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/bvec.h>
18 #include <linux/cache.h>
19 #include <linux/rbtree.h>
20 #include <linux/socket.h>
21 #include <linux/refcount.h>
23 #include <linux/atomic.h>
24 #include <asm/types.h>
25 #include <linux/spinlock.h>
26 #include <linux/net.h>
27 #include <linux/textsearch.h>
28 #include <net/checksum.h>
29 #include <linux/rcupdate.h>
30 #include <linux/hrtimer.h>
31 #include <linux/dma-mapping.h>
32 #include <linux/netdev_features.h>
33 #include <linux/sched.h>
34 #include <linux/sched/clock.h>
35 #include <net/flow_dissector.h>
36 #include <linux/splice.h>
37 #include <linux/in6.h>
38 #include <linux/if_packet.h>
41 /* The interface for checksum offload between the stack and networking drivers
44 * A. IP checksum related features
46 * Drivers advertise checksum offload capabilities in the features of a device.
47 * From the stack's point of view these are capabilities offered by the driver,
48 * a driver typically only advertises features that it is capable of offloading
51 * The checksum related features are:
53 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
54 * IP (one's complement) checksum for any combination
55 * of protocols or protocol layering. The checksum is
56 * computed and set in a packet per the CHECKSUM_PARTIAL
57 * interface (see below).
59 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
60 * TCP or UDP packets over IPv4. These are specifically
61 * unencapsulated packets of the form IPv4|TCP or
62 * IPv4|UDP where the Protocol field in the IPv4 header
63 * is TCP or UDP. The IPv4 header may contain IP options
64 * This feature cannot be set in features for a device
65 * with NETIF_F_HW_CSUM also set. This feature is being
66 * DEPRECATED (see below).
68 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
69 * TCP or UDP packets over IPv6. These are specifically
70 * unencapsulated packets of the form IPv6|TCP or
71 * IPv4|UDP where the Next Header field in the IPv6
72 * header is either TCP or UDP. IPv6 extension headers
73 * are not supported with this feature. This feature
74 * cannot be set in features for a device with
75 * NETIF_F_HW_CSUM also set. This feature is being
76 * DEPRECATED (see below).
78 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
79 * This flag is used only used to disable the RX checksum
80 * feature for a device. The stack will accept receive
81 * checksum indication in packets received on a device
82 * regardless of whether NETIF_F_RXCSUM is set.
84 * B. Checksumming of received packets by device. Indication of checksum
85 * verification is in set skb->ip_summed. Possible values are:
89 * Device did not checksum this packet e.g. due to lack of capabilities.
90 * The packet contains full (though not verified) checksum in packet but
91 * not in skb->csum. Thus, skb->csum is undefined in this case.
93 * CHECKSUM_UNNECESSARY:
95 * The hardware you're dealing with doesn't calculate the full checksum
96 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
97 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
98 * if their checksums are okay. skb->csum is still undefined in this case
99 * though. A driver or device must never modify the checksum field in the
100 * packet even if checksum is verified.
102 * CHECKSUM_UNNECESSARY is applicable to following protocols:
103 * TCP: IPv6 and IPv4.
104 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
105 * zero UDP checksum for either IPv4 or IPv6, the networking stack
106 * may perform further validation in this case.
107 * GRE: only if the checksum is present in the header.
108 * SCTP: indicates the CRC in SCTP header has been validated.
109 * FCOE: indicates the CRC in FC frame has been validated.
111 * skb->csum_level indicates the number of consecutive checksums found in
112 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
113 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
114 * and a device is able to verify the checksums for UDP (possibly zero),
115 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
116 * two. If the device were only able to verify the UDP checksum and not
117 * GRE, either because it doesn't support GRE checksum of because GRE
118 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
119 * not considered in this case).
123 * This is the most generic way. The device supplied checksum of the _whole_
124 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
125 * hardware doesn't need to parse L3/L4 headers to implement this.
128 * - Even if device supports only some protocols, but is able to produce
129 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
130 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
134 * A checksum is set up to be offloaded to a device as described in the
135 * output description for CHECKSUM_PARTIAL. This may occur on a packet
136 * received directly from another Linux OS, e.g., a virtualized Linux kernel
137 * on the same host, or it may be set in the input path in GRO or remote
138 * checksum offload. For the purposes of checksum verification, the checksum
139 * referred to by skb->csum_start + skb->csum_offset and any preceding
140 * checksums in the packet are considered verified. Any checksums in the
141 * packet that are after the checksum being offloaded are not considered to
144 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
145 * in the skb->ip_summed for a packet. Values are:
149 * The driver is required to checksum the packet as seen by hard_start_xmit()
150 * from skb->csum_start up to the end, and to record/write the checksum at
151 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
152 * csum_start and csum_offset values are valid values given the length and
153 * offset of the packet, however they should not attempt to validate that the
154 * checksum refers to a legitimate transport layer checksum-- it is the
155 * purview of the stack to validate that csum_start and csum_offset are set
158 * When the stack requests checksum offload for a packet, the driver MUST
159 * ensure that the checksum is set correctly. A driver can either offload the
160 * checksum calculation to the device, or call skb_checksum_help (in the case
161 * that the device does not support offload for a particular checksum).
163 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
164 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
165 * checksum offload capability.
166 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
167 * on network device checksumming capabilities: if a packet does not match
168 * them, skb_checksum_help or skb_crc32c_help (depending on the value of
169 * csum_not_inet, see item D.) is called to resolve the checksum.
173 * The skb was already checksummed by the protocol, or a checksum is not
176 * CHECKSUM_UNNECESSARY:
178 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
182 * Not used in checksum output. If a driver observes a packet with this value
183 * set in skbuff, if should treat as CHECKSUM_NONE being set.
185 * D. Non-IP checksum (CRC) offloads
187 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
188 * offloading the SCTP CRC in a packet. To perform this offload the stack
189 * will set set csum_start and csum_offset accordingly, set ip_summed to
190 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
191 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
192 * A driver that supports both IP checksum offload and SCTP CRC32c offload
193 * must verify which offload is configured for a packet by testing the
194 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
195 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
197 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
198 * offloading the FCOE CRC in a packet. To perform this offload the stack
199 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
200 * accordingly. Note the there is no indication in the skbuff that the
201 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
202 * both IP checksum offload and FCOE CRC offload must verify which offload
203 * is configured for a packet presumably by inspecting packet headers.
205 * E. Checksumming on output with GSO.
207 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
208 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
209 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
210 * part of the GSO operation is implied. If a checksum is being offloaded
211 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
212 * are set to refer to the outermost checksum being offload (two offloaded
213 * checksums are possible with UDP encapsulation).
216 /* Don't change this without changing skb_csum_unnecessary! */
217 #define CHECKSUM_NONE 0
218 #define CHECKSUM_UNNECESSARY 1
219 #define CHECKSUM_COMPLETE 2
220 #define CHECKSUM_PARTIAL 3
222 /* Maximum value in skb->csum_level */
223 #define SKB_MAX_CSUM_LEVEL 3
225 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
226 #define SKB_WITH_OVERHEAD(X) \
227 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
228 #define SKB_MAX_ORDER(X, ORDER) \
229 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
230 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
231 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
233 /* return minimum truesize of one skb containing X bytes of data */
234 #define SKB_TRUESIZE(X) ((X) + \
235 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
236 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
240 struct pipe_inode_info;
247 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
248 struct nf_conntrack {
253 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
254 struct nf_bridge_info {
256 BRNF_PROTO_UNCHANGED,
264 struct net_device *physindev;
266 /* always valid & non-NULL from FORWARD on, for physdev match */
267 struct net_device *physoutdev;
269 /* prerouting: detect dnat in orig/reply direction */
271 struct in6_addr ipv6_daddr;
273 /* after prerouting + nat detected: store original source
274 * mac since neigh resolution overwrites it, only used while
275 * skb is out in neigh layer.
277 char neigh_header[8];
282 struct sk_buff_head {
283 /* These two members must be first. */
284 struct sk_buff *next;
285 struct sk_buff *prev;
293 /* To allow 64K frame to be packed as single skb without frag_list we
294 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
295 * buffers which do not start on a page boundary.
297 * Since GRO uses frags we allocate at least 16 regardless of page
300 #if (65536/PAGE_SIZE + 1) < 16
301 #define MAX_SKB_FRAGS 16UL
303 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
305 extern int sysctl_max_skb_frags;
307 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
308 * segment using its current segmentation instead.
310 #define GSO_BY_FRAGS 0xFFFF
312 typedef struct bio_vec skb_frag_t;
315 * skb_frag_size() - Returns the size of a skb fragment
316 * @frag: skb fragment
318 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
324 * skb_frag_size_set() - Sets the size of a skb fragment
325 * @frag: skb fragment
326 * @size: size of fragment
328 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
334 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
335 * @frag: skb fragment
336 * @delta: value to add
338 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
340 frag->bv_len += delta;
344 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
345 * @frag: skb fragment
346 * @delta: value to subtract
348 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
350 frag->bv_len -= delta;
354 * skb_frag_must_loop - Test if %p is a high memory page
355 * @p: fragment's page
357 static inline bool skb_frag_must_loop(struct page *p)
359 #if defined(CONFIG_HIGHMEM)
367 * skb_frag_foreach_page - loop over pages in a fragment
369 * @f: skb frag to operate on
370 * @f_off: offset from start of f->bv_page
371 * @f_len: length from f_off to loop over
372 * @p: (temp var) current page
373 * @p_off: (temp var) offset from start of current page,
374 * non-zero only on first page.
375 * @p_len: (temp var) length in current page,
376 * < PAGE_SIZE only on first and last page.
377 * @copied: (temp var) length so far, excluding current p_len.
379 * A fragment can hold a compound page, in which case per-page
380 * operations, notably kmap_atomic, must be called for each
383 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
384 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
385 p_off = (f_off) & (PAGE_SIZE - 1), \
386 p_len = skb_frag_must_loop(p) ? \
387 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
390 copied += p_len, p++, p_off = 0, \
391 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
393 #define HAVE_HW_TIME_STAMP
396 * struct skb_shared_hwtstamps - hardware time stamps
397 * @hwtstamp: hardware time stamp transformed into duration
398 * since arbitrary point in time
400 * Software time stamps generated by ktime_get_real() are stored in
403 * hwtstamps can only be compared against other hwtstamps from
406 * This structure is attached to packets as part of the
407 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
409 struct skb_shared_hwtstamps {
413 /* Definitions for tx_flags in struct skb_shared_info */
415 /* generate hardware time stamp */
416 SKBTX_HW_TSTAMP = 1 << 0,
418 /* generate software time stamp when queueing packet to NIC */
419 SKBTX_SW_TSTAMP = 1 << 1,
421 /* device driver is going to provide hardware time stamp */
422 SKBTX_IN_PROGRESS = 1 << 2,
424 /* device driver supports TX zero-copy buffers */
425 SKBTX_DEV_ZEROCOPY = 1 << 3,
427 /* generate wifi status information (where possible) */
428 SKBTX_WIFI_STATUS = 1 << 4,
430 /* This indicates at least one fragment might be overwritten
431 * (as in vmsplice(), sendfile() ...)
432 * If we need to compute a TX checksum, we'll need to copy
433 * all frags to avoid possible bad checksum
435 SKBTX_SHARED_FRAG = 1 << 5,
437 /* generate software time stamp when entering packet scheduling */
438 SKBTX_SCHED_TSTAMP = 1 << 6,
441 #define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
442 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
444 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
447 * The callback notifies userspace to release buffers when skb DMA is done in
448 * lower device, the skb last reference should be 0 when calling this.
449 * The zerocopy_success argument is true if zero copy transmit occurred,
450 * false on data copy or out of memory error caused by data copy attempt.
451 * The ctx field is used to track device context.
452 * The desc field is used to track userspace buffer index.
