1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * Definitions for the 'struct sk_buff' memory handlers.
10 #ifndef _LINUX_SKBUFF_H
11 #define _LINUX_SKBUFF_H
13 #include <linux/kernel.h>
14 #include <linux/compiler.h>
15 #include <linux/time.h>
16 #include <linux/bug.h>
17 #include <linux/bvec.h>
18 #include <linux/cache.h>
19 #include <linux/rbtree.h>
20 #include <linux/socket.h>
21 #include <linux/refcount.h>
23 #include <linux/atomic.h>
24 #include <asm/types.h>
25 #include <linux/spinlock.h>
26 #include <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 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
357 * segment using its current segmentation instead.
359 #define GSO_BY_FRAGS 0xFFFF
361 typedef struct skb_frag {
368 * skb_frag_size() - Returns the size of a skb fragment
369 * @frag: skb fragment
371 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
377 * skb_frag_size_set() - Sets the size of a skb fragment
378 * @frag: skb fragment
379 * @size: size of fragment
381 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
387 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
388 * @frag: skb fragment
389 * @delta: value to add
391 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
397 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
398 * @frag: skb fragment
399 * @delta: value to subtract
401 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
407 * skb_frag_must_loop - Test if %p is a high memory page
408 * @p: fragment's page
410 static inline bool skb_frag_must_loop(struct page *p)
412 #if defined(CONFIG_HIGHMEM)
413 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
420 * skb_frag_foreach_page - loop over pages in a fragment
422 * @f: skb frag to operate on
423 * @f_off: offset from start of f->netmem
424 * @f_len: length from f_off to loop over
425 * @p: (temp var) current page
426 * @p_off: (temp var) offset from start of current page,
427 * non-zero only on first page.
428 * @p_len: (temp var) length in current page,
429 * < PAGE_SIZE only on first and last page.
430 * @copied: (temp var) length so far, excluding current p_len.
432 * A fragment can hold a compound page, in which case per-page
433 * operations, notably kmap_atomic, must be called for each
436 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
437 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
438 p_off = (f_off) & (PAGE_SIZE - 1), \
439 p_len = skb_frag_must_loop(p) ? \
440 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
443 copied += p_len, p++, p_off = 0, \
444 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
447 * struct skb_shared_hwtstamps - hardware time stamps
448 * @hwtstamp: hardware time stamp transformed into duration
449 * since arbitrary point in time
450 * @netdev_data: address/cookie of network device driver used as
451 * reference to actual hardware time stamp
453 * Software time stamps generated by ktime_get_real() are stored in
456 * hwtstamps can only be compared against other hwtstamps from
459 * This structure is attached to packets as part of the
460 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
462 struct skb_shared_hwtstamps {
469 /* Definitions for tx_flags in struct skb_shared_info */
471 /* generate hardware time stamp */
472 SKBTX_HW_TSTAMP = 1 << 0,
474 /* generate software time stamp when queueing packet to NIC */
475 SKBTX_SW_TSTAMP = 1 << 1,
477 /* device driver is going to provide hardware time stamp */
478 SKBTX_IN_PROGRESS = 1 << 2,
480 /* generate hardware time stamp based on cycles if supported */
481 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3,
483 /* generate wifi status information (where possible) */
484 SKBTX_WIFI_STATUS = 1 << 4,
486 /* determine hardware time stamp based on time or cycles */
487 SKBTX_HW_TSTAMP_NETDEV = 1 << 5,
489 /* generate software time stamp when entering packet scheduling */
490 SKBTX_SCHED_TSTAMP = 1 << 6,
493 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
495 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \
496 SKBTX_HW_TSTAMP_USE_CYCLES | \
499 /* Definitions for flags in struct skb_shared_info */
501 /* use zcopy routines */
502 SKBFL_ZEROCOPY_ENABLE = BIT(0),
504 /* This indicates at least one fragment might be overwritten
505 * (as in vmsplice(), sendfile() ...)
506 * If we need to compute a TX checksum, we'll need to copy
507 * all frags to avoid possible bad checksum
509 SKBFL_SHARED_FRAG = BIT(1),
511 /* segment contains only zerocopy data and should not be
512 * charged to the kernel memory.
514 SKBFL_PURE_ZEROCOPY = BIT(2),
516 SKBFL_DONT_ORPHAN = BIT(3),
518 /* page references are managed by the ubuf_info, so it's safe to
519 * use frags only up until ubuf_info is released
521 SKBFL_MANAGED_FRAG_REFS = BIT(4),
524 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
525 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
526 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
528 struct ubuf_info_ops {
529 void (*complete)(struct sk_buff *, struct ubuf_info *,
530 bool zerocopy_success);
531 /* has to be compatible with skb_zcopy_set() */
532 int (*link_skb)(struct sk_buff *skb, struct ubuf_info *uarg);
536 * The callback notifies userspace to release buffers when skb DMA is done in
537 * lower device, the skb last reference should be 0 when calling this.
538 * The zerocopy_success argument is true if zero copy transmit occurred,
539 * false on data copy or out of memory error caused by data copy attempt.
540 * The ctx field is used to track device context.
541 * The desc field is used to track userspace buffer index.
544 const struct ubuf_info_ops *ops;
549 struct ubuf_info_msgzc {
550 struct ubuf_info ubuf;
566 struct user_struct *user;
571 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
572 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \
575 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
576 void mm_unaccount_pinned_pages(struct mmpin *mmp);
578 /* Preserve some data across TX submission and completion.
580 * Note, this state is stored in the driver. Extending the layout
581 * might need some special care.
583 struct xsk_tx_metadata_compl {
587 /* This data is invariant across clones and lives at
588 * the end of the header data, ie. at skb->end.
590 struct skb_shared_info {
595 unsigned short gso_size;
596 /* Warning: this field is not always filled in (UFO)! */
597 unsigned short gso_segs;
598 struct sk_buff *frag_list;
600 struct skb_shared_hwtstamps hwtstamps;
601 struct xsk_tx_metadata_compl xsk_meta;
603 unsigned int gso_type;
607 * Warning : all fields before dataref are cleared in __alloc_skb()
610 unsigned int xdp_frags_size;
612 /* Intermediate layers must ensure that destructor_arg
613 * remains valid until skb destructor */
614 void * destructor_arg;
616 /* must be last field, see pskb_expand_head() */
617 skb_frag_t frags[MAX_SKB_FRAGS];
621 * DOC: dataref and headerless skbs
623 * Transport layers send out clones of payload skbs they hold for
624 * retransmissions. To allow lower layers of the stack to prepend their headers
625 * we split &skb_shared_info.dataref into two halves.
626 * The lower 16 bits count the overall number of references.
627 * The higher 16 bits indicate how many of the references are payload-only.
628 * skb_header_cloned() checks if skb is allowed to add / write the headers.
630 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
631 * (via __skb_header_release()). Any clone created from marked skb will get
632 * &sk_buff.hdr_len populated with the available headroom.
633 * If there's the only clone in existence it's able to modify the headroom
634 * at will. The sequence of calls inside the transport layer is::
638 * __skb_header_release()
640 * // send the clone down the stack
642 * This is not a very generic construct and it depends on the transport layers
643 * doing the right thing. In practice there's usually only one payload-only skb.
644 * Having multiple payload-only skbs with different lengths of hdr_len is not
645 * possible. The payload-only skbs should never leave their owner.
647 #define SKB_DATAREF_SHIFT 16
648 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
652 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
653 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
654 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
658 SKB_GSO_TCPV4 = 1 << 0,
660 /* This indicates the skb is from an untrusted source. */
661 SKB_GSO_DODGY = 1 << 1,
663 /* This indicates the tcp segment has CWR set. */
664 SKB_GSO_TCP_ECN = 1 << 2,
666 SKB_GSO_TCP_FIXEDID = 1 << 3,
668 SKB_GSO_TCPV6 = 1 << 4,
670 SKB_GSO_FCOE = 1 << 5,
672 SKB_GSO_GRE = 1 << 6,
674 SKB_GSO_GRE_CSUM = 1 << 7,
676 SKB_GSO_IPXIP4 = 1 << 8,
678 SKB_GSO_IPXIP6 = 1 << 9,
680 SKB_GSO_UDP_TUNNEL = 1 << 10,
682 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
684 SKB_GSO_PARTIAL = 1 << 12,
686 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
688 SKB_GSO_SCTP = 1 << 14,
690 SKB_GSO_ESP = 1 << 15,
692 SKB_GSO_UDP = 1 << 16,
694 SKB_GSO_UDP_L4 = 1 << 17,
696 SKB_GSO_FRAGLIST = 1 << 18,
699 #if BITS_PER_LONG > 32
700 #define NET_SKBUFF_DATA_USES_OFFSET 1
703 #ifdef NET_SKBUFF_DATA_USES_OFFSET
704 typedef unsigned int sk_buff_data_t;
706 typedef unsigned char *sk_buff_data_t;
709 enum skb_tstamp_type {
713 __SKB_CLOCK_MAX = SKB_CLOCK_TAI,
717 * DOC: Basic sk_buff geometry
719 * struct sk_buff itself is a metadata structure and does not hold any packet
720 * data. All the data is held in associated buffers.
722 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
725 * - data buffer, containing headers and sometimes payload;
726 * this is the part of the skb operated on by the common helpers
727 * such as skb_put() or skb_pull();
728 * - shared info (struct skb_shared_info) which holds an array of pointers
729 * to read-only data in the (page, offset, length) format.
731 * Optionally &skb_shared_info.frag_list may point to another skb.
733 * Basic diagram may look like this::
738 * ,--------------------------- + head
739 * / ,----------------- + data
740 * / / ,----------- + tail
744 * -----------------------------------------------
745 * | headroom | data | tailroom | skb_shared_info |
746 * -----------------------------------------------
750 * + [page frag] ---------
751 * + frag_list --> | sk_buff |
757 * struct sk_buff - socket buffer
758 * @next: Next buffer in list
759 * @prev: Previous buffer in list
760 * @tstamp: Time we arrived/left
761 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
762 * for retransmit timer
763 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
765 * @ll_node: anchor in an llist (eg socket defer_list)
766 * @sk: Socket we are owned by
767 * @dev: Device we arrived on/are leaving by
768 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
769 * @cb: Control buffer. Free for use by every layer. Put private vars here
770 * @_skb_refdst: destination entry (with norefcount bit)
771 * @len: Length of actual data
772 * @data_len: Data length
773 * @mac_len: Length of link layer header
774 * @hdr_len: writable header length of cloned skb
775 * @csum: Checksum (must include start/offset pair)
776 * @csum_start: Offset from skb->head where checksumming should start
777 * @csum_offset: Offset from csum_start where checksum should be stored
778 * @priority: Packet queueing priority
779 * @ignore_df: allow local fragmentation
780 * @cloned: Head may be cloned (check refcnt to be sure)
781 * @ip_summed: Driver fed us an IP checksum
782 * @nohdr: Payload reference only, must not modify header
783 * @pkt_type: Packet class
784 * @fclone: skbuff clone status
785 * @ipvs_property: skbuff is owned by ipvs
786 * @inner_protocol_type: whether the inner protocol is
787 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
788 * @remcsum_offload: remote checksum offload is enabled
789 * @offload_fwd_mark: Packet was L2-forwarded in hardware
790 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
791 * @tc_skip_classify: do not classify packet. set by IFB device
792 * @tc_at_ingress: used within tc_classify to distinguish in/egress
793 * @redirected: packet was redirected by packet classifier
794 * @from_ingress: packet was redirected from the ingress path
795 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
796 * @peeked: this packet has been seen already, so stats have been
797 * done for it, don't do them again
798 * @nf_trace: netfilter packet trace flag
799 * @protocol: Packet protocol from driver
800 * @destructor: Destruct function
801 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
802 * @_sk_redir: socket redirection information for skmsg
803 * @_nfct: Associated connection, if any (with nfctinfo bits)
804 * @skb_iif: ifindex of device we arrived on
805 * @tc_index: Traffic control index
806 * @hash: the packet hash
807 * @queue_mapping: Queue mapping for multiqueue devices
808 * @head_frag: skb was allocated from page fragments,
809 * not allocated by kmalloc() or vmalloc().
810 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
811 * @pp_recycle: mark the packet for recycling instead of freeing (implies
812 * page_pool support on driver)
813 * @active_extensions: active extensions (skb_ext_id types)
814 * @ndisc_nodetype: router type (from link layer)
815 * @ooo_okay: allow the mapping of a socket to a queue to be changed
816 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
818 * @sw_hash: indicates hash was computed in software stack
819 * @wifi_acked_valid: wifi_acked was set
820 * @wifi_acked: whether frame was acked on wifi or not
821 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
822 * @encapsulation: indicates the inner headers in the skbuff are valid
823 * @encap_hdr_csum: software checksum is needed
824 * @csum_valid: checksum is already valid
825 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
826 * @csum_complete_sw: checksum was completed by software
827 * @csum_level: indicates the number of consecutive checksums found in
828 * the packet minus one that have been verified as
829 * CHECKSUM_UNNECESSARY (max 3)
830 * @dst_pending_confirm: need to confirm neighbour
831 * @decrypted: Decrypted SKB
832 * @slow_gro: state present at GRO time, slower prepare step required
833 * @tstamp_type: When set, skb->tstamp has the
834 * delivery_time clock base of skb->tstamp.
