1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * Definitions for the 'struct sk_buff' memory handlers.
6 * Alan Cox, <gw4pts@gw4pts.ampr.org>
7 * Florian La Roche, <rzsfl@rz.uni-sb.de>
10 #ifndef _LINUX_SKBUFF_H
11 #define _LINUX_SKBUFF_H
13 #include <linux/kernel.h>
14 #include <linux/compiler.h>
15 #include <linux/time.h>
16 #include <linux/bug.h>
17 #include <linux/cache.h>
18 #include <linux/rbtree.h>
19 #include <linux/socket.h>
20 #include <linux/refcount.h>
22 #include <linux/atomic.h>
23 #include <asm/types.h>
24 #include <linux/spinlock.h>
25 #include <linux/net.h>
26 #include <linux/textsearch.h>
27 #include <net/checksum.h>
28 #include <linux/rcupdate.h>
29 #include <linux/hrtimer.h>
30 #include <linux/dma-mapping.h>
31 #include <linux/netdev_features.h>
32 #include <linux/sched.h>
33 #include <linux/sched/clock.h>
34 #include <net/flow_dissector.h>
35 #include <linux/splice.h>
36 #include <linux/in6.h>
37 #include <linux/if_packet.h>
40 /* The interface for checksum offload between the stack and networking drivers
43 * A. IP checksum related features
45 * Drivers advertise checksum offload capabilities in the features of a device.
46 * From the stack's point of view these are capabilities offered by the driver,
47 * a driver typically only advertises features that it is capable of offloading
50 * The checksum related features are:
52 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
53 * IP (one's complement) checksum for any combination
54 * of protocols or protocol layering. The checksum is
55 * computed and set in a packet per the CHECKSUM_PARTIAL
56 * interface (see below).
58 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
59 * TCP or UDP packets over IPv4. These are specifically
60 * unencapsulated packets of the form IPv4|TCP or
61 * IPv4|UDP where the Protocol field in the IPv4 header
62 * is TCP or UDP. The IPv4 header may contain IP options
63 * This feature cannot be set in features for a device
64 * with NETIF_F_HW_CSUM also set. This feature is being
65 * DEPRECATED (see below).
67 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
68 * TCP or UDP packets over IPv6. These are specifically
69 * unencapsulated packets of the form IPv6|TCP or
70 * IPv4|UDP where the Next Header field in the IPv6
71 * header is either TCP or UDP. IPv6 extension headers
72 * are not supported with this feature. This feature
73 * cannot be set in features for a device with
74 * NETIF_F_HW_CSUM also set. This feature is being
75 * DEPRECATED (see below).
77 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
78 * This flag is used only used to disable the RX checksum
79 * feature for a device. The stack will accept receive
80 * checksum indication in packets received on a device
81 * regardless of whether NETIF_F_RXCSUM is set.
83 * B. Checksumming of received packets by device. Indication of checksum
84 * verification is in set skb->ip_summed. Possible values are:
88 * Device did not checksum this packet e.g. due to lack of capabilities.
89 * The packet contains full (though not verified) checksum in packet but
90 * not in skb->csum. Thus, skb->csum is undefined in this case.
92 * CHECKSUM_UNNECESSARY:
94 * The hardware you're dealing with doesn't calculate the full checksum
95 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
96 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
97 * if their checksums are okay. skb->csum is still undefined in this case
98 * though. A driver or device must never modify the checksum field in the
99 * packet even if checksum is verified.
101 * CHECKSUM_UNNECESSARY is applicable to following protocols:
102 * TCP: IPv6 and IPv4.
103 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
104 * zero UDP checksum for either IPv4 or IPv6, the networking stack
105 * may perform further validation in this case.
106 * GRE: only if the checksum is present in the header.
107 * SCTP: indicates the CRC in SCTP header has been validated.
108 * FCOE: indicates the CRC in FC frame has been validated.
110 * skb->csum_level indicates the number of consecutive checksums found in
111 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
112 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
113 * and a device is able to verify the checksums for UDP (possibly zero),
114 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
115 * two. If the device were only able to verify the UDP checksum and not
116 * GRE, either because it doesn't support GRE checksum of because GRE
117 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
118 * not considered in this case).
122 * This is the most generic way. The device supplied checksum of the _whole_
123 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
124 * hardware doesn't need to parse L3/L4 headers to implement this.
127 * - Even if device supports only some protocols, but is able to produce
128 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
129 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
133 * A checksum is set up to be offloaded to a device as described in the
134 * output description for CHECKSUM_PARTIAL. This may occur on a packet
135 * received directly from another Linux OS, e.g., a virtualized Linux kernel
136 * on the same host, or it may be set in the input path in GRO or remote
137 * checksum offload. For the purposes of checksum verification, the checksum
138 * referred to by skb->csum_start + skb->csum_offset and any preceding
139 * checksums in the packet are considered verified. Any checksums in the
140 * packet that are after the checksum being offloaded are not considered to
143 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
144 * in the skb->ip_summed for a packet. Values are:
148 * The driver is required to checksum the packet as seen by hard_start_xmit()
149 * from skb->csum_start up to the end, and to record/write the checksum at
150 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
151 * csum_start and csum_offset values are valid values given the length and
152 * offset of the packet, however they should not attempt to validate that the
153 * checksum refers to a legitimate transport layer checksum-- it is the
154 * purview of the stack to validate that csum_start and csum_offset are set
157 * When the stack requests checksum offload for a packet, the driver MUST
158 * ensure that the checksum is set correctly. A driver can either offload the
159 * checksum calculation to the device, or call skb_checksum_help (in the case
160 * that the device does not support offload for a particular checksum).
162 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
163 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
164 * checksum offload capability.
165 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
166 * on network device checksumming capabilities: if a packet does not match
167 * them, skb_checksum_help or skb_crc32c_help (depending on the value of
168 * csum_not_inet, see item D.) is called to resolve the checksum.
172 * The skb was already checksummed by the protocol, or a checksum is not
175 * CHECKSUM_UNNECESSARY:
177 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
181 * Not used in checksum output. If a driver observes a packet with this value
182 * set in skbuff, if should treat as CHECKSUM_NONE being set.
184 * D. Non-IP checksum (CRC) offloads
186 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
187 * offloading the SCTP CRC in a packet. To perform this offload the stack
188 * will set set csum_start and csum_offset accordingly, set ip_summed to
189 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
190 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
191 * A driver that supports both IP checksum offload and SCTP CRC32c offload
192 * must verify which offload is configured for a packet by testing the
193 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
194 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
196 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
197 * offloading the FCOE CRC in a packet. To perform this offload the stack
198 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
199 * accordingly. Note the there is no indication in the skbuff that the
200 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
201 * both IP checksum offload and FCOE CRC offload must verify which offload
202 * is configured for a packet presumably by inspecting packet headers.
204 * E. Checksumming on output with GSO.
206 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
207 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
208 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
209 * part of the GSO operation is implied. If a checksum is being offloaded
210 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
211 * are set to refer to the outermost checksum being offload (two offloaded
212 * checksums are possible with UDP encapsulation).
215 /* Don't change this without changing skb_csum_unnecessary! */
216 #define CHECKSUM_NONE 0
217 #define CHECKSUM_UNNECESSARY 1
218 #define CHECKSUM_COMPLETE 2
219 #define CHECKSUM_PARTIAL 3
221 /* Maximum value in skb->csum_level */
222 #define SKB_MAX_CSUM_LEVEL 3
224 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
225 #define SKB_WITH_OVERHEAD(X) \
226 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
227 #define SKB_MAX_ORDER(X, ORDER) \
228 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
229 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
230 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
232 /* return minimum truesize of one skb containing X bytes of data */
233 #define SKB_TRUESIZE(X) ((X) + \
234 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
235 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
239 struct pipe_inode_info;
246 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
247 struct nf_conntrack {
252 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
253 struct nf_bridge_info {
255 BRNF_PROTO_UNCHANGED,
263 struct net_device *physindev;
265 /* always valid & non-NULL from FORWARD on, for physdev match */
266 struct net_device *physoutdev;
268 /* prerouting: detect dnat in orig/reply direction */
270 struct in6_addr ipv6_daddr;
272 /* after prerouting + nat detected: store original source
273 * mac since neigh resolution overwrites it, only used while
274 * skb is out in neigh layer.
276 char neigh_header[8];
281 struct sk_buff_head {
282 /* These two members must be first. */
283 struct sk_buff *next;
284 struct sk_buff *prev;
292 /* To allow 64K frame to be packed as single skb without frag_list we
293 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
294 * buffers which do not start on a page boundary.
296 * Since GRO uses frags we allocate at least 16 regardless of page
299 #if (65536/PAGE_SIZE + 1) < 16
300 #define MAX_SKB_FRAGS 16UL
302 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
304 extern int sysctl_max_skb_frags;
306 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
307 * segment using its current segmentation instead.
309 #define GSO_BY_FRAGS 0xFFFF
311 typedef struct skb_frag_struct skb_frag_t;
313 struct skb_frag_struct {
317 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
327 * skb_frag_size - Returns the size of a skb fragment
328 * @frag: skb fragment
330 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
336 * skb_frag_size_set - Sets the size of a skb fragment
337 * @frag: skb fragment
338 * @size: size of fragment
340 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
346 * skb_frag_size_add - Incrementes the size of a skb fragment by %delta
347 * @frag: skb fragment
348 * @delta: value to add
350 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
356 * skb_frag_size_sub - Decrements the size of a skb fragment by %delta
357 * @frag: skb fragment
358 * @delta: value to subtract
360 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
366 * skb_frag_must_loop - Test if %p is a high memory page
367 * @p: fragment's page
369 static inline bool skb_frag_must_loop(struct page *p)
371 #if defined(CONFIG_HIGHMEM)
379 * skb_frag_foreach_page - loop over pages in a fragment
381 * @f: skb frag to operate on
382 * @f_off: offset from start of f->page.p
383 * @f_len: length from f_off to loop over
384 * @p: (temp var) current page
385 * @p_off: (temp var) offset from start of current page,
386 * non-zero only on first page.
387 * @p_len: (temp var) length in current page,
388 * < PAGE_SIZE only on first and last page.
389 * @copied: (temp var) length so far, excluding current p_len.
391 * A fragment can hold a compound page, in which case per-page
392 * operations, notably kmap_atomic, must be called for each
395 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
396 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
397 p_off = (f_off) & (PAGE_SIZE - 1), \
398 p_len = skb_frag_must_loop(p) ? \
399 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
402 copied += p_len, p++, p_off = 0, \
403 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
405 #define HAVE_HW_TIME_STAMP
408 * struct skb_shared_hwtstamps - hardware time stamps
409 * @hwtstamp: hardware time stamp transformed into duration
410 * since arbitrary point in time
412 * Software time stamps generated by ktime_get_real() are stored in
415 * hwtstamps can only be compared against other hwtstamps from
418 * This structure is attached to packets as part of the
419 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
421 struct skb_shared_hwtstamps {
425 /* Definitions for tx_flags in struct skb_shared_info */
427 /* generate hardware time stamp */
428 SKBTX_HW_TSTAMP = 1 << 0,
430 /* generate software time stamp when queueing packet to NIC */
431 SKBTX_SW_TSTAMP = 1 << 1,
433 /* device driver is going to provide hardware time stamp */
434 SKBTX_IN_PROGRESS = 1 << 2,
436 /* device driver supports TX zero-copy buffers */
437 SKBTX_DEV_ZEROCOPY = 1 << 3,
439 /* generate wifi status information (where possible) */
440 SKBTX_WIFI_STATUS = 1 << 4,
442 /* This indicates at least one fragment might be overwritten
443 * (as in vmsplice(), sendfile() ...)
