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 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 set in 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 or 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 in skb->csum. This means 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, but it 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 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, it should treat the packet as if CHECKSUM_NONE were 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 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 that there is no indication in the skbuff that the
204 * CHECKSUM_PARTIAL refers to an FCOE checksum, so 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, and both csum_start and
215 * csum_offset are set to refer to the outermost checksum being offloaded
216 * (two offloaded checksums are possible with UDP encapsulation).
219 /* Don't change this without changing skb_csum_unnecessary! */
220 #define CHECKSUM_NONE 0
221 #define CHECKSUM_UNNECESSARY 1
222 #define CHECKSUM_COMPLETE 2
223 #define CHECKSUM_PARTIAL 3
225 /* Maximum value in skb->csum_level */
226 #define SKB_MAX_CSUM_LEVEL 3
228 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
229 #define SKB_WITH_OVERHEAD(X) \
230 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
231 #define SKB_MAX_ORDER(X, ORDER) \
232 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
233 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
234 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
236 /* return minimum truesize of one skb containing X bytes of data */
237 #define SKB_TRUESIZE(X) ((X) + \
238 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
239 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
243 struct pipe_inode_info;
250 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
251 struct nf_bridge_info {
253 BRNF_PROTO_UNCHANGED,
261 struct net_device *physindev;
263 /* always valid & non-NULL from FORWARD on, for physdev match */
264 struct net_device *physoutdev;
266 /* prerouting: detect dnat in orig/reply direction */
268 struct in6_addr ipv6_daddr;
270 /* after prerouting + nat detected: store original source
271 * mac since neigh resolution overwrites it, only used while
272 * skb is out in neigh layer.
274 char neigh_header[8];
279 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
280 /* Chain in tc_skb_ext will be used to share the tc chain with
281 * ovs recirc_id. It will be set to the current chain by tc
282 * and read by ovs to recirc_id.
289 struct sk_buff_head {
290 /* These two members must be first. */
291 struct sk_buff *next;
292 struct sk_buff *prev;
300 /* To allow 64K frame to be packed as single skb without frag_list we
301 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
302 * buffers which do not start on a page boundary.
304 * Since GRO uses frags we allocate at least 16 regardless of page
307 #if (65536/PAGE_SIZE + 1) < 16
308 #define MAX_SKB_FRAGS 16UL
310 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
312 extern int sysctl_max_skb_frags;
314 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
315 * segment using its current segmentation instead.
317 #define GSO_BY_FRAGS 0xFFFF
319 typedef struct bio_vec skb_frag_t;
322 * skb_frag_size() - Returns the size of a skb fragment
323 * @frag: skb fragment
325 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
331 * skb_frag_size_set() - Sets the size of a skb fragment
332 * @frag: skb fragment
333 * @size: size of fragment
335 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
341 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
342 * @frag: skb fragment
343 * @delta: value to add
345 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
347 frag->bv_len += delta;
351 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
352 * @frag: skb fragment
353 * @delta: value to subtract
355 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
357 frag->bv_len -= delta;
361 * skb_frag_must_loop - Test if %p is a high memory page
362 * @p: fragment's page
364 static inline bool skb_frag_must_loop(struct page *p)
366 #if defined(CONFIG_HIGHMEM)
374 * skb_frag_foreach_page - loop over pages in a fragment
376 * @f: skb frag to operate on
377 * @f_off: offset from start of f->bv_page
378 * @f_len: length from f_off to loop over
379 * @p: (temp var) current page
380 * @p_off: (temp var) offset from start of current page,
381 * non-zero only on first page.
382 * @p_len: (temp var) length in current page,
383 * < PAGE_SIZE only on first and last page.
384 * @copied: (temp var) length so far, excluding current p_len.
386 * A fragment can hold a compound page, in which case per-page
387 * operations, notably kmap_atomic, must be called for each
390 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
391 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
392 p_off = (f_off) & (PAGE_SIZE - 1), \
393 p_len = skb_frag_must_loop(p) ? \
394 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
397 copied += p_len, p++, p_off = 0, \
398 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
400 #define HAVE_HW_TIME_STAMP
403 * struct skb_shared_hwtstamps - hardware time stamps
404 * @hwtstamp: hardware time stamp transformed into duration
405 * since arbitrary point in time
407 * Software time stamps generated by ktime_get_real() are stored in
410 * hwtstamps can only be compared against other hwtstamps from
413 * This structure is attached to packets as part of the
414 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
416 struct skb_shared_hwtstamps {
420 /* Definitions for tx_flags in struct skb_shared_info */
422 /* generate hardware time stamp */
423 SKBTX_HW_TSTAMP = 1 << 0,
425 /* generate software time stamp when queueing packet to NIC */
426 SKBTX_SW_TSTAMP = 1 << 1,
428 /* device driver is going to provide hardware time stamp */
429 SKBTX_IN_PROGRESS = 1 << 2,
431 /* device driver supports TX zero-copy buffers */
432 SKBTX_DEV_ZEROCOPY = 1 << 3,
434 /* generate wifi status information (where possible) */
435 SKBTX_WIFI_STATUS = 1 << 4,
437 /* This indicates at least one fragment might be overwritten
438 * (as in vmsplice(), sendfile() ...)
439 * If we need to compute a TX checksum, we'll need to copy
440 * all frags to avoid possible bad checksum
442 SKBTX_SHARED_FRAG = 1 << 5,
444 /* generate software time stamp when entering packet scheduling */
445 SKBTX_SCHED_TSTAMP = 1 << 6,
448 #define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
449 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
451 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
454 * The callback notifies userspace to release buffers when skb DMA is done in
455 * lower device, the skb last reference should be 0 when calling this.
456 * The zerocopy_success argument is true if zero copy transmit occurred,
457 * false on data copy or out of memory error caused by data copy attempt.
458 * The ctx field is used to track device context.
459 * The desc field is used to track userspace buffer index.
462 void (*callback)(struct ubuf_info *, bool zerocopy_success);
478 struct user_struct *user;
483 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
485 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
486 void mm_unaccount_pinned_pages(struct mmpin *mmp);
488 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
489 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
490 struct ubuf_info *uarg);
492 static inline void sock_zerocopy_get(struct ubuf_info *uarg)
494 refcount_inc(&uarg->refcnt);
497 void sock_zerocopy_put(struct ubuf_info *uarg);
498 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
500 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
502 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len);
503 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
504 struct msghdr *msg, int len,
505 struct ubuf_info *uarg);
507 /* This data is invariant across clones and lives at
508 * the end of the header data, ie. at skb->end.
510 struct skb_shared_info {
515 unsigned short gso_size;
516 /* Warning: this field is not always filled in (UFO)! */
517 unsigned short gso_segs;
518 struct sk_buff *frag_list;
519 struct skb_shared_hwtstamps hwtstamps;
520 unsigned int gso_type;
524 * Warning : all fields before dataref are cleared in __alloc_skb()
528 /* Intermediate layers must ensure that destructor_arg
529 * remains valid until skb destructor */
530 void * destructor_arg;
532 /* must be last field, see pskb_expand_head() */
533 skb_frag_t frags[MAX_SKB_FRAGS];
536 /* We divide dataref into two halves. The higher 16 bits hold references
537 * to the payload part of skb->data. The lower 16 bits hold references to
538 * the entire skb->data. A clone of a headerless skb holds the length of
539 * the header in skb->hdr_len.
541 * All users must obey the rule that the skb->data reference count must be
542 * greater than or equal to the payload reference count.
544 * Holding a reference to the payload part means that the user does not
545 * care about modifications to the header part of skb->data.
547 #define SKB_DATAREF_SHIFT 16
548 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
552 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
553 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
554 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
558 SKB_GSO_TCPV4 = 1 << 0,
560 /* This indicates the skb is from an untrusted source. */
561 SKB_GSO_DODGY = 1 << 1,
563 /* This indicates the tcp segment has CWR set. */
564 SKB_GSO_TCP_ECN = 1 << 2,
566 SKB_GSO_TCP_FIXEDID = 1 << 3,
568 SKB_GSO_TCPV6 = 1 << 4,
570 SKB_GSO_FCOE = 1 << 5,
572 SKB_GSO_GRE = 1 << 6,
574 SKB_GSO_GRE_CSUM = 1 << 7,
576 SKB_GSO_IPXIP4 = 1 << 8,
578 SKB_GSO_IPXIP6 = 1 << 9,
580 SKB_GSO_UDP_TUNNEL = 1 << 10,
582 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
584 SKB_GSO_PARTIAL = 1 << 12,
586 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
588 SKB_GSO_SCTP = 1 << 14,
590 SKB_GSO_ESP = 1 << 15,
592 SKB_GSO_UDP = 1 << 16,
594 SKB_GSO_UDP_L4 = 1 << 17,
596 SKB_GSO_FRAGLIST = 1 << 18,
599 #if BITS_PER_LONG > 32
600 #define NET_SKBUFF_DATA_USES_OFFSET 1
603 #ifdef NET_SKBUFF_DATA_USES_OFFSET
604 typedef unsigned int sk_buff_data_t;
606 typedef unsigned char *sk_buff_data_t;
610 * struct sk_buff - socket buffer
611 * @next: Next buffer in list
612 * @prev: Previous buffer in list
613 * @tstamp: Time we arrived/left
614 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
615 * for retransmit timer
616 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
618 * @sk: Socket we are owned by
619 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
620 * fragmentation management
621 * @dev: Device we arrived on/are leaving by
622 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
623 * @cb: Control buffer. Free for use by every layer. Put private vars here
624 * @_skb_refdst: destination entry (with norefcount bit)
625 * @sp: the security path, used for xfrm
626 * @len: Length of actual data
627 * @data_len: Data length
628 * @mac_len: Length of link layer header
629 * @hdr_len: writable header length of cloned skb
630 * @csum: Checksum (must include start/offset pair)
631 * @csum_start: Offset from skb->head where checksumming should start
632 * @csum_offset: Offset from csum_start where checksum should be stored
633 * @priority: Packet queueing priority
634 * @ignore_df: allow local fragmentation
635 * @cloned: Head may be cloned (check refcnt to be sure)
636 * @ip_summed: Driver fed us an IP checksum
637 * @nohdr: Payload reference only, must not modify header
638 * @pkt_type: Packet class
639 * @fclone: skbuff clone status
640 * @ipvs_property: skbuff is owned by ipvs
641 * @inner_protocol_type: whether the inner protocol is
642 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
643 * @remcsum_offload: remote checksum offload is enabled
644 * @offload_fwd_mark: Packet was L2-forwarded in hardware
645 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
646 * @tc_skip_classify: do not classify packet. set by IFB device
647 * @tc_at_ingress: used within tc_classify to distinguish in/egress
648 * @redirected: packet was redirected by packet classifier
649 * @from_ingress: packet was redirected from the ingress path
650 * @peeked: this packet has been seen already, so stats have been
651 * done for it, don't do them again
652 * @nf_trace: netfilter packet trace flag
653 * @protocol: Packet protocol from driver
654 * @destructor: Destruct function
655 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
656 * @_nfct: Associated connection, if any (with nfctinfo bits)
657 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
658 * @skb_iif: ifindex of device we arrived on
659 * @tc_index: Traffic control index
660 * @hash: the packet hash
661 * @queue_mapping: Queue mapping for multiqueue devices
662 * @head_frag: skb was allocated from page fragments,
663 * not allocated by kmalloc() or vmalloc().
