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/compiler.h>
19 #include <linux/time.h>
20 #include <linux/bug.h>
21 #include <linux/cache.h>
22 #include <linux/rbtree.h>
23 #include <linux/socket.h>
24 #include <linux/refcount.h>
26 #include <linux/atomic.h>
27 #include <asm/types.h>
28 #include <linux/spinlock.h>
29 #include <linux/net.h>
30 #include <linux/textsearch.h>
31 #include <net/checksum.h>
32 #include <linux/rcupdate.h>
33 #include <linux/hrtimer.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/netdev_features.h>
36 #include <linux/sched.h>
37 #include <linux/sched/clock.h>
38 #include <net/flow_dissector.h>
39 #include <linux/splice.h>
40 #include <linux/in6.h>
41 #include <linux/if_packet.h>
44 /* The interface for checksum offload between the stack and networking drivers
47 * A. IP checksum related features
49 * Drivers advertise checksum offload capabilities in the features of a device.
50 * From the stack's point of view these are capabilities offered by the driver,
51 * a driver typically only advertises features that it is capable of offloading
54 * The checksum related features are:
56 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
57 * IP (one's complement) checksum for any combination
58 * of protocols or protocol layering. The checksum is
59 * computed and set in a packet per the CHECKSUM_PARTIAL
60 * interface (see below).
62 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
63 * TCP or UDP packets over IPv4. These are specifically
64 * unencapsulated packets of the form IPv4|TCP or
65 * IPv4|UDP where the Protocol field in the IPv4 header
66 * is TCP or UDP. The IPv4 header may contain IP options
67 * This feature cannot be set in features for a device
68 * with NETIF_F_HW_CSUM also set. This feature is being
69 * DEPRECATED (see below).
71 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
72 * TCP or UDP packets over IPv6. These are specifically
73 * unencapsulated packets of the form IPv6|TCP or
74 * IPv4|UDP where the Next Header field in the IPv6
75 * header is either TCP or UDP. IPv6 extension headers
76 * are not supported with this feature. This feature
77 * cannot be set in features for a device with
78 * NETIF_F_HW_CSUM also set. This feature is being
79 * DEPRECATED (see below).
81 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
82 * This flag is used only used to disable the RX checksum
83 * feature for a device. The stack will accept receive
84 * checksum indication in packets received on a device
85 * regardless of whether NETIF_F_RXCSUM is set.
87 * B. Checksumming of received packets by device. Indication of checksum
88 * verification is in set skb->ip_summed. Possible values are:
92 * Device did not checksum this packet e.g. due to lack of capabilities.
93 * The packet contains full (though not verified) checksum in packet but
94 * not in skb->csum. Thus, skb->csum is undefined in this case.
96 * CHECKSUM_UNNECESSARY:
98 * The hardware you're dealing with doesn't calculate the full checksum
99 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
100 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
101 * if their checksums are okay. skb->csum is still undefined in this case
102 * though. A driver or device must never modify the checksum field in the
103 * packet even if checksum is verified.
105 * CHECKSUM_UNNECESSARY is applicable to following protocols:
106 * TCP: IPv6 and IPv4.
107 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
108 * zero UDP checksum for either IPv4 or IPv6, the networking stack
109 * may perform further validation in this case.
110 * GRE: only if the checksum is present in the header.
111 * SCTP: indicates the CRC in SCTP header has been validated.
112 * FCOE: indicates the CRC in FC frame has been validated.
114 * skb->csum_level indicates the number of consecutive checksums found in
115 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
116 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
117 * and a device is able to verify the checksums for UDP (possibly zero),
118 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
119 * two. If the device were only able to verify the UDP checksum and not
120 * GRE, either because it doesn't support GRE checksum of because GRE
121 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
122 * not considered in this case).
126 * This is the most generic way. The device supplied checksum of the _whole_
127 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
128 * hardware doesn't need to parse L3/L4 headers to implement this.
131 * - Even if device supports only some protocols, but is able to produce
132 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
133 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
137 * A checksum is set up to be offloaded to a device as described in the
138 * output description for CHECKSUM_PARTIAL. This may occur on a packet
139 * received directly from another Linux OS, e.g., a virtualized Linux kernel
140 * on the same host, or it may be set in the input path in GRO or remote
141 * checksum offload. For the purposes of checksum verification, the checksum
142 * referred to by skb->csum_start + skb->csum_offset and any preceding
143 * checksums in the packet are considered verified. Any checksums in the
144 * packet that are after the checksum being offloaded are not considered to
147 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
148 * in the skb->ip_summed for a packet. Values are:
152 * The driver is required to checksum the packet as seen by hard_start_xmit()
153 * from skb->csum_start up to the end, and to record/write the checksum at
154 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
155 * csum_start and csum_offset values are valid values given the length and
156 * offset of the packet, however they should not attempt to validate that the
157 * checksum refers to a legitimate transport layer checksum-- it is the
158 * purview of the stack to validate that csum_start and csum_offset are set
161 * When the stack requests checksum offload for a packet, the driver MUST
162 * ensure that the checksum is set correctly. A driver can either offload the
163 * checksum calculation to the device, or call skb_checksum_help (in the case
164 * that the device does not support offload for a particular checksum).
166 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
167 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
168 * checksum offload capability.
169 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
170 * on network device checksumming capabilities: if a packet does not match
171 * them, skb_checksum_help or skb_crc32c_help (depending on the value of
172 * csum_not_inet, see item D.) is called to resolve the checksum.
176 * The skb was already checksummed by the protocol, or a checksum is not
179 * CHECKSUM_UNNECESSARY:
181 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
185 * Not used in checksum output. If a driver observes a packet with this value
186 * set in skbuff, if should treat as CHECKSUM_NONE being set.
188 * D. Non-IP checksum (CRC) offloads
190 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
191 * offloading the SCTP CRC in a packet. To perform this offload the stack
192 * will set set csum_start and csum_offset accordingly, set ip_summed to
193 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
194 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
195 * A driver that supports both IP checksum offload and SCTP CRC32c offload
196 * must verify which offload is configured for a packet by testing the
197 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
198 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
200 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
201 * offloading the FCOE CRC in a packet. To perform this offload the stack
202 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
203 * accordingly. Note the there is no indication in the skbuff that the
204 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
205 * both IP checksum offload and FCOE CRC offload must verify which offload
206 * is configured for a packet presumably by inspecting packet headers.
208 * E. Checksumming on output with GSO.
210 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
211 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
212 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
213 * part of the GSO operation is implied. If a checksum is being offloaded
214 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
215 * are set to refer to the outermost checksum being offload (two offloaded
216 * checksums are possible with UDP encapsulation).
219 /* Don't change this without changing skb_csum_unnecessary! */
220 #define CHECKSUM_NONE 0
221 #define CHECKSUM_UNNECESSARY 1
222 #define CHECKSUM_COMPLETE 2
223 #define CHECKSUM_PARTIAL 3
225 /* Maximum value in skb->csum_level */
226 #define SKB_MAX_CSUM_LEVEL 3
228 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
229 #define SKB_WITH_OVERHEAD(X) \
230 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
231 #define SKB_MAX_ORDER(X, ORDER) \
232 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
233 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
234 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
236 /* return minimum truesize of one skb containing X bytes of data */
237 #define SKB_TRUESIZE(X) ((X) + \
238 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
239 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
243 struct pipe_inode_info;
247 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
248 struct nf_conntrack {
253 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
254 struct nf_bridge_info {
257 BRNF_PROTO_UNCHANGED,
265 struct net_device *physindev;
267 /* always valid & non-NULL from FORWARD on, for physdev match */
268 struct net_device *physoutdev;
270 /* prerouting: detect dnat in orig/reply direction */
272 struct in6_addr ipv6_daddr;
274 /* after prerouting + nat detected: store original source
275 * mac since neigh resolution overwrites it, only used while
276 * skb is out in neigh layer.
278 char neigh_header[8];
283 struct sk_buff_head {
284 /* These two members must be first. */
285 struct sk_buff *next;
286 struct sk_buff *prev;
294 /* To allow 64K frame to be packed as single skb without frag_list we
295 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
296 * buffers which do not start on a page boundary.
298 * Since GRO uses frags we allocate at least 16 regardless of page
301 #if (65536/PAGE_SIZE + 1) < 16
302 #define MAX_SKB_FRAGS 16UL
304 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
306 extern int sysctl_max_skb_frags;
308 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
309 * segment using its current segmentation instead.
311 #define GSO_BY_FRAGS 0xFFFF
313 typedef struct skb_frag_struct skb_frag_t;
315 struct skb_frag_struct {
319 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
328 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
333 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
338 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
343 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
348 static inline bool skb_frag_must_loop(struct page *p)
350 #if defined(CONFIG_HIGHMEM)
358 * skb_frag_foreach_page - loop over pages in a fragment
360 * @f: skb frag to operate on
361 * @f_off: offset from start of f->page.p
362 * @f_len: length from f_off to loop over
363 * @p: (temp var) current page
364 * @p_off: (temp var) offset from start of current page,
365 * non-zero only on first page.
366 * @p_len: (temp var) length in current page,
367 * < PAGE_SIZE only on first and last page.
368 * @copied: (temp var) length so far, excluding current p_len.
370 * A fragment can hold a compound page, in which case per-page
371 * operations, notably kmap_atomic, must be called for each
374 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
375 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
376 p_off = (f_off) & (PAGE_SIZE - 1), \
377 p_len = skb_frag_must_loop(p) ? \
378 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
381 copied += p_len, p++, p_off = 0, \
382 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
384 #define HAVE_HW_TIME_STAMP
387 * struct skb_shared_hwtstamps - hardware time stamps
388 * @hwtstamp: hardware time stamp transformed into duration
389 * since arbitrary point in time
391 * Software time stamps generated by ktime_get_real() are stored in
394 * hwtstamps can only be compared against other hwtstamps from
397 * This structure is attached to packets as part of the
398 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
400 struct skb_shared_hwtstamps {
404 /* Definitions for tx_flags in struct skb_shared_info */
406 /* generate hardware time stamp */
407 SKBTX_HW_TSTAMP = 1 << 0,
409 /* generate software time stamp when queueing packet to NIC */
410 SKBTX_SW_TSTAMP = 1 << 1,
412 /* device driver is going to provide hardware time stamp */
413 SKBTX_IN_PROGRESS = 1 << 2,
415 /* device driver supports TX zero-copy buffers */
416 SKBTX_DEV_ZEROCOPY = 1 << 3,
418 /* generate wifi status information (where possible) */
419 SKBTX_WIFI_STATUS = 1 << 4,
421 /* This indicates at least one fragment might be overwritten
422 * (as in vmsplice(), sendfile() ...)
423 * If we need to compute a TX checksum, we'll need to copy
424 * all frags to avoid possible bad checksum
426 SKBTX_SHARED_FRAG = 1 << 5,
428 /* generate software time stamp when entering packet scheduling */
429 SKBTX_SCHED_TSTAMP = 1 << 6,
432 #define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
433 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
435 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
438 * The callback notifies userspace to release buffers when skb DMA is done in
439 * lower device, the skb last reference should be 0 when calling this.
440 * The zerocopy_success argument is true if zero copy transmit occurred,
441 * false on data copy or out of memory error caused by data copy attempt.
442 * The ctx field is used to track device context.
443 * The desc field is used to track userspace buffer index.
