1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * Definitions for the 'struct sk_buff' memory handlers.
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)))
241 struct ahash_request;
244 struct pipe_inode_info;
251 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
252 struct nf_bridge_info {
254 BRNF_PROTO_UNCHANGED,
262 struct net_device *physindev;
264 /* always valid & non-NULL from FORWARD on, for physdev match */
265 struct net_device *physoutdev;
267 /* prerouting: detect dnat in orig/reply direction */
269 struct in6_addr ipv6_daddr;
271 /* after prerouting + nat detected: store original source
272 * mac since neigh resolution overwrites it, only used while
273 * skb is out in neigh layer.
275 char neigh_header[8];
280 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
281 /* Chain in tc_skb_ext will be used to share the tc chain with
282 * ovs recirc_id. It will be set to the current chain by tc
283 * and read by ovs to recirc_id.
291 struct sk_buff_head {
292 /* These two members must be first. */
293 struct sk_buff *next;
294 struct sk_buff *prev;
302 /* To allow 64K frame to be packed as single skb without frag_list we
303 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
304 * buffers which do not start on a page boundary.
306 * Since GRO uses frags we allocate at least 16 regardless of page
309 #if (65536/PAGE_SIZE + 1) < 16
310 #define MAX_SKB_FRAGS 16UL
312 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
314 extern int sysctl_max_skb_frags;
316 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
317 * segment using its current segmentation instead.
319 #define GSO_BY_FRAGS 0xFFFF
321 typedef struct bio_vec skb_frag_t;
324 * skb_frag_size() - Returns the size of a skb fragment
325 * @frag: skb fragment
327 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
333 * skb_frag_size_set() - Sets the size of a skb fragment
334 * @frag: skb fragment
335 * @size: size of fragment
337 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
343 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
344 * @frag: skb fragment
345 * @delta: value to add
347 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
349 frag->bv_len += delta;
353 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
354 * @frag: skb fragment
355 * @delta: value to subtract
357 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
359 frag->bv_len -= delta;
363 * skb_frag_must_loop - Test if %p is a high memory page
364 * @p: fragment's page
366 static inline bool skb_frag_must_loop(struct page *p)
368 #if defined(CONFIG_HIGHMEM)
376 * skb_frag_foreach_page - loop over pages in a fragment
378 * @f: skb frag to operate on
379 * @f_off: offset from start of f->bv_page
380 * @f_len: length from f_off to loop over
381 * @p: (temp var) current page
382 * @p_off: (temp var) offset from start of current page,
383 * non-zero only on first page.
384 * @p_len: (temp var) length in current page,
385 * < PAGE_SIZE only on first and last page.
386 * @copied: (temp var) length so far, excluding current p_len.
388 * A fragment can hold a compound page, in which case per-page
389 * operations, notably kmap_atomic, must be called for each
392 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
393 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
394 p_off = (f_off) & (PAGE_SIZE - 1), \
395 p_len = skb_frag_must_loop(p) ? \
396 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
399 copied += p_len, p++, p_off = 0, \
400 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
402 #define HAVE_HW_TIME_STAMP
405 * struct skb_shared_hwtstamps - hardware time stamps
406 * @hwtstamp: hardware time stamp transformed into duration
407 * since arbitrary point in time
409 * Software time stamps generated by ktime_get_real() are stored in
412 * hwtstamps can only be compared against other hwtstamps from
415 * This structure is attached to packets as part of the
416 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
418 struct skb_shared_hwtstamps {
422 /* Definitions for tx_flags in struct skb_shared_info */
424 /* generate hardware time stamp */
425 SKBTX_HW_TSTAMP = 1 << 0,
427 /* generate software time stamp when queueing packet to NIC */
428 SKBTX_SW_TSTAMP = 1 << 1,
430 /* device driver is going to provide hardware time stamp */
431 SKBTX_IN_PROGRESS = 1 << 2,
433 /* device driver supports TX zero-copy buffers */
434 SKBTX_DEV_ZEROCOPY = 1 << 3,
436 /* generate wifi status information (where possible) */
437 SKBTX_WIFI_STATUS = 1 << 4,
439 /* This indicates at least one fragment might be overwritten
440 * (as in vmsplice(), sendfile() ...)
441 * If we need to compute a TX checksum, we'll need to copy
442 * all frags to avoid possible bad checksum
444 SKBTX_SHARED_FRAG = 1 << 5,
446 /* generate software time stamp when entering packet scheduling */
447 SKBTX_SCHED_TSTAMP = 1 << 6,
450 #define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
451 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
453 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
456 * The callback notifies userspace to release buffers when skb DMA is done in
457 * lower device, the skb last reference should be 0 when calling this.
458 * The zerocopy_success argument is true if zero copy transmit occurred,
459 * false on data copy or out of memory error caused by data copy attempt.
460 * The ctx field is used to track device context.
461 * The desc field is used to track userspace buffer index.
464 void (*callback)(struct ubuf_info *, bool zerocopy_success);
480 struct user_struct *user;
485 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
487 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
488 void mm_unaccount_pinned_pages(struct mmpin *mmp);
490 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
491 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
492 struct ubuf_info *uarg);
494 static inline void sock_zerocopy_get(struct ubuf_info *uarg)
496 refcount_inc(&uarg->refcnt);
499 void sock_zerocopy_put(struct ubuf_info *uarg);
500 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
502 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
504 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len);
505 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
506 struct msghdr *msg, int len,
507 struct ubuf_info *uarg);
509 /* This data is invariant across clones and lives at
510 * the end of the header data, ie. at skb->end.
512 struct skb_shared_info {
517 unsigned short gso_size;
518 /* Warning: this field is not always filled in (UFO)! */
519 unsigned short gso_segs;
520 struct sk_buff *frag_list;
521 struct skb_shared_hwtstamps hwtstamps;
522 unsigned int gso_type;
526 * Warning : all fields before dataref are cleared in __alloc_skb()
530 /* Intermediate layers must ensure that destructor_arg
531 * remains valid until skb destructor */
532 void * destructor_arg;
534 /* must be last field, see pskb_expand_head() */
535 skb_frag_t frags[MAX_SKB_FRAGS];
538 /* We divide dataref into two halves. The higher 16 bits hold references
539 * to the payload part of skb->data. The lower 16 bits hold references to
540 * the entire skb->data. A clone of a headerless skb holds the length of
541 * the header in skb->hdr_len.
543 * All users must obey the rule that the skb->data reference count must be
544 * greater than or equal to the payload reference count.
546 * Holding a reference to the payload part means that the user does not
547 * care about modifications to the header part of skb->data.
549 #define SKB_DATAREF_SHIFT 16
550 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
554 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
555 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
556 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
560 SKB_GSO_TCPV4 = 1 << 0,
562 /* This indicates the skb is from an untrusted source. */
563 SKB_GSO_DODGY = 1 << 1,
565 /* This indicates the tcp segment has CWR set. */
566 SKB_GSO_TCP_ECN = 1 << 2,
568 SKB_GSO_TCP_FIXEDID = 1 << 3,
570 SKB_GSO_TCPV6 = 1 << 4,
572 SKB_GSO_FCOE = 1 << 5,
574 SKB_GSO_GRE = 1 << 6,
576 SKB_GSO_GRE_CSUM = 1 << 7,
578 SKB_GSO_IPXIP4 = 1 << 8,
580 SKB_GSO_IPXIP6 = 1 << 9,
582 SKB_GSO_UDP_TUNNEL = 1 << 10,
584 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
586 SKB_GSO_PARTIAL = 1 << 12,
588 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
590 SKB_GSO_SCTP = 1 << 14,
592 SKB_GSO_ESP = 1 << 15,
594 SKB_GSO_UDP = 1 << 16,
596 SKB_GSO_UDP_L4 = 1 << 17,
598 SKB_GSO_FRAGLIST = 1 << 18,
601 #if BITS_PER_LONG > 32
602 #define NET_SKBUFF_DATA_USES_OFFSET 1
605 #ifdef NET_SKBUFF_DATA_USES_OFFSET
606 typedef unsigned int sk_buff_data_t;
608 typedef unsigned char *sk_buff_data_t;
612 * struct sk_buff - socket buffer
613 * @next: Next buffer in list
614 * @prev: Previous buffer in list
615 * @tstamp: Time we arrived/left
616 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
617 * for retransmit timer
618 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
620 * @sk: Socket we are owned by
621 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
622 * fragmentation management
623 * @dev: Device we arrived on/are leaving by
624 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
625 * @cb: Control buffer. Free for use by every layer. Put private vars here
626 * @_skb_refdst: destination entry (with norefcount bit)
627 * @sp: the security path, used for xfrm
628 * @len: Length of actual data
629 * @data_len: Data length
630 * @mac_len: Length of link layer header
631 * @hdr_len: writable header length of cloned skb
632 * @csum: Checksum (must include start/offset pair)
633 * @csum_start: Offset from skb->head where checksumming should start
634 * @csum_offset: Offset from csum_start where checksum should be stored
635 * @priority: Packet queueing priority
636 * @ignore_df: allow local fragmentation
637 * @cloned: Head may be cloned (check refcnt to be sure)
638 * @ip_summed: Driver fed us an IP checksum
639 * @nohdr: Payload reference only, must not modify header
640 * @pkt_type: Packet class
641 * @fclone: skbuff clone status
642 * @ipvs_property: skbuff is owned by ipvs
643 * @inner_protocol_type: whether the inner protocol is
644 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
645 * @remcsum_offload: remote checksum offload is enabled
646 * @offload_fwd_mark: Packet was L2-forwarded in hardware
647 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
648 * @tc_skip_classify: do not classify packet. set by IFB device
649 * @tc_at_ingress: used within tc_classify to distinguish in/egress
650 * @redirected: packet was redirected by packet classifier
651 * @from_ingress: packet was redirected from the ingress path
652 * @peeked: this packet has been seen already, so stats have been
653 * done for it, don't do them again
654 * @nf_trace: netfilter packet trace flag
655 * @protocol: Packet protocol from driver
656 * @destructor: Destruct function
657 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
658 * @_nfct: Associated connection, if any (with nfctinfo bits)
659 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
660 * @skb_iif: ifindex of device we arrived on
661 * @tc_index: Traffic control index
662 * @hash: the packet hash
663 * @queue_mapping: Queue mapping for multiqueue devices
664 * @head_frag: skb was allocated from page fragments,
665 * not allocated by kmalloc() or vmalloc().
