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 #include <net/page_pool.h>
36 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
37 #include <linux/netfilter/nf_conntrack_common.h>
39 #include <net/net_debug.h>
40 #include <net/dropreason.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,
298 struct net_device *physindev;
300 /* always valid & non-NULL from FORWARD on, for physdev match */
301 struct net_device *physoutdev;
303 /* prerouting: detect dnat in orig/reply direction */
305 struct in6_addr ipv6_daddr;
307 /* after prerouting + nat detected: store original source
308 * mac since neigh resolution overwrites it, only used while
309 * skb is out in neigh layer.
311 char neigh_header[8];
316 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
317 /* Chain in tc_skb_ext will be used to share the tc chain with
318 * ovs recirc_id. It will be set to the current chain by tc
319 * and read by ovs to recirc_id.
331 u8 act_miss:1; /* Set if act_miss_cookie is used */
335 struct sk_buff_head {
336 /* These two members must be first to match sk_buff. */
337 struct_group_tagged(sk_buff_list, list,
338 struct sk_buff *next;
339 struct sk_buff *prev;
348 /* To allow 64K frame to be packed as single skb without frag_list we
349 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
350 * buffers which do not start on a page boundary.
352 * Since GRO uses frags we allocate at least 16 regardless of page
355 #if (65536/PAGE_SIZE + 1) < 16
356 #define MAX_SKB_FRAGS 16UL
358 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
360 extern int sysctl_max_skb_frags;
362 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
363 * segment using its current segmentation instead.
365 #define GSO_BY_FRAGS 0xFFFF
367 typedef struct bio_vec skb_frag_t;
370 * skb_frag_size() - Returns the size of a skb fragment
371 * @frag: skb fragment
373 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
379 * skb_frag_size_set() - Sets the size of a skb fragment
380 * @frag: skb fragment
381 * @size: size of fragment
383 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
389 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
390 * @frag: skb fragment
391 * @delta: value to add
393 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
395 frag->bv_len += delta;
399 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
400 * @frag: skb fragment
401 * @delta: value to subtract
403 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
405 frag->bv_len -= delta;
409 * skb_frag_must_loop - Test if %p is a high memory page
410 * @p: fragment's page
412 static inline bool skb_frag_must_loop(struct page *p)
414 #if defined(CONFIG_HIGHMEM)
415 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
422 * skb_frag_foreach_page - loop over pages in a fragment
424 * @f: skb frag to operate on
425 * @f_off: offset from start of f->bv_page
426 * @f_len: length from f_off to loop over
427 * @p: (temp var) current page
428 * @p_off: (temp var) offset from start of current page,
429 * non-zero only on first page.
430 * @p_len: (temp var) length in current page,
431 * < PAGE_SIZE only on first and last page.
432 * @copied: (temp var) length so far, excluding current p_len.
434 * A fragment can hold a compound page, in which case per-page
435 * operations, notably kmap_atomic, must be called for each
438 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
439 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
440 p_off = (f_off) & (PAGE_SIZE - 1), \
441 p_len = skb_frag_must_loop(p) ? \
442 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
445 copied += p_len, p++, p_off = 0, \
446 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
448 #define HAVE_HW_TIME_STAMP
451 * struct skb_shared_hwtstamps - hardware time stamps
452 * @hwtstamp: hardware time stamp transformed into duration
453 * since arbitrary point in time
454 * @netdev_data: address/cookie of network device driver used as
455 * reference to actual hardware time stamp
457 * Software time stamps generated by ktime_get_real() are stored in
460 * hwtstamps can only be compared against other hwtstamps from
463 * This structure is attached to packets as part of the
464 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
466 struct skb_shared_hwtstamps {
473 /* Definitions for tx_flags in struct skb_shared_info */
475 /* generate hardware time stamp */
476 SKBTX_HW_TSTAMP = 1 << 0,
478 /* generate software time stamp when queueing packet to NIC */
479 SKBTX_SW_TSTAMP = 1 << 1,
481 /* device driver is going to provide hardware time stamp */
482 SKBTX_IN_PROGRESS = 1 << 2,
484 /* generate hardware time stamp based on cycles if supported */
485 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3,
487 /* generate wifi status information (where possible) */
488 SKBTX_WIFI_STATUS = 1 << 4,
490 /* determine hardware time stamp based on time or cycles */
491 SKBTX_HW_TSTAMP_NETDEV = 1 << 5,
493 /* generate software time stamp when entering packet scheduling */
494 SKBTX_SCHED_TSTAMP = 1 << 6,
497 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
499 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \
500 SKBTX_HW_TSTAMP_USE_CYCLES | \
503 /* Definitions for flags in struct skb_shared_info */
505 /* use zcopy routines */
506 SKBFL_ZEROCOPY_ENABLE = BIT(0),
508 /* This indicates at least one fragment might be overwritten
509 * (as in vmsplice(), sendfile() ...)
510 * If we need to compute a TX checksum, we'll need to copy
511 * all frags to avoid possible bad checksum
513 SKBFL_SHARED_FRAG = BIT(1),
515 /* segment contains only zerocopy data and should not be
516 * charged to the kernel memory.
518 SKBFL_PURE_ZEROCOPY = BIT(2),
520 SKBFL_DONT_ORPHAN = BIT(3),
522 /* page references are managed by the ubuf_info, so it's safe to
523 * use frags only up until ubuf_info is released
525 SKBFL_MANAGED_FRAG_REFS = BIT(4),
528 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
529 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
530 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
533 * The callback notifies userspace to release buffers when skb DMA is done in
534 * lower device, the skb last reference should be 0 when calling this.
535 * The zerocopy_success argument is true if zero copy transmit occurred,
536 * false on data copy or out of memory error caused by data copy attempt.
537 * The ctx field is used to track device context.
538 * The desc field is used to track userspace buffer index.
541 void (*callback)(struct sk_buff *, struct ubuf_info *,
542 bool zerocopy_success);
547 struct ubuf_info_msgzc {
548 struct ubuf_info ubuf;
564 struct user_struct *user;
569 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
570 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \
573 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
574 void mm_unaccount_pinned_pages(struct mmpin *mmp);
576 /* This data is invariant across clones and lives at
577 * the end of the header data, ie. at skb->end.
579 struct skb_shared_info {
584 unsigned short gso_size;
585 /* Warning: this field is not always filled in (UFO)! */
586 unsigned short gso_segs;
587 struct sk_buff *frag_list;
588 struct skb_shared_hwtstamps hwtstamps;
589 unsigned int gso_type;
593 * Warning : all fields before dataref are cleared in __alloc_skb()
596 unsigned int xdp_frags_size;
598 /* Intermediate layers must ensure that destructor_arg
599 * remains valid until skb destructor */
600 void * destructor_arg;
602 /* must be last field, see pskb_expand_head() */
603 skb_frag_t frags[MAX_SKB_FRAGS];
607 * DOC: dataref and headerless skbs
609 * Transport layers send out clones of payload skbs they hold for
610 * retransmissions. To allow lower layers of the stack to prepend their headers
611 * we split &skb_shared_info.dataref into two halves.
