2 * Definitions for the 'struct sk_buff' memory handlers.
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
17 #include <linux/kernel.h>
18 #include <linux/kmemcheck.h>
19 #include <linux/compiler.h>
20 #include <linux/time.h>
21 #include <linux/bug.h>
22 #include <linux/cache.h>
23 #include <linux/rbtree.h>
24 #include <linux/socket.h>
25 #include <linux/refcount.h>
27 #include <linux/atomic.h>
28 #include <asm/types.h>
29 #include <linux/spinlock.h>
30 #include <linux/net.h>
31 #include <linux/textsearch.h>
32 #include <net/checksum.h>
33 #include <linux/rcupdate.h>
34 #include <linux/hrtimer.h>
35 #include <linux/dma-mapping.h>
36 #include <linux/netdev_features.h>
37 #include <linux/sched.h>
38 #include <linux/sched/clock.h>
39 #include <net/flow_dissector.h>
40 #include <linux/splice.h>
41 #include <linux/in6.h>
42 #include <linux/if_packet.h>
45 /* The interface for checksum offload between the stack and networking drivers
48 * A. IP checksum related features
50 * Drivers advertise checksum offload capabilities in the features of a device.
51 * From the stack's point of view these are capabilities offered by the driver,
52 * a driver typically only advertises features that it is capable of offloading
55 * The checksum related features are:
57 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
58 * IP (one's complement) checksum for any combination
59 * of protocols or protocol layering. The checksum is
60 * computed and set in a packet per the CHECKSUM_PARTIAL
61 * interface (see below).
63 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
64 * TCP or UDP packets over IPv4. These are specifically
65 * unencapsulated packets of the form IPv4|TCP or
66 * IPv4|UDP where the Protocol field in the IPv4 header
67 * is TCP or UDP. The IPv4 header may contain IP options
68 * This feature cannot be set in features for a device
69 * with NETIF_F_HW_CSUM also set. This feature is being
70 * DEPRECATED (see below).
72 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
73 * TCP or UDP packets over IPv6. These are specifically
74 * unencapsulated packets of the form IPv6|TCP or
75 * IPv4|UDP where the Next Header field in the IPv6
76 * header is either TCP or UDP. IPv6 extension headers
77 * are not supported with this feature. This feature
78 * cannot be set in features for a device with
79 * NETIF_F_HW_CSUM also set. This feature is being
80 * DEPRECATED (see below).
82 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
83 * This flag is used only used to disable the RX checksum
84 * feature for a device. The stack will accept receive
85 * checksum indication in packets received on a device
86 * regardless of whether NETIF_F_RXCSUM is set.
88 * B. Checksumming of received packets by device. Indication of checksum
89 * verification is in set skb->ip_summed. Possible values are:
93 * Device did not checksum this packet e.g. due to lack of capabilities.
94 * The packet contains full (though not verified) checksum in packet but
95 * not in skb->csum. Thus, skb->csum is undefined in this case.
97 * CHECKSUM_UNNECESSARY:
99 * The hardware you're dealing with doesn't calculate the full checksum
100 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
101 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
102 * if their checksums are okay. skb->csum is still undefined in this case
103 * though. A driver or device must never modify the checksum field in the
104 * packet even if checksum is verified.
106 * CHECKSUM_UNNECESSARY is applicable to following protocols:
107 * TCP: IPv6 and IPv4.
108 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
109 * zero UDP checksum for either IPv4 or IPv6, the networking stack
110 * may perform further validation in this case.
111 * GRE: only if the checksum is present in the header.
112 * SCTP: indicates the CRC in SCTP header has been validated.
113 * FCOE: indicates the CRC in FC frame has been validated.
115 * skb->csum_level indicates the number of consecutive checksums found in
116 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
117 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
118 * and a device is able to verify the checksums for UDP (possibly zero),
119 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
120 * two. If the device were only able to verify the UDP checksum and not
121 * GRE, either because it doesn't support GRE checksum of because GRE
122 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
123 * not considered in this case).
127 * This is the most generic way. The device supplied checksum of the _whole_
128 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
129 * hardware doesn't need to parse L3/L4 headers to implement this.
132 * - Even if device supports only some protocols, but is able to produce
133 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
134 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
138 * A checksum is set up to be offloaded to a device as described in the
139 * output description for CHECKSUM_PARTIAL. This may occur on a packet
140 * received directly from another Linux OS, e.g., a virtualized Linux kernel
141 * on the same host, or it may be set in the input path in GRO or remote
142 * checksum offload. For the purposes of checksum verification, the checksum
143 * referred to by skb->csum_start + skb->csum_offset and any preceding
144 * checksums in the packet are considered verified. Any checksums in the
145 * packet that are after the checksum being offloaded are not considered to
148 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
149 * in the skb->ip_summed for a packet. Values are:
153 * The driver is required to checksum the packet as seen by hard_start_xmit()
154 * from skb->csum_start up to the end, and to record/write the checksum at
155 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
156 * csum_start and csum_offset values are valid values given the length and
157 * offset of the packet, however they should not attempt to validate that the
158 * checksum refers to a legitimate transport layer checksum-- it is the
159 * purview of the stack to validate that csum_start and csum_offset are set
162 * When the stack requests checksum offload for a packet, the driver MUST
163 * ensure that the checksum is set correctly. A driver can either offload the
164 * checksum calculation to the device, or call skb_checksum_help (in the case
165 * that the device does not support offload for a particular checksum).
167 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
168 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
169 * checksum offload capability.
170 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
171 * on network device checksumming capabilities: if a packet does not match
172 * them, skb_checksum_help or skb_crc32c_help (depending on the value of
173 * csum_not_inet, see item D.) is called to resolve the checksum.
177 * The skb was already checksummed by the protocol, or a checksum is not
180 * CHECKSUM_UNNECESSARY:
182 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
186 * Not used in checksum output. If a driver observes a packet with this value
187 * set in skbuff, if should treat as CHECKSUM_NONE being set.
189 * D. Non-IP checksum (CRC) offloads
191 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
192 * offloading the SCTP CRC in a packet. To perform this offload the stack
193 * will set set csum_start and csum_offset accordingly, set ip_summed to
194 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
195 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
196 * A driver that supports both IP checksum offload and SCTP CRC32c offload
197 * must verify which offload is configured for a packet by testing the
198 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
199 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
201 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
202 * offloading the FCOE CRC in a packet. To perform this offload the stack
203 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
204 * accordingly. Note the there is no indication in the skbuff that the
205 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
206 * both IP checksum offload and FCOE CRC offload must verify which offload
207 * is configured for a packet presumably by inspecting packet headers.
209 * E. Checksumming on output with GSO.
211 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
212 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
213 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
214 * part of the GSO operation is implied. If a checksum is being offloaded
215 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
216 * are set to refer to the outermost checksum being offload (two offloaded
217 * checksums are possible with UDP encapsulation).
220 /* Don't change this without changing skb_csum_unnecessary! */
221 #define CHECKSUM_NONE 0
222 #define CHECKSUM_UNNECESSARY 1
223 #define CHECKSUM_COMPLETE 2
224 #define CHECKSUM_PARTIAL 3
226 /* Maximum value in skb->csum_level */
227 #define SKB_MAX_CSUM_LEVEL 3
229 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
230 #define SKB_WITH_OVERHEAD(X) \
231 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
232 #define SKB_MAX_ORDER(X, ORDER) \
233 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
234 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
235 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
237 /* return minimum truesize of one skb containing X bytes of data */
238 #define SKB_TRUESIZE(X) ((X) + \
239 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
240 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
244 struct pipe_inode_info;
248 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
249 struct nf_conntrack {
254 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
255 struct nf_bridge_info {
258 BRNF_PROTO_UNCHANGED,
266 struct net_device *physindev;
268 /* always valid & non-NULL from FORWARD on, for physdev match */
269 struct net_device *physoutdev;
271 /* prerouting: detect dnat in orig/reply direction */
273 struct in6_addr ipv6_daddr;
275 /* after prerouting + nat detected: store original source
276 * mac since neigh resolution overwrites it, only used while
277 * skb is out in neigh layer.
279 char neigh_header[8];
284 struct sk_buff_head {
285 /* These two members must be first. */
286 struct sk_buff *next;
287 struct sk_buff *prev;
295 /* To allow 64K frame to be packed as single skb without frag_list we
296 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
297 * buffers which do not start on a page boundary.
299 * Since GRO uses frags we allocate at least 16 regardless of page
302 #if (65536/PAGE_SIZE + 1) < 16
303 #define MAX_SKB_FRAGS 16UL
305 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
307 extern int sysctl_max_skb_frags;
309 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
310 * segment using its current segmentation instead.
312 #define GSO_BY_FRAGS 0xFFFF
314 typedef struct skb_frag_struct skb_frag_t;
316 struct skb_frag_struct {
320 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
329 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
334 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
339 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
344 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
349 static inline bool skb_frag_must_loop(struct page *p)
351 #if defined(CONFIG_HIGHMEM)
359 * skb_frag_foreach_page - loop over pages in a fragment
361 * @f: skb frag to operate on
362 * @f_off: offset from start of f->page.p
363 * @f_len: length from f_off to loop over
364 * @p: (temp var) current page
365 * @p_off: (temp var) offset from start of current page,
366 * non-zero only on first page.
367 * @p_len: (temp var) length in current page,
368 * < PAGE_SIZE only on first and last page.
369 * @copied: (temp var) length so far, excluding current p_len.
371 * A fragment can hold a compound page, in which case per-page
372 * operations, notably kmap_atomic, must be called for each
375 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
376 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
377 p_off = (f_off) & (PAGE_SIZE - 1), \
378 p_len = skb_frag_must_loop(p) ? \
379 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
382 copied += p_len, p++, p_off = 0, \
383 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
385 #define HAVE_HW_TIME_STAMP
388 * struct skb_shared_hwtstamps - hardware time stamps
389 * @hwtstamp: hardware time stamp transformed into duration
390 * since arbitrary point in time
392 * Software time stamps generated by ktime_get_real() are stored in
395 * hwtstamps can only be compared against other hwtstamps from
398 * This structure is attached to packets as part of the
399 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
401 struct skb_shared_hwtstamps {
405 /* Definitions for tx_flags in struct skb_shared_info */
407 /* generate hardware time stamp */
408 SKBTX_HW_TSTAMP = 1 << 0,
410 /* generate software time stamp when queueing packet to NIC */
411 SKBTX_SW_TSTAMP = 1 << 1,
413 /* device driver is going to provide hardware time stamp */
414 SKBTX_IN_PROGRESS = 1 << 2,
416 /* device driver supports TX zero-copy buffers */
417 SKBTX_DEV_ZEROCOPY = 1 << 3,
419 /* generate wifi status information (where possible) */
420 SKBTX_WIFI_STATUS = 1 << 4,
422 /* This indicates at least one fragment might be overwritten
423 * (as in vmsplice(), sendfile() ...)