666 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
667 * @active_extensions: active extensions (skb_ext_id types)
668 * @ndisc_nodetype: router type (from link layer)
669 * @ooo_okay: allow the mapping of a socket to a queue to be changed
670 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
672 * @sw_hash: indicates hash was computed in software stack
673 * @wifi_acked_valid: wifi_acked was set
674 * @wifi_acked: whether frame was acked on wifi or not
675 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
676 * @encapsulation: indicates the inner headers in the skbuff are valid
677 * @encap_hdr_csum: software checksum is needed
678 * @csum_valid: checksum is already valid
679 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
680 * @csum_complete_sw: checksum was completed by software
681 * @csum_level: indicates the number of consecutive checksums found in
682 * the packet minus one that have been verified as
683 * CHECKSUM_UNNECESSARY (max 3)
684 * @dst_pending_confirm: need to confirm neighbour
685 * @decrypted: Decrypted SKB
686 * @napi_id: id of the NAPI struct this skb came from
687 * @sender_cpu: (aka @napi_id) source CPU in XPS
688 * @secmark: security marking
689 * @mark: Generic packet mark
690 * @reserved_tailroom: (aka @mark) number of bytes of free space available
691 * at the tail of an sk_buff
692 * @vlan_present: VLAN tag is present
693 * @vlan_proto: vlan encapsulation protocol
694 * @vlan_tci: vlan tag control information
695 * @inner_protocol: Protocol (encapsulation)
696 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
697 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
698 * @inner_transport_header: Inner transport layer header (encapsulation)
699 * @inner_network_header: Network layer header (encapsulation)
700 * @inner_mac_header: Link layer header (encapsulation)
701 * @transport_header: Transport layer header
702 * @network_header: Network layer header
703 * @mac_header: Link layer header
704 * @tail: Tail pointer
706 * @head: Head of buffer
707 * @data: Data head pointer
708 * @truesize: Buffer size
709 * @users: User count - see {datagram,tcp}.c
710 * @extensions: allocated extensions, valid if active_extensions is nonzero
716 /* These two members must be first. */
717 struct sk_buff *next;
718 struct sk_buff *prev;
721 struct net_device *dev;
722 /* Some protocols might use this space to store information,
723 * while device pointer would be NULL.
724 * UDP receive path is one user.
726 unsigned long dev_scratch;
729 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
730 struct list_head list;
735 int ip_defrag_offset;
740 u64 skb_mstamp_ns; /* earliest departure time */
743 * This is the control buffer. It is free to use for every
744 * layer. Please put your private variables there. If you
745 * want to keep them across layers you have to do a skb_clone()
746 * first. This is owned by whoever has the skb queued ATM.
748 char cb[48] __aligned(8);
752 unsigned long _skb_refdst;
753 void (*destructor)(struct sk_buff *skb);
755 struct list_head tcp_tsorted_anchor;
758 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
766 /* Following fields are _not_ copied in __copy_skb_header()
767 * Note that queue_mapping is here mostly to fill a hole.
771 /* if you move cloned around you also must adapt those constants */
772 #ifdef __BIG_ENDIAN_BITFIELD
773 #define CLONED_MASK (1 << 7)
775 #define CLONED_MASK 1
777 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
780 __u8 __cloned_offset[0];
788 #ifdef CONFIG_SKB_EXTENSIONS
789 __u8 active_extensions;
791 /* fields enclosed in headers_start/headers_end are copied
792 * using a single memcpy() in __copy_skb_header()
795 __u32 headers_start[0];
798 /* if you move pkt_type around you also must adapt those constants */
799 #ifdef __BIG_ENDIAN_BITFIELD
800 #define PKT_TYPE_MAX (7 << 5)
802 #define PKT_TYPE_MAX 7
804 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
807 __u8 __pkt_type_offset[0];
817 __u8 wifi_acked_valid:1;
820 /* Indicates the inner headers are valid in the skbuff. */
821 __u8 encapsulation:1;
822 __u8 encap_hdr_csum:1;
825 #ifdef __BIG_ENDIAN_BITFIELD
826 #define PKT_VLAN_PRESENT_BIT 7
828 #define PKT_VLAN_PRESENT_BIT 0
830 #define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset)
832 __u8 __pkt_vlan_present_offset[0];
835 __u8 csum_complete_sw:1;
837 __u8 csum_not_inet:1;
838 __u8 dst_pending_confirm:1;
839 #ifdef CONFIG_IPV6_NDISC_NODETYPE
840 __u8 ndisc_nodetype:2;
843 __u8 ipvs_property:1;
844 __u8 inner_protocol_type:1;
845 __u8 remcsum_offload:1;
846 #ifdef CONFIG_NET_SWITCHDEV
847 __u8 offload_fwd_mark:1;
848 __u8 offload_l3_fwd_mark:1;
850 #ifdef CONFIG_NET_CLS_ACT
851 __u8 tc_skip_classify:1;
852 __u8 tc_at_ingress:1;
854 #ifdef CONFIG_NET_REDIRECT
858 #ifdef CONFIG_TLS_DEVICE
862 #ifdef CONFIG_NET_SCHED
863 __u16 tc_index; /* traffic control index */
878 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
880 unsigned int napi_id;
881 unsigned int sender_cpu;
884 #ifdef CONFIG_NETWORK_SECMARK
890 __u32 reserved_tailroom;
894 __be16 inner_protocol;
898 __u16 inner_transport_header;
899 __u16 inner_network_header;
900 __u16 inner_mac_header;
903 __u16 transport_header;
904 __u16 network_header;
908 __u32 headers_end[0];
911 /* These elements must be at the end, see alloc_skb() for details. */
916 unsigned int truesize;
919 #ifdef CONFIG_SKB_EXTENSIONS
920 /* only useable after checking ->active_extensions != 0 */
921 struct skb_ext *extensions;
927 * Handling routines are only of interest to the kernel
930 #define SKB_ALLOC_FCLONE 0x01
931 #define SKB_ALLOC_RX 0x02
932 #define SKB_ALLOC_NAPI 0x04
935 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
938 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
940 return unlikely(skb->pfmemalloc);
944 * skb might have a dst pointer attached, refcounted or not.
945 * _skb_refdst low order bit is set if refcount was _not_ taken
947 #define SKB_DST_NOREF 1UL
948 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
951 * skb_dst - returns skb dst_entry
954 * Returns skb dst_entry, regardless of reference taken or not.
956 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
958 /* If refdst was not refcounted, check we still are in a
959 * rcu_read_lock section
961 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
962 !rcu_read_lock_held() &&
963 !rcu_read_lock_bh_held());
964 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
968 * skb_dst_set - sets skb dst
972 * Sets skb dst, assuming a reference was taken on dst and should
973 * be released by skb_dst_drop()
975 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
977 skb->_skb_refdst = (unsigned long)dst;
981 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
985 * Sets skb dst, assuming a reference was not taken on dst.
986 * If dst entry is cached, we do not take reference and dst_release
987 * will be avoided by refdst_drop. If dst entry is not cached, we take
988 * reference, so that last dst_release can destroy the dst immediately.
990 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
992 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
993 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
997 * skb_dst_is_noref - Test if skb dst isn't refcounted
1000 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1002 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1006 * skb_rtable - Returns the skb &rtable
1009 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1011 return (struct rtable *)skb_dst(skb);
1014 /* For mangling skb->pkt_type from user space side from applications
1015 * such as nft, tc, etc, we only allow a conservative subset of
1016 * possible pkt_types to be set.
1018 static inline bool skb_pkt_type_ok(u32 ptype)
1020 return ptype <= PACKET_OTHERHOST;
1024 * skb_napi_id - Returns the skb's NAPI id
1027 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1029 #ifdef CONFIG_NET_RX_BUSY_POLL
1030 return skb->napi_id;
1037 * skb_unref - decrement the skb's reference count
1040 * Returns true if we can free the skb.
1042 static inline bool skb_unref(struct sk_buff *skb)
1046 if (likely(refcount_read(&skb->users) == 1))
1048 else if (likely(!refcount_dec_and_test(&skb->users)))
1054 void skb_release_head_state(struct sk_buff *skb);
1055 void kfree_skb(struct sk_buff *skb);
1056 void kfree_skb_list(struct sk_buff *segs);
1057 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1058 void skb_tx_error(struct sk_buff *skb);
1059 void consume_skb(struct sk_buff *skb);
1060 void __consume_stateless_skb(struct sk_buff *skb);
1061 void __kfree_skb(struct sk_buff *skb);
1062 extern struct kmem_cache *skbuff_head_cache;
1064 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1065 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1066 bool *fragstolen, int *delta_truesize);
1068 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1070 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1071 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1072 struct sk_buff *build_skb_around(struct sk_buff *skb,
1073 void *data, unsigned int frag_size);
1076 * alloc_skb - allocate a network buffer
1077 * @size: size to allocate
1078 * @priority: allocation mask
1080 * This function is a convenient wrapper around __alloc_skb().
1082 static inline struct sk_buff *alloc_skb(unsigned int size,
1085 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1088 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1089 unsigned long data_len,
1093 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1095 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1096 struct sk_buff_fclones {
1097 struct sk_buff skb1;
1099 struct sk_buff skb2;
1101 refcount_t fclone_ref;
1105 * skb_fclone_busy - check if fclone is busy
1109 * Returns true if skb is a fast clone, and its clone is not freed.
1110 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1111 * so we also check that this didnt happen.
1113 static inline bool skb_fclone_busy(const struct sock *sk,
1114 const struct sk_buff *skb)
1116 const struct sk_buff_fclones *fclones;
1118 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1120 return skb->fclone == SKB_FCLONE_ORIG &&
1121 refcount_read(&fclones->fclone_ref) > 1 &&
1122 fclones->skb2.sk == sk;
1126 * alloc_skb_fclone - allocate a network buffer from fclone cache
1127 * @size: size to allocate
1128 * @priority: allocation mask
1130 * This function is a convenient wrapper around __alloc_skb().
1132 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1135 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1138 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1139 void skb_headers_offset_update(struct sk_buff *skb, int off);
1140 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1141 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1142 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1143 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1144 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1145 gfp_t gfp_mask, bool fclone);
1146 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1149 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1152 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1153 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1154 unsigned int headroom);
1155 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1156 int newtailroom, gfp_t priority);
1157 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1158 int offset, int len);
1159 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1160 int offset, int len);
1161 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1162 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1165 * skb_pad - zero pad the tail of an skb
1166 * @skb: buffer to pad
1167 * @pad: space to pad
1169 * Ensure that a buffer is followed by a padding area that is zero
1170 * filled. Used by network drivers which may DMA or transfer data
1171 * beyond the buffer end onto the wire.
1173 * May return error in out of memory cases. The skb is freed on error.
1175 static inline int skb_pad(struct sk_buff *skb, int pad)
1177 return __skb_pad(skb, pad, true);
1179 #define dev_kfree_skb(a) consume_skb(a)
1181 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1182 int offset, size_t size);
1184 struct skb_seq_state {
1188 __u32 stepped_offset;
1189 struct sk_buff *root_skb;
1190 struct sk_buff *cur_skb;
1194 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1195 unsigned int to, struct skb_seq_state *st);
1196 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1197 struct skb_seq_state *st);
1198 void skb_abort_seq_read(struct skb_seq_state *st);
1200 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1201 unsigned int to, struct ts_config *config);
1204 * Packet hash types specify the type of hash in skb_set_hash.
1206 * Hash types refer to the protocol layer addresses which are used to
1207 * construct a packet's hash. The hashes are used to differentiate or identify
1208 * flows of the protocol layer for the hash type. Hash types are either
1209 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1211 * Properties of hashes:
1213 * 1) Two packets in different flows have different hash values
1214 * 2) Two packets in the same flow should have the same hash value
1216 * A hash at a higher layer is considered to be more specific. A driver should
1217 * set the most specific hash possible.