612 * The lower 16 bits count the overall number of references.
613 * The higher 16 bits indicate how many of the references are payload-only.
614 * skb_header_cloned() checks if skb is allowed to add / write the headers.
616 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
617 * (via __skb_header_release()). Any clone created from marked skb will get
618 * &sk_buff.hdr_len populated with the available headroom.
619 * If there's the only clone in existence it's able to modify the headroom
620 * at will. The sequence of calls inside the transport layer is::
624 * __skb_header_release()
626 * // send the clone down the stack
628 * This is not a very generic construct and it depends on the transport layers
629 * doing the right thing. In practice there's usually only one payload-only skb.
630 * Having multiple payload-only skbs with different lengths of hdr_len is not
631 * possible. The payload-only skbs should never leave their owner.
633 #define SKB_DATAREF_SHIFT 16
634 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
638 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
639 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
640 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
644 SKB_GSO_TCPV4 = 1 << 0,
646 /* This indicates the skb is from an untrusted source. */
647 SKB_GSO_DODGY = 1 << 1,
649 /* This indicates the tcp segment has CWR set. */
650 SKB_GSO_TCP_ECN = 1 << 2,
652 SKB_GSO_TCP_FIXEDID = 1 << 3,
654 SKB_GSO_TCPV6 = 1 << 4,
656 SKB_GSO_FCOE = 1 << 5,
658 SKB_GSO_GRE = 1 << 6,
660 SKB_GSO_GRE_CSUM = 1 << 7,
662 SKB_GSO_IPXIP4 = 1 << 8,
664 SKB_GSO_IPXIP6 = 1 << 9,
666 SKB_GSO_UDP_TUNNEL = 1 << 10,
668 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
670 SKB_GSO_PARTIAL = 1 << 12,
672 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
674 SKB_GSO_SCTP = 1 << 14,
676 SKB_GSO_ESP = 1 << 15,
678 SKB_GSO_UDP = 1 << 16,
680 SKB_GSO_UDP_L4 = 1 << 17,
682 SKB_GSO_FRAGLIST = 1 << 18,
685 #if BITS_PER_LONG > 32
686 #define NET_SKBUFF_DATA_USES_OFFSET 1
689 #ifdef NET_SKBUFF_DATA_USES_OFFSET
690 typedef unsigned int sk_buff_data_t;
692 typedef unsigned char *sk_buff_data_t;
696 * DOC: Basic sk_buff geometry
698 * struct sk_buff itself is a metadata structure and does not hold any packet
699 * data. All the data is held in associated buffers.
701 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
704 * - data buffer, containing headers and sometimes payload;
705 * this is the part of the skb operated on by the common helpers
706 * such as skb_put() or skb_pull();
707 * - shared info (struct skb_shared_info) which holds an array of pointers
708 * to read-only data in the (page, offset, length) format.
710 * Optionally &skb_shared_info.frag_list may point to another skb.
712 * Basic diagram may look like this::
717 * ,--------------------------- + head
718 * / ,----------------- + data
719 * / / ,----------- + tail
723 * -----------------------------------------------
724 * | headroom | data | tailroom | skb_shared_info |
725 * -----------------------------------------------
729 * + [page frag] ---------
730 * + frag_list --> | sk_buff |
736 * struct sk_buff - socket buffer
737 * @next: Next buffer in list
738 * @prev: Previous buffer in list
739 * @tstamp: Time we arrived/left
740 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
741 * for retransmit timer
742 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
744 * @ll_node: anchor in an llist (eg socket defer_list)
745 * @sk: Socket we are owned by
746 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
747 * fragmentation management
748 * @dev: Device we arrived on/are leaving by
749 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
750 * @cb: Control buffer. Free for use by every layer. Put private vars here
751 * @_skb_refdst: destination entry (with norefcount bit)
752 * @sp: the security path, used for xfrm
753 * @len: Length of actual data
754 * @data_len: Data length
755 * @mac_len: Length of link layer header
756 * @hdr_len: writable header length of cloned skb
757 * @csum: Checksum (must include start/offset pair)
758 * @csum_start: Offset from skb->head where checksumming should start
759 * @csum_offset: Offset from csum_start where checksum should be stored
760 * @priority: Packet queueing priority
761 * @ignore_df: allow local fragmentation
762 * @cloned: Head may be cloned (check refcnt to be sure)
763 * @ip_summed: Driver fed us an IP checksum
764 * @nohdr: Payload reference only, must not modify header
765 * @pkt_type: Packet class
766 * @fclone: skbuff clone status
767 * @ipvs_property: skbuff is owned by ipvs
768 * @inner_protocol_type: whether the inner protocol is
769 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
770 * @remcsum_offload: remote checksum offload is enabled
771 * @offload_fwd_mark: Packet was L2-forwarded in hardware
772 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
773 * @tc_skip_classify: do not classify packet. set by IFB device
774 * @tc_at_ingress: used within tc_classify to distinguish in/egress
775 * @redirected: packet was redirected by packet classifier
776 * @from_ingress: packet was redirected from the ingress path
777 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
778 * @peeked: this packet has been seen already, so stats have been
779 * done for it, don't do them again
780 * @nf_trace: netfilter packet trace flag
781 * @protocol: Packet protocol from driver
782 * @destructor: Destruct function
783 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
784 * @_sk_redir: socket redirection information for skmsg
785 * @_nfct: Associated connection, if any (with nfctinfo bits)
786 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
787 * @skb_iif: ifindex of device we arrived on
788 * @tc_index: Traffic control index
789 * @hash: the packet hash
790 * @queue_mapping: Queue mapping for multiqueue devices
791 * @head_frag: skb was allocated from page fragments,
792 * not allocated by kmalloc() or vmalloc().
793 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
794 * @pp_recycle: mark the packet for recycling instead of freeing (implies
795 * page_pool support on driver)
796 * @active_extensions: active extensions (skb_ext_id types)
797 * @ndisc_nodetype: router type (from link layer)
798 * @ooo_okay: allow the mapping of a socket to a queue to be changed
799 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
801 * @sw_hash: indicates hash was computed in software stack
802 * @wifi_acked_valid: wifi_acked was set
803 * @wifi_acked: whether frame was acked on wifi or not
804 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
805 * @encapsulation: indicates the inner headers in the skbuff are valid
806 * @encap_hdr_csum: software checksum is needed
807 * @csum_valid: checksum is already valid
808 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
809 * @csum_complete_sw: checksum was completed by software
810 * @csum_level: indicates the number of consecutive checksums found in
811 * the packet minus one that have been verified as
812 * CHECKSUM_UNNECESSARY (max 3)
813 * @dst_pending_confirm: need to confirm neighbour
814 * @decrypted: Decrypted SKB
815 * @slow_gro: state present at GRO time, slower prepare step required
816 * @mono_delivery_time: When set, skb->tstamp has the
817 * delivery_time in mono clock base (i.e. EDT). Otherwise, the
818 * skb->tstamp has the (rcv) timestamp at ingress and
819 * delivery_time at egress.
820 * @napi_id: id of the NAPI struct this skb came from
821 * @sender_cpu: (aka @napi_id) source CPU in XPS
822 * @alloc_cpu: CPU which did the skb allocation.
