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,
597 #if BITS_PER_LONG > 32
598 #define NET_SKBUFF_DATA_USES_OFFSET 1
601 #ifdef NET_SKBUFF_DATA_USES_OFFSET
602 typedef unsigned int sk_buff_data_t;
604 typedef unsigned char *sk_buff_data_t;
608 * struct sk_buff - socket buffer
609 * @next: Next buffer in list
610 * @prev: Previous buffer in list
611 * @tstamp: Time we arrived/left
612 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
613 * @sk: Socket we are owned by
614 * @dev: Device we arrived on/are leaving by
615 * @cb: Control buffer. Free for use by every layer. Put private vars here
616 * @_skb_refdst: destination entry (with norefcount bit)
617 * @sp: the security path, used for xfrm
618 * @len: Length of actual data
619 * @data_len: Data length
620 * @mac_len: Length of link layer header
621 * @hdr_len: writable header length of cloned skb
622 * @csum: Checksum (must include start/offset pair)
623 * @csum_start: Offset from skb->head where checksumming should start
624 * @csum_offset: Offset from csum_start where checksum should be stored
625 * @priority: Packet queueing priority
626 * @ignore_df: allow local fragmentation
627 * @cloned: Head may be cloned (check refcnt to be sure)
628 * @ip_summed: Driver fed us an IP checksum
629 * @nohdr: Payload reference only, must not modify header
630 * @pkt_type: Packet class
631 * @fclone: skbuff clone status
632 * @ipvs_property: skbuff is owned by ipvs
633 * @offload_fwd_mark: Packet was L2-forwarded in hardware
634 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
635 * @tc_skip_classify: do not classify packet. set by IFB device
636 * @tc_at_ingress: used within tc_classify to distinguish in/egress
637 * @redirected: packet was redirected by packet classifier
638 * @from_ingress: packet was redirected from the ingress path
639 * @peeked: this packet has been seen already, so stats have been
640 * done for it, don't do them again
641 * @nf_trace: netfilter packet trace flag
642 * @protocol: Packet protocol from driver
643 * @destructor: Destruct function
644 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
645 * @_nfct: Associated connection, if any (with nfctinfo bits)
646 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
647 * @skb_iif: ifindex of device we arrived on
648 * @tc_index: Traffic control index
649 * @hash: the packet hash
650 * @queue_mapping: Queue mapping for multiqueue devices
651 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
652 * @active_extensions: active extensions (skb_ext_id types)
653 * @ndisc_nodetype: router type (from link layer)
654 * @ooo_okay: allow the mapping of a socket to a queue to be changed
655 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
657 * @sw_hash: indicates hash was computed in software stack
658 * @wifi_acked_valid: wifi_acked was set
659 * @wifi_acked: whether frame was acked on wifi or not
660 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
661 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
662 * @scm_io_uring: SKB holds io_uring registered files
663 * @dst_pending_confirm: need to confirm neighbour
664 * @decrypted: Decrypted SKB
665 * @napi_id: id of the NAPI struct this skb came from
666 * @secmark: security marking
667 * @mark: Generic packet mark
668 * @vlan_proto: vlan encapsulation protocol
669 * @vlan_tci: vlan tag control information
670 * @inner_protocol: Protocol (encapsulation)
671 * @inner_transport_header: Inner transport layer header (encapsulation)
672 * @inner_network_header: Network layer header (encapsulation)
673 * @inner_mac_header: Link layer header (encapsulation)
674 * @transport_header: Transport layer header
675 * @network_header: Network layer header
676 * @mac_header: Link layer header
677 * @tail: Tail pointer
679 * @head: Head of buffer
680 * @data: Data head pointer
681 * @truesize: Buffer size
682 * @users: User count - see {datagram,tcp}.c
683 * @extensions: allocated extensions, valid if active_extensions is nonzero
689 /* These two members must be first. */
690 struct sk_buff *next;
691 struct sk_buff *prev;
694 struct net_device *dev;
695 /* Some protocols might use this space to store information,
696 * while device pointer would be NULL.
697 * UDP receive path is one user.
699 unsigned long dev_scratch;
702 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
703 struct list_head list;
708 int ip_defrag_offset;
713 u64 skb_mstamp_ns; /* earliest departure time */
716 * This is the control buffer. It is free to use for every
717 * layer. Please put your private variables there. If you
718 * want to keep them across layers you have to do a skb_clone()
719 * first. This is owned by whoever has the skb queued ATM.
721 char cb[48] __aligned(8);
725 unsigned long _skb_refdst;
726 void (*destructor)(struct sk_buff *skb);
728 struct list_head tcp_tsorted_anchor;
731 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
739 /* Following fields are _not_ copied in __copy_skb_header()
740 * Note that queue_mapping is here mostly to fill a hole.
744 /* if you move cloned around you also must adapt those constants */
745 #ifdef __BIG_ENDIAN_BITFIELD
746 #define CLONED_MASK (1 << 7)
748 #define CLONED_MASK 1
750 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
752 __u8 __cloned_offset[0];
759 #ifdef CONFIG_SKB_EXTENSIONS
760 __u8 active_extensions;
762 /* fields enclosed in headers_start/headers_end are copied
763 * using a single memcpy() in __copy_skb_header()
766 __u32 headers_start[0];
769 /* if you move pkt_type around you also must adapt those constants */
770 #ifdef __BIG_ENDIAN_BITFIELD
771 #define PKT_TYPE_MAX (7 << 5)
773 #define PKT_TYPE_MAX 7
775 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
777 __u8 __pkt_type_offset[0];
786 __u8 wifi_acked_valid:1;
789 /* Indicates the inner headers are valid in the skbuff. */
790 __u8 encapsulation:1;
791 __u8 encap_hdr_csum:1;
794 #ifdef __BIG_ENDIAN_BITFIELD
795 #define PKT_VLAN_PRESENT_BIT 7
797 #define PKT_VLAN_PRESENT_BIT 0
799 #define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset)
800 __u8 __pkt_vlan_present_offset[0];
802 __u8 csum_complete_sw:1;
804 __u8 csum_not_inet:1;
805 __u8 dst_pending_confirm:1;
806 #ifdef CONFIG_IPV6_NDISC_NODETYPE
807 __u8 ndisc_nodetype:2;
810 __u8 ipvs_property:1;
811 __u8 inner_protocol_type:1;
812 __u8 remcsum_offload:1;
813 #ifdef CONFIG_NET_SWITCHDEV
814 __u8 offload_fwd_mark:1;
815 __u8 offload_l3_fwd_mark:1;
817 #ifdef CONFIG_NET_CLS_ACT
818 __u8 tc_skip_classify:1;
819 __u8 tc_at_ingress:1;
821 #ifdef CONFIG_NET_REDIRECT
825 #ifdef CONFIG_TLS_DEVICE
830 #ifdef CONFIG_NET_SCHED
831 __u16 tc_index; /* traffic control index */
846 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
848 unsigned int napi_id;
849 unsigned int sender_cpu;
852 #ifdef CONFIG_NETWORK_SECMARK
858 __u32 reserved_tailroom;
862 __be16 inner_protocol;
866 __u16 inner_transport_header;
867 __u16 inner_network_header;
868 __u16 inner_mac_header;
871 __u16 transport_header;
872 __u16 network_header;
876 __u32 headers_end[0];
879 /* These elements must be at the end, see alloc_skb() for details. */
884 unsigned int truesize;
887 #ifdef CONFIG_SKB_EXTENSIONS
888 /* only useable after checking ->active_extensions != 0 */
889 struct skb_ext *extensions;
895 * Handling routines are only of interest to the kernel
898 #define SKB_ALLOC_FCLONE 0x01
899 #define SKB_ALLOC_RX 0x02
900 #define SKB_ALLOC_NAPI 0x04
903 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
906 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
908 return unlikely(skb->pfmemalloc);
912 * skb might have a dst pointer attached, refcounted or not.
913 * _skb_refdst low order bit is set if refcount was _not_ taken
915 #define SKB_DST_NOREF 1UL
916 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
919 * skb_dst - returns skb dst_entry
922 * Returns skb dst_entry, regardless of reference taken or not.
924 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
926 /* If refdst was not refcounted, check we still are in a
927 * rcu_read_lock section
929 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
930 !rcu_read_lock_held() &&
931 !rcu_read_lock_bh_held());
932 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
936 * skb_dst_set - sets skb dst
940 * Sets skb dst, assuming a reference was taken on dst and should
941 * be released by skb_dst_drop()
943 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
945 skb->_skb_refdst = (unsigned long)dst;
949 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
953 * Sets skb dst, assuming a reference was not taken on dst.
954 * If dst entry is cached, we do not take reference and dst_release
955 * will be avoided by refdst_drop. If dst entry is not cached, we take
956 * reference, so that last dst_release can destroy the dst immediately.
958 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
960 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
961 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
965 * skb_dst_is_noref - Test if skb dst isn't refcounted
968 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
970 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
974 * skb_rtable - Returns the skb &rtable
977 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
979 return (struct rtable *)skb_dst(skb);
982 /* For mangling skb->pkt_type from user space side from applications
983 * such as nft, tc, etc, we only allow a conservative subset of
984 * possible pkt_types to be set.
986 static inline bool skb_pkt_type_ok(u32 ptype)
988 return ptype <= PACKET_OTHERHOST;
992 * skb_napi_id - Returns the skb's NAPI id
995 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
997 #ifdef CONFIG_NET_RX_BUSY_POLL
1005 * skb_unref - decrement the skb's reference count
1008 * Returns true if we can free the skb.
1010 static inline bool skb_unref(struct sk_buff *skb)
1014 if (likely(refcount_read(&skb->users) == 1))
1016 else if (likely(!refcount_dec_and_test(&skb->users)))
1022 void skb_release_head_state(struct sk_buff *skb);
1023 void kfree_skb(struct sk_buff *skb);
1024 void kfree_skb_list(struct sk_buff *segs);
1025 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1026 void skb_tx_error(struct sk_buff *skb);
1027 void consume_skb(struct sk_buff *skb);
1028 void __consume_stateless_skb(struct sk_buff *skb);
1029 void __kfree_skb(struct sk_buff *skb);
1030 extern struct kmem_cache *skbuff_head_cache;
1032 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1033 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1034 bool *fragstolen, int *delta_truesize);
1036 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1038 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1039 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1040 struct sk_buff *build_skb_around(struct sk_buff *skb,
1041 void *data, unsigned int frag_size);
1044 * alloc_skb - allocate a network buffer
1045 * @size: size to allocate
1046 * @priority: allocation mask
1048 * This function is a convenient wrapper around __alloc_skb().