835 * @napi_id: id of the NAPI struct this skb came from
836 * @sender_cpu: (aka @napi_id) source CPU in XPS
837 * @alloc_cpu: CPU which did the skb allocation.
838 * @secmark: security marking
839 * @mark: Generic packet mark
840 * @reserved_tailroom: (aka @mark) number of bytes of free space available
841 * at the tail of an sk_buff
842 * @vlan_all: vlan fields (proto & tci)
843 * @vlan_proto: vlan encapsulation protocol
844 * @vlan_tci: vlan tag control information
845 * @inner_protocol: Protocol (encapsulation)
846 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
847 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
848 * @inner_transport_header: Inner transport layer header (encapsulation)
849 * @inner_network_header: Network layer header (encapsulation)
850 * @inner_mac_header: Link layer header (encapsulation)
851 * @transport_header: Transport layer header
852 * @network_header: Network layer header
853 * @mac_header: Link layer header
854 * @kcov_handle: KCOV remote handle for remote coverage collection
855 * @tail: Tail pointer
857 * @head: Head of buffer
858 * @data: Data head pointer
859 * @truesize: Buffer size
860 * @users: User count - see {datagram,tcp}.c
861 * @extensions: allocated extensions, valid if active_extensions is nonzero
867 /* These two members must be first to match sk_buff_head. */
868 struct sk_buff *next;
869 struct sk_buff *prev;
872 struct net_device *dev;
873 /* Some protocols might use this space to store information,
874 * while device pointer would be NULL.
875 * UDP receive path is one user.
877 unsigned long dev_scratch;
880 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
881 struct list_head list;
882 struct llist_node ll_node;
889 u64 skb_mstamp_ns; /* earliest departure time */
892 * This is the control buffer. It is free to use for every
893 * layer. Please put your private variables there. If you
894 * want to keep them across layers you have to do a skb_clone()
895 * first. This is owned by whoever has the skb queued ATM.
897 char cb[48] __aligned(8);
901 unsigned long _skb_refdst;
902 void (*destructor)(struct sk_buff *skb);
904 struct list_head tcp_tsorted_anchor;
905 #ifdef CONFIG_NET_SOCK_MSG
906 unsigned long _sk_redir;
910 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
918 /* Following fields are _not_ copied in __copy_skb_header()
919 * Note that queue_mapping is here mostly to fill a hole.
923 /* if you move cloned around you also must adapt those constants */
924 #ifdef __BIG_ENDIAN_BITFIELD
925 #define CLONED_MASK (1 << 7)
927 #define CLONED_MASK 1
929 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
932 __u8 __cloned_offset[0];
940 pp_recycle:1; /* page_pool recycle indicator */
941 #ifdef CONFIG_SKB_EXTENSIONS
942 __u8 active_extensions;
945 /* Fields enclosed in headers group are copied
946 * using a single memcpy() in __copy_skb_header()
948 struct_group(headers,
951 __u8 __pkt_type_offset[0];
953 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
955 __u8 dst_pending_confirm:1;
960 __u8 __mono_tc_offset[0];
962 __u8 tstamp_type:2; /* See skb_tstamp_type */
963 #ifdef CONFIG_NET_XGRESS
964 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
965 __u8 tc_skip_classify:1;
967 __u8 remcsum_offload:1;
968 __u8 csum_complete_sw:1;
970 __u8 inner_protocol_type:1;
974 #ifdef CONFIG_WIRELESS
975 __u8 wifi_acked_valid:1;
979 /* Indicates the inner headers are valid in the skbuff. */
980 __u8 encapsulation:1;
981 __u8 encap_hdr_csum:1;
983 #ifdef CONFIG_IPV6_NDISC_NODETYPE
984 __u8 ndisc_nodetype:2;
987 #if IS_ENABLED(CONFIG_IP_VS)
988 __u8 ipvs_property:1;
990 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
993 #ifdef CONFIG_NET_SWITCHDEV
994 __u8 offload_fwd_mark:1;
995 __u8 offload_l3_fwd_mark:1;
998 #ifdef CONFIG_NET_REDIRECT
1001 #ifdef CONFIG_NETFILTER_SKIP_EGRESS
1002 __u8 nf_skip_egress:1;
1004 #ifdef CONFIG_SKB_DECRYPTED
1008 #if IS_ENABLED(CONFIG_IP_SCTP)
1009 __u8 csum_not_inet:1;
1012 #if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS)
1013 __u16 tc_index; /* traffic control index */
1035 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
1037 unsigned int napi_id;
1038 unsigned int sender_cpu;
1041 #ifdef CONFIG_NETWORK_SECMARK
1047 __u32 reserved_tailroom;
1051 __be16 inner_protocol;
1055 __u16 inner_transport_header;
1056 __u16 inner_network_header;
1057 __u16 inner_mac_header;
1060 __u16 transport_header;
1061 __u16 network_header;
1068 ); /* end headers group */
1070 /* These elements must be at the end, see alloc_skb() for details. */
1071 sk_buff_data_t tail;
1073 unsigned char *head,
1075 unsigned int truesize;
1078 #ifdef CONFIG_SKB_EXTENSIONS
1079 /* only usable after checking ->active_extensions != 0 */
1080 struct skb_ext *extensions;
1084 /* if you move pkt_type around you also must adapt those constants */
1085 #ifdef __BIG_ENDIAN_BITFIELD
1086 #define PKT_TYPE_MAX (7 << 5)
1088 #define PKT_TYPE_MAX 7
1090 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1092 /* if you move tc_at_ingress or tstamp_type
1093 * around, you also must adapt these constants.
1095 #ifdef __BIG_ENDIAN_BITFIELD
1096 #define SKB_TSTAMP_TYPE_MASK (3 << 6)
1097 #define SKB_TSTAMP_TYPE_RSHIFT (6)
1098 #define TC_AT_INGRESS_MASK (1 << 5)
1100 #define SKB_TSTAMP_TYPE_MASK (3)
1101 #define TC_AT_INGRESS_MASK (1 << 2)
1103 #define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset)
1107 * Handling routines are only of interest to the kernel
1110 #define SKB_ALLOC_FCLONE 0x01
1111 #define SKB_ALLOC_RX 0x02
1112 #define SKB_ALLOC_NAPI 0x04
1115 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1118 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1120 return unlikely(skb->pfmemalloc);
1124 * skb might have a dst pointer attached, refcounted or not.
1125 * _skb_refdst low order bit is set if refcount was _not_ taken
1127 #define SKB_DST_NOREF 1UL
1128 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1131 * skb_dst - returns skb dst_entry
1134 * Returns skb dst_entry, regardless of reference taken or not.
1136 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1138 /* If refdst was not refcounted, check we still are in a
1139 * rcu_read_lock section
1141 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1142 !rcu_read_lock_held() &&
1143 !rcu_read_lock_bh_held());
1144 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1148 * skb_dst_set - sets skb dst
1152 * Sets skb dst, assuming a reference was taken on dst and should
1153 * be released by skb_dst_drop()
1155 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1157 skb->slow_gro |= !!dst;
1158 skb->_skb_refdst = (unsigned long)dst;
1162 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1166 * Sets skb dst, assuming a reference was not taken on dst.
1167 * If dst entry is cached, we do not take reference and dst_release
1168 * will be avoided by refdst_drop. If dst entry is not cached, we take
1169 * reference, so that last dst_release can destroy the dst immediately.
1171 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1173 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1174 skb->slow_gro |= !!dst;
1175 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1179 * skb_dst_is_noref - Test if skb dst isn't refcounted
1182 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1184 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1187 /* For mangling skb->pkt_type from user space side from applications
1188 * such as nft, tc, etc, we only allow a conservative subset of
1189 * possible pkt_types to be set.
1191 static inline bool skb_pkt_type_ok(u32 ptype)
1193 return ptype <= PACKET_OTHERHOST;
1197 * skb_napi_id - Returns the skb's NAPI id
1200 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1202 #ifdef CONFIG_NET_RX_BUSY_POLL
1203 return skb->napi_id;
1209 static inline bool skb_wifi_acked_valid(const struct sk_buff *skb)
1211 #ifdef CONFIG_WIRELESS
1212 return skb->wifi_acked_valid;
1219 * skb_unref - decrement the skb's reference count
1222 * Returns true if we can free the skb.
1224 static inline bool skb_unref(struct sk_buff *skb)
1228 if (likely(refcount_read(&skb->users) == 1))
1230 else if (likely(!refcount_dec_and_test(&skb->users)))
1236 static inline bool skb_data_unref(const struct sk_buff *skb,
1237 struct skb_shared_info *shinfo)
1244 bias = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
1246 if (atomic_read(&shinfo->dataref) == bias)
1248 else if (atomic_sub_return(bias, &shinfo->dataref))
1254 void __fix_address sk_skb_reason_drop(struct sock *sk, struct sk_buff *skb,
1255 enum skb_drop_reason reason);
1258 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason)
1260 sk_skb_reason_drop(NULL, skb, reason);
1264 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1265 * @skb: buffer to free
1267 static inline void kfree_skb(struct sk_buff *skb)
1269 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1272 void skb_release_head_state(struct sk_buff *skb);
1273 void kfree_skb_list_reason(struct sk_buff *segs,
1274 enum skb_drop_reason reason);
1275 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1276 void skb_tx_error(struct sk_buff *skb);
1278 static inline void kfree_skb_list(struct sk_buff *segs)
1280 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1283 #ifdef CONFIG_TRACEPOINTS
1284 void consume_skb(struct sk_buff *skb);
1286 static inline void consume_skb(struct sk_buff *skb)
1288 return kfree_skb(skb);
1292 void __consume_stateless_skb(struct sk_buff *skb);
1293 void __kfree_skb(struct sk_buff *skb);
1295 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1296 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1297 bool *fragstolen, int *delta_truesize);
1299 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1301 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1302 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1303 struct sk_buff *build_skb_around(struct sk_buff *skb,
1304 void *data, unsigned int frag_size);
1305 void skb_attempt_defer_free(struct sk_buff *skb);
1307 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1308 struct sk_buff *slab_build_skb(void *data);
1311 * alloc_skb - allocate a network buffer
1312 * @size: size to allocate
1313 * @priority: allocation mask
1315 * This function is a convenient wrapper around __alloc_skb().
1317 static inline struct sk_buff *alloc_skb(unsigned int size,
1320 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1323 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1324 unsigned long data_len,
1328 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1330 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1331 struct sk_buff_fclones {
1332 struct sk_buff skb1;
1334 struct sk_buff skb2;
1336 refcount_t fclone_ref;
1340 * skb_fclone_busy - check if fclone is busy
1344 * Returns true if skb is a fast clone, and its clone is not freed.
1345 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1346 * so we also check that didn't happen.
1348 static inline bool skb_fclone_busy(const struct sock *sk,
1349 const struct sk_buff *skb)
1351 const struct sk_buff_fclones *fclones;
1353 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1355 return skb->fclone == SKB_FCLONE_ORIG &&
1356 refcount_read(&fclones->fclone_ref) > 1 &&
1357 READ_ONCE(fclones->skb2.sk) == sk;
1361 * alloc_skb_fclone - allocate a network buffer from fclone cache
1362 * @size: size to allocate
1363 * @priority: allocation mask
1365 * This function is a convenient wrapper around __alloc_skb().
1367 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1370 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1373 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1374 void skb_headers_offset_update(struct sk_buff *skb, int off);
1375 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1376 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1377 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1378 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1379 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1380 gfp_t gfp_mask, bool fclone);
1381 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1384 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1387 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1388 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1389 unsigned int headroom);
1390 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1391 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1392 int newtailroom, gfp_t priority);
1393 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1394 int offset, int len);
1395 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1396 int offset, int len);
1397 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1398 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1401 * skb_pad - zero pad the tail of an skb
1402 * @skb: buffer to pad
1403 * @pad: space to pad
1405 * Ensure that a buffer is followed by a padding area that is zero
1406 * filled. Used by network drivers which may DMA or transfer data
1407 * beyond the buffer end onto the wire.
1409 * May return error in out of memory cases. The skb is freed on error.
1411 static inline int skb_pad(struct sk_buff *skb, int pad)
1413 return __skb_pad(skb, pad, true);
1415 #define dev_kfree_skb(a) consume_skb(a)
1417 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1418 int offset, size_t size, size_t max_frags);
1420 struct skb_seq_state {
1424 __u32 stepped_offset;
1425 struct sk_buff *root_skb;
1426 struct sk_buff *cur_skb;
1431 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1432 unsigned int to, struct skb_seq_state *st);
1433 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1434 struct skb_seq_state *st);
1435 void skb_abort_seq_read(struct skb_seq_state *st);
1437 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1438 unsigned int to, struct ts_config *config);
1441 * Packet hash types specify the type of hash in skb_set_hash.