455 void (*callback)(struct ubuf_info *, bool zerocopy_success);
471 struct user_struct *user;
476 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
478 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
479 void mm_unaccount_pinned_pages(struct mmpin *mmp);
481 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
482 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
483 struct ubuf_info *uarg);
485 static inline void sock_zerocopy_get(struct ubuf_info *uarg)
487 refcount_inc(&uarg->refcnt);
490 void sock_zerocopy_put(struct ubuf_info *uarg);
491 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
493 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
495 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len);
496 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
497 struct msghdr *msg, int len,
498 struct ubuf_info *uarg);
500 /* This data is invariant across clones and lives at
501 * the end of the header data, ie. at skb->end.
503 struct skb_shared_info {
508 unsigned short gso_size;
509 /* Warning: this field is not always filled in (UFO)! */
510 unsigned short gso_segs;
511 struct sk_buff *frag_list;
512 struct skb_shared_hwtstamps hwtstamps;
513 unsigned int gso_type;
517 * Warning : all fields before dataref are cleared in __alloc_skb()
521 /* Intermediate layers must ensure that destructor_arg
522 * remains valid until skb destructor */
523 void * destructor_arg;
525 /* must be last field, see pskb_expand_head() */
526 skb_frag_t frags[MAX_SKB_FRAGS];
529 /* We divide dataref into two halves. The higher 16 bits hold references
530 * to the payload part of skb->data. The lower 16 bits hold references to
531 * the entire skb->data. A clone of a headerless skb holds the length of
532 * the header in skb->hdr_len.
534 * All users must obey the rule that the skb->data reference count must be
535 * greater than or equal to the payload reference count.
537 * Holding a reference to the payload part means that the user does not
538 * care about modifications to the header part of skb->data.
540 #define SKB_DATAREF_SHIFT 16
541 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
545 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
546 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
547 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
551 SKB_GSO_TCPV4 = 1 << 0,
553 /* This indicates the skb is from an untrusted source. */
554 SKB_GSO_DODGY = 1 << 1,
556 /* This indicates the tcp segment has CWR set. */
557 SKB_GSO_TCP_ECN = 1 << 2,
559 SKB_GSO_TCP_FIXEDID = 1 << 3,
561 SKB_GSO_TCPV6 = 1 << 4,
563 SKB_GSO_FCOE = 1 << 5,
565 SKB_GSO_GRE = 1 << 6,
567 SKB_GSO_GRE_CSUM = 1 << 7,
569 SKB_GSO_IPXIP4 = 1 << 8,
571 SKB_GSO_IPXIP6 = 1 << 9,
573 SKB_GSO_UDP_TUNNEL = 1 << 10,
575 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
577 SKB_GSO_PARTIAL = 1 << 12,
579 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
581 SKB_GSO_SCTP = 1 << 14,
583 SKB_GSO_ESP = 1 << 15,
585 SKB_GSO_UDP = 1 << 16,
587 SKB_GSO_UDP_L4 = 1 << 17,
590 #if BITS_PER_LONG > 32
591 #define NET_SKBUFF_DATA_USES_OFFSET 1
594 #ifdef NET_SKBUFF_DATA_USES_OFFSET
595 typedef unsigned int sk_buff_data_t;
597 typedef unsigned char *sk_buff_data_t;
601 * struct sk_buff - socket buffer
602 * @next: Next buffer in list
603 * @prev: Previous buffer in list
604 * @tstamp: Time we arrived/left
605 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
606 * @sk: Socket we are owned by
607 * @dev: Device we arrived on/are leaving by
608 * @cb: Control buffer. Free for use by every layer. Put private vars here
609 * @_skb_refdst: destination entry (with norefcount bit)
610 * @sp: the security path, used for xfrm
611 * @len: Length of actual data
612 * @data_len: Data length
613 * @mac_len: Length of link layer header
614 * @hdr_len: writable header length of cloned skb
615 * @csum: Checksum (must include start/offset pair)
616 * @csum_start: Offset from skb->head where checksumming should start
617 * @csum_offset: Offset from csum_start where checksum should be stored
618 * @priority: Packet queueing priority
619 * @ignore_df: allow local fragmentation
620 * @cloned: Head may be cloned (check refcnt to be sure)
621 * @ip_summed: Driver fed us an IP checksum
622 * @nohdr: Payload reference only, must not modify header
623 * @pkt_type: Packet class
624 * @fclone: skbuff clone status
625 * @ipvs_property: skbuff is owned by ipvs
626 * @offload_fwd_mark: Packet was L2-forwarded in hardware
627 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
628 * @tc_skip_classify: do not classify packet. set by IFB device
629 * @tc_at_ingress: used within tc_classify to distinguish in/egress
630 * @tc_redirected: packet was redirected by a tc action
631 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
632 * @peeked: this packet has been seen already, so stats have been
633 * done for it, don't do them again
634 * @nf_trace: netfilter packet trace flag
635 * @protocol: Packet protocol from driver
636 * @destructor: Destruct function
637 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
638 * @_nfct: Associated connection, if any (with nfctinfo bits)
639 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
640 * @skb_iif: ifindex of device we arrived on
641 * @tc_index: Traffic control index
642 * @hash: the packet hash
643 * @queue_mapping: Queue mapping for multiqueue devices
644 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
645 * @active_extensions: active extensions (skb_ext_id types)
646 * @ndisc_nodetype: router type (from link layer)
647 * @ooo_okay: allow the mapping of a socket to a queue to be changed
648 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
650 * @sw_hash: indicates hash was computed in software stack
651 * @wifi_acked_valid: wifi_acked was set
652 * @wifi_acked: whether frame was acked on wifi or not
653 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
654 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
655 * @dst_pending_confirm: need to confirm neighbour
656 * @decrypted: Decrypted SKB
657 * @napi_id: id of the NAPI struct this skb came from
658 * @secmark: security marking
659 * @mark: Generic packet mark
660 * @vlan_proto: vlan encapsulation protocol
661 * @vlan_tci: vlan tag control information
662 * @inner_protocol: Protocol (encapsulation)
663 * @inner_transport_header: Inner transport layer header (encapsulation)
664 * @inner_network_header: Network layer header (encapsulation)
665 * @inner_mac_header: Link layer header (encapsulation)
666 * @transport_header: Transport layer header
667 * @network_header: Network layer header
668 * @mac_header: Link layer header
669 * @tail: Tail pointer
671 * @head: Head of buffer
672 * @data: Data head pointer
673 * @truesize: Buffer size
674 * @users: User count - see {datagram,tcp}.c
675 * @extensions: allocated extensions, valid if active_extensions is nonzero
681 /* These two members must be first. */
682 struct sk_buff *next;
683 struct sk_buff *prev;
686 struct net_device *dev;
687 /* Some protocols might use this space to store information,
688 * while device pointer would be NULL.
689 * UDP receive path is one user.
691 unsigned long dev_scratch;
694 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
695 struct list_head list;
700 int ip_defrag_offset;
705 u64 skb_mstamp_ns; /* earliest departure time */
708 * This is the control buffer. It is free to use for every
709 * layer. Please put your private variables there. If you
710 * want to keep them across layers you have to do a skb_clone()
711 * first. This is owned by whoever has the skb queued ATM.
713 char cb[48] __aligned(8);
717 unsigned long _skb_refdst;
718 void (*destructor)(struct sk_buff *skb);
720 struct list_head tcp_tsorted_anchor;
723 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
731 /* Following fields are _not_ copied in __copy_skb_header()
732 * Note that queue_mapping is here mostly to fill a hole.
736 /* if you move cloned around you also must adapt those constants */
737 #ifdef __BIG_ENDIAN_BITFIELD
738 #define CLONED_MASK (1 << 7)
740 #define CLONED_MASK 1
742 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
744 __u8 __cloned_offset[0];
751 #ifdef CONFIG_SKB_EXTENSIONS
752 __u8 active_extensions;
754 /* fields enclosed in headers_start/headers_end are copied
755 * using a single memcpy() in __copy_skb_header()
758 __u32 headers_start[0];
761 /* if you move pkt_type around you also must adapt those constants */
762 #ifdef __BIG_ENDIAN_BITFIELD
763 #define PKT_TYPE_MAX (7 << 5)
765 #define PKT_TYPE_MAX 7
767 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
769 __u8 __pkt_type_offset[0];
778 __u8 wifi_acked_valid:1;
781 /* Indicates the inner headers are valid in the skbuff. */
782 __u8 encapsulation:1;
783 __u8 encap_hdr_csum:1;
786 #ifdef __BIG_ENDIAN_BITFIELD
787 #define PKT_VLAN_PRESENT_BIT 7
789 #define PKT_VLAN_PRESENT_BIT 0
791 #define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset)
792 __u8 __pkt_vlan_present_offset[0];
794 __u8 csum_complete_sw:1;
796 __u8 csum_not_inet:1;
797 __u8 dst_pending_confirm:1;
798 #ifdef CONFIG_IPV6_NDISC_NODETYPE
799 __u8 ndisc_nodetype:2;
802 __u8 ipvs_property:1;
803 __u8 inner_protocol_type:1;
804 __u8 remcsum_offload:1;
805 #ifdef CONFIG_NET_SWITCHDEV
806 __u8 offload_fwd_mark:1;
807 __u8 offload_l3_fwd_mark:1;
809 #ifdef CONFIG_NET_CLS_ACT
810 __u8 tc_skip_classify:1;
811 __u8 tc_at_ingress:1;
812 __u8 tc_redirected:1;
813 __u8 tc_from_ingress:1;
815 #ifdef CONFIG_TLS_DEVICE
819 #ifdef CONFIG_NET_SCHED
820 __u16 tc_index; /* traffic control index */
835 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
837 unsigned int napi_id;
838 unsigned int sender_cpu;
841 #ifdef CONFIG_NETWORK_SECMARK
847 __u32 reserved_tailroom;
851 __be16 inner_protocol;
855 __u16 inner_transport_header;
856 __u16 inner_network_header;
857 __u16 inner_mac_header;
860 __u16 transport_header;
861 __u16 network_header;
865 __u32 headers_end[0];
868 /* These elements must be at the end, see alloc_skb() for details. */
873 unsigned int truesize;
876 #ifdef CONFIG_SKB_EXTENSIONS
877 /* only useable after checking ->active_extensions != 0 */
878 struct skb_ext *extensions;
884 * Handling routines are only of interest to the kernel
887 #define SKB_ALLOC_FCLONE 0x01
888 #define SKB_ALLOC_RX 0x02
889 #define SKB_ALLOC_NAPI 0x04
892 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
895 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
897 return unlikely(skb->pfmemalloc);
901 * skb might have a dst pointer attached, refcounted or not.
902 * _skb_refdst low order bit is set if refcount was _not_ taken
904 #define SKB_DST_NOREF 1UL
905 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
907 #define SKB_NFCT_PTRMASK ~(7UL)
909 * skb_dst - returns skb dst_entry
912 * Returns skb dst_entry, regardless of reference taken or not.
914 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
916 /* If refdst was not refcounted, check we still are in a
917 * rcu_read_lock section
919 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
920 !rcu_read_lock_held() &&
921 !rcu_read_lock_bh_held());
922 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
926 * skb_dst_set - sets skb dst
930 * Sets skb dst, assuming a reference was taken on dst and should
931 * be released by skb_dst_drop()
933 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
935 skb->_skb_refdst = (unsigned long)dst;
939 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
943 * Sets skb dst, assuming a reference was not taken on dst.
944 * If dst entry is cached, we do not take reference and dst_release
945 * will be avoided by refdst_drop. If dst entry is not cached, we take
946 * reference, so that last dst_release can destroy the dst immediately.
948 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
950 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
951 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
955 * skb_dst_is_noref - Test if skb dst isn't refcounted
958 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
960 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
964 * skb_rtable - Returns the skb &rtable
967 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
969 return (struct rtable *)skb_dst(skb);
972 /* For mangling skb->pkt_type from user space side from applications
973 * such as nft, tc, etc, we only allow a conservative subset of
974 * possible pkt_types to be set.
