1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * Definitions for the 'struct sk_buff' memory handlers.
6 * Alan Cox, <gw4pts@gw4pts.ampr.org>
7 * Florian La Roche, <rzsfl@rz.uni-sb.de>
10 #ifndef _LINUX_SKBUFF_H
11 #define _LINUX_SKBUFF_H
13 #include <linux/kernel.h>
14 #include <linux/compiler.h>
15 #include <linux/time.h>
16 #include <linux/bug.h>
17 #include <linux/bvec.h>
18 #include <linux/cache.h>
19 #include <linux/rbtree.h>
20 #include <linux/socket.h>
21 #include <linux/refcount.h>
23 #include <linux/atomic.h>
24 #include <asm/types.h>
25 #include <linux/spinlock.h>
26 #include <linux/net.h>
27 #include <linux/textsearch.h>
28 #include <net/checksum.h>
29 #include <linux/rcupdate.h>
30 #include <linux/hrtimer.h>
31 #include <linux/dma-mapping.h>
32 #include <linux/netdev_features.h>
33 #include <linux/sched.h>
34 #include <linux/sched/clock.h>
35 #include <net/flow_dissector.h>
36 #include <linux/splice.h>
37 #include <linux/in6.h>
38 #include <linux/if_packet.h>
40 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
41 #include <linux/netfilter/nf_conntrack_common.h>
44 /* The interface for checksum offload between the stack and networking drivers
47 * A. IP checksum related features
49 * Drivers advertise checksum offload capabilities in the features of a device.
50 * From the stack's point of view these are capabilities offered by the driver.
51 * A driver typically only advertises features that it is capable of offloading
54 * The checksum related features are:
56 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
57 * IP (one's complement) checksum for any combination
58 * of protocols or protocol layering. The checksum is
59 * computed and set in a packet per the CHECKSUM_PARTIAL
60 * interface (see below).
62 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
63 * TCP or UDP packets over IPv4. These are specifically
64 * unencapsulated packets of the form IPv4|TCP or
65 * IPv4|UDP where the Protocol field in the IPv4 header
66 * is TCP or UDP. The IPv4 header may contain IP options.
67 * This feature cannot be set in features for a device
68 * with NETIF_F_HW_CSUM also set. This feature is being
69 * DEPRECATED (see below).
71 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
72 * TCP or UDP packets over IPv6. These are specifically
73 * unencapsulated packets of the form IPv6|TCP or
74 * IPv6|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,
261 u8 sabotage_in_done:1;
263 struct net_device *physindev;
265 /* always valid & non-NULL from FORWARD on, for physdev match */
266 struct net_device *physoutdev;
268 /* prerouting: detect dnat in orig/reply direction */
270 struct in6_addr ipv6_daddr;
272 /* after prerouting + nat detected: store original source
273 * mac since neigh resolution overwrites it, only used while
274 * skb is out in neigh layer.
276 char neigh_header[8];
281 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
282 /* Chain in tc_skb_ext will be used to share the tc chain with
283 * ovs recirc_id. It will be set to the current chain by tc
284 * and read by ovs to recirc_id.
292 struct sk_buff_head {
293 /* These two members must be first. */
294 struct sk_buff *next;
295 struct sk_buff *prev;
303 /* To allow 64K frame to be packed as single skb without frag_list we
304 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
305 * buffers which do not start on a page boundary.
307 * Since GRO uses frags we allocate at least 16 regardless of page
310 #if (65536/PAGE_SIZE + 1) < 16
311 #define MAX_SKB_FRAGS 16UL
313 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
315 extern int sysctl_max_skb_frags;
317 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
318 * segment using its current segmentation instead.
320 #define GSO_BY_FRAGS 0xFFFF
322 typedef struct bio_vec skb_frag_t;
325 * skb_frag_size() - Returns the size of a skb fragment
326 * @frag: skb fragment
328 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
334 * skb_frag_size_set() - Sets the size of a skb fragment
335 * @frag: skb fragment
336 * @size: size of fragment
338 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
344 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
345 * @frag: skb fragment
346 * @delta: value to add
348 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
350 frag->bv_len += delta;
354 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
355 * @frag: skb fragment
356 * @delta: value to subtract
358 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
360 frag->bv_len -= delta;
364 * skb_frag_must_loop - Test if %p is a high memory page
365 * @p: fragment's page
367 static inline bool skb_frag_must_loop(struct page *p)
369 #if defined(CONFIG_HIGHMEM)
377 * skb_frag_foreach_page - loop over pages in a fragment
379 * @f: skb frag to operate on
380 * @f_off: offset from start of f->bv_page
381 * @f_len: length from f_off to loop over
382 * @p: (temp var) current page
383 * @p_off: (temp var) offset from start of current page,
384 * non-zero only on first page.
385 * @p_len: (temp var) length in current page,
386 * < PAGE_SIZE only on first and last page.
387 * @copied: (temp var) length so far, excluding current p_len.
389 * A fragment can hold a compound page, in which case per-page
390 * operations, notably kmap_atomic, must be called for each
393 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
394 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
395 p_off = (f_off) & (PAGE_SIZE - 1), \
396 p_len = skb_frag_must_loop(p) ? \
397 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
400 copied += p_len, p++, p_off = 0, \
401 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
403 #define HAVE_HW_TIME_STAMP
406 * struct skb_shared_hwtstamps - hardware time stamps
407 * @hwtstamp: hardware time stamp transformed into duration
408 * since arbitrary point in time
410 * Software time stamps generated by ktime_get_real() are stored in
413 * hwtstamps can only be compared against other hwtstamps from
416 * This structure is attached to packets as part of the
417 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
419 struct skb_shared_hwtstamps {
423 /* Definitions for tx_flags in struct skb_shared_info */
425 /* generate hardware time stamp */
426 SKBTX_HW_TSTAMP = 1 << 0,
428 /* generate software time stamp when queueing packet to NIC */
429 SKBTX_SW_TSTAMP = 1 << 1,
431 /* device driver is going to provide hardware time stamp */
432 SKBTX_IN_PROGRESS = 1 << 2,
434 /* device driver supports TX zero-copy buffers */
435 SKBTX_DEV_ZEROCOPY = 1 << 3,
437 /* generate wifi status information (where possible) */
438 SKBTX_WIFI_STATUS = 1 << 4,
440 /* This indicates at least one fragment might be overwritten
441 * (as in vmsplice(), sendfile() ...)
442 * If we need to compute a TX checksum, we'll need to copy
443 * all frags to avoid possible bad checksum
445 SKBTX_SHARED_FRAG = 1 << 5,
447 /* generate software time stamp when entering packet scheduling */
448 SKBTX_SCHED_TSTAMP = 1 << 6,
451 #define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
452 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
454 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
457 * The callback notifies userspace to release buffers when skb DMA is done in
458 * lower device, the skb last reference should be 0 when calling this.
459 * The zerocopy_success argument is true if zero copy transmit occurred,
460 * false on data copy or out of memory error caused by data copy attempt.
461 * The ctx field is used to track device context.
462 * The desc field is used to track userspace buffer index.
465 void (*callback)(struct ubuf_info *, bool zerocopy_success);
481 struct user_struct *user;
486 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
488 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
489 void mm_unaccount_pinned_pages(struct mmpin *mmp);
491 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
492 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
493 struct ubuf_info *uarg);
495 static inline void sock_zerocopy_get(struct ubuf_info *uarg)
497 refcount_inc(&uarg->refcnt);
500 void sock_zerocopy_put(struct ubuf_info *uarg);
501 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
503 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
505 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len);
506 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
507 struct msghdr *msg, int len,
508 struct ubuf_info *uarg);
510 /* This data is invariant across clones and lives at
511 * the end of the header data, ie. at skb->end.
513 struct skb_shared_info {
518 unsigned short gso_size;
519 /* Warning: this field is not always filled in (UFO)! */
520 unsigned short gso_segs;
521 struct sk_buff *frag_list;
522 struct skb_shared_hwtstamps hwtstamps;
523 unsigned int gso_type;
527 * Warning : all fields before dataref are cleared in __alloc_skb()
531 /* Intermediate layers must ensure that destructor_arg
532 * remains valid until skb destructor */
533 void * destructor_arg;
535 /* must be last field, see pskb_expand_head() */
536 skb_frag_t frags[MAX_SKB_FRAGS];
539 /* We divide dataref into two halves. The higher 16 bits hold references
540 * to the payload part of skb->data. The lower 16 bits hold references to
541 * the entire skb->data. A clone of a headerless skb holds the length of
542 * the header in skb->hdr_len.
544 * All users must obey the rule that the skb->data reference count must be
545 * greater than or equal to the payload reference count.
547 * Holding a reference to the payload part means that the user does not
548 * care about modifications to the header part of skb->data.
550 #define SKB_DATAREF_SHIFT 16
551 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
555 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
556 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
557 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
561 SKB_GSO_TCPV4 = 1 << 0,
563 /* This indicates the skb is from an untrusted source. */
564 SKB_GSO_DODGY = 1 << 1,
566 /* This indicates the tcp segment has CWR set. */
567 SKB_GSO_TCP_ECN = 1 << 2,
569 SKB_GSO_TCP_FIXEDID = 1 << 3,
571 SKB_GSO_TCPV6 = 1 << 4,
573 SKB_GSO_FCOE = 1 << 5,
575 SKB_GSO_GRE = 1 << 6,
577 SKB_GSO_GRE_CSUM = 1 << 7,
579 SKB_GSO_IPXIP4 = 1 << 8,
581 SKB_GSO_IPXIP6 = 1 << 9,
583 SKB_GSO_UDP_TUNNEL = 1 << 10,
585 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
587 SKB_GSO_PARTIAL = 1 << 12,
589 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
591 SKB_GSO_SCTP = 1 << 14,
593 SKB_GSO_ESP = 1 << 15,
595 SKB_GSO_UDP = 1 << 16,
597 SKB_GSO_UDP_L4 = 1 << 17,
599 SKB_GSO_FRAGLIST = 1 << 18,
602 #if BITS_PER_LONG > 32
603 #define NET_SKBUFF_DATA_USES_OFFSET 1
606 #ifdef NET_SKBUFF_DATA_USES_OFFSET
607 typedef unsigned int sk_buff_data_t;
609 typedef unsigned char *sk_buff_data_t;
613 * struct sk_buff - socket buffer
614 * @next: Next buffer in list
615 * @prev: Previous buffer in list
616 * @tstamp: Time we arrived/left
617 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
618 * for retransmit timer
619 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
621 * @sk: Socket we are owned by
622 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
623 * fragmentation management
624 * @dev: Device we arrived on/are leaving by
625 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
626 * @cb: Control buffer. Free for use by every layer. Put private vars here
627 * @_skb_refdst: destination entry (with norefcount bit)
628 * @sp: the security path, used for xfrm
629 * @len: Length of actual data
630 * @data_len: Data length
631 * @mac_len: Length of link layer header
632 * @hdr_len: writable header length of cloned skb
633 * @csum: Checksum (must include start/offset pair)
634 * @csum_start: Offset from skb->head where checksumming should start
635 * @csum_offset: Offset from csum_start where checksum should be stored
636 * @priority: Packet queueing priority
637 * @ignore_df: allow local fragmentation
638 * @cloned: Head may be cloned (check refcnt to be sure)
639 * @ip_summed: Driver fed us an IP checksum
640 * @nohdr: Payload reference only, must not modify header
641 * @pkt_type: Packet class
642 * @fclone: skbuff clone status
643 * @ipvs_property: skbuff is owned by ipvs
644 * @inner_protocol_type: whether the inner protocol is
645 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
646 * @remcsum_offload: remote checksum offload is enabled
647 * @offload_fwd_mark: Packet was L2-forwarded in hardware
648 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
649 * @tc_skip_classify: do not classify packet. set by IFB device
650 * @tc_at_ingress: used within tc_classify to distinguish in/egress
651 * @redirected: packet was redirected by packet classifier
652 * @from_ingress: packet was redirected from the ingress path
653 * @peeked: this packet has been seen already, so stats have been
654 * done for it, don't do them again
655 * @nf_trace: netfilter packet trace flag
656 * @protocol: Packet protocol from driver
657 * @destructor: Destruct function
658 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
659 * @_nfct: Associated connection, if any (with nfctinfo bits)
660 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
661 * @skb_iif: ifindex of device we arrived on
662 * @tc_index: Traffic control index
663 * @hash: the packet hash
664 * @queue_mapping: Queue mapping for multiqueue devices
665 * @head_frag: skb was allocated from page fragments,
666 * not allocated by kmalloc() or vmalloc().
