in this case.
110 * GRE: only if the checksum is present in the header.
111 * SCTP: indicates the CRC in SCTP header has been validated.
112 * FCOE: indicates the CRC in FC frame has been validated.
114 * skb->csum_level indicates the number of consecutive checksums found in
115 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
116 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
117 * and a device is able to verify the checksums for UDP (possibly zero),
118 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
119 * two. If the device were only able to verify the UDP checksum and not
120 * GRE, either because it doesn't support GRE checksum of because GRE
121 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
122 * not considered in this case).
126 * This is the most generic way. The device supplied checksum of the _whole_
127 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
128 * hardware doesn't need to parse L3/L4 headers to implement this.
131 * - Even if device supports only some protocols, but is able to produce
132 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
133 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
137 * A checksum is set up to be offloaded to a device as described in the
138 * output description for CHECKSUM_PARTIAL. This may occur on a packet
139 * received directly from another Linux OS, e.g., a virtualized Linux kernel
140 * on the same host, or it may be set in the input path in GRO or remote
141 * checksum offload. For the purposes of checksum verification, the checksum
142 * referred to by skb->csum_start + skb->csum_offset and any preceding
143 * checksums in the packet are considered verified. Any checksums in the
144 * packet that are after the checksum being offloaded are not considered to
147 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
148 * in the skb->ip_summed for a packet. Values are:
152 * The driver is required to checksum the packet as seen by hard_start_xmit()
153 * from skb->csum_start up to the end, and to record/write the checksum at
154 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
155 * csum_start and csum_offset values are valid values given the length and
156 * offset of the packet, however they should not attempt to validate that the
157 * checksum refers to a legitimate transport layer checksum-- it is the
158 * purview of the stack to validate that csum_start and csum_offset are set
161 * When the stack requests checksum offload for a packet, the driver MUST
162 * ensure that the checksum is set correctly. A driver can either offload the
163 * checksum calculation to the device, or call skb_checksum_help (in the case
164 * that the device does not support offload for a particular checksum).
166 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
167 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
168 * checksum offload capability.
169 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
170 * on network device checksumming capabilities: if a packet does not match
171 * them, skb_checksum_help or skb_crc32c_help (depending on the value of
172 * csum_not_inet, see item D.) is called to resolve the checksum.
176 * The skb was already checksummed by the protocol, or a checksum is not
179 * CHECKSUM_UNNECESSARY:
181 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
185 * Not used in checksum output. If a driver observes a packet with this value
186 * set in skbuff, if should treat as CHECKSUM_NONE being set.
188 * D. Non-IP checksum (CRC) offloads
190 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
191 * offloading the SCTP CRC in a packet. To perform this offload the stack
192 * will set set csum_start and csum_offset accordingly, set ip_summed to
193 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
194 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
195 * A driver that supports both IP checksum offload and SCTP CRC32c offload
196 * must verify which offload is configured for a packet by testing the
197 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
198 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
200 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
201 * offloading the FCOE CRC in a packet. To perform this offload the stack
202 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
203 * accordingly. Note the there is no indication in the skbuff that the
204 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
205 * both IP checksum offload and FCOE CRC offload must verify which offload
206 * is configured for a packet presumably by inspecting packet headers.
208 * E. Checksumming on output with GSO.
210 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
211 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
212 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
213 * part of the GSO operation is implied. If a checksum is being offloaded
214 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
215 * are set to refer to the outermost checksum being offload (two offloaded
216 * checksums are possible with UDP encapsulation).
219 /* Don't change this without changing skb_csum_unnecessary! */
220 #define CHECKSUM_NONE 0
221 #define CHECKSUM_UNNECESSARY 1
222 #define CHECKSUM_COMPLETE 2
223 #define CHECKSUM_PARTIAL 3
225 /* Maximum value in skb->csum_level */
226 #define SKB_MAX_CSUM_LEVEL 3
228 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
229 #define SKB_WITH_OVERHEAD(X) \
230 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
231 #define SKB_MAX_ORDER(X, ORDER) \
232 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
233 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
234 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
236 /* return minimum truesize of one skb containing X bytes of data */
237 #define SKB_TRUESIZE(X) ((X) + \
238 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
239 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
243 struct pipe_inode_info;
247 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
248 struct nf_conntrack {
253 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
254 struct nf_bridge_info {
257 BRNF_PROTO_UNCHANGED,
265 struct net_device *physindev;
267 /* always valid & non-NULL from FORWARD on, for physdev match */
268 struct net_device *physoutdev;
270 /* prerouting: detect dnat in orig/reply direction */
272 struct in6_addr ipv6_daddr;
274 /* after prerouting + nat detected: store original source
275 * mac since neigh resolution overwrites it, only used while
276 * skb is out in neigh layer.
278 char neigh_header[8];
283 struct sk_buff_head {
284 /* These two members must be first. */
285 struct sk_buff *next;
286 struct sk_buff *prev;
294 /* To allow 64K frame to be packed as single skb without frag_list we
295 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
296 * buffers which do not start on a page boundary.
298 * Since GRO uses frags we allocate at least 16 regardless of page
301 #if (65536/PAGE_SIZE + 1) < 16
302 #define MAX_SKB_FRAGS 16UL
304 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
306 extern int sysctl_max_skb_frags;
308 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
309 * segment using its current segmentation instead.
311 #define GSO_BY_FRAGS 0xFFFF
313 typedef struct skb_frag_struct skb_frag_t;
315 struct skb_frag_struct {
319 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
328 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
333 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
338 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
343 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
348 static inline bool skb_frag_must_loop(struct page *p)
350 #if defined(CONFIG_HIGHMEM)
358 * skb_frag_foreach_page - loop over pages in a fragment
360 * @f: skb frag to operate on
361 * @f_off: offset from start of f->page.p
362 * @f_len: length from f_off to loop over
363 * @p: (temp var) current page
364 * @p_off: (temp var) offset from start of current page,
365 * non-zero only on first page.
366 * @p_len: (temp var) length in current page,
367 * < PAGE_SIZE only on first and last page.
368 * @copied: (temp var) length so far, excluding current p_len.
370 * A fragment can hold a compound page, in which case per-page
371 * operations, notably kmap_atomic, must be called for each
374 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
375 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
376 p_off = (f_off) & (PAGE_SIZE - 1), \
377 p_len = skb_frag_must_loop(p) ? \
378 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
381 copied += p_len, p++, p_off = 0, \
382 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
384 #define HAVE_HW_TIME_STAMP
387 * struct skb_shared_hwtstamps - hardware time stamps
388 * @hwtstamp: hardware time stamp transformed into duration
389 * since arbitrary point in time
391 * Software time stamps generated by ktime_get_real() are stored in
394 * hwtstamps can only be compared against other hwtstamps from
397 * This structure is attached to packets as part of the
398 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
400 struct skb_shared_hwtstamps {
404 /* Definitions for tx_flags in struct skb_shared_info */
406 /* generate hardware time stamp */
407 SKBTX_HW_TSTAMP = 1 << 0,
409 /* generate software time stamp when queueing packet to NIC */
410 SKBTX_SW_TSTAMP = 1 << 1,
412 /* device driver is going to provide hardware time stamp */
413 SKBTX_IN_PROGRESS = 1 << 2,
415 /* device driver supports TX zero-copy buffers */
416 SKBTX_DEV_ZEROCOPY = 1 << 3,
418 /* generate wifi status information (where possible) */
419 SKBTX_WIFI_STATUS = 1 << 4,
421 /* This indicates at least one fragment might be overwritten
422 * (as in vmsplice(), sendfile() ...)
423 * If we need to compute a TX checksum, we'll need to copy
424 * all frags to avoid possible bad checksum
426 SKBTX_SHARED_FRAG = 1 << 5,
428 /* generate software time stamp when entering packet scheduling */
429 SKBTX_SCHED_TSTAMP = 1 << 6,
432 #define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
433 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
435 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
438 * The callback notifies userspace to release buffers when skb DMA is done in
439 * lower device, the skb last reference should be 0 when calling this.
440 * The zerocopy_success argument is true if zero copy transmit occurred,
441 * false on data copy or out of memory error caused by data copy attempt.
442 * The ctx field is used to track device context.
443 * The desc field is used to track userspace buffer index.
446 void (*callback)(struct ubuf_info *, bool zerocopy_success);
462 struct user_struct *user;
467 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
469 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
470 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
471 struct ubuf_info *uarg);
473 static inline void sock_zerocopy_get(struct ubuf_info *uarg)
475 refcount_inc(&uarg->refcnt);
478 void sock_zerocopy_put(struct ubuf_info *uarg);
479 void sock_zerocopy_put_abort(struct ubuf_info *uarg);
481 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
483 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
484 struct msghdr *msg, int len,
485 struct ubuf_info *uarg);
487 /* This data is invariant across clones and lives at
488 * the end of the header data, ie. at skb->end.
490 struct skb_shared_info {
495 unsigned short gso_size;
496 /* Warning: this field is not always filled in (UFO)! */
497 unsigned short gso_segs;
498 struct sk_buff *frag_list;
499 struct skb_shared_hwtstamps hwtstamps;
500 unsigned int gso_type;
504 * Warning : all fields before dataref are cleared in __alloc_skb()
508 /* Intermediate layers must ensure that destructor_arg
509 * remains valid until skb destructor */
510 void * destructor_arg;
512 /* must be last field, see pskb_expand_head() */
513 skb_frag_t frags[MAX_SKB_FRAGS];
516 /* We divide dataref into two halves. The higher 16 bits hold references
517 * to the payload part of skb->data. The lower 16 bits hold references to
518 * the entire skb->data. A clone of a headerless skb holds the length of
519 * the header in skb->hdr_len.
521 * All users must obey the rule that the skb->data reference count must be
522 * greater than or equal to the payload reference count.
524 * Holding a reference to the payload part means that the user does not
525 * care about modifications to the header part of skb->data.