1443 * Hash types refer to the protocol layer addresses which are used to
1444 * construct a packet's hash. The hashes are used to differentiate or identify
1445 * flows of the protocol layer for the hash type. Hash types are either
1446 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1448 * Properties of hashes:
1450 * 1) Two packets in different flows have different hash values
1451 * 2) Two packets in the same flow should have the same hash value
1453 * A hash at a higher layer is considered to be more specific. A driver should
1454 * set the most specific hash possible.
1456 * A driver cannot indicate a more specific hash than the layer at which a hash
1457 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1459 * A driver may indicate a hash level which is less specific than the
1460 * actual layer the hash was computed on. For instance, a hash computed
1461 * at L4 may be considered an L3 hash. This should only be done if the
1462 * driver can't unambiguously determine that the HW computed the hash at
1463 * the higher layer. Note that the "should" in the second property above
1466 enum pkt_hash_types {
1467 PKT_HASH_TYPE_NONE, /* Undefined type */
1468 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1469 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1470 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1473 static inline void skb_clear_hash(struct sk_buff *skb)
1480 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1483 skb_clear_hash(skb);
1487 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1489 skb->l4_hash = is_l4;
1490 skb->sw_hash = is_sw;
1495 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1497 /* Used by drivers to set hash from HW */
1498 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1502 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1504 __skb_set_hash(skb, hash, true, is_l4);
1507 u32 __skb_get_hash_symmetric_net(const struct net *net, const struct sk_buff *skb);
1509 static inline u32 __skb_get_hash_symmetric(const struct sk_buff *skb)
1511 return __skb_get_hash_symmetric_net(NULL, skb);
1514 void __skb_get_hash_net(const struct net *net, struct sk_buff *skb);
1515 u32 skb_get_poff(const struct sk_buff *skb);
1516 u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1517 const struct flow_keys_basic *keys, int hlen);
1518 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1519 const void *data, int hlen_proto);
1521 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1522 int thoff, u8 ip_proto)
1524 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1527 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1528 const struct flow_dissector_key *key,
1529 unsigned int key_count);
1531 struct bpf_flow_dissector;
1532 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1533 __be16 proto, int nhoff, int hlen, unsigned int flags);
1535 bool __skb_flow_dissect(const struct net *net,
1536 const struct sk_buff *skb,
1537 struct flow_dissector *flow_dissector,
1538 void *target_container, const void *data,
1539 __be16 proto, int nhoff, int hlen, unsigned int flags);
1541 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1542 struct flow_dissector *flow_dissector,
1543 void *target_container, unsigned int flags)
1545 return __skb_flow_dissect(NULL, skb, flow_dissector,
1546 target_container, NULL, 0, 0, 0, flags);
1549 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1550 struct flow_keys *flow,
1553 memset(flow, 0, sizeof(*flow));
1554 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1555 flow, NULL, 0, 0, 0, flags);
1559 skb_flow_dissect_flow_keys_basic(const struct net *net,
1560 const struct sk_buff *skb,
1561 struct flow_keys_basic *flow,
1562 const void *data, __be16 proto,
1563 int nhoff, int hlen, unsigned int flags)
1565 memset(flow, 0, sizeof(*flow));
1566 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1567 data, proto, nhoff, hlen, flags);
1570 void skb_flow_dissect_meta(const struct sk_buff *skb,
1571 struct flow_dissector *flow_dissector,
1572 void *target_container);
1574 /* Gets a skb connection tracking info, ctinfo map should be a
1575 * map of mapsize to translate enum ip_conntrack_info states
1579 skb_flow_dissect_ct(const struct sk_buff *skb,
1580 struct flow_dissector *flow_dissector,
1581 void *target_container,
1582 u16 *ctinfo_map, size_t mapsize,
1583 bool post_ct, u16 zone);
1585 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1586 struct flow_dissector *flow_dissector,
1587 void *target_container);
1589 void skb_flow_dissect_hash(const struct sk_buff *skb,
1590 struct flow_dissector *flow_dissector,
1591 void *target_container);
1593 static inline __u32 skb_get_hash_net(const struct net *net, struct sk_buff *skb)
1595 if (!skb->l4_hash && !skb->sw_hash)
1596 __skb_get_hash_net(net, skb);
1601 static inline __u32 skb_get_hash(struct sk_buff *skb)
1603 if (!skb->l4_hash && !skb->sw_hash)
1604 __skb_get_hash_net(NULL, skb);
1609 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1611 if (!skb->l4_hash && !skb->sw_hash) {
1612 struct flow_keys keys;
1613 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1615 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1621 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1622 const siphash_key_t *perturb);
1624 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1629 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1631 to->hash = from->hash;
1632 to->sw_hash = from->sw_hash;
1633 to->l4_hash = from->l4_hash;
1636 static inline int skb_cmp_decrypted(const struct sk_buff *skb1,
1637 const struct sk_buff *skb2)
1639 #ifdef CONFIG_SKB_DECRYPTED
1640 return skb2->decrypted - skb1->decrypted;
1646 static inline bool skb_is_decrypted(const struct sk_buff *skb)
1648 #ifdef CONFIG_SKB_DECRYPTED
1649 return skb->decrypted;
1655 static inline void skb_copy_decrypted(struct sk_buff *to,
1656 const struct sk_buff *from)
1658 #ifdef CONFIG_SKB_DECRYPTED
1659 to->decrypted = from->decrypted;
1663 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1664 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1666 return skb->head + skb->end;
1669 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1674 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1679 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1684 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1686 return skb->end - skb->head;
1689 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1691 skb->end = skb->head + offset;
1695 extern const struct ubuf_info_ops msg_zerocopy_ubuf_ops;
1697 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1698 struct ubuf_info *uarg);
1700 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1702 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
1703 struct sk_buff *skb, struct iov_iter *from,
1706 int zerocopy_fill_skb_from_iter(struct sk_buff *skb,
1707 struct iov_iter *from, size_t length);
1709 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1710 struct msghdr *msg, int len)
1712 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len);
1715 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1716 struct msghdr *msg, int len,
1717 struct ubuf_info *uarg);
1720 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1722 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1724 return &skb_shinfo(skb)->hwtstamps;
1727 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1729 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1731 return is_zcopy ? skb_uarg(skb) : NULL;
1734 static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1736 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1739 static inline bool skb_zcopy_managed(const struct sk_buff *skb)
1741 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS;
1744 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1745 const struct sk_buff *skb2)
1747 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1750 static inline void net_zcopy_get(struct ubuf_info *uarg)
1752 refcount_inc(&uarg->refcnt);
1755 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1757 skb_shinfo(skb)->destructor_arg = uarg;
1758 skb_shinfo(skb)->flags |= uarg->flags;
1761 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1764 if (skb && uarg && !skb_zcopy(skb)) {
1765 if (unlikely(have_ref && *have_ref))
1768 net_zcopy_get(uarg);
1769 skb_zcopy_init(skb, uarg);
1773 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1775 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1776 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1779 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1781 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1784 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1786 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1789 static inline void net_zcopy_put(struct ubuf_info *uarg)
1792 uarg->ops->complete(NULL, uarg, true);
1795 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1798 if (uarg->ops == &msg_zerocopy_ubuf_ops)
1799 msg_zerocopy_put_abort(uarg, have_uref);
1801 net_zcopy_put(uarg);
1805 /* Release a reference on a zerocopy structure */
1806 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1808 struct ubuf_info *uarg = skb_zcopy(skb);
1811 if (!skb_zcopy_is_nouarg(skb))
1812 uarg->ops->complete(skb, uarg, zerocopy_success);
1814 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1818 void __skb_zcopy_downgrade_managed(struct sk_buff *skb);
1820 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb)
1822 if (unlikely(skb_zcopy_managed(skb)))
1823 __skb_zcopy_downgrade_managed(skb);
1826 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1831 static inline void skb_poison_list(struct sk_buff *skb)
1833 #ifdef CONFIG_DEBUG_NET
1834 skb->next = SKB_LIST_POISON_NEXT;
1838 /* Iterate through singly-linked GSO fragments of an skb. */
1839 #define skb_list_walk_safe(first, skb, next_skb) \
1840 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1841 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1843 static inline void skb_list_del_init(struct sk_buff *skb)
1845 __list_del_entry(&skb->list);
1846 skb_mark_not_on_list(skb);
1850 * skb_queue_empty - check if a queue is empty
1853 * Returns true if the queue is empty, false otherwise.
1855 static inline int skb_queue_empty(const struct sk_buff_head *list)
1857 return list->next == (const struct sk_buff *) list;
1861 * skb_queue_empty_lockless - check if a queue is empty
1864 * Returns true if the queue is empty, false otherwise.
1865 * This variant can be used in lockless contexts.
1867 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1869 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1874 * skb_queue_is_last - check if skb is the last entry in the queue
1878 * Returns true if @skb is the last buffer on the list.
1880 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1881 const struct sk_buff *skb)
1883 return skb->next == (const struct sk_buff *) list;
1887 * skb_queue_is_first - check if skb is the first entry in the queue
1891 * Returns true if @skb is the first buffer on the list.
1893 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1894 const struct sk_buff *skb)
1896 return skb->prev == (const struct sk_buff *) list;
1900 * skb_queue_next - return the next packet in the queue
1902 * @skb: current buffer
1904 * Return the next packet in @list after @skb. It is only valid to
1905 * call this if skb_queue_is_last() evaluates to false.
1907 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1908 const struct sk_buff *skb)
1910 /* This BUG_ON may seem severe, but if we just return then we
1911 * are going to dereference garbage.
1913 BUG_ON(skb_queue_is_last(list, skb));
1918 * skb_queue_prev - return the prev packet in the queue
1920 * @skb: current buffer
1922 * Return the prev packet in @list before @skb. It is only valid to
1923 * call this if skb_queue_is_first() evaluates to false.
1925 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1926 const struct sk_buff *skb)
1928 /* This BUG_ON may seem severe, but if we just return then we
1929 * are going to dereference garbage.
1931 BUG_ON(skb_queue_is_first(list, skb));
1936 * skb_get - reference buffer
1937 * @skb: buffer to reference
1939 * Makes another reference to a socket buffer and returns a pointer
1942 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1944 refcount_inc(&skb->users);
1949 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1953 * skb_cloned - is the buffer a clone
1954 * @skb: buffer to check
1956 * Returns true if the buffer was generated with skb_clone() and is
1957 * one of multiple shared copies of the buffer. Cloned buffers are
1958 * shared data so must not be written to under normal circumstances.
1960 static inline int skb_cloned(const struct sk_buff *skb)
1962 return skb->cloned &&
1963 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1966 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1968 might_sleep_if(gfpflags_allow_blocking(pri));
1970 if (skb_cloned(skb))
1971 return pskb_expand_head(skb, 0, 0, pri);
1976 /* This variant of skb_unclone() makes sure skb->truesize
1977 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
1979 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
1980 * when various debugging features are in place.
1982 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
1983 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1985 might_sleep_if(gfpflags_allow_blocking(pri));
1987 if (skb_cloned(skb))
1988 return __skb_unclone_keeptruesize(skb, pri);
1993 * skb_header_cloned - is the header a clone
1994 * @skb: buffer to check
1996 * Returns true if modifying the header part of the buffer requires
1997 * the data to be copied.
1999 static inline int skb_header_cloned(const struct sk_buff *skb)
2006 dataref = atomic_read(&skb_shinfo(skb)->dataref);
2007 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
2008 return dataref != 1;
2011 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
2013 might_sleep_if(gfpflags_allow_blocking(pri));
2015 if (skb_header_cloned(skb))
2016 return pskb_expand_head(skb, 0, 0, pri);
2022 * __skb_header_release() - allow clones to use the headroom
2023 * @skb: buffer to operate on
2025 * See "DOC: dataref and headerless skbs".
2027 static inline void __skb_header_release(struct sk_buff *skb)
2030 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
2035 * skb_shared - is the buffer shared
2036 * @skb: buffer to check
2038 * Returns true if more than one person has a reference to this
2041 static inline int skb_shared(const struct sk_buff *skb)
2043 return refcount_read(&skb->users) != 1;
2047 * skb_share_check - check if buffer is shared and if so clone it
2048 * @skb: buffer to check
2049 * @pri: priority for memory allocation
2051 * If the buffer is shared the buffer is cloned and the old copy
2052 * drops a reference. A new clone with a single reference is returned.
2053 * If the buffer is not shared the original buffer is returned. When
2054 * being called from interrupt status or with spinlocks held pri must
2057 * NULL is returned on a memory allocation failure.
2059 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
2061 might_sleep_if(gfpflags_allow_blocking(pri));
2062 if (skb_shared(skb)) {
2063 struct sk_buff *nskb = skb_clone(skb, pri);
2075 * Copy shared buffers into a new sk_buff. We effectively do COW on
2076 * packets to handle cases where we have a local reader and forward
2077 * and a couple of other messy ones. The normal one is tcpdumping
2078 * a packet that's being forwarded.