1219 * A driver cannot indicate a more specific hash than the layer at which a hash
1220 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1222 * A driver may indicate a hash level which is less specific than the
1223 * actual layer the hash was computed on. For instance, a hash computed
1224 * at L4 may be considered an L3 hash. This should only be done if the
1225 * driver can't unambiguously determine that the HW computed the hash at
1226 * the higher layer. Note that the "should" in the second property above
1229 enum pkt_hash_types {
1230 PKT_HASH_TYPE_NONE, /* Undefined type */
1231 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1232 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1233 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1236 static inline void skb_clear_hash(struct sk_buff *skb)
1243 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1246 skb_clear_hash(skb);
1250 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1252 skb->l4_hash = is_l4;
1253 skb->sw_hash = is_sw;
1258 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1260 /* Used by drivers to set hash from HW */
1261 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1265 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1267 __skb_set_hash(skb, hash, true, is_l4);
1270 void __skb_get_hash(struct sk_buff *skb);
1271 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1272 u32 skb_get_poff(const struct sk_buff *skb);
1273 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1274 const struct flow_keys_basic *keys, int hlen);
1275 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1276 void *data, int hlen_proto);
1278 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1279 int thoff, u8 ip_proto)
1281 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1284 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1285 const struct flow_dissector_key *key,
1286 unsigned int key_count);
1288 struct bpf_flow_dissector;
1289 bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1290 __be16 proto, int nhoff, int hlen, unsigned int flags);
1292 bool __skb_flow_dissect(const struct net *net,
1293 const struct sk_buff *skb,
1294 struct flow_dissector *flow_dissector,
1295 void *target_container,
1296 void *data, __be16 proto, int nhoff, int hlen,
1297 unsigned int flags);
1299 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1300 struct flow_dissector *flow_dissector,
1301 void *target_container, unsigned int flags)
1303 return __skb_flow_dissect(NULL, skb, flow_dissector,
1304 target_container, NULL, 0, 0, 0, flags);
1307 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1308 struct flow_keys *flow,
1311 memset(flow, 0, sizeof(*flow));
1312 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1313 flow, NULL, 0, 0, 0, flags);
1317 skb_flow_dissect_flow_keys_basic(const struct net *net,
1318 const struct sk_buff *skb,
1319 struct flow_keys_basic *flow, void *data,
1320 __be16 proto, int nhoff, int hlen,
1323 memset(flow, 0, sizeof(*flow));
1324 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1325 data, proto, nhoff, hlen, flags);
1328 void skb_flow_dissect_meta(const struct sk_buff *skb,
1329 struct flow_dissector *flow_dissector,
1330 void *target_container);
1332 /* Gets a skb connection tracking info, ctinfo map should be a
1333 * map of mapsize to translate enum ip_conntrack_info states
1337 skb_flow_dissect_ct(const struct sk_buff *skb,
1338 struct flow_dissector *flow_dissector,
1339 void *target_container,
1343 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1344 struct flow_dissector *flow_dissector,
1345 void *target_container);
1347 void skb_flow_dissect_hash(const struct sk_buff *skb,
1348 struct flow_dissector *flow_dissector,
1349 void *target_container);
1351 static inline __u32 skb_get_hash(struct sk_buff *skb)
1353 if (!skb->l4_hash && !skb->sw_hash)
1354 __skb_get_hash(skb);
1359 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1361 if (!skb->l4_hash && !skb->sw_hash) {
1362 struct flow_keys keys;
1363 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1365 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1371 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1372 const siphash_key_t *perturb);
1374 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1379 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1381 to->hash = from->hash;
1382 to->sw_hash = from->sw_hash;
1383 to->l4_hash = from->l4_hash;
1386 static inline void skb_copy_decrypted(struct sk_buff *to,
1387 const struct sk_buff *from)
1389 #ifdef CONFIG_TLS_DEVICE
1390 to->decrypted = from->decrypted;
1394 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1395 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1397 return skb->head + skb->end;
1400 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1405 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1410 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1412 return skb->end - skb->head;
1417 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1419 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1421 return &skb_shinfo(skb)->hwtstamps;
1424 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1426 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1428 return is_zcopy ? skb_uarg(skb) : NULL;
1431 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1434 if (skb && uarg && !skb_zcopy(skb)) {
1435 if (unlikely(have_ref && *have_ref))
1438 sock_zerocopy_get(uarg);
1439 skb_shinfo(skb)->destructor_arg = uarg;
1440 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1444 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1446 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1447 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1450 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1452 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1455 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1457 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1460 /* Release a reference on a zerocopy structure */
1461 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1463 struct ubuf_info *uarg = skb_zcopy(skb);
1466 if (skb_zcopy_is_nouarg(skb)) {
1467 /* no notification callback */
1468 } else if (uarg->callback == sock_zerocopy_callback) {
1469 uarg->zerocopy = uarg->zerocopy && zerocopy;
1470 sock_zerocopy_put(uarg);
1472 uarg->callback(uarg, zerocopy);
1475 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1479 /* Abort a zerocopy operation and revert zckey on error in send syscall */
1480 static inline void skb_zcopy_abort(struct sk_buff *skb)
1482 struct ubuf_info *uarg = skb_zcopy(skb);
1485 sock_zerocopy_put_abort(uarg, false);
1486 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1490 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1495 /* Iterate through singly-linked GSO fragments of an skb. */
1496 #define skb_list_walk_safe(first, skb, next_skb) \
1497 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1498 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1500 static inline void skb_list_del_init(struct sk_buff *skb)
1502 __list_del_entry(&skb->list);
1503 skb_mark_not_on_list(skb);
1507 * skb_queue_empty - check if a queue is empty
1510 * Returns true if the queue is empty, false otherwise.
1512 static inline int skb_queue_empty(const struct sk_buff_head *list)
1514 return list->next == (const struct sk_buff *) list;
1518 * skb_queue_empty_lockless - check if a queue is empty
1521 * Returns true if the queue is empty, false otherwise.
1522 * This variant can be used in lockless contexts.
1524 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1526 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1531 * skb_queue_is_last - check if skb is the last entry in the queue
1535 * Returns true if @skb is the last buffer on the list.
1537 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1538 const struct sk_buff *skb)
1540 return skb->next == (const struct sk_buff *) list;
1544 * skb_queue_is_first - check if skb is the first entry in the queue
1548 * Returns true if @skb is the first buffer on the list.
1550 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1551 const struct sk_buff *skb)
1553 return skb->prev == (const struct sk_buff *) list;
1557 * skb_queue_next - return the next packet in the queue
1559 * @skb: current buffer
1561 * Return the next packet in @list after @skb. It is only valid to
1562 * call this if skb_queue_is_last() evaluates to false.
1564 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1565 const struct sk_buff *skb)
1567 /* This BUG_ON may seem severe, but if we just return then we
1568 * are going to dereference garbage.
1570 BUG_ON(skb_queue_is_last(list, skb));
1575 * skb_queue_prev - return the prev packet in the queue
1577 * @skb: current buffer
1579 * Return the prev packet in @list before @skb. It is only valid to
1580 * call this if skb_queue_is_first() evaluates to false.
1582 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1583 const struct sk_buff *skb)
1585 /* This BUG_ON may seem severe, but if we just return then we
1586 * are going to dereference garbage.
1588 BUG_ON(skb_queue_is_first(list, skb));
1593 * skb_get - reference buffer
1594 * @skb: buffer to reference
1596 * Makes another reference to a socket buffer and returns a pointer
1599 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1601 refcount_inc(&skb->users);
1606 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1610 * skb_cloned - is the buffer a clone
1611 * @skb: buffer to check
1613 * Returns true if the buffer was generated with skb_clone() and is
1614 * one of multiple shared copies of the buffer. Cloned buffers are
1615 * shared data so must not be written to under normal circumstances.
1617 static inline int skb_cloned(const struct sk_buff *skb)
1619 return skb->cloned &&
1620 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1623 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1625 might_sleep_if(gfpflags_allow_blocking(pri));
1627 if (skb_cloned(skb))
1628 return pskb_expand_head(skb, 0, 0, pri);
1634 * skb_header_cloned - is the header a clone
1635 * @skb: buffer to check
1637 * Returns true if modifying the header part of the buffer requires
1638 * the data to be copied.
1640 static inline int skb_header_cloned(const struct sk_buff *skb)
1647 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1648 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1649 return dataref != 1;
1652 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1654 might_sleep_if(gfpflags_allow_blocking(pri));
1656 if (skb_header_cloned(skb))
1657 return pskb_expand_head(skb, 0, 0, pri);
1663 * __skb_header_release - release reference to header
1664 * @skb: buffer to operate on
1666 static inline void __skb_header_release(struct sk_buff *skb)
1669 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1674 * skb_shared - is the buffer shared
1675 * @skb: buffer to check
1677 * Returns true if more than one person has a reference to this
1680 static inline int skb_shared(const struct sk_buff *skb)
1682 return refcount_read(&skb->users) != 1;
1686 * skb_share_check - check if buffer is shared and if so clone it
1687 * @skb: buffer to check
1688 * @pri: priority for memory allocation
1690 * If the buffer is shared the buffer is cloned and the old copy
1691 * drops a reference. A new clone with a single reference is returned.
1692 * If the buffer is not shared the original buffer is returned. When
1693 * being called from interrupt status or with spinlocks held pri must
1696 * NULL is returned on a memory allocation failure.
1698 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1700 might_sleep_if(gfpflags_allow_blocking(pri));
1701 if (skb_shared(skb)) {
1702 struct sk_buff *nskb = skb_clone(skb, pri);
1714 * Copy shared buffers into a new sk_buff. We effectively do COW on
1715 * packets to handle cases where we have a local reader and forward
1716 * and a couple of other messy ones. The normal one is tcpdumping
1717 * a packet thats being forwarded.
1721 * skb_unshare - make a copy of a shared buffer
1722 * @skb: buffer to check
1723 * @pri: priority for memory allocation
1725 * If the socket buffer is a clone then this function creates a new
1726 * copy of the data, drops a reference count on the old copy and returns
1727 * the new copy with the reference count at 1. If the buffer is not a clone
1728 * the original buffer is returned. When called with a spinlock held or
1729 * from interrupt state @pri must be %GFP_ATOMIC
1731 * %NULL is returned on a memory allocation failure.
1733 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1736 might_sleep_if(gfpflags_allow_blocking(pri));
1737 if (skb_cloned(skb)) {
1738 struct sk_buff *nskb = skb_copy(skb, pri);
1740 /* Free our shared copy */
1751 * skb_peek - peek at the head of an &sk_buff_head
1752 * @list_: list to peek at
1754 * Peek an &sk_buff. Unlike most other operations you _MUST_
1755 * be careful with this one. A peek leaves the buffer on the
1756 * list and someone else may run off with it. You must hold
1757 * the appropriate locks or have a private queue to do this.
1759 * Returns %NULL for an empty list or a pointer to the head element.
1760 * The reference count is not incremented and the reference is therefore
1761 * volatile. Use with caution.
1763 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1765 struct sk_buff *skb = list_->next;
1767 if (skb == (struct sk_buff *)list_)
1773 * __skb_peek - peek at the head of a non-empty &sk_buff_head
1774 * @list_: list to peek at
1776 * Like skb_peek(), but the caller knows that the list is not empty.
1778 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
1784 * skb_peek_next - peek skb following the given one from a queue
1785 * @skb: skb to start from
1786 * @list_: list to peek at
1788 * Returns %NULL when the end of the list is met or a pointer to the
1789 * next element. The reference count is not incremented and the
1790 * reference is therefore volatile. Use with caution.
1792 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1793 const struct sk_buff_head *list_)
1795 struct sk_buff *next = skb->next;
1797 if (next == (struct sk_buff *)list_)
1803 * skb_peek_tail - peek at the tail of an &sk_buff_head
1804 * @list_: list to peek at
1806 * Peek an &sk_buff. Unlike most other operations you _MUST_
1807 * be careful with this one. A peek leaves the buffer on the
1808 * list and someone else may run off with it. You must hold
1809 * the appropriate locks or have a private queue to do this.