664 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
665 * @active_extensions: active extensions (skb_ext_id types)
666 * @ndisc_nodetype: router type (from link layer)
667 * @ooo_okay: allow the mapping of a socket to a queue to be changed
668 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
670 * @sw_hash: indicates hash was computed in software stack
671 * @wifi_acked_valid: wifi_acked was set
672 * @wifi_acked: whether frame was acked on wifi or not
673 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
674 * @encapsulation: indicates the inner headers in the skbuff are valid
675 * @encap_hdr_csum: software checksum is needed
676 * @csum_valid: checksum is already valid
677 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
678 * @csum_complete_sw: checksum was completed by software
679 * @csum_level: indicates the number of consecutive checksums found in
680 * the packet minus one that have been verified as
681 * CHECKSUM_UNNECESSARY (max 3)
682 * @dst_pending_confirm: need to confirm neighbour
683 * @decrypted: Decrypted SKB
684 * @napi_id: id of the NAPI struct this skb came from
685 * @sender_cpu: (aka @napi_id) source CPU in XPS
686 * @secmark: security marking
687 * @mark: Generic packet mark
688 * @reserved_tailroom: (aka @mark) number of bytes of free space available
689 * at the tail of an sk_buff
690 * @vlan_present: VLAN tag is present
691 * @vlan_proto: vlan encapsulation protocol
692 * @vlan_tci: vlan tag control information
693 * @inner_protocol: Protocol (encapsulation)
694 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
695 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
696 * @inner_transport_header: Inner transport layer header (encapsulation)
697 * @inner_network_header: Network layer header (encapsulation)
698 * @inner_mac_header: Link layer header (encapsulation)
699 * @transport_header: Transport layer header
700 * @network_header: Network layer header
701 * @mac_header: Link layer header
702 * @tail: Tail pointer
704 * @head: Head of buffer
705 * @data: Data head pointer
706 * @truesize: Buffer size
707 * @users: User count - see {datagram,tcp}.c
708 * @extensions: allocated extensions, valid if active_extensions is nonzero
714 /* These two members must be first. */
715 struct sk_buff *next;
716 struct sk_buff *prev;
719 struct net_device *dev;
720 /* Some protocols might use this space to store information,
721 * while device pointer would be NULL.
722 * UDP receive path is one user.
724 unsigned long dev_scratch;
727 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
728 struct list_head list;
733 int ip_defrag_offset;
738 u64 skb_mstamp_ns; /* earliest departure time */
741 * This is the control buffer. It is free to use for every
742 * layer. Please put your private variables there. If you
743 * want to keep them across layers you have to do a skb_clone()
744 * first. This is owned by whoever has the skb queued ATM.
746 char cb[48] __aligned(8);
750 unsigned long _skb_refdst;
751 void (*destructor)(struct sk_buff *skb);
753 struct list_head tcp_tsorted_anchor;
756 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
764 /* Following fields are _not_ copied in __copy_skb_header()
765 * Note that queue_mapping is here mostly to fill a hole.
769 /* if you move cloned around you also must adapt those constants */
770 #ifdef __BIG_ENDIAN_BITFIELD
771 #define CLONED_MASK (1 << 7)
773 #define CLONED_MASK 1
775 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
778 __u8 __cloned_offset[0];
786 #ifdef CONFIG_SKB_EXTENSIONS
787 __u8 active_extensions;
789 /* fields enclosed in headers_start/headers_end are copied
790 * using a single memcpy() in __copy_skb_header()
793 __u32 headers_start[0];
796 /* if you move pkt_type around you also must adapt those constants */
797 #ifdef __BIG_ENDIAN_BITFIELD
798 #define PKT_TYPE_MAX (7 << 5)
800 #define PKT_TYPE_MAX 7
802 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
805 __u8 __pkt_type_offset[0];
815 __u8 wifi_acked_valid:1;
818 /* Indicates the inner headers are valid in the skbuff. */
819 __u8 encapsulation:1;
820 __u8 encap_hdr_csum:1;
823 #ifdef __BIG_ENDIAN_BITFIELD
824 #define PKT_VLAN_PRESENT_BIT 7
826 #define PKT_VLAN_PRESENT_BIT 0
828 #define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset)
830 __u8 __pkt_vlan_present_offset[0];
833 __u8 csum_complete_sw:1;
835 __u8 csum_not_inet:1;
836 __u8 dst_pending_confirm:1;
837 #ifdef CONFIG_IPV6_NDISC_NODETYPE
838 __u8 ndisc_nodetype:2;
841 __u8 ipvs_property:1;
842 __u8 inner_protocol_type:1;
843 __u8 remcsum_offload:1;
844 #ifdef CONFIG_NET_SWITCHDEV
845 __u8 offload_fwd_mark:1;
846 __u8 offload_l3_fwd_mark:1;
848 #ifdef CONFIG_NET_CLS_ACT
849 __u8 tc_skip_classify:1;
850 __u8 tc_at_ingress:1;
852 #ifdef CONFIG_NET_REDIRECT
856 #ifdef CONFIG_TLS_DEVICE
860 #ifdef CONFIG_NET_SCHED
861 __u16 tc_index; /* traffic control index */
876 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
878 unsigned int napi_id;
879 unsigned int sender_cpu;
882 #ifdef CONFIG_NETWORK_SECMARK
888 __u32 reserved_tailroom;
892 __be16 inner_protocol;
896 __u16 inner_transport_header;
897 __u16 inner_network_header;
898 __u16 inner_mac_header;
901 __u16 transport_header;
902 __u16 network_header;
906 __u32 headers_end[0];
909 /* These elements must be at the end, see alloc_skb() for details. */
914 unsigned int truesize;
917 #ifdef CONFIG_SKB_EXTENSIONS
918 /* only useable after checking ->active_extensions != 0 */
919 struct skb_ext *extensions;
925 * Handling routines are only of interest to the kernel
928 #define SKB_ALLOC_FCLONE 0x01
929 #define SKB_ALLOC_RX 0x02
930 #define SKB_ALLOC_NAPI 0x04
933 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
936 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
938 return unlikely(skb->pfmemalloc);
942 * skb might have a dst pointer attached, refcounted or not.
943 * _skb_refdst low order bit is set if refcount was _not_ taken
945 #define SKB_DST_NOREF 1UL
946 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
949 * skb_dst - returns skb dst_entry
952 * Returns skb dst_entry, regardless of reference taken or not.
954 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
956 /* If refdst was not refcounted, check we still are in a
957 * rcu_read_lock section
959 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
960 !rcu_read_lock_held() &&
961 !rcu_read_lock_bh_held());
962 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
966 * skb_dst_set - sets skb dst
970 * Sets skb dst, assuming a reference was taken on dst and should
971 * be released by skb_dst_drop()
973 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
975 skb->_skb_refdst = (unsigned long)dst;
979 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
983 * Sets skb dst, assuming a reference was not taken on dst.
984 * If dst entry is cached, we do not take reference and dst_release
985 * will be avoided by refdst_drop. If dst entry is not cached, we take
986 * reference, so that last dst_release can destroy the dst immediately.
988 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
990 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
991 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
995 * skb_dst_is_noref - Test if skb dst isn't refcounted
998 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1000 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1004 * skb_rtable - Returns the skb &rtable
1007 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1009 return (struct rtable *)skb_dst(skb);
1012 /* For mangling skb->pkt_type from user space side from applications
1013 * such as nft, tc, etc, we only allow a conservative subset of
1014 * possible pkt_types to be set.
1016 static inline bool skb_pkt_type_ok(u32 ptype)
1018 return ptype <= PACKET_OTHERHOST;
1022 * skb_napi_id - Returns the skb's NAPI id
1025 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1027 #ifdef CONFIG_NET_RX_BUSY_POLL
1028 return skb->napi_id;
1035 * skb_unref - decrement the skb's reference count
1038 * Returns true if we can free the skb.
1040 static inline bool skb_unref(struct sk_buff *skb)
1044 if (likely(refcount_read(&skb->users) == 1))
1046 else if (likely(!refcount_dec_and_test(&skb->users)))
1052 void skb_release_head_state(struct sk_buff *skb);
1053 void kfree_skb(struct sk_buff *skb);
1054 void kfree_skb_list(struct sk_buff *segs);
1055 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1056 void skb_tx_error(struct sk_buff *skb);
1057 void consume_skb(struct sk_buff *skb);
1058 void __consume_stateless_skb(struct sk_buff *skb);
1059 void __kfree_skb(struct sk_buff *skb);
1060 extern struct kmem_cache *skbuff_head_cache;
1062 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1063 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1064 bool *fragstolen, int *delta_truesize);
1066 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1068 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1069 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1070 struct sk_buff *build_skb_around(struct sk_buff *skb,
1071 void *data, unsigned int frag_size);
1074 * alloc_skb - allocate a network buffer
1075 * @size: size to allocate
1076 * @priority: allocation mask
1078 * This function is a convenient wrapper around __alloc_skb().
1080 static inline struct sk_buff *alloc_skb(unsigned int size,
1083 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1086 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1087 unsigned long data_len,
1091 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1093 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1094 struct sk_buff_fclones {
1095 struct sk_buff skb1;
1097 struct sk_buff skb2;
1099 refcount_t fclone_ref;
1103 * skb_fclone_busy - check if fclone is busy
1107 * Returns true if skb is a fast clone, and its clone is not freed.
1108 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1109 * so we also check that this didnt happen.
1111 static inline bool skb_fclone_busy(const struct sock *sk,
1112 const struct sk_buff *skb)
1114 const struct sk_buff_fclones *fclones;
1116 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1118 return skb->fclone == SKB_FCLONE_ORIG &&
1119 refcount_read(&fclones->fclone_ref) > 1 &&
1120 fclones->skb2.sk == sk;
1124 * alloc_skb_fclone - allocate a network buffer from fclone cache
1125 * @size: size to allocate
1126 * @priority: allocation mask
1128 * This function is a convenient wrapper around __alloc_skb().
1130 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1133 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1136 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1137 void skb_headers_offset_update(struct sk_buff *skb, int off);
1138 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1139 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1140 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1141 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1142 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1143 gfp_t gfp_mask, bool fclone);
1144 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1147 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1150 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1151 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1152 unsigned int headroom);
1153 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1154 int newtailroom, gfp_t priority);
1155 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1156 int offset, int len);
1157 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1158 int offset, int len);
1159 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1160 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1163 * skb_pad - zero pad the tail of an skb
1164 * @skb: buffer to pad
1165 * @pad: space to pad
1167 * Ensure that a buffer is followed by a padding area that is zero
1168 * filled. Used by network drivers which may DMA or transfer data
1169 * beyond the buffer end onto the wire.
1171 * May return error in out of memory cases. The skb is freed on error.
1173 static inline int skb_pad(struct sk_buff *skb, int pad)
1175 return __skb_pad(skb, pad, true);
1177 #define dev_kfree_skb(a) consume_skb(a)
1179 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1180 int offset, size_t size);
1182 struct skb_seq_state {
1186 __u32 stepped_offset;
1187 struct sk_buff *root_skb;
1188 struct sk_buff *cur_skb;
1192 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1193 unsigned int to, struct skb_seq_state *st);
1194 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1195 struct skb_seq_state *st);
1196 void skb_abort_seq_read(struct skb_seq_state *st);
1198 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1199 unsigned int to, struct ts_config *config);
1202 * Packet hash types specify the type of hash in skb_set_hash.