667 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
668 * @active_extensions: active extensions (skb_ext_id types)
669 * @ndisc_nodetype: router type (from link layer)
670 * @ooo_okay: allow the mapping of a socket to a queue to be changed
671 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
673 * @sw_hash: indicates hash was computed in software stack
674 * @wifi_acked_valid: wifi_acked was set
675 * @wifi_acked: whether frame was acked on wifi or not
676 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
677 * @encapsulation: indicates the inner headers in the skbuff are valid
678 * @encap_hdr_csum: software checksum is needed
679 * @csum_valid: checksum is already valid
680 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
681 * @csum_complete_sw: checksum was completed by software
682 * @csum_level: indicates the number of consecutive checksums found in
683 * the packet minus one that have been verified as
684 * CHECKSUM_UNNECESSARY (max 3)
685 * @scm_io_uring: SKB holds io_uring registered files
686 * @dst_pending_confirm: need to confirm neighbour
687 * @decrypted: Decrypted SKB
688 * @napi_id: id of the NAPI struct this skb came from
689 * @sender_cpu: (aka @napi_id) source CPU in XPS
690 * @secmark: security marking
691 * @mark: Generic packet mark
692 * @reserved_tailroom: (aka @mark) number of bytes of free space available
693 * at the tail of an sk_buff
694 * @vlan_present: VLAN tag is present
695 * @vlan_proto: vlan encapsulation protocol
696 * @vlan_tci: vlan tag control information
697 * @inner_protocol: Protocol (encapsulation)
698 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
699 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
700 * @inner_transport_header: Inner transport layer header (encapsulation)
701 * @inner_network_header: Network layer header (encapsulation)
702 * @inner_mac_header: Link layer header (encapsulation)
703 * @transport_header: Transport layer header
704 * @network_header: Network layer header
705 * @mac_header: Link layer header
706 * @kcov_handle: KCOV remote handle for remote coverage collection
707 * @tail: Tail pointer
709 * @head: Head of buffer
710 * @data: Data head pointer
711 * @truesize: Buffer size
712 * @users: User count - see {datagram,tcp}.c
713 * @extensions: allocated extensions, valid if active_extensions is nonzero
719 /* These two members must be first. */
720 struct sk_buff *next;
721 struct sk_buff *prev;
724 struct net_device *dev;
725 /* Some protocols might use this space to store information,
726 * while device pointer would be NULL.
727 * UDP receive path is one user.
729 unsigned long dev_scratch;
732 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
733 struct list_head list;
738 int ip_defrag_offset;
743 u64 skb_mstamp_ns; /* earliest departure time */
746 * This is the control buffer. It is free to use for every
747 * layer. Please put your private variables there. If you
748 * want to keep them across layers you have to do a skb_clone()
749 * first. This is owned by whoever has the skb queued ATM.
751 char cb[48] __aligned(8);
755 unsigned long _skb_refdst;
756 void (*destructor)(struct sk_buff *skb);
758 struct list_head tcp_tsorted_anchor;
761 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
769 /* Following fields are _not_ copied in __copy_skb_header()
770 * Note that queue_mapping is here mostly to fill a hole.
774 /* if you move cloned around you also must adapt those constants */
775 #ifdef __BIG_ENDIAN_BITFIELD
776 #define CLONED_MASK (1 << 7)
778 #define CLONED_MASK 1
780 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
783 __u8 __cloned_offset[0];
791 #ifdef CONFIG_SKB_EXTENSIONS
792 __u8 active_extensions;
794 /* fields enclosed in headers_start/headers_end are copied
795 * using a single memcpy() in __copy_skb_header()
798 __u32 headers_start[0];
801 /* if you move pkt_type around you also must adapt those constants */
802 #ifdef __BIG_ENDIAN_BITFIELD
803 #define PKT_TYPE_MAX (7 << 5)
805 #define PKT_TYPE_MAX 7
807 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
810 __u8 __pkt_type_offset[0];
820 __u8 wifi_acked_valid:1;
823 /* Indicates the inner headers are valid in the skbuff. */
824 __u8 encapsulation:1;
825 __u8 encap_hdr_csum:1;
828 #ifdef __BIG_ENDIAN_BITFIELD
829 #define PKT_VLAN_PRESENT_BIT 7
831 #define PKT_VLAN_PRESENT_BIT 0
833 #define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset)
835 __u8 __pkt_vlan_present_offset[0];
838 __u8 csum_complete_sw:1;
840 __u8 csum_not_inet:1;
841 __u8 dst_pending_confirm:1;
842 #ifdef CONFIG_IPV6_NDISC_NODETYPE
843 __u8 ndisc_nodetype:2;
846 __u8 ipvs_property:1;
847 __u8 inner_protocol_type:1;
848 __u8 remcsum_offload:1;
849 #ifdef CONFIG_NET_SWITCHDEV
850 __u8 offload_fwd_mark:1;
851 __u8 offload_l3_fwd_mark:1;
853 #ifdef CONFIG_NET_CLS_ACT
854 __u8 tc_skip_classify:1;
855 __u8 tc_at_ingress:1;
857 #ifdef CONFIG_NET_REDIRECT
861 #ifdef CONFIG_TLS_DEVICE
866 #ifdef CONFIG_NET_SCHED
867 __u16 tc_index; /* traffic control index */
882 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
884 unsigned int napi_id;
885 unsigned int sender_cpu;
888 #ifdef CONFIG_NETWORK_SECMARK
894 __u32 reserved_tailroom;
898 __be16 inner_protocol;
902 __u16 inner_transport_header;
903 __u16 inner_network_header;
904 __u16 inner_mac_header;
907 __u16 transport_header;
908 __u16 network_header;
916 __u32 headers_end[0];
919 /* These elements must be at the end, see alloc_skb() for details. */
924 unsigned int truesize;
927 #ifdef CONFIG_SKB_EXTENSIONS
928 /* only useable after checking ->active_extensions != 0 */
929 struct skb_ext *extensions;
935 * Handling routines are only of interest to the kernel
938 #define SKB_ALLOC_FCLONE 0x01
939 #define SKB_ALLOC_RX 0x02
940 #define SKB_ALLOC_NAPI 0x04
943 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
946 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
948 return unlikely(skb->pfmemalloc);
952 * skb might have a dst pointer attached, refcounted or not.
953 * _skb_refdst low order bit is set if refcount was _not_ taken
955 #define SKB_DST_NOREF 1UL
956 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
959 * skb_dst - returns skb dst_entry
962 * Returns skb dst_entry, regardless of reference taken or not.
964 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
966 /* If refdst was not refcounted, check we still are in a
967 * rcu_read_lock section
969 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
970 !rcu_read_lock_held() &&
971 !rcu_read_lock_bh_held());
972 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
976 * skb_dst_set - sets skb dst
980 * Sets skb dst, assuming a reference was taken on dst and should
981 * be released by skb_dst_drop()
983 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
985 skb->_skb_refdst = (unsigned long)dst;
989 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
993 * Sets skb dst, assuming a reference was not taken on dst.
994 * If dst entry is cached, we do not take reference and dst_release
995 * will be avoided by refdst_drop. If dst entry is not cached, we take
996 * reference, so that last dst_release can destroy the dst immediately.
998 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1000 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1001 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1005 * skb_dst_is_noref - Test if skb dst isn't refcounted
1008 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1010 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1014 * skb_rtable - Returns the skb &rtable
1017 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1019 return (struct rtable *)skb_dst(skb);
1022 /* For mangling skb->pkt_type from user space side from applications
1023 * such as nft, tc, etc, we only allow a conservative subset of
1024 * possible pkt_types to be set.
1026 static inline bool skb_pkt_type_ok(u32 ptype)
1028 return ptype <= PACKET_OTHERHOST;
1032 * skb_napi_id - Returns the skb's NAPI id
1035 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1037 #ifdef CONFIG_NET_RX_BUSY_POLL
1038 return skb->napi_id;
1045 * skb_unref - decrement the skb's reference count
1048 * Returns true if we can free the skb.
1050 static inline bool skb_unref(struct sk_buff *skb)
1054 if (likely(refcount_read(&skb->users) == 1))
1056 else if (likely(!refcount_dec_and_test(&skb->users)))
1062 void skb_release_head_state(struct sk_buff *skb);
1063 void kfree_skb(struct sk_buff *skb);
1064 void kfree_skb_list(struct sk_buff *segs);
1065 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1066 void skb_tx_error(struct sk_buff *skb);
1068 #ifdef CONFIG_TRACEPOINTS
1069 void consume_skb(struct sk_buff *skb);
1071 static inline void consume_skb(struct sk_buff *skb)
1073 return kfree_skb(skb);
1077 void __consume_stateless_skb(struct sk_buff *skb);
1078 void __kfree_skb(struct sk_buff *skb);
1079 extern struct kmem_cache *skbuff_head_cache;
1081 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1082 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1083 bool *fragstolen, int *delta_truesize);
1085 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1087 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1088 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1089 struct sk_buff *build_skb_around(struct sk_buff *skb,
1090 void *data, unsigned int frag_size);
1093 * alloc_skb - allocate a network buffer
1094 * @size: size to allocate
1095 * @priority: allocation mask
1097 * This function is a convenient wrapper around __alloc_skb().
1099 static inline struct sk_buff *alloc_skb(unsigned int size,
1102 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1105 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1106 unsigned long data_len,
1110 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1112 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1113 struct sk_buff_fclones {
1114 struct sk_buff skb1;
1116 struct sk_buff skb2;
1118 refcount_t fclone_ref;
1122 * skb_fclone_busy - check if fclone is busy
1126 * Returns true if skb is a fast clone, and its clone is not freed.
1127 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1128 * so we also check that this didnt happen.
1130 static inline bool skb_fclone_busy(const struct sock *sk,
1131 const struct sk_buff *skb)
1133 const struct sk_buff_fclones *fclones;
1135 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1137 return skb->fclone == SKB_FCLONE_ORIG &&
1138 refcount_read(&fclones->fclone_ref) > 1 &&
1139 fclones->skb2.sk == sk;
1143 * alloc_skb_fclone - allocate a network buffer from fclone cache
1144 * @size: size to allocate
1145 * @priority: allocation mask
1147 * This function is a convenient wrapper around __alloc_skb().
1149 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1152 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1155 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1156 void skb_headers_offset_update(struct sk_buff *skb, int off);
1157 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1158 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1159 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1160 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1161 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1162 gfp_t gfp_mask, bool fclone);
1163 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1166 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1169 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1170 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1171 unsigned int headroom);
1172 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1173 int newtailroom, gfp_t priority);
1174 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1175 int offset, int len);
1176 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1177 int offset, int len);
1178 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1179 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1182 * skb_pad - zero pad the tail of an skb
1183 * @skb: buffer to pad
1184 * @pad: space to pad
1186 * Ensure that a buffer is followed by a padding area that is zero
1187 * filled. Used by network drivers which may DMA or transfer data
1188 * beyond the buffer end onto the wire.
1190 * May return error in out of memory cases. The skb is freed on error.
