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 <net/checksum.h>
27 #include <linux/rcupdate.h>
28 #include <linux/hrtimer.h>
29 #include <linux/dma-mapping.h>
30 #include <linux/netdev_features.h>
31 #include <linux/sched.h>
32 #include <linux/sched/clock.h>
33 #include <net/flow_dissector.h>
34 #include <linux/splice.h>
35 #include <linux/in6.h>
36 #include <linux/if_packet.h>
37 #include <linux/llist.h>
39 #include <net/page_pool.h>
40 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
41 #include <linux/netfilter/nf_conntrack_common.h>
43 #include <net/net_debug.h>
44 #include <net/dropreason.h>
49 * The interface for checksum offload between the stack and networking drivers
52 * IP checksum related features
53 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
55 * Drivers advertise checksum offload capabilities in the features of a device.
56 * From the stack's point of view these are capabilities offered by the driver.
57 * A driver typically only advertises features that it is capable of offloading
60 * .. flat-table:: Checksum related device features
63 * * - %NETIF_F_HW_CSUM
64 * - The driver (or its device) is able to compute one
65 * IP (one's complement) checksum for any combination
66 * of protocols or protocol layering. The checksum is
67 * computed and set in a packet per the CHECKSUM_PARTIAL
68 * interface (see below).
70 * * - %NETIF_F_IP_CSUM
71 * - Driver (device) is only able to checksum plain
72 * TCP or UDP packets over IPv4. These are specifically
73 * unencapsulated packets of the form IPv4|TCP or
74 * IPv4|UDP where the Protocol field in the IPv4 header
75 * is TCP or UDP. The IPv4 header may contain IP options.
76 * This feature cannot be set in features for a device
77 * with NETIF_F_HW_CSUM also set. This feature is being
78 * DEPRECATED (see below).
80 * * - %NETIF_F_IPV6_CSUM
81 * - Driver (device) is only able to checksum plain
82 * TCP or UDP packets over IPv6. These are specifically
83 * unencapsulated packets of the form IPv6|TCP or
84 * IPv6|UDP where the Next Header field in the IPv6
85 * header is either TCP or UDP. IPv6 extension headers
86 * are not supported with this feature. This feature
87 * cannot be set in features for a device with
88 * NETIF_F_HW_CSUM also set. This feature is being
89 * DEPRECATED (see below).
92 * - Driver (device) performs receive checksum offload.
93 * This flag is only used to disable the RX checksum
94 * feature for a device. The stack will accept receive
95 * checksum indication in packets received on a device
96 * regardless of whether NETIF_F_RXCSUM is set.
98 * Checksumming of received packets by device
99 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
101 * Indication of checksum verification is set in &sk_buff.ip_summed.
102 * Possible values are:
106 * Device did not checksum this packet e.g. due to lack of capabilities.
107 * The packet contains full (though not verified) checksum in packet but
108 * not in skb->csum. Thus, skb->csum is undefined in this case.
110 * - %CHECKSUM_UNNECESSARY
112 * The hardware you're dealing with doesn't calculate the full checksum
113 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums
114 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY
115 * if their checksums are okay. &sk_buff.csum is still undefined in this case
116 * though. A driver or device must never modify the checksum field in the
117 * packet even if checksum is verified.
119 * %CHECKSUM_UNNECESSARY is applicable to following protocols:
121 * - TCP: IPv6 and IPv4.
122 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
123 * zero UDP checksum for either IPv4 or IPv6, the networking stack
124 * may perform further validation in this case.
125 * - GRE: only if the checksum is present in the header.
126 * - SCTP: indicates the CRC in SCTP header has been validated.
127 * - FCOE: indicates the CRC in FC frame has been validated.
129 * &sk_buff.csum_level indicates the number of consecutive checksums found in
130 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY.
131 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
132 * and a device is able to verify the checksums for UDP (possibly zero),
133 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to
134 * two. If the device were only able to verify the UDP checksum and not
135 * GRE, either because it doesn't support GRE checksum or because GRE
136 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
137 * not considered in this case).
139 * - %CHECKSUM_COMPLETE
141 * This is the most generic way. The device supplied checksum of the _whole_
142 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the
143 * hardware doesn't need to parse L3/L4 headers to implement this.
147 * - Even if device supports only some protocols, but is able to produce
148 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
149 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
151 * - %CHECKSUM_PARTIAL
153 * A checksum is set up to be offloaded to a device as described in the
154 * output description for CHECKSUM_PARTIAL. This may occur on a packet
155 * received directly from another Linux OS, e.g., a virtualized Linux kernel
156 * on the same host, or it may be set in the input path in GRO or remote
157 * checksum offload. For the purposes of checksum verification, the checksum
158 * referred to by skb->csum_start + skb->csum_offset and any preceding
159 * checksums in the packet are considered verified. Any checksums in the
160 * packet that are after the checksum being offloaded are not considered to
163 * Checksumming on transmit for non-GSO
164 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
166 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet.
169 * - %CHECKSUM_PARTIAL
171 * The driver is required to checksum the packet as seen by hard_start_xmit()
172 * from &sk_buff.csum_start up to the end, and to record/write the checksum at
173 * offset &sk_buff.csum_start + &sk_buff.csum_offset.
174 * A driver may verify that the
175 * csum_start and csum_offset values are valid values given the length and
176 * offset of the packet, but it should not attempt to validate that the
177 * checksum refers to a legitimate transport layer checksum -- it is the
178 * purview of the stack to validate that csum_start and csum_offset are set
181 * When the stack requests checksum offload for a packet, the driver MUST
182 * ensure that the checksum is set correctly. A driver can either offload the
183 * checksum calculation to the device, or call skb_checksum_help (in the case
184 * that the device does not support offload for a particular checksum).
186 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of
187 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate
188 * checksum offload capability.
189 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based
190 * on network device checksumming capabilities: if a packet does not match
191 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of
192 * &sk_buff.csum_not_inet, see :ref:`crc`)
193 * is called to resolve the checksum.
197 * The skb was already checksummed by the protocol, or a checksum is not
200 * - %CHECKSUM_UNNECESSARY
202 * This has the same meaning as CHECKSUM_NONE for checksum offload on
205 * - %CHECKSUM_COMPLETE
207 * Not used in checksum output. If a driver observes a packet with this value
208 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set.
212 * Non-IP checksum (CRC) offloads
213 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
218 * * - %NETIF_F_SCTP_CRC
219 * - This feature indicates that a device is capable of
220 * offloading the SCTP CRC in a packet. To perform this offload the stack
221 * will set csum_start and csum_offset accordingly, set ip_summed to
222 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication
223 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c.
224 * A driver that supports both IP checksum offload and SCTP CRC32c offload
225 * must verify which offload is configured for a packet by testing the
226 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to
227 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
229 * * - %NETIF_F_FCOE_CRC
230 * - This feature indicates that a device is capable of offloading the FCOE
231 * CRC in a packet. To perform this offload the stack will set ip_summed
232 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset
233 * accordingly. Note that there is no indication in the skbuff that the
234 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports
235 * both IP checksum offload and FCOE CRC offload must verify which offload
236 * is configured for a packet, presumably by inspecting packet headers.
238 * Checksumming on output with GSO
239 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
241 * In the case of a GSO packet (skb_is_gso() is true), checksum offload
242 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
243 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as
244 * part of the GSO operation is implied. If a checksum is being offloaded
245 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and
246 * csum_offset are set to refer to the outermost checksum being offloaded
247 * (two offloaded checksums are possible with UDP encapsulation).
250 /* Don't change this without changing skb_csum_unnecessary! */
251 #define CHECKSUM_NONE 0
252 #define CHECKSUM_UNNECESSARY 1
253 #define CHECKSUM_COMPLETE 2
254 #define CHECKSUM_PARTIAL 3
256 /* Maximum value in skb->csum_level */
257 #define SKB_MAX_CSUM_LEVEL 3
259 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
260 #define SKB_WITH_OVERHEAD(X) \
261 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
262 #define SKB_MAX_ORDER(X, ORDER) \
263 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
264 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
265 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
267 /* return minimum truesize of one skb containing X bytes of data */
268 #define SKB_TRUESIZE(X) ((X) + \
269 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
270 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
272 struct ahash_request;
275 struct pipe_inode_info;
283 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
284 struct nf_bridge_info {
286 BRNF_PROTO_UNCHANGED,
294 struct net_device *physindev;
296 /* always valid & non-NULL from FORWARD on, for physdev match */
297 struct net_device *physoutdev;
299 /* prerouting: detect dnat in orig/reply direction */
301 struct in6_addr ipv6_daddr;
303 /* after prerouting + nat detected: store original source
304 * mac since neigh resolution overwrites it, only used while
305 * skb is out in neigh layer.
307 char neigh_header[8];
312 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
313 /* Chain in tc_skb_ext will be used to share the tc chain with
314 * ovs recirc_id. It will be set to the current chain by tc
315 * and read by ovs to recirc_id.
327 struct sk_buff_head {
328 /* These two members must be first to match sk_buff. */
329 struct_group_tagged(sk_buff_list, list,
330 struct sk_buff *next;
331 struct sk_buff *prev;
340 /* To allow 64K frame to be packed as single skb without frag_list we
341 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
342 * buffers which do not start on a page boundary.
344 * Since GRO uses frags we allocate at least 16 regardless of page
347 #if (65536/PAGE_SIZE + 1) < 16
348 #define MAX_SKB_FRAGS 16UL
350 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
352 extern int sysctl_max_skb_frags;
354 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
355 * segment using its current segmentation instead.
357 #define GSO_BY_FRAGS 0xFFFF
359 typedef struct bio_vec skb_frag_t;
362 * skb_frag_size() - Returns the size of a skb fragment
363 * @frag: skb fragment
365 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
371 * skb_frag_size_set() - Sets the size of a skb fragment
372 * @frag: skb fragment
373 * @size: size of fragment
375 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
381 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
382 * @frag: skb fragment
383 * @delta: value to add
385 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
387 frag->bv_len += delta;
391 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
392 * @frag: skb fragment
393 * @delta: value to subtract
395 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
397 frag->bv_len -= delta;
401 * skb_frag_must_loop - Test if %p is a high memory page
402 * @p: fragment's page
404 static inline bool skb_frag_must_loop(struct page *p)
406 #if defined(CONFIG_HIGHMEM)
407 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
414 * skb_frag_foreach_page - loop over pages in a fragment
416 * @f: skb frag to operate on
417 * @f_off: offset from start of f->bv_page
418 * @f_len: length from f_off to loop over
419 * @p: (temp var) current page
420 * @p_off: (temp var) offset from start of current page,
421 * non-zero only on first page.
422 * @p_len: (temp var) length in current page,
423 * < PAGE_SIZE only on first and last page.
424 * @copied: (temp var) length so far, excluding current p_len.
426 * A fragment can hold a compound page, in which case per-page
427 * operations, notably kmap_atomic, must be called for each
430 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
431 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
432 p_off = (f_off) & (PAGE_SIZE - 1), \
433 p_len = skb_frag_must_loop(p) ? \
434 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
437 copied += p_len, p++, p_off = 0, \
438 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
440 #define HAVE_HW_TIME_STAMP
443 * struct skb_shared_hwtstamps - hardware time stamps
444 * @hwtstamp: hardware time stamp transformed into duration
445 * since arbitrary point in time
446 * @netdev_data: address/cookie of network device driver used as
447 * reference to actual hardware time stamp
449 * Software time stamps generated by ktime_get_real() are stored in
452 * hwtstamps can only be compared against other hwtstamps from
455 * This structure is attached to packets as part of the
456 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
458 struct skb_shared_hwtstamps {
465 /* Definitions for tx_flags in struct skb_shared_info */
467 /* generate hardware time stamp */
468 SKBTX_HW_TSTAMP = 1 << 0,
470 /* generate software time stamp when queueing packet to NIC */
471 SKBTX_SW_TSTAMP = 1 << 1,
473 /* device driver is going to provide hardware time stamp */
474 SKBTX_IN_PROGRESS = 1 << 2,
476 /* generate hardware time stamp based on cycles if supported */
477 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3,
479 /* generate wifi status information (where possible) */
480 SKBTX_WIFI_STATUS = 1 << 4,
482 /* determine hardware time stamp based on time or cycles */
483 SKBTX_HW_TSTAMP_NETDEV = 1 << 5,
485 /* generate software time stamp when entering packet scheduling */
486 SKBTX_SCHED_TSTAMP = 1 << 6,
489 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
491 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \
492 SKBTX_HW_TSTAMP_USE_CYCLES | \
495 /* Definitions for flags in struct skb_shared_info */
497 /* use zcopy routines */
498 SKBFL_ZEROCOPY_ENABLE = BIT(0),
500 /* This indicates at least one fragment might be overwritten
501 * (as in vmsplice(), sendfile() ...)
502 * If we need to compute a TX checksum, we'll need to copy
503 * all frags to avoid possible bad checksum
505 SKBFL_SHARED_FRAG = BIT(1),
507 /* segment contains only zerocopy data and should not be
508 * charged to the kernel memory.
510 SKBFL_PURE_ZEROCOPY = BIT(2),
512 SKBFL_DONT_ORPHAN = BIT(3),
514 /* page references are managed by the ubuf_info, so it's safe to
515 * use frags only up until ubuf_info is released
517 SKBFL_MANAGED_FRAG_REFS = BIT(4),
520 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
521 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
522 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
525 * The callback notifies userspace to release buffers when skb DMA is done in
526 * lower device, the skb last reference should be 0 when calling this.
527 * The zerocopy_success argument is true if zero copy transmit occurred,
528 * false on data copy or out of memory error caused by data copy attempt.
529 * The ctx field is used to track device context.