1811 * Returns %NULL for an empty list or a pointer to the tail element.
1812 * The reference count is not incremented and the reference is therefore
1813 * volatile. Use with caution.
1815 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1817 struct sk_buff *skb = READ_ONCE(list_->prev);
1819 if (skb == (struct sk_buff *)list_)
1826 * skb_queue_len - get queue length
1827 * @list_: list to measure
1829 * Return the length of an &sk_buff queue.
1831 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1837 * skb_queue_len_lockless - get queue length
1838 * @list_: list to measure
1840 * Return the length of an &sk_buff queue.
1841 * This variant can be used in lockless contexts.
1843 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
1845 return READ_ONCE(list_->qlen);
1849 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1850 * @list: queue to initialize
1852 * This initializes only the list and queue length aspects of
1853 * an sk_buff_head object. This allows to initialize the list
1854 * aspects of an sk_buff_head without reinitializing things like
1855 * the spinlock. It can also be used for on-stack sk_buff_head
1856 * objects where the spinlock is known to not be used.
1858 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1860 list->prev = list->next = (struct sk_buff *)list;
1865 * This function creates a split out lock class for each invocation;
1866 * this is needed for now since a whole lot of users of the skb-queue
1867 * infrastructure in drivers have different locking usage (in hardirq)
1868 * than the networking core (in softirq only). In the long run either the
1869 * network layer or drivers should need annotation to consolidate the
1870 * main types of usage into 3 classes.
1872 static inline void skb_queue_head_init(struct sk_buff_head *list)
1874 spin_lock_init(&list->lock);
1875 __skb_queue_head_init(list);
1878 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1879 struct lock_class_key *class)
1881 skb_queue_head_init(list);
1882 lockdep_set_class(&list->lock, class);
1886 * Insert an sk_buff on a list.
1888 * The "__skb_xxxx()" functions are the non-atomic ones that
1889 * can only be called with interrupts disabled.
1891 static inline void __skb_insert(struct sk_buff *newsk,
1892 struct sk_buff *prev, struct sk_buff *next,
1893 struct sk_buff_head *list)
1895 /* See skb_queue_empty_lockless() and skb_peek_tail()
1896 * for the opposite READ_ONCE()
1898 WRITE_ONCE(newsk->next, next);
1899 WRITE_ONCE(newsk->prev, prev);
1900 WRITE_ONCE(next->prev, newsk);
1901 WRITE_ONCE(prev->next, newsk);
1905 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1906 struct sk_buff *prev,
1907 struct sk_buff *next)
1909 struct sk_buff *first = list->next;
1910 struct sk_buff *last = list->prev;
1912 WRITE_ONCE(first->prev, prev);
1913 WRITE_ONCE(prev->next, first);
1915 WRITE_ONCE(last->next, next);
1916 WRITE_ONCE(next->prev, last);
1920 * skb_queue_splice - join two skb lists, this is designed for stacks
1921 * @list: the new list to add
1922 * @head: the place to add it in the first list
1924 static inline void skb_queue_splice(const struct sk_buff_head *list,
1925 struct sk_buff_head *head)
1927 if (!skb_queue_empty(list)) {
1928 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1929 head->qlen += list->qlen;
1934 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1935 * @list: the new list to add
1936 * @head: the place to add it in the first list
1938 * The list at @list is reinitialised
1940 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1941 struct sk_buff_head *head)
1943 if (!skb_queue_empty(list)) {
1944 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1945 head->qlen += list->qlen;
1946 __skb_queue_head_init(list);
1951 * skb_queue_splice_tail - join two skb lists, each list being a queue
1952 * @list: the new list to add
1953 * @head: the place to add it in the first list
1955 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1956 struct sk_buff_head *head)
1958 if (!skb_queue_empty(list)) {
1959 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1960 head->qlen += list->qlen;
1965 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1966 * @list: the new list to add
1967 * @head: the place to add it in the first list
1969 * Each of the lists is a queue.
1970 * The list at @list is reinitialised
1972 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1973 struct sk_buff_head *head)
1975 if (!skb_queue_empty(list)) {
1976 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1977 head->qlen += list->qlen;
1978 __skb_queue_head_init(list);
1983 * __skb_queue_after - queue a buffer at the list head
1984 * @list: list to use
1985 * @prev: place after this buffer
1986 * @newsk: buffer to queue
1988 * Queue a buffer int the middle of a list. This function takes no locks
1989 * and you must therefore hold required locks before calling it.
1991 * A buffer cannot be placed on two lists at the same time.
1993 static inline void __skb_queue_after(struct sk_buff_head *list,
1994 struct sk_buff *prev,
1995 struct sk_buff *newsk)
1997 __skb_insert(newsk, prev, prev->next, list);
2000 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2001 struct sk_buff_head *list);
2003 static inline void __skb_queue_before(struct sk_buff_head *list,
2004 struct sk_buff *next,
2005 struct sk_buff *newsk)
2007 __skb_insert(newsk, next->prev, next, list);
2011 * __skb_queue_head - queue a buffer at the list head
2012 * @list: list to use
2013 * @newsk: buffer to queue
2015 * Queue a buffer at the start of a list. This function takes no locks
2016 * and you must therefore hold required locks before calling it.
2018 * A buffer cannot be placed on two lists at the same time.
2020 static inline void __skb_queue_head(struct sk_buff_head *list,
2021 struct sk_buff *newsk)
2023 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2025 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2028 * __skb_queue_tail - queue a buffer at the list tail
2029 * @list: list to use
2030 * @newsk: buffer to queue
2032 * Queue a buffer at the end of a list. This function takes no locks
2033 * and you must therefore hold required locks before calling it.
2035 * A buffer cannot be placed on two lists at the same time.
2037 static inline void __skb_queue_tail(struct sk_buff_head *list,
2038 struct sk_buff *newsk)
2040 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2042 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2045 * remove sk_buff from list. _Must_ be called atomically, and with
2048 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2049 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2051 struct sk_buff *next, *prev;
2053 WRITE_ONCE(list->qlen, list->qlen - 1);
2056 skb->next = skb->prev = NULL;
2057 WRITE_ONCE(next->prev, prev);
2058 WRITE_ONCE(prev->next, next);
2062 * __skb_dequeue - remove from the head of the queue
2063 * @list: list to dequeue from
2065 * Remove the head of the list. This function does not take any locks
2066 * so must be used with appropriate locks held only. The head item is
2067 * returned or %NULL if the list is empty.
2069 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2071 struct sk_buff *skb = skb_peek(list);
2073 __skb_unlink(skb, list);
2076 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2079 * __skb_dequeue_tail - remove from the tail of the queue
2080 * @list: list to dequeue from
2082 * Remove the tail of the list. This function does not take any locks
2083 * so must be used with appropriate locks held only. The tail item is
2084 * returned or %NULL if the list is empty.
2086 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2088 struct sk_buff *skb = skb_peek_tail(list);
2090 __skb_unlink(skb, list);
2093 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2096 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2098 return skb->data_len;
2101 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2103 return skb->len - skb->data_len;
2106 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2108 unsigned int i, len = 0;
2110 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2111 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2115 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2117 return skb_headlen(skb) + __skb_pagelen(skb);
2121 * __skb_fill_page_desc - initialise a paged fragment in an skb
2122 * @skb: buffer containing fragment to be initialised
2123 * @i: paged fragment index to initialise
2124 * @page: the page to use for this fragment
2125 * @off: the offset to the data with @page
2126 * @size: the length of the data
2128 * Initialises the @i'th fragment of @skb to point to &size bytes at
2129 * offset @off within @page.
2131 * Does not take any additional reference on the fragment.
2133 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2134 struct page *page, int off, int size)
2136 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2139 * Propagate page pfmemalloc to the skb if we can. The problem is
2140 * that not all callers have unique ownership of the page but rely
2141 * on page_is_pfmemalloc doing the right thing(tm).
2143 frag->bv_page = page;
2144 frag->bv_offset = off;
2145 skb_frag_size_set(frag, size);
2147 page = compound_head(page);
2148 if (page_is_pfmemalloc(page))
2149 skb->pfmemalloc = true;
2153 * skb_fill_page_desc - initialise a paged fragment in an skb
2154 * @skb: buffer containing fragment to be initialised
2155 * @i: paged fragment index to initialise
2156 * @page: the page to use for this fragment
2157 * @off: the offset to the data with @page
2158 * @size: the length of the data
2160 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2161 * @skb to point to @size bytes at offset @off within @page. In
2162 * addition updates @skb such that @i is the last fragment.
2164 * Does not take any additional reference on the fragment.
2166 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2167 struct page *page, int off, int size)
2169 __skb_fill_page_desc(skb, i, page, off, size);
2170 skb_shinfo(skb)->nr_frags = i + 1;
2173 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2174 int size, unsigned int truesize);
2176 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2177 unsigned int truesize);
2179 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2181 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2182 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2184 return skb->head + skb->tail;
2187 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2189 skb->tail = skb->data - skb->head;
2192 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2194 skb_reset_tail_pointer(skb);
2195 skb->tail += offset;
2198 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2199 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2204 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2206 skb->tail = skb->data;
2209 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2211 skb->tail = skb->data + offset;
2214 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2217 * Add data to an sk_buff
2219 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2220 void *skb_put(struct sk_buff *skb, unsigned int len);
2221 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2223 void *tmp = skb_tail_pointer(skb);
2224 SKB_LINEAR_ASSERT(skb);
2230 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2232 void *tmp = __skb_put(skb, len);
2234 memset(tmp, 0, len);
2238 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2241 void *tmp = __skb_put(skb, len);
2243 memcpy(tmp, data, len);
2247 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2249 *(u8 *)__skb_put(skb, 1) = val;
2252 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2254 void *tmp = skb_put(skb, len);
2256 memset(tmp, 0, len);
2261 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2264 void *tmp = skb_put(skb, len);
2266 memcpy(tmp, data, len);
2271 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2273 *(u8 *)skb_put(skb, 1) = val;
2276 void *skb_push(struct sk_buff *skb, unsigned int len);
2277 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2284 void *skb_pull(struct sk_buff *skb, unsigned int len);
2285 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2288 BUG_ON(skb->len < skb->data_len);
2289 return skb->data += len;
2292 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2294 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2297 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2299 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2301 if (len > skb_headlen(skb) &&
2302 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2305 return skb->data += len;
2308 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2310 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2313 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2315 if (likely(len <= skb_headlen(skb)))
2317 if (unlikely(len > skb->len))
2319 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2322 void skb_condense(struct sk_buff *skb);
2325 * skb_headroom - bytes at buffer head
2326 * @skb: buffer to check
2328 * Return the number of bytes of free space at the head of an &sk_buff.
2330 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2332 return skb->data - skb->head;
2336 * skb_tailroom - bytes at buffer end
2337 * @skb: buffer to check
2339 * Return the number of bytes of free space at the tail of an sk_buff
2341 static inline int skb_tailroom(const struct sk_buff *skb)
2343 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2347 * skb_availroom - bytes at buffer end
2348 * @skb: buffer to check
2350 * Return the number of bytes of free space at the tail of an sk_buff
2351 * allocated by sk_stream_alloc()
2353 static inline int skb_availroom(const struct sk_buff *skb)
2355 if (skb_is_nonlinear(skb))
2358 return skb->end - skb->tail - skb->reserved_tailroom;
2362 * skb_reserve - adjust headroom
2363 * @skb: buffer to alter
2364 * @len: bytes to move
2366 * Increase the headroom of an empty &sk_buff by reducing the tail
2367 * room. This is only allowed for an empty buffer.