527 #define SKB_DATAREF_SHIFT 16
528 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
532 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
533 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
534 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
538 SKB_GSO_TCPV4 = 1 << 0,
540 /* This indicates the skb is from an untrusted source. */
541 SKB_GSO_DODGY = 1 << 1,
543 /* This indicates the tcp segment has CWR set. */
544 SKB_GSO_TCP_ECN = 1 << 2,
546 SKB_GSO_TCP_FIXEDID = 1 << 3,
548 SKB_GSO_TCPV6 = 1 << 4,
550 SKB_GSO_FCOE = 1 << 5,
552 SKB_GSO_GRE = 1 << 6,
554 SKB_GSO_GRE_CSUM = 1 << 7,
556 SKB_GSO_IPXIP4 = 1 << 8,
558 SKB_GSO_IPXIP6 = 1 << 9,
560 SKB_GSO_UDP_TUNNEL = 1 << 10,
562 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
564 SKB_GSO_PARTIAL = 1 << 12,
566 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
568 SKB_GSO_SCTP = 1 << 14,
570 SKB_GSO_ESP = 1 << 15,
573 #if BITS_PER_LONG > 32
574 #define NET_SKBUFF_DATA_USES_OFFSET 1
577 #ifdef NET_SKBUFF_DATA_USES_OFFSET
578 typedef unsigned int sk_buff_data_t;
580 typedef unsigned char *sk_buff_data_t;
584 * struct sk_buff - socket buffer
585 * @next: Next buffer in list
586 * @prev: Previous buffer in list
587 * @tstamp: Time we arrived/left
588 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
589 * @sk: Socket we are owned by
590 * @dev: Device we arrived on/are leaving by
591 * @cb: Control buffer. Free for use by every layer. Put private vars here
592 * @_skb_refdst: destination entry (with norefcount bit)
593 * @sp: the security path, used for xfrm
594 * @len: Length of actual data
595 * @data_len: Data length
596 * @mac_len: Length of link layer header
597 * @hdr_len: writable header length of cloned skb
598 * @csum: Checksum (must include start/offset pair)
599 * @csum_start: Offset from skb->head where checksumming should start
600 * @csum_offset: Offset from csum_start where checksum should be stored
601 * @priority: Packet queueing priority
602 * @ignore_df: allow local fragmentation
603 * @cloned: Head may be cloned (check refcnt to be sure)
604 * @ip_summed: Driver fed us an IP checksum
605 * @nohdr: Payload reference only, must not modify header
606 * @pkt_type: Packet class
607 * @fclone: skbuff clone status
608 * @ipvs_property: skbuff is owned by ipvs
609 * @tc_skip_classify: do not classify packet. set by IFB device
610 * @tc_at_ingress: used within tc_classify to distinguish in/egress
611 * @tc_redirected: packet was redirected by a tc action
612 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
613 * @peeked: this packet has been seen already, so stats have been
614 * done for it, don't do them again
615 * @nf_trace: netfilter packet trace flag
616 * @protocol: Packet protocol from driver
617 * @destructor: Destruct function
618 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
619 * @_nfct: Associated connection, if any (with nfctinfo bits)
620 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
621 * @skb_iif: ifindex of device we arrived on
622 * @tc_index: Traffic control index
623 * @hash: the packet hash
624 * @queue_mapping: Queue mapping for multiqueue devices
625 * @xmit_more: More SKBs are pending for this queue
626 * @ndisc_nodetype: router type (from link layer)
627 * @ooo_okay: allow the mapping of a socket to a queue to be changed
628 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
630 * @sw_hash: indicates hash was computed in software stack
631 * @wifi_acked_valid: wifi_acked was set
632 * @wifi_acked: whether frame was acked on wifi or not
633 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
634 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
635 * @dst_pending_confirm: need to confirm neighbour
636 * @napi_id: id of the NAPI struct this skb came from
637 * @secmark: security marking
638 * @mark: Generic packet mark
639 * @vlan_proto: vlan encapsulation protocol
640 * @vlan_tci: vlan tag control information
641 * @inner_protocol: Protocol (encapsulation)
642 * @inner_transport_header: Inner transport layer header (encapsulation)
643 * @inner_network_header: Network layer header (encapsulation)
644 * @inner_mac_header: Link layer header (encapsulation)
645 * @transport_header: Transport layer header
646 * @network_header: Network layer header
647 * @mac_header: Link layer header
648 * @tail: Tail pointer
650 * @head: Head of buffer
651 * @data: Data head pointer
652 * @truesize: Buffer size
653 * @users: User count - see {datagram,tcp}.c
659 /* These two members must be first. */
660 struct sk_buff *next;
661 struct sk_buff *prev;
664 struct net_device *dev;
665 /* Some protocols might use this space to store information,
666 * while device pointer would be NULL.
667 * UDP receive path is one user.
669 unsigned long dev_scratch;
672 struct rb_node rbnode; /* used in netem & tcp stack */
681 * This is the control buffer. It is free to use for every
682 * layer. Please put your private variables there. If you
683 * want to keep them across layers you have to do a skb_clone()
684 * first. This is owned by whoever has the skb queued ATM.
686 char cb[48] __aligned(8);
690 unsigned long _skb_refdst;
691 void (*destructor)(struct sk_buff *skb);
693 struct list_head tcp_tsorted_anchor;
699 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
702 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
703 struct nf_bridge_info *nf_bridge;
710 /* Following fields are _not_ copied in __copy_skb_header()
711 * Note that queue_mapping is here mostly to fill a hole.
715 /* if you move cloned around you also must adapt those constants */
716 #ifdef __BIG_ENDIAN_BITFIELD
717 #define CLONED_MASK (1 << 7)
719 #define CLONED_MASK 1
721 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
723 __u8 __cloned_offset[0];
730 __unused:1; /* one bit hole */
732 /* fields enclosed in headers_start/headers_end are copied
733 * using a single memcpy() in __copy_skb_header()
736 __u32 headers_start[0];
739 /* if you move pkt_type around you also must adapt those constants */
740 #ifdef __BIG_ENDIAN_BITFIELD
741 #define PKT_TYPE_MAX (7 << 5)
743 #define PKT_TYPE_MAX 7
745 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
747 __u8 __pkt_type_offset[0];
757 __u8 wifi_acked_valid:1;
761 /* Indicates the inner headers are valid in the skbuff. */
762 __u8 encapsulation:1;
763 __u8 encap_hdr_csum:1;
765 __u8 csum_complete_sw:1;
767 __u8 csum_not_inet:1;
769 __u8 dst_pending_confirm:1;
770 #ifdef CONFIG_IPV6_NDISC_NODETYPE
771 __u8 ndisc_nodetype:2;
773 __u8 ipvs_property:1;
774 __u8 inner_protocol_type:1;
775 __u8 remcsum_offload:1;
776 #ifdef CONFIG_NET_SWITCHDEV
777 __u8 offload_fwd_mark:1;
778 __u8 offload_mr_fwd_mark:1;
780 #ifdef CONFIG_NET_CLS_ACT
781 __u8 tc_skip_classify:1;
782 __u8 tc_at_ingress:1;
783 __u8 tc_redirected:1;
784 __u8 tc_from_ingress:1;
787 #ifdef CONFIG_NET_SCHED
788 __u16 tc_index; /* traffic control index */
803 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
805 unsigned int napi_id;
806 unsigned int sender_cpu;
809 #ifdef CONFIG_NETWORK_SECMARK
815 __u32 reserved_tailroom;
819 __be16 inner_protocol;
823 __u16 inner_transport_header;
824 __u16 inner_network_header;
825 __u16 inner_mac_header;
828 __u16 transport_header;
829 __u16 network_header;
833 __u32 headers_end[0];
836 /* These elements must be at the end, see alloc_skb() for details. */
841 unsigned int truesize;
847 * Handling routines are only of interest to the kernel
849 #include <linux/slab.h>
852 #define SKB_ALLOC_FCLONE 0x01
853 #define SKB_ALLOC_RX 0x02
854 #define SKB_ALLOC_NAPI 0x04
856 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
857 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
859 return unlikely(skb->pfmemalloc);
863 * skb might have a dst pointer attached, refcounted or not.
864 * _skb_refdst low order bit is set if refcount was _not_ taken
866 #define SKB_DST_NOREF 1UL
867 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
869 #define SKB_NFCT_PTRMASK ~(7UL)
871 * skb_dst - returns skb dst_entry
874 * Returns skb dst_entry, regardless of reference taken or not.
876 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
878 /* If refdst was not refcounted, check we still are in a
879 * rcu_read_lock section
881 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
882 !rcu_read_lock_held() &&
883 !rcu_read_lock_bh_held());
884 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
888 * skb_dst_set - sets skb dst
892 * Sets skb dst, assuming a reference was taken on dst and should
893 * be released by skb_dst_drop()
895 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
897 skb->_skb_refdst = (unsigned long)dst;
901 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
905 * Sets skb dst, assuming a reference was not taken on dst.
906 * If dst entry is cached, we do not take reference and dst_release
907 * will be avoided by refdst_drop. If dst entry is not cached, we take
908 * reference, so that last dst_release can destroy the dst immediately.
910 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
912 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
913 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
917 * skb_dst_is_noref - Test if skb dst isn't refcounted
920 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
922 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
925 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
927 return (struct rtable *)skb_dst(skb);
930 /* For mangling skb->pkt_type from user space side from applications
931 * such as nft, tc, etc, we only allow a conservative subset of
932 * possible pkt_types to be set.
934 static inline bool skb_pkt_type_ok(u32 ptype)
936 return ptype <= PACKET_OTHERHOST;
939 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
941 #ifdef CONFIG_NET_RX_BUSY_POLL
948 /* decrement the reference count and return true if we can free the skb */
949 static inline bool skb_unref(struct sk_buff *skb)
953 if (likely(refcount_read(&skb->users) == 1))
955 else if (likely(!refcount_dec_and_test(&skb->users)))
961 void skb_release_head_state(struct sk_buff *skb);
962 void kfree_skb(struct sk_buff *skb);
963 void kfree_skb_list(struct sk_buff *segs);
964 void skb_tx_error(struct sk_buff *skb);
965 void consume_skb(struct sk_buff *skb);
966 void __consume_stateless_skb(struct sk_buff *skb);
967 void __kfree_skb(struct sk_buff *skb);
968 extern struct kmem_cache *skbuff_head_cache;
970 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
971 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
972 bool *fragstolen, int *delta_truesize);
974 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
976 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
977 struct sk_buff *build_skb(void *data, unsigned int frag_size);
978 static inline struct sk_buff *alloc_skb(unsigned int size,
981 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
984 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
985 unsigned long data_len,
990 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
991 struct sk_buff_fclones {
996 refcount_t fclone_ref;
1000 * skb_fclone_busy - check if fclone is busy
1004 * Returns true if skb is a fast clone, and its clone is not freed.
1005 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1006 * so we also check that this didnt happen.
1008 static inline bool skb_fclone_busy(const struct sock *sk,
1009 const struct sk_buff *skb)
1011 const struct sk_buff_fclones *fclones;
1013 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1015 return skb->fclone == SKB_FCLONE_ORIG &&
1016 refcount_read(&fclones->fclone_ref) > 1 &&
1017 fclones->skb2.sk == sk;
1020 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1023 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1026 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1027 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1028 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1029 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1030 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1031 gfp_t gfp_mask, bool fclone);
1032 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1035 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1038 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1039 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1040 unsigned int headroom);
1041 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1042 int newtailroom, gfp_t priority);
1043 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1044 int offset, int len);
1045 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1046 int offset, int len);
1047 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1048 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1051 * skb_pad - zero pad the tail of an skb
1052 * @skb: buffer to pad
1053 * @pad: space to pad
1055 * Ensure that a buffer is followed by a padding area that is zero
1056 * filled. Used by network drivers which may DMA or transfer data
1057 * beyond the buffer end onto the wire.