1050 static inline struct sk_buff *alloc_skb(unsigned int size,
1053 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1056 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1057 unsigned long data_len,
1061 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1063 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1064 struct sk_buff_fclones {
1065 struct sk_buff skb1;
1067 struct sk_buff skb2;
1069 refcount_t fclone_ref;
1073 * skb_fclone_busy - check if fclone is busy
1077 * Returns true if skb is a fast clone, and its clone is not freed.
1078 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1079 * so we also check that this didnt happen.
1081 static inline bool skb_fclone_busy(const struct sock *sk,
1082 const struct sk_buff *skb)
1084 const struct sk_buff_fclones *fclones;
1086 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1088 return skb->fclone == SKB_FCLONE_ORIG &&
1089 refcount_read(&fclones->fclone_ref) > 1 &&
1090 fclones->skb2.sk == sk;
1094 * alloc_skb_fclone - allocate a network buffer from fclone cache
1095 * @size: size to allocate
1096 * @priority: allocation mask
1098 * This function is a convenient wrapper around __alloc_skb().
1100 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1103 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1106 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1107 void skb_headers_offset_update(struct sk_buff *skb, int off);
1108 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1109 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1110 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1111 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1112 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1113 gfp_t gfp_mask, bool fclone);
1114 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1117 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1120 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1121 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1122 unsigned int headroom);
1123 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1124 int newtailroom, gfp_t priority);
1125 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1126 int offset, int len);
1127 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1128 int offset, int len);
1129 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1130 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1133 * skb_pad - zero pad the tail of an skb
1134 * @skb: buffer to pad
1135 * @pad: space to pad
1137 * Ensure that a buffer is followed by a padding area that is zero
1138 * filled. Used by network drivers which may DMA or transfer data
1139 * beyond the buffer end onto the wire.
1141 * May return error in out of memory cases. The skb is freed on error.
1143 static inline int skb_pad(struct sk_buff *skb, int pad)
1145 return __skb_pad(skb, pad, true);
1147 #define dev_kfree_skb(a) consume_skb(a)
1149 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1150 int offset, size_t size);
1152 struct skb_seq_state {
1156 __u32 stepped_offset;
1157 struct sk_buff *root_skb;
1158 struct sk_buff *cur_skb;
1162 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1163 unsigned int to, struct skb_seq_state *st);
1164 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1165 struct skb_seq_state *st);
1166 void skb_abort_seq_read(struct skb_seq_state *st);
1168 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1169 unsigned int to, struct ts_config *config);
1172 * Packet hash types specify the type of hash in skb_set_hash.
1174 * Hash types refer to the protocol layer addresses which are used to
1175 * construct a packet's hash. The hashes are used to differentiate or identify
1176 * flows of the protocol layer for the hash type. Hash types are either
1177 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1179 * Properties of hashes:
1181 * 1) Two packets in different flows have different hash values
1182 * 2) Two packets in the same flow should have the same hash value
1184 * A hash at a higher layer is considered to be more specific. A driver should
1185 * set the most specific hash possible.
1187 * A driver cannot indicate a more specific hash than the layer at which a hash
1188 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1190 * A driver may indicate a hash level which is less specific than the
1191 * actual layer the hash was computed on. For instance, a hash computed
1192 * at L4 may be considered an L3 hash. This should only be done if the
1193 * driver can't unambiguously determine that the HW computed the hash at
1194 * the higher layer. Note that the "should" in the second property above
1197 enum pkt_hash_types {
1198 PKT_HASH_TYPE_NONE, /* Undefined type */
1199 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1200 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1201 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1204 static inline void skb_clear_hash(struct sk_buff *skb)
1211 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1214 skb_clear_hash(skb);
1218 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1220 skb->l4_hash = is_l4;
1221 skb->sw_hash = is_sw;
1226 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1228 /* Used by drivers to set hash from HW */
1229 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1233 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1235 __skb_set_hash(skb, hash, true, is_l4);
1238 void __skb_get_hash(struct sk_buff *skb);
1239 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1240 u32 skb_get_poff(const struct sk_buff *skb);
1241 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1242 const struct flow_keys_basic *keys, int hlen);
1243 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1244 void *data, int hlen_proto);
1246 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1247 int thoff, u8 ip_proto)
1249 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1252 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1253 const struct flow_dissector_key *key,
1254 unsigned int key_count);
1257 int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1258 union bpf_attr __user *uattr);
1259 int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1260 struct bpf_prog *prog);
1262 int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr);
1264 static inline int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1265 union bpf_attr __user *uattr)
1270 static inline int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1271 struct bpf_prog *prog)
1276 static inline int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr)
1282 struct bpf_flow_dissector;
1283 bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1284 __be16 proto, int nhoff, int hlen, unsigned int flags);
1286 bool __skb_flow_dissect(const struct net *net,
1287 const struct sk_buff *skb,
1288 struct flow_dissector *flow_dissector,
1289 void *target_container,
1290 void *data, __be16 proto, int nhoff, int hlen,
1291 unsigned int flags);
1293 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1294 struct flow_dissector *flow_dissector,
1295 void *target_container, unsigned int flags)
1297 return __skb_flow_dissect(NULL, skb, flow_dissector,
1298 target_container, NULL, 0, 0, 0, flags);
1301 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1302 struct flow_keys *flow,
1305 memset(flow, 0, sizeof(*flow));
1306 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1307 flow, NULL, 0, 0, 0, flags);
1311 skb_flow_dissect_flow_keys_basic(const struct net *net,
1312 const struct sk_buff *skb,
1313 struct flow_keys_basic *flow, void *data,
1314 __be16 proto, int nhoff, int hlen,
1317 memset(flow, 0, sizeof(*flow));
1318 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1319 data, proto, nhoff, hlen, flags);
1322 void skb_flow_dissect_meta(const struct sk_buff *skb,
1323 struct flow_dissector *flow_dissector,
1324 void *target_container);
1326 /* Gets a skb connection tracking info, ctinfo map should be a
1327 * a map of mapsize to translate enum ip_conntrack_info states
1331 skb_flow_dissect_ct(const struct sk_buff *skb,
1332 struct flow_dissector *flow_dissector,
1333 void *target_container,
1337 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1338 struct flow_dissector *flow_dissector,
1339 void *target_container);
1341 static inline __u32 skb_get_hash(struct sk_buff *skb)
1343 if (!skb->l4_hash && !skb->sw_hash)
1344 __skb_get_hash(skb);
1349 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1351 if (!skb->l4_hash && !skb->sw_hash) {
1352 struct flow_keys keys;
1353 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1355 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1361 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1362 const siphash_key_t *perturb);
1364 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1369 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1371 to->hash = from->hash;
1372 to->sw_hash = from->sw_hash;
1373 to->l4_hash = from->l4_hash;
1376 static inline void skb_copy_decrypted(struct sk_buff *to,
1377 const struct sk_buff *from)
1379 #ifdef CONFIG_TLS_DEVICE
1380 to->decrypted = from->decrypted;
1384 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1385 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1387 return skb->head + skb->end;
1390 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1395 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1400 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1402 return skb->end - skb->head;
1407 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1409 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1411 return &skb_shinfo(skb)->hwtstamps;
1414 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1416 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1418 return is_zcopy ? skb_uarg(skb) : NULL;
1421 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1424 if (skb && uarg && !skb_zcopy(skb)) {
1425 if (unlikely(have_ref && *have_ref))
1428 sock_zerocopy_get(uarg);
1429 skb_shinfo(skb)->destructor_arg = uarg;
1430 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1434 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1436 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1437 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1440 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1442 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1445 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1447 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1450 /* Release a reference on a zerocopy structure */
1451 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1453 struct ubuf_info *uarg = skb_zcopy(skb);
1456 if (skb_zcopy_is_nouarg(skb)) {
1457 /* no notification callback */
1458 } else if (uarg->callback == sock_zerocopy_callback) {
1459 uarg->zerocopy = uarg->zerocopy && zerocopy;
1460 sock_zerocopy_put(uarg);
1462 uarg->callback(uarg, zerocopy);
1465 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1469 /* Abort a zerocopy operation and revert zckey on error in send syscall */
1470 static inline void skb_zcopy_abort(struct sk_buff *skb)
1472 struct ubuf_info *uarg = skb_zcopy(skb);
1475 sock_zerocopy_put_abort(uarg, false);
1476 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1480 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1485 /* Iterate through singly-linked GSO fragments of an skb. */
1486 #define skb_list_walk_safe(first, skb, next_skb) \
1487 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1488 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1490 static inline void skb_list_del_init(struct sk_buff *skb)
1492 __list_del_entry(&skb->list);
1493 skb_mark_not_on_list(skb);
1497 * skb_queue_empty - check if a queue is empty
1500 * Returns true if the queue is empty, false otherwise.
1502 static inline int skb_queue_empty(const struct sk_buff_head *list)
1504 return list->next == (const struct sk_buff *) list;
1508 * skb_queue_empty_lockless - check if a queue is empty
1511 * Returns true if the queue is empty, false otherwise.
1512 * This variant can be used in lockless contexts.
1514 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1516 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1521 * skb_queue_is_last - check if skb is the last entry in the queue
1525 * Returns true if @skb is the last buffer on the list.
1527 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1528 const struct sk_buff *skb)
1530 return skb->next == (const struct sk_buff *) list;
1534 * skb_queue_is_first - check if skb is the first entry in the queue
1538 * Returns true if @skb is the first buffer on the list.
1540 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1541 const struct sk_buff *skb)
1543 return skb->prev == (const struct sk_buff *) list;
1547 * skb_queue_next - return the next packet in the queue
1549 * @skb: current buffer
1551 * Return the next packet in @list after @skb. It is only valid to
1552 * call this if skb_queue_is_last() evaluates to false.
1554 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1555 const struct sk_buff *skb)
1557 /* This BUG_ON may seem severe, but if we just return then we
1558 * are going to dereference garbage.