424 * If we need to compute a TX checksum, we'll need to copy
425 * all frags to avoid possible bad checksum
427 SKBTX_SHARED_FRAG = 1 << 5,
429 /* generate software time stamp when entering packet scheduling */
430 SKBTX_SCHED_TSTAMP = 1 << 6,
433 #define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
434 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
436 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
439 * The callback notifies userspace to release buffers when skb DMA is done in
440 * lower device, the skb last reference should be 0 when calling this.
441 * The zerocopy_success argument is true if zero copy transmit occurred,
442 * false on data copy or out of memory error caused by data copy attempt.
443 * The ctx field is used to track device context.
444 * The desc field is used to track userspace buffer index.
447 void (*callback)(struct ubuf_info *, bool zerocopy_success);
463 struct user_struct *user;
468 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
470 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
471 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
472 struct ubuf_info *uarg);
474 static inline void sock_zerocopy_get(struct ubuf_info *uarg)
476 refcount_inc(&uarg->refcnt);
479 void sock_zerocopy_put(struct ubuf_info *uarg);
480 void sock_zerocopy_put_abort(struct ubuf_info *uarg);
482 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
484 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
485 struct msghdr *msg, int len,
486 struct ubuf_info *uarg);
488 /* This data is invariant across clones and lives at
489 * the end of the header data, ie. at skb->end.
491 struct skb_shared_info {
496 unsigned short gso_size;
497 /* Warning: this field is not always filled in (UFO)! */
498 unsigned short gso_segs;
499 struct sk_buff *frag_list;
500 struct skb_shared_hwtstamps hwtstamps;
501 unsigned int gso_type;
505 * Warning : all fields before dataref are cleared in __alloc_skb()
509 /* Intermediate layers must ensure that destructor_arg
510 * remains valid until skb destructor */
511 void * destructor_arg;
513 /* must be last field, see pskb_expand_head() */
514 skb_frag_t frags[MAX_SKB_FRAGS];
517 /* We divide dataref into two halves. The higher 16 bits hold references
518 * to the payload part of skb->data. The lower 16 bits hold references to
519 * the entire skb->data. A clone of a headerless skb holds the length of
520 * the header in skb->hdr_len.
522 * All users must obey the rule that the skb->data reference count must be
523 * greater than or equal to the payload reference count.
525 * Holding a reference to the payload part means that the user does not
526 * care about modifications to the header part of skb->data.
528 #define SKB_DATAREF_SHIFT 16
529 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
533 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
534 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
535 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
539 SKB_GSO_TCPV4 = 1 << 0,
541 /* This indicates the skb is from an untrusted source. */
542 SKB_GSO_DODGY = 1 << 1,
544 /* This indicates the tcp segment has CWR set. */
545 SKB_GSO_TCP_ECN = 1 << 2,
547 SKB_GSO_TCP_FIXEDID = 1 << 3,
549 SKB_GSO_TCPV6 = 1 << 4,
551 SKB_GSO_FCOE = 1 << 5,
553 SKB_GSO_GRE = 1 << 6,
555 SKB_GSO_GRE_CSUM = 1 << 7,
557 SKB_GSO_IPXIP4 = 1 << 8,
559 SKB_GSO_IPXIP6 = 1 << 9,
561 SKB_GSO_UDP_TUNNEL = 1 << 10,
563 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
565 SKB_GSO_PARTIAL = 1 << 12,
567 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
569 SKB_GSO_SCTP = 1 << 14,
571 SKB_GSO_ESP = 1 << 15,
574 #if BITS_PER_LONG > 32
575 #define NET_SKBUFF_DATA_USES_OFFSET 1
578 #ifdef NET_SKBUFF_DATA_USES_OFFSET
579 typedef unsigned int sk_buff_data_t;
581 typedef unsigned char *sk_buff_data_t;
585 * struct sk_buff - socket buffer
586 * @next: Next buffer in list
587 * @prev: Previous buffer in list
588 * @tstamp: Time we arrived/left
589 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
590 * @sk: Socket we are owned by
591 * @dev: Device we arrived on/are leaving by
592 * @cb: Control buffer. Free for use by every layer. Put private vars here
593 * @_skb_refdst: destination entry (with norefcount bit)
594 * @sp: the security path, used for xfrm
595 * @len: Length of actual data
596 * @data_len: Data length
597 * @mac_len: Length of link layer header
598 * @hdr_len: writable header length of cloned skb
599 * @csum: Checksum (must include start/offset pair)
600 * @csum_start: Offset from skb->head where checksumming should start
601 * @csum_offset: Offset from csum_start where checksum should be stored
602 * @priority: Packet queueing priority
603 * @ignore_df: allow local fragmentation
604 * @cloned: Head may be cloned (check refcnt to be sure)
605 * @ip_summed: Driver fed us an IP checksum
606 * @nohdr: Payload reference only, must not modify header
607 * @pkt_type: Packet class
608 * @fclone: skbuff clone status
609 * @ipvs_property: skbuff is owned by ipvs
610 * @tc_skip_classify: do not classify packet. set by IFB device
611 * @tc_at_ingress: used within tc_classify to distinguish in/egress
612 * @tc_redirected: packet was redirected by a tc action
613 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
614 * @peeked: this packet has been seen already, so stats have been
615 * done for it, don't do them again
616 * @nf_trace: netfilter packet trace flag
617 * @protocol: Packet protocol from driver
618 * @destructor: Destruct function
619 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
620 * @_nfct: Associated connection, if any (with nfctinfo bits)
621 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
622 * @skb_iif: ifindex of device we arrived on
623 * @tc_index: Traffic control index
624 * @hash: the packet hash
625 * @queue_mapping: Queue mapping for multiqueue devices
626 * @xmit_more: More SKBs are pending for this queue
627 * @ndisc_nodetype: router type (from link layer)
628 * @ooo_okay: allow the mapping of a socket to a queue to be changed
629 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
631 * @sw_hash: indicates hash was computed in software stack
632 * @wifi_acked_valid: wifi_acked was set
633 * @wifi_acked: whether frame was acked on wifi or not
634 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
635 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
636 * @dst_pending_confirm: need to confirm neighbour
637 * @napi_id: id of the NAPI struct this skb came from
638 * @secmark: security marking
639 * @mark: Generic packet mark
640 * @vlan_proto: vlan encapsulation protocol
641 * @vlan_tci: vlan tag control information
642 * @inner_protocol: Protocol (encapsulation)
643 * @inner_transport_header: Inner transport layer header (encapsulation)
644 * @inner_network_header: Network layer header (encapsulation)
645 * @inner_mac_header: Link layer header (encapsulation)
646 * @transport_header: Transport layer header
647 * @network_header: Network layer header
648 * @mac_header: Link layer header
649 * @tail: Tail pointer
651 * @head: Head of buffer
652 * @data: Data head pointer
653 * @truesize: Buffer size
654 * @users: User count - see {datagram,tcp}.c
660 /* These two members must be first. */
661 struct sk_buff *next;
662 struct sk_buff *prev;
665 struct net_device *dev;
666 /* Some protocols might use this space to store information,
667 * while device pointer would be NULL.
668 * UDP receive path is one user.
670 unsigned long dev_scratch;
673 struct rb_node rbnode; /* used in netem & tcp stack */
682 * This is the control buffer. It is free to use for every
683 * layer. Please put your private variables there. If you
684 * want to keep them across layers you have to do a skb_clone()
685 * first. This is owned by whoever has the skb queued ATM.
687 char cb[48] __aligned(8);
691 unsigned long _skb_refdst;
692 void (*destructor)(struct sk_buff *skb);
694 struct list_head tcp_tsorted_anchor;
700 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
703 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
704 struct nf_bridge_info *nf_bridge;
711 /* Following fields are _not_ copied in __copy_skb_header()
712 * Note that queue_mapping is here mostly to fill a hole.
714 kmemcheck_bitfield_begin(flags1);
717 /* if you move cloned around you also must adapt those constants */
718 #ifdef __BIG_ENDIAN_BITFIELD
719 #define CLONED_MASK (1 << 7)
721 #define CLONED_MASK 1
723 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
725 __u8 __cloned_offset[0];
732 __unused:1; /* one bit hole */
733 kmemcheck_bitfield_end(flags1);
735 /* fields enclosed in headers_start/headers_end are copied
736 * using a single memcpy() in __copy_skb_header()
739 __u32 headers_start[0];
742 /* if you move pkt_type around you also must adapt those constants */
743 #ifdef __BIG_ENDIAN_BITFIELD
744 #define PKT_TYPE_MAX (7 << 5)
746 #define PKT_TYPE_MAX 7
748 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
750 __u8 __pkt_type_offset[0];
760 __u8 wifi_acked_valid:1;
764 /* Indicates the inner headers are valid in the skbuff. */
765 __u8 encapsulation:1;
766 __u8 encap_hdr_csum:1;
768 __u8 csum_complete_sw:1;
770 __u8 csum_not_inet:1;
772 __u8 dst_pending_confirm:1;
773 #ifdef CONFIG_IPV6_NDISC_NODETYPE
774 __u8 ndisc_nodetype:2;
776 __u8 ipvs_property:1;
777 __u8 inner_protocol_type:1;
778 __u8 remcsum_offload:1;
779 #ifdef CONFIG_NET_SWITCHDEV
780 __u8 offload_fwd_mark:1;
781 __u8 offload_mr_fwd_mark:1;
783 #ifdef CONFIG_NET_CLS_ACT
784 __u8 tc_skip_classify:1;
785 __u8 tc_at_ingress:1;
786 __u8 tc_redirected:1;
787 __u8 tc_from_ingress:1;
790 #ifdef CONFIG_NET_SCHED
791 __u16 tc_index; /* traffic control index */
806 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
808 unsigned int napi_id;
809 unsigned int sender_cpu;
812 #ifdef CONFIG_NETWORK_SECMARK
818 __u32 reserved_tailroom;
822 __be16 inner_protocol;
826 __u16 inner_transport_header;
827 __u16 inner_network_header;
828 __u16 inner_mac_header;
831 __u16 transport_header;
832 __u16 network_header;
836 __u32 headers_end[0];
839 /* These elements must be at the end, see alloc_skb() for details. */
844 unsigned int truesize;
850 * Handling routines are only of interest to the kernel
852 #include <linux/slab.h>
855 #define SKB_ALLOC_FCLONE 0x01
856 #define SKB_ALLOC_RX 0x02
857 #define SKB_ALLOC_NAPI 0x04
859 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
860 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
862 return unlikely(skb->pfmemalloc);
866 * skb might have a dst pointer attached, refcounted or not.
867 * _skb_refdst low order bit is set if refcount was _not_ taken
869 #define SKB_DST_NOREF 1UL
870 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
872 #define SKB_NFCT_PTRMASK ~(7UL)
874 * skb_dst - returns skb dst_entry
877 * Returns skb dst_entry, regardless of reference taken or not.
879 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
881 /* If refdst was not refcounted, check we still are in a
882 * rcu_read_lock section
884 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
885 !rcu_read_lock_held() &&
886 !rcu_read_lock_bh_held());
887 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
891 * skb_dst_set - sets skb dst
895 * Sets skb dst, assuming a reference was taken on dst and should
896 * be released by skb_dst_drop()
898 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
900 skb->_skb_refdst = (unsigned long)dst;
904 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
908 * Sets skb dst, assuming a reference was not taken on dst.
