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>
40 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
41 #include <linux/netfilter/nf_conntrack_common.h>
44 /* The interface for checksum offload between the stack and networking drivers
47 * A. IP checksum related features
49 * Drivers advertise checksum offload capabilities in the features of a device.
50 * From the stack's point of view these are capabilities offered by the driver,
51 * a driver typically only advertises features that it is capable of offloading
54 * The checksum related features are:
56 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
57 * IP (one's complement) checksum for any combination
58 * of protocols or protocol layering. The checksum is
59 * computed and set in a packet per the CHECKSUM_PARTIAL
60 * interface (see below).
62 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
63 * TCP or UDP packets over IPv4. These are specifically
64 * unencapsulated packets of the form IPv4|TCP or
65 * IPv4|UDP where the Protocol field in the IPv4 header
66 * is TCP or UDP. The IPv4 header may contain IP options
67 * This feature cannot be set in features for a device
68 * with NETIF_F_HW_CSUM also set. This feature is being
69 * DEPRECATED (see below).
71 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
72 * TCP or UDP packets over IPv6. These are specifically
73 * unencapsulated packets of the form IPv6|TCP or
74 * IPv4|UDP where the Next Header field in the IPv6
75 * header is either TCP or UDP. IPv6 extension headers
76 * are not supported with this feature. This feature
77 * cannot be set in features for a device with
78 * NETIF_F_HW_CSUM also set. This feature is being
79 * DEPRECATED (see below).
81 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
82 * This flag is used only used to disable the RX checksum
83 * feature for a device. The stack will accept receive
84 * checksum indication in packets received on a device
85 * regardless of whether NETIF_F_RXCSUM is set.
87 * B. Checksumming of received packets by device. Indication of checksum
88 * verification is in set skb->ip_summed. Possible values are:
92 * Device did not checksum this packet e.g. due to lack of capabilities.
93 * The packet contains full (though not verified) checksum in packet but
94 * not in skb->csum. Thus, skb->csum is undefined in this case.
96 * CHECKSUM_UNNECESSARY:
98 * The hardware you're dealing with doesn't calculate the full checksum
99 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
100 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
101 * if their checksums are okay. skb->csum is still undefined in this case
102 * though. A driver or device must never modify the checksum field in the
103 * packet even if checksum is verified.
105 * CHECKSUM_UNNECESSARY is applicable to following protocols:
106 * TCP: IPv6 and IPv4.
107 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
108 * zero UDP checksum for either IPv4 or IPv6, the networking stack
109 * may perform further validation in this case.
110 * GRE: only if the checksum is present in the header.
111 * SCTP: indicates the CRC in SCTP header has been validated.
112 * FCOE: indicates the CRC in FC frame has been validated.
114 * skb->csum_level indicates the number of consecutive checksums found in
115 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
116 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
117 * and a device is able to verify the checksums for UDP (possibly zero),
118 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
119 * two. If the device were only able to verify the UDP checksum and not
120 * GRE, either because it doesn't support GRE checksum of because GRE
121 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
122 * not considered in this case).
126 * This is the most generic way. The device supplied checksum of the _whole_
127 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
128 * hardware doesn't need to parse L3/L4 headers to implement this.
131 * - Even if device supports only some protocols, but is able to produce
132 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
133 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
137 * A checksum is set up to be offloaded to a device as described in the
138 * output description for CHECKSUM_PARTIAL. This may occur on a packet
139 * received directly from another Linux OS, e.g., a virtualized Linux kernel
140 * on the same host, or it may be set in the input path in GRO or remote
141 * checksum offload. For the purposes of checksum verification, the checksum
142 * referred to by skb->csum_start + skb->csum_offset and any preceding
143 * checksums in the packet are considered verified. Any checksums in the
144 * packet that are after the checksum being offloaded are not considered to
147 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
148 * in the skb->ip_summed for a packet. Values are:
152 * The driver is required to checksum the packet as seen by hard_start_xmit()
153 * from skb->csum_start up to the end, and to record/write the checksum at
154 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
155 * csum_start and csum_offset values are valid values given the length and
156 * offset of the packet, however they should not attempt to validate that the
157 * checksum refers to a legitimate transport layer checksum-- it is the
158 * purview of the stack to validate that csum_start and csum_offset are set
161 * When the stack requests checksum offload for a packet, the driver MUST
162 * ensure that the checksum is set correctly. A driver can either offload the
163 * checksum calculation to the device, or call skb_checksum_help (in the case
164 * that the device does not support offload for a particular checksum).
166 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
167 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
168 * checksum offload capability.
169 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
170 * on network device checksumming capabilities: if a packet does not match
171 * them, skb_checksum_help or skb_crc32c_help (depending on the value of
172 * csum_not_inet, see item D.) is called to resolve the checksum.
176 * The skb was already checksummed by the protocol, or a checksum is not
179 * CHECKSUM_UNNECESSARY:
181 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
185 * Not used in checksum output. If a driver observes a packet with this value
186 * set in skbuff, if should treat as CHECKSUM_NONE being set.
188 * D. Non-IP checksum (CRC) offloads
190 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
191 * offloading the SCTP CRC in a packet. To perform this offload the stack
192 * will set set csum_start and csum_offset accordingly, set ip_summed to
193 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
194 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
195 * A driver that supports both IP checksum offload and SCTP CRC32c offload
196 * must verify which offload is configured for a packet by testing the
197 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
198 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
200 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
201 * offloading the FCOE CRC in a packet. To perform this offload the stack
202 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
203 * accordingly. Note the there is no indication in the skbuff that the
204 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
205 * both IP checksum offload and FCOE CRC offload must verify which offload
206 * is configured for a packet presumably by inspecting packet headers.
208 * E. Checksumming on output with GSO.
210 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
211 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
212 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
213 * part of the GSO operation is implied. If a checksum is being offloaded
214 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
215 * are set to refer to the outermost checksum being offload (two offloaded
216 * checksums are possible with UDP encapsulation).
219 /* Don't change this without changing skb_csum_unnecessary! */
220 #define CHECKSUM_NONE 0
221 #define CHECKSUM_UNNECESSARY 1
222 #define CHECKSUM_COMPLETE 2
223 #define CHECKSUM_PARTIAL 3
225 /* Maximum value in skb->csum_level */
226 #define SKB_MAX_CSUM_LEVEL 3
228 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
229 #define SKB_WITH_OVERHEAD(X) \
230 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
231 #define SKB_MAX_ORDER(X, ORDER) \
232 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
233 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
234 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
236 /* return minimum truesize of one skb containing X bytes of data */
237 #define SKB_TRUESIZE(X) ((X) + \
238 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
239 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
243 struct pipe_inode_info;
250 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
251 struct nf_bridge_info {
253 BRNF_PROTO_UNCHANGED,
261 struct net_device *physindev;
263 /* always valid & non-NULL from FORWARD on, for physdev match */
264 struct net_device *physoutdev;
266 /* prerouting: detect dnat in orig/reply direction */
268 struct in6_addr ipv6_daddr;
270 /* after prerouting + nat detected: store original source
271 * mac since neigh resolution overwrites it, only used while
272 * skb is out in neigh layer.
274 char neigh_header[8];
279 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
280 /* Chain in tc_skb_ext will be used to share the tc chain with
281 * ovs recirc_id. It will be set to the current chain by tc
282 * and read by ovs to recirc_id.
289 struct sk_buff_head {
290 /* These two members must be first. */
291 struct sk_buff *next;
292 struct sk_buff *prev;
300 /* To allow 64K frame to be packed as single skb without frag_list we
301 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
302 * buffers which do not start on a page boundary.
304 * Since GRO uses frags we allocate at least 16 regardless of page
307 #if (65536/PAGE_SIZE + 1) < 16
308 #define MAX_SKB_FRAGS 16UL
310 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
312 extern int sysctl_max_skb_frags;
314 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
315 * segment using its current segmentation instead.
317 #define GSO_BY_FRAGS 0xFFFF
319 typedef struct bio_vec skb_frag_t;
322 * skb_frag_size() - Returns the size of a skb fragment
323 * @frag: skb fragment
325 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
331 * skb_frag_size_set() - Sets the size of a skb fragment
332 * @frag: skb fragment
333 * @size: size of fragment
335 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
341 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
342 * @frag: skb fragment
343 * @delta: value to add
345 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
347 frag->bv_len += delta;
351 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
352 * @frag: skb fragment
353 * @delta: value to subtract
355 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
357 frag->bv_len -= delta;
361 * skb_frag_must_loop - Test if %p is a high memory page
362 * @p: fragment's page
364 static inline bool skb_frag_must_loop(struct page *p)
366 #if defined(CONFIG_HIGHMEM)
374 * skb_frag_foreach_page - loop over pages in a fragment
376 * @f: skb frag to operate on
377 * @f_off: offset from start of f->bv_page
378 * @f_len: length from f_off to loop over
379 * @p: (temp var) current page
380 * @p_off: (temp var) offset from start of current page,
381 * non-zero only on first page.
382 * @p_len: (temp var) length in current page,
383 * < PAGE_SIZE only on first and last page.
384 * @copied: (temp var) length so far, excluding current p_len.
386 * A fragment can hold a compound page, in which case per-page
387 * operations, notably kmap_atomic, must be called for each
390 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
391 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
392 p_off = (f_off) & (PAGE_SIZE - 1), \
393 p_len = skb_frag_must_loop(p) ? \
394 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
397 copied += p_len, p++, p_off = 0, \
398 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
400 #define HAVE_HW_TIME_STAMP
403 * struct skb_shared_hwtstamps - hardware time stamps
404 * @hwtstamp: hardware time stamp transformed into duration
405 * since arbitrary point in time
407 * Software time stamps generated by ktime_get_real() are stored in
410 * hwtstamps can only be compared against other hwtstamps from
413 * This structure is attached to packets as part of the
414 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
416 struct skb_shared_hwtstamps {
420 /* Definitions for tx_flags in struct skb_shared_info */
422 /* generate hardware time stamp */
423 SKBTX_HW_TSTAMP = 1 << 0,
425 /* generate software time stamp when queueing packet to NIC */
426 SKBTX_SW_TSTAMP = 1 << 1,
428 /* device driver is going to provide hardware time stamp */
429 SKBTX_IN_PROGRESS = 1 << 2,
431 /* device driver supports TX zero-copy buffers */
432 SKBTX_DEV_ZEROCOPY = 1 << 3,
434 /* generate wifi status information (where possible) */
435 SKBTX_WIFI_STATUS = 1 << 4,
437 /* This indicates at least one fragment might be overwritten
438 * (as in vmsplice(), sendfile() ...)
439 * If we need to compute a TX checksum, we'll need to copy
440 * all frags to avoid possible bad checksum
442 SKBTX_SHARED_FRAG = 1 << 5,
444 /* generate software time stamp when entering packet scheduling */
445 SKBTX_SCHED_TSTAMP = 1 << 6,
448 #define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
449 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
451 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
454 * The callback notifies userspace to release buffers when skb DMA is done in
455 * lower device, the skb last reference should be 0 when calling this.
456 * The zerocopy_success argument is true if zero copy transmit occurred,
457 * false on data copy or out of memory error caused by data copy attempt.
458 * The ctx field is used to track device context.
459 * The desc field is used to track userspace buffer index.
462 void (*callback)(struct ubuf_info *, bool zerocopy_success);
478 struct user_struct *user;
483 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
485 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
486 void mm_unaccount_pinned_pages(struct mmpin *mmp);
488 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
489 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
490 struct ubuf_info *uarg);
492 static inline void sock_zerocopy_get(struct ubuf_info *uarg)
494 refcount_inc(&uarg->refcnt);
497 void sock_zerocopy_put(struct ubuf_info *uarg);
498 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
500 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
502 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len);
503 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
504 struct msghdr *msg, int len,
505 struct ubuf_info *uarg);
507 /* This data is invariant across clones and lives at
508 * the end of the header data, ie. at skb->end.