2082 * skb_unshare - make a copy of a shared buffer
2083 * @skb: buffer to check
2084 * @pri: priority for memory allocation
2086 * If the socket buffer is a clone then this function creates a new
2087 * copy of the data, drops a reference count on the old copy and returns
2088 * the new copy with the reference count at 1. If the buffer is not a clone
2089 * the original buffer is returned. When called with a spinlock held or
2090 * from interrupt state @pri must be %GFP_ATOMIC
2092 * %NULL is returned on a memory allocation failure.
2094 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
2097 might_sleep_if(gfpflags_allow_blocking(pri));
2098 if (skb_cloned(skb)) {
2099 struct sk_buff *nskb = skb_copy(skb, pri);
2101 /* Free our shared copy */
2112 * skb_peek - peek at the head of an &sk_buff_head
2113 * @list_: list to peek at
2115 * Peek an &sk_buff. Unlike most other operations you _MUST_
2116 * be careful with this one. A peek leaves the buffer on the
2117 * list and someone else may run off with it. You must hold
2118 * the appropriate locks or have a private queue to do this.
2120 * Returns %NULL for an empty list or a pointer to the head element.
2121 * The reference count is not incremented and the reference is therefore
2122 * volatile. Use with caution.
2124 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2126 struct sk_buff *skb = list_->next;
2128 if (skb == (struct sk_buff *)list_)
2134 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2135 * @list_: list to peek at
2137 * Like skb_peek(), but the caller knows that the list is not empty.
2139 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2145 * skb_peek_next - peek skb following the given one from a queue
2146 * @skb: skb to start from
2147 * @list_: list to peek at
2149 * Returns %NULL when the end of the list is met or a pointer to the
2150 * next element. The reference count is not incremented and the
2151 * reference is therefore volatile. Use with caution.
2153 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2154 const struct sk_buff_head *list_)
2156 struct sk_buff *next = skb->next;
2158 if (next == (struct sk_buff *)list_)
2164 * skb_peek_tail - peek at the tail of an &sk_buff_head
2165 * @list_: list to peek at
2167 * Peek an &sk_buff. Unlike most other operations you _MUST_
2168 * be careful with this one. A peek leaves the buffer on the
2169 * list and someone else may run off with it. You must hold
2170 * the appropriate locks or have a private queue to do this.
2172 * Returns %NULL for an empty list or a pointer to the tail element.
2173 * The reference count is not incremented and the reference is therefore
2174 * volatile. Use with caution.
2176 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2178 struct sk_buff *skb = READ_ONCE(list_->prev);
2180 if (skb == (struct sk_buff *)list_)
2187 * skb_queue_len - get queue length
2188 * @list_: list to measure
2190 * Return the length of an &sk_buff queue.
2192 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2198 * skb_queue_len_lockless - get queue length
2199 * @list_: list to measure
2201 * Return the length of an &sk_buff queue.
2202 * This variant can be used in lockless contexts.
2204 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2206 return READ_ONCE(list_->qlen);
2210 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2211 * @list: queue to initialize
2213 * This initializes only the list and queue length aspects of
2214 * an sk_buff_head object. This allows to initialize the list
2215 * aspects of an sk_buff_head without reinitializing things like
2216 * the spinlock. It can also be used for on-stack sk_buff_head
2217 * objects where the spinlock is known to not be used.
2219 static inline void __skb_queue_head_init(struct sk_buff_head *list)
2221 list->prev = list->next = (struct sk_buff *)list;
2226 * This function creates a split out lock class for each invocation;
2227 * this is needed for now since a whole lot of users of the skb-queue
2228 * infrastructure in drivers have different locking usage (in hardirq)
2229 * than the networking core (in softirq only). In the long run either the
2230 * network layer or drivers should need annotation to consolidate the
2231 * main types of usage into 3 classes.
2233 static inline void skb_queue_head_init(struct sk_buff_head *list)
2235 spin_lock_init(&list->lock);
2236 __skb_queue_head_init(list);
2239 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2240 struct lock_class_key *class)
2242 skb_queue_head_init(list);
2243 lockdep_set_class(&list->lock, class);
2247 * Insert an sk_buff on a list.
2249 * The "__skb_xxxx()" functions are the non-atomic ones that
2250 * can only be called with interrupts disabled.
2252 static inline void __skb_insert(struct sk_buff *newsk,
2253 struct sk_buff *prev, struct sk_buff *next,
2254 struct sk_buff_head *list)
2256 /* See skb_queue_empty_lockless() and skb_peek_tail()
2257 * for the opposite READ_ONCE()
2259 WRITE_ONCE(newsk->next, next);
2260 WRITE_ONCE(newsk->prev, prev);
2261 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2262 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2263 WRITE_ONCE(list->qlen, list->qlen + 1);
2266 static inline void __skb_queue_splice(const struct sk_buff_head *list,
2267 struct sk_buff *prev,
2268 struct sk_buff *next)
2270 struct sk_buff *first = list->next;
2271 struct sk_buff *last = list->prev;
2273 WRITE_ONCE(first->prev, prev);
2274 WRITE_ONCE(prev->next, first);
2276 WRITE_ONCE(last->next, next);
2277 WRITE_ONCE(next->prev, last);
2281 * skb_queue_splice - join two skb lists, this is designed for stacks
2282 * @list: the new list to add
2283 * @head: the place to add it in the first list
2285 static inline void skb_queue_splice(const struct sk_buff_head *list,
2286 struct sk_buff_head *head)
2288 if (!skb_queue_empty(list)) {
2289 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2290 head->qlen += list->qlen;
2295 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2296 * @list: the new list to add
2297 * @head: the place to add it in the first list
2299 * The list at @list is reinitialised
2301 static inline void skb_queue_splice_init(struct sk_buff_head *list,
2302 struct sk_buff_head *head)
2304 if (!skb_queue_empty(list)) {
2305 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2306 head->qlen += list->qlen;
2307 __skb_queue_head_init(list);
2312 * skb_queue_splice_tail - join two skb lists, each list being a queue
2313 * @list: the new list to add
2314 * @head: the place to add it in the first list
2316 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2317 struct sk_buff_head *head)
2319 if (!skb_queue_empty(list)) {
2320 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2321 head->qlen += list->qlen;
2326 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2327 * @list: the new list to add
2328 * @head: the place to add it in the first list
2330 * Each of the lists is a queue.
2331 * The list at @list is reinitialised
2333 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2334 struct sk_buff_head *head)
2336 if (!skb_queue_empty(list)) {
2337 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2338 head->qlen += list->qlen;
2339 __skb_queue_head_init(list);
2344 * __skb_queue_after - queue a buffer at the list head
2345 * @list: list to use
2346 * @prev: place after this buffer
2347 * @newsk: buffer to queue
2349 * Queue a buffer int the middle 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_after(struct sk_buff_head *list,
2355 struct sk_buff *prev,
2356 struct sk_buff *newsk)
2358 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2361 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2362 struct sk_buff_head *list);
2364 static inline void __skb_queue_before(struct sk_buff_head *list,
2365 struct sk_buff *next,
2366 struct sk_buff *newsk)
2368 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2372 * __skb_queue_head - queue a buffer at the list head
2373 * @list: list to use
2374 * @newsk: buffer to queue
2376 * Queue a buffer at the start of a list. This function takes no locks
2377 * and you must therefore hold required locks before calling it.
2379 * A buffer cannot be placed on two lists at the same time.
2381 static inline void __skb_queue_head(struct sk_buff_head *list,
2382 struct sk_buff *newsk)
2384 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2386 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2389 * __skb_queue_tail - queue a buffer at the list tail
2390 * @list: list to use
2391 * @newsk: buffer to queue
2393 * Queue a buffer at the end of a list. This function takes no locks
2394 * and you must therefore hold required locks before calling it.
2396 * A buffer cannot be placed on two lists at the same time.
2398 static inline void __skb_queue_tail(struct sk_buff_head *list,
2399 struct sk_buff *newsk)
2401 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2403 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2406 * remove sk_buff from list. _Must_ be called atomically, and with
2409 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2410 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2412 struct sk_buff *next, *prev;
2414 WRITE_ONCE(list->qlen, list->qlen - 1);
2417 skb->next = skb->prev = NULL;
2418 WRITE_ONCE(next->prev, prev);
2419 WRITE_ONCE(prev->next, next);
2423 * __skb_dequeue - remove from the head of the queue
2424 * @list: list to dequeue from
2426 * Remove the head of the list. This function does not take any locks
2427 * so must be used with appropriate locks held only. The head item is
2428 * returned or %NULL if the list is empty.
2430 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2432 struct sk_buff *skb = skb_peek(list);
2434 __skb_unlink(skb, list);
2437 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2440 * __skb_dequeue_tail - remove from the tail of the queue
2441 * @list: list to dequeue from
2443 * Remove the tail of the list. This function does not take any locks
2444 * so must be used with appropriate locks held only. The tail item is
2445 * returned or %NULL if the list is empty.
2447 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2449 struct sk_buff *skb = skb_peek_tail(list);
2451 __skb_unlink(skb, list);
2454 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2457 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2459 return skb->data_len;
2462 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2464 return skb->len - skb->data_len;
2467 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2469 unsigned int i, len = 0;
2471 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2472 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2476 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2478 return skb_headlen(skb) + __skb_pagelen(skb);
2481 static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag,
2482 netmem_ref netmem, int off,
2485 frag->netmem = netmem;
2487 skb_frag_size_set(frag, size);
2490 static inline void skb_frag_fill_page_desc(skb_frag_t *frag,
2494 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size);
2497 static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo,
2498 int i, netmem_ref netmem,
2501 skb_frag_t *frag = &shinfo->frags[i];
2503 skb_frag_fill_netmem_desc(frag, netmem, off, size);
2506 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo,
2507 int i, struct page *page,
2510 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off,
2515 * skb_len_add - adds a number to len fields of skb
2516 * @skb: buffer to add len to
2517 * @delta: number of bytes to add
2519 static inline void skb_len_add(struct sk_buff *skb, int delta)
2522 skb->data_len += delta;
2523 skb->truesize += delta;
2527 * __skb_fill_netmem_desc - initialise a fragment in an skb
2528 * @skb: buffer containing fragment to be initialised
2529 * @i: fragment index to initialise
2530 * @netmem: the netmem to use for this fragment
2531 * @off: the offset to the data with @page
2532 * @size: the length of the data
2534 * Initialises the @i'th fragment of @skb to point to &size bytes at
2535 * offset @off within @page.
2537 * Does not take any additional reference on the fragment.
2539 static inline void __skb_fill_netmem_desc(struct sk_buff *skb, int i,
2540 netmem_ref netmem, int off, int size)
2542 struct page *page = netmem_to_page(netmem);
2544 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size);
2546 /* Propagate page pfmemalloc to the skb if we can. The problem is
2547 * that not all callers have unique ownership of the page but rely
2548 * on page_is_pfmemalloc doing the right thing(tm).
2550 page = compound_head(page);
2551 if (page_is_pfmemalloc(page))
2552 skb->pfmemalloc = true;
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);
2561 static inline void skb_fill_netmem_desc(struct sk_buff *skb, int i,
2562 netmem_ref netmem, int off, int size)
2564 __skb_fill_netmem_desc(skb, i, netmem, off, size);
2565 skb_shinfo(skb)->nr_frags = i + 1;
2569 * skb_fill_page_desc - initialise a paged fragment in an skb
2570 * @skb: buffer containing fragment to be initialised
2571 * @i: paged fragment index to initialise
2572 * @page: the page to use for this fragment
2573 * @off: the offset to the data with @page
2574 * @size: the length of the data
2576 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2577 * @skb to point to @size bytes at offset @off within @page. In
2578 * addition updates @skb such that @i is the last fragment.
2580 * Does not take any additional reference on the fragment.
2582 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2583 struct page *page, int off, int size)
2585 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size);
2589 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb
2590 * @skb: buffer containing fragment to be initialised
2591 * @i: paged fragment index to initialise
2592 * @page: the page to use for this fragment
2593 * @off: the offset to the data with @page
2594 * @size: the length of the data
2596 * Variant of skb_fill_page_desc() which does not deal with
2597 * pfmemalloc, if page is not owned by us.