909 * If dst entry is cached, we do not take reference and dst_release
910 * will be avoided by refdst_drop. If dst entry is not cached, we take
911 * reference, so that last dst_release can destroy the dst immediately.
913 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
915 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
916 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
920 * skb_dst_is_noref - Test if skb dst isn't refcounted
923 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
925 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
928 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
930 return (struct rtable *)skb_dst(skb);
933 /* For mangling skb->pkt_type from user space side from applications
934 * such as nft, tc, etc, we only allow a conservative subset of
935 * possible pkt_types to be set.
937 static inline bool skb_pkt_type_ok(u32 ptype)
939 return ptype <= PACKET_OTHERHOST;
942 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
944 #ifdef CONFIG_NET_RX_BUSY_POLL
951 /* decrement the reference count and return true if we can free the skb */
952 static inline bool skb_unref(struct sk_buff *skb)
956 if (likely(refcount_read(&skb->users) == 1))
958 else if (likely(!refcount_dec_and_test(&skb->users)))
964 void skb_release_head_state(struct sk_buff *skb);
965 void kfree_skb(struct sk_buff *skb);
966 void kfree_skb_list(struct sk_buff *segs);
967 void skb_tx_error(struct sk_buff *skb);
968 void consume_skb(struct sk_buff *skb);
969 void __consume_stateless_skb(struct sk_buff *skb);
970 void __kfree_skb(struct sk_buff *skb);
971 extern struct kmem_cache *skbuff_head_cache;
973 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
974 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
975 bool *fragstolen, int *delta_truesize);
977 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
979 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
980 struct sk_buff *build_skb(void *data, unsigned int frag_size);
981 static inline struct sk_buff *alloc_skb(unsigned int size,
984 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
987 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
988 unsigned long data_len,
993 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
994 struct sk_buff_fclones {
999 refcount_t fclone_ref;
1003 * skb_fclone_busy - check if fclone is busy
1007 * Returns true if skb is a fast clone, and its clone is not freed.
1008 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1009 * so we also check that this didnt happen.
1011 static inline bool skb_fclone_busy(const struct sock *sk,
1012 const struct sk_buff *skb)
1014 const struct sk_buff_fclones *fclones;
1016 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1018 return skb->fclone == SKB_FCLONE_ORIG &&
1019 refcount_read(&fclones->fclone_ref) > 1 &&
1020 fclones->skb2.sk == sk;
1023 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1026 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1029 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1030 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1031 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1032 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1033 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1034 gfp_t gfp_mask, bool fclone);
1035 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1038 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1041 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1042 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1043 unsigned int headroom);
1044 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1045 int newtailroom, gfp_t priority);
1046 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1047 int offset, int len);
1048 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1049 int offset, int len);
1050 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1051 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1054 * skb_pad - zero pad the tail of an skb
1055 * @skb: buffer to pad
1056 * @pad: space to pad
1058 * Ensure that a buffer is followed by a padding area that is zero
1059 * filled. Used by network drivers which may DMA or transfer data
1060 * beyond the buffer end onto the wire.
1062 * May return error in out of memory cases. The skb is freed on error.
1064 static inline int skb_pad(struct sk_buff *skb, int pad)
1066 return __skb_pad(skb, pad, true);
1068 #define dev_kfree_skb(a) consume_skb(a)
1070 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1071 int getfrag(void *from, char *to, int offset,
1072 int len, int odd, struct sk_buff *skb),
1073 void *from, int length);
1075 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1076 int offset, size_t size);
1078 struct skb_seq_state {
1082 __u32 stepped_offset;
1083 struct sk_buff *root_skb;
1084 struct sk_buff *cur_skb;
1088 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1089 unsigned int to, struct skb_seq_state *st);
1090 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1091 struct skb_seq_state *st);
1092 void skb_abort_seq_read(struct skb_seq_state *st);
1094 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1095 unsigned int to, struct ts_config *config);
1098 * Packet hash types specify the type of hash in skb_set_hash.
1100 * Hash types refer to the protocol layer addresses which are used to
1101 * construct a packet's hash. The hashes are used to differentiate or identify
1102 * flows of the protocol layer for the hash type. Hash types are either
1103 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1105 * Properties of hashes:
1107 * 1) Two packets in different flows have different hash values
1108 * 2) Two packets in the same flow should have the same hash value
1110 * A hash at a higher layer is considered to be more specific. A driver should
1111 * set the most specific hash possible.
1113 * A driver cannot indicate a more specific hash than the layer at which a hash
1114 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1116 * A driver may indicate a hash level which is less specific than the
1117 * actual layer the hash was computed on. For instance, a hash computed
1118 * at L4 may be considered an L3 hash. This should only be done if the
1119 * driver can't unambiguously determine that the HW computed the hash at
1120 * the higher layer. Note that the "should" in the second property above
1123 enum pkt_hash_types {
1124 PKT_HASH_TYPE_NONE, /* Undefined type */
1125 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1126 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1127 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1130 static inline void skb_clear_hash(struct sk_buff *skb)
1137 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1140 skb_clear_hash(skb);
1144 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1146 skb->l4_hash = is_l4;
1147 skb->sw_hash = is_sw;
1152 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1154 /* Used by drivers to set hash from HW */
1155 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1159 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1161 __skb_set_hash(skb, hash, true, is_l4);
1164 void __skb_get_hash(struct sk_buff *skb);
1165 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1166 u32 skb_get_poff(const struct sk_buff *skb);
1167 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1168 const struct flow_keys *keys, int hlen);
1169 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1170 void *data, int hlen_proto);
1172 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1173 int thoff, u8 ip_proto)
1175 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1178 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1179 const struct flow_dissector_key *key,
1180 unsigned int key_count);
1182 bool __skb_flow_dissect(const struct sk_buff *skb,
1183 struct flow_dissector *flow_dissector,
1184 void *target_container,
1185 void *data, __be16 proto, int nhoff, int hlen,
1186 unsigned int flags);
1188 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1189 struct flow_dissector *flow_dissector,
1190 void *target_container, unsigned int flags)
1192 return __skb_flow_dissect(skb, flow_dissector, target_container,
1193 NULL, 0, 0, 0, flags);
1196 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1197 struct flow_keys *flow,
1200 memset(flow, 0, sizeof(*flow));
1201 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1202 NULL, 0, 0, 0, flags);
1205 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1206 void *data, __be16 proto,
1207 int nhoff, int hlen,
1210 memset(flow, 0, sizeof(*flow));
1211 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1212 data, proto, nhoff, hlen, flags);
1215 static inline __u32 skb_get_hash(struct sk_buff *skb)
1217 if (!skb->l4_hash && !skb->sw_hash)
1218 __skb_get_hash(skb);
1223 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1225 if (!skb->l4_hash && !skb->sw_hash) {
1226 struct flow_keys keys;
1227 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1229 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1235 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1237 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1242 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1244 to->hash = from->hash;
1245 to->sw_hash = from->sw_hash;
1246 to->l4_hash = from->l4_hash;
1249 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1250 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1252 return skb->head + skb->end;
1255 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1260 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1265 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1267 return skb->end - skb->head;
1272 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1274 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1276 return &skb_shinfo(skb)->hwtstamps;
1279 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1281 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1283 return is_zcopy ? skb_uarg(skb) : NULL;
1286 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg)
1288 if (skb && uarg && !skb_zcopy(skb)) {
1289 sock_zerocopy_get(uarg);
1290 skb_shinfo(skb)->destructor_arg = uarg;
1291 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1295 /* Release a reference on a zerocopy structure */
1296 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1298 struct ubuf_info *uarg = skb_zcopy(skb);
1301 if (uarg->callback == sock_zerocopy_callback) {
1302 uarg->zerocopy = uarg->zerocopy && zerocopy;
1303 sock_zerocopy_put(uarg);
1305 uarg->callback(uarg, zerocopy);
1308 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1312 /* Abort a zerocopy operation and revert zckey on error in send syscall */
1313 static inline void skb_zcopy_abort(struct sk_buff *skb)
1315 struct ubuf_info *uarg = skb_zcopy(skb);
1318 sock_zerocopy_put_abort(uarg);
1319 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1324 * skb_queue_empty - check if a queue is empty
1327 * Returns true if the queue is empty, false otherwise.
1329 static inline int skb_queue_empty(const struct sk_buff_head *list)
1331 return list->next == (const struct sk_buff *) list;
1335 * skb_queue_is_last - check if skb is the last entry in the queue
1339 * Returns true if @skb is the last buffer on the list.
1341 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1342 const struct sk_buff *skb)
1344 return skb->next == (const struct sk_buff *) list;
1348 * skb_queue_is_first - check if skb is the first entry in the queue
1352 * Returns true if @skb is the first buffer on the list.
1354 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1355 const struct sk_buff *skb)
1357 return skb->prev == (const struct sk_buff *) list;
1361 * skb_queue_next - return the next packet in the queue
1363 * @skb: current buffer
1365 * Return the next packet in @list after @skb. It is only valid to
1366 * call this if skb_queue_is_last() evaluates to false.
1368 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1369 const struct sk_buff *skb)
1371 /* This BUG_ON may seem severe, but if we just return then we
1372 * are going to dereference garbage.
1374 BUG_ON(skb_queue_is_last(list, skb));
1379 * skb_queue_prev - return the prev packet in the queue
1381 * @skb: current buffer
1383 * Return the prev packet in @list before @skb. It is only valid to
1384 * call this if skb_queue_is_first() evaluates to false.
1386 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1387 const struct sk_buff *skb)
1389 /* This BUG_ON may seem severe, but if we just return then we
1390 * are going to dereference garbage.
1392 BUG_ON(skb_queue_is_first(list, skb));
1397 * skb_get - reference buffer
1398 * @skb: buffer to reference
1400 * Makes another reference to a socket buffer and returns a pointer
1403 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1405 refcount_inc(&skb->users);
1410 * If users == 1, we are the only owner and are can avoid redundant
1415 * skb_cloned - is the buffer a clone
1416 * @skb: buffer to check
1418 * Returns true if the buffer was generated with skb_clone() and is
1419 * one of multiple shared copies of the buffer. Cloned buffers are
1420 * shared data so must not be written to under normal circumstances.
1422 static inline int skb_cloned(const struct sk_buff *skb)
1424 return skb->cloned &&
1425 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1428 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1430 might_sleep_if(gfpflags_allow_blocking(pri));
1432 if (skb_cloned(skb))
1433 return pskb_expand_head(skb, 0, 0, pri);
1439 * skb_header_cloned - is the header a clone
1440 * @skb: buffer to check
1442 * Returns true if modifying the header part of the buffer requires
1443 * the data to be copied.