444 * If we need to compute a TX checksum, we'll need to copy
445 * all frags to avoid possible bad checksum
447 SKBTX_SHARED_FRAG = 1 << 5,
449 /* generate software time stamp when entering packet scheduling */
450 SKBTX_SCHED_TSTAMP = 1 << 6,
453 #define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
454 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
456 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
459 * The callback notifies userspace to release buffers when skb DMA is done in
460 * lower device, the skb last reference should be 0 when calling this.
461 * The zerocopy_success argument is true if zero copy transmit occurred,
462 * false on data copy or out of memory error caused by data copy attempt.
463 * The ctx field is used to track device context.
464 * The desc field is used to track userspace buffer index.
467 void (*callback)(struct ubuf_info *, bool zerocopy_success);
483 struct user_struct *user;
488 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
490 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
491 void mm_unaccount_pinned_pages(struct mmpin *mmp);
493 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
494 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
495 struct ubuf_info *uarg);
497 static inline void sock_zerocopy_get(struct ubuf_info *uarg)
499 refcount_inc(&uarg->refcnt);
502 void sock_zerocopy_put(struct ubuf_info *uarg);
503 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
505 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
507 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len);
508 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
509 struct msghdr *msg, int len,
510 struct ubuf_info *uarg);
512 /* This data is invariant across clones and lives at
513 * the end of the header data, ie. at skb->end.
515 struct skb_shared_info {
520 unsigned short gso_size;
521 /* Warning: this field is not always filled in (UFO)! */
522 unsigned short gso_segs;
523 struct sk_buff *frag_list;
524 struct skb_shared_hwtstamps hwtstamps;
525 unsigned int gso_type;
529 * Warning : all fields before dataref are cleared in __alloc_skb()
533 /* Intermediate layers must ensure that destructor_arg
534 * remains valid until skb destructor */
535 void * destructor_arg;
537 /* must be last field, see pskb_expand_head() */
538 skb_frag_t frags[MAX_SKB_FRAGS];
541 /* We divide dataref into two halves. The higher 16 bits hold references
542 * to the payload part of skb->data. The lower 16 bits hold references to
543 * the entire skb->data. A clone of a headerless skb holds the length of
544 * the header in skb->hdr_len.
546 * All users must obey the rule that the skb->data reference count must be
547 * greater than or equal to the payload reference count.
549 * Holding a reference to the payload part means that the user does not
550 * care about modifications to the header part of skb->data.
552 #define SKB_DATAREF_SHIFT 16
553 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
557 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
558 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
559 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
563 SKB_GSO_TCPV4 = 1 << 0,
565 /* This indicates the skb is from an untrusted source. */
566 SKB_GSO_DODGY = 1 << 1,
568 /* This indicates the tcp segment has CWR set. */
569 SKB_GSO_TCP_ECN = 1 << 2,
571 SKB_GSO_TCP_FIXEDID = 1 << 3,
573 SKB_GSO_TCPV6 = 1 << 4,
575 SKB_GSO_FCOE = 1 << 5,
577 SKB_GSO_GRE = 1 << 6,
579 SKB_GSO_GRE_CSUM = 1 << 7,
581 SKB_GSO_IPXIP4 = 1 << 8,
583 SKB_GSO_IPXIP6 = 1 << 9,
585 SKB_GSO_UDP_TUNNEL = 1 << 10,
587 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
589 SKB_GSO_PARTIAL = 1 << 12,
591 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
593 SKB_GSO_SCTP = 1 << 14,
595 SKB_GSO_ESP = 1 << 15,
597 SKB_GSO_UDP = 1 << 16,
599 SKB_GSO_UDP_L4 = 1 << 17,
602 #if BITS_PER_LONG > 32
603 #define NET_SKBUFF_DATA_USES_OFFSET 1
606 #ifdef NET_SKBUFF_DATA_USES_OFFSET
607 typedef unsigned int sk_buff_data_t;
609 typedef unsigned char *sk_buff_data_t;
613 * struct sk_buff - socket buffer
614 * @next: Next buffer in list
615 * @prev: Previous buffer in list
616 * @tstamp: Time we arrived/left
617 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
618 * @sk: Socket we are owned by
619 * @dev: Device we arrived on/are leaving by
620 * @cb: Control buffer. Free for use by every layer. Put private vars here
621 * @_skb_refdst: destination entry (with norefcount bit)
622 * @sp: the security path, used for xfrm
623 * @len: Length of actual data
624 * @data_len: Data length
625 * @mac_len: Length of link layer header
626 * @hdr_len: writable header length of cloned skb
627 * @csum: Checksum (must include start/offset pair)
628 * @csum_start: Offset from skb->head where checksumming should start
629 * @csum_offset: Offset from csum_start where checksum should be stored
630 * @priority: Packet queueing priority
631 * @ignore_df: allow local fragmentation
632 * @cloned: Head may be cloned (check refcnt to be sure)
633 * @ip_summed: Driver fed us an IP checksum
634 * @nohdr: Payload reference only, must not modify header
635 * @pkt_type: Packet class
636 * @fclone: skbuff clone status
637 * @ipvs_property: skbuff is owned by ipvs
638 * @offload_fwd_mark: Packet was L2-forwarded in hardware
639 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
640 * @tc_skip_classify: do not classify packet. set by IFB device
641 * @tc_at_ingress: used within tc_classify to distinguish in/egress
642 * @tc_redirected: packet was redirected by a tc action
643 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
644 * @peeked: this packet has been seen already, so stats have been
645 * done for it, don't do them again
646 * @nf_trace: netfilter packet trace flag
647 * @protocol: Packet protocol from driver
648 * @destructor: Destruct function
649 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
650 * @_nfct: Associated connection, if any (with nfctinfo bits)
651 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
652 * @skb_iif: ifindex of device we arrived on
653 * @tc_index: Traffic control index
654 * @hash: the packet hash
655 * @queue_mapping: Queue mapping for multiqueue devices
656 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
657 * @active_extensions: active extensions (skb_ext_id types)
658 * @ndisc_nodetype: router type (from link layer)
659 * @ooo_okay: allow the mapping of a socket to a queue to be changed
660 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
662 * @sw_hash: indicates hash was computed in software stack
663 * @wifi_acked_valid: wifi_acked was set
664 * @wifi_acked: whether frame was acked on wifi or not
665 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
666 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
667 * @dst_pending_confirm: need to confirm neighbour
668 * @decrypted: Decrypted SKB
669 * @napi_id: id of the NAPI struct this skb came from
670 * @secmark: security marking
671 * @mark: Generic packet mark
672 * @vlan_proto: vlan encapsulation protocol
673 * @vlan_tci: vlan tag control information
674 * @inner_protocol: Protocol (encapsulation)
675 * @inner_transport_header: Inner transport layer header (encapsulation)
676 * @inner_network_header: Network layer header (encapsulation)
677 * @inner_mac_header: Link layer header (encapsulation)
678 * @transport_header: Transport layer header
679 * @network_header: Network layer header
680 * @mac_header: Link layer header
681 * @tail: Tail pointer
683 * @head: Head of buffer
684 * @data: Data head pointer
685 * @truesize: Buffer size
686 * @users: User count - see {datagram,tcp}.c
687 * @extensions: allocated extensions, valid if active_extensions is nonzero
693 /* These two members must be first. */
694 struct sk_buff *next;
695 struct sk_buff *prev;
698 struct net_device *dev;
699 /* Some protocols might use this space to store information,
700 * while device pointer would be NULL.
701 * UDP receive path is one user.
703 unsigned long dev_scratch;
706 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
707 struct list_head list;
712 int ip_defrag_offset;
717 u64 skb_mstamp_ns; /* earliest departure time */
720 * This is the control buffer. It is free to use for every
721 * layer. Please put your private variables there. If you
722 * want to keep them across layers you have to do a skb_clone()
723 * first. This is owned by whoever has the skb queued ATM.
725 char cb[48] __aligned(8);
729 unsigned long _skb_refdst;
730 void (*destructor)(struct sk_buff *skb);
732 struct list_head tcp_tsorted_anchor;
735 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
743 /* Following fields are _not_ copied in __copy_skb_header()
744 * Note that queue_mapping is here mostly to fill a hole.
748 /* if you move cloned around you also must adapt those constants */
749 #ifdef __BIG_ENDIAN_BITFIELD
750 #define CLONED_MASK (1 << 7)
752 #define CLONED_MASK 1
754 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
756 __u8 __cloned_offset[0];
763 #ifdef CONFIG_SKB_EXTENSIONS
764 __u8 active_extensions;
766 /* fields enclosed in headers_start/headers_end are copied
767 * using a single memcpy() in __copy_skb_header()
770 __u32 headers_start[0];
773 /* if you move pkt_type around you also must adapt those constants */
774 #ifdef __BIG_ENDIAN_BITFIELD
775 #define PKT_TYPE_MAX (7 << 5)
777 #define PKT_TYPE_MAX 7
779 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
781 __u8 __pkt_type_offset[0];
790 __u8 wifi_acked_valid:1;
793 /* Indicates the inner headers are valid in the skbuff. */
794 __u8 encapsulation:1;
795 __u8 encap_hdr_csum:1;
798 #ifdef __BIG_ENDIAN_BITFIELD
799 #define PKT_VLAN_PRESENT_BIT 7
801 #define PKT_VLAN_PRESENT_BIT 0
803 #define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset)
804 __u8 __pkt_vlan_present_offset[0];
806 __u8 csum_complete_sw:1;
808 __u8 csum_not_inet:1;
809 __u8 dst_pending_confirm:1;
810 #ifdef CONFIG_IPV6_NDISC_NODETYPE
811 __u8 ndisc_nodetype:2;
814 __u8 ipvs_property:1;
815 __u8 inner_protocol_type:1;
816 __u8 remcsum_offload:1;
817 #ifdef CONFIG_NET_SWITCHDEV
818 __u8 offload_fwd_mark:1;
819 __u8 offload_l3_fwd_mark:1;
821 #ifdef CONFIG_NET_CLS_ACT
822 __u8 tc_skip_classify:1;
823 __u8 tc_at_ingress:1;
824 __u8 tc_redirected:1;
825 __u8 tc_from_ingress:1;
827 #ifdef CONFIG_TLS_DEVICE
831 #ifdef CONFIG_NET_SCHED
832 __u16 tc_index; /* traffic control index */
847 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
849 unsigned int napi_id;
850 unsigned int sender_cpu;
853 #ifdef CONFIG_NETWORK_SECMARK
859 __u32 reserved_tailroom;
863 __be16 inner_protocol;
867 __u16 inner_transport_header;
868 __u16 inner_network_header;
869 __u16 inner_mac_header;
872 __u16 transport_header;
873 __u16 network_header;
877 __u32 headers_end[0];
880 /* These elements must be at the end, see alloc_skb() for details. */
885 unsigned int truesize;
888 #ifdef CONFIG_SKB_EXTENSIONS
889 /* only useable after checking ->active_extensions != 0 */
890 struct skb_ext *extensions;
896 * Handling routines are only of interest to the kernel
899 #define SKB_ALLOC_FCLONE 0x01
900 #define SKB_ALLOC_RX 0x02
901 #define SKB_ALLOC_NAPI 0x04
904 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
907 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
909 return unlikely(skb->pfmemalloc);
913 * skb might have a dst pointer attached, refcounted or not.
914 * _skb_refdst low order bit is set if refcount was _not_ taken
916 #define SKB_DST_NOREF 1UL
917 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
919 #define SKB_NFCT_PTRMASK ~(7UL)
921 * skb_dst - returns skb dst_entry
924 * Returns skb dst_entry, regardless of reference taken or not.