1204 * Hash types refer to the protocol layer addresses which are used to
1205 * construct a packet's hash. The hashes are used to differentiate or identify
1206 * flows of the protocol layer for the hash type. Hash types are either
1207 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1209 * Properties of hashes:
1211 * 1) Two packets in different flows have different hash values
1212 * 2) Two packets in the same flow should have the same hash value
1214 * A hash at a higher layer is considered to be more specific. A driver should
1215 * set the most specific hash possible.
1217 * A driver cannot indicate a more specific hash than the layer at which a hash
1218 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1220 * A driver may indicate a hash level which is less specific than the
1221 * actual layer the hash was computed on. For instance, a hash computed
1222 * at L4 may be considered an L3 hash. This should only be done if the
1223 * driver can't unambiguously determine that the HW computed the hash at
1224 * the higher layer. Note that the "should" in the second property above
1227 enum pkt_hash_types {
1228 PKT_HASH_TYPE_NONE, /* Undefined type */
1229 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1230 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1231 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1234 static inline void skb_clear_hash(struct sk_buff *skb)
1241 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1244 skb_clear_hash(skb);
1248 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1250 skb->l4_hash = is_l4;
1251 skb->sw_hash = is_sw;
1256 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1258 /* Used by drivers to set hash from HW */
1259 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1263 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1265 __skb_set_hash(skb, hash, true, is_l4);
1268 void __skb_get_hash(struct sk_buff *skb);
1269 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1270 u32 skb_get_poff(const struct sk_buff *skb);
1271 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1272 const struct flow_keys_basic *keys, int hlen);
1273 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1274 void *data, int hlen_proto);
1276 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1277 int thoff, u8 ip_proto)
1279 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1282 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1283 const struct flow_dissector_key *key,
1284 unsigned int key_count);
1286 struct bpf_flow_dissector;
1287 bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1288 __be16 proto, int nhoff, int hlen, unsigned int flags);
1290 bool __skb_flow_dissect(const struct net *net,
1291 const struct sk_buff *skb,
1292 struct flow_dissector *flow_dissector,
1293 void *target_container,
1294 void *data, __be16 proto, int nhoff, int hlen,
1295 unsigned int flags);
1297 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1298 struct flow_dissector *flow_dissector,
1299 void *target_container, unsigned int flags)
1301 return __skb_flow_dissect(NULL, skb, flow_dissector,
1302 target_container, NULL, 0, 0, 0, flags);
1305 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1306 struct flow_keys *flow,
1309 memset(flow, 0, sizeof(*flow));
1310 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1311 flow, NULL, 0, 0, 0, flags);
1315 skb_flow_dissect_flow_keys_basic(const struct net *net,
1316 const struct sk_buff *skb,
1317 struct flow_keys_basic *flow, void *data,
1318 __be16 proto, int nhoff, int hlen,
1321 memset(flow, 0, sizeof(*flow));
1322 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1323 data, proto, nhoff, hlen, flags);
1326 void skb_flow_dissect_meta(const struct sk_buff *skb,
1327 struct flow_dissector *flow_dissector,
1328 void *target_container);
1330 /* Gets a skb connection tracking info, ctinfo map should be a
1331 * a map of mapsize to translate enum ip_conntrack_info states
1335 skb_flow_dissect_ct(const struct sk_buff *skb,
1336 struct flow_dissector *flow_dissector,
1337 void *target_container,
1341 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1342 struct flow_dissector *flow_dissector,
1343 void *target_container);
1345 static inline __u32 skb_get_hash(struct sk_buff *skb)
1347 if (!skb->l4_hash && !skb->sw_hash)
1348 __skb_get_hash(skb);
1353 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1355 if (!skb->l4_hash && !skb->sw_hash) {
1356 struct flow_keys keys;
1357 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1359 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1365 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1366 const siphash_key_t *perturb);
1368 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1373 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1375 to->hash = from->hash;
1376 to->sw_hash = from->sw_hash;
1377 to->l4_hash = from->l4_hash;
1380 static inline void skb_copy_decrypted(struct sk_buff *to,
1381 const struct sk_buff *from)
1383 #ifdef CONFIG_TLS_DEVICE
1384 to->decrypted = from->decrypted;
1388 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1389 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1391 return skb->head + skb->end;
1394 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1399 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1404 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1406 return skb->end - skb->head;
1411 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1413 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1415 return &skb_shinfo(skb)->hwtstamps;
1418 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1420 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1422 return is_zcopy ? skb_uarg(skb) : NULL;
1425 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1428 if (skb && uarg && !skb_zcopy(skb)) {
1429 if (unlikely(have_ref && *have_ref))
1432 sock_zerocopy_get(uarg);
1433 skb_shinfo(skb)->destructor_arg = uarg;
1434 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1438 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1440 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1441 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1444 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1446 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1449 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1451 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1454 /* Release a reference on a zerocopy structure */
1455 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1457 struct ubuf_info *uarg = skb_zcopy(skb);
1460 if (skb_zcopy_is_nouarg(skb)) {
1461 /* no notification callback */
1462 } else if (uarg->callback == sock_zerocopy_callback) {
1463 uarg->zerocopy = uarg->zerocopy && zerocopy;
1464 sock_zerocopy_put(uarg);
1466 uarg->callback(uarg, zerocopy);
1469 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1473 /* Abort a zerocopy operation and revert zckey on error in send syscall */
1474 static inline void skb_zcopy_abort(struct sk_buff *skb)
1476 struct ubuf_info *uarg = skb_zcopy(skb);
1479 sock_zerocopy_put_abort(uarg, false);
1480 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1484 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1489 /* Iterate through singly-linked GSO fragments of an skb. */
1490 #define skb_list_walk_safe(first, skb, next_skb) \
1491 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1492 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1494 static inline void skb_list_del_init(struct sk_buff *skb)
1496 __list_del_entry(&skb->list);
1497 skb_mark_not_on_list(skb);
1501 * skb_queue_empty - check if a queue is empty
1504 * Returns true if the queue is empty, false otherwise.
1506 static inline int skb_queue_empty(const struct sk_buff_head *list)
1508 return list->next == (const struct sk_buff *) list;
1512 * skb_queue_empty_lockless - check if a queue is empty
1515 * Returns true if the queue is empty, false otherwise.
1516 * This variant can be used in lockless contexts.
1518 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1520 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1525 * skb_queue_is_last - check if skb is the last entry in the queue
1529 * Returns true if @skb is the last buffer on the list.
1531 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1532 const struct sk_buff *skb)
1534 return skb->next == (const struct sk_buff *) list;
1538 * skb_queue_is_first - check if skb is the first entry in the queue
1542 * Returns true if @skb is the first buffer on the list.
1544 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1545 const struct sk_buff *skb)
1547 return skb->prev == (const struct sk_buff *) list;
1551 * skb_queue_next - return the next packet in the queue
1553 * @skb: current buffer
1555 * Return the next packet in @list after @skb. It is only valid to
1556 * call this if skb_queue_is_last() evaluates to false.
1558 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1559 const struct sk_buff *skb)
1561 /* This BUG_ON may seem severe, but if we just return then we
1562 * are going to dereference garbage.
1564 BUG_ON(skb_queue_is_last(list, skb));
1569 * skb_queue_prev - return the prev packet in the queue
1571 * @skb: current buffer
1573 * Return the prev packet in @list before @skb. It is only valid to
1574 * call this if skb_queue_is_first() evaluates to false.
1576 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1577 const struct sk_buff *skb)
1579 /* This BUG_ON may seem severe, but if we just return then we
1580 * are going to dereference garbage.
1582 BUG_ON(skb_queue_is_first(list, skb));
1587 * skb_get - reference buffer
1588 * @skb: buffer to reference
1590 * Makes another reference to a socket buffer and returns a pointer
1593 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1595 refcount_inc(&skb->users);
1600 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1604 * skb_cloned - is the buffer a clone
1605 * @skb: buffer to check
1607 * Returns true if the buffer was generated with skb_clone() and is
1608 * one of multiple shared copies of the buffer. Cloned buffers are
1609 * shared data so must not be written to under normal circumstances.
1611 static inline int skb_cloned(const struct sk_buff *skb)
1613 return skb->cloned &&
1614 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1617 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1619 might_sleep_if(gfpflags_allow_blocking(pri));
1621 if (skb_cloned(skb))
1622 return pskb_expand_head(skb, 0, 0, pri);
1628 * skb_header_cloned - is the header a clone
1629 * @skb: buffer to check
1631 * Returns true if modifying the header part of the buffer requires
1632 * the data to be copied.
1634 static inline int skb_header_cloned(const struct sk_buff *skb)
1641 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1642 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1643 return dataref != 1;
1646 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1648 might_sleep_if(gfpflags_allow_blocking(pri));
1650 if (skb_header_cloned(skb))
1651 return pskb_expand_head(skb, 0, 0, pri);
1657 * __skb_header_release - release reference to header
1658 * @skb: buffer to operate on
1660 static inline void __skb_header_release(struct sk_buff *skb)
1663 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1668 * skb_shared - is the buffer shared
1669 * @skb: buffer to check
1671 * Returns true if more than one person has a reference to this
1674 static inline int skb_shared(const struct sk_buff *skb)
1676 return refcount_read(&skb->users) != 1;
1680 * skb_share_check - check if buffer is shared and if so clone it
1681 * @skb: buffer to check
1682 * @pri: priority for memory allocation
1684 * If the buffer is shared the buffer is cloned and the old copy
1685 * drops a reference. A new clone with a single reference is returned.
1686 * If the buffer is not shared the original buffer is returned. When
1687 * being called from interrupt status or with spinlocks held pri must
1690 * NULL is returned on a memory allocation failure.
1692 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1694 might_sleep_if(gfpflags_allow_blocking(pri));
1695 if (skb_shared(skb)) {
1696 struct sk_buff *nskb = skb_clone(skb, pri);
1708 * Copy shared buffers into a new sk_buff. We effectively do COW on
1709 * packets to handle cases where we have a local reader and forward
1710 * and a couple of other messy ones. The normal one is tcpdumping
1711 * a packet thats being forwarded.
1715 * skb_unshare - make a copy of a shared buffer
1716 * @skb: buffer to check
1717 * @pri: priority for memory allocation
1719 * If the socket buffer is a clone then this function creates a new
1720 * copy of the data, drops a reference count on the old copy and returns
1721 * the new copy with the reference count at 1. If the buffer is not a clone
1722 * the original buffer is returned. When called with a spinlock held or
1723 * from interrupt state @pri must be %GFP_ATOMIC
1725 * %NULL is returned on a memory allocation failure.
1727 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1730 might_sleep_if(gfpflags_allow_blocking(pri));
1731 if (skb_cloned(skb)) {
1732 struct sk_buff *nskb = skb_copy(skb, pri);
1734 /* Free our shared copy */
1745 * skb_peek - peek at the head of an &sk_buff_head
1746 * @list_: list to peek at
1748 * Peek an &sk_buff. Unlike most other operations you _MUST_
1749 * be careful with this one. A peek leaves the buffer on the
1750 * list and someone else may run off with it. You must hold
1751 * the appropriate locks or have a private queue to do this.
1753 * Returns %NULL for an empty list or a pointer to the head element.
1754 * The reference count is not incremented and the reference is therefore
1755 * volatile. Use with caution.
1757 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1759 struct sk_buff *skb = list_->next;
1761 if (skb == (struct sk_buff *)list_)
1767 * __skb_peek - peek at the head of a non-empty &sk_buff_head
1768 * @list_: list to peek at
1770 * Like skb_peek(), but the caller knows that the list is not empty.