976 static inline bool skb_pkt_type_ok(u32 ptype)
978 return ptype <= PACKET_OTHERHOST;
982 * skb_napi_id - Returns the skb's NAPI id
985 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
987 #ifdef CONFIG_NET_RX_BUSY_POLL
995 * skb_unref - decrement the skb's reference count
998 * Returns true if we can free the skb.
1000 static inline bool skb_unref(struct sk_buff *skb)
1004 if (likely(refcount_read(&skb->users) == 1))
1006 else if (likely(!refcount_dec_and_test(&skb->users)))
1012 void skb_release_head_state(struct sk_buff *skb);
1013 void kfree_skb(struct sk_buff *skb);
1014 void kfree_skb_list(struct sk_buff *segs);
1015 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1016 void skb_tx_error(struct sk_buff *skb);
1017 void consume_skb(struct sk_buff *skb);
1018 void __consume_stateless_skb(struct sk_buff *skb);
1019 void __kfree_skb(struct sk_buff *skb);
1020 extern struct kmem_cache *skbuff_head_cache;
1022 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1023 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1024 bool *fragstolen, int *delta_truesize);
1026 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1028 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1029 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1030 struct sk_buff *build_skb_around(struct sk_buff *skb,
1031 void *data, unsigned int frag_size);
1034 * alloc_skb - allocate a network buffer
1035 * @size: size to allocate
1036 * @priority: allocation mask
1038 * This function is a convenient wrapper around __alloc_skb().
1040 static inline struct sk_buff *alloc_skb(unsigned int size,
1043 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1046 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1047 unsigned long data_len,
1051 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1053 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1054 struct sk_buff_fclones {
1055 struct sk_buff skb1;
1057 struct sk_buff skb2;
1059 refcount_t fclone_ref;
1063 * skb_fclone_busy - check if fclone is busy
1067 * Returns true if skb is a fast clone, and its clone is not freed.
1068 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1069 * so we also check that this didnt happen.
1071 static inline bool skb_fclone_busy(const struct sock *sk,
1072 const struct sk_buff *skb)
1074 const struct sk_buff_fclones *fclones;
1076 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1078 return skb->fclone == SKB_FCLONE_ORIG &&
1079 refcount_read(&fclones->fclone_ref) > 1 &&
1080 fclones->skb2.sk == sk;
1084 * alloc_skb_fclone - allocate a network buffer from fclone cache
1085 * @size: size to allocate
1086 * @priority: allocation mask
1088 * This function is a convenient wrapper around __alloc_skb().
1090 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1093 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1096 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1097 void skb_headers_offset_update(struct sk_buff *skb, int off);
1098 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1099 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1100 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1101 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1102 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1103 gfp_t gfp_mask, bool fclone);
1104 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1107 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1110 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1111 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1112 unsigned int headroom);
1113 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1114 int newtailroom, gfp_t priority);
1115 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1116 int offset, int len);
1117 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1118 int offset, int len);
1119 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1120 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1123 * skb_pad - zero pad the tail of an skb
1124 * @skb: buffer to pad
1125 * @pad: space to pad
1127 * Ensure that a buffer is followed by a padding area that is zero
1128 * filled. Used by network drivers which may DMA or transfer data
1129 * beyond the buffer end onto the wire.
1131 * May return error in out of memory cases. The skb is freed on error.
1133 static inline int skb_pad(struct sk_buff *skb, int pad)
1135 return __skb_pad(skb, pad, true);
1137 #define dev_kfree_skb(a) consume_skb(a)
1139 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1140 int offset, size_t size);
1142 struct skb_seq_state {
1146 __u32 stepped_offset;
1147 struct sk_buff *root_skb;
1148 struct sk_buff *cur_skb;
1152 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1153 unsigned int to, struct skb_seq_state *st);
1154 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1155 struct skb_seq_state *st);
1156 void skb_abort_seq_read(struct skb_seq_state *st);
1158 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1159 unsigned int to, struct ts_config *config);
1162 * Packet hash types specify the type of hash in skb_set_hash.
1164 * Hash types refer to the protocol layer addresses which are used to
1165 * construct a packet's hash. The hashes are used to differentiate or identify
1166 * flows of the protocol layer for the hash type. Hash types are either
1167 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1169 * Properties of hashes:
1171 * 1) Two packets in different flows have different hash values
1172 * 2) Two packets in the same flow should have the same hash value
1174 * A hash at a higher layer is considered to be more specific. A driver should
1175 * set the most specific hash possible.
1177 * A driver cannot indicate a more specific hash than the layer at which a hash
1178 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1180 * A driver may indicate a hash level which is less specific than the
1181 * actual layer the hash was computed on. For instance, a hash computed
1182 * at L4 may be considered an L3 hash. This should only be done if the
1183 * driver can't unambiguously determine that the HW computed the hash at
1184 * the higher layer. Note that the "should" in the second property above
1187 enum pkt_hash_types {
1188 PKT_HASH_TYPE_NONE, /* Undefined type */
1189 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1190 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1191 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1194 static inline void skb_clear_hash(struct sk_buff *skb)
1201 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1204 skb_clear_hash(skb);
1208 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1210 skb->l4_hash = is_l4;
1211 skb->sw_hash = is_sw;
1216 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1218 /* Used by drivers to set hash from HW */
1219 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1223 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1225 __skb_set_hash(skb, hash, true, is_l4);
1228 void __skb_get_hash(struct sk_buff *skb);
1229 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1230 u32 skb_get_poff(const struct sk_buff *skb);
1231 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1232 const struct flow_keys_basic *keys, int hlen);
1233 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1234 void *data, int hlen_proto);
1236 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1237 int thoff, u8 ip_proto)
1239 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1242 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1243 const struct flow_dissector_key *key,
1244 unsigned int key_count);
1247 int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1248 union bpf_attr __user *uattr);
1249 int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1250 struct bpf_prog *prog);
1252 int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr);
1254 static inline int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1255 union bpf_attr __user *uattr)
1260 static inline int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1261 struct bpf_prog *prog)
1266 static inline int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr)
1272 struct bpf_flow_dissector;
1273 bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1274 __be16 proto, int nhoff, int hlen, unsigned int flags);
1276 bool __skb_flow_dissect(const struct net *net,
1277 const struct sk_buff *skb,
1278 struct flow_dissector *flow_dissector,
1279 void *target_container,
1280 void *data, __be16 proto, int nhoff, int hlen,
1281 unsigned int flags);
1283 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1284 struct flow_dissector *flow_dissector,
1285 void *target_container, unsigned int flags)
1287 return __skb_flow_dissect(NULL, skb, flow_dissector,
1288 target_container, NULL, 0, 0, 0, flags);
1291 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1292 struct flow_keys *flow,
1295 memset(flow, 0, sizeof(*flow));
1296 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1297 flow, NULL, 0, 0, 0, flags);
1301 skb_flow_dissect_flow_keys_basic(const struct net *net,
1302 const struct sk_buff *skb,
1303 struct flow_keys_basic *flow, void *data,
1304 __be16 proto, int nhoff, int hlen,
1307 memset(flow, 0, sizeof(*flow));
1308 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1309 data, proto, nhoff, hlen, flags);
1312 void skb_flow_dissect_meta(const struct sk_buff *skb,
1313 struct flow_dissector *flow_dissector,
1314 void *target_container);
1316 /* Gets a skb connection tracking info, ctinfo map should be a
1317 * a map of mapsize to translate enum ip_conntrack_info states
1321 skb_flow_dissect_ct(const struct sk_buff *skb,
1322 struct flow_dissector *flow_dissector,
1323 void *target_container,
1327 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1328 struct flow_dissector *flow_dissector,
1329 void *target_container);
1331 static inline __u32 skb_get_hash(struct sk_buff *skb)
1333 if (!skb->l4_hash && !skb->sw_hash)
1334 __skb_get_hash(skb);
1339 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1341 if (!skb->l4_hash && !skb->sw_hash) {
1342 struct flow_keys keys;
1343 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1345 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1351 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1353 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1358 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1360 to->hash = from->hash;
1361 to->sw_hash = from->sw_hash;
1362 to->l4_hash = from->l4_hash;
1365 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1366 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1368 return skb->head + skb->end;
1371 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1376 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1381 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1383 return skb->end - skb->head;
1388 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1390 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1392 return &skb_shinfo(skb)->hwtstamps;
1395 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1397 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1399 return is_zcopy ? skb_uarg(skb) : NULL;
1402 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1405 if (skb && uarg && !skb_zcopy(skb)) {
1406 if (unlikely(have_ref && *have_ref))
1409 sock_zerocopy_get(uarg);
1410 skb_shinfo(skb)->destructor_arg = uarg;
1411 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1415 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1417 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1418 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1421 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1423 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1426 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1428 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1431 /* Release a reference on a zerocopy structure */
1432 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1434 struct ubuf_info *uarg = skb_zcopy(skb);
1437 if (skb_zcopy_is_nouarg(skb)) {
1438 /* no notification callback */
1439 } else if (uarg->callback == sock_zerocopy_callback) {
1440 uarg->zerocopy = uarg->zerocopy && zerocopy;
1441 sock_zerocopy_put(uarg);
1443 uarg->callback(uarg, zerocopy);
1446 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1450 /* Abort a zerocopy operation and revert zckey on error in send syscall */
1451 static inline void skb_zcopy_abort(struct sk_buff *skb)
1453 struct ubuf_info *uarg = skb_zcopy(skb);
1456 sock_zerocopy_put_abort(uarg, false);
1457 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1461 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1466 static inline void skb_list_del_init(struct sk_buff *skb)
1468 __list_del_entry(&skb->list);
1469 skb_mark_not_on_list(skb);
1473 * skb_queue_empty - check if a queue is empty
1476 * Returns true if the queue is empty, false otherwise.
1478 static inline int skb_queue_empty(const struct sk_buff_head *list)
1480 return list->next == (const struct sk_buff *) list;
1484 * skb_queue_is_last - check if skb is the last entry in the queue
1488 * Returns true if @skb is the last buffer on the list.
1490 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1491 const struct sk_buff *skb)
1493 return skb->next == (const struct sk_buff *) list;
1497 * skb_queue_is_first - check if skb is the first entry in the queue
1501 * Returns true if @skb is the first buffer on the list.
1503 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1504 const struct sk_buff *skb)
1506 return skb->prev == (const struct sk_buff *) list;
1510 * skb_queue_next - return the next packet in the queue
1512 * @skb: current buffer
1514 * Return the next packet in @list after @skb. It is only valid to
1515 * call this if skb_queue_is_last() evaluates to false.
1517 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1518 const struct sk_buff *skb)
1520 /* This BUG_ON may seem severe, but if we just return then we
1521 * are going to dereference garbage.
1523 BUG_ON(skb_queue_is_last(list, skb));
1528 * skb_queue_prev - return the prev packet in the queue
1530 * @skb: current buffer
1532 * Return the prev packet in @list before @skb. It is only valid to
1533 * call this if skb_queue_is_first() evaluates to false.
1535 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1536 const struct sk_buff *skb)
1538 /* This BUG_ON may seem severe, but if we just return then we
1539 * are going to dereference garbage.
1541 BUG_ON(skb_queue_is_first(list, skb));
1546 * skb_get - reference buffer
1547 * @skb: buffer to reference
1549 * Makes another reference to a socket buffer and returns a pointer
1552 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1554 refcount_inc(&skb->users);
1559 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1563 * skb_cloned - is the buffer a clone
1564 * @skb: buffer to check
1566 * Returns true if the buffer was generated with skb_clone() and is
1567 * one of multiple shared copies of the buffer. Cloned buffers are
1568 * shared data so must not be written to under normal circumstances.