2369 static inline void skb_reserve(struct sk_buff *skb, int len)
2376 * skb_tailroom_reserve - adjust reserved_tailroom
2377 * @skb: buffer to alter
2378 * @mtu: maximum amount of headlen permitted
2379 * @needed_tailroom: minimum amount of reserved_tailroom
2381 * Set reserved_tailroom so that headlen can be as large as possible but
2382 * not larger than mtu and tailroom cannot be smaller than
2384 * The required headroom should already have been reserved before using
2387 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2388 unsigned int needed_tailroom)
2390 SKB_LINEAR_ASSERT(skb);
2391 if (mtu < skb_tailroom(skb) - needed_tailroom)
2392 /* use at most mtu */
2393 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2395 /* use up to all available space */
2396 skb->reserved_tailroom = needed_tailroom;
2399 #define ENCAP_TYPE_ETHER 0
2400 #define ENCAP_TYPE_IPPROTO 1
2402 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2405 skb->inner_protocol = protocol;
2406 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2409 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2412 skb->inner_ipproto = ipproto;
2413 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2416 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2418 skb->inner_mac_header = skb->mac_header;
2419 skb->inner_network_header = skb->network_header;
2420 skb->inner_transport_header = skb->transport_header;
2423 static inline void skb_reset_mac_len(struct sk_buff *skb)
2425 skb->mac_len = skb->network_header - skb->mac_header;
2428 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2431 return skb->head + skb->inner_transport_header;
2434 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2436 return skb_inner_transport_header(skb) - skb->data;
2439 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2441 skb->inner_transport_header = skb->data - skb->head;
2444 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2447 skb_reset_inner_transport_header(skb);
2448 skb->inner_transport_header += offset;
2451 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2453 return skb->head + skb->inner_network_header;
2456 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2458 skb->inner_network_header = skb->data - skb->head;
2461 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2464 skb_reset_inner_network_header(skb);
2465 skb->inner_network_header += offset;
2468 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2470 return skb->head + skb->inner_mac_header;
2473 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2475 skb->inner_mac_header = skb->data - skb->head;
2478 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2481 skb_reset_inner_mac_header(skb);
2482 skb->inner_mac_header += offset;
2484 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2486 return skb->transport_header != (typeof(skb->transport_header))~0U;
2489 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2491 return skb->head + skb->transport_header;
2494 static inline void skb_reset_transport_header(struct sk_buff *skb)
2496 skb->transport_header = skb->data - skb->head;
2499 static inline void skb_set_transport_header(struct sk_buff *skb,
2502 skb_reset_transport_header(skb);
2503 skb->transport_header += offset;
2506 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2508 return skb->head + skb->network_header;
2511 static inline void skb_reset_network_header(struct sk_buff *skb)
2513 skb->network_header = skb->data - skb->head;
2516 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2518 skb_reset_network_header(skb);
2519 skb->network_header += offset;
2522 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2524 return skb->head + skb->mac_header;
2527 static inline int skb_mac_offset(const struct sk_buff *skb)
2529 return skb_mac_header(skb) - skb->data;
2532 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2534 return skb->network_header - skb->mac_header;
2537 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2539 return skb->mac_header != (typeof(skb->mac_header))~0U;
2542 static inline void skb_reset_mac_header(struct sk_buff *skb)
2544 skb->mac_header = skb->data - skb->head;
2547 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2549 skb_reset_mac_header(skb);
2550 skb->mac_header += offset;
2553 static inline void skb_pop_mac_header(struct sk_buff *skb)
2555 skb->mac_header = skb->network_header;
2558 static inline void skb_probe_transport_header(struct sk_buff *skb)
2560 struct flow_keys_basic keys;
2562 if (skb_transport_header_was_set(skb))
2565 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2567 skb_set_transport_header(skb, keys.control.thoff);
2570 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2572 if (skb_mac_header_was_set(skb)) {
2573 const unsigned char *old_mac = skb_mac_header(skb);
2575 skb_set_mac_header(skb, -skb->mac_len);
2576 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2580 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2582 return skb->csum_start - skb_headroom(skb);
2585 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2587 return skb->head + skb->csum_start;
2590 static inline int skb_transport_offset(const struct sk_buff *skb)
2592 return skb_transport_header(skb) - skb->data;
2595 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2597 return skb->transport_header - skb->network_header;
2600 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2602 return skb->inner_transport_header - skb->inner_network_header;
2605 static inline int skb_network_offset(const struct sk_buff *skb)
2607 return skb_network_header(skb) - skb->data;
2610 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2612 return skb_inner_network_header(skb) - skb->data;
2615 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2617 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2621 * CPUs often take a performance hit when accessing unaligned memory
2622 * locations. The actual performance hit varies, it can be small if the
2623 * hardware handles it or large if we have to take an exception and fix it
2626 * Since an ethernet header is 14 bytes network drivers often end up with
2627 * the IP header at an unaligned offset. The IP header can be aligned by
2628 * shifting the start of the packet by 2 bytes. Drivers should do this
2631 * skb_reserve(skb, NET_IP_ALIGN);
2633 * The downside to this alignment of the IP header is that the DMA is now
2634 * unaligned. On some architectures the cost of an unaligned DMA is high
2635 * and this cost outweighs the gains made by aligning the IP header.
2637 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2640 #ifndef NET_IP_ALIGN
2641 #define NET_IP_ALIGN 2
2645 * The networking layer reserves some headroom in skb data (via
2646 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2647 * the header has to grow. In the default case, if the header has to grow
2648 * 32 bytes or less we avoid the reallocation.
2650 * Unfortunately this headroom changes the DMA alignment of the resulting
2651 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2652 * on some architectures. An architecture can override this value,
2653 * perhaps setting it to a cacheline in size (since that will maintain
2654 * cacheline alignment of the DMA). It must be a power of 2.
2656 * Various parts of the networking layer expect at least 32 bytes of
2657 * headroom, you should not reduce this.
2659 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2660 * to reduce average number of cache lines per packet.
2661 * get_rps_cpus() for example only access one 64 bytes aligned block :
2662 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2665 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2668 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2670 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2672 if (WARN_ON(skb_is_nonlinear(skb)))
2675 skb_set_tail_pointer(skb, len);
2678 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2680 __skb_set_length(skb, len);
2683 void skb_trim(struct sk_buff *skb, unsigned int len);
2685 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2688 return ___pskb_trim(skb, len);
2689 __skb_trim(skb, len);
2693 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2695 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2699 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2700 * @skb: buffer to alter
2703 * This is identical to pskb_trim except that the caller knows that
2704 * the skb is not cloned so we should never get an error due to out-
2707 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2709 int err = pskb_trim(skb, len);
2713 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2715 unsigned int diff = len - skb->len;
2717 if (skb_tailroom(skb) < diff) {
2718 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2723 __skb_set_length(skb, len);
2728 * skb_orphan - orphan a buffer
2729 * @skb: buffer to orphan
2731 * If a buffer currently has an owner then we call the owner's
2732 * destructor function and make the @skb unowned. The buffer continues
2733 * to exist but is no longer charged to its former owner.
2735 static inline void skb_orphan(struct sk_buff *skb)
2737 if (skb->destructor) {
2738 skb->destructor(skb);
2739 skb->destructor = NULL;
2747 * skb_orphan_frags - orphan the frags contained in a buffer
2748 * @skb: buffer to orphan frags from
2749 * @gfp_mask: allocation mask for replacement pages
2751 * For each frag in the SKB which needs a destructor (i.e. has an
2752 * owner) create a copy of that frag and release the original
2753 * page by calling the destructor.
2755 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2757 if (likely(!skb_zcopy(skb)))
2759 if (!skb_zcopy_is_nouarg(skb) &&
2760 skb_uarg(skb)->callback == sock_zerocopy_callback)
2762 return skb_copy_ubufs(skb, gfp_mask);
2765 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2766 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2768 if (likely(!skb_zcopy(skb)))
2770 return skb_copy_ubufs(skb, gfp_mask);
2774 * __skb_queue_purge - empty a list
2775 * @list: list to empty
2777 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2778 * the list and one reference dropped. This function does not take the
2779 * list lock and the caller must hold the relevant locks to use it.
2781 static inline void __skb_queue_purge(struct sk_buff_head *list)
2783 struct sk_buff *skb;
2784 while ((skb = __skb_dequeue(list)) != NULL)
2787 void skb_queue_purge(struct sk_buff_head *list);
2789 unsigned int skb_rbtree_purge(struct rb_root *root);
2791 void *netdev_alloc_frag(unsigned int fragsz);
2793 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2797 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2798 * @dev: network device to receive on
2799 * @length: length to allocate
2801 * Allocate a new &sk_buff and assign it a usage count of one. The
2802 * buffer has unspecified headroom built in. Users should allocate
2803 * the headroom they think they need without accounting for the
2804 * built in space. The built in space is used for optimisations.
2806 * %NULL is returned if there is no free memory. Although this function
2807 * allocates memory it can be called from an interrupt.
2809 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2810 unsigned int length)
2812 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2815 /* legacy helper around __netdev_alloc_skb() */
2816 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2819 return __netdev_alloc_skb(NULL, length, gfp_mask);
2822 /* legacy helper around netdev_alloc_skb() */
2823 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2825 return netdev_alloc_skb(NULL, length);
2829 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2830 unsigned int length, gfp_t gfp)
2832 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2834 if (NET_IP_ALIGN && skb)
2835 skb_reserve(skb, NET_IP_ALIGN);
2839 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2840 unsigned int length)
2842 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2845 static inline void skb_free_frag(void *addr)
2847 page_frag_free(addr);
2850 void *napi_alloc_frag(unsigned int fragsz);
2851 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2852 unsigned int length, gfp_t gfp_mask);
2853 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2854 unsigned int length)
2856 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2858 void napi_consume_skb(struct sk_buff *skb, int budget);
2860 void __kfree_skb_flush(void);
2861 void __kfree_skb_defer(struct sk_buff *skb);
2864 * __dev_alloc_pages - allocate page for network Rx
2865 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2866 * @order: size of the allocation
2868 * Allocate a new page.
2870 * %NULL is returned if there is no free memory.
2872 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2875 /* This piece of code contains several assumptions.
2876 * 1. This is for device Rx, therefor a cold page is preferred.
2877 * 2. The expectation is the user wants a compound page.
2878 * 3. If requesting a order 0 page it will not be compound
2879 * due to the check to see if order has a value in prep_new_page
2880 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2881 * code in gfp_to_alloc_flags that should be enforcing this.
2883 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2885 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2888 static inline struct page *dev_alloc_pages(unsigned int order)
2890 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2894 * __dev_alloc_page - allocate a page for network Rx
2895 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2897 * Allocate a new page.
2899 * %NULL is returned if there is no free memory.