1445 static inline int skb_header_cloned(const struct sk_buff *skb)
1452 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1453 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1454 return dataref != 1;
1457 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1459 might_sleep_if(gfpflags_allow_blocking(pri));
1461 if (skb_header_cloned(skb))
1462 return pskb_expand_head(skb, 0, 0, pri);
1468 * __skb_header_release - release reference to header
1469 * @skb: buffer to operate on
1471 static inline void __skb_header_release(struct sk_buff *skb)
1474 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1479 * skb_shared - is the buffer shared
1480 * @skb: buffer to check
1482 * Returns true if more than one person has a reference to this
1485 static inline int skb_shared(const struct sk_buff *skb)
1487 return refcount_read(&skb->users) != 1;
1491 * skb_share_check - check if buffer is shared and if so clone it
1492 * @skb: buffer to check
1493 * @pri: priority for memory allocation
1495 * If the buffer is shared the buffer is cloned and the old copy
1496 * drops a reference. A new clone with a single reference is returned.
1497 * If the buffer is not shared the original buffer is returned. When
1498 * being called from interrupt status or with spinlocks held pri must
1501 * NULL is returned on a memory allocation failure.
1503 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1505 might_sleep_if(gfpflags_allow_blocking(pri));
1506 if (skb_shared(skb)) {
1507 struct sk_buff *nskb = skb_clone(skb, pri);
1519 * Copy shared buffers into a new sk_buff. We effectively do COW on
1520 * packets to handle cases where we have a local reader and forward
1521 * and a couple of other messy ones. The normal one is tcpdumping
1522 * a packet thats being forwarded.
1526 * skb_unshare - make a copy of a shared buffer
1527 * @skb: buffer to check
1528 * @pri: priority for memory allocation
1530 * If the socket buffer is a clone then this function creates a new
1531 * copy of the data, drops a reference count on the old copy and returns
1532 * the new copy with the reference count at 1. If the buffer is not a clone
1533 * the original buffer is returned. When called with a spinlock held or
1534 * from interrupt state @pri must be %GFP_ATOMIC
1536 * %NULL is returned on a memory allocation failure.
1538 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1541 might_sleep_if(gfpflags_allow_blocking(pri));
1542 if (skb_cloned(skb)) {
1543 struct sk_buff *nskb = skb_copy(skb, pri);
1545 /* Free our shared copy */
1556 * skb_peek - peek at the head of an &sk_buff_head
1557 * @list_: list to peek at
1559 * Peek an &sk_buff. Unlike most other operations you _MUST_
1560 * be careful with this one. A peek leaves the buffer on the
1561 * list and someone else may run off with it. You must hold
1562 * the appropriate locks or have a private queue to do this.
1564 * Returns %NULL for an empty list or a pointer to the head element.
1565 * The reference count is not incremented and the reference is therefore
1566 * volatile. Use with caution.
1568 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1570 struct sk_buff *skb = list_->next;
1572 if (skb == (struct sk_buff *)list_)
1578 * skb_peek_next - peek skb following the given one from a queue
1579 * @skb: skb to start from
1580 * @list_: list to peek at
1582 * Returns %NULL when the end of the list is met or a pointer to the
1583 * next element. The reference count is not incremented and the
1584 * reference is therefore volatile. Use with caution.
1586 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1587 const struct sk_buff_head *list_)
1589 struct sk_buff *next = skb->next;
1591 if (next == (struct sk_buff *)list_)
1597 * skb_peek_tail - peek at the tail of an &sk_buff_head
1598 * @list_: list to peek at
1600 * Peek an &sk_buff. Unlike most other operations you _MUST_
1601 * be careful with this one. A peek leaves the buffer on the
1602 * list and someone else may run off with it. You must hold
1603 * the appropriate locks or have a private queue to do this.
1605 * Returns %NULL for an empty list or a pointer to the tail element.
1606 * The reference count is not incremented and the reference is therefore
1607 * volatile. Use with caution.
1609 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1611 struct sk_buff *skb = list_->prev;
1613 if (skb == (struct sk_buff *)list_)
1620 * skb_queue_len - get queue length
1621 * @list_: list to measure
1623 * Return the length of an &sk_buff queue.
1625 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1631 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1632 * @list: queue to initialize
1634 * This initializes only the list and queue length aspects of
1635 * an sk_buff_head object. This allows to initialize the list
1636 * aspects of an sk_buff_head without reinitializing things like
1637 * the spinlock. It can also be used for on-stack sk_buff_head
1638 * objects where the spinlock is known to not be used.
1640 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1642 list->prev = list->next = (struct sk_buff *)list;
1647 * This function creates a split out lock class for each invocation;
1648 * this is needed for now since a whole lot of users of the skb-queue
1649 * infrastructure in drivers have different locking usage (in hardirq)
1650 * than the networking core (in softirq only). In the long run either the
1651 * network layer or drivers should need annotation to consolidate the
1652 * main types of usage into 3 classes.
1654 static inline void skb_queue_head_init(struct sk_buff_head *list)
1656 spin_lock_init(&list->lock);
1657 __skb_queue_head_init(list);
1660 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1661 struct lock_class_key *class)
1663 skb_queue_head_init(list);
1664 lockdep_set_class(&list->lock, class);
1668 * Insert an sk_buff on a list.
1670 * The "__skb_xxxx()" functions are the non-atomic ones that
1671 * can only be called with interrupts disabled.
1673 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1674 struct sk_buff_head *list);
1675 static inline void __skb_insert(struct sk_buff *newsk,
1676 struct sk_buff *prev, struct sk_buff *next,
1677 struct sk_buff_head *list)
1681 next->prev = prev->next = newsk;
1685 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1686 struct sk_buff *prev,
1687 struct sk_buff *next)
1689 struct sk_buff *first = list->next;
1690 struct sk_buff *last = list->prev;
1700 * skb_queue_splice - join two skb lists, this is designed for stacks
1701 * @list: the new list to add
1702 * @head: the place to add it in the first list
1704 static inline void skb_queue_splice(const struct sk_buff_head *list,
1705 struct sk_buff_head *head)
1707 if (!skb_queue_empty(list)) {
1708 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1709 head->qlen += list->qlen;
1714 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1715 * @list: the new list to add
1716 * @head: the place to add it in the first list
1718 * The list at @list is reinitialised
1720 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1721 struct sk_buff_head *head)
1723 if (!skb_queue_empty(list)) {
1724 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1725 head->qlen += list->qlen;
1726 __skb_queue_head_init(list);
1731 * skb_queue_splice_tail - join two skb lists, each list being a queue
1732 * @list: the new list to add
1733 * @head: the place to add it in the first list
1735 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1736 struct sk_buff_head *head)
1738 if (!skb_queue_empty(list)) {
1739 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1740 head->qlen += list->qlen;
1745 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1746 * @list: the new list to add
1747 * @head: the place to add it in the first list
1749 * Each of the lists is a queue.
1750 * The list at @list is reinitialised
1752 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1753 struct sk_buff_head *head)
1755 if (!skb_queue_empty(list)) {
1756 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1757 head->qlen += list->qlen;
1758 __skb_queue_head_init(list);
1763 * __skb_queue_after - queue a buffer at the list head
1764 * @list: list to use
1765 * @prev: place after this buffer
1766 * @newsk: buffer to queue
1768 * Queue a buffer int the middle of a list. This function takes no locks
1769 * and you must therefore hold required locks before calling it.
1771 * A buffer cannot be placed on two lists at the same time.
1773 static inline void __skb_queue_after(struct sk_buff_head *list,
1774 struct sk_buff *prev,
1775 struct sk_buff *newsk)
1777 __skb_insert(newsk, prev, prev->next, list);
1780 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1781 struct sk_buff_head *list);
1783 static inline void __skb_queue_before(struct sk_buff_head *list,
1784 struct sk_buff *next,
1785 struct sk_buff *newsk)
1787 __skb_insert(newsk, next->prev, next, list);
1791 * __skb_queue_head - queue a buffer at the list head
1792 * @list: list to use
1793 * @newsk: buffer to queue
1795 * Queue a buffer at the start of a list. This function takes no locks
1796 * and you must therefore hold required locks before calling it.
1798 * A buffer cannot be placed on two lists at the same time.
1800 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1801 static inline void __skb_queue_head(struct sk_buff_head *list,
1802 struct sk_buff *newsk)
1804 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1808 * __skb_queue_tail - queue a buffer at the list tail
1809 * @list: list to use
1810 * @newsk: buffer to queue
1812 * Queue a buffer at the end of a list. This function takes no locks
1813 * and you must therefore hold required locks before calling it.
1815 * A buffer cannot be placed on two lists at the same time.
1817 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1818 static inline void __skb_queue_tail(struct sk_buff_head *list,
1819 struct sk_buff *newsk)
1821 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1825 * remove sk_buff from list. _Must_ be called atomically, and with
1828 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1829 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1831 struct sk_buff *next, *prev;
1836 skb->next = skb->prev = NULL;
1842 * __skb_dequeue - remove from the head of the queue
1843 * @list: list to dequeue from
1845 * Remove the head of the list. This function does not take any locks
1846 * so must be used with appropriate locks held only. The head item is
1847 * returned or %NULL if the list is empty.
1849 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1850 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1852 struct sk_buff *skb = skb_peek(list);
1854 __skb_unlink(skb, list);
1859 * __skb_dequeue_tail - remove from the tail of the queue
1860 * @list: list to dequeue from
1862 * Remove the tail of the list. This function does not take any locks
1863 * so must be used with appropriate locks held only. The tail item is
1864 * returned or %NULL if the list is empty.
1866 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1867 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1869 struct sk_buff *skb = skb_peek_tail(list);
1871 __skb_unlink(skb, list);
1876 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1878 return skb->data_len;
1881 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1883 return skb->len - skb->data_len;
1886 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
1888 unsigned int i, len = 0;
1890 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1891 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1895 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1897 return skb_headlen(skb) + __skb_pagelen(skb);
1901 * __skb_fill_page_desc - initialise a paged fragment in an skb
1902 * @skb: buffer containing fragment to be initialised
1903 * @i: paged fragment index to initialise
1904 * @page: the page to use for this fragment
1905 * @off: the offset to the data with @page
1906 * @size: the length of the data
1908 * Initialises the @i'th fragment of @skb to point to &size bytes at
1909 * offset @off within @page.
1911 * Does not take any additional reference on the fragment.
1913 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1914 struct page *page, int off, int size)
1916 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1919 * Propagate page pfmemalloc to the skb if we can. The problem is
1920 * that not all callers have unique ownership of the page but rely
1921 * on page_is_pfmemalloc doing the right thing(tm).
1923 frag->page.p = page;
1924 frag->page_offset = off;
1925 skb_frag_size_set(frag, size);
1927 page = compound_head(page);
1928 if (page_is_pfmemalloc(page))
1929 skb->pfmemalloc = true;
1933 * skb_fill_page_desc - initialise a paged fragment in an skb
1934 * @skb: buffer containing fragment to be initialised
1935 * @i: paged fragment index to initialise
1936 * @page: the page to use for this fragment
1937 * @off: the offset to the data with @page
1938 * @size: the length of the data
1940 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1941 * @skb to point to @size bytes at offset @off within @page. In
1942 * addition updates @skb such that @i is the last fragment.
1944 * Does not take any additional reference on the fragment.