926 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
928 /* If refdst was not refcounted, check we still are in a
929 * rcu_read_lock section
931 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
932 !rcu_read_lock_held() &&
933 !rcu_read_lock_bh_held());
934 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
938 * skb_dst_set - sets skb dst
942 * Sets skb dst, assuming a reference was taken on dst and should
943 * be released by skb_dst_drop()
945 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
947 skb->_skb_refdst = (unsigned long)dst;
951 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
955 * Sets skb dst, assuming a reference was not taken on dst.
956 * If dst entry is cached, we do not take reference and dst_release
957 * will be avoided by refdst_drop. If dst entry is not cached, we take
958 * reference, so that last dst_release can destroy the dst immediately.
960 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
962 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
963 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
967 * skb_dst_is_noref - Test if skb dst isn't refcounted
970 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
972 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
976 * skb_rtable - Returns the skb &rtable
979 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
981 return (struct rtable *)skb_dst(skb);
984 /* For mangling skb->pkt_type from user space side from applications
985 * such as nft, tc, etc, we only allow a conservative subset of
986 * possible pkt_types to be set.
988 static inline bool skb_pkt_type_ok(u32 ptype)
990 return ptype <= PACKET_OTHERHOST;
994 * skb_napi_id - Returns the skb's NAPI id
997 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
999 #ifdef CONFIG_NET_RX_BUSY_POLL
1000 return skb->napi_id;
1007 * skb_unref - decrement the skb's reference count
1010 * Returns true if we can free the skb.
1012 static inline bool skb_unref(struct sk_buff *skb)
1016 if (likely(refcount_read(&skb->users) == 1))
1018 else if (likely(!refcount_dec_and_test(&skb->users)))
1024 void skb_release_head_state(struct sk_buff *skb);
1025 void kfree_skb(struct sk_buff *skb);
1026 void kfree_skb_list(struct sk_buff *segs);
1027 void skb_tx_error(struct sk_buff *skb);
1028 void consume_skb(struct sk_buff *skb);
1029 void __consume_stateless_skb(struct sk_buff *skb);
1030 void __kfree_skb(struct sk_buff *skb);
1031 extern struct kmem_cache *skbuff_head_cache;
1033 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1034 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1035 bool *fragstolen, int *delta_truesize);
1037 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1039 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1040 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1041 struct sk_buff *build_skb_around(struct sk_buff *skb,
1042 void *data, unsigned int frag_size);
1045 * alloc_skb - allocate a network buffer
1046 * @size: size to allocate
1047 * @priority: allocation mask
1049 * This function is a convenient wrapper around __alloc_skb().
1051 static inline struct sk_buff *alloc_skb(unsigned int size,
1054 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1057 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1058 unsigned long data_len,
1062 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1064 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1065 struct sk_buff_fclones {
1066 struct sk_buff skb1;
1068 struct sk_buff skb2;
1070 refcount_t fclone_ref;
1074 * skb_fclone_busy - check if fclone is busy
1078 * Returns true if skb is a fast clone, and its clone is not freed.
1079 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1080 * so we also check that this didnt happen.
1082 static inline bool skb_fclone_busy(const struct sock *sk,
1083 const struct sk_buff *skb)
1085 const struct sk_buff_fclones *fclones;
1087 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1089 return skb->fclone == SKB_FCLONE_ORIG &&
1090 refcount_read(&fclones->fclone_ref) > 1 &&
1091 fclones->skb2.sk == sk;
1095 * alloc_skb_fclone - allocate a network buffer from fclone cache
1096 * @size: size to allocate
1097 * @priority: allocation mask
1099 * This function is a convenient wrapper around __alloc_skb().
1101 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1104 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1107 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1108 void skb_headers_offset_update(struct sk_buff *skb, int off);
1109 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1110 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1111 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1112 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1113 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1114 gfp_t gfp_mask, bool fclone);
1115 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1118 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1121 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1122 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1123 unsigned int headroom);
1124 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1125 int newtailroom, gfp_t priority);
1126 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1127 int offset, int len);
1128 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1129 int offset, int len);
1130 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1131 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1134 * skb_pad - zero pad the tail of an skb
1135 * @skb: buffer to pad
1136 * @pad: space to pad
1138 * Ensure that a buffer is followed by a padding area that is zero
1139 * filled. Used by network drivers which may DMA or transfer data
1140 * beyond the buffer end onto the wire.
1142 * May return error in out of memory cases. The skb is freed on error.
1144 static inline int skb_pad(struct sk_buff *skb, int pad)
1146 return __skb_pad(skb, pad, true);
1148 #define dev_kfree_skb(a) consume_skb(a)
1150 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1151 int offset, size_t size);
1153 struct skb_seq_state {
1157 __u32 stepped_offset;
1158 struct sk_buff *root_skb;
1159 struct sk_buff *cur_skb;
1163 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1164 unsigned int to, struct skb_seq_state *st);
1165 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1166 struct skb_seq_state *st);
1167 void skb_abort_seq_read(struct skb_seq_state *st);
1169 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1170 unsigned int to, struct ts_config *config);
1173 * Packet hash types specify the type of hash in skb_set_hash.
1175 * Hash types refer to the protocol layer addresses which are used to
1176 * construct a packet's hash. The hashes are used to differentiate or identify
1177 * flows of the protocol layer for the hash type. Hash types are either
1178 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1180 * Properties of hashes:
1182 * 1) Two packets in different flows have different hash values
1183 * 2) Two packets in the same flow should have the same hash value
1185 * A hash at a higher layer is considered to be more specific. A driver should
1186 * set the most specific hash possible.
1188 * A driver cannot indicate a more specific hash than the layer at which a hash
1189 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1191 * A driver may indicate a hash level which is less specific than the
1192 * actual layer the hash was computed on. For instance, a hash computed
1193 * at L4 may be considered an L3 hash. This should only be done if the
1194 * driver can't unambiguously determine that the HW computed the hash at
1195 * the higher layer. Note that the "should" in the second property above
1198 enum pkt_hash_types {
1199 PKT_HASH_TYPE_NONE, /* Undefined type */
1200 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1201 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1202 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1205 static inline void skb_clear_hash(struct sk_buff *skb)
1212 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1215 skb_clear_hash(skb);
1219 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1221 skb->l4_hash = is_l4;
1222 skb->sw_hash = is_sw;
1227 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1229 /* Used by drivers to set hash from HW */
1230 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1234 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1236 __skb_set_hash(skb, hash, true, is_l4);
1239 void __skb_get_hash(struct sk_buff *skb);
1240 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1241 u32 skb_get_poff(const struct sk_buff *skb);
1242 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1243 const struct flow_keys_basic *keys, int hlen);
1244 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1245 void *data, int hlen_proto);
1247 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1248 int thoff, u8 ip_proto)
1250 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1253 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1254 const struct flow_dissector_key *key,
1255 unsigned int key_count);
1258 int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1259 union bpf_attr __user *uattr);
1260 int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1261 struct bpf_prog *prog);
1263 int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr);
1265 static inline int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1266 union bpf_attr __user *uattr)
1271 static inline int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1272 struct bpf_prog *prog)
1277 static inline int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr)
1283 struct bpf_flow_dissector;
1284 bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1285 __be16 proto, int nhoff, int hlen);
1287 bool __skb_flow_dissect(const struct net *net,
1288 const struct sk_buff *skb,
1289 struct flow_dissector *flow_dissector,
1290 void *target_container,
1291 void *data, __be16 proto, int nhoff, int hlen,
1292 unsigned int flags);
1294 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1295 struct flow_dissector *flow_dissector,
1296 void *target_container, unsigned int flags)
1298 return __skb_flow_dissect(NULL, skb, flow_dissector,
1299 target_container, NULL, 0, 0, 0, flags);
1302 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1303 struct flow_keys *flow,
1306 memset(flow, 0, sizeof(*flow));
1307 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1308 flow, NULL, 0, 0, 0, flags);
1312 skb_flow_dissect_flow_keys_basic(const struct net *net,
1313 const struct sk_buff *skb,
1314 struct flow_keys_basic *flow, void *data,
1315 __be16 proto, int nhoff, int hlen,
1318 memset(flow, 0, sizeof(*flow));
1319 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1320 data, proto, nhoff, hlen, flags);
1323 void skb_flow_dissect_meta(const struct sk_buff *skb,
1324 struct flow_dissector *flow_dissector,
1325 void *target_container);
1328 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1329 struct flow_dissector *flow_dissector,
1330 void *target_container);
1332 static inline __u32 skb_get_hash(struct sk_buff *skb)
1334 if (!skb->l4_hash && !skb->sw_hash)
1335 __skb_get_hash(skb);
1340 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1342 if (!skb->l4_hash && !skb->sw_hash) {
1343 struct flow_keys keys;
1344 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1346 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1352 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1354 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1359 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1361 to->hash = from->hash;
1362 to->sw_hash = from->sw_hash;
1363 to->l4_hash = from->l4_hash;
1366 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1367 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1369 return skb->head + skb->end;
1372 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1377 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1382 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1384 return skb->end - skb->head;
1389 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1391 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1393 return &skb_shinfo(skb)->hwtstamps;
1396 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1398 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1400 return is_zcopy ? skb_uarg(skb) : NULL;
1403 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1406 if (skb && uarg && !skb_zcopy(skb)) {
1407 if (unlikely(have_ref && *have_ref))
1410 sock_zerocopy_get(uarg);
1411 skb_shinfo(skb)->destructor_arg = uarg;
1412 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1416 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1418 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1419 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1422 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1424 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1427 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1429 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1432 /* Release a reference on a zerocopy structure */
1433 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1435 struct ubuf_info *uarg = skb_zcopy(skb);
1438 if (skb_zcopy_is_nouarg(skb)) {
1439 /* no notification callback */
1440 } else if (uarg->callback == sock_zerocopy_callback) {
1441 uarg->zerocopy = uarg->zerocopy && zerocopy;
1442 sock_zerocopy_put(uarg);
1444 uarg->callback(uarg, zerocopy);
1447 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1451 /* Abort a zerocopy operation and revert zckey on error in send syscall */
1452 static inline void skb_zcopy_abort(struct sk_buff *skb)
1454 struct ubuf_info *uarg = skb_zcopy(skb);
1457 sock_zerocopy_put_abort(uarg, false);
1458 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1462 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1467 static inline void skb_list_del_init(struct sk_buff *skb)
1469 __list_del_entry(&skb->list);
1470 skb_mark_not_on_list(skb);
1474 * skb_queue_empty - check if a queue is empty
1477 * Returns true if the queue is empty, false otherwise.
1479 static inline int skb_queue_empty(const struct sk_buff_head *list)
1481 return list->next == (const struct sk_buff *) list;
1485 * skb_queue_is_last - check if skb is the last entry in the queue
1489 * Returns true if @skb is the last buffer on the list.
1491 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1492 const struct sk_buff *skb)
1494 return skb->next == (const struct sk_buff *) list;
1498 * skb_queue_is_first - check if skb is the first entry in the queue
1502 * Returns true if @skb is the first buffer on the list.
1504 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1505 const struct sk_buff *skb)
1507 return skb->prev == (const struct sk_buff *) list;
1511 * skb_queue_next - return the next packet in the queue
1513 * @skb: current buffer
1515 * Return the next packet in @list after @skb. It is only valid to
1516 * call this if skb_queue_is_last() evaluates to false.
1518 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1519 const struct sk_buff *skb)
1521 /* This BUG_ON may seem severe, but if we just return then we
1522 * are going to dereference garbage.
1524 BUG_ON(skb_queue_is_last(list, skb));
1529 * skb_queue_prev - return the prev packet in the queue
1531 * @skb: current buffer
1533 * Return the prev packet in @list before @skb. It is only valid to
1534 * call this if skb_queue_is_first() evaluates to false.
1536 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1537 const struct sk_buff *skb)
1539 /* This BUG_ON may seem severe, but if we just return then we
1540 * are going to dereference garbage.