530 * The desc field is used to track userspace buffer index.
533 void (*callback)(struct sk_buff *, struct ubuf_info *,
534 bool zerocopy_success);
539 struct ubuf_info_msgzc {
540 struct ubuf_info ubuf;
556 struct user_struct *user;
561 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
562 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \
565 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
566 void mm_unaccount_pinned_pages(struct mmpin *mmp);
568 /* This data is invariant across clones and lives at
569 * the end of the header data, ie. at skb->end.
571 struct skb_shared_info {
576 unsigned short gso_size;
577 /* Warning: this field is not always filled in (UFO)! */
578 unsigned short gso_segs;
579 struct sk_buff *frag_list;
580 struct skb_shared_hwtstamps hwtstamps;
581 unsigned int gso_type;
585 * Warning : all fields before dataref are cleared in __alloc_skb()
588 unsigned int xdp_frags_size;
590 /* Intermediate layers must ensure that destructor_arg
591 * remains valid until skb destructor */
592 void * destructor_arg;
594 /* must be last field, see pskb_expand_head() */
595 skb_frag_t frags[MAX_SKB_FRAGS];
599 * DOC: dataref and headerless skbs
601 * Transport layers send out clones of payload skbs they hold for
602 * retransmissions. To allow lower layers of the stack to prepend their headers
603 * we split &skb_shared_info.dataref into two halves.
604 * The lower 16 bits count the overall number of references.
605 * The higher 16 bits indicate how many of the references are payload-only.
606 * skb_header_cloned() checks if skb is allowed to add / write the headers.
608 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
609 * (via __skb_header_release()). Any clone created from marked skb will get
610 * &sk_buff.hdr_len populated with the available headroom.
611 * If there's the only clone in existence it's able to modify the headroom
612 * at will. The sequence of calls inside the transport layer is::
616 * __skb_header_release()
618 * // send the clone down the stack
620 * This is not a very generic construct and it depends on the transport layers
621 * doing the right thing. In practice there's usually only one payload-only skb.
622 * Having multiple payload-only skbs with different lengths of hdr_len is not
623 * possible. The payload-only skbs should never leave their owner.
625 #define SKB_DATAREF_SHIFT 16
626 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
630 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
631 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
632 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
636 SKB_GSO_TCPV4 = 1 << 0,
638 /* This indicates the skb is from an untrusted source. */
639 SKB_GSO_DODGY = 1 << 1,
641 /* This indicates the tcp segment has CWR set. */
642 SKB_GSO_TCP_ECN = 1 << 2,
644 SKB_GSO_TCP_FIXEDID = 1 << 3,
646 SKB_GSO_TCPV6 = 1 << 4,
648 SKB_GSO_FCOE = 1 << 5,
650 SKB_GSO_GRE = 1 << 6,
652 SKB_GSO_GRE_CSUM = 1 << 7,
654 SKB_GSO_IPXIP4 = 1 << 8,
656 SKB_GSO_IPXIP6 = 1 << 9,
658 SKB_GSO_UDP_TUNNEL = 1 << 10,
660 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
662 SKB_GSO_PARTIAL = 1 << 12,
664 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
666 SKB_GSO_SCTP = 1 << 14,
668 SKB_GSO_ESP = 1 << 15,
670 SKB_GSO_UDP = 1 << 16,
672 SKB_GSO_UDP_L4 = 1 << 17,
674 SKB_GSO_FRAGLIST = 1 << 18,
677 #if BITS_PER_LONG > 32
678 #define NET_SKBUFF_DATA_USES_OFFSET 1
681 #ifdef NET_SKBUFF_DATA_USES_OFFSET
682 typedef unsigned int sk_buff_data_t;
684 typedef unsigned char *sk_buff_data_t;
688 * DOC: Basic sk_buff geometry
690 * struct sk_buff itself is a metadata structure and does not hold any packet
691 * data. All the data is held in associated buffers.
693 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
696 * - data buffer, containing headers and sometimes payload;
697 * this is the part of the skb operated on by the common helpers
698 * such as skb_put() or skb_pull();
699 * - shared info (struct skb_shared_info) which holds an array of pointers
700 * to read-only data in the (page, offset, length) format.
702 * Optionally &skb_shared_info.frag_list may point to another skb.
704 * Basic diagram may look like this::
709 * ,--------------------------- + head
710 * / ,----------------- + data
711 * / / ,----------- + tail
715 * -----------------------------------------------
716 * | headroom | data | tailroom | skb_shared_info |
717 * -----------------------------------------------
721 * + [page frag] ---------
722 * + frag_list --> | sk_buff |
728 * struct sk_buff - socket buffer
729 * @next: Next buffer in list
730 * @prev: Previous buffer in list
731 * @tstamp: Time we arrived/left
732 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
733 * for retransmit timer
734 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
736 * @ll_node: anchor in an llist (eg socket defer_list)
737 * @sk: Socket we are owned by
738 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
739 * fragmentation management
740 * @dev: Device we arrived on/are leaving by
741 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
742 * @cb: Control buffer. Free for use by every layer. Put private vars here
743 * @_skb_refdst: destination entry (with norefcount bit)
744 * @sp: the security path, used for xfrm
745 * @len: Length of actual data
746 * @data_len: Data length
747 * @mac_len: Length of link layer header
748 * @hdr_len: writable header length of cloned skb
749 * @csum: Checksum (must include start/offset pair)
750 * @csum_start: Offset from skb->head where checksumming should start
751 * @csum_offset: Offset from csum_start where checksum should be stored
752 * @priority: Packet queueing priority
753 * @ignore_df: allow local fragmentation
754 * @cloned: Head may be cloned (check refcnt to be sure)
755 * @ip_summed: Driver fed us an IP checksum
756 * @nohdr: Payload reference only, must not modify header
757 * @pkt_type: Packet class
758 * @fclone: skbuff clone status
759 * @ipvs_property: skbuff is owned by ipvs
760 * @inner_protocol_type: whether the inner protocol is
761 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
762 * @remcsum_offload: remote checksum offload is enabled
763 * @offload_fwd_mark: Packet was L2-forwarded in hardware
764 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
765 * @tc_skip_classify: do not classify packet. set by IFB device
766 * @tc_at_ingress: used within tc_classify to distinguish in/egress
767 * @redirected: packet was redirected by packet classifier
768 * @from_ingress: packet was redirected from the ingress path
769 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
770 * @peeked: this packet has been seen already, so stats have been
771 * done for it, don't do them again
772 * @nf_trace: netfilter packet trace flag
773 * @protocol: Packet protocol from driver
774 * @destructor: Destruct function
775 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
776 * @_sk_redir: socket redirection information for skmsg
777 * @_nfct: Associated connection, if any (with nfctinfo bits)
778 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
779 * @skb_iif: ifindex of device we arrived on
780 * @tc_index: Traffic control index
781 * @hash: the packet hash
782 * @queue_mapping: Queue mapping for multiqueue devices
783 * @head_frag: skb was allocated from page fragments,
784 * not allocated by kmalloc() or vmalloc().
785 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
786 * @pp_recycle: mark the packet for recycling instead of freeing (implies
787 * page_pool support on driver)
788 * @active_extensions: active extensions (skb_ext_id types)
789 * @ndisc_nodetype: router type (from link layer)
790 * @ooo_okay: allow the mapping of a socket to a queue to be changed
791 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
793 * @sw_hash: indicates hash was computed in software stack
794 * @wifi_acked_valid: wifi_acked was set
795 * @wifi_acked: whether frame was acked on wifi or not
796 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
797 * @encapsulation: indicates the inner headers in the skbuff are valid
798 * @encap_hdr_csum: software checksum is needed
799 * @csum_valid: checksum is already valid
800 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
801 * @csum_complete_sw: checksum was completed by software
802 * @csum_level: indicates the number of consecutive checksums found in
803 * the packet minus one that have been verified as
804 * CHECKSUM_UNNECESSARY (max 3)
805 * @scm_io_uring: SKB holds io_uring registered files
806 * @dst_pending_confirm: need to confirm neighbour
807 * @decrypted: Decrypted SKB
808 * @slow_gro: state present at GRO time, slower prepare step required
809 * @mono_delivery_time: When set, skb->tstamp has the
810 * delivery_time in mono clock base (i.e. EDT). Otherwise, the
811 * skb->tstamp has the (rcv) timestamp at ingress and
812 * delivery_time at egress.
813 * @napi_id: id of the NAPI struct this skb came from
814 * @sender_cpu: (aka @napi_id) source CPU in XPS
815 * @alloc_cpu: CPU which did the skb allocation.
816 * @secmark: security marking
817 * @mark: Generic packet mark
818 * @reserved_tailroom: (aka @mark) number of bytes of free space available
819 * at the tail of an sk_buff
820 * @vlan_all: vlan fields (proto & tci)
821 * @vlan_proto: vlan encapsulation protocol
822 * @vlan_tci: vlan tag control information
823 * @inner_protocol: Protocol (encapsulation)
824 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
825 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
826 * @inner_transport_header: Inner transport layer header (encapsulation)
827 * @inner_network_header: Network layer header (encapsulation)
828 * @inner_mac_header: Link layer header (encapsulation)
829 * @transport_header: Transport layer header
830 * @network_header: Network layer header
831 * @mac_header: Link layer header
832 * @kcov_handle: KCOV remote handle for remote coverage collection
833 * @tail: Tail pointer
835 * @head: Head of buffer
836 * @data: Data head pointer
837 * @truesize: Buffer size
838 * @users: User count - see {datagram,tcp}.c
839 * @extensions: allocated extensions, valid if active_extensions is nonzero
845 /* These two members must be first to match sk_buff_head. */
846 struct sk_buff *next;
847 struct sk_buff *prev;
850 struct net_device *dev;
851 /* Some protocols might use this space to store information,
852 * while device pointer would be NULL.
853 * UDP receive path is one user.
855 unsigned long dev_scratch;
858 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
859 struct list_head list;
860 struct llist_node ll_node;
865 int ip_defrag_offset;
870 u64 skb_mstamp_ns; /* earliest departure time */
873 * This is the control buffer. It is free to use for every
874 * layer. Please put your private variables there. If you
875 * want to keep them across layers you have to do a skb_clone()
876 * first. This is owned by whoever has the skb queued ATM.
878 char cb[48] __aligned(8);
882 unsigned long _skb_refdst;
883 void (*destructor)(struct sk_buff *skb);
885 struct list_head tcp_tsorted_anchor;
886 #ifdef CONFIG_NET_SOCK_MSG
887 unsigned long _sk_redir;
891 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
899 /* Following fields are _not_ copied in __copy_skb_header()
900 * Note that queue_mapping is here mostly to fill a hole.
904 /* if you move cloned around you also must adapt those constants */
905 #ifdef __BIG_ENDIAN_BITFIELD
906 #define CLONED_MASK (1 << 7)
908 #define CLONED_MASK 1
910 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
913 __u8 __cloned_offset[0];
921 pp_recycle:1; /* page_pool recycle indicator */
922 #ifdef CONFIG_SKB_EXTENSIONS
923 __u8 active_extensions;
926 /* Fields enclosed in headers group are copied
927 * using a single memcpy() in __copy_skb_header()
929 struct_group(headers,
932 __u8 __pkt_type_offset[0];
934 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
942 __u8 wifi_acked_valid:1;
945 /* Indicates the inner headers are valid in the skbuff. */
946 __u8 encapsulation:1;
947 __u8 encap_hdr_csum:1;
951 __u8 __pkt_vlan_present_offset[0];
953 __u8 remcsum_offload:1;
954 __u8 csum_complete_sw:1;
956 __u8 dst_pending_confirm:1;
957 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */
958 #ifdef CONFIG_NET_CLS_ACT
959 __u8 tc_skip_classify:1;
960 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
962 #ifdef CONFIG_IPV6_NDISC_NODETYPE
963 __u8 ndisc_nodetype:2;
966 __u8 ipvs_property:1;
967 __u8 inner_protocol_type:1;
968 #ifdef CONFIG_NET_SWITCHDEV
969 __u8 offload_fwd_mark:1;
970 __u8 offload_l3_fwd_mark:1;
973 #ifdef CONFIG_NET_REDIRECT
976 #ifdef CONFIG_NETFILTER_SKIP_EGRESS
977 __u8 nf_skip_egress:1;
979 #ifdef CONFIG_TLS_DEVICE
983 __u8 csum_not_inet:1;
986 #ifdef CONFIG_NET_SCHED
987 __u16 tc_index; /* traffic control index */
1007 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
1009 unsigned int napi_id;
1010 unsigned int sender_cpu;
1014 #ifdef CONFIG_NETWORK_SECMARK
1020 __u32 reserved_tailroom;
1024 __be16 inner_protocol;
1028 __u16 inner_transport_header;
1029 __u16 inner_network_header;
1030 __u16 inner_mac_header;
1033 __u16 transport_header;
1034 __u16 network_header;
1041 ); /* end headers group */
1043 /* These elements must be at the end, see alloc_skb() for details. */
1044 sk_buff_data_t tail;
1046 unsigned char *head,
1048 unsigned int truesize;
1051 #ifdef CONFIG_SKB_EXTENSIONS
1052 /* only useable after checking ->active_extensions != 0 */
1053 struct skb_ext *extensions;
1057 /* if you move pkt_type around you also must adapt those constants */
1058 #ifdef __BIG_ENDIAN_BITFIELD
1059 #define PKT_TYPE_MAX (7 << 5)
1061 #define PKT_TYPE_MAX 7
1063 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1065 /* if you move tc_at_ingress or mono_delivery_time
1066 * around, you also must adapt these constants.