1560 BUG_ON(skb_queue_is_last(list, skb));
1565 * skb_queue_prev - return the prev packet in the queue
1567 * @skb: current buffer
1569 * Return the prev packet in @list before @skb. It is only valid to
1570 * call this if skb_queue_is_first() evaluates to false.
1572 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1573 const struct sk_buff *skb)
1575 /* This BUG_ON may seem severe, but if we just return then we
1576 * are going to dereference garbage.
1578 BUG_ON(skb_queue_is_first(list, skb));
1583 * skb_get - reference buffer
1584 * @skb: buffer to reference
1586 * Makes another reference to a socket buffer and returns a pointer
1589 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1591 refcount_inc(&skb->users);
1596 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1600 * skb_cloned - is the buffer a clone
1601 * @skb: buffer to check
1603 * Returns true if the buffer was generated with skb_clone() and is
1604 * one of multiple shared copies of the buffer. Cloned buffers are
1605 * shared data so must not be written to under normal circumstances.
1607 static inline int skb_cloned(const struct sk_buff *skb)
1609 return skb->cloned &&
1610 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1613 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1615 might_sleep_if(gfpflags_allow_blocking(pri));
1617 if (skb_cloned(skb))
1618 return pskb_expand_head(skb, 0, 0, pri);
1624 * skb_header_cloned - is the header a clone
1625 * @skb: buffer to check
1627 * Returns true if modifying the header part of the buffer requires
1628 * the data to be copied.
1630 static inline int skb_header_cloned(const struct sk_buff *skb)
1637 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1638 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1639 return dataref != 1;
1642 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1644 might_sleep_if(gfpflags_allow_blocking(pri));
1646 if (skb_header_cloned(skb))
1647 return pskb_expand_head(skb, 0, 0, pri);
1653 * __skb_header_release - release reference to header
1654 * @skb: buffer to operate on
1656 static inline void __skb_header_release(struct sk_buff *skb)
1659 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1664 * skb_shared - is the buffer shared
1665 * @skb: buffer to check
1667 * Returns true if more than one person has a reference to this
1670 static inline int skb_shared(const struct sk_buff *skb)
1672 return refcount_read(&skb->users) != 1;
1676 * skb_share_check - check if buffer is shared and if so clone it
1677 * @skb: buffer to check
1678 * @pri: priority for memory allocation
1680 * If the buffer is shared the buffer is cloned and the old copy
1681 * drops a reference. A new clone with a single reference is returned.
1682 * If the buffer is not shared the original buffer is returned. When
1683 * being called from interrupt status or with spinlocks held pri must
1686 * NULL is returned on a memory allocation failure.
1688 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1690 might_sleep_if(gfpflags_allow_blocking(pri));
1691 if (skb_shared(skb)) {
1692 struct sk_buff *nskb = skb_clone(skb, pri);
1704 * Copy shared buffers into a new sk_buff. We effectively do COW on
1705 * packets to handle cases where we have a local reader and forward
1706 * and a couple of other messy ones. The normal one is tcpdumping
1707 * a packet thats being forwarded.
1711 * skb_unshare - make a copy of a shared buffer
1712 * @skb: buffer to check
1713 * @pri: priority for memory allocation
1715 * If the socket buffer is a clone then this function creates a new
1716 * copy of the data, drops a reference count on the old copy and returns
1717 * the new copy with the reference count at 1. If the buffer is not a clone
1718 * the original buffer is returned. When called with a spinlock held or
1719 * from interrupt state @pri must be %GFP_ATOMIC
1721 * %NULL is returned on a memory allocation failure.
1723 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1726 might_sleep_if(gfpflags_allow_blocking(pri));
1727 if (skb_cloned(skb)) {
1728 struct sk_buff *nskb = skb_copy(skb, pri);
1730 /* Free our shared copy */
1741 * skb_peek - peek at the head of an &sk_buff_head
1742 * @list_: list to peek at
1744 * Peek an &sk_buff. Unlike most other operations you _MUST_
1745 * be careful with this one. A peek leaves the buffer on the
1746 * list and someone else may run off with it. You must hold
1747 * the appropriate locks or have a private queue to do this.
1749 * Returns %NULL for an empty list or a pointer to the head element.
1750 * The reference count is not incremented and the reference is therefore
1751 * volatile. Use with caution.
1753 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1755 struct sk_buff *skb = list_->next;
1757 if (skb == (struct sk_buff *)list_)
1763 * __skb_peek - peek at the head of a non-empty &sk_buff_head
1764 * @list_: list to peek at
1766 * Like skb_peek(), but the caller knows that the list is not empty.
1768 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
1774 * skb_peek_next - peek skb following the given one from a queue
1775 * @skb: skb to start from
1776 * @list_: list to peek at
1778 * Returns %NULL when the end of the list is met or a pointer to the
1779 * next element. The reference count is not incremented and the
1780 * reference is therefore volatile. Use with caution.
1782 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1783 const struct sk_buff_head *list_)
1785 struct sk_buff *next = skb->next;
1787 if (next == (struct sk_buff *)list_)
1793 * skb_peek_tail - peek at the tail of an &sk_buff_head
1794 * @list_: list to peek at
1796 * Peek an &sk_buff. Unlike most other operations you _MUST_
1797 * be careful with this one. A peek leaves the buffer on the
1798 * list and someone else may run off with it. You must hold
1799 * the appropriate locks or have a private queue to do this.
1801 * Returns %NULL for an empty list or a pointer to the tail element.
1802 * The reference count is not incremented and the reference is therefore
1803 * volatile. Use with caution.
1805 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1807 struct sk_buff *skb = READ_ONCE(list_->prev);
1809 if (skb == (struct sk_buff *)list_)
1816 * skb_queue_len - get queue length
1817 * @list_: list to measure
1819 * Return the length of an &sk_buff queue.
1821 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1827 * skb_queue_len_lockless - get queue length
1828 * @list_: list to measure
1830 * Return the length of an &sk_buff queue.
1831 * This variant can be used in lockless contexts.
1833 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
1835 return READ_ONCE(list_->qlen);
1839 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1840 * @list: queue to initialize
1842 * This initializes only the list and queue length aspects of
1843 * an sk_buff_head object. This allows to initialize the list
1844 * aspects of an sk_buff_head without reinitializing things like
1845 * the spinlock. It can also be used for on-stack sk_buff_head
1846 * objects where the spinlock is known to not be used.
1848 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1850 list->prev = list->next = (struct sk_buff *)list;
1855 * This function creates a split out lock class for each invocation;
1856 * this is needed for now since a whole lot of users of the skb-queue
1857 * infrastructure in drivers have different locking usage (in hardirq)
1858 * than the networking core (in softirq only). In the long run either the
1859 * network layer or drivers should need annotation to consolidate the
1860 * main types of usage into 3 classes.
1862 static inline void skb_queue_head_init(struct sk_buff_head *list)
1864 spin_lock_init(&list->lock);
1865 __skb_queue_head_init(list);
1868 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1869 struct lock_class_key *class)
1871 skb_queue_head_init(list);
1872 lockdep_set_class(&list->lock, class);
1876 * Insert an sk_buff on a list.
1878 * The "__skb_xxxx()" functions are the non-atomic ones that
1879 * can only be called with interrupts disabled.
1881 static inline void __skb_insert(struct sk_buff *newsk,
1882 struct sk_buff *prev, struct sk_buff *next,
1883 struct sk_buff_head *list)
1885 /* See skb_queue_empty_lockless() and skb_peek_tail()
1886 * for the opposite READ_ONCE()
1888 WRITE_ONCE(newsk->next, next);
1889 WRITE_ONCE(newsk->prev, prev);
1890 WRITE_ONCE(next->prev, newsk);
1891 WRITE_ONCE(prev->next, newsk);
1892 WRITE_ONCE(list->qlen, list->qlen + 1);
1895 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1896 struct sk_buff *prev,
1897 struct sk_buff *next)
1899 struct sk_buff *first = list->next;
1900 struct sk_buff *last = list->prev;
1902 WRITE_ONCE(first->prev, prev);
1903 WRITE_ONCE(prev->next, first);
1905 WRITE_ONCE(last->next, next);
1906 WRITE_ONCE(next->prev, last);
1910 * skb_queue_splice - join two skb lists, this is designed for stacks
1911 * @list: the new list to add
1912 * @head: the place to add it in the first list
1914 static inline void skb_queue_splice(const struct sk_buff_head *list,
1915 struct sk_buff_head *head)
1917 if (!skb_queue_empty(list)) {
1918 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1919 head->qlen += list->qlen;
1924 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1925 * @list: the new list to add
1926 * @head: the place to add it in the first list
1928 * The list at @list is reinitialised
1930 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1931 struct sk_buff_head *head)
1933 if (!skb_queue_empty(list)) {
1934 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1935 head->qlen += list->qlen;
1936 __skb_queue_head_init(list);
1941 * skb_queue_splice_tail - join two skb lists, each list being a queue
1942 * @list: the new list to add
1943 * @head: the place to add it in the first list
1945 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1946 struct sk_buff_head *head)
1948 if (!skb_queue_empty(list)) {
1949 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1950 head->qlen += list->qlen;
1955 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1956 * @list: the new list to add
1957 * @head: the place to add it in the first list
1959 * Each of the lists is a queue.
1960 * The list at @list is reinitialised
1962 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1963 struct sk_buff_head *head)
1965 if (!skb_queue_empty(list)) {
1966 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1967 head->qlen += list->qlen;
1968 __skb_queue_head_init(list);
1973 * __skb_queue_after - queue a buffer at the list head
1974 * @list: list to use
1975 * @prev: place after this buffer
1976 * @newsk: buffer to queue
1978 * Queue a buffer int the middle of a list. This function takes no locks
1979 * and you must therefore hold required locks before calling it.
1981 * A buffer cannot be placed on two lists at the same time.
1983 static inline void __skb_queue_after(struct sk_buff_head *list,
1984 struct sk_buff *prev,
1985 struct sk_buff *newsk)
1987 __skb_insert(newsk, prev, prev->next, list);
1990 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1991 struct sk_buff_head *list);
1993 static inline void __skb_queue_before(struct sk_buff_head *list,
1994 struct sk_buff *next,
1995 struct sk_buff *newsk)
1997 __skb_insert(newsk, next->prev, next, list);
2001 * __skb_queue_head - queue a buffer at the list head
2002 * @list: list to use
2003 * @newsk: buffer to queue
2005 * Queue a buffer at the start of a list. This function takes no locks
2006 * and you must therefore hold required locks before calling it.