1946 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1947 struct page *page, int off, int size)
1949 __skb_fill_page_desc(skb, i, page, off, size);
1950 skb_shinfo(skb)->nr_frags = i + 1;
1953 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1954 int size, unsigned int truesize);
1956 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1957 unsigned int truesize);
1959 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1960 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1961 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1963 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1964 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1966 return skb->head + skb->tail;
1969 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1971 skb->tail = skb->data - skb->head;
1974 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1976 skb_reset_tail_pointer(skb);
1977 skb->tail += offset;
1980 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1981 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1986 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1988 skb->tail = skb->data;
1991 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1993 skb->tail = skb->data + offset;
1996 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1999 * Add data to an sk_buff
2001 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2002 void *skb_put(struct sk_buff *skb, unsigned int len);
2003 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2005 void *tmp = skb_tail_pointer(skb);
2006 SKB_LINEAR_ASSERT(skb);
2012 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2014 void *tmp = __skb_put(skb, len);
2016 memset(tmp, 0, len);
2020 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2023 void *tmp = __skb_put(skb, len);
2025 memcpy(tmp, data, len);
2029 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2031 *(u8 *)__skb_put(skb, 1) = val;
2034 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2036 void *tmp = skb_put(skb, len);
2038 memset(tmp, 0, len);
2043 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2046 void *tmp = skb_put(skb, len);
2048 memcpy(tmp, data, len);
2053 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2055 *(u8 *)skb_put(skb, 1) = val;
2058 void *skb_push(struct sk_buff *skb, unsigned int len);
2059 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2066 void *skb_pull(struct sk_buff *skb, unsigned int len);
2067 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2070 BUG_ON(skb->len < skb->data_len);
2071 return skb->data += len;
2074 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2076 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2079 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2081 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2083 if (len > skb_headlen(skb) &&
2084 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2087 return skb->data += len;
2090 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2092 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2095 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2097 if (likely(len <= skb_headlen(skb)))
2099 if (unlikely(len > skb->len))
2101 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2104 void skb_condense(struct sk_buff *skb);
2107 * skb_headroom - bytes at buffer head
2108 * @skb: buffer to check
2110 * Return the number of bytes of free space at the head of an &sk_buff.
2112 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2114 return skb->data - skb->head;
2118 * skb_tailroom - bytes at buffer end
2119 * @skb: buffer to check
2121 * Return the number of bytes of free space at the tail of an sk_buff
2123 static inline int skb_tailroom(const struct sk_buff *skb)
2125 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2129 * skb_availroom - bytes at buffer end
2130 * @skb: buffer to check
2132 * Return the number of bytes of free space at the tail of an sk_buff
2133 * allocated by sk_stream_alloc()
2135 static inline int skb_availroom(const struct sk_buff *skb)
2137 if (skb_is_nonlinear(skb))
2140 return skb->end - skb->tail - skb->reserved_tailroom;
2144 * skb_reserve - adjust headroom
2145 * @skb: buffer to alter
2146 * @len: bytes to move
2148 * Increase the headroom of an empty &sk_buff by reducing the tail
2149 * room. This is only allowed for an empty buffer.
2151 static inline void skb_reserve(struct sk_buff *skb, int len)
2158 * skb_tailroom_reserve - adjust reserved_tailroom
2159 * @skb: buffer to alter
2160 * @mtu: maximum amount of headlen permitted
2161 * @needed_tailroom: minimum amount of reserved_tailroom
2163 * Set reserved_tailroom so that headlen can be as large as possible but
2164 * not larger than mtu and tailroom cannot be smaller than
2166 * The required headroom should already have been reserved before using
2169 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2170 unsigned int needed_tailroom)
2172 SKB_LINEAR_ASSERT(skb);
2173 if (mtu < skb_tailroom(skb) - needed_tailroom)
2174 /* use at most mtu */
2175 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2177 /* use up to all available space */
2178 skb->reserved_tailroom = needed_tailroom;
2181 #define ENCAP_TYPE_ETHER 0
2182 #define ENCAP_TYPE_IPPROTO 1
2184 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2187 skb->inner_protocol = protocol;
2188 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2191 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2194 skb->inner_ipproto = ipproto;
2195 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2198 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2200 skb->inner_mac_header = skb->mac_header;
2201 skb->inner_network_header = skb->network_header;
2202 skb->inner_transport_header = skb->transport_header;
2205 static inline void skb_reset_mac_len(struct sk_buff *skb)
2207 skb->mac_len = skb->network_header - skb->mac_header;
2210 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2213 return skb->head + skb->inner_transport_header;
2216 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2218 return skb_inner_transport_header(skb) - skb->data;
2221 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2223 skb->inner_transport_header = skb->data - skb->head;
2226 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2229 skb_reset_inner_transport_header(skb);
2230 skb->inner_transport_header += offset;
2233 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2235 return skb->head + skb->inner_network_header;
2238 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2240 skb->inner_network_header = skb->data - skb->head;
2243 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2246 skb_reset_inner_network_header(skb);
2247 skb->inner_network_header += offset;
2250 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2252 return skb->head + skb->inner_mac_header;
2255 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2257 skb->inner_mac_header = skb->data - skb->head;
2260 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2263 skb_reset_inner_mac_header(skb);
2264 skb->inner_mac_header += offset;
2266 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2268 return skb->transport_header != (typeof(skb->transport_header))~0U;
2271 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2273 return skb->head + skb->transport_header;
2276 static inline void skb_reset_transport_header(struct sk_buff *skb)
2278 skb->transport_header = skb->data - skb->head;
2281 static inline void skb_set_transport_header(struct sk_buff *skb,
2284 skb_reset_transport_header(skb);
2285 skb->transport_header += offset;
2288 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2290 return skb->head + skb->network_header;
2293 static inline void skb_reset_network_header(struct sk_buff *skb)
2295 skb->network_header = skb->data - skb->head;
2298 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2300 skb_reset_network_header(skb);
2301 skb->network_header += offset;
2304 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2306 return skb->head + skb->mac_header;
2309 static inline int skb_mac_offset(const struct sk_buff *skb)
2311 return skb_mac_header(skb) - skb->data;
2314 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2316 return skb->network_header - skb->mac_header;
2319 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2321 return skb->mac_header != (typeof(skb->mac_header))~0U;
2324 static inline void skb_reset_mac_header(struct sk_buff *skb)
2326 skb->mac_header = skb->data - skb->head;
2329 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2331 skb_reset_mac_header(skb);
2332 skb->mac_header += offset;
2335 static inline void skb_pop_mac_header(struct sk_buff *skb)
2337 skb->mac_header = skb->network_header;
2340 static inline void skb_probe_transport_header(struct sk_buff *skb,
2341 const int offset_hint)
2343 struct flow_keys keys;
2345 if (skb_transport_header_was_set(skb))
2347 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2348 skb_set_transport_header(skb, keys.control.thoff);
2350 skb_set_transport_header(skb, offset_hint);
2353 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2355 if (skb_mac_header_was_set(skb)) {
2356 const unsigned char *old_mac = skb_mac_header(skb);
2358 skb_set_mac_header(skb, -skb->mac_len);
2359 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2363 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2365 return skb->csum_start - skb_headroom(skb);
2368 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2370 return skb->head + skb->csum_start;
2373 static inline int skb_transport_offset(const struct sk_buff *skb)
2375 return skb_transport_header(skb) - skb->data;
2378 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2380 return skb->transport_header - skb->network_header;
2383 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2385 return skb->inner_transport_header - skb->inner_network_header;
2388 static inline int skb_network_offset(const struct sk_buff *skb)
2390 return skb_network_header(skb) - skb->data;
2393 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2395 return skb_inner_network_header(skb) - skb->data;
2398 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2400 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2404 * CPUs often take a performance hit when accessing unaligned memory
2405 * locations. The actual performance hit varies, it can be small if the
2406 * hardware handles it or large if we have to take an exception and fix it
2409 * Since an ethernet header is 14 bytes network drivers often end up with
2410 * the IP header at an unaligned offset. The IP header can be aligned by
2411 * shifting the start of the packet by 2 bytes. Drivers should do this
2414 * skb_reserve(skb, NET_IP_ALIGN);
2416 * The downside to this alignment of the IP header is that the DMA is now
2417 * unaligned. On some architectures the cost of an unaligned DMA is high
2418 * and this cost outweighs the gains made by aligning the IP header.
2420 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2423 #ifndef NET_IP_ALIGN
2424 #define NET_IP_ALIGN 2
2428 * The networking layer reserves some headroom in skb data (via
2429 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2430 * the header has to grow. In the default case, if the header has to grow
2431 * 32 bytes or less we avoid the reallocation.
2433 * Unfortunately this headroom changes the DMA alignment of the resulting
2434 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2435 * on some architectures. An architecture can override this value,
2436 * perhaps setting it to a cacheline in size (since that will maintain
2437 * cacheline alignment of the DMA). It must be a power of 2.
2439 * Various parts of the networking layer expect at least 32 bytes of
2440 * headroom, you should not reduce this.
2442 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2443 * to reduce average number of cache lines per packet.
2444 * get_rps_cpus() for example only access one 64 bytes aligned block :
2445 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2448 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2451 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2453 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2455 if (unlikely(skb_is_nonlinear(skb))) {
2460 skb_set_tail_pointer(skb, len);
2463 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2465 __skb_set_length(skb, len);
2468 void skb_trim(struct sk_buff *skb, unsigned int len);
2470 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2473 return ___pskb_trim(skb, len);
2474 __skb_trim(skb, len);
2478 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2480 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2484 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2485 * @skb: buffer to alter
2488 * This is identical to pskb_trim except that the caller knows that
2489 * the skb is not cloned so we should never get an error due to out-
2492 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2494 int err = pskb_trim(skb, len);
2498 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2500 unsigned int diff = len - skb->len;
2502 if (skb_tailroom(skb) < diff) {
2503 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2508 __skb_set_length(skb, len);
2513 * skb_orphan - orphan a buffer
2514 * @skb: buffer to orphan
2516 * If a buffer currently has an owner then we call the owner's
2517 * destructor function and make the @skb unowned. The buffer continues
2518 * to exist but is no longer charged to its former owner.
2520 static inline void skb_orphan(struct sk_buff *skb)
2522 if (skb->destructor) {
2523 skb->destructor(skb);
2524 skb->destructor = NULL;
2532 * skb_orphan_frags - orphan the frags contained in a buffer
2533 * @skb: buffer to orphan frags from
2534 * @gfp_mask: allocation mask for replacement pages
2536 * For each frag in the SKB which needs a destructor (i.e. has an
2537 * owner) create a copy of that frag and release the original
2538 * page by calling the destructor.
2540 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2542 if (likely(!skb_zcopy(skb)))
2544 if (skb_uarg(skb)->callback == sock_zerocopy_callback)
2546 return skb_copy_ubufs(skb, gfp_mask);
2549 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2550 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2552 if (likely(!skb_zcopy(skb)))
2554 return skb_copy_ubufs(skb, gfp_mask);
2558 * __skb_queue_purge - empty a list
2559 * @list: list to empty
2561 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2562 * the list and one reference dropped. This function does not take the
2563 * list lock and the caller must hold the relevant locks to use it.