2901 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2903 return __dev_alloc_pages(gfp_mask, 0);
2906 static inline struct page *dev_alloc_page(void)
2908 return dev_alloc_pages(0);
2912 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2913 * @page: The page that was allocated from skb_alloc_page
2914 * @skb: The skb that may need pfmemalloc set
2916 static inline void skb_propagate_pfmemalloc(struct page *page,
2917 struct sk_buff *skb)
2919 if (page_is_pfmemalloc(page))
2920 skb->pfmemalloc = true;
2924 * skb_frag_off() - Returns the offset of a skb fragment
2925 * @frag: the paged fragment
2927 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
2929 return frag->bv_offset;
2933 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
2934 * @frag: skb fragment
2935 * @delta: value to add
2937 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
2939 frag->bv_offset += delta;
2943 * skb_frag_off_set() - Sets the offset of a skb fragment
2944 * @frag: skb fragment
2945 * @offset: offset of fragment
2947 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
2949 frag->bv_offset = offset;
2953 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
2954 * @fragto: skb fragment where offset is set
2955 * @fragfrom: skb fragment offset is copied from
2957 static inline void skb_frag_off_copy(skb_frag_t *fragto,
2958 const skb_frag_t *fragfrom)
2960 fragto->bv_offset = fragfrom->bv_offset;
2964 * skb_frag_page - retrieve the page referred to by a paged fragment
2965 * @frag: the paged fragment
2967 * Returns the &struct page associated with @frag.
2969 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2971 return frag->bv_page;
2975 * __skb_frag_ref - take an addition reference on a paged fragment.
2976 * @frag: the paged fragment
2978 * Takes an additional reference on the paged fragment @frag.
2980 static inline void __skb_frag_ref(skb_frag_t *frag)
2982 get_page(skb_frag_page(frag));
2986 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2988 * @f: the fragment offset.
2990 * Takes an additional reference on the @f'th paged fragment of @skb.
2992 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2994 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2998 * __skb_frag_unref - release a reference on a paged fragment.
2999 * @frag: the paged fragment
3001 * Releases a reference on the paged fragment @frag.
3003 static inline void __skb_frag_unref(skb_frag_t *frag)
3005 put_page(skb_frag_page(frag));
3009 * skb_frag_unref - release a reference on a paged fragment of an skb.
3011 * @f: the fragment offset
3013 * Releases a reference on the @f'th paged fragment of @skb.
3015 static inline void skb_frag_unref(struct sk_buff *skb, int f)
3017 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
3021 * skb_frag_address - gets the address of the data contained in a paged fragment
3022 * @frag: the paged fragment buffer
3024 * Returns the address of the data within @frag. The page must already
3027 static inline void *skb_frag_address(const skb_frag_t *frag)
3029 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3033 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3034 * @frag: the paged fragment buffer
3036 * Returns the address of the data within @frag. Checks that the page
3037 * is mapped and returns %NULL otherwise.
3039 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3041 void *ptr = page_address(skb_frag_page(frag));
3045 return ptr + skb_frag_off(frag);
3049 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3050 * @fragto: skb fragment where page is set
3051 * @fragfrom: skb fragment page is copied from
3053 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3054 const skb_frag_t *fragfrom)
3056 fragto->bv_page = fragfrom->bv_page;
3060 * __skb_frag_set_page - sets the page contained in a paged fragment
3061 * @frag: the paged fragment
3062 * @page: the page to set
3064 * Sets the fragment @frag to contain @page.
3066 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3068 frag->bv_page = page;
3072 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3074 * @f: the fragment offset
3075 * @page: the page to set
3077 * Sets the @f'th fragment of @skb to contain @page.
3079 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3082 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3085 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3088 * skb_frag_dma_map - maps a paged fragment via the DMA API
3089 * @dev: the device to map the fragment to
3090 * @frag: the paged fragment to map
3091 * @offset: the offset within the fragment (starting at the
3092 * fragment's own offset)
3093 * @size: the number of bytes to map
3094 * @dir: the direction of the mapping (``PCI_DMA_*``)
3096 * Maps the page associated with @frag to @device.
3098 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3099 const skb_frag_t *frag,
3100 size_t offset, size_t size,
3101 enum dma_data_direction dir)
3103 return dma_map_page(dev, skb_frag_page(frag),
3104 skb_frag_off(frag) + offset, size, dir);
3107 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3110 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3114 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3117 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3122 * skb_clone_writable - is the header of a clone writable
3123 * @skb: buffer to check
3124 * @len: length up to which to write
3126 * Returns true if modifying the header part of the cloned buffer
3127 * does not requires the data to be copied.
3129 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3131 return !skb_header_cloned(skb) &&
3132 skb_headroom(skb) + len <= skb->hdr_len;
3135 static inline int skb_try_make_writable(struct sk_buff *skb,
3136 unsigned int write_len)
3138 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3139 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3142 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3147 if (headroom > skb_headroom(skb))
3148 delta = headroom - skb_headroom(skb);
3150 if (delta || cloned)
3151 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3157 * skb_cow - copy header of skb when it is required
3158 * @skb: buffer to cow
3159 * @headroom: needed headroom
3161 * If the skb passed lacks sufficient headroom or its data part
3162 * is shared, data is reallocated. If reallocation fails, an error
3163 * is returned and original skb is not changed.
3165 * The result is skb with writable area skb->head...skb->tail
3166 * and at least @headroom of space at head.
3168 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3170 return __skb_cow(skb, headroom, skb_cloned(skb));
3174 * skb_cow_head - skb_cow but only making the head writable
3175 * @skb: buffer to cow
3176 * @headroom: needed headroom
3178 * This function is identical to skb_cow except that we replace the
3179 * skb_cloned check by skb_header_cloned. It should be used when
3180 * you only need to push on some header and do not need to modify
3183 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3185 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3189 * skb_padto - pad an skbuff up to a minimal size
3190 * @skb: buffer to pad
3191 * @len: minimal length
3193 * Pads up a buffer to ensure the trailing bytes exist and are
3194 * blanked. If the buffer already contains sufficient data it
3195 * is untouched. Otherwise it is extended. Returns zero on
3196 * success. The skb is freed on error.
3198 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3200 unsigned int size = skb->len;
3201 if (likely(size >= len))
3203 return skb_pad(skb, len - size);
3207 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3208 * @skb: buffer to pad
3209 * @len: minimal length
3210 * @free_on_error: free buffer on error
3212 * Pads up a buffer to ensure the trailing bytes exist and are
3213 * blanked. If the buffer already contains sufficient data it
3214 * is untouched. Otherwise it is extended. Returns zero on
3215 * success. The skb is freed on error if @free_on_error is true.
3217 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
3220 unsigned int size = skb->len;
3222 if (unlikely(size < len)) {
3224 if (__skb_pad(skb, len, free_on_error))
3226 __skb_put(skb, len);
3232 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3233 * @skb: buffer to pad
3234 * @len: minimal length
3236 * Pads up a buffer to ensure the trailing bytes exist and are
3237 * blanked. If the buffer already contains sufficient data it
3238 * is untouched. Otherwise it is extended. Returns zero on
3239 * success. The skb is freed on error.
3241 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
3243 return __skb_put_padto(skb, len, true);
3246 static inline int skb_add_data(struct sk_buff *skb,
3247 struct iov_iter *from, int copy)
3249 const int off = skb->len;
3251 if (skb->ip_summed == CHECKSUM_NONE) {
3253 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3255 skb->csum = csum_block_add(skb->csum, csum, off);
3258 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3261 __skb_trim(skb, off);
3265 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3266 const struct page *page, int off)
3271 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3273 return page == skb_frag_page(frag) &&
3274 off == skb_frag_off(frag) + skb_frag_size(frag);
3279 static inline int __skb_linearize(struct sk_buff *skb)
3281 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3285 * skb_linearize - convert paged skb to linear one
3286 * @skb: buffer to linarize
3288 * If there is no free memory -ENOMEM is returned, otherwise zero
3289 * is returned and the old skb data released.
3291 static inline int skb_linearize(struct sk_buff *skb)
3293 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3297 * skb_has_shared_frag - can any frag be overwritten
3298 * @skb: buffer to test
3300 * Return true if the skb has at least one frag that might be modified
3301 * by an external entity (as in vmsplice()/sendfile())
3303 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3305 return skb_is_nonlinear(skb) &&
3306 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3310 * skb_linearize_cow - make sure skb is linear and writable
3311 * @skb: buffer to process
3313 * If there is no free memory -ENOMEM is returned, otherwise zero
3314 * is returned and the old skb data released.
3316 static inline int skb_linearize_cow(struct sk_buff *skb)
3318 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3319 __skb_linearize(skb) : 0;
3322 static __always_inline void
3323 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3326 if (skb->ip_summed == CHECKSUM_COMPLETE)
3327 skb->csum = csum_block_sub(skb->csum,
3328 csum_partial(start, len, 0), off);
3329 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3330 skb_checksum_start_offset(skb) < 0)
3331 skb->ip_summed = CHECKSUM_NONE;
3335 * skb_postpull_rcsum - update checksum for received skb after pull
3336 * @skb: buffer to update
3337 * @start: start of data before pull
3338 * @len: length of data pulled
3340 * After doing a pull on a received packet, you need to call this to
3341 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3342 * CHECKSUM_NONE so that it can be recomputed from scratch.
3344 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3345 const void *start, unsigned int len)
3347 __skb_postpull_rcsum(skb, start, len, 0);
3350 static __always_inline void
3351 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3354 if (skb->ip_summed == CHECKSUM_COMPLETE)
3355 skb->csum = csum_block_add(skb->csum,
3356 csum_partial(start, len, 0), off);
3360 * skb_postpush_rcsum - update checksum for received skb after push
3361 * @skb: buffer to update
3362 * @start: start of data after push
3363 * @len: length of data pushed
3365 * After doing a push on a received packet, you need to call this to
3366 * update the CHECKSUM_COMPLETE checksum.
3368 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3369 const void *start, unsigned int len)
3371 __skb_postpush_rcsum(skb, start, len, 0);
3374 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3377 * skb_push_rcsum - push skb and update receive checksum
3378 * @skb: buffer to update
3379 * @len: length of data pulled
3381 * This function performs an skb_push on the packet and updates
3382 * the CHECKSUM_COMPLETE checksum. It should be used on
3383 * receive path processing instead of skb_push unless you know
3384 * that the checksum difference is zero (e.g., a valid IP header)
3385 * or you are setting ip_summed to CHECKSUM_NONE.
3387 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3390 skb_postpush_rcsum(skb, skb->data, len);
3394 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3396 * pskb_trim_rcsum - trim received skb and update checksum
3397 * @skb: buffer to trim
3400 * This is exactly the same as pskb_trim except that it ensures the
3401 * checksum of received packets are still valid after the operation.
3402 * It can change skb pointers.