1059 * May return error in out of memory cases. The skb is freed on error.
1061 static inline int skb_pad(struct sk_buff *skb, int pad)
1063 return __skb_pad(skb, pad, true);
1065 #define dev_kfree_skb(a) consume_skb(a)
1067 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1068 int getfrag(void *from, char *to, int offset,
1069 int len, int odd, struct sk_buff *skb),
1070 void *from, int length);
1072 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1073 int offset, size_t size);
1075 struct skb_seq_state {
1079 __u32 stepped_offset;
1080 struct sk_buff *root_skb;
1081 struct sk_buff *cur_skb;
1085 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1086 unsigned int to, struct skb_seq_state *st);
1087 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1088 struct skb_seq_state *st);
1089 void skb_abort_seq_read(struct skb_seq_state *st);
1091 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1092 unsigned int to, struct ts_config *config);
1095 * Packet hash types specify the type of hash in skb_set_hash.
1097 * Hash types refer to the protocol layer addresses which are used to
1098 * construct a packet's hash. The hashes are used to differentiate or identify
1099 * flows of the protocol layer for the hash type. Hash types are either
1100 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1102 * Properties of hashes:
1104 * 1) Two packets in different flows have different hash values
1105 * 2) Two packets in the same flow should have the same hash value
1107 * A hash at a higher layer is considered to be more specific. A driver should
1108 * set the most specific hash possible.
1110 * A driver cannot indicate a more specific hash than the layer at which a hash
1111 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1113 * A driver may indicate a hash level which is less specific than the
1114 * actual layer the hash was computed on. For instance, a hash computed
1115 * at L4 may be considered an L3 hash. This should only be done if the
1116 * driver can't unambiguously determine that the HW computed the hash at
1117 * the higher layer. Note that the "should" in the second property above
1120 enum pkt_hash_types {
1121 PKT_HASH_TYPE_NONE, /* Undefined type */
1122 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1123 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1124 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1127 static inline void skb_clear_hash(struct sk_buff *skb)
1134 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1137 skb_clear_hash(skb);
1141 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1143 skb->l4_hash = is_l4;
1144 skb->sw_hash = is_sw;
1149 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1151 /* Used by drivers to set hash from HW */
1152 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1156 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1158 __skb_set_hash(skb, hash, true, is_l4);
1161 void __skb_get_hash(struct sk_buff *skb);
1162 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1163 u32 skb_get_poff(const struct sk_buff *skb);
1164 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1165 const struct flow_keys *keys, int hlen);
1166 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1167 void *data, int hlen_proto);
1169 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1170 int thoff, u8 ip_proto)
1172 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1175 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1176 const struct flow_dissector_key *key,
1177 unsigned int key_count);
1179 bool __skb_flow_dissect(const struct sk_buff *skb,
1180 struct flow_dissector *flow_dissector,
1181 void *target_container,
1182 void *data, __be16 proto, int nhoff, int hlen,
1183 unsigned int flags);
1185 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1186 struct flow_dissector *flow_dissector,
1187 void *target_container, unsigned int flags)
1189 return __skb_flow_dissect(skb, flow_dissector, target_container,
1190 NULL, 0, 0, 0, flags);
1193 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1194 struct flow_keys *flow,
1197 memset(flow, 0, sizeof(*flow));
1198 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1199 NULL, 0, 0, 0, flags);
1202 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1203 void *data, __be16 proto,
1204 int nhoff, int hlen,
1207 memset(flow, 0, sizeof(*flow));
1208 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1209 data, proto, nhoff, hlen, flags);
1212 static inline __u32 skb_get_hash(struct sk_buff *skb)
1214 if (!skb->l4_hash && !skb->sw_hash)
1215 __skb_get_hash(skb);
1220 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1222 if (!skb->l4_hash && !skb->sw_hash) {
1223 struct flow_keys keys;
1224 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1226 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1232 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1234 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1239 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1241 to->hash = from->hash;
1242 to->sw_hash = from->sw_hash;
1243 to->l4_hash = from->l4_hash;
1246 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1247 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1249 return skb->head + skb->end;
1252 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1257 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1262 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1264 return skb->end - skb->head;
1269 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1271 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1273 return &skb_shinfo(skb)->hwtstamps;
1276 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1278 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1280 return is_zcopy ? skb_uarg(skb) : NULL;
1283 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg)
1285 if (skb && uarg && !skb_zcopy(skb)) {
1286 sock_zerocopy_get(uarg);
1287 skb_shinfo(skb)->destructor_arg = uarg;
1288 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1292 /* Release a reference on a zerocopy structure */
1293 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1295 struct ubuf_info *uarg = skb_zcopy(skb);
1298 if (uarg->callback == sock_zerocopy_callback) {
1299 uarg->zerocopy = uarg->zerocopy && zerocopy;
1300 sock_zerocopy_put(uarg);
1302 uarg->callback(uarg, zerocopy);
1305 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1309 /* Abort a zerocopy operation and revert zckey on error in send syscall */
1310 static inline void skb_zcopy_abort(struct sk_buff *skb)
1312 struct ubuf_info *uarg = skb_zcopy(skb);
1315 sock_zerocopy_put_abort(uarg);
1316 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1321 * skb_queue_empty - check if a queue is empty
1324 * Returns true if the queue is empty, false otherwise.
1326 static inline int skb_queue_empty(const struct sk_buff_head *list)
1328 return list->next == (const struct sk_buff *) list;
1332 * skb_queue_is_last - check if skb is the last entry in the queue
1336 * Returns true if @skb is the last buffer on the list.
1338 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1339 const struct sk_buff *skb)
1341 return skb->next == (const struct sk_buff *) list;
1345 * skb_queue_is_first - check if skb is the first entry in the queue
1349 * Returns true if @skb is the first buffer on the list.
1351 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1352 const struct sk_buff *skb)
1354 return skb->prev == (const struct sk_buff *) list;
1358 * skb_queue_next - return the next packet in the queue
1360 * @skb: current buffer
1362 * Return the next packet in @list after @skb. It is only valid to
1363 * call this if skb_queue_is_last() evaluates to false.
1365 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1366 const struct sk_buff *skb)
1368 /* This BUG_ON may seem severe, but if we just return then we
1369 * are going to dereference garbage.
1371 BUG_ON(skb_queue_is_last(list, skb));
1376 * skb_queue_prev - return the prev packet in the queue
1378 * @skb: current buffer
1380 * Return the prev packet in @list before @skb. It is only valid to
1381 * call this if skb_queue_is_first() evaluates to false.
1383 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1384 const struct sk_buff *skb)
1386 /* This BUG_ON may seem severe, but if we just return then we
1387 * are going to dereference garbage.
1389 BUG_ON(skb_queue_is_first(list, skb));
1394 * skb_get - reference buffer
1395 * @skb: buffer to reference
1397 * Makes another reference to a socket buffer and returns a pointer
1400 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1402 refcount_inc(&skb->users);
1407 * If users == 1, we are the only owner and are can avoid redundant
1412 * skb_cloned - is the buffer a clone
1413 * @skb: buffer to check
1415 * Returns true if the buffer was generated with skb_clone() and is
1416 * one of multiple shared copies of the buffer. Cloned buffers are
1417 * shared data so must not be written to under normal circumstances.
1419 static inline int skb_cloned(const struct sk_buff *skb)
1421 return skb->cloned &&
1422 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1425 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1427 might_sleep_if(gfpflags_allow_blocking(pri));
1429 if (skb_cloned(skb))
1430 return pskb_expand_head(skb, 0, 0, pri);
1436 * skb_header_cloned - is the header a clone
1437 * @skb: buffer to check
1439 * Returns true if modifying the header part of the buffer requires
1440 * the data to be copied.
1442 static inline int skb_header_cloned(const struct sk_buff *skb)
1449 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1450 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1451 return dataref != 1;
1454 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1456 might_sleep_if(gfpflags_allow_blocking(pri));
1458 if (skb_header_cloned(skb))
1459 return pskb_expand_head(skb, 0, 0, pri);
1465 * __skb_header_release - release reference to header
1466 * @skb: buffer to operate on
1468 static inline void __skb_header_release(struct sk_buff *skb)
1471 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1476 * skb_shared - is the buffer shared
1477 * @skb: buffer to check
1479 * Returns true if more than one person has a reference to this
1482 static inline int skb_shared(const struct sk_buff *skb)
1484 return refcount_read(&skb->users) != 1;
1488 * skb_share_check - check if buffer is shared and if so clone it
1489 * @skb: buffer to check
1490 * @pri: priority for memory allocation
1492 * If the buffer is shared the buffer is cloned and the old copy
1493 * drops a reference. A new clone with a single reference is returned.
1494 * If the buffer is not shared the original buffer is returned. When
1495 * being called from interrupt status or with spinlocks held pri must
1498 * NULL is returned on a memory allocation failure.
1500 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1502 might_sleep_if(gfpflags_allow_blocking(pri));
1503 if (skb_shared(skb)) {
1504 struct sk_buff *nskb = skb_clone(skb, pri);
1516 * Copy shared buffers into a new sk_buff. We effectively do COW on
1517 * packets to handle cases where we have a local reader and forward
1518 * and a couple of other messy ones. The normal one is tcpdumping
1519 * a packet thats being forwarded.
1523 * skb_unshare - make a copy of a shared buffer
1524 * @skb: buffer to check
1525 * @pri: priority for memory allocation
1527 * If the socket buffer is a clone then this function creates a new
1528 * copy of the data, drops a reference count on the old copy and returns
1529 * the new copy with the reference count at 1. If the buffer is not a clone
1530 * the original buffer is returned. When called with a spinlock held or
1531 * from interrupt state @pri must be %GFP_ATOMIC
1533 * %NULL is returned on a memory allocation failure.
1535 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1538 might_sleep_if(gfpflags_allow_blocking(pri));
1539 if (skb_cloned(skb)) {
1540 struct sk_buff *nskb = skb_copy(skb, pri);
1542 /* Free our shared copy */
1553 * skb_peek - peek at the head of an &sk_buff_head
1554 * @list_: list to peek at
1556 * Peek an &sk_buff. Unlike most other operations you _MUST_
1557 * be careful with this one. A peek leaves the buffer on the
1558 * list and someone else may run off with it. You must hold
1559 * the appropriate locks or have a private queue to do this.