2565 void skb_queue_purge(struct sk_buff_head *list);
2566 static inline void __skb_queue_purge(struct sk_buff_head *list)
2568 struct sk_buff *skb;
2569 while ((skb = __skb_dequeue(list)) != NULL)
2573 void skb_rbtree_purge(struct rb_root *root);
2575 void *netdev_alloc_frag(unsigned int fragsz);
2577 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2581 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2582 * @dev: network device to receive on
2583 * @length: length to allocate
2585 * Allocate a new &sk_buff and assign it a usage count of one. The
2586 * buffer has unspecified headroom built in. Users should allocate
2587 * the headroom they think they need without accounting for the
2588 * built in space. The built in space is used for optimisations.
2590 * %NULL is returned if there is no free memory. Although this function
2591 * allocates memory it can be called from an interrupt.
2593 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2594 unsigned int length)
2596 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2599 /* legacy helper around __netdev_alloc_skb() */
2600 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2603 return __netdev_alloc_skb(NULL, length, gfp_mask);
2606 /* legacy helper around netdev_alloc_skb() */
2607 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2609 return netdev_alloc_skb(NULL, length);
2613 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2614 unsigned int length, gfp_t gfp)
2616 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2618 if (NET_IP_ALIGN && skb)
2619 skb_reserve(skb, NET_IP_ALIGN);
2623 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2624 unsigned int length)
2626 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2629 static inline void skb_free_frag(void *addr)
2631 page_frag_free(addr);
2634 void *napi_alloc_frag(unsigned int fragsz);
2635 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2636 unsigned int length, gfp_t gfp_mask);
2637 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2638 unsigned int length)
2640 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2642 void napi_consume_skb(struct sk_buff *skb, int budget);
2644 void __kfree_skb_flush(void);
2645 void __kfree_skb_defer(struct sk_buff *skb);
2648 * __dev_alloc_pages - allocate page for network Rx
2649 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2650 * @order: size of the allocation
2652 * Allocate a new page.
2654 * %NULL is returned if there is no free memory.
2656 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2659 /* This piece of code contains several assumptions.
2660 * 1. This is for device Rx, therefor a cold page is preferred.
2661 * 2. The expectation is the user wants a compound page.
2662 * 3. If requesting a order 0 page it will not be compound
2663 * due to the check to see if order has a value in prep_new_page
2664 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2665 * code in gfp_to_alloc_flags that should be enforcing this.
2667 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2669 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2672 static inline struct page *dev_alloc_pages(unsigned int order)
2674 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2678 * __dev_alloc_page - allocate a page for network Rx
2679 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2681 * Allocate a new page.
2683 * %NULL is returned if there is no free memory.
2685 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2687 return __dev_alloc_pages(gfp_mask, 0);
2690 static inline struct page *dev_alloc_page(void)
2692 return dev_alloc_pages(0);
2696 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2697 * @page: The page that was allocated from skb_alloc_page
2698 * @skb: The skb that may need pfmemalloc set
2700 static inline void skb_propagate_pfmemalloc(struct page *page,
2701 struct sk_buff *skb)
2703 if (page_is_pfmemalloc(page))
2704 skb->pfmemalloc = true;
2708 * skb_frag_page - retrieve the page referred to by a paged fragment
2709 * @frag: the paged fragment
2711 * Returns the &struct page associated with @frag.
2713 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2715 return frag->page.p;
2719 * __skb_frag_ref - take an addition reference on a paged fragment.
2720 * @frag: the paged fragment
2722 * Takes an additional reference on the paged fragment @frag.
2724 static inline void __skb_frag_ref(skb_frag_t *frag)
2726 get_page(skb_frag_page(frag));
2730 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2732 * @f: the fragment offset.
2734 * Takes an additional reference on the @f'th paged fragment of @skb.
2736 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2738 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2742 * __skb_frag_unref - release a reference on a paged fragment.
2743 * @frag: the paged fragment
2745 * Releases a reference on the paged fragment @frag.
2747 static inline void __skb_frag_unref(skb_frag_t *frag)
2749 put_page(skb_frag_page(frag));
2753 * skb_frag_unref - release a reference on a paged fragment of an skb.
2755 * @f: the fragment offset
2757 * Releases a reference on the @f'th paged fragment of @skb.
2759 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2761 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2765 * skb_frag_address - gets the address of the data contained in a paged fragment
2766 * @frag: the paged fragment buffer
2768 * Returns the address of the data within @frag. The page must already
2771 static inline void *skb_frag_address(const skb_frag_t *frag)
2773 return page_address(skb_frag_page(frag)) + frag->page_offset;
2777 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2778 * @frag: the paged fragment buffer
2780 * Returns the address of the data within @frag. Checks that the page
2781 * is mapped and returns %NULL otherwise.
2783 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2785 void *ptr = page_address(skb_frag_page(frag));
2789 return ptr + frag->page_offset;
2793 * __skb_frag_set_page - sets the page contained in a paged fragment
2794 * @frag: the paged fragment
2795 * @page: the page to set
2797 * Sets the fragment @frag to contain @page.
2799 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2801 frag->page.p = page;
2805 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2807 * @f: the fragment offset
2808 * @page: the page to set
2810 * Sets the @f'th fragment of @skb to contain @page.
2812 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2815 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2818 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2821 * skb_frag_dma_map - maps a paged fragment via the DMA API
2822 * @dev: the device to map the fragment to
2823 * @frag: the paged fragment to map
2824 * @offset: the offset within the fragment (starting at the
2825 * fragment's own offset)
2826 * @size: the number of bytes to map
2827 * @dir: the direction of the mapping (``PCI_DMA_*``)
2829 * Maps the page associated with @frag to @device.
2831 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2832 const skb_frag_t *frag,
2833 size_t offset, size_t size,
2834 enum dma_data_direction dir)
2836 return dma_map_page(dev, skb_frag_page(frag),
2837 frag->page_offset + offset, size, dir);
2840 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2843 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2847 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2850 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2855 * skb_clone_writable - is the header of a clone writable
2856 * @skb: buffer to check
2857 * @len: length up to which to write
2859 * Returns true if modifying the header part of the cloned buffer
2860 * does not requires the data to be copied.
2862 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2864 return !skb_header_cloned(skb) &&
2865 skb_headroom(skb) + len <= skb->hdr_len;
2868 static inline int skb_try_make_writable(struct sk_buff *skb,
2869 unsigned int write_len)
2871 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2872 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2875 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2880 if (headroom > skb_headroom(skb))
2881 delta = headroom - skb_headroom(skb);
2883 if (delta || cloned)
2884 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2890 * skb_cow - copy header of skb when it is required
2891 * @skb: buffer to cow
2892 * @headroom: needed headroom
2894 * If the skb passed lacks sufficient headroom or its data part
2895 * is shared, data is reallocated. If reallocation fails, an error
2896 * is returned and original skb is not changed.
2898 * The result is skb with writable area skb->head...skb->tail
2899 * and at least @headroom of space at head.
2901 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2903 return __skb_cow(skb, headroom, skb_cloned(skb));
2907 * skb_cow_head - skb_cow but only making the head writable
2908 * @skb: buffer to cow
2909 * @headroom: needed headroom
2911 * This function is identical to skb_cow except that we replace the
2912 * skb_cloned check by skb_header_cloned. It should be used when
2913 * you only need to push on some header and do not need to modify
2916 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2918 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2922 * skb_padto - pad an skbuff up to a minimal size
2923 * @skb: buffer to pad
2924 * @len: minimal length
2926 * Pads up a buffer to ensure the trailing bytes exist and are
2927 * blanked. If the buffer already contains sufficient data it
2928 * is untouched. Otherwise it is extended. Returns zero on
2929 * success. The skb is freed on error.
2931 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2933 unsigned int size = skb->len;
2934 if (likely(size >= len))
2936 return skb_pad(skb, len - size);
2940 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2941 * @skb: buffer to pad
2942 * @len: minimal length
2943 * @free_on_error: free buffer on error
2945 * Pads up a buffer to ensure the trailing bytes exist and are
2946 * blanked. If the buffer already contains sufficient data it
2947 * is untouched. Otherwise it is extended. Returns zero on
2948 * success. The skb is freed on error if @free_on_error is true.
2950 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
2953 unsigned int size = skb->len;
2955 if (unlikely(size < len)) {
2957 if (__skb_pad(skb, len, free_on_error))
2959 __skb_put(skb, len);
2965 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2966 * @skb: buffer to pad
2967 * @len: minimal length
2969 * Pads up a buffer to ensure the trailing bytes exist and are
2970 * blanked. If the buffer already contains sufficient data it
2971 * is untouched. Otherwise it is extended. Returns zero on
2972 * success. The skb is freed on error.
2974 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2976 return __skb_put_padto(skb, len, true);
2979 static inline int skb_add_data(struct sk_buff *skb,
2980 struct iov_iter *from, int copy)
2982 const int off = skb->len;
2984 if (skb->ip_summed == CHECKSUM_NONE) {
2986 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2988 skb->csum = csum_block_add(skb->csum, csum, off);
2991 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
2994 __skb_trim(skb, off);
2998 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2999 const struct page *page, int off)
3004 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
3006 return page == skb_frag_page(frag) &&
3007 off == frag->page_offset + skb_frag_size(frag);
3012 static inline int __skb_linearize(struct sk_buff *skb)
3014 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3018 * skb_linearize - convert paged skb to linear one
3019 * @skb: buffer to linarize
3021 * If there is no free memory -ENOMEM is returned, otherwise zero
3022 * is returned and the old skb data released.
3024 static inline int skb_linearize(struct sk_buff *skb)
3026 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3030 * skb_has_shared_frag - can any frag be overwritten
3031 * @skb: buffer to test
3033 * Return true if the skb has at least one frag that might be modified
3034 * by an external entity (as in vmsplice()/sendfile())
3036 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3038 return skb_is_nonlinear(skb) &&
3039 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3043 * skb_linearize_cow - make sure skb is linear and writable
3044 * @skb: buffer to process
3046 * If there is no free memory -ENOMEM is returned, otherwise zero
3047 * is returned and the old skb data released.
3049 static inline int skb_linearize_cow(struct sk_buff *skb)
3051 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3052 __skb_linearize(skb) : 0;
3055 static __always_inline void
3056 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3059 if (skb->ip_summed == CHECKSUM_COMPLETE)
3060 skb->csum = csum_block_sub(skb->csum,
3061 csum_partial(start, len, 0), off);
3062 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3063 skb_checksum_start_offset(skb) < 0)
3064 skb->ip_summed = CHECKSUM_NONE;
3068 * skb_postpull_rcsum - update checksum for received skb after pull
3069 * @skb: buffer to update
3070 * @start: start of data before pull
3071 * @len: length of data pulled
3073 * After doing a pull on a received packet, you need to call this to
3074 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3075 * CHECKSUM_NONE so that it can be recomputed from scratch.
3077 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3078 const void *start, unsigned int len)
3080 __skb_postpull_rcsum(skb, start, len, 0);
3083 static __always_inline void
3084 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3087 if (skb->ip_summed == CHECKSUM_COMPLETE)
3088 skb->csum = csum_block_add(skb->csum,
3089 csum_partial(start, len, 0), off);
3093 * skb_postpush_rcsum - update checksum for received skb after push
3094 * @skb: buffer to update
3095 * @start: start of data after push
3096 * @len: length of data pushed
3098 * After doing a push on a received packet, you need to call this to
3099 * update the CHECKSUM_COMPLETE checksum.