1192 static inline int skb_pad(struct sk_buff *skb, int pad)
1194 return __skb_pad(skb, pad, true);
1196 #define dev_kfree_skb(a) consume_skb(a)
1198 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1199 int offset, size_t size);
1201 struct skb_seq_state {
1205 __u32 stepped_offset;
1206 struct sk_buff *root_skb;
1207 struct sk_buff *cur_skb;
1211 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1212 unsigned int to, struct skb_seq_state *st);
1213 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1214 struct skb_seq_state *st);
1215 void skb_abort_seq_read(struct skb_seq_state *st);
1217 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1218 unsigned int to, struct ts_config *config);
1221 * Packet hash types specify the type of hash in skb_set_hash.
1223 * Hash types refer to the protocol layer addresses which are used to
1224 * construct a packet's hash. The hashes are used to differentiate or identify
1225 * flows of the protocol layer for the hash type. Hash types are either
1226 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1228 * Properties of hashes:
1230 * 1) Two packets in different flows have different hash values
1231 * 2) Two packets in the same flow should have the same hash value
1233 * A hash at a higher layer is considered to be more specific. A driver should
1234 * set the most specific hash possible.
1236 * A driver cannot indicate a more specific hash than the layer at which a hash
1237 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1239 * A driver may indicate a hash level which is less specific than the
1240 * actual layer the hash was computed on. For instance, a hash computed
1241 * at L4 may be considered an L3 hash. This should only be done if the
1242 * driver can't unambiguously determine that the HW computed the hash at
1243 * the higher layer. Note that the "should" in the second property above
1246 enum pkt_hash_types {
1247 PKT_HASH_TYPE_NONE, /* Undefined type */
1248 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1249 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1250 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1253 static inline void skb_clear_hash(struct sk_buff *skb)
1260 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1263 skb_clear_hash(skb);
1267 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1269 skb->l4_hash = is_l4;
1270 skb->sw_hash = is_sw;
1275 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1277 /* Used by drivers to set hash from HW */
1278 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1282 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1284 __skb_set_hash(skb, hash, true, is_l4);
1287 void __skb_get_hash(struct sk_buff *skb);
1288 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1289 u32 skb_get_poff(const struct sk_buff *skb);
1290 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1291 const struct flow_keys_basic *keys, int hlen);
1292 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1293 void *data, int hlen_proto);
1295 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1296 int thoff, u8 ip_proto)
1298 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1301 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1302 const struct flow_dissector_key *key,
1303 unsigned int key_count);
1305 struct bpf_flow_dissector;
1306 bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1307 __be16 proto, int nhoff, int hlen, unsigned int flags);
1309 bool __skb_flow_dissect(const struct net *net,
1310 const struct sk_buff *skb,
1311 struct flow_dissector *flow_dissector,
1312 void *target_container,
1313 void *data, __be16 proto, int nhoff, int hlen,
1314 unsigned int flags);
1316 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1317 struct flow_dissector *flow_dissector,
1318 void *target_container, unsigned int flags)
1320 return __skb_flow_dissect(NULL, skb, flow_dissector,
1321 target_container, NULL, 0, 0, 0, flags);
1324 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1325 struct flow_keys *flow,
1328 memset(flow, 0, sizeof(*flow));
1329 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1330 flow, NULL, 0, 0, 0, flags);
1334 skb_flow_dissect_flow_keys_basic(const struct net *net,
1335 const struct sk_buff *skb,
1336 struct flow_keys_basic *flow, void *data,
1337 __be16 proto, int nhoff, int hlen,
1340 memset(flow, 0, sizeof(*flow));
1341 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1342 data, proto, nhoff, hlen, flags);
1345 void skb_flow_dissect_meta(const struct sk_buff *skb,
1346 struct flow_dissector *flow_dissector,
1347 void *target_container);
1349 /* Gets a skb connection tracking info, ctinfo map should be a
1350 * map of mapsize to translate enum ip_conntrack_info states
1354 skb_flow_dissect_ct(const struct sk_buff *skb,
1355 struct flow_dissector *flow_dissector,
1356 void *target_container,
1360 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1361 struct flow_dissector *flow_dissector,
1362 void *target_container);
1364 void skb_flow_dissect_hash(const struct sk_buff *skb,
1365 struct flow_dissector *flow_dissector,
1366 void *target_container);
1368 static inline __u32 skb_get_hash(struct sk_buff *skb)
1370 if (!skb->l4_hash && !skb->sw_hash)
1371 __skb_get_hash(skb);
1376 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1378 if (!skb->l4_hash && !skb->sw_hash) {
1379 struct flow_keys keys;
1380 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1382 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1388 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1389 const siphash_key_t *perturb);
1391 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1396 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1398 to->hash = from->hash;
1399 to->sw_hash = from->sw_hash;
1400 to->l4_hash = from->l4_hash;
1403 static inline void skb_copy_decrypted(struct sk_buff *to,
1404 const struct sk_buff *from)
1406 #ifdef CONFIG_TLS_DEVICE
1407 to->decrypted = from->decrypted;
1411 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1412 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1414 return skb->head + skb->end;
1417 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1422 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1427 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1429 return skb->end - skb->head;
1434 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1436 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1438 return &skb_shinfo(skb)->hwtstamps;
1441 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1443 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1445 return is_zcopy ? skb_uarg(skb) : NULL;
1448 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1451 if (skb && uarg && !skb_zcopy(skb)) {
1452 if (unlikely(have_ref && *have_ref))
1455 sock_zerocopy_get(uarg);
1456 skb_shinfo(skb)->destructor_arg = uarg;
1457 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1461 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1463 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1464 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1467 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1469 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1472 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1474 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1477 /* Release a reference on a zerocopy structure */
1478 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1480 struct ubuf_info *uarg = skb_zcopy(skb);
1483 if (skb_zcopy_is_nouarg(skb)) {
1484 /* no notification callback */
1485 } else if (uarg->callback == sock_zerocopy_callback) {
1486 uarg->zerocopy = uarg->zerocopy && zerocopy;
1487 sock_zerocopy_put(uarg);
1489 uarg->callback(uarg, zerocopy);
1492 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1496 /* Abort a zerocopy operation and revert zckey on error in send syscall */
1497 static inline void skb_zcopy_abort(struct sk_buff *skb)
1499 struct ubuf_info *uarg = skb_zcopy(skb);
1502 sock_zerocopy_put_abort(uarg, false);
1503 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1507 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1512 /* Iterate through singly-linked GSO fragments of an skb. */
1513 #define skb_list_walk_safe(first, skb, next_skb) \
1514 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1515 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1517 static inline void skb_list_del_init(struct sk_buff *skb)
1519 __list_del_entry(&skb->list);
1520 skb_mark_not_on_list(skb);
1524 * skb_queue_empty - check if a queue is empty
1527 * Returns true if the queue is empty, false otherwise.
1529 static inline int skb_queue_empty(const struct sk_buff_head *list)
1531 return list->next == (const struct sk_buff *) list;
1535 * skb_queue_empty_lockless - check if a queue is empty
1538 * Returns true if the queue is empty, false otherwise.
1539 * This variant can be used in lockless contexts.
1541 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1543 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1548 * skb_queue_is_last - check if skb is the last entry in the queue
1552 * Returns true if @skb is the last buffer on the list.
1554 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1555 const struct sk_buff *skb)
1557 return skb->next == (const struct sk_buff *) list;
1561 * skb_queue_is_first - check if skb is the first entry in the queue
1565 * Returns true if @skb is the first buffer on the list.
1567 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1568 const struct sk_buff *skb)
1570 return skb->prev == (const struct sk_buff *) list;
1574 * skb_queue_next - return the next packet in the queue
1576 * @skb: current buffer
1578 * Return the next packet in @list after @skb. It is only valid to
1579 * call this if skb_queue_is_last() evaluates to false.
1581 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1582 const struct sk_buff *skb)
1584 /* This BUG_ON may seem severe, but if we just return then we
1585 * are going to dereference garbage.
1587 BUG_ON(skb_queue_is_last(list, skb));
1592 * skb_queue_prev - return the prev packet in the queue
1594 * @skb: current buffer
1596 * Return the prev packet in @list before @skb. It is only valid to
1597 * call this if skb_queue_is_first() evaluates to false.
1599 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1600 const struct sk_buff *skb)
1602 /* This BUG_ON may seem severe, but if we just return then we
1603 * are going to dereference garbage.
1605 BUG_ON(skb_queue_is_first(list, skb));
1610 * skb_get - reference buffer
1611 * @skb: buffer to reference
1613 * Makes another reference to a socket buffer and returns a pointer
1616 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1618 refcount_inc(&skb->users);
1623 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1627 * skb_cloned - is the buffer a clone
1628 * @skb: buffer to check
1630 * Returns true if the buffer was generated with skb_clone() and is
1631 * one of multiple shared copies of the buffer. Cloned buffers are
1632 * shared data so must not be written to under normal circumstances.
1634 static inline int skb_cloned(const struct sk_buff *skb)
1636 return skb->cloned &&
1637 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1640 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1642 might_sleep_if(gfpflags_allow_blocking(pri));
1644 if (skb_cloned(skb))
1645 return pskb_expand_head(skb, 0, 0, pri);
1651 * skb_header_cloned - is the header a clone
1652 * @skb: buffer to check
1654 * Returns true if modifying the header part of the buffer requires
1655 * the data to be copied.
1657 static inline int skb_header_cloned(const struct sk_buff *skb)
1664 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1665 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1666 return dataref != 1;
1669 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1671 might_sleep_if(gfpflags_allow_blocking(pri));
1673 if (skb_header_cloned(skb))
1674 return pskb_expand_head(skb, 0, 0, pri);
1680 * __skb_header_release - release reference to header
1681 * @skb: buffer to operate on
1683 static inline void __skb_header_release(struct sk_buff *skb)
1686 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1691 * skb_shared - is the buffer shared
1692 * @skb: buffer to check
1694 * Returns true if more than one person has a reference to this
1697 static inline int skb_shared(const struct sk_buff *skb)
1699 return refcount_read(&skb->users) != 1;
1703 * skb_share_check - check if buffer is shared and if so clone it
1704 * @skb: buffer to check
1705 * @pri: priority for memory allocation
1707 * If the buffer is shared the buffer is cloned and the old copy
1708 * drops a reference. A new clone with a single reference is returned.
1709 * If the buffer is not shared the original buffer is returned. When
1710 * being called from interrupt status or with spinlocks held pri must
1713 * NULL is returned on a memory allocation failure.
1715 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1717 might_sleep_if(gfpflags_allow_blocking(pri));
1718 if (skb_shared(skb)) {
1719 struct sk_buff *nskb = skb_clone(skb, pri);
1731 * Copy shared buffers into a new sk_buff. We effectively do COW on
1732 * packets to handle cases where we have a local reader and forward
1733 * and a couple of other messy ones. The normal one is tcpdumping
1734 * a packet thats being forwarded.
1738 * skb_unshare - make a copy of a shared buffer
1739 * @skb: buffer to check
1740 * @pri: priority for memory allocation
1742 * If the socket buffer is a clone then this function creates a new
1743 * copy of the data, drops a reference count on the old copy and returns
1744 * the new copy with the reference count at 1. If the buffer is not a clone
1745 * the original buffer is returned. When called with a spinlock held or
1746 * from interrupt state @pri must be %GFP_ATOMIC
1748 * %NULL is returned on a memory allocation failure.
1750 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1753 might_sleep_if(gfpflags_allow_blocking(pri));
1754 if (skb_cloned(skb)) {
1755 struct sk_buff *nskb = skb_copy(skb, pri);
1757 /* Free our shared copy */
1768 * skb_peek - peek at the head of an &sk_buff_head
1769 * @list_: list to peek at
1771 * Peek an &sk_buff. Unlike most other operations you _MUST_
1772 * be careful with this one. A peek leaves the buffer on the
1773 * list and someone else may run off with it. You must hold
1774 * the appropriate locks or have a private queue to do this.