1772 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
1778 * skb_peek_next - peek skb following the given one from a queue
1779 * @skb: skb to start from
1780 * @list_: list to peek at
1782 * Returns %NULL when the end of the list is met or a pointer to the
1783 * next element. The reference count is not incremented and the
1784 * reference is therefore volatile. Use with caution.
1786 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1787 const struct sk_buff_head *list_)
1789 struct sk_buff *next = skb->next;
1791 if (next == (struct sk_buff *)list_)
1797 * skb_peek_tail - peek at the tail of an &sk_buff_head
1798 * @list_: list to peek at
1800 * Peek an &sk_buff. Unlike most other operations you _MUST_
1801 * be careful with this one. A peek leaves the buffer on the
1802 * list and someone else may run off with it. You must hold
1803 * the appropriate locks or have a private queue to do this.
1805 * Returns %NULL for an empty list or a pointer to the tail element.
1806 * The reference count is not incremented and the reference is therefore
1807 * volatile. Use with caution.
1809 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1811 struct sk_buff *skb = READ_ONCE(list_->prev);
1813 if (skb == (struct sk_buff *)list_)
1820 * skb_queue_len - get queue length
1821 * @list_: list to measure
1823 * Return the length of an &sk_buff queue.
1825 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1831 * skb_queue_len_lockless - get queue length
1832 * @list_: list to measure
1834 * Return the length of an &sk_buff queue.
1835 * This variant can be used in lockless contexts.
1837 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
1839 return READ_ONCE(list_->qlen);
1843 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1844 * @list: queue to initialize
1846 * This initializes only the list and queue length aspects of
1847 * an sk_buff_head object. This allows to initialize the list
1848 * aspects of an sk_buff_head without reinitializing things like
1849 * the spinlock. It can also be used for on-stack sk_buff_head
1850 * objects where the spinlock is known to not be used.
1852 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1854 list->prev = list->next = (struct sk_buff *)list;
1859 * This function creates a split out lock class for each invocation;
1860 * this is needed for now since a whole lot of users of the skb-queue
1861 * infrastructure in drivers have different locking usage (in hardirq)
1862 * than the networking core (in softirq only). In the long run either the
1863 * network layer or drivers should need annotation to consolidate the
1864 * main types of usage into 3 classes.
1866 static inline void skb_queue_head_init(struct sk_buff_head *list)
1868 spin_lock_init(&list->lock);
1869 __skb_queue_head_init(list);
1872 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1873 struct lock_class_key *class)
1875 skb_queue_head_init(list);
1876 lockdep_set_class(&list->lock, class);
1880 * Insert an sk_buff on a list.
1882 * The "__skb_xxxx()" functions are the non-atomic ones that
1883 * can only be called with interrupts disabled.
1885 static inline void __skb_insert(struct sk_buff *newsk,
1886 struct sk_buff *prev, struct sk_buff *next,
1887 struct sk_buff_head *list)
1889 /* See skb_queue_empty_lockless() and skb_peek_tail()
1890 * for the opposite READ_ONCE()
1892 WRITE_ONCE(newsk->next, next);
1893 WRITE_ONCE(newsk->prev, prev);
1894 WRITE_ONCE(next->prev, newsk);
1895 WRITE_ONCE(prev->next, newsk);
1899 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1900 struct sk_buff *prev,
1901 struct sk_buff *next)
1903 struct sk_buff *first = list->next;
1904 struct sk_buff *last = list->prev;
1906 WRITE_ONCE(first->prev, prev);
1907 WRITE_ONCE(prev->next, first);
1909 WRITE_ONCE(last->next, next);
1910 WRITE_ONCE(next->prev, last);
1914 * skb_queue_splice - join two skb lists, this is designed for stacks
1915 * @list: the new list to add
1916 * @head: the place to add it in the first list
1918 static inline void skb_queue_splice(const struct sk_buff_head *list,
1919 struct sk_buff_head *head)
1921 if (!skb_queue_empty(list)) {
1922 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1923 head->qlen += list->qlen;
1928 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1929 * @list: the new list to add
1930 * @head: the place to add it in the first list
1932 * The list at @list is reinitialised
1934 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1935 struct sk_buff_head *head)
1937 if (!skb_queue_empty(list)) {
1938 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1939 head->qlen += list->qlen;
1940 __skb_queue_head_init(list);
1945 * skb_queue_splice_tail - join two skb lists, each list being a queue
1946 * @list: the new list to add
1947 * @head: the place to add it in the first list
1949 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1950 struct sk_buff_head *head)
1952 if (!skb_queue_empty(list)) {
1953 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1954 head->qlen += list->qlen;
1959 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1960 * @list: the new list to add
1961 * @head: the place to add it in the first list
1963 * Each of the lists is a queue.
1964 * The list at @list is reinitialised
1966 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1967 struct sk_buff_head *head)
1969 if (!skb_queue_empty(list)) {
1970 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1971 head->qlen += list->qlen;
1972 __skb_queue_head_init(list);
1977 * __skb_queue_after - queue a buffer at the list head
1978 * @list: list to use
1979 * @prev: place after this buffer
1980 * @newsk: buffer to queue
1982 * Queue a buffer int the middle of a list. This function takes no locks
1983 * and you must therefore hold required locks before calling it.
1985 * A buffer cannot be placed on two lists at the same time.
1987 static inline void __skb_queue_after(struct sk_buff_head *list,
1988 struct sk_buff *prev,
1989 struct sk_buff *newsk)
1991 __skb_insert(newsk, prev, prev->next, list);
1994 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1995 struct sk_buff_head *list);
1997 static inline void __skb_queue_before(struct sk_buff_head *list,
1998 struct sk_buff *next,
1999 struct sk_buff *newsk)
2001 __skb_insert(newsk, next->prev, next, list);
2005 * __skb_queue_head - queue a buffer at the list head
2006 * @list: list to use
2007 * @newsk: buffer to queue
2009 * Queue a buffer at the start of a list. This function takes no locks
2010 * and you must therefore hold required locks before calling it.
2012 * A buffer cannot be placed on two lists at the same time.
2014 static inline void __skb_queue_head(struct sk_buff_head *list,
2015 struct sk_buff *newsk)
2017 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2019 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2022 * __skb_queue_tail - queue a buffer at the list tail
2023 * @list: list to use
2024 * @newsk: buffer to queue
2026 * Queue a buffer at the end of a list. This function takes no locks
2027 * and you must therefore hold required locks before calling it.
2029 * A buffer cannot be placed on two lists at the same time.
2031 static inline void __skb_queue_tail(struct sk_buff_head *list,
2032 struct sk_buff *newsk)
2034 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2036 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2039 * remove sk_buff from list. _Must_ be called atomically, and with
2042 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2043 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2045 struct sk_buff *next, *prev;
2047 WRITE_ONCE(list->qlen, list->qlen - 1);
2050 skb->next = skb->prev = NULL;
2051 WRITE_ONCE(next->prev, prev);
2052 WRITE_ONCE(prev->next, next);
2056 * __skb_dequeue - remove from the head of the queue
2057 * @list: list to dequeue from
2059 * Remove the head of the list. This function does not take any locks
2060 * so must be used with appropriate locks held only. The head item is
2061 * returned or %NULL if the list is empty.
2063 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2065 struct sk_buff *skb = skb_peek(list);
2067 __skb_unlink(skb, list);
2070 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2073 * __skb_dequeue_tail - remove from the tail of the queue
2074 * @list: list to dequeue from
2076 * Remove the tail of the list. This function does not take any locks
2077 * so must be used with appropriate locks held only. The tail item is
2078 * returned or %NULL if the list is empty.
2080 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2082 struct sk_buff *skb = skb_peek_tail(list);
2084 __skb_unlink(skb, list);
2087 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2090 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2092 return skb->data_len;
2095 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2097 return skb->len - skb->data_len;
2100 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2102 unsigned int i, len = 0;
2104 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2105 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2109 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2111 return skb_headlen(skb) + __skb_pagelen(skb);
2115 * __skb_fill_page_desc - initialise a paged fragment in an skb
2116 * @skb: buffer containing fragment to be initialised
2117 * @i: paged fragment index to initialise
2118 * @page: the page to use for this fragment
2119 * @off: the offset to the data with @page
2120 * @size: the length of the data
2122 * Initialises the @i'th fragment of @skb to point to &size bytes at
2123 * offset @off within @page.
2125 * Does not take any additional reference on the fragment.
2127 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2128 struct page *page, int off, int size)
2130 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2133 * Propagate page pfmemalloc to the skb if we can. The problem is
2134 * that not all callers have unique ownership of the page but rely
2135 * on page_is_pfmemalloc doing the right thing(tm).
2137 frag->bv_page = page;
2138 frag->bv_offset = off;
2139 skb_frag_size_set(frag, size);
2141 page = compound_head(page);
2142 if (page_is_pfmemalloc(page))
2143 skb->pfmemalloc = true;
2147 * skb_fill_page_desc - initialise a paged fragment in an skb
2148 * @skb: buffer containing fragment to be initialised
2149 * @i: paged fragment index to initialise
2150 * @page: the page to use for this fragment
2151 * @off: the offset to the data with @page
2152 * @size: the length of the data
2154 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2155 * @skb to point to @size bytes at offset @off within @page. In
2156 * addition updates @skb such that @i is the last fragment.
2158 * Does not take any additional reference on the fragment.
2160 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2161 struct page *page, int off, int size)
2163 __skb_fill_page_desc(skb, i, page, off, size);
2164 skb_shinfo(skb)->nr_frags = i + 1;
2167 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2168 int size, unsigned int truesize);
2170 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2171 unsigned int truesize);
2173 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2175 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2176 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2178 return skb->head + skb->tail;
2181 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2183 skb->tail = skb->data - skb->head;
2186 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2188 skb_reset_tail_pointer(skb);
2189 skb->tail += offset;
2192 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2193 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2198 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2200 skb->tail = skb->data;
2203 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2205 skb->tail = skb->data + offset;
2208 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2211 * Add data to an sk_buff
2213 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2214 void *skb_put(struct sk_buff *skb, unsigned int len);
2215 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2217 void *tmp = skb_tail_pointer(skb);
2218 SKB_LINEAR_ASSERT(skb);
2224 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2226 void *tmp = __skb_put(skb, len);
2228 memset(tmp, 0, len);
2232 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2235 void *tmp = __skb_put(skb, len);
2237 memcpy(tmp, data, len);
2241 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2243 *(u8 *)__skb_put(skb, 1) = val;
2246 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2248 void *tmp = skb_put(skb, len);
2250 memset(tmp, 0, len);
2255 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2258 void *tmp = skb_put(skb, len);
2260 memcpy(tmp, data, len);
2265 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2267 *(u8 *)skb_put(skb, 1) = val;
2270 void *skb_push(struct sk_buff *skb, unsigned int len);
2271 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2278 void *skb_pull(struct sk_buff *skb, unsigned int len);
2279 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2282 BUG_ON(skb->len < skb->data_len);
2283 return skb->data += len;
2286 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2288 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2291 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2293 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2295 if (len > skb_headlen(skb) &&
2296 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2299 return skb->data += len;
2302 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2304 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2307 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2309 if (likely(len <= skb_headlen(skb)))
2311 if (unlikely(len > skb->len))
2313 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2316 void skb_condense(struct sk_buff *skb);
2319 * skb_headroom - bytes at buffer head
2320 * @skb: buffer to check
2322 * Return the number of bytes of free space at the head of an &sk_buff.