1561 * Returns %NULL for an empty list or a pointer to the head element.
1562 * The reference count is not incremented and the reference is therefore
1563 * volatile. Use with caution.
1565 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1567 struct sk_buff *skb = list_->next;
1569 if (skb == (struct sk_buff *)list_)
1575 * skb_peek_next - peek skb following the given one from a queue
1576 * @skb: skb to start from
1577 * @list_: list to peek at
1579 * Returns %NULL when the end of the list is met or a pointer to the
1580 * next element. The reference count is not incremented and the
1581 * reference is therefore volatile. Use with caution.
1583 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1584 const struct sk_buff_head *list_)
1586 struct sk_buff *next = skb->next;
1588 if (next == (struct sk_buff *)list_)
1594 * skb_peek_tail - peek at the tail of an &sk_buff_head
1595 * @list_: list to peek at
1597 * Peek an &sk_buff. Unlike most other operations you _MUST_
1598 * be careful with this one. A peek leaves the buffer on the
1599 * list and someone else may run off with it. You must hold
1600 * the appropriate locks or have a private queue to do this.
1602 * Returns %NULL for an empty list or a pointer to the tail element.
1603 * The reference count is not incremented and the reference is therefore
1604 * volatile. Use with caution.
1606 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1608 struct sk_buff *skb = list_->prev;
1610 if (skb == (struct sk_buff *)list_)
1617 * skb_queue_len - get queue length
1618 * @list_: list to measure
1620 * Return the length of an &sk_buff queue.
1622 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1628 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1629 * @list: queue to initialize
1631 * This initializes only the list and queue length aspects of
1632 * an sk_buff_head object. This allows to initialize the list
1633 * aspects of an sk_buff_head without reinitializing things like
1634 * the spinlock. It can also be used for on-stack sk_buff_head
1635 * objects where the spinlock is known to not be used.
1637 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1639 list->prev = list->next = (struct sk_buff *)list;
1644 * This function creates a split out lock class for each invocation;
1645 * this is needed for now since a whole lot of users of the skb-queue
1646 * infrastructure in drivers have different locking usage (in hardirq)
1647 * than the networking core (in softirq only). In the long run either the
1648 * network layer or drivers should need annotation to consolidate the
1649 * main types of usage into 3 classes.
1651 static inline void skb_queue_head_init(struct sk_buff_head *list)
1653 spin_lock_init(&list->lock);
1654 __skb_queue_head_init(list);
1657 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1658 struct lock_class_key *class)
1660 skb_queue_head_init(list);
1661 lockdep_set_class(&list->lock, class);
1665 * Insert an sk_buff on a list.
1667 * The "__skb_xxxx()" functions are the non-atomic ones that
1668 * can only be called with interrupts disabled.
1670 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1671 struct sk_buff_head *list);
1672 static inline void __skb_insert(struct sk_buff *newsk,
1673 struct sk_buff *prev, struct sk_buff *next,
1674 struct sk_buff_head *list)
1678 next->prev = prev->next = newsk;
1682 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1683 struct sk_buff *prev,
1684 struct sk_buff *next)
1686 struct sk_buff *first = list->next;
1687 struct sk_buff *last = list->prev;
1697 * skb_queue_splice - join two skb lists, this is designed for stacks
1698 * @list: the new list to add
1699 * @head: the place to add it in the first list
1701 static inline void skb_queue_splice(const struct sk_buff_head *list,
1702 struct sk_buff_head *head)
1704 if (!skb_queue_empty(list)) {
1705 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1706 head->qlen += list->qlen;
1711 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1712 * @list: the new list to add
1713 * @head: the place to add it in the first list
1715 * The list at @list is reinitialised
1717 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1718 struct sk_buff_head *head)
1720 if (!skb_queue_empty(list)) {
1721 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1722 head->qlen += list->qlen;
1723 __skb_queue_head_init(list);
1728 * skb_queue_splice_tail - join two skb lists, each list being a queue
1729 * @list: the new list to add
1730 * @head: the place to add it in the first list
1732 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1733 struct sk_buff_head *head)
1735 if (!skb_queue_empty(list)) {
1736 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1737 head->qlen += list->qlen;
1742 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1743 * @list: the new list to add
1744 * @head: the place to add it in the first list
1746 * Each of the lists is a queue.
1747 * The list at @list is reinitialised
1749 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1750 struct sk_buff_head *head)
1752 if (!skb_queue_empty(list)) {
1753 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1754 head->qlen += list->qlen;
1755 __skb_queue_head_init(list);
1760 * __skb_queue_after - queue a buffer at the list head
1761 * @list: list to use
1762 * @prev: place after this buffer
1763 * @newsk: buffer to queue
1765 * Queue a buffer int the middle of a list. This function takes no locks
1766 * and you must therefore hold required locks before calling it.
1768 * A buffer cannot be placed on two lists at the same time.
1770 static inline void __skb_queue_after(struct sk_buff_head *list,
1771 struct sk_buff *prev,
1772 struct sk_buff *newsk)
1774 __skb_insert(newsk, prev, prev->next, list);
1777 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1778 struct sk_buff_head *list);
1780 static inline void __skb_queue_before(struct sk_buff_head *list,
1781 struct sk_buff *next,
1782 struct sk_buff *newsk)
1784 __skb_insert(newsk, next->prev, next, list);
1788 * __skb_queue_head - queue a buffer at the list head
1789 * @list: list to use
1790 * @newsk: buffer to queue
1792 * Queue a buffer at the start of a list. This function takes no locks
1793 * and you must therefore hold required locks before calling it.
1795 * A buffer cannot be placed on two lists at the same time.
1797 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1798 static inline void __skb_queue_head(struct sk_buff_head *list,
1799 struct sk_buff *newsk)
1801 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1805 * __skb_queue_tail - queue a buffer at the list tail
1806 * @list: list to use
1807 * @newsk: buffer to queue
1809 * Queue a buffer at the end of a list. This function takes no locks
1810 * and you must therefore hold required locks before calling it.
1812 * A buffer cannot be placed on two lists at the same time.
1814 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1815 static inline void __skb_queue_tail(struct sk_buff_head *list,
1816 struct sk_buff *newsk)
1818 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1822 * remove sk_buff from list. _Must_ be called atomically, and with
1825 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1826 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1828 struct sk_buff *next, *prev;
1833 skb->next = skb->prev = NULL;
1839 * __skb_dequeue - remove from the head of the queue
1840 * @list: list to dequeue from
1842 * Remove the head of the list. This function does not take any locks
1843 * so must be used with appropriate locks held only. The head item is
1844 * returned or %NULL if the list is empty.
1846 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1847 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1849 struct sk_buff *skb = skb_peek(list);
1851 __skb_unlink(skb, list);
1856 * __skb_dequeue_tail - remove from the tail of the queue
1857 * @list: list to dequeue from
1859 * Remove the tail of the list. This function does not take any locks
1860 * so must be used with appropriate locks held only. The tail item is
1861 * returned or %NULL if the list is empty.
1863 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1864 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1866 struct sk_buff *skb = skb_peek_tail(list);
1868 __skb_unlink(skb, list);
1873 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1875 return skb->data_len;
1878 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1880 return skb->len - skb->data_len;
1883 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
1885 unsigned int i, len = 0;
1887 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1888 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1892 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1894 return skb_headlen(skb) + __skb_pagelen(skb);
1898 * __skb_fill_page_desc - initialise a paged fragment in an skb
1899 * @skb: buffer containing fragment to be initialised
1900 * @i: paged fragment index to initialise
1901 * @page: the page to use for this fragment
1902 * @off: the offset to the data with @page
1903 * @size: the length of the data
1905 * Initialises the @i'th fragment of @skb to point to &size bytes at
1906 * offset @off within @page.
1908 * Does not take any additional reference on the fragment.
1910 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1911 struct page *page, int off, int size)
1913 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1916 * Propagate page pfmemalloc to the skb if we can. The problem is
1917 * that not all callers have unique ownership of the page but rely
1918 * on page_is_pfmemalloc doing the right thing(tm).
1920 frag->page.p = page;
1921 frag->page_offset = off;
1922 skb_frag_size_set(frag, size);
1924 page = compound_head(page);
1925 if (page_is_pfmemalloc(page))
1926 skb->pfmemalloc = true;
1930 * skb_fill_page_desc - initialise a paged fragment in an skb
1931 * @skb: buffer containing fragment to be initialised
1932 * @i: paged fragment index to initialise
1933 * @page: the page to use for this fragment
1934 * @off: the offset to the data with @page
1935 * @size: the length of the data
1937 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1938 * @skb to point to @size bytes at offset @off within @page. In
1939 * addition updates @skb such that @i is the last fragment.
1941 * Does not take any additional reference on the fragment.
1943 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1944 struct page *page, int off, int size)
1946 __skb_fill_page_desc(skb, i, page, off, size);
1947 skb_shinfo(skb)->nr_frags = i + 1;
1950 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1951 int size, unsigned int truesize);
1953 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1954 unsigned int truesize);
1956 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1957 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1958 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1960 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1961 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1963 return skb->head + skb->tail;
1966 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1968 skb->tail = skb->data - skb->head;
1971 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1973 skb_reset_tail_pointer(skb);
1974 skb->tail += offset;
1977 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1978 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1983 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1985 skb->tail = skb->data;
1988 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1990 skb->tail = skb->data + offset;
1993 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1996 * Add data to an sk_buff
1998 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1999 void *skb_put(struct sk_buff *skb, unsigned int len);
2000 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2002 void *tmp = skb_tail_pointer(skb);
2003 SKB_LINEAR_ASSERT(skb);
2009 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2011 void *tmp = __skb_put(skb, len);
2013 memset(tmp, 0, len);
2017 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2020 void *tmp = __skb_put(skb, len);
2022 memcpy(tmp, data, len);
2026 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2028 *(u8 *)__skb_put(skb, 1) = val;
2031 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2033 void *tmp = skb_put(skb, len);
2035 memset(tmp, 0, len);
2040 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2043 void *tmp = skb_put(skb, len);
2045 memcpy(tmp, data, len);
2050 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2052 *(u8 *)skb_put(skb, 1) = val;
2055 void *skb_push(struct sk_buff *skb, unsigned int len);
2056 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2063 void *skb_pull(struct sk_buff *skb, unsigned int len);
2064 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2067 BUG_ON(skb->len < skb->data_len);
2068 return skb->data += len;
2071 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2073 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2076 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2078 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2080 if (len > skb_headlen(skb) &&
2081 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2084 return skb->data += len;
2087 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2089 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2092 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2094 if (likely(len <= skb_headlen(skb)))
2096 if (unlikely(len > skb->len))
2098 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2101 void skb_condense(struct sk_buff *skb);
2104 * skb_headroom - bytes at buffer head
2105 * @skb: buffer to check
2107 * Return the number of bytes of free space at the head of an &sk_buff.