510 struct skb_shared_info {
515 unsigned short gso_size;
516 /* Warning: this field is not always filled in (UFO)! */
517 unsigned short gso_segs;
518 struct sk_buff *frag_list;
519 struct skb_shared_hwtstamps hwtstamps;
520 unsigned int gso_type;
524 * Warning : all fields before dataref are cleared in __alloc_skb()
528 /* Intermediate layers must ensure that destructor_arg
529 * remains valid until skb destructor */
530 void * destructor_arg;
532 /* must be last field, see pskb_expand_head() */
533 skb_frag_t frags[MAX_SKB_FRAGS];
536 /* We divide dataref into two halves. The higher 16 bits hold references
537 * to the payload part of skb->data. The lower 16 bits hold references to
538 * the entire skb->data. A clone of a headerless skb holds the length of
539 * the header in skb->hdr_len.
541 * All users must obey the rule that the skb->data reference count must be
542 * greater than or equal to the payload reference count.
544 * Holding a reference to the payload part means that the user does not
545 * care about modifications to the header part of skb->data.
547 #define SKB_DATAREF_SHIFT 16
548 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
552 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
553 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
554 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
558 SKB_GSO_TCPV4 = 1 << 0,
560 /* This indicates the skb is from an untrusted source. */
561 SKB_GSO_DODGY = 1 << 1,
563 /* This indicates the tcp segment has CWR set. */
564 SKB_GSO_TCP_ECN = 1 << 2,
566 SKB_GSO_TCP_FIXEDID = 1 << 3,
568 SKB_GSO_TCPV6 = 1 << 4,
570 SKB_GSO_FCOE = 1 << 5,
572 SKB_GSO_GRE = 1 << 6,
574 SKB_GSO_GRE_CSUM = 1 << 7,
576 SKB_GSO_IPXIP4 = 1 << 8,
578 SKB_GSO_IPXIP6 = 1 << 9,
580 SKB_GSO_UDP_TUNNEL = 1 << 10,
582 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
584 SKB_GSO_PARTIAL = 1 << 12,
586 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
588 SKB_GSO_SCTP = 1 << 14,
590 SKB_GSO_ESP = 1 << 15,
592 SKB_GSO_UDP = 1 << 16,
594 SKB_GSO_UDP_L4 = 1 << 17,
596 SKB_GSO_FRAGLIST = 1 << 18,
599 #if BITS_PER_LONG > 32
600 #define NET_SKBUFF_DATA_USES_OFFSET 1
603 #ifdef NET_SKBUFF_DATA_USES_OFFSET
604 typedef unsigned int sk_buff_data_t;
606 typedef unsigned char *sk_buff_data_t;
610 * struct sk_buff - socket buffer
611 * @next: Next buffer in list
612 * @prev: Previous buffer in list
613 * @tstamp: Time we arrived/left
614 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
615 * @sk: Socket we are owned by
616 * @dev: Device we arrived on/are leaving by
617 * @cb: Control buffer. Free for use by every layer. Put private vars here
618 * @_skb_refdst: destination entry (with norefcount bit)
619 * @sp: the security path, used for xfrm
620 * @len: Length of actual data
621 * @data_len: Data length
622 * @mac_len: Length of link layer header
623 * @hdr_len: writable header length of cloned skb
624 * @csum: Checksum (must include start/offset pair)
625 * @csum_start: Offset from skb->head where checksumming should start
626 * @csum_offset: Offset from csum_start where checksum should be stored
627 * @priority: Packet queueing priority
628 * @ignore_df: allow local fragmentation
629 * @cloned: Head may be cloned (check refcnt to be sure)
630 * @ip_summed: Driver fed us an IP checksum
631 * @nohdr: Payload reference only, must not modify header
632 * @pkt_type: Packet class
633 * @fclone: skbuff clone status
634 * @ipvs_property: skbuff is owned by ipvs
635 * @offload_fwd_mark: Packet was L2-forwarded in hardware
636 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
637 * @tc_skip_classify: do not classify packet. set by IFB device
638 * @tc_at_ingress: used within tc_classify to distinguish in/egress
639 * @tc_redirected: packet was redirected by a tc action
640 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
641 * @peeked: this packet has been seen already, so stats have been
642 * done for it, don't do them again
643 * @nf_trace: netfilter packet trace flag
644 * @protocol: Packet protocol from driver
645 * @destructor: Destruct function
646 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
647 * @_nfct: Associated connection, if any (with nfctinfo bits)
648 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
649 * @skb_iif: ifindex of device we arrived on
650 * @tc_index: Traffic control index
651 * @hash: the packet hash
652 * @queue_mapping: Queue mapping for multiqueue devices
653 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
654 * @active_extensions: active extensions (skb_ext_id types)
655 * @ndisc_nodetype: router type (from link layer)
656 * @ooo_okay: allow the mapping of a socket to a queue to be changed
657 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
659 * @sw_hash: indicates hash was computed in software stack
660 * @wifi_acked_valid: wifi_acked was set
661 * @wifi_acked: whether frame was acked on wifi or not
662 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
663 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
664 * @dst_pending_confirm: need to confirm neighbour
665 * @decrypted: Decrypted SKB
666 * @napi_id: id of the NAPI struct this skb came from
667 * @secmark: security marking
668 * @mark: Generic packet mark
669 * @vlan_proto: vlan encapsulation protocol
670 * @vlan_tci: vlan tag control information
671 * @inner_protocol: Protocol (encapsulation)
672 * @inner_transport_header: Inner transport layer header (encapsulation)
673 * @inner_network_header: Network layer header (encapsulation)
674 * @inner_mac_header: Link layer header (encapsulation)
675 * @transport_header: Transport layer header
676 * @network_header: Network layer header
677 * @mac_header: Link layer header
678 * @tail: Tail pointer
680 * @head: Head of buffer
681 * @data: Data head pointer
682 * @truesize: Buffer size
683 * @users: User count - see {datagram,tcp}.c
684 * @extensions: allocated extensions, valid if active_extensions is nonzero
690 /* These two members must be first. */
691 struct sk_buff *next;
692 struct sk_buff *prev;
695 struct net_device *dev;
696 /* Some protocols might use this space to store information,
697 * while device pointer would be NULL.
698 * UDP receive path is one user.
700 unsigned long dev_scratch;
703 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
704 struct list_head list;
709 int ip_defrag_offset;
714 u64 skb_mstamp_ns; /* earliest departure time */
717 * This is the control buffer. It is free to use for every
718 * layer. Please put your private variables there. If you
719 * want to keep them across layers you have to do a skb_clone()
720 * first. This is owned by whoever has the skb queued ATM.
722 char cb[48] __aligned(8);
726 unsigned long _skb_refdst;
727 void (*destructor)(struct sk_buff *skb);
729 struct list_head tcp_tsorted_anchor;
732 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
740 /* Following fields are _not_ copied in __copy_skb_header()
741 * Note that queue_mapping is here mostly to fill a hole.
745 /* if you move cloned around you also must adapt those constants */
746 #ifdef __BIG_ENDIAN_BITFIELD
747 #define CLONED_MASK (1 << 7)
749 #define CLONED_MASK 1
751 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
753 __u8 __cloned_offset[0];
760 #ifdef CONFIG_SKB_EXTENSIONS
761 __u8 active_extensions;
763 /* fields enclosed in headers_start/headers_end are copied
764 * using a single memcpy() in __copy_skb_header()
767 __u32 headers_start[0];
770 /* if you move pkt_type around you also must adapt those constants */
771 #ifdef __BIG_ENDIAN_BITFIELD
772 #define PKT_TYPE_MAX (7 << 5)
774 #define PKT_TYPE_MAX 7
776 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
778 __u8 __pkt_type_offset[0];
787 __u8 wifi_acked_valid:1;
790 /* Indicates the inner headers are valid in the skbuff. */
791 __u8 encapsulation:1;
792 __u8 encap_hdr_csum:1;
795 #ifdef __BIG_ENDIAN_BITFIELD
796 #define PKT_VLAN_PRESENT_BIT 7
798 #define PKT_VLAN_PRESENT_BIT 0
800 #define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset)
801 __u8 __pkt_vlan_present_offset[0];
803 __u8 csum_complete_sw:1;
805 __u8 csum_not_inet:1;
806 __u8 dst_pending_confirm:1;
807 #ifdef CONFIG_IPV6_NDISC_NODETYPE
808 __u8 ndisc_nodetype:2;
811 __u8 ipvs_property:1;
812 __u8 inner_protocol_type:1;
813 __u8 remcsum_offload:1;
814 #ifdef CONFIG_NET_SWITCHDEV
815 __u8 offload_fwd_mark:1;
816 __u8 offload_l3_fwd_mark:1;
818 #ifdef CONFIG_NET_CLS_ACT
819 __u8 tc_skip_classify:1;
820 __u8 tc_at_ingress:1;
821 __u8 tc_redirected:1;
822 __u8 tc_from_ingress:1;
824 #ifdef CONFIG_TLS_DEVICE
828 #ifdef CONFIG_NET_SCHED
829 __u16 tc_index; /* traffic control index */
844 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
846 unsigned int napi_id;
847 unsigned int sender_cpu;
850 #ifdef CONFIG_NETWORK_SECMARK
856 __u32 reserved_tailroom;
860 __be16 inner_protocol;
864 __u16 inner_transport_header;
865 __u16 inner_network_header;
866 __u16 inner_mac_header;
869 __u16 transport_header;
870 __u16 network_header;
874 __u32 headers_end[0];
877 /* These elements must be at the end, see alloc_skb() for details. */
882 unsigned int truesize;
885 #ifdef CONFIG_SKB_EXTENSIONS
886 /* only useable after checking ->active_extensions != 0 */
887 struct skb_ext *extensions;
893 * Handling routines are only of interest to the kernel
896 #define SKB_ALLOC_FCLONE 0x01
897 #define SKB_ALLOC_RX 0x02
898 #define SKB_ALLOC_NAPI 0x04
901 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
904 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
906 return unlikely(skb->pfmemalloc);
910 * skb might have a dst pointer attached, refcounted or not.
911 * _skb_refdst low order bit is set if refcount was _not_ taken
913 #define SKB_DST_NOREF 1UL
914 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
917 * skb_dst - returns skb dst_entry
920 * Returns skb dst_entry, regardless of reference taken or not.
922 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
924 /* If refdst was not refcounted, check we still are in a
925 * rcu_read_lock section
927 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
928 !rcu_read_lock_held() &&
929 !rcu_read_lock_bh_held());
930 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
934 * skb_dst_set - sets skb dst
938 * Sets skb dst, assuming a reference was taken on dst and should
939 * be released by skb_dst_drop()
941 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
943 skb->_skb_refdst = (unsigned long)dst;
947 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
951 * Sets skb dst, assuming a reference was not taken on dst.
952 * If dst entry is cached, we do not take reference and dst_release
953 * will be avoided by refdst_drop. If dst entry is not cached, we take
954 * reference, so that last dst_release can destroy the dst immediately.
956 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
958 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
959 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
963 * skb_dst_is_noref - Test if skb dst isn't refcounted
966 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
968 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
972 * skb_rtable - Returns the skb &rtable
975 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
977 return (struct rtable *)skb_dst(skb);
980 /* For mangling skb->pkt_type from user space side from applications
981 * such as nft, tc, etc, we only allow a conservative subset of
982 * possible pkt_types to be set.
984 static inline bool skb_pkt_type_ok(u32 ptype)
986 return ptype <= PACKET_OTHERHOST;
990 * skb_napi_id - Returns the skb's NAPI id
993 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
995 #ifdef CONFIG_NET_RX_BUSY_POLL
1003 * skb_unref - decrement the skb's reference count
1006 * Returns true if we can free the skb.