446 void (*callback)(struct ubuf_info *, bool zerocopy_success);
462 struct user_struct *user;
467 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
469 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
470 void mm_unaccount_pinned_pages(struct mmpin *mmp);
472 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
473 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
474 struct ubuf_info *uarg);
476 static inline void sock_zerocopy_get(struct ubuf_info *uarg)
478 refcount_inc(&uarg->refcnt);
481 void sock_zerocopy_put(struct ubuf_info *uarg);
482 void sock_zerocopy_put_abort(struct ubuf_info *uarg);
484 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
486 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
487 struct msghdr *msg, int len,
488 struct ubuf_info *uarg);
490 /* This data is invariant across clones and lives at
491 * the end of the header data, ie. at skb->end.
493 struct skb_shared_info {
498 unsigned short gso_size;
499 /* Warning: this field is not always filled in (UFO)! */
500 unsigned short gso_segs;
501 struct sk_buff *frag_list;
502 struct skb_shared_hwtstamps hwtstamps;
503 unsigned int gso_type;
507 * Warning : all fields before dataref are cleared in __alloc_skb()
511 /* Intermediate layers must ensure that destructor_arg
512 * remains valid until skb destructor */
513 void * destructor_arg;
515 /* must be last field, see pskb_expand_head() */
516 skb_frag_t frags[MAX_SKB_FRAGS];
519 /* We divide dataref into two halves. The higher 16 bits hold references
520 * to the payload part of skb->data. The lower 16 bits hold references to
521 * the entire skb->data. A clone of a headerless skb holds the length of
522 * the header in skb->hdr_len.
524 * All users must obey the rule that the skb->data reference count must be
525 * greater than or equal to the payload reference count.
527 * Holding a reference to the payload part means that the user does not
528 * care about modifications to the header part of skb->data.
530 #define SKB_DATAREF_SHIFT 16
531 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
535 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
536 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
537 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
541 SKB_GSO_TCPV4 = 1 << 0,
543 /* This indicates the skb is from an untrusted source. */
544 SKB_GSO_DODGY = 1 << 1,
546 /* This indicates the tcp segment has CWR set. */
547 SKB_GSO_TCP_ECN = 1 << 2,
549 SKB_GSO_TCP_FIXEDID = 1 << 3,
551 SKB_GSO_TCPV6 = 1 << 4,
553 SKB_GSO_FCOE = 1 << 5,
555 SKB_GSO_GRE = 1 << 6,
557 SKB_GSO_GRE_CSUM = 1 << 7,
559 SKB_GSO_IPXIP4 = 1 << 8,
561 SKB_GSO_IPXIP6 = 1 << 9,
563 SKB_GSO_UDP_TUNNEL = 1 << 10,
565 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
567 SKB_GSO_PARTIAL = 1 << 12,
569 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
571 SKB_GSO_SCTP = 1 << 14,
573 SKB_GSO_ESP = 1 << 15,
575 SKB_GSO_UDP = 1 << 16,
577 SKB_GSO_UDP_L4 = 1 << 17,
580 #if BITS_PER_LONG > 32
581 #define NET_SKBUFF_DATA_USES_OFFSET 1
584 #ifdef NET_SKBUFF_DATA_USES_OFFSET
585 typedef unsigned int sk_buff_data_t;
587 typedef unsigned char *sk_buff_data_t;
591 * struct sk_buff - socket buffer
592 * @next: Next buffer in list
593 * @prev: Previous buffer in list
594 * @tstamp: Time we arrived/left
595 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
596 * @sk: Socket we are owned by
597 * @dev: Device we arrived on/are leaving by
598 * @cb: Control buffer. Free for use by every layer. Put private vars here
599 * @_skb_refdst: destination entry (with norefcount bit)
600 * @sp: the security path, used for xfrm
601 * @len: Length of actual data
602 * @data_len: Data length
603 * @mac_len: Length of link layer header
604 * @hdr_len: writable header length of cloned skb
605 * @csum: Checksum (must include start/offset pair)
606 * @csum_start: Offset from skb->head where checksumming should start
607 * @csum_offset: Offset from csum_start where checksum should be stored
608 * @priority: Packet queueing priority
609 * @ignore_df: allow local fragmentation
610 * @cloned: Head may be cloned (check refcnt to be sure)
611 * @ip_summed: Driver fed us an IP checksum
612 * @nohdr: Payload reference only, must not modify header
613 * @pkt_type: Packet class
614 * @fclone: skbuff clone status
615 * @ipvs_property: skbuff is owned by ipvs
616 * @tc_skip_classify: do not classify packet. set by IFB device
617 * @tc_at_ingress: used within tc_classify to distinguish in/egress
618 * @tc_redirected: packet was redirected by a tc action
619 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
620 * @peeked: this packet has been seen already, so stats have been
621 * done for it, don't do them again
622 * @nf_trace: netfilter packet trace flag
623 * @protocol: Packet protocol from driver
624 * @destructor: Destruct function
625 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
626 * @_nfct: Associated connection, if any (with nfctinfo bits)
627 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
628 * @skb_iif: ifindex of device we arrived on
629 * @tc_index: Traffic control index
630 * @hash: the packet hash
631 * @queue_mapping: Queue mapping for multiqueue devices
632 * @xmit_more: More SKBs are pending for this queue
633 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
634 * @ndisc_nodetype: router type (from link layer)
635 * @ooo_okay: allow the mapping of a socket to a queue to be changed
636 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
638 * @sw_hash: indicates hash was computed in software stack
639 * @wifi_acked_valid: wifi_acked was set
640 * @wifi_acked: whether frame was acked on wifi or not
641 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
642 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
643 * @dst_pending_confirm: need to confirm neighbour
644 * @decrypted: Decrypted SKB
645 * @napi_id: id of the NAPI struct this skb came from
646 * @secmark: security marking
647 * @mark: Generic packet mark
648 * @vlan_proto: vlan encapsulation protocol
649 * @vlan_tci: vlan tag control information
650 * @inner_protocol: Protocol (encapsulation)
651 * @inner_transport_header: Inner transport layer header (encapsulation)
652 * @inner_network_header: Network layer header (encapsulation)
653 * @inner_mac_header: Link layer header (encapsulation)
654 * @transport_header: Transport layer header
655 * @network_header: Network layer header
656 * @mac_header: Link layer header
657 * @tail: Tail pointer
659 * @head: Head of buffer
660 * @data: Data head pointer
661 * @truesize: Buffer size
662 * @users: User count - see {datagram,tcp}.c
668 /* These two members must be first. */
669 struct sk_buff *next;
670 struct sk_buff *prev;
673 struct net_device *dev;
674 /* Some protocols might use this space to store information,
675 * while device pointer would be NULL.
676 * UDP receive path is one user.
678 unsigned long dev_scratch;
681 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
682 struct list_head list;
687 int ip_defrag_offset;
695 * This is the control buffer. It is free to use for every
696 * layer. Please put your private variables there. If you
697 * want to keep them across layers you have to do a skb_clone()
698 * first. This is owned by whoever has the skb queued ATM.
700 char cb[48] __aligned(8);
704 unsigned long _skb_refdst;
705 void (*destructor)(struct sk_buff *skb);
707 struct list_head tcp_tsorted_anchor;
713 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
716 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
717 struct nf_bridge_info *nf_bridge;
724 /* Following fields are _not_ copied in __copy_skb_header()
725 * Note that queue_mapping is here mostly to fill a hole.
729 /* if you move cloned around you also must adapt those constants */
730 #ifdef __BIG_ENDIAN_BITFIELD
731 #define CLONED_MASK (1 << 7)
733 #define CLONED_MASK 1
735 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
737 __u8 __cloned_offset[0];
746 /* fields enclosed in headers_start/headers_end are copied
747 * using a single memcpy() in __copy_skb_header()
750 __u32 headers_start[0];
753 /* if you move pkt_type around you also must adapt those constants */
754 #ifdef __BIG_ENDIAN_BITFIELD
755 #define PKT_TYPE_MAX (7 << 5)
757 #define PKT_TYPE_MAX 7
759 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
761 __u8 __pkt_type_offset[0];
770 __u8 wifi_acked_valid:1;
773 /* Indicates the inner headers are valid in the skbuff. */
774 __u8 encapsulation:1;
775 __u8 encap_hdr_csum:1;
778 __u8 csum_complete_sw:1;
780 __u8 csum_not_inet:1;
781 __u8 dst_pending_confirm:1;
782 #ifdef CONFIG_IPV6_NDISC_NODETYPE
783 __u8 ndisc_nodetype:2;
785 __u8 ipvs_property:1;
787 __u8 inner_protocol_type:1;
788 __u8 remcsum_offload:1;
789 #ifdef CONFIG_NET_SWITCHDEV
790 __u8 offload_fwd_mark:1;
791 __u8 offload_mr_fwd_mark:1;
793 #ifdef CONFIG_NET_CLS_ACT
794 __u8 tc_skip_classify:1;
795 __u8 tc_at_ingress:1;
796 __u8 tc_redirected:1;
797 __u8 tc_from_ingress:1;
799 #ifdef CONFIG_TLS_DEVICE
803 #ifdef CONFIG_NET_SCHED
804 __u16 tc_index; /* traffic control index */
819 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
821 unsigned int napi_id;
822 unsigned int sender_cpu;
825 #ifdef CONFIG_NETWORK_SECMARK
831 __u32 reserved_tailroom;
835 __be16 inner_protocol;
839 __u16 inner_transport_header;
840 __u16 inner_network_header;
841 __u16 inner_mac_header;
844 __u16 transport_header;
845 __u16 network_header;
849 __u32 headers_end[0];
852 /* These elements must be at the end, see alloc_skb() for details. */
857 unsigned int truesize;
863 * Handling routines are only of interest to the kernel
866 #define SKB_ALLOC_FCLONE 0x01
867 #define SKB_ALLOC_RX 0x02
868 #define SKB_ALLOC_NAPI 0x04
870 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
871 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
873 return unlikely(skb->pfmemalloc);
877 * skb might have a dst pointer attached, refcounted or not.
878 * _skb_refdst low order bit is set if refcount was _not_ taken
880 #define SKB_DST_NOREF 1UL
881 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
883 #define SKB_NFCT_PTRMASK ~(7UL)
885 * skb_dst - returns skb dst_entry
888 * Returns skb dst_entry, regardless of reference taken or not.
890 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
892 /* If refdst was not refcounted, check we still are in a
893 * rcu_read_lock section
895 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
896 !rcu_read_lock_held() &&
897 !rcu_read_lock_bh_held());
898 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
902 * skb_dst_set - sets skb dst
906 * Sets skb dst, assuming a reference was taken on dst and should
907 * be released by skb_dst_drop()
909 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
911 skb->_skb_refdst = (unsigned long)dst;
915 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
919 * Sets skb dst, assuming a reference was not taken on dst.
920 * If dst entry is cached, we do not take reference and dst_release
921 * will be avoided by refdst_drop. If dst entry is not cached, we take
922 * reference, so that last dst_release can destroy the dst immediately.
924 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
926 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
927 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
931 * skb_dst_is_noref - Test if skb dst isn't refcounted
934 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
936 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
939 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
941 return (struct rtable *)skb_dst(skb);
944 /* For mangling skb->pkt_type from user space side from applications
945 * such as nft, tc, etc, we only allow a conservative subset of
946 * possible pkt_types to be set.