1570 static inline int skb_cloned(const struct sk_buff *skb)
1572 return skb->cloned &&
1573 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1576 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1578 might_sleep_if(gfpflags_allow_blocking(pri));
1580 if (skb_cloned(skb))
1581 return pskb_expand_head(skb, 0, 0, pri);
1587 * skb_header_cloned - is the header a clone
1588 * @skb: buffer to check
1590 * Returns true if modifying the header part of the buffer requires
1591 * the data to be copied.
1593 static inline int skb_header_cloned(const struct sk_buff *skb)
1600 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1601 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1602 return dataref != 1;
1605 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1607 might_sleep_if(gfpflags_allow_blocking(pri));
1609 if (skb_header_cloned(skb))
1610 return pskb_expand_head(skb, 0, 0, pri);
1616 * __skb_header_release - release reference to header
1617 * @skb: buffer to operate on
1619 static inline void __skb_header_release(struct sk_buff *skb)
1622 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1627 * skb_shared - is the buffer shared
1628 * @skb: buffer to check
1630 * Returns true if more than one person has a reference to this
1633 static inline int skb_shared(const struct sk_buff *skb)
1635 return refcount_read(&skb->users) != 1;
1639 * skb_share_check - check if buffer is shared and if so clone it
1640 * @skb: buffer to check
1641 * @pri: priority for memory allocation
1643 * If the buffer is shared the buffer is cloned and the old copy
1644 * drops a reference. A new clone with a single reference is returned.
1645 * If the buffer is not shared the original buffer is returned. When
1646 * being called from interrupt status or with spinlocks held pri must
1649 * NULL is returned on a memory allocation failure.
1651 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1653 might_sleep_if(gfpflags_allow_blocking(pri));
1654 if (skb_shared(skb)) {
1655 struct sk_buff *nskb = skb_clone(skb, pri);
1667 * Copy shared buffers into a new sk_buff. We effectively do COW on
1668 * packets to handle cases where we have a local reader and forward
1669 * and a couple of other messy ones. The normal one is tcpdumping
1670 * a packet thats being forwarded.
1674 * skb_unshare - make a copy of a shared buffer
1675 * @skb: buffer to check
1676 * @pri: priority for memory allocation
1678 * If the socket buffer is a clone then this function creates a new
1679 * copy of the data, drops a reference count on the old copy and returns
1680 * the new copy with the reference count at 1. If the buffer is not a clone
1681 * the original buffer is returned. When called with a spinlock held or
1682 * from interrupt state @pri must be %GFP_ATOMIC
1684 * %NULL is returned on a memory allocation failure.
1686 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1689 might_sleep_if(gfpflags_allow_blocking(pri));
1690 if (skb_cloned(skb)) {
1691 struct sk_buff *nskb = skb_copy(skb, pri);
1693 /* Free our shared copy */
1704 * skb_peek - peek at the head of an &sk_buff_head
1705 * @list_: list to peek at
1707 * Peek an &sk_buff. Unlike most other operations you _MUST_
1708 * be careful with this one. A peek leaves the buffer on the
1709 * list and someone else may run off with it. You must hold
1710 * the appropriate locks or have a private queue to do this.
1712 * Returns %NULL for an empty list or a pointer to the head element.
1713 * The reference count is not incremented and the reference is therefore
1714 * volatile. Use with caution.
1716 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1718 struct sk_buff *skb = list_->next;
1720 if (skb == (struct sk_buff *)list_)
1726 * __skb_peek - peek at the head of a non-empty &sk_buff_head
1727 * @list_: list to peek at
1729 * Like skb_peek(), but the caller knows that the list is not empty.
1731 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
1737 * skb_peek_next - peek skb following the given one from a queue
1738 * @skb: skb to start from
1739 * @list_: list to peek at
1741 * Returns %NULL when the end of the list is met or a pointer to the
1742 * next element. The reference count is not incremented and the
1743 * reference is therefore volatile. Use with caution.
1745 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1746 const struct sk_buff_head *list_)
1748 struct sk_buff *next = skb->next;
1750 if (next == (struct sk_buff *)list_)
1756 * skb_peek_tail - peek at the tail of an &sk_buff_head
1757 * @list_: list to peek at
1759 * Peek an &sk_buff. Unlike most other operations you _MUST_
1760 * be careful with this one. A peek leaves the buffer on the
1761 * list and someone else may run off with it. You must hold
1762 * the appropriate locks or have a private queue to do this.
1764 * Returns %NULL for an empty list or a pointer to the tail element.
1765 * The reference count is not incremented and the reference is therefore
1766 * volatile. Use with caution.
1768 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1770 struct sk_buff *skb = list_->prev;
1772 if (skb == (struct sk_buff *)list_)
1779 * skb_queue_len - get queue length
1780 * @list_: list to measure
1782 * Return the length of an &sk_buff queue.
1784 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1790 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1791 * @list: queue to initialize
1793 * This initializes only the list and queue length aspects of
1794 * an sk_buff_head object. This allows to initialize the list
1795 * aspects of an sk_buff_head without reinitializing things like
1796 * the spinlock. It can also be used for on-stack sk_buff_head
1797 * objects where the spinlock is known to not be used.
1799 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1801 list->prev = list->next = (struct sk_buff *)list;
1806 * This function creates a split out lock class for each invocation;
1807 * this is needed for now since a whole lot of users of the skb-queue
1808 * infrastructure in drivers have different locking usage (in hardirq)
1809 * than the networking core (in softirq only). In the long run either the
1810 * network layer or drivers should need annotation to consolidate the
1811 * main types of usage into 3 classes.
1813 static inline void skb_queue_head_init(struct sk_buff_head *list)
1815 spin_lock_init(&list->lock);
1816 __skb_queue_head_init(list);
1819 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1820 struct lock_class_key *class)
1822 skb_queue_head_init(list);
1823 lockdep_set_class(&list->lock, class);
1827 * Insert an sk_buff on a list.
1829 * The "__skb_xxxx()" functions are the non-atomic ones that
1830 * can only be called with interrupts disabled.
1832 static inline void __skb_insert(struct sk_buff *newsk,
1833 struct sk_buff *prev, struct sk_buff *next,
1834 struct sk_buff_head *list)
1838 next->prev = prev->next = newsk;
1842 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1843 struct sk_buff *prev,
1844 struct sk_buff *next)
1846 struct sk_buff *first = list->next;
1847 struct sk_buff *last = list->prev;
1857 * skb_queue_splice - join two skb lists, this is designed for stacks
1858 * @list: the new list to add
1859 * @head: the place to add it in the first list
1861 static inline void skb_queue_splice(const struct sk_buff_head *list,
1862 struct sk_buff_head *head)
1864 if (!skb_queue_empty(list)) {
1865 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1866 head->qlen += list->qlen;
1871 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1872 * @list: the new list to add
1873 * @head: the place to add it in the first list
1875 * The list at @list is reinitialised
1877 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1878 struct sk_buff_head *head)
1880 if (!skb_queue_empty(list)) {
1881 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1882 head->qlen += list->qlen;
1883 __skb_queue_head_init(list);
1888 * skb_queue_splice_tail - join two skb lists, each list being a queue
1889 * @list: the new list to add
1890 * @head: the place to add it in the first list
1892 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1893 struct sk_buff_head *head)
1895 if (!skb_queue_empty(list)) {
1896 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1897 head->qlen += list->qlen;
1902 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1903 * @list: the new list to add
1904 * @head: the place to add it in the first list
1906 * Each of the lists is a queue.
1907 * The list at @list is reinitialised
1909 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1910 struct sk_buff_head *head)
1912 if (!skb_queue_empty(list)) {
1913 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1914 head->qlen += list->qlen;
1915 __skb_queue_head_init(list);
1920 * __skb_queue_after - queue a buffer at the list head
1921 * @list: list to use
1922 * @prev: place after this buffer
1923 * @newsk: buffer to queue
1925 * Queue a buffer int the middle of a list. This function takes no locks
1926 * and you must therefore hold required locks before calling it.
1928 * A buffer cannot be placed on two lists at the same time.
1930 static inline void __skb_queue_after(struct sk_buff_head *list,
1931 struct sk_buff *prev,
1932 struct sk_buff *newsk)
1934 __skb_insert(newsk, prev, prev->next, list);
1937 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1938 struct sk_buff_head *list);
1940 static inline void __skb_queue_before(struct sk_buff_head *list,
1941 struct sk_buff *next,
1942 struct sk_buff *newsk)
1944 __skb_insert(newsk, next->prev, next, list);
1948 * __skb_queue_head - queue a buffer at the list head
1949 * @list: list to use
1950 * @newsk: buffer to queue
1952 * Queue a buffer at the start of a list. This function takes no locks
1953 * and you must therefore hold required locks before calling it.
1955 * A buffer cannot be placed on two lists at the same time.
1957 static inline void __skb_queue_head(struct sk_buff_head *list,
1958 struct sk_buff *newsk)
1960 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1962 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1965 * __skb_queue_tail - queue a buffer at the list tail
1966 * @list: list to use
1967 * @newsk: buffer to queue
1969 * Queue a buffer at the end of a list. This function takes no locks
1970 * and you must therefore hold required locks before calling it.
1972 * A buffer cannot be placed on two lists at the same time.
1974 static inline void __skb_queue_tail(struct sk_buff_head *list,
1975 struct sk_buff *newsk)
1977 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1979 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1982 * remove sk_buff from list. _Must_ be called atomically, and with
1985 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1986 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1988 struct sk_buff *next, *prev;
1993 skb->next = skb->prev = NULL;
1999 * __skb_dequeue - remove from the head of the queue
2000 * @list: list to dequeue from
2002 * Remove the head of the list. This function does not take any locks
2003 * so must be used with appropriate locks held only. The head item is
2004 * returned or %NULL if the list is empty.
2006 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2008 struct sk_buff *skb = skb_peek(list);
2010 __skb_unlink(skb, list);
2013 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2016 * __skb_dequeue_tail - remove from the tail of the queue
2017 * @list: list to dequeue from
2019 * Remove the tail of the list. This function does not take any locks
2020 * so must be used with appropriate locks held only. The tail item is
2021 * returned or %NULL if the list is empty.
2023 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2025 struct sk_buff *skb = skb_peek_tail(list);
2027 __skb_unlink(skb, list);
2030 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2033 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2035 return skb->data_len;
2038 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2040 return skb->len - skb->data_len;
2043 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2045 unsigned int i, len = 0;
2047 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2048 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2052 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2054 return skb_headlen(skb) + __skb_pagelen(skb);
2058 * __skb_fill_page_desc - initialise a paged fragment in an skb
2059 * @skb: buffer containing fragment to be initialised
2060 * @i: paged fragment index to initialise
2061 * @page: the page to use for this fragment
2062 * @off: the offset to the data with @page
2063 * @size: the length of the data
2065 * Initialises the @i'th fragment of @skb to point to &size bytes at
2066 * offset @off within @page.
2068 * Does not take any additional reference on the fragment.
2070 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2071 struct page *page, int off, int size)
2073 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2076 * Propagate page pfmemalloc to the skb if we can. The problem is
2077 * that not all callers have unique ownership of the page but rely
2078 * on page_is_pfmemalloc doing the right thing(tm).
2080 frag->bv_page = page;
2081 frag->bv_offset = off;
2082 skb_frag_size_set(frag, size);
2084 page = compound_head(page);
2085 if (page_is_pfmemalloc(page))
2086 skb->pfmemalloc = true;
2090 * skb_fill_page_desc - initialise a paged fragment in an skb
2091 * @skb: buffer containing fragment to be initialised
2092 * @i: paged fragment index to initialise
2093 * @page: the page to use for this fragment
2094 * @off: the offset to the data with @page
2095 * @size: the length of the data
2097 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2098 * @skb to point to @size bytes at offset @off within @page. In
2099 * addition updates @skb such that @i is the last fragment.