2008 * A buffer cannot be placed on two lists at the same time.
2010 static inline void __skb_queue_head(struct sk_buff_head *list,
2011 struct sk_buff *newsk)
2013 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2015 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2018 * __skb_queue_tail - queue a buffer at the list tail
2019 * @list: list to use
2020 * @newsk: buffer to queue
2022 * Queue a buffer at the end of a list. This function takes no locks
2023 * and you must therefore hold required locks before calling it.
2025 * A buffer cannot be placed on two lists at the same time.
2027 static inline void __skb_queue_tail(struct sk_buff_head *list,
2028 struct sk_buff *newsk)
2030 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2032 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2035 * remove sk_buff from list. _Must_ be called atomically, and with
2038 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2039 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2041 struct sk_buff *next, *prev;
2043 WRITE_ONCE(list->qlen, list->qlen - 1);
2046 skb->next = skb->prev = NULL;
2047 WRITE_ONCE(next->prev, prev);
2048 WRITE_ONCE(prev->next, next);
2052 * __skb_dequeue - remove from the head of the queue
2053 * @list: list to dequeue from
2055 * Remove the head of the list. This function does not take any locks
2056 * so must be used with appropriate locks held only. The head item is
2057 * returned or %NULL if the list is empty.
2059 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2061 struct sk_buff *skb = skb_peek(list);
2063 __skb_unlink(skb, list);
2066 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2069 * __skb_dequeue_tail - remove from the tail of the queue
2070 * @list: list to dequeue from
2072 * Remove the tail of the list. This function does not take any locks
2073 * so must be used with appropriate locks held only. The tail item is
2074 * returned or %NULL if the list is empty.
2076 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2078 struct sk_buff *skb = skb_peek_tail(list);
2080 __skb_unlink(skb, list);
2083 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2086 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2088 return skb->data_len;
2091 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2093 return skb->len - skb->data_len;
2096 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2098 unsigned int i, len = 0;
2100 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2101 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2105 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2107 return skb_headlen(skb) + __skb_pagelen(skb);
2111 * __skb_fill_page_desc - initialise a paged fragment in an skb
2112 * @skb: buffer containing fragment to be initialised
2113 * @i: paged fragment index to initialise
2114 * @page: the page to use for this fragment
2115 * @off: the offset to the data with @page
2116 * @size: the length of the data
2118 * Initialises the @i'th fragment of @skb to point to &size bytes at
2119 * offset @off within @page.
2121 * Does not take any additional reference on the fragment.
2123 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2124 struct page *page, int off, int size)
2126 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2129 * Propagate page pfmemalloc to the skb if we can. The problem is
2130 * that not all callers have unique ownership of the page but rely
2131 * on page_is_pfmemalloc doing the right thing(tm).
2133 frag->bv_page = page;
2134 frag->bv_offset = off;
2135 skb_frag_size_set(frag, size);
2137 page = compound_head(page);
2138 if (page_is_pfmemalloc(page))
2139 skb->pfmemalloc = true;
2143 * skb_fill_page_desc - initialise a paged fragment in an skb
2144 * @skb: buffer containing fragment to be initialised
2145 * @i: paged fragment index to initialise
2146 * @page: the page to use for this fragment
2147 * @off: the offset to the data with @page
2148 * @size: the length of the data
2150 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2151 * @skb to point to @size bytes at offset @off within @page. In
2152 * addition updates @skb such that @i is the last fragment.
2154 * Does not take any additional reference on the fragment.
2156 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2157 struct page *page, int off, int size)
2159 __skb_fill_page_desc(skb, i, page, off, size);
2160 skb_shinfo(skb)->nr_frags = i + 1;
2163 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2164 int size, unsigned int truesize);
2166 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2167 unsigned int truesize);
2169 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2171 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2172 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2174 return skb->head + skb->tail;
2177 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2179 skb->tail = skb->data - skb->head;
2182 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2184 skb_reset_tail_pointer(skb);
2185 skb->tail += offset;
2188 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2189 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2194 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2196 skb->tail = skb->data;
2199 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2201 skb->tail = skb->data + offset;
2204 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2206 static inline void skb_assert_len(struct sk_buff *skb)
2208 #ifdef CONFIG_DEBUG_NET
2209 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2210 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2211 #endif /* CONFIG_DEBUG_NET */
2215 * Add data to an sk_buff
2217 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2218 void *skb_put(struct sk_buff *skb, unsigned int len);
2219 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2221 void *tmp = skb_tail_pointer(skb);
2222 SKB_LINEAR_ASSERT(skb);
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);
2236 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2239 void *tmp = __skb_put(skb, len);
2241 memcpy(tmp, data, len);
2245 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2247 *(u8 *)__skb_put(skb, 1) = val;
2250 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2252 void *tmp = skb_put(skb, len);
2254 memset(tmp, 0, len);
2259 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2262 void *tmp = skb_put(skb, len);
2264 memcpy(tmp, data, len);
2269 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2271 *(u8 *)skb_put(skb, 1) = val;
2274 void *skb_push(struct sk_buff *skb, unsigned int len);
2275 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2282 void *skb_pull(struct sk_buff *skb, unsigned int len);
2283 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2286 BUG_ON(skb->len < skb->data_len);
2287 return skb->data += len;
2290 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2292 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2295 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2297 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2299 if (len > skb_headlen(skb) &&
2300 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2303 return skb->data += len;
2306 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2308 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2311 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2313 if (likely(len <= skb_headlen(skb)))
2315 if (unlikely(len > skb->len))
2317 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2320 void skb_condense(struct sk_buff *skb);
2323 * skb_headroom - bytes at buffer head
2324 * @skb: buffer to check
2326 * Return the number of bytes of free space at the head of an &sk_buff.
2328 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2330 return skb->data - skb->head;
2334 * skb_tailroom - bytes at buffer end
2335 * @skb: buffer to check
2337 * Return the number of bytes of free space at the tail of an sk_buff
2339 static inline int skb_tailroom(const struct sk_buff *skb)
2341 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2345 * skb_availroom - bytes at buffer end
2346 * @skb: buffer to check
2348 * Return the number of bytes of free space at the tail of an sk_buff
2349 * allocated by sk_stream_alloc()
2351 static inline int skb_availroom(const struct sk_buff *skb)
2353 if (skb_is_nonlinear(skb))
2356 return skb->end - skb->tail - skb->reserved_tailroom;
2360 * skb_reserve - adjust headroom
2361 * @skb: buffer to alter
2362 * @len: bytes to move
2364 * Increase the headroom of an empty &sk_buff by reducing the tail
2365 * room. This is only allowed for an empty buffer.
2367 static inline void skb_reserve(struct sk_buff *skb, int len)
2374 * skb_tailroom_reserve - adjust reserved_tailroom
2375 * @skb: buffer to alter
2376 * @mtu: maximum amount of headlen permitted
2377 * @needed_tailroom: minimum amount of reserved_tailroom
2379 * Set reserved_tailroom so that headlen can be as large as possible but
2380 * not larger than mtu and tailroom cannot be smaller than
2382 * The required headroom should already have been reserved before using
2385 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2386 unsigned int needed_tailroom)
2388 SKB_LINEAR_ASSERT(skb);
2389 if (mtu < skb_tailroom(skb) - needed_tailroom)
2390 /* use at most mtu */
2391 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2393 /* use up to all available space */
2394 skb->reserved_tailroom = needed_tailroom;
2397 #define ENCAP_TYPE_ETHER 0
2398 #define ENCAP_TYPE_IPPROTO 1
2400 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2403 skb->inner_protocol = protocol;
2404 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2407 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2410 skb->inner_ipproto = ipproto;
2411 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2414 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2416 skb->inner_mac_header = skb->mac_header;
2417 skb->inner_network_header = skb->network_header;
2418 skb->inner_transport_header = skb->transport_header;
2421 static inline void skb_reset_mac_len(struct sk_buff *skb)
2423 skb->mac_len = skb->network_header - skb->mac_header;
2426 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2429 return skb->head + skb->inner_transport_header;
2432 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2434 return skb_inner_transport_header(skb) - skb->data;
2437 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2439 skb->inner_transport_header = skb->data - skb->head;
2442 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2445 skb_reset_inner_transport_header(skb);
2446 skb->inner_transport_header += offset;
2449 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2451 return skb->head + skb->inner_network_header;
2454 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2456 skb->inner_network_header = skb->data - skb->head;
2459 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2462 skb_reset_inner_network_header(skb);
2463 skb->inner_network_header += offset;
2466 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2468 return skb->head + skb->inner_mac_header;
2471 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2473 skb->inner_mac_header = skb->data - skb->head;
2476 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2479 skb_reset_inner_mac_header(skb);
2480 skb->inner_mac_header += offset;
2482 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2484 return skb->transport_header != (typeof(skb->transport_header))~0U;
2487 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2489 return skb->head + skb->transport_header;
2492 static inline void skb_reset_transport_header(struct sk_buff *skb)
2494 skb->transport_header = skb->data - skb->head;
2497 static inline void skb_set_transport_header(struct sk_buff *skb,
2500 skb_reset_transport_header(skb);
2501 skb->transport_header += offset;
2504 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2506 return skb->head + skb->network_header;
2509 static inline void skb_reset_network_header(struct sk_buff *skb)
2511 skb->network_header = skb->data - skb->head;
2514 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2516 skb_reset_network_header(skb);
2517 skb->network_header += offset;
2520 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2522 return skb->head + skb->mac_header;
2525 static inline int skb_mac_offset(const struct sk_buff *skb)
2527 return skb_mac_header(skb) - skb->data;
2530 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2532 return skb->network_header - skb->mac_header;
2535 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2537 return skb->mac_header != (typeof(skb->mac_header))~0U;
2540 static inline void skb_reset_mac_header(struct sk_buff *skb)
2542 skb->mac_header = skb->data - skb->head;
2545 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2547 skb_reset_mac_header(skb);
2548 skb->mac_header += offset;
2551 static inline void skb_pop_mac_header(struct sk_buff *skb)
2553 skb->mac_header = skb->network_header;
2556 static inline void skb_probe_transport_header(struct sk_buff *skb)
2558 struct flow_keys_basic keys;
2560 if (skb_transport_header_was_set(skb))
2563 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2565 skb_set_transport_header(skb, keys.control.thoff);
2568 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2570 if (skb_mac_header_was_set(skb)) {
2571 const unsigned char *old_mac = skb_mac_header(skb);
2573 skb_set_mac_header(skb, -skb->mac_len);
2574 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2578 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2580 return skb->csum_start - skb_headroom(skb);
2583 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2585 return skb->head + skb->csum_start;
2588 static inline int skb_transport_offset(const struct sk_buff *skb)
2590 return skb_transport_header(skb) - skb->data;
2593 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2595 return skb->transport_header - skb->network_header;
2598 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2600 return skb->inner_transport_header - skb->inner_network_header;
2603 static inline int skb_network_offset(const struct sk_buff *skb)
2605 return skb_network_header(skb) - skb->data;
2608 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2610 return skb_inner_network_header(skb) - skb->data;
2613 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2615 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2619 * CPUs often take a performance hit when accessing unaligned memory
2620 * locations. The actual performance hit varies, it can be small if the
2621 * hardware handles it or large if we have to take an exception and fix it
2624 * Since an ethernet header is 14 bytes network drivers often end up with
2625 * the IP header at an unaligned offset. The IP header can be aligned by
2626 * shifting the start of the packet by 2 bytes. Drivers should do this
2629 * skb_reserve(skb, NET_IP_ALIGN);
2631 * The downside to this alignment of the IP header is that the DMA is now
2632 * unaligned. On some architectures the cost of an unaligned DMA is high
2633 * and this cost outweighs the gains made by aligning the IP header.