2599 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i,
2600 struct page *page, int off,
2603 struct skb_shared_info *shinfo = skb_shinfo(skb);
2605 __skb_fill_page_desc_noacc(shinfo, i, page, off, size);
2606 shinfo->nr_frags = i + 1;
2609 void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem,
2610 int off, int size, unsigned int truesize);
2612 static inline void skb_add_rx_frag(struct sk_buff *skb, int i,
2613 struct page *page, int off, int size,
2614 unsigned int truesize)
2616 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size,
2620 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2621 unsigned int truesize);
2623 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2625 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2626 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2628 return skb->head + skb->tail;
2631 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2633 skb->tail = skb->data - skb->head;
2636 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2638 skb_reset_tail_pointer(skb);
2639 skb->tail += offset;
2642 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2643 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2648 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2650 skb->tail = skb->data;
2653 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2655 skb->tail = skb->data + offset;
2658 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2660 static inline void skb_assert_len(struct sk_buff *skb)
2662 #ifdef CONFIG_DEBUG_NET
2663 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2664 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2665 #endif /* CONFIG_DEBUG_NET */
2669 * Add data to an sk_buff
2671 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2672 void *skb_put(struct sk_buff *skb, unsigned int len);
2673 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2675 void *tmp = skb_tail_pointer(skb);
2676 SKB_LINEAR_ASSERT(skb);
2682 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2684 void *tmp = __skb_put(skb, len);
2686 memset(tmp, 0, len);
2690 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2693 void *tmp = __skb_put(skb, len);
2695 memcpy(tmp, data, len);
2699 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2701 *(u8 *)__skb_put(skb, 1) = val;
2704 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2706 void *tmp = skb_put(skb, len);
2708 memset(tmp, 0, len);
2713 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2716 void *tmp = skb_put(skb, len);
2718 memcpy(tmp, data, len);
2723 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2725 *(u8 *)skb_put(skb, 1) = val;
2728 void *skb_push(struct sk_buff *skb, unsigned int len);
2729 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2731 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2738 void *skb_pull(struct sk_buff *skb, unsigned int len);
2739 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2741 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2744 if (unlikely(skb->len < skb->data_len)) {
2745 #if defined(CONFIG_DEBUG_NET)
2747 pr_err("__skb_pull(len=%u)\n", len);
2748 skb_dump(KERN_ERR, skb, false);
2752 return skb->data += len;
2755 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2757 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2760 void *skb_pull_data(struct sk_buff *skb, size_t len);
2762 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2764 static inline enum skb_drop_reason
2765 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len)
2767 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2769 if (likely(len <= skb_headlen(skb)))
2770 return SKB_NOT_DROPPED_YET;
2772 if (unlikely(len > skb->len))
2773 return SKB_DROP_REASON_PKT_TOO_SMALL;
2775 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb))))
2776 return SKB_DROP_REASON_NOMEM;
2778 return SKB_NOT_DROPPED_YET;
2781 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2783 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET;
2786 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2788 if (!pskb_may_pull(skb, len))
2792 return skb->data += len;
2795 void skb_condense(struct sk_buff *skb);
2798 * skb_headroom - bytes at buffer head
2799 * @skb: buffer to check
2801 * Return the number of bytes of free space at the head of an &sk_buff.
2803 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2805 return skb->data - skb->head;
2809 * skb_tailroom - bytes at buffer end
2810 * @skb: buffer to check
2812 * Return the number of bytes of free space at the tail of an sk_buff
2814 static inline int skb_tailroom(const struct sk_buff *skb)
2816 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2820 * skb_availroom - bytes at buffer end
2821 * @skb: buffer to check
2823 * Return the number of bytes of free space at the tail of an sk_buff
2824 * allocated by sk_stream_alloc()
2826 static inline int skb_availroom(const struct sk_buff *skb)
2828 if (skb_is_nonlinear(skb))
2831 return skb->end - skb->tail - skb->reserved_tailroom;
2835 * skb_reserve - adjust headroom
2836 * @skb: buffer to alter
2837 * @len: bytes to move
2839 * Increase the headroom of an empty &sk_buff by reducing the tail
2840 * room. This is only allowed for an empty buffer.
2842 static inline void skb_reserve(struct sk_buff *skb, int len)
2849 * skb_tailroom_reserve - adjust reserved_tailroom
2850 * @skb: buffer to alter
2851 * @mtu: maximum amount of headlen permitted
2852 * @needed_tailroom: minimum amount of reserved_tailroom
2854 * Set reserved_tailroom so that headlen can be as large as possible but
2855 * not larger than mtu and tailroom cannot be smaller than
2857 * The required headroom should already have been reserved before using
2860 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2861 unsigned int needed_tailroom)
2863 SKB_LINEAR_ASSERT(skb);
2864 if (mtu < skb_tailroom(skb) - needed_tailroom)
2865 /* use at most mtu */
2866 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2868 /* use up to all available space */
2869 skb->reserved_tailroom = needed_tailroom;
2872 #define ENCAP_TYPE_ETHER 0
2873 #define ENCAP_TYPE_IPPROTO 1
2875 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2878 skb->inner_protocol = protocol;
2879 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2882 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2885 skb->inner_ipproto = ipproto;
2886 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2889 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2891 skb->inner_mac_header = skb->mac_header;
2892 skb->inner_network_header = skb->network_header;
2893 skb->inner_transport_header = skb->transport_header;
2896 static inline void skb_reset_mac_len(struct sk_buff *skb)
2898 skb->mac_len = skb->network_header - skb->mac_header;
2901 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2904 return skb->head + skb->inner_transport_header;
2907 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2909 return skb_inner_transport_header(skb) - skb->data;
2912 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2914 skb->inner_transport_header = skb->data - skb->head;
2917 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2920 skb_reset_inner_transport_header(skb);
2921 skb->inner_transport_header += offset;
2924 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2926 return skb->head + skb->inner_network_header;
2929 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2931 skb->inner_network_header = skb->data - skb->head;
2934 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2937 skb_reset_inner_network_header(skb);
2938 skb->inner_network_header += offset;
2941 static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb)
2943 return skb->inner_network_header > 0;
2946 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2948 return skb->head + skb->inner_mac_header;
2951 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2953 skb->inner_mac_header = skb->data - skb->head;
2956 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2959 skb_reset_inner_mac_header(skb);
2960 skb->inner_mac_header += offset;
2962 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2964 return skb->transport_header != (typeof(skb->transport_header))~0U;
2967 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2969 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
2970 return skb->head + skb->transport_header;
2973 static inline void skb_reset_transport_header(struct sk_buff *skb)
2975 skb->transport_header = skb->data - skb->head;
2978 static inline void skb_set_transport_header(struct sk_buff *skb,
2981 skb_reset_transport_header(skb);
2982 skb->transport_header += offset;
2985 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2987 return skb->head + skb->network_header;
2990 static inline void skb_reset_network_header(struct sk_buff *skb)
2992 skb->network_header = skb->data - skb->head;
2995 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2997 skb_reset_network_header(skb);
2998 skb->network_header += offset;
3001 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
3003 return skb->mac_header != (typeof(skb->mac_header))~0U;
3006 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
3008 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
3009 return skb->head + skb->mac_header;
3012 static inline int skb_mac_offset(const struct sk_buff *skb)
3014 return skb_mac_header(skb) - skb->data;
3017 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
3019 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
3020 return skb->network_header - skb->mac_header;
3023 static inline void skb_unset_mac_header(struct sk_buff *skb)
3025 skb->mac_header = (typeof(skb->mac_header))~0U;
3028 static inline void skb_reset_mac_header(struct sk_buff *skb)
3030 skb->mac_header = skb->data - skb->head;
3033 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
3035 skb_reset_mac_header(skb);
3036 skb->mac_header += offset;
3039 static inline void skb_pop_mac_header(struct sk_buff *skb)
3041 skb->mac_header = skb->network_header;
3044 static inline void skb_probe_transport_header(struct sk_buff *skb)
3046 struct flow_keys_basic keys;
3048 if (skb_transport_header_was_set(skb))
3051 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
3053 skb_set_transport_header(skb, keys.control.thoff);
3056 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
3058 if (skb_mac_header_was_set(skb)) {
3059 const unsigned char *old_mac = skb_mac_header(skb);
3061 skb_set_mac_header(skb, -skb->mac_len);
3062 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
3066 /* Move the full mac header up to current network_header.
3067 * Leaves skb->data pointing at offset skb->mac_len into the mac_header.
3068 * Must be provided the complete mac header length.
3070 static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len)
3072 if (skb_mac_header_was_set(skb)) {
3073 const unsigned char *old_mac = skb_mac_header(skb);
3075 skb_set_mac_header(skb, -full_mac_len);
3076 memmove(skb_mac_header(skb), old_mac, full_mac_len);
3077 __skb_push(skb, full_mac_len - skb->mac_len);
3081 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
3083 return skb->csum_start - skb_headroom(skb);
3086 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
3088 return skb->head + skb->csum_start;
3091 static inline int skb_transport_offset(const struct sk_buff *skb)
3093 return skb_transport_header(skb) - skb->data;
3096 static inline u32 skb_network_header_len(const struct sk_buff *skb)
3098 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
3099 return skb->transport_header - skb->network_header;
3102 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
3104 return skb->inner_transport_header - skb->inner_network_header;
3107 static inline int skb_network_offset(const struct sk_buff *skb)
3109 return skb_network_header(skb) - skb->data;
3112 static inline int skb_inner_network_offset(const struct sk_buff *skb)
3114 return skb_inner_network_header(skb) - skb->data;
3117 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
3119 return pskb_may_pull(skb, skb_network_offset(skb) + len);
3123 * CPUs often take a performance hit when accessing unaligned memory
3124 * locations. The actual performance hit varies, it can be small if the
3125 * hardware handles it or large if we have to take an exception and fix it
3128 * Since an ethernet header is 14 bytes network drivers often end up with
3129 * the IP header at an unaligned offset. The IP header can be aligned by
3130 * shifting the start of the packet by 2 bytes. Drivers should do this
3133 * skb_reserve(skb, NET_IP_ALIGN);
3135 * The downside to this alignment of the IP header is that the DMA is now
3136 * unaligned. On some architectures the cost of an unaligned DMA is high
3137 * and this cost outweighs the gains made by aligning the IP header.
3139 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
3142 #ifndef NET_IP_ALIGN
3143 #define NET_IP_ALIGN 2
3147 * The networking layer reserves some headroom in skb data (via
3148 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
3149 * the header has to grow. In the default case, if the header has to grow
3150 * 32 bytes or less we avoid the reallocation.
3152 * Unfortunately this headroom changes the DMA alignment of the resulting
3153 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
3154 * on some architectures. An architecture can override this value,
3155 * perhaps setting it to a cacheline in size (since that will maintain
3156 * cacheline alignment of the DMA). It must be a power of 2.
3158 * Various parts of the networking layer expect at least 32 bytes of
3159 * headroom, you should not reduce this.
3161 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
3162 * to reduce average number of cache lines per packet.
3163 * get_rps_cpu() for example only access one 64 bytes aligned block :
3164 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
3167 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
3170 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
3172 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
3174 if (WARN_ON(skb_is_nonlinear(skb)))
3177 skb_set_tail_pointer(skb, len);
3180 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
3182 __skb_set_length(skb, len);
3185 void skb_trim(struct sk_buff *skb, unsigned int len);
3187 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
3190 return ___pskb_trim(skb, len);
3191 __skb_trim(skb, len);
3195 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
3197 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
3201 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
3202 * @skb: buffer to alter
3205 * This is identical to pskb_trim except that the caller knows that
3206 * the skb is not cloned so we should never get an error due to out-
3209 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3211 int err = pskb_trim(skb, len);
3215 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3217 unsigned int diff = len - skb->len;
3219 if (skb_tailroom(skb) < diff) {
3220 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3225 __skb_set_length(skb, len);
3230 * skb_orphan - orphan a buffer
3231 * @skb: buffer to orphan
3233 * If a buffer currently has an owner then we call the owner's
3234 * destructor function and make the @skb unowned. The buffer continues
3235 * to exist but is no longer charged to its former owner.
3237 static inline void skb_orphan(struct sk_buff *skb)
3239 if (skb->destructor) {
3240 skb->destructor(skb);
3241 skb->destructor = NULL;
3249 * skb_orphan_frags - orphan the frags contained in a buffer
3250 * @skb: buffer to orphan frags from
3251 * @gfp_mask: allocation mask for replacement pages
3253 * For each frag in the SKB which needs a destructor (i.e. has an
3254 * owner) create a copy of that frag and release the original
3255 * page by calling the destructor.
3257 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3259 if (likely(!skb_zcopy(skb)))
3261 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN)
3263 return skb_copy_ubufs(skb, gfp_mask);
3266 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3267 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3269 if (likely(!skb_zcopy(skb)))
3271 return skb_copy_ubufs(skb, gfp_mask);
3275 * __skb_queue_purge_reason - empty a list
3276 * @list: list to empty
3277 * @reason: drop reason
3279 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3280 * the list and one reference dropped. This function does not take the
3281 * list lock and the caller must hold the relevant locks to use it.
3283 static inline void __skb_queue_purge_reason(struct sk_buff_head *list,
3284 enum skb_drop_reason reason)
3286 struct sk_buff *skb;
3288 while ((skb = __skb_dequeue(list)) != NULL)
3289 kfree_skb_reason(skb, reason);
3292 static inline void __skb_queue_purge(struct sk_buff_head *list)
3294 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE);
3297 void skb_queue_purge_reason(struct sk_buff_head *list,
3298 enum skb_drop_reason reason);
3300 static inline void skb_queue_purge(struct sk_buff_head *list)
3302 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE);
3305 unsigned int skb_rbtree_purge(struct rb_root *root);
3306 void skb_errqueue_purge(struct sk_buff_head *list);
3308 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3311 * netdev_alloc_frag - allocate a page fragment
3312 * @fragsz: fragment size
3314 * Allocates a frag from a page for receive buffer.