1542 BUG_ON(skb_queue_is_first(list, skb));
1547 * skb_get - reference buffer
1548 * @skb: buffer to reference
1550 * Makes another reference to a socket buffer and returns a pointer
1553 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1555 refcount_inc(&skb->users);
1560 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1564 * skb_cloned - is the buffer a clone
1565 * @skb: buffer to check
1567 * Returns true if the buffer was generated with skb_clone() and is
1568 * one of multiple shared copies of the buffer. Cloned buffers are
1569 * shared data so must not be written to under normal circumstances.
1571 static inline int skb_cloned(const struct sk_buff *skb)
1573 return skb->cloned &&
1574 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1577 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1579 might_sleep_if(gfpflags_allow_blocking(pri));
1581 if (skb_cloned(skb))
1582 return pskb_expand_head(skb, 0, 0, pri);
1588 * skb_header_cloned - is the header a clone
1589 * @skb: buffer to check
1591 * Returns true if modifying the header part of the buffer requires
1592 * the data to be copied.
1594 static inline int skb_header_cloned(const struct sk_buff *skb)
1601 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1602 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1603 return dataref != 1;
1606 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1608 might_sleep_if(gfpflags_allow_blocking(pri));
1610 if (skb_header_cloned(skb))
1611 return pskb_expand_head(skb, 0, 0, pri);
1617 * __skb_header_release - release reference to header
1618 * @skb: buffer to operate on
1620 static inline void __skb_header_release(struct sk_buff *skb)
1623 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1628 * skb_shared - is the buffer shared
1629 * @skb: buffer to check
1631 * Returns true if more than one person has a reference to this
1634 static inline int skb_shared(const struct sk_buff *skb)
1636 return refcount_read(&skb->users) != 1;
1640 * skb_share_check - check if buffer is shared and if so clone it
1641 * @skb: buffer to check
1642 * @pri: priority for memory allocation
1644 * If the buffer is shared the buffer is cloned and the old copy
1645 * drops a reference. A new clone with a single reference is returned.
1646 * If the buffer is not shared the original buffer is returned. When
1647 * being called from interrupt status or with spinlocks held pri must
1650 * NULL is returned on a memory allocation failure.
1652 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1654 might_sleep_if(gfpflags_allow_blocking(pri));
1655 if (skb_shared(skb)) {
1656 struct sk_buff *nskb = skb_clone(skb, pri);
1668 * Copy shared buffers into a new sk_buff. We effectively do COW on
1669 * packets to handle cases where we have a local reader and forward
1670 * and a couple of other messy ones. The normal one is tcpdumping
1671 * a packet thats being forwarded.
1675 * skb_unshare - make a copy of a shared buffer
1676 * @skb: buffer to check
1677 * @pri: priority for memory allocation
1679 * If the socket buffer is a clone then this function creates a new
1680 * copy of the data, drops a reference count on the old copy and returns
1681 * the new copy with the reference count at 1. If the buffer is not a clone
1682 * the original buffer is returned. When called with a spinlock held or
1683 * from interrupt state @pri must be %GFP_ATOMIC
1685 * %NULL is returned on a memory allocation failure.
1687 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1690 might_sleep_if(gfpflags_allow_blocking(pri));
1691 if (skb_cloned(skb)) {
1692 struct sk_buff *nskb = skb_copy(skb, pri);
1694 /* Free our shared copy */
1705 * skb_peek - peek at the head of an &sk_buff_head
1706 * @list_: list to peek at
1708 * Peek an &sk_buff. Unlike most other operations you _MUST_
1709 * be careful with this one. A peek leaves the buffer on the
1710 * list and someone else may run off with it. You must hold
1711 * the appropriate locks or have a private queue to do this.
1713 * Returns %NULL for an empty list or a pointer to the head element.
1714 * The reference count is not incremented and the reference is therefore
1715 * volatile. Use with caution.
1717 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1719 struct sk_buff *skb = list_->next;
1721 if (skb == (struct sk_buff *)list_)
1727 * __skb_peek - peek at the head of a non-empty &sk_buff_head
1728 * @list_: list to peek at
1730 * Like skb_peek(), but the caller knows that the list is not empty.
1732 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
1738 * skb_peek_next - peek skb following the given one from a queue
1739 * @skb: skb to start from
1740 * @list_: list to peek at
1742 * Returns %NULL when the end of the list is met or a pointer to the
1743 * next element. The reference count is not incremented and the
1744 * reference is therefore volatile. Use with caution.
1746 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1747 const struct sk_buff_head *list_)
1749 struct sk_buff *next = skb->next;
1751 if (next == (struct sk_buff *)list_)
1757 * skb_peek_tail - peek at the tail of an &sk_buff_head
1758 * @list_: list to peek at
1760 * Peek an &sk_buff. Unlike most other operations you _MUST_
1761 * be careful with this one. A peek leaves the buffer on the
1762 * list and someone else may run off with it. You must hold
1763 * the appropriate locks or have a private queue to do this.
1765 * Returns %NULL for an empty list or a pointer to the tail element.
1766 * The reference count is not incremented and the reference is therefore
1767 * volatile. Use with caution.
1769 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1771 struct sk_buff *skb = list_->prev;
1773 if (skb == (struct sk_buff *)list_)
1780 * skb_queue_len - get queue length
1781 * @list_: list to measure
1783 * Return the length of an &sk_buff queue.
1785 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1791 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1792 * @list: queue to initialize
1794 * This initializes only the list and queue length aspects of
1795 * an sk_buff_head object. This allows to initialize the list
1796 * aspects of an sk_buff_head without reinitializing things like
1797 * the spinlock. It can also be used for on-stack sk_buff_head
1798 * objects where the spinlock is known to not be used.
1800 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1802 list->prev = list->next = (struct sk_buff *)list;
1807 * This function creates a split out lock class for each invocation;
1808 * this is needed for now since a whole lot of users of the skb-queue
1809 * infrastructure in drivers have different locking usage (in hardirq)
1810 * than the networking core (in softirq only). In the long run either the
1811 * network layer or drivers should need annotation to consolidate the
1812 * main types of usage into 3 classes.
1814 static inline void skb_queue_head_init(struct sk_buff_head *list)
1816 spin_lock_init(&list->lock);
1817 __skb_queue_head_init(list);
1820 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1821 struct lock_class_key *class)
1823 skb_queue_head_init(list);
1824 lockdep_set_class(&list->lock, class);
1828 * Insert an sk_buff on a list.
1830 * The "__skb_xxxx()" functions are the non-atomic ones that
1831 * can only be called with interrupts disabled.
1833 static inline void __skb_insert(struct sk_buff *newsk,
1834 struct sk_buff *prev, struct sk_buff *next,
1835 struct sk_buff_head *list)
1839 next->prev = prev->next = newsk;
1843 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1844 struct sk_buff *prev,
1845 struct sk_buff *next)
1847 struct sk_buff *first = list->next;
1848 struct sk_buff *last = list->prev;
1858 * skb_queue_splice - join two skb lists, this is designed for stacks
1859 * @list: the new list to add
1860 * @head: the place to add it in the first list
1862 static inline void skb_queue_splice(const struct sk_buff_head *list,
1863 struct sk_buff_head *head)
1865 if (!skb_queue_empty(list)) {
1866 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1867 head->qlen += list->qlen;
1872 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1873 * @list: the new list to add
1874 * @head: the place to add it in the first list
1876 * The list at @list is reinitialised
1878 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1879 struct sk_buff_head *head)
1881 if (!skb_queue_empty(list)) {
1882 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1883 head->qlen += list->qlen;
1884 __skb_queue_head_init(list);
1889 * skb_queue_splice_tail - join two skb lists, each list being a queue
1890 * @list: the new list to add
1891 * @head: the place to add it in the first list
1893 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1894 struct sk_buff_head *head)
1896 if (!skb_queue_empty(list)) {
1897 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1898 head->qlen += list->qlen;
1903 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1904 * @list: the new list to add
1905 * @head: the place to add it in the first list
1907 * Each of the lists is a queue.
1908 * The list at @list is reinitialised
1910 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1911 struct sk_buff_head *head)
1913 if (!skb_queue_empty(list)) {
1914 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1915 head->qlen += list->qlen;
1916 __skb_queue_head_init(list);
1921 * __skb_queue_after - queue a buffer at the list head
1922 * @list: list to use
1923 * @prev: place after this buffer
1924 * @newsk: buffer to queue
1926 * Queue a buffer int the middle of a list. This function takes no locks
1927 * and you must therefore hold required locks before calling it.
1929 * A buffer cannot be placed on two lists at the same time.
1931 static inline void __skb_queue_after(struct sk_buff_head *list,
1932 struct sk_buff *prev,
1933 struct sk_buff *newsk)
1935 __skb_insert(newsk, prev, prev->next, list);
1938 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1939 struct sk_buff_head *list);
1941 static inline void __skb_queue_before(struct sk_buff_head *list,
1942 struct sk_buff *next,
1943 struct sk_buff *newsk)
1945 __skb_insert(newsk, next->prev, next, list);
1949 * __skb_queue_head - queue a buffer at the list head
1950 * @list: list to use
1951 * @newsk: buffer to queue
1953 * Queue a buffer at the start of a list. This function takes no locks
1954 * and you must therefore hold required locks before calling it.
1956 * A buffer cannot be placed on two lists at the same time.
1958 static inline void __skb_queue_head(struct sk_buff_head *list,
1959 struct sk_buff *newsk)
1961 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1963 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1966 * __skb_queue_tail - queue a buffer at the list tail
1967 * @list: list to use
1968 * @newsk: buffer to queue
1970 * Queue a buffer at the end of a list. This function takes no locks
1971 * and you must therefore hold required locks before calling it.
1973 * A buffer cannot be placed on two lists at the same time.
1975 static inline void __skb_queue_tail(struct sk_buff_head *list,
1976 struct sk_buff *newsk)
1978 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1980 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1983 * remove sk_buff from list. _Must_ be called atomically, and with
1986 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1987 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1989 struct sk_buff *next, *prev;
1994 skb->next = skb->prev = NULL;
2000 * __skb_dequeue - remove from the head of the queue
2001 * @list: list to dequeue from
2003 * Remove the head of the list. This function does not take any locks
2004 * so must be used with appropriate locks held only. The head item is
2005 * returned or %NULL if the list is empty.
2007 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2009 struct sk_buff *skb = skb_peek(list);
2011 __skb_unlink(skb, list);
2014 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2017 * __skb_dequeue_tail - remove from the tail of the queue
2018 * @list: list to dequeue from
2020 * Remove the tail of the list. This function does not take any locks
2021 * so must be used with appropriate locks held only. The tail item is
2022 * returned or %NULL if the list is empty.
2024 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2026 struct sk_buff *skb = skb_peek_tail(list);
2028 __skb_unlink(skb, list);
2031 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2034 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2036 return skb->data_len;
2039 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2041 return skb->len - skb->data_len;
2044 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2046 unsigned int i, len = 0;
2048 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2049 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2053 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2055 return skb_headlen(skb) + __skb_pagelen(skb);
2059 * __skb_fill_page_desc - initialise a paged fragment in an skb
2060 * @skb: buffer containing fragment to be initialised
2061 * @i: paged fragment index to initialise
2062 * @page: the page to use for this fragment
2063 * @off: the offset to the data with @page
2064 * @size: the length of the data
2066 * Initialises the @i'th fragment of @skb to point to &size bytes at
2067 * offset @off within @page.
2069 * Does not take any additional reference on the fragment.
2071 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2072 struct page *page, int off, int size)
2074 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2077 * Propagate page pfmemalloc to the skb if we can. The problem is
2078 * that not all callers have unique ownership of the page but rely
2079 * on page_is_pfmemalloc doing the right thing(tm).