3405 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3407 if (likely(len >= skb->len))
3409 return pskb_trim_rcsum_slow(skb, len);
3412 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3414 if (skb->ip_summed == CHECKSUM_COMPLETE)
3415 skb->ip_summed = CHECKSUM_NONE;
3416 __skb_trim(skb, len);
3420 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3422 if (skb->ip_summed == CHECKSUM_COMPLETE)
3423 skb->ip_summed = CHECKSUM_NONE;
3424 return __skb_grow(skb, len);
3427 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3428 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3429 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3430 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3431 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3433 #define skb_queue_walk(queue, skb) \
3434 for (skb = (queue)->next; \
3435 skb != (struct sk_buff *)(queue); \
3438 #define skb_queue_walk_safe(queue, skb, tmp) \
3439 for (skb = (queue)->next, tmp = skb->next; \
3440 skb != (struct sk_buff *)(queue); \
3441 skb = tmp, tmp = skb->next)
3443 #define skb_queue_walk_from(queue, skb) \
3444 for (; skb != (struct sk_buff *)(queue); \
3447 #define skb_rbtree_walk(skb, root) \
3448 for (skb = skb_rb_first(root); skb != NULL; \
3449 skb = skb_rb_next(skb))
3451 #define skb_rbtree_walk_from(skb) \
3452 for (; skb != NULL; \
3453 skb = skb_rb_next(skb))
3455 #define skb_rbtree_walk_from_safe(skb, tmp) \
3456 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3459 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3460 for (tmp = skb->next; \
3461 skb != (struct sk_buff *)(queue); \
3462 skb = tmp, tmp = skb->next)
3464 #define skb_queue_reverse_walk(queue, skb) \
3465 for (skb = (queue)->prev; \
3466 skb != (struct sk_buff *)(queue); \
3469 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3470 for (skb = (queue)->prev, tmp = skb->prev; \
3471 skb != (struct sk_buff *)(queue); \
3472 skb = tmp, tmp = skb->prev)
3474 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3475 for (tmp = skb->prev; \
3476 skb != (struct sk_buff *)(queue); \
3477 skb = tmp, tmp = skb->prev)
3479 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3481 return skb_shinfo(skb)->frag_list != NULL;
3484 static inline void skb_frag_list_init(struct sk_buff *skb)
3486 skb_shinfo(skb)->frag_list = NULL;
3489 #define skb_walk_frags(skb, iter) \
3490 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3493 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3494 int *err, long *timeo_p,
3495 const struct sk_buff *skb);
3496 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3497 struct sk_buff_head *queue,
3500 struct sk_buff **last);
3501 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
3502 struct sk_buff_head *queue,
3503 unsigned int flags, int *off, int *err,
3504 struct sk_buff **last);
3505 struct sk_buff *__skb_recv_datagram(struct sock *sk,
3506 struct sk_buff_head *sk_queue,
3507 unsigned int flags, int *off, int *err);
3508 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3510 __poll_t datagram_poll(struct file *file, struct socket *sock,
3511 struct poll_table_struct *wait);
3512 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3513 struct iov_iter *to, int size);
3514 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3515 struct msghdr *msg, int size)
3517 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3519 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3520 struct msghdr *msg);
3521 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3522 struct iov_iter *to, int len,
3523 struct ahash_request *hash);
3524 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3525 struct iov_iter *from, int len);
3526 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3527 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3528 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3529 static inline void skb_free_datagram_locked(struct sock *sk,
3530 struct sk_buff *skb)
3532 __skb_free_datagram_locked(sk, skb, 0);
3534 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3535 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3536 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3537 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3538 int len, __wsum csum);
3539 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3540 struct pipe_inode_info *pipe, unsigned int len,
3541 unsigned int flags);
3542 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3544 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3545 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3546 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3548 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3549 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3550 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3551 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3552 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3553 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3554 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
3555 unsigned int offset);
3556 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3557 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3558 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3559 int skb_vlan_pop(struct sk_buff *skb);
3560 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3561 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
3562 int mac_len, bool ethernet);
3563 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
3565 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3566 int skb_mpls_dec_ttl(struct sk_buff *skb);
3567 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3570 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3572 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3575 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3577 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3580 struct skb_checksum_ops {
3581 __wsum (*update)(const void *mem, int len, __wsum wsum);
3582 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3585 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3587 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3588 __wsum csum, const struct skb_checksum_ops *ops);
3589 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3592 static inline void * __must_check
3593 __skb_header_pointer(const struct sk_buff *skb, int offset,
3594 int len, void *data, int hlen, void *buffer)
3596 if (hlen - offset >= len)
3597 return data + offset;
3600 skb_copy_bits(skb, offset, buffer, len) < 0)
3606 static inline void * __must_check
3607 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3609 return __skb_header_pointer(skb, offset, len, skb->data,
3610 skb_headlen(skb), buffer);
3614 * skb_needs_linearize - check if we need to linearize a given skb
3615 * depending on the given device features.
3616 * @skb: socket buffer to check
3617 * @features: net device features
3619 * Returns true if either:
3620 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3621 * 2. skb is fragmented and the device does not support SG.
3623 static inline bool skb_needs_linearize(struct sk_buff *skb,
3624 netdev_features_t features)
3626 return skb_is_nonlinear(skb) &&
3627 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3628 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3631 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3633 const unsigned int len)
3635 memcpy(to, skb->data, len);
3638 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3639 const int offset, void *to,
3640 const unsigned int len)
3642 memcpy(to, skb->data + offset, len);
3645 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3647 const unsigned int len)
3649 memcpy(skb->data, from, len);
3652 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3655 const unsigned int len)
3657 memcpy(skb->data + offset, from, len);
3660 void skb_init(void);
3662 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3668 * skb_get_timestamp - get timestamp from a skb
3669 * @skb: skb to get stamp from
3670 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
3672 * Timestamps are stored in the skb as offsets to a base timestamp.
3673 * This function converts the offset back to a struct timeval and stores
3676 static inline void skb_get_timestamp(const struct sk_buff *skb,
3677 struct __kernel_old_timeval *stamp)
3679 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
3682 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
3683 struct __kernel_sock_timeval *stamp)
3685 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3687 stamp->tv_sec = ts.tv_sec;
3688 stamp->tv_usec = ts.tv_nsec / 1000;
3691 static inline void skb_get_timestampns(const struct sk_buff *skb,
3692 struct __kernel_old_timespec *stamp)
3694 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3696 stamp->tv_sec = ts.tv_sec;
3697 stamp->tv_nsec = ts.tv_nsec;
3700 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
3701 struct __kernel_timespec *stamp)
3703 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3705 stamp->tv_sec = ts.tv_sec;
3706 stamp->tv_nsec = ts.tv_nsec;
3709 static inline void __net_timestamp(struct sk_buff *skb)
3711 skb->tstamp = ktime_get_real();
3714 static inline ktime_t net_timedelta(ktime_t t)
3716 return ktime_sub(ktime_get_real(), t);
3719 static inline ktime_t net_invalid_timestamp(void)
3724 static inline u8 skb_metadata_len(const struct sk_buff *skb)
3726 return skb_shinfo(skb)->meta_len;
3729 static inline void *skb_metadata_end(const struct sk_buff *skb)
3731 return skb_mac_header(skb);
3734 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3735 const struct sk_buff *skb_b,
3738 const void *a = skb_metadata_end(skb_a);
3739 const void *b = skb_metadata_end(skb_b);
3740 /* Using more efficient varaiant than plain call to memcmp(). */
3741 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3745 #define __it(x, op) (x -= sizeof(u##op))
3746 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3747 case 32: diffs |= __it_diff(a, b, 64);
3749 case 24: diffs |= __it_diff(a, b, 64);
3751 case 16: diffs |= __it_diff(a, b, 64);
3753 case 8: diffs |= __it_diff(a, b, 64);
3755 case 28: diffs |= __it_diff(a, b, 64);
3757 case 20: diffs |= __it_diff(a, b, 64);
3759 case 12: diffs |= __it_diff(a, b, 64);
3761 case 4: diffs |= __it_diff(a, b, 32);
3766 return memcmp(a - meta_len, b - meta_len, meta_len);
3770 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3771 const struct sk_buff *skb_b)
3773 u8 len_a = skb_metadata_len(skb_a);
3774 u8 len_b = skb_metadata_len(skb_b);
3776 if (!(len_a | len_b))
3779 return len_a != len_b ?
3780 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3783 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3785 skb_shinfo(skb)->meta_len = meta_len;
3788 static inline void skb_metadata_clear(struct sk_buff *skb)
3790 skb_metadata_set(skb, 0);
3793 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3795 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3797 void skb_clone_tx_timestamp(struct sk_buff *skb);
3798 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3800 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3802 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3806 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3811 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3814 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3816 * PHY drivers may accept clones of transmitted packets for
3817 * timestamping via their phy_driver.txtstamp method. These drivers
3818 * must call this function to return the skb back to the stack with a
3821 * @skb: clone of the original outgoing packet
3822 * @hwtstamps: hardware time stamps
3825 void skb_complete_tx_timestamp(struct sk_buff *skb,
3826 struct skb_shared_hwtstamps *hwtstamps);
3828 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3829 struct skb_shared_hwtstamps *hwtstamps,
3830 struct sock *sk, int tstype);
3833 * skb_tstamp_tx - queue clone of skb with send time stamps
3834 * @orig_skb: the original outgoing packet
3835 * @hwtstamps: hardware time stamps, may be NULL if not available
3837 * If the skb has a socket associated, then this function clones the
3838 * skb (thus sharing the actual data and optional structures), stores
3839 * the optional hardware time stamping information (if non NULL) or
3840 * generates a software time stamp (otherwise), then queues the clone
3841 * to the error queue of the socket. Errors are silently ignored.
3843 void skb_tstamp_tx(struct sk_buff *orig_skb,
3844 struct skb_shared_hwtstamps *hwtstamps);
3847 * skb_tx_timestamp() - Driver hook for transmit timestamping
3849 * Ethernet MAC Drivers should call this function in their hard_xmit()
3850 * function immediately before giving the sk_buff to the MAC hardware.
3852 * Specifically, one should make absolutely sure that this function is
3853 * called before TX completion of this packet can trigger. Otherwise
3854 * the packet could potentially already be freed.
3856 * @skb: A socket buffer.
3858 static inline void skb_tx_timestamp(struct sk_buff *skb)
3860 skb_clone_tx_timestamp(skb);
3861 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3862 skb_tstamp_tx(skb, NULL);
3866 * skb_complete_wifi_ack - deliver skb with wifi status
3868 * @skb: the original outgoing packet
3869 * @acked: ack status
3872 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3874 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3875 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3877 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3879 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3881 (skb->ip_summed == CHECKSUM_PARTIAL &&
3882 skb_checksum_start_offset(skb) >= 0));
3886 * skb_checksum_complete - Calculate checksum of an entire packet
3887 * @skb: packet to process
3889 * This function calculates the checksum over the entire packet plus
3890 * the value of skb->csum. The latter can be used to supply the
3891 * checksum of a pseudo header as used by TCP/UDP. It returns the
3894 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3895 * this function can be used to verify that checksum on received
3896 * packets. In that case the function should return zero if the
3897 * checksum is correct. In particular, this function will return zero
3898 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3899 * hardware has already verified the correctness of the checksum.
3901 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3903 return skb_csum_unnecessary(skb) ?
3904 0 : __skb_checksum_complete(skb);
3907 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3909 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3910 if (skb->csum_level == 0)
3911 skb->ip_summed = CHECKSUM_NONE;
3917 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3919 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3920 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3922 } else if (skb->ip_summed == CHECKSUM_NONE) {
3923 skb->ip_summed = CHECKSUM_UNNECESSARY;
3924 skb->csum_level = 0;
3928 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
3930 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3931 skb->ip_summed = CHECKSUM_NONE;
3932 skb->csum_level = 0;
3936 /* Check if we need to perform checksum complete validation.
3938 * Returns true if checksum complete is needed, false otherwise
3939 * (either checksum is unnecessary or zero checksum is allowed).