948 static inline bool skb_pkt_type_ok(u32 ptype)
950 return ptype <= PACKET_OTHERHOST;
953 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
955 #ifdef CONFIG_NET_RX_BUSY_POLL
962 /* decrement the reference count and return true if we can free the skb */
963 static inline bool skb_unref(struct sk_buff *skb)
967 if (likely(refcount_read(&skb->users) == 1))
969 else if (likely(!refcount_dec_and_test(&skb->users)))
975 void skb_release_head_state(struct sk_buff *skb);
976 void kfree_skb(struct sk_buff *skb);
977 void kfree_skb_list(struct sk_buff *segs);
978 void skb_tx_error(struct sk_buff *skb);
979 void consume_skb(struct sk_buff *skb);
980 void __consume_stateless_skb(struct sk_buff *skb);
981 void __kfree_skb(struct sk_buff *skb);
982 extern struct kmem_cache *skbuff_head_cache;
984 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
985 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
986 bool *fragstolen, int *delta_truesize);
988 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
990 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
991 struct sk_buff *build_skb(void *data, unsigned int frag_size);
992 static inline struct sk_buff *alloc_skb(unsigned int size,
995 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
998 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
999 unsigned long data_len,
1004 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1005 struct sk_buff_fclones {
1006 struct sk_buff skb1;
1008 struct sk_buff skb2;
1010 refcount_t fclone_ref;
1014 * skb_fclone_busy - check if fclone is busy
1018 * Returns true if skb is a fast clone, and its clone is not freed.
1019 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1020 * so we also check that this didnt happen.
1022 static inline bool skb_fclone_busy(const struct sock *sk,
1023 const struct sk_buff *skb)
1025 const struct sk_buff_fclones *fclones;
1027 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1029 return skb->fclone == SKB_FCLONE_ORIG &&
1030 refcount_read(&fclones->fclone_ref) > 1 &&
1031 fclones->skb2.sk == sk;
1034 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1037 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1040 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1041 void skb_headers_offset_update(struct sk_buff *skb, int off);
1042 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1043 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1044 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1045 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1046 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1047 gfp_t gfp_mask, bool fclone);
1048 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1051 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1054 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1055 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1056 unsigned int headroom);
1057 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1058 int newtailroom, gfp_t priority);
1059 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1060 int offset, int len);
1061 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1062 int offset, int len);
1063 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1064 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1067 * skb_pad - zero pad the tail of an skb
1068 * @skb: buffer to pad
1069 * @pad: space to pad
1071 * Ensure that a buffer is followed by a padding area that is zero
1072 * filled. Used by network drivers which may DMA or transfer data
1073 * beyond the buffer end onto the wire.
1075 * May return error in out of memory cases. The skb is freed on error.
1077 static inline int skb_pad(struct sk_buff *skb, int pad)
1079 return __skb_pad(skb, pad, true);
1081 #define dev_kfree_skb(a) consume_skb(a)
1083 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1084 int getfrag(void *from, char *to, int offset,
1085 int len, int odd, struct sk_buff *skb),
1086 void *from, int length);
1088 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1089 int offset, size_t size);
1091 struct skb_seq_state {
1095 __u32 stepped_offset;
1096 struct sk_buff *root_skb;
1097 struct sk_buff *cur_skb;
1101 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1102 unsigned int to, struct skb_seq_state *st);
1103 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1104 struct skb_seq_state *st);
1105 void skb_abort_seq_read(struct skb_seq_state *st);
1107 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1108 unsigned int to, struct ts_config *config);
1111 * Packet hash types specify the type of hash in skb_set_hash.
1113 * Hash types refer to the protocol layer addresses which are used to
1114 * construct a packet's hash. The hashes are used to differentiate or identify
1115 * flows of the protocol layer for the hash type. Hash types are either
1116 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1118 * Properties of hashes:
1120 * 1) Two packets in different flows have different hash values
1121 * 2) Two packets in the same flow should have the same hash value
1123 * A hash at a higher layer is considered to be more specific. A driver should
1124 * set the most specific hash possible.
1126 * A driver cannot indicate a more specific hash than the layer at which a hash
1127 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1129 * A driver may indicate a hash level which is less specific than the
1130 * actual layer the hash was computed on. For instance, a hash computed
1131 * at L4 may be considered an L3 hash. This should only be done if the
1132 * driver can't unambiguously determine that the HW computed the hash at
1133 * the higher layer. Note that the "should" in the second property above
1136 enum pkt_hash_types {
1137 PKT_HASH_TYPE_NONE, /* Undefined type */
1138 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1139 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1140 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1143 static inline void skb_clear_hash(struct sk_buff *skb)
1150 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1153 skb_clear_hash(skb);
1157 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1159 skb->l4_hash = is_l4;
1160 skb->sw_hash = is_sw;
1165 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1167 /* Used by drivers to set hash from HW */
1168 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1172 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1174 __skb_set_hash(skb, hash, true, is_l4);
1177 void __skb_get_hash(struct sk_buff *skb);
1178 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1179 u32 skb_get_poff(const struct sk_buff *skb);
1180 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1181 const struct flow_keys_basic *keys, int hlen);
1182 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1183 void *data, int hlen_proto);
1185 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1186 int thoff, u8 ip_proto)
1188 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1191 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1192 const struct flow_dissector_key *key,
1193 unsigned int key_count);
1195 bool __skb_flow_dissect(const struct sk_buff *skb,
1196 struct flow_dissector *flow_dissector,
1197 void *target_container,
1198 void *data, __be16 proto, int nhoff, int hlen,
1199 unsigned int flags);
1201 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1202 struct flow_dissector *flow_dissector,
1203 void *target_container, unsigned int flags)
1205 return __skb_flow_dissect(skb, flow_dissector, target_container,
1206 NULL, 0, 0, 0, flags);
1209 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1210 struct flow_keys *flow,
1213 memset(flow, 0, sizeof(*flow));
1214 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1215 NULL, 0, 0, 0, flags);
1219 skb_flow_dissect_flow_keys_basic(const struct sk_buff *skb,
1220 struct flow_keys_basic *flow, void *data,
1221 __be16 proto, int nhoff, int hlen,
1224 memset(flow, 0, sizeof(*flow));
1225 return __skb_flow_dissect(skb, &flow_keys_basic_dissector, flow,
1226 data, proto, nhoff, hlen, flags);
1230 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1231 struct flow_dissector *flow_dissector,
1232 void *target_container);
1234 static inline __u32 skb_get_hash(struct sk_buff *skb)
1236 if (!skb->l4_hash && !skb->sw_hash)
1237 __skb_get_hash(skb);
1242 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1244 if (!skb->l4_hash && !skb->sw_hash) {
1245 struct flow_keys keys;
1246 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1248 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1254 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1255 const siphash_key_t *perturb);
1257 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1262 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1264 to->hash = from->hash;
1265 to->sw_hash = from->sw_hash;
1266 to->l4_hash = from->l4_hash;
1269 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1270 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1272 return skb->head + skb->end;
1275 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1280 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1285 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1287 return skb->end - skb->head;
1292 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1294 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1296 return &skb_shinfo(skb)->hwtstamps;
1299 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1301 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1303 return is_zcopy ? skb_uarg(skb) : NULL;
1306 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg)
1308 if (skb && uarg && !skb_zcopy(skb)) {
1309 sock_zerocopy_get(uarg);
1310 skb_shinfo(skb)->destructor_arg = uarg;
1311 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1315 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1317 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1318 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1321 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1323 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1326 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1328 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1331 /* Release a reference on a zerocopy structure */
1332 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1334 struct ubuf_info *uarg = skb_zcopy(skb);
1337 if (skb_zcopy_is_nouarg(skb)) {
1338 /* no notification callback */
1339 } else if (uarg->callback == sock_zerocopy_callback) {
1340 uarg->zerocopy = uarg->zerocopy && zerocopy;
1341 sock_zerocopy_put(uarg);
1343 uarg->callback(uarg, zerocopy);
1346 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1350 /* Abort a zerocopy operation and revert zckey on error in send syscall */
1351 static inline void skb_zcopy_abort(struct sk_buff *skb)
1353 struct ubuf_info *uarg = skb_zcopy(skb);
1356 sock_zerocopy_put_abort(uarg);
1357 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1361 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1366 /* Iterate through singly-linked GSO fragments of an skb. */
1367 #define skb_list_walk_safe(first, skb, next_skb) \
1368 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1369 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1371 static inline void skb_list_del_init(struct sk_buff *skb)
1373 __list_del_entry(&skb->list);
1374 skb_mark_not_on_list(skb);
1378 * skb_queue_empty - check if a queue is empty
1381 * Returns true if the queue is empty, false otherwise.
1383 static inline int skb_queue_empty(const struct sk_buff_head *list)
1385 return list->next == (const struct sk_buff *) list;
1389 * skb_queue_empty_lockless - check if a queue is empty
1392 * Returns true if the queue is empty, false otherwise.
1393 * This variant can be used in lockless contexts.
1395 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1397 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1402 * skb_queue_is_last - check if skb is the last entry in the queue
1406 * Returns true if @skb is the last buffer on the list.
1408 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1409 const struct sk_buff *skb)
1411 return skb->next == (const struct sk_buff *) list;
1415 * skb_queue_is_first - check if skb is the first entry in the queue
1419 * Returns true if @skb is the first buffer on the list.
1421 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1422 const struct sk_buff *skb)
1424 return skb->prev == (const struct sk_buff *) list;
1428 * skb_queue_next - return the next packet in the queue
1430 * @skb: current buffer
1432 * Return the next packet in @list after @skb. It is only valid to
1433 * call this if skb_queue_is_last() evaluates to false.
1435 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1436 const struct sk_buff *skb)
1438 /* This BUG_ON may seem severe, but if we just return then we
1439 * are going to dereference garbage.
1441 BUG_ON(skb_queue_is_last(list, skb));
1446 * skb_queue_prev - return the prev packet in the queue
1448 * @skb: current buffer
1450 * Return the prev packet in @list before @skb. It is only valid to
1451 * call this if skb_queue_is_first() evaluates to false.
1453 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1454 const struct sk_buff *skb)
1456 /* This BUG_ON may seem severe, but if we just return then we
1457 * are going to dereference garbage.
1459 BUG_ON(skb_queue_is_first(list, skb));
1464 * skb_get - reference buffer
1465 * @skb: buffer to reference
1467 * Makes another reference to a socket buffer and returns a pointer
1470 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1472 refcount_inc(&skb->users);
1477 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1481 * skb_cloned - is the buffer a clone
1482 * @skb: buffer to check
1484 * Returns true if the buffer was generated with skb_clone() and is
1485 * one of multiple shared copies of the buffer. Cloned buffers are
1486 * shared data so must not be written to under normal circumstances.
1488 static inline int skb_cloned(const struct sk_buff *skb)
1490 return skb->cloned &&
1491 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1494 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1496 might_sleep_if(gfpflags_allow_blocking(pri));
1498 if (skb_cloned(skb))
1499 return pskb_expand_head(skb, 0, 0, pri);
1505 * skb_header_cloned - is the header a clone
1506 * @skb: buffer to check
1508 * Returns true if modifying the header part of the buffer requires
1509 * the data to be copied.