2081 frag->page.p = page;
2082 frag->page_offset = off;
2083 skb_frag_size_set(frag, size);
2085 page = compound_head(page);
2086 if (page_is_pfmemalloc(page))
2087 skb->pfmemalloc = true;
2091 * skb_fill_page_desc - initialise a paged fragment in an skb
2092 * @skb: buffer containing fragment to be initialised
2093 * @i: paged fragment index to initialise
2094 * @page: the page to use for this fragment
2095 * @off: the offset to the data with @page
2096 * @size: the length of the data
2098 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2099 * @skb to point to @size bytes at offset @off within @page. In
2100 * addition updates @skb such that @i is the last fragment.
2102 * Does not take any additional reference on the fragment.
2104 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2105 struct page *page, int off, int size)
2107 __skb_fill_page_desc(skb, i, page, off, size);
2108 skb_shinfo(skb)->nr_frags = i + 1;
2111 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2112 int size, unsigned int truesize);
2114 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2115 unsigned int truesize);
2117 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2119 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2120 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2122 return skb->head + skb->tail;
2125 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2127 skb->tail = skb->data - skb->head;
2130 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2132 skb_reset_tail_pointer(skb);
2133 skb->tail += offset;
2136 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2137 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2142 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2144 skb->tail = skb->data;
2147 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2149 skb->tail = skb->data + offset;
2152 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2155 * Add data to an sk_buff
2157 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2158 void *skb_put(struct sk_buff *skb, unsigned int len);
2159 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2161 void *tmp = skb_tail_pointer(skb);
2162 SKB_LINEAR_ASSERT(skb);
2168 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2170 void *tmp = __skb_put(skb, len);
2172 memset(tmp, 0, len);
2176 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2179 void *tmp = __skb_put(skb, len);
2181 memcpy(tmp, data, len);
2185 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2187 *(u8 *)__skb_put(skb, 1) = val;
2190 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2192 void *tmp = skb_put(skb, len);
2194 memset(tmp, 0, len);
2199 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2202 void *tmp = skb_put(skb, len);
2204 memcpy(tmp, data, len);
2209 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2211 *(u8 *)skb_put(skb, 1) = val;
2214 void *skb_push(struct sk_buff *skb, unsigned int len);
2215 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2222 void *skb_pull(struct sk_buff *skb, unsigned int len);
2223 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2226 BUG_ON(skb->len < skb->data_len);
2227 return skb->data += len;
2230 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2232 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2235 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2237 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2239 if (len > skb_headlen(skb) &&
2240 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2243 return skb->data += len;
2246 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2248 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2251 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2253 if (likely(len <= skb_headlen(skb)))
2255 if (unlikely(len > skb->len))
2257 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2260 void skb_condense(struct sk_buff *skb);
2263 * skb_headroom - bytes at buffer head
2264 * @skb: buffer to check
2266 * Return the number of bytes of free space at the head of an &sk_buff.
2268 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2270 return skb->data - skb->head;
2274 * skb_tailroom - bytes at buffer end
2275 * @skb: buffer to check
2277 * Return the number of bytes of free space at the tail of an sk_buff
2279 static inline int skb_tailroom(const struct sk_buff *skb)
2281 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2285 * skb_availroom - bytes at buffer end
2286 * @skb: buffer to check
2288 * Return the number of bytes of free space at the tail of an sk_buff
2289 * allocated by sk_stream_alloc()
2291 static inline int skb_availroom(const struct sk_buff *skb)
2293 if (skb_is_nonlinear(skb))
2296 return skb->end - skb->tail - skb->reserved_tailroom;
2300 * skb_reserve - adjust headroom
2301 * @skb: buffer to alter
2302 * @len: bytes to move
2304 * Increase the headroom of an empty &sk_buff by reducing the tail
2305 * room. This is only allowed for an empty buffer.
2307 static inline void skb_reserve(struct sk_buff *skb, int len)
2314 * skb_tailroom_reserve - adjust reserved_tailroom
2315 * @skb: buffer to alter
2316 * @mtu: maximum amount of headlen permitted
2317 * @needed_tailroom: minimum amount of reserved_tailroom
2319 * Set reserved_tailroom so that headlen can be as large as possible but
2320 * not larger than mtu and tailroom cannot be smaller than
2322 * The required headroom should already have been reserved before using
2325 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2326 unsigned int needed_tailroom)
2328 SKB_LINEAR_ASSERT(skb);
2329 if (mtu < skb_tailroom(skb) - needed_tailroom)
2330 /* use at most mtu */
2331 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2333 /* use up to all available space */
2334 skb->reserved_tailroom = needed_tailroom;
2337 #define ENCAP_TYPE_ETHER 0
2338 #define ENCAP_TYPE_IPPROTO 1
2340 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2343 skb->inner_protocol = protocol;
2344 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2347 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2350 skb->inner_ipproto = ipproto;
2351 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2354 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2356 skb->inner_mac_header = skb->mac_header;
2357 skb->inner_network_header = skb->network_header;
2358 skb->inner_transport_header = skb->transport_header;
2361 static inline void skb_reset_mac_len(struct sk_buff *skb)
2363 skb->mac_len = skb->network_header - skb->mac_header;
2366 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2369 return skb->head + skb->inner_transport_header;
2372 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2374 return skb_inner_transport_header(skb) - skb->data;
2377 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2379 skb->inner_transport_header = skb->data - skb->head;
2382 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2385 skb_reset_inner_transport_header(skb);
2386 skb->inner_transport_header += offset;
2389 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2391 return skb->head + skb->inner_network_header;
2394 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2396 skb->inner_network_header = skb->data - skb->head;
2399 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2402 skb_reset_inner_network_header(skb);
2403 skb->inner_network_header += offset;
2406 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2408 return skb->head + skb->inner_mac_header;
2411 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2413 skb->inner_mac_header = skb->data - skb->head;
2416 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2419 skb_reset_inner_mac_header(skb);
2420 skb->inner_mac_header += offset;
2422 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2424 return skb->transport_header != (typeof(skb->transport_header))~0U;
2427 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2429 return skb->head + skb->transport_header;
2432 static inline void skb_reset_transport_header(struct sk_buff *skb)
2434 skb->transport_header = skb->data - skb->head;
2437 static inline void skb_set_transport_header(struct sk_buff *skb,
2440 skb_reset_transport_header(skb);
2441 skb->transport_header += offset;
2444 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2446 return skb->head + skb->network_header;
2449 static inline void skb_reset_network_header(struct sk_buff *skb)
2451 skb->network_header = skb->data - skb->head;
2454 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2456 skb_reset_network_header(skb);
2457 skb->network_header += offset;
2460 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2462 return skb->head + skb->mac_header;
2465 static inline int skb_mac_offset(const struct sk_buff *skb)
2467 return skb_mac_header(skb) - skb->data;
2470 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2472 return skb->network_header - skb->mac_header;
2475 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2477 return skb->mac_header != (typeof(skb->mac_header))~0U;
2480 static inline void skb_reset_mac_header(struct sk_buff *skb)
2482 skb->mac_header = skb->data - skb->head;
2485 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2487 skb_reset_mac_header(skb);
2488 skb->mac_header += offset;
2491 static inline void skb_pop_mac_header(struct sk_buff *skb)
2493 skb->mac_header = skb->network_header;
2496 static inline void skb_probe_transport_header(struct sk_buff *skb)
2498 struct flow_keys_basic keys;
2500 if (skb_transport_header_was_set(skb))
2503 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2505 skb_set_transport_header(skb, keys.control.thoff);
2508 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2510 if (skb_mac_header_was_set(skb)) {
2511 const unsigned char *old_mac = skb_mac_header(skb);
2513 skb_set_mac_header(skb, -skb->mac_len);
2514 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2518 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2520 return skb->csum_start - skb_headroom(skb);
2523 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2525 return skb->head + skb->csum_start;
2528 static inline int skb_transport_offset(const struct sk_buff *skb)
2530 return skb_transport_header(skb) - skb->data;
2533 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2535 return skb->transport_header - skb->network_header;
2538 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2540 return skb->inner_transport_header - skb->inner_network_header;
2543 static inline int skb_network_offset(const struct sk_buff *skb)
2545 return skb_network_header(skb) - skb->data;
2548 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2550 return skb_inner_network_header(skb) - skb->data;
2553 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2555 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2559 * CPUs often take a performance hit when accessing unaligned memory
2560 * locations. The actual performance hit varies, it can be small if the
2561 * hardware handles it or large if we have to take an exception and fix it
2564 * Since an ethernet header is 14 bytes network drivers often end up with
2565 * the IP header at an unaligned offset. The IP header can be aligned by
2566 * shifting the start of the packet by 2 bytes. Drivers should do this
2569 * skb_reserve(skb, NET_IP_ALIGN);
2571 * The downside to this alignment of the IP header is that the DMA is now
2572 * unaligned. On some architectures the cost of an unaligned DMA is high
2573 * and this cost outweighs the gains made by aligning the IP header.
2575 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2578 #ifndef NET_IP_ALIGN
2579 #define NET_IP_ALIGN 2
2583 * The networking layer reserves some headroom in skb data (via
2584 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2585 * the header has to grow. In the default case, if the header has to grow
2586 * 32 bytes or less we avoid the reallocation.
2588 * Unfortunately this headroom changes the DMA alignment of the resulting
2589 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2590 * on some architectures. An architecture can override this value,
2591 * perhaps setting it to a cacheline in size (since that will maintain
2592 * cacheline alignment of the DMA). It must be a power of 2.
2594 * Various parts of the networking layer expect at least 32 bytes of
2595 * headroom, you should not reduce this.
2597 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2598 * to reduce average number of cache lines per packet.
2599 * get_rps_cpus() for example only access one 64 bytes aligned block :
2600 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2603 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2606 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2608 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2610 if (WARN_ON(skb_is_nonlinear(skb)))
2613 skb_set_tail_pointer(skb, len);
2616 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2618 __skb_set_length(skb, len);
2621 void skb_trim(struct sk_buff *skb, unsigned int len);
2623 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2626 return ___pskb_trim(skb, len);
2627 __skb_trim(skb, len);
2631 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2633 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2637 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2638 * @skb: buffer to alter
2641 * This is identical to pskb_trim except that the caller knows that
2642 * the skb is not cloned so we should never get an error due to out-
2645 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2647 int err = pskb_trim(skb, len);
2651 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2653 unsigned int diff = len - skb->len;
2655 if (skb_tailroom(skb) < diff) {
2656 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2661 __skb_set_length(skb, len);
2666 * skb_orphan - orphan a buffer
2667 * @skb: buffer to orphan
2669 * If a buffer currently has an owner then we call the owner's
2670 * destructor function and make the @skb unowned. The buffer continues
2671 * to exist but is no longer charged to its former owner.
2673 static inline void skb_orphan(struct sk_buff *skb)
2675 if (skb->destructor) {
2676 skb->destructor(skb);
2677 skb->destructor = NULL;
2685 * skb_orphan_frags - orphan the frags contained in a buffer
2686 * @skb: buffer to orphan frags from
2687 * @gfp_mask: allocation mask for replacement pages
2689 * For each frag in the SKB which needs a destructor (i.e. has an
2690 * owner) create a copy of that frag and release the original
2691 * page by calling the destructor.
2693 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2695 if (likely(!skb_zcopy(skb)))
2697 if (!skb_zcopy_is_nouarg(skb) &&
2698 skb_uarg(skb)->callback == sock_zerocopy_callback)
2700 return skb_copy_ubufs(skb, gfp_mask);
2703 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2704 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2706 if (likely(!skb_zcopy(skb)))
2708 return skb_copy_ubufs(skb, gfp_mask);
2712 * __skb_queue_purge - empty a list
2713 * @list: list to empty
2715 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2716 * the list and one reference dropped. This function does not take the
2717 * list lock and the caller must hold the relevant locks to use it.