2109 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2111 return skb->data - skb->head;
2115 * skb_tailroom - bytes at buffer end
2116 * @skb: buffer to check
2118 * Return the number of bytes of free space at the tail of an sk_buff
2120 static inline int skb_tailroom(const struct sk_buff *skb)
2122 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2126 * skb_availroom - bytes at buffer end
2127 * @skb: buffer to check
2129 * Return the number of bytes of free space at the tail of an sk_buff
2130 * allocated by sk_stream_alloc()
2132 static inline int skb_availroom(const struct sk_buff *skb)
2134 if (skb_is_nonlinear(skb))
2137 return skb->end - skb->tail - skb->reserved_tailroom;
2141 * skb_reserve - adjust headroom
2142 * @skb: buffer to alter
2143 * @len: bytes to move
2145 * Increase the headroom of an empty &sk_buff by reducing the tail
2146 * room. This is only allowed for an empty buffer.
2148 static inline void skb_reserve(struct sk_buff *skb, int len)
2155 * skb_tailroom_reserve - adjust reserved_tailroom
2156 * @skb: buffer to alter
2157 * @mtu: maximum amount of headlen permitted
2158 * @needed_tailroom: minimum amount of reserved_tailroom
2160 * Set reserved_tailroom so that headlen can be as large as possible but
2161 * not larger than mtu and tailroom cannot be smaller than
2163 * The required headroom should already have been reserved before using
2166 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2167 unsigned int needed_tailroom)
2169 SKB_LINEAR_ASSERT(skb);
2170 if (mtu < skb_tailroom(skb) - needed_tailroom)
2171 /* use at most mtu */
2172 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2174 /* use up to all available space */
2175 skb->reserved_tailroom = needed_tailroom;
2178 #define ENCAP_TYPE_ETHER 0
2179 #define ENCAP_TYPE_IPPROTO 1
2181 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2184 skb->inner_protocol = protocol;
2185 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2188 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2191 skb->inner_ipproto = ipproto;
2192 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2195 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2197 skb->inner_mac_header = skb->mac_header;
2198 skb->inner_network_header = skb->network_header;
2199 skb->inner_transport_header = skb->transport_header;
2202 static inline void skb_reset_mac_len(struct sk_buff *skb)
2204 skb->mac_len = skb->network_header - skb->mac_header;
2207 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2210 return skb->head + skb->inner_transport_header;
2213 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2215 return skb_inner_transport_header(skb) - skb->data;
2218 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2220 skb->inner_transport_header = skb->data - skb->head;
2223 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2226 skb_reset_inner_transport_header(skb);
2227 skb->inner_transport_header += offset;
2230 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2232 return skb->head + skb->inner_network_header;
2235 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2237 skb->inner_network_header = skb->data - skb->head;
2240 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2243 skb_reset_inner_network_header(skb);
2244 skb->inner_network_header += offset;
2247 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2249 return skb->head + skb->inner_mac_header;
2252 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2254 skb->inner_mac_header = skb->data - skb->head;
2257 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2260 skb_reset_inner_mac_header(skb);
2261 skb->inner_mac_header += offset;
2263 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2265 return skb->transport_header != (typeof(skb->transport_header))~0U;
2268 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2270 return skb->head + skb->transport_header;
2273 static inline void skb_reset_transport_header(struct sk_buff *skb)
2275 skb->transport_header = skb->data - skb->head;
2278 static inline void skb_set_transport_header(struct sk_buff *skb,
2281 skb_reset_transport_header(skb);
2282 skb->transport_header += offset;
2285 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2287 return skb->head + skb->network_header;
2290 static inline void skb_reset_network_header(struct sk_buff *skb)
2292 skb->network_header = skb->data - skb->head;
2295 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2297 skb_reset_network_header(skb);
2298 skb->network_header += offset;
2301 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2303 return skb->head + skb->mac_header;
2306 static inline int skb_mac_offset(const struct sk_buff *skb)
2308 return skb_mac_header(skb) - skb->data;
2311 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2313 return skb->network_header - skb->mac_header;
2316 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2318 return skb->mac_header != (typeof(skb->mac_header))~0U;
2321 static inline void skb_reset_mac_header(struct sk_buff *skb)
2323 skb->mac_header = skb->data - skb->head;
2326 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2328 skb_reset_mac_header(skb);
2329 skb->mac_header += offset;
2332 static inline void skb_pop_mac_header(struct sk_buff *skb)
2334 skb->mac_header = skb->network_header;
2337 static inline void skb_probe_transport_header(struct sk_buff *skb,
2338 const int offset_hint)
2340 struct flow_keys keys;
2342 if (skb_transport_header_was_set(skb))
2344 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2345 skb_set_transport_header(skb, keys.control.thoff);
2347 skb_set_transport_header(skb, offset_hint);
2350 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2352 if (skb_mac_header_was_set(skb)) {
2353 const unsigned char *old_mac = skb_mac_header(skb);
2355 skb_set_mac_header(skb, -skb->mac_len);
2356 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2360 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2362 return skb->csum_start - skb_headroom(skb);
2365 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2367 return skb->head + skb->csum_start;
2370 static inline int skb_transport_offset(const struct sk_buff *skb)
2372 return skb_transport_header(skb) - skb->data;
2375 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2377 return skb->transport_header - skb->network_header;
2380 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2382 return skb->inner_transport_header - skb->inner_network_header;
2385 static inline int skb_network_offset(const struct sk_buff *skb)
2387 return skb_network_header(skb) - skb->data;
2390 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2392 return skb_inner_network_header(skb) - skb->data;
2395 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2397 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2401 * CPUs often take a performance hit when accessing unaligned memory
2402 * locations. The actual performance hit varies, it can be small if the
2403 * hardware handles it or large if we have to take an exception and fix it
2406 * Since an ethernet header is 14 bytes network drivers often end up with
2407 * the IP header at an unaligned offset. The IP header can be aligned by
2408 * shifting the start of the packet by 2 bytes. Drivers should do this
2411 * skb_reserve(skb, NET_IP_ALIGN);
2413 * The downside to this alignment of the IP header is that the DMA is now
2414 * unaligned. On some architectures the cost of an unaligned DMA is high
2415 * and this cost outweighs the gains made by aligning the IP header.
2417 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2420 #ifndef NET_IP_ALIGN
2421 #define NET_IP_ALIGN 2
2425 * The networking layer reserves some headroom in skb data (via
2426 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2427 * the header has to grow. In the default case, if the header has to grow
2428 * 32 bytes or less we avoid the reallocation.
2430 * Unfortunately this headroom changes the DMA alignment of the resulting
2431 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2432 * on some architectures. An architecture can override this value,
2433 * perhaps setting it to a cacheline in size (since that will maintain
2434 * cacheline alignment of the DMA). It must be a power of 2.
2436 * Various parts of the networking layer expect at least 32 bytes of
2437 * headroom, you should not reduce this.
2439 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2440 * to reduce average number of cache lines per packet.
2441 * get_rps_cpus() for example only access one 64 bytes aligned block :
2442 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2445 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2448 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2450 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2452 if (unlikely(skb_is_nonlinear(skb))) {
2457 skb_set_tail_pointer(skb, len);
2460 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2462 __skb_set_length(skb, len);
2465 void skb_trim(struct sk_buff *skb, unsigned int len);
2467 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2470 return ___pskb_trim(skb, len);
2471 __skb_trim(skb, len);
2475 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2477 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2481 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2482 * @skb: buffer to alter
2485 * This is identical to pskb_trim except that the caller knows that
2486 * the skb is not cloned so we should never get an error due to out-
2489 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2491 int err = pskb_trim(skb, len);
2495 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2497 unsigned int diff = len - skb->len;
2499 if (skb_tailroom(skb) < diff) {
2500 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2505 __skb_set_length(skb, len);
2510 * skb_orphan - orphan a buffer
2511 * @skb: buffer to orphan
2513 * If a buffer currently has an owner then we call the owner's
2514 * destructor function and make the @skb unowned. The buffer continues
2515 * to exist but is no longer charged to its former owner.
2517 static inline void skb_orphan(struct sk_buff *skb)
2519 if (skb->destructor) {
2520 skb->destructor(skb);
2521 skb->destructor = NULL;
2529 * skb_orphan_frags - orphan the frags contained in a buffer
2530 * @skb: buffer to orphan frags from
2531 * @gfp_mask: allocation mask for replacement pages
2533 * For each frag in the SKB which needs a destructor (i.e. has an
2534 * owner) create a copy of that frag and release the original
2535 * page by calling the destructor.
2537 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2539 if (likely(!skb_zcopy(skb)))
2541 if (skb_uarg(skb)->callback == sock_zerocopy_callback)
2543 return skb_copy_ubufs(skb, gfp_mask);
2546 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2547 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2549 if (likely(!skb_zcopy(skb)))
2551 return skb_copy_ubufs(skb, gfp_mask);
2555 * __skb_queue_purge - empty a list
2556 * @list: list to empty
2558 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2559 * the list and one reference dropped. This function does not take the
2560 * list lock and the caller must hold the relevant locks to use it.
2562 void skb_queue_purge(struct sk_buff_head *list);
2563 static inline void __skb_queue_purge(struct sk_buff_head *list)
2565 struct sk_buff *skb;
2566 while ((skb = __skb_dequeue(list)) != NULL)
2570 void skb_rbtree_purge(struct rb_root *root);
2572 void *netdev_alloc_frag(unsigned int fragsz);
2574 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2578 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2579 * @dev: network device to receive on
2580 * @length: length to allocate
2582 * Allocate a new &sk_buff and assign it a usage count of one. The
2583 * buffer has unspecified headroom built in. Users should allocate
2584 * the headroom they think they need without accounting for the
2585 * built in space. The built in space is used for optimisations.
2587 * %NULL is returned if there is no free memory. Although this function
2588 * allocates memory it can be called from an interrupt.
2590 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2591 unsigned int length)
2593 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2596 /* legacy helper around __netdev_alloc_skb() */
2597 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2600 return __netdev_alloc_skb(NULL, length, gfp_mask);
2603 /* legacy helper around netdev_alloc_skb() */
2604 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2606 return netdev_alloc_skb(NULL, length);
2610 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2611 unsigned int length, gfp_t gfp)
2613 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2615 if (NET_IP_ALIGN && skb)
2616 skb_reserve(skb, NET_IP_ALIGN);
2620 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2621 unsigned int length)
2623 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2626 static inline void skb_free_frag(void *addr)
2628 page_frag_free(addr);
2631 void *napi_alloc_frag(unsigned int fragsz);
2632 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2633 unsigned int length, gfp_t gfp_mask);
2634 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2635 unsigned int length)
2637 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2639 void napi_consume_skb(struct sk_buff *skb, int budget);
2641 void __kfree_skb_flush(void);
2642 void __kfree_skb_defer(struct sk_buff *skb);
2645 * __dev_alloc_pages - allocate page for network Rx
2646 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2647 * @order: size of the allocation
2649 * Allocate a new page.