1511 static inline int skb_header_cloned(const struct sk_buff *skb)
1518 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1519 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1520 return dataref != 1;
1523 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1525 might_sleep_if(gfpflags_allow_blocking(pri));
1527 if (skb_header_cloned(skb))
1528 return pskb_expand_head(skb, 0, 0, pri);
1534 * __skb_header_release - release reference to header
1535 * @skb: buffer to operate on
1537 static inline void __skb_header_release(struct sk_buff *skb)
1540 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1545 * skb_shared - is the buffer shared
1546 * @skb: buffer to check
1548 * Returns true if more than one person has a reference to this
1551 static inline int skb_shared(const struct sk_buff *skb)
1553 return refcount_read(&skb->users) != 1;
1557 * skb_share_check - check if buffer is shared and if so clone it
1558 * @skb: buffer to check
1559 * @pri: priority for memory allocation
1561 * If the buffer is shared the buffer is cloned and the old copy
1562 * drops a reference. A new clone with a single reference is returned.
1563 * If the buffer is not shared the original buffer is returned. When
1564 * being called from interrupt status or with spinlocks held pri must
1567 * NULL is returned on a memory allocation failure.
1569 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1571 might_sleep_if(gfpflags_allow_blocking(pri));
1572 if (skb_shared(skb)) {
1573 struct sk_buff *nskb = skb_clone(skb, pri);
1585 * Copy shared buffers into a new sk_buff. We effectively do COW on
1586 * packets to handle cases where we have a local reader and forward
1587 * and a couple of other messy ones. The normal one is tcpdumping
1588 * a packet thats being forwarded.
1592 * skb_unshare - make a copy of a shared buffer
1593 * @skb: buffer to check
1594 * @pri: priority for memory allocation
1596 * If the socket buffer is a clone then this function creates a new
1597 * copy of the data, drops a reference count on the old copy and returns
1598 * the new copy with the reference count at 1. If the buffer is not a clone
1599 * the original buffer is returned. When called with a spinlock held or
1600 * from interrupt state @pri must be %GFP_ATOMIC
1602 * %NULL is returned on a memory allocation failure.
1604 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1607 might_sleep_if(gfpflags_allow_blocking(pri));
1608 if (skb_cloned(skb)) {
1609 struct sk_buff *nskb = skb_copy(skb, pri);
1611 /* Free our shared copy */
1622 * skb_peek - peek at the head of an &sk_buff_head
1623 * @list_: list to peek at
1625 * Peek an &sk_buff. Unlike most other operations you _MUST_
1626 * be careful with this one. A peek leaves the buffer on the
1627 * list and someone else may run off with it. You must hold
1628 * the appropriate locks or have a private queue to do this.
1630 * Returns %NULL for an empty list or a pointer to the head element.
1631 * The reference count is not incremented and the reference is therefore
1632 * volatile. Use with caution.
1634 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1636 struct sk_buff *skb = list_->next;
1638 if (skb == (struct sk_buff *)list_)
1644 * skb_peek_next - peek skb following the given one from a queue
1645 * @skb: skb to start from
1646 * @list_: list to peek at
1648 * Returns %NULL when the end of the list is met or a pointer to the
1649 * next element. The reference count is not incremented and the
1650 * reference is therefore volatile. Use with caution.
1652 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1653 const struct sk_buff_head *list_)
1655 struct sk_buff *next = skb->next;
1657 if (next == (struct sk_buff *)list_)
1663 * skb_peek_tail - peek at the tail of an &sk_buff_head
1664 * @list_: list to peek at
1666 * Peek an &sk_buff. Unlike most other operations you _MUST_
1667 * be careful with this one. A peek leaves the buffer on the
1668 * list and someone else may run off with it. You must hold
1669 * the appropriate locks or have a private queue to do this.
1671 * Returns %NULL for an empty list or a pointer to the tail element.
1672 * The reference count is not incremented and the reference is therefore
1673 * volatile. Use with caution.
1675 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1677 struct sk_buff *skb = READ_ONCE(list_->prev);
1679 if (skb == (struct sk_buff *)list_)
1686 * skb_queue_len - get queue length
1687 * @list_: list to measure
1689 * Return the length of an &sk_buff queue.
1691 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1697 * skb_queue_len_lockless - get queue length
1698 * @list_: list to measure
1700 * Return the length of an &sk_buff queue.
1701 * This variant can be used in lockless contexts.
1703 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
1705 return READ_ONCE(list_->qlen);
1709 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1710 * @list: queue to initialize
1712 * This initializes only the list and queue length aspects of
1713 * an sk_buff_head object. This allows to initialize the list
1714 * aspects of an sk_buff_head without reinitializing things like
1715 * the spinlock. It can also be used for on-stack sk_buff_head
1716 * objects where the spinlock is known to not be used.
1718 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1720 list->prev = list->next = (struct sk_buff *)list;
1725 * This function creates a split out lock class for each invocation;
1726 * this is needed for now since a whole lot of users of the skb-queue
1727 * infrastructure in drivers have different locking usage (in hardirq)
1728 * than the networking core (in softirq only). In the long run either the
1729 * network layer or drivers should need annotation to consolidate the
1730 * main types of usage into 3 classes.
1732 static inline void skb_queue_head_init(struct sk_buff_head *list)
1734 spin_lock_init(&list->lock);
1735 __skb_queue_head_init(list);
1738 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1739 struct lock_class_key *class)
1741 skb_queue_head_init(list);
1742 lockdep_set_class(&list->lock, class);
1746 * Insert an sk_buff on a list.
1748 * The "__skb_xxxx()" functions are the non-atomic ones that
1749 * can only be called with interrupts disabled.
1751 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1752 struct sk_buff_head *list);
1753 static inline void __skb_insert(struct sk_buff *newsk,
1754 struct sk_buff *prev, struct sk_buff *next,
1755 struct sk_buff_head *list)
1757 /* See skb_queue_empty_lockless() and skb_peek_tail()
1758 * for the opposite READ_ONCE()
1760 WRITE_ONCE(newsk->next, next);
1761 WRITE_ONCE(newsk->prev, prev);
1762 WRITE_ONCE(next->prev, newsk);
1763 WRITE_ONCE(prev->next, newsk);
1764 WRITE_ONCE(list->qlen, list->qlen + 1);
1767 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1768 struct sk_buff *prev,
1769 struct sk_buff *next)
1771 struct sk_buff *first = list->next;
1772 struct sk_buff *last = list->prev;
1774 WRITE_ONCE(first->prev, prev);
1775 WRITE_ONCE(prev->next, first);
1777 WRITE_ONCE(last->next, next);
1778 WRITE_ONCE(next->prev, last);
1782 * skb_queue_splice - join two skb lists, this is designed for stacks
1783 * @list: the new list to add
1784 * @head: the place to add it in the first list
1786 static inline void skb_queue_splice(const struct sk_buff_head *list,
1787 struct sk_buff_head *head)
1789 if (!skb_queue_empty(list)) {
1790 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1791 head->qlen += list->qlen;
1796 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1797 * @list: the new list to add
1798 * @head: the place to add it in the first list
1800 * The list at @list is reinitialised
1802 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1803 struct sk_buff_head *head)
1805 if (!skb_queue_empty(list)) {
1806 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1807 head->qlen += list->qlen;
1808 __skb_queue_head_init(list);
1813 * skb_queue_splice_tail - join two skb lists, each list being a queue
1814 * @list: the new list to add
1815 * @head: the place to add it in the first list
1817 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1818 struct sk_buff_head *head)
1820 if (!skb_queue_empty(list)) {
1821 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1822 head->qlen += list->qlen;
1827 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1828 * @list: the new list to add
1829 * @head: the place to add it in the first list
1831 * Each of the lists is a queue.
1832 * The list at @list is reinitialised
1834 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1835 struct sk_buff_head *head)
1837 if (!skb_queue_empty(list)) {
1838 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1839 head->qlen += list->qlen;
1840 __skb_queue_head_init(list);
1845 * __skb_queue_after - queue a buffer at the list head
1846 * @list: list to use
1847 * @prev: place after this buffer
1848 * @newsk: buffer to queue
1850 * Queue a buffer int the middle of a list. This function takes no locks
1851 * and you must therefore hold required locks before calling it.
1853 * A buffer cannot be placed on two lists at the same time.
1855 static inline void __skb_queue_after(struct sk_buff_head *list,
1856 struct sk_buff *prev,
1857 struct sk_buff *newsk)
1859 __skb_insert(newsk, prev, prev->next, list);
1862 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1863 struct sk_buff_head *list);
1865 static inline void __skb_queue_before(struct sk_buff_head *list,
1866 struct sk_buff *next,
1867 struct sk_buff *newsk)
1869 __skb_insert(newsk, next->prev, next, list);
1873 * __skb_queue_head - queue a buffer at the list head
1874 * @list: list to use
1875 * @newsk: buffer to queue
1877 * Queue a buffer at the start of a list. This function takes no locks
1878 * and you must therefore hold required locks before calling it.
1880 * A buffer cannot be placed on two lists at the same time.
1882 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1883 static inline void __skb_queue_head(struct sk_buff_head *list,
1884 struct sk_buff *newsk)
1886 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1890 * __skb_queue_tail - queue a buffer at the list tail
1891 * @list: list to use
1892 * @newsk: buffer to queue
1894 * Queue a buffer at the end of a list. This function takes no locks
1895 * and you must therefore hold required locks before calling it.
1897 * A buffer cannot be placed on two lists at the same time.
1899 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1900 static inline void __skb_queue_tail(struct sk_buff_head *list,
1901 struct sk_buff *newsk)
1903 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1907 * remove sk_buff from list. _Must_ be called atomically, and with
1910 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1911 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1913 struct sk_buff *next, *prev;
1915 WRITE_ONCE(list->qlen, list->qlen - 1);
1918 skb->next = skb->prev = NULL;
1919 WRITE_ONCE(next->prev, prev);
1920 WRITE_ONCE(prev->next, next);
1924 * __skb_dequeue - remove from the head of the queue
1925 * @list: list to dequeue from
1927 * Remove the head of the list. This function does not take any locks
1928 * so must be used with appropriate locks held only. The head item is
1929 * returned or %NULL if the list is empty.
1931 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1932 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1934 struct sk_buff *skb = skb_peek(list);
1936 __skb_unlink(skb, list);
1941 * __skb_dequeue_tail - remove from the tail of the queue
1942 * @list: list to dequeue from
1944 * Remove the tail of the list. This function does not take any locks
1945 * so must be used with appropriate locks held only. The tail item is
1946 * returned or %NULL if the list is empty.
1948 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1949 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1951 struct sk_buff *skb = skb_peek_tail(list);
1953 __skb_unlink(skb, list);
1958 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1960 return skb->data_len;
1963 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1965 return skb->len - skb->data_len;
1968 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
1970 unsigned int i, len = 0;
1972 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1973 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1977 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1979 return skb_headlen(skb) + __skb_pagelen(skb);
1983 * __skb_fill_page_desc - initialise a paged fragment in an skb
1984 * @skb: buffer containing fragment to be initialised
1985 * @i: paged fragment index to initialise
1986 * @page: the page to use for this fragment
1987 * @off: the offset to the data with @page
1988 * @size: the length of the data
1990 * Initialises the @i'th fragment of @skb to point to &size bytes at
1991 * offset @off within @page.
1993 * Does not take any additional reference on the fragment.
1995 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1996 struct page *page, int off, int size)
1998 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2001 * Propagate page pfmemalloc to the skb if we can. The problem is
2002 * that not all callers have unique ownership of the page but rely
2003 * on page_is_pfmemalloc doing the right thing(tm).