2635 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2638 #ifndef NET_IP_ALIGN
2639 #define NET_IP_ALIGN 2
2643 * The networking layer reserves some headroom in skb data (via
2644 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2645 * the header has to grow. In the default case, if the header has to grow
2646 * 32 bytes or less we avoid the reallocation.
2648 * Unfortunately this headroom changes the DMA alignment of the resulting
2649 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2650 * on some architectures. An architecture can override this value,
2651 * perhaps setting it to a cacheline in size (since that will maintain
2652 * cacheline alignment of the DMA). It must be a power of 2.
2654 * Various parts of the networking layer expect at least 32 bytes of
2655 * headroom, you should not reduce this.
2657 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2658 * to reduce average number of cache lines per packet.
2659 * get_rps_cpus() for example only access one 64 bytes aligned block :
2660 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2663 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2666 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2668 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2670 if (WARN_ON(skb_is_nonlinear(skb)))
2673 skb_set_tail_pointer(skb, len);
2676 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2678 __skb_set_length(skb, len);
2681 void skb_trim(struct sk_buff *skb, unsigned int len);
2683 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2686 return ___pskb_trim(skb, len);
2687 __skb_trim(skb, len);
2691 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2693 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2697 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2698 * @skb: buffer to alter
2701 * This is identical to pskb_trim except that the caller knows that
2702 * the skb is not cloned so we should never get an error due to out-
2705 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2707 int err = pskb_trim(skb, len);
2711 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2713 unsigned int diff = len - skb->len;
2715 if (skb_tailroom(skb) < diff) {
2716 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2721 __skb_set_length(skb, len);
2726 * skb_orphan - orphan a buffer
2727 * @skb: buffer to orphan
2729 * If a buffer currently has an owner then we call the owner's
2730 * destructor function and make the @skb unowned. The buffer continues
2731 * to exist but is no longer charged to its former owner.
2733 static inline void skb_orphan(struct sk_buff *skb)
2735 if (skb->destructor) {
2736 skb->destructor(skb);
2737 skb->destructor = NULL;
2745 * skb_orphan_frags - orphan the frags contained in a buffer
2746 * @skb: buffer to orphan frags from
2747 * @gfp_mask: allocation mask for replacement pages
2749 * For each frag in the SKB which needs a destructor (i.e. has an
2750 * owner) create a copy of that frag and release the original
2751 * page by calling the destructor.
2753 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2755 if (likely(!skb_zcopy(skb)))
2757 if (!skb_zcopy_is_nouarg(skb) &&
2758 skb_uarg(skb)->callback == sock_zerocopy_callback)
2760 return skb_copy_ubufs(skb, gfp_mask);
2763 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2764 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2766 if (likely(!skb_zcopy(skb)))
2768 return skb_copy_ubufs(skb, gfp_mask);
2772 * __skb_queue_purge - empty a list
2773 * @list: list to empty
2775 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2776 * the list and one reference dropped. This function does not take the
2777 * list lock and the caller must hold the relevant locks to use it.
2779 static inline void __skb_queue_purge(struct sk_buff_head *list)
2781 struct sk_buff *skb;
2782 while ((skb = __skb_dequeue(list)) != NULL)
2785 void skb_queue_purge(struct sk_buff_head *list);
2787 unsigned int skb_rbtree_purge(struct rb_root *root);
2789 void *netdev_alloc_frag(unsigned int fragsz);
2791 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2795 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2796 * @dev: network device to receive on
2797 * @length: length to allocate
2799 * Allocate a new &sk_buff and assign it a usage count of one. The
2800 * buffer has unspecified headroom built in. Users should allocate
2801 * the headroom they think they need without accounting for the
2802 * built in space. The built in space is used for optimisations.
2804 * %NULL is returned if there is no free memory. Although this function
2805 * allocates memory it can be called from an interrupt.
2807 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2808 unsigned int length)
2810 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2813 /* legacy helper around __netdev_alloc_skb() */
2814 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2817 return __netdev_alloc_skb(NULL, length, gfp_mask);
2820 /* legacy helper around netdev_alloc_skb() */
2821 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2823 return netdev_alloc_skb(NULL, length);
2827 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2828 unsigned int length, gfp_t gfp)
2830 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2832 if (NET_IP_ALIGN && skb)
2833 skb_reserve(skb, NET_IP_ALIGN);
2837 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2838 unsigned int length)
2840 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2843 static inline void skb_free_frag(void *addr)
2845 page_frag_free(addr);
2848 void *napi_alloc_frag(unsigned int fragsz);
2849 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2850 unsigned int length, gfp_t gfp_mask);
2851 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2852 unsigned int length)
2854 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2856 void napi_consume_skb(struct sk_buff *skb, int budget);
2858 void __kfree_skb_flush(void);
2859 void __kfree_skb_defer(struct sk_buff *skb);
2862 * __dev_alloc_pages - allocate page for network Rx
2863 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2864 * @order: size of the allocation
2866 * Allocate a new page.
2868 * %NULL is returned if there is no free memory.
2870 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2873 /* This piece of code contains several assumptions.
2874 * 1. This is for device Rx, therefor a cold page is preferred.
2875 * 2. The expectation is the user wants a compound page.
2876 * 3. If requesting a order 0 page it will not be compound
2877 * due to the check to see if order has a value in prep_new_page
2878 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2879 * code in gfp_to_alloc_flags that should be enforcing this.
2881 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2883 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2886 static inline struct page *dev_alloc_pages(unsigned int order)
2888 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2892 * __dev_alloc_page - allocate a page for network Rx
2893 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2895 * Allocate a new page.
2897 * %NULL is returned if there is no free memory.
2899 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2901 return __dev_alloc_pages(gfp_mask, 0);
2904 static inline struct page *dev_alloc_page(void)
2906 return dev_alloc_pages(0);
2910 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2911 * @page: The page that was allocated from skb_alloc_page
2912 * @skb: The skb that may need pfmemalloc set
2914 static inline void skb_propagate_pfmemalloc(struct page *page,
2915 struct sk_buff *skb)
2917 if (page_is_pfmemalloc(page))
2918 skb->pfmemalloc = true;
2922 * skb_frag_off() - Returns the offset of a skb fragment
2923 * @frag: the paged fragment
2925 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
2927 return frag->bv_offset;
2931 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
2932 * @frag: skb fragment
2933 * @delta: value to add
2935 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
2937 frag->bv_offset += delta;
2941 * skb_frag_off_set() - Sets the offset of a skb fragment
2942 * @frag: skb fragment
2943 * @offset: offset of fragment
2945 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
2947 frag->bv_offset = offset;
2951 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
2952 * @fragto: skb fragment where offset is set
2953 * @fragfrom: skb fragment offset is copied from
2955 static inline void skb_frag_off_copy(skb_frag_t *fragto,
2956 const skb_frag_t *fragfrom)
2958 fragto->bv_offset = fragfrom->bv_offset;
2962 * skb_frag_page - retrieve the page referred to by a paged fragment
2963 * @frag: the paged fragment
2965 * Returns the &struct page associated with @frag.
2967 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2969 return frag->bv_page;
2973 * __skb_frag_ref - take an addition reference on a paged fragment.
2974 * @frag: the paged fragment
2976 * Takes an additional reference on the paged fragment @frag.
2978 static inline void __skb_frag_ref(skb_frag_t *frag)
2980 get_page(skb_frag_page(frag));
2984 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2986 * @f: the fragment offset.
2988 * Takes an additional reference on the @f'th paged fragment of @skb.
2990 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2992 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2996 * __skb_frag_unref - release a reference on a paged fragment.
2997 * @frag: the paged fragment
2999 * Releases a reference on the paged fragment @frag.
3001 static inline void __skb_frag_unref(skb_frag_t *frag)
3003 put_page(skb_frag_page(frag));
3007 * skb_frag_unref - release a reference on a paged fragment of an skb.
3009 * @f: the fragment offset
3011 * Releases a reference on the @f'th paged fragment of @skb.
3013 static inline void skb_frag_unref(struct sk_buff *skb, int f)
3015 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
3019 * skb_frag_address - gets the address of the data contained in a paged fragment
3020 * @frag: the paged fragment buffer
3022 * Returns the address of the data within @frag. The page must already
3025 static inline void *skb_frag_address(const skb_frag_t *frag)
3027 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3031 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3032 * @frag: the paged fragment buffer
3034 * Returns the address of the data within @frag. Checks that the page
3035 * is mapped and returns %NULL otherwise.
3037 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3039 void *ptr = page_address(skb_frag_page(frag));
3043 return ptr + skb_frag_off(frag);
3047 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3048 * @fragto: skb fragment where page is set
3049 * @fragfrom: skb fragment page is copied from
3051 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3052 const skb_frag_t *fragfrom)
3054 fragto->bv_page = fragfrom->bv_page;
3058 * __skb_frag_set_page - sets the page contained in a paged fragment
3059 * @frag: the paged fragment
3060 * @page: the page to set
3062 * Sets the fragment @frag to contain @page.
