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/dma-mapping.h>
29 #include <linux/netdev_features.h>
30 #include <net/flow_dissector.h>
31 #include <linux/in6.h>
32 #include <linux/if_packet.h>
33 #include <linux/llist.h>
35 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
36 #include <linux/netfilter/nf_conntrack_common.h>
38 #include <net/net_debug.h>
39 #include <net/dropreason-core.h>
40 #include <net/netmem.h>
45 * The interface for checksum offload between the stack and networking drivers
48 * IP checksum related features
49 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
51 * Drivers advertise checksum offload capabilities in the features of a device.
52 * From the stack's point of view these are capabilities offered by the driver.
53 * A driver typically only advertises features that it is capable of offloading
56 * .. flat-table:: Checksum related device features
59 * * - %NETIF_F_HW_CSUM
60 * - The driver (or its device) is able to compute one
61 * IP (one's complement) checksum for any combination
62 * of protocols or protocol layering. The checksum is
63 * computed and set in a packet per the CHECKSUM_PARTIAL
64 * interface (see below).
66 * * - %NETIF_F_IP_CSUM
67 * - Driver (device) is only able to checksum plain
68 * TCP or UDP packets over IPv4. These are specifically
69 * unencapsulated packets of the form IPv4|TCP or
70 * IPv4|UDP where the Protocol field in the IPv4 header
71 * is TCP or UDP. The IPv4 header may contain IP options.
72 * This feature cannot be set in features for a device
73 * with NETIF_F_HW_CSUM also set. This feature is being
74 * DEPRECATED (see below).
76 * * - %NETIF_F_IPV6_CSUM
77 * - Driver (device) is only able to checksum plain
78 * TCP or UDP packets over IPv6. These are specifically
79 * unencapsulated packets of the form IPv6|TCP or
80 * IPv6|UDP where the Next Header field in the IPv6
81 * header is either TCP or UDP. IPv6 extension headers
82 * are not supported with this feature. This feature
83 * cannot be set in features for a device with
84 * NETIF_F_HW_CSUM also set. This feature is being
85 * DEPRECATED (see below).
88 * - Driver (device) performs receive checksum offload.
89 * This flag is only used to disable the RX checksum
90 * feature for a device. The stack will accept receive
91 * checksum indication in packets received on a device
92 * regardless of whether NETIF_F_RXCSUM is set.
94 * Checksumming of received packets by device
95 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
97 * Indication of checksum verification is set in &sk_buff.ip_summed.
98 * Possible values are:
102 * Device did not checksum this packet e.g. due to lack of capabilities.
103 * The packet contains full (though not verified) checksum in packet but
104 * not in skb->csum. Thus, skb->csum is undefined in this case.
106 * - %CHECKSUM_UNNECESSARY
108 * The hardware you're dealing with doesn't calculate the full checksum
109 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums
110 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY
111 * if their checksums are okay. &sk_buff.csum is still undefined in this case
112 * though. A driver or device must never modify the checksum field in the
113 * packet even if checksum is verified.
115 * %CHECKSUM_UNNECESSARY is applicable to following protocols:
117 * - TCP: IPv6 and IPv4.
118 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
119 * zero UDP checksum for either IPv4 or IPv6, the networking stack
120 * may perform further validation in this case.
121 * - GRE: only if the checksum is present in the header.
122 * - SCTP: indicates the CRC in SCTP header has been validated.
123 * - FCOE: indicates the CRC in FC frame has been validated.
125 * &sk_buff.csum_level indicates the number of consecutive checksums found in
126 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY.
127 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
128 * and a device is able to verify the checksums for UDP (possibly zero),
129 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to
130 * two. If the device were only able to verify the UDP checksum and not
131 * GRE, either because it doesn't support GRE checksum or because GRE
132 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
133 * not considered in this case).
135 * - %CHECKSUM_COMPLETE
137 * This is the most generic way. The device supplied checksum of the _whole_
138 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the
139 * hardware doesn't need to parse L3/L4 headers to implement this.
143 * - Even if device supports only some protocols, but is able to produce
144 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
145 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
147 * - %CHECKSUM_PARTIAL
149 * A checksum is set up to be offloaded to a device as described in the
150 * output description for CHECKSUM_PARTIAL. This may occur on a packet
151 * received directly from another Linux OS, e.g., a virtualized Linux kernel
152 * on the same host, or it may be set in the input path in GRO or remote
153 * checksum offload. For the purposes of checksum verification, the checksum
154 * referred to by skb->csum_start + skb->csum_offset and any preceding
155 * checksums in the packet are considered verified. Any checksums in the
156 * packet that are after the checksum being offloaded are not considered to
159 * Checksumming on transmit for non-GSO
160 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
162 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet.
165 * - %CHECKSUM_PARTIAL
167 * The driver is required to checksum the packet as seen by hard_start_xmit()
168 * from &sk_buff.csum_start up to the end, and to record/write the checksum at
169 * offset &sk_buff.csum_start + &sk_buff.csum_offset.
170 * A driver may verify that the
171 * csum_start and csum_offset values are valid values given the length and
172 * offset of the packet, but it should not attempt to validate that the
173 * checksum refers to a legitimate transport layer checksum -- it is the
174 * purview of the stack to validate that csum_start and csum_offset are set
177 * When the stack requests checksum offload for a packet, the driver MUST
178 * ensure that the checksum is set correctly. A driver can either offload the
179 * checksum calculation to the device, or call skb_checksum_help (in the case
180 * that the device does not support offload for a particular checksum).
182 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of
183 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate
184 * checksum offload capability.
185 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based
186 * on network device checksumming capabilities: if a packet does not match
187 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of
188 * &sk_buff.csum_not_inet, see :ref:`crc`)
189 * is called to resolve the checksum.
193 * The skb was already checksummed by the protocol, or a checksum is not
196 * - %CHECKSUM_UNNECESSARY
198 * This has the same meaning as CHECKSUM_NONE for checksum offload on
201 * - %CHECKSUM_COMPLETE
203 * Not used in checksum output. If a driver observes a packet with this value
204 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set.
208 * Non-IP checksum (CRC) offloads
209 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
214 * * - %NETIF_F_SCTP_CRC
215 * - This feature indicates that a device is capable of
216 * offloading the SCTP CRC in a packet. To perform this offload the stack
217 * will set csum_start and csum_offset accordingly, set ip_summed to
218 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication
219 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c.
220 * A driver that supports both IP checksum offload and SCTP CRC32c offload
221 * must verify which offload is configured for a packet by testing the
222 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to
223 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
225 * * - %NETIF_F_FCOE_CRC
226 * - This feature indicates that a device is capable of offloading the FCOE
227 * CRC in a packet. To perform this offload the stack will set ip_summed
228 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset
229 * accordingly. Note that there is no indication in the skbuff that the
230 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports
231 * both IP checksum offload and FCOE CRC offload must verify which offload
232 * is configured for a packet, presumably by inspecting packet headers.
234 * Checksumming on output with GSO
235 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
237 * In the case of a GSO packet (skb_is_gso() is true), checksum offload
238 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
239 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as
240 * part of the GSO operation is implied. If a checksum is being offloaded
241 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and
242 * csum_offset are set to refer to the outermost checksum being offloaded
243 * (two offloaded checksums are possible with UDP encapsulation).
246 /* Don't change this without changing skb_csum_unnecessary! */
247 #define CHECKSUM_NONE 0
248 #define CHECKSUM_UNNECESSARY 1
249 #define CHECKSUM_COMPLETE 2
250 #define CHECKSUM_PARTIAL 3
252 /* Maximum value in skb->csum_level */
253 #define SKB_MAX_CSUM_LEVEL 3
255 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
256 #define SKB_WITH_OVERHEAD(X) \
257 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
259 /* For X bytes available in skb->head, what is the minimal
260 * allocation needed, knowing struct skb_shared_info needs
263 #define SKB_HEAD_ALIGN(X) (SKB_DATA_ALIGN(X) + \
264 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
266 #define SKB_MAX_ORDER(X, ORDER) \
267 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
268 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
269 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
271 /* return minimum truesize of one skb containing X bytes of data */
272 #define SKB_TRUESIZE(X) ((X) + \
273 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
274 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
276 struct ahash_request;
279 struct pipe_inode_info;
287 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
288 struct nf_bridge_info {
290 BRNF_PROTO_UNCHANGED,
297 u8 sabotage_in_done:1;
301 /* always valid & non-NULL from FORWARD on, for physdev match */
302 struct net_device *physoutdev;
304 /* prerouting: detect dnat in orig/reply direction */
306 struct in6_addr ipv6_daddr;
308 /* after prerouting + nat detected: store original source
309 * mac since neigh resolution overwrites it, only used while
310 * skb is out in neigh layer.
312 char neigh_header[8];
317 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
318 /* Chain in tc_skb_ext will be used to share the tc chain with
319 * ovs recirc_id. It will be set to the current chain by tc
320 * and read by ovs to recirc_id.
332 u8 act_miss:1; /* Set if act_miss_cookie is used */
333 u8 l2_miss:1; /* Set by bridge upon FDB or MDB miss */
337 struct sk_buff_head {
338 /* These two members must be first to match sk_buff. */
339 struct_group_tagged(sk_buff_list, list,
340 struct sk_buff *next;
341 struct sk_buff *prev;
350 #ifndef CONFIG_MAX_SKB_FRAGS
351 # define CONFIG_MAX_SKB_FRAGS 17
354 #define MAX_SKB_FRAGS CONFIG_MAX_SKB_FRAGS
356 extern int sysctl_max_skb_frags;
358 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
359 * segment using its current segmentation instead.
361 #define GSO_BY_FRAGS 0xFFFF
363 typedef struct skb_frag {
370 * skb_frag_size() - Returns the size of a skb fragment
371 * @frag: skb fragment
373 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
379 * skb_frag_size_set() - Sets the size of a skb fragment
380 * @frag: skb fragment
381 * @size: size of fragment
383 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
389 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
390 * @frag: skb fragment
391 * @delta: value to add
393 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
399 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
400 * @frag: skb fragment
401 * @delta: value to subtract
403 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
409 * skb_frag_must_loop - Test if %p is a high memory page
410 * @p: fragment's page
412 static inline bool skb_frag_must_loop(struct page *p)
414 #if defined(CONFIG_HIGHMEM)
415 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
422 * skb_frag_foreach_page - loop over pages in a fragment
424 * @f: skb frag to operate on
425 * @f_off: offset from start of f->netmem
426 * @f_len: length from f_off to loop over
427 * @p: (temp var) current page
428 * @p_off: (temp var) offset from start of current page,
429 * non-zero only on first page.
430 * @p_len: (temp var) length in current page,
431 * < PAGE_SIZE only on first and last page.
432 * @copied: (temp var) length so far, excluding current p_len.
434 * A fragment can hold a compound page, in which case per-page
435 * operations, notably kmap_atomic, must be called for each
438 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
439 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
440 p_off = (f_off) & (PAGE_SIZE - 1), \
441 p_len = skb_frag_must_loop(p) ? \
442 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
445 copied += p_len, p++, p_off = 0, \
446 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
449 * struct skb_shared_hwtstamps - hardware time stamps
450 * @hwtstamp: hardware time stamp transformed into duration
451 * since arbitrary point in time
452 * @netdev_data: address/cookie of network device driver used as
453 * reference to actual hardware time stamp
455 * Software time stamps generated by ktime_get_real() are stored in
458 * hwtstamps can only be compared against other hwtstamps from
461 * This structure is attached to packets as part of the
462 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
464 struct skb_shared_hwtstamps {
471 /* Definitions for tx_flags in struct skb_shared_info */
473 /* generate hardware time stamp */
474 SKBTX_HW_TSTAMP = 1 << 0,
476 /* generate software time stamp when queueing packet to NIC */
477 SKBTX_SW_TSTAMP = 1 << 1,
479 /* device driver is going to provide hardware time stamp */
480 SKBTX_IN_PROGRESS = 1 << 2,
482 /* generate hardware time stamp based on cycles if supported */
483 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3,
485 /* generate wifi status information (where possible) */
486 SKBTX_WIFI_STATUS = 1 << 4,
488 /* determine hardware time stamp based on time or cycles */
489 SKBTX_HW_TSTAMP_NETDEV = 1 << 5,
491 /* generate software time stamp when entering packet scheduling */
492 SKBTX_SCHED_TSTAMP = 1 << 6,
495 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
497 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \
498 SKBTX_HW_TSTAMP_USE_CYCLES | \
501 /* Definitions for flags in struct skb_shared_info */
503 /* use zcopy routines */
504 SKBFL_ZEROCOPY_ENABLE = BIT(0),
506 /* This indicates at least one fragment might be overwritten
507 * (as in vmsplice(), sendfile() ...)
508 * If we need to compute a TX checksum, we'll need to copy
509 * all frags to avoid possible bad checksum
511 SKBFL_SHARED_FRAG = BIT(1),
513 /* segment contains only zerocopy data and should not be
514 * charged to the kernel memory.
516 SKBFL_PURE_ZEROCOPY = BIT(2),
518 SKBFL_DONT_ORPHAN = BIT(3),
520 /* page references are managed by the ubuf_info, so it's safe to
521 * use frags only up until ubuf_info is released
523 SKBFL_MANAGED_FRAG_REFS = BIT(4),
526 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
527 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
528 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
531 * The callback notifies userspace to release buffers when skb DMA is done in
532 * lower device, the skb last reference should be 0 when calling this.
533 * The zerocopy_success argument is true if zero copy transmit occurred,
534 * false on data copy or out of memory error caused by data copy attempt.
535 * The ctx field is used to track device context.
536 * The desc field is used to track userspace buffer index.
539 void (*callback)(struct sk_buff *, struct ubuf_info *,
540 bool zerocopy_success);
545 struct ubuf_info_msgzc {
546 struct ubuf_info ubuf;
562 struct user_struct *user;
567 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
568 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \
571 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
572 void mm_unaccount_pinned_pages(struct mmpin *mmp);
574 /* Preserve some data across TX submission and completion.