2005 frag->page.p = page;
2006 frag->page_offset = off;
2007 skb_frag_size_set(frag, size);
2009 page = compound_head(page);
2010 if (page_is_pfmemalloc(page))
2011 skb->pfmemalloc = true;
2015 * skb_fill_page_desc - initialise a paged fragment in an skb
2016 * @skb: buffer containing fragment to be initialised
2017 * @i: paged fragment index to initialise
2018 * @page: the page to use for this fragment
2019 * @off: the offset to the data with @page
2020 * @size: the length of the data
2022 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2023 * @skb to point to @size bytes at offset @off within @page. In
2024 * addition updates @skb such that @i is the last fragment.
2026 * Does not take any additional reference on the fragment.
2028 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2029 struct page *page, int off, int size)
2031 __skb_fill_page_desc(skb, i, page, off, size);
2032 skb_shinfo(skb)->nr_frags = i + 1;
2035 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2036 int size, unsigned int truesize);
2038 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2039 unsigned int truesize);
2041 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
2042 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
2043 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2045 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2046 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2048 return skb->head + skb->tail;
2051 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2053 skb->tail = skb->data - skb->head;
2056 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2058 skb_reset_tail_pointer(skb);
2059 skb->tail += offset;
2062 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2063 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2068 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2070 skb->tail = skb->data;
2073 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2075 skb->tail = skb->data + offset;
2078 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2081 * Add data to an sk_buff
2083 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2084 void *skb_put(struct sk_buff *skb, unsigned int len);
2085 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2087 void *tmp = skb_tail_pointer(skb);
2088 SKB_LINEAR_ASSERT(skb);
2094 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2096 void *tmp = __skb_put(skb, len);
2098 memset(tmp, 0, len);
2102 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2105 void *tmp = __skb_put(skb, len);
2107 memcpy(tmp, data, len);
2111 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2113 *(u8 *)__skb_put(skb, 1) = val;
2116 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2118 void *tmp = skb_put(skb, len);
2120 memset(tmp, 0, len);
2125 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2128 void *tmp = skb_put(skb, len);
2130 memcpy(tmp, data, len);
2135 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2137 *(u8 *)skb_put(skb, 1) = val;
2140 void *skb_push(struct sk_buff *skb, unsigned int len);
2141 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2148 void *skb_pull(struct sk_buff *skb, unsigned int len);
2149 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2152 BUG_ON(skb->len < skb->data_len);
2153 return skb->data += len;
2156 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2158 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2161 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2163 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2165 if (len > skb_headlen(skb) &&
2166 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2169 return skb->data += len;
2172 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2174 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2177 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2179 if (likely(len <= skb_headlen(skb)))
2181 if (unlikely(len > skb->len))
2183 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2186 void skb_condense(struct sk_buff *skb);
2189 * skb_headroom - bytes at buffer head
2190 * @skb: buffer to check
2192 * Return the number of bytes of free space at the head of an &sk_buff.
2194 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2196 return skb->data - skb->head;
2200 * skb_tailroom - bytes at buffer end
2201 * @skb: buffer to check
2203 * Return the number of bytes of free space at the tail of an sk_buff
2205 static inline int skb_tailroom(const struct sk_buff *skb)
2207 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2211 * skb_availroom - bytes at buffer end
2212 * @skb: buffer to check
2214 * Return the number of bytes of free space at the tail of an sk_buff
2215 * allocated by sk_stream_alloc()
2217 static inline int skb_availroom(const struct sk_buff *skb)
2219 if (skb_is_nonlinear(skb))
2222 return skb->end - skb->tail - skb->reserved_tailroom;
2226 * skb_reserve - adjust headroom
2227 * @skb: buffer to alter
2228 * @len: bytes to move
2230 * Increase the headroom of an empty &sk_buff by reducing the tail
2231 * room. This is only allowed for an empty buffer.
2233 static inline void skb_reserve(struct sk_buff *skb, int len)
2240 * skb_tailroom_reserve - adjust reserved_tailroom
2241 * @skb: buffer to alter
2242 * @mtu: maximum amount of headlen permitted
2243 * @needed_tailroom: minimum amount of reserved_tailroom
2245 * Set reserved_tailroom so that headlen can be as large as possible but
2246 * not larger than mtu and tailroom cannot be smaller than
2248 * The required headroom should already have been reserved before using
2251 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2252 unsigned int needed_tailroom)
2254 SKB_LINEAR_ASSERT(skb);
2255 if (mtu < skb_tailroom(skb) - needed_tailroom)
2256 /* use at most mtu */
2257 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2259 /* use up to all available space */
2260 skb->reserved_tailroom = needed_tailroom;
2263 #define ENCAP_TYPE_ETHER 0
2264 #define ENCAP_TYPE_IPPROTO 1
2266 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2269 skb->inner_protocol = protocol;
2270 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2273 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2276 skb->inner_ipproto = ipproto;
2277 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2280 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2282 skb->inner_mac_header = skb->mac_header;
2283 skb->inner_network_header = skb->network_header;
2284 skb->inner_transport_header = skb->transport_header;
2287 static inline void skb_reset_mac_len(struct sk_buff *skb)
2289 skb->mac_len = skb->network_header - skb->mac_header;
2292 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2295 return skb->head + skb->inner_transport_header;
2298 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2300 return skb_inner_transport_header(skb) - skb->data;
2303 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2305 skb->inner_transport_header = skb->data - skb->head;
2308 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2311 skb_reset_inner_transport_header(skb);
2312 skb->inner_transport_header += offset;
2315 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2317 return skb->head + skb->inner_network_header;
2320 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2322 skb->inner_network_header = skb->data - skb->head;
2325 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2328 skb_reset_inner_network_header(skb);
2329 skb->inner_network_header += offset;
2332 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2334 return skb->head + skb->inner_mac_header;
2337 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2339 skb->inner_mac_header = skb->data - skb->head;
2342 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2345 skb_reset_inner_mac_header(skb);
2346 skb->inner_mac_header += offset;
2348 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2350 return skb->transport_header != (typeof(skb->transport_header))~0U;
2353 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2355 return skb->head + skb->transport_header;
2358 static inline void skb_reset_transport_header(struct sk_buff *skb)
2360 skb->transport_header = skb->data - skb->head;
2363 static inline void skb_set_transport_header(struct sk_buff *skb,
2366 skb_reset_transport_header(skb);
2367 skb->transport_header += offset;
2370 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2372 return skb->head + skb->network_header;
2375 static inline void skb_reset_network_header(struct sk_buff *skb)
2377 skb->network_header = skb->data - skb->head;
2380 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2382 skb_reset_network_header(skb);
2383 skb->network_header += offset;
2386 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2388 return skb->head + skb->mac_header;
2391 static inline int skb_mac_offset(const struct sk_buff *skb)
2393 return skb_mac_header(skb) - skb->data;
2396 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2398 return skb->network_header - skb->mac_header;
2401 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2403 return skb->mac_header != (typeof(skb->mac_header))~0U;
2406 static inline void skb_reset_mac_header(struct sk_buff *skb)
2408 skb->mac_header = skb->data - skb->head;
2411 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2413 skb_reset_mac_header(skb);
2414 skb->mac_header += offset;
2417 static inline void skb_pop_mac_header(struct sk_buff *skb)
2419 skb->mac_header = skb->network_header;
2422 static inline void skb_probe_transport_header(struct sk_buff *skb,
2423 const int offset_hint)
2425 struct flow_keys_basic keys;
2427 if (skb_transport_header_was_set(skb))
2430 if (skb_flow_dissect_flow_keys_basic(skb, &keys, NULL, 0, 0, 0, 0))
2431 skb_set_transport_header(skb, keys.control.thoff);
2432 else if (offset_hint >= 0)
2433 skb_set_transport_header(skb, offset_hint);
2436 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2438 if (skb_mac_header_was_set(skb)) {
2439 const unsigned char *old_mac = skb_mac_header(skb);
2441 skb_set_mac_header(skb, -skb->mac_len);
2442 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2446 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2448 return skb->csum_start - skb_headroom(skb);
2451 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2453 return skb->head + skb->csum_start;
2456 static inline int skb_transport_offset(const struct sk_buff *skb)
2458 return skb_transport_header(skb) - skb->data;
2461 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2463 return skb->transport_header - skb->network_header;
2466 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2468 return skb->inner_transport_header - skb->inner_network_header;
2471 static inline int skb_network_offset(const struct sk_buff *skb)
2473 return skb_network_header(skb) - skb->data;
2476 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2478 return skb_inner_network_header(skb) - skb->data;
2481 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2483 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2487 * CPUs often take a performance hit when accessing unaligned memory
2488 * locations. The actual performance hit varies, it can be small if the
2489 * hardware handles it or large if we have to take an exception and fix it
2492 * Since an ethernet header is 14 bytes network drivers often end up with
2493 * the IP header at an unaligned offset. The IP header can be aligned by
2494 * shifting the start of the packet by 2 bytes. Drivers should do this
2497 * skb_reserve(skb, NET_IP_ALIGN);
2499 * The downside to this alignment of the IP header is that the DMA is now
2500 * unaligned. On some architectures the cost of an unaligned DMA is high
2501 * and this cost outweighs the gains made by aligning the IP header.
2503 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2506 #ifndef NET_IP_ALIGN
2507 #define NET_IP_ALIGN 2
2511 * The networking layer reserves some headroom in skb data (via
2512 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2513 * the header has to grow. In the default case, if the header has to grow
2514 * 32 bytes or less we avoid the reallocation.
2516 * Unfortunately this headroom changes the DMA alignment of the resulting
2517 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2518 * on some architectures. An architecture can override this value,
2519 * perhaps setting it to a cacheline in size (since that will maintain
2520 * cacheline alignment of the DMA). It must be a power of 2.
2522 * Various parts of the networking layer expect at least 32 bytes of
2523 * headroom, you should not reduce this.
2525 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2526 * to reduce average number of cache lines per packet.
2527 * get_rps_cpus() for example only access one 64 bytes aligned block :
2528 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2531 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2534 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2536 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2538 if (unlikely(skb_is_nonlinear(skb))) {
2543 skb_set_tail_pointer(skb, len);
2546 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2548 __skb_set_length(skb, len);
2551 void skb_trim(struct sk_buff *skb, unsigned int len);
2553 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2556 return ___pskb_trim(skb, len);
2557 __skb_trim(skb, len);
2561 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2563 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2567 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2568 * @skb: buffer to alter
2571 * This is identical to pskb_trim except that the caller knows that
2572 * the skb is not cloned so we should never get an error due to out-
2575 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2577 int err = pskb_trim(skb, len);
2581 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2583 unsigned int diff = len - skb->len;
2585 if (skb_tailroom(skb) < diff) {
2586 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2591 __skb_set_length(skb, len);
2596 * skb_orphan - orphan a buffer
2597 * @skb: buffer to orphan
2599 * If a buffer currently has an owner then we call the owner's
2600 * destructor function and make the @skb unowned. The buffer continues
2601 * to exist but is no longer charged to its former owner.
2603 static inline void skb_orphan(struct sk_buff *skb)
2605 if (skb->destructor) {
2606 skb->destructor(skb);
2607 skb->destructor = NULL;
2615 * skb_orphan_frags - orphan the frags contained in a buffer
2616 * @skb: buffer to orphan frags from
2617 * @gfp_mask: allocation mask for replacement pages
2619 * For each frag in the SKB which needs a destructor (i.e. has an
2620 * owner) create a copy of that frag and release the original
2621 * page by calling the destructor.