1068 #ifdef __BIG_ENDIAN_BITFIELD
1069 #define TC_AT_INGRESS_MASK (1 << 0)
1070 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 2)
1072 #define TC_AT_INGRESS_MASK (1 << 7)
1073 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 5)
1075 #define PKT_VLAN_PRESENT_OFFSET offsetof(struct sk_buff, __pkt_vlan_present_offset)
1079 * Handling routines are only of interest to the kernel
1082 #define SKB_ALLOC_FCLONE 0x01
1083 #define SKB_ALLOC_RX 0x02
1084 #define SKB_ALLOC_NAPI 0x04
1087 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1090 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1092 return unlikely(skb->pfmemalloc);
1096 * skb might have a dst pointer attached, refcounted or not.
1097 * _skb_refdst low order bit is set if refcount was _not_ taken
1099 #define SKB_DST_NOREF 1UL
1100 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1103 * skb_dst - returns skb dst_entry
1106 * Returns skb dst_entry, regardless of reference taken or not.
1108 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1110 /* If refdst was not refcounted, check we still are in a
1111 * rcu_read_lock section
1113 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1114 !rcu_read_lock_held() &&
1115 !rcu_read_lock_bh_held());
1116 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1120 * skb_dst_set - sets skb dst
1124 * Sets skb dst, assuming a reference was taken on dst and should
1125 * be released by skb_dst_drop()
1127 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1129 skb->slow_gro |= !!dst;
1130 skb->_skb_refdst = (unsigned long)dst;
1134 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1138 * Sets skb dst, assuming a reference was not taken on dst.
1139 * If dst entry is cached, we do not take reference and dst_release
1140 * will be avoided by refdst_drop. If dst entry is not cached, we take
1141 * reference, so that last dst_release can destroy the dst immediately.
1143 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1145 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1146 skb->slow_gro |= !!dst;
1147 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1151 * skb_dst_is_noref - Test if skb dst isn't refcounted
1154 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1156 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1160 * skb_rtable - Returns the skb &rtable
1163 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1165 return (struct rtable *)skb_dst(skb);
1168 /* For mangling skb->pkt_type from user space side from applications
1169 * such as nft, tc, etc, we only allow a conservative subset of
1170 * possible pkt_types to be set.
1172 static inline bool skb_pkt_type_ok(u32 ptype)
1174 return ptype <= PACKET_OTHERHOST;
1178 * skb_napi_id - Returns the skb's NAPI id
1181 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1183 #ifdef CONFIG_NET_RX_BUSY_POLL
1184 return skb->napi_id;
1191 * skb_unref - decrement the skb's reference count
1194 * Returns true if we can free the skb.
1196 static inline bool skb_unref(struct sk_buff *skb)
1200 if (likely(refcount_read(&skb->users) == 1))
1202 else if (likely(!refcount_dec_and_test(&skb->users)))
1209 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason);
1212 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1213 * @skb: buffer to free
1215 static inline void kfree_skb(struct sk_buff *skb)
1217 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1220 void skb_release_head_state(struct sk_buff *skb);
1221 void kfree_skb_list_reason(struct sk_buff *segs,
1222 enum skb_drop_reason reason);
1223 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1224 void skb_tx_error(struct sk_buff *skb);
1226 static inline void kfree_skb_list(struct sk_buff *segs)
1228 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1231 #ifdef CONFIG_TRACEPOINTS
1232 void consume_skb(struct sk_buff *skb);
1234 static inline void consume_skb(struct sk_buff *skb)
1236 return kfree_skb(skb);
1240 void __consume_stateless_skb(struct sk_buff *skb);
1241 void __kfree_skb(struct sk_buff *skb);
1242 extern struct kmem_cache *skbuff_head_cache;
1244 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1245 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1246 bool *fragstolen, int *delta_truesize);
1248 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1250 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1251 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1252 struct sk_buff *build_skb_around(struct sk_buff *skb,
1253 void *data, unsigned int frag_size);
1254 void skb_attempt_defer_free(struct sk_buff *skb);
1256 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1257 struct sk_buff *slab_build_skb(void *data);
1260 * alloc_skb - allocate a network buffer
1261 * @size: size to allocate
1262 * @priority: allocation mask
1264 * This function is a convenient wrapper around __alloc_skb().
1266 static inline struct sk_buff *alloc_skb(unsigned int size,
1269 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1272 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1273 unsigned long data_len,
1277 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1279 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1280 struct sk_buff_fclones {
1281 struct sk_buff skb1;
1283 struct sk_buff skb2;
1285 refcount_t fclone_ref;
1289 * skb_fclone_busy - check if fclone is busy
1293 * Returns true if skb is a fast clone, and its clone is not freed.
1294 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1295 * so we also check that this didnt happen.
1297 static inline bool skb_fclone_busy(const struct sock *sk,
1298 const struct sk_buff *skb)
1300 const struct sk_buff_fclones *fclones;
1302 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1304 return skb->fclone == SKB_FCLONE_ORIG &&
1305 refcount_read(&fclones->fclone_ref) > 1 &&
1306 READ_ONCE(fclones->skb2.sk) == sk;
1310 * alloc_skb_fclone - allocate a network buffer from fclone cache
1311 * @size: size to allocate
1312 * @priority: allocation mask
1314 * This function is a convenient wrapper around __alloc_skb().
1316 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1319 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1322 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1323 void skb_headers_offset_update(struct sk_buff *skb, int off);
1324 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1325 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1326 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1327 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1328 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1329 gfp_t gfp_mask, bool fclone);
1330 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1333 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1336 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1337 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1338 unsigned int headroom);
1339 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1340 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1341 int newtailroom, gfp_t priority);
1342 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1343 int offset, int len);
1344 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1345 int offset, int len);
1346 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1347 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1350 * skb_pad - zero pad the tail of an skb
1351 * @skb: buffer to pad
1352 * @pad: space to pad
1354 * Ensure that a buffer is followed by a padding area that is zero
1355 * filled. Used by network drivers which may DMA or transfer data
1356 * beyond the buffer end onto the wire.
1358 * May return error in out of memory cases. The skb is freed on error.
1360 static inline int skb_pad(struct sk_buff *skb, int pad)
1362 return __skb_pad(skb, pad, true);
1364 #define dev_kfree_skb(a) consume_skb(a)
1366 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1367 int offset, size_t size);
1369 struct skb_seq_state {
1373 __u32 stepped_offset;
1374 struct sk_buff *root_skb;
1375 struct sk_buff *cur_skb;
1380 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1381 unsigned int to, struct skb_seq_state *st);
1382 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1383 struct skb_seq_state *st);
1384 void skb_abort_seq_read(struct skb_seq_state *st);
1386 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1387 unsigned int to, struct ts_config *config);
1390 * Packet hash types specify the type of hash in skb_set_hash.
1392 * Hash types refer to the protocol layer addresses which are used to
1393 * construct a packet's hash. The hashes are used to differentiate or identify
1394 * flows of the protocol layer for the hash type. Hash types are either
1395 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1397 * Properties of hashes:
1399 * 1) Two packets in different flows have different hash values
1400 * 2) Two packets in the same flow should have the same hash value
1402 * A hash at a higher layer is considered to be more specific. A driver should
1403 * set the most specific hash possible.
1405 * A driver cannot indicate a more specific hash than the layer at which a hash
1406 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1408 * A driver may indicate a hash level which is less specific than the
1409 * actual layer the hash was computed on. For instance, a hash computed
1410 * at L4 may be considered an L3 hash. This should only be done if the
1411 * driver can't unambiguously determine that the HW computed the hash at
1412 * the higher layer. Note that the "should" in the second property above
1415 enum pkt_hash_types {
1416 PKT_HASH_TYPE_NONE, /* Undefined type */
1417 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1418 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1419 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1422 static inline void skb_clear_hash(struct sk_buff *skb)
1429 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1432 skb_clear_hash(skb);
1436 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1438 skb->l4_hash = is_l4;
1439 skb->sw_hash = is_sw;
1444 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1446 /* Used by drivers to set hash from HW */
1447 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1451 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1453 __skb_set_hash(skb, hash, true, is_l4);
1456 void __skb_get_hash(struct sk_buff *skb);
1457 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1458 u32 skb_get_poff(const struct sk_buff *skb);
1459 u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1460 const struct flow_keys_basic *keys, int hlen);
1461 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1462 const void *data, int hlen_proto);
1464 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1465 int thoff, u8 ip_proto)
1467 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1470 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1471 const struct flow_dissector_key *key,
1472 unsigned int key_count);
1474 struct bpf_flow_dissector;
1475 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1476 __be16 proto, int nhoff, int hlen, unsigned int flags);
1478 bool __skb_flow_dissect(const struct net *net,
1479 const struct sk_buff *skb,
1480 struct flow_dissector *flow_dissector,
1481 void *target_container, const void *data,
1482 __be16 proto, int nhoff, int hlen, unsigned int flags);
1484 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1485 struct flow_dissector *flow_dissector,
1486 void *target_container, unsigned int flags)
1488 return __skb_flow_dissect(NULL, skb, flow_dissector,
1489 target_container, NULL, 0, 0, 0, flags);
1492 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1493 struct flow_keys *flow,
1496 memset(flow, 0, sizeof(*flow));
1497 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1498 flow, NULL, 0, 0, 0, flags);
1502 skb_flow_dissect_flow_keys_basic(const struct net *net,
1503 const struct sk_buff *skb,
1504 struct flow_keys_basic *flow,
1505 const void *data, __be16 proto,
1506 int nhoff, int hlen, unsigned int flags)
1508 memset(flow, 0, sizeof(*flow));
1509 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1510 data, proto, nhoff, hlen, flags);
1513 void skb_flow_dissect_meta(const struct sk_buff *skb,
1514 struct flow_dissector *flow_dissector,
1515 void *target_container);
1517 /* Gets a skb connection tracking info, ctinfo map should be a
1518 * map of mapsize to translate enum ip_conntrack_info states
1522 skb_flow_dissect_ct(const struct sk_buff *skb,
1523 struct flow_dissector *flow_dissector,
1524 void *target_container,
1525 u16 *ctinfo_map, size_t mapsize,
1526 bool post_ct, u16 zone);
1528 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1529 struct flow_dissector *flow_dissector,
1530 void *target_container);
1532 void skb_flow_dissect_hash(const struct sk_buff *skb,
1533 struct flow_dissector *flow_dissector,
1534 void *target_container);
1536 static inline __u32 skb_get_hash(struct sk_buff *skb)
1538 if (!skb->l4_hash && !skb->sw_hash)
1539 __skb_get_hash(skb);
1544 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1546 if (!skb->l4_hash && !skb->sw_hash) {
1547 struct flow_keys keys;
1548 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1550 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1556 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1557 const siphash_key_t *perturb);
1559 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1564 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1566 to->hash = from->hash;
1567 to->sw_hash = from->sw_hash;
1568 to->l4_hash = from->l4_hash;
1571 static inline void skb_copy_decrypted(struct sk_buff *to,
1572 const struct sk_buff *from)
1574 #ifdef CONFIG_TLS_DEVICE
1575 to->decrypted = from->decrypted;
1579 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1580 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1582 return skb->head + skb->end;
1585 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1590 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1595 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1600 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1602 return skb->end - skb->head;
1605 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1607 skb->end = skb->head + offset;
1611 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1612 struct ubuf_info *uarg);
1614 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1616 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1619 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
1620 struct sk_buff *skb, struct iov_iter *from,
1623 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1624 struct msghdr *msg, int len)
1626 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len);
1629 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1630 struct msghdr *msg, int len,
1631 struct ubuf_info *uarg);
1634 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1636 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1638 return &skb_shinfo(skb)->hwtstamps;
1641 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1643 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1645 return is_zcopy ? skb_uarg(skb) : NULL;
1648 static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1650 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1653 static inline bool skb_zcopy_managed(const struct sk_buff *skb)
1655 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS;
1658 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1659 const struct sk_buff *skb2)
1661 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1664 static inline void net_zcopy_get(struct ubuf_info *uarg)
1666 refcount_inc(&uarg->refcnt);
1669 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1671 skb_shinfo(skb)->destructor_arg = uarg;
1672 skb_shinfo(skb)->flags |= uarg->flags;
1675 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1678 if (skb && uarg && !skb_zcopy(skb)) {
1679 if (unlikely(have_ref && *have_ref))
1682 net_zcopy_get(uarg);
1683 skb_zcopy_init(skb, uarg);
1687 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1689 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1690 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1693 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1695 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1698 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1700 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1703 static inline void net_zcopy_put(struct ubuf_info *uarg)
1706 uarg->callback(NULL, uarg, true);
1709 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1712 if (uarg->callback == msg_zerocopy_callback)
1713 msg_zerocopy_put_abort(uarg, have_uref);
1715 net_zcopy_put(uarg);
1719 /* Release a reference on a zerocopy structure */
1720 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1722 struct ubuf_info *uarg = skb_zcopy(skb);
1725 if (!skb_zcopy_is_nouarg(skb))
1726 uarg->callback(skb, uarg, zerocopy_success);
1728 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1732 void __skb_zcopy_downgrade_managed(struct sk_buff *skb);
1734 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb)
1736 if (unlikely(skb_zcopy_managed(skb)))
1737 __skb_zcopy_downgrade_managed(skb);
1740 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1745 /* Iterate through singly-linked GSO fragments of an skb. */
1746 #define skb_list_walk_safe(first, skb, next_skb) \
1747 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1748 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1750 static inline void skb_list_del_init(struct sk_buff *skb)
1752 __list_del_entry(&skb->list);
1753 skb_mark_not_on_list(skb);
1757 * skb_queue_empty - check if a queue is empty
1760 * Returns true if the queue is empty, false otherwise.