3941 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3945 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3946 skb->csum_valid = 1;
3947 __skb_decr_checksum_unnecessary(skb);
3954 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
3957 #define CHECKSUM_BREAK 76
3959 /* Unset checksum-complete
3961 * Unset checksum complete can be done when packet is being modified
3962 * (uncompressed for instance) and checksum-complete value is
3965 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3967 if (skb->ip_summed == CHECKSUM_COMPLETE)
3968 skb->ip_summed = CHECKSUM_NONE;
3971 /* Validate (init) checksum based on checksum complete.
3974 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3975 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3976 * checksum is stored in skb->csum for use in __skb_checksum_complete
3977 * non-zero: value of invalid checksum
3980 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3984 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3985 if (!csum_fold(csum_add(psum, skb->csum))) {
3986 skb->csum_valid = 1;
3993 if (complete || skb->len <= CHECKSUM_BREAK) {
3996 csum = __skb_checksum_complete(skb);
3997 skb->csum_valid = !csum;
4004 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4009 /* Perform checksum validate (init). Note that this is a macro since we only
4010 * want to calculate the pseudo header which is an input function if necessary.
4011 * First we try to validate without any computation (checksum unnecessary) and
4012 * then calculate based on checksum complete calling the function to compute
4016 * 0: checksum is validated or try to in skb_checksum_complete
4017 * non-zero: value of invalid checksum
4019 #define __skb_checksum_validate(skb, proto, complete, \
4020 zero_okay, check, compute_pseudo) \
4022 __sum16 __ret = 0; \
4023 skb->csum_valid = 0; \
4024 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4025 __ret = __skb_checksum_validate_complete(skb, \
4026 complete, compute_pseudo(skb, proto)); \
4030 #define skb_checksum_init(skb, proto, compute_pseudo) \
4031 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4033 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4034 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4036 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4037 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4039 #define skb_checksum_validate_zero_check(skb, proto, check, \
4041 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4043 #define skb_checksum_simple_validate(skb) \
4044 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4046 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4048 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4051 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4053 skb->csum = ~pseudo;
4054 skb->ip_summed = CHECKSUM_COMPLETE;
4057 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4059 if (__skb_checksum_convert_check(skb)) \
4060 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4063 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4064 u16 start, u16 offset)
4066 skb->ip_summed = CHECKSUM_PARTIAL;
4067 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4068 skb->csum_offset = offset - start;
4071 /* Update skbuf and packet to reflect the remote checksum offload operation.
4072 * When called, ptr indicates the starting point for skb->csum when
4073 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4074 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4076 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4077 int start, int offset, bool nopartial)
4082 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4086 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4087 __skb_checksum_complete(skb);
4088 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4091 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4093 /* Adjust skb->csum since we changed the packet */
4094 skb->csum = csum_add(skb->csum, delta);
4097 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4099 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4100 return (void *)(skb->_nfct & NFCT_PTRMASK);
4106 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4108 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4115 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4117 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4122 #ifdef CONFIG_SKB_EXTENSIONS
4124 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4130 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4133 #if IS_ENABLED(CONFIG_MPTCP)
4136 SKB_EXT_NUM, /* must be last */
4140 * struct skb_ext - sk_buff extensions
4141 * @refcnt: 1 on allocation, deallocated on 0
4142 * @offset: offset to add to @data to obtain extension address
4143 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4144 * @data: start of extension data, variable sized
4146 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4147 * to use 'u8' types while allowing up to 2kb worth of extension data.
4151 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4152 u8 chunks; /* same */
4153 char data[] __aligned(8);
4156 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4157 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4158 struct skb_ext *ext);
4159 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4160 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4161 void __skb_ext_put(struct skb_ext *ext);
4163 static inline void skb_ext_put(struct sk_buff *skb)
4165 if (skb->active_extensions)
4166 __skb_ext_put(skb->extensions);
4169 static inline void __skb_ext_copy(struct sk_buff *dst,
4170 const struct sk_buff *src)
4172 dst->active_extensions = src->active_extensions;
4174 if (src->active_extensions) {
4175 struct skb_ext *ext = src->extensions;
4177 refcount_inc(&ext->refcnt);
4178 dst->extensions = ext;
4182 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4185 __skb_ext_copy(dst, src);
4188 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4190 return !!ext->offset[i];
4193 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4195 return skb->active_extensions & (1 << id);
4198 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4200 if (skb_ext_exist(skb, id))
4201 __skb_ext_del(skb, id);
4204 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4206 if (skb_ext_exist(skb, id)) {
4207 struct skb_ext *ext = skb->extensions;
4209 return (void *)ext + (ext->offset[id] << 3);
4215 static inline void skb_ext_reset(struct sk_buff *skb)
4217 if (unlikely(skb->active_extensions)) {
4218 __skb_ext_put(skb->extensions);
4219 skb->active_extensions = 0;
4223 static inline bool skb_has_extensions(struct sk_buff *skb)
4225 return unlikely(skb->active_extensions);
4228 static inline void skb_ext_put(struct sk_buff *skb) {}
4229 static inline void skb_ext_reset(struct sk_buff *skb) {}
4230 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4231 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4232 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4233 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4234 #endif /* CONFIG_SKB_EXTENSIONS */
4236 static inline void nf_reset_ct(struct sk_buff *skb)
4238 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4239 nf_conntrack_put(skb_nfct(skb));
4244 static inline void nf_reset_trace(struct sk_buff *skb)
4246 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4251 static inline void ipvs_reset(struct sk_buff *skb)
4253 #if IS_ENABLED(CONFIG_IP_VS)
4254 skb->ipvs_property = 0;
4258 /* Note: This doesn't put any conntrack info in dst. */
4259 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4262 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4263 dst->_nfct = src->_nfct;
4264 nf_conntrack_get(skb_nfct(src));
4266 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4268 dst->nf_trace = src->nf_trace;
4272 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4274 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4275 nf_conntrack_put(skb_nfct(dst));
4277 __nf_copy(dst, src, true);
4280 #ifdef CONFIG_NETWORK_SECMARK
4281 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4283 to->secmark = from->secmark;
4286 static inline void skb_init_secmark(struct sk_buff *skb)
4291 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4294 static inline void skb_init_secmark(struct sk_buff *skb)
4298 static inline int secpath_exists(const struct sk_buff *skb)
4301 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4307 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4309 return !skb->destructor &&
4310 !secpath_exists(skb) &&
4312 !skb->_skb_refdst &&
4313 !skb_has_frag_list(skb);
4316 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4318 skb->queue_mapping = queue_mapping;
4321 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4323 return skb->queue_mapping;
4326 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4328 to->queue_mapping = from->queue_mapping;
4331 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4333 skb->queue_mapping = rx_queue + 1;
4336 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4338 return skb->queue_mapping - 1;
4341 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4343 return skb->queue_mapping != 0;
4346 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4348 skb->dst_pending_confirm = val;
4351 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4353 return skb->dst_pending_confirm != 0;
4356 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4359 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4365 /* Keeps track of mac header offset relative to skb->head.
4366 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4367 * For non-tunnel skb it points to skb_mac_header() and for
4368 * tunnel skb it points to outer mac header.
4369 * Keeps track of level of encapsulation of network headers.
4380 #define SKB_GSO_CB_OFFSET 32
4381 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET))
4383 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4385 return (skb_mac_header(inner_skb) - inner_skb->head) -
4386 SKB_GSO_CB(inner_skb)->mac_offset;
4389 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4391 int new_headroom, headroom;
4394 headroom = skb_headroom(skb);
4395 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4399 new_headroom = skb_headroom(skb);
4400 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4404 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4406 /* Do not update partial checksums if remote checksum is enabled. */
4407 if (skb->remcsum_offload)
4410 SKB_GSO_CB(skb)->csum = res;
4411 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4414 /* Compute the checksum for a gso segment. First compute the checksum value
4415 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4416 * then add in skb->csum (checksum from csum_start to end of packet).
4417 * skb->csum and csum_start are then updated to reflect the checksum of the
4418 * resultant packet starting from the transport header-- the resultant checksum
4419 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4422 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4424 unsigned char *csum_start = skb_transport_header(skb);
4425 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4426 __wsum partial = SKB_GSO_CB(skb)->csum;
4428 SKB_GSO_CB(skb)->csum = res;
4429 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4431 return csum_fold(csum_partial(csum_start, plen, partial));
4434 static inline bool skb_is_gso(const struct sk_buff *skb)
4436 return skb_shinfo(skb)->gso_size;
4439 /* Note: Should be called only if skb_is_gso(skb) is true */
4440 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4442 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4445 /* Note: Should be called only if skb_is_gso(skb) is true */
4446 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4448 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4451 /* Note: Should be called only if skb_is_gso(skb) is true */
4452 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4454 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4457 static inline void skb_gso_reset(struct sk_buff *skb)
4459 skb_shinfo(skb)->gso_size = 0;
4460 skb_shinfo(skb)->gso_segs = 0;
4461 skb_shinfo(skb)->gso_type = 0;
4464 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4467 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4469 shinfo->gso_size += increment;
4472 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4475 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4477 shinfo->gso_size -= decrement;
4480 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4482 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4484 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4485 * wanted then gso_type will be set. */
4486 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4488 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4489 unlikely(shinfo->gso_type == 0)) {
4490 __skb_warn_lro_forwarding(skb);
4496 static inline void skb_forward_csum(struct sk_buff *skb)
4498 /* Unfortunately we don't support this one. Any brave souls? */
4499 if (skb->ip_summed == CHECKSUM_COMPLETE)
4500 skb->ip_summed = CHECKSUM_NONE;
4504 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4505 * @skb: skb to check
4507 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4508 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4509 * use this helper, to document places where we make this assertion.
4511 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4514 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4518 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4520 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4521 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4522 unsigned int transport_len,
4523 __sum16(*skb_chkf)(struct sk_buff *skb));
4526 * skb_head_is_locked - Determine if the skb->head is locked down
4527 * @skb: skb to check
4529 * The head on skbs build around a head frag can be removed if they are
4530 * not cloned. This function returns true if the skb head is locked down
4531 * due to either being allocated via kmalloc, or by being a clone with
4532 * multiple references to the head.
4534 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4536 return !skb->head_frag || skb_cloned(skb);
4539 /* Local Checksum Offload.
4540 * Compute outer checksum based on the assumption that the
4541 * inner checksum will be offloaded later.
4542 * See Documentation/networking/checksum-offloads.rst for
4543 * explanation of how this works.
4544 * Fill in outer checksum adjustment (e.g. with sum of outer
4545 * pseudo-header) before calling.
4546 * Also ensure that inner checksum is in linear data area.
4548 static inline __wsum lco_csum(struct sk_buff *skb)
4550 unsigned char *csum_start = skb_checksum_start(skb);
4551 unsigned char *l4_hdr = skb_transport_header(skb);
4554 /* Start with complement of inner checksum adjustment */
4555 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4558 /* Add in checksum of our headers (incl. outer checksum
4559 * adjustment filled in by caller) and return result.
4561 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4564 static inline bool skb_is_redirected(const struct sk_buff *skb)
4566 #ifdef CONFIG_NET_REDIRECT
4567 return skb->redirected;
4573 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
4575 #ifdef CONFIG_NET_REDIRECT
4576 skb->redirected = 1;
4577 skb->from_ingress = from_ingress;
4578 if (skb->from_ingress)
4583 static inline void skb_reset_redirect(struct sk_buff *skb)
4585 #ifdef CONFIG_NET_REDIRECT
4586 skb->redirected = 0;
4590 #endif /* __KERNEL__ */
4591 #endif /* _LINUX_SKBUFF_H */