576 * Note, this state is stored in the driver. Extending the layout
577 * might need some special care.
579 struct xsk_tx_metadata_compl {
583 /* This data is invariant across clones and lives at
584 * the end of the header data, ie. at skb->end.
586 struct skb_shared_info {
591 unsigned short gso_size;
592 /* Warning: this field is not always filled in (UFO)! */
593 unsigned short gso_segs;
594 struct sk_buff *frag_list;
596 struct skb_shared_hwtstamps hwtstamps;
597 struct xsk_tx_metadata_compl xsk_meta;
599 unsigned int gso_type;
603 * Warning : all fields before dataref are cleared in __alloc_skb()
606 unsigned int xdp_frags_size;
608 /* Intermediate layers must ensure that destructor_arg
609 * remains valid until skb destructor */
610 void * destructor_arg;
612 /* must be last field, see pskb_expand_head() */
613 skb_frag_t frags[MAX_SKB_FRAGS];
617 * DOC: dataref and headerless skbs
619 * Transport layers send out clones of payload skbs they hold for
620 * retransmissions. To allow lower layers of the stack to prepend their headers
621 * we split &skb_shared_info.dataref into two halves.
622 * The lower 16 bits count the overall number of references.
623 * The higher 16 bits indicate how many of the references are payload-only.
624 * skb_header_cloned() checks if skb is allowed to add / write the headers.
626 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
627 * (via __skb_header_release()). Any clone created from marked skb will get
628 * &sk_buff.hdr_len populated with the available headroom.
629 * If there's the only clone in existence it's able to modify the headroom
630 * at will. The sequence of calls inside the transport layer is::
634 * __skb_header_release()
636 * // send the clone down the stack
638 * This is not a very generic construct and it depends on the transport layers
639 * doing the right thing. In practice there's usually only one payload-only skb.
640 * Having multiple payload-only skbs with different lengths of hdr_len is not
641 * possible. The payload-only skbs should never leave their owner.
643 #define SKB_DATAREF_SHIFT 16
644 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
648 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
649 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
650 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
654 SKB_GSO_TCPV4 = 1 << 0,
656 /* This indicates the skb is from an untrusted source. */
657 SKB_GSO_DODGY = 1 << 1,
659 /* This indicates the tcp segment has CWR set. */
660 SKB_GSO_TCP_ECN = 1 << 2,
662 SKB_GSO_TCP_FIXEDID = 1 << 3,
664 SKB_GSO_TCPV6 = 1 << 4,
666 SKB_GSO_FCOE = 1 << 5,
668 SKB_GSO_GRE = 1 << 6,
670 SKB_GSO_GRE_CSUM = 1 << 7,
672 SKB_GSO_IPXIP4 = 1 << 8,
674 SKB_GSO_IPXIP6 = 1 << 9,
676 SKB_GSO_UDP_TUNNEL = 1 << 10,
678 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
680 SKB_GSO_PARTIAL = 1 << 12,
682 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
684 SKB_GSO_SCTP = 1 << 14,
686 SKB_GSO_ESP = 1 << 15,
688 SKB_GSO_UDP = 1 << 16,
690 SKB_GSO_UDP_L4 = 1 << 17,
692 SKB_GSO_FRAGLIST = 1 << 18,
695 #if BITS_PER_LONG > 32
696 #define NET_SKBUFF_DATA_USES_OFFSET 1
699 #ifdef NET_SKBUFF_DATA_USES_OFFSET
700 typedef unsigned int sk_buff_data_t;
702 typedef unsigned char *sk_buff_data_t;
706 * DOC: Basic sk_buff geometry
708 * struct sk_buff itself is a metadata structure and does not hold any packet
709 * data. All the data is held in associated buffers.
711 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
714 * - data buffer, containing headers and sometimes payload;
715 * this is the part of the skb operated on by the common helpers
716 * such as skb_put() or skb_pull();
717 * - shared info (struct skb_shared_info) which holds an array of pointers
718 * to read-only data in the (page, offset, length) format.
720 * Optionally &skb_shared_info.frag_list may point to another skb.
722 * Basic diagram may look like this::
727 * ,--------------------------- + head
728 * / ,----------------- + data
729 * / / ,----------- + tail
733 * -----------------------------------------------
734 * | headroom | data | tailroom | skb_shared_info |
735 * -----------------------------------------------
739 * + [page frag] ---------
740 * + frag_list --> | sk_buff |
746 * struct sk_buff - socket buffer
747 * @next: Next buffer in list
748 * @prev: Previous buffer in list
749 * @tstamp: Time we arrived/left
750 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
751 * for retransmit timer
752 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
754 * @ll_node: anchor in an llist (eg socket defer_list)
755 * @sk: Socket we are owned by
756 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
757 * fragmentation management
758 * @dev: Device we arrived on/are leaving by
759 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
760 * @cb: Control buffer. Free for use by every layer. Put private vars here
761 * @_skb_refdst: destination entry (with norefcount bit)
762 * @len: Length of actual data
763 * @data_len: Data length
764 * @mac_len: Length of link layer header
765 * @hdr_len: writable header length of cloned skb
766 * @csum: Checksum (must include start/offset pair)
767 * @csum_start: Offset from skb->head where checksumming should start
768 * @csum_offset: Offset from csum_start where checksum should be stored
769 * @priority: Packet queueing priority
770 * @ignore_df: allow local fragmentation
771 * @cloned: Head may be cloned (check refcnt to be sure)
772 * @ip_summed: Driver fed us an IP checksum
773 * @nohdr: Payload reference only, must not modify header
774 * @pkt_type: Packet class
775 * @fclone: skbuff clone status
776 * @ipvs_property: skbuff is owned by ipvs
777 * @inner_protocol_type: whether the inner protocol is
778 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
779 * @remcsum_offload: remote checksum offload is enabled
780 * @offload_fwd_mark: Packet was L2-forwarded in hardware
781 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
782 * @tc_skip_classify: do not classify packet. set by IFB device
783 * @tc_at_ingress: used within tc_classify to distinguish in/egress
784 * @redirected: packet was redirected by packet classifier
785 * @from_ingress: packet was redirected from the ingress path
786 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
787 * @peeked: this packet has been seen already, so stats have been
788 * done for it, don't do them again
789 * @nf_trace: netfilter packet trace flag
790 * @protocol: Packet protocol from driver
791 * @destructor: Destruct function
792 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
793 * @_sk_redir: socket redirection information for skmsg
794 * @_nfct: Associated connection, if any (with nfctinfo bits)
795 * @skb_iif: ifindex of device we arrived on
796 * @tc_index: Traffic control index
797 * @hash: the packet hash
798 * @queue_mapping: Queue mapping for multiqueue devices
799 * @head_frag: skb was allocated from page fragments,
800 * not allocated by kmalloc() or vmalloc().
801 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
802 * @pp_recycle: mark the packet for recycling instead of freeing (implies
803 * page_pool support on driver)
804 * @active_extensions: active extensions (skb_ext_id types)
805 * @ndisc_nodetype: router type (from link layer)
806 * @ooo_okay: allow the mapping of a socket to a queue to be changed
807 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
809 * @sw_hash: indicates hash was computed in software stack
810 * @wifi_acked_valid: wifi_acked was set
811 * @wifi_acked: whether frame was acked on wifi or not
812 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
813 * @encapsulation: indicates the inner headers in the skbuff are valid
814 * @encap_hdr_csum: software checksum is needed
815 * @csum_valid: checksum is already valid
816 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
817 * @csum_complete_sw: checksum was completed by software
818 * @csum_level: indicates the number of consecutive checksums found in
819 * the packet minus one that have been verified as
820 * CHECKSUM_UNNECESSARY (max 3)
821 * @dst_pending_confirm: need to confirm neighbour
822 * @decrypted: Decrypted SKB
823 * @slow_gro: state present at GRO time, slower prepare step required
824 * @mono_delivery_time: When set, skb->tstamp has the
825 * delivery_time in mono clock base (i.e., EDT) or a clock base chosen
826 * by SO_TXTIME. If zero, skb->tstamp has the (rcv) timestamp at
828 * @napi_id: id of the NAPI struct this skb came from
829 * @sender_cpu: (aka @napi_id) source CPU in XPS
830 * @alloc_cpu: CPU which did the skb allocation.
831 * @secmark: security marking
832 * @mark: Generic packet mark
833 * @reserved_tailroom: (aka @mark) number of bytes of free space available
834 * at the tail of an sk_buff
835 * @vlan_all: vlan fields (proto & tci)
836 * @vlan_proto: vlan encapsulation protocol
837 * @vlan_tci: vlan tag control information
838 * @inner_protocol: Protocol (encapsulation)
839 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
840 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
841 * @inner_transport_header: Inner transport layer header (encapsulation)
842 * @inner_network_header: Network layer header (encapsulation)
843 * @inner_mac_header: Link layer header (encapsulation)
844 * @transport_header: Transport layer header
845 * @network_header: Network layer header
846 * @mac_header: Link layer header
847 * @kcov_handle: KCOV remote handle for remote coverage collection
848 * @tail: Tail pointer
850 * @head: Head of buffer
851 * @data: Data head pointer
852 * @truesize: Buffer size
853 * @users: User count - see {datagram,tcp}.c
854 * @extensions: allocated extensions, valid if active_extensions is nonzero
860 /* These two members must be first to match sk_buff_head. */
861 struct sk_buff *next;
862 struct sk_buff *prev;
865 struct net_device *dev;
866 /* Some protocols might use this space to store information,
867 * while device pointer would be NULL.
868 * UDP receive path is one user.
870 unsigned long dev_scratch;
873 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
874 struct list_head list;
875 struct llist_node ll_node;
880 int ip_defrag_offset;
885 u64 skb_mstamp_ns; /* earliest departure time */
888 * This is the control buffer. It is free to use for every
889 * layer. Please put your private variables there. If you
890 * want to keep them across layers you have to do a skb_clone()
891 * first. This is owned by whoever has the skb queued ATM.
893 char cb[48] __aligned(8);
897 unsigned long _skb_refdst;
898 void (*destructor)(struct sk_buff *skb);
900 struct list_head tcp_tsorted_anchor;
901 #ifdef CONFIG_NET_SOCK_MSG
902 unsigned long _sk_redir;
906 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
914 /* Following fields are _not_ copied in __copy_skb_header()
915 * Note that queue_mapping is here mostly to fill a hole.
919 /* if you move cloned around you also must adapt those constants */
920 #ifdef __BIG_ENDIAN_BITFIELD
921 #define CLONED_MASK (1 << 7)
923 #define CLONED_MASK 1
925 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
928 __u8 __cloned_offset[0];
936 pp_recycle:1; /* page_pool recycle indicator */
937 #ifdef CONFIG_SKB_EXTENSIONS
938 __u8 active_extensions;
941 /* Fields enclosed in headers group are copied
942 * using a single memcpy() in __copy_skb_header()
944 struct_group(headers,
947 __u8 __pkt_type_offset[0];
949 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
951 __u8 dst_pending_confirm:1;
956 __u8 __mono_tc_offset[0];
958 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */
959 #ifdef CONFIG_NET_XGRESS
960 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
961 __u8 tc_skip_classify:1;
963 __u8 remcsum_offload:1;
964 __u8 csum_complete_sw:1;
966 __u8 inner_protocol_type:1;
970 #ifdef CONFIG_WIRELESS
971 __u8 wifi_acked_valid:1;
975 /* Indicates the inner headers are valid in the skbuff. */
976 __u8 encapsulation:1;
977 __u8 encap_hdr_csum:1;
979 #ifdef CONFIG_IPV6_NDISC_NODETYPE
980 __u8 ndisc_nodetype:2;
983 #if IS_ENABLED(CONFIG_IP_VS)
984 __u8 ipvs_property:1;
986 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
989 #ifdef CONFIG_NET_SWITCHDEV
990 __u8 offload_fwd_mark:1;
991 __u8 offload_l3_fwd_mark:1;
994 #ifdef CONFIG_NET_REDIRECT
997 #ifdef CONFIG_NETFILTER_SKIP_EGRESS
998 __u8 nf_skip_egress:1;
1000 #ifdef CONFIG_TLS_DEVICE
1004 #if IS_ENABLED(CONFIG_IP_SCTP)
1005 __u8 csum_not_inet:1;
1008 #if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS)
1009 __u16 tc_index; /* traffic control index */
1031 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
1033 unsigned int napi_id;
1034 unsigned int sender_cpu;
1037 #ifdef CONFIG_NETWORK_SECMARK
1043 __u32 reserved_tailroom;
1047 __be16 inner_protocol;
1051 __u16 inner_transport_header;
1052 __u16 inner_network_header;
1053 __u16 inner_mac_header;
1056 __u16 transport_header;
1057 __u16 network_header;
1064 ); /* end headers group */
1066 /* These elements must be at the end, see alloc_skb() for details. */
1067 sk_buff_data_t tail;
1069 unsigned char *head,
1071 unsigned int truesize;
1074 #ifdef CONFIG_SKB_EXTENSIONS
1075 /* only usable after checking ->active_extensions != 0 */
1076 struct skb_ext *extensions;
1080 /* if you move pkt_type around you also must adapt those constants */
1081 #ifdef __BIG_ENDIAN_BITFIELD
1082 #define PKT_TYPE_MAX (7 << 5)
1084 #define PKT_TYPE_MAX 7
1086 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1088 /* if you move tc_at_ingress or mono_delivery_time
1089 * around, you also must adapt these constants.
1091 #ifdef __BIG_ENDIAN_BITFIELD
1092 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 7)
1093 #define TC_AT_INGRESS_MASK (1 << 6)
1095 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 0)
1096 #define TC_AT_INGRESS_MASK (1 << 1)
1098 #define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset)
1102 * Handling routines are only of interest to the kernel
1105 #define SKB_ALLOC_FCLONE 0x01
1106 #define SKB_ALLOC_RX 0x02
1107 #define SKB_ALLOC_NAPI 0x04
1110 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1113 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1115 return unlikely(skb->pfmemalloc);
1119 * skb might have a dst pointer attached, refcounted or not.