1008 static inline bool skb_unref(struct sk_buff *skb)
1012 if (likely(refcount_read(&skb->users) == 1))
1014 else if (likely(!refcount_dec_and_test(&skb->users)))
1020 void skb_release_head_state(struct sk_buff *skb);
1021 void kfree_skb(struct sk_buff *skb);
1022 void kfree_skb_list(struct sk_buff *segs);
1023 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1024 void skb_tx_error(struct sk_buff *skb);
1025 void consume_skb(struct sk_buff *skb);
1026 void __consume_stateless_skb(struct sk_buff *skb);
1027 void __kfree_skb(struct sk_buff *skb);
1028 extern struct kmem_cache *skbuff_head_cache;
1030 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1031 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1032 bool *fragstolen, int *delta_truesize);
1034 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1036 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1037 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1038 struct sk_buff *build_skb_around(struct sk_buff *skb,
1039 void *data, unsigned int frag_size);
1042 * alloc_skb - allocate a network buffer
1043 * @size: size to allocate
1044 * @priority: allocation mask
1046 * This function is a convenient wrapper around __alloc_skb().
1048 static inline struct sk_buff *alloc_skb(unsigned int size,
1051 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1054 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1055 unsigned long data_len,
1059 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1061 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1062 struct sk_buff_fclones {
1063 struct sk_buff skb1;
1065 struct sk_buff skb2;
1067 refcount_t fclone_ref;
1071 * skb_fclone_busy - check if fclone is busy
1075 * Returns true if skb is a fast clone, and its clone is not freed.
1076 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1077 * so we also check that this didnt happen.
1079 static inline bool skb_fclone_busy(const struct sock *sk,
1080 const struct sk_buff *skb)
1082 const struct sk_buff_fclones *fclones;
1084 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1086 return skb->fclone == SKB_FCLONE_ORIG &&
1087 refcount_read(&fclones->fclone_ref) > 1 &&
1088 fclones->skb2.sk == sk;
1092 * alloc_skb_fclone - allocate a network buffer from fclone cache
1093 * @size: size to allocate
1094 * @priority: allocation mask
1096 * This function is a convenient wrapper around __alloc_skb().
1098 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1101 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1104 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1105 void skb_headers_offset_update(struct sk_buff *skb, int off);
1106 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1107 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1108 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1109 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1110 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1111 gfp_t gfp_mask, bool fclone);
1112 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1115 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1118 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1119 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1120 unsigned int headroom);
1121 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1122 int newtailroom, gfp_t priority);
1123 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1124 int offset, int len);
1125 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1126 int offset, int len);
1127 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1128 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1131 * skb_pad - zero pad the tail of an skb
1132 * @skb: buffer to pad
1133 * @pad: space to pad
1135 * Ensure that a buffer is followed by a padding area that is zero
1136 * filled. Used by network drivers which may DMA or transfer data
1137 * beyond the buffer end onto the wire.
1139 * May return error in out of memory cases. The skb is freed on error.
1141 static inline int skb_pad(struct sk_buff *skb, int pad)
1143 return __skb_pad(skb, pad, true);
1145 #define dev_kfree_skb(a) consume_skb(a)
1147 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1148 int offset, size_t size);
1150 struct skb_seq_state {
1154 __u32 stepped_offset;
1155 struct sk_buff *root_skb;
1156 struct sk_buff *cur_skb;
1160 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1161 unsigned int to, struct skb_seq_state *st);
1162 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1163 struct skb_seq_state *st);
1164 void skb_abort_seq_read(struct skb_seq_state *st);
1166 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1167 unsigned int to, struct ts_config *config);
1170 * Packet hash types specify the type of hash in skb_set_hash.
1172 * Hash types refer to the protocol layer addresses which are used to
1173 * construct a packet's hash. The hashes are used to differentiate or identify
1174 * flows of the protocol layer for the hash type. Hash types are either
1175 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1177 * Properties of hashes:
1179 * 1) Two packets in different flows have different hash values
1180 * 2) Two packets in the same flow should have the same hash value
1182 * A hash at a higher layer is considered to be more specific. A driver should
1183 * set the most specific hash possible.
1185 * A driver cannot indicate a more specific hash than the layer at which a hash
1186 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1188 * A driver may indicate a hash level which is less specific than the
1189 * actual layer the hash was computed on. For instance, a hash computed
1190 * at L4 may be considered an L3 hash. This should only be done if the
1191 * driver can't unambiguously determine that the HW computed the hash at
1192 * the higher layer. Note that the "should" in the second property above
1195 enum pkt_hash_types {
1196 PKT_HASH_TYPE_NONE, /* Undefined type */
1197 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1198 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1199 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1202 static inline void skb_clear_hash(struct sk_buff *skb)
1209 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1212 skb_clear_hash(skb);
1216 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1218 skb->l4_hash = is_l4;
1219 skb->sw_hash = is_sw;
1224 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1226 /* Used by drivers to set hash from HW */
1227 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1231 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1233 __skb_set_hash(skb, hash, true, is_l4);
1236 void __skb_get_hash(struct sk_buff *skb);
1237 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1238 u32 skb_get_poff(const struct sk_buff *skb);
1239 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1240 const struct flow_keys_basic *keys, int hlen);
1241 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1242 void *data, int hlen_proto);
1244 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1245 int thoff, u8 ip_proto)
1247 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1250 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1251 const struct flow_dissector_key *key,
1252 unsigned int key_count);
1255 int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1256 union bpf_attr __user *uattr);
1257 int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1258 struct bpf_prog *prog);
1260 int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr);
1262 static inline int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1263 union bpf_attr __user *uattr)
1268 static inline int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1269 struct bpf_prog *prog)
1274 static inline int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr)
1280 struct bpf_flow_dissector;
1281 bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1282 __be16 proto, int nhoff, int hlen, unsigned int flags);
1284 bool __skb_flow_dissect(const struct net *net,
1285 const struct sk_buff *skb,
1286 struct flow_dissector *flow_dissector,
1287 void *target_container,
1288 void *data, __be16 proto, int nhoff, int hlen,
1289 unsigned int flags);
1291 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1292 struct flow_dissector *flow_dissector,
1293 void *target_container, unsigned int flags)
1295 return __skb_flow_dissect(NULL, skb, flow_dissector,
1296 target_container, NULL, 0, 0, 0, flags);
1299 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1300 struct flow_keys *flow,
1303 memset(flow, 0, sizeof(*flow));
1304 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1305 flow, NULL, 0, 0, 0, flags);
1309 skb_flow_dissect_flow_keys_basic(const struct net *net,
1310 const struct sk_buff *skb,
1311 struct flow_keys_basic *flow, void *data,
1312 __be16 proto, int nhoff, int hlen,
1315 memset(flow, 0, sizeof(*flow));
1316 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1317 data, proto, nhoff, hlen, flags);
1320 void skb_flow_dissect_meta(const struct sk_buff *skb,
1321 struct flow_dissector *flow_dissector,
1322 void *target_container);
1324 /* Gets a skb connection tracking info, ctinfo map should be a
1325 * a map of mapsize to translate enum ip_conntrack_info states
1329 skb_flow_dissect_ct(const struct sk_buff *skb,
1330 struct flow_dissector *flow_dissector,
1331 void *target_container,
1335 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1336 struct flow_dissector *flow_dissector,
1337 void *target_container);
1339 static inline __u32 skb_get_hash(struct sk_buff *skb)
1341 if (!skb->l4_hash && !skb->sw_hash)
1342 __skb_get_hash(skb);
1347 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1349 if (!skb->l4_hash && !skb->sw_hash) {
1350 struct flow_keys keys;
1351 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1353 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1359 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1360 const siphash_key_t *perturb);
1362 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1367 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1369 to->hash = from->hash;
1370 to->sw_hash = from->sw_hash;
1371 to->l4_hash = from->l4_hash;
1374 static inline void skb_copy_decrypted(struct sk_buff *to,
1375 const struct sk_buff *from)
1377 #ifdef CONFIG_TLS_DEVICE
1378 to->decrypted = from->decrypted;
1382 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1383 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1385 return skb->head + skb->end;
1388 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1393 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1398 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1400 return skb->end - skb->head;
1405 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1407 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1409 return &skb_shinfo(skb)->hwtstamps;
1412 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1414 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1416 return is_zcopy ? skb_uarg(skb) : NULL;
1419 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1422 if (skb && uarg && !skb_zcopy(skb)) {
1423 if (unlikely(have_ref && *have_ref))
1426 sock_zerocopy_get(uarg);
1427 skb_shinfo(skb)->destructor_arg = uarg;
1428 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1432 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1434 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1435 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1438 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1440 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1443 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1445 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1448 /* Release a reference on a zerocopy structure */
1449 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1451 struct ubuf_info *uarg = skb_zcopy(skb);
1454 if (skb_zcopy_is_nouarg(skb)) {
1455 /* no notification callback */
1456 } else if (uarg->callback == sock_zerocopy_callback) {
1457 uarg->zerocopy = uarg->zerocopy && zerocopy;
1458 sock_zerocopy_put(uarg);
1460 uarg->callback(uarg, zerocopy);
1463 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1467 /* Abort a zerocopy operation and revert zckey on error in send syscall */
1468 static inline void skb_zcopy_abort(struct sk_buff *skb)
1470 struct ubuf_info *uarg = skb_zcopy(skb);
1473 sock_zerocopy_put_abort(uarg, false);
1474 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1478 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1483 /* Iterate through singly-linked GSO fragments of an skb. */
1484 #define skb_list_walk_safe(first, skb, next_skb) \
1485 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1486 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1488 static inline void skb_list_del_init(struct sk_buff *skb)
1490 __list_del_entry(&skb->list);
1491 skb_mark_not_on_list(skb);
1495 * skb_queue_empty - check if a queue is empty
1498 * Returns true if the queue is empty, false otherwise.
1500 static inline int skb_queue_empty(const struct sk_buff_head *list)
1502 return list->next == (const struct sk_buff *) list;
1506 * skb_queue_empty_lockless - check if a queue is empty
1509 * Returns true if the queue is empty, false otherwise.
1510 * This variant can be used in lockless contexts.
1512 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1514 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1519 * skb_queue_is_last - check if skb is the last entry in the queue
1523 * Returns true if @skb is the last buffer on the list.
1525 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1526 const struct sk_buff *skb)
1528 return skb->next == (const struct sk_buff *) list;
1532 * skb_queue_is_first - check if skb is the first entry in the queue
1536 * Returns true if @skb is the first buffer on the list.
1538 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1539 const struct sk_buff *skb)
1541 return skb->prev == (const struct sk_buff *) list;
1545 * skb_queue_next - return the next packet in the queue
1547 * @skb: current buffer
1549 * Return the next packet in @list after @skb. It is only valid to
1550 * call this if skb_queue_is_last() evaluates to false.
1552 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1553 const struct sk_buff *skb)
1555 /* This BUG_ON may seem severe, but if we just return then we
1556 * are going to dereference garbage.
1558 BUG_ON(skb_queue_is_last(list, skb));
1563 * skb_queue_prev - return the prev packet in the queue
1565 * @skb: current buffer
1567 * Return the prev packet in @list before @skb. It is only valid to
1568 * call this if skb_queue_is_first() evaluates to false.
1570 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1571 const struct sk_buff *skb)
1573 /* This BUG_ON may seem severe, but if we just return then we
1574 * are going to dereference garbage.
1576 BUG_ON(skb_queue_is_first(list, skb));
1581 * skb_get - reference buffer
1582 * @skb: buffer to reference
1584 * Makes another reference to a socket buffer and returns a pointer
1587 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1589 refcount_inc(&skb->users);
1594 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1598 * skb_cloned - is the buffer a clone
1599 * @skb: buffer to check
1601 * Returns true if the buffer was generated with skb_clone() and is
1602 * one of multiple shared copies of the buffer. Cloned buffers are
1603 * shared data so must not be written to under normal circumstances.
1605 static inline int skb_cloned(const struct sk_buff *skb)
1607 return skb->cloned &&
1608 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1611 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1613 might_sleep_if(gfpflags_allow_blocking(pri));
1615 if (skb_cloned(skb))
1616 return pskb_expand_head(skb, 0, 0, pri);
1622 * skb_header_cloned - is the header a clone
1623 * @skb: buffer to check
1625 * Returns true if modifying the header part of the buffer requires
1626 * the data to be copied.