1776 * Returns %NULL for an empty list or a pointer to the head element.
1777 * The reference count is not incremented and the reference is therefore
1778 * volatile. Use with caution.
1780 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1782 struct sk_buff *skb = list_->next;
1784 if (skb == (struct sk_buff *)list_)
1790 * __skb_peek - peek at the head of a non-empty &sk_buff_head
1791 * @list_: list to peek at
1793 * Like skb_peek(), but the caller knows that the list is not empty.
1795 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
1801 * skb_peek_next - peek skb following the given one from a queue
1802 * @skb: skb to start from
1803 * @list_: list to peek at
1805 * Returns %NULL when the end of the list is met or a pointer to the
1806 * next element. The reference count is not incremented and the
1807 * reference is therefore volatile. Use with caution.
1809 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1810 const struct sk_buff_head *list_)
1812 struct sk_buff *next = skb->next;
1814 if (next == (struct sk_buff *)list_)
1820 * skb_peek_tail - peek at the tail of an &sk_buff_head
1821 * @list_: list to peek at
1823 * Peek an &sk_buff. Unlike most other operations you _MUST_
1824 * be careful with this one. A peek leaves the buffer on the
1825 * list and someone else may run off with it. You must hold
1826 * the appropriate locks or have a private queue to do this.
1828 * Returns %NULL for an empty list or a pointer to the tail element.
1829 * The reference count is not incremented and the reference is therefore
1830 * volatile. Use with caution.
1832 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1834 struct sk_buff *skb = READ_ONCE(list_->prev);
1836 if (skb == (struct sk_buff *)list_)
1843 * skb_queue_len - get queue length
1844 * @list_: list to measure
1846 * Return the length of an &sk_buff queue.
1848 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1854 * skb_queue_len_lockless - get queue length
1855 * @list_: list to measure
1857 * Return the length of an &sk_buff queue.
1858 * This variant can be used in lockless contexts.
1860 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
1862 return READ_ONCE(list_->qlen);
1866 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1867 * @list: queue to initialize
1869 * This initializes only the list and queue length aspects of
1870 * an sk_buff_head object. This allows to initialize the list
1871 * aspects of an sk_buff_head without reinitializing things like
1872 * the spinlock. It can also be used for on-stack sk_buff_head
1873 * objects where the spinlock is known to not be used.
1875 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1877 list->prev = list->next = (struct sk_buff *)list;
1882 * This function creates a split out lock class for each invocation;
1883 * this is needed for now since a whole lot of users of the skb-queue
1884 * infrastructure in drivers have different locking usage (in hardirq)
1885 * than the networking core (in softirq only). In the long run either the
1886 * network layer or drivers should need annotation to consolidate the
1887 * main types of usage into 3 classes.
1889 static inline void skb_queue_head_init(struct sk_buff_head *list)
1891 spin_lock_init(&list->lock);
1892 __skb_queue_head_init(list);
1895 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1896 struct lock_class_key *class)
1898 skb_queue_head_init(list);
1899 lockdep_set_class(&list->lock, class);
1903 * Insert an sk_buff on a list.
1905 * The "__skb_xxxx()" functions are the non-atomic ones that
1906 * can only be called with interrupts disabled.
1908 static inline void __skb_insert(struct sk_buff *newsk,
1909 struct sk_buff *prev, struct sk_buff *next,
1910 struct sk_buff_head *list)
1912 /* See skb_queue_empty_lockless() and skb_peek_tail()
1913 * for the opposite READ_ONCE()
1915 WRITE_ONCE(newsk->next, next);
1916 WRITE_ONCE(newsk->prev, prev);
1917 WRITE_ONCE(next->prev, newsk);
1918 WRITE_ONCE(prev->next, newsk);
1919 WRITE_ONCE(list->qlen, list->qlen + 1);
1922 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1923 struct sk_buff *prev,
1924 struct sk_buff *next)
1926 struct sk_buff *first = list->next;
1927 struct sk_buff *last = list->prev;
1929 WRITE_ONCE(first->prev, prev);
1930 WRITE_ONCE(prev->next, first);
1932 WRITE_ONCE(last->next, next);
1933 WRITE_ONCE(next->prev, last);
1937 * skb_queue_splice - join two skb lists, this is designed for stacks
1938 * @list: the new list to add
1939 * @head: the place to add it in the first list
1941 static inline void skb_queue_splice(const struct sk_buff_head *list,
1942 struct sk_buff_head *head)
1944 if (!skb_queue_empty(list)) {
1945 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1946 head->qlen += list->qlen;
1951 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1952 * @list: the new list to add
1953 * @head: the place to add it in the first list
1955 * The list at @list is reinitialised
1957 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1958 struct sk_buff_head *head)
1960 if (!skb_queue_empty(list)) {
1961 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1962 head->qlen += list->qlen;
1963 __skb_queue_head_init(list);
1968 * skb_queue_splice_tail - join two skb lists, each list being a queue
1969 * @list: the new list to add
1970 * @head: the place to add it in the first list
1972 static inline void skb_queue_splice_tail(const 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;
1982 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1983 * @list: the new list to add
1984 * @head: the place to add it in the first list
1986 * Each of the lists is a queue.
1987 * The list at @list is reinitialised
1989 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1990 struct sk_buff_head *head)
1992 if (!skb_queue_empty(list)) {
1993 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1994 head->qlen += list->qlen;
1995 __skb_queue_head_init(list);
2000 * __skb_queue_after - queue a buffer at the list head
2001 * @list: list to use
2002 * @prev: place after this buffer
2003 * @newsk: buffer to queue
2005 * Queue a buffer int the middle of a list. This function takes no locks
2006 * and you must therefore hold required locks before calling it.
2008 * A buffer cannot be placed on two lists at the same time.
2010 static inline void __skb_queue_after(struct sk_buff_head *list,
2011 struct sk_buff *prev,
2012 struct sk_buff *newsk)
2014 __skb_insert(newsk, prev, prev->next, list);
2017 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2018 struct sk_buff_head *list);
2020 static inline void __skb_queue_before(struct sk_buff_head *list,
2021 struct sk_buff *next,
2022 struct sk_buff *newsk)
2024 __skb_insert(newsk, next->prev, next, list);
2028 * __skb_queue_head - queue a buffer at the list head
2029 * @list: list to use
2030 * @newsk: buffer to queue
2032 * Queue a buffer at the start 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_head(struct sk_buff_head *list,
2038 struct sk_buff *newsk)
2040 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2042 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2045 * __skb_queue_tail - queue a buffer at the list tail
2046 * @list: list to use
2047 * @newsk: buffer to queue
2049 * Queue a buffer at the end of a list. This function takes no locks
2050 * and you must therefore hold required locks before calling it.
2052 * A buffer cannot be placed on two lists at the same time.
2054 static inline void __skb_queue_tail(struct sk_buff_head *list,
2055 struct sk_buff *newsk)
2057 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2059 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2062 * remove sk_buff from list. _Must_ be called atomically, and with
2065 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2066 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2068 struct sk_buff *next, *prev;
2070 WRITE_ONCE(list->qlen, list->qlen - 1);
2073 skb->next = skb->prev = NULL;
2074 WRITE_ONCE(next->prev, prev);
2075 WRITE_ONCE(prev->next, next);
2079 * __skb_dequeue - remove from the head of the queue
2080 * @list: list to dequeue from
2082 * Remove the head of the list. This function does not take any locks
2083 * so must be used with appropriate locks held only. The head item is
2084 * returned or %NULL if the list is empty.
2086 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2088 struct sk_buff *skb = skb_peek(list);
2090 __skb_unlink(skb, list);
2093 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2096 * __skb_dequeue_tail - remove from the tail of the queue
2097 * @list: list to dequeue from
2099 * Remove the tail of the list. This function does not take any locks
2100 * so must be used with appropriate locks held only. The tail item is
2101 * returned or %NULL if the list is empty.
2103 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2105 struct sk_buff *skb = skb_peek_tail(list);
2107 __skb_unlink(skb, list);
2110 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2113 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2115 return skb->data_len;
2118 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2120 return skb->len - skb->data_len;
2123 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2125 unsigned int i, len = 0;
2127 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2128 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2132 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2134 return skb_headlen(skb) + __skb_pagelen(skb);
2138 * __skb_fill_page_desc - initialise a paged fragment in an skb
2139 * @skb: buffer containing fragment to be initialised
2140 * @i: paged fragment index to initialise
2141 * @page: the page to use for this fragment
2142 * @off: the offset to the data with @page
2143 * @size: the length of the data
2145 * Initialises the @i'th fragment of @skb to point to &size bytes at
2146 * offset @off within @page.
2148 * Does not take any additional reference on the fragment.
2150 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2151 struct page *page, int off, int size)
2153 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2156 * Propagate page pfmemalloc to the skb if we can. The problem is
2157 * that not all callers have unique ownership of the page but rely
2158 * on page_is_pfmemalloc doing the right thing(tm).
2160 frag->bv_page = page;
2161 frag->bv_offset = off;
2162 skb_frag_size_set(frag, size);
2164 page = compound_head(page);
2165 if (page_is_pfmemalloc(page))
2166 skb->pfmemalloc = true;
2170 * skb_fill_page_desc - initialise a paged fragment in an skb
2171 * @skb: buffer containing fragment to be initialised
2172 * @i: paged fragment index to initialise
2173 * @page: the page to use for this fragment
2174 * @off: the offset to the data with @page
2175 * @size: the length of the data
2177 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2178 * @skb to point to @size bytes at offset @off within @page. In
2179 * addition updates @skb such that @i is the last fragment.
2181 * Does not take any additional reference on the fragment.
2183 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2184 struct page *page, int off, int size)
2186 __skb_fill_page_desc(skb, i, page, off, size);
2187 skb_shinfo(skb)->nr_frags = i + 1;
2190 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2191 int size, unsigned int truesize);
2193 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2194 unsigned int truesize);
2196 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2198 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2199 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2201 return skb->head + skb->tail;
2204 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2206 skb->tail = skb->data - skb->head;
2209 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2211 skb_reset_tail_pointer(skb);
2212 skb->tail += offset;
2215 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2216 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2221 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2223 skb->tail = skb->data;
2226 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2228 skb->tail = skb->data + offset;
2231 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2233 static inline void skb_assert_len(struct sk_buff *skb)
2235 #ifdef CONFIG_DEBUG_NET
2236 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2237 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2238 #endif /* CONFIG_DEBUG_NET */
2242 * Add data to an sk_buff
2244 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2245 void *skb_put(struct sk_buff *skb, unsigned int len);
2246 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2248 void *tmp = skb_tail_pointer(skb);
2249 SKB_LINEAR_ASSERT(skb);
2255 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2257 void *tmp = __skb_put(skb, len);
2259 memset(tmp, 0, len);
2263 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2266 void *tmp = __skb_put(skb, len);
2268 memcpy(tmp, data, len);
2272 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2274 *(u8 *)__skb_put(skb, 1) = val;
2277 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2279 void *tmp = skb_put(skb, len);
2281 memset(tmp, 0, len);
2286 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2289 void *tmp = skb_put(skb, len);
2291 memcpy(tmp, data, len);
2296 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2298 *(u8 *)skb_put(skb, 1) = val;
2301 void *skb_push(struct sk_buff *skb, unsigned int len);
2302 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2309 void *skb_pull(struct sk_buff *skb, unsigned int len);
2310 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2313 BUG_ON(skb->len < skb->data_len);
2314 return skb->data += len;
2317 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2319 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2322 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2324 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2326 if (len > skb_headlen(skb) &&
2327 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2330 return skb->data += len;
2333 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2335 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2338 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2340 if (likely(len <= skb_headlen(skb)))
2342 if (unlikely(len > skb->len))
2344 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2347 void skb_condense(struct sk_buff *skb);
2350 * skb_headroom - bytes at buffer head
2351 * @skb: buffer to check
2353 * Return the number of bytes of free space at the head of an &sk_buff.