2101 * Does not take any additional reference on the fragment.
2103 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2104 struct page *page, int off, int size)
2106 __skb_fill_page_desc(skb, i, page, off, size);
2107 skb_shinfo(skb)->nr_frags = i + 1;
2110 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2111 int size, unsigned int truesize);
2113 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2114 unsigned int truesize);
2116 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2118 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2119 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2121 return skb->head + skb->tail;
2124 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2126 skb->tail = skb->data - skb->head;
2129 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2131 skb_reset_tail_pointer(skb);
2132 skb->tail += offset;
2135 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2136 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2141 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2143 skb->tail = skb->data;
2146 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2148 skb->tail = skb->data + offset;
2151 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2154 * Add data to an sk_buff
2156 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2157 void *skb_put(struct sk_buff *skb, unsigned int len);
2158 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2160 void *tmp = skb_tail_pointer(skb);
2161 SKB_LINEAR_ASSERT(skb);
2167 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2169 void *tmp = __skb_put(skb, len);
2171 memset(tmp, 0, len);
2175 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2178 void *tmp = __skb_put(skb, len);
2180 memcpy(tmp, data, len);
2184 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2186 *(u8 *)__skb_put(skb, 1) = val;
2189 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2191 void *tmp = skb_put(skb, len);
2193 memset(tmp, 0, len);
2198 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2201 void *tmp = skb_put(skb, len);
2203 memcpy(tmp, data, len);
2208 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2210 *(u8 *)skb_put(skb, 1) = val;
2213 void *skb_push(struct sk_buff *skb, unsigned int len);
2214 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2221 void *skb_pull(struct sk_buff *skb, unsigned int len);
2222 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2225 BUG_ON(skb->len < skb->data_len);
2226 return skb->data += len;
2229 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2231 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2234 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2236 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2238 if (len > skb_headlen(skb) &&
2239 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2242 return skb->data += len;
2245 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2247 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2250 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2252 if (likely(len <= skb_headlen(skb)))
2254 if (unlikely(len > skb->len))
2256 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2259 void skb_condense(struct sk_buff *skb);
2262 * skb_headroom - bytes at buffer head
2263 * @skb: buffer to check
2265 * Return the number of bytes of free space at the head of an &sk_buff.
2267 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2269 return skb->data - skb->head;
2273 * skb_tailroom - bytes at buffer end
2274 * @skb: buffer to check
2276 * Return the number of bytes of free space at the tail of an sk_buff
2278 static inline int skb_tailroom(const struct sk_buff *skb)
2280 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2284 * skb_availroom - bytes at buffer end
2285 * @skb: buffer to check
2287 * Return the number of bytes of free space at the tail of an sk_buff
2288 * allocated by sk_stream_alloc()
2290 static inline int skb_availroom(const struct sk_buff *skb)
2292 if (skb_is_nonlinear(skb))
2295 return skb->end - skb->tail - skb->reserved_tailroom;
2299 * skb_reserve - adjust headroom
2300 * @skb: buffer to alter
2301 * @len: bytes to move
2303 * Increase the headroom of an empty &sk_buff by reducing the tail
2304 * room. This is only allowed for an empty buffer.
2306 static inline void skb_reserve(struct sk_buff *skb, int len)
2313 * skb_tailroom_reserve - adjust reserved_tailroom
2314 * @skb: buffer to alter
2315 * @mtu: maximum amount of headlen permitted
2316 * @needed_tailroom: minimum amount of reserved_tailroom
2318 * Set reserved_tailroom so that headlen can be as large as possible but
2319 * not larger than mtu and tailroom cannot be smaller than
2321 * The required headroom should already have been reserved before using
2324 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2325 unsigned int needed_tailroom)
2327 SKB_LINEAR_ASSERT(skb);
2328 if (mtu < skb_tailroom(skb) - needed_tailroom)
2329 /* use at most mtu */
2330 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2332 /* use up to all available space */
2333 skb->reserved_tailroom = needed_tailroom;
2336 #define ENCAP_TYPE_ETHER 0
2337 #define ENCAP_TYPE_IPPROTO 1
2339 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2342 skb->inner_protocol = protocol;
2343 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2346 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2349 skb->inner_ipproto = ipproto;
2350 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2353 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2355 skb->inner_mac_header = skb->mac_header;
2356 skb->inner_network_header = skb->network_header;
2357 skb->inner_transport_header = skb->transport_header;
2360 static inline void skb_reset_mac_len(struct sk_buff *skb)
2362 skb->mac_len = skb->network_header - skb->mac_header;
2365 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2368 return skb->head + skb->inner_transport_header;
2371 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2373 return skb_inner_transport_header(skb) - skb->data;
2376 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2378 skb->inner_transport_header = skb->data - skb->head;
2381 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2384 skb_reset_inner_transport_header(skb);
2385 skb->inner_transport_header += offset;
2388 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2390 return skb->head + skb->inner_network_header;
2393 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2395 skb->inner_network_header = skb->data - skb->head;
2398 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2401 skb_reset_inner_network_header(skb);
2402 skb->inner_network_header += offset;
2405 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2407 return skb->head + skb->inner_mac_header;
2410 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2412 skb->inner_mac_header = skb->data - skb->head;
2415 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2418 skb_reset_inner_mac_header(skb);
2419 skb->inner_mac_header += offset;
2421 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2423 return skb->transport_header != (typeof(skb->transport_header))~0U;
2426 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2428 return skb->head + skb->transport_header;
2431 static inline void skb_reset_transport_header(struct sk_buff *skb)
2433 skb->transport_header = skb->data - skb->head;
2436 static inline void skb_set_transport_header(struct sk_buff *skb,
2439 skb_reset_transport_header(skb);
2440 skb->transport_header += offset;
2443 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2445 return skb->head + skb->network_header;
2448 static inline void skb_reset_network_header(struct sk_buff *skb)
2450 skb->network_header = skb->data - skb->head;
2453 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2455 skb_reset_network_header(skb);
2456 skb->network_header += offset;
2459 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2461 return skb->head + skb->mac_header;
2464 static inline int skb_mac_offset(const struct sk_buff *skb)
2466 return skb_mac_header(skb) - skb->data;
2469 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2471 return skb->network_header - skb->mac_header;
2474 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2476 return skb->mac_header != (typeof(skb->mac_header))~0U;
2479 static inline void skb_reset_mac_header(struct sk_buff *skb)
2481 skb->mac_header = skb->data - skb->head;
2484 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2486 skb_reset_mac_header(skb);
2487 skb->mac_header += offset;
2490 static inline void skb_pop_mac_header(struct sk_buff *skb)
2492 skb->mac_header = skb->network_header;
2495 static inline void skb_probe_transport_header(struct sk_buff *skb)
2497 struct flow_keys_basic keys;
2499 if (skb_transport_header_was_set(skb))
2502 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2504 skb_set_transport_header(skb, keys.control.thoff);
2507 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2509 if (skb_mac_header_was_set(skb)) {
2510 const unsigned char *old_mac = skb_mac_header(skb);
2512 skb_set_mac_header(skb, -skb->mac_len);
2513 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2517 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2519 return skb->csum_start - skb_headroom(skb);
2522 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2524 return skb->head + skb->csum_start;
2527 static inline int skb_transport_offset(const struct sk_buff *skb)
2529 return skb_transport_header(skb) - skb->data;
2532 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2534 return skb->transport_header - skb->network_header;
2537 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2539 return skb->inner_transport_header - skb->inner_network_header;
2542 static inline int skb_network_offset(const struct sk_buff *skb)
2544 return skb_network_header(skb) - skb->data;
2547 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2549 return skb_inner_network_header(skb) - skb->data;
2552 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2554 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2558 * CPUs often take a performance hit when accessing unaligned memory
2559 * locations. The actual performance hit varies, it can be small if the
2560 * hardware handles it or large if we have to take an exception and fix it
2563 * Since an ethernet header is 14 bytes network drivers often end up with
2564 * the IP header at an unaligned offset. The IP header can be aligned by
2565 * shifting the start of the packet by 2 bytes. Drivers should do this
2568 * skb_reserve(skb, NET_IP_ALIGN);
2570 * The downside to this alignment of the IP header is that the DMA is now
2571 * unaligned. On some architectures the cost of an unaligned DMA is high
2572 * and this cost outweighs the gains made by aligning the IP header.
2574 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2577 #ifndef NET_IP_ALIGN
2578 #define NET_IP_ALIGN 2
2582 * The networking layer reserves some headroom in skb data (via
2583 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2584 * the header has to grow. In the default case, if the header has to grow
2585 * 32 bytes or less we avoid the reallocation.
2587 * Unfortunately this headroom changes the DMA alignment of the resulting
2588 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2589 * on some architectures. An architecture can override this value,
2590 * perhaps setting it to a cacheline in size (since that will maintain
2591 * cacheline alignment of the DMA). It must be a power of 2.
2593 * Various parts of the networking layer expect at least 32 bytes of
2594 * headroom, you should not reduce this.
2596 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2597 * to reduce average number of cache lines per packet.
2598 * get_rps_cpus() for example only access one 64 bytes aligned block :
2599 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2602 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2605 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2607 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2609 if (WARN_ON(skb_is_nonlinear(skb)))
2612 skb_set_tail_pointer(skb, len);
2615 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2617 __skb_set_length(skb, len);
2620 void skb_trim(struct sk_buff *skb, unsigned int len);
2622 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2625 return ___pskb_trim(skb, len);
2626 __skb_trim(skb, len);
2630 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2632 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2636 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2637 * @skb: buffer to alter
2640 * This is identical to pskb_trim except that the caller knows that
2641 * the skb is not cloned so we should never get an error due to out-
2644 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2646 int err = pskb_trim(skb, len);
2650 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2652 unsigned int diff = len - skb->len;
2654 if (skb_tailroom(skb) < diff) {
2655 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2660 __skb_set_length(skb, len);
2665 * skb_orphan - orphan a buffer
2666 * @skb: buffer to orphan
2668 * If a buffer currently has an owner then we call the owner's
2669 * destructor function and make the @skb unowned. The buffer continues
2670 * to exist but is no longer charged to its former owner.
2672 static inline void skb_orphan(struct sk_buff *skb)
2674 if (skb->destructor) {
2675 skb->destructor(skb);
2676 skb->destructor = NULL;
2684 * skb_orphan_frags - orphan the frags contained in a buffer
2685 * @skb: buffer to orphan frags from
2686 * @gfp_mask: allocation mask for replacement pages
2688 * For each frag in the SKB which needs a destructor (i.e. has an
2689 * owner) create a copy of that frag and release the original
2690 * page by calling the destructor.
2692 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2694 if (likely(!skb_zcopy(skb)))
2696 if (!skb_zcopy_is_nouarg(skb) &&
2697 skb_uarg(skb)->callback == sock_zerocopy_callback)
2699 return skb_copy_ubufs(skb, gfp_mask);
2702 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2703 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2705 if (likely(!skb_zcopy(skb)))
2707 return skb_copy_ubufs(skb, gfp_mask);
2711 * __skb_queue_purge - empty a list
2712 * @list: list to empty
2714 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2715 * the list and one reference dropped. This function does not take the
2716 * list lock and the caller must hold the relevant locks to use it.
2718 static inline void __skb_queue_purge(struct sk_buff_head *list)
2720 struct sk_buff *skb;
2721 while ((skb = __skb_dequeue(list)) != NULL)
2724 void skb_queue_purge(struct sk_buff_head *list);
2726 unsigned int skb_rbtree_purge(struct rb_root *root);
2728 void *netdev_alloc_frag(unsigned int fragsz);
2730 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2734 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2735 * @dev: network device to receive on
2736 * @length: length to allocate
2738 * Allocate a new &sk_buff and assign it a usage count of one. The
2739 * buffer has unspecified headroom built in. Users should allocate
2740 * the headroom they think they need without accounting for the
2741 * built in space. The built in space is used for optimisations.