3064 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3066 frag->bv_page = page;
3070 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3072 * @f: the fragment offset
3073 * @page: the page to set
3075 * Sets the @f'th fragment of @skb to contain @page.
3077 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3080 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3083 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3086 * skb_frag_dma_map - maps a paged fragment via the DMA API
3087 * @dev: the device to map the fragment to
3088 * @frag: the paged fragment to map
3089 * @offset: the offset within the fragment (starting at the
3090 * fragment's own offset)
3091 * @size: the number of bytes to map
3092 * @dir: the direction of the mapping (``PCI_DMA_*``)
3094 * Maps the page associated with @frag to @device.
3096 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3097 const skb_frag_t *frag,
3098 size_t offset, size_t size,
3099 enum dma_data_direction dir)
3101 return dma_map_page(dev, skb_frag_page(frag),
3102 skb_frag_off(frag) + offset, size, dir);
3105 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3108 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3112 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3115 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3120 * skb_clone_writable - is the header of a clone writable
3121 * @skb: buffer to check
3122 * @len: length up to which to write
3124 * Returns true if modifying the header part of the cloned buffer
3125 * does not requires the data to be copied.
3127 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3129 return !skb_header_cloned(skb) &&
3130 skb_headroom(skb) + len <= skb->hdr_len;
3133 static inline int skb_try_make_writable(struct sk_buff *skb,
3134 unsigned int write_len)
3136 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3137 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3140 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3145 if (headroom > skb_headroom(skb))
3146 delta = headroom - skb_headroom(skb);
3148 if (delta || cloned)
3149 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3155 * skb_cow - copy header of skb when it is required
3156 * @skb: buffer to cow
3157 * @headroom: needed headroom
3159 * If the skb passed lacks sufficient headroom or its data part
3160 * is shared, data is reallocated. If reallocation fails, an error
3161 * is returned and original skb is not changed.
3163 * The result is skb with writable area skb->head...skb->tail
3164 * and at least @headroom of space at head.
3166 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3168 return __skb_cow(skb, headroom, skb_cloned(skb));
3172 * skb_cow_head - skb_cow but only making the head writable
3173 * @skb: buffer to cow
3174 * @headroom: needed headroom
3176 * This function is identical to skb_cow except that we replace the
3177 * skb_cloned check by skb_header_cloned. It should be used when
3178 * you only need to push on some header and do not need to modify
3181 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3183 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3187 * skb_padto - pad an skbuff up to a minimal size
3188 * @skb: buffer to pad
3189 * @len: minimal length
3191 * Pads up a buffer to ensure the trailing bytes exist and are
3192 * blanked. If the buffer already contains sufficient data it
3193 * is untouched. Otherwise it is extended. Returns zero on
3194 * success. The skb is freed on error.
3196 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3198 unsigned int size = skb->len;
3199 if (likely(size >= len))
3201 return skb_pad(skb, len - size);
3205 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3206 * @skb: buffer to pad
3207 * @len: minimal length
3208 * @free_on_error: free buffer on error
3210 * Pads up a buffer to ensure the trailing bytes exist and are
3211 * blanked. If the buffer already contains sufficient data it
3212 * is untouched. Otherwise it is extended. Returns zero on
3213 * success. The skb is freed on error if @free_on_error is true.
3215 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3219 unsigned int size = skb->len;
3221 if (unlikely(size < len)) {
3223 if (__skb_pad(skb, len, free_on_error))
3225 __skb_put(skb, len);
3231 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3232 * @skb: buffer to pad
3233 * @len: minimal length
3235 * Pads up a buffer to ensure the trailing bytes exist and are
3236 * blanked. If the buffer already contains sufficient data it
3237 * is untouched. Otherwise it is extended. Returns zero on
3238 * success. The skb is freed on error.
3240 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3242 return __skb_put_padto(skb, len, true);
3245 static inline int skb_add_data(struct sk_buff *skb,
3246 struct iov_iter *from, int copy)
3248 const int off = skb->len;
3250 if (skb->ip_summed == CHECKSUM_NONE) {
3252 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3254 skb->csum = csum_block_add(skb->csum, csum, off);
3257 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3260 __skb_trim(skb, off);
3264 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3265 const struct page *page, int off)
3270 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3272 return page == skb_frag_page(frag) &&
3273 off == skb_frag_off(frag) + skb_frag_size(frag);
3278 static inline int __skb_linearize(struct sk_buff *skb)
3280 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3284 * skb_linearize - convert paged skb to linear one
3285 * @skb: buffer to linarize
3287 * If there is no free memory -ENOMEM is returned, otherwise zero
3288 * is returned and the old skb data released.
3290 static inline int skb_linearize(struct sk_buff *skb)
3292 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3296 * skb_has_shared_frag - can any frag be overwritten
3297 * @skb: buffer to test
3299 * Return true if the skb has at least one frag that might be modified
3300 * by an external entity (as in vmsplice()/sendfile())
3302 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3304 return skb_is_nonlinear(skb) &&
3305 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3309 * skb_linearize_cow - make sure skb is linear and writable
3310 * @skb: buffer to process
3312 * If there is no free memory -ENOMEM is returned, otherwise zero
3313 * is returned and the old skb data released.
3315 static inline int skb_linearize_cow(struct sk_buff *skb)
3317 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3318 __skb_linearize(skb) : 0;
3321 static __always_inline void
3322 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3325 if (skb->ip_summed == CHECKSUM_COMPLETE)
3326 skb->csum = csum_block_sub(skb->csum,
3327 csum_partial(start, len, 0), off);
3328 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3329 skb_checksum_start_offset(skb) < 0)
3330 skb->ip_summed = CHECKSUM_NONE;
3334 * skb_postpull_rcsum - update checksum for received skb after pull
3335 * @skb: buffer to update
3336 * @start: start of data before pull
3337 * @len: length of data pulled
3339 * After doing a pull on a received packet, you need to call this to
3340 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3341 * CHECKSUM_NONE so that it can be recomputed from scratch.
3343 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3344 const void *start, unsigned int len)
3346 __skb_postpull_rcsum(skb, start, len, 0);
3349 static __always_inline void
3350 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3353 if (skb->ip_summed == CHECKSUM_COMPLETE)
3354 skb->csum = csum_block_add(skb->csum,
3355 csum_partial(start, len, 0), off);
3359 * skb_postpush_rcsum - update checksum for received skb after push
3360 * @skb: buffer to update
3361 * @start: start of data after push
3362 * @len: length of data pushed
3364 * After doing a push on a received packet, you need to call this to
3365 * update the CHECKSUM_COMPLETE checksum.
3367 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3368 const void *start, unsigned int len)
3370 __skb_postpush_rcsum(skb, start, len, 0);
3373 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3376 * skb_push_rcsum - push skb and update receive checksum
3377 * @skb: buffer to update
3378 * @len: length of data pulled
3380 * This function performs an skb_push on the packet and updates
3381 * the CHECKSUM_COMPLETE checksum. It should be used on
3382 * receive path processing instead of skb_push unless you know
3383 * that the checksum difference is zero (e.g., a valid IP header)
3384 * or you are setting ip_summed to CHECKSUM_NONE.
3386 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3389 skb_postpush_rcsum(skb, skb->data, len);
3393 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3395 * pskb_trim_rcsum - trim received skb and update checksum
3396 * @skb: buffer to trim
3399 * This is exactly the same as pskb_trim except that it ensures the
3400 * checksum of received packets are still valid after the operation.
3401 * It can change skb pointers.
3404 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3406 if (likely(len >= skb->len))
3408 return pskb_trim_rcsum_slow(skb, len);
3411 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3413 if (skb->ip_summed == CHECKSUM_COMPLETE)
3414 skb->ip_summed = CHECKSUM_NONE;
3415 __skb_trim(skb, len);
3419 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3421 if (skb->ip_summed == CHECKSUM_COMPLETE)
3422 skb->ip_summed = CHECKSUM_NONE;
3423 return __skb_grow(skb, len);
3426 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3427 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3428 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3429 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3430 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3432 #define skb_queue_walk(queue, skb) \
3433 for (skb = (queue)->next; \
3434 skb != (struct sk_buff *)(queue); \
3437 #define skb_queue_walk_safe(queue, skb, tmp) \
3438 for (skb = (queue)->next, tmp = skb->next; \
3439 skb != (struct sk_buff *)(queue); \
3440 skb = tmp, tmp = skb->next)
3442 #define skb_queue_walk_from(queue, skb) \
3443 for (; skb != (struct sk_buff *)(queue); \
3446 #define skb_rbtree_walk(skb, root) \
3447 for (skb = skb_rb_first(root); skb != NULL; \
3448 skb = skb_rb_next(skb))
3450 #define skb_rbtree_walk_from(skb) \
3451 for (; skb != NULL; \
3452 skb = skb_rb_next(skb))
3454 #define skb_rbtree_walk_from_safe(skb, tmp) \
3455 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3458 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3459 for (tmp = skb->next; \
3460 skb != (struct sk_buff *)(queue); \
3461 skb = tmp, tmp = skb->next)
3463 #define skb_queue_reverse_walk(queue, skb) \
3464 for (skb = (queue)->prev; \
3465 skb != (struct sk_buff *)(queue); \
3468 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3469 for (skb = (queue)->prev, tmp = skb->prev; \
3470 skb != (struct sk_buff *)(queue); \
3471 skb = tmp, tmp = skb->prev)
3473 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3474 for (tmp = skb->prev; \
3475 skb != (struct sk_buff *)(queue); \
3476 skb = tmp, tmp = skb->prev)
3478 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3480 return skb_shinfo(skb)->frag_list != NULL;
3483 static inline void skb_frag_list_init(struct sk_buff *skb)
3485 skb_shinfo(skb)->frag_list = NULL;
3488 #define skb_walk_frags(skb, iter) \
3489 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3492 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3493 const struct sk_buff *skb);
3494 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3495 struct sk_buff_head *queue,
3497 void (*destructor)(struct sock *sk,
3498 struct sk_buff *skb),
3500 struct sk_buff **last);
3501 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3502 void (*destructor)(struct sock *sk,
3503 struct sk_buff *skb),
3505 struct sk_buff **last);
3506 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3507 void (*destructor)(struct sock *sk,
3508 struct sk_buff *skb),
3509 int *off, int *err);
3510 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3512 __poll_t datagram_poll(struct file *file, struct socket *sock,
3513 struct poll_table_struct *wait);
3514 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3515 struct iov_iter *to, int size);
3516 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3517 struct msghdr *msg, int size)
3519 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3521 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3522 struct msghdr *msg);
3523 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3524 struct iov_iter *to, int len,
3525 struct ahash_request *hash);
3526 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3527 struct iov_iter *from, int len);
3528 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3529 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3530 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3531 static inline void skb_free_datagram_locked(struct sock *sk,
3532 struct sk_buff *skb)
3534 __skb_free_datagram_locked(sk, skb, 0);
3536 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3537 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3538 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3539 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3540 int len, __wsum csum);
3541 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3542 struct pipe_inode_info *pipe, unsigned int len,
3543 unsigned int flags);
3544 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3546 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3547 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3548 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3550 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3551 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3552 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3553 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3554 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3555 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3556 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3557 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3558 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3559 int skb_vlan_pop(struct sk_buff *skb);
3560 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3561 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
3562 int mac_len, bool ethernet);
3563 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
3565 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3566 int skb_mpls_dec_ttl(struct sk_buff *skb);
3567 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3570 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3572 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3575 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3577 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3580 struct skb_checksum_ops {
3581 __wsum (*update)(const void *mem, int len, __wsum wsum);
3582 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3585 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3587 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3588 __wsum csum, const struct skb_checksum_ops *ops);
3589 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3592 static inline void * __must_check
3593 __skb_header_pointer(const struct sk_buff *skb, int offset,
3594 int len, void *data, int hlen, void *buffer)
3596 if (hlen - offset >= len)
3597 return data + offset;
3600 skb_copy_bits(skb, offset, buffer, len) < 0)
3606 static inline void * __must_check
3607 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3609 return __skb_header_pointer(skb, offset, len, skb->data,
3610 skb_headlen(skb), buffer);
3614 * skb_needs_linearize - check if we need to linearize a given skb
3615 * depending on the given device features.