1628 static inline int skb_header_cloned(const struct sk_buff *skb)
1635 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1636 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1637 return dataref != 1;
1640 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1642 might_sleep_if(gfpflags_allow_blocking(pri));
1644 if (skb_header_cloned(skb))
1645 return pskb_expand_head(skb, 0, 0, pri);
1651 * __skb_header_release - release reference to header
1652 * @skb: buffer to operate on
1654 static inline void __skb_header_release(struct sk_buff *skb)
1657 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1662 * skb_shared - is the buffer shared
1663 * @skb: buffer to check
1665 * Returns true if more than one person has a reference to this
1668 static inline int skb_shared(const struct sk_buff *skb)
1670 return refcount_read(&skb->users) != 1;
1674 * skb_share_check - check if buffer is shared and if so clone it
1675 * @skb: buffer to check
1676 * @pri: priority for memory allocation
1678 * If the buffer is shared the buffer is cloned and the old copy
1679 * drops a reference. A new clone with a single reference is returned.
1680 * If the buffer is not shared the original buffer is returned. When
1681 * being called from interrupt status or with spinlocks held pri must
1684 * NULL is returned on a memory allocation failure.
1686 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1688 might_sleep_if(gfpflags_allow_blocking(pri));
1689 if (skb_shared(skb)) {
1690 struct sk_buff *nskb = skb_clone(skb, pri);
1702 * Copy shared buffers into a new sk_buff. We effectively do COW on
1703 * packets to handle cases where we have a local reader and forward
1704 * and a couple of other messy ones. The normal one is tcpdumping
1705 * a packet thats being forwarded.
1709 * skb_unshare - make a copy of a shared buffer
1710 * @skb: buffer to check
1711 * @pri: priority for memory allocation
1713 * If the socket buffer is a clone then this function creates a new
1714 * copy of the data, drops a reference count on the old copy and returns
1715 * the new copy with the reference count at 1. If the buffer is not a clone
1716 * the original buffer is returned. When called with a spinlock held or
1717 * from interrupt state @pri must be %GFP_ATOMIC
1719 * %NULL is returned on a memory allocation failure.
1721 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1724 might_sleep_if(gfpflags_allow_blocking(pri));
1725 if (skb_cloned(skb)) {
1726 struct sk_buff *nskb = skb_copy(skb, pri);
1728 /* Free our shared copy */
1739 * skb_peek - peek at the head of an &sk_buff_head
1740 * @list_: list to peek at
1742 * Peek an &sk_buff. Unlike most other operations you _MUST_
1743 * be careful with this one. A peek leaves the buffer on the
1744 * list and someone else may run off with it. You must hold
1745 * the appropriate locks or have a private queue to do this.
1747 * Returns %NULL for an empty list or a pointer to the head element.
1748 * The reference count is not incremented and the reference is therefore
1749 * volatile. Use with caution.
1751 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1753 struct sk_buff *skb = list_->next;
1755 if (skb == (struct sk_buff *)list_)
1761 * __skb_peek - peek at the head of a non-empty &sk_buff_head
1762 * @list_: list to peek at
1764 * Like skb_peek(), but the caller knows that the list is not empty.
1766 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
1772 * skb_peek_next - peek skb following the given one from a queue
1773 * @skb: skb to start from
1774 * @list_: list to peek at
1776 * Returns %NULL when the end of the list is met or a pointer to the
1777 * next element. The reference count is not incremented and the
1778 * reference is therefore volatile. Use with caution.
1780 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1781 const struct sk_buff_head *list_)
1783 struct sk_buff *next = skb->next;
1785 if (next == (struct sk_buff *)list_)
1791 * skb_peek_tail - peek at the tail of an &sk_buff_head
1792 * @list_: list to peek at
1794 * Peek an &sk_buff. Unlike most other operations you _MUST_
1795 * be careful with this one. A peek leaves the buffer on the
1796 * list and someone else may run off with it. You must hold
1797 * the appropriate locks or have a private queue to do this.
1799 * Returns %NULL for an empty list or a pointer to the tail element.
1800 * The reference count is not incremented and the reference is therefore
1801 * volatile. Use with caution.
1803 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1805 struct sk_buff *skb = READ_ONCE(list_->prev);
1807 if (skb == (struct sk_buff *)list_)
1814 * skb_queue_len - get queue length
1815 * @list_: list to measure
1817 * Return the length of an &sk_buff queue.
1819 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1825 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1826 * @list: queue to initialize
1828 * This initializes only the list and queue length aspects of
1829 * an sk_buff_head object. This allows to initialize the list
1830 * aspects of an sk_buff_head without reinitializing things like
1831 * the spinlock. It can also be used for on-stack sk_buff_head
1832 * objects where the spinlock is known to not be used.
1834 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1836 list->prev = list->next = (struct sk_buff *)list;
1841 * This function creates a split out lock class for each invocation;
1842 * this is needed for now since a whole lot of users of the skb-queue
1843 * infrastructure in drivers have different locking usage (in hardirq)
1844 * than the networking core (in softirq only). In the long run either the
1845 * network layer or drivers should need annotation to consolidate the
1846 * main types of usage into 3 classes.
1848 static inline void skb_queue_head_init(struct sk_buff_head *list)
1850 spin_lock_init(&list->lock);
1851 __skb_queue_head_init(list);
1854 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1855 struct lock_class_key *class)
1857 skb_queue_head_init(list);
1858 lockdep_set_class(&list->lock, class);
1862 * Insert an sk_buff on a list.
1864 * The "__skb_xxxx()" functions are the non-atomic ones that
1865 * can only be called with interrupts disabled.
1867 static inline void __skb_insert(struct sk_buff *newsk,
1868 struct sk_buff *prev, struct sk_buff *next,
1869 struct sk_buff_head *list)
1871 /* See skb_queue_empty_lockless() and skb_peek_tail()
1872 * for the opposite READ_ONCE()
1874 WRITE_ONCE(newsk->next, next);
1875 WRITE_ONCE(newsk->prev, prev);
1876 WRITE_ONCE(next->prev, newsk);
1877 WRITE_ONCE(prev->next, newsk);
1881 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1882 struct sk_buff *prev,
1883 struct sk_buff *next)
1885 struct sk_buff *first = list->next;
1886 struct sk_buff *last = list->prev;
1888 WRITE_ONCE(first->prev, prev);
1889 WRITE_ONCE(prev->next, first);
1891 WRITE_ONCE(last->next, next);
1892 WRITE_ONCE(next->prev, last);
1896 * skb_queue_splice - join two skb lists, this is designed for stacks
1897 * @list: the new list to add
1898 * @head: the place to add it in the first list
1900 static inline void skb_queue_splice(const struct sk_buff_head *list,
1901 struct sk_buff_head *head)
1903 if (!skb_queue_empty(list)) {
1904 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1905 head->qlen += list->qlen;
1910 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1911 * @list: the new list to add
1912 * @head: the place to add it in the first list
1914 * The list at @list is reinitialised
1916 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1917 struct sk_buff_head *head)
1919 if (!skb_queue_empty(list)) {
1920 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1921 head->qlen += list->qlen;
1922 __skb_queue_head_init(list);
1927 * skb_queue_splice_tail - join two skb lists, each list being a queue
1928 * @list: the new list to add
1929 * @head: the place to add it in the first list
1931 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1932 struct sk_buff_head *head)
1934 if (!skb_queue_empty(list)) {
1935 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1936 head->qlen += list->qlen;
1941 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1942 * @list: the new list to add
1943 * @head: the place to add it in the first list
1945 * Each of the lists is a queue.
1946 * The list at @list is reinitialised
1948 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1949 struct sk_buff_head *head)
1951 if (!skb_queue_empty(list)) {
1952 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1953 head->qlen += list->qlen;
1954 __skb_queue_head_init(list);
1959 * __skb_queue_after - queue a buffer at the list head
1960 * @list: list to use
1961 * @prev: place after this buffer
1962 * @newsk: buffer to queue
1964 * Queue a buffer int the middle of a list. This function takes no locks
1965 * and you must therefore hold required locks before calling it.
1967 * A buffer cannot be placed on two lists at the same time.
1969 static inline void __skb_queue_after(struct sk_buff_head *list,
1970 struct sk_buff *prev,
1971 struct sk_buff *newsk)
1973 __skb_insert(newsk, prev, prev->next, list);
1976 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1977 struct sk_buff_head *list);
1979 static inline void __skb_queue_before(struct sk_buff_head *list,
1980 struct sk_buff *next,
1981 struct sk_buff *newsk)
1983 __skb_insert(newsk, next->prev, next, list);
1987 * __skb_queue_head - queue a buffer at the list head
1988 * @list: list to use
1989 * @newsk: buffer to queue
1991 * Queue a buffer at the start of a list. This function takes no locks
1992 * and you must therefore hold required locks before calling it.
1994 * A buffer cannot be placed on two lists at the same time.
1996 static inline void __skb_queue_head(struct sk_buff_head *list,
1997 struct sk_buff *newsk)
1999 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2001 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2004 * __skb_queue_tail - queue a buffer at the list tail
2005 * @list: list to use
2006 * @newsk: buffer to queue
2008 * Queue a buffer at the end of a list. This function takes no locks
2009 * and you must therefore hold required locks before calling it.
2011 * A buffer cannot be placed on two lists at the same time.
2013 static inline void __skb_queue_tail(struct sk_buff_head *list,
2014 struct sk_buff *newsk)
2016 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2018 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2021 * remove sk_buff from list. _Must_ be called atomically, and with
2024 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2025 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2027 struct sk_buff *next, *prev;
2032 skb->next = skb->prev = NULL;
2033 WRITE_ONCE(next->prev, prev);
2034 WRITE_ONCE(prev->next, next);
2038 * __skb_dequeue - remove from the head of the queue
2039 * @list: list to dequeue from
2041 * Remove the head of the list. This function does not take any locks
2042 * so must be used with appropriate locks held only. The head item is
2043 * returned or %NULL if the list is empty.
2045 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2047 struct sk_buff *skb = skb_peek(list);
2049 __skb_unlink(skb, list);
2052 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2055 * __skb_dequeue_tail - remove from the tail of the queue
2056 * @list: list to dequeue from
2058 * Remove the tail of the list. This function does not take any locks
2059 * so must be used with appropriate locks held only. The tail item is
2060 * returned or %NULL if the list is empty.
2062 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2064 struct sk_buff *skb = skb_peek_tail(list);
2066 __skb_unlink(skb, list);
2069 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2072 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2074 return skb->data_len;
2077 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2079 return skb->len - skb->data_len;
2082 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2084 unsigned int i, len = 0;
2086 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2087 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2091 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2093 return skb_headlen(skb) + __skb_pagelen(skb);
2097 * __skb_fill_page_desc - initialise a paged fragment in an skb
2098 * @skb: buffer containing fragment to be initialised
2099 * @i: paged fragment index to initialise
2100 * @page: the page to use for this fragment
2101 * @off: the offset to the data with @page
2102 * @size: the length of the data
2104 * Initialises the @i'th fragment of @skb to point to &size bytes at
2105 * offset @off within @page.
2107 * Does not take any additional reference on the fragment.
2109 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2110 struct page *page, int off, int size)
2112 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2115 * Propagate page pfmemalloc to the skb if we can. The problem is
2116 * that not all callers have unique ownership of the page but rely
2117 * on page_is_pfmemalloc doing the right thing(tm).
2119 frag->bv_page = page;
2120 frag->bv_offset = off;
2121 skb_frag_size_set(frag, size);
2123 page = compound_head(page);
2124 if (page_is_pfmemalloc(page))
2125 skb->pfmemalloc = true;
2129 * skb_fill_page_desc - initialise a paged fragment in an skb
2130 * @skb: buffer containing fragment to be initialised
2131 * @i: paged fragment index to initialise
2132 * @page: the page to use for this fragment
2133 * @off: the offset to the data with @page
2134 * @size: the length of the data
2136 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2137 * @skb to point to @size bytes at offset @off within @page. In
2138 * addition updates @skb such that @i is the last fragment.