2324 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2326 return skb->data - skb->head;
2330 * skb_tailroom - bytes at buffer end
2331 * @skb: buffer to check
2333 * Return the number of bytes of free space at the tail of an sk_buff
2335 static inline int skb_tailroom(const struct sk_buff *skb)
2337 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2341 * skb_availroom - bytes at buffer end
2342 * @skb: buffer to check
2344 * Return the number of bytes of free space at the tail of an sk_buff
2345 * allocated by sk_stream_alloc()
2347 static inline int skb_availroom(const struct sk_buff *skb)
2349 if (skb_is_nonlinear(skb))
2352 return skb->end - skb->tail - skb->reserved_tailroom;
2356 * skb_reserve - adjust headroom
2357 * @skb: buffer to alter
2358 * @len: bytes to move
2360 * Increase the headroom of an empty &sk_buff by reducing the tail
2361 * room. This is only allowed for an empty buffer.
2363 static inline void skb_reserve(struct sk_buff *skb, int len)
2370 * skb_tailroom_reserve - adjust reserved_tailroom
2371 * @skb: buffer to alter
2372 * @mtu: maximum amount of headlen permitted
2373 * @needed_tailroom: minimum amount of reserved_tailroom
2375 * Set reserved_tailroom so that headlen can be as large as possible but
2376 * not larger than mtu and tailroom cannot be smaller than
2378 * The required headroom should already have been reserved before using
2381 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2382 unsigned int needed_tailroom)
2384 SKB_LINEAR_ASSERT(skb);
2385 if (mtu < skb_tailroom(skb) - needed_tailroom)
2386 /* use at most mtu */
2387 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2389 /* use up to all available space */
2390 skb->reserved_tailroom = needed_tailroom;
2393 #define ENCAP_TYPE_ETHER 0
2394 #define ENCAP_TYPE_IPPROTO 1
2396 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2399 skb->inner_protocol = protocol;
2400 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2403 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2406 skb->inner_ipproto = ipproto;
2407 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2410 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2412 skb->inner_mac_header = skb->mac_header;
2413 skb->inner_network_header = skb->network_header;
2414 skb->inner_transport_header = skb->transport_header;
2417 static inline void skb_reset_mac_len(struct sk_buff *skb)
2419 skb->mac_len = skb->network_header - skb->mac_header;
2422 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2425 return skb->head + skb->inner_transport_header;
2428 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2430 return skb_inner_transport_header(skb) - skb->data;
2433 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2435 skb->inner_transport_header = skb->data - skb->head;
2438 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2441 skb_reset_inner_transport_header(skb);
2442 skb->inner_transport_header += offset;
2445 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2447 return skb->head + skb->inner_network_header;
2450 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2452 skb->inner_network_header = skb->data - skb->head;
2455 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2458 skb_reset_inner_network_header(skb);
2459 skb->inner_network_header += offset;
2462 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2464 return skb->head + skb->inner_mac_header;
2467 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2469 skb->inner_mac_header = skb->data - skb->head;
2472 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2475 skb_reset_inner_mac_header(skb);
2476 skb->inner_mac_header += offset;
2478 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2480 return skb->transport_header != (typeof(skb->transport_header))~0U;
2483 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2485 return skb->head + skb->transport_header;
2488 static inline void skb_reset_transport_header(struct sk_buff *skb)
2490 skb->transport_header = skb->data - skb->head;
2493 static inline void skb_set_transport_header(struct sk_buff *skb,
2496 skb_reset_transport_header(skb);
2497 skb->transport_header += offset;
2500 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2502 return skb->head + skb->network_header;
2505 static inline void skb_reset_network_header(struct sk_buff *skb)
2507 skb->network_header = skb->data - skb->head;
2510 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2512 skb_reset_network_header(skb);
2513 skb->network_header += offset;
2516 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2518 return skb->head + skb->mac_header;
2521 static inline int skb_mac_offset(const struct sk_buff *skb)
2523 return skb_mac_header(skb) - skb->data;
2526 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2528 return skb->network_header - skb->mac_header;
2531 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2533 return skb->mac_header != (typeof(skb->mac_header))~0U;
2536 static inline void skb_reset_mac_header(struct sk_buff *skb)
2538 skb->mac_header = skb->data - skb->head;
2541 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2543 skb_reset_mac_header(skb);
2544 skb->mac_header += offset;
2547 static inline void skb_pop_mac_header(struct sk_buff *skb)
2549 skb->mac_header = skb->network_header;
2552 static inline void skb_probe_transport_header(struct sk_buff *skb)
2554 struct flow_keys_basic keys;
2556 if (skb_transport_header_was_set(skb))
2559 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2561 skb_set_transport_header(skb, keys.control.thoff);
2564 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2566 if (skb_mac_header_was_set(skb)) {
2567 const unsigned char *old_mac = skb_mac_header(skb);
2569 skb_set_mac_header(skb, -skb->mac_len);
2570 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2574 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2576 return skb->csum_start - skb_headroom(skb);
2579 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2581 return skb->head + skb->csum_start;
2584 static inline int skb_transport_offset(const struct sk_buff *skb)
2586 return skb_transport_header(skb) - skb->data;
2589 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2591 return skb->transport_header - skb->network_header;
2594 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2596 return skb->inner_transport_header - skb->inner_network_header;
2599 static inline int skb_network_offset(const struct sk_buff *skb)
2601 return skb_network_header(skb) - skb->data;
2604 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2606 return skb_inner_network_header(skb) - skb->data;
2609 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2611 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2615 * CPUs often take a performance hit when accessing unaligned memory
2616 * locations. The actual performance hit varies, it can be small if the
2617 * hardware handles it or large if we have to take an exception and fix it
2620 * Since an ethernet header is 14 bytes network drivers often end up with
2621 * the IP header at an unaligned offset. The IP header can be aligned by
2622 * shifting the start of the packet by 2 bytes. Drivers should do this
2625 * skb_reserve(skb, NET_IP_ALIGN);
2627 * The downside to this alignment of the IP header is that the DMA is now
2628 * unaligned. On some architectures the cost of an unaligned DMA is high
2629 * and this cost outweighs the gains made by aligning the IP header.
2631 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2634 #ifndef NET_IP_ALIGN
2635 #define NET_IP_ALIGN 2
2639 * The networking layer reserves some headroom in skb data (via
2640 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2641 * the header has to grow. In the default case, if the header has to grow
2642 * 32 bytes or less we avoid the reallocation.
2644 * Unfortunately this headroom changes the DMA alignment of the resulting
2645 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2646 * on some architectures. An architecture can override this value,
2647 * perhaps setting it to a cacheline in size (since that will maintain
2648 * cacheline alignment of the DMA). It must be a power of 2.
2650 * Various parts of the networking layer expect at least 32 bytes of
2651 * headroom, you should not reduce this.
2653 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2654 * to reduce average number of cache lines per packet.
2655 * get_rps_cpus() for example only access one 64 bytes aligned block :
2656 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2659 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2662 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2664 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2666 if (WARN_ON(skb_is_nonlinear(skb)))
2669 skb_set_tail_pointer(skb, len);
2672 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2674 __skb_set_length(skb, len);
2677 void skb_trim(struct sk_buff *skb, unsigned int len);
2679 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2682 return ___pskb_trim(skb, len);
2683 __skb_trim(skb, len);
2687 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2689 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2693 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2694 * @skb: buffer to alter
2697 * This is identical to pskb_trim except that the caller knows that
2698 * the skb is not cloned so we should never get an error due to out-
2701 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2703 int err = pskb_trim(skb, len);
2707 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2709 unsigned int diff = len - skb->len;
2711 if (skb_tailroom(skb) < diff) {
2712 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2717 __skb_set_length(skb, len);
2722 * skb_orphan - orphan a buffer
2723 * @skb: buffer to orphan
2725 * If a buffer currently has an owner then we call the owner's
2726 * destructor function and make the @skb unowned. The buffer continues
2727 * to exist but is no longer charged to its former owner.
2729 static inline void skb_orphan(struct sk_buff *skb)
2731 if (skb->destructor) {
2732 skb->destructor(skb);
2733 skb->destructor = NULL;
2741 * skb_orphan_frags - orphan the frags contained in a buffer
2742 * @skb: buffer to orphan frags from
2743 * @gfp_mask: allocation mask for replacement pages
2745 * For each frag in the SKB which needs a destructor (i.e. has an
2746 * owner) create a copy of that frag and release the original
2747 * page by calling the destructor.
2749 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2751 if (likely(!skb_zcopy(skb)))
2753 if (!skb_zcopy_is_nouarg(skb) &&
2754 skb_uarg(skb)->callback == sock_zerocopy_callback)
2756 return skb_copy_ubufs(skb, gfp_mask);
2759 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2760 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2762 if (likely(!skb_zcopy(skb)))
2764 return skb_copy_ubufs(skb, gfp_mask);
2768 * __skb_queue_purge - empty a list
2769 * @list: list to empty
2771 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2772 * the list and one reference dropped. This function does not take the
2773 * list lock and the caller must hold the relevant locks to use it.
2775 static inline void __skb_queue_purge(struct sk_buff_head *list)
2777 struct sk_buff *skb;
2778 while ((skb = __skb_dequeue(list)) != NULL)
2781 void skb_queue_purge(struct sk_buff_head *list);
2783 unsigned int skb_rbtree_purge(struct rb_root *root);
2785 void *netdev_alloc_frag(unsigned int fragsz);
2787 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2791 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2792 * @dev: network device to receive on
2793 * @length: length to allocate
2795 * Allocate a new &sk_buff and assign it a usage count of one. The
2796 * buffer has unspecified headroom built in. Users should allocate
2797 * the headroom they think they need without accounting for the
2798 * built in space. The built in space is used for optimisations.
2800 * %NULL is returned if there is no free memory. Although this function
2801 * allocates memory it can be called from an interrupt.
2803 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2804 unsigned int length)
2806 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2809 /* legacy helper around __netdev_alloc_skb() */
2810 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2813 return __netdev_alloc_skb(NULL, length, gfp_mask);
2816 /* legacy helper around netdev_alloc_skb() */
2817 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2819 return netdev_alloc_skb(NULL, length);
2823 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2824 unsigned int length, gfp_t gfp)
2826 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2828 if (NET_IP_ALIGN && skb)
2829 skb_reserve(skb, NET_IP_ALIGN);
2833 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2834 unsigned int length)
2836 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2839 static inline void skb_free_frag(void *addr)
2841 page_frag_free(addr);
2844 void *napi_alloc_frag(unsigned int fragsz);
2845 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2846 unsigned int length, gfp_t gfp_mask);
2847 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2848 unsigned int length)
2850 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2852 void napi_consume_skb(struct sk_buff *skb, int budget);
2854 void __kfree_skb_flush(void);
2855 void __kfree_skb_defer(struct sk_buff *skb);
2858 * __dev_alloc_pages - allocate page for network Rx
2859 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2860 * @order: size of the allocation
2862 * Allocate a new page.
2864 * %NULL is returned if there is no free memory.
2866 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2869 /* This piece of code contains several assumptions.
2870 * 1. This is for device Rx, therefor a cold page is preferred.
2871 * 2. The expectation is the user wants a compound page.
2872 * 3. If requesting a order 0 page it will not be compound
2873 * due to the check to see if order has a value in prep_new_page
2874 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2875 * code in gfp_to_alloc_flags that should be enforcing this.
2877 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2879 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2882 static inline struct page *dev_alloc_pages(unsigned int order)
2884 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2888 * __dev_alloc_page - allocate a page for network Rx
2889 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2891 * Allocate a new page.
2893 * %NULL is returned if there is no free memory.