3616 * @skb: socket buffer to check
3617 * @features: net device features
3619 * Returns true if either:
3620 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3621 * 2. skb is fragmented and the device does not support SG.
3623 static inline bool skb_needs_linearize(struct sk_buff *skb,
3624 netdev_features_t features)
3626 return skb_is_nonlinear(skb) &&
3627 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3628 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3631 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3633 const unsigned int len)
3635 memcpy(to, skb->data, len);
3638 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3639 const int offset, void *to,
3640 const unsigned int len)
3642 memcpy(to, skb->data + offset, len);
3645 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3647 const unsigned int len)
3649 memcpy(skb->data, from, len);
3652 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3655 const unsigned int len)
3657 memcpy(skb->data + offset, from, len);
3660 void skb_init(void);
3662 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3668 * skb_get_timestamp - get timestamp from a skb
3669 * @skb: skb to get stamp from
3670 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
3672 * Timestamps are stored in the skb as offsets to a base timestamp.
3673 * This function converts the offset back to a struct timeval and stores
3676 static inline void skb_get_timestamp(const struct sk_buff *skb,
3677 struct __kernel_old_timeval *stamp)
3679 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
3682 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
3683 struct __kernel_sock_timeval *stamp)
3685 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3687 stamp->tv_sec = ts.tv_sec;
3688 stamp->tv_usec = ts.tv_nsec / 1000;
3691 static inline void skb_get_timestampns(const struct sk_buff *skb,
3692 struct timespec *stamp)
3694 *stamp = ktime_to_timespec(skb->tstamp);
3697 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
3698 struct __kernel_timespec *stamp)
3700 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3702 stamp->tv_sec = ts.tv_sec;
3703 stamp->tv_nsec = ts.tv_nsec;
3706 static inline void __net_timestamp(struct sk_buff *skb)
3708 skb->tstamp = ktime_get_real();
3711 static inline ktime_t net_timedelta(ktime_t t)
3713 return ktime_sub(ktime_get_real(), t);
3716 static inline ktime_t net_invalid_timestamp(void)
3721 static inline u8 skb_metadata_len(const struct sk_buff *skb)
3723 return skb_shinfo(skb)->meta_len;
3726 static inline void *skb_metadata_end(const struct sk_buff *skb)
3728 return skb_mac_header(skb);
3731 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3732 const struct sk_buff *skb_b,
3735 const void *a = skb_metadata_end(skb_a);
3736 const void *b = skb_metadata_end(skb_b);
3737 /* Using more efficient varaiant than plain call to memcmp(). */
3738 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3742 #define __it(x, op) (x -= sizeof(u##op))
3743 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3744 case 32: diffs |= __it_diff(a, b, 64);
3746 case 24: diffs |= __it_diff(a, b, 64);
3748 case 16: diffs |= __it_diff(a, b, 64);
3750 case 8: diffs |= __it_diff(a, b, 64);
3752 case 28: diffs |= __it_diff(a, b, 64);
3754 case 20: diffs |= __it_diff(a, b, 64);
3756 case 12: diffs |= __it_diff(a, b, 64);
3758 case 4: diffs |= __it_diff(a, b, 32);
3763 return memcmp(a - meta_len, b - meta_len, meta_len);
3767 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3768 const struct sk_buff *skb_b)
3770 u8 len_a = skb_metadata_len(skb_a);
3771 u8 len_b = skb_metadata_len(skb_b);
3773 if (!(len_a | len_b))
3776 return len_a != len_b ?
3777 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3780 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3782 skb_shinfo(skb)->meta_len = meta_len;
3785 static inline void skb_metadata_clear(struct sk_buff *skb)
3787 skb_metadata_set(skb, 0);
3790 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3792 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3794 void skb_clone_tx_timestamp(struct sk_buff *skb);
3795 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3797 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3799 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3803 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3808 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3811 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3813 * PHY drivers may accept clones of transmitted packets for
3814 * timestamping via their phy_driver.txtstamp method. These drivers
3815 * must call this function to return the skb back to the stack with a
3818 * @skb: clone of the the original outgoing packet
3819 * @hwtstamps: hardware time stamps
3822 void skb_complete_tx_timestamp(struct sk_buff *skb,
3823 struct skb_shared_hwtstamps *hwtstamps);
3825 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3826 struct skb_shared_hwtstamps *hwtstamps,
3827 struct sock *sk, int tstype);
3830 * skb_tstamp_tx - queue clone of skb with send time stamps
3831 * @orig_skb: the original outgoing packet
3832 * @hwtstamps: hardware time stamps, may be NULL if not available
3834 * If the skb has a socket associated, then this function clones the
3835 * skb (thus sharing the actual data and optional structures), stores
3836 * the optional hardware time stamping information (if non NULL) or
3837 * generates a software time stamp (otherwise), then queues the clone
3838 * to the error queue of the socket. Errors are silently ignored.
3840 void skb_tstamp_tx(struct sk_buff *orig_skb,
3841 struct skb_shared_hwtstamps *hwtstamps);
3844 * skb_tx_timestamp() - Driver hook for transmit timestamping
3846 * Ethernet MAC Drivers should call this function in their hard_xmit()
3847 * function immediately before giving the sk_buff to the MAC hardware.
3849 * Specifically, one should make absolutely sure that this function is
3850 * called before TX completion of this packet can trigger. Otherwise
3851 * the packet could potentially already be freed.
3853 * @skb: A socket buffer.
3855 static inline void skb_tx_timestamp(struct sk_buff *skb)
3857 skb_clone_tx_timestamp(skb);
3858 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3859 skb_tstamp_tx(skb, NULL);
3863 * skb_complete_wifi_ack - deliver skb with wifi status
3865 * @skb: the original outgoing packet
3866 * @acked: ack status
3869 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3871 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3872 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3874 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3876 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3878 (skb->ip_summed == CHECKSUM_PARTIAL &&
3879 skb_checksum_start_offset(skb) >= 0));
3883 * skb_checksum_complete - Calculate checksum of an entire packet
3884 * @skb: packet to process
3886 * This function calculates the checksum over the entire packet plus
3887 * the value of skb->csum. The latter can be used to supply the
3888 * checksum of a pseudo header as used by TCP/UDP. It returns the
3891 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3892 * this function can be used to verify that checksum on received
3893 * packets. In that case the function should return zero if the
3894 * checksum is correct. In particular, this function will return zero
3895 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3896 * hardware has already verified the correctness of the checksum.
3898 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3900 return skb_csum_unnecessary(skb) ?
3901 0 : __skb_checksum_complete(skb);
3904 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3906 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3907 if (skb->csum_level == 0)
3908 skb->ip_summed = CHECKSUM_NONE;
3914 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3916 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3917 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3919 } else if (skb->ip_summed == CHECKSUM_NONE) {
3920 skb->ip_summed = CHECKSUM_UNNECESSARY;
3921 skb->csum_level = 0;
3925 /* Check if we need to perform checksum complete validation.
3927 * Returns true if checksum complete is needed, false otherwise
3928 * (either checksum is unnecessary or zero checksum is allowed).
3930 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3934 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3935 skb->csum_valid = 1;
3936 __skb_decr_checksum_unnecessary(skb);
3943 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
3946 #define CHECKSUM_BREAK 76
3948 /* Unset checksum-complete
3950 * Unset checksum complete can be done when packet is being modified
3951 * (uncompressed for instance) and checksum-complete value is
3954 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3956 if (skb->ip_summed == CHECKSUM_COMPLETE)
3957 skb->ip_summed = CHECKSUM_NONE;
3960 /* Validate (init) checksum based on checksum complete.
3963 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3964 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3965 * checksum is stored in skb->csum for use in __skb_checksum_complete
3966 * non-zero: value of invalid checksum
3969 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3973 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3974 if (!csum_fold(csum_add(psum, skb->csum))) {
3975 skb->csum_valid = 1;
3982 if (complete || skb->len <= CHECKSUM_BREAK) {
3985 csum = __skb_checksum_complete(skb);
3986 skb->csum_valid = !csum;
3993 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3998 /* Perform checksum validate (init). Note that this is a macro since we only
3999 * want to calculate the pseudo header which is an input function if necessary.