2651 * %NULL is returned if there is no free memory.
2653 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2656 /* This piece of code contains several assumptions.
2657 * 1. This is for device Rx, therefor a cold page is preferred.
2658 * 2. The expectation is the user wants a compound page.
2659 * 3. If requesting a order 0 page it will not be compound
2660 * due to the check to see if order has a value in prep_new_page
2661 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2662 * code in gfp_to_alloc_flags that should be enforcing this.
2664 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2666 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2669 static inline struct page *dev_alloc_pages(unsigned int order)
2671 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2675 * __dev_alloc_page - allocate a page for network Rx
2676 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2678 * Allocate a new page.
2680 * %NULL is returned if there is no free memory.
2682 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2684 return __dev_alloc_pages(gfp_mask, 0);
2687 static inline struct page *dev_alloc_page(void)
2689 return dev_alloc_pages(0);
2693 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2694 * @page: The page that was allocated from skb_alloc_page
2695 * @skb: The skb that may need pfmemalloc set
2697 static inline void skb_propagate_pfmemalloc(struct page *page,
2698 struct sk_buff *skb)
2700 if (page_is_pfmemalloc(page))
2701 skb->pfmemalloc = true;
2705 * skb_frag_page - retrieve the page referred to by a paged fragment
2706 * @frag: the paged fragment
2708 * Returns the &struct page associated with @frag.
2710 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2712 return frag->page.p;
2716 * __skb_frag_ref - take an addition reference on a paged fragment.
2717 * @frag: the paged fragment
2719 * Takes an additional reference on the paged fragment @frag.
2721 static inline void __skb_frag_ref(skb_frag_t *frag)
2723 get_page(skb_frag_page(frag));
2727 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2729 * @f: the fragment offset.
2731 * Takes an additional reference on the @f'th paged fragment of @skb.
2733 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2735 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2739 * __skb_frag_unref - release a reference on a paged fragment.
2740 * @frag: the paged fragment
2742 * Releases a reference on the paged fragment @frag.
2744 static inline void __skb_frag_unref(skb_frag_t *frag)
2746 put_page(skb_frag_page(frag));
2750 * skb_frag_unref - release a reference on a paged fragment of an skb.
2752 * @f: the fragment offset
2754 * Releases a reference on the @f'th paged fragment of @skb.
2756 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2758 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2762 * skb_frag_address - gets the address of the data contained in a paged fragment
2763 * @frag: the paged fragment buffer
2765 * Returns the address of the data within @frag. The page must already
2768 static inline void *skb_frag_address(const skb_frag_t *frag)
2770 return page_address(skb_frag_page(frag)) + frag->page_offset;
2774 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2775 * @frag: the paged fragment buffer
2777 * Returns the address of the data within @frag. Checks that the page
2778 * is mapped and returns %NULL otherwise.
2780 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2782 void *ptr = page_address(skb_frag_page(frag));
2786 return ptr + frag->page_offset;
2790 * __skb_frag_set_page - sets the page contained in a paged fragment
2791 * @frag: the paged fragment
2792 * @page: the page to set
2794 * Sets the fragment @frag to contain @page.
2796 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2798 frag->page.p = page;
2802 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2804 * @f: the fragment offset
2805 * @page: the page to set
2807 * Sets the @f'th fragment of @skb to contain @page.
2809 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2812 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2815 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2818 * skb_frag_dma_map - maps a paged fragment via the DMA API
2819 * @dev: the device to map the fragment to
2820 * @frag: the paged fragment to map
2821 * @offset: the offset within the fragment (starting at the
2822 * fragment's own offset)
2823 * @size: the number of bytes to map
2824 * @dir: the direction of the mapping (``PCI_DMA_*``)
2826 * Maps the page associated with @frag to @device.
2828 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2829 const skb_frag_t *frag,
2830 size_t offset, size_t size,
2831 enum dma_data_direction dir)
2833 return dma_map_page(dev, skb_frag_page(frag),
2834 frag->page_offset + offset, size, dir);
2837 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2840 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2844 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2847 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2852 * skb_clone_writable - is the header of a clone writable
2853 * @skb: buffer to check
2854 * @len: length up to which to write
2856 * Returns true if modifying the header part of the cloned buffer
2857 * does not requires the data to be copied.
2859 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2861 return !skb_header_cloned(skb) &&
2862 skb_headroom(skb) + len <= skb->hdr_len;
2865 static inline int skb_try_make_writable(struct sk_buff *skb,
2866 unsigned int write_len)
2868 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2869 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2872 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2877 if (headroom > skb_headroom(skb))
2878 delta = headroom - skb_headroom(skb);
2880 if (delta || cloned)
2881 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2887 * skb_cow - copy header of skb when it is required
2888 * @skb: buffer to cow
2889 * @headroom: needed headroom
2891 * If the skb passed lacks sufficient headroom or its data part
2892 * is shared, data is reallocated. If reallocation fails, an error
2893 * is returned and original skb is not changed.
2895 * The result is skb with writable area skb->head...skb->tail
2896 * and at least @headroom of space at head.
2898 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2900 return __skb_cow(skb, headroom, skb_cloned(skb));
2904 * skb_cow_head - skb_cow but only making the head writable
2905 * @skb: buffer to cow
2906 * @headroom: needed headroom
2908 * This function is identical to skb_cow except that we replace the
2909 * skb_cloned check by skb_header_cloned. It should be used when
2910 * you only need to push on some header and do not need to modify
2913 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2915 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2919 * skb_padto - pad an skbuff up to a minimal size
2920 * @skb: buffer to pad
2921 * @len: minimal length
2923 * Pads up a buffer to ensure the trailing bytes exist and are
2924 * blanked. If the buffer already contains sufficient data it
2925 * is untouched. Otherwise it is extended. Returns zero on
2926 * success. The skb is freed on error.
2928 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2930 unsigned int size = skb->len;
2931 if (likely(size >= len))
2933 return skb_pad(skb, len - size);
2937 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2938 * @skb: buffer to pad
2939 * @len: minimal length
2940 * @free_on_error: free buffer on error
2942 * Pads up a buffer to ensure the trailing bytes exist and are
2943 * blanked. If the buffer already contains sufficient data it
2944 * is untouched. Otherwise it is extended. Returns zero on
2945 * success. The skb is freed on error if @free_on_error is true.
2947 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
2950 unsigned int size = skb->len;
2952 if (unlikely(size < len)) {
2954 if (__skb_pad(skb, len, free_on_error))
2956 __skb_put(skb, len);
2962 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2963 * @skb: buffer to pad
2964 * @len: minimal length
2966 * Pads up a buffer to ensure the trailing bytes exist and are
2967 * blanked. If the buffer already contains sufficient data it
2968 * is untouched. Otherwise it is extended. Returns zero on
2969 * success. The skb is freed on error.
2971 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2973 return __skb_put_padto(skb, len, true);
2976 static inline int skb_add_data(struct sk_buff *skb,
2977 struct iov_iter *from, int copy)
2979 const int off = skb->len;
2981 if (skb->ip_summed == CHECKSUM_NONE) {
2983 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2985 skb->csum = csum_block_add(skb->csum, csum, off);
2988 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
2991 __skb_trim(skb, off);
2995 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2996 const struct page *page, int off)
3001 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
3003 return page == skb_frag_page(frag) &&
3004 off == frag->page_offset + skb_frag_size(frag);
3009 static inline int __skb_linearize(struct sk_buff *skb)
3011 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3015 * skb_linearize - convert paged skb to linear one
3016 * @skb: buffer to linarize
3018 * If there is no free memory -ENOMEM is returned, otherwise zero
3019 * is returned and the old skb data released.
3021 static inline int skb_linearize(struct sk_buff *skb)
3023 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3027 * skb_has_shared_frag - can any frag be overwritten
3028 * @skb: buffer to test
3030 * Return true if the skb has at least one frag that might be modified
3031 * by an external entity (as in vmsplice()/sendfile())
3033 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3035 return skb_is_nonlinear(skb) &&
3036 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3040 * skb_linearize_cow - make sure skb is linear and writable
3041 * @skb: buffer to process
3043 * If there is no free memory -ENOMEM is returned, otherwise zero
3044 * is returned and the old skb data released.
3046 static inline int skb_linearize_cow(struct sk_buff *skb)
3048 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3049 __skb_linearize(skb) : 0;
3052 static __always_inline void
3053 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3056 if (skb->ip_summed == CHECKSUM_COMPLETE)
3057 skb->csum = csum_block_sub(skb->csum,
3058 csum_partial(start, len, 0), off);
3059 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3060 skb_checksum_start_offset(skb) < 0)
3061 skb->ip_summed = CHECKSUM_NONE;
3065 * skb_postpull_rcsum - update checksum for received skb after pull
3066 * @skb: buffer to update
3067 * @start: start of data before pull
3068 * @len: length of data pulled
3070 * After doing a pull on a received packet, you need to call this to
3071 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3072 * CHECKSUM_NONE so that it can be recomputed from scratch.
3074 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3075 const void *start, unsigned int len)
3077 __skb_postpull_rcsum(skb, start, len, 0);
3080 static __always_inline void
3081 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3084 if (skb->ip_summed == CHECKSUM_COMPLETE)
3085 skb->csum = csum_block_add(skb->csum,
3086 csum_partial(start, len, 0), off);
3090 * skb_postpush_rcsum - update checksum for received skb after push
3091 * @skb: buffer to update
3092 * @start: start of data after push
3093 * @len: length of data pushed
3095 * After doing a push on a received packet, you need to call this to
3096 * update the CHECKSUM_COMPLETE checksum.
3098 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3099 const void *start, unsigned int len)
3101 __skb_postpush_rcsum(skb, start, len, 0);
3104 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3107 * skb_push_rcsum - push skb and update receive checksum
3108 * @skb: buffer to update
3109 * @len: length of data pulled
3111 * This function performs an skb_push on the packet and updates
3112 * the CHECKSUM_COMPLETE checksum. It should be used on
3113 * receive path processing instead of skb_push unless you know
3114 * that the checksum difference is zero (e.g., a valid IP header)
3115 * or you are setting ip_summed to CHECKSUM_NONE.