2743 * %NULL is returned if there is no free memory. Although this function
2744 * allocates memory it can be called from an interrupt.
2746 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2747 unsigned int length)
2749 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2752 /* legacy helper around __netdev_alloc_skb() */
2753 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2756 return __netdev_alloc_skb(NULL, length, gfp_mask);
2759 /* legacy helper around netdev_alloc_skb() */
2760 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2762 return netdev_alloc_skb(NULL, length);
2766 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2767 unsigned int length, gfp_t gfp)
2769 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2771 if (NET_IP_ALIGN && skb)
2772 skb_reserve(skb, NET_IP_ALIGN);
2776 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2777 unsigned int length)
2779 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2782 static inline void skb_free_frag(void *addr)
2784 page_frag_free(addr);
2787 void *napi_alloc_frag(unsigned int fragsz);
2788 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2789 unsigned int length, gfp_t gfp_mask);
2790 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2791 unsigned int length)
2793 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2795 void napi_consume_skb(struct sk_buff *skb, int budget);
2797 void __kfree_skb_flush(void);
2798 void __kfree_skb_defer(struct sk_buff *skb);
2801 * __dev_alloc_pages - allocate page for network Rx
2802 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2803 * @order: size of the allocation
2805 * Allocate a new page.
2807 * %NULL is returned if there is no free memory.
2809 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2812 /* This piece of code contains several assumptions.
2813 * 1. This is for device Rx, therefor a cold page is preferred.
2814 * 2. The expectation is the user wants a compound page.
2815 * 3. If requesting a order 0 page it will not be compound
2816 * due to the check to see if order has a value in prep_new_page
2817 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2818 * code in gfp_to_alloc_flags that should be enforcing this.
2820 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2822 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2825 static inline struct page *dev_alloc_pages(unsigned int order)
2827 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2831 * __dev_alloc_page - allocate a page for network Rx
2832 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2834 * Allocate a new page.
2836 * %NULL is returned if there is no free memory.
2838 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2840 return __dev_alloc_pages(gfp_mask, 0);
2843 static inline struct page *dev_alloc_page(void)
2845 return dev_alloc_pages(0);
2849 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2850 * @page: The page that was allocated from skb_alloc_page
2851 * @skb: The skb that may need pfmemalloc set
2853 static inline void skb_propagate_pfmemalloc(struct page *page,
2854 struct sk_buff *skb)
2856 if (page_is_pfmemalloc(page))
2857 skb->pfmemalloc = true;
2861 * skb_frag_off() - Returns the offset of a skb fragment
2862 * @frag: the paged fragment
2864 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
2866 return frag->bv_offset;
2870 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
2871 * @frag: skb fragment
2872 * @delta: value to add
2874 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
2876 frag->bv_offset += delta;
2880 * skb_frag_off_set() - Sets the offset of a skb fragment
2881 * @frag: skb fragment
2882 * @offset: offset of fragment
2884 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
2886 frag->bv_offset = offset;
2890 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
2891 * @fragto: skb fragment where offset is set
2892 * @fragfrom: skb fragment offset is copied from
2894 static inline void skb_frag_off_copy(skb_frag_t *fragto,
2895 const skb_frag_t *fragfrom)
2897 fragto->bv_offset = fragfrom->bv_offset;
2901 * skb_frag_page - retrieve the page referred to by a paged fragment
2902 * @frag: the paged fragment
2904 * Returns the &struct page associated with @frag.
2906 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2908 return frag->bv_page;
2912 * __skb_frag_ref - take an addition reference on a paged fragment.
2913 * @frag: the paged fragment
2915 * Takes an additional reference on the paged fragment @frag.
2917 static inline void __skb_frag_ref(skb_frag_t *frag)
2919 get_page(skb_frag_page(frag));
2923 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2925 * @f: the fragment offset.
2927 * Takes an additional reference on the @f'th paged fragment of @skb.
2929 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2931 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2935 * __skb_frag_unref - release a reference on a paged fragment.
2936 * @frag: the paged fragment
2938 * Releases a reference on the paged fragment @frag.
2940 static inline void __skb_frag_unref(skb_frag_t *frag)
2942 put_page(skb_frag_page(frag));
2946 * skb_frag_unref - release a reference on a paged fragment of an skb.
2948 * @f: the fragment offset
2950 * Releases a reference on the @f'th paged fragment of @skb.
2952 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2954 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2958 * skb_frag_address - gets the address of the data contained in a paged fragment
2959 * @frag: the paged fragment buffer
2961 * Returns the address of the data within @frag. The page must already
2964 static inline void *skb_frag_address(const skb_frag_t *frag)
2966 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
2970 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2971 * @frag: the paged fragment buffer
2973 * Returns the address of the data within @frag. Checks that the page
2974 * is mapped and returns %NULL otherwise.
2976 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2978 void *ptr = page_address(skb_frag_page(frag));
2982 return ptr + skb_frag_off(frag);
2986 * skb_frag_page_copy() - sets the page in a fragment from another fragment
2987 * @fragto: skb fragment where page is set
2988 * @fragfrom: skb fragment page is copied from
2990 static inline void skb_frag_page_copy(skb_frag_t *fragto,
2991 const skb_frag_t *fragfrom)
2993 fragto->bv_page = fragfrom->bv_page;
2997 * __skb_frag_set_page - sets the page contained in a paged fragment
2998 * @frag: the paged fragment
2999 * @page: the page to set
3001 * Sets the fragment @frag to contain @page.
3003 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3005 frag->bv_page = page;
3009 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3011 * @f: the fragment offset
3012 * @page: the page to set
3014 * Sets the @f'th fragment of @skb to contain @page.
3016 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3019 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3022 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3025 * skb_frag_dma_map - maps a paged fragment via the DMA API
3026 * @dev: the device to map the fragment to
3027 * @frag: the paged fragment to map
3028 * @offset: the offset within the fragment (starting at the
3029 * fragment's own offset)
3030 * @size: the number of bytes to map
3031 * @dir: the direction of the mapping (``PCI_DMA_*``)
3033 * Maps the page associated with @frag to @device.
3035 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3036 const skb_frag_t *frag,
3037 size_t offset, size_t size,
3038 enum dma_data_direction dir)
3040 return dma_map_page(dev, skb_frag_page(frag),
3041 skb_frag_off(frag) + offset, size, dir);
3044 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3047 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3051 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3054 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3059 * skb_clone_writable - is the header of a clone writable
3060 * @skb: buffer to check
3061 * @len: length up to which to write
3063 * Returns true if modifying the header part of the cloned buffer
3064 * does not requires the data to be copied.
3066 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3068 return !skb_header_cloned(skb) &&
3069 skb_headroom(skb) + len <= skb->hdr_len;
3072 static inline int skb_try_make_writable(struct sk_buff *skb,
3073 unsigned int write_len)
3075 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3076 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3079 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3084 if (headroom > skb_headroom(skb))
3085 delta = headroom - skb_headroom(skb);
3087 if (delta || cloned)
3088 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3094 * skb_cow - copy header of skb when it is required
3095 * @skb: buffer to cow
3096 * @headroom: needed headroom
3098 * If the skb passed lacks sufficient headroom or its data part
3099 * is shared, data is reallocated. If reallocation fails, an error
3100 * is returned and original skb is not changed.
3102 * The result is skb with writable area skb->head...skb->tail
3103 * and at least @headroom of space at head.
3105 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3107 return __skb_cow(skb, headroom, skb_cloned(skb));
3111 * skb_cow_head - skb_cow but only making the head writable
3112 * @skb: buffer to cow
3113 * @headroom: needed headroom
3115 * This function is identical to skb_cow except that we replace the
3116 * skb_cloned check by skb_header_cloned. It should be used when
3117 * you only need to push on some header and do not need to modify
3120 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3122 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3126 * skb_padto - pad an skbuff up to a minimal size
3127 * @skb: buffer to pad
3128 * @len: minimal length
3130 * Pads up a buffer to ensure the trailing bytes exist and are
3131 * blanked. If the buffer already contains sufficient data it
3132 * is untouched. Otherwise it is extended. Returns zero on
3133 * success. The skb is freed on error.
3135 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3137 unsigned int size = skb->len;
3138 if (likely(size >= len))
3140 return skb_pad(skb, len - size);
3144 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3145 * @skb: buffer to pad
3146 * @len: minimal length
3147 * @free_on_error: free buffer on error
3149 * Pads up a buffer to ensure the trailing bytes exist and are
3150 * blanked. If the buffer already contains sufficient data it
3151 * is untouched. Otherwise it is extended. Returns zero on
3152 * success. The skb is freed on error if @free_on_error is true.
3154 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
3157 unsigned int size = skb->len;
3159 if (unlikely(size < len)) {
3161 if (__skb_pad(skb, len, free_on_error))
3163 __skb_put(skb, len);
3169 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3170 * @skb: buffer to pad
3171 * @len: minimal length
3173 * Pads up a buffer to ensure the trailing bytes exist and are
3174 * blanked. If the buffer already contains sufficient data it
3175 * is untouched. Otherwise it is extended. Returns zero on
3176 * success. The skb is freed on error.
3178 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
3180 return __skb_put_padto(skb, len, true);
3183 static inline int skb_add_data(struct sk_buff *skb,
3184 struct iov_iter *from, int copy)
3186 const int off = skb->len;
3188 if (skb->ip_summed == CHECKSUM_NONE) {
3190 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3192 skb->csum = csum_block_add(skb->csum, csum, off);
3195 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3198 __skb_trim(skb, off);
3202 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3203 const struct page *page, int off)
3208 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3210 return page == skb_frag_page(frag) &&
3211 off == skb_frag_off(frag) + skb_frag_size(frag);
3216 static inline int __skb_linearize(struct sk_buff *skb)
3218 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3222 * skb_linearize - convert paged skb to linear one
3223 * @skb: buffer to linarize
3225 * If there is no free memory -ENOMEM is returned, otherwise zero
3226 * is returned and the old skb data released.
3228 static inline int skb_linearize(struct sk_buff *skb)
3230 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3234 * skb_has_shared_frag - can any frag be overwritten
3235 * @skb: buffer to test
3237 * Return true if the skb has at least one frag that might be modified
3238 * by an external entity (as in vmsplice()/sendfile())
3240 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3242 return skb_is_nonlinear(skb) &&
3243 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3247 * skb_linearize_cow - make sure skb is linear and writable
3248 * @skb: buffer to process
3250 * If there is no free memory -ENOMEM is returned, otherwise zero
3251 * is returned and the old skb data released.
3253 static inline int skb_linearize_cow(struct sk_buff *skb)
3255 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3256 __skb_linearize(skb) : 0;
3259 static __always_inline void
3260 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3263 if (skb->ip_summed == CHECKSUM_COMPLETE)
3264 skb->csum = csum_block_sub(skb->csum,
3265 csum_partial(start, len, 0), off);
3266 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3267 skb_checksum_start_offset(skb) < 0)
3268 skb->ip_summed = CHECKSUM_NONE;
3272 * skb_postpull_rcsum - update checksum for received skb after pull
3273 * @skb: buffer to update
3274 * @start: start of data before pull
3275 * @len: length of data pulled
3277 * After doing a pull on a received packet, you need to call this to
3278 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3279 * CHECKSUM_NONE so that it can be recomputed from scratch.
3281 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3282 const void *start, unsigned int len)
3284 __skb_postpull_rcsum(skb, start, len, 0);
3287 static __always_inline void
3288 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3291 if (skb->ip_summed == CHECKSUM_COMPLETE)
3292 skb->csum = csum_block_add(skb->csum,
3293 csum_partial(start, len, 0), off);
3297 * skb_postpush_rcsum - update checksum for received skb after push
3298 * @skb: buffer to update
3299 * @start: start of data after push
3300 * @len: length of data pushed
3302 * After doing a push on a received packet, you need to call this to
3303 * update the CHECKSUM_COMPLETE checksum.