1120 * _skb_refdst low order bit is set if refcount was _not_ taken
1122 #define SKB_DST_NOREF 1UL
1123 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1126 * skb_dst - returns skb dst_entry
1129 * Returns skb dst_entry, regardless of reference taken or not.
1131 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1133 /* If refdst was not refcounted, check we still are in a
1134 * rcu_read_lock section
1136 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1137 !rcu_read_lock_held() &&
1138 !rcu_read_lock_bh_held());
1139 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1143 * skb_dst_set - sets skb dst
1147 * Sets skb dst, assuming a reference was taken on dst and should
1148 * be released by skb_dst_drop()
1150 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1152 skb->slow_gro |= !!dst;
1153 skb->_skb_refdst = (unsigned long)dst;
1157 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1161 * Sets skb dst, assuming a reference was not taken on dst.
1162 * If dst entry is cached, we do not take reference and dst_release
1163 * will be avoided by refdst_drop. If dst entry is not cached, we take
1164 * reference, so that last dst_release can destroy the dst immediately.
1166 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1168 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1169 skb->slow_gro |= !!dst;
1170 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1174 * skb_dst_is_noref - Test if skb dst isn't refcounted
1177 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1179 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1183 * skb_rtable - Returns the skb &rtable
1186 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1188 return (struct rtable *)skb_dst(skb);
1191 /* For mangling skb->pkt_type from user space side from applications
1192 * such as nft, tc, etc, we only allow a conservative subset of
1193 * possible pkt_types to be set.
1195 static inline bool skb_pkt_type_ok(u32 ptype)
1197 return ptype <= PACKET_OTHERHOST;
1201 * skb_napi_id - Returns the skb's NAPI id
1204 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1206 #ifdef CONFIG_NET_RX_BUSY_POLL
1207 return skb->napi_id;
1213 static inline bool skb_wifi_acked_valid(const struct sk_buff *skb)
1215 #ifdef CONFIG_WIRELESS
1216 return skb->wifi_acked_valid;
1223 * skb_unref - decrement the skb's reference count
1226 * Returns true if we can free the skb.
1228 static inline bool skb_unref(struct sk_buff *skb)
1232 if (likely(refcount_read(&skb->users) == 1))
1234 else if (likely(!refcount_dec_and_test(&skb->users)))
1240 static inline bool skb_data_unref(const struct sk_buff *skb,
1241 struct skb_shared_info *shinfo)
1248 bias = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
1250 if (atomic_read(&shinfo->dataref) == bias)
1252 else if (atomic_sub_return(bias, &shinfo->dataref))
1259 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason);
1262 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1263 * @skb: buffer to free
1265 static inline void kfree_skb(struct sk_buff *skb)
1267 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1270 void skb_release_head_state(struct sk_buff *skb);
1271 void kfree_skb_list_reason(struct sk_buff *segs,
1272 enum skb_drop_reason reason);
1273 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1274 void skb_tx_error(struct sk_buff *skb);
1276 static inline void kfree_skb_list(struct sk_buff *segs)
1278 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1281 #ifdef CONFIG_TRACEPOINTS
1282 void consume_skb(struct sk_buff *skb);
1284 static inline void consume_skb(struct sk_buff *skb)
1286 return kfree_skb(skb);
1290 void __consume_stateless_skb(struct sk_buff *skb);
1291 void __kfree_skb(struct sk_buff *skb);
1293 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1294 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1295 bool *fragstolen, int *delta_truesize);
1297 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1299 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1300 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1301 struct sk_buff *build_skb_around(struct sk_buff *skb,
1302 void *data, unsigned int frag_size);
1303 void skb_attempt_defer_free(struct sk_buff *skb);
1305 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1306 struct sk_buff *slab_build_skb(void *data);
1309 * alloc_skb - allocate a network buffer
1310 * @size: size to allocate
1311 * @priority: allocation mask
1313 * This function is a convenient wrapper around __alloc_skb().
1315 static inline struct sk_buff *alloc_skb(unsigned int size,
1318 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1321 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1322 unsigned long data_len,
1326 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1328 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1329 struct sk_buff_fclones {
1330 struct sk_buff skb1;
1332 struct sk_buff skb2;
1334 refcount_t fclone_ref;
1338 * skb_fclone_busy - check if fclone is busy
1342 * Returns true if skb is a fast clone, and its clone is not freed.
1343 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1344 * so we also check that didn't happen.
1346 static inline bool skb_fclone_busy(const struct sock *sk,
1347 const struct sk_buff *skb)
1349 const struct sk_buff_fclones *fclones;
1351 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1353 return skb->fclone == SKB_FCLONE_ORIG &&
1354 refcount_read(&fclones->fclone_ref) > 1 &&
1355 READ_ONCE(fclones->skb2.sk) == sk;
1359 * alloc_skb_fclone - allocate a network buffer from fclone cache
1360 * @size: size to allocate
1361 * @priority: allocation mask
1363 * This function is a convenient wrapper around __alloc_skb().
1365 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1368 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1371 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1372 void skb_headers_offset_update(struct sk_buff *skb, int off);
1373 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1374 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1375 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1376 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1377 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1378 gfp_t gfp_mask, bool fclone);
1379 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1382 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1385 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1386 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1387 unsigned int headroom);
1388 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1389 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1390 int newtailroom, gfp_t priority);
1391 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1392 int offset, int len);
1393 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1394 int offset, int len);
1395 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1396 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1399 * skb_pad - zero pad the tail of an skb
1400 * @skb: buffer to pad
1401 * @pad: space to pad
1403 * Ensure that a buffer is followed by a padding area that is zero
1404 * filled. Used by network drivers which may DMA or transfer data
1405 * beyond the buffer end onto the wire.
1407 * May return error in out of memory cases. The skb is freed on error.
1409 static inline int skb_pad(struct sk_buff *skb, int pad)
1411 return __skb_pad(skb, pad, true);
1413 #define dev_kfree_skb(a) consume_skb(a)
1415 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1416 int offset, size_t size, size_t max_frags);
1418 struct skb_seq_state {
1422 __u32 stepped_offset;
1423 struct sk_buff *root_skb;
1424 struct sk_buff *cur_skb;
1429 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1430 unsigned int to, struct skb_seq_state *st);
1431 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1432 struct skb_seq_state *st);
1433 void skb_abort_seq_read(struct skb_seq_state *st);
1435 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1436 unsigned int to, struct ts_config *config);
1439 * Packet hash types specify the type of hash in skb_set_hash.
1441 * Hash types refer to the protocol layer addresses which are used to
1442 * construct a packet's hash. The hashes are used to differentiate or identify
1443 * flows of the protocol layer for the hash type. Hash types are either
1444 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1446 * Properties of hashes:
1448 * 1) Two packets in different flows have different hash values
1449 * 2) Two packets in the same flow should have the same hash value
1451 * A hash at a higher layer is considered to be more specific. A driver should
1452 * set the most specific hash possible.
1454 * A driver cannot indicate a more specific hash than the layer at which a hash
1455 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1457 * A driver may indicate a hash level which is less specific than the
1458 * actual layer the hash was computed on. For instance, a hash computed
1459 * at L4 may be considered an L3 hash. This should only be done if the
1460 * driver can't unambiguously determine that the HW computed the hash at
1461 * the higher layer. Note that the "should" in the second property above
1464 enum pkt_hash_types {
1465 PKT_HASH_TYPE_NONE, /* Undefined type */
1466 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1467 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1468 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1471 static inline void skb_clear_hash(struct sk_buff *skb)
1478 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1481 skb_clear_hash(skb);
1485 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1487 skb->l4_hash = is_l4;
1488 skb->sw_hash = is_sw;
1493 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1495 /* Used by drivers to set hash from HW */
1496 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1500 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1502 __skb_set_hash(skb, hash, true, is_l4);
1505 void __skb_get_hash(struct sk_buff *skb);
1506 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1507 u32 skb_get_poff(const struct sk_buff *skb);
1508 u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1509 const struct flow_keys_basic *keys, int hlen);
1510 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1511 const void *data, int hlen_proto);
1513 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1514 int thoff, u8 ip_proto)
1516 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1519 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1520 const struct flow_dissector_key *key,
1521 unsigned int key_count);
1523 struct bpf_flow_dissector;
1524 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1525 __be16 proto, int nhoff, int hlen, unsigned int flags);
1527 bool __skb_flow_dissect(const struct net *net,
1528 const struct sk_buff *skb,
1529 struct flow_dissector *flow_dissector,
1530 void *target_container, const void *data,
1531 __be16 proto, int nhoff, int hlen, unsigned int flags);
1533 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1534 struct flow_dissector *flow_dissector,
1535 void *target_container, unsigned int flags)
1537 return __skb_flow_dissect(NULL, skb, flow_dissector,
1538 target_container, NULL, 0, 0, 0, flags);
1541 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1542 struct flow_keys *flow,
1545 memset(flow, 0, sizeof(*flow));
1546 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1547 flow, NULL, 0, 0, 0, flags);
1551 skb_flow_dissect_flow_keys_basic(const struct net *net,
1552 const struct sk_buff *skb,
1553 struct flow_keys_basic *flow,
1554 const void *data, __be16 proto,
1555 int nhoff, int hlen, unsigned int flags)
1557 memset(flow, 0, sizeof(*flow));
1558 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1559 data, proto, nhoff, hlen, flags);
1562 void skb_flow_dissect_meta(const struct sk_buff *skb,
1563 struct flow_dissector *flow_dissector,
1564 void *target_container);
1566 /* Gets a skb connection tracking info, ctinfo map should be a
1567 * map of mapsize to translate enum ip_conntrack_info states
1571 skb_flow_dissect_ct(const struct sk_buff *skb,
1572 struct flow_dissector *flow_dissector,
1573 void *target_container,
1574 u16 *ctinfo_map, size_t mapsize,
1575 bool post_ct, u16 zone);
1577 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1578 struct flow_dissector *flow_dissector,
1579 void *target_container);
1581 void skb_flow_dissect_hash(const struct sk_buff *skb,
1582 struct flow_dissector *flow_dissector,
1583 void *target_container);
1585 static inline __u32 skb_get_hash(struct sk_buff *skb)
1587 if (!skb->l4_hash && !skb->sw_hash)
1588 __skb_get_hash(skb);
1593 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1595 if (!skb->l4_hash && !skb->sw_hash) {
1596 struct flow_keys keys;
1597 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1599 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1605 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1606 const siphash_key_t *perturb);
1608 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1613 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1615 to->hash = from->hash;
1616 to->sw_hash = from->sw_hash;
1617 to->l4_hash = from->l4_hash;
1620 static inline int skb_cmp_decrypted(const struct sk_buff *skb1,
1621 const struct sk_buff *skb2)
1623 #ifdef CONFIG_TLS_DEVICE
1624 return skb2->decrypted - skb1->decrypted;
1630 static inline void skb_copy_decrypted(struct sk_buff *to,
1631 const struct sk_buff *from)
1633 #ifdef CONFIG_TLS_DEVICE
1634 to->decrypted = from->decrypted;
1638 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1639 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1641 return skb->head + skb->end;
1644 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1649 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1654 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1659 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1661 return skb->end - skb->head;
1664 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1666 skb->end = skb->head + offset;
1670 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1671 struct ubuf_info *uarg);
1673 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1675 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1678 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
1679 struct sk_buff *skb, struct iov_iter *from,
1682 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1683 struct msghdr *msg, int len)
1685 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len);
1688 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1689 struct msghdr *msg, int len,
1690 struct ubuf_info *uarg);
1693 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1695 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1697 return &skb_shinfo(skb)->hwtstamps;
1700 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1702 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1704 return is_zcopy ? skb_uarg(skb) : NULL;
1707 static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1709 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1712 static inline bool skb_zcopy_managed(const struct sk_buff *skb)
1714 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS;
1717 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1718 const struct sk_buff *skb2)
1720 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1723 static inline void net_zcopy_get(struct ubuf_info *uarg)
1725 refcount_inc(&uarg->refcnt);
1728 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1730 skb_shinfo(skb)->destructor_arg = uarg;
1731 skb_shinfo(skb)->flags |= uarg->flags;
1734 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1737 if (skb && uarg && !skb_zcopy(skb)) {
1738 if (unlikely(have_ref && *have_ref))
1741 net_zcopy_get(uarg);
1742 skb_zcopy_init(skb, uarg);
1746 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1748 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1749 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1752 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1754 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1757 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1759 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1762 static inline void net_zcopy_put(struct ubuf_info *uarg)
1765 uarg->callback(NULL, uarg, true);
1768 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1771 if (uarg->callback == msg_zerocopy_callback)
1772 msg_zerocopy_put_abort(uarg, have_uref);
1774 net_zcopy_put(uarg);
1778 /* Release a reference on a zerocopy structure */
1779 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1781 struct ubuf_info *uarg = skb_zcopy(skb);
1784 if (!skb_zcopy_is_nouarg(skb))
1785 uarg->callback(skb, uarg, zerocopy_success);
1787 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1791 void __skb_zcopy_downgrade_managed(struct sk_buff *skb);
1793 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb)
1795 if (unlikely(skb_zcopy_managed(skb)))
1796 __skb_zcopy_downgrade_managed(skb);
1799 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1804 static inline void skb_poison_list(struct sk_buff *skb)
1806 #ifdef CONFIG_DEBUG_NET
1807 skb->next = SKB_LIST_POISON_NEXT;
1811 /* Iterate through singly-linked GSO fragments of an skb. */
1812 #define skb_list_walk_safe(first, skb, next_skb) \
1813 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1814 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1816 static inline void skb_list_del_init(struct sk_buff *skb)
1818 __list_del_entry(&skb->list);
1819 skb_mark_not_on_list(skb);
1823 * skb_queue_empty - check if a queue is empty
1826 * Returns true if the queue is empty, false otherwise.