1762 static inline int skb_queue_empty(const struct sk_buff_head *list)
1764 return list->next == (const struct sk_buff *) list;
1768 * skb_queue_empty_lockless - check if a queue is empty
1771 * Returns true if the queue is empty, false otherwise.
1772 * This variant can be used in lockless contexts.
1774 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1776 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1781 * skb_queue_is_last - check if skb is the last entry in the queue
1785 * Returns true if @skb is the last buffer on the list.
1787 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1788 const struct sk_buff *skb)
1790 return skb->next == (const struct sk_buff *) list;
1794 * skb_queue_is_first - check if skb is the first entry in the queue
1798 * Returns true if @skb is the first buffer on the list.
1800 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1801 const struct sk_buff *skb)
1803 return skb->prev == (const struct sk_buff *) list;
1807 * skb_queue_next - return the next packet in the queue
1809 * @skb: current buffer
1811 * Return the next packet in @list after @skb. It is only valid to
1812 * call this if skb_queue_is_last() evaluates to false.
1814 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1815 const struct sk_buff *skb)
1817 /* This BUG_ON may seem severe, but if we just return then we
1818 * are going to dereference garbage.
1820 BUG_ON(skb_queue_is_last(list, skb));
1825 * skb_queue_prev - return the prev packet in the queue
1827 * @skb: current buffer
1829 * Return the prev packet in @list before @skb. It is only valid to
1830 * call this if skb_queue_is_first() evaluates to false.
1832 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1833 const struct sk_buff *skb)
1835 /* This BUG_ON may seem severe, but if we just return then we
1836 * are going to dereference garbage.
1838 BUG_ON(skb_queue_is_first(list, skb));
1843 * skb_get - reference buffer
1844 * @skb: buffer to reference
1846 * Makes another reference to a socket buffer and returns a pointer
1849 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1851 refcount_inc(&skb->users);
1856 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1860 * skb_cloned - is the buffer a clone
1861 * @skb: buffer to check
1863 * Returns true if the buffer was generated with skb_clone() and is
1864 * one of multiple shared copies of the buffer. Cloned buffers are
1865 * shared data so must not be written to under normal circumstances.
1867 static inline int skb_cloned(const struct sk_buff *skb)
1869 return skb->cloned &&
1870 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1873 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1875 might_sleep_if(gfpflags_allow_blocking(pri));
1877 if (skb_cloned(skb))
1878 return pskb_expand_head(skb, 0, 0, pri);
1883 /* This variant of skb_unclone() makes sure skb->truesize
1884 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
1886 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
1887 * when various debugging features are in place.
1889 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
1890 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1892 might_sleep_if(gfpflags_allow_blocking(pri));
1894 if (skb_cloned(skb))
1895 return __skb_unclone_keeptruesize(skb, pri);
1900 * skb_header_cloned - is the header a clone
1901 * @skb: buffer to check
1903 * Returns true if modifying the header part of the buffer requires
1904 * the data to be copied.
1906 static inline int skb_header_cloned(const struct sk_buff *skb)
1913 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1914 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1915 return dataref != 1;
1918 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1920 might_sleep_if(gfpflags_allow_blocking(pri));
1922 if (skb_header_cloned(skb))
1923 return pskb_expand_head(skb, 0, 0, pri);
1929 * __skb_header_release() - allow clones to use the headroom
1930 * @skb: buffer to operate on
1932 * See "DOC: dataref and headerless skbs".
1934 static inline void __skb_header_release(struct sk_buff *skb)
1937 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1942 * skb_shared - is the buffer shared
1943 * @skb: buffer to check
1945 * Returns true if more than one person has a reference to this
1948 static inline int skb_shared(const struct sk_buff *skb)
1950 return refcount_read(&skb->users) != 1;
1954 * skb_share_check - check if buffer is shared and if so clone it
1955 * @skb: buffer to check
1956 * @pri: priority for memory allocation
1958 * If the buffer is shared the buffer is cloned and the old copy
1959 * drops a reference. A new clone with a single reference is returned.
1960 * If the buffer is not shared the original buffer is returned. When
1961 * being called from interrupt status or with spinlocks held pri must
1964 * NULL is returned on a memory allocation failure.
1966 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1968 might_sleep_if(gfpflags_allow_blocking(pri));
1969 if (skb_shared(skb)) {
1970 struct sk_buff *nskb = skb_clone(skb, pri);
1982 * Copy shared buffers into a new sk_buff. We effectively do COW on
1983 * packets to handle cases where we have a local reader and forward
1984 * and a couple of other messy ones. The normal one is tcpdumping
1985 * a packet thats being forwarded.
1989 * skb_unshare - make a copy of a shared buffer
1990 * @skb: buffer to check
1991 * @pri: priority for memory allocation
1993 * If the socket buffer is a clone then this function creates a new
1994 * copy of the data, drops a reference count on the old copy and returns
1995 * the new copy with the reference count at 1. If the buffer is not a clone
1996 * the original buffer is returned. When called with a spinlock held or
1997 * from interrupt state @pri must be %GFP_ATOMIC
1999 * %NULL is returned on a memory allocation failure.
2001 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
2004 might_sleep_if(gfpflags_allow_blocking(pri));
2005 if (skb_cloned(skb)) {
2006 struct sk_buff *nskb = skb_copy(skb, pri);
2008 /* Free our shared copy */
2019 * skb_peek - peek at the head of an &sk_buff_head
2020 * @list_: list to peek at
2022 * Peek an &sk_buff. Unlike most other operations you _MUST_
2023 * be careful with this one. A peek leaves the buffer on the
2024 * list and someone else may run off with it. You must hold
2025 * the appropriate locks or have a private queue to do this.
2027 * Returns %NULL for an empty list or a pointer to the head element.
2028 * The reference count is not incremented and the reference is therefore
2029 * volatile. Use with caution.
2031 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2033 struct sk_buff *skb = list_->next;
2035 if (skb == (struct sk_buff *)list_)
2041 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2042 * @list_: list to peek at
2044 * Like skb_peek(), but the caller knows that the list is not empty.
2046 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2052 * skb_peek_next - peek skb following the given one from a queue
2053 * @skb: skb to start from
2054 * @list_: list to peek at
2056 * Returns %NULL when the end of the list is met or a pointer to the
2057 * next element. The reference count is not incremented and the
2058 * reference is therefore volatile. Use with caution.
2060 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2061 const struct sk_buff_head *list_)
2063 struct sk_buff *next = skb->next;
2065 if (next == (struct sk_buff *)list_)
2071 * skb_peek_tail - peek at the tail of an &sk_buff_head
2072 * @list_: list to peek at
2074 * Peek an &sk_buff. Unlike most other operations you _MUST_
2075 * be careful with this one. A peek leaves the buffer on the
2076 * list and someone else may run off with it. You must hold
2077 * the appropriate locks or have a private queue to do this.
2079 * Returns %NULL for an empty list or a pointer to the tail element.
2080 * The reference count is not incremented and the reference is therefore
2081 * volatile. Use with caution.
2083 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2085 struct sk_buff *skb = READ_ONCE(list_->prev);
2087 if (skb == (struct sk_buff *)list_)
2094 * skb_queue_len - get queue length
2095 * @list_: list to measure
2097 * Return the length of an &sk_buff queue.
2099 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2105 * skb_queue_len_lockless - get queue length
2106 * @list_: list to measure
2108 * Return the length of an &sk_buff queue.
2109 * This variant can be used in lockless contexts.
2111 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2113 return READ_ONCE(list_->qlen);
2117 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2118 * @list: queue to initialize
2120 * This initializes only the list and queue length aspects of
2121 * an sk_buff_head object. This allows to initialize the list
2122 * aspects of an sk_buff_head without reinitializing things like
2123 * the spinlock. It can also be used for on-stack sk_buff_head
2124 * objects where the spinlock is known to not be used.
2126 static inline void __skb_queue_head_init(struct sk_buff_head *list)
2128 list->prev = list->next = (struct sk_buff *)list;
2133 * This function creates a split out lock class for each invocation;
2134 * this is needed for now since a whole lot of users of the skb-queue
2135 * infrastructure in drivers have different locking usage (in hardirq)
2136 * than the networking core (in softirq only). In the long run either the
2137 * network layer or drivers should need annotation to consolidate the
2138 * main types of usage into 3 classes.
2140 static inline void skb_queue_head_init(struct sk_buff_head *list)
2142 spin_lock_init(&list->lock);
2143 __skb_queue_head_init(list);
2146 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2147 struct lock_class_key *class)
2149 skb_queue_head_init(list);
2150 lockdep_set_class(&list->lock, class);
2154 * Insert an sk_buff on a list.
2156 * The "__skb_xxxx()" functions are the non-atomic ones that
2157 * can only be called with interrupts disabled.
2159 static inline void __skb_insert(struct sk_buff *newsk,
2160 struct sk_buff *prev, struct sk_buff *next,
2161 struct sk_buff_head *list)
2163 /* See skb_queue_empty_lockless() and skb_peek_tail()
2164 * for the opposite READ_ONCE()
2166 WRITE_ONCE(newsk->next, next);
2167 WRITE_ONCE(newsk->prev, prev);
2168 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2169 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2170 WRITE_ONCE(list->qlen, list->qlen + 1);
2173 static inline void __skb_queue_splice(const struct sk_buff_head *list,
2174 struct sk_buff *prev,
2175 struct sk_buff *next)
2177 struct sk_buff *first = list->next;
2178 struct sk_buff *last = list->prev;
2180 WRITE_ONCE(first->prev, prev);
2181 WRITE_ONCE(prev->next, first);
2183 WRITE_ONCE(last->next, next);
2184 WRITE_ONCE(next->prev, last);
2188 * skb_queue_splice - join two skb lists, this is designed for stacks
2189 * @list: the new list to add
2190 * @head: the place to add it in the first list
2192 static inline void skb_queue_splice(const struct sk_buff_head *list,
2193 struct sk_buff_head *head)
2195 if (!skb_queue_empty(list)) {
2196 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2197 head->qlen += list->qlen;
2202 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2203 * @list: the new list to add
2204 * @head: the place to add it in the first list
2206 * The list at @list is reinitialised
2208 static inline void skb_queue_splice_init(struct sk_buff_head *list,
2209 struct sk_buff_head *head)
2211 if (!skb_queue_empty(list)) {
2212 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2213 head->qlen += list->qlen;
2214 __skb_queue_head_init(list);
2219 * skb_queue_splice_tail - join two skb lists, each list being a queue
2220 * @list: the new list to add
2221 * @head: the place to add it in the first list
2223 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2224 struct sk_buff_head *head)
2226 if (!skb_queue_empty(list)) {
2227 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2228 head->qlen += list->qlen;
2233 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2234 * @list: the new list to add
2235 * @head: the place to add it in the first list
2237 * Each of the lists is a queue.
2238 * The list at @list is reinitialised
2240 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2241 struct sk_buff_head *head)
2243 if (!skb_queue_empty(list)) {
2244 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2245 head->qlen += list->qlen;
2246 __skb_queue_head_init(list);
2251 * __skb_queue_after - queue a buffer at the list head
2252 * @list: list to use
2253 * @prev: place after this buffer
2254 * @newsk: buffer to queue
2256 * Queue a buffer int the middle of a list. This function takes no locks
2257 * and you must therefore hold required locks before calling it.
2259 * A buffer cannot be placed on two lists at the same time.
2261 static inline void __skb_queue_after(struct sk_buff_head *list,
2262 struct sk_buff *prev,
2263 struct sk_buff *newsk)
2265 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2268 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2269 struct sk_buff_head *list);
2271 static inline void __skb_queue_before(struct sk_buff_head *list,
2272 struct sk_buff *next,
2273 struct sk_buff *newsk)
2275 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2279 * __skb_queue_head - queue a buffer at the list head
2280 * @list: list to use
2281 * @newsk: buffer to queue
2283 * Queue a buffer at the start of a list. This function takes no locks
2284 * and you must therefore hold required locks before calling it.
2286 * A buffer cannot be placed on two lists at the same time.
2288 static inline void __skb_queue_head(struct sk_buff_head *list,
2289 struct sk_buff *newsk)
2291 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2293 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2296 * __skb_queue_tail - queue a buffer at the list tail
2297 * @list: list to use
2298 * @newsk: buffer to queue
2300 * Queue a buffer at the end of a list. This function takes no locks
2301 * and you must therefore hold required locks before calling it.
2303 * A buffer cannot be placed on two lists at the same time.
2305 static inline void __skb_queue_tail(struct sk_buff_head *list,
2306 struct sk_buff *newsk)
2308 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2310 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2313 * remove sk_buff from list. _Must_ be called atomically, and with
2316 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2317 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2319 struct sk_buff *next, *prev;
2321 WRITE_ONCE(list->qlen, list->qlen - 1);
2324 skb->next = skb->prev = NULL;
2325 WRITE_ONCE(next->prev, prev);
2326 WRITE_ONCE(prev->next, next);
2330 * __skb_dequeue - remove from the head of the queue
2331 * @list: list to dequeue from
2333 * Remove the head of the list. This function does not take any locks
2334 * so must be used with appropriate locks held only. The head item is
2335 * returned or %NULL if the list is empty.
2337 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2339 struct sk_buff *skb = skb_peek(list);
2341 __skb_unlink(skb, list);
2344 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2347 * __skb_dequeue_tail - remove from the tail of the queue
2348 * @list: list to dequeue from
2350 * Remove the tail of the list. This function does not take any locks
2351 * so must be used with appropriate locks held only. The tail item is
2352 * returned or %NULL if the list is empty.