2140 * Does not take any additional reference on the fragment.
2142 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2143 struct page *page, int off, int size)
2145 __skb_fill_page_desc(skb, i, page, off, size);
2146 skb_shinfo(skb)->nr_frags = i + 1;
2149 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2150 int size, unsigned int truesize);
2152 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2153 unsigned int truesize);
2155 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2157 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2158 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2160 return skb->head + skb->tail;
2163 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2165 skb->tail = skb->data - skb->head;
2168 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2170 skb_reset_tail_pointer(skb);
2171 skb->tail += offset;
2174 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2175 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2180 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2182 skb->tail = skb->data;
2185 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2187 skb->tail = skb->data + offset;
2190 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2193 * Add data to an sk_buff
2195 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2196 void *skb_put(struct sk_buff *skb, unsigned int len);
2197 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2199 void *tmp = skb_tail_pointer(skb);
2200 SKB_LINEAR_ASSERT(skb);
2206 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2208 void *tmp = __skb_put(skb, len);
2210 memset(tmp, 0, len);
2214 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2217 void *tmp = __skb_put(skb, len);
2219 memcpy(tmp, data, len);
2223 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2225 *(u8 *)__skb_put(skb, 1) = val;
2228 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2230 void *tmp = skb_put(skb, len);
2232 memset(tmp, 0, len);
2237 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2240 void *tmp = skb_put(skb, len);
2242 memcpy(tmp, data, len);
2247 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2249 *(u8 *)skb_put(skb, 1) = val;
2252 void *skb_push(struct sk_buff *skb, unsigned int len);
2253 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2260 void *skb_pull(struct sk_buff *skb, unsigned int len);
2261 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2264 BUG_ON(skb->len < skb->data_len);
2265 return skb->data += len;
2268 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2270 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2273 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2275 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2277 if (len > skb_headlen(skb) &&
2278 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2281 return skb->data += len;
2284 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2286 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2289 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2291 if (likely(len <= skb_headlen(skb)))
2293 if (unlikely(len > skb->len))
2295 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2298 void skb_condense(struct sk_buff *skb);
2301 * skb_headroom - bytes at buffer head
2302 * @skb: buffer to check
2304 * Return the number of bytes of free space at the head of an &sk_buff.
2306 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2308 return skb->data - skb->head;
2312 * skb_tailroom - bytes at buffer end
2313 * @skb: buffer to check
2315 * Return the number of bytes of free space at the tail of an sk_buff
2317 static inline int skb_tailroom(const struct sk_buff *skb)
2319 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2323 * skb_availroom - bytes at buffer end
2324 * @skb: buffer to check
2326 * Return the number of bytes of free space at the tail of an sk_buff
2327 * allocated by sk_stream_alloc()
2329 static inline int skb_availroom(const struct sk_buff *skb)
2331 if (skb_is_nonlinear(skb))
2334 return skb->end - skb->tail - skb->reserved_tailroom;
2338 * skb_reserve - adjust headroom
2339 * @skb: buffer to alter
2340 * @len: bytes to move
2342 * Increase the headroom of an empty &sk_buff by reducing the tail
2343 * room. This is only allowed for an empty buffer.
2345 static inline void skb_reserve(struct sk_buff *skb, int len)
2352 * skb_tailroom_reserve - adjust reserved_tailroom
2353 * @skb: buffer to alter
2354 * @mtu: maximum amount of headlen permitted
2355 * @needed_tailroom: minimum amount of reserved_tailroom
2357 * Set reserved_tailroom so that headlen can be as large as possible but
2358 * not larger than mtu and tailroom cannot be smaller than
2360 * The required headroom should already have been reserved before using
2363 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2364 unsigned int needed_tailroom)
2366 SKB_LINEAR_ASSERT(skb);
2367 if (mtu < skb_tailroom(skb) - needed_tailroom)
2368 /* use at most mtu */
2369 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2371 /* use up to all available space */
2372 skb->reserved_tailroom = needed_tailroom;
2375 #define ENCAP_TYPE_ETHER 0
2376 #define ENCAP_TYPE_IPPROTO 1
2378 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2381 skb->inner_protocol = protocol;
2382 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2385 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2388 skb->inner_ipproto = ipproto;
2389 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2392 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2394 skb->inner_mac_header = skb->mac_header;
2395 skb->inner_network_header = skb->network_header;
2396 skb->inner_transport_header = skb->transport_header;
2399 static inline void skb_reset_mac_len(struct sk_buff *skb)
2401 skb->mac_len = skb->network_header - skb->mac_header;
2404 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2407 return skb->head + skb->inner_transport_header;
2410 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2412 return skb_inner_transport_header(skb) - skb->data;
2415 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2417 skb->inner_transport_header = skb->data - skb->head;
2420 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2423 skb_reset_inner_transport_header(skb);
2424 skb->inner_transport_header += offset;
2427 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2429 return skb->head + skb->inner_network_header;
2432 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2434 skb->inner_network_header = skb->data - skb->head;
2437 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2440 skb_reset_inner_network_header(skb);
2441 skb->inner_network_header += offset;
2444 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2446 return skb->head + skb->inner_mac_header;
2449 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2451 skb->inner_mac_header = skb->data - skb->head;
2454 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2457 skb_reset_inner_mac_header(skb);
2458 skb->inner_mac_header += offset;
2460 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2462 return skb->transport_header != (typeof(skb->transport_header))~0U;
2465 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2467 return skb->head + skb->transport_header;
2470 static inline void skb_reset_transport_header(struct sk_buff *skb)
2472 skb->transport_header = skb->data - skb->head;
2475 static inline void skb_set_transport_header(struct sk_buff *skb,
2478 skb_reset_transport_header(skb);
2479 skb->transport_header += offset;
2482 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2484 return skb->head + skb->network_header;
2487 static inline void skb_reset_network_header(struct sk_buff *skb)
2489 skb->network_header = skb->data - skb->head;
2492 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2494 skb_reset_network_header(skb);
2495 skb->network_header += offset;
2498 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2500 return skb->head + skb->mac_header;
2503 static inline int skb_mac_offset(const struct sk_buff *skb)
2505 return skb_mac_header(skb) - skb->data;
2508 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2510 return skb->network_header - skb->mac_header;
2513 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2515 return skb->mac_header != (typeof(skb->mac_header))~0U;
2518 static inline void skb_reset_mac_header(struct sk_buff *skb)
2520 skb->mac_header = skb->data - skb->head;
2523 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2525 skb_reset_mac_header(skb);
2526 skb->mac_header += offset;
2529 static inline void skb_pop_mac_header(struct sk_buff *skb)
2531 skb->mac_header = skb->network_header;
2534 static inline void skb_probe_transport_header(struct sk_buff *skb)
2536 struct flow_keys_basic keys;
2538 if (skb_transport_header_was_set(skb))
2541 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2543 skb_set_transport_header(skb, keys.control.thoff);
2546 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2548 if (skb_mac_header_was_set(skb)) {
2549 const unsigned char *old_mac = skb_mac_header(skb);
2551 skb_set_mac_header(skb, -skb->mac_len);
2552 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2556 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2558 return skb->csum_start - skb_headroom(skb);
2561 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2563 return skb->head + skb->csum_start;
2566 static inline int skb_transport_offset(const struct sk_buff *skb)
2568 return skb_transport_header(skb) - skb->data;
2571 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2573 return skb->transport_header - skb->network_header;
2576 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2578 return skb->inner_transport_header - skb->inner_network_header;
2581 static inline int skb_network_offset(const struct sk_buff *skb)
2583 return skb_network_header(skb) - skb->data;
2586 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2588 return skb_inner_network_header(skb) - skb->data;
2591 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2593 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2597 * CPUs often take a performance hit when accessing unaligned memory
2598 * locations. The actual performance hit varies, it can be small if the
2599 * hardware handles it or large if we have to take an exception and fix it
2602 * Since an ethernet header is 14 bytes network drivers often end up with
2603 * the IP header at an unaligned offset. The IP header can be aligned by
2604 * shifting the start of the packet by 2 bytes. Drivers should do this
2607 * skb_reserve(skb, NET_IP_ALIGN);
2609 * The downside to this alignment of the IP header is that the DMA is now
2610 * unaligned. On some architectures the cost of an unaligned DMA is high
2611 * and this cost outweighs the gains made by aligning the IP header.
2613 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2616 #ifndef NET_IP_ALIGN
2617 #define NET_IP_ALIGN 2
2621 * The networking layer reserves some headroom in skb data (via
2622 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2623 * the header has to grow. In the default case, if the header has to grow
2624 * 32 bytes or less we avoid the reallocation.
2626 * Unfortunately this headroom changes the DMA alignment of the resulting
2627 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2628 * on some architectures. An architecture can override this value,
2629 * perhaps setting it to a cacheline in size (since that will maintain
2630 * cacheline alignment of the DMA). It must be a power of 2.
2632 * Various parts of the networking layer expect at least 32 bytes of
2633 * headroom, you should not reduce this.
2635 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2636 * to reduce average number of cache lines per packet.
2637 * get_rps_cpus() for example only access one 64 bytes aligned block :
2638 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2641 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2644 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2646 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2648 if (WARN_ON(skb_is_nonlinear(skb)))
2651 skb_set_tail_pointer(skb, len);
2654 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2656 __skb_set_length(skb, len);
2659 void skb_trim(struct sk_buff *skb, unsigned int len);
2661 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2664 return ___pskb_trim(skb, len);
2665 __skb_trim(skb, len);
2669 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2671 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2675 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2676 * @skb: buffer to alter
2679 * This is identical to pskb_trim except that the caller knows that
2680 * the skb is not cloned so we should never get an error due to out-
2683 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2685 int err = pskb_trim(skb, len);
2689 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2691 unsigned int diff = len - skb->len;
2693 if (skb_tailroom(skb) < diff) {
2694 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2699 __skb_set_length(skb, len);
2704 * skb_orphan - orphan a buffer
2705 * @skb: buffer to orphan
2707 * If a buffer currently has an owner then we call the owner's
2708 * destructor function and make the @skb unowned. The buffer continues
2709 * to exist but is no longer charged to its former owner.
2711 static inline void skb_orphan(struct sk_buff *skb)
2713 if (skb->destructor) {
2714 skb->destructor(skb);
2715 skb->destructor = NULL;
2723 * skb_orphan_frags - orphan the frags contained in a buffer
2724 * @skb: buffer to orphan frags from
2725 * @gfp_mask: allocation mask for replacement pages
2727 * For each frag in the SKB which needs a destructor (i.e. has an
2728 * owner) create a copy of that frag and release the original
2729 * page by calling the destructor.
2731 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2733 if (likely(!skb_zcopy(skb)))
2735 if (!skb_zcopy_is_nouarg(skb) &&
2736 skb_uarg(skb)->callback == sock_zerocopy_callback)
2738 return skb_copy_ubufs(skb, gfp_mask);
2741 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2742 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2744 if (likely(!skb_zcopy(skb)))
2746 return skb_copy_ubufs(skb, gfp_mask);
2750 * __skb_queue_purge - empty a list
2751 * @list: list to empty
2753 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2754 * the list and one reference dropped. This function does not take the
2755 * list lock and the caller must hold the relevant locks to use it.
2757 static inline void __skb_queue_purge(struct sk_buff_head *list)
2759 struct sk_buff *skb;
2760 while ((skb = __skb_dequeue(list)) != NULL)
2763 void skb_queue_purge(struct sk_buff_head *list);
2765 unsigned int skb_rbtree_purge(struct rb_root *root);
2767 void *netdev_alloc_frag(unsigned int fragsz);
2769 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2773 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2774 * @dev: network device to receive on
2775 * @length: length to allocate
2777 * Allocate a new &sk_buff and assign it a usage count of one. The
2778 * buffer has unspecified headroom built in. Users should allocate
2779 * the headroom they think they need without accounting for the
2780 * built in space. The built in space is used for optimisations.