823 * @secmark: security marking
824 * @mark: Generic packet mark
825 * @reserved_tailroom: (aka @mark) number of bytes of free space available
826 * at the tail of an sk_buff
827 * @vlan_all: vlan fields (proto & tci)
828 * @vlan_proto: vlan encapsulation protocol
829 * @vlan_tci: vlan tag control information
830 * @inner_protocol: Protocol (encapsulation)
831 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
832 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
833 * @inner_transport_header: Inner transport layer header (encapsulation)
834 * @inner_network_header: Network layer header (encapsulation)
835 * @inner_mac_header: Link layer header (encapsulation)
836 * @transport_header: Transport layer header
837 * @network_header: Network layer header
838 * @mac_header: Link layer header
839 * @kcov_handle: KCOV remote handle for remote coverage collection
840 * @tail: Tail pointer
842 * @head: Head of buffer
843 * @data: Data head pointer
844 * @truesize: Buffer size
845 * @users: User count - see {datagram,tcp}.c
846 * @extensions: allocated extensions, valid if active_extensions is nonzero
852 /* These two members must be first to match sk_buff_head. */
853 struct sk_buff *next;
854 struct sk_buff *prev;
857 struct net_device *dev;
858 /* Some protocols might use this space to store information,
859 * while device pointer would be NULL.
860 * UDP receive path is one user.
862 unsigned long dev_scratch;
865 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
866 struct list_head list;
867 struct llist_node ll_node;
872 int ip_defrag_offset;
877 u64 skb_mstamp_ns; /* earliest departure time */
880 * This is the control buffer. It is free to use for every
881 * layer. Please put your private variables there. If you
882 * want to keep them across layers you have to do a skb_clone()
883 * first. This is owned by whoever has the skb queued ATM.
885 char cb[48] __aligned(8);
889 unsigned long _skb_refdst;
890 void (*destructor)(struct sk_buff *skb);
892 struct list_head tcp_tsorted_anchor;
893 #ifdef CONFIG_NET_SOCK_MSG
894 unsigned long _sk_redir;
898 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
906 /* Following fields are _not_ copied in __copy_skb_header()
907 * Note that queue_mapping is here mostly to fill a hole.
911 /* if you move cloned around you also must adapt those constants */
912 #ifdef __BIG_ENDIAN_BITFIELD
913 #define CLONED_MASK (1 << 7)
915 #define CLONED_MASK 1
917 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
920 __u8 __cloned_offset[0];
928 pp_recycle:1; /* page_pool recycle indicator */
929 #ifdef CONFIG_SKB_EXTENSIONS
930 __u8 active_extensions;
933 /* Fields enclosed in headers group are copied
934 * using a single memcpy() in __copy_skb_header()
936 struct_group(headers,
939 __u8 __pkt_type_offset[0];
941 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
949 __u8 wifi_acked_valid:1;
952 /* Indicates the inner headers are valid in the skbuff. */
953 __u8 encapsulation:1;
954 __u8 encap_hdr_csum:1;
958 __u8 __pkt_vlan_present_offset[0];
960 __u8 remcsum_offload:1;
961 __u8 csum_complete_sw:1;
963 __u8 dst_pending_confirm:1;
964 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */
965 #ifdef CONFIG_NET_CLS_ACT
966 __u8 tc_skip_classify:1;
967 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
969 #ifdef CONFIG_IPV6_NDISC_NODETYPE
970 __u8 ndisc_nodetype:2;
973 __u8 ipvs_property:1;
974 __u8 inner_protocol_type:1;
975 #ifdef CONFIG_NET_SWITCHDEV
976 __u8 offload_fwd_mark:1;
977 __u8 offload_l3_fwd_mark:1;
980 #ifdef CONFIG_NET_REDIRECT
983 #ifdef CONFIG_NETFILTER_SKIP_EGRESS
984 __u8 nf_skip_egress:1;
986 #ifdef CONFIG_TLS_DEVICE
990 __u8 csum_not_inet:1;
992 #ifdef CONFIG_NET_SCHED
993 __u16 tc_index; /* traffic control index */
1013 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
1015 unsigned int napi_id;
1016 unsigned int sender_cpu;
1020 #ifdef CONFIG_NETWORK_SECMARK
1026 __u32 reserved_tailroom;
1030 __be16 inner_protocol;
1034 __u16 inner_transport_header;
1035 __u16 inner_network_header;
1036 __u16 inner_mac_header;
1039 __u16 transport_header;
1040 __u16 network_header;
1047 ); /* end headers group */
1049 /* These elements must be at the end, see alloc_skb() for details. */
1050 sk_buff_data_t tail;
1052 unsigned char *head,
1054 unsigned int truesize;
1057 #ifdef CONFIG_SKB_EXTENSIONS
1058 /* only useable after checking ->active_extensions != 0 */
1059 struct skb_ext *extensions;
1063 /* if you move pkt_type around you also must adapt those constants */
1064 #ifdef __BIG_ENDIAN_BITFIELD
1065 #define PKT_TYPE_MAX (7 << 5)
1067 #define PKT_TYPE_MAX 7
1069 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1071 /* if you move tc_at_ingress or mono_delivery_time
1072 * around, you also must adapt these constants.
1074 #ifdef __BIG_ENDIAN_BITFIELD
1075 #define TC_AT_INGRESS_MASK (1 << 0)
1076 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 2)
1078 #define TC_AT_INGRESS_MASK (1 << 7)
1079 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 5)
1081 #define PKT_VLAN_PRESENT_OFFSET offsetof(struct sk_buff, __pkt_vlan_present_offset)
1085 * Handling routines are only of interest to the kernel
1088 #define SKB_ALLOC_FCLONE 0x01
1089 #define SKB_ALLOC_RX 0x02
1090 #define SKB_ALLOC_NAPI 0x04
1093 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1096 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1098 return unlikely(skb->pfmemalloc);
1102 * skb might have a dst pointer attached, refcounted or not.
1103 * _skb_refdst low order bit is set if refcount was _not_ taken
1105 #define SKB_DST_NOREF 1UL
1106 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1109 * skb_dst - returns skb dst_entry
1112 * Returns skb dst_entry, regardless of reference taken or not.
1114 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1116 /* If refdst was not refcounted, check we still are in a
1117 * rcu_read_lock section
1119 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1120 !rcu_read_lock_held() &&
1121 !rcu_read_lock_bh_held());
1122 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1126 * skb_dst_set - sets skb dst
1130 * Sets skb dst, assuming a reference was taken on dst and should
1131 * be released by skb_dst_drop()
1133 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1135 skb->slow_gro |= !!dst;
1136 skb->_skb_refdst = (unsigned long)dst;
1140 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1144 * Sets skb dst, assuming a reference was not taken on dst.
1145 * If dst entry is cached, we do not take reference and dst_release
1146 * will be avoided by refdst_drop. If dst entry is not cached, we take
1147 * reference, so that last dst_release can destroy the dst immediately.