3315 * Uses GFP_ATOMIC allocations.
3317 static inline void *netdev_alloc_frag(unsigned int fragsz)
3319 return __netdev_alloc_frag_align(fragsz, ~0u);
3322 static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3325 WARN_ON_ONCE(!is_power_of_2(align));
3326 return __netdev_alloc_frag_align(fragsz, -align);
3329 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3333 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3334 * @dev: network device to receive on
3335 * @length: length to allocate
3337 * Allocate a new &sk_buff and assign it a usage count of one. The
3338 * buffer has unspecified headroom built in. Users should allocate
3339 * the headroom they think they need without accounting for the
3340 * built in space. The built in space is used for optimisations.
3342 * %NULL is returned if there is no free memory. Although this function
3343 * allocates memory it can be called from an interrupt.
3345 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3346 unsigned int length)
3348 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3351 /* legacy helper around __netdev_alloc_skb() */
3352 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3355 return __netdev_alloc_skb(NULL, length, gfp_mask);
3358 /* legacy helper around netdev_alloc_skb() */
3359 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3361 return netdev_alloc_skb(NULL, length);
3365 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3366 unsigned int length, gfp_t gfp)
3368 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3370 if (NET_IP_ALIGN && skb)
3371 skb_reserve(skb, NET_IP_ALIGN);
3375 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3376 unsigned int length)
3378 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3381 static inline void skb_free_frag(void *addr)
3383 page_frag_free(addr);
3386 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3388 static inline void *napi_alloc_frag(unsigned int fragsz)
3390 return __napi_alloc_frag_align(fragsz, ~0u);
3393 static inline void *napi_alloc_frag_align(unsigned int fragsz,
3396 WARN_ON_ONCE(!is_power_of_2(align));
3397 return __napi_alloc_frag_align(fragsz, -align);
3400 struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int length);
3401 void napi_consume_skb(struct sk_buff *skb, int budget);
3403 void napi_skb_free_stolen_head(struct sk_buff *skb);
3404 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason);
3407 * __dev_alloc_pages - allocate page for network Rx
3408 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3409 * @order: size of the allocation
3411 * Allocate a new page.
3413 * %NULL is returned if there is no free memory.
3415 static inline struct page *__dev_alloc_pages_noprof(gfp_t gfp_mask,
3418 /* This piece of code contains several assumptions.
3419 * 1. This is for device Rx, therefore a cold page is preferred.
3420 * 2. The expectation is the user wants a compound page.
3421 * 3. If requesting a order 0 page it will not be compound
3422 * due to the check to see if order has a value in prep_new_page
3423 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3424 * code in gfp_to_alloc_flags that should be enforcing this.
3426 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3428 return alloc_pages_node_noprof(NUMA_NO_NODE, gfp_mask, order);
3430 #define __dev_alloc_pages(...) alloc_hooks(__dev_alloc_pages_noprof(__VA_ARGS__))
3433 * This specialized allocator has to be a macro for its allocations to be
3434 * accounted separately (to have a separate alloc_tag).
3436 #define dev_alloc_pages(_order) __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, _order)
3439 * __dev_alloc_page - allocate a page for network Rx
3440 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3442 * Allocate a new page.
3444 * %NULL is returned if there is no free memory.
3446 static inline struct page *__dev_alloc_page_noprof(gfp_t gfp_mask)
3448 return __dev_alloc_pages_noprof(gfp_mask, 0);
3450 #define __dev_alloc_page(...) alloc_hooks(__dev_alloc_page_noprof(__VA_ARGS__))
3453 * This specialized allocator has to be a macro for its allocations to be
3454 * accounted separately (to have a separate alloc_tag).
3456 #define dev_alloc_page() dev_alloc_pages(0)
3459 * dev_page_is_reusable - check whether a page can be reused for network Rx
3460 * @page: the page to test
3462 * A page shouldn't be considered for reusing/recycling if it was allocated
3463 * under memory pressure or at a distant memory node.
3465 * Returns false if this page should be returned to page allocator, true
3468 static inline bool dev_page_is_reusable(const struct page *page)
3470 return likely(page_to_nid(page) == numa_mem_id() &&
3471 !page_is_pfmemalloc(page));
3475 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3476 * @page: The page that was allocated from skb_alloc_page
3477 * @skb: The skb that may need pfmemalloc set
3479 static inline void skb_propagate_pfmemalloc(const struct page *page,
3480 struct sk_buff *skb)
3482 if (page_is_pfmemalloc(page))
3483 skb->pfmemalloc = true;
3487 * skb_frag_off() - Returns the offset of a skb fragment
3488 * @frag: the paged fragment
3490 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3492 return frag->offset;
3496 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3497 * @frag: skb fragment
3498 * @delta: value to add
3500 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3502 frag->offset += delta;
3506 * skb_frag_off_set() - Sets the offset of a skb fragment
3507 * @frag: skb fragment
3508 * @offset: offset of fragment
3510 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3512 frag->offset = offset;
3516 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3517 * @fragto: skb fragment where offset is set
3518 * @fragfrom: skb fragment offset is copied from
3520 static inline void skb_frag_off_copy(skb_frag_t *fragto,
3521 const skb_frag_t *fragfrom)
3523 fragto->offset = fragfrom->offset;
3527 * skb_frag_page - retrieve the page referred to by a paged fragment
3528 * @frag: the paged fragment
3530 * Returns the &struct page associated with @frag.
3532 static inline struct page *skb_frag_page(const skb_frag_t *frag)
3534 return netmem_to_page(frag->netmem);
3537 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb,
3538 unsigned int headroom);
3539 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb,
3540 struct bpf_prog *prog);
3542 * skb_frag_address - gets the address of the data contained in a paged fragment
3543 * @frag: the paged fragment buffer
3545 * Returns the address of the data within @frag. The page must already
3548 static inline void *skb_frag_address(const skb_frag_t *frag)
3550 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3554 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3555 * @frag: the paged fragment buffer
3557 * Returns the address of the data within @frag. Checks that the page
3558 * is mapped and returns %NULL otherwise.
3560 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3562 void *ptr = page_address(skb_frag_page(frag));
3566 return ptr + skb_frag_off(frag);
3570 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3571 * @fragto: skb fragment where page is set
3572 * @fragfrom: skb fragment page is copied from
3574 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3575 const skb_frag_t *fragfrom)
3577 fragto->netmem = fragfrom->netmem;
3580 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3583 * skb_frag_dma_map - maps a paged fragment via the DMA API
3584 * @dev: the device to map the fragment to
3585 * @frag: the paged fragment to map
3586 * @offset: the offset within the fragment (starting at the
3587 * fragment's own offset)
3588 * @size: the number of bytes to map
3589 * @dir: the direction of the mapping (``PCI_DMA_*``)
3591 * Maps the page associated with @frag to @device.
3593 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3594 const skb_frag_t *frag,
3595 size_t offset, size_t size,
3596 enum dma_data_direction dir)
3598 return dma_map_page(dev, skb_frag_page(frag),
3599 skb_frag_off(frag) + offset, size, dir);
3602 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3605 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3609 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3612 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3617 * skb_clone_writable - is the header of a clone writable
3618 * @skb: buffer to check
3619 * @len: length up to which to write
3621 * Returns true if modifying the header part of the cloned buffer
3622 * does not requires the data to be copied.
3624 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3626 return !skb_header_cloned(skb) &&
3627 skb_headroom(skb) + len <= skb->hdr_len;
3630 static inline int skb_try_make_writable(struct sk_buff *skb,
3631 unsigned int write_len)
3633 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3634 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3637 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3642 if (headroom > skb_headroom(skb))
3643 delta = headroom - skb_headroom(skb);
3645 if (delta || cloned)
3646 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3652 * skb_cow - copy header of skb when it is required
3653 * @skb: buffer to cow
3654 * @headroom: needed headroom
3656 * If the skb passed lacks sufficient headroom or its data part
3657 * is shared, data is reallocated. If reallocation fails, an error
3658 * is returned and original skb is not changed.
3660 * The result is skb with writable area skb->head...skb->tail
3661 * and at least @headroom of space at head.
3663 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3665 return __skb_cow(skb, headroom, skb_cloned(skb));
3669 * skb_cow_head - skb_cow but only making the head writable
3670 * @skb: buffer to cow
3671 * @headroom: needed headroom
3673 * This function is identical to skb_cow except that we replace the
3674 * skb_cloned check by skb_header_cloned. It should be used when
3675 * you only need to push on some header and do not need to modify
3678 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3680 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3684 * skb_padto - pad an skbuff up to a minimal size
3685 * @skb: buffer to pad
3686 * @len: minimal length
3688 * Pads up a buffer to ensure the trailing bytes exist and are
3689 * blanked. If the buffer already contains sufficient data it
3690 * is untouched. Otherwise it is extended. Returns zero on
3691 * success. The skb is freed on error.
3693 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3695 unsigned int size = skb->len;
3696 if (likely(size >= len))
3698 return skb_pad(skb, len - size);
3702 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3703 * @skb: buffer to pad
3704 * @len: minimal length
3705 * @free_on_error: free buffer on error
3707 * Pads up a buffer to ensure the trailing bytes exist and are
3708 * blanked. If the buffer already contains sufficient data it
3709 * is untouched. Otherwise it is extended. Returns zero on
3710 * success. The skb is freed on error if @free_on_error is true.
3712 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3716 unsigned int size = skb->len;
3718 if (unlikely(size < len)) {
3720 if (__skb_pad(skb, len, free_on_error))
3722 __skb_put(skb, len);
3728 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3729 * @skb: buffer to pad
3730 * @len: minimal length
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.
3737 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3739 return __skb_put_padto(skb, len, true);
3742 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i)
3745 static inline int skb_add_data(struct sk_buff *skb,
3746 struct iov_iter *from, int copy)
3748 const int off = skb->len;
3750 if (skb->ip_summed == CHECKSUM_NONE) {
3752 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3754 skb->csum = csum_block_add(skb->csum, csum, off);
3757 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3760 __skb_trim(skb, off);
3764 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3765 const struct page *page, int off)
3770 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3772 return page == skb_frag_page(frag) &&
3773 off == skb_frag_off(frag) + skb_frag_size(frag);
3778 static inline int __skb_linearize(struct sk_buff *skb)
3780 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3784 * skb_linearize - convert paged skb to linear one
3785 * @skb: buffer to linarize
3787 * If there is no free memory -ENOMEM is returned, otherwise zero
3788 * is returned and the old skb data released.
3790 static inline int skb_linearize(struct sk_buff *skb)
3792 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3796 * skb_has_shared_frag - can any frag be overwritten
3797 * @skb: buffer to test
3799 * Return true if the skb has at least one frag that might be modified
3800 * by an external entity (as in vmsplice()/sendfile())
3802 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3804 return skb_is_nonlinear(skb) &&
3805 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3809 * skb_linearize_cow - make sure skb is linear and writable
3810 * @skb: buffer to process
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_cow(struct sk_buff *skb)
3817 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3818 __skb_linearize(skb) : 0;
3821 static __always_inline void
3822 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3825 if (skb->ip_summed == CHECKSUM_COMPLETE)
3826 skb->csum = csum_block_sub(skb->csum,
3827 csum_partial(start, len, 0), off);
3828 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3829 skb_checksum_start_offset(skb) < 0)
3830 skb->ip_summed = CHECKSUM_NONE;
3834 * skb_postpull_rcsum - update checksum for received skb after pull
3835 * @skb: buffer to update
3836 * @start: start of data before pull
3837 * @len: length of data pulled
3839 * After doing a pull on a received packet, you need to call this to
3840 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3841 * CHECKSUM_NONE so that it can be recomputed from scratch.
3843 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3844 const void *start, unsigned int len)
3846 if (skb->ip_summed == CHECKSUM_COMPLETE)
3847 skb->csum = wsum_negate(csum_partial(start, len,
3848 wsum_negate(skb->csum)));
3849 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3850 skb_checksum_start_offset(skb) < 0)
3851 skb->ip_summed = CHECKSUM_NONE;
3854 static __always_inline void
3855 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3858 if (skb->ip_summed == CHECKSUM_COMPLETE)
3859 skb->csum = csum_block_add(skb->csum,
3860 csum_partial(start, len, 0), off);
3864 * skb_postpush_rcsum - update checksum for received skb after push
3865 * @skb: buffer to update
3866 * @start: start of data after push
3867 * @len: length of data pushed
3869 * After doing a push on a received packet, you need to call this to
3870 * update the CHECKSUM_COMPLETE checksum.