2782 * %NULL is returned if there is no free memory. Although this function
2783 * allocates memory it can be called from an interrupt.
2785 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2786 unsigned int length)
2788 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2791 /* legacy helper around __netdev_alloc_skb() */
2792 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2795 return __netdev_alloc_skb(NULL, length, gfp_mask);
2798 /* legacy helper around netdev_alloc_skb() */
2799 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2801 return netdev_alloc_skb(NULL, length);
2805 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2806 unsigned int length, gfp_t gfp)
2808 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2810 if (NET_IP_ALIGN && skb)
2811 skb_reserve(skb, NET_IP_ALIGN);
2815 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2816 unsigned int length)
2818 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2821 static inline void skb_free_frag(void *addr)
2823 page_frag_free(addr);
2826 void *napi_alloc_frag(unsigned int fragsz);
2827 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2828 unsigned int length, gfp_t gfp_mask);
2829 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2830 unsigned int length)
2832 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2834 void napi_consume_skb(struct sk_buff *skb, int budget);
2836 void __kfree_skb_flush(void);
2837 void __kfree_skb_defer(struct sk_buff *skb);
2840 * __dev_alloc_pages - allocate page for network Rx
2841 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2842 * @order: size of the allocation
2844 * Allocate a new page.
2846 * %NULL is returned if there is no free memory.
2848 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2851 /* This piece of code contains several assumptions.
2852 * 1. This is for device Rx, therefor a cold page is preferred.
2853 * 2. The expectation is the user wants a compound page.
2854 * 3. If requesting a order 0 page it will not be compound
2855 * due to the check to see if order has a value in prep_new_page
2856 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2857 * code in gfp_to_alloc_flags that should be enforcing this.
2859 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2861 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2864 static inline struct page *dev_alloc_pages(unsigned int order)
2866 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2870 * __dev_alloc_page - allocate a page for network Rx
2871 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2873 * Allocate a new page.
2875 * %NULL is returned if there is no free memory.
2877 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2879 return __dev_alloc_pages(gfp_mask, 0);
2882 static inline struct page *dev_alloc_page(void)
2884 return dev_alloc_pages(0);
2888 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2889 * @page: The page that was allocated from skb_alloc_page
2890 * @skb: The skb that may need pfmemalloc set
2892 static inline void skb_propagate_pfmemalloc(struct page *page,
2893 struct sk_buff *skb)
2895 if (page_is_pfmemalloc(page))
2896 skb->pfmemalloc = true;
2900 * skb_frag_off() - Returns the offset of a skb fragment
2901 * @frag: the paged fragment
2903 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
2905 return frag->bv_offset;
2909 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
2910 * @frag: skb fragment
2911 * @delta: value to add
2913 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
2915 frag->bv_offset += delta;
2919 * skb_frag_off_set() - Sets the offset of a skb fragment
2920 * @frag: skb fragment
2921 * @offset: offset of fragment
2923 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
2925 frag->bv_offset = offset;
2929 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
2930 * @fragto: skb fragment where offset is set
2931 * @fragfrom: skb fragment offset is copied from
2933 static inline void skb_frag_off_copy(skb_frag_t *fragto,
2934 const skb_frag_t *fragfrom)
2936 fragto->bv_offset = fragfrom->bv_offset;
2940 * skb_frag_page - retrieve the page referred to by a paged fragment
2941 * @frag: the paged fragment
2943 * Returns the &struct page associated with @frag.
2945 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2947 return frag->bv_page;
2951 * __skb_frag_ref - take an addition reference on a paged fragment.
2952 * @frag: the paged fragment
2954 * Takes an additional reference on the paged fragment @frag.
2956 static inline void __skb_frag_ref(skb_frag_t *frag)
2958 get_page(skb_frag_page(frag));
2962 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2964 * @f: the fragment offset.
2966 * Takes an additional reference on the @f'th paged fragment of @skb.
2968 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2970 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2974 * __skb_frag_unref - release a reference on a paged fragment.
2975 * @frag: the paged fragment
2977 * Releases a reference on the paged fragment @frag.
2979 static inline void __skb_frag_unref(skb_frag_t *frag)
2981 put_page(skb_frag_page(frag));
2985 * skb_frag_unref - release a reference on a paged fragment of an skb.
2987 * @f: the fragment offset
2989 * Releases a reference on the @f'th paged fragment of @skb.
2991 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2993 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2997 * skb_frag_address - gets the address of the data contained in a paged fragment
2998 * @frag: the paged fragment buffer
3000 * Returns the address of the data within @frag. The page must already
3003 static inline void *skb_frag_address(const skb_frag_t *frag)
3005 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3009 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3010 * @frag: the paged fragment buffer
3012 * Returns the address of the data within @frag. Checks that the page
3013 * is mapped and returns %NULL otherwise.
3015 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3017 void *ptr = page_address(skb_frag_page(frag));
3021 return ptr + skb_frag_off(frag);
3025 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3026 * @fragto: skb fragment where page is set
3027 * @fragfrom: skb fragment page is copied from
3029 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3030 const skb_frag_t *fragfrom)
3032 fragto->bv_page = fragfrom->bv_page;
3036 * __skb_frag_set_page - sets the page contained in a paged fragment
3037 * @frag: the paged fragment
3038 * @page: the page to set
3040 * Sets the fragment @frag to contain @page.
3042 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3044 frag->bv_page = page;
3048 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3050 * @f: the fragment offset
3051 * @page: the page to set
3053 * Sets the @f'th fragment of @skb to contain @page.
3055 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3058 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3061 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3064 * skb_frag_dma_map - maps a paged fragment via the DMA API
3065 * @dev: the device to map the fragment to
3066 * @frag: the paged fragment to map
3067 * @offset: the offset within the fragment (starting at the
3068 * fragment's own offset)
3069 * @size: the number of bytes to map
3070 * @dir: the direction of the mapping (``PCI_DMA_*``)
3072 * Maps the page associated with @frag to @device.
3074 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3075 const skb_frag_t *frag,
3076 size_t offset, size_t size,
3077 enum dma_data_direction dir)
3079 return dma_map_page(dev, skb_frag_page(frag),
3080 skb_frag_off(frag) + offset, size, dir);
3083 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3086 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3090 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3093 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3098 * skb_clone_writable - is the header of a clone writable
3099 * @skb: buffer to check
3100 * @len: length up to which to write
3102 * Returns true if modifying the header part of the cloned buffer
3103 * does not requires the data to be copied.
3105 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3107 return !skb_header_cloned(skb) &&
3108 skb_headroom(skb) + len <= skb->hdr_len;
3111 static inline int skb_try_make_writable(struct sk_buff *skb,
3112 unsigned int write_len)
3114 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3115 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3118 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3123 if (headroom > skb_headroom(skb))
3124 delta = headroom - skb_headroom(skb);
3126 if (delta || cloned)
3127 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3133 * skb_cow - copy header of skb when it is required
3134 * @skb: buffer to cow
3135 * @headroom: needed headroom
3137 * If the skb passed lacks sufficient headroom or its data part
3138 * is shared, data is reallocated. If reallocation fails, an error
3139 * is returned and original skb is not changed.
3141 * The result is skb with writable area skb->head...skb->tail
3142 * and at least @headroom of space at head.
3144 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3146 return __skb_cow(skb, headroom, skb_cloned(skb));
3150 * skb_cow_head - skb_cow but only making the head writable
3151 * @skb: buffer to cow
3152 * @headroom: needed headroom
3154 * This function is identical to skb_cow except that we replace the
3155 * skb_cloned check by skb_header_cloned. It should be used when
3156 * you only need to push on some header and do not need to modify
3159 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3161 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3165 * skb_padto - pad an skbuff up to a minimal size
3166 * @skb: buffer to pad
3167 * @len: minimal length
3169 * Pads up a buffer to ensure the trailing bytes exist and are
3170 * blanked. If the buffer already contains sufficient data it
3171 * is untouched. Otherwise it is extended. Returns zero on
3172 * success. The skb is freed on error.
3174 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3176 unsigned int size = skb->len;
3177 if (likely(size >= len))
3179 return skb_pad(skb, len - size);
3183 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3184 * @skb: buffer to pad
3185 * @len: minimal length
3186 * @free_on_error: free buffer on error
3188 * Pads up a buffer to ensure the trailing bytes exist and are
3189 * blanked. If the buffer already contains sufficient data it
3190 * is untouched. Otherwise it is extended. Returns zero on
3191 * success. The skb is freed on error if @free_on_error is true.
3193 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
3196 unsigned int size = skb->len;
3198 if (unlikely(size < len)) {
3200 if (__skb_pad(skb, len, free_on_error))
3202 __skb_put(skb, len);
3208 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3209 * @skb: buffer to pad
3210 * @len: minimal length
3212 * Pads up a buffer to ensure the trailing bytes exist and are
3213 * blanked. If the buffer already contains sufficient data it
3214 * is untouched. Otherwise it is extended. Returns zero on
3215 * success. The skb is freed on error.
3217 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
3219 return __skb_put_padto(skb, len, true);
3222 static inline int skb_add_data(struct sk_buff *skb,
3223 struct iov_iter *from, int copy)
3225 const int off = skb->len;
3227 if (skb->ip_summed == CHECKSUM_NONE) {
3229 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3231 skb->csum = csum_block_add(skb->csum, csum, off);
3234 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3237 __skb_trim(skb, off);
3241 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3242 const struct page *page, int off)
3247 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3249 return page == skb_frag_page(frag) &&
3250 off == skb_frag_off(frag) + skb_frag_size(frag);
3255 static inline int __skb_linearize(struct sk_buff *skb)
3257 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3261 * skb_linearize - convert paged skb to linear one
3262 * @skb: buffer to linarize
3264 * If there is no free memory -ENOMEM is returned, otherwise zero
3265 * is returned and the old skb data released.
3267 static inline int skb_linearize(struct sk_buff *skb)
3269 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3273 * skb_has_shared_frag - can any frag be overwritten
3274 * @skb: buffer to test
3276 * Return true if the skb has at least one frag that might be modified
3277 * by an external entity (as in vmsplice()/sendfile())
3279 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3281 return skb_is_nonlinear(skb) &&
3282 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3286 * skb_linearize_cow - make sure skb is linear and writable
3287 * @skb: buffer to process
3289 * If there is no free memory -ENOMEM is returned, otherwise zero
3290 * is returned and the old skb data released.
3292 static inline int skb_linearize_cow(struct sk_buff *skb)
3294 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3295 __skb_linearize(skb) : 0;
3298 static __always_inline void
3299 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3302 if (skb->ip_summed == CHECKSUM_COMPLETE)
3303 skb->csum = csum_block_sub(skb->csum,
3304 csum_partial(start, len, 0), off);
3305 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3306 skb_checksum_start_offset(skb) < 0)
3307 skb->ip_summed = CHECKSUM_NONE;
3311 * skb_postpull_rcsum - update checksum for received skb after pull
3312 * @skb: buffer to update
3313 * @start: start of data before pull
3314 * @len: length of data pulled
3316 * After doing a pull on a received packet, you need to call this to
3317 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3318 * CHECKSUM_NONE so that it can be recomputed from scratch.
3320 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3321 const void *start, unsigned int len)
3323 __skb_postpull_rcsum(skb, start, len, 0);
3326 static __always_inline void
3327 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3330 if (skb->ip_summed == CHECKSUM_COMPLETE)
3331 skb->csum = csum_block_add(skb->csum,
3332 csum_partial(start, len, 0), off);
3336 * skb_postpush_rcsum - update checksum for received skb after push
3337 * @skb: buffer to update
3338 * @start: start of data after push
3339 * @len: length of data pushed
3341 * After doing a push on a received packet, you need to call this to
3342 * update the CHECKSUM_COMPLETE checksum.