2719 static inline void __skb_queue_purge(struct sk_buff_head *list)
2721 struct sk_buff *skb;
2722 while ((skb = __skb_dequeue(list)) != NULL)
2725 void skb_queue_purge(struct sk_buff_head *list);
2727 unsigned int skb_rbtree_purge(struct rb_root *root);
2729 void *netdev_alloc_frag(unsigned int fragsz);
2731 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2735 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2736 * @dev: network device to receive on
2737 * @length: length to allocate
2739 * Allocate a new &sk_buff and assign it a usage count of one. The
2740 * buffer has unspecified headroom built in. Users should allocate
2741 * the headroom they think they need without accounting for the
2742 * built in space. The built in space is used for optimisations.
2744 * %NULL is returned if there is no free memory. Although this function
2745 * allocates memory it can be called from an interrupt.
2747 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2748 unsigned int length)
2750 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2753 /* legacy helper around __netdev_alloc_skb() */
2754 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2757 return __netdev_alloc_skb(NULL, length, gfp_mask);
2760 /* legacy helper around netdev_alloc_skb() */
2761 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2763 return netdev_alloc_skb(NULL, length);
2767 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2768 unsigned int length, gfp_t gfp)
2770 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2772 if (NET_IP_ALIGN && skb)
2773 skb_reserve(skb, NET_IP_ALIGN);
2777 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2778 unsigned int length)
2780 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2783 static inline void skb_free_frag(void *addr)
2785 page_frag_free(addr);
2788 void *napi_alloc_frag(unsigned int fragsz);
2789 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2790 unsigned int length, gfp_t gfp_mask);
2791 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2792 unsigned int length)
2794 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2796 void napi_consume_skb(struct sk_buff *skb, int budget);
2798 void __kfree_skb_flush(void);
2799 void __kfree_skb_defer(struct sk_buff *skb);
2802 * __dev_alloc_pages - allocate page for network Rx
2803 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2804 * @order: size of the allocation
2806 * Allocate a new page.
2808 * %NULL is returned if there is no free memory.
2810 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2813 /* This piece of code contains several assumptions.
2814 * 1. This is for device Rx, therefor a cold page is preferred.
2815 * 2. The expectation is the user wants a compound page.
2816 * 3. If requesting a order 0 page it will not be compound
2817 * due to the check to see if order has a value in prep_new_page
2818 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2819 * code in gfp_to_alloc_flags that should be enforcing this.
2821 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2823 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2826 static inline struct page *dev_alloc_pages(unsigned int order)
2828 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2832 * __dev_alloc_page - allocate a page for network Rx
2833 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2835 * Allocate a new page.
2837 * %NULL is returned if there is no free memory.
2839 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2841 return __dev_alloc_pages(gfp_mask, 0);
2844 static inline struct page *dev_alloc_page(void)
2846 return dev_alloc_pages(0);
2850 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2851 * @page: The page that was allocated from skb_alloc_page
2852 * @skb: The skb that may need pfmemalloc set
2854 static inline void skb_propagate_pfmemalloc(struct page *page,
2855 struct sk_buff *skb)
2857 if (page_is_pfmemalloc(page))
2858 skb->pfmemalloc = true;
2862 * skb_frag_page - retrieve the page referred to by a paged fragment
2863 * @frag: the paged fragment
2865 * Returns the &struct page associated with @frag.
2867 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2869 return frag->page.p;
2873 * __skb_frag_ref - take an addition reference on a paged fragment.
2874 * @frag: the paged fragment
2876 * Takes an additional reference on the paged fragment @frag.
2878 static inline void __skb_frag_ref(skb_frag_t *frag)
2880 get_page(skb_frag_page(frag));
2884 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2886 * @f: the fragment offset.
2888 * Takes an additional reference on the @f'th paged fragment of @skb.
2890 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2892 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2896 * __skb_frag_unref - release a reference on a paged fragment.
2897 * @frag: the paged fragment
2899 * Releases a reference on the paged fragment @frag.
2901 static inline void __skb_frag_unref(skb_frag_t *frag)
2903 put_page(skb_frag_page(frag));
2907 * skb_frag_unref - release a reference on a paged fragment of an skb.
2909 * @f: the fragment offset
2911 * Releases a reference on the @f'th paged fragment of @skb.
2913 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2915 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2919 * skb_frag_address - gets the address of the data contained in a paged fragment
2920 * @frag: the paged fragment buffer
2922 * Returns the address of the data within @frag. The page must already
2925 static inline void *skb_frag_address(const skb_frag_t *frag)
2927 return page_address(skb_frag_page(frag)) + frag->page_offset;
2931 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2932 * @frag: the paged fragment buffer
2934 * Returns the address of the data within @frag. Checks that the page
2935 * is mapped and returns %NULL otherwise.
2937 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2939 void *ptr = page_address(skb_frag_page(frag));
2943 return ptr + frag->page_offset;
2947 * __skb_frag_set_page - sets the page contained in a paged fragment
2948 * @frag: the paged fragment
2949 * @page: the page to set
2951 * Sets the fragment @frag to contain @page.
2953 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2955 frag->page.p = page;
2959 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2961 * @f: the fragment offset
2962 * @page: the page to set
2964 * Sets the @f'th fragment of @skb to contain @page.
2966 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2969 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2972 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2975 * skb_frag_dma_map - maps a paged fragment via the DMA API
2976 * @dev: the device to map the fragment to
2977 * @frag: the paged fragment to map
2978 * @offset: the offset within the fragment (starting at the
2979 * fragment's own offset)
2980 * @size: the number of bytes to map
2981 * @dir: the direction of the mapping (``PCI_DMA_*``)
2983 * Maps the page associated with @frag to @device.
2985 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2986 const skb_frag_t *frag,
2987 size_t offset, size_t size,
2988 enum dma_data_direction dir)
2990 return dma_map_page(dev, skb_frag_page(frag),
2991 frag->page_offset + offset, size, dir);
2994 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2997 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3001 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3004 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3009 * skb_clone_writable - is the header of a clone writable
3010 * @skb: buffer to check
3011 * @len: length up to which to write
3013 * Returns true if modifying the header part of the cloned buffer
3014 * does not requires the data to be copied.
3016 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3018 return !skb_header_cloned(skb) &&
3019 skb_headroom(skb) + len <= skb->hdr_len;
3022 static inline int skb_try_make_writable(struct sk_buff *skb,
3023 unsigned int write_len)
3025 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3026 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3029 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3034 if (headroom > skb_headroom(skb))
3035 delta = headroom - skb_headroom(skb);
3037 if (delta || cloned)
3038 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3044 * skb_cow - copy header of skb when it is required
3045 * @skb: buffer to cow
3046 * @headroom: needed headroom
3048 * If the skb passed lacks sufficient headroom or its data part
3049 * is shared, data is reallocated. If reallocation fails, an error
3050 * is returned and original skb is not changed.
3052 * The result is skb with writable area skb->head...skb->tail
3053 * and at least @headroom of space at head.
3055 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3057 return __skb_cow(skb, headroom, skb_cloned(skb));
3061 * skb_cow_head - skb_cow but only making the head writable
3062 * @skb: buffer to cow
3063 * @headroom: needed headroom
3065 * This function is identical to skb_cow except that we replace the
3066 * skb_cloned check by skb_header_cloned. It should be used when
3067 * you only need to push on some header and do not need to modify
3070 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3072 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3076 * skb_padto - pad an skbuff up to a minimal size
3077 * @skb: buffer to pad
3078 * @len: minimal length
3080 * Pads up a buffer to ensure the trailing bytes exist and are
3081 * blanked. If the buffer already contains sufficient data it
3082 * is untouched. Otherwise it is extended. Returns zero on
3083 * success. The skb is freed on error.
3085 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3087 unsigned int size = skb->len;
3088 if (likely(size >= len))
3090 return skb_pad(skb, len - size);
3094 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3095 * @skb: buffer to pad
3096 * @len: minimal length
3097 * @free_on_error: free buffer on error
3099 * Pads up a buffer to ensure the trailing bytes exist and are
3100 * blanked. If the buffer already contains sufficient data it
3101 * is untouched. Otherwise it is extended. Returns zero on
3102 * success. The skb is freed on error if @free_on_error is true.
3104 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
3107 unsigned int size = skb->len;
3109 if (unlikely(size < len)) {
3111 if (__skb_pad(skb, len, free_on_error))
3113 __skb_put(skb, len);
3119 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3120 * @skb: buffer to pad
3121 * @len: minimal length
3123 * Pads up a buffer to ensure the trailing bytes exist and are
3124 * blanked. If the buffer already contains sufficient data it
3125 * is untouched. Otherwise it is extended. Returns zero on
3126 * success. The skb is freed on error.
3128 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
3130 return __skb_put_padto(skb, len, true);
3133 static inline int skb_add_data(struct sk_buff *skb,
3134 struct iov_iter *from, int copy)
3136 const int off = skb->len;
3138 if (skb->ip_summed == CHECKSUM_NONE) {
3140 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3142 skb->csum = csum_block_add(skb->csum, csum, off);
3145 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3148 __skb_trim(skb, off);
3152 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3153 const struct page *page, int off)
3158 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
3160 return page == skb_frag_page(frag) &&
3161 off == frag->page_offset + skb_frag_size(frag);
3166 static inline int __skb_linearize(struct sk_buff *skb)
3168 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3172 * skb_linearize - convert paged skb to linear one
3173 * @skb: buffer to linarize
3175 * If there is no free memory -ENOMEM is returned, otherwise zero
3176 * is returned and the old skb data released.
3178 static inline int skb_linearize(struct sk_buff *skb)
3180 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3184 * skb_has_shared_frag - can any frag be overwritten
3185 * @skb: buffer to test
3187 * Return true if the skb has at least one frag that might be modified
3188 * by an external entity (as in vmsplice()/sendfile())
3190 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3192 return skb_is_nonlinear(skb) &&
3193 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3197 * skb_linearize_cow - make sure skb is linear and writable
3198 * @skb: buffer to process
3200 * If there is no free memory -ENOMEM is returned, otherwise zero
3201 * is returned and the old skb data released.
3203 static inline int skb_linearize_cow(struct sk_buff *skb)
3205 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3206 __skb_linearize(skb) : 0;
3209 static __always_inline void
3210 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3213 if (skb->ip_summed == CHECKSUM_COMPLETE)
3214 skb->csum = csum_block_sub(skb->csum,
3215 csum_partial(start, len, 0), off);
3216 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3217 skb_checksum_start_offset(skb) < 0)
3218 skb->ip_summed = CHECKSUM_NONE;
3222 * skb_postpull_rcsum - update checksum for received skb after pull
3223 * @skb: buffer to update
3224 * @start: start of data before pull
3225 * @len: length of data pulled
3227 * After doing a pull on a received packet, you need to call this to
3228 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3229 * CHECKSUM_NONE so that it can be recomputed from scratch.
3231 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3232 const void *start, unsigned int len)
3234 __skb_postpull_rcsum(skb, start, len, 0);
3237 static __always_inline void
3238 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3241 if (skb->ip_summed == CHECKSUM_COMPLETE)
3242 skb->csum = csum_block_add(skb->csum,
3243 csum_partial(start, len, 0), off);
3247 * skb_postpush_rcsum - update checksum for received skb after push
3248 * @skb: buffer to update
3249 * @start: start of data after push
3250 * @len: length of data pushed
3252 * After doing a push on a received packet, you need to call this to
3253 * update the CHECKSUM_COMPLETE checksum.