3872 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3873 const void *start, unsigned int len)
3875 __skb_postpush_rcsum(skb, start, len, 0);
3878 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3881 * skb_push_rcsum - push skb and update receive checksum
3882 * @skb: buffer to update
3883 * @len: length of data pulled
3885 * This function performs an skb_push on the packet and updates
3886 * the CHECKSUM_COMPLETE checksum. It should be used on
3887 * receive path processing instead of skb_push unless you know
3888 * that the checksum difference is zero (e.g., a valid IP header)
3889 * or you are setting ip_summed to CHECKSUM_NONE.
3891 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3894 skb_postpush_rcsum(skb, skb->data, len);
3898 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3900 * pskb_trim_rcsum - trim received skb and update checksum
3901 * @skb: buffer to trim
3904 * This is exactly the same as pskb_trim except that it ensures the
3905 * checksum of received packets are still valid after the operation.
3906 * It can change skb pointers.
3909 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3911 if (likely(len >= skb->len))
3913 return pskb_trim_rcsum_slow(skb, len);
3916 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3918 if (skb->ip_summed == CHECKSUM_COMPLETE)
3919 skb->ip_summed = CHECKSUM_NONE;
3920 __skb_trim(skb, len);
3924 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3926 if (skb->ip_summed == CHECKSUM_COMPLETE)
3927 skb->ip_summed = CHECKSUM_NONE;
3928 return __skb_grow(skb, len);
3931 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3932 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3933 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3934 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3935 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3937 #define skb_queue_walk(queue, skb) \
3938 for (skb = (queue)->next; \
3939 skb != (struct sk_buff *)(queue); \
3942 #define skb_queue_walk_safe(queue, skb, tmp) \
3943 for (skb = (queue)->next, tmp = skb->next; \
3944 skb != (struct sk_buff *)(queue); \
3945 skb = tmp, tmp = skb->next)
3947 #define skb_queue_walk_from(queue, skb) \
3948 for (; skb != (struct sk_buff *)(queue); \
3951 #define skb_rbtree_walk(skb, root) \
3952 for (skb = skb_rb_first(root); skb != NULL; \
3953 skb = skb_rb_next(skb))
3955 #define skb_rbtree_walk_from(skb) \
3956 for (; skb != NULL; \
3957 skb = skb_rb_next(skb))
3959 #define skb_rbtree_walk_from_safe(skb, tmp) \
3960 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3963 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3964 for (tmp = skb->next; \
3965 skb != (struct sk_buff *)(queue); \
3966 skb = tmp, tmp = skb->next)
3968 #define skb_queue_reverse_walk(queue, skb) \
3969 for (skb = (queue)->prev; \
3970 skb != (struct sk_buff *)(queue); \
3973 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3974 for (skb = (queue)->prev, tmp = skb->prev; \
3975 skb != (struct sk_buff *)(queue); \
3976 skb = tmp, tmp = skb->prev)
3978 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3979 for (tmp = skb->prev; \
3980 skb != (struct sk_buff *)(queue); \
3981 skb = tmp, tmp = skb->prev)
3983 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3985 return skb_shinfo(skb)->frag_list != NULL;
3988 static inline void skb_frag_list_init(struct sk_buff *skb)
3990 skb_shinfo(skb)->frag_list = NULL;
3993 #define skb_walk_frags(skb, iter) \
3994 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3997 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3998 int *err, long *timeo_p,
3999 const struct sk_buff *skb);
4000 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
4001 struct sk_buff_head *queue,
4004 struct sk_buff **last);
4005 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
4006 struct sk_buff_head *queue,
4007 unsigned int flags, int *off, int *err,
4008 struct sk_buff **last);
4009 struct sk_buff *__skb_recv_datagram(struct sock *sk,
4010 struct sk_buff_head *sk_queue,
4011 unsigned int flags, int *off, int *err);
4012 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
4013 __poll_t datagram_poll(struct file *file, struct socket *sock,
4014 struct poll_table_struct *wait);
4015 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
4016 struct iov_iter *to, int size);
4017 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
4018 struct msghdr *msg, int size)
4020 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
4022 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
4023 struct msghdr *msg);
4024 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
4025 struct iov_iter *to, int len,
4026 struct ahash_request *hash);
4027 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
4028 struct iov_iter *from, int len);
4029 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
4030 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
4031 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
4032 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
4033 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
4034 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
4036 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
4037 struct pipe_inode_info *pipe, unsigned int len,
4038 unsigned int flags);
4039 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
4041 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
4042 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
4043 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
4044 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
4046 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
4047 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
4048 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
4049 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
4050 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
4051 unsigned int offset);
4052 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
4053 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
4054 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev);
4055 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
4056 int skb_vlan_pop(struct sk_buff *skb);
4057 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
4058 int skb_eth_pop(struct sk_buff *skb);
4059 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
4060 const unsigned char *src);
4061 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
4062 int mac_len, bool ethernet);
4063 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
4065 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
4066 int skb_mpls_dec_ttl(struct sk_buff *skb);
4067 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
4070 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
4072 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
4075 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
4077 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
4080 struct skb_checksum_ops {
4081 __wsum (*update)(const void *mem, int len, __wsum wsum);
4082 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
4085 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
4087 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
4088 __wsum csum, const struct skb_checksum_ops *ops);
4089 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
4092 static inline void * __must_check
4093 __skb_header_pointer(const struct sk_buff *skb, int offset, int len,
4094 const void *data, int hlen, void *buffer)
4096 if (likely(hlen - offset >= len))
4097 return (void *)data + offset;
4099 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
4105 static inline void * __must_check
4106 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
4108 return __skb_header_pointer(skb, offset, len, skb->data,
4109 skb_headlen(skb), buffer);
4112 static inline void * __must_check
4113 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len)
4115 if (likely(skb_headlen(skb) - offset >= len))
4116 return skb->data + offset;
4121 * skb_needs_linearize - check if we need to linearize a given skb
4122 * depending on the given device features.
4123 * @skb: socket buffer to check
4124 * @features: net device features
4126 * Returns true if either:
4127 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
4128 * 2. skb is fragmented and the device does not support SG.
4130 static inline bool skb_needs_linearize(struct sk_buff *skb,
4131 netdev_features_t features)
4133 return skb_is_nonlinear(skb) &&
4134 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
4135 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
4138 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
4140 const unsigned int len)
4142 memcpy(to, skb->data, len);
4145 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4146 const int offset, void *to,
4147 const unsigned int len)
4149 memcpy(to, skb->data + offset, len);
4152 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4154 const unsigned int len)
4156 memcpy(skb->data, from, len);
4159 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4162 const unsigned int len)
4164 memcpy(skb->data + offset, from, len);
4167 void skb_init(void);
4169 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4175 * skb_get_timestamp - get timestamp from a skb
4176 * @skb: skb to get stamp from
4177 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4179 * Timestamps are stored in the skb as offsets to a base timestamp.
4180 * This function converts the offset back to a struct timeval and stores
4183 static inline void skb_get_timestamp(const struct sk_buff *skb,
4184 struct __kernel_old_timeval *stamp)
4186 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4189 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4190 struct __kernel_sock_timeval *stamp)
4192 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4194 stamp->tv_sec = ts.tv_sec;
4195 stamp->tv_usec = ts.tv_nsec / 1000;
4198 static inline void skb_get_timestampns(const struct sk_buff *skb,
4199 struct __kernel_old_timespec *stamp)
4201 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4203 stamp->tv_sec = ts.tv_sec;
4204 stamp->tv_nsec = ts.tv_nsec;
4207 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4208 struct __kernel_timespec *stamp)
4210 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4212 stamp->tv_sec = ts.tv_sec;
4213 stamp->tv_nsec = ts.tv_nsec;
4216 static inline void __net_timestamp(struct sk_buff *skb)
4218 skb->tstamp = ktime_get_real();
4219 skb->tstamp_type = SKB_CLOCK_REALTIME;
4222 static inline ktime_t net_timedelta(ktime_t t)
4224 return ktime_sub(ktime_get_real(), t);
4227 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4233 skb->tstamp_type = tstamp_type;
4235 skb->tstamp_type = SKB_CLOCK_REALTIME;
4238 static inline void skb_set_delivery_type_by_clockid(struct sk_buff *skb,
4239 ktime_t kt, clockid_t clockid)
4241 u8 tstamp_type = SKB_CLOCK_REALTIME;
4244 case CLOCK_REALTIME:
4246 case CLOCK_MONOTONIC:
4247 tstamp_type = SKB_CLOCK_MONOTONIC;
4250 tstamp_type = SKB_CLOCK_TAI;
4257 skb_set_delivery_time(skb, kt, tstamp_type);
4260 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4262 /* It is used in the ingress path to clear the delivery_time.
4263 * If needed, set the skb->tstamp to the (rcv) timestamp.
4265 static inline void skb_clear_delivery_time(struct sk_buff *skb)
4267 if (skb->tstamp_type) {
4268 skb->tstamp_type = SKB_CLOCK_REALTIME;
4269 if (static_branch_unlikely(&netstamp_needed_key))
4270 skb->tstamp = ktime_get_real();
4276 static inline void skb_clear_tstamp(struct sk_buff *skb)
4278 if (skb->tstamp_type)
4284 static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4286 if (skb->tstamp_type)
4292 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4294 if (skb->tstamp_type != SKB_CLOCK_MONOTONIC && skb->tstamp)
4297 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4298 return ktime_get_real();
4303 static inline u8 skb_metadata_len(const struct sk_buff *skb)
4305 return skb_shinfo(skb)->meta_len;
4308 static inline void *skb_metadata_end(const struct sk_buff *skb)
4310 return skb_mac_header(skb);
4313 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4314 const struct sk_buff *skb_b,
4317 const void *a = skb_metadata_end(skb_a);
4318 const void *b = skb_metadata_end(skb_b);
4321 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ||
4322 BITS_PER_LONG != 64)
4325 /* Using more efficient variant than plain call to memcmp(). */
4327 #define __it(x, op) (x -= sizeof(u##op))
4328 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4329 case 32: diffs |= __it_diff(a, b, 64);
4331 case 24: diffs |= __it_diff(a, b, 64);
4333 case 16: diffs |= __it_diff(a, b, 64);
4335 case 8: diffs |= __it_diff(a, b, 64);
4337 case 28: diffs |= __it_diff(a, b, 64);
4339 case 20: diffs |= __it_diff(a, b, 64);
4341 case 12: diffs |= __it_diff(a, b, 64);
4343 case 4: diffs |= __it_diff(a, b, 32);
4347 return memcmp(a - meta_len, b - meta_len, meta_len);
4352 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4353 const struct sk_buff *skb_b)
4355 u8 len_a = skb_metadata_len(skb_a);
4356 u8 len_b = skb_metadata_len(skb_b);
4358 if (!(len_a | len_b))
4361 return len_a != len_b ?
4362 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4365 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4367 skb_shinfo(skb)->meta_len = meta_len;
4370 static inline void skb_metadata_clear(struct sk_buff *skb)
4372 skb_metadata_set(skb, 0);
4375 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4377 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4379 void skb_clone_tx_timestamp(struct sk_buff *skb);
4380 bool skb_defer_rx_timestamp(struct sk_buff *skb);
4382 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4384 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4388 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4393 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4396 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4398 * PHY drivers may accept clones of transmitted packets for
4399 * timestamping via their phy_driver.txtstamp method. These drivers
4400 * must call this function to return the skb back to the stack with a
4403 * @skb: clone of the original outgoing packet
4404 * @hwtstamps: hardware time stamps
4407 void skb_complete_tx_timestamp(struct sk_buff *skb,
4408 struct skb_shared_hwtstamps *hwtstamps);
4410 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4411 struct skb_shared_hwtstamps *hwtstamps,
4412 struct sock *sk, int tstype);
4415 * skb_tstamp_tx - queue clone of skb with send time stamps
4416 * @orig_skb: the original outgoing packet
4417 * @hwtstamps: hardware time stamps, may be NULL if not available
4419 * If the skb has a socket associated, then this function clones the
4420 * skb (thus sharing the actual data and optional structures), stores
4421 * the optional hardware time stamping information (if non NULL) or
4422 * generates a software time stamp (otherwise), then queues the clone
4423 * to the error queue of the socket. Errors are silently ignored.
4425 void skb_tstamp_tx(struct sk_buff *orig_skb,
4426 struct skb_shared_hwtstamps *hwtstamps);
4429 * skb_tx_timestamp() - Driver hook for transmit timestamping
4431 * Ethernet MAC Drivers should call this function in their hard_xmit()
4432 * function immediately before giving the sk_buff to the MAC hardware.
4434 * Specifically, one should make absolutely sure that this function is
4435 * called before TX completion of this packet can trigger. Otherwise
4436 * the packet could potentially already be freed.
4438 * @skb: A socket buffer.