1149 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1151 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1152 skb->slow_gro |= !!dst;
1153 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1157 * skb_dst_is_noref - Test if skb dst isn't refcounted
1160 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1162 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1166 * skb_rtable - Returns the skb &rtable
1169 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1171 return (struct rtable *)skb_dst(skb);
1174 /* For mangling skb->pkt_type from user space side from applications
1175 * such as nft, tc, etc, we only allow a conservative subset of
1176 * possible pkt_types to be set.
1178 static inline bool skb_pkt_type_ok(u32 ptype)
1180 return ptype <= PACKET_OTHERHOST;
1184 * skb_napi_id - Returns the skb's NAPI id
1187 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1189 #ifdef CONFIG_NET_RX_BUSY_POLL
1190 return skb->napi_id;
1197 * skb_unref - decrement the skb's reference count
1200 * Returns true if we can free the skb.
1202 static inline bool skb_unref(struct sk_buff *skb)
1206 if (likely(refcount_read(&skb->users) == 1))
1208 else if (likely(!refcount_dec_and_test(&skb->users)))
1215 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason);
1218 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1219 * @skb: buffer to free
1221 static inline void kfree_skb(struct sk_buff *skb)
1223 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1226 void skb_release_head_state(struct sk_buff *skb);
1227 void kfree_skb_list_reason(struct sk_buff *segs,
1228 enum skb_drop_reason reason);
1229 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1230 void skb_tx_error(struct sk_buff *skb);
1232 static inline void kfree_skb_list(struct sk_buff *segs)
1234 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1237 #ifdef CONFIG_TRACEPOINTS
1238 void consume_skb(struct sk_buff *skb);
1240 static inline void consume_skb(struct sk_buff *skb)
1242 return kfree_skb(skb);
1246 void __consume_stateless_skb(struct sk_buff *skb);
1247 void __kfree_skb(struct sk_buff *skb);
1248 extern struct kmem_cache *skbuff_cache;
1250 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1251 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1252 bool *fragstolen, int *delta_truesize);
1254 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1256 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1257 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1258 struct sk_buff *build_skb_around(struct sk_buff *skb,
1259 void *data, unsigned int frag_size);
1260 void skb_attempt_defer_free(struct sk_buff *skb);
1262 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1263 struct sk_buff *slab_build_skb(void *data);
1266 * alloc_skb - allocate a network buffer
1267 * @size: size to allocate
1268 * @priority: allocation mask
1270 * This function is a convenient wrapper around __alloc_skb().
1272 static inline struct sk_buff *alloc_skb(unsigned int size,
1275 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1278 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1279 unsigned long data_len,
1283 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1285 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1286 struct sk_buff_fclones {
1287 struct sk_buff skb1;
1289 struct sk_buff skb2;
1291 refcount_t fclone_ref;
1295 * skb_fclone_busy - check if fclone is busy
1299 * Returns true if skb is a fast clone, and its clone is not freed.
1300 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1301 * so we also check that this didnt happen.
1303 static inline bool skb_fclone_busy(const struct sock *sk,
1304 const struct sk_buff *skb)
1306 const struct sk_buff_fclones *fclones;
1308 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1310 return skb->fclone == SKB_FCLONE_ORIG &&
1311 refcount_read(&fclones->fclone_ref) > 1 &&
1312 READ_ONCE(fclones->skb2.sk) == sk;
1316 * alloc_skb_fclone - allocate a network buffer from fclone cache
1317 * @size: size to allocate
1318 * @priority: allocation mask
1320 * This function is a convenient wrapper around __alloc_skb().
1322 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1325 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1328 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1329 void skb_headers_offset_update(struct sk_buff *skb, int off);
1330 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1331 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1332 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1333 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1334 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1335 gfp_t gfp_mask, bool fclone);
1336 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1339 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1342 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1343 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1344 unsigned int headroom);
1345 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1346 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1347 int newtailroom, gfp_t priority);
1348 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1349 int offset, int len);
1350 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1351 int offset, int len);
1352 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1353 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1356 * skb_pad - zero pad the tail of an skb
1357 * @skb: buffer to pad
1358 * @pad: space to pad
1360 * Ensure that a buffer is followed by a padding area that is zero
1361 * filled. Used by network drivers which may DMA or transfer data
1362 * beyond the buffer end onto the wire.
1364 * May return error in out of memory cases. The skb is freed on error.
1366 static inline int skb_pad(struct sk_buff *skb, int pad)
1368 return __skb_pad(skb, pad, true);
1370 #define dev_kfree_skb(a) consume_skb(a)
1372 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1373 int offset, size_t size);
1375 struct skb_seq_state {
1379 __u32 stepped_offset;
1380 struct sk_buff *root_skb;
1381 struct sk_buff *cur_skb;
1386 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1387 unsigned int to, struct skb_seq_state *st);
1388 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1389 struct skb_seq_state *st);
1390 void skb_abort_seq_read(struct skb_seq_state *st);
1392 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1393 unsigned int to, struct ts_config *config);
1396 * Packet hash types specify the type of hash in skb_set_hash.
1398 * Hash types refer to the protocol layer addresses which are used to
1399 * construct a packet's hash. The hashes are used to differentiate or identify
1400 * flows of the protocol layer for the hash type. Hash types are either
1401 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1403 * Properties of hashes:
1405 * 1) Two packets in different flows have different hash values
1406 * 2) Two packets in the same flow should have the same hash value
1408 * A hash at a higher layer is considered to be more specific. A driver should
1409 * set the most specific hash possible.