2355 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2357 return skb->data - skb->head;
2361 * skb_tailroom - bytes at buffer end
2362 * @skb: buffer to check
2364 * Return the number of bytes of free space at the tail of an sk_buff
2366 static inline int skb_tailroom(const struct sk_buff *skb)
2368 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2372 * skb_availroom - bytes at buffer end
2373 * @skb: buffer to check
2375 * Return the number of bytes of free space at the tail of an sk_buff
2376 * allocated by sk_stream_alloc()
2378 static inline int skb_availroom(const struct sk_buff *skb)
2380 if (skb_is_nonlinear(skb))
2383 return skb->end - skb->tail - skb->reserved_tailroom;
2387 * skb_reserve - adjust headroom
2388 * @skb: buffer to alter
2389 * @len: bytes to move
2391 * Increase the headroom of an empty &sk_buff by reducing the tail
2392 * room. This is only allowed for an empty buffer.
2394 static inline void skb_reserve(struct sk_buff *skb, int len)
2401 * skb_tailroom_reserve - adjust reserved_tailroom
2402 * @skb: buffer to alter
2403 * @mtu: maximum amount of headlen permitted
2404 * @needed_tailroom: minimum amount of reserved_tailroom
2406 * Set reserved_tailroom so that headlen can be as large as possible but
2407 * not larger than mtu and tailroom cannot be smaller than
2409 * The required headroom should already have been reserved before using
2412 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2413 unsigned int needed_tailroom)
2415 SKB_LINEAR_ASSERT(skb);
2416 if (mtu < skb_tailroom(skb) - needed_tailroom)
2417 /* use at most mtu */
2418 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2420 /* use up to all available space */
2421 skb->reserved_tailroom = needed_tailroom;
2424 #define ENCAP_TYPE_ETHER 0
2425 #define ENCAP_TYPE_IPPROTO 1
2427 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2430 skb->inner_protocol = protocol;
2431 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2434 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2437 skb->inner_ipproto = ipproto;
2438 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2441 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2443 skb->inner_mac_header = skb->mac_header;
2444 skb->inner_network_header = skb->network_header;
2445 skb->inner_transport_header = skb->transport_header;
2448 static inline void skb_reset_mac_len(struct sk_buff *skb)
2450 skb->mac_len = skb->network_header - skb->mac_header;
2453 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2456 return skb->head + skb->inner_transport_header;
2459 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2461 return skb_inner_transport_header(skb) - skb->data;
2464 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2466 skb->inner_transport_header = skb->data - skb->head;
2469 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2472 skb_reset_inner_transport_header(skb);
2473 skb->inner_transport_header += offset;
2476 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2478 return skb->head + skb->inner_network_header;
2481 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2483 skb->inner_network_header = skb->data - skb->head;
2486 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2489 skb_reset_inner_network_header(skb);
2490 skb->inner_network_header += offset;
2493 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2495 return skb->head + skb->inner_mac_header;
2498 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2500 skb->inner_mac_header = skb->data - skb->head;
2503 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2506 skb_reset_inner_mac_header(skb);
2507 skb->inner_mac_header += offset;
2509 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2511 return skb->transport_header != (typeof(skb->transport_header))~0U;
2514 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2516 return skb->head + skb->transport_header;
2519 static inline void skb_reset_transport_header(struct sk_buff *skb)
2521 skb->transport_header = skb->data - skb->head;
2524 static inline void skb_set_transport_header(struct sk_buff *skb,
2527 skb_reset_transport_header(skb);
2528 skb->transport_header += offset;
2531 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2533 return skb->head + skb->network_header;
2536 static inline void skb_reset_network_header(struct sk_buff *skb)
2538 skb->network_header = skb->data - skb->head;
2541 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2543 skb_reset_network_header(skb);
2544 skb->network_header += offset;
2547 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2549 return skb->head + skb->mac_header;
2552 static inline int skb_mac_offset(const struct sk_buff *skb)
2554 return skb_mac_header(skb) - skb->data;
2557 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2559 return skb->network_header - skb->mac_header;
2562 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2564 return skb->mac_header != (typeof(skb->mac_header))~0U;
2567 static inline void skb_unset_mac_header(struct sk_buff *skb)
2569 skb->mac_header = (typeof(skb->mac_header))~0U;
2572 static inline void skb_reset_mac_header(struct sk_buff *skb)
2574 skb->mac_header = skb->data - skb->head;
2577 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2579 skb_reset_mac_header(skb);
2580 skb->mac_header += offset;
2583 static inline void skb_pop_mac_header(struct sk_buff *skb)
2585 skb->mac_header = skb->network_header;
2588 static inline void skb_probe_transport_header(struct sk_buff *skb)
2590 struct flow_keys_basic keys;
2592 if (skb_transport_header_was_set(skb))
2595 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2597 skb_set_transport_header(skb, keys.control.thoff);
2600 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2602 if (skb_mac_header_was_set(skb)) {
2603 const unsigned char *old_mac = skb_mac_header(skb);
2605 skb_set_mac_header(skb, -skb->mac_len);
2606 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2610 /* Move the full mac header up to current network_header.
2611 * Leaves skb->data pointing at offset skb->mac_len into the mac_header.
2612 * Must be provided the complete mac header length.
2614 static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len)
2616 if (skb_mac_header_was_set(skb)) {
2617 const unsigned char *old_mac = skb_mac_header(skb);
2619 skb_set_mac_header(skb, -full_mac_len);
2620 memmove(skb_mac_header(skb), old_mac, full_mac_len);
2621 __skb_push(skb, full_mac_len - skb->mac_len);
2625 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2627 return skb->csum_start - skb_headroom(skb);
2630 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2632 return skb->head + skb->csum_start;
2635 static inline int skb_transport_offset(const struct sk_buff *skb)
2637 return skb_transport_header(skb) - skb->data;
2640 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2642 return skb->transport_header - skb->network_header;
2645 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2647 return skb->inner_transport_header - skb->inner_network_header;
2650 static inline int skb_network_offset(const struct sk_buff *skb)
2652 return skb_network_header(skb) - skb->data;
2655 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2657 return skb_inner_network_header(skb) - skb->data;
2660 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2662 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2666 * CPUs often take a performance hit when accessing unaligned memory
2667 * locations. The actual performance hit varies, it can be small if the
2668 * hardware handles it or large if we have to take an exception and fix it
2671 * Since an ethernet header is 14 bytes network drivers often end up with
2672 * the IP header at an unaligned offset. The IP header can be aligned by
2673 * shifting the start of the packet by 2 bytes. Drivers should do this
2676 * skb_reserve(skb, NET_IP_ALIGN);
2678 * The downside to this alignment of the IP header is that the DMA is now
2679 * unaligned. On some architectures the cost of an unaligned DMA is high
2680 * and this cost outweighs the gains made by aligning the IP header.
2682 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2685 #ifndef NET_IP_ALIGN
2686 #define NET_IP_ALIGN 2
2690 * The networking layer reserves some headroom in skb data (via
2691 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2692 * the header has to grow. In the default case, if the header has to grow
2693 * 32 bytes or less we avoid the reallocation.
2695 * Unfortunately this headroom changes the DMA alignment of the resulting
2696 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2697 * on some architectures. An architecture can override this value,
2698 * perhaps setting it to a cacheline in size (since that will maintain
2699 * cacheline alignment of the DMA). It must be a power of 2.
2701 * Various parts of the networking layer expect at least 32 bytes of
2702 * headroom, you should not reduce this.
2704 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2705 * to reduce average number of cache lines per packet.
2706 * get_rps_cpu() for example only access one 64 bytes aligned block :
2707 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2710 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2713 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2715 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2717 if (WARN_ON(skb_is_nonlinear(skb)))
2720 skb_set_tail_pointer(skb, len);
2723 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2725 __skb_set_length(skb, len);
2728 void skb_trim(struct sk_buff *skb, unsigned int len);
2730 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2733 return ___pskb_trim(skb, len);
2734 __skb_trim(skb, len);
2738 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2740 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2744 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2745 * @skb: buffer to alter
2748 * This is identical to pskb_trim except that the caller knows that
2749 * the skb is not cloned so we should never get an error due to out-
2752 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2754 int err = pskb_trim(skb, len);
2758 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2760 unsigned int diff = len - skb->len;
2762 if (skb_tailroom(skb) < diff) {
2763 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2768 __skb_set_length(skb, len);
2773 * skb_orphan - orphan a buffer
2774 * @skb: buffer to orphan
2776 * If a buffer currently has an owner then we call the owner's
2777 * destructor function and make the @skb unowned. The buffer continues
2778 * to exist but is no longer charged to its former owner.
2780 static inline void skb_orphan(struct sk_buff *skb)
2782 if (skb->destructor) {
2783 skb->destructor(skb);
2784 skb->destructor = NULL;
2792 * skb_orphan_frags - orphan the frags contained in a buffer
2793 * @skb: buffer to orphan frags from
2794 * @gfp_mask: allocation mask for replacement pages
2796 * For each frag in the SKB which needs a destructor (i.e. has an
2797 * owner) create a copy of that frag and release the original
2798 * page by calling the destructor.
2800 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2802 if (likely(!skb_zcopy(skb)))
2804 if (!skb_zcopy_is_nouarg(skb) &&
2805 skb_uarg(skb)->callback == sock_zerocopy_callback)
2807 return skb_copy_ubufs(skb, gfp_mask);
2810 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2811 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2813 if (likely(!skb_zcopy(skb)))
2815 return skb_copy_ubufs(skb, gfp_mask);
2819 * __skb_queue_purge - empty a list
2820 * @list: list to empty
2822 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2823 * the list and one reference dropped. This function does not take the
2824 * list lock and the caller must hold the relevant locks to use it.
2826 static inline void __skb_queue_purge(struct sk_buff_head *list)
2828 struct sk_buff *skb;
2829 while ((skb = __skb_dequeue(list)) != NULL)
2832 void skb_queue_purge(struct sk_buff_head *list);
2834 unsigned int skb_rbtree_purge(struct rb_root *root);
2836 void *netdev_alloc_frag(unsigned int fragsz);
2838 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2842 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2843 * @dev: network device to receive on
2844 * @length: length to allocate
2846 * Allocate a new &sk_buff and assign it a usage count of one. The
2847 * buffer has unspecified headroom built in. Users should allocate
2848 * the headroom they think they need without accounting for the
2849 * built in space. The built in space is used for optimisations.
2851 * %NULL is returned if there is no free memory. Although this function
2852 * allocates memory it can be called from an interrupt.
2854 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2855 unsigned int length)
2857 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2860 /* legacy helper around __netdev_alloc_skb() */
2861 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2864 return __netdev_alloc_skb(NULL, length, gfp_mask);
2867 /* legacy helper around netdev_alloc_skb() */
2868 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2870 return netdev_alloc_skb(NULL, length);
2874 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2875 unsigned int length, gfp_t gfp)
2877 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2879 if (NET_IP_ALIGN && skb)
2880 skb_reserve(skb, NET_IP_ALIGN);
2884 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2885 unsigned int length)
2887 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2890 static inline void skb_free_frag(void *addr)
2892 page_frag_free(addr);
2895 void *napi_alloc_frag(unsigned int fragsz);
2896 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2897 unsigned int length, gfp_t gfp_mask);
2898 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2899 unsigned int length)
2901 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2903 void napi_consume_skb(struct sk_buff *skb, int budget);
2905 void __kfree_skb_flush(void);
2906 void __kfree_skb_defer(struct sk_buff *skb);
2909 * __dev_alloc_pages - allocate page for network Rx
2910 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2911 * @order: size of the allocation
2913 * Allocate a new page.
2915 * %NULL is returned if there is no free memory.
2917 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2920 /* This piece of code contains several assumptions.