4440 static inline void skb_tx_timestamp(struct sk_buff *skb)
4442 skb_clone_tx_timestamp(skb);
4443 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4444 skb_tstamp_tx(skb, NULL);
4448 * skb_complete_wifi_ack - deliver skb with wifi status
4450 * @skb: the original outgoing packet
4451 * @acked: ack status
4454 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4456 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4457 __sum16 __skb_checksum_complete(struct sk_buff *skb);
4459 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4461 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4463 (skb->ip_summed == CHECKSUM_PARTIAL &&
4464 skb_checksum_start_offset(skb) >= 0));
4468 * skb_checksum_complete - Calculate checksum of an entire packet
4469 * @skb: packet to process
4471 * This function calculates the checksum over the entire packet plus
4472 * the value of skb->csum. The latter can be used to supply the
4473 * checksum of a pseudo header as used by TCP/UDP. It returns the
4476 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4477 * this function can be used to verify that checksum on received
4478 * packets. In that case the function should return zero if the
4479 * checksum is correct. In particular, this function will return zero
4480 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4481 * hardware has already verified the correctness of the checksum.
4483 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4485 return skb_csum_unnecessary(skb) ?
4486 0 : __skb_checksum_complete(skb);
4489 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4491 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4492 if (skb->csum_level == 0)
4493 skb->ip_summed = CHECKSUM_NONE;
4499 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4501 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4502 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4504 } else if (skb->ip_summed == CHECKSUM_NONE) {
4505 skb->ip_summed = CHECKSUM_UNNECESSARY;
4506 skb->csum_level = 0;
4510 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4512 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4513 skb->ip_summed = CHECKSUM_NONE;
4514 skb->csum_level = 0;
4518 /* Check if we need to perform checksum complete validation.
4520 * Returns true if checksum complete is needed, false otherwise
4521 * (either checksum is unnecessary or zero checksum is allowed).
4523 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4527 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4528 skb->csum_valid = 1;
4529 __skb_decr_checksum_unnecessary(skb);
4536 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4539 #define CHECKSUM_BREAK 76
4541 /* Unset checksum-complete
4543 * Unset checksum complete can be done when packet is being modified
4544 * (uncompressed for instance) and checksum-complete value is
4547 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4549 if (skb->ip_summed == CHECKSUM_COMPLETE)
4550 skb->ip_summed = CHECKSUM_NONE;
4553 /* Validate (init) checksum based on checksum complete.
4556 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4557 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4558 * checksum is stored in skb->csum for use in __skb_checksum_complete
4559 * non-zero: value of invalid checksum
4562 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4566 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4567 if (!csum_fold(csum_add(psum, skb->csum))) {
4568 skb->csum_valid = 1;
4575 if (complete || skb->len <= CHECKSUM_BREAK) {
4578 csum = __skb_checksum_complete(skb);
4579 skb->csum_valid = !csum;
4586 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4591 /* Perform checksum validate (init). Note that this is a macro since we only
4592 * want to calculate the pseudo header which is an input function if necessary.
4593 * First we try to validate without any computation (checksum unnecessary) and
4594 * then calculate based on checksum complete calling the function to compute
4598 * 0: checksum is validated or try to in skb_checksum_complete
4599 * non-zero: value of invalid checksum
4601 #define __skb_checksum_validate(skb, proto, complete, \
4602 zero_okay, check, compute_pseudo) \
4604 __sum16 __ret = 0; \
4605 skb->csum_valid = 0; \
4606 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4607 __ret = __skb_checksum_validate_complete(skb, \
4608 complete, compute_pseudo(skb, proto)); \
4612 #define skb_checksum_init(skb, proto, compute_pseudo) \
4613 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4615 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4616 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4618 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4619 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4621 #define skb_checksum_validate_zero_check(skb, proto, check, \
4623 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4625 #define skb_checksum_simple_validate(skb) \
4626 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4628 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4630 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4633 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4635 skb->csum = ~pseudo;
4636 skb->ip_summed = CHECKSUM_COMPLETE;
4639 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4641 if (__skb_checksum_convert_check(skb)) \
4642 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4645 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4646 u16 start, u16 offset)
4648 skb->ip_summed = CHECKSUM_PARTIAL;
4649 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4650 skb->csum_offset = offset - start;
4653 /* Update skbuf and packet to reflect the remote checksum offload operation.
4654 * When called, ptr indicates the starting point for skb->csum when
4655 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4656 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4658 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4659 int start, int offset, bool nopartial)
4664 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4668 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4669 __skb_checksum_complete(skb);
4670 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4673 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4675 /* Adjust skb->csum since we changed the packet */
4676 skb->csum = csum_add(skb->csum, delta);
4679 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4681 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4682 return (void *)(skb->_nfct & NFCT_PTRMASK);
4688 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4690 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4697 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4699 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4700 skb->slow_gro |= !!nfct;
4705 #ifdef CONFIG_SKB_EXTENSIONS
4707 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4713 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4716 #if IS_ENABLED(CONFIG_MPTCP)
4719 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4722 SKB_EXT_NUM, /* must be last */
4726 * struct skb_ext - sk_buff extensions
4727 * @refcnt: 1 on allocation, deallocated on 0
4728 * @offset: offset to add to @data to obtain extension address
4729 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4730 * @data: start of extension data, variable sized
4732 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4733 * to use 'u8' types while allowing up to 2kb worth of extension data.
4737 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4738 u8 chunks; /* same */
4739 char data[] __aligned(8);
4742 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4743 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4744 struct skb_ext *ext);
4745 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4746 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4747 void __skb_ext_put(struct skb_ext *ext);
4749 static inline void skb_ext_put(struct sk_buff *skb)
4751 if (skb->active_extensions)
4752 __skb_ext_put(skb->extensions);
4755 static inline void __skb_ext_copy(struct sk_buff *dst,
4756 const struct sk_buff *src)
4758 dst->active_extensions = src->active_extensions;
4760 if (src->active_extensions) {
4761 struct skb_ext *ext = src->extensions;
4763 refcount_inc(&ext->refcnt);
4764 dst->extensions = ext;
4768 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4771 __skb_ext_copy(dst, src);
4774 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4776 return !!ext->offset[i];
4779 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4781 return skb->active_extensions & (1 << id);
4784 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4786 if (skb_ext_exist(skb, id))
4787 __skb_ext_del(skb, id);
4790 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4792 if (skb_ext_exist(skb, id)) {
4793 struct skb_ext *ext = skb->extensions;
4795 return (void *)ext + (ext->offset[id] << 3);
4801 static inline void skb_ext_reset(struct sk_buff *skb)
4803 if (unlikely(skb->active_extensions)) {
4804 __skb_ext_put(skb->extensions);
4805 skb->active_extensions = 0;
4809 static inline bool skb_has_extensions(struct sk_buff *skb)
4811 return unlikely(skb->active_extensions);
4814 static inline void skb_ext_put(struct sk_buff *skb) {}
4815 static inline void skb_ext_reset(struct sk_buff *skb) {}
4816 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4817 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4818 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4819 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4820 #endif /* CONFIG_SKB_EXTENSIONS */
4822 static inline void nf_reset_ct(struct sk_buff *skb)
4824 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4825 nf_conntrack_put(skb_nfct(skb));
4830 static inline void nf_reset_trace(struct sk_buff *skb)
4832 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4837 static inline void ipvs_reset(struct sk_buff *skb)
4839 #if IS_ENABLED(CONFIG_IP_VS)
4840 skb->ipvs_property = 0;
4844 /* Note: This doesn't put any conntrack info in dst. */
4845 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4848 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4849 dst->_nfct = src->_nfct;
4850 nf_conntrack_get(skb_nfct(src));
4852 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4854 dst->nf_trace = src->nf_trace;
4858 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4860 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4861 nf_conntrack_put(skb_nfct(dst));
4863 dst->slow_gro = src->slow_gro;
4864 __nf_copy(dst, src, true);
4867 #ifdef CONFIG_NETWORK_SECMARK
4868 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4870 to->secmark = from->secmark;
4873 static inline void skb_init_secmark(struct sk_buff *skb)
4878 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4881 static inline void skb_init_secmark(struct sk_buff *skb)
4885 static inline int secpath_exists(const struct sk_buff *skb)
4888 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4894 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4896 return !skb->destructor &&
4897 !secpath_exists(skb) &&
4899 !skb->_skb_refdst &&
4900 !skb_has_frag_list(skb);
4903 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4905 skb->queue_mapping = queue_mapping;
4908 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4910 return skb->queue_mapping;
4913 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4915 to->queue_mapping = from->queue_mapping;
4918 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4920 skb->queue_mapping = rx_queue + 1;
4923 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4925 return skb->queue_mapping - 1;
4928 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4930 return skb->queue_mapping != 0;
4933 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4935 skb->dst_pending_confirm = val;
4938 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4940 return skb->dst_pending_confirm != 0;
4943 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4946 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4952 static inline bool skb_is_gso(const struct sk_buff *skb)
4954 return skb_shinfo(skb)->gso_size;
4957 /* Note: Should be called only if skb_is_gso(skb) is true */
4958 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4960 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4963 /* Note: Should be called only if skb_is_gso(skb) is true */
4964 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4966 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4969 /* Note: Should be called only if skb_is_gso(skb) is true */
4970 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4972 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4975 static inline void skb_gso_reset(struct sk_buff *skb)
4977 skb_shinfo(skb)->gso_size = 0;
4978 skb_shinfo(skb)->gso_segs = 0;
4979 skb_shinfo(skb)->gso_type = 0;
4982 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4985 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4987 shinfo->gso_size += increment;
4990 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4993 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4995 shinfo->gso_size -= decrement;
4998 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
5000 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
5002 /* LRO sets gso_size but not gso_type, whereas if GSO is really
5003 * wanted then gso_type will be set. */
5004 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5006 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
5007 unlikely(shinfo->gso_type == 0)) {
5008 __skb_warn_lro_forwarding(skb);
5014 static inline void skb_forward_csum(struct sk_buff *skb)
5016 /* Unfortunately we don't support this one. Any brave souls? */
5017 if (skb->ip_summed == CHECKSUM_COMPLETE)
5018 skb->ip_summed = CHECKSUM_NONE;
5022 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
5023 * @skb: skb to check
5025 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
5026 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
5027 * use this helper, to document places where we make this assertion.
5029 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
5031 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE);
5034 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
5036 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
5037 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5038 unsigned int transport_len,
5039 __sum16(*skb_chkf)(struct sk_buff *skb));
5042 * skb_head_is_locked - Determine if the skb->head is locked down
5043 * @skb: skb to check
5045 * The head on skbs build around a head frag can be removed if they are
5046 * not cloned. This function returns true if the skb head is locked down
5047 * due to either being allocated via kmalloc, or by being a clone with
5048 * multiple references to the head.
5050 static inline bool skb_head_is_locked(const struct sk_buff *skb)
5052 return !skb->head_frag || skb_cloned(skb);
5055 /* Local Checksum Offload.
5056 * Compute outer checksum based on the assumption that the
5057 * inner checksum will be offloaded later.
5058 * See Documentation/networking/checksum-offloads.rst for
5059 * explanation of how this works.
5060 * Fill in outer checksum adjustment (e.g. with sum of outer
5061 * pseudo-header) before calling.
5062 * Also ensure that inner checksum is in linear data area.
5064 static inline __wsum lco_csum(struct sk_buff *skb)
5066 unsigned char *csum_start = skb_checksum_start(skb);
5067 unsigned char *l4_hdr = skb_transport_header(skb);
5070 /* Start with complement of inner checksum adjustment */
5071 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
5074 /* Add in checksum of our headers (incl. outer checksum
5075 * adjustment filled in by caller) and return result.
5077 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
5080 static inline bool skb_is_redirected(const struct sk_buff *skb)
5082 return skb->redirected;
5085 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
5087 skb->redirected = 1;
5088 #ifdef CONFIG_NET_REDIRECT
5089 skb->from_ingress = from_ingress;
5090 if (skb->from_ingress)
5091 skb_clear_tstamp(skb);
5095 static inline void skb_reset_redirect(struct sk_buff *skb)
5097 skb->redirected = 0;
5100 static inline void skb_set_redirected_noclear(struct sk_buff *skb,
5103 skb->redirected = 1;
5104 #ifdef CONFIG_NET_REDIRECT
5105 skb->from_ingress = from_ingress;
5109 static inline bool skb_csum_is_sctp(struct sk_buff *skb)
5111 #if IS_ENABLED(CONFIG_IP_SCTP)
5112 return skb->csum_not_inet;
5118 static inline void skb_reset_csum_not_inet(struct sk_buff *skb)
5120 skb->ip_summed = CHECKSUM_NONE;
5121 #if IS_ENABLED(CONFIG_IP_SCTP)
5122 skb->csum_not_inet = 0;
5126 static inline void skb_set_kcov_handle(struct sk_buff *skb,
5127 const u64 kcov_handle)
5130 skb->kcov_handle = kcov_handle;
5134 static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
5137 return skb->kcov_handle;
5143 static inline void skb_mark_for_recycle(struct sk_buff *skb)
5145 #ifdef CONFIG_PAGE_POOL
5146 skb->pp_recycle = 1;
5150 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter,
5151 ssize_t maxsize, gfp_t gfp);
5153 #endif /* __KERNEL__ */
5154 #endif /* _LINUX_SKBUFF_H */