1411 * A driver cannot indicate a more specific hash than the layer at which a hash
1412 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1414 * A driver may indicate a hash level which is less specific than the
1415 * actual layer the hash was computed on. For instance, a hash computed
1416 * at L4 may be considered an L3 hash. This should only be done if the
1417 * driver can't unambiguously determine that the HW computed the hash at
1418 * the higher layer. Note that the "should" in the second property above
1421 enum pkt_hash_types {
1422 PKT_HASH_TYPE_NONE, /* Undefined type */
1423 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1424 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1425 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1428 static inline void skb_clear_hash(struct sk_buff *skb)
1435 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1438 skb_clear_hash(skb);
1442 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1444 skb->l4_hash = is_l4;
1445 skb->sw_hash = is_sw;
1450 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1452 /* Used by drivers to set hash from HW */
1453 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1457 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1459 __skb_set_hash(skb, hash, true, is_l4);
1462 void __skb_get_hash(struct sk_buff *skb);
1463 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1464 u32 skb_get_poff(const struct sk_buff *skb);
1465 u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1466 const struct flow_keys_basic *keys, int hlen);
1467 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1468 const void *data, int hlen_proto);
1470 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1471 int thoff, u8 ip_proto)
1473 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1476 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1477 const struct flow_dissector_key *key,
1478 unsigned int key_count);
1480 struct bpf_flow_dissector;
1481 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1482 __be16 proto, int nhoff, int hlen, unsigned int flags);
1484 bool __skb_flow_dissect(const struct net *net,
1485 const struct sk_buff *skb,
1486 struct flow_dissector *flow_dissector,
1487 void *target_container, const void *data,
1488 __be16 proto, int nhoff, int hlen, unsigned int flags);
1490 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1491 struct flow_dissector *flow_dissector,
1492 void *target_container, unsigned int flags)
1494 return __skb_flow_dissect(NULL, skb, flow_dissector,
1495 target_container, NULL, 0, 0, 0, flags);
1498 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1499 struct flow_keys *flow,
1502 memset(flow, 0, sizeof(*flow));
1503 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1504 flow, NULL, 0, 0, 0, flags);
1508 skb_flow_dissect_flow_keys_basic(const struct net *net,
1509 const struct sk_buff *skb,
1510 struct flow_keys_basic *flow,
1511 const void *data, __be16 proto,
1512 int nhoff, int hlen, unsigned int flags)
1514 memset(flow, 0, sizeof(*flow));
1515 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1516 data, proto, nhoff, hlen, flags);
1519 void skb_flow_dissect_meta(const struct sk_buff *skb,
1520 struct flow_dissector *flow_dissector,
1521 void *target_container);
1523 /* Gets a skb connection tracking info, ctinfo map should be a
1524 * map of mapsize to translate enum ip_conntrack_info states
1528 skb_flow_dissect_ct(const struct sk_buff *skb,
1529 struct flow_dissector *flow_dissector,
1530 void *target_container,
1531 u16 *ctinfo_map, size_t mapsize,
1532 bool post_ct, u16 zone);
1534 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1535 struct flow_dissector *flow_dissector,
1536 void *target_container);
1538 void skb_flow_dissect_hash(const struct sk_buff *skb,
1539 struct flow_dissector *flow_dissector,
1540 void *target_container);
1542 static inline __u32 skb_get_hash(struct sk_buff *skb)
1544 if (!skb->l4_hash && !skb->sw_hash)
1545 __skb_get_hash(skb);
1550 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1552 if (!skb->l4_hash && !skb->sw_hash) {
1553 struct flow_keys keys;
1554 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1556 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1562 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1563 const siphash_key_t *perturb);
1565 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1570 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1572 to->hash = from->hash;
1573 to->sw_hash = from->sw_hash;
1574 to->l4_hash = from->l4_hash;
1577 static inline void skb_copy_decrypted(struct sk_buff *to,
1578 const struct sk_buff *from)
1580 #ifdef CONFIG_TLS_DEVICE
1581 to->decrypted = from->decrypted;
1585 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1586 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1588 return skb->head + skb->end;
1591 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1596 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1601 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1606 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1608 return skb->end - skb->head;
1611 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1613 skb->end = skb->head + offset;
1617 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1618 struct ubuf_info *uarg);
1620 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1622 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1625 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
1626 struct sk_buff *skb, struct iov_iter *from,
1629 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1630 struct msghdr *msg, int len)
1632 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len);
1635 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1636 struct msghdr *msg, int len,
1637 struct ubuf_info *uarg);
1640 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1642 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1644 return &skb_shinfo(skb)->hwtstamps;
1647 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1649 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1651 return is_zcopy ? skb_uarg(skb) : NULL;
1654 static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1656 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1659 static inline bool skb_zcopy_managed(const struct sk_buff *skb)
1661 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS;
1664 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1665 const struct sk_buff *skb2)
1667 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1670 static inline void net_zcopy_get(struct ubuf_info *uarg)
1672 refcount_inc(&uarg->refcnt);
1675 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1677 skb_shinfo(skb)->destructor_arg = uarg;
1678 skb_shinfo(skb)->flags |= uarg->flags;
1681 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1684 if (skb && uarg && !skb_zcopy(skb)) {
1685 if (unlikely(have_ref && *have_ref))
1688 net_zcopy_get(uarg);
1689 skb_zcopy_init(skb, uarg);
1693 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1695 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1696 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1699 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1701 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1704 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1706 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1709 static inline void net_zcopy_put(struct ubuf_info *uarg)
1712 uarg->callback(NULL, uarg, true);
1715 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1718 if (uarg->callback == msg_zerocopy_callback)
1719 msg_zerocopy_put_abort(uarg, have_uref);
1721 net_zcopy_put(uarg);
1725 /* Release a reference on a zerocopy structure */
1726 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1728 struct ubuf_info *uarg = skb_zcopy(skb);
1731 if (!skb_zcopy_is_nouarg(skb))
1732 uarg->callback(skb, uarg, zerocopy_success);
1734 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1738 void __skb_zcopy_downgrade_managed(struct sk_buff *skb);
1740 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb)
1742 if (unlikely(skb_zcopy_managed(skb)))
1743 __skb_zcopy_downgrade_managed(skb);
1746 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1751 static inline void skb_poison_list(struct sk_buff *skb)
1753 #ifdef CONFIG_DEBUG_NET
1754 skb->next = SKB_LIST_POISON_NEXT;
1758 /* Iterate through singly-linked GSO fragments of an skb. */
1759 #define skb_list_walk_safe(first, skb, next_skb) \
1760 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1761 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1763 static inline void skb_list_del_init(struct sk_buff *skb)
1765 __list_del_entry(&skb->list);
1766 skb_mark_not_on_list(skb);
1770 * skb_queue_empty - check if a queue is empty
1773 * Returns true if the queue is empty, false otherwise.
1775 static inline int skb_queue_empty(const struct sk_buff_head *list)
1777 return list->next == (const struct sk_buff *) list;
1781 * skb_queue_empty_lockless - check if a queue is empty
1784 * Returns true if the queue is empty, false otherwise.
1785 * This variant can be used in lockless contexts.
1787 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1789 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1794 * skb_queue_is_last - check if skb is the last entry in the queue
1798 * Returns true if @skb is the last buffer on the list.
1800 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1801 const struct sk_buff *skb)
1803 return skb->next == (const struct sk_buff *) list;
1807 * skb_queue_is_first - check if skb is the first entry in the queue
1811 * Returns true if @skb is the first buffer on the list.
1813 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1814 const struct sk_buff *skb)
1816 return skb->prev == (const struct sk_buff *) list;
1820 * skb_queue_next - return the next packet in the queue
1822 * @skb: current buffer
1824 * Return the next packet in @list after @skb. It is only valid to
1825 * call this if skb_queue_is_last() evaluates to false.
1827 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1828 const struct sk_buff *skb)
1830 /* This BUG_ON may seem severe, but if we just return then we
1831 * are going to dereference garbage.
1833 BUG_ON(skb_queue_is_last(list, skb));
1838 * skb_queue_prev - return the prev packet in the queue
1840 * @skb: current buffer
1842 * Return the prev packet in @list before @skb. It is only valid to
1843 * call this if skb_queue_is_first() evaluates to false.
1845 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1846 const struct sk_buff *skb)
1848 /* This BUG_ON may seem severe, but if we just return then we
1849 * are going to dereference garbage.
1851 BUG_ON(skb_queue_is_first(list, skb));
1856 * skb_get - reference buffer
1857 * @skb: buffer to reference
1859 * Makes another reference to a socket buffer and returns a pointer
1862 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1864 refcount_inc(&skb->users);
1869 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1873 * skb_cloned - is the buffer a clone
1874 * @skb: buffer to check
1876 * Returns true if the buffer was generated with skb_clone() and is
1877 * one of multiple shared copies of the buffer. Cloned buffers are
1878 * shared data so must not be written to under normal circumstances.