3344 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3345 const void *start, unsigned int len)
3347 __skb_postpush_rcsum(skb, start, len, 0);
3350 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3353 * skb_push_rcsum - push skb and update receive checksum
3354 * @skb: buffer to update
3355 * @len: length of data pulled
3357 * This function performs an skb_push on the packet and updates
3358 * the CHECKSUM_COMPLETE checksum. It should be used on
3359 * receive path processing instead of skb_push unless you know
3360 * that the checksum difference is zero (e.g., a valid IP header)
3361 * or you are setting ip_summed to CHECKSUM_NONE.
3363 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3366 skb_postpush_rcsum(skb, skb->data, len);
3370 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3372 * pskb_trim_rcsum - trim received skb and update checksum
3373 * @skb: buffer to trim
3376 * This is exactly the same as pskb_trim except that it ensures the
3377 * checksum of received packets are still valid after the operation.
3378 * It can change skb pointers.
3381 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3383 if (likely(len >= skb->len))
3385 return pskb_trim_rcsum_slow(skb, len);
3388 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3390 if (skb->ip_summed == CHECKSUM_COMPLETE)
3391 skb->ip_summed = CHECKSUM_NONE;
3392 __skb_trim(skb, len);
3396 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3398 if (skb->ip_summed == CHECKSUM_COMPLETE)
3399 skb->ip_summed = CHECKSUM_NONE;
3400 return __skb_grow(skb, len);
3403 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3404 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3405 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3406 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3407 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3409 #define skb_queue_walk(queue, skb) \
3410 for (skb = (queue)->next; \
3411 skb != (struct sk_buff *)(queue); \
3414 #define skb_queue_walk_safe(queue, skb, tmp) \
3415 for (skb = (queue)->next, tmp = skb->next; \
3416 skb != (struct sk_buff *)(queue); \
3417 skb = tmp, tmp = skb->next)
3419 #define skb_queue_walk_from(queue, skb) \
3420 for (; skb != (struct sk_buff *)(queue); \
3423 #define skb_rbtree_walk(skb, root) \
3424 for (skb = skb_rb_first(root); skb != NULL; \
3425 skb = skb_rb_next(skb))
3427 #define skb_rbtree_walk_from(skb) \
3428 for (; skb != NULL; \
3429 skb = skb_rb_next(skb))
3431 #define skb_rbtree_walk_from_safe(skb, tmp) \
3432 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3435 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3436 for (tmp = skb->next; \
3437 skb != (struct sk_buff *)(queue); \
3438 skb = tmp, tmp = skb->next)
3440 #define skb_queue_reverse_walk(queue, skb) \
3441 for (skb = (queue)->prev; \
3442 skb != (struct sk_buff *)(queue); \
3445 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3446 for (skb = (queue)->prev, tmp = skb->prev; \
3447 skb != (struct sk_buff *)(queue); \
3448 skb = tmp, tmp = skb->prev)
3450 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3451 for (tmp = skb->prev; \
3452 skb != (struct sk_buff *)(queue); \
3453 skb = tmp, tmp = skb->prev)
3455 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3457 return skb_shinfo(skb)->frag_list != NULL;
3460 static inline void skb_frag_list_init(struct sk_buff *skb)
3462 skb_shinfo(skb)->frag_list = NULL;
3465 #define skb_walk_frags(skb, iter) \
3466 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3469 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3470 int *err, long *timeo_p,
3471 const struct sk_buff *skb);
3472 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3473 struct sk_buff_head *queue,
3475 void (*destructor)(struct sock *sk,
3476 struct sk_buff *skb),
3478 struct sk_buff **last);
3479 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
3480 struct sk_buff_head *queue,
3482 void (*destructor)(struct sock *sk,
3483 struct sk_buff *skb),
3485 struct sk_buff **last);
3486 struct sk_buff *__skb_recv_datagram(struct sock *sk,
3487 struct sk_buff_head *sk_queue,
3489 void (*destructor)(struct sock *sk,
3490 struct sk_buff *skb),
3491 int *off, int *err);
3492 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3494 __poll_t datagram_poll(struct file *file, struct socket *sock,
3495 struct poll_table_struct *wait);
3496 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3497 struct iov_iter *to, int size);
3498 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3499 struct msghdr *msg, int size)
3501 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3503 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3504 struct msghdr *msg);
3505 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3506 struct iov_iter *to, int len,
3507 struct ahash_request *hash);
3508 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3509 struct iov_iter *from, int len);
3510 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3511 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3512 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3513 static inline void skb_free_datagram_locked(struct sock *sk,
3514 struct sk_buff *skb)
3516 __skb_free_datagram_locked(sk, skb, 0);
3518 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3519 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3520 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3521 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3522 int len, __wsum csum);
3523 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3524 struct pipe_inode_info *pipe, unsigned int len,
3525 unsigned int flags);
3526 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3528 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3529 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3530 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3532 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3533 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3534 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3535 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3536 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3537 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3538 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3539 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3540 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3541 int skb_vlan_pop(struct sk_buff *skb);
3542 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3543 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
3544 int mac_len, bool ethernet);
3545 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
3547 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3548 int skb_mpls_dec_ttl(struct sk_buff *skb);
3549 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3552 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3554 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3557 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3559 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3562 struct skb_checksum_ops {
3563 __wsum (*update)(const void *mem, int len, __wsum wsum);
3564 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3567 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3569 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3570 __wsum csum, const struct skb_checksum_ops *ops);
3571 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3574 static inline void * __must_check
3575 __skb_header_pointer(const struct sk_buff *skb, int offset,
3576 int len, void *data, int hlen, void *buffer)
3578 if (hlen - offset >= len)
3579 return data + offset;
3582 skb_copy_bits(skb, offset, buffer, len) < 0)
3588 static inline void * __must_check
3589 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3591 return __skb_header_pointer(skb, offset, len, skb->data,
3592 skb_headlen(skb), buffer);
3596 * skb_needs_linearize - check if we need to linearize a given skb
3597 * depending on the given device features.
3598 * @skb: socket buffer to check
3599 * @features: net device features
3601 * Returns true if either:
3602 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3603 * 2. skb is fragmented and the device does not support SG.
3605 static inline bool skb_needs_linearize(struct sk_buff *skb,
3606 netdev_features_t features)
3608 return skb_is_nonlinear(skb) &&
3609 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3610 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3613 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3615 const unsigned int len)
3617 memcpy(to, skb->data, len);
3620 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3621 const int offset, void *to,
3622 const unsigned int len)
3624 memcpy(to, skb->data + offset, len);
3627 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3629 const unsigned int len)
3631 memcpy(skb->data, from, len);
3634 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3637 const unsigned int len)
3639 memcpy(skb->data + offset, from, len);
3642 void skb_init(void);
3644 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3650 * skb_get_timestamp - get timestamp from a skb
3651 * @skb: skb to get stamp from
3652 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
3654 * Timestamps are stored in the skb as offsets to a base timestamp.
3655 * This function converts the offset back to a struct timeval and stores
3658 static inline void skb_get_timestamp(const struct sk_buff *skb,
3659 struct __kernel_old_timeval *stamp)
3661 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
3664 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
3665 struct __kernel_sock_timeval *stamp)
3667 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3669 stamp->tv_sec = ts.tv_sec;
3670 stamp->tv_usec = ts.tv_nsec / 1000;
3673 static inline void skb_get_timestampns(const struct sk_buff *skb,
3674 struct __kernel_old_timespec *stamp)
3676 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3678 stamp->tv_sec = ts.tv_sec;
3679 stamp->tv_nsec = ts.tv_nsec;
3682 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
3683 struct __kernel_timespec *stamp)
3685 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3687 stamp->tv_sec = ts.tv_sec;
3688 stamp->tv_nsec = ts.tv_nsec;
3691 static inline void __net_timestamp(struct sk_buff *skb)
3693 skb->tstamp = ktime_get_real();
3696 static inline ktime_t net_timedelta(ktime_t t)
3698 return ktime_sub(ktime_get_real(), t);
3701 static inline ktime_t net_invalid_timestamp(void)
3706 static inline u8 skb_metadata_len(const struct sk_buff *skb)
3708 return skb_shinfo(skb)->meta_len;
3711 static inline void *skb_metadata_end(const struct sk_buff *skb)
3713 return skb_mac_header(skb);
3716 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3717 const struct sk_buff *skb_b,
3720 const void *a = skb_metadata_end(skb_a);
3721 const void *b = skb_metadata_end(skb_b);
3722 /* Using more efficient varaiant than plain call to memcmp(). */
3723 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3727 #define __it(x, op) (x -= sizeof(u##op))
3728 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3729 case 32: diffs |= __it_diff(a, b, 64);
3731 case 24: diffs |= __it_diff(a, b, 64);
3733 case 16: diffs |= __it_diff(a, b, 64);
3735 case 8: diffs |= __it_diff(a, b, 64);
3737 case 28: diffs |= __it_diff(a, b, 64);
3739 case 20: diffs |= __it_diff(a, b, 64);
3741 case 12: diffs |= __it_diff(a, b, 64);
3743 case 4: diffs |= __it_diff(a, b, 32);
3748 return memcmp(a - meta_len, b - meta_len, meta_len);
3752 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3753 const struct sk_buff *skb_b)
3755 u8 len_a = skb_metadata_len(skb_a);
3756 u8 len_b = skb_metadata_len(skb_b);
3758 if (!(len_a | len_b))
3761 return len_a != len_b ?
3762 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3765 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3767 skb_shinfo(skb)->meta_len = meta_len;
3770 static inline void skb_metadata_clear(struct sk_buff *skb)
3772 skb_metadata_set(skb, 0);
3775 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3777 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3779 void skb_clone_tx_timestamp(struct sk_buff *skb);
3780 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3782 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3784 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3788 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3793 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3796 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3798 * PHY drivers may accept clones of transmitted packets for
3799 * timestamping via their phy_driver.txtstamp method. These drivers
3800 * must call this function to return the skb back to the stack with a
3803 * @skb: clone of the the original outgoing packet
3804 * @hwtstamps: hardware time stamps
3807 void skb_complete_tx_timestamp(struct sk_buff *skb,
3808 struct skb_shared_hwtstamps *hwtstamps);
3810 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3811 struct skb_shared_hwtstamps *hwtstamps,
3812 struct sock *sk, int tstype);
3815 * skb_tstamp_tx - queue clone of skb with send time stamps
3816 * @orig_skb: the original outgoing packet
3817 * @hwtstamps: hardware time stamps, may be NULL if not available
3819 * If the skb has a socket associated, then this function clones the
3820 * skb (thus sharing the actual data and optional structures), stores
3821 * the optional hardware time stamping information (if non NULL) or
3822 * generates a software time stamp (otherwise), then queues the clone
3823 * to the error queue of the socket. Errors are silently ignored.
3825 void skb_tstamp_tx(struct sk_buff *orig_skb,
3826 struct skb_shared_hwtstamps *hwtstamps);
3829 * skb_tx_timestamp() - Driver hook for transmit timestamping
3831 * Ethernet MAC Drivers should call this function in their hard_xmit()
3832 * function immediately before giving the sk_buff to the MAC hardware.
3834 * Specifically, one should make absolutely sure that this function is
3835 * called before TX completion of this packet can trigger. Otherwise
3836 * the packet could potentially already be freed.
3838 * @skb: A socket buffer.
3840 static inline void skb_tx_timestamp(struct sk_buff *skb)
3842 skb_clone_tx_timestamp(skb);
3843 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3844 skb_tstamp_tx(skb, NULL);
3848 * skb_complete_wifi_ack - deliver skb with wifi status
3850 * @skb: the original outgoing packet
3851 * @acked: ack status
3854 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3856 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3857 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3859 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3861 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3863 (skb->ip_summed == CHECKSUM_PARTIAL &&
3864 skb_checksum_start_offset(skb) >= 0));
3868 * skb_checksum_complete - Calculate checksum of an entire packet
3869 * @skb: packet to process
3871 * This function calculates the checksum over the entire packet plus
3872 * the value of skb->csum. The latter can be used to supply the
3873 * checksum of a pseudo header as used by TCP/UDP. It returns the
3876 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3877 * this function can be used to verify that checksum on received
3878 * packets. In that case the function should return zero if the
3879 * checksum is correct. In particular, this function will return zero
3880 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3881 * hardware has already verified the correctness of the checksum.