2895 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2897 return __dev_alloc_pages(gfp_mask, 0);
2900 static inline struct page *dev_alloc_page(void)
2902 return dev_alloc_pages(0);
2906 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2907 * @page: The page that was allocated from skb_alloc_page
2908 * @skb: The skb that may need pfmemalloc set
2910 static inline void skb_propagate_pfmemalloc(struct page *page,
2911 struct sk_buff *skb)
2913 if (page_is_pfmemalloc(page))
2914 skb->pfmemalloc = true;
2918 * skb_frag_off() - Returns the offset of a skb fragment
2919 * @frag: the paged fragment
2921 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
2923 return frag->bv_offset;
2927 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
2928 * @frag: skb fragment
2929 * @delta: value to add
2931 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
2933 frag->bv_offset += delta;
2937 * skb_frag_off_set() - Sets the offset of a skb fragment
2938 * @frag: skb fragment
2939 * @offset: offset of fragment
2941 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
2943 frag->bv_offset = offset;
2947 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
2948 * @fragto: skb fragment where offset is set
2949 * @fragfrom: skb fragment offset is copied from
2951 static inline void skb_frag_off_copy(skb_frag_t *fragto,
2952 const skb_frag_t *fragfrom)
2954 fragto->bv_offset = fragfrom->bv_offset;
2958 * skb_frag_page - retrieve the page referred to by a paged fragment
2959 * @frag: the paged fragment
2961 * Returns the &struct page associated with @frag.
2963 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2965 return frag->bv_page;
2969 * __skb_frag_ref - take an addition reference on a paged fragment.
2970 * @frag: the paged fragment
2972 * Takes an additional reference on the paged fragment @frag.
2974 static inline void __skb_frag_ref(skb_frag_t *frag)
2976 get_page(skb_frag_page(frag));
2980 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2982 * @f: the fragment offset.
2984 * Takes an additional reference on the @f'th paged fragment of @skb.
2986 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2988 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2992 * __skb_frag_unref - release a reference on a paged fragment.
2993 * @frag: the paged fragment
2995 * Releases a reference on the paged fragment @frag.
2997 static inline void __skb_frag_unref(skb_frag_t *frag)
2999 put_page(skb_frag_page(frag));
3003 * skb_frag_unref - release a reference on a paged fragment of an skb.
3005 * @f: the fragment offset
3007 * Releases a reference on the @f'th paged fragment of @skb.
3009 static inline void skb_frag_unref(struct sk_buff *skb, int f)
3011 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
3015 * skb_frag_address - gets the address of the data contained in a paged fragment
3016 * @frag: the paged fragment buffer
3018 * Returns the address of the data within @frag. The page must already
3021 static inline void *skb_frag_address(const skb_frag_t *frag)
3023 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3027 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3028 * @frag: the paged fragment buffer
3030 * Returns the address of the data within @frag. Checks that the page
3031 * is mapped and returns %NULL otherwise.
3033 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3035 void *ptr = page_address(skb_frag_page(frag));
3039 return ptr + skb_frag_off(frag);
3043 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3044 * @fragto: skb fragment where page is set
3045 * @fragfrom: skb fragment page is copied from
3047 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3048 const skb_frag_t *fragfrom)
3050 fragto->bv_page = fragfrom->bv_page;
3054 * __skb_frag_set_page - sets the page contained in a paged fragment
3055 * @frag: the paged fragment
3056 * @page: the page to set
3058 * Sets the fragment @frag to contain @page.
3060 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3062 frag->bv_page = page;
3066 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3068 * @f: the fragment offset
3069 * @page: the page to set
3071 * Sets the @f'th fragment of @skb to contain @page.
3073 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3076 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3079 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3082 * skb_frag_dma_map - maps a paged fragment via the DMA API
3083 * @dev: the device to map the fragment to
3084 * @frag: the paged fragment to map
3085 * @offset: the offset within the fragment (starting at the
3086 * fragment's own offset)
3087 * @size: the number of bytes to map
3088 * @dir: the direction of the mapping (``PCI_DMA_*``)
3090 * Maps the page associated with @frag to @device.
3092 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3093 const skb_frag_t *frag,
3094 size_t offset, size_t size,
3095 enum dma_data_direction dir)
3097 return dma_map_page(dev, skb_frag_page(frag),
3098 skb_frag_off(frag) + offset, size, dir);
3101 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3104 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3108 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3111 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3116 * skb_clone_writable - is the header of a clone writable
3117 * @skb: buffer to check
3118 * @len: length up to which to write
3120 * Returns true if modifying the header part of the cloned buffer
3121 * does not requires the data to be copied.
3123 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3125 return !skb_header_cloned(skb) &&
3126 skb_headroom(skb) + len <= skb->hdr_len;
3129 static inline int skb_try_make_writable(struct sk_buff *skb,
3130 unsigned int write_len)
3132 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3133 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3136 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3141 if (headroom > skb_headroom(skb))
3142 delta = headroom - skb_headroom(skb);
3144 if (delta || cloned)
3145 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3151 * skb_cow - copy header of skb when it is required
3152 * @skb: buffer to cow
3153 * @headroom: needed headroom
3155 * If the skb passed lacks sufficient headroom or its data part
3156 * is shared, data is reallocated. If reallocation fails, an error
3157 * is returned and original skb is not changed.
3159 * The result is skb with writable area skb->head...skb->tail
3160 * and at least @headroom of space at head.
3162 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3164 return __skb_cow(skb, headroom, skb_cloned(skb));
3168 * skb_cow_head - skb_cow but only making the head writable
3169 * @skb: buffer to cow
3170 * @headroom: needed headroom
3172 * This function is identical to skb_cow except that we replace the
3173 * skb_cloned check by skb_header_cloned. It should be used when
3174 * you only need to push on some header and do not need to modify
3177 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3179 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3183 * skb_padto - pad an skbuff up to a minimal size
3184 * @skb: buffer to pad
3185 * @len: minimal length
3187 * Pads up a buffer to ensure the trailing bytes exist and are
3188 * blanked. If the buffer already contains sufficient data it
3189 * is untouched. Otherwise it is extended. Returns zero on
3190 * success. The skb is freed on error.
3192 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3194 unsigned int size = skb->len;
3195 if (likely(size >= len))
3197 return skb_pad(skb, len - size);
3201 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3202 * @skb: buffer to pad
3203 * @len: minimal length
3204 * @free_on_error: free buffer on error
3206 * Pads up a buffer to ensure the trailing bytes exist and are
3207 * blanked. If the buffer already contains sufficient data it
3208 * is untouched. Otherwise it is extended. Returns zero on
3209 * success. The skb is freed on error if @free_on_error is true.
3211 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
3214 unsigned int size = skb->len;
3216 if (unlikely(size < len)) {
3218 if (__skb_pad(skb, len, free_on_error))
3220 __skb_put(skb, len);
3226 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3227 * @skb: buffer to pad
3228 * @len: minimal length
3230 * Pads up a buffer to ensure the trailing bytes exist and are
3231 * blanked. If the buffer already contains sufficient data it
3232 * is untouched. Otherwise it is extended. Returns zero on
3233 * success. The skb is freed on error.
3235 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
3237 return __skb_put_padto(skb, len, true);
3240 static inline int skb_add_data(struct sk_buff *skb,
3241 struct iov_iter *from, int copy)
3243 const int off = skb->len;
3245 if (skb->ip_summed == CHECKSUM_NONE) {
3247 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3249 skb->csum = csum_block_add(skb->csum, csum, off);
3252 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3255 __skb_trim(skb, off);
3259 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3260 const struct page *page, int off)
3265 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3267 return page == skb_frag_page(frag) &&
3268 off == skb_frag_off(frag) + skb_frag_size(frag);
3273 static inline int __skb_linearize(struct sk_buff *skb)
3275 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3279 * skb_linearize - convert paged skb to linear one
3280 * @skb: buffer to linarize
3282 * If there is no free memory -ENOMEM is returned, otherwise zero
3283 * is returned and the old skb data released.
3285 static inline int skb_linearize(struct sk_buff *skb)
3287 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3291 * skb_has_shared_frag - can any frag be overwritten
3292 * @skb: buffer to test
3294 * Return true if the skb has at least one frag that might be modified
3295 * by an external entity (as in vmsplice()/sendfile())
3297 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3299 return skb_is_nonlinear(skb) &&
3300 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3304 * skb_linearize_cow - make sure skb is linear and writable
3305 * @skb: buffer to process
3307 * If there is no free memory -ENOMEM is returned, otherwise zero
3308 * is returned and the old skb data released.
3310 static inline int skb_linearize_cow(struct sk_buff *skb)
3312 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3313 __skb_linearize(skb) : 0;
3316 static __always_inline void
3317 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3320 if (skb->ip_summed == CHECKSUM_COMPLETE)
3321 skb->csum = csum_block_sub(skb->csum,
3322 csum_partial(start, len, 0), off);
3323 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3324 skb_checksum_start_offset(skb) < 0)
3325 skb->ip_summed = CHECKSUM_NONE;
3329 * skb_postpull_rcsum - update checksum for received skb after pull
3330 * @skb: buffer to update
3331 * @start: start of data before pull
3332 * @len: length of data pulled
3334 * After doing a pull on a received packet, you need to call this to
3335 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3336 * CHECKSUM_NONE so that it can be recomputed from scratch.
3338 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3339 const void *start, unsigned int len)
3341 __skb_postpull_rcsum(skb, start, len, 0);
3344 static __always_inline void
3345 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3348 if (skb->ip_summed == CHECKSUM_COMPLETE)
3349 skb->csum = csum_block_add(skb->csum,
3350 csum_partial(start, len, 0), off);
3354 * skb_postpush_rcsum - update checksum for received skb after push
3355 * @skb: buffer to update
3356 * @start: start of data after push
3357 * @len: length of data pushed
3359 * After doing a push on a received packet, you need to call this to
3360 * update the CHECKSUM_COMPLETE checksum.
3362 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3363 const void *start, unsigned int len)
3365 __skb_postpush_rcsum(skb, start, len, 0);
3368 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3371 * skb_push_rcsum - push skb and update receive checksum
3372 * @skb: buffer to update
3373 * @len: length of data pulled
3375 * This function performs an skb_push on the packet and updates
3376 * the CHECKSUM_COMPLETE checksum. It should be used on
3377 * receive path processing instead of skb_push unless you know
3378 * that the checksum difference is zero (e.g., a valid IP header)
3379 * or you are setting ip_summed to CHECKSUM_NONE.
3381 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3384 skb_postpush_rcsum(skb, skb->data, len);
3388 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3390 * pskb_trim_rcsum - trim received skb and update checksum
3391 * @skb: buffer to trim
3394 * This is exactly the same as pskb_trim except that it ensures the
3395 * checksum of received packets are still valid after the operation.
3396 * It can change skb pointers.