2354 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2356 struct sk_buff *skb = skb_peek_tail(list);
2358 __skb_unlink(skb, list);
2361 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2364 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2366 return skb->data_len;
2369 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2371 return skb->len - skb->data_len;
2374 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2376 unsigned int i, len = 0;
2378 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2379 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2383 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2385 return skb_headlen(skb) + __skb_pagelen(skb);
2388 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo,
2389 int i, struct page *page,
2392 skb_frag_t *frag = &shinfo->frags[i];
2395 * Propagate page pfmemalloc to the skb if we can. The problem is
2396 * that not all callers have unique ownership of the page but rely
2397 * on page_is_pfmemalloc doing the right thing(tm).
2399 frag->bv_page = page;
2400 frag->bv_offset = off;
2401 skb_frag_size_set(frag, size);
2405 * skb_len_add - adds a number to len fields of skb
2406 * @skb: buffer to add len to
2407 * @delta: number of bytes to add
2409 static inline void skb_len_add(struct sk_buff *skb, int delta)
2412 skb->data_len += delta;
2413 skb->truesize += delta;
2417 * __skb_fill_page_desc - initialise a paged fragment in an skb
2418 * @skb: buffer containing fragment to be initialised
2419 * @i: paged fragment index to initialise
2420 * @page: the page to use for this fragment
2421 * @off: the offset to the data with @page
2422 * @size: the length of the data
2424 * Initialises the @i'th fragment of @skb to point to &size bytes at
2425 * offset @off within @page.
2427 * Does not take any additional reference on the fragment.
2429 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2430 struct page *page, int off, int size)
2432 __skb_fill_page_desc_noacc(skb_shinfo(skb), i, page, off, size);
2433 page = compound_head(page);
2434 if (page_is_pfmemalloc(page))
2435 skb->pfmemalloc = true;
2439 * skb_fill_page_desc - initialise a paged fragment in an skb
2440 * @skb: buffer containing fragment to be initialised
2441 * @i: paged fragment index to initialise
2442 * @page: the page to use for this fragment
2443 * @off: the offset to the data with @page
2444 * @size: the length of the data
2446 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2447 * @skb to point to @size bytes at offset @off within @page. In
2448 * addition updates @skb such that @i is the last fragment.
2450 * Does not take any additional reference on the fragment.
2452 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2453 struct page *page, int off, int size)
2455 __skb_fill_page_desc(skb, i, page, off, size);
2456 skb_shinfo(skb)->nr_frags = i + 1;
2460 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb
2461 * @skb: buffer containing fragment to be initialised
2462 * @i: paged fragment index to initialise
2463 * @page: the page to use for this fragment
2464 * @off: the offset to the data with @page
2465 * @size: the length of the data
2467 * Variant of skb_fill_page_desc() which does not deal with
2468 * pfmemalloc, if page is not owned by us.
2470 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i,
2471 struct page *page, int off,
2474 struct skb_shared_info *shinfo = skb_shinfo(skb);
2476 __skb_fill_page_desc_noacc(shinfo, i, page, off, size);
2477 shinfo->nr_frags = i + 1;
2480 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2481 int size, unsigned int truesize);
2483 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2484 unsigned int truesize);
2486 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2488 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2489 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2491 return skb->head + skb->tail;
2494 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2496 skb->tail = skb->data - skb->head;
2499 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2501 skb_reset_tail_pointer(skb);
2502 skb->tail += offset;
2505 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2506 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2511 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2513 skb->tail = skb->data;
2516 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2518 skb->tail = skb->data + offset;
2521 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2523 static inline void skb_assert_len(struct sk_buff *skb)
2525 #ifdef CONFIG_DEBUG_NET
2526 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2527 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2528 #endif /* CONFIG_DEBUG_NET */
2532 * Add data to an sk_buff
2534 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2535 void *skb_put(struct sk_buff *skb, unsigned int len);
2536 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2538 void *tmp = skb_tail_pointer(skb);
2539 SKB_LINEAR_ASSERT(skb);
2545 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2547 void *tmp = __skb_put(skb, len);
2549 memset(tmp, 0, len);
2553 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2556 void *tmp = __skb_put(skb, len);
2558 memcpy(tmp, data, len);
2562 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2564 *(u8 *)__skb_put(skb, 1) = val;
2567 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2569 void *tmp = skb_put(skb, len);
2571 memset(tmp, 0, len);
2576 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2579 void *tmp = skb_put(skb, len);
2581 memcpy(tmp, data, len);
2586 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2588 *(u8 *)skb_put(skb, 1) = val;
2591 void *skb_push(struct sk_buff *skb, unsigned int len);
2592 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2599 void *skb_pull(struct sk_buff *skb, unsigned int len);
2600 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2603 if (unlikely(skb->len < skb->data_len)) {
2604 #if defined(CONFIG_DEBUG_NET)
2606 pr_err("__skb_pull(len=%u)\n", len);
2607 skb_dump(KERN_ERR, skb, false);
2611 return skb->data += len;
2614 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2616 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2619 void *skb_pull_data(struct sk_buff *skb, size_t len);
2621 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2623 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2625 if (likely(len <= skb_headlen(skb)))
2627 if (unlikely(len > skb->len))
2629 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2632 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2634 if (!pskb_may_pull(skb, len))
2638 return skb->data += len;
2641 void skb_condense(struct sk_buff *skb);
2644 * skb_headroom - bytes at buffer head
2645 * @skb: buffer to check
2647 * Return the number of bytes of free space at the head of an &sk_buff.
2649 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2651 return skb->data - skb->head;
2655 * skb_tailroom - bytes at buffer end
2656 * @skb: buffer to check
2658 * Return the number of bytes of free space at the tail of an sk_buff
2660 static inline int skb_tailroom(const struct sk_buff *skb)
2662 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2666 * skb_availroom - bytes at buffer end
2667 * @skb: buffer to check
2669 * Return the number of bytes of free space at the tail of an sk_buff
2670 * allocated by sk_stream_alloc()
2672 static inline int skb_availroom(const struct sk_buff *skb)
2674 if (skb_is_nonlinear(skb))
2677 return skb->end - skb->tail - skb->reserved_tailroom;
2681 * skb_reserve - adjust headroom
2682 * @skb: buffer to alter
2683 * @len: bytes to move
2685 * Increase the headroom of an empty &sk_buff by reducing the tail
2686 * room. This is only allowed for an empty buffer.
2688 static inline void skb_reserve(struct sk_buff *skb, int len)
2695 * skb_tailroom_reserve - adjust reserved_tailroom
2696 * @skb: buffer to alter
2697 * @mtu: maximum amount of headlen permitted
2698 * @needed_tailroom: minimum amount of reserved_tailroom
2700 * Set reserved_tailroom so that headlen can be as large as possible but
2701 * not larger than mtu and tailroom cannot be smaller than
2703 * The required headroom should already have been reserved before using
2706 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2707 unsigned int needed_tailroom)
2709 SKB_LINEAR_ASSERT(skb);
2710 if (mtu < skb_tailroom(skb) - needed_tailroom)
2711 /* use at most mtu */
2712 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2714 /* use up to all available space */
2715 skb->reserved_tailroom = needed_tailroom;
2718 #define ENCAP_TYPE_ETHER 0
2719 #define ENCAP_TYPE_IPPROTO 1
2721 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2724 skb->inner_protocol = protocol;
2725 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2728 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2731 skb->inner_ipproto = ipproto;
2732 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2735 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2737 skb->inner_mac_header = skb->mac_header;
2738 skb->inner_network_header = skb->network_header;
2739 skb->inner_transport_header = skb->transport_header;
2742 static inline void skb_reset_mac_len(struct sk_buff *skb)
2744 skb->mac_len = skb->network_header - skb->mac_header;
2747 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2750 return skb->head + skb->inner_transport_header;
2753 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2755 return skb_inner_transport_header(skb) - skb->data;
2758 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2760 skb->inner_transport_header = skb->data - skb->head;
2763 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2766 skb_reset_inner_transport_header(skb);
2767 skb->inner_transport_header += offset;
2770 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2772 return skb->head + skb->inner_network_header;
2775 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2777 skb->inner_network_header = skb->data - skb->head;
2780 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2783 skb_reset_inner_network_header(skb);
2784 skb->inner_network_header += offset;
2787 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2789 return skb->head + skb->inner_mac_header;
2792 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2794 skb->inner_mac_header = skb->data - skb->head;
2797 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2800 skb_reset_inner_mac_header(skb);
2801 skb->inner_mac_header += offset;
2803 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2805 return skb->transport_header != (typeof(skb->transport_header))~0U;
2808 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2810 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
2811 return skb->head + skb->transport_header;
2814 static inline void skb_reset_transport_header(struct sk_buff *skb)
2816 skb->transport_header = skb->data - skb->head;
2819 static inline void skb_set_transport_header(struct sk_buff *skb,
2822 skb_reset_transport_header(skb);
2823 skb->transport_header += offset;
2826 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2828 return skb->head + skb->network_header;
2831 static inline void skb_reset_network_header(struct sk_buff *skb)
2833 skb->network_header = skb->data - skb->head;
2836 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2838 skb_reset_network_header(skb);
2839 skb->network_header += offset;
2842 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2844 return skb->mac_header != (typeof(skb->mac_header))~0U;
2847 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2849 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2850 return skb->head + skb->mac_header;
2853 static inline int skb_mac_offset(const struct sk_buff *skb)
2855 return skb_mac_header(skb) - skb->data;
2858 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2860 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2861 return skb->network_header - skb->mac_header;
2864 static inline void skb_unset_mac_header(struct sk_buff *skb)
2866 skb->mac_header = (typeof(skb->mac_header))~0U;
2869 static inline void skb_reset_mac_header(struct sk_buff *skb)
2871 skb->mac_header = skb->data - skb->head;
2874 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2876 skb_reset_mac_header(skb);
2877 skb->mac_header += offset;
2880 static inline void skb_pop_mac_header(struct sk_buff *skb)
2882 skb->mac_header = skb->network_header;
2885 static inline void skb_probe_transport_header(struct sk_buff *skb)
2887 struct flow_keys_basic keys;
2889 if (skb_transport_header_was_set(skb))
2892 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2894 skb_set_transport_header(skb, keys.control.thoff);
2897 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2899 if (skb_mac_header_was_set(skb)) {
2900 const unsigned char *old_mac = skb_mac_header(skb);
2902 skb_set_mac_header(skb, -skb->mac_len);
2903 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2907 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2909 return skb->csum_start - skb_headroom(skb);
2912 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2914 return skb->head + skb->csum_start;
2917 static inline int skb_transport_offset(const struct sk_buff *skb)
2919 return skb_transport_header(skb) - skb->data;
2922 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2924 return skb->transport_header - skb->network_header;
2927 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2929 return skb->inner_transport_header - skb->inner_network_header;
2932 static inline int skb_network_offset(const struct sk_buff *skb)
2934 return skb_network_header(skb) - skb->data;
2937 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2939 return skb_inner_network_header(skb) - skb->data;
2942 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2944 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2948 * CPUs often take a performance hit when accessing unaligned memory
2949 * locations. The actual performance hit varies, it can be small if the
2950 * hardware handles it or large if we have to take an exception and fix it
2953 * Since an ethernet header is 14 bytes network drivers often end up with
2954 * the IP header at an unaligned offset. The IP header can be aligned by
2955 * shifting the start of the packet by 2 bytes. Drivers should do this
2958 * skb_reserve(skb, NET_IP_ALIGN);
2960 * The downside to this alignment of the IP header is that the DMA is now
2961 * unaligned. On some architectures the cost of an unaligned DMA is high
2962 * and this cost outweighs the gains made by aligning the IP header.
2964 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2967 #ifndef NET_IP_ALIGN
2968 #define NET_IP_ALIGN 2
2972 * The networking layer reserves some headroom in skb data (via
2973 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2974 * the header has to grow. In the default case, if the header has to grow
2975 * 32 bytes or less we avoid the reallocation.
2977 * Unfortunately this headroom changes the DMA alignment of the resulting
2978 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2979 * on some architectures. An architecture can override this value,
2980 * perhaps setting it to a cacheline in size (since that will maintain
2981 * cacheline alignment of the DMA). It must be a power of 2.
2983 * Various parts of the networking layer expect at least 32 bytes of
2984 * headroom, you should not reduce this.
2986 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2987 * to reduce average number of cache lines per packet.
2988 * get_rps_cpu() for example only access one 64 bytes aligned block :
2989 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2992 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2995 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2997 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2999 if (WARN_ON(skb_is_nonlinear(skb)))
3002 skb_set_tail_pointer(skb, len);
3005 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
3007 __skb_set_length(skb, len);
3010 void skb_trim(struct sk_buff *skb, unsigned int len);
3012 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
3015 return ___pskb_trim(skb, len);
3016 __skb_trim(skb, len);
3020 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
3022 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
3026 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
3027 * @skb: buffer to alter
3030 * This is identical to pskb_trim except that the caller knows that
3031 * the skb is not cloned so we should never get an error due to out-
3034 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3036 int err = pskb_trim(skb, len);
3040 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3042 unsigned int diff = len - skb->len;
3044 if (skb_tailroom(skb) < diff) {
3045 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3050 __skb_set_length(skb, len);
3055 * skb_orphan - orphan a buffer
3056 * @skb: buffer to orphan
3058 * If a buffer currently has an owner then we call the owner's
3059 * destructor function and make the @skb unowned. The buffer continues
3060 * to exist but is no longer charged to its former owner.
3062 static inline void skb_orphan(struct sk_buff *skb)
3064 if (skb->destructor) {
3065 skb->destructor(skb);
3066 skb->destructor = NULL;
3074 * skb_orphan_frags - orphan the frags contained in a buffer
3075 * @skb: buffer to orphan frags from
3076 * @gfp_mask: allocation mask for replacement pages
3078 * For each frag in the SKB which needs a destructor (i.e. has an
3079 * owner) create a copy of that frag and release the original
3080 * page by calling the destructor.