1828 static inline int skb_queue_empty(const struct sk_buff_head *list)
1830 return list->next == (const struct sk_buff *) list;
1834 * skb_queue_empty_lockless - check if a queue is empty
1837 * Returns true if the queue is empty, false otherwise.
1838 * This variant can be used in lockless contexts.
1840 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1842 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1847 * skb_queue_is_last - check if skb is the last entry in the queue
1851 * Returns true if @skb is the last buffer on the list.
1853 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1854 const struct sk_buff *skb)
1856 return skb->next == (const struct sk_buff *) list;
1860 * skb_queue_is_first - check if skb is the first entry in the queue
1864 * Returns true if @skb is the first buffer on the list.
1866 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1867 const struct sk_buff *skb)
1869 return skb->prev == (const struct sk_buff *) list;
1873 * skb_queue_next - return the next packet in the queue
1875 * @skb: current buffer
1877 * Return the next packet in @list after @skb. It is only valid to
1878 * call this if skb_queue_is_last() evaluates to false.
1880 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1881 const struct sk_buff *skb)
1883 /* This BUG_ON may seem severe, but if we just return then we
1884 * are going to dereference garbage.
1886 BUG_ON(skb_queue_is_last(list, skb));
1891 * skb_queue_prev - return the prev packet in the queue
1893 * @skb: current buffer
1895 * Return the prev packet in @list before @skb. It is only valid to
1896 * call this if skb_queue_is_first() evaluates to false.
1898 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1899 const struct sk_buff *skb)
1901 /* This BUG_ON may seem severe, but if we just return then we
1902 * are going to dereference garbage.
1904 BUG_ON(skb_queue_is_first(list, skb));
1909 * skb_get - reference buffer
1910 * @skb: buffer to reference
1912 * Makes another reference to a socket buffer and returns a pointer
1915 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1917 refcount_inc(&skb->users);
1922 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1926 * skb_cloned - is the buffer a clone
1927 * @skb: buffer to check
1929 * Returns true if the buffer was generated with skb_clone() and is
1930 * one of multiple shared copies of the buffer. Cloned buffers are
1931 * shared data so must not be written to under normal circumstances.
1933 static inline int skb_cloned(const struct sk_buff *skb)
1935 return skb->cloned &&
1936 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1939 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1941 might_sleep_if(gfpflags_allow_blocking(pri));
1943 if (skb_cloned(skb))
1944 return pskb_expand_head(skb, 0, 0, pri);
1949 /* This variant of skb_unclone() makes sure skb->truesize
1950 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
1952 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
1953 * when various debugging features are in place.
1955 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
1956 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1958 might_sleep_if(gfpflags_allow_blocking(pri));
1960 if (skb_cloned(skb))
1961 return __skb_unclone_keeptruesize(skb, pri);
1966 * skb_header_cloned - is the header a clone
1967 * @skb: buffer to check
1969 * Returns true if modifying the header part of the buffer requires
1970 * the data to be copied.
1972 static inline int skb_header_cloned(const struct sk_buff *skb)
1979 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1980 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1981 return dataref != 1;
1984 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1986 might_sleep_if(gfpflags_allow_blocking(pri));
1988 if (skb_header_cloned(skb))
1989 return pskb_expand_head(skb, 0, 0, pri);
1995 * __skb_header_release() - allow clones to use the headroom
1996 * @skb: buffer to operate on
1998 * See "DOC: dataref and headerless skbs".
2000 static inline void __skb_header_release(struct sk_buff *skb)
2003 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
2008 * skb_shared - is the buffer shared
2009 * @skb: buffer to check
2011 * Returns true if more than one person has a reference to this
2014 static inline int skb_shared(const struct sk_buff *skb)
2016 return refcount_read(&skb->users) != 1;
2020 * skb_share_check - check if buffer is shared and if so clone it
2021 * @skb: buffer to check
2022 * @pri: priority for memory allocation
2024 * If the buffer is shared the buffer is cloned and the old copy
2025 * drops a reference. A new clone with a single reference is returned.
2026 * If the buffer is not shared the original buffer is returned. When
2027 * being called from interrupt status or with spinlocks held pri must
2030 * NULL is returned on a memory allocation failure.
2032 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
2034 might_sleep_if(gfpflags_allow_blocking(pri));
2035 if (skb_shared(skb)) {
2036 struct sk_buff *nskb = skb_clone(skb, pri);
2048 * Copy shared buffers into a new sk_buff. We effectively do COW on
2049 * packets to handle cases where we have a local reader and forward
2050 * and a couple of other messy ones. The normal one is tcpdumping
2051 * a packet that's being forwarded.
2055 * skb_unshare - make a copy of a shared buffer
2056 * @skb: buffer to check
2057 * @pri: priority for memory allocation
2059 * If the socket buffer is a clone then this function creates a new
2060 * copy of the data, drops a reference count on the old copy and returns
2061 * the new copy with the reference count at 1. If the buffer is not a clone
2062 * the original buffer is returned. When called with a spinlock held or
2063 * from interrupt state @pri must be %GFP_ATOMIC
2065 * %NULL is returned on a memory allocation failure.
2067 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
2070 might_sleep_if(gfpflags_allow_blocking(pri));
2071 if (skb_cloned(skb)) {
2072 struct sk_buff *nskb = skb_copy(skb, pri);
2074 /* Free our shared copy */
2085 * skb_peek - peek at the head of an &sk_buff_head
2086 * @list_: list to peek at
2088 * Peek an &sk_buff. Unlike most other operations you _MUST_
2089 * be careful with this one. A peek leaves the buffer on the
2090 * list and someone else may run off with it. You must hold
2091 * the appropriate locks or have a private queue to do this.
2093 * Returns %NULL for an empty list or a pointer to the head element.
2094 * The reference count is not incremented and the reference is therefore
2095 * volatile. Use with caution.
2097 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2099 struct sk_buff *skb = list_->next;
2101 if (skb == (struct sk_buff *)list_)
2107 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2108 * @list_: list to peek at
2110 * Like skb_peek(), but the caller knows that the list is not empty.
2112 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2118 * skb_peek_next - peek skb following the given one from a queue
2119 * @skb: skb to start from
2120 * @list_: list to peek at
2122 * Returns %NULL when the end of the list is met or a pointer to the
2123 * next element. The reference count is not incremented and the
2124 * reference is therefore volatile. Use with caution.
2126 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2127 const struct sk_buff_head *list_)
2129 struct sk_buff *next = skb->next;
2131 if (next == (struct sk_buff *)list_)
2137 * skb_peek_tail - peek at the tail of an &sk_buff_head
2138 * @list_: list to peek at
2140 * Peek an &sk_buff. Unlike most other operations you _MUST_
2141 * be careful with this one. A peek leaves the buffer on the
2142 * list and someone else may run off with it. You must hold
2143 * the appropriate locks or have a private queue to do this.
2145 * Returns %NULL for an empty list or a pointer to the tail element.
2146 * The reference count is not incremented and the reference is therefore
2147 * volatile. Use with caution.
2149 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2151 struct sk_buff *skb = READ_ONCE(list_->prev);
2153 if (skb == (struct sk_buff *)list_)
2160 * skb_queue_len - get queue length
2161 * @list_: list to measure
2163 * Return the length of an &sk_buff queue.
2165 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2171 * skb_queue_len_lockless - get queue length
2172 * @list_: list to measure
2174 * Return the length of an &sk_buff queue.
2175 * This variant can be used in lockless contexts.
2177 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2179 return READ_ONCE(list_->qlen);
2183 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2184 * @list: queue to initialize
2186 * This initializes only the list and queue length aspects of
2187 * an sk_buff_head object. This allows to initialize the list
2188 * aspects of an sk_buff_head without reinitializing things like
2189 * the spinlock. It can also be used for on-stack sk_buff_head
2190 * objects where the spinlock is known to not be used.
2192 static inline void __skb_queue_head_init(struct sk_buff_head *list)
2194 list->prev = list->next = (struct sk_buff *)list;
2199 * This function creates a split out lock class for each invocation;
2200 * this is needed for now since a whole lot of users of the skb-queue
2201 * infrastructure in drivers have different locking usage (in hardirq)
2202 * than the networking core (in softirq only). In the long run either the
2203 * network layer or drivers should need annotation to consolidate the
2204 * main types of usage into 3 classes.
2206 static inline void skb_queue_head_init(struct sk_buff_head *list)
2208 spin_lock_init(&list->lock);
2209 __skb_queue_head_init(list);
2212 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2213 struct lock_class_key *class)
2215 skb_queue_head_init(list);
2216 lockdep_set_class(&list->lock, class);
2220 * Insert an sk_buff on a list.
2222 * The "__skb_xxxx()" functions are the non-atomic ones that
2223 * can only be called with interrupts disabled.
2225 static inline void __skb_insert(struct sk_buff *newsk,
2226 struct sk_buff *prev, struct sk_buff *next,
2227 struct sk_buff_head *list)
2229 /* See skb_queue_empty_lockless() and skb_peek_tail()
2230 * for the opposite READ_ONCE()
2232 WRITE_ONCE(newsk->next, next);
2233 WRITE_ONCE(newsk->prev, prev);
2234 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2235 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2236 WRITE_ONCE(list->qlen, list->qlen + 1);
2239 static inline void __skb_queue_splice(const struct sk_buff_head *list,
2240 struct sk_buff *prev,
2241 struct sk_buff *next)
2243 struct sk_buff *first = list->next;
2244 struct sk_buff *last = list->prev;
2246 WRITE_ONCE(first->prev, prev);
2247 WRITE_ONCE(prev->next, first);
2249 WRITE_ONCE(last->next, next);
2250 WRITE_ONCE(next->prev, last);
2254 * skb_queue_splice - join two skb lists, this is designed for stacks
2255 * @list: the new list to add
2256 * @head: the place to add it in the first list
2258 static inline void skb_queue_splice(const struct sk_buff_head *list,
2259 struct sk_buff_head *head)
2261 if (!skb_queue_empty(list)) {
2262 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2263 head->qlen += list->qlen;
2268 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2269 * @list: the new list to add
2270 * @head: the place to add it in the first list
2272 * The list at @list is reinitialised
2274 static inline void skb_queue_splice_init(struct sk_buff_head *list,
2275 struct sk_buff_head *head)
2277 if (!skb_queue_empty(list)) {
2278 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2279 head->qlen += list->qlen;
2280 __skb_queue_head_init(list);
2285 * skb_queue_splice_tail - join two skb lists, each list being a queue
2286 * @list: the new list to add
2287 * @head: the place to add it in the first list
2289 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2290 struct sk_buff_head *head)
2292 if (!skb_queue_empty(list)) {
2293 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2294 head->qlen += list->qlen;
2299 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2300 * @list: the new list to add
2301 * @head: the place to add it in the first list
2303 * Each of the lists is a queue.
2304 * The list at @list is reinitialised
2306 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2307 struct sk_buff_head *head)
2309 if (!skb_queue_empty(list)) {
2310 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2311 head->qlen += list->qlen;
2312 __skb_queue_head_init(list);
2317 * __skb_queue_after - queue a buffer at the list head
2318 * @list: list to use
2319 * @prev: place after this buffer
2320 * @newsk: buffer to queue
2322 * Queue a buffer int the middle of a list. This function takes no locks
2323 * and you must therefore hold required locks before calling it.
2325 * A buffer cannot be placed on two lists at the same time.
2327 static inline void __skb_queue_after(struct sk_buff_head *list,
2328 struct sk_buff *prev,
2329 struct sk_buff *newsk)
2331 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2334 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2335 struct sk_buff_head *list);
2337 static inline void __skb_queue_before(struct sk_buff_head *list,
2338 struct sk_buff *next,
2339 struct sk_buff *newsk)
2341 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2345 * __skb_queue_head - queue a buffer at the list head
2346 * @list: list to use
2347 * @newsk: buffer to queue
2349 * Queue a buffer at the start of a list. This function takes no locks
2350 * and you must therefore hold required locks before calling it.
2352 * A buffer cannot be placed on two lists at the same time.
2354 static inline void __skb_queue_head(struct sk_buff_head *list,
2355 struct sk_buff *newsk)
2357 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2359 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2362 * __skb_queue_tail - queue a buffer at the list tail
2363 * @list: list to use
2364 * @newsk: buffer to queue
2366 * Queue a buffer at the end of a list. This function takes no locks
2367 * and you must therefore hold required locks before calling it.
2369 * A buffer cannot be placed on two lists at the same time.
2371 static inline void __skb_queue_tail(struct sk_buff_head *list,
2372 struct sk_buff *newsk)
2374 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2376 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2379 * remove sk_buff from list. _Must_ be called atomically, and with
2382 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2383 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2385 struct sk_buff *next, *prev;
2387 WRITE_ONCE(list->qlen, list->qlen - 1);
2390 skb->next = skb->prev = NULL;
2391 WRITE_ONCE(next->prev, prev);
2392 WRITE_ONCE(prev->next, next);
2396 * __skb_dequeue - remove from the head of the queue
2397 * @list: list to dequeue from
2399 * Remove the head of the list. This function does not take any locks
2400 * so must be used with appropriate locks held only. The head item is
2401 * returned or %NULL if the list is empty.
2403 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2405 struct sk_buff *skb = skb_peek(list);
2407 __skb_unlink(skb, list);
2410 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2413 * __skb_dequeue_tail - remove from the tail of the queue
2414 * @list: list to dequeue from
2416 * Remove the tail of the list. This function does not take any locks
2417 * so must be used with appropriate locks held only. The tail item is
2418 * returned or %NULL if the list is empty.