3101 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3102 const void *start, unsigned int len)
3104 __skb_postpush_rcsum(skb, start, len, 0);
3107 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3110 * skb_push_rcsum - push skb and update receive checksum
3111 * @skb: buffer to update
3112 * @len: length of data pulled
3114 * This function performs an skb_push on the packet and updates
3115 * the CHECKSUM_COMPLETE checksum. It should be used on
3116 * receive path processing instead of skb_push unless you know
3117 * that the checksum difference is zero (e.g., a valid IP header)
3118 * or you are setting ip_summed to CHECKSUM_NONE.
3120 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3123 skb_postpush_rcsum(skb, skb->data, len);
3128 * pskb_trim_rcsum - trim received skb and update checksum
3129 * @skb: buffer to trim
3132 * This is exactly the same as pskb_trim except that it ensures the
3133 * checksum of received packets are still valid after the operation.
3136 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3138 if (likely(len >= skb->len))
3140 if (skb->ip_summed == CHECKSUM_COMPLETE)
3141 skb->ip_summed = CHECKSUM_NONE;
3142 return __pskb_trim(skb, len);
3145 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3147 if (skb->ip_summed == CHECKSUM_COMPLETE)
3148 skb->ip_summed = CHECKSUM_NONE;
3149 __skb_trim(skb, len);
3153 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3155 if (skb->ip_summed == CHECKSUM_COMPLETE)
3156 skb->ip_summed = CHECKSUM_NONE;
3157 return __skb_grow(skb, len);
3160 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3161 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3162 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3163 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3164 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3166 #define skb_queue_walk(queue, skb) \
3167 for (skb = (queue)->next; \
3168 skb != (struct sk_buff *)(queue); \
3171 #define skb_queue_walk_safe(queue, skb, tmp) \
3172 for (skb = (queue)->next, tmp = skb->next; \
3173 skb != (struct sk_buff *)(queue); \
3174 skb = tmp, tmp = skb->next)
3176 #define skb_queue_walk_from(queue, skb) \
3177 for (; skb != (struct sk_buff *)(queue); \
3180 #define skb_rbtree_walk(skb, root) \
3181 for (skb = skb_rb_first(root); skb != NULL; \
3182 skb = skb_rb_next(skb))
3184 #define skb_rbtree_walk_from(skb) \
3185 for (; skb != NULL; \
3186 skb = skb_rb_next(skb))
3188 #define skb_rbtree_walk_from_safe(skb, tmp) \
3189 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3192 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3193 for (tmp = skb->next; \
3194 skb != (struct sk_buff *)(queue); \
3195 skb = tmp, tmp = skb->next)
3197 #define skb_queue_reverse_walk(queue, skb) \
3198 for (skb = (queue)->prev; \
3199 skb != (struct sk_buff *)(queue); \
3202 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3203 for (skb = (queue)->prev, tmp = skb->prev; \
3204 skb != (struct sk_buff *)(queue); \
3205 skb = tmp, tmp = skb->prev)
3207 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3208 for (tmp = skb->prev; \
3209 skb != (struct sk_buff *)(queue); \
3210 skb = tmp, tmp = skb->prev)
3212 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3214 return skb_shinfo(skb)->frag_list != NULL;
3217 static inline void skb_frag_list_init(struct sk_buff *skb)
3219 skb_shinfo(skb)->frag_list = NULL;
3222 #define skb_walk_frags(skb, iter) \
3223 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3226 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3227 const struct sk_buff *skb);
3228 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3229 struct sk_buff_head *queue,
3231 void (*destructor)(struct sock *sk,
3232 struct sk_buff *skb),
3233 int *peeked, int *off, int *err,
3234 struct sk_buff **last);
3235 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3236 void (*destructor)(struct sock *sk,
3237 struct sk_buff *skb),
3238 int *peeked, int *off, int *err,
3239 struct sk_buff **last);
3240 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3241 void (*destructor)(struct sock *sk,
3242 struct sk_buff *skb),
3243 int *peeked, int *off, int *err);
3244 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3246 unsigned int datagram_poll(struct file *file, struct socket *sock,
3247 struct poll_table_struct *wait);
3248 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3249 struct iov_iter *to, int size);
3250 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3251 struct msghdr *msg, int size)
3253 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3255 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3256 struct msghdr *msg);
3257 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3258 struct iov_iter *from, int len);
3259 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3260 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3261 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3262 static inline void skb_free_datagram_locked(struct sock *sk,
3263 struct sk_buff *skb)
3265 __skb_free_datagram_locked(sk, skb, 0);
3267 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3268 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3269 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3270 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3271 int len, __wsum csum);
3272 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3273 struct pipe_inode_info *pipe, unsigned int len,
3274 unsigned int flags);
3275 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3277 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3278 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3279 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3280 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3282 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3283 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3284 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3285 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3286 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3287 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3288 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3289 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3290 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3291 int skb_vlan_pop(struct sk_buff *skb);
3292 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3293 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3296 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3298 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3301 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3303 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3306 struct skb_checksum_ops {
3307 __wsum (*update)(const void *mem, int len, __wsum wsum);
3308 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3311 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3313 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3314 __wsum csum, const struct skb_checksum_ops *ops);
3315 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3318 static inline void * __must_check
3319 __skb_header_pointer(const struct sk_buff *skb, int offset,
3320 int len, void *data, int hlen, void *buffer)
3322 if (hlen - offset >= len)
3323 return data + offset;
3326 skb_copy_bits(skb, offset, buffer, len) < 0)
3332 static inline void * __must_check
3333 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3335 return __skb_header_pointer(skb, offset, len, skb->data,
3336 skb_headlen(skb), buffer);
3340 * skb_needs_linearize - check if we need to linearize a given skb
3341 * depending on the given device features.
3342 * @skb: socket buffer to check
3343 * @features: net device features
3345 * Returns true if either:
3346 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3347 * 2. skb is fragmented and the device does not support SG.
3349 static inline bool skb_needs_linearize(struct sk_buff *skb,
3350 netdev_features_t features)
3352 return skb_is_nonlinear(skb) &&
3353 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3354 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3357 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3359 const unsigned int len)
3361 memcpy(to, skb->data, len);
3364 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3365 const int offset, void *to,
3366 const unsigned int len)
3368 memcpy(to, skb->data + offset, len);
3371 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3373 const unsigned int len)
3375 memcpy(skb->data, from, len);
3378 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3381 const unsigned int len)
3383 memcpy(skb->data + offset, from, len);
3386 void skb_init(void);
3388 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3394 * skb_get_timestamp - get timestamp from a skb
3395 * @skb: skb to get stamp from
3396 * @stamp: pointer to struct timeval to store stamp in
3398 * Timestamps are stored in the skb as offsets to a base timestamp.
3399 * This function converts the offset back to a struct timeval and stores
3402 static inline void skb_get_timestamp(const struct sk_buff *skb,
3403 struct timeval *stamp)
3405 *stamp = ktime_to_timeval(skb->tstamp);
3408 static inline void skb_get_timestampns(const struct sk_buff *skb,
3409 struct timespec *stamp)
3411 *stamp = ktime_to_timespec(skb->tstamp);
3414 static inline void __net_timestamp(struct sk_buff *skb)
3416 skb->tstamp = ktime_get_real();
3419 static inline ktime_t net_timedelta(ktime_t t)
3421 return ktime_sub(ktime_get_real(), t);
3424 static inline ktime_t net_invalid_timestamp(void)
3429 static inline u8 skb_metadata_len(const struct sk_buff *skb)
3431 return skb_shinfo(skb)->meta_len;
3434 static inline void *skb_metadata_end(const struct sk_buff *skb)
3436 return skb_mac_header(skb);
3439 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3440 const struct sk_buff *skb_b,
3443 const void *a = skb_metadata_end(skb_a);
3444 const void *b = skb_metadata_end(skb_b);
3445 /* Using more efficient varaiant than plain call to memcmp(). */
3446 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3450 #define __it(x, op) (x -= sizeof(u##op))
3451 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3452 case 32: diffs |= __it_diff(a, b, 64);
3453 case 24: diffs |= __it_diff(a, b, 64);
3454 case 16: diffs |= __it_diff(a, b, 64);
3455 case 8: diffs |= __it_diff(a, b, 64);
3457 case 28: diffs |= __it_diff(a, b, 64);
3458 case 20: diffs |= __it_diff(a, b, 64);
3459 case 12: diffs |= __it_diff(a, b, 64);
3460 case 4: diffs |= __it_diff(a, b, 32);
3465 return memcmp(a - meta_len, b - meta_len, meta_len);
3469 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3470 const struct sk_buff *skb_b)
3472 u8 len_a = skb_metadata_len(skb_a);
3473 u8 len_b = skb_metadata_len(skb_b);
3475 if (!(len_a | len_b))
3478 return len_a != len_b ?
3479 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3482 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3484 skb_shinfo(skb)->meta_len = meta_len;
3487 static inline void skb_metadata_clear(struct sk_buff *skb)
3489 skb_metadata_set(skb, 0);
3492 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3494 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3496 void skb_clone_tx_timestamp(struct sk_buff *skb);
3497 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3499 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3501 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3505 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3510 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3513 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3515 * PHY drivers may accept clones of transmitted packets for
3516 * timestamping via their phy_driver.txtstamp method. These drivers
3517 * must call this function to return the skb back to the stack with a
3520 * @skb: clone of the the original outgoing packet
3521 * @hwtstamps: hardware time stamps
3524 void skb_complete_tx_timestamp(struct sk_buff *skb,
3525 struct skb_shared_hwtstamps *hwtstamps);
3527 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3528 struct skb_shared_hwtstamps *hwtstamps,
3529 struct sock *sk, int tstype);
3532 * skb_tstamp_tx - queue clone of skb with send time stamps
3533 * @orig_skb: the original outgoing packet
3534 * @hwtstamps: hardware time stamps, may be NULL if not available
3536 * If the skb has a socket associated, then this function clones the
3537 * skb (thus sharing the actual data and optional structures), stores
3538 * the optional hardware time stamping information (if non NULL) or
3539 * generates a software time stamp (otherwise), then queues the clone
3540 * to the error queue of the socket. Errors are silently ignored.
3542 void skb_tstamp_tx(struct sk_buff *orig_skb,
3543 struct skb_shared_hwtstamps *hwtstamps);
3546 * skb_tx_timestamp() - Driver hook for transmit timestamping
3548 * Ethernet MAC Drivers should call this function in their hard_xmit()
3549 * function immediately before giving the sk_buff to the MAC hardware.
3551 * Specifically, one should make absolutely sure that this function is
3552 * called before TX completion of this packet can trigger. Otherwise
3553 * the packet could potentially already be freed.
3555 * @skb: A socket buffer.