3082 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3084 if (likely(!skb_zcopy(skb)))
3086 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN)
3088 return skb_copy_ubufs(skb, gfp_mask);
3091 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3092 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3094 if (likely(!skb_zcopy(skb)))
3096 return skb_copy_ubufs(skb, gfp_mask);
3100 * __skb_queue_purge - empty a list
3101 * @list: list to empty
3103 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3104 * the list and one reference dropped. This function does not take the
3105 * list lock and the caller must hold the relevant locks to use it.
3107 static inline void __skb_queue_purge(struct sk_buff_head *list)
3109 struct sk_buff *skb;
3110 while ((skb = __skb_dequeue(list)) != NULL)
3113 void skb_queue_purge(struct sk_buff_head *list);
3115 unsigned int skb_rbtree_purge(struct rb_root *root);
3117 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3120 * netdev_alloc_frag - allocate a page fragment
3121 * @fragsz: fragment size
3123 * Allocates a frag from a page for receive buffer.
3124 * Uses GFP_ATOMIC allocations.
3126 static inline void *netdev_alloc_frag(unsigned int fragsz)
3128 return __netdev_alloc_frag_align(fragsz, ~0u);
3131 static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3134 WARN_ON_ONCE(!is_power_of_2(align));
3135 return __netdev_alloc_frag_align(fragsz, -align);
3138 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3142 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3143 * @dev: network device to receive on
3144 * @length: length to allocate
3146 * Allocate a new &sk_buff and assign it a usage count of one. The
3147 * buffer has unspecified headroom built in. Users should allocate
3148 * the headroom they think they need without accounting for the
3149 * built in space. The built in space is used for optimisations.
3151 * %NULL is returned if there is no free memory. Although this function
3152 * allocates memory it can be called from an interrupt.
3154 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3155 unsigned int length)
3157 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3160 /* legacy helper around __netdev_alloc_skb() */
3161 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3164 return __netdev_alloc_skb(NULL, length, gfp_mask);
3167 /* legacy helper around netdev_alloc_skb() */
3168 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3170 return netdev_alloc_skb(NULL, length);
3174 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3175 unsigned int length, gfp_t gfp)
3177 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3179 if (NET_IP_ALIGN && skb)
3180 skb_reserve(skb, NET_IP_ALIGN);
3184 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3185 unsigned int length)
3187 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3190 static inline void skb_free_frag(void *addr)
3192 page_frag_free(addr);
3195 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3197 static inline void *napi_alloc_frag(unsigned int fragsz)
3199 return __napi_alloc_frag_align(fragsz, ~0u);
3202 static inline void *napi_alloc_frag_align(unsigned int fragsz,
3205 WARN_ON_ONCE(!is_power_of_2(align));
3206 return __napi_alloc_frag_align(fragsz, -align);
3209 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
3210 unsigned int length, gfp_t gfp_mask);
3211 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
3212 unsigned int length)
3214 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
3216 void napi_consume_skb(struct sk_buff *skb, int budget);
3218 void napi_skb_free_stolen_head(struct sk_buff *skb);
3219 void __kfree_skb_defer(struct sk_buff *skb);
3222 * __dev_alloc_pages - allocate page for network Rx
3223 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3224 * @order: size of the allocation
3226 * Allocate a new page.
3228 * %NULL is returned if there is no free memory.
3230 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
3233 /* This piece of code contains several assumptions.
3234 * 1. This is for device Rx, therefor a cold page is preferred.
3235 * 2. The expectation is the user wants a compound page.
3236 * 3. If requesting a order 0 page it will not be compound
3237 * due to the check to see if order has a value in prep_new_page
3238 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3239 * code in gfp_to_alloc_flags that should be enforcing this.
3241 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3243 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
3246 static inline struct page *dev_alloc_pages(unsigned int order)
3248 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
3252 * __dev_alloc_page - allocate a page for network Rx
3253 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3255 * Allocate a new page.
3257 * %NULL is returned if there is no free memory.
3259 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
3261 return __dev_alloc_pages(gfp_mask, 0);
3264 static inline struct page *dev_alloc_page(void)
3266 return dev_alloc_pages(0);
3270 * dev_page_is_reusable - check whether a page can be reused for network Rx
3271 * @page: the page to test
3273 * A page shouldn't be considered for reusing/recycling if it was allocated
3274 * under memory pressure or at a distant memory node.
3276 * Returns false if this page should be returned to page allocator, true
3279 static inline bool dev_page_is_reusable(const struct page *page)
3281 return likely(page_to_nid(page) == numa_mem_id() &&
3282 !page_is_pfmemalloc(page));
3286 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3287 * @page: The page that was allocated from skb_alloc_page
3288 * @skb: The skb that may need pfmemalloc set
3290 static inline void skb_propagate_pfmemalloc(const struct page *page,
3291 struct sk_buff *skb)
3293 if (page_is_pfmemalloc(page))
3294 skb->pfmemalloc = true;
3298 * skb_frag_off() - Returns the offset of a skb fragment
3299 * @frag: the paged fragment
3301 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3303 return frag->bv_offset;
3307 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3308 * @frag: skb fragment
3309 * @delta: value to add
3311 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3313 frag->bv_offset += delta;
3317 * skb_frag_off_set() - Sets the offset of a skb fragment
3318 * @frag: skb fragment
3319 * @offset: offset of fragment
3321 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3323 frag->bv_offset = offset;
3327 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3328 * @fragto: skb fragment where offset is set
3329 * @fragfrom: skb fragment offset is copied from
3331 static inline void skb_frag_off_copy(skb_frag_t *fragto,
3332 const skb_frag_t *fragfrom)
3334 fragto->bv_offset = fragfrom->bv_offset;
3338 * skb_frag_page - retrieve the page referred to by a paged fragment
3339 * @frag: the paged fragment
3341 * Returns the &struct page associated with @frag.
3343 static inline struct page *skb_frag_page(const skb_frag_t *frag)
3345 return frag->bv_page;
3349 * __skb_frag_ref - take an addition reference on a paged fragment.
3350 * @frag: the paged fragment
3352 * Takes an additional reference on the paged fragment @frag.
3354 static inline void __skb_frag_ref(skb_frag_t *frag)
3356 get_page(skb_frag_page(frag));
3360 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
3362 * @f: the fragment offset.
3364 * Takes an additional reference on the @f'th paged fragment of @skb.
3366 static inline void skb_frag_ref(struct sk_buff *skb, int f)
3368 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
3372 * __skb_frag_unref - release a reference on a paged fragment.
3373 * @frag: the paged fragment
3374 * @recycle: recycle the page if allocated via page_pool
3376 * Releases a reference on the paged fragment @frag
3377 * or recycles the page via the page_pool API.
3379 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle)
3381 struct page *page = skb_frag_page(frag);
3383 #ifdef CONFIG_PAGE_POOL
3384 if (recycle && page_pool_return_skb_page(page))
3391 * skb_frag_unref - release a reference on a paged fragment of an skb.
3393 * @f: the fragment offset
3395 * Releases a reference on the @f'th paged fragment of @skb.
3397 static inline void skb_frag_unref(struct sk_buff *skb, int f)
3399 struct skb_shared_info *shinfo = skb_shinfo(skb);
3401 if (!skb_zcopy_managed(skb))
3402 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle);
3406 * skb_frag_address - gets the address of the data contained in a paged fragment
3407 * @frag: the paged fragment buffer
3409 * Returns the address of the data within @frag. The page must already
3412 static inline void *skb_frag_address(const skb_frag_t *frag)
3414 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3418 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3419 * @frag: the paged fragment buffer
3421 * Returns the address of the data within @frag. Checks that the page
3422 * is mapped and returns %NULL otherwise.
3424 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3426 void *ptr = page_address(skb_frag_page(frag));
3430 return ptr + skb_frag_off(frag);
3434 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3435 * @fragto: skb fragment where page is set
3436 * @fragfrom: skb fragment page is copied from
3438 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3439 const skb_frag_t *fragfrom)
3441 fragto->bv_page = fragfrom->bv_page;
3445 * __skb_frag_set_page - sets the page contained in a paged fragment
3446 * @frag: the paged fragment
3447 * @page: the page to set
3449 * Sets the fragment @frag to contain @page.
3451 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3453 frag->bv_page = page;
3457 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3459 * @f: the fragment offset
3460 * @page: the page to set
3462 * Sets the @f'th fragment of @skb to contain @page.
3464 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3467 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3470 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3473 * skb_frag_dma_map - maps a paged fragment via the DMA API
3474 * @dev: the device to map the fragment to
3475 * @frag: the paged fragment to map
3476 * @offset: the offset within the fragment (starting at the
3477 * fragment's own offset)
3478 * @size: the number of bytes to map
3479 * @dir: the direction of the mapping (``PCI_DMA_*``)
3481 * Maps the page associated with @frag to @device.
3483 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3484 const skb_frag_t *frag,
3485 size_t offset, size_t size,
3486 enum dma_data_direction dir)
3488 return dma_map_page(dev, skb_frag_page(frag),
3489 skb_frag_off(frag) + offset, size, dir);
3492 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3495 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3499 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3502 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3507 * skb_clone_writable - is the header of a clone writable
3508 * @skb: buffer to check
3509 * @len: length up to which to write
3511 * Returns true if modifying the header part of the cloned buffer
3512 * does not requires the data to be copied.
3514 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3516 return !skb_header_cloned(skb) &&
3517 skb_headroom(skb) + len <= skb->hdr_len;
3520 static inline int skb_try_make_writable(struct sk_buff *skb,
3521 unsigned int write_len)
3523 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3524 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3527 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3532 if (headroom > skb_headroom(skb))
3533 delta = headroom - skb_headroom(skb);
3535 if (delta || cloned)
3536 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3542 * skb_cow - copy header of skb when it is required
3543 * @skb: buffer to cow
3544 * @headroom: needed headroom
3546 * If the skb passed lacks sufficient headroom or its data part
3547 * is shared, data is reallocated. If reallocation fails, an error
3548 * is returned and original skb is not changed.
3550 * The result is skb with writable area skb->head...skb->tail
3551 * and at least @headroom of space at head.
3553 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3555 return __skb_cow(skb, headroom, skb_cloned(skb));
3559 * skb_cow_head - skb_cow but only making the head writable
3560 * @skb: buffer to cow
3561 * @headroom: needed headroom
3563 * This function is identical to skb_cow except that we replace the
3564 * skb_cloned check by skb_header_cloned. It should be used when
3565 * you only need to push on some header and do not need to modify
3568 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3570 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3574 * skb_padto - pad an skbuff up to a minimal size
3575 * @skb: buffer to pad
3576 * @len: minimal length
3578 * Pads up a buffer to ensure the trailing bytes exist and are
3579 * blanked. If the buffer already contains sufficient data it
3580 * is untouched. Otherwise it is extended. Returns zero on
3581 * success. The skb is freed on error.
3583 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3585 unsigned int size = skb->len;
3586 if (likely(size >= len))
3588 return skb_pad(skb, len - size);
3592 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3593 * @skb: buffer to pad
3594 * @len: minimal length
3595 * @free_on_error: free buffer on error
3597 * Pads up a buffer to ensure the trailing bytes exist and are
3598 * blanked. If the buffer already contains sufficient data it
3599 * is untouched. Otherwise it is extended. Returns zero on
3600 * success. The skb is freed on error if @free_on_error is true.
3602 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3606 unsigned int size = skb->len;
3608 if (unlikely(size < len)) {
3610 if (__skb_pad(skb, len, free_on_error))
3612 __skb_put(skb, len);
3618 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3619 * @skb: buffer to pad
3620 * @len: minimal length
3622 * Pads up a buffer to ensure the trailing bytes exist and are
3623 * blanked. If the buffer already contains sufficient data it
3624 * is untouched. Otherwise it is extended. Returns zero on
3625 * success. The skb is freed on error.
3627 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3629 return __skb_put_padto(skb, len, true);
3632 static inline int skb_add_data(struct sk_buff *skb,
3633 struct iov_iter *from, int copy)
3635 const int off = skb->len;
3637 if (skb->ip_summed == CHECKSUM_NONE) {
3639 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3641 skb->csum = csum_block_add(skb->csum, csum, off);
3644 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3647 __skb_trim(skb, off);
3651 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3652 const struct page *page, int off)
3657 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3659 return page == skb_frag_page(frag) &&
3660 off == skb_frag_off(frag) + skb_frag_size(frag);
3665 static inline int __skb_linearize(struct sk_buff *skb)
3667 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3671 * skb_linearize - convert paged skb to linear one
3672 * @skb: buffer to linarize
3674 * If there is no free memory -ENOMEM is returned, otherwise zero
3675 * is returned and the old skb data released.
3677 static inline int skb_linearize(struct sk_buff *skb)
3679 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3683 * skb_has_shared_frag - can any frag be overwritten
3684 * @skb: buffer to test
3686 * Return true if the skb has at least one frag that might be modified
3687 * by an external entity (as in vmsplice()/sendfile())
3689 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3691 return skb_is_nonlinear(skb) &&
3692 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3696 * skb_linearize_cow - make sure skb is linear and writable
3697 * @skb: buffer to process
3699 * If there is no free memory -ENOMEM is returned, otherwise zero
3700 * is returned and the old skb data released.