3399 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3401 if (likely(len >= skb->len))
3403 return pskb_trim_rcsum_slow(skb, len);
3406 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3408 if (skb->ip_summed == CHECKSUM_COMPLETE)
3409 skb->ip_summed = CHECKSUM_NONE;
3410 __skb_trim(skb, len);
3414 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3416 if (skb->ip_summed == CHECKSUM_COMPLETE)
3417 skb->ip_summed = CHECKSUM_NONE;
3418 return __skb_grow(skb, len);
3421 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3422 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3423 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3424 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3425 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3427 #define skb_queue_walk(queue, skb) \
3428 for (skb = (queue)->next; \
3429 skb != (struct sk_buff *)(queue); \
3432 #define skb_queue_walk_safe(queue, skb, tmp) \
3433 for (skb = (queue)->next, tmp = skb->next; \
3434 skb != (struct sk_buff *)(queue); \
3435 skb = tmp, tmp = skb->next)
3437 #define skb_queue_walk_from(queue, skb) \
3438 for (; skb != (struct sk_buff *)(queue); \
3441 #define skb_rbtree_walk(skb, root) \
3442 for (skb = skb_rb_first(root); skb != NULL; \
3443 skb = skb_rb_next(skb))
3445 #define skb_rbtree_walk_from(skb) \
3446 for (; skb != NULL; \
3447 skb = skb_rb_next(skb))
3449 #define skb_rbtree_walk_from_safe(skb, tmp) \
3450 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3453 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3454 for (tmp = skb->next; \
3455 skb != (struct sk_buff *)(queue); \
3456 skb = tmp, tmp = skb->next)
3458 #define skb_queue_reverse_walk(queue, skb) \
3459 for (skb = (queue)->prev; \
3460 skb != (struct sk_buff *)(queue); \
3463 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3464 for (skb = (queue)->prev, tmp = skb->prev; \
3465 skb != (struct sk_buff *)(queue); \
3466 skb = tmp, tmp = skb->prev)
3468 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3469 for (tmp = skb->prev; \
3470 skb != (struct sk_buff *)(queue); \
3471 skb = tmp, tmp = skb->prev)
3473 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3475 return skb_shinfo(skb)->frag_list != NULL;
3478 static inline void skb_frag_list_init(struct sk_buff *skb)
3480 skb_shinfo(skb)->frag_list = NULL;
3483 #define skb_walk_frags(skb, iter) \
3484 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3487 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3488 int *err, long *timeo_p,
3489 const struct sk_buff *skb);
3490 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3491 struct sk_buff_head *queue,
3494 struct sk_buff **last);
3495 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
3496 struct sk_buff_head *queue,
3497 unsigned int flags, int *off, int *err,
3498 struct sk_buff **last);
3499 struct sk_buff *__skb_recv_datagram(struct sock *sk,
3500 struct sk_buff_head *sk_queue,
3501 unsigned int flags, int *off, int *err);
3502 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3504 __poll_t datagram_poll(struct file *file, struct socket *sock,
3505 struct poll_table_struct *wait);
3506 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3507 struct iov_iter *to, int size);
3508 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3509 struct msghdr *msg, int size)
3511 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3513 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3514 struct msghdr *msg);
3515 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3516 struct iov_iter *to, int len,
3517 struct ahash_request *hash);
3518 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3519 struct iov_iter *from, int len);
3520 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3521 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3522 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3523 static inline void skb_free_datagram_locked(struct sock *sk,
3524 struct sk_buff *skb)
3526 __skb_free_datagram_locked(sk, skb, 0);
3528 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3529 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3530 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3531 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3532 int len, __wsum csum);
3533 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3534 struct pipe_inode_info *pipe, unsigned int len,
3535 unsigned int flags);
3536 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3538 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3539 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3540 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3542 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3543 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3544 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3545 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3546 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3547 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3548 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
3549 unsigned int offset);
3550 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3551 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3552 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3553 int skb_vlan_pop(struct sk_buff *skb);
3554 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3555 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
3556 int mac_len, bool ethernet);
3557 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
3559 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3560 int skb_mpls_dec_ttl(struct sk_buff *skb);
3561 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3564 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3566 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3569 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3571 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3574 struct skb_checksum_ops {
3575 __wsum (*update)(const void *mem, int len, __wsum wsum);
3576 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3579 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3581 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3582 __wsum csum, const struct skb_checksum_ops *ops);
3583 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3586 static inline void * __must_check
3587 __skb_header_pointer(const struct sk_buff *skb, int offset,
3588 int len, void *data, int hlen, void *buffer)
3590 if (hlen - offset >= len)
3591 return data + offset;
3594 skb_copy_bits(skb, offset, buffer, len) < 0)
3600 static inline void * __must_check
3601 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3603 return __skb_header_pointer(skb, offset, len, skb->data,
3604 skb_headlen(skb), buffer);
3608 * skb_needs_linearize - check if we need to linearize a given skb
3609 * depending on the given device features.
3610 * @skb: socket buffer to check
3611 * @features: net device features
3613 * Returns true if either:
3614 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3615 * 2. skb is fragmented and the device does not support SG.
3617 static inline bool skb_needs_linearize(struct sk_buff *skb,
3618 netdev_features_t features)
3620 return skb_is_nonlinear(skb) &&
3621 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3622 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3625 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3627 const unsigned int len)
3629 memcpy(to, skb->data, len);
3632 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3633 const int offset, void *to,
3634 const unsigned int len)
3636 memcpy(to, skb->data + offset, len);
3639 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3641 const unsigned int len)
3643 memcpy(skb->data, from, len);
3646 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3649 const unsigned int len)
3651 memcpy(skb->data + offset, from, len);
3654 void skb_init(void);
3656 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3662 * skb_get_timestamp - get timestamp from a skb
3663 * @skb: skb to get stamp from
3664 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
3666 * Timestamps are stored in the skb as offsets to a base timestamp.
3667 * This function converts the offset back to a struct timeval and stores
3670 static inline void skb_get_timestamp(const struct sk_buff *skb,
3671 struct __kernel_old_timeval *stamp)
3673 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
3676 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
3677 struct __kernel_sock_timeval *stamp)
3679 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3681 stamp->tv_sec = ts.tv_sec;
3682 stamp->tv_usec = ts.tv_nsec / 1000;
3685 static inline void skb_get_timestampns(const struct sk_buff *skb,
3686 struct __kernel_old_timespec *stamp)
3688 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3690 stamp->tv_sec = ts.tv_sec;
3691 stamp->tv_nsec = ts.tv_nsec;
3694 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
3695 struct __kernel_timespec *stamp)
3697 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3699 stamp->tv_sec = ts.tv_sec;
3700 stamp->tv_nsec = ts.tv_nsec;
3703 static inline void __net_timestamp(struct sk_buff *skb)
3705 skb->tstamp = ktime_get_real();
3708 static inline ktime_t net_timedelta(ktime_t t)
3710 return ktime_sub(ktime_get_real(), t);
3713 static inline ktime_t net_invalid_timestamp(void)
3718 static inline u8 skb_metadata_len(const struct sk_buff *skb)
3720 return skb_shinfo(skb)->meta_len;
3723 static inline void *skb_metadata_end(const struct sk_buff *skb)
3725 return skb_mac_header(skb);
3728 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3729 const struct sk_buff *skb_b,
3732 const void *a = skb_metadata_end(skb_a);
3733 const void *b = skb_metadata_end(skb_b);
3734 /* Using more efficient varaiant than plain call to memcmp(). */
3735 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3739 #define __it(x, op) (x -= sizeof(u##op))
3740 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3741 case 32: diffs |= __it_diff(a, b, 64);
3743 case 24: diffs |= __it_diff(a, b, 64);
3745 case 16: diffs |= __it_diff(a, b, 64);
3747 case 8: diffs |= __it_diff(a, b, 64);
3749 case 28: diffs |= __it_diff(a, b, 64);
3751 case 20: diffs |= __it_diff(a, b, 64);
3753 case 12: diffs |= __it_diff(a, b, 64);
3755 case 4: diffs |= __it_diff(a, b, 32);
3760 return memcmp(a - meta_len, b - meta_len, meta_len);
3764 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3765 const struct sk_buff *skb_b)
3767 u8 len_a = skb_metadata_len(skb_a);
3768 u8 len_b = skb_metadata_len(skb_b);
3770 if (!(len_a | len_b))
3773 return len_a != len_b ?
3774 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3777 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3779 skb_shinfo(skb)->meta_len = meta_len;
3782 static inline void skb_metadata_clear(struct sk_buff *skb)
3784 skb_metadata_set(skb, 0);
3787 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3789 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3791 void skb_clone_tx_timestamp(struct sk_buff *skb);
3792 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3794 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3796 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3800 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3805 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3808 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3810 * PHY drivers may accept clones of transmitted packets for
3811 * timestamping via their phy_driver.txtstamp method. These drivers
3812 * must call this function to return the skb back to the stack with a
3815 * @skb: clone of the the original outgoing packet
3816 * @hwtstamps: hardware time stamps
3819 void skb_complete_tx_timestamp(struct sk_buff *skb,
3820 struct skb_shared_hwtstamps *hwtstamps);
3822 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3823 struct skb_shared_hwtstamps *hwtstamps,
3824 struct sock *sk, int tstype);
3827 * skb_tstamp_tx - queue clone of skb with send time stamps
3828 * @orig_skb: the original outgoing packet
3829 * @hwtstamps: hardware time stamps, may be NULL if not available
3831 * If the skb has a socket associated, then this function clones the
3832 * skb (thus sharing the actual data and optional structures), stores
3833 * the optional hardware time stamping information (if non NULL) or
3834 * generates a software time stamp (otherwise), then queues the clone
3835 * to the error queue of the socket. Errors are silently ignored.
3837 void skb_tstamp_tx(struct sk_buff *orig_skb,
3838 struct skb_shared_hwtstamps *hwtstamps);
3841 * skb_tx_timestamp() - Driver hook for transmit timestamping
3843 * Ethernet MAC Drivers should call this function in their hard_xmit()
3844 * function immediately before giving the sk_buff to the MAC hardware.
3846 * Specifically, one should make absolutely sure that this function is
3847 * called before TX completion of this packet can trigger. Otherwise
3848 * the packet could potentially already be freed.
3850 * @skb: A socket buffer.
3852 static inline void skb_tx_timestamp(struct sk_buff *skb)
3854 skb_clone_tx_timestamp(skb);
3855 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3856 skb_tstamp_tx(skb, NULL);
3860 * skb_complete_wifi_ack - deliver skb with wifi status
3862 * @skb: the original outgoing packet
3863 * @acked: ack status
3866 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3868 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3869 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3871 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3873 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3875 (skb->ip_summed == CHECKSUM_PARTIAL &&
3876 skb_checksum_start_offset(skb) >= 0));
3880 * skb_checksum_complete - Calculate checksum of an entire packet
3881 * @skb: packet to process
3883 * This function calculates the checksum over the entire packet plus
3884 * the value of skb->csum. The latter can be used to supply the
3885 * checksum of a pseudo header as used by TCP/UDP. It returns the
3888 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3889 * this function can be used to verify that checksum on received
3890 * packets. In that case the function should return zero if the
3891 * checksum is correct. In particular, this function will return zero
3892 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3893 * hardware has already verified the correctness of the checksum.
3895 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3897 return skb_csum_unnecessary(skb) ?
3898 0 : __skb_checksum_complete(skb);
3901 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3903 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3904 if (skb->csum_level == 0)
3905 skb->ip_summed = CHECKSUM_NONE;
3911 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3913 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3914 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3916 } else if (skb->ip_summed == CHECKSUM_NONE) {
3917 skb->ip_summed = CHECKSUM_UNNECESSARY;
3918 skb->csum_level = 0;
3922 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
3924 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3925 skb->ip_summed = CHECKSUM_NONE;
3926 skb->csum_level = 0;
3930 /* Check if we need to perform checksum complete validation.
3932 * Returns true if checksum complete is needed, false otherwise
3933 * (either checksum is unnecessary or zero checksum is allowed).