1880 static inline int skb_cloned(const struct sk_buff *skb)
1882 return skb->cloned &&
1883 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1886 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1888 might_sleep_if(gfpflags_allow_blocking(pri));
1890 if (skb_cloned(skb))
1891 return pskb_expand_head(skb, 0, 0, pri);
1896 /* This variant of skb_unclone() makes sure skb->truesize
1897 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
1899 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
1900 * when various debugging features are in place.
1902 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
1903 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1905 might_sleep_if(gfpflags_allow_blocking(pri));
1907 if (skb_cloned(skb))
1908 return __skb_unclone_keeptruesize(skb, pri);
1913 * skb_header_cloned - is the header a clone
1914 * @skb: buffer to check
1916 * Returns true if modifying the header part of the buffer requires
1917 * the data to be copied.
1919 static inline int skb_header_cloned(const struct sk_buff *skb)
1926 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1927 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1928 return dataref != 1;
1931 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1933 might_sleep_if(gfpflags_allow_blocking(pri));
1935 if (skb_header_cloned(skb))
1936 return pskb_expand_head(skb, 0, 0, pri);
1942 * __skb_header_release() - allow clones to use the headroom
1943 * @skb: buffer to operate on
1945 * See "DOC: dataref and headerless skbs".
1947 static inline void __skb_header_release(struct sk_buff *skb)
1950 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1955 * skb_shared - is the buffer shared
1956 * @skb: buffer to check
1958 * Returns true if more than one person has a reference to this
1961 static inline int skb_shared(const struct sk_buff *skb)
1963 return refcount_read(&skb->users) != 1;
1967 * skb_share_check - check if buffer is shared and if so clone it
1968 * @skb: buffer to check
1969 * @pri: priority for memory allocation
1971 * If the buffer is shared the buffer is cloned and the old copy
1972 * drops a reference. A new clone with a single reference is returned.
1973 * If the buffer is not shared the original buffer is returned. When
1974 * being called from interrupt status or with spinlocks held pri must
1977 * NULL is returned on a memory allocation failure.
1979 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1981 might_sleep_if(gfpflags_allow_blocking(pri));
1982 if (skb_shared(skb)) {
1983 struct sk_buff *nskb = skb_clone(skb, pri);
1995 * Copy shared buffers into a new sk_buff. We effectively do COW on
1996 * packets to handle cases where we have a local reader and forward
1997 * and a couple of other messy ones. The normal one is tcpdumping
1998 * a packet thats being forwarded.
2002 * skb_unshare - make a copy of a shared buffer
2003 * @skb: buffer to check
2004 * @pri: priority for memory allocation
2006 * If the socket buffer is a clone then this function creates a new
2007 * copy of the data, drops a reference count on the old copy and returns
2008 * the new copy with the reference count at 1. If the buffer is not a clone
2009 * the original buffer is returned. When called with a spinlock held or
2010 * from interrupt state @pri must be %GFP_ATOMIC
2012 * %NULL is returned on a memory allocation failure.
2014 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
2017 might_sleep_if(gfpflags_allow_blocking(pri));
2018 if (skb_cloned(skb)) {
2019 struct sk_buff *nskb = skb_copy(skb, pri);
2021 /* Free our shared copy */
2032 * skb_peek - peek at the head of an &sk_buff_head
2033 * @list_: list to peek at
2035 * Peek an &sk_buff. Unlike most other operations you _MUST_
2036 * be careful with this one. A peek leaves the buffer on the
2037 * list and someone else may run off with it. You must hold
2038 * the appropriate locks or have a private queue to do this.
2040 * Returns %NULL for an empty list or a pointer to the head element.
2041 * The reference count is not incremented and the reference is therefore
2042 * volatile. Use with caution.
2044 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2046 struct sk_buff *skb = list_->next;
2048 if (skb == (struct sk_buff *)list_)
2054 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2055 * @list_: list to peek at
2057 * Like skb_peek(), but the caller knows that the list is not empty.
2059 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2065 * skb_peek_next - peek skb following the given one from a queue
2066 * @skb: skb to start from
2067 * @list_: list to peek at
2069 * Returns %NULL when the end of the list is met or a pointer to the
2070 * next element. The reference count is not incremented and the
2071 * reference is therefore volatile. Use with caution.
2073 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2074 const struct sk_buff_head *list_)
2076 struct sk_buff *next = skb->next;
2078 if (next == (struct sk_buff *)list_)
2084 * skb_peek_tail - peek at the tail of an &sk_buff_head
2085 * @list_: list to peek at
2087 * Peek an &sk_buff. Unlike most other operations you _MUST_
2088 * be careful with this one. A peek leaves the buffer on the
2089 * list and someone else may run off with it. You must hold
2090 * the appropriate locks or have a private queue to do this.
2092 * Returns %NULL for an empty list or a pointer to the tail element.
2093 * The reference count is not incremented and the reference is therefore
2094 * volatile. Use with caution.
2096 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2098 struct sk_buff *skb = READ_ONCE(list_->prev);
2100 if (skb == (struct sk_buff *)list_)
2107 * skb_queue_len - get queue length
2108 * @list_: list to measure
2110 * Return the length of an &sk_buff queue.
2112 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2118 * skb_queue_len_lockless - get queue length
2119 * @list_: list to measure
2121 * Return the length of an &sk_buff queue.
2122 * This variant can be used in lockless contexts.
2124 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2126 return READ_ONCE(list_->qlen);
2130 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2131 * @list: queue to initialize
2133 * This initializes only the list and queue length aspects of
2134 * an sk_buff_head object. This allows to initialize the list
2135 * aspects of an sk_buff_head without reinitializing things like
2136 * the spinlock. It can also be used for on-stack sk_buff_head
2137 * objects where the spinlock is known to not be used.
2139 static inline void __skb_queue_head_init(struct sk_buff_head *list)
2141 list->prev = list->next = (struct sk_buff *)list;
2146 * This function creates a split out lock class for each invocation;
2147 * this is needed for now since a whole lot of users of the skb-queue
2148 * infrastructure in drivers have different locking usage (in hardirq)
2149 * than the networking core (in softirq only). In the long run either the
2150 * network layer or drivers should need annotation to consolidate the
2151 * main types of usage into 3 classes.
2153 static inline void skb_queue_head_init(struct sk_buff_head *list)
2155 spin_lock_init(&list->lock);
2156 __skb_queue_head_init(list);
2159 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2160 struct lock_class_key *class)
2162 skb_queue_head_init(list);
2163 lockdep_set_class(&list->lock, class);
2167 * Insert an sk_buff on a list.
2169 * The "__skb_xxxx()" functions are the non-atomic ones that
2170 * can only be called with interrupts disabled.