3702 static inline int skb_linearize_cow(struct sk_buff *skb)
3704 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3705 __skb_linearize(skb) : 0;
3708 static __always_inline void
3709 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3712 if (skb->ip_summed == CHECKSUM_COMPLETE)
3713 skb->csum = csum_block_sub(skb->csum,
3714 csum_partial(start, len, 0), off);
3715 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3716 skb_checksum_start_offset(skb) < 0)
3717 skb->ip_summed = CHECKSUM_NONE;
3721 * skb_postpull_rcsum - update checksum for received skb after pull
3722 * @skb: buffer to update
3723 * @start: start of data before pull
3724 * @len: length of data pulled
3726 * After doing a pull on a received packet, you need to call this to
3727 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3728 * CHECKSUM_NONE so that it can be recomputed from scratch.
3730 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3731 const void *start, unsigned int len)
3733 if (skb->ip_summed == CHECKSUM_COMPLETE)
3734 skb->csum = wsum_negate(csum_partial(start, len,
3735 wsum_negate(skb->csum)));
3736 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3737 skb_checksum_start_offset(skb) < 0)
3738 skb->ip_summed = CHECKSUM_NONE;
3741 static __always_inline void
3742 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3745 if (skb->ip_summed == CHECKSUM_COMPLETE)
3746 skb->csum = csum_block_add(skb->csum,
3747 csum_partial(start, len, 0), off);
3751 * skb_postpush_rcsum - update checksum for received skb after push
3752 * @skb: buffer to update
3753 * @start: start of data after push
3754 * @len: length of data pushed
3756 * After doing a push on a received packet, you need to call this to
3757 * update the CHECKSUM_COMPLETE checksum.
3759 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3760 const void *start, unsigned int len)
3762 __skb_postpush_rcsum(skb, start, len, 0);
3765 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3768 * skb_push_rcsum - push skb and update receive checksum
3769 * @skb: buffer to update
3770 * @len: length of data pulled
3772 * This function performs an skb_push on the packet and updates
3773 * the CHECKSUM_COMPLETE checksum. It should be used on
3774 * receive path processing instead of skb_push unless you know
3775 * that the checksum difference is zero (e.g., a valid IP header)
3776 * or you are setting ip_summed to CHECKSUM_NONE.
3778 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3781 skb_postpush_rcsum(skb, skb->data, len);
3785 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3787 * pskb_trim_rcsum - trim received skb and update checksum
3788 * @skb: buffer to trim
3791 * This is exactly the same as pskb_trim except that it ensures the
3792 * checksum of received packets are still valid after the operation.
3793 * It can change skb pointers.
3796 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3798 if (likely(len >= skb->len))
3800 return pskb_trim_rcsum_slow(skb, len);
3803 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3805 if (skb->ip_summed == CHECKSUM_COMPLETE)
3806 skb->ip_summed = CHECKSUM_NONE;
3807 __skb_trim(skb, len);
3811 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3813 if (skb->ip_summed == CHECKSUM_COMPLETE)
3814 skb->ip_summed = CHECKSUM_NONE;
3815 return __skb_grow(skb, len);
3818 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3819 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3820 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3821 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3822 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3824 #define skb_queue_walk(queue, skb) \
3825 for (skb = (queue)->next; \
3826 skb != (struct sk_buff *)(queue); \
3829 #define skb_queue_walk_safe(queue, skb, tmp) \
3830 for (skb = (queue)->next, tmp = skb->next; \
3831 skb != (struct sk_buff *)(queue); \
3832 skb = tmp, tmp = skb->next)
3834 #define skb_queue_walk_from(queue, skb) \
3835 for (; skb != (struct sk_buff *)(queue); \
3838 #define skb_rbtree_walk(skb, root) \
3839 for (skb = skb_rb_first(root); skb != NULL; \
3840 skb = skb_rb_next(skb))
3842 #define skb_rbtree_walk_from(skb) \
3843 for (; skb != NULL; \
3844 skb = skb_rb_next(skb))
3846 #define skb_rbtree_walk_from_safe(skb, tmp) \
3847 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3850 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3851 for (tmp = skb->next; \
3852 skb != (struct sk_buff *)(queue); \
3853 skb = tmp, tmp = skb->next)
3855 #define skb_queue_reverse_walk(queue, skb) \
3856 for (skb = (queue)->prev; \
3857 skb != (struct sk_buff *)(queue); \
3860 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3861 for (skb = (queue)->prev, tmp = skb->prev; \
3862 skb != (struct sk_buff *)(queue); \
3863 skb = tmp, tmp = skb->prev)
3865 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3866 for (tmp = skb->prev; \
3867 skb != (struct sk_buff *)(queue); \
3868 skb = tmp, tmp = skb->prev)
3870 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3872 return skb_shinfo(skb)->frag_list != NULL;
3875 static inline void skb_frag_list_init(struct sk_buff *skb)
3877 skb_shinfo(skb)->frag_list = NULL;
3880 #define skb_walk_frags(skb, iter) \
3881 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3884 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3885 int *err, long *timeo_p,
3886 const struct sk_buff *skb);
3887 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3888 struct sk_buff_head *queue,
3891 struct sk_buff **last);
3892 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
3893 struct sk_buff_head *queue,
3894 unsigned int flags, int *off, int *err,
3895 struct sk_buff **last);
3896 struct sk_buff *__skb_recv_datagram(struct sock *sk,
3897 struct sk_buff_head *sk_queue,
3898 unsigned int flags, int *off, int *err);
3899 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
3900 __poll_t datagram_poll(struct file *file, struct socket *sock,
3901 struct poll_table_struct *wait);
3902 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3903 struct iov_iter *to, int size);
3904 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3905 struct msghdr *msg, int size)
3907 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3909 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3910 struct msghdr *msg);
3911 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3912 struct iov_iter *to, int len,
3913 struct ahash_request *hash);
3914 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3915 struct iov_iter *from, int len);
3916 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3917 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3918 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3919 static inline void skb_free_datagram_locked(struct sock *sk,
3920 struct sk_buff *skb)
3922 __skb_free_datagram_locked(sk, skb, 0);
3924 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3925 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3926 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3927 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3929 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3930 struct pipe_inode_info *pipe, unsigned int len,
3931 unsigned int flags);
3932 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3934 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3935 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3936 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3937 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3939 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3940 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3941 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3942 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3943 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3944 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3945 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
3946 unsigned int offset);
3947 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3948 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
3949 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3950 int skb_vlan_pop(struct sk_buff *skb);
3951 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3952 int skb_eth_pop(struct sk_buff *skb);
3953 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
3954 const unsigned char *src);
3955 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
3956 int mac_len, bool ethernet);
3957 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
3959 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3960 int skb_mpls_dec_ttl(struct sk_buff *skb);
3961 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3964 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3966 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3969 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3971 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3974 struct skb_checksum_ops {
3975 __wsum (*update)(const void *mem, int len, __wsum wsum);
3976 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3979 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3981 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3982 __wsum csum, const struct skb_checksum_ops *ops);
3983 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3986 static inline void * __must_check
3987 __skb_header_pointer(const struct sk_buff *skb, int offset, int len,
3988 const void *data, int hlen, void *buffer)
3990 if (likely(hlen - offset >= len))
3991 return (void *)data + offset;
3993 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
3999 static inline void * __must_check
4000 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
4002 return __skb_header_pointer(skb, offset, len, skb->data,
4003 skb_headlen(skb), buffer);
4007 * skb_needs_linearize - check if we need to linearize a given skb
4008 * depending on the given device features.
4009 * @skb: socket buffer to check
4010 * @features: net device features
4012 * Returns true if either:
4013 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
4014 * 2. skb is fragmented and the device does not support SG.
4016 static inline bool skb_needs_linearize(struct sk_buff *skb,
4017 netdev_features_t features)
4019 return skb_is_nonlinear(skb) &&
4020 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
4021 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
4024 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
4026 const unsigned int len)
4028 memcpy(to, skb->data, len);
4031 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4032 const int offset, void *to,
4033 const unsigned int len)
4035 memcpy(to, skb->data + offset, len);
4038 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4040 const unsigned int len)
4042 memcpy(skb->data, from, len);
4045 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4048 const unsigned int len)
4050 memcpy(skb->data + offset, from, len);
4053 void skb_init(void);
4055 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4061 * skb_get_timestamp - get timestamp from a skb
4062 * @skb: skb to get stamp from
4063 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4065 * Timestamps are stored in the skb as offsets to a base timestamp.
4066 * This function converts the offset back to a struct timeval and stores
4069 static inline void skb_get_timestamp(const struct sk_buff *skb,
4070 struct __kernel_old_timeval *stamp)
4072 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4075 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4076 struct __kernel_sock_timeval *stamp)
4078 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4080 stamp->tv_sec = ts.tv_sec;
4081 stamp->tv_usec = ts.tv_nsec / 1000;
4084 static inline void skb_get_timestampns(const struct sk_buff *skb,
4085 struct __kernel_old_timespec *stamp)
4087 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4089 stamp->tv_sec = ts.tv_sec;
4090 stamp->tv_nsec = ts.tv_nsec;
4093 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4094 struct __kernel_timespec *stamp)
4096 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4098 stamp->tv_sec = ts.tv_sec;
4099 stamp->tv_nsec = ts.tv_nsec;
4102 static inline void __net_timestamp(struct sk_buff *skb)
4104 skb->tstamp = ktime_get_real();
4105 skb->mono_delivery_time = 0;
4108 static inline ktime_t net_timedelta(ktime_t t)
4110 return ktime_sub(ktime_get_real(), t);
4113 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4117 skb->mono_delivery_time = kt && mono;
4120 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4122 /* It is used in the ingress path to clear the delivery_time.
4123 * If needed, set the skb->tstamp to the (rcv) timestamp.
4125 static inline void skb_clear_delivery_time(struct sk_buff *skb)
4127 if (skb->mono_delivery_time) {
4128 skb->mono_delivery_time = 0;
4129 if (static_branch_unlikely(&netstamp_needed_key))
4130 skb->tstamp = ktime_get_real();
4136 static inline void skb_clear_tstamp(struct sk_buff *skb)
4138 if (skb->mono_delivery_time)
4144 static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4146 if (skb->mono_delivery_time)
4152 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4154 if (!skb->mono_delivery_time && skb->tstamp)
4157 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4158 return ktime_get_real();
4163 static inline u8 skb_metadata_len(const struct sk_buff *skb)
4165 return skb_shinfo(skb)->meta_len;
4168 static inline void *skb_metadata_end(const struct sk_buff *skb)
4170 return skb_mac_header(skb);
4173 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4174 const struct sk_buff *skb_b,
4177 const void *a = skb_metadata_end(skb_a);
4178 const void *b = skb_metadata_end(skb_b);
4179 /* Using more efficient varaiant than plain call to memcmp(). */
4180 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
4184 #define __it(x, op) (x -= sizeof(u##op))
4185 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4186 case 32: diffs |= __it_diff(a, b, 64);
4188 case 24: diffs |= __it_diff(a, b, 64);
4190 case 16: diffs |= __it_diff(a, b, 64);
4192 case 8: diffs |= __it_diff(a, b, 64);
4194 case 28: diffs |= __it_diff(a, b, 64);
4196 case 20: diffs |= __it_diff(a, b, 64);
4198 case 12: diffs |= __it_diff(a, b, 64);
4200 case 4: diffs |= __it_diff(a, b, 32);
4205 return memcmp(a - meta_len, b - meta_len, meta_len);
4209 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4210 const struct sk_buff *skb_b)
4212 u8 len_a = skb_metadata_len(skb_a);
4213 u8 len_b = skb_metadata_len(skb_b);
4215 if (!(len_a | len_b))
4218 return len_a != len_b ?
4219 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4222 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4224 skb_shinfo(skb)->meta_len = meta_len;
4227 static inline void skb_metadata_clear(struct sk_buff *skb)
4229 skb_metadata_set(skb, 0);
4232 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4234 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4236 void skb_clone_tx_timestamp(struct sk_buff *skb);
4237 bool skb_defer_rx_timestamp(struct sk_buff *skb);
4239 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4241 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4245 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4250 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4253 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4255 * PHY drivers may accept clones of transmitted packets for
4256 * timestamping via their phy_driver.txtstamp method. These drivers
4257 * must call this function to return the skb back to the stack with a
4260 * @skb: clone of the original outgoing packet
4261 * @hwtstamps: hardware time stamps
4264 void skb_complete_tx_timestamp(struct sk_buff *skb,
4265 struct skb_shared_hwtstamps *hwtstamps);
4267 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4268 struct skb_shared_hwtstamps *hwtstamps,
4269 struct sock *sk, int tstype);
4272 * skb_tstamp_tx - queue clone of skb with send time stamps
4273 * @orig_skb: the original outgoing packet
4274 * @hwtstamps: hardware time stamps, may be NULL if not available
4276 * If the skb has a socket associated, then this function clones the
4277 * skb (thus sharing the actual data and optional structures), stores
4278 * the optional hardware time stamping information (if non NULL) or
4279 * generates a software time stamp (otherwise), then queues the clone
4280 * to the error queue of the socket. Errors are silently ignored.
4282 void skb_tstamp_tx(struct sk_buff *orig_skb,
4283 struct skb_shared_hwtstamps *hwtstamps);
4286 * skb_tx_timestamp() - Driver hook for transmit timestamping
4288 * Ethernet MAC Drivers should call this function in their hard_xmit()
4289 * function immediately before giving the sk_buff to the MAC hardware.
4291 * Specifically, one should make absolutely sure that this function is
4292 * called before TX completion of this packet can trigger. Otherwise
4293 * the packet could potentially already be freed.
4295 * @skb: A socket buffer.