3255 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3256 const void *start, unsigned int len)
3258 __skb_postpush_rcsum(skb, start, len, 0);
3261 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3264 * skb_push_rcsum - push skb and update receive checksum
3265 * @skb: buffer to update
3266 * @len: length of data pulled
3268 * This function performs an skb_push on the packet and updates
3269 * the CHECKSUM_COMPLETE checksum. It should be used on
3270 * receive path processing instead of skb_push unless you know
3271 * that the checksum difference is zero (e.g., a valid IP header)
3272 * or you are setting ip_summed to CHECKSUM_NONE.
3274 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3277 skb_postpush_rcsum(skb, skb->data, len);
3281 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3283 * pskb_trim_rcsum - trim received skb and update checksum
3284 * @skb: buffer to trim
3287 * This is exactly the same as pskb_trim except that it ensures the
3288 * checksum of received packets are still valid after the operation.
3289 * It can change skb pointers.
3292 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3294 if (likely(len >= skb->len))
3296 return pskb_trim_rcsum_slow(skb, len);
3299 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3301 if (skb->ip_summed == CHECKSUM_COMPLETE)
3302 skb->ip_summed = CHECKSUM_NONE;
3303 __skb_trim(skb, len);
3307 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3309 if (skb->ip_summed == CHECKSUM_COMPLETE)
3310 skb->ip_summed = CHECKSUM_NONE;
3311 return __skb_grow(skb, len);
3314 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3315 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3316 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3317 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3318 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3320 #define skb_queue_walk(queue, skb) \
3321 for (skb = (queue)->next; \
3322 skb != (struct sk_buff *)(queue); \
3325 #define skb_queue_walk_safe(queue, skb, tmp) \
3326 for (skb = (queue)->next, tmp = skb->next; \
3327 skb != (struct sk_buff *)(queue); \
3328 skb = tmp, tmp = skb->next)
3330 #define skb_queue_walk_from(queue, skb) \
3331 for (; skb != (struct sk_buff *)(queue); \
3334 #define skb_rbtree_walk(skb, root) \
3335 for (skb = skb_rb_first(root); skb != NULL; \
3336 skb = skb_rb_next(skb))
3338 #define skb_rbtree_walk_from(skb) \
3339 for (; skb != NULL; \
3340 skb = skb_rb_next(skb))
3342 #define skb_rbtree_walk_from_safe(skb, tmp) \
3343 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3346 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3347 for (tmp = skb->next; \
3348 skb != (struct sk_buff *)(queue); \
3349 skb = tmp, tmp = skb->next)
3351 #define skb_queue_reverse_walk(queue, skb) \
3352 for (skb = (queue)->prev; \
3353 skb != (struct sk_buff *)(queue); \
3356 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3357 for (skb = (queue)->prev, tmp = skb->prev; \
3358 skb != (struct sk_buff *)(queue); \
3359 skb = tmp, tmp = skb->prev)
3361 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3362 for (tmp = skb->prev; \
3363 skb != (struct sk_buff *)(queue); \
3364 skb = tmp, tmp = skb->prev)
3366 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3368 return skb_shinfo(skb)->frag_list != NULL;
3371 static inline void skb_frag_list_init(struct sk_buff *skb)
3373 skb_shinfo(skb)->frag_list = NULL;
3376 #define skb_walk_frags(skb, iter) \
3377 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3380 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3381 const struct sk_buff *skb);
3382 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3383 struct sk_buff_head *queue,
3385 void (*destructor)(struct sock *sk,
3386 struct sk_buff *skb),
3388 struct sk_buff **last);
3389 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3390 void (*destructor)(struct sock *sk,
3391 struct sk_buff *skb),
3393 struct sk_buff **last);
3394 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3395 void (*destructor)(struct sock *sk,
3396 struct sk_buff *skb),
3397 int *off, int *err);
3398 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3400 __poll_t datagram_poll(struct file *file, struct socket *sock,
3401 struct poll_table_struct *wait);
3402 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3403 struct iov_iter *to, int size);
3404 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3405 struct msghdr *msg, int size)
3407 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3409 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3410 struct msghdr *msg);
3411 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3412 struct iov_iter *to, int len,
3413 struct ahash_request *hash);
3414 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3415 struct iov_iter *from, int len);
3416 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3417 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3418 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3419 static inline void skb_free_datagram_locked(struct sock *sk,
3420 struct sk_buff *skb)
3422 __skb_free_datagram_locked(sk, skb, 0);
3424 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3425 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3426 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3427 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3428 int len, __wsum csum);
3429 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3430 struct pipe_inode_info *pipe, unsigned int len,
3431 unsigned int flags);
3432 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3434 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3435 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3436 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3438 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3439 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3440 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3441 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3442 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3443 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3444 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3445 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3446 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3447 int skb_vlan_pop(struct sk_buff *skb);
3448 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3449 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3452 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3454 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3457 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3459 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3462 struct skb_checksum_ops {
3463 __wsum (*update)(const void *mem, int len, __wsum wsum);
3464 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3467 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3469 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3470 __wsum csum, const struct skb_checksum_ops *ops);
3471 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3474 static inline void * __must_check
3475 __skb_header_pointer(const struct sk_buff *skb, int offset,
3476 int len, void *data, int hlen, void *buffer)
3478 if (hlen - offset >= len)
3479 return data + offset;
3482 skb_copy_bits(skb, offset, buffer, len) < 0)
3488 static inline void * __must_check
3489 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3491 return __skb_header_pointer(skb, offset, len, skb->data,
3492 skb_headlen(skb), buffer);
3496 * skb_needs_linearize - check if we need to linearize a given skb
3497 * depending on the given device features.
3498 * @skb: socket buffer to check
3499 * @features: net device features
3501 * Returns true if either:
3502 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3503 * 2. skb is fragmented and the device does not support SG.
3505 static inline bool skb_needs_linearize(struct sk_buff *skb,
3506 netdev_features_t features)
3508 return skb_is_nonlinear(skb) &&
3509 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3510 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3513 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3515 const unsigned int len)
3517 memcpy(to, skb->data, len);
3520 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3521 const int offset, void *to,
3522 const unsigned int len)
3524 memcpy(to, skb->data + offset, len);
3527 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3529 const unsigned int len)
3531 memcpy(skb->data, from, len);
3534 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3537 const unsigned int len)
3539 memcpy(skb->data + offset, from, len);
3542 void skb_init(void);
3544 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3550 * skb_get_timestamp - get timestamp from a skb
3551 * @skb: skb to get stamp from
3552 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
3554 * Timestamps are stored in the skb as offsets to a base timestamp.
3555 * This function converts the offset back to a struct timeval and stores
3558 static inline void skb_get_timestamp(const struct sk_buff *skb,
3559 struct __kernel_old_timeval *stamp)
3561 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
3564 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
3565 struct __kernel_sock_timeval *stamp)
3567 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3569 stamp->tv_sec = ts.tv_sec;
3570 stamp->tv_usec = ts.tv_nsec / 1000;
3573 static inline void skb_get_timestampns(const struct sk_buff *skb,
3574 struct timespec *stamp)
3576 *stamp = ktime_to_timespec(skb->tstamp);
3579 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
3580 struct __kernel_timespec *stamp)
3582 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3584 stamp->tv_sec = ts.tv_sec;
3585 stamp->tv_nsec = ts.tv_nsec;
3588 static inline void __net_timestamp(struct sk_buff *skb)
3590 skb->tstamp = ktime_get_real();
3593 static inline ktime_t net_timedelta(ktime_t t)
3595 return ktime_sub(ktime_get_real(), t);
3598 static inline ktime_t net_invalid_timestamp(void)
3603 static inline u8 skb_metadata_len(const struct sk_buff *skb)
3605 return skb_shinfo(skb)->meta_len;
3608 static inline void *skb_metadata_end(const struct sk_buff *skb)
3610 return skb_mac_header(skb);
3613 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3614 const struct sk_buff *skb_b,
3617 const void *a = skb_metadata_end(skb_a);
3618 const void *b = skb_metadata_end(skb_b);
3619 /* Using more efficient varaiant than plain call to memcmp(). */
3620 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3624 #define __it(x, op) (x -= sizeof(u##op))
3625 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3626 case 32: diffs |= __it_diff(a, b, 64);
3628 case 24: diffs |= __it_diff(a, b, 64);
3630 case 16: diffs |= __it_diff(a, b, 64);
3632 case 8: diffs |= __it_diff(a, b, 64);
3634 case 28: diffs |= __it_diff(a, b, 64);
3636 case 20: diffs |= __it_diff(a, b, 64);
3638 case 12: diffs |= __it_diff(a, b, 64);
3640 case 4: diffs |= __it_diff(a, b, 32);
3645 return memcmp(a - meta_len, b - meta_len, meta_len);
3649 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3650 const struct sk_buff *skb_b)
3652 u8 len_a = skb_metadata_len(skb_a);
3653 u8 len_b = skb_metadata_len(skb_b);
3655 if (!(len_a | len_b))
3658 return len_a != len_b ?
3659 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3662 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3664 skb_shinfo(skb)->meta_len = meta_len;
3667 static inline void skb_metadata_clear(struct sk_buff *skb)
3669 skb_metadata_set(skb, 0);
3672 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3674 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3676 void skb_clone_tx_timestamp(struct sk_buff *skb);
3677 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3679 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3681 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3685 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3690 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3693 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3695 * PHY drivers may accept clones of transmitted packets for
3696 * timestamping via their phy_driver.txtstamp method. These drivers
3697 * must call this function to return the skb back to the stack with a
3700 * @skb: clone of the the original outgoing packet
3701 * @hwtstamps: hardware time stamps
3704 void skb_complete_tx_timestamp(struct sk_buff *skb,
3705 struct skb_shared_hwtstamps *hwtstamps);
3707 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3708 struct skb_shared_hwtstamps *hwtstamps,
3709 struct sock *sk, int tstype);
3712 * skb_tstamp_tx - queue clone of skb with send time stamps
3713 * @orig_skb: the original outgoing packet
3714 * @hwtstamps: hardware time stamps, may be NULL if not available
3716 * If the skb has a socket associated, then this function clones the
3717 * skb (thus sharing the actual data and optional structures), stores
3718 * the optional hardware time stamping information (if non NULL) or
3719 * generates a software time stamp (otherwise), then queues the clone
3720 * to the error queue of the socket. Errors are silently ignored.
3722 void skb_tstamp_tx(struct sk_buff *orig_skb,
3723 struct skb_shared_hwtstamps *hwtstamps);
3726 * skb_tx_timestamp() - Driver hook for transmit timestamping
3728 * Ethernet MAC Drivers should call this function in their hard_xmit()
3729 * function immediately before giving the sk_buff to the MAC hardware.
3731 * Specifically, one should make absolutely sure that this function is
3732 * called before TX completion of this packet can trigger. Otherwise
3733 * the packet could potentially already be freed.
3735 * @skb: A socket buffer.
3737 static inline void skb_tx_timestamp(struct sk_buff *skb)
3739 skb_clone_tx_timestamp(skb);
3740 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3741 skb_tstamp_tx(skb, NULL);
3745 * skb_complete_wifi_ack - deliver skb with wifi status
3747 * @skb: the original outgoing packet
3748 * @acked: ack status
3751 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3753 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3754 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3756 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3758 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3760 (skb->ip_summed == CHECKSUM_PARTIAL &&
3761 skb_checksum_start_offset(skb) >= 0));
3765 * skb_checksum_complete - Calculate checksum of an entire packet
3766 * @skb: packet to process
3768 * This function calculates the checksum over the entire packet plus
3769 * the value of skb->csum. The latter can be used to supply the
3770 * checksum of a pseudo header as used by TCP/UDP. It returns the
3773 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3774 * this function can be used to verify that checksum on received
3775 * packets. In that case the function should return zero if the
3776 * checksum is correct. In particular, this function will return zero
3777 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3778 * hardware has already verified the correctness of the checksum.