4000 * First we try to validate without any computation (checksum unnecessary) and
4001 * then calculate based on checksum complete calling the function to compute
4005 * 0: checksum is validated or try to in skb_checksum_complete
4006 * non-zero: value of invalid checksum
4008 #define __skb_checksum_validate(skb, proto, complete, \
4009 zero_okay, check, compute_pseudo) \
4011 __sum16 __ret = 0; \
4012 skb->csum_valid = 0; \
4013 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4014 __ret = __skb_checksum_validate_complete(skb, \
4015 complete, compute_pseudo(skb, proto)); \
4019 #define skb_checksum_init(skb, proto, compute_pseudo) \
4020 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4022 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4023 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4025 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4026 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4028 #define skb_checksum_validate_zero_check(skb, proto, check, \
4030 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4032 #define skb_checksum_simple_validate(skb) \
4033 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4035 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4037 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4040 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4042 skb->csum = ~pseudo;
4043 skb->ip_summed = CHECKSUM_COMPLETE;
4046 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4048 if (__skb_checksum_convert_check(skb)) \
4049 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4052 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4053 u16 start, u16 offset)
4055 skb->ip_summed = CHECKSUM_PARTIAL;
4056 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4057 skb->csum_offset = offset - start;
4060 /* Update skbuf and packet to reflect the remote checksum offload operation.
4061 * When called, ptr indicates the starting point for skb->csum when
4062 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4063 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4065 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4066 int start, int offset, bool nopartial)
4071 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4075 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4076 __skb_checksum_complete(skb);
4077 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4080 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4082 /* Adjust skb->csum since we changed the packet */
4083 skb->csum = csum_add(skb->csum, delta);
4086 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4088 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4089 return (void *)(skb->_nfct & NFCT_PTRMASK);
4095 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4097 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4104 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4106 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4111 #ifdef CONFIG_SKB_EXTENSIONS
4113 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4119 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4122 SKB_EXT_NUM, /* must be last */
4126 * struct skb_ext - sk_buff extensions
4127 * @refcnt: 1 on allocation, deallocated on 0
4128 * @offset: offset to add to @data to obtain extension address
4129 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4130 * @data: start of extension data, variable sized
4132 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4133 * to use 'u8' types while allowing up to 2kb worth of extension data.
4137 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4138 u8 chunks; /* same */
4139 char data[0] __aligned(8);
4142 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4143 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4144 void __skb_ext_put(struct skb_ext *ext);
4146 static inline void skb_ext_put(struct sk_buff *skb)
4148 if (skb->active_extensions)
4149 __skb_ext_put(skb->extensions);
4152 static inline void __skb_ext_copy(struct sk_buff *dst,
4153 const struct sk_buff *src)
4155 dst->active_extensions = src->active_extensions;
4157 if (src->active_extensions) {
4158 struct skb_ext *ext = src->extensions;
4160 refcount_inc(&ext->refcnt);
4161 dst->extensions = ext;
4165 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4168 __skb_ext_copy(dst, src);
4171 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4173 return !!ext->offset[i];
4176 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4178 return skb->active_extensions & (1 << id);
4181 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4183 if (skb_ext_exist(skb, id))
4184 __skb_ext_del(skb, id);
4187 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4189 if (skb_ext_exist(skb, id)) {
4190 struct skb_ext *ext = skb->extensions;
4192 return (void *)ext + (ext->offset[id] << 3);
4198 static inline void skb_ext_reset(struct sk_buff *skb)
4200 if (unlikely(skb->active_extensions)) {
4201 __skb_ext_put(skb->extensions);
4202 skb->active_extensions = 0;
4206 static inline bool skb_has_extensions(struct sk_buff *skb)
4208 return unlikely(skb->active_extensions);
4211 static inline void skb_ext_put(struct sk_buff *skb) {}
4212 static inline void skb_ext_reset(struct sk_buff *skb) {}
4213 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4214 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4215 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4216 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4217 #endif /* CONFIG_SKB_EXTENSIONS */
4219 static inline void nf_reset_ct(struct sk_buff *skb)
4221 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4222 nf_conntrack_put(skb_nfct(skb));
4227 static inline void nf_reset_trace(struct sk_buff *skb)
4229 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4234 static inline void ipvs_reset(struct sk_buff *skb)
4236 #if IS_ENABLED(CONFIG_IP_VS)
4237 skb->ipvs_property = 0;
4241 /* Note: This doesn't put any conntrack info in dst. */
4242 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4245 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4246 dst->_nfct = src->_nfct;
4247 nf_conntrack_get(skb_nfct(src));
4249 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4251 dst->nf_trace = src->nf_trace;
4255 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4257 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4258 nf_conntrack_put(skb_nfct(dst));
4260 __nf_copy(dst, src, true);
4263 #ifdef CONFIG_NETWORK_SECMARK
4264 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4266 to->secmark = from->secmark;
4269 static inline void skb_init_secmark(struct sk_buff *skb)
4274 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4277 static inline void skb_init_secmark(struct sk_buff *skb)
4281 static inline int secpath_exists(const struct sk_buff *skb)
4284 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4290 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4292 return !skb->destructor &&
4293 !secpath_exists(skb) &&
4295 !skb->_skb_refdst &&
4296 !skb_has_frag_list(skb);
4299 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4301 skb->queue_mapping = queue_mapping;
4304 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4306 return skb->queue_mapping;
4309 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4311 to->queue_mapping = from->queue_mapping;
4314 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4316 skb->queue_mapping = rx_queue + 1;
4319 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4321 return skb->queue_mapping - 1;
4324 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4326 return skb->queue_mapping != 0;
4329 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4331 skb->dst_pending_confirm = val;
4334 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4336 return skb->dst_pending_confirm != 0;
4339 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4342 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4348 /* Keeps track of mac header offset relative to skb->head.
4349 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4350 * For non-tunnel skb it points to skb_mac_header() and for
4351 * tunnel skb it points to outer mac header.
4352 * Keeps track of level of encapsulation of network headers.
4363 #define SKB_SGO_CB_OFFSET 32
4364 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
4366 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4368 return (skb_mac_header(inner_skb) - inner_skb->head) -
4369 SKB_GSO_CB(inner_skb)->mac_offset;
4372 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4374 int new_headroom, headroom;
4377 headroom = skb_headroom(skb);
4378 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4382 new_headroom = skb_headroom(skb);
4383 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4387 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4389 /* Do not update partial checksums if remote checksum is enabled. */
4390 if (skb->remcsum_offload)
4393 SKB_GSO_CB(skb)->csum = res;
4394 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4397 /* Compute the checksum for a gso segment. First compute the checksum value
4398 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4399 * then add in skb->csum (checksum from csum_start to end of packet).
4400 * skb->csum and csum_start are then updated to reflect the checksum of the
4401 * resultant packet starting from the transport header-- the resultant checksum
4402 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4405 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4407 unsigned char *csum_start = skb_transport_header(skb);
4408 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4409 __wsum partial = SKB_GSO_CB(skb)->csum;
4411 SKB_GSO_CB(skb)->csum = res;
4412 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4414 return csum_fold(csum_partial(csum_start, plen, partial));
4417 static inline bool skb_is_gso(const struct sk_buff *skb)
4419 return skb_shinfo(skb)->gso_size;
4422 /* Note: Should be called only if skb_is_gso(skb) is true */
4423 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4425 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4428 /* Note: Should be called only if skb_is_gso(skb) is true */
4429 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4431 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4434 /* Note: Should be called only if skb_is_gso(skb) is true */
4435 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4437 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4440 static inline void skb_gso_reset(struct sk_buff *skb)
4442 skb_shinfo(skb)->gso_size = 0;
4443 skb_shinfo(skb)->gso_segs = 0;
4444 skb_shinfo(skb)->gso_type = 0;
4447 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4450 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4452 shinfo->gso_size += increment;
4455 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4458 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4460 shinfo->gso_size -= decrement;
4463 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4465 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4467 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4468 * wanted then gso_type will be set. */
4469 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4471 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4472 unlikely(shinfo->gso_type == 0)) {
4473 __skb_warn_lro_forwarding(skb);
4479 static inline void skb_forward_csum(struct sk_buff *skb)
4481 /* Unfortunately we don't support this one. Any brave souls? */
4482 if (skb->ip_summed == CHECKSUM_COMPLETE)
4483 skb->ip_summed = CHECKSUM_NONE;
4487 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4488 * @skb: skb to check
4490 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4491 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4492 * use this helper, to document places where we make this assertion.
4494 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4497 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4501 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4503 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4504 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4505 unsigned int transport_len,
4506 __sum16(*skb_chkf)(struct sk_buff *skb));
4509 * skb_head_is_locked - Determine if the skb->head is locked down
4510 * @skb: skb to check
4512 * The head on skbs build around a head frag can be removed if they are
4513 * not cloned. This function returns true if the skb head is locked down
4514 * due to either being allocated via kmalloc, or by being a clone with
4515 * multiple references to the head.
4517 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4519 return !skb->head_frag || skb_cloned(skb);
4522 /* Local Checksum Offload.
4523 * Compute outer checksum based on the assumption that the
4524 * inner checksum will be offloaded later.
4525 * See Documentation/networking/checksum-offloads.rst for
4526 * explanation of how this works.
4527 * Fill in outer checksum adjustment (e.g. with sum of outer
4528 * pseudo-header) before calling.
4529 * Also ensure that inner checksum is in linear data area.
4531 static inline __wsum lco_csum(struct sk_buff *skb)
4533 unsigned char *csum_start = skb_checksum_start(skb);
4534 unsigned char *l4_hdr = skb_transport_header(skb);
4537 /* Start with complement of inner checksum adjustment */
4538 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4541 /* Add in checksum of our headers (incl. outer checksum
4542 * adjustment filled in by caller) and return result.
4544 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4547 static inline bool skb_is_redirected(const struct sk_buff *skb)
4549 #ifdef CONFIG_NET_REDIRECT
4550 return skb->redirected;
4556 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
4558 #ifdef CONFIG_NET_REDIRECT
4559 skb->redirected = 1;
4560 skb->from_ingress = from_ingress;
4561 if (skb->from_ingress)
4566 static inline void skb_reset_redirect(struct sk_buff *skb)
4568 #ifdef CONFIG_NET_REDIRECT
4569 skb->redirected = 0;
4573 #endif /* __KERNEL__ */
4574 #endif /* _LINUX_SKBUFF_H */