3117 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3120 skb_postpush_rcsum(skb, skb->data, len);
3125 * pskb_trim_rcsum - trim received skb and update checksum
3126 * @skb: buffer to trim
3129 * This is exactly the same as pskb_trim except that it ensures the
3130 * checksum of received packets are still valid after the operation.
3133 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3135 if (likely(len >= skb->len))
3137 if (skb->ip_summed == CHECKSUM_COMPLETE)
3138 skb->ip_summed = CHECKSUM_NONE;
3139 return __pskb_trim(skb, len);
3142 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3144 if (skb->ip_summed == CHECKSUM_COMPLETE)
3145 skb->ip_summed = CHECKSUM_NONE;
3146 __skb_trim(skb, len);
3150 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3152 if (skb->ip_summed == CHECKSUM_COMPLETE)
3153 skb->ip_summed = CHECKSUM_NONE;
3154 return __skb_grow(skb, len);
3157 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3158 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3159 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3160 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3161 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3163 #define skb_queue_walk(queue, skb) \
3164 for (skb = (queue)->next; \
3165 skb != (struct sk_buff *)(queue); \
3168 #define skb_queue_walk_safe(queue, skb, tmp) \
3169 for (skb = (queue)->next, tmp = skb->next; \
3170 skb != (struct sk_buff *)(queue); \
3171 skb = tmp, tmp = skb->next)
3173 #define skb_queue_walk_from(queue, skb) \
3174 for (; skb != (struct sk_buff *)(queue); \
3177 #define skb_rbtree_walk(skb, root) \
3178 for (skb = skb_rb_first(root); skb != NULL; \
3179 skb = skb_rb_next(skb))
3181 #define skb_rbtree_walk_from(skb) \
3182 for (; skb != NULL; \
3183 skb = skb_rb_next(skb))
3185 #define skb_rbtree_walk_from_safe(skb, tmp) \
3186 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3189 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3190 for (tmp = skb->next; \
3191 skb != (struct sk_buff *)(queue); \
3192 skb = tmp, tmp = skb->next)
3194 #define skb_queue_reverse_walk(queue, skb) \
3195 for (skb = (queue)->prev; \
3196 skb != (struct sk_buff *)(queue); \
3199 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3200 for (skb = (queue)->prev, tmp = skb->prev; \
3201 skb != (struct sk_buff *)(queue); \
3202 skb = tmp, tmp = skb->prev)
3204 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3205 for (tmp = skb->prev; \
3206 skb != (struct sk_buff *)(queue); \
3207 skb = tmp, tmp = skb->prev)
3209 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3211 return skb_shinfo(skb)->frag_list != NULL;
3214 static inline void skb_frag_list_init(struct sk_buff *skb)
3216 skb_shinfo(skb)->frag_list = NULL;
3219 #define skb_walk_frags(skb, iter) \
3220 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3223 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3224 const struct sk_buff *skb);
3225 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3226 struct sk_buff_head *queue,
3228 void (*destructor)(struct sock *sk,
3229 struct sk_buff *skb),
3230 int *peeked, int *off, int *err,
3231 struct sk_buff **last);
3232 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3233 void (*destructor)(struct sock *sk,
3234 struct sk_buff *skb),
3235 int *peeked, int *off, int *err,
3236 struct sk_buff **last);
3237 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3238 void (*destructor)(struct sock *sk,
3239 struct sk_buff *skb),
3240 int *peeked, int *off, int *err);
3241 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3243 unsigned int datagram_poll(struct file *file, struct socket *sock,
3244 struct poll_table_struct *wait);
3245 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3246 struct iov_iter *to, int size);
3247 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3248 struct msghdr *msg, int size)
3250 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3252 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3253 struct msghdr *msg);
3254 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3255 struct iov_iter *from, int len);
3256 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3257 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3258 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3259 static inline void skb_free_datagram_locked(struct sock *sk,
3260 struct sk_buff *skb)
3262 __skb_free_datagram_locked(sk, skb, 0);
3264 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3265 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3266 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3267 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3268 int len, __wsum csum);
3269 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3270 struct pipe_inode_info *pipe, unsigned int len,
3271 unsigned int flags);
3272 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3274 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3275 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3276 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3277 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3279 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3280 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3281 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3282 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3283 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3284 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3285 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3286 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3287 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3288 int skb_vlan_pop(struct sk_buff *skb);
3289 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3290 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3293 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3295 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3298 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3300 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3303 struct skb_checksum_ops {
3304 __wsum (*update)(const void *mem, int len, __wsum wsum);
3305 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3308 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3310 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3311 __wsum csum, const struct skb_checksum_ops *ops);
3312 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3315 static inline void * __must_check
3316 __skb_header_pointer(const struct sk_buff *skb, int offset,
3317 int len, void *data, int hlen, void *buffer)
3319 if (hlen - offset >= len)
3320 return data + offset;
3323 skb_copy_bits(skb, offset, buffer, len) < 0)
3329 static inline void * __must_check
3330 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3332 return __skb_header_pointer(skb, offset, len, skb->data,
3333 skb_headlen(skb), buffer);
3337 * skb_needs_linearize - check if we need to linearize a given skb
3338 * depending on the given device features.
3339 * @skb: socket buffer to check
3340 * @features: net device features
3342 * Returns true if either:
3343 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3344 * 2. skb is fragmented and the device does not support SG.
3346 static inline bool skb_needs_linearize(struct sk_buff *skb,
3347 netdev_features_t features)
3349 return skb_is_nonlinear(skb) &&
3350 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3351 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3354 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3356 const unsigned int len)
3358 memcpy(to, skb->data, len);
3361 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3362 const int offset, void *to,
3363 const unsigned int len)
3365 memcpy(to, skb->data + offset, len);
3368 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3370 const unsigned int len)
3372 memcpy(skb->data, from, len);
3375 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3378 const unsigned int len)
3380 memcpy(skb->data + offset, from, len);
3383 void skb_init(void);
3385 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3391 * skb_get_timestamp - get timestamp from a skb
3392 * @skb: skb to get stamp from
3393 * @stamp: pointer to struct timeval to store stamp in
3395 * Timestamps are stored in the skb as offsets to a base timestamp.
3396 * This function converts the offset back to a struct timeval and stores
3399 static inline void skb_get_timestamp(const struct sk_buff *skb,
3400 struct timeval *stamp)
3402 *stamp = ktime_to_timeval(skb->tstamp);
3405 static inline void skb_get_timestampns(const struct sk_buff *skb,
3406 struct timespec *stamp)
3408 *stamp = ktime_to_timespec(skb->tstamp);
3411 static inline void __net_timestamp(struct sk_buff *skb)
3413 skb->tstamp = ktime_get_real();
3416 static inline ktime_t net_timedelta(ktime_t t)
3418 return ktime_sub(ktime_get_real(), t);
3421 static inline ktime_t net_invalid_timestamp(void)
3426 static inline u8 skb_metadata_len(const struct sk_buff *skb)
3428 return skb_shinfo(skb)->meta_len;
3431 static inline void *skb_metadata_end(const struct sk_buff *skb)
3433 return skb_mac_header(skb);
3436 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3437 const struct sk_buff *skb_b,
3440 const void *a = skb_metadata_end(skb_a);
3441 const void *b = skb_metadata_end(skb_b);
3442 /* Using more efficient varaiant than plain call to memcmp(). */
3443 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3447 #define __it(x, op) (x -= sizeof(u##op))
3448 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3449 case 32: diffs |= __it_diff(a, b, 64);
3450 case 24: diffs |= __it_diff(a, b, 64);
3451 case 16: diffs |= __it_diff(a, b, 64);
3452 case 8: diffs |= __it_diff(a, b, 64);
3454 case 28: diffs |= __it_diff(a, b, 64);
3455 case 20: diffs |= __it_diff(a, b, 64);
3456 case 12: diffs |= __it_diff(a, b, 64);
3457 case 4: diffs |= __it_diff(a, b, 32);
3462 return memcmp(a - meta_len, b - meta_len, meta_len);
3466 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3467 const struct sk_buff *skb_b)
3469 u8 len_a = skb_metadata_len(skb_a);
3470 u8 len_b = skb_metadata_len(skb_b);
3472 if (!(len_a | len_b))
3475 return len_a != len_b ?
3476 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3479 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3481 skb_shinfo(skb)->meta_len = meta_len;
3484 static inline void skb_metadata_clear(struct sk_buff *skb)
3486 skb_metadata_set(skb, 0);
3489 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3491 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3493 void skb_clone_tx_timestamp(struct sk_buff *skb);
3494 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3496 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3498 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3502 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3507 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3510 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3512 * PHY drivers may accept clones of transmitted packets for
3513 * timestamping via their phy_driver.txtstamp method. These drivers
3514 * must call this function to return the skb back to the stack with a
3517 * @skb: clone of the the original outgoing packet
3518 * @hwtstamps: hardware time stamps
3521 void skb_complete_tx_timestamp(struct sk_buff *skb,
3522 struct skb_shared_hwtstamps *hwtstamps);
3524 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3525 struct skb_shared_hwtstamps *hwtstamps,
3526 struct sock *sk, int tstype);
3529 * skb_tstamp_tx - queue clone of skb with send time stamps
3530 * @orig_skb: the original outgoing packet
3531 * @hwtstamps: hardware time stamps, may be NULL if not available
3533 * If the skb has a socket associated, then this function clones the
3534 * skb (thus sharing the actual data and optional structures), stores
3535 * the optional hardware time stamping information (if non NULL) or
3536 * generates a software time stamp (otherwise), then queues the clone
3537 * to the error queue of the socket. Errors are silently ignored.
3539 void skb_tstamp_tx(struct sk_buff *orig_skb,
3540 struct skb_shared_hwtstamps *hwtstamps);
3543 * skb_tx_timestamp() - Driver hook for transmit timestamping
3545 * Ethernet MAC Drivers should call this function in their hard_xmit()
3546 * function immediately before giving the sk_buff to the MAC hardware.
3548 * Specifically, one should make absolutely sure that this function is
3549 * called before TX completion of this packet can trigger. Otherwise
3550 * the packet could potentially already be freed.
3552 * @skb: A socket buffer.
3554 static inline void skb_tx_timestamp(struct sk_buff *skb)
3556 skb_clone_tx_timestamp(skb);
3557 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3558 skb_tstamp_tx(skb, NULL);
3562 * skb_complete_wifi_ack - deliver skb with wifi status
3564 * @skb: the original outgoing packet
3565 * @acked: ack status
3568 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3570 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3571 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3573 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3575 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3577 (skb->ip_summed == CHECKSUM_PARTIAL &&
3578 skb_checksum_start_offset(skb) >= 0));
3582 * skb_checksum_complete - Calculate checksum of an entire packet
3583 * @skb: packet to process
3585 * This function calculates the checksum over the entire packet plus
3586 * the value of skb->csum. The latter can be used to supply the
3587 * checksum of a pseudo header as used by TCP/UDP. It returns the
3590 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3591 * this function can be used to verify that checksum on received
3592 * packets. In that case the function should return zero if the
3593 * checksum is correct. In particular, this function will return zero
3594 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3595 * hardware has already verified the correctness of the checksum.