3780 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3782 return skb_csum_unnecessary(skb) ?
3783 0 : __skb_checksum_complete(skb);
3786 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3788 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3789 if (skb->csum_level == 0)
3790 skb->ip_summed = CHECKSUM_NONE;
3796 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3798 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3799 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3801 } else if (skb->ip_summed == CHECKSUM_NONE) {
3802 skb->ip_summed = CHECKSUM_UNNECESSARY;
3803 skb->csum_level = 0;
3807 /* Check if we need to perform checksum complete validation.
3809 * Returns true if checksum complete is needed, false otherwise
3810 * (either checksum is unnecessary or zero checksum is allowed).
3812 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3816 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3817 skb->csum_valid = 1;
3818 __skb_decr_checksum_unnecessary(skb);
3825 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
3828 #define CHECKSUM_BREAK 76
3830 /* Unset checksum-complete
3832 * Unset checksum complete can be done when packet is being modified
3833 * (uncompressed for instance) and checksum-complete value is
3836 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3838 if (skb->ip_summed == CHECKSUM_COMPLETE)
3839 skb->ip_summed = CHECKSUM_NONE;
3842 /* Validate (init) checksum based on checksum complete.
3845 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3846 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3847 * checksum is stored in skb->csum for use in __skb_checksum_complete
3848 * non-zero: value of invalid checksum
3851 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3855 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3856 if (!csum_fold(csum_add(psum, skb->csum))) {
3857 skb->csum_valid = 1;
3864 if (complete || skb->len <= CHECKSUM_BREAK) {
3867 csum = __skb_checksum_complete(skb);
3868 skb->csum_valid = !csum;
3875 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3880 /* Perform checksum validate (init). Note that this is a macro since we only
3881 * want to calculate the pseudo header which is an input function if necessary.
3882 * First we try to validate without any computation (checksum unnecessary) and
3883 * then calculate based on checksum complete calling the function to compute
3887 * 0: checksum is validated or try to in skb_checksum_complete
3888 * non-zero: value of invalid checksum
3890 #define __skb_checksum_validate(skb, proto, complete, \
3891 zero_okay, check, compute_pseudo) \
3893 __sum16 __ret = 0; \
3894 skb->csum_valid = 0; \
3895 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3896 __ret = __skb_checksum_validate_complete(skb, \
3897 complete, compute_pseudo(skb, proto)); \
3901 #define skb_checksum_init(skb, proto, compute_pseudo) \
3902 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3904 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3905 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3907 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3908 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3910 #define skb_checksum_validate_zero_check(skb, proto, check, \
3912 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3914 #define skb_checksum_simple_validate(skb) \
3915 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3917 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3919 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3922 static inline void __skb_checksum_convert(struct sk_buff *skb,
3923 __sum16 check, __wsum pseudo)
3925 skb->csum = ~pseudo;
3926 skb->ip_summed = CHECKSUM_COMPLETE;
3929 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3931 if (__skb_checksum_convert_check(skb)) \
3932 __skb_checksum_convert(skb, check, \
3933 compute_pseudo(skb, proto)); \
3936 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3937 u16 start, u16 offset)
3939 skb->ip_summed = CHECKSUM_PARTIAL;
3940 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3941 skb->csum_offset = offset - start;
3944 /* Update skbuf and packet to reflect the remote checksum offload operation.
3945 * When called, ptr indicates the starting point for skb->csum when
3946 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3947 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3949 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3950 int start, int offset, bool nopartial)
3955 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3959 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3960 __skb_checksum_complete(skb);
3961 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3964 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3966 /* Adjust skb->csum since we changed the packet */
3967 skb->csum = csum_add(skb->csum, delta);
3970 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3972 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3973 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3979 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3980 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3981 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3983 if (nfct && atomic_dec_and_test(&nfct->use))
3984 nf_conntrack_destroy(nfct);
3986 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3989 atomic_inc(&nfct->use);
3993 #ifdef CONFIG_SKB_EXTENSIONS
3995 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4001 SKB_EXT_NUM, /* must be last */
4005 * struct skb_ext - sk_buff extensions
4006 * @refcnt: 1 on allocation, deallocated on 0
4007 * @offset: offset to add to @data to obtain extension address
4008 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4009 * @data: start of extension data, variable sized
4011 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4012 * to use 'u8' types while allowing up to 2kb worth of extension data.
4016 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4017 u8 chunks; /* same */
4018 char data[0] __aligned(8);
4021 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4022 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4023 void __skb_ext_put(struct skb_ext *ext);
4025 static inline void skb_ext_put(struct sk_buff *skb)
4027 if (skb->active_extensions)
4028 __skb_ext_put(skb->extensions);
4031 static inline void __skb_ext_copy(struct sk_buff *dst,
4032 const struct sk_buff *src)
4034 dst->active_extensions = src->active_extensions;
4036 if (src->active_extensions) {
4037 struct skb_ext *ext = src->extensions;
4039 refcount_inc(&ext->refcnt);
4040 dst->extensions = ext;
4044 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4047 __skb_ext_copy(dst, src);
4050 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4052 return !!ext->offset[i];
4055 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4057 return skb->active_extensions & (1 << id);
4060 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4062 if (skb_ext_exist(skb, id))
4063 __skb_ext_del(skb, id);
4066 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4068 if (skb_ext_exist(skb, id)) {
4069 struct skb_ext *ext = skb->extensions;
4071 return (void *)ext + (ext->offset[id] << 3);
4077 static inline void skb_ext_put(struct sk_buff *skb) {}
4078 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4079 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4080 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4081 #endif /* CONFIG_SKB_EXTENSIONS */
4083 static inline void nf_reset(struct sk_buff *skb)
4085 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4086 nf_conntrack_put(skb_nfct(skb));
4089 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4090 skb_ext_del(skb, SKB_EXT_BRIDGE_NF);
4094 static inline void nf_reset_trace(struct sk_buff *skb)
4096 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4101 static inline void ipvs_reset(struct sk_buff *skb)
4103 #if IS_ENABLED(CONFIG_IP_VS)
4104 skb->ipvs_property = 0;
4108 /* Note: This doesn't put any conntrack info in dst. */
4109 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4112 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4113 dst->_nfct = src->_nfct;
4114 nf_conntrack_get(skb_nfct(src));
4116 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4118 dst->nf_trace = src->nf_trace;
4122 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4124 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4125 nf_conntrack_put(skb_nfct(dst));
4127 __nf_copy(dst, src, true);
4130 #ifdef CONFIG_NETWORK_SECMARK
4131 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4133 to->secmark = from->secmark;
4136 static inline void skb_init_secmark(struct sk_buff *skb)
4141 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4144 static inline void skb_init_secmark(struct sk_buff *skb)
4148 static inline int secpath_exists(const struct sk_buff *skb)
4151 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4157 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4159 return !skb->destructor &&
4160 !secpath_exists(skb) &&
4162 !skb->_skb_refdst &&
4163 !skb_has_frag_list(skb);
4166 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4168 skb->queue_mapping = queue_mapping;
4171 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4173 return skb->queue_mapping;
4176 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4178 to->queue_mapping = from->queue_mapping;
4181 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4183 skb->queue_mapping = rx_queue + 1;
4186 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4188 return skb->queue_mapping - 1;
4191 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4193 return skb->queue_mapping != 0;
4196 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4198 skb->dst_pending_confirm = val;
4201 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4203 return skb->dst_pending_confirm != 0;
4206 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4209 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4215 /* Keeps track of mac header offset relative to skb->head.
4216 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4217 * For non-tunnel skb it points to skb_mac_header() and for
4218 * tunnel skb it points to outer mac header.
4219 * Keeps track of level of encapsulation of network headers.
4230 #define SKB_SGO_CB_OFFSET 32
4231 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
4233 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4235 return (skb_mac_header(inner_skb) - inner_skb->head) -
4236 SKB_GSO_CB(inner_skb)->mac_offset;
4239 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4241 int new_headroom, headroom;
4244 headroom = skb_headroom(skb);
4245 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4249 new_headroom = skb_headroom(skb);
4250 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4254 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4256 /* Do not update partial checksums if remote checksum is enabled. */
4257 if (skb->remcsum_offload)
4260 SKB_GSO_CB(skb)->csum = res;
4261 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4264 /* Compute the checksum for a gso segment. First compute the checksum value
4265 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4266 * then add in skb->csum (checksum from csum_start to end of packet).
4267 * skb->csum and csum_start are then updated to reflect the checksum of the
4268 * resultant packet starting from the transport header-- the resultant checksum
4269 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4272 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4274 unsigned char *csum_start = skb_transport_header(skb);
4275 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4276 __wsum partial = SKB_GSO_CB(skb)->csum;
4278 SKB_GSO_CB(skb)->csum = res;
4279 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4281 return csum_fold(csum_partial(csum_start, plen, partial));
4284 static inline bool skb_is_gso(const struct sk_buff *skb)
4286 return skb_shinfo(skb)->gso_size;
4289 /* Note: Should be called only if skb_is_gso(skb) is true */
4290 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4292 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4295 /* Note: Should be called only if skb_is_gso(skb) is true */
4296 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4298 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4301 /* Note: Should be called only if skb_is_gso(skb) is true */
4302 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4304 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4307 static inline void skb_gso_reset(struct sk_buff *skb)
4309 skb_shinfo(skb)->gso_size = 0;
4310 skb_shinfo(skb)->gso_segs = 0;
4311 skb_shinfo(skb)->gso_type = 0;
4314 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4317 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4319 shinfo->gso_size += increment;
4322 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4325 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4327 shinfo->gso_size -= decrement;
4330 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4332 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4334 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4335 * wanted then gso_type will be set. */
4336 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4338 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4339 unlikely(shinfo->gso_type == 0)) {
4340 __skb_warn_lro_forwarding(skb);
4346 static inline void skb_forward_csum(struct sk_buff *skb)
4348 /* Unfortunately we don't support this one. Any brave souls? */
4349 if (skb->ip_summed == CHECKSUM_COMPLETE)
4350 skb->ip_summed = CHECKSUM_NONE;
4354 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4355 * @skb: skb to check
4357 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4358 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4359 * use this helper, to document places where we make this assertion.
4361 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4364 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4368 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4370 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4371 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4372 unsigned int transport_len,
4373 __sum16(*skb_chkf)(struct sk_buff *skb));
4376 * skb_head_is_locked - Determine if the skb->head is locked down
4377 * @skb: skb to check
4379 * The head on skbs build around a head frag can be removed if they are
4380 * not cloned. This function returns true if the skb head is locked down
4381 * due to either being allocated via kmalloc, or by being a clone with
4382 * multiple references to the head.
4384 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4386 return !skb->head_frag || skb_cloned(skb);
4389 /* Local Checksum Offload.
4390 * Compute outer checksum based on the assumption that the
4391 * inner checksum will be offloaded later.
4392 * See Documentation/networking/checksum-offloads.rst for
4393 * explanation of how this works.
4394 * Fill in outer checksum adjustment (e.g. with sum of outer
4395 * pseudo-header) before calling.
4396 * Also ensure that inner checksum is in linear data area.
4398 static inline __wsum lco_csum(struct sk_buff *skb)
4400 unsigned char *csum_start = skb_checksum_start(skb);
4401 unsigned char *l4_hdr = skb_transport_header(skb);
4404 /* Start with complement of inner checksum adjustment */
4405 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4408 /* Add in checksum of our headers (incl. outer checksum
4409 * adjustment filled in by caller) and return result.
4411 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4414 #endif /* __KERNEL__ */
4415 #endif /* _LINUX_SKBUFF_H */