2921 * 1. This is for device Rx, therefor a cold page is preferred.
2922 * 2. The expectation is the user wants a compound page.
2923 * 3. If requesting a order 0 page it will not be compound
2924 * due to the check to see if order has a value in prep_new_page
2925 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2926 * code in gfp_to_alloc_flags that should be enforcing this.
2928 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2930 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2933 static inline struct page *dev_alloc_pages(unsigned int order)
2935 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2939 * __dev_alloc_page - allocate a page for network Rx
2940 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2942 * Allocate a new page.
2944 * %NULL is returned if there is no free memory.
2946 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2948 return __dev_alloc_pages(gfp_mask, 0);
2951 static inline struct page *dev_alloc_page(void)
2953 return dev_alloc_pages(0);
2957 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2958 * @page: The page that was allocated from skb_alloc_page
2959 * @skb: The skb that may need pfmemalloc set
2961 static inline void skb_propagate_pfmemalloc(struct page *page,
2962 struct sk_buff *skb)
2964 if (page_is_pfmemalloc(page))
2965 skb->pfmemalloc = true;
2969 * skb_frag_off() - Returns the offset of a skb fragment
2970 * @frag: the paged fragment
2972 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
2974 return frag->bv_offset;
2978 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
2979 * @frag: skb fragment
2980 * @delta: value to add
2982 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
2984 frag->bv_offset += delta;
2988 * skb_frag_off_set() - Sets the offset of a skb fragment
2989 * @frag: skb fragment
2990 * @offset: offset of fragment
2992 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
2994 frag->bv_offset = offset;
2998 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
2999 * @fragto: skb fragment where offset is set
3000 * @fragfrom: skb fragment offset is copied from
3002 static inline void skb_frag_off_copy(skb_frag_t *fragto,
3003 const skb_frag_t *fragfrom)
3005 fragto->bv_offset = fragfrom->bv_offset;
3009 * skb_frag_page - retrieve the page referred to by a paged fragment
3010 * @frag: the paged fragment
3012 * Returns the &struct page associated with @frag.
3014 static inline struct page *skb_frag_page(const skb_frag_t *frag)
3016 return frag->bv_page;
3020 * __skb_frag_ref - take an addition reference on a paged fragment.
3021 * @frag: the paged fragment
3023 * Takes an additional reference on the paged fragment @frag.
3025 static inline void __skb_frag_ref(skb_frag_t *frag)
3027 get_page(skb_frag_page(frag));
3031 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
3033 * @f: the fragment offset.
3035 * Takes an additional reference on the @f'th paged fragment of @skb.
3037 static inline void skb_frag_ref(struct sk_buff *skb, int f)
3039 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
3043 * __skb_frag_unref - release a reference on a paged fragment.
3044 * @frag: the paged fragment
3046 * Releases a reference on the paged fragment @frag.
3048 static inline void __skb_frag_unref(skb_frag_t *frag)
3050 put_page(skb_frag_page(frag));
3054 * skb_frag_unref - release a reference on a paged fragment of an skb.
3056 * @f: the fragment offset
3058 * Releases a reference on the @f'th paged fragment of @skb.
3060 static inline void skb_frag_unref(struct sk_buff *skb, int f)
3062 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
3066 * skb_frag_address - gets the address of the data contained in a paged fragment
3067 * @frag: the paged fragment buffer
3069 * Returns the address of the data within @frag. The page must already
3072 static inline void *skb_frag_address(const skb_frag_t *frag)
3074 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3078 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3079 * @frag: the paged fragment buffer
3081 * Returns the address of the data within @frag. Checks that the page
3082 * is mapped and returns %NULL otherwise.
3084 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3086 void *ptr = page_address(skb_frag_page(frag));
3090 return ptr + skb_frag_off(frag);
3094 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3095 * @fragto: skb fragment where page is set
3096 * @fragfrom: skb fragment page is copied from
3098 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3099 const skb_frag_t *fragfrom)
3101 fragto->bv_page = fragfrom->bv_page;
3105 * __skb_frag_set_page - sets the page contained in a paged fragment
3106 * @frag: the paged fragment
3107 * @page: the page to set
3109 * Sets the fragment @frag to contain @page.
3111 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3113 frag->bv_page = page;
3117 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3119 * @f: the fragment offset
3120 * @page: the page to set
3122 * Sets the @f'th fragment of @skb to contain @page.
3124 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3127 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3130 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3133 * skb_frag_dma_map - maps a paged fragment via the DMA API
3134 * @dev: the device to map the fragment to
3135 * @frag: the paged fragment to map
3136 * @offset: the offset within the fragment (starting at the
3137 * fragment's own offset)
3138 * @size: the number of bytes to map
3139 * @dir: the direction of the mapping (``PCI_DMA_*``)
3141 * Maps the page associated with @frag to @device.
3143 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3144 const skb_frag_t *frag,
3145 size_t offset, size_t size,
3146 enum dma_data_direction dir)
3148 return dma_map_page(dev, skb_frag_page(frag),
3149 skb_frag_off(frag) + offset, size, dir);
3152 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3155 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3159 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3162 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3167 * skb_clone_writable - is the header of a clone writable
3168 * @skb: buffer to check
3169 * @len: length up to which to write
3171 * Returns true if modifying the header part of the cloned buffer
3172 * does not requires the data to be copied.
3174 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3176 return !skb_header_cloned(skb) &&
3177 skb_headroom(skb) + len <= skb->hdr_len;
3180 static inline int skb_try_make_writable(struct sk_buff *skb,
3181 unsigned int write_len)
3183 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3184 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3187 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3192 if (headroom > skb_headroom(skb))
3193 delta = headroom - skb_headroom(skb);
3195 if (delta || cloned)
3196 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3202 * skb_cow - copy header of skb when it is required
3203 * @skb: buffer to cow
3204 * @headroom: needed headroom
3206 * If the skb passed lacks sufficient headroom or its data part
3207 * is shared, data is reallocated. If reallocation fails, an error
3208 * is returned and original skb is not changed.
3210 * The result is skb with writable area skb->head...skb->tail
3211 * and at least @headroom of space at head.
3213 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3215 return __skb_cow(skb, headroom, skb_cloned(skb));
3219 * skb_cow_head - skb_cow but only making the head writable
3220 * @skb: buffer to cow
3221 * @headroom: needed headroom
3223 * This function is identical to skb_cow except that we replace the
3224 * skb_cloned check by skb_header_cloned. It should be used when
3225 * you only need to push on some header and do not need to modify
3228 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3230 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3234 * skb_padto - pad an skbuff up to a minimal size
3235 * @skb: buffer to pad
3236 * @len: minimal length
3238 * Pads up a buffer to ensure the trailing bytes exist and are
3239 * blanked. If the buffer already contains sufficient data it
3240 * is untouched. Otherwise it is extended. Returns zero on
3241 * success. The skb is freed on error.
3243 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3245 unsigned int size = skb->len;
3246 if (likely(size >= len))
3248 return skb_pad(skb, len - size);
3252 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3253 * @skb: buffer to pad
3254 * @len: minimal length
3255 * @free_on_error: free buffer on error
3257 * Pads up a buffer to ensure the trailing bytes exist and are
3258 * blanked. If the buffer already contains sufficient data it
3259 * is untouched. Otherwise it is extended. Returns zero on
3260 * success. The skb is freed on error if @free_on_error is true.
3262 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3266 unsigned int size = skb->len;
3268 if (unlikely(size < len)) {
3270 if (__skb_pad(skb, len, free_on_error))
3272 __skb_put(skb, len);
3278 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3279 * @skb: buffer to pad
3280 * @len: minimal length
3282 * Pads up a buffer to ensure the trailing bytes exist and are
3283 * blanked. If the buffer already contains sufficient data it
3284 * is untouched. Otherwise it is extended. Returns zero on
3285 * success. The skb is freed on error.
3287 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3289 return __skb_put_padto(skb, len, true);
3292 static inline int skb_add_data(struct sk_buff *skb,
3293 struct iov_iter *from, int copy)
3295 const int off = skb->len;
3297 if (skb->ip_summed == CHECKSUM_NONE) {
3299 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3301 skb->csum = csum_block_add(skb->csum, csum, off);
3304 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3307 __skb_trim(skb, off);
3311 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3312 const struct page *page, int off)
3317 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3319 return page == skb_frag_page(frag) &&
3320 off == skb_frag_off(frag) + skb_frag_size(frag);
3325 static inline int __skb_linearize(struct sk_buff *skb)
3327 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3331 * skb_linearize - convert paged skb to linear one
3332 * @skb: buffer to linarize
3334 * If there is no free memory -ENOMEM is returned, otherwise zero
3335 * is returned and the old skb data released.
3337 static inline int skb_linearize(struct sk_buff *skb)
3339 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3343 * skb_has_shared_frag - can any frag be overwritten
3344 * @skb: buffer to test
3346 * Return true if the skb has at least one frag that might be modified
3347 * by an external entity (as in vmsplice()/sendfile())
3349 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3351 return skb_is_nonlinear(skb) &&
3352 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3356 * skb_linearize_cow - make sure skb is linear and writable
3357 * @skb: buffer to process
3359 * If there is no free memory -ENOMEM is returned, otherwise zero
3360 * is returned and the old skb data released.
3362 static inline int skb_linearize_cow(struct sk_buff *skb)
3364 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3365 __skb_linearize(skb) : 0;
3368 static __always_inline void
3369 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3372 if (skb->ip_summed == CHECKSUM_COMPLETE)
3373 skb->csum = csum_block_sub(skb->csum,
3374 csum_partial(start, len, 0), off);
3375 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3376 skb_checksum_start_offset(skb) < 0)
3377 skb->ip_summed = CHECKSUM_NONE;
3381 * skb_postpull_rcsum - update checksum for received skb after pull
3382 * @skb: buffer to update
3383 * @start: start of data before pull
3384 * @len: length of data pulled
3386 * After doing a pull on a received packet, you need to call this to
3387 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3388 * CHECKSUM_NONE so that it can be recomputed from scratch.
3390 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3391 const void *start, unsigned int len)
3393 __skb_postpull_rcsum(skb, start, len, 0);
3396 static __always_inline void
3397 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3400 if (skb->ip_summed == CHECKSUM_COMPLETE)
3401 skb->csum = csum_block_add(skb->csum,
3402 csum_partial(start, len, 0), off);
3406 * skb_postpush_rcsum - update checksum for received skb after push
3407 * @skb: buffer to update
3408 * @start: start of data after push
3409 * @len: length of data pushed
3411 * After doing a push on a received packet, you need to call this to
3412 * update the CHECKSUM_COMPLETE checksum.
3414 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3415 const void *start, unsigned int len)
3417 __skb_postpush_rcsum(skb, start, len, 0);
3420 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3423 * skb_push_rcsum - push skb and update receive checksum
3424 * @skb: buffer to update
3425 * @len: length of data pulled
3427 * This function performs an skb_push on the packet and updates
3428 * the CHECKSUM_COMPLETE checksum. It should be used on
3429 * receive path processing instead of skb_push unless you know
3430 * that the checksum difference is zero (e.g., a valid IP header)
3431 * or you are setting ip_summed to CHECKSUM_NONE.
3433 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3436 skb_postpush_rcsum(skb, skb->data, len);
3440 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3442 * pskb_trim_rcsum - trim received skb and update checksum
3443 * @skb: buffer to trim
3446 * This is exactly the same as pskb_trim except that it ensures the
3447 * checksum of received packets are still valid after the operation.
3448 * It can change skb pointers.