2623 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2625 if (likely(!skb_zcopy(skb)))
2627 if (!skb_zcopy_is_nouarg(skb) &&
2628 skb_uarg(skb)->callback == sock_zerocopy_callback)
2630 return skb_copy_ubufs(skb, gfp_mask);
2633 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2634 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2636 if (likely(!skb_zcopy(skb)))
2638 return skb_copy_ubufs(skb, gfp_mask);
2642 * __skb_queue_purge - empty a list
2643 * @list: list to empty
2645 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2646 * the list and one reference dropped. This function does not take the
2647 * list lock and the caller must hold the relevant locks to use it.
2649 void skb_queue_purge(struct sk_buff_head *list);
2650 static inline void __skb_queue_purge(struct sk_buff_head *list)
2652 struct sk_buff *skb;
2653 while ((skb = __skb_dequeue(list)) != NULL)
2657 unsigned int skb_rbtree_purge(struct rb_root *root);
2659 void *netdev_alloc_frag(unsigned int fragsz);
2661 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2665 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2666 * @dev: network device to receive on
2667 * @length: length to allocate
2669 * Allocate a new &sk_buff and assign it a usage count of one. The
2670 * buffer has unspecified headroom built in. Users should allocate
2671 * the headroom they think they need without accounting for the
2672 * built in space. The built in space is used for optimisations.
2674 * %NULL is returned if there is no free memory. Although this function
2675 * allocates memory it can be called from an interrupt.
2677 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2678 unsigned int length)
2680 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2683 /* legacy helper around __netdev_alloc_skb() */
2684 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2687 return __netdev_alloc_skb(NULL, length, gfp_mask);
2690 /* legacy helper around netdev_alloc_skb() */
2691 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2693 return netdev_alloc_skb(NULL, length);
2697 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2698 unsigned int length, gfp_t gfp)
2700 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2702 if (NET_IP_ALIGN && skb)
2703 skb_reserve(skb, NET_IP_ALIGN);
2707 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2708 unsigned int length)
2710 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2713 static inline void skb_free_frag(void *addr)
2715 page_frag_free(addr);
2718 void *napi_alloc_frag(unsigned int fragsz);
2719 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2720 unsigned int length, gfp_t gfp_mask);
2721 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2722 unsigned int length)
2724 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2726 void napi_consume_skb(struct sk_buff *skb, int budget);
2728 void __kfree_skb_flush(void);
2729 void __kfree_skb_defer(struct sk_buff *skb);
2732 * __dev_alloc_pages - allocate page for network Rx
2733 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2734 * @order: size of the allocation
2736 * Allocate a new page.
2738 * %NULL is returned if there is no free memory.
2740 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2743 /* This piece of code contains several assumptions.
2744 * 1. This is for device Rx, therefor a cold page is preferred.
2745 * 2. The expectation is the user wants a compound page.
2746 * 3. If requesting a order 0 page it will not be compound
2747 * due to the check to see if order has a value in prep_new_page
2748 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2749 * code in gfp_to_alloc_flags that should be enforcing this.
2751 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2753 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2756 static inline struct page *dev_alloc_pages(unsigned int order)
2758 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2762 * __dev_alloc_page - allocate a page for network Rx
2763 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2765 * Allocate a new page.
2767 * %NULL is returned if there is no free memory.
2769 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2771 return __dev_alloc_pages(gfp_mask, 0);
2774 static inline struct page *dev_alloc_page(void)
2776 return dev_alloc_pages(0);
2780 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2781 * @page: The page that was allocated from skb_alloc_page
2782 * @skb: The skb that may need pfmemalloc set
2784 static inline void skb_propagate_pfmemalloc(struct page *page,
2785 struct sk_buff *skb)
2787 if (page_is_pfmemalloc(page))
2788 skb->pfmemalloc = true;
2792 * skb_frag_off() - Returns the offset of a skb fragment
2793 * @frag: the paged fragment
2795 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
2797 return frag->page_offset;
2801 * skb_frag_page - retrieve the page referred to by a paged fragment
2802 * @frag: the paged fragment
2804 * Returns the &struct page associated with @frag.
2806 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2808 return frag->page.p;
2812 * __skb_frag_ref - take an addition reference on a paged fragment.
2813 * @frag: the paged fragment
2815 * Takes an additional reference on the paged fragment @frag.
2817 static inline void __skb_frag_ref(skb_frag_t *frag)
2819 get_page(skb_frag_page(frag));
2823 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2825 * @f: the fragment offset.
2827 * Takes an additional reference on the @f'th paged fragment of @skb.
2829 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2831 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2835 * __skb_frag_unref - release a reference on a paged fragment.
2836 * @frag: the paged fragment
2838 * Releases a reference on the paged fragment @frag.
2840 static inline void __skb_frag_unref(skb_frag_t *frag)
2842 put_page(skb_frag_page(frag));
2846 * skb_frag_unref - release a reference on a paged fragment of an skb.
2848 * @f: the fragment offset
2850 * Releases a reference on the @f'th paged fragment of @skb.
2852 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2854 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2858 * skb_frag_address - gets the address of the data contained in a paged fragment
2859 * @frag: the paged fragment buffer
2861 * Returns the address of the data within @frag. The page must already
2864 static inline void *skb_frag_address(const skb_frag_t *frag)
2866 return page_address(skb_frag_page(frag)) + frag->page_offset;
2870 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2871 * @frag: the paged fragment buffer
2873 * Returns the address of the data within @frag. Checks that the page
2874 * is mapped and returns %NULL otherwise.
2876 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2878 void *ptr = page_address(skb_frag_page(frag));
2882 return ptr + frag->page_offset;
2886 * __skb_frag_set_page - sets the page contained in a paged fragment
2887 * @frag: the paged fragment
2888 * @page: the page to set
2890 * Sets the fragment @frag to contain @page.
2892 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2894 frag->page.p = page;
2898 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2900 * @f: the fragment offset
2901 * @page: the page to set
2903 * Sets the @f'th fragment of @skb to contain @page.
2905 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2908 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2911 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2914 * skb_frag_dma_map - maps a paged fragment via the DMA API
2915 * @dev: the device to map the fragment to
2916 * @frag: the paged fragment to map
2917 * @offset: the offset within the fragment (starting at the
2918 * fragment's own offset)
2919 * @size: the number of bytes to map
2920 * @dir: the direction of the mapping (``PCI_DMA_*``)
2922 * Maps the page associated with @frag to @device.
2924 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2925 const skb_frag_t *frag,
2926 size_t offset, size_t size,
2927 enum dma_data_direction dir)
2929 return dma_map_page(dev, skb_frag_page(frag),
2930 frag->page_offset + offset, size, dir);
2933 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2936 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2940 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2943 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2948 * skb_clone_writable - is the header of a clone writable
2949 * @skb: buffer to check
2950 * @len: length up to which to write
2952 * Returns true if modifying the header part of the cloned buffer
2953 * does not requires the data to be copied.
2955 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2957 return !skb_header_cloned(skb) &&
2958 skb_headroom(skb) + len <= skb->hdr_len;
2961 static inline int skb_try_make_writable(struct sk_buff *skb,
2962 unsigned int write_len)
2964 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2965 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2968 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2973 if (headroom > skb_headroom(skb))
2974 delta = headroom - skb_headroom(skb);
2976 if (delta || cloned)
2977 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2983 * skb_cow - copy header of skb when it is required
2984 * @skb: buffer to cow
2985 * @headroom: needed headroom
2987 * If the skb passed lacks sufficient headroom or its data part
2988 * is shared, data is reallocated. If reallocation fails, an error
2989 * is returned and original skb is not changed.
2991 * The result is skb with writable area skb->head...skb->tail
2992 * and at least @headroom of space at head.
2994 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2996 return __skb_cow(skb, headroom, skb_cloned(skb));
3000 * skb_cow_head - skb_cow but only making the head writable
3001 * @skb: buffer to cow
3002 * @headroom: needed headroom
3004 * This function is identical to skb_cow except that we replace the
3005 * skb_cloned check by skb_header_cloned. It should be used when
3006 * you only need to push on some header and do not need to modify
3009 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3011 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3015 * skb_padto - pad an skbuff up to a minimal size
3016 * @skb: buffer to pad
3017 * @len: minimal length
3019 * Pads up a buffer to ensure the trailing bytes exist and are
3020 * blanked. If the buffer already contains sufficient data it
3021 * is untouched. Otherwise it is extended. Returns zero on
3022 * success. The skb is freed on error.
3024 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3026 unsigned int size = skb->len;
3027 if (likely(size >= len))
3029 return skb_pad(skb, len - size);
3033 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3034 * @skb: buffer to pad
3035 * @len: minimal length
3036 * @free_on_error: free buffer on error
3038 * Pads up a buffer to ensure the trailing bytes exist and are
3039 * blanked. If the buffer already contains sufficient data it
3040 * is untouched. Otherwise it is extended. Returns zero on
3041 * success. The skb is freed on error if @free_on_error is true.
3043 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3047 unsigned int size = skb->len;
3049 if (unlikely(size < len)) {
3051 if (__skb_pad(skb, len, free_on_error))
3053 __skb_put(skb, len);
3059 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3060 * @skb: buffer to pad
3061 * @len: minimal length
3063 * Pads up a buffer to ensure the trailing bytes exist and are
3064 * blanked. If the buffer already contains sufficient data it
3065 * is untouched. Otherwise it is extended. Returns zero on
3066 * success. The skb is freed on error.
3068 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3070 return __skb_put_padto(skb, len, true);
3073 static inline int skb_add_data(struct sk_buff *skb,
3074 struct iov_iter *from, int copy)
3076 const int off = skb->len;
3078 if (skb->ip_summed == CHECKSUM_NONE) {
3080 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3082 skb->csum = csum_block_add(skb->csum, csum, off);
3085 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3088 __skb_trim(skb, off);
3092 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3093 const struct page *page, int off)
3098 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
3100 return page == skb_frag_page(frag) &&
3101 off == frag->page_offset + skb_frag_size(frag);
3106 static inline int __skb_linearize(struct sk_buff *skb)
3108 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3112 * skb_linearize - convert paged skb to linear one
3113 * @skb: buffer to linarize
3115 * If there is no free memory -ENOMEM is returned, otherwise zero
3116 * is returned and the old skb data released.
3118 static inline int skb_linearize(struct sk_buff *skb)
3120 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3124 * skb_has_shared_frag - can any frag be overwritten
3125 * @skb: buffer to test
3127 * Return true if the skb has at least one frag that might be modified
3128 * by an external entity (as in vmsplice()/sendfile())
3130 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3132 return skb_is_nonlinear(skb) &&
3133 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3137 * skb_linearize_cow - make sure skb is linear and writable
3138 * @skb: buffer to process
3140 * If there is no free memory -ENOMEM is returned, otherwise zero
3141 * is returned and the old skb data released.
3143 static inline int skb_linearize_cow(struct sk_buff *skb)
3145 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3146 __skb_linearize(skb) : 0;
3149 static __always_inline void
3150 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3153 if (skb->ip_summed == CHECKSUM_COMPLETE)
3154 skb->csum = csum_block_sub(skb->csum,
3155 csum_partial(start, len, 0), off);
3156 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3157 skb_checksum_start_offset(skb) < 0)
3158 skb->ip_summed = CHECKSUM_NONE;
3162 * skb_postpull_rcsum - update checksum for received skb after pull
3163 * @skb: buffer to update
3164 * @start: start of data before pull
3165 * @len: length of data pulled
3167 * After doing a pull on a received packet, you need to call this to
3168 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3169 * CHECKSUM_NONE so that it can be recomputed from scratch.