2172 static inline void __skb_insert(struct sk_buff *newsk,
2173 struct sk_buff *prev, struct sk_buff *next,
2174 struct sk_buff_head *list)
2176 /* See skb_queue_empty_lockless() and skb_peek_tail()
2177 * for the opposite READ_ONCE()
2179 WRITE_ONCE(newsk->next, next);
2180 WRITE_ONCE(newsk->prev, prev);
2181 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2182 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2183 WRITE_ONCE(list->qlen, list->qlen + 1);
2186 static inline void __skb_queue_splice(const struct sk_buff_head *list,
2187 struct sk_buff *prev,
2188 struct sk_buff *next)
2190 struct sk_buff *first = list->next;
2191 struct sk_buff *last = list->prev;
2193 WRITE_ONCE(first->prev, prev);
2194 WRITE_ONCE(prev->next, first);
2196 WRITE_ONCE(last->next, next);
2197 WRITE_ONCE(next->prev, last);
2201 * skb_queue_splice - join two skb lists, this is designed for stacks
2202 * @list: the new list to add
2203 * @head: the place to add it in the first list
2205 static inline void skb_queue_splice(const struct sk_buff_head *list,
2206 struct sk_buff_head *head)
2208 if (!skb_queue_empty(list)) {
2209 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2210 head->qlen += list->qlen;
2215 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2216 * @list: the new list to add
2217 * @head: the place to add it in the first list
2219 * The list at @list is reinitialised
2221 static inline void skb_queue_splice_init(struct sk_buff_head *list,
2222 struct sk_buff_head *head)
2224 if (!skb_queue_empty(list)) {
2225 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2226 head->qlen += list->qlen;
2227 __skb_queue_head_init(list);
2232 * skb_queue_splice_tail - join two skb lists, each list being a queue
2233 * @list: the new list to add
2234 * @head: the place to add it in the first list
2236 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2237 struct sk_buff_head *head)
2239 if (!skb_queue_empty(list)) {
2240 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2241 head->qlen += list->qlen;
2246 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2247 * @list: the new list to add
2248 * @head: the place to add it in the first list
2250 * Each of the lists is a queue.
2251 * The list at @list is reinitialised
2253 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2254 struct sk_buff_head *head)
2256 if (!skb_queue_empty(list)) {
2257 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2258 head->qlen += list->qlen;
2259 __skb_queue_head_init(list);
2264 * __skb_queue_after - queue a buffer at the list head
2265 * @list: list to use
2266 * @prev: place after this buffer
2267 * @newsk: buffer to queue
2269 * Queue a buffer int the middle of a list. This function takes no locks
2270 * and you must therefore hold required locks before calling it.
2272 * A buffer cannot be placed on two lists at the same time.
2274 static inline void __skb_queue_after(struct sk_buff_head *list,
2275 struct sk_buff *prev,
2276 struct sk_buff *newsk)
2278 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2281 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2282 struct sk_buff_head *list);
2284 static inline void __skb_queue_before(struct sk_buff_head *list,
2285 struct sk_buff *next,
2286 struct sk_buff *newsk)
2288 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2292 * __skb_queue_head - queue a buffer at the list head
2293 * @list: list to use
2294 * @newsk: buffer to queue
2296 * Queue a buffer at the start of a list. This function takes no locks
2297 * and you must therefore hold required locks before calling it.
2299 * A buffer cannot be placed on two lists at the same time.
2301 static inline void __skb_queue_head(struct sk_buff_head *list,
2302 struct sk_buff *newsk)
2304 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2306 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2309 * __skb_queue_tail - queue a buffer at the list tail
2310 * @list: list to use
2311 * @newsk: buffer to queue
2313 * Queue a buffer at the end of a list. This function takes no locks
2314 * and you must therefore hold required locks before calling it.
2316 * A buffer cannot be placed on two lists at the same time.
2318 static inline void __skb_queue_tail(struct sk_buff_head *list,
2319 struct sk_buff *newsk)
2321 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2323 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2326 * remove sk_buff from list. _Must_ be called atomically, and with
2329 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2330 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2332 struct sk_buff *next, *prev;
2334 WRITE_ONCE(list->qlen, list->qlen - 1);
2337 skb->next = skb->prev = NULL;
2338 WRITE_ONCE(next->prev, prev);
2339 WRITE_ONCE(prev->next, next);
2343 * __skb_dequeue - remove from the head of the queue
2344 * @list: list to dequeue from
2346 * Remove the head of the list. This function does not take any locks
2347 * so must be used with appropriate locks held only. The head item is
2348 * returned or %NULL if the list is empty.
2350 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2352 struct sk_buff *skb = skb_peek(list);
2354 __skb_unlink(skb, list);
2357 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2360 * __skb_dequeue_tail - remove from the tail of the queue
2361 * @list: list to dequeue from
2363 * Remove the tail of the list. This function does not take any locks
2364 * so must be used with appropriate locks held only. The tail item is
2365 * returned or %NULL if the list is empty.
2367 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2369 struct sk_buff *skb = skb_peek_tail(list);
2371 __skb_unlink(skb, list);
2374 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2377 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2379 return skb->data_len;
2382 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2384 return skb->len - skb->data_len;
2387 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2389 unsigned int i, len = 0;
2391 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2392 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2396 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2398 return skb_headlen(skb) + __skb_pagelen(skb);
2401 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo,
2402 int i, struct page *page,
2405 skb_frag_t *frag = &shinfo->frags[i];
2408 * Propagate page pfmemalloc to the skb if we can. The problem is
2409 * that not all callers have unique ownership of the page but rely
2410 * on page_is_pfmemalloc doing the right thing(tm).
2412 frag->bv_page = page;
2413 frag->bv_offset = off;
2414 skb_frag_size_set(frag, size);
2418 * skb_len_add - adds a number to len fields of skb
2419 * @skb: buffer to add len to
2420 * @delta: number of bytes to add
2422 static inline void skb_len_add(struct sk_buff *skb, int delta)
2425 skb->data_len += delta;
2426 skb->truesize += delta;
2430 * __skb_fill_page_desc - initialise a paged fragment in an skb
2431 * @skb: buffer containing fragment to be initialised
2432 * @i: paged fragment index to initialise
2433 * @page: the page to use for this fragment
2434 * @off: the offset to the data with @page
2435 * @size: the length of the data
2437 * Initialises the @i'th fragment of @skb to point to &size bytes at
2438 * offset @off within @page.
2440 * Does not take any additional reference on the fragment.
2442 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2443 struct page *page, int off, int size)
2445 __skb_fill_page_desc_noacc(skb_shinfo(skb), i, page, off, size);
2446 page = compound_head(page);
2447 if (page_is_pfmemalloc(page))
2448 skb->pfmemalloc = true;
2452 * skb_fill_page_desc - initialise a paged fragment in an skb
2453 * @skb: buffer containing fragment to be initialised
2454 * @i: paged fragment index to initialise
2455 * @page: the page to use for this fragment
2456 * @off: the offset to the data with @page
2457 * @size: the length of the data
2459 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2460 * @skb to point to @size bytes at offset @off within @page. In
2461 * addition updates @skb such that @i is the last fragment.
2463 * Does not take any additional reference on the fragment.