2420 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2422 struct sk_buff *skb = skb_peek_tail(list);
2424 __skb_unlink(skb, list);
2427 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2430 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2432 return skb->data_len;
2435 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2437 return skb->len - skb->data_len;
2440 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2442 unsigned int i, len = 0;
2444 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2445 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2449 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2451 return skb_headlen(skb) + __skb_pagelen(skb);
2454 static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag,
2455 netmem_ref netmem, int off,
2458 frag->netmem = netmem;
2460 skb_frag_size_set(frag, size);
2463 static inline void skb_frag_fill_page_desc(skb_frag_t *frag,
2467 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size);
2470 static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo,
2471 int i, netmem_ref netmem,
2474 skb_frag_t *frag = &shinfo->frags[i];
2476 skb_frag_fill_netmem_desc(frag, netmem, off, size);
2479 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo,
2480 int i, struct page *page,
2483 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off,
2488 * skb_len_add - adds a number to len fields of skb
2489 * @skb: buffer to add len to
2490 * @delta: number of bytes to add
2492 static inline void skb_len_add(struct sk_buff *skb, int delta)
2495 skb->data_len += delta;
2496 skb->truesize += delta;
2500 * __skb_fill_netmem_desc - initialise a fragment in an skb
2501 * @skb: buffer containing fragment to be initialised
2502 * @i: fragment index to initialise
2503 * @netmem: the netmem to use for this fragment
2504 * @off: the offset to the data with @page
2505 * @size: the length of the data
2507 * Initialises the @i'th fragment of @skb to point to &size bytes at
2508 * offset @off within @page.
2510 * Does not take any additional reference on the fragment.
2512 static inline void __skb_fill_netmem_desc(struct sk_buff *skb, int i,
2513 netmem_ref netmem, int off, int size)
2515 struct page *page = netmem_to_page(netmem);
2517 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size);
2519 /* Propagate page pfmemalloc to the skb if we can. The problem is
2520 * that not all callers have unique ownership of the page but rely
2521 * on page_is_pfmemalloc doing the right thing(tm).
2523 page = compound_head(page);
2524 if (page_is_pfmemalloc(page))
2525 skb->pfmemalloc = true;
2528 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2529 struct page *page, int off, int size)
2531 __skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size);
2534 static inline void skb_fill_netmem_desc(struct sk_buff *skb, int i,
2535 netmem_ref netmem, int off, int size)
2537 __skb_fill_netmem_desc(skb, i, netmem, off, size);
2538 skb_shinfo(skb)->nr_frags = i + 1;
2542 * skb_fill_page_desc - initialise a paged fragment in an skb
2543 * @skb: buffer containing fragment to be initialised
2544 * @i: paged fragment index to initialise
2545 * @page: the page to use for this fragment
2546 * @off: the offset to the data with @page
2547 * @size: the length of the data
2549 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2550 * @skb to point to @size bytes at offset @off within @page. In
2551 * addition updates @skb such that @i is the last fragment.
2553 * Does not take any additional reference on the fragment.
2555 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2556 struct page *page, int off, int size)
2558 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size);
2562 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb
2563 * @skb: buffer containing fragment to be initialised
2564 * @i: paged fragment index to initialise
2565 * @page: the page to use for this fragment
2566 * @off: the offset to the data with @page
2567 * @size: the length of the data
2569 * Variant of skb_fill_page_desc() which does not deal with
2570 * pfmemalloc, if page is not owned by us.
2572 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i,
2573 struct page *page, int off,
2576 struct skb_shared_info *shinfo = skb_shinfo(skb);
2578 __skb_fill_page_desc_noacc(shinfo, i, page, off, size);
2579 shinfo->nr_frags = i + 1;
2582 void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem,
2583 int off, int size, unsigned int truesize);
2585 static inline void skb_add_rx_frag(struct sk_buff *skb, int i,
2586 struct page *page, int off, int size,
2587 unsigned int truesize)
2589 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size,
2593 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2594 unsigned int truesize);
2596 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2598 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2599 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2601 return skb->head + skb->tail;
2604 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2606 skb->tail = skb->data - skb->head;
2609 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2611 skb_reset_tail_pointer(skb);
2612 skb->tail += offset;
2615 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2616 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2621 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2623 skb->tail = skb->data;
2626 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2628 skb->tail = skb->data + offset;
2631 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2633 static inline void skb_assert_len(struct sk_buff *skb)
2635 #ifdef CONFIG_DEBUG_NET
2636 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2637 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2638 #endif /* CONFIG_DEBUG_NET */
2642 * Add data to an sk_buff
2644 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2645 void *skb_put(struct sk_buff *skb, unsigned int len);
2646 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2648 void *tmp = skb_tail_pointer(skb);
2649 SKB_LINEAR_ASSERT(skb);
2655 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2657 void *tmp = __skb_put(skb, len);
2659 memset(tmp, 0, len);
2663 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2666 void *tmp = __skb_put(skb, len);
2668 memcpy(tmp, data, len);
2672 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2674 *(u8 *)__skb_put(skb, 1) = val;
2677 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2679 void *tmp = skb_put(skb, len);
2681 memset(tmp, 0, len);
2686 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2689 void *tmp = skb_put(skb, len);
2691 memcpy(tmp, data, len);
2696 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2698 *(u8 *)skb_put(skb, 1) = val;
2701 void *skb_push(struct sk_buff *skb, unsigned int len);
2702 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2704 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2711 void *skb_pull(struct sk_buff *skb, unsigned int len);
2712 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2714 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2717 if (unlikely(skb->len < skb->data_len)) {
2718 #if defined(CONFIG_DEBUG_NET)
2720 pr_err("__skb_pull(len=%u)\n", len);
2721 skb_dump(KERN_ERR, skb, false);
2725 return skb->data += len;
2728 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2730 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2733 void *skb_pull_data(struct sk_buff *skb, size_t len);
2735 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2737 static inline enum skb_drop_reason
2738 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len)
2740 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2742 if (likely(len <= skb_headlen(skb)))
2743 return SKB_NOT_DROPPED_YET;
2745 if (unlikely(len > skb->len))
2746 return SKB_DROP_REASON_PKT_TOO_SMALL;
2748 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb))))
2749 return SKB_DROP_REASON_NOMEM;
2751 return SKB_NOT_DROPPED_YET;
2754 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2756 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET;
2759 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2761 if (!pskb_may_pull(skb, len))
2765 return skb->data += len;
2768 void skb_condense(struct sk_buff *skb);
2771 * skb_headroom - bytes at buffer head
2772 * @skb: buffer to check
2774 * Return the number of bytes of free space at the head of an &sk_buff.
2776 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2778 return skb->data - skb->head;
2782 * skb_tailroom - bytes at buffer end
2783 * @skb: buffer to check
2785 * Return the number of bytes of free space at the tail of an sk_buff
2787 static inline int skb_tailroom(const struct sk_buff *skb)
2789 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2793 * skb_availroom - bytes at buffer end
2794 * @skb: buffer to check
2796 * Return the number of bytes of free space at the tail of an sk_buff
2797 * allocated by sk_stream_alloc()
2799 static inline int skb_availroom(const struct sk_buff *skb)
2801 if (skb_is_nonlinear(skb))
2804 return skb->end - skb->tail - skb->reserved_tailroom;
2808 * skb_reserve - adjust headroom
2809 * @skb: buffer to alter
2810 * @len: bytes to move
2812 * Increase the headroom of an empty &sk_buff by reducing the tail
2813 * room. This is only allowed for an empty buffer.
2815 static inline void skb_reserve(struct sk_buff *skb, int len)
2822 * skb_tailroom_reserve - adjust reserved_tailroom
2823 * @skb: buffer to alter
2824 * @mtu: maximum amount of headlen permitted
2825 * @needed_tailroom: minimum amount of reserved_tailroom
2827 * Set reserved_tailroom so that headlen can be as large as possible but
2828 * not larger than mtu and tailroom cannot be smaller than
2830 * The required headroom should already have been reserved before using
2833 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2834 unsigned int needed_tailroom)
2836 SKB_LINEAR_ASSERT(skb);
2837 if (mtu < skb_tailroom(skb) - needed_tailroom)
2838 /* use at most mtu */
2839 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2841 /* use up to all available space */
2842 skb->reserved_tailroom = needed_tailroom;
2845 #define ENCAP_TYPE_ETHER 0
2846 #define ENCAP_TYPE_IPPROTO 1
2848 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2851 skb->inner_protocol = protocol;
2852 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2855 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2858 skb->inner_ipproto = ipproto;
2859 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2862 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2864 skb->inner_mac_header = skb->mac_header;
2865 skb->inner_network_header = skb->network_header;
2866 skb->inner_transport_header = skb->transport_header;
2869 static inline void skb_reset_mac_len(struct sk_buff *skb)
2871 skb->mac_len = skb->network_header - skb->mac_header;
2874 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2877 return skb->head + skb->inner_transport_header;
2880 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2882 return skb_inner_transport_header(skb) - skb->data;
2885 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2887 skb->inner_transport_header = skb->data - skb->head;
2890 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2893 skb_reset_inner_transport_header(skb);
2894 skb->inner_transport_header += offset;
2897 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2899 return skb->head + skb->inner_network_header;
2902 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2904 skb->inner_network_header = skb->data - skb->head;
2907 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2910 skb_reset_inner_network_header(skb);
2911 skb->inner_network_header += offset;
2914 static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb)
2916 return skb->inner_network_header > 0;
2919 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2921 return skb->head + skb->inner_mac_header;
2924 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2926 skb->inner_mac_header = skb->data - skb->head;
2929 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2932 skb_reset_inner_mac_header(skb);
2933 skb->inner_mac_header += offset;
2935 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2937 return skb->transport_header != (typeof(skb->transport_header))~0U;
2940 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2942 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
2943 return skb->head + skb->transport_header;
2946 static inline void skb_reset_transport_header(struct sk_buff *skb)
2948 skb->transport_header = skb->data - skb->head;
2951 static inline void skb_set_transport_header(struct sk_buff *skb,
2954 skb_reset_transport_header(skb);
2955 skb->transport_header += offset;
2958 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2960 return skb->head + skb->network_header;
2963 static inline void skb_reset_network_header(struct sk_buff *skb)
2965 skb->network_header = skb->data - skb->head;
2968 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2970 skb_reset_network_header(skb);
2971 skb->network_header += offset;
2974 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2976 return skb->mac_header != (typeof(skb->mac_header))~0U;
2979 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2981 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2982 return skb->head + skb->mac_header;
2985 static inline int skb_mac_offset(const struct sk_buff *skb)
2987 return skb_mac_header(skb) - skb->data;
2990 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2992 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2993 return skb->network_header - skb->mac_header;
2996 static inline void skb_unset_mac_header(struct sk_buff *skb)
2998 skb->mac_header = (typeof(skb->mac_header))~0U;
3001 static inline void skb_reset_mac_header(struct sk_buff *skb)
3003 skb->mac_header = skb->data - skb->head;
3006 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
3008 skb_reset_mac_header(skb);
3009 skb->mac_header += offset;
3012 static inline void skb_pop_mac_header(struct sk_buff *skb)
3014 skb->mac_header = skb->network_header;
3017 static inline void skb_probe_transport_header(struct sk_buff *skb)
3019 struct flow_keys_basic keys;
3021 if (skb_transport_header_was_set(skb))
3024 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
3026 skb_set_transport_header(skb, keys.control.thoff);
3029 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
3031 if (skb_mac_header_was_set(skb)) {
3032 const unsigned char *old_mac = skb_mac_header(skb);
3034 skb_set_mac_header(skb, -skb->mac_len);
3035 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
3039 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
3041 return skb->csum_start - skb_headroom(skb);
3044 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
3046 return skb->head + skb->csum_start;
3049 static inline int skb_transport_offset(const struct sk_buff *skb)
3051 return skb_transport_header(skb) - skb->data;
3054 static inline u32 skb_network_header_len(const struct sk_buff *skb)
3056 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
3057 return skb->transport_header - skb->network_header;
3060 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
3062 return skb->inner_transport_header - skb->inner_network_header;
3065 static inline int skb_network_offset(const struct sk_buff *skb)
3067 return skb_network_header(skb) - skb->data;
3070 static inline int skb_inner_network_offset(const struct sk_buff *skb)
3072 return skb_inner_network_header(skb) - skb->data;
3075 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
3077 return pskb_may_pull(skb, skb_network_offset(skb) + len);
3081 * CPUs often take a performance hit when accessing unaligned memory
3082 * locations. The actual performance hit varies, it can be small if the
3083 * hardware handles it or large if we have to take an exception and fix it
3086 * Since an ethernet header is 14 bytes network drivers often end up with
3087 * the IP header at an unaligned offset. The IP header can be aligned by
3088 * shifting the start of the packet by 2 bytes. Drivers should do this
3091 * skb_reserve(skb, NET_IP_ALIGN);
3093 * The downside to this alignment of the IP header is that the DMA is now
3094 * unaligned. On some architectures the cost of an unaligned DMA is high
3095 * and this cost outweighs the gains made by aligning the IP header.
3097 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
3100 #ifndef NET_IP_ALIGN
3101 #define NET_IP_ALIGN 2
3105 * The networking layer reserves some headroom in skb data (via
3106 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
3107 * the header has to grow. In the default case, if the header has to grow
3108 * 32 bytes or less we avoid the reallocation.
3110 * Unfortunately this headroom changes the DMA alignment of the resulting
3111 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
3112 * on some architectures. An architecture can override this value,
3113 * perhaps setting it to a cacheline in size (since that will maintain
3114 * cacheline alignment of the DMA). It must be a power of 2.
3116 * Various parts of the networking layer expect at least 32 bytes of
3117 * headroom, you should not reduce this.
3119 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
3120 * to reduce average number of cache lines per packet.
3121 * get_rps_cpu() for example only access one 64 bytes aligned block :
3122 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
3125 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
3128 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
3130 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
3132 if (WARN_ON(skb_is_nonlinear(skb)))
3135 skb_set_tail_pointer(skb, len);
3138 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
3140 __skb_set_length(skb, len);
3143 void skb_trim(struct sk_buff *skb, unsigned int len);
3145 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
3148 return ___pskb_trim(skb, len);
3149 __skb_trim(skb, len);
3153 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
3155 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
3159 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
3160 * @skb: buffer to alter
3163 * This is identical to pskb_trim except that the caller knows that
3164 * the skb is not cloned so we should never get an error due to out-
3167 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3169 int err = pskb_trim(skb, len);
3173 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3175 unsigned int diff = len - skb->len;
3177 if (skb_tailroom(skb) < diff) {
3178 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3183 __skb_set_length(skb, len);
3188 * skb_orphan - orphan a buffer
3189 * @skb: buffer to orphan
3191 * If a buffer currently has an owner then we call the owner's
3192 * destructor function and make the @skb unowned. The buffer continues
3193 * to exist but is no longer charged to its former owner.