4297 static inline void skb_tx_timestamp(struct sk_buff *skb)
4299 skb_clone_tx_timestamp(skb);
4300 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4301 skb_tstamp_tx(skb, NULL);
4305 * skb_complete_wifi_ack - deliver skb with wifi status
4307 * @skb: the original outgoing packet
4308 * @acked: ack status
4311 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4313 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4314 __sum16 __skb_checksum_complete(struct sk_buff *skb);
4316 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4318 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4320 (skb->ip_summed == CHECKSUM_PARTIAL &&
4321 skb_checksum_start_offset(skb) >= 0));
4325 * skb_checksum_complete - Calculate checksum of an entire packet
4326 * @skb: packet to process
4328 * This function calculates the checksum over the entire packet plus
4329 * the value of skb->csum. The latter can be used to supply the
4330 * checksum of a pseudo header as used by TCP/UDP. It returns the
4333 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4334 * this function can be used to verify that checksum on received
4335 * packets. In that case the function should return zero if the
4336 * checksum is correct. In particular, this function will return zero
4337 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4338 * hardware has already verified the correctness of the checksum.
4340 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4342 return skb_csum_unnecessary(skb) ?
4343 0 : __skb_checksum_complete(skb);
4346 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4348 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4349 if (skb->csum_level == 0)
4350 skb->ip_summed = CHECKSUM_NONE;
4356 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4358 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4359 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4361 } else if (skb->ip_summed == CHECKSUM_NONE) {
4362 skb->ip_summed = CHECKSUM_UNNECESSARY;
4363 skb->csum_level = 0;
4367 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4369 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4370 skb->ip_summed = CHECKSUM_NONE;
4371 skb->csum_level = 0;
4375 /* Check if we need to perform checksum complete validation.
4377 * Returns true if checksum complete is needed, false otherwise
4378 * (either checksum is unnecessary or zero checksum is allowed).
4380 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4384 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4385 skb->csum_valid = 1;
4386 __skb_decr_checksum_unnecessary(skb);
4393 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4396 #define CHECKSUM_BREAK 76
4398 /* Unset checksum-complete
4400 * Unset checksum complete can be done when packet is being modified
4401 * (uncompressed for instance) and checksum-complete value is
4404 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4406 if (skb->ip_summed == CHECKSUM_COMPLETE)
4407 skb->ip_summed = CHECKSUM_NONE;
4410 /* Validate (init) checksum based on checksum complete.
4413 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4414 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4415 * checksum is stored in skb->csum for use in __skb_checksum_complete
4416 * non-zero: value of invalid checksum
4419 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4423 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4424 if (!csum_fold(csum_add(psum, skb->csum))) {
4425 skb->csum_valid = 1;
4432 if (complete || skb->len <= CHECKSUM_BREAK) {
4435 csum = __skb_checksum_complete(skb);
4436 skb->csum_valid = !csum;
4443 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4448 /* Perform checksum validate (init). Note that this is a macro since we only
4449 * want to calculate the pseudo header which is an input function if necessary.
4450 * First we try to validate without any computation (checksum unnecessary) and
4451 * then calculate based on checksum complete calling the function to compute
4455 * 0: checksum is validated or try to in skb_checksum_complete
4456 * non-zero: value of invalid checksum
4458 #define __skb_checksum_validate(skb, proto, complete, \
4459 zero_okay, check, compute_pseudo) \
4461 __sum16 __ret = 0; \
4462 skb->csum_valid = 0; \
4463 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4464 __ret = __skb_checksum_validate_complete(skb, \
4465 complete, compute_pseudo(skb, proto)); \
4469 #define skb_checksum_init(skb, proto, compute_pseudo) \
4470 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4472 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4473 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4475 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4476 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4478 #define skb_checksum_validate_zero_check(skb, proto, check, \
4480 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4482 #define skb_checksum_simple_validate(skb) \
4483 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4485 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4487 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4490 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4492 skb->csum = ~pseudo;
4493 skb->ip_summed = CHECKSUM_COMPLETE;
4496 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4498 if (__skb_checksum_convert_check(skb)) \
4499 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4502 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4503 u16 start, u16 offset)
4505 skb->ip_summed = CHECKSUM_PARTIAL;
4506 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4507 skb->csum_offset = offset - start;
4510 /* Update skbuf and packet to reflect the remote checksum offload operation.
4511 * When called, ptr indicates the starting point for skb->csum when
4512 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4513 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4515 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4516 int start, int offset, bool nopartial)
4521 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4525 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4526 __skb_checksum_complete(skb);
4527 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4530 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4532 /* Adjust skb->csum since we changed the packet */
4533 skb->csum = csum_add(skb->csum, delta);
4536 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4538 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4539 return (void *)(skb->_nfct & NFCT_PTRMASK);
4545 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4547 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4554 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4556 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4557 skb->slow_gro |= !!nfct;
4562 #ifdef CONFIG_SKB_EXTENSIONS
4564 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4570 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4573 #if IS_ENABLED(CONFIG_MPTCP)
4576 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4579 SKB_EXT_NUM, /* must be last */
4583 * struct skb_ext - sk_buff extensions
4584 * @refcnt: 1 on allocation, deallocated on 0
4585 * @offset: offset to add to @data to obtain extension address
4586 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4587 * @data: start of extension data, variable sized
4589 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4590 * to use 'u8' types while allowing up to 2kb worth of extension data.
4594 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4595 u8 chunks; /* same */
4596 char data[] __aligned(8);
4599 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4600 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4601 struct skb_ext *ext);
4602 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4603 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4604 void __skb_ext_put(struct skb_ext *ext);
4606 static inline void skb_ext_put(struct sk_buff *skb)
4608 if (skb->active_extensions)
4609 __skb_ext_put(skb->extensions);
4612 static inline void __skb_ext_copy(struct sk_buff *dst,
4613 const struct sk_buff *src)
4615 dst->active_extensions = src->active_extensions;
4617 if (src->active_extensions) {
4618 struct skb_ext *ext = src->extensions;
4620 refcount_inc(&ext->refcnt);
4621 dst->extensions = ext;
4625 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4628 __skb_ext_copy(dst, src);
4631 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4633 return !!ext->offset[i];
4636 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4638 return skb->active_extensions & (1 << id);
4641 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4643 if (skb_ext_exist(skb, id))
4644 __skb_ext_del(skb, id);
4647 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4649 if (skb_ext_exist(skb, id)) {
4650 struct skb_ext *ext = skb->extensions;
4652 return (void *)ext + (ext->offset[id] << 3);
4658 static inline void skb_ext_reset(struct sk_buff *skb)
4660 if (unlikely(skb->active_extensions)) {
4661 __skb_ext_put(skb->extensions);
4662 skb->active_extensions = 0;
4666 static inline bool skb_has_extensions(struct sk_buff *skb)
4668 return unlikely(skb->active_extensions);
4671 static inline void skb_ext_put(struct sk_buff *skb) {}
4672 static inline void skb_ext_reset(struct sk_buff *skb) {}
4673 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4674 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4675 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4676 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4677 #endif /* CONFIG_SKB_EXTENSIONS */
4679 static inline void nf_reset_ct(struct sk_buff *skb)
4681 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4682 nf_conntrack_put(skb_nfct(skb));
4687 static inline void nf_reset_trace(struct sk_buff *skb)
4689 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4694 static inline void ipvs_reset(struct sk_buff *skb)
4696 #if IS_ENABLED(CONFIG_IP_VS)
4697 skb->ipvs_property = 0;
4701 /* Note: This doesn't put any conntrack info in dst. */
4702 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4705 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4706 dst->_nfct = src->_nfct;
4707 nf_conntrack_get(skb_nfct(src));
4709 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4711 dst->nf_trace = src->nf_trace;
4715 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4717 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4718 nf_conntrack_put(skb_nfct(dst));
4720 dst->slow_gro = src->slow_gro;
4721 __nf_copy(dst, src, true);
4724 #ifdef CONFIG_NETWORK_SECMARK
4725 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4727 to->secmark = from->secmark;
4730 static inline void skb_init_secmark(struct sk_buff *skb)
4735 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4738 static inline void skb_init_secmark(struct sk_buff *skb)
4742 static inline int secpath_exists(const struct sk_buff *skb)
4745 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4751 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4753 return !skb->destructor &&
4754 !secpath_exists(skb) &&
4756 !skb->_skb_refdst &&
4757 !skb_has_frag_list(skb);
4760 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4762 skb->queue_mapping = queue_mapping;
4765 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4767 return skb->queue_mapping;
4770 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4772 to->queue_mapping = from->queue_mapping;
4775 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4777 skb->queue_mapping = rx_queue + 1;
4780 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4782 return skb->queue_mapping - 1;
4785 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4787 return skb->queue_mapping != 0;
4790 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4792 skb->dst_pending_confirm = val;
4795 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4797 return skb->dst_pending_confirm != 0;
4800 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4803 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4809 /* Keeps track of mac header offset relative to skb->head.
4810 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4811 * For non-tunnel skb it points to skb_mac_header() and for
4812 * tunnel skb it points to outer mac header.
4813 * Keeps track of level of encapsulation of network headers.
4824 #define SKB_GSO_CB_OFFSET 32
4825 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET))
4827 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4829 return (skb_mac_header(inner_skb) - inner_skb->head) -
4830 SKB_GSO_CB(inner_skb)->mac_offset;
4833 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4835 int new_headroom, headroom;
4838 headroom = skb_headroom(skb);
4839 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4843 new_headroom = skb_headroom(skb);
4844 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4848 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4850 /* Do not update partial checksums if remote checksum is enabled. */
4851 if (skb->remcsum_offload)
4854 SKB_GSO_CB(skb)->csum = res;
4855 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4858 /* Compute the checksum for a gso segment. First compute the checksum value
4859 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4860 * then add in skb->csum (checksum from csum_start to end of packet).
4861 * skb->csum and csum_start are then updated to reflect the checksum of the
4862 * resultant packet starting from the transport header-- the resultant checksum
4863 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4866 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4868 unsigned char *csum_start = skb_transport_header(skb);
4869 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4870 __wsum partial = SKB_GSO_CB(skb)->csum;
4872 SKB_GSO_CB(skb)->csum = res;
4873 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4875 return csum_fold(csum_partial(csum_start, plen, partial));
4878 static inline bool skb_is_gso(const struct sk_buff *skb)
4880 return skb_shinfo(skb)->gso_size;
4883 /* Note: Should be called only if skb_is_gso(skb) is true */
4884 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4886 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4889 /* Note: Should be called only if skb_is_gso(skb) is true */
4890 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4892 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4895 /* Note: Should be called only if skb_is_gso(skb) is true */
4896 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4898 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4901 static inline void skb_gso_reset(struct sk_buff *skb)
4903 skb_shinfo(skb)->gso_size = 0;
4904 skb_shinfo(skb)->gso_segs = 0;
4905 skb_shinfo(skb)->gso_type = 0;
4908 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4911 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4913 shinfo->gso_size += increment;
4916 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4919 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4921 shinfo->gso_size -= decrement;
4924 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4926 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4928 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4929 * wanted then gso_type will be set. */
4930 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4932 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4933 unlikely(shinfo->gso_type == 0)) {
4934 __skb_warn_lro_forwarding(skb);
4940 static inline void skb_forward_csum(struct sk_buff *skb)
4942 /* Unfortunately we don't support this one. Any brave souls? */
4943 if (skb->ip_summed == CHECKSUM_COMPLETE)
4944 skb->ip_summed = CHECKSUM_NONE;
4948 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4949 * @skb: skb to check
4951 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4952 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4953 * use this helper, to document places where we make this assertion.
4955 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4957 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE);
4960 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4962 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4963 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4964 unsigned int transport_len,
4965 __sum16(*skb_chkf)(struct sk_buff *skb));
4968 * skb_head_is_locked - Determine if the skb->head is locked down
4969 * @skb: skb to check
4971 * The head on skbs build around a head frag can be removed if they are
4972 * not cloned. This function returns true if the skb head is locked down
4973 * due to either being allocated via kmalloc, or by being a clone with
4974 * multiple references to the head.
4976 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4978 return !skb->head_frag || skb_cloned(skb);
4981 /* Local Checksum Offload.
4982 * Compute outer checksum based on the assumption that the
4983 * inner checksum will be offloaded later.
4984 * See Documentation/networking/checksum-offloads.rst for
4985 * explanation of how this works.
4986 * Fill in outer checksum adjustment (e.g. with sum of outer
4987 * pseudo-header) before calling.
4988 * Also ensure that inner checksum is in linear data area.
4990 static inline __wsum lco_csum(struct sk_buff *skb)
4992 unsigned char *csum_start = skb_checksum_start(skb);
4993 unsigned char *l4_hdr = skb_transport_header(skb);
4996 /* Start with complement of inner checksum adjustment */
4997 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
5000 /* Add in checksum of our headers (incl. outer checksum
5001 * adjustment filled in by caller) and return result.
5003 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
5006 static inline bool skb_is_redirected(const struct sk_buff *skb)
5008 return skb->redirected;
5011 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
5013 skb->redirected = 1;
5014 #ifdef CONFIG_NET_REDIRECT
5015 skb->from_ingress = from_ingress;
5016 if (skb->from_ingress)
5017 skb_clear_tstamp(skb);
5021 static inline void skb_reset_redirect(struct sk_buff *skb)
5023 skb->redirected = 0;
5026 static inline bool skb_csum_is_sctp(struct sk_buff *skb)
5028 return skb->csum_not_inet;
5031 static inline void skb_set_kcov_handle(struct sk_buff *skb,
5032 const u64 kcov_handle)
5035 skb->kcov_handle = kcov_handle;
5039 static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
5042 return skb->kcov_handle;
5048 #ifdef CONFIG_PAGE_POOL
5049 static inline void skb_mark_for_recycle(struct sk_buff *skb)
5051 skb->pp_recycle = 1;
5055 #endif /* __KERNEL__ */
5056 #endif /* _LINUX_SKBUFF_H */