3451 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3453 if (likely(len >= skb->len))
3455 return pskb_trim_rcsum_slow(skb, len);
3458 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3460 if (skb->ip_summed == CHECKSUM_COMPLETE)
3461 skb->ip_summed = CHECKSUM_NONE;
3462 __skb_trim(skb, len);
3466 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3468 if (skb->ip_summed == CHECKSUM_COMPLETE)
3469 skb->ip_summed = CHECKSUM_NONE;
3470 return __skb_grow(skb, len);
3473 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3474 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3475 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3476 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3477 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3479 #define skb_queue_walk(queue, skb) \
3480 for (skb = (queue)->next; \
3481 skb != (struct sk_buff *)(queue); \
3484 #define skb_queue_walk_safe(queue, skb, tmp) \
3485 for (skb = (queue)->next, tmp = skb->next; \
3486 skb != (struct sk_buff *)(queue); \
3487 skb = tmp, tmp = skb->next)
3489 #define skb_queue_walk_from(queue, skb) \
3490 for (; skb != (struct sk_buff *)(queue); \
3493 #define skb_rbtree_walk(skb, root) \
3494 for (skb = skb_rb_first(root); skb != NULL; \
3495 skb = skb_rb_next(skb))
3497 #define skb_rbtree_walk_from(skb) \
3498 for (; skb != NULL; \
3499 skb = skb_rb_next(skb))
3501 #define skb_rbtree_walk_from_safe(skb, tmp) \
3502 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3505 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3506 for (tmp = skb->next; \
3507 skb != (struct sk_buff *)(queue); \
3508 skb = tmp, tmp = skb->next)
3510 #define skb_queue_reverse_walk(queue, skb) \
3511 for (skb = (queue)->prev; \
3512 skb != (struct sk_buff *)(queue); \
3515 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3516 for (skb = (queue)->prev, tmp = skb->prev; \
3517 skb != (struct sk_buff *)(queue); \
3518 skb = tmp, tmp = skb->prev)
3520 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3521 for (tmp = skb->prev; \
3522 skb != (struct sk_buff *)(queue); \
3523 skb = tmp, tmp = skb->prev)
3525 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3527 return skb_shinfo(skb)->frag_list != NULL;
3530 static inline void skb_frag_list_init(struct sk_buff *skb)
3532 skb_shinfo(skb)->frag_list = NULL;
3535 #define skb_walk_frags(skb, iter) \
3536 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3539 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3540 int *err, long *timeo_p,
3541 const struct sk_buff *skb);
3542 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3543 struct sk_buff_head *queue,
3546 struct sk_buff **last);
3547 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
3548 struct sk_buff_head *queue,
3549 unsigned int flags, int *off, int *err,
3550 struct sk_buff **last);
3551 struct sk_buff *__skb_recv_datagram(struct sock *sk,
3552 struct sk_buff_head *sk_queue,
3553 unsigned int flags, int *off, int *err);
3554 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3556 __poll_t datagram_poll(struct file *file, struct socket *sock,
3557 struct poll_table_struct *wait);
3558 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3559 struct iov_iter *to, int size);
3560 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3561 struct msghdr *msg, int size)
3563 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3565 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3566 struct msghdr *msg);
3567 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3568 struct iov_iter *to, int len,
3569 struct ahash_request *hash);
3570 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3571 struct iov_iter *from, int len);
3572 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3573 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3574 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3575 static inline void skb_free_datagram_locked(struct sock *sk,
3576 struct sk_buff *skb)
3578 __skb_free_datagram_locked(sk, skb, 0);
3580 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3581 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3582 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3583 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3585 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3586 struct pipe_inode_info *pipe, unsigned int len,
3587 unsigned int flags);
3588 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3590 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3591 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3592 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3594 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3595 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3596 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3597 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3598 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3599 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3600 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
3601 unsigned int offset);
3602 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3603 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3604 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3605 int skb_vlan_pop(struct sk_buff *skb);
3606 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3607 int skb_eth_pop(struct sk_buff *skb);
3608 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
3609 const unsigned char *src);
3610 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
3611 int mac_len, bool ethernet);
3612 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
3614 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3615 int skb_mpls_dec_ttl(struct sk_buff *skb);
3616 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3619 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3621 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3624 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3626 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3629 struct skb_checksum_ops {
3630 __wsum (*update)(const void *mem, int len, __wsum wsum);
3631 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3634 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3636 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3637 __wsum csum, const struct skb_checksum_ops *ops);
3638 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3641 static inline void * __must_check
3642 __skb_header_pointer(const struct sk_buff *skb, int offset,
3643 int len, void *data, int hlen, void *buffer)
3645 if (hlen - offset >= len)
3646 return data + offset;
3649 skb_copy_bits(skb, offset, buffer, len) < 0)
3655 static inline void * __must_check
3656 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3658 return __skb_header_pointer(skb, offset, len, skb->data,
3659 skb_headlen(skb), buffer);
3663 * skb_needs_linearize - check if we need to linearize a given skb
3664 * depending on the given device features.
3665 * @skb: socket buffer to check
3666 * @features: net device features
3668 * Returns true if either:
3669 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3670 * 2. skb is fragmented and the device does not support SG.
3672 static inline bool skb_needs_linearize(struct sk_buff *skb,
3673 netdev_features_t features)
3675 return skb_is_nonlinear(skb) &&
3676 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3677 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3680 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3682 const unsigned int len)
3684 memcpy(to, skb->data, len);
3687 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3688 const int offset, void *to,
3689 const unsigned int len)
3691 memcpy(to, skb->data + offset, len);
3694 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3696 const unsigned int len)
3698 memcpy(skb->data, from, len);
3701 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3704 const unsigned int len)
3706 memcpy(skb->data + offset, from, len);
3709 void skb_init(void);
3711 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3717 * skb_get_timestamp - get timestamp from a skb
3718 * @skb: skb to get stamp from
3719 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
3721 * Timestamps are stored in the skb as offsets to a base timestamp.
3722 * This function converts the offset back to a struct timeval and stores
3725 static inline void skb_get_timestamp(const struct sk_buff *skb,
3726 struct __kernel_old_timeval *stamp)
3728 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
3731 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
3732 struct __kernel_sock_timeval *stamp)
3734 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3736 stamp->tv_sec = ts.tv_sec;
3737 stamp->tv_usec = ts.tv_nsec / 1000;
3740 static inline void skb_get_timestampns(const struct sk_buff *skb,
3741 struct __kernel_old_timespec *stamp)
3743 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3745 stamp->tv_sec = ts.tv_sec;
3746 stamp->tv_nsec = ts.tv_nsec;
3749 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
3750 struct __kernel_timespec *stamp)
3752 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3754 stamp->tv_sec = ts.tv_sec;
3755 stamp->tv_nsec = ts.tv_nsec;
3758 static inline void __net_timestamp(struct sk_buff *skb)
3760 skb->tstamp = ktime_get_real();
3763 static inline ktime_t net_timedelta(ktime_t t)
3765 return ktime_sub(ktime_get_real(), t);
3768 static inline ktime_t net_invalid_timestamp(void)
3773 static inline u8 skb_metadata_len(const struct sk_buff *skb)
3775 return skb_shinfo(skb)->meta_len;
3778 static inline void *skb_metadata_end(const struct sk_buff *skb)
3780 return skb_mac_header(skb);
3783 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3784 const struct sk_buff *skb_b,
3787 const void *a = skb_metadata_end(skb_a);
3788 const void *b = skb_metadata_end(skb_b);
3789 /* Using more efficient varaiant than plain call to memcmp(). */
3790 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3794 #define __it(x, op) (x -= sizeof(u##op))
3795 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3796 case 32: diffs |= __it_diff(a, b, 64);
3798 case 24: diffs |= __it_diff(a, b, 64);
3800 case 16: diffs |= __it_diff(a, b, 64);
3802 case 8: diffs |= __it_diff(a, b, 64);
3804 case 28: diffs |= __it_diff(a, b, 64);
3806 case 20: diffs |= __it_diff(a, b, 64);
3808 case 12: diffs |= __it_diff(a, b, 64);
3810 case 4: diffs |= __it_diff(a, b, 32);
3815 return memcmp(a - meta_len, b - meta_len, meta_len);
3819 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3820 const struct sk_buff *skb_b)
3822 u8 len_a = skb_metadata_len(skb_a);
3823 u8 len_b = skb_metadata_len(skb_b);
3825 if (!(len_a | len_b))
3828 return len_a != len_b ?
3829 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3832 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3834 skb_shinfo(skb)->meta_len = meta_len;
3837 static inline void skb_metadata_clear(struct sk_buff *skb)
3839 skb_metadata_set(skb, 0);
3842 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3844 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3846 void skb_clone_tx_timestamp(struct sk_buff *skb);
3847 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3849 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3851 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3855 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3860 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3863 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3865 * PHY drivers may accept clones of transmitted packets for
3866 * timestamping via their phy_driver.txtstamp method. These drivers
3867 * must call this function to return the skb back to the stack with a
3870 * @skb: clone of the original outgoing packet
3871 * @hwtstamps: hardware time stamps
3874 void skb_complete_tx_timestamp(struct sk_buff *skb,
3875 struct skb_shared_hwtstamps *hwtstamps);
3877 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3878 struct skb_shared_hwtstamps *hwtstamps,
3879 struct sock *sk, int tstype);
3882 * skb_tstamp_tx - queue clone of skb with send time stamps
3883 * @orig_skb: the original outgoing packet
3884 * @hwtstamps: hardware time stamps, may be NULL if not available
3886 * If the skb has a socket associated, then this function clones the
3887 * skb (thus sharing the actual data and optional structures), stores
3888 * the optional hardware time stamping information (if non NULL) or
3889 * generates a software time stamp (otherwise), then queues the clone
3890 * to the error queue of the socket. Errors are silently ignored.
3892 void skb_tstamp_tx(struct sk_buff *orig_skb,
3893 struct skb_shared_hwtstamps *hwtstamps);
3896 * skb_tx_timestamp() - Driver hook for transmit timestamping
3898 * Ethernet MAC Drivers should call this function in their hard_xmit()
3899 * function immediately before giving the sk_buff to the MAC hardware.
3901 * Specifically, one should make absolutely sure that this function is
3902 * called before TX completion of this packet can trigger. Otherwise
3903 * the packet could potentially already be freed.
3905 * @skb: A socket buffer.
3907 static inline void skb_tx_timestamp(struct sk_buff *skb)
3909 skb_clone_tx_timestamp(skb);
3910 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3911 skb_tstamp_tx(skb, NULL);
3915 * skb_complete_wifi_ack - deliver skb with wifi status
3917 * @skb: the original outgoing packet
3918 * @acked: ack status
3921 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3923 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3924 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3926 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3928 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3930 (skb->ip_summed == CHECKSUM_PARTIAL &&
3931 skb_checksum_start_offset(skb) >= 0));
3935 * skb_checksum_complete - Calculate checksum of an entire packet
3936 * @skb: packet to process
3938 * This function calculates the checksum over the entire packet plus
3939 * the value of skb->csum. The latter can be used to supply the
3940 * checksum of a pseudo header as used by TCP/UDP. It returns the
3943 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3944 * this function can be used to verify that checksum on received
3945 * packets. In that case the function should return zero if the
3946 * checksum is correct. In particular, this function will return zero
3947 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3948 * hardware has already verified the correctness of the checksum.
3950 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3952 return skb_csum_unnecessary(skb) ?
3953 0 : __skb_checksum_complete(skb);
3956 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3958 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3959 if (skb->csum_level == 0)
3960 skb->ip_summed = CHECKSUM_NONE;
3966 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3968 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3969 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3971 } else if (skb->ip_summed == CHECKSUM_NONE) {
3972 skb->ip_summed = CHECKSUM_UNNECESSARY;
3973 skb->csum_level = 0;
3977 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
3979 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3980 skb->ip_summed = CHECKSUM_NONE;
3981 skb->csum_level = 0;
3985 /* Check if we need to perform checksum complete validation.
3987 * Returns true if checksum complete is needed, false otherwise
3988 * (either checksum is unnecessary or zero checksum is allowed).