3195 static inline void skb_orphan(struct sk_buff *skb)
3197 if (skb->destructor) {
3198 skb->destructor(skb);
3199 skb->destructor = NULL;
3207 * skb_orphan_frags - orphan the frags contained in a buffer
3208 * @skb: buffer to orphan frags from
3209 * @gfp_mask: allocation mask for replacement pages
3211 * For each frag in the SKB which needs a destructor (i.e. has an
3212 * owner) create a copy of that frag and release the original
3213 * page by calling the destructor.
3215 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3217 if (likely(!skb_zcopy(skb)))
3219 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN)
3221 return skb_copy_ubufs(skb, gfp_mask);
3224 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3225 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3227 if (likely(!skb_zcopy(skb)))
3229 return skb_copy_ubufs(skb, gfp_mask);
3233 * __skb_queue_purge_reason - empty a list
3234 * @list: list to empty
3235 * @reason: drop reason
3237 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3238 * the list and one reference dropped. This function does not take the
3239 * list lock and the caller must hold the relevant locks to use it.
3241 static inline void __skb_queue_purge_reason(struct sk_buff_head *list,
3242 enum skb_drop_reason reason)
3244 struct sk_buff *skb;
3246 while ((skb = __skb_dequeue(list)) != NULL)
3247 kfree_skb_reason(skb, reason);
3250 static inline void __skb_queue_purge(struct sk_buff_head *list)
3252 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE);
3255 void skb_queue_purge_reason(struct sk_buff_head *list,
3256 enum skb_drop_reason reason);
3258 static inline void skb_queue_purge(struct sk_buff_head *list)
3260 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE);
3263 unsigned int skb_rbtree_purge(struct rb_root *root);
3264 void skb_errqueue_purge(struct sk_buff_head *list);
3266 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3269 * netdev_alloc_frag - allocate a page fragment
3270 * @fragsz: fragment size
3272 * Allocates a frag from a page for receive buffer.
3273 * Uses GFP_ATOMIC allocations.
3275 static inline void *netdev_alloc_frag(unsigned int fragsz)
3277 return __netdev_alloc_frag_align(fragsz, ~0u);
3280 static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3283 WARN_ON_ONCE(!is_power_of_2(align));
3284 return __netdev_alloc_frag_align(fragsz, -align);
3287 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3291 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3292 * @dev: network device to receive on
3293 * @length: length to allocate
3295 * Allocate a new &sk_buff and assign it a usage count of one. The
3296 * buffer has unspecified headroom built in. Users should allocate
3297 * the headroom they think they need without accounting for the
3298 * built in space. The built in space is used for optimisations.
3300 * %NULL is returned if there is no free memory. Although this function
3301 * allocates memory it can be called from an interrupt.
3303 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3304 unsigned int length)
3306 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3309 /* legacy helper around __netdev_alloc_skb() */
3310 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3313 return __netdev_alloc_skb(NULL, length, gfp_mask);
3316 /* legacy helper around netdev_alloc_skb() */
3317 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3319 return netdev_alloc_skb(NULL, length);
3323 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3324 unsigned int length, gfp_t gfp)
3326 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3328 if (NET_IP_ALIGN && skb)
3329 skb_reserve(skb, NET_IP_ALIGN);
3333 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3334 unsigned int length)
3336 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3339 static inline void skb_free_frag(void *addr)
3341 page_frag_free(addr);
3344 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3346 static inline void *napi_alloc_frag(unsigned int fragsz)
3348 return __napi_alloc_frag_align(fragsz, ~0u);
3351 static inline void *napi_alloc_frag_align(unsigned int fragsz,
3354 WARN_ON_ONCE(!is_power_of_2(align));
3355 return __napi_alloc_frag_align(fragsz, -align);
3358 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
3359 unsigned int length, gfp_t gfp_mask);
3360 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
3361 unsigned int length)
3363 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
3365 void napi_consume_skb(struct sk_buff *skb, int budget);
3367 void napi_skb_free_stolen_head(struct sk_buff *skb);
3368 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason);
3371 * __dev_alloc_pages - allocate page for network Rx
3372 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3373 * @order: size of the allocation
3375 * Allocate a new page.
3377 * %NULL is returned if there is no free memory.
3379 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
3382 /* This piece of code contains several assumptions.
3383 * 1. This is for device Rx, therefore a cold page is preferred.
3384 * 2. The expectation is the user wants a compound page.
3385 * 3. If requesting a order 0 page it will not be compound
3386 * due to the check to see if order has a value in prep_new_page
3387 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3388 * code in gfp_to_alloc_flags that should be enforcing this.
3390 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3392 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
3395 static inline struct page *dev_alloc_pages(unsigned int order)
3397 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
3401 * __dev_alloc_page - allocate a page for network Rx
3402 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3404 * Allocate a new page.
3406 * %NULL is returned if there is no free memory.
3408 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
3410 return __dev_alloc_pages(gfp_mask, 0);
3413 static inline struct page *dev_alloc_page(void)
3415 return dev_alloc_pages(0);
3419 * dev_page_is_reusable - check whether a page can be reused for network Rx
3420 * @page: the page to test
3422 * A page shouldn't be considered for reusing/recycling if it was allocated
3423 * under memory pressure or at a distant memory node.
3425 * Returns false if this page should be returned to page allocator, true
3428 static inline bool dev_page_is_reusable(const struct page *page)
3430 return likely(page_to_nid(page) == numa_mem_id() &&
3431 !page_is_pfmemalloc(page));
3435 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3436 * @page: The page that was allocated from skb_alloc_page
3437 * @skb: The skb that may need pfmemalloc set
3439 static inline void skb_propagate_pfmemalloc(const struct page *page,
3440 struct sk_buff *skb)
3442 if (page_is_pfmemalloc(page))
3443 skb->pfmemalloc = true;
3447 * skb_frag_off() - Returns the offset of a skb fragment
3448 * @frag: the paged fragment
3450 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3452 return frag->offset;
3456 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3457 * @frag: skb fragment
3458 * @delta: value to add
3460 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3462 frag->offset += delta;
3466 * skb_frag_off_set() - Sets the offset of a skb fragment
3467 * @frag: skb fragment
3468 * @offset: offset of fragment
3470 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3472 frag->offset = offset;
3476 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3477 * @fragto: skb fragment where offset is set
3478 * @fragfrom: skb fragment offset is copied from
3480 static inline void skb_frag_off_copy(skb_frag_t *fragto,
3481 const skb_frag_t *fragfrom)
3483 fragto->offset = fragfrom->offset;
3487 * skb_frag_page - retrieve the page referred to by a paged fragment
3488 * @frag: the paged fragment
3490 * Returns the &struct page associated with @frag.
3492 static inline struct page *skb_frag_page(const skb_frag_t *frag)
3494 return netmem_to_page(frag->netmem);
3498 * __skb_frag_ref - take an addition reference on a paged fragment.
3499 * @frag: the paged fragment
3501 * Takes an additional reference on the paged fragment @frag.
3503 static inline void __skb_frag_ref(skb_frag_t *frag)
3505 get_page(skb_frag_page(frag));
3509 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
3511 * @f: the fragment offset.
3513 * Takes an additional reference on the @f'th paged fragment of @skb.
3515 static inline void skb_frag_ref(struct sk_buff *skb, int f)
3517 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
3520 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb,
3521 unsigned int headroom);
3522 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb,
3523 struct bpf_prog *prog);
3524 bool napi_pp_put_page(struct page *page, bool napi_safe);
3527 napi_frag_unref(skb_frag_t *frag, bool recycle, bool napi_safe)
3529 struct page *page = skb_frag_page(frag);
3531 #ifdef CONFIG_PAGE_POOL
3532 if (recycle && napi_pp_put_page(page, napi_safe))
3539 * __skb_frag_unref - release a reference on a paged fragment.
3540 * @frag: the paged fragment
3541 * @recycle: recycle the page if allocated via page_pool
3543 * Releases a reference on the paged fragment @frag
3544 * or recycles the page via the page_pool API.
3546 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle)
3548 napi_frag_unref(frag, recycle, false);
3552 * skb_frag_unref - release a reference on a paged fragment of an skb.
3554 * @f: the fragment offset
3556 * Releases a reference on the @f'th paged fragment of @skb.
3558 static inline void skb_frag_unref(struct sk_buff *skb, int f)
3560 struct skb_shared_info *shinfo = skb_shinfo(skb);
3562 if (!skb_zcopy_managed(skb))
3563 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle);
3567 * skb_frag_address - gets the address of the data contained in a paged fragment
3568 * @frag: the paged fragment buffer
3570 * Returns the address of the data within @frag. The page must already
3573 static inline void *skb_frag_address(const skb_frag_t *frag)
3575 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3579 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3580 * @frag: the paged fragment buffer
3582 * Returns the address of the data within @frag. Checks that the page
3583 * is mapped and returns %NULL otherwise.
3585 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3587 void *ptr = page_address(skb_frag_page(frag));
3591 return ptr + skb_frag_off(frag);
3595 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3596 * @fragto: skb fragment where page is set
3597 * @fragfrom: skb fragment page is copied from
3599 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3600 const skb_frag_t *fragfrom)
3602 fragto->netmem = fragfrom->netmem;
3605 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3608 * skb_frag_dma_map - maps a paged fragment via the DMA API
3609 * @dev: the device to map the fragment to
3610 * @frag: the paged fragment to map
3611 * @offset: the offset within the fragment (starting at the
3612 * fragment's own offset)
3613 * @size: the number of bytes to map
3614 * @dir: the direction of the mapping (``PCI_DMA_*``)
3616 * Maps the page associated with @frag to @device.
3618 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3619 const skb_frag_t *frag,
3620 size_t offset, size_t size,
3621 enum dma_data_direction dir)
3623 return dma_map_page(dev, skb_frag_page(frag),
3624 skb_frag_off(frag) + offset, size, dir);
3627 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3630 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3634 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3637 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3642 * skb_clone_writable - is the header of a clone writable
3643 * @skb: buffer to check
3644 * @len: length up to which to write
3646 * Returns true if modifying the header part of the cloned buffer
3647 * does not requires the data to be copied.
3649 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3651 return !skb_header_cloned(skb) &&
3652 skb_headroom(skb) + len <= skb->hdr_len;
3655 static inline int skb_try_make_writable(struct sk_buff *skb,
3656 unsigned int write_len)
3658 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3659 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3662 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3667 if (headroom > skb_headroom(skb))
3668 delta = headroom - skb_headroom(skb);
3670 if (delta || cloned)
3671 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3677 * skb_cow - copy header of skb when it is required
3678 * @skb: buffer to cow
3679 * @headroom: needed headroom
3681 * If the skb passed lacks sufficient headroom or its data part
3682 * is shared, data is reallocated. If reallocation fails, an error
3683 * is returned and original skb is not changed.
3685 * The result is skb with writable area skb->head...skb->tail
3686 * and at least @headroom of space at head.
3688 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3690 return __skb_cow(skb, headroom, skb_cloned(skb));
3694 * skb_cow_head - skb_cow but only making the head writable
3695 * @skb: buffer to cow
3696 * @headroom: needed headroom
3698 * This function is identical to skb_cow except that we replace the
3699 * skb_cloned check by skb_header_cloned. It should be used when
3700 * you only need to push on some header and do not need to modify
3703 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3705 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3709 * skb_padto - pad an skbuff up to a minimal size
3710 * @skb: buffer to pad
3711 * @len: minimal length
3713 * Pads up a buffer to ensure the trailing bytes exist and are
3714 * blanked. If the buffer already contains sufficient data it
3715 * is untouched. Otherwise it is extended. Returns zero on
3716 * success. The skb is freed on error.
3718 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3720 unsigned int size = skb->len;
3721 if (likely(size >= len))
3723 return skb_pad(skb, len - size);
3727 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3728 * @skb: buffer to pad
3729 * @len: minimal length
3730 * @free_on_error: free buffer on error
3732 * Pads up a buffer to ensure the trailing bytes exist and are
3733 * blanked. If the buffer already contains sufficient data it
3734 * is untouched. Otherwise it is extended. Returns zero on
3735 * success. The skb is freed on error if @free_on_error is true.
3737 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3741 unsigned int size = skb->len;
3743 if (unlikely(size < len)) {
3745 if (__skb_pad(skb, len, free_on_error))
3747 __skb_put(skb, len);
3753 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3754 * @skb: buffer to pad
3755 * @len: minimal length
3757 * Pads up a buffer to ensure the trailing bytes exist and are
3758 * blanked. If the buffer already contains sufficient data it
3759 * is untouched. Otherwise it is extended. Returns zero on
3760 * success. The skb is freed on error.
3762 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3764 return __skb_put_padto(skb, len, true);
3767 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i)
3770 static inline int skb_add_data(struct sk_buff *skb,
3771 struct iov_iter *from, int copy)
3773 const int off = skb->len;
3775 if (skb->ip_summed == CHECKSUM_NONE) {
3777 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3779 skb->csum = csum_block_add(skb->csum, csum, off);
3782 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3785 __skb_trim(skb, off);
3789 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3790 const struct page *page, int off)
3795 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3797 return page == skb_frag_page(frag) &&
3798 off == skb_frag_off(frag) + skb_frag_size(frag);
3803 static inline int __skb_linearize(struct sk_buff *skb)
3805 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3809 * skb_linearize - convert paged skb to linear one
3810 * @skb: buffer to linarize
3812 * If there is no free memory -ENOMEM is returned, otherwise zero
3813 * is returned and the old skb data released.