3883 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3885 return skb_csum_unnecessary(skb) ?
3886 0 : __skb_checksum_complete(skb);
3889 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3891 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3892 if (skb->csum_level == 0)
3893 skb->ip_summed = CHECKSUM_NONE;
3899 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3901 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3902 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3904 } else if (skb->ip_summed == CHECKSUM_NONE) {
3905 skb->ip_summed = CHECKSUM_UNNECESSARY;
3906 skb->csum_level = 0;
3910 /* Check if we need to perform checksum complete validation.
3912 * Returns true if checksum complete is needed, false otherwise
3913 * (either checksum is unnecessary or zero checksum is allowed).
3915 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3919 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3920 skb->csum_valid = 1;
3921 __skb_decr_checksum_unnecessary(skb);
3928 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
3931 #define CHECKSUM_BREAK 76
3933 /* Unset checksum-complete
3935 * Unset checksum complete can be done when packet is being modified
3936 * (uncompressed for instance) and checksum-complete value is
3939 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3941 if (skb->ip_summed == CHECKSUM_COMPLETE)
3942 skb->ip_summed = CHECKSUM_NONE;
3945 /* Validate (init) checksum based on checksum complete.
3948 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3949 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3950 * checksum is stored in skb->csum for use in __skb_checksum_complete
3951 * non-zero: value of invalid checksum
3954 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3958 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3959 if (!csum_fold(csum_add(psum, skb->csum))) {
3960 skb->csum_valid = 1;
3967 if (complete || skb->len <= CHECKSUM_BREAK) {
3970 csum = __skb_checksum_complete(skb);
3971 skb->csum_valid = !csum;
3978 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3983 /* Perform checksum validate (init). Note that this is a macro since we only
3984 * want to calculate the pseudo header which is an input function if necessary.
3985 * First we try to validate without any computation (checksum unnecessary) and
3986 * then calculate based on checksum complete calling the function to compute
3990 * 0: checksum is validated or try to in skb_checksum_complete
3991 * non-zero: value of invalid checksum
3993 #define __skb_checksum_validate(skb, proto, complete, \
3994 zero_okay, check, compute_pseudo) \
3996 __sum16 __ret = 0; \
3997 skb->csum_valid = 0; \
3998 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3999 __ret = __skb_checksum_validate_complete(skb, \
4000 complete, compute_pseudo(skb, proto)); \
4004 #define skb_checksum_init(skb, proto, compute_pseudo) \
4005 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4007 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4008 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4010 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4011 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4013 #define skb_checksum_validate_zero_check(skb, proto, check, \
4015 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4017 #define skb_checksum_simple_validate(skb) \
4018 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4020 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4022 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4025 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4027 skb->csum = ~pseudo;
4028 skb->ip_summed = CHECKSUM_COMPLETE;
4031 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4033 if (__skb_checksum_convert_check(skb)) \
4034 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4037 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4038 u16 start, u16 offset)
4040 skb->ip_summed = CHECKSUM_PARTIAL;
4041 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4042 skb->csum_offset = offset - start;
4045 /* Update skbuf and packet to reflect the remote checksum offload operation.
4046 * When called, ptr indicates the starting point for skb->csum when
4047 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4048 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4050 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4051 int start, int offset, bool nopartial)
4056 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4060 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4061 __skb_checksum_complete(skb);
4062 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4065 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4067 /* Adjust skb->csum since we changed the packet */
4068 skb->csum = csum_add(skb->csum, delta);
4071 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4073 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4074 return (void *)(skb->_nfct & NFCT_PTRMASK);
4080 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4082 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4089 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4091 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4096 #ifdef CONFIG_SKB_EXTENSIONS
4098 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4104 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4107 #if IS_ENABLED(CONFIG_MPTCP)
4110 SKB_EXT_NUM, /* must be last */
4114 * struct skb_ext - sk_buff extensions
4115 * @refcnt: 1 on allocation, deallocated on 0
4116 * @offset: offset to add to @data to obtain extension address
4117 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4118 * @data: start of extension data, variable sized
4120 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4121 * to use 'u8' types while allowing up to 2kb worth of extension data.
4125 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4126 u8 chunks; /* same */
4127 char data[0] __aligned(8);
4130 struct skb_ext *__skb_ext_alloc(void);
4131 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4132 struct skb_ext *ext);
4133 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4134 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4135 void __skb_ext_put(struct skb_ext *ext);
4137 static inline void skb_ext_put(struct sk_buff *skb)
4139 if (skb->active_extensions)
4140 __skb_ext_put(skb->extensions);
4143 static inline void __skb_ext_copy(struct sk_buff *dst,
4144 const struct sk_buff *src)
4146 dst->active_extensions = src->active_extensions;
4148 if (src->active_extensions) {
4149 struct skb_ext *ext = src->extensions;
4151 refcount_inc(&ext->refcnt);
4152 dst->extensions = ext;
4156 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4159 __skb_ext_copy(dst, src);
4162 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4164 return !!ext->offset[i];
4167 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4169 return skb->active_extensions & (1 << id);
4172 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4174 if (skb_ext_exist(skb, id))
4175 __skb_ext_del(skb, id);
4178 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4180 if (skb_ext_exist(skb, id)) {
4181 struct skb_ext *ext = skb->extensions;
4183 return (void *)ext + (ext->offset[id] << 3);
4189 static inline void skb_ext_reset(struct sk_buff *skb)
4191 if (unlikely(skb->active_extensions)) {
4192 __skb_ext_put(skb->extensions);
4193 skb->active_extensions = 0;
4197 static inline bool skb_has_extensions(struct sk_buff *skb)
4199 return unlikely(skb->active_extensions);
4202 static inline void skb_ext_put(struct sk_buff *skb) {}
4203 static inline void skb_ext_reset(struct sk_buff *skb) {}
4204 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4205 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4206 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4207 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4208 #endif /* CONFIG_SKB_EXTENSIONS */
4210 static inline void nf_reset_ct(struct sk_buff *skb)
4212 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4213 nf_conntrack_put(skb_nfct(skb));
4218 static inline void nf_reset_trace(struct sk_buff *skb)
4220 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4225 static inline void ipvs_reset(struct sk_buff *skb)
4227 #if IS_ENABLED(CONFIG_IP_VS)
4228 skb->ipvs_property = 0;
4232 /* Note: This doesn't put any conntrack info in dst. */
4233 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4236 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4237 dst->_nfct = src->_nfct;
4238 nf_conntrack_get(skb_nfct(src));
4240 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4242 dst->nf_trace = src->nf_trace;
4246 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4248 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4249 nf_conntrack_put(skb_nfct(dst));
4251 __nf_copy(dst, src, true);
4254 #ifdef CONFIG_NETWORK_SECMARK
4255 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4257 to->secmark = from->secmark;
4260 static inline void skb_init_secmark(struct sk_buff *skb)
4265 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4268 static inline void skb_init_secmark(struct sk_buff *skb)
4272 static inline int secpath_exists(const struct sk_buff *skb)
4275 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4281 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4283 return !skb->destructor &&
4284 !secpath_exists(skb) &&
4286 !skb->_skb_refdst &&
4287 !skb_has_frag_list(skb);
4290 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4292 skb->queue_mapping = queue_mapping;
4295 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4297 return skb->queue_mapping;
4300 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4302 to->queue_mapping = from->queue_mapping;
4305 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4307 skb->queue_mapping = rx_queue + 1;
4310 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4312 return skb->queue_mapping - 1;
4315 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4317 return skb->queue_mapping != 0;
4320 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4322 skb->dst_pending_confirm = val;
4325 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4327 return skb->dst_pending_confirm != 0;
4330 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4333 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4339 /* Keeps track of mac header offset relative to skb->head.
4340 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4341 * For non-tunnel skb it points to skb_mac_header() and for
4342 * tunnel skb it points to outer mac header.
4343 * Keeps track of level of encapsulation of network headers.
4354 #define SKB_SGO_CB_OFFSET 32
4355 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
4357 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4359 return (skb_mac_header(inner_skb) - inner_skb->head) -
4360 SKB_GSO_CB(inner_skb)->mac_offset;
4363 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4365 int new_headroom, headroom;
4368 headroom = skb_headroom(skb);
4369 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4373 new_headroom = skb_headroom(skb);
4374 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4378 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4380 /* Do not update partial checksums if remote checksum is enabled. */
4381 if (skb->remcsum_offload)
4384 SKB_GSO_CB(skb)->csum = res;
4385 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4388 /* Compute the checksum for a gso segment. First compute the checksum value
4389 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4390 * then add in skb->csum (checksum from csum_start to end of packet).
4391 * skb->csum and csum_start are then updated to reflect the checksum of the
4392 * resultant packet starting from the transport header-- the resultant checksum
4393 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4396 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4398 unsigned char *csum_start = skb_transport_header(skb);
4399 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4400 __wsum partial = SKB_GSO_CB(skb)->csum;
4402 SKB_GSO_CB(skb)->csum = res;
4403 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4405 return csum_fold(csum_partial(csum_start, plen, partial));
4408 static inline bool skb_is_gso(const struct sk_buff *skb)
4410 return skb_shinfo(skb)->gso_size;
4413 /* Note: Should be called only if skb_is_gso(skb) is true */
4414 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4416 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4419 /* Note: Should be called only if skb_is_gso(skb) is true */
4420 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4422 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4425 /* Note: Should be called only if skb_is_gso(skb) is true */
4426 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4428 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4431 static inline void skb_gso_reset(struct sk_buff *skb)
4433 skb_shinfo(skb)->gso_size = 0;
4434 skb_shinfo(skb)->gso_segs = 0;
4435 skb_shinfo(skb)->gso_type = 0;
4438 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4441 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4443 shinfo->gso_size += increment;
4446 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4449 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4451 shinfo->gso_size -= decrement;
4454 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4456 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4458 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4459 * wanted then gso_type will be set. */
4460 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4462 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4463 unlikely(shinfo->gso_type == 0)) {
4464 __skb_warn_lro_forwarding(skb);
4470 static inline void skb_forward_csum(struct sk_buff *skb)
4472 /* Unfortunately we don't support this one. Any brave souls? */
4473 if (skb->ip_summed == CHECKSUM_COMPLETE)
4474 skb->ip_summed = CHECKSUM_NONE;
4478 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4479 * @skb: skb to check
4481 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4482 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4483 * use this helper, to document places where we make this assertion.
4485 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4488 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4492 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4494 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4495 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4496 unsigned int transport_len,
4497 __sum16(*skb_chkf)(struct sk_buff *skb));
4500 * skb_head_is_locked - Determine if the skb->head is locked down
4501 * @skb: skb to check
4503 * The head on skbs build around a head frag can be removed if they are
4504 * not cloned. This function returns true if the skb head is locked down
4505 * due to either being allocated via kmalloc, or by being a clone with
4506 * multiple references to the head.
4508 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4510 return !skb->head_frag || skb_cloned(skb);
4513 /* Local Checksum Offload.
4514 * Compute outer checksum based on the assumption that the
4515 * inner checksum will be offloaded later.
4516 * See Documentation/networking/checksum-offloads.rst for
4517 * explanation of how this works.
4518 * Fill in outer checksum adjustment (e.g. with sum of outer
4519 * pseudo-header) before calling.
4520 * Also ensure that inner checksum is in linear data area.
4522 static inline __wsum lco_csum(struct sk_buff *skb)
4524 unsigned char *csum_start = skb_checksum_start(skb);
4525 unsigned char *l4_hdr = skb_transport_header(skb);
4528 /* Start with complement of inner checksum adjustment */
4529 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4532 /* Add in checksum of our headers (incl. outer checksum
4533 * adjustment filled in by caller) and return result.
4535 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4538 #endif /* __KERNEL__ */
4539 #endif /* _LINUX_SKBUFF_H */