3557 static inline void skb_tx_timestamp(struct sk_buff *skb)
3559 skb_clone_tx_timestamp(skb);
3560 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3561 skb_tstamp_tx(skb, NULL);
3565 * skb_complete_wifi_ack - deliver skb with wifi status
3567 * @skb: the original outgoing packet
3568 * @acked: ack status
3571 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3573 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3574 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3576 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3578 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3580 (skb->ip_summed == CHECKSUM_PARTIAL &&
3581 skb_checksum_start_offset(skb) >= 0));
3585 * skb_checksum_complete - Calculate checksum of an entire packet
3586 * @skb: packet to process
3588 * This function calculates the checksum over the entire packet plus
3589 * the value of skb->csum. The latter can be used to supply the
3590 * checksum of a pseudo header as used by TCP/UDP. It returns the
3593 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3594 * this function can be used to verify that checksum on received
3595 * packets. In that case the function should return zero if the
3596 * checksum is correct. In particular, this function will return zero
3597 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3598 * hardware has already verified the correctness of the checksum.
3600 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3602 return skb_csum_unnecessary(skb) ?
3603 0 : __skb_checksum_complete(skb);
3606 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3608 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3609 if (skb->csum_level == 0)
3610 skb->ip_summed = CHECKSUM_NONE;
3616 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3618 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3619 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3621 } else if (skb->ip_summed == CHECKSUM_NONE) {
3622 skb->ip_summed = CHECKSUM_UNNECESSARY;
3623 skb->csum_level = 0;
3627 /* Check if we need to perform checksum complete validation.
3629 * Returns true if checksum complete is needed, false otherwise
3630 * (either checksum is unnecessary or zero checksum is allowed).
3632 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3636 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3637 skb->csum_valid = 1;
3638 __skb_decr_checksum_unnecessary(skb);
3645 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3648 #define CHECKSUM_BREAK 76
3650 /* Unset checksum-complete
3652 * Unset checksum complete can be done when packet is being modified
3653 * (uncompressed for instance) and checksum-complete value is
3656 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3658 if (skb->ip_summed == CHECKSUM_COMPLETE)
3659 skb->ip_summed = CHECKSUM_NONE;
3662 /* Validate (init) checksum based on checksum complete.
3665 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3666 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3667 * checksum is stored in skb->csum for use in __skb_checksum_complete
3668 * non-zero: value of invalid checksum
3671 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3675 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3676 if (!csum_fold(csum_add(psum, skb->csum))) {
3677 skb->csum_valid = 1;
3684 if (complete || skb->len <= CHECKSUM_BREAK) {
3687 csum = __skb_checksum_complete(skb);
3688 skb->csum_valid = !csum;
3695 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3700 /* Perform checksum validate (init). Note that this is a macro since we only
3701 * want to calculate the pseudo header which is an input function if necessary.
3702 * First we try to validate without any computation (checksum unnecessary) and
3703 * then calculate based on checksum complete calling the function to compute
3707 * 0: checksum is validated or try to in skb_checksum_complete
3708 * non-zero: value of invalid checksum
3710 #define __skb_checksum_validate(skb, proto, complete, \
3711 zero_okay, check, compute_pseudo) \
3713 __sum16 __ret = 0; \
3714 skb->csum_valid = 0; \
3715 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3716 __ret = __skb_checksum_validate_complete(skb, \
3717 complete, compute_pseudo(skb, proto)); \
3721 #define skb_checksum_init(skb, proto, compute_pseudo) \
3722 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3724 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3725 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3727 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3728 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3730 #define skb_checksum_validate_zero_check(skb, proto, check, \
3732 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3734 #define skb_checksum_simple_validate(skb) \
3735 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3737 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3739 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3742 static inline void __skb_checksum_convert(struct sk_buff *skb,
3743 __sum16 check, __wsum pseudo)
3745 skb->csum = ~pseudo;
3746 skb->ip_summed = CHECKSUM_COMPLETE;
3749 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3751 if (__skb_checksum_convert_check(skb)) \
3752 __skb_checksum_convert(skb, check, \
3753 compute_pseudo(skb, proto)); \
3756 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3757 u16 start, u16 offset)
3759 skb->ip_summed = CHECKSUM_PARTIAL;
3760 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3761 skb->csum_offset = offset - start;
3764 /* Update skbuf and packet to reflect the remote checksum offload operation.
3765 * When called, ptr indicates the starting point for skb->csum when
3766 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3767 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3769 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3770 int start, int offset, bool nopartial)
3775 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3779 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3780 __skb_checksum_complete(skb);
3781 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3784 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3786 /* Adjust skb->csum since we changed the packet */
3787 skb->csum = csum_add(skb->csum, delta);
3790 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3792 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3793 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3799 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3800 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3801 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3803 if (nfct && atomic_dec_and_test(&nfct->use))
3804 nf_conntrack_destroy(nfct);
3806 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3809 atomic_inc(&nfct->use);
3812 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3813 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3815 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use))
3818 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3821 refcount_inc(&nf_bridge->use);
3823 #endif /* CONFIG_BRIDGE_NETFILTER */
3824 static inline void nf_reset(struct sk_buff *skb)
3826 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3827 nf_conntrack_put(skb_nfct(skb));
3830 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3831 nf_bridge_put(skb->nf_bridge);
3832 skb->nf_bridge = NULL;
3836 static inline void nf_reset_trace(struct sk_buff *skb)
3838 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3843 static inline void ipvs_reset(struct sk_buff *skb)
3845 #if IS_ENABLED(CONFIG_IP_VS)
3846 skb->ipvs_property = 0;
3850 /* Note: This doesn't put any conntrack and bridge info in dst. */
3851 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3854 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3855 dst->_nfct = src->_nfct;
3856 nf_conntrack_get(skb_nfct(src));
3858 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3859 dst->nf_bridge = src->nf_bridge;
3860 nf_bridge_get(src->nf_bridge);
3862 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3864 dst->nf_trace = src->nf_trace;
3868 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3870 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3871 nf_conntrack_put(skb_nfct(dst));
3873 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3874 nf_bridge_put(dst->nf_bridge);
3876 __nf_copy(dst, src, true);
3879 #ifdef CONFIG_NETWORK_SECMARK
3880 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3882 to->secmark = from->secmark;
3885 static inline void skb_init_secmark(struct sk_buff *skb)
3890 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3893 static inline void skb_init_secmark(struct sk_buff *skb)
3897 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3899 return !skb->destructor &&
3900 #if IS_ENABLED(CONFIG_XFRM)
3904 !skb->_skb_refdst &&
3905 !skb_has_frag_list(skb);
3908 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3910 skb->queue_mapping = queue_mapping;
3913 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3915 return skb->queue_mapping;
3918 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3920 to->queue_mapping = from->queue_mapping;
3923 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3925 skb->queue_mapping = rx_queue + 1;
3928 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3930 return skb->queue_mapping - 1;
3933 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3935 return skb->queue_mapping != 0;
3938 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3940 skb->dst_pending_confirm = val;
3943 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3945 return skb->dst_pending_confirm != 0;
3948 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3957 /* Keeps track of mac header offset relative to skb->head.
3958 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3959 * For non-tunnel skb it points to skb_mac_header() and for
3960 * tunnel skb it points to outer mac header.
3961 * Keeps track of level of encapsulation of network headers.
3972 #define SKB_SGO_CB_OFFSET 32
3973 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3975 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3977 return (skb_mac_header(inner_skb) - inner_skb->head) -
3978 SKB_GSO_CB(inner_skb)->mac_offset;
3981 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3983 int new_headroom, headroom;
3986 headroom = skb_headroom(skb);
3987 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3991 new_headroom = skb_headroom(skb);
3992 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3996 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3998 /* Do not update partial checksums if remote checksum is enabled. */
3999 if (skb->remcsum_offload)
4002 SKB_GSO_CB(skb)->csum = res;
4003 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4006 /* Compute the checksum for a gso segment. First compute the checksum value
4007 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4008 * then add in skb->csum (checksum from csum_start to end of packet).
4009 * skb->csum and csum_start are then updated to reflect the checksum of the
4010 * resultant packet starting from the transport header-- the resultant checksum
4011 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4014 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4016 unsigned char *csum_start = skb_transport_header(skb);
4017 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4018 __wsum partial = SKB_GSO_CB(skb)->csum;
4020 SKB_GSO_CB(skb)->csum = res;
4021 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4023 return csum_fold(csum_partial(csum_start, plen, partial));
4026 static inline bool skb_is_gso(const struct sk_buff *skb)
4028 return skb_shinfo(skb)->gso_size;
4031 /* Note: Should be called only if skb_is_gso(skb) is true */
4032 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4034 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4037 static inline void skb_gso_reset(struct sk_buff *skb)
4039 skb_shinfo(skb)->gso_size = 0;
4040 skb_shinfo(skb)->gso_segs = 0;
4041 skb_shinfo(skb)->gso_type = 0;
4044 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4046 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4048 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4049 * wanted then gso_type will be set. */
4050 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4052 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4053 unlikely(shinfo->gso_type == 0)) {
4054 __skb_warn_lro_forwarding(skb);
4060 static inline void skb_forward_csum(struct sk_buff *skb)
4062 /* Unfortunately we don't support this one. Any brave souls? */
4063 if (skb->ip_summed == CHECKSUM_COMPLETE)
4064 skb->ip_summed = CHECKSUM_NONE;
4068 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4069 * @skb: skb to check
4071 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4072 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4073 * use this helper, to document places where we make this assertion.
4075 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4078 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4082 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4084 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4085 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4086 unsigned int transport_len,
4087 __sum16(*skb_chkf)(struct sk_buff *skb));
4090 * skb_head_is_locked - Determine if the skb->head is locked down
4091 * @skb: skb to check
4093 * The head on skbs build around a head frag can be removed if they are
4094 * not cloned. This function returns true if the skb head is locked down
4095 * due to either being allocated via kmalloc, or by being a clone with
4096 * multiple references to the head.
4098 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4100 return !skb->head_frag || skb_cloned(skb);
4104 * skb_gso_network_seglen - Return length of individual segments of a gso packet
4108 * skb_gso_network_seglen is used to determine the real size of the
4109 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
4111 * The MAC/L2 header is not accounted for.
4113 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
4115 unsigned int hdr_len = skb_transport_header(skb) -
4116 skb_network_header(skb);
4117 return hdr_len + skb_gso_transport_seglen(skb);
4120 /* Local Checksum Offload.
4121 * Compute outer checksum based on the assumption that the
4122 * inner checksum will be offloaded later.
4123 * See Documentation/networking/checksum-offloads.txt for
4124 * explanation of how this works.
4125 * Fill in outer checksum adjustment (e.g. with sum of outer
4126 * pseudo-header) before calling.
4127 * Also ensure that inner checksum is in linear data area.
4129 static inline __wsum lco_csum(struct sk_buff *skb)
4131 unsigned char *csum_start = skb_checksum_start(skb);
4132 unsigned char *l4_hdr = skb_transport_header(skb);
4135 /* Start with complement of inner checksum adjustment */
4136 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4139 /* Add in checksum of our headers (incl. outer checksum
4140 * adjustment filled in by caller) and return result.
4142 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4145 #endif /* __KERNEL__ */
4146 #endif /* _LINUX_SKBUFF_H */