3171 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3172 const void *start, unsigned int len)
3174 __skb_postpull_rcsum(skb, start, len, 0);
3177 static __always_inline void
3178 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3181 if (skb->ip_summed == CHECKSUM_COMPLETE)
3182 skb->csum = csum_block_add(skb->csum,
3183 csum_partial(start, len, 0), off);
3187 * skb_postpush_rcsum - update checksum for received skb after push
3188 * @skb: buffer to update
3189 * @start: start of data after push
3190 * @len: length of data pushed
3192 * After doing a push on a received packet, you need to call this to
3193 * update the CHECKSUM_COMPLETE checksum.
3195 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3196 const void *start, unsigned int len)
3198 __skb_postpush_rcsum(skb, start, len, 0);
3201 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3204 * skb_push_rcsum - push skb and update receive checksum
3205 * @skb: buffer to update
3206 * @len: length of data pulled
3208 * This function performs an skb_push on the packet and updates
3209 * the CHECKSUM_COMPLETE checksum. It should be used on
3210 * receive path processing instead of skb_push unless you know
3211 * that the checksum difference is zero (e.g., a valid IP header)
3212 * or you are setting ip_summed to CHECKSUM_NONE.
3214 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3217 skb_postpush_rcsum(skb, skb->data, len);
3221 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3223 * pskb_trim_rcsum - trim received skb and update checksum
3224 * @skb: buffer to trim
3227 * This is exactly the same as pskb_trim except that it ensures the
3228 * checksum of received packets are still valid after the operation.
3229 * It can change skb pointers.
3232 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3234 if (likely(len >= skb->len))
3236 return pskb_trim_rcsum_slow(skb, len);
3239 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3241 if (skb->ip_summed == CHECKSUM_COMPLETE)
3242 skb->ip_summed = CHECKSUM_NONE;
3243 __skb_trim(skb, len);
3247 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3249 if (skb->ip_summed == CHECKSUM_COMPLETE)
3250 skb->ip_summed = CHECKSUM_NONE;
3251 return __skb_grow(skb, len);
3254 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3255 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3256 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3257 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3258 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3260 #define skb_queue_walk(queue, skb) \
3261 for (skb = (queue)->next; \
3262 skb != (struct sk_buff *)(queue); \
3265 #define skb_queue_walk_safe(queue, skb, tmp) \
3266 for (skb = (queue)->next, tmp = skb->next; \
3267 skb != (struct sk_buff *)(queue); \
3268 skb = tmp, tmp = skb->next)
3270 #define skb_queue_walk_from(queue, skb) \
3271 for (; skb != (struct sk_buff *)(queue); \
3274 #define skb_rbtree_walk(skb, root) \
3275 for (skb = skb_rb_first(root); skb != NULL; \
3276 skb = skb_rb_next(skb))
3278 #define skb_rbtree_walk_from(skb) \
3279 for (; skb != NULL; \
3280 skb = skb_rb_next(skb))
3282 #define skb_rbtree_walk_from_safe(skb, tmp) \
3283 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3286 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3287 for (tmp = skb->next; \
3288 skb != (struct sk_buff *)(queue); \
3289 skb = tmp, tmp = skb->next)
3291 #define skb_queue_reverse_walk(queue, skb) \
3292 for (skb = (queue)->prev; \
3293 skb != (struct sk_buff *)(queue); \
3296 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3297 for (skb = (queue)->prev, tmp = skb->prev; \
3298 skb != (struct sk_buff *)(queue); \
3299 skb = tmp, tmp = skb->prev)
3301 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3302 for (tmp = skb->prev; \
3303 skb != (struct sk_buff *)(queue); \
3304 skb = tmp, tmp = skb->prev)
3306 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3308 return skb_shinfo(skb)->frag_list != NULL;
3311 static inline void skb_frag_list_init(struct sk_buff *skb)
3313 skb_shinfo(skb)->frag_list = NULL;
3316 #define skb_walk_frags(skb, iter) \
3317 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3320 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3321 const struct sk_buff *skb);
3322 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3323 struct sk_buff_head *queue,
3325 void (*destructor)(struct sock *sk,
3326 struct sk_buff *skb),
3327 int *peeked, int *off, int *err,
3328 struct sk_buff **last);
3329 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3330 void (*destructor)(struct sock *sk,
3331 struct sk_buff *skb),
3332 int *peeked, int *off, int *err,
3333 struct sk_buff **last);
3334 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3335 void (*destructor)(struct sock *sk,
3336 struct sk_buff *skb),
3337 int *peeked, int *off, int *err);
3338 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3340 __poll_t datagram_poll(struct file *file, struct socket *sock,
3341 struct poll_table_struct *wait);
3342 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3343 struct iov_iter *to, int size);
3344 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3345 struct msghdr *msg, int size)
3347 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3349 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3350 struct msghdr *msg);
3351 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3352 struct iov_iter *from, int len);
3353 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3354 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3355 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3356 static inline void skb_free_datagram_locked(struct sock *sk,
3357 struct sk_buff *skb)
3359 __skb_free_datagram_locked(sk, skb, 0);
3361 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3362 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3363 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3364 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3365 int len, __wsum csum);
3366 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3367 struct pipe_inode_info *pipe, unsigned int len,
3368 unsigned int flags);
3369 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3371 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3372 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3373 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3374 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3376 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3377 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3378 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3379 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3380 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3381 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3382 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3383 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3384 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3385 int skb_vlan_pop(struct sk_buff *skb);
3386 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3387 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3390 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3392 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3395 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3397 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3400 struct skb_checksum_ops {
3401 __wsum (*update)(const void *mem, int len, __wsum wsum);
3402 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3405 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3407 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3408 __wsum csum, const struct skb_checksum_ops *ops);
3409 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3412 static inline void * __must_check
3413 __skb_header_pointer(const struct sk_buff *skb, int offset,
3414 int len, void *data, int hlen, void *buffer)
3416 if (hlen - offset >= len)
3417 return data + offset;
3420 skb_copy_bits(skb, offset, buffer, len) < 0)
3426 static inline void * __must_check
3427 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3429 return __skb_header_pointer(skb, offset, len, skb->data,
3430 skb_headlen(skb), buffer);
3434 * skb_needs_linearize - check if we need to linearize a given skb
3435 * depending on the given device features.
3436 * @skb: socket buffer to check
3437 * @features: net device features
3439 * Returns true if either:
3440 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3441 * 2. skb is fragmented and the device does not support SG.
3443 static inline bool skb_needs_linearize(struct sk_buff *skb,
3444 netdev_features_t features)
3446 return skb_is_nonlinear(skb) &&
3447 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3448 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3451 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3453 const unsigned int len)
3455 memcpy(to, skb->data, len);
3458 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3459 const int offset, void *to,
3460 const unsigned int len)
3462 memcpy(to, skb->data + offset, len);
3465 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3467 const unsigned int len)
3469 memcpy(skb->data, from, len);
3472 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3475 const unsigned int len)
3477 memcpy(skb->data + offset, from, len);
3480 void skb_init(void);
3482 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3488 * skb_get_timestamp - get timestamp from a skb
3489 * @skb: skb to get stamp from
3490 * @stamp: pointer to struct timeval to store stamp in
3492 * Timestamps are stored in the skb as offsets to a base timestamp.
3493 * This function converts the offset back to a struct timeval and stores
3496 static inline void skb_get_timestamp(const struct sk_buff *skb,
3497 struct timeval *stamp)
3499 *stamp = ktime_to_timeval(skb->tstamp);
3502 static inline void skb_get_timestampns(const struct sk_buff *skb,
3503 struct timespec *stamp)
3505 *stamp = ktime_to_timespec(skb->tstamp);
3508 static inline void __net_timestamp(struct sk_buff *skb)
3510 skb->tstamp = ktime_get_real();
3513 static inline ktime_t net_timedelta(ktime_t t)
3515 return ktime_sub(ktime_get_real(), t);
3518 static inline ktime_t net_invalid_timestamp(void)
3523 static inline u8 skb_metadata_len(const struct sk_buff *skb)
3525 return skb_shinfo(skb)->meta_len;
3528 static inline void *skb_metadata_end(const struct sk_buff *skb)
3530 return skb_mac_header(skb);
3533 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3534 const struct sk_buff *skb_b,
3537 const void *a = skb_metadata_end(skb_a);
3538 const void *b = skb_metadata_end(skb_b);
3539 /* Using more efficient varaiant than plain call to memcmp(). */
3540 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3544 #define __it(x, op) (x -= sizeof(u##op))
3545 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3546 case 32: diffs |= __it_diff(a, b, 64);
3547 case 24: diffs |= __it_diff(a, b, 64);
3548 case 16: diffs |= __it_diff(a, b, 64);
3549 case 8: diffs |= __it_diff(a, b, 64);
3551 case 28: diffs |= __it_diff(a, b, 64);
3552 case 20: diffs |= __it_diff(a, b, 64);
3553 case 12: diffs |= __it_diff(a, b, 64);
3554 case 4: diffs |= __it_diff(a, b, 32);
3559 return memcmp(a - meta_len, b - meta_len, meta_len);
3563 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3564 const struct sk_buff *skb_b)
3566 u8 len_a = skb_metadata_len(skb_a);
3567 u8 len_b = skb_metadata_len(skb_b);
3569 if (!(len_a | len_b))
3572 return len_a != len_b ?
3573 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3576 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3578 skb_shinfo(skb)->meta_len = meta_len;
3581 static inline void skb_metadata_clear(struct sk_buff *skb)
3583 skb_metadata_set(skb, 0);
3586 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3588 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3590 void skb_clone_tx_timestamp(struct sk_buff *skb);
3591 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3593 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3595 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3599 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3604 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3607 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3609 * PHY drivers may accept clones of transmitted packets for
3610 * timestamping via their phy_driver.txtstamp method. These drivers
3611 * must call this function to return the skb back to the stack with a
3614 * @skb: clone of the the original outgoing packet
3615 * @hwtstamps: hardware time stamps
3618 void skb_complete_tx_timestamp(struct sk_buff *skb,
3619 struct skb_shared_hwtstamps *hwtstamps);
3621 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3622 struct skb_shared_hwtstamps *hwtstamps,
3623 struct sock *sk, int tstype);
3626 * skb_tstamp_tx - queue clone of skb with send time stamps
3627 * @orig_skb: the original outgoing packet
3628 * @hwtstamps: hardware time stamps, may be NULL if not available
3630 * If the skb has a socket associated, then this function clones the
3631 * skb (thus sharing the actual data and optional structures), stores
3632 * the optional hardware time stamping information (if non NULL) or
3633 * generates a software time stamp (otherwise), then queues the clone
3634 * to the error queue of the socket. Errors are silently ignored.
3636 void skb_tstamp_tx(struct sk_buff *orig_skb,
3637 struct skb_shared_hwtstamps *hwtstamps);
3640 * skb_tx_timestamp() - Driver hook for transmit timestamping
3642 * Ethernet MAC Drivers should call this function in their hard_xmit()
3643 * function immediately before giving the sk_buff to the MAC hardware.