3815 static inline int skb_linearize(struct sk_buff *skb)
3817 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3821 * skb_has_shared_frag - can any frag be overwritten
3822 * @skb: buffer to test
3824 * Return true if the skb has at least one frag that might be modified
3825 * by an external entity (as in vmsplice()/sendfile())
3827 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3829 return skb_is_nonlinear(skb) &&
3830 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3834 * skb_linearize_cow - make sure skb is linear and writable
3835 * @skb: buffer to process
3837 * If there is no free memory -ENOMEM is returned, otherwise zero
3838 * is returned and the old skb data released.
3840 static inline int skb_linearize_cow(struct sk_buff *skb)
3842 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3843 __skb_linearize(skb) : 0;
3846 static __always_inline void
3847 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3850 if (skb->ip_summed == CHECKSUM_COMPLETE)
3851 skb->csum = csum_block_sub(skb->csum,
3852 csum_partial(start, len, 0), off);
3853 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3854 skb_checksum_start_offset(skb) < 0)
3855 skb->ip_summed = CHECKSUM_NONE;
3859 * skb_postpull_rcsum - update checksum for received skb after pull
3860 * @skb: buffer to update
3861 * @start: start of data before pull
3862 * @len: length of data pulled
3864 * After doing a pull on a received packet, you need to call this to
3865 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3866 * CHECKSUM_NONE so that it can be recomputed from scratch.
3868 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3869 const void *start, unsigned int len)
3871 if (skb->ip_summed == CHECKSUM_COMPLETE)
3872 skb->csum = wsum_negate(csum_partial(start, len,
3873 wsum_negate(skb->csum)));
3874 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3875 skb_checksum_start_offset(skb) < 0)
3876 skb->ip_summed = CHECKSUM_NONE;
3879 static __always_inline void
3880 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3883 if (skb->ip_summed == CHECKSUM_COMPLETE)
3884 skb->csum = csum_block_add(skb->csum,
3885 csum_partial(start, len, 0), off);
3889 * skb_postpush_rcsum - update checksum for received skb after push
3890 * @skb: buffer to update
3891 * @start: start of data after push
3892 * @len: length of data pushed
3894 * After doing a push on a received packet, you need to call this to
3895 * update the CHECKSUM_COMPLETE checksum.
3897 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3898 const void *start, unsigned int len)
3900 __skb_postpush_rcsum(skb, start, len, 0);
3903 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3906 * skb_push_rcsum - push skb and update receive checksum
3907 * @skb: buffer to update
3908 * @len: length of data pulled
3910 * This function performs an skb_push on the packet and updates
3911 * the CHECKSUM_COMPLETE checksum. It should be used on
3912 * receive path processing instead of skb_push unless you know
3913 * that the checksum difference is zero (e.g., a valid IP header)
3914 * or you are setting ip_summed to CHECKSUM_NONE.
3916 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3919 skb_postpush_rcsum(skb, skb->data, len);
3923 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3925 * pskb_trim_rcsum - trim received skb and update checksum
3926 * @skb: buffer to trim
3929 * This is exactly the same as pskb_trim except that it ensures the
3930 * checksum of received packets are still valid after the operation.
3931 * It can change skb pointers.
3934 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3936 if (likely(len >= skb->len))
3938 return pskb_trim_rcsum_slow(skb, len);
3941 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3943 if (skb->ip_summed == CHECKSUM_COMPLETE)
3944 skb->ip_summed = CHECKSUM_NONE;
3945 __skb_trim(skb, len);
3949 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3951 if (skb->ip_summed == CHECKSUM_COMPLETE)
3952 skb->ip_summed = CHECKSUM_NONE;
3953 return __skb_grow(skb, len);
3956 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3957 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3958 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3959 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3960 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3962 #define skb_queue_walk(queue, skb) \
3963 for (skb = (queue)->next; \
3964 skb != (struct sk_buff *)(queue); \
3967 #define skb_queue_walk_safe(queue, skb, tmp) \
3968 for (skb = (queue)->next, tmp = skb->next; \
3969 skb != (struct sk_buff *)(queue); \
3970 skb = tmp, tmp = skb->next)
3972 #define skb_queue_walk_from(queue, skb) \
3973 for (; skb != (struct sk_buff *)(queue); \
3976 #define skb_rbtree_walk(skb, root) \
3977 for (skb = skb_rb_first(root); skb != NULL; \
3978 skb = skb_rb_next(skb))
3980 #define skb_rbtree_walk_from(skb) \
3981 for (; skb != NULL; \
3982 skb = skb_rb_next(skb))
3984 #define skb_rbtree_walk_from_safe(skb, tmp) \
3985 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3988 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3989 for (tmp = skb->next; \
3990 skb != (struct sk_buff *)(queue); \
3991 skb = tmp, tmp = skb->next)
3993 #define skb_queue_reverse_walk(queue, skb) \
3994 for (skb = (queue)->prev; \
3995 skb != (struct sk_buff *)(queue); \
3998 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3999 for (skb = (queue)->prev, tmp = skb->prev; \
4000 skb != (struct sk_buff *)(queue); \
4001 skb = tmp, tmp = skb->prev)
4003 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
4004 for (tmp = skb->prev; \
4005 skb != (struct sk_buff *)(queue); \
4006 skb = tmp, tmp = skb->prev)
4008 static inline bool skb_has_frag_list(const struct sk_buff *skb)
4010 return skb_shinfo(skb)->frag_list != NULL;
4013 static inline void skb_frag_list_init(struct sk_buff *skb)
4015 skb_shinfo(skb)->frag_list = NULL;
4018 #define skb_walk_frags(skb, iter) \
4019 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
4022 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
4023 int *err, long *timeo_p,
4024 const struct sk_buff *skb);
4025 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
4026 struct sk_buff_head *queue,
4029 struct sk_buff **last);
4030 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
4031 struct sk_buff_head *queue,
4032 unsigned int flags, int *off, int *err,
4033 struct sk_buff **last);
4034 struct sk_buff *__skb_recv_datagram(struct sock *sk,
4035 struct sk_buff_head *sk_queue,
4036 unsigned int flags, int *off, int *err);
4037 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
4038 __poll_t datagram_poll(struct file *file, struct socket *sock,
4039 struct poll_table_struct *wait);
4040 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
4041 struct iov_iter *to, int size);
4042 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
4043 struct msghdr *msg, int size)
4045 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
4047 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
4048 struct msghdr *msg);
4049 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
4050 struct iov_iter *to, int len,
4051 struct ahash_request *hash);
4052 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
4053 struct iov_iter *from, int len);
4054 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
4055 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
4056 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
4057 static inline void skb_free_datagram_locked(struct sock *sk,
4058 struct sk_buff *skb)
4060 __skb_free_datagram_locked(sk, skb, 0);
4062 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
4063 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
4064 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
4065 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
4067 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
4068 struct pipe_inode_info *pipe, unsigned int len,
4069 unsigned int flags);
4070 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
4072 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
4073 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
4074 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
4075 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
4077 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
4078 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
4079 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
4080 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
4081 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
4082 unsigned int offset);
4083 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
4084 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
4085 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev);
4086 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
4087 int skb_vlan_pop(struct sk_buff *skb);
4088 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
4089 int skb_eth_pop(struct sk_buff *skb);
4090 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
4091 const unsigned char *src);
4092 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
4093 int mac_len, bool ethernet);
4094 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
4096 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
4097 int skb_mpls_dec_ttl(struct sk_buff *skb);
4098 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
4101 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
4103 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
4106 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
4108 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
4111 struct skb_checksum_ops {
4112 __wsum (*update)(const void *mem, int len, __wsum wsum);
4113 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
4116 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
4118 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
4119 __wsum csum, const struct skb_checksum_ops *ops);
4120 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
4123 static inline void * __must_check
4124 __skb_header_pointer(const struct sk_buff *skb, int offset, int len,
4125 const void *data, int hlen, void *buffer)
4127 if (likely(hlen - offset >= len))
4128 return (void *)data + offset;
4130 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
4136 static inline void * __must_check
4137 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
4139 return __skb_header_pointer(skb, offset, len, skb->data,
4140 skb_headlen(skb), buffer);
4143 static inline void * __must_check
4144 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len)
4146 if (likely(skb_headlen(skb) - offset >= len))
4147 return skb->data + offset;
4152 * skb_needs_linearize - check if we need to linearize a given skb
4153 * depending on the given device features.
4154 * @skb: socket buffer to check
4155 * @features: net device features
4157 * Returns true if either:
4158 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
4159 * 2. skb is fragmented and the device does not support SG.
4161 static inline bool skb_needs_linearize(struct sk_buff *skb,
4162 netdev_features_t features)
4164 return skb_is_nonlinear(skb) &&
4165 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
4166 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
4169 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
4171 const unsigned int len)
4173 memcpy(to, skb->data, len);
4176 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4177 const int offset, void *to,
4178 const unsigned int len)
4180 memcpy(to, skb->data + offset, len);
4183 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4185 const unsigned int len)
4187 memcpy(skb->data, from, len);
4190 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4193 const unsigned int len)
4195 memcpy(skb->data + offset, from, len);
4198 void skb_init(void);
4200 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4206 * skb_get_timestamp - get timestamp from a skb
4207 * @skb: skb to get stamp from
4208 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4210 * Timestamps are stored in the skb as offsets to a base timestamp.
4211 * This function converts the offset back to a struct timeval and stores
4214 static inline void skb_get_timestamp(const struct sk_buff *skb,
4215 struct __kernel_old_timeval *stamp)
4217 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4220 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4221 struct __kernel_sock_timeval *stamp)
4223 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4225 stamp->tv_sec = ts.tv_sec;
4226 stamp->tv_usec = ts.tv_nsec / 1000;
4229 static inline void skb_get_timestampns(const struct sk_buff *skb,
4230 struct __kernel_old_timespec *stamp)
4232 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4234 stamp->tv_sec = ts.tv_sec;
4235 stamp->tv_nsec = ts.tv_nsec;
4238 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4239 struct __kernel_timespec *stamp)
4241 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4243 stamp->tv_sec = ts.tv_sec;
4244 stamp->tv_nsec = ts.tv_nsec;
4247 static inline void __net_timestamp(struct sk_buff *skb)
4249 skb->tstamp = ktime_get_real();
4250 skb->mono_delivery_time = 0;
4253 static inline ktime_t net_timedelta(ktime_t t)
4255 return ktime_sub(ktime_get_real(), t);
4258 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4262 skb->mono_delivery_time = kt && mono;
4265 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4267 /* It is used in the ingress path to clear the delivery_time.
4268 * If needed, set the skb->tstamp to the (rcv) timestamp.
4270 static inline void skb_clear_delivery_time(struct sk_buff *skb)
4272 if (skb->mono_delivery_time) {
4273 skb->mono_delivery_time = 0;
4274 if (static_branch_unlikely(&netstamp_needed_key))
4275 skb->tstamp = ktime_get_real();
4281 static inline void skb_clear_tstamp(struct sk_buff *skb)
4283 if (skb->mono_delivery_time)
4289 static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4291 if (skb->mono_delivery_time)
4297 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4299 if (!skb->mono_delivery_time && skb->tstamp)
4302 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4303 return ktime_get_real();
4308 static inline u8 skb_metadata_len(const struct sk_buff *skb)
4310 return skb_shinfo(skb)->meta_len;
4313 static inline void *skb_metadata_end(const struct sk_buff *skb)
4315 return skb_mac_header(skb);
4318 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4319 const struct sk_buff *skb_b,
4322 const void *a = skb_metadata_end(skb_a);
4323 const void *b = skb_metadata_end(skb_b);
4326 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ||
4327 BITS_PER_LONG != 64)
4330 /* Using more efficient variant than plain call to memcmp(). */
4332 #define __it(x, op) (x -= sizeof(u##op))
4333 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4334 case 32: diffs |= __it_diff(a, b, 64);
4336 case 24: diffs |= __it_diff(a, b, 64);
4338 case 16: diffs |= __it_diff(a, b, 64);
4340 case 8: diffs |= __it_diff(a, b, 64);
4342 case 28: diffs |= __it_diff(a, b, 64);
4344 case 20: diffs |= __it_diff(a, b, 64);
4346 case 12: diffs |= __it_diff(a, b, 64);
4348 case 4: diffs |= __it_diff(a, b, 32);
4352 return memcmp(a - meta_len, b - meta_len, meta_len);
4357 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4358 const struct sk_buff *skb_b)
4360 u8 len_a = skb_metadata_len(skb_a);
4361 u8 len_b = skb_metadata_len(skb_b);
4363 if (!(len_a | len_b))
4366 return len_a != len_b ?
4367 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4370 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4372 skb_shinfo(skb)->meta_len = meta_len;
4375 static inline void skb_metadata_clear(struct sk_buff *skb)
4377 skb_metadata_set(skb, 0);
4380 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4382 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4384 void skb_clone_tx_timestamp(struct sk_buff *skb);
4385 bool skb_defer_rx_timestamp(struct sk_buff *skb);
4387 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4389 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4393 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4398 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4401 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4403 * PHY drivers may accept clones of transmitted packets for
4404 * timestamping via their phy_driver.txtstamp method. These drivers
4405 * must call this function to return the skb back to the stack with a
4408 * @skb: clone of the original outgoing packet
4409 * @hwtstamps: hardware time stamps
4412 void skb_complete_tx_timestamp(struct sk_buff *skb,
4413 struct skb_shared_hwtstamps *hwtstamps);
4415 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4416 struct skb_shared_hwtstamps *hwtstamps,
4417 struct sock *sk, int tstype);
4420 * skb_tstamp_tx - queue clone of skb with send time stamps
4421 * @orig_skb: the original outgoing packet
4422 * @hwtstamps: hardware time stamps, may be NULL if not available
4424 * If the skb has a socket associated, then this function clones the
4425 * skb (thus sharing the actual data and optional structures), stores
4426 * the optional hardware time stamping information (if non NULL) or
4427 * generates a software time stamp (otherwise), then queues the clone
4428 * to the error queue of the socket. Errors are silently ignored.