3305 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3306 const void *start, unsigned int len)
3308 __skb_postpush_rcsum(skb, start, len, 0);
3311 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3314 * skb_push_rcsum - push skb and update receive checksum
3315 * @skb: buffer to update
3316 * @len: length of data pulled
3318 * This function performs an skb_push on the packet and updates
3319 * the CHECKSUM_COMPLETE checksum. It should be used on
3320 * receive path processing instead of skb_push unless you know
3321 * that the checksum difference is zero (e.g., a valid IP header)
3322 * or you are setting ip_summed to CHECKSUM_NONE.
3324 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3327 skb_postpush_rcsum(skb, skb->data, len);
3331 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3333 * pskb_trim_rcsum - trim received skb and update checksum
3334 * @skb: buffer to trim
3337 * This is exactly the same as pskb_trim except that it ensures the
3338 * checksum of received packets are still valid after the operation.
3339 * It can change skb pointers.
3342 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3344 if (likely(len >= skb->len))
3346 return pskb_trim_rcsum_slow(skb, len);
3349 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3351 if (skb->ip_summed == CHECKSUM_COMPLETE)
3352 skb->ip_summed = CHECKSUM_NONE;
3353 __skb_trim(skb, len);
3357 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3359 if (skb->ip_summed == CHECKSUM_COMPLETE)
3360 skb->ip_summed = CHECKSUM_NONE;
3361 return __skb_grow(skb, len);
3364 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3365 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3366 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3367 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3368 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3370 #define skb_queue_walk(queue, skb) \
3371 for (skb = (queue)->next; \
3372 skb != (struct sk_buff *)(queue); \
3375 #define skb_queue_walk_safe(queue, skb, tmp) \
3376 for (skb = (queue)->next, tmp = skb->next; \
3377 skb != (struct sk_buff *)(queue); \
3378 skb = tmp, tmp = skb->next)
3380 #define skb_queue_walk_from(queue, skb) \
3381 for (; skb != (struct sk_buff *)(queue); \
3384 #define skb_rbtree_walk(skb, root) \
3385 for (skb = skb_rb_first(root); skb != NULL; \
3386 skb = skb_rb_next(skb))
3388 #define skb_rbtree_walk_from(skb) \
3389 for (; skb != NULL; \
3390 skb = skb_rb_next(skb))
3392 #define skb_rbtree_walk_from_safe(skb, tmp) \
3393 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3396 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3397 for (tmp = skb->next; \
3398 skb != (struct sk_buff *)(queue); \
3399 skb = tmp, tmp = skb->next)
3401 #define skb_queue_reverse_walk(queue, skb) \
3402 for (skb = (queue)->prev; \
3403 skb != (struct sk_buff *)(queue); \
3406 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3407 for (skb = (queue)->prev, tmp = skb->prev; \
3408 skb != (struct sk_buff *)(queue); \
3409 skb = tmp, tmp = skb->prev)
3411 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3412 for (tmp = skb->prev; \
3413 skb != (struct sk_buff *)(queue); \
3414 skb = tmp, tmp = skb->prev)
3416 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3418 return skb_shinfo(skb)->frag_list != NULL;
3421 static inline void skb_frag_list_init(struct sk_buff *skb)
3423 skb_shinfo(skb)->frag_list = NULL;
3426 #define skb_walk_frags(skb, iter) \
3427 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3430 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3431 const struct sk_buff *skb);
3432 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3433 struct sk_buff_head *queue,
3435 void (*destructor)(struct sock *sk,
3436 struct sk_buff *skb),
3438 struct sk_buff **last);
3439 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3440 void (*destructor)(struct sock *sk,
3441 struct sk_buff *skb),
3443 struct sk_buff **last);
3444 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3445 void (*destructor)(struct sock *sk,
3446 struct sk_buff *skb),
3447 int *off, int *err);
3448 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3450 __poll_t datagram_poll(struct file *file, struct socket *sock,
3451 struct poll_table_struct *wait);
3452 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3453 struct iov_iter *to, int size);
3454 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3455 struct msghdr *msg, int size)
3457 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3459 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3460 struct msghdr *msg);
3461 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3462 struct iov_iter *to, int len,
3463 struct ahash_request *hash);
3464 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3465 struct iov_iter *from, int len);
3466 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3467 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3468 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3469 static inline void skb_free_datagram_locked(struct sock *sk,
3470 struct sk_buff *skb)
3472 __skb_free_datagram_locked(sk, skb, 0);
3474 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3475 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3476 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3477 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3478 int len, __wsum csum);
3479 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3480 struct pipe_inode_info *pipe, unsigned int len,
3481 unsigned int flags);
3482 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3484 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3485 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3486 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3488 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3489 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3490 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3491 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3492 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3493 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3494 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3495 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3496 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3497 int skb_vlan_pop(struct sk_buff *skb);
3498 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3499 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto);
3500 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto);
3501 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3502 int skb_mpls_dec_ttl(struct sk_buff *skb);
3503 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3506 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3508 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3511 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3513 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3516 struct skb_checksum_ops {
3517 __wsum (*update)(const void *mem, int len, __wsum wsum);
3518 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3521 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3523 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3524 __wsum csum, const struct skb_checksum_ops *ops);
3525 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3528 static inline void * __must_check
3529 __skb_header_pointer(const struct sk_buff *skb, int offset,
3530 int len, void *data, int hlen, void *buffer)
3532 if (hlen - offset >= len)
3533 return data + offset;
3536 skb_copy_bits(skb, offset, buffer, len) < 0)
3542 static inline void * __must_check
3543 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3545 return __skb_header_pointer(skb, offset, len, skb->data,
3546 skb_headlen(skb), buffer);
3550 * skb_needs_linearize - check if we need to linearize a given skb
3551 * depending on the given device features.
3552 * @skb: socket buffer to check
3553 * @features: net device features
3555 * Returns true if either:
3556 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3557 * 2. skb is fragmented and the device does not support SG.
3559 static inline bool skb_needs_linearize(struct sk_buff *skb,
3560 netdev_features_t features)
3562 return skb_is_nonlinear(skb) &&
3563 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3564 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3567 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3569 const unsigned int len)
3571 memcpy(to, skb->data, len);
3574 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3575 const int offset, void *to,
3576 const unsigned int len)
3578 memcpy(to, skb->data + offset, len);
3581 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3583 const unsigned int len)
3585 memcpy(skb->data, from, len);
3588 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3591 const unsigned int len)
3593 memcpy(skb->data + offset, from, len);
3596 void skb_init(void);
3598 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3604 * skb_get_timestamp - get timestamp from a skb
3605 * @skb: skb to get stamp from
3606 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
3608 * Timestamps are stored in the skb as offsets to a base timestamp.
3609 * This function converts the offset back to a struct timeval and stores
3612 static inline void skb_get_timestamp(const struct sk_buff *skb,
3613 struct __kernel_old_timeval *stamp)
3615 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
3618 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
3619 struct __kernel_sock_timeval *stamp)
3621 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3623 stamp->tv_sec = ts.tv_sec;
3624 stamp->tv_usec = ts.tv_nsec / 1000;
3627 static inline void skb_get_timestampns(const struct sk_buff *skb,
3628 struct timespec *stamp)
3630 *stamp = ktime_to_timespec(skb->tstamp);
3633 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
3634 struct __kernel_timespec *stamp)
3636 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3638 stamp->tv_sec = ts.tv_sec;
3639 stamp->tv_nsec = ts.tv_nsec;
3642 static inline void __net_timestamp(struct sk_buff *skb)
3644 skb->tstamp = ktime_get_real();
3647 static inline ktime_t net_timedelta(ktime_t t)
3649 return ktime_sub(ktime_get_real(), t);
3652 static inline ktime_t net_invalid_timestamp(void)
3657 static inline u8 skb_metadata_len(const struct sk_buff *skb)
3659 return skb_shinfo(skb)->meta_len;
3662 static inline void *skb_metadata_end(const struct sk_buff *skb)
3664 return skb_mac_header(skb);
3667 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3668 const struct sk_buff *skb_b,
3671 const void *a = skb_metadata_end(skb_a);
3672 const void *b = skb_metadata_end(skb_b);
3673 /* Using more efficient varaiant than plain call to memcmp(). */
3674 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3678 #define __it(x, op) (x -= sizeof(u##op))
3679 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3680 case 32: diffs |= __it_diff(a, b, 64);
3682 case 24: diffs |= __it_diff(a, b, 64);
3684 case 16: diffs |= __it_diff(a, b, 64);
3686 case 8: diffs |= __it_diff(a, b, 64);
3688 case 28: diffs |= __it_diff(a, b, 64);
3690 case 20: diffs |= __it_diff(a, b, 64);
3692 case 12: diffs |= __it_diff(a, b, 64);
3694 case 4: diffs |= __it_diff(a, b, 32);
3699 return memcmp(a - meta_len, b - meta_len, meta_len);
3703 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3704 const struct sk_buff *skb_b)
3706 u8 len_a = skb_metadata_len(skb_a);
3707 u8 len_b = skb_metadata_len(skb_b);
3709 if (!(len_a | len_b))
3712 return len_a != len_b ?
3713 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3716 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3718 skb_shinfo(skb)->meta_len = meta_len;
3721 static inline void skb_metadata_clear(struct sk_buff *skb)
3723 skb_metadata_set(skb, 0);
3726 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3728 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3730 void skb_clone_tx_timestamp(struct sk_buff *skb);
3731 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3733 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3735 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3739 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3744 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3747 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3749 * PHY drivers may accept clones of transmitted packets for
3750 * timestamping via their phy_driver.txtstamp method. These drivers
3751 * must call this function to return the skb back to the stack with a
3754 * @skb: clone of the the original outgoing packet
3755 * @hwtstamps: hardware time stamps
3758 void skb_complete_tx_timestamp(struct sk_buff *skb,
3759 struct skb_shared_hwtstamps *hwtstamps);
3761 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3762 struct skb_shared_hwtstamps *hwtstamps,
3763 struct sock *sk, int tstype);
3766 * skb_tstamp_tx - queue clone of skb with send time stamps
3767 * @orig_skb: the original outgoing packet
3768 * @hwtstamps: hardware time stamps, may be NULL if not available
3770 * If the skb has a socket associated, then this function clones the
3771 * skb (thus sharing the actual data and optional structures), stores
3772 * the optional hardware time stamping information (if non NULL) or
3773 * generates a software time stamp (otherwise), then queues the clone
3774 * to the error queue of the socket. Errors are silently ignored.
3776 void skb_tstamp_tx(struct sk_buff *orig_skb,
3777 struct skb_shared_hwtstamps *hwtstamps);
3780 * skb_tx_timestamp() - Driver hook for transmit timestamping
3782 * Ethernet MAC Drivers should call this function in their hard_xmit()
3783 * function immediately before giving the sk_buff to the MAC hardware.
3785 * Specifically, one should make absolutely sure that this function is
3786 * called before TX completion of this packet can trigger. Otherwise
3787 * the packet could potentially already be freed.
3789 * @skb: A socket buffer.
3791 static inline void skb_tx_timestamp(struct sk_buff *skb)
3793 skb_clone_tx_timestamp(skb);
3794 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3795 skb_tstamp_tx(skb, NULL);
3799 * skb_complete_wifi_ack - deliver skb with wifi status
3801 * @skb: the original outgoing packet
3802 * @acked: ack status
3805 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3807 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3808 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3810 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3812 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3814 (skb->ip_summed == CHECKSUM_PARTIAL &&
3815 skb_checksum_start_offset(skb) >= 0));
3819 * skb_checksum_complete - Calculate checksum of an entire packet
3820 * @skb: packet to process
3822 * This function calculates the checksum over the entire packet plus
3823 * the value of skb->csum. The latter can be used to supply the
3824 * checksum of a pseudo header as used by TCP/UDP. It returns the
3827 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3828 * this function can be used to verify that checksum on received
3829 * packets. In that case the function should return zero if the
3830 * checksum is correct. In particular, this function will return zero
3831 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3832 * hardware has already verified the correctness of the checksum.