3597 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3599 return skb_csum_unnecessary(skb) ?
3600 0 : __skb_checksum_complete(skb);
3603 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3605 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3606 if (skb->csum_level == 0)
3607 skb->ip_summed = CHECKSUM_NONE;
3613 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3615 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3616 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3618 } else if (skb->ip_summed == CHECKSUM_NONE) {
3619 skb->ip_summed = CHECKSUM_UNNECESSARY;
3620 skb->csum_level = 0;
3624 /* Check if we need to perform checksum complete validation.
3626 * Returns true if checksum complete is needed, false otherwise
3627 * (either checksum is unnecessary or zero checksum is allowed).
3629 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3633 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3634 skb->csum_valid = 1;
3635 __skb_decr_checksum_unnecessary(skb);
3642 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3645 #define CHECKSUM_BREAK 76
3647 /* Unset checksum-complete
3649 * Unset checksum complete can be done when packet is being modified
3650 * (uncompressed for instance) and checksum-complete value is
3653 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3655 if (skb->ip_summed == CHECKSUM_COMPLETE)
3656 skb->ip_summed = CHECKSUM_NONE;
3659 /* Validate (init) checksum based on checksum complete.
3662 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3663 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3664 * checksum is stored in skb->csum for use in __skb_checksum_complete
3665 * non-zero: value of invalid checksum
3668 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3672 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3673 if (!csum_fold(csum_add(psum, skb->csum))) {
3674 skb->csum_valid = 1;
3681 if (complete || skb->len <= CHECKSUM_BREAK) {
3684 csum = __skb_checksum_complete(skb);
3685 skb->csum_valid = !csum;
3692 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3697 /* Perform checksum validate (init). Note that this is a macro since we only
3698 * want to calculate the pseudo header which is an input function if necessary.
3699 * First we try to validate without any computation (checksum unnecessary) and
3700 * then calculate based on checksum complete calling the function to compute
3704 * 0: checksum is validated or try to in skb_checksum_complete
3705 * non-zero: value of invalid checksum
3707 #define __skb_checksum_validate(skb, proto, complete, \
3708 zero_okay, check, compute_pseudo) \
3710 __sum16 __ret = 0; \
3711 skb->csum_valid = 0; \
3712 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3713 __ret = __skb_checksum_validate_complete(skb, \
3714 complete, compute_pseudo(skb, proto)); \
3718 #define skb_checksum_init(skb, proto, compute_pseudo) \
3719 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3721 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3722 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3724 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3725 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3727 #define skb_checksum_validate_zero_check(skb, proto, check, \
3729 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3731 #define skb_checksum_simple_validate(skb) \
3732 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3734 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3736 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3739 static inline void __skb_checksum_convert(struct sk_buff *skb,
3740 __sum16 check, __wsum pseudo)
3742 skb->csum = ~pseudo;
3743 skb->ip_summed = CHECKSUM_COMPLETE;
3746 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3748 if (__skb_checksum_convert_check(skb)) \
3749 __skb_checksum_convert(skb, check, \
3750 compute_pseudo(skb, proto)); \
3753 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3754 u16 start, u16 offset)
3756 skb->ip_summed = CHECKSUM_PARTIAL;
3757 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3758 skb->csum_offset = offset - start;
3761 /* Update skbuf and packet to reflect the remote checksum offload operation.
3762 * When called, ptr indicates the starting point for skb->csum when
3763 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3764 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3766 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3767 int start, int offset, bool nopartial)
3772 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3776 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3777 __skb_checksum_complete(skb);
3778 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3781 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3783 /* Adjust skb->csum since we changed the packet */
3784 skb->csum = csum_add(skb->csum, delta);
3787 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3789 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3790 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3796 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3797 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3798 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3800 if (nfct && atomic_dec_and_test(&nfct->use))
3801 nf_conntrack_destroy(nfct);
3803 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3806 atomic_inc(&nfct->use);
3809 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3810 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3812 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use))
3815 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3818 refcount_inc(&nf_bridge->use);
3820 #endif /* CONFIG_BRIDGE_NETFILTER */
3821 static inline void nf_reset(struct sk_buff *skb)
3823 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3824 nf_conntrack_put(skb_nfct(skb));
3827 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3828 nf_bridge_put(skb->nf_bridge);
3829 skb->nf_bridge = NULL;
3833 static inline void nf_reset_trace(struct sk_buff *skb)
3835 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3840 static inline void ipvs_reset(struct sk_buff *skb)
3842 #if IS_ENABLED(CONFIG_IP_VS)
3843 skb->ipvs_property = 0;
3847 /* Note: This doesn't put any conntrack and bridge info in dst. */
3848 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3851 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3852 dst->_nfct = src->_nfct;
3853 nf_conntrack_get(skb_nfct(src));
3855 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3856 dst->nf_bridge = src->nf_bridge;
3857 nf_bridge_get(src->nf_bridge);
3859 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3861 dst->nf_trace = src->nf_trace;
3865 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3867 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3868 nf_conntrack_put(skb_nfct(dst));
3870 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3871 nf_bridge_put(dst->nf_bridge);
3873 __nf_copy(dst, src, true);
3876 #ifdef CONFIG_NETWORK_SECMARK
3877 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3879 to->secmark = from->secmark;
3882 static inline void skb_init_secmark(struct sk_buff *skb)
3887 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3890 static inline void skb_init_secmark(struct sk_buff *skb)
3894 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3896 return !skb->destructor &&
3897 #if IS_ENABLED(CONFIG_XFRM)
3901 !skb->_skb_refdst &&
3902 !skb_has_frag_list(skb);
3905 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3907 skb->queue_mapping = queue_mapping;
3910 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3912 return skb->queue_mapping;
3915 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3917 to->queue_mapping = from->queue_mapping;
3920 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3922 skb->queue_mapping = rx_queue + 1;
3925 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3927 return skb->queue_mapping - 1;
3930 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3932 return skb->queue_mapping != 0;
3935 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3937 skb->dst_pending_confirm = val;
3940 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3942 return skb->dst_pending_confirm != 0;
3945 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3954 /* Keeps track of mac header offset relative to skb->head.
3955 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3956 * For non-tunnel skb it points to skb_mac_header() and for
3957 * tunnel skb it points to outer mac header.
3958 * Keeps track of level of encapsulation of network headers.
3969 #define SKB_SGO_CB_OFFSET 32
3970 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3972 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3974 return (skb_mac_header(inner_skb) - inner_skb->head) -
3975 SKB_GSO_CB(inner_skb)->mac_offset;
3978 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3980 int new_headroom, headroom;
3983 headroom = skb_headroom(skb);
3984 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3988 new_headroom = skb_headroom(skb);
3989 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3993 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3995 /* Do not update partial checksums if remote checksum is enabled. */
3996 if (skb->remcsum_offload)
3999 SKB_GSO_CB(skb)->csum = res;
4000 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4003 /* Compute the checksum for a gso segment. First compute the checksum value
4004 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4005 * then add in skb->csum (checksum from csum_start to end of packet).
4006 * skb->csum and csum_start are then updated to reflect the checksum of the
4007 * resultant packet starting from the transport header-- the resultant checksum
4008 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4011 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4013 unsigned char *csum_start = skb_transport_header(skb);
4014 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4015 __wsum partial = SKB_GSO_CB(skb)->csum;
4017 SKB_GSO_CB(skb)->csum = res;
4018 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4020 return csum_fold(csum_partial(csum_start, plen, partial));
4023 static inline bool skb_is_gso(const struct sk_buff *skb)
4025 return skb_shinfo(skb)->gso_size;
4028 /* Note: Should be called only if skb_is_gso(skb) is true */
4029 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4031 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4034 static inline void skb_gso_reset(struct sk_buff *skb)
4036 skb_shinfo(skb)->gso_size = 0;
4037 skb_shinfo(skb)->gso_segs = 0;
4038 skb_shinfo(skb)->gso_type = 0;
4041 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4043 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4045 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4046 * wanted then gso_type will be set. */
4047 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4049 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4050 unlikely(shinfo->gso_type == 0)) {
4051 __skb_warn_lro_forwarding(skb);
4057 static inline void skb_forward_csum(struct sk_buff *skb)
4059 /* Unfortunately we don't support this one. Any brave souls? */
4060 if (skb->ip_summed == CHECKSUM_COMPLETE)
4061 skb->ip_summed = CHECKSUM_NONE;
4065 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4066 * @skb: skb to check
4068 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4069 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4070 * use this helper, to document places where we make this assertion.
4072 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4075 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4079 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4081 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4082 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4083 unsigned int transport_len,
4084 __sum16(*skb_chkf)(struct sk_buff *skb));
4087 * skb_head_is_locked - Determine if the skb->head is locked down
4088 * @skb: skb to check
4090 * The head on skbs build around a head frag can be removed if they are
4091 * not cloned. This function returns true if the skb head is locked down
4092 * due to either being allocated via kmalloc, or by being a clone with
4093 * multiple references to the head.
4095 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4097 return !skb->head_frag || skb_cloned(skb);
4101 * skb_gso_network_seglen - Return length of individual segments of a gso packet
4105 * skb_gso_network_seglen is used to determine the real size of the
4106 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
4108 * The MAC/L2 header is not accounted for.
4110 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
4112 unsigned int hdr_len = skb_transport_header(skb) -
4113 skb_network_header(skb);
4114 return hdr_len + skb_gso_transport_seglen(skb);
4117 /* Local Checksum Offload.
4118 * Compute outer checksum based on the assumption that the
4119 * inner checksum will be offloaded later.
4120 * See Documentation/networking/checksum-offloads.txt for
4121 * explanation of how this works.
4122 * Fill in outer checksum adjustment (e.g. with sum of outer
4123 * pseudo-header) before calling.
4124 * Also ensure that inner checksum is in linear data area.
4126 static inline __wsum lco_csum(struct sk_buff *skb)
4128 unsigned char *csum_start = skb_checksum_start(skb);
4129 unsigned char *l4_hdr = skb_transport_header(skb);
4132 /* Start with complement of inner checksum adjustment */
4133 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4136 /* Add in checksum of our headers (incl. outer checksum
4137 * adjustment filled in by caller) and return result.
4139 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4142 #endif /* __KERNEL__ */
4143 #endif /* _LINUX_SKBUFF_H */