3645 * Specifically, one should make absolutely sure that this function is
3646 * called before TX completion of this packet can trigger. Otherwise
3647 * the packet could potentially already be freed.
3649 * @skb: A socket buffer.
3651 static inline void skb_tx_timestamp(struct sk_buff *skb)
3653 skb_clone_tx_timestamp(skb);
3654 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3655 skb_tstamp_tx(skb, NULL);
3659 * skb_complete_wifi_ack - deliver skb with wifi status
3661 * @skb: the original outgoing packet
3662 * @acked: ack status
3665 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3667 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3668 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3670 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3672 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3674 (skb->ip_summed == CHECKSUM_PARTIAL &&
3675 skb_checksum_start_offset(skb) >= 0));
3679 * skb_checksum_complete - Calculate checksum of an entire packet
3680 * @skb: packet to process
3682 * This function calculates the checksum over the entire packet plus
3683 * the value of skb->csum. The latter can be used to supply the
3684 * checksum of a pseudo header as used by TCP/UDP. It returns the
3687 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3688 * this function can be used to verify that checksum on received
3689 * packets. In that case the function should return zero if the
3690 * checksum is correct. In particular, this function will return zero
3691 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3692 * hardware has already verified the correctness of the checksum.
3694 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3696 return skb_csum_unnecessary(skb) ?
3697 0 : __skb_checksum_complete(skb);
3700 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3702 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3703 if (skb->csum_level == 0)
3704 skb->ip_summed = CHECKSUM_NONE;
3710 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3712 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3713 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3715 } else if (skb->ip_summed == CHECKSUM_NONE) {
3716 skb->ip_summed = CHECKSUM_UNNECESSARY;
3717 skb->csum_level = 0;
3721 /* Check if we need to perform checksum complete validation.
3723 * Returns true if checksum complete is needed, false otherwise
3724 * (either checksum is unnecessary or zero checksum is allowed).
3726 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3730 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3731 skb->csum_valid = 1;
3732 __skb_decr_checksum_unnecessary(skb);
3739 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
3742 #define CHECKSUM_BREAK 76
3744 /* Unset checksum-complete
3746 * Unset checksum complete can be done when packet is being modified
3747 * (uncompressed for instance) and checksum-complete value is
3750 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3752 if (skb->ip_summed == CHECKSUM_COMPLETE)
3753 skb->ip_summed = CHECKSUM_NONE;
3756 /* Validate (init) checksum based on checksum complete.
3759 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3760 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3761 * checksum is stored in skb->csum for use in __skb_checksum_complete
3762 * non-zero: value of invalid checksum
3765 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3769 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3770 if (!csum_fold(csum_add(psum, skb->csum))) {
3771 skb->csum_valid = 1;
3778 if (complete || skb->len <= CHECKSUM_BREAK) {
3781 csum = __skb_checksum_complete(skb);
3782 skb->csum_valid = !csum;
3789 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3794 /* Perform checksum validate (init). Note that this is a macro since we only
3795 * want to calculate the pseudo header which is an input function if necessary.
3796 * First we try to validate without any computation (checksum unnecessary) and
3797 * then calculate based on checksum complete calling the function to compute
3801 * 0: checksum is validated or try to in skb_checksum_complete
3802 * non-zero: value of invalid checksum
3804 #define __skb_checksum_validate(skb, proto, complete, \
3805 zero_okay, check, compute_pseudo) \
3807 __sum16 __ret = 0; \
3808 skb->csum_valid = 0; \
3809 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3810 __ret = __skb_checksum_validate_complete(skb, \
3811 complete, compute_pseudo(skb, proto)); \
3815 #define skb_checksum_init(skb, proto, compute_pseudo) \
3816 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3818 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3819 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3821 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3822 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3824 #define skb_checksum_validate_zero_check(skb, proto, check, \
3826 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3828 #define skb_checksum_simple_validate(skb) \
3829 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3831 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3833 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3836 static inline void __skb_checksum_convert(struct sk_buff *skb,
3837 __sum16 check, __wsum pseudo)
3839 skb->csum = ~pseudo;
3840 skb->ip_summed = CHECKSUM_COMPLETE;
3843 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3845 if (__skb_checksum_convert_check(skb)) \
3846 __skb_checksum_convert(skb, check, \
3847 compute_pseudo(skb, proto)); \
3850 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3851 u16 start, u16 offset)
3853 skb->ip_summed = CHECKSUM_PARTIAL;
3854 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3855 skb->csum_offset = offset - start;
3858 /* Update skbuf and packet to reflect the remote checksum offload operation.
3859 * When called, ptr indicates the starting point for skb->csum when
3860 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3861 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3863 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3864 int start, int offset, bool nopartial)
3869 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3873 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3874 __skb_checksum_complete(skb);
3875 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3878 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3880 /* Adjust skb->csum since we changed the packet */
3881 skb->csum = csum_add(skb->csum, delta);
3884 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3886 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3887 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3893 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3894 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3895 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3897 if (nfct && atomic_dec_and_test(&nfct->use))
3898 nf_conntrack_destroy(nfct);
3900 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3903 atomic_inc(&nfct->use);
3906 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3907 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3909 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use))
3912 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3915 refcount_inc(&nf_bridge->use);
3917 #endif /* CONFIG_BRIDGE_NETFILTER */
3918 static inline void nf_reset(struct sk_buff *skb)
3920 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3921 nf_conntrack_put(skb_nfct(skb));
3924 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3925 nf_bridge_put(skb->nf_bridge);
3926 skb->nf_bridge = NULL;
3930 static inline void nf_reset_trace(struct sk_buff *skb)
3932 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3937 static inline void ipvs_reset(struct sk_buff *skb)
3939 #if IS_ENABLED(CONFIG_IP_VS)
3940 skb->ipvs_property = 0;
3944 /* Note: This doesn't put any conntrack and bridge info in dst. */
3945 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3948 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3949 dst->_nfct = src->_nfct;
3950 nf_conntrack_get(skb_nfct(src));
3952 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3953 dst->nf_bridge = src->nf_bridge;
3954 nf_bridge_get(src->nf_bridge);
3956 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3958 dst->nf_trace = src->nf_trace;
3962 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3964 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3965 nf_conntrack_put(skb_nfct(dst));
3967 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3968 nf_bridge_put(dst->nf_bridge);
3970 __nf_copy(dst, src, true);
3973 #ifdef CONFIG_NETWORK_SECMARK
3974 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3976 to->secmark = from->secmark;
3979 static inline void skb_init_secmark(struct sk_buff *skb)
3984 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3987 static inline void skb_init_secmark(struct sk_buff *skb)
3991 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3993 return !skb->destructor &&
3994 #if IS_ENABLED(CONFIG_XFRM)
3998 !skb->_skb_refdst &&
3999 !skb_has_frag_list(skb);
4002 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4004 skb->queue_mapping = queue_mapping;
4007 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4009 return skb->queue_mapping;
4012 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4014 to->queue_mapping = from->queue_mapping;
4017 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4019 skb->queue_mapping = rx_queue + 1;
4022 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4024 return skb->queue_mapping - 1;
4027 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4029 return skb->queue_mapping != 0;
4032 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4034 skb->dst_pending_confirm = val;
4037 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4039 return skb->dst_pending_confirm != 0;
4042 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
4051 /* Keeps track of mac header offset relative to skb->head.
4052 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4053 * For non-tunnel skb it points to skb_mac_header() and for
4054 * tunnel skb it points to outer mac header.
4055 * Keeps track of level of encapsulation of network headers.
4066 #define SKB_SGO_CB_OFFSET 32
4067 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
4069 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4071 return (skb_mac_header(inner_skb) - inner_skb->head) -
4072 SKB_GSO_CB(inner_skb)->mac_offset;
4075 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4077 int new_headroom, headroom;
4080 headroom = skb_headroom(skb);
4081 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4085 new_headroom = skb_headroom(skb);
4086 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4090 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4092 /* Do not update partial checksums if remote checksum is enabled. */
4093 if (skb->remcsum_offload)
4096 SKB_GSO_CB(skb)->csum = res;
4097 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4100 /* Compute the checksum for a gso segment. First compute the checksum value
4101 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4102 * then add in skb->csum (checksum from csum_start to end of packet).
4103 * skb->csum and csum_start are then updated to reflect the checksum of the
4104 * resultant packet starting from the transport header-- the resultant checksum
4105 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4108 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4110 unsigned char *csum_start = skb_transport_header(skb);
4111 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4112 __wsum partial = SKB_GSO_CB(skb)->csum;
4114 SKB_GSO_CB(skb)->csum = res;
4115 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4117 return csum_fold(csum_partial(csum_start, plen, partial));
4120 static inline bool skb_is_gso(const struct sk_buff *skb)
4122 return skb_shinfo(skb)->gso_size;
4125 /* Note: Should be called only if skb_is_gso(skb) is true */
4126 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4128 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4131 /* Note: Should be called only if skb_is_gso(skb) is true */
4132 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4134 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4137 /* Note: Should be called only if skb_is_gso(skb) is true */
4138 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4140 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4143 static inline void skb_gso_reset(struct sk_buff *skb)
4145 skb_shinfo(skb)->gso_size = 0;
4146 skb_shinfo(skb)->gso_segs = 0;
4147 skb_shinfo(skb)->gso_type = 0;
4150 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4153 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4155 shinfo->gso_size += increment;
4158 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4161 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4163 shinfo->gso_size -= decrement;
4166 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4168 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4170 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4171 * wanted then gso_type will be set. */
4172 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4174 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4175 unlikely(shinfo->gso_type == 0)) {
4176 __skb_warn_lro_forwarding(skb);
4182 static inline void skb_forward_csum(struct sk_buff *skb)
4184 /* Unfortunately we don't support this one. Any brave souls? */
4185 if (skb->ip_summed == CHECKSUM_COMPLETE)
4186 skb->ip_summed = CHECKSUM_NONE;
4190 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4191 * @skb: skb to check
4193 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4194 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4195 * use this helper, to document places where we make this assertion.
4197 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4200 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4204 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4206 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4207 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4208 unsigned int transport_len,
4209 __sum16(*skb_chkf)(struct sk_buff *skb));
4212 * skb_head_is_locked - Determine if the skb->head is locked down
4213 * @skb: skb to check
4215 * The head on skbs build around a head frag can be removed if they are
4216 * not cloned. This function returns true if the skb head is locked down
4217 * due to either being allocated via kmalloc, or by being a clone with
4218 * multiple references to the head.
4220 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4222 return !skb->head_frag || skb_cloned(skb);
4225 /* Local Checksum Offload.
4226 * Compute outer checksum based on the assumption that the
4227 * inner checksum will be offloaded later.
4228 * See Documentation/networking/checksum-offloads.txt for
4229 * explanation of how this works.
4230 * Fill in outer checksum adjustment (e.g. with sum of outer
4231 * pseudo-header) before calling.
4232 * Also ensure that inner checksum is in linear data area.
4234 static inline __wsum lco_csum(struct sk_buff *skb)
4236 unsigned char *csum_start = skb_checksum_start(skb);
4237 unsigned char *l4_hdr = skb_transport_header(skb);
4240 /* Start with complement of inner checksum adjustment */
4241 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4244 /* Add in checksum of our headers (incl. outer checksum
4245 * adjustment filled in by caller) and return result.
4247 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4250 #endif /* __KERNEL__ */
4251 #endif /* _LINUX_SKBUFF_H */