3990 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3994 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3995 skb->csum_valid = 1;
3996 __skb_decr_checksum_unnecessary(skb);
4003 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4006 #define CHECKSUM_BREAK 76
4008 /* Unset checksum-complete
4010 * Unset checksum complete can be done when packet is being modified
4011 * (uncompressed for instance) and checksum-complete value is
4014 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4016 if (skb->ip_summed == CHECKSUM_COMPLETE)
4017 skb->ip_summed = CHECKSUM_NONE;
4020 /* Validate (init) checksum based on checksum complete.
4023 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4024 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4025 * checksum is stored in skb->csum for use in __skb_checksum_complete
4026 * non-zero: value of invalid checksum
4029 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4033 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4034 if (!csum_fold(csum_add(psum, skb->csum))) {
4035 skb->csum_valid = 1;
4042 if (complete || skb->len <= CHECKSUM_BREAK) {
4045 csum = __skb_checksum_complete(skb);
4046 skb->csum_valid = !csum;
4053 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4058 /* Perform checksum validate (init). Note that this is a macro since we only
4059 * want to calculate the pseudo header which is an input function if necessary.
4060 * First we try to validate without any computation (checksum unnecessary) and
4061 * then calculate based on checksum complete calling the function to compute
4065 * 0: checksum is validated or try to in skb_checksum_complete
4066 * non-zero: value of invalid checksum
4068 #define __skb_checksum_validate(skb, proto, complete, \
4069 zero_okay, check, compute_pseudo) \
4071 __sum16 __ret = 0; \
4072 skb->csum_valid = 0; \
4073 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4074 __ret = __skb_checksum_validate_complete(skb, \
4075 complete, compute_pseudo(skb, proto)); \
4079 #define skb_checksum_init(skb, proto, compute_pseudo) \
4080 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4082 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4083 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4085 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4086 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4088 #define skb_checksum_validate_zero_check(skb, proto, check, \
4090 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4092 #define skb_checksum_simple_validate(skb) \
4093 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4095 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4097 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4100 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4102 skb->csum = ~pseudo;
4103 skb->ip_summed = CHECKSUM_COMPLETE;
4106 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4108 if (__skb_checksum_convert_check(skb)) \
4109 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4112 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4113 u16 start, u16 offset)
4115 skb->ip_summed = CHECKSUM_PARTIAL;
4116 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4117 skb->csum_offset = offset - start;
4120 /* Update skbuf and packet to reflect the remote checksum offload operation.
4121 * When called, ptr indicates the starting point for skb->csum when
4122 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4123 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4125 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4126 int start, int offset, bool nopartial)
4131 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4135 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4136 __skb_checksum_complete(skb);
4137 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4140 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4142 /* Adjust skb->csum since we changed the packet */
4143 skb->csum = csum_add(skb->csum, delta);
4146 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4148 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4149 return (void *)(skb->_nfct & NFCT_PTRMASK);
4155 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4157 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4164 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4166 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4171 #ifdef CONFIG_SKB_EXTENSIONS
4173 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4179 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4182 #if IS_ENABLED(CONFIG_MPTCP)
4185 SKB_EXT_NUM, /* must be last */
4189 * struct skb_ext - sk_buff extensions
4190 * @refcnt: 1 on allocation, deallocated on 0
4191 * @offset: offset to add to @data to obtain extension address
4192 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4193 * @data: start of extension data, variable sized
4195 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4196 * to use 'u8' types while allowing up to 2kb worth of extension data.
4200 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4201 u8 chunks; /* same */
4202 char data[] __aligned(8);
4205 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4206 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4207 struct skb_ext *ext);
4208 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4209 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4210 void __skb_ext_put(struct skb_ext *ext);
4212 static inline void skb_ext_put(struct sk_buff *skb)
4214 if (skb->active_extensions)
4215 __skb_ext_put(skb->extensions);
4218 static inline void __skb_ext_copy(struct sk_buff *dst,
4219 const struct sk_buff *src)
4221 dst->active_extensions = src->active_extensions;
4223 if (src->active_extensions) {
4224 struct skb_ext *ext = src->extensions;
4226 refcount_inc(&ext->refcnt);
4227 dst->extensions = ext;
4231 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4234 __skb_ext_copy(dst, src);
4237 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4239 return !!ext->offset[i];
4242 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4244 return skb->active_extensions & (1 << id);
4247 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4249 if (skb_ext_exist(skb, id))
4250 __skb_ext_del(skb, id);
4253 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4255 if (skb_ext_exist(skb, id)) {
4256 struct skb_ext *ext = skb->extensions;
4258 return (void *)ext + (ext->offset[id] << 3);
4264 static inline void skb_ext_reset(struct sk_buff *skb)
4266 if (unlikely(skb->active_extensions)) {
4267 __skb_ext_put(skb->extensions);
4268 skb->active_extensions = 0;
4272 static inline bool skb_has_extensions(struct sk_buff *skb)
4274 return unlikely(skb->active_extensions);
4277 static inline void skb_ext_put(struct sk_buff *skb) {}
4278 static inline void skb_ext_reset(struct sk_buff *skb) {}
4279 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4280 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4281 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4282 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4283 #endif /* CONFIG_SKB_EXTENSIONS */
4285 static inline void nf_reset_ct(struct sk_buff *skb)
4287 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4288 nf_conntrack_put(skb_nfct(skb));
4293 static inline void nf_reset_trace(struct sk_buff *skb)
4295 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4300 static inline void ipvs_reset(struct sk_buff *skb)
4302 #if IS_ENABLED(CONFIG_IP_VS)
4303 skb->ipvs_property = 0;
4307 /* Note: This doesn't put any conntrack info in dst. */
4308 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4311 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4312 dst->_nfct = src->_nfct;
4313 nf_conntrack_get(skb_nfct(src));
4315 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4317 dst->nf_trace = src->nf_trace;
4321 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4323 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4324 nf_conntrack_put(skb_nfct(dst));
4326 __nf_copy(dst, src, true);
4329 #ifdef CONFIG_NETWORK_SECMARK
4330 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4332 to->secmark = from->secmark;
4335 static inline void skb_init_secmark(struct sk_buff *skb)
4340 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4343 static inline void skb_init_secmark(struct sk_buff *skb)
4347 static inline int secpath_exists(const struct sk_buff *skb)
4350 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4356 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4358 return !skb->destructor &&
4359 !secpath_exists(skb) &&
4361 !skb->_skb_refdst &&
4362 !skb_has_frag_list(skb);
4365 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4367 skb->queue_mapping = queue_mapping;
4370 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4372 return skb->queue_mapping;
4375 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4377 to->queue_mapping = from->queue_mapping;
4380 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4382 skb->queue_mapping = rx_queue + 1;
4385 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4387 return skb->queue_mapping - 1;
4390 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4392 return skb->queue_mapping != 0;
4395 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4397 skb->dst_pending_confirm = val;
4400 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4402 return skb->dst_pending_confirm != 0;
4405 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4408 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4414 /* Keeps track of mac header offset relative to skb->head.
4415 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4416 * For non-tunnel skb it points to skb_mac_header() and for
4417 * tunnel skb it points to outer mac header.
4418 * Keeps track of level of encapsulation of network headers.
4429 #define SKB_GSO_CB_OFFSET 32
4430 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET))
4432 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4434 return (skb_mac_header(inner_skb) - inner_skb->head) -
4435 SKB_GSO_CB(inner_skb)->mac_offset;
4438 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4440 int new_headroom, headroom;
4443 headroom = skb_headroom(skb);
4444 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4448 new_headroom = skb_headroom(skb);
4449 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4453 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4455 /* Do not update partial checksums if remote checksum is enabled. */
4456 if (skb->remcsum_offload)
4459 SKB_GSO_CB(skb)->csum = res;
4460 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4463 /* Compute the checksum for a gso segment. First compute the checksum value
4464 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4465 * then add in skb->csum (checksum from csum_start to end of packet).
4466 * skb->csum and csum_start are then updated to reflect the checksum of the
4467 * resultant packet starting from the transport header-- the resultant checksum
4468 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4471 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4473 unsigned char *csum_start = skb_transport_header(skb);
4474 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4475 __wsum partial = SKB_GSO_CB(skb)->csum;
4477 SKB_GSO_CB(skb)->csum = res;
4478 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4480 return csum_fold(csum_partial(csum_start, plen, partial));
4483 static inline bool skb_is_gso(const struct sk_buff *skb)
4485 return skb_shinfo(skb)->gso_size;
4488 /* Note: Should be called only if skb_is_gso(skb) is true */
4489 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4491 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4494 /* Note: Should be called only if skb_is_gso(skb) is true */
4495 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4497 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4500 /* Note: Should be called only if skb_is_gso(skb) is true */
4501 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4503 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4506 static inline void skb_gso_reset(struct sk_buff *skb)
4508 skb_shinfo(skb)->gso_size = 0;
4509 skb_shinfo(skb)->gso_segs = 0;
4510 skb_shinfo(skb)->gso_type = 0;
4513 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4516 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4518 shinfo->gso_size += increment;
4521 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4524 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4526 shinfo->gso_size -= decrement;
4529 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4531 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4533 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4534 * wanted then gso_type will be set. */
4535 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4537 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4538 unlikely(shinfo->gso_type == 0)) {
4539 __skb_warn_lro_forwarding(skb);
4545 static inline void skb_forward_csum(struct sk_buff *skb)
4547 /* Unfortunately we don't support this one. Any brave souls? */
4548 if (skb->ip_summed == CHECKSUM_COMPLETE)
4549 skb->ip_summed = CHECKSUM_NONE;
4553 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4554 * @skb: skb to check
4556 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4557 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4558 * use this helper, to document places where we make this assertion.
4560 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4563 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4567 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4569 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4570 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4571 unsigned int transport_len,
4572 __sum16(*skb_chkf)(struct sk_buff *skb));
4575 * skb_head_is_locked - Determine if the skb->head is locked down
4576 * @skb: skb to check
4578 * The head on skbs build around a head frag can be removed if they are
4579 * not cloned. This function returns true if the skb head is locked down
4580 * due to either being allocated via kmalloc, or by being a clone with
4581 * multiple references to the head.
4583 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4585 return !skb->head_frag || skb_cloned(skb);
4588 /* Local Checksum Offload.
4589 * Compute outer checksum based on the assumption that the
4590 * inner checksum will be offloaded later.
4591 * See Documentation/networking/checksum-offloads.rst for
4592 * explanation of how this works.
4593 * Fill in outer checksum adjustment (e.g. with sum of outer
4594 * pseudo-header) before calling.
4595 * Also ensure that inner checksum is in linear data area.
4597 static inline __wsum lco_csum(struct sk_buff *skb)
4599 unsigned char *csum_start = skb_checksum_start(skb);
4600 unsigned char *l4_hdr = skb_transport_header(skb);
4603 /* Start with complement of inner checksum adjustment */
4604 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4607 /* Add in checksum of our headers (incl. outer checksum
4608 * adjustment filled in by caller) and return result.
4610 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4613 static inline bool skb_is_redirected(const struct sk_buff *skb)
4615 #ifdef CONFIG_NET_REDIRECT
4616 return skb->redirected;
4622 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
4624 #ifdef CONFIG_NET_REDIRECT
4625 skb->redirected = 1;
4626 skb->from_ingress = from_ingress;
4627 if (skb->from_ingress)
4632 static inline void skb_reset_redirect(struct sk_buff *skb)
4634 #ifdef CONFIG_NET_REDIRECT
4635 skb->redirected = 0;
4639 static inline bool skb_csum_is_sctp(struct sk_buff *skb)
4641 return skb->csum_not_inet;
4644 static inline void skb_set_kcov_handle(struct sk_buff *skb,
4645 const u64 kcov_handle)
4648 skb->kcov_handle = kcov_handle;
4652 static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
4655 return skb->kcov_handle;
4661 #endif /* __KERNEL__ */
4662 #endif /* _LINUX_SKBUFF_H */