4430 void skb_tstamp_tx(struct sk_buff *orig_skb,
4431 struct skb_shared_hwtstamps *hwtstamps);
4434 * skb_tx_timestamp() - Driver hook for transmit timestamping
4436 * Ethernet MAC Drivers should call this function in their hard_xmit()
4437 * function immediately before giving the sk_buff to the MAC hardware.
4439 * Specifically, one should make absolutely sure that this function is
4440 * called before TX completion of this packet can trigger. Otherwise
4441 * the packet could potentially already be freed.
4443 * @skb: A socket buffer.
4445 static inline void skb_tx_timestamp(struct sk_buff *skb)
4447 skb_clone_tx_timestamp(skb);
4448 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4449 skb_tstamp_tx(skb, NULL);
4453 * skb_complete_wifi_ack - deliver skb with wifi status
4455 * @skb: the original outgoing packet
4456 * @acked: ack status
4459 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4461 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4462 __sum16 __skb_checksum_complete(struct sk_buff *skb);
4464 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4466 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4468 (skb->ip_summed == CHECKSUM_PARTIAL &&
4469 skb_checksum_start_offset(skb) >= 0));
4473 * skb_checksum_complete - Calculate checksum of an entire packet
4474 * @skb: packet to process
4476 * This function calculates the checksum over the entire packet plus
4477 * the value of skb->csum. The latter can be used to supply the
4478 * checksum of a pseudo header as used by TCP/UDP. It returns the
4481 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4482 * this function can be used to verify that checksum on received
4483 * packets. In that case the function should return zero if the
4484 * checksum is correct. In particular, this function will return zero
4485 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4486 * hardware has already verified the correctness of the checksum.
4488 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4490 return skb_csum_unnecessary(skb) ?
4491 0 : __skb_checksum_complete(skb);
4494 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4496 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4497 if (skb->csum_level == 0)
4498 skb->ip_summed = CHECKSUM_NONE;
4504 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4506 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4507 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4509 } else if (skb->ip_summed == CHECKSUM_NONE) {
4510 skb->ip_summed = CHECKSUM_UNNECESSARY;
4511 skb->csum_level = 0;
4515 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4517 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4518 skb->ip_summed = CHECKSUM_NONE;
4519 skb->csum_level = 0;
4523 /* Check if we need to perform checksum complete validation.
4525 * Returns true if checksum complete is needed, false otherwise
4526 * (either checksum is unnecessary or zero checksum is allowed).
4528 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4532 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4533 skb->csum_valid = 1;
4534 __skb_decr_checksum_unnecessary(skb);
4541 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4544 #define CHECKSUM_BREAK 76
4546 /* Unset checksum-complete
4548 * Unset checksum complete can be done when packet is being modified
4549 * (uncompressed for instance) and checksum-complete value is
4552 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4554 if (skb->ip_summed == CHECKSUM_COMPLETE)
4555 skb->ip_summed = CHECKSUM_NONE;
4558 /* Validate (init) checksum based on checksum complete.
4561 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4562 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4563 * checksum is stored in skb->csum for use in __skb_checksum_complete
4564 * non-zero: value of invalid checksum
4567 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4571 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4572 if (!csum_fold(csum_add(psum, skb->csum))) {
4573 skb->csum_valid = 1;
4580 if (complete || skb->len <= CHECKSUM_BREAK) {
4583 csum = __skb_checksum_complete(skb);
4584 skb->csum_valid = !csum;
4591 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4596 /* Perform checksum validate (init). Note that this is a macro since we only
4597 * want to calculate the pseudo header which is an input function if necessary.
4598 * First we try to validate without any computation (checksum unnecessary) and
4599 * then calculate based on checksum complete calling the function to compute
4603 * 0: checksum is validated or try to in skb_checksum_complete
4604 * non-zero: value of invalid checksum
4606 #define __skb_checksum_validate(skb, proto, complete, \
4607 zero_okay, check, compute_pseudo) \
4609 __sum16 __ret = 0; \
4610 skb->csum_valid = 0; \
4611 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4612 __ret = __skb_checksum_validate_complete(skb, \
4613 complete, compute_pseudo(skb, proto)); \
4617 #define skb_checksum_init(skb, proto, compute_pseudo) \
4618 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4620 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4621 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4623 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4624 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4626 #define skb_checksum_validate_zero_check(skb, proto, check, \
4628 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4630 #define skb_checksum_simple_validate(skb) \
4631 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4633 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4635 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4638 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4640 skb->csum = ~pseudo;
4641 skb->ip_summed = CHECKSUM_COMPLETE;
4644 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4646 if (__skb_checksum_convert_check(skb)) \
4647 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4650 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4651 u16 start, u16 offset)
4653 skb->ip_summed = CHECKSUM_PARTIAL;
4654 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4655 skb->csum_offset = offset - start;
4658 /* Update skbuf and packet to reflect the remote checksum offload operation.
4659 * When called, ptr indicates the starting point for skb->csum when
4660 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4661 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4663 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4664 int start, int offset, bool nopartial)
4669 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4673 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4674 __skb_checksum_complete(skb);
4675 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4678 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4680 /* Adjust skb->csum since we changed the packet */
4681 skb->csum = csum_add(skb->csum, delta);
4684 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4686 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4687 return (void *)(skb->_nfct & NFCT_PTRMASK);
4693 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4695 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4702 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4704 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4705 skb->slow_gro |= !!nfct;
4710 #ifdef CONFIG_SKB_EXTENSIONS
4712 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4718 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4721 #if IS_ENABLED(CONFIG_MPTCP)
4724 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4727 SKB_EXT_NUM, /* must be last */
4731 * struct skb_ext - sk_buff extensions
4732 * @refcnt: 1 on allocation, deallocated on 0
4733 * @offset: offset to add to @data to obtain extension address
4734 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4735 * @data: start of extension data, variable sized
4737 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4738 * to use 'u8' types while allowing up to 2kb worth of extension data.
4742 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4743 u8 chunks; /* same */
4744 char data[] __aligned(8);
4747 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4748 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4749 struct skb_ext *ext);
4750 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4751 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4752 void __skb_ext_put(struct skb_ext *ext);
4754 static inline void skb_ext_put(struct sk_buff *skb)
4756 if (skb->active_extensions)
4757 __skb_ext_put(skb->extensions);
4760 static inline void __skb_ext_copy(struct sk_buff *dst,
4761 const struct sk_buff *src)
4763 dst->active_extensions = src->active_extensions;
4765 if (src->active_extensions) {
4766 struct skb_ext *ext = src->extensions;
4768 refcount_inc(&ext->refcnt);
4769 dst->extensions = ext;
4773 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4776 __skb_ext_copy(dst, src);
4779 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4781 return !!ext->offset[i];
4784 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4786 return skb->active_extensions & (1 << id);
4789 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4791 if (skb_ext_exist(skb, id))
4792 __skb_ext_del(skb, id);
4795 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4797 if (skb_ext_exist(skb, id)) {
4798 struct skb_ext *ext = skb->extensions;
4800 return (void *)ext + (ext->offset[id] << 3);
4806 static inline void skb_ext_reset(struct sk_buff *skb)
4808 if (unlikely(skb->active_extensions)) {
4809 __skb_ext_put(skb->extensions);
4810 skb->active_extensions = 0;
4814 static inline bool skb_has_extensions(struct sk_buff *skb)
4816 return unlikely(skb->active_extensions);
4819 static inline void skb_ext_put(struct sk_buff *skb) {}
4820 static inline void skb_ext_reset(struct sk_buff *skb) {}
4821 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4822 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4823 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4824 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4825 #endif /* CONFIG_SKB_EXTENSIONS */
4827 static inline void nf_reset_ct(struct sk_buff *skb)
4829 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4830 nf_conntrack_put(skb_nfct(skb));
4835 static inline void nf_reset_trace(struct sk_buff *skb)
4837 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4842 static inline void ipvs_reset(struct sk_buff *skb)
4844 #if IS_ENABLED(CONFIG_IP_VS)
4845 skb->ipvs_property = 0;
4849 /* Note: This doesn't put any conntrack info in dst. */
4850 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4853 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4854 dst->_nfct = src->_nfct;
4855 nf_conntrack_get(skb_nfct(src));
4857 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4859 dst->nf_trace = src->nf_trace;
4863 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4865 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4866 nf_conntrack_put(skb_nfct(dst));
4868 dst->slow_gro = src->slow_gro;
4869 __nf_copy(dst, src, true);
4872 #ifdef CONFIG_NETWORK_SECMARK
4873 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4875 to->secmark = from->secmark;
4878 static inline void skb_init_secmark(struct sk_buff *skb)
4883 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4886 static inline void skb_init_secmark(struct sk_buff *skb)
4890 static inline int secpath_exists(const struct sk_buff *skb)
4893 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4899 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4901 return !skb->destructor &&
4902 !secpath_exists(skb) &&
4904 !skb->_skb_refdst &&
4905 !skb_has_frag_list(skb);
4908 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4910 skb->queue_mapping = queue_mapping;
4913 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4915 return skb->queue_mapping;
4918 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4920 to->queue_mapping = from->queue_mapping;
4923 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4925 skb->queue_mapping = rx_queue + 1;
4928 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4930 return skb->queue_mapping - 1;
4933 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4935 return skb->queue_mapping != 0;
4938 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4940 skb->dst_pending_confirm = val;
4943 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4945 return skb->dst_pending_confirm != 0;
4948 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4951 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4957 static inline bool skb_is_gso(const struct sk_buff *skb)
4959 return skb_shinfo(skb)->gso_size;
4962 /* Note: Should be called only if skb_is_gso(skb) is true */
4963 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4965 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4968 /* Note: Should be called only if skb_is_gso(skb) is true */
4969 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4971 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4974 /* Note: Should be called only if skb_is_gso(skb) is true */
4975 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4977 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4980 static inline void skb_gso_reset(struct sk_buff *skb)
4982 skb_shinfo(skb)->gso_size = 0;
4983 skb_shinfo(skb)->gso_segs = 0;
4984 skb_shinfo(skb)->gso_type = 0;
4987 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4990 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4992 shinfo->gso_size += increment;
4995 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4998 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
5000 shinfo->gso_size -= decrement;
5003 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
5005 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
5007 /* LRO sets gso_size but not gso_type, whereas if GSO is really
5008 * wanted then gso_type will be set. */
5009 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5011 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
5012 unlikely(shinfo->gso_type == 0)) {
5013 __skb_warn_lro_forwarding(skb);
5019 static inline void skb_forward_csum(struct sk_buff *skb)
5021 /* Unfortunately we don't support this one. Any brave souls? */
5022 if (skb->ip_summed == CHECKSUM_COMPLETE)
5023 skb->ip_summed = CHECKSUM_NONE;
5027 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
5028 * @skb: skb to check
5030 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
5031 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
5032 * use this helper, to document places where we make this assertion.
5034 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
5036 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE);
5039 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
5041 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
5042 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5043 unsigned int transport_len,
5044 __sum16(*skb_chkf)(struct sk_buff *skb));
5047 * skb_head_is_locked - Determine if the skb->head is locked down
5048 * @skb: skb to check
5050 * The head on skbs build around a head frag can be removed if they are
5051 * not cloned. This function returns true if the skb head is locked down
5052 * due to either being allocated via kmalloc, or by being a clone with
5053 * multiple references to the head.
5055 static inline bool skb_head_is_locked(const struct sk_buff *skb)
5057 return !skb->head_frag || skb_cloned(skb);
5060 /* Local Checksum Offload.
5061 * Compute outer checksum based on the assumption that the
5062 * inner checksum will be offloaded later.
5063 * See Documentation/networking/checksum-offloads.rst for
5064 * explanation of how this works.
5065 * Fill in outer checksum adjustment (e.g. with sum of outer
5066 * pseudo-header) before calling.
5067 * Also ensure that inner checksum is in linear data area.
5069 static inline __wsum lco_csum(struct sk_buff *skb)
5071 unsigned char *csum_start = skb_checksum_start(skb);
5072 unsigned char *l4_hdr = skb_transport_header(skb);
5075 /* Start with complement of inner checksum adjustment */
5076 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
5079 /* Add in checksum of our headers (incl. outer checksum
5080 * adjustment filled in by caller) and return result.
5082 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
5085 static inline bool skb_is_redirected(const struct sk_buff *skb)
5087 return skb->redirected;
5090 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
5092 skb->redirected = 1;
5093 #ifdef CONFIG_NET_REDIRECT
5094 skb->from_ingress = from_ingress;
5095 if (skb->from_ingress)
5096 skb_clear_tstamp(skb);
5100 static inline void skb_reset_redirect(struct sk_buff *skb)
5102 skb->redirected = 0;
5105 static inline void skb_set_redirected_noclear(struct sk_buff *skb,
5108 skb->redirected = 1;
5109 #ifdef CONFIG_NET_REDIRECT
5110 skb->from_ingress = from_ingress;
5114 static inline bool skb_csum_is_sctp(struct sk_buff *skb)
5116 #if IS_ENABLED(CONFIG_IP_SCTP)
5117 return skb->csum_not_inet;
5123 static inline void skb_reset_csum_not_inet(struct sk_buff *skb)
5125 skb->ip_summed = CHECKSUM_NONE;
5126 #if IS_ENABLED(CONFIG_IP_SCTP)
5127 skb->csum_not_inet = 0;
5131 static inline void skb_set_kcov_handle(struct sk_buff *skb,
5132 const u64 kcov_handle)
5135 skb->kcov_handle = kcov_handle;
5139 static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
5142 return skb->kcov_handle;
5148 static inline void skb_mark_for_recycle(struct sk_buff *skb)
5150 #ifdef CONFIG_PAGE_POOL
5151 skb->pp_recycle = 1;
5155 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter,
5156 ssize_t maxsize, gfp_t gfp);
5158 #endif /* __KERNEL__ */
5159 #endif /* _LINUX_SKBUFF_H */