3935 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3939 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3940 skb->csum_valid = 1;
3941 __skb_decr_checksum_unnecessary(skb);
3948 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
3951 #define CHECKSUM_BREAK 76
3953 /* Unset checksum-complete
3955 * Unset checksum complete can be done when packet is being modified
3956 * (uncompressed for instance) and checksum-complete value is
3959 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3961 if (skb->ip_summed == CHECKSUM_COMPLETE)
3962 skb->ip_summed = CHECKSUM_NONE;
3965 /* Validate (init) checksum based on checksum complete.
3968 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3969 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3970 * checksum is stored in skb->csum for use in __skb_checksum_complete
3971 * non-zero: value of invalid checksum
3974 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3978 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3979 if (!csum_fold(csum_add(psum, skb->csum))) {
3980 skb->csum_valid = 1;
3987 if (complete || skb->len <= CHECKSUM_BREAK) {
3990 csum = __skb_checksum_complete(skb);
3991 skb->csum_valid = !csum;
3998 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4003 /* Perform checksum validate (init). Note that this is a macro since we only
4004 * want to calculate the pseudo header which is an input function if necessary.
4005 * First we try to validate without any computation (checksum unnecessary) and
4006 * then calculate based on checksum complete calling the function to compute
4010 * 0: checksum is validated or try to in skb_checksum_complete
4011 * non-zero: value of invalid checksum
4013 #define __skb_checksum_validate(skb, proto, complete, \
4014 zero_okay, check, compute_pseudo) \
4016 __sum16 __ret = 0; \
4017 skb->csum_valid = 0; \
4018 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4019 __ret = __skb_checksum_validate_complete(skb, \
4020 complete, compute_pseudo(skb, proto)); \
4024 #define skb_checksum_init(skb, proto, compute_pseudo) \
4025 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4027 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4028 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4030 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4031 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4033 #define skb_checksum_validate_zero_check(skb, proto, check, \
4035 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4037 #define skb_checksum_simple_validate(skb) \
4038 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4040 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4042 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4045 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4047 skb->csum = ~pseudo;
4048 skb->ip_summed = CHECKSUM_COMPLETE;
4051 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4053 if (__skb_checksum_convert_check(skb)) \
4054 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4057 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4058 u16 start, u16 offset)
4060 skb->ip_summed = CHECKSUM_PARTIAL;
4061 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4062 skb->csum_offset = offset - start;
4065 /* Update skbuf and packet to reflect the remote checksum offload operation.
4066 * When called, ptr indicates the starting point for skb->csum when
4067 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4068 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4070 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4071 int start, int offset, bool nopartial)
4076 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4080 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4081 __skb_checksum_complete(skb);
4082 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4085 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4087 /* Adjust skb->csum since we changed the packet */
4088 skb->csum = csum_add(skb->csum, delta);
4091 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4093 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4094 return (void *)(skb->_nfct & NFCT_PTRMASK);
4100 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4102 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4109 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4111 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4116 #ifdef CONFIG_SKB_EXTENSIONS
4118 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4124 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4127 #if IS_ENABLED(CONFIG_MPTCP)
4130 SKB_EXT_NUM, /* must be last */
4134 * struct skb_ext - sk_buff extensions
4135 * @refcnt: 1 on allocation, deallocated on 0
4136 * @offset: offset to add to @data to obtain extension address
4137 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4138 * @data: start of extension data, variable sized
4140 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4141 * to use 'u8' types while allowing up to 2kb worth of extension data.
4145 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4146 u8 chunks; /* same */
4147 char data[] __aligned(8);
4150 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4151 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4152 struct skb_ext *ext);
4153 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4154 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4155 void __skb_ext_put(struct skb_ext *ext);
4157 static inline void skb_ext_put(struct sk_buff *skb)
4159 if (skb->active_extensions)
4160 __skb_ext_put(skb->extensions);
4163 static inline void __skb_ext_copy(struct sk_buff *dst,
4164 const struct sk_buff *src)
4166 dst->active_extensions = src->active_extensions;
4168 if (src->active_extensions) {
4169 struct skb_ext *ext = src->extensions;
4171 refcount_inc(&ext->refcnt);
4172 dst->extensions = ext;
4176 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4179 __skb_ext_copy(dst, src);
4182 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4184 return !!ext->offset[i];
4187 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4189 return skb->active_extensions & (1 << id);
4192 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4194 if (skb_ext_exist(skb, id))
4195 __skb_ext_del(skb, id);
4198 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4200 if (skb_ext_exist(skb, id)) {
4201 struct skb_ext *ext = skb->extensions;
4203 return (void *)ext + (ext->offset[id] << 3);
4209 static inline void skb_ext_reset(struct sk_buff *skb)
4211 if (unlikely(skb->active_extensions)) {
4212 __skb_ext_put(skb->extensions);
4213 skb->active_extensions = 0;
4217 static inline bool skb_has_extensions(struct sk_buff *skb)
4219 return unlikely(skb->active_extensions);
4222 static inline void skb_ext_put(struct sk_buff *skb) {}
4223 static inline void skb_ext_reset(struct sk_buff *skb) {}
4224 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4225 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4226 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4227 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4228 #endif /* CONFIG_SKB_EXTENSIONS */
4230 static inline void nf_reset_ct(struct sk_buff *skb)
4232 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4233 nf_conntrack_put(skb_nfct(skb));
4238 static inline void nf_reset_trace(struct sk_buff *skb)
4240 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4245 static inline void ipvs_reset(struct sk_buff *skb)
4247 #if IS_ENABLED(CONFIG_IP_VS)
4248 skb->ipvs_property = 0;
4252 /* Note: This doesn't put any conntrack info in dst. */
4253 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4256 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4257 dst->_nfct = src->_nfct;
4258 nf_conntrack_get(skb_nfct(src));
4260 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4262 dst->nf_trace = src->nf_trace;
4266 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4268 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4269 nf_conntrack_put(skb_nfct(dst));
4271 __nf_copy(dst, src, true);
4274 #ifdef CONFIG_NETWORK_SECMARK
4275 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4277 to->secmark = from->secmark;
4280 static inline void skb_init_secmark(struct sk_buff *skb)
4285 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4288 static inline void skb_init_secmark(struct sk_buff *skb)
4292 static inline int secpath_exists(const struct sk_buff *skb)
4295 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4301 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4303 return !skb->destructor &&
4304 !secpath_exists(skb) &&
4306 !skb->_skb_refdst &&
4307 !skb_has_frag_list(skb);
4310 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4312 skb->queue_mapping = queue_mapping;
4315 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4317 return skb->queue_mapping;
4320 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4322 to->queue_mapping = from->queue_mapping;
4325 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4327 skb->queue_mapping = rx_queue + 1;
4330 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4332 return skb->queue_mapping - 1;
4335 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4337 return skb->queue_mapping != 0;
4340 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4342 skb->dst_pending_confirm = val;
4345 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4347 return skb->dst_pending_confirm != 0;
4350 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4353 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4359 /* Keeps track of mac header offset relative to skb->head.
4360 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4361 * For non-tunnel skb it points to skb_mac_header() and for
4362 * tunnel skb it points to outer mac header.
4363 * Keeps track of level of encapsulation of network headers.
4374 #define SKB_GSO_CB_OFFSET 32
4375 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET))
4377 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4379 return (skb_mac_header(inner_skb) - inner_skb->head) -
4380 SKB_GSO_CB(inner_skb)->mac_offset;
4383 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4385 int new_headroom, headroom;
4388 headroom = skb_headroom(skb);
4389 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4393 new_headroom = skb_headroom(skb);
4394 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4398 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4400 /* Do not update partial checksums if remote checksum is enabled. */
4401 if (skb->remcsum_offload)
4404 SKB_GSO_CB(skb)->csum = res;
4405 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4408 /* Compute the checksum for a gso segment. First compute the checksum value
4409 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4410 * then add in skb->csum (checksum from csum_start to end of packet).
4411 * skb->csum and csum_start are then updated to reflect the checksum of the
4412 * resultant packet starting from the transport header-- the resultant checksum
4413 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4416 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4418 unsigned char *csum_start = skb_transport_header(skb);
4419 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4420 __wsum partial = SKB_GSO_CB(skb)->csum;
4422 SKB_GSO_CB(skb)->csum = res;
4423 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4425 return csum_fold(csum_partial(csum_start, plen, partial));
4428 static inline bool skb_is_gso(const struct sk_buff *skb)
4430 return skb_shinfo(skb)->gso_size;
4433 /* Note: Should be called only if skb_is_gso(skb) is true */
4434 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4436 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4439 /* Note: Should be called only if skb_is_gso(skb) is true */
4440 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4442 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4445 /* Note: Should be called only if skb_is_gso(skb) is true */
4446 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4448 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4451 static inline void skb_gso_reset(struct sk_buff *skb)
4453 skb_shinfo(skb)->gso_size = 0;
4454 skb_shinfo(skb)->gso_segs = 0;
4455 skb_shinfo(skb)->gso_type = 0;
4458 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4461 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4463 shinfo->gso_size += increment;
4466 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4469 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4471 shinfo->gso_size -= decrement;
4474 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4476 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4478 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4479 * wanted then gso_type will be set. */
4480 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4482 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4483 unlikely(shinfo->gso_type == 0)) {
4484 __skb_warn_lro_forwarding(skb);
4490 static inline void skb_forward_csum(struct sk_buff *skb)
4492 /* Unfortunately we don't support this one. Any brave souls? */
4493 if (skb->ip_summed == CHECKSUM_COMPLETE)
4494 skb->ip_summed = CHECKSUM_NONE;
4498 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4499 * @skb: skb to check
4501 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4502 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4503 * use this helper, to document places where we make this assertion.
4505 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4508 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4512 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4514 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4515 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4516 unsigned int transport_len,
4517 __sum16(*skb_chkf)(struct sk_buff *skb));
4520 * skb_head_is_locked - Determine if the skb->head is locked down
4521 * @skb: skb to check
4523 * The head on skbs build around a head frag can be removed if they are
4524 * not cloned. This function returns true if the skb head is locked down
4525 * due to either being allocated via kmalloc, or by being a clone with
4526 * multiple references to the head.
4528 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4530 return !skb->head_frag || skb_cloned(skb);
4533 /* Local Checksum Offload.
4534 * Compute outer checksum based on the assumption that the
4535 * inner checksum will be offloaded later.
4536 * See Documentation/networking/checksum-offloads.rst for
4537 * explanation of how this works.
4538 * Fill in outer checksum adjustment (e.g. with sum of outer
4539 * pseudo-header) before calling.
4540 * Also ensure that inner checksum is in linear data area.
4542 static inline __wsum lco_csum(struct sk_buff *skb)
4544 unsigned char *csum_start = skb_checksum_start(skb);
4545 unsigned char *l4_hdr = skb_transport_header(skb);
4548 /* Start with complement of inner checksum adjustment */
4549 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4552 /* Add in checksum of our headers (incl. outer checksum
4553 * adjustment filled in by caller) and return result.
4555 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4558 static inline bool skb_is_redirected(const struct sk_buff *skb)
4560 #ifdef CONFIG_NET_REDIRECT
4561 return skb->redirected;
4567 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
4569 #ifdef CONFIG_NET_REDIRECT
4570 skb->redirected = 1;
4571 skb->from_ingress = from_ingress;
4572 if (skb->from_ingress)
4577 static inline void skb_reset_redirect(struct sk_buff *skb)
4579 #ifdef CONFIG_NET_REDIRECT
4580 skb->redirected = 0;
4584 #endif /* __KERNEL__ */
4585 #endif /* _LINUX_SKBUFF_H */