3834 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3836 return skb_csum_unnecessary(skb) ?
3837 0 : __skb_checksum_complete(skb);
3840 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3842 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3843 if (skb->csum_level == 0)
3844 skb->ip_summed = CHECKSUM_NONE;
3850 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3852 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3853 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3855 } else if (skb->ip_summed == CHECKSUM_NONE) {
3856 skb->ip_summed = CHECKSUM_UNNECESSARY;
3857 skb->csum_level = 0;
3861 /* Check if we need to perform checksum complete validation.
3863 * Returns true if checksum complete is needed, false otherwise
3864 * (either checksum is unnecessary or zero checksum is allowed).
3866 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3870 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3871 skb->csum_valid = 1;
3872 __skb_decr_checksum_unnecessary(skb);
3879 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
3882 #define CHECKSUM_BREAK 76
3884 /* Unset checksum-complete
3886 * Unset checksum complete can be done when packet is being modified
3887 * (uncompressed for instance) and checksum-complete value is
3890 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3892 if (skb->ip_summed == CHECKSUM_COMPLETE)
3893 skb->ip_summed = CHECKSUM_NONE;
3896 /* Validate (init) checksum based on checksum complete.
3899 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3900 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3901 * checksum is stored in skb->csum for use in __skb_checksum_complete
3902 * non-zero: value of invalid checksum
3905 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3909 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3910 if (!csum_fold(csum_add(psum, skb->csum))) {
3911 skb->csum_valid = 1;
3918 if (complete || skb->len <= CHECKSUM_BREAK) {
3921 csum = __skb_checksum_complete(skb);
3922 skb->csum_valid = !csum;
3929 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3934 /* Perform checksum validate (init). Note that this is a macro since we only
3935 * want to calculate the pseudo header which is an input function if necessary.
3936 * First we try to validate without any computation (checksum unnecessary) and
3937 * then calculate based on checksum complete calling the function to compute
3941 * 0: checksum is validated or try to in skb_checksum_complete
3942 * non-zero: value of invalid checksum
3944 #define __skb_checksum_validate(skb, proto, complete, \
3945 zero_okay, check, compute_pseudo) \
3947 __sum16 __ret = 0; \
3948 skb->csum_valid = 0; \
3949 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3950 __ret = __skb_checksum_validate_complete(skb, \
3951 complete, compute_pseudo(skb, proto)); \
3955 #define skb_checksum_init(skb, proto, compute_pseudo) \
3956 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3958 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3959 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3961 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3962 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3964 #define skb_checksum_validate_zero_check(skb, proto, check, \
3966 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3968 #define skb_checksum_simple_validate(skb) \
3969 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3971 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3973 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3976 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
3978 skb->csum = ~pseudo;
3979 skb->ip_summed = CHECKSUM_COMPLETE;
3982 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
3984 if (__skb_checksum_convert_check(skb)) \
3985 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
3988 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3989 u16 start, u16 offset)
3991 skb->ip_summed = CHECKSUM_PARTIAL;
3992 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3993 skb->csum_offset = offset - start;
3996 /* Update skbuf and packet to reflect the remote checksum offload operation.
3997 * When called, ptr indicates the starting point for skb->csum when
3998 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3999 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4001 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4002 int start, int offset, bool nopartial)
4007 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4011 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4012 __skb_checksum_complete(skb);
4013 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4016 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4018 /* Adjust skb->csum since we changed the packet */
4019 skb->csum = csum_add(skb->csum, delta);
4022 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4024 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4025 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
4031 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4032 void nf_conntrack_destroy(struct nf_conntrack *nfct);
4033 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
4035 if (nfct && atomic_dec_and_test(&nfct->use))
4036 nf_conntrack_destroy(nfct);
4038 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
4041 atomic_inc(&nfct->use);
4045 #ifdef CONFIG_SKB_EXTENSIONS
4047 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4053 SKB_EXT_NUM, /* must be last */
4057 * struct skb_ext - sk_buff extensions
4058 * @refcnt: 1 on allocation, deallocated on 0
4059 * @offset: offset to add to @data to obtain extension address
4060 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4061 * @data: start of extension data, variable sized
4063 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4064 * to use 'u8' types while allowing up to 2kb worth of extension data.
4068 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4069 u8 chunks; /* same */
4070 char data[0] __aligned(8);
4073 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4074 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4075 void __skb_ext_put(struct skb_ext *ext);
4077 static inline void skb_ext_put(struct sk_buff *skb)
4079 if (skb->active_extensions)
4080 __skb_ext_put(skb->extensions);
4083 static inline void __skb_ext_copy(struct sk_buff *dst,
4084 const struct sk_buff *src)
4086 dst->active_extensions = src->active_extensions;
4088 if (src->active_extensions) {
4089 struct skb_ext *ext = src->extensions;
4091 refcount_inc(&ext->refcnt);
4092 dst->extensions = ext;
4096 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4099 __skb_ext_copy(dst, src);
4102 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4104 return !!ext->offset[i];
4107 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4109 return skb->active_extensions & (1 << id);
4112 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4114 if (skb_ext_exist(skb, id))
4115 __skb_ext_del(skb, id);
4118 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4120 if (skb_ext_exist(skb, id)) {
4121 struct skb_ext *ext = skb->extensions;
4123 return (void *)ext + (ext->offset[id] << 3);
4129 static inline void skb_ext_put(struct sk_buff *skb) {}
4130 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4131 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4132 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4133 #endif /* CONFIG_SKB_EXTENSIONS */
4135 static inline void nf_reset(struct sk_buff *skb)
4137 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4138 nf_conntrack_put(skb_nfct(skb));
4141 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4142 skb_ext_del(skb, SKB_EXT_BRIDGE_NF);
4146 static inline void nf_reset_trace(struct sk_buff *skb)
4148 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4153 static inline void ipvs_reset(struct sk_buff *skb)
4155 #if IS_ENABLED(CONFIG_IP_VS)
4156 skb->ipvs_property = 0;
4160 /* Note: This doesn't put any conntrack info in dst. */
4161 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4164 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4165 dst->_nfct = src->_nfct;
4166 nf_conntrack_get(skb_nfct(src));
4168 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4170 dst->nf_trace = src->nf_trace;
4174 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4176 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4177 nf_conntrack_put(skb_nfct(dst));
4179 __nf_copy(dst, src, true);
4182 #ifdef CONFIG_NETWORK_SECMARK
4183 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4185 to->secmark = from->secmark;
4188 static inline void skb_init_secmark(struct sk_buff *skb)
4193 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4196 static inline void skb_init_secmark(struct sk_buff *skb)
4200 static inline int secpath_exists(const struct sk_buff *skb)
4203 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4209 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4211 return !skb->destructor &&
4212 !secpath_exists(skb) &&
4214 !skb->_skb_refdst &&
4215 !skb_has_frag_list(skb);
4218 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4220 skb->queue_mapping = queue_mapping;
4223 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4225 return skb->queue_mapping;
4228 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4230 to->queue_mapping = from->queue_mapping;
4233 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4235 skb->queue_mapping = rx_queue + 1;
4238 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4240 return skb->queue_mapping - 1;
4243 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4245 return skb->queue_mapping != 0;
4248 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4250 skb->dst_pending_confirm = val;
4253 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4255 return skb->dst_pending_confirm != 0;
4258 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4261 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4267 /* Keeps track of mac header offset relative to skb->head.
4268 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4269 * For non-tunnel skb it points to skb_mac_header() and for
4270 * tunnel skb it points to outer mac header.
4271 * Keeps track of level of encapsulation of network headers.
4282 #define SKB_SGO_CB_OFFSET 32
4283 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
4285 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4287 return (skb_mac_header(inner_skb) - inner_skb->head) -
4288 SKB_GSO_CB(inner_skb)->mac_offset;
4291 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4293 int new_headroom, headroom;
4296 headroom = skb_headroom(skb);
4297 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4301 new_headroom = skb_headroom(skb);
4302 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4306 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4308 /* Do not update partial checksums if remote checksum is enabled. */
4309 if (skb->remcsum_offload)
4312 SKB_GSO_CB(skb)->csum = res;
4313 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4316 /* Compute the checksum for a gso segment. First compute the checksum value
4317 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4318 * then add in skb->csum (checksum from csum_start to end of packet).
4319 * skb->csum and csum_start are then updated to reflect the checksum of the
4320 * resultant packet starting from the transport header-- the resultant checksum
4321 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4324 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4326 unsigned char *csum_start = skb_transport_header(skb);
4327 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4328 __wsum partial = SKB_GSO_CB(skb)->csum;
4330 SKB_GSO_CB(skb)->csum = res;
4331 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4333 return csum_fold(csum_partial(csum_start, plen, partial));
4336 static inline bool skb_is_gso(const struct sk_buff *skb)
4338 return skb_shinfo(skb)->gso_size;
4341 /* Note: Should be called only if skb_is_gso(skb) is true */
4342 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4344 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4347 /* Note: Should be called only if skb_is_gso(skb) is true */
4348 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4350 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4353 /* Note: Should be called only if skb_is_gso(skb) is true */
4354 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4356 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4359 static inline void skb_gso_reset(struct sk_buff *skb)
4361 skb_shinfo(skb)->gso_size = 0;
4362 skb_shinfo(skb)->gso_segs = 0;
4363 skb_shinfo(skb)->gso_type = 0;
4366 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4369 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4371 shinfo->gso_size += increment;
4374 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4377 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4379 shinfo->gso_size -= decrement;
4382 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4384 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4386 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4387 * wanted then gso_type will be set. */
4388 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4390 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4391 unlikely(shinfo->gso_type == 0)) {
4392 __skb_warn_lro_forwarding(skb);
4398 static inline void skb_forward_csum(struct sk_buff *skb)
4400 /* Unfortunately we don't support this one. Any brave souls? */
4401 if (skb->ip_summed == CHECKSUM_COMPLETE)
4402 skb->ip_summed = CHECKSUM_NONE;
4406 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4407 * @skb: skb to check
4409 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4410 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4411 * use this helper, to document places where we make this assertion.
4413 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4416 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4420 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4422 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4423 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4424 unsigned int transport_len,
4425 __sum16(*skb_chkf)(struct sk_buff *skb));
4428 * skb_head_is_locked - Determine if the skb->head is locked down
4429 * @skb: skb to check
4431 * The head on skbs build around a head frag can be removed if they are
4432 * not cloned. This function returns true if the skb head is locked down
4433 * due to either being allocated via kmalloc, or by being a clone with
4434 * multiple references to the head.
4436 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4438 return !skb->head_frag || skb_cloned(skb);
4441 /* Local Checksum Offload.
4442 * Compute outer checksum based on the assumption that the
4443 * inner checksum will be offloaded later.
4444 * See Documentation/networking/checksum-offloads.rst for
4445 * explanation of how this works.
4446 * Fill in outer checksum adjustment (e.g. with sum of outer
4447 * pseudo-header) before calling.
4448 * Also ensure that inner checksum is in linear data area.
4450 static inline __wsum lco_csum(struct sk_buff *skb)
4452 unsigned char *csum_start = skb_checksum_start(skb);
4453 unsigned char *l4_hdr = skb_transport_header(skb);
4456 /* Start with complement of inner checksum adjustment */
4457 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4460 /* Add in checksum of our headers (incl. outer checksum
4461 * adjustment filled in by caller) and return result.
4463 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4466 #endif /* __KERNEL__ */
4467 #endif /* _LINUX_SKBUFF_H */