2465 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2466 struct page *page, int off, int size)
2468 __skb_fill_page_desc(skb, i, page, off, size);
2469 skb_shinfo(skb)->nr_frags = i + 1;
2473 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb
2474 * @skb: buffer containing fragment to be initialised
2475 * @i: paged fragment index to initialise
2476 * @page: the page to use for this fragment
2477 * @off: the offset to the data with @page
2478 * @size: the length of the data
2480 * Variant of skb_fill_page_desc() which does not deal with
2481 * pfmemalloc, if page is not owned by us.
2483 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i,
2484 struct page *page, int off,
2487 struct skb_shared_info *shinfo = skb_shinfo(skb);
2489 __skb_fill_page_desc_noacc(shinfo, i, page, off, size);
2490 shinfo->nr_frags = i + 1;
2493 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2494 int size, unsigned int truesize);
2496 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2497 unsigned int truesize);
2499 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2501 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2502 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2504 return skb->head + skb->tail;
2507 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2509 skb->tail = skb->data - skb->head;
2512 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2514 skb_reset_tail_pointer(skb);
2515 skb->tail += offset;
2518 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2519 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2524 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2526 skb->tail = skb->data;
2529 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2531 skb->tail = skb->data + offset;
2534 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2536 static inline void skb_assert_len(struct sk_buff *skb)
2538 #ifdef CONFIG_DEBUG_NET
2539 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2540 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2541 #endif /* CONFIG_DEBUG_NET */
2545 * Add data to an sk_buff
2547 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2548 void *skb_put(struct sk_buff *skb, unsigned int len);
2549 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2551 void *tmp = skb_tail_pointer(skb);
2552 SKB_LINEAR_ASSERT(skb);
2558 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2560 void *tmp = __skb_put(skb, len);
2562 memset(tmp, 0, len);
2566 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2569 void *tmp = __skb_put(skb, len);
2571 memcpy(tmp, data, len);
2575 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2577 *(u8 *)__skb_put(skb, 1) = val;
2580 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2582 void *tmp = skb_put(skb, len);
2584 memset(tmp, 0, len);
2589 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2592 void *tmp = skb_put(skb, len);
2594 memcpy(tmp, data, len);
2599 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2601 *(u8 *)skb_put(skb, 1) = val;
2604 void *skb_push(struct sk_buff *skb, unsigned int len);
2605 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2612 void *skb_pull(struct sk_buff *skb, unsigned int len);
2613 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2616 if (unlikely(skb->len < skb->data_len)) {
2617 #if defined(CONFIG_DEBUG_NET)
2619 pr_err("__skb_pull(len=%u)\n", len);
2620 skb_dump(KERN_ERR, skb, false);
2624 return skb->data += len;
2627 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2629 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2632 void *skb_pull_data(struct sk_buff *skb, size_t len);
2634 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2636 static inline enum skb_drop_reason
2637 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len)
2639 if (likely(len <= skb_headlen(skb)))
2640 return SKB_NOT_DROPPED_YET;
2642 if (unlikely(len > skb->len))
2643 return SKB_DROP_REASON_PKT_TOO_SMALL;
2645 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb))))
2646 return SKB_DROP_REASON_NOMEM;
2648 return SKB_NOT_DROPPED_YET;
2651 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2653 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET;
2656 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2658 if (!pskb_may_pull(skb, len))
2662 return skb->data += len;
2665 void skb_condense(struct sk_buff *skb);
2668 * skb_headroom - bytes at buffer head
2669 * @skb: buffer to check
2671 * Return the number of bytes of free space at the head of an &sk_buff.
2673 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2675 return skb->data - skb->head;
2679 * skb_tailroom - bytes at buffer end
2680 * @skb: buffer to check
2682 * Return the number of bytes of free space at the tail of an sk_buff
2684 static inline int skb_tailroom(const struct sk_buff *skb)
2686 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2690 * skb_availroom - bytes at buffer end
2691 * @skb: buffer to check
2693 * Return the number of bytes of free space at the tail of an sk_buff
2694 * allocated by sk_stream_alloc()
2696 static inline int skb_availroom(const struct sk_buff *skb)
2698 if (skb_is_nonlinear(skb))
2701 return skb->end - skb->tail - skb->reserved_tailroom;
2705 * skb_reserve - adjust headroom
2706 * @skb: buffer to alter
2707 * @len: bytes to move
2709 * Increase the headroom of an empty &sk_buff by reducing the tail
2710 * room. This is only allowed for an empty buffer.
2712 static inline void skb_reserve(struct sk_buff *skb, int len)
2719 * skb_tailroom_reserve - adjust reserved_tailroom
2720 * @skb: buffer to alter
2721 * @mtu: maximum amount of headlen permitted
2722 * @needed_tailroom: minimum amount of reserved_tailroom
2724 * Set reserved_tailroom so that headlen can be as large as possible but
2725 * not larger than mtu and tailroom cannot be smaller than
2727 * The required headroom should already have been reserved before using
2730 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2731 unsigned int needed_tailroom)
2733 SKB_LINEAR_ASSERT(skb);
2734 if (mtu < skb_tailroom(skb) - needed_tailroom)
2735 /* use at most mtu */
2736 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2738 /* use up to all available space */
2739 skb->reserved_tailroom = needed_tailroom;
2742 #define ENCAP_TYPE_ETHER 0
2743 #define ENCAP_TYPE_IPPROTO 1
2745 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2748 skb->inner_protocol = protocol;
2749 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2752 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2755 skb->inner_ipproto = ipproto;
2756 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2759 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2761 skb->inner_mac_header = skb->mac_header;
2762 skb->inner_network_header = skb->network_header;
2763 skb->inner_transport_header = skb->transport_header;
2766 static inline void skb_reset_mac_len(struct sk_buff *skb)
2768 skb->mac_len = skb->network_header - skb->mac_header;
2771 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2774 return skb->head + skb->inner_transport_header;
2777 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2779 return skb_inner_transport_header(skb) - skb->data;
2782 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2784 skb->inner_transport_header = skb->data - skb->head;
2787 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2790 skb_reset_inner_transport_header(skb);
2791 skb->inner_transport_header += offset;
2794 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2796 return skb->head + skb->inner_network_header;
2799 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2801 skb->inner_network_header = skb->data - skb->head;
2804 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2807 skb_reset_inner_network_header(skb);
2808 skb->inner_network_header += offset;
2811 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2813 return skb->head + skb->inner_mac_header;
2816 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2818 skb->inner_mac_header = skb->data - skb->head;
2821 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2824 skb_reset_inner_mac_header(skb);
2825 skb->inner_mac_header += offset;
2827 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2829 return skb->transport_header != (typeof(skb->transport_header))~0U;
2832 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2834 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
2835 return skb->head + skb->transport_header;
2838 static inline void skb_reset_transport_header(struct sk_buff *skb)
2840 skb->transport_header = skb->data - skb->head;
2843 static inline void skb_set_transport_header(struct sk_buff *skb,
2846 skb_reset_transport_header(skb);
2847 skb->transport_header += offset;
2850 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2852 return skb->head + skb->network_header;
2855 static inline void skb_reset_network_header(struct sk_buff *skb)
2857 skb->network_header = skb->data - skb->head;
2860 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2862 skb_reset_network_header(skb);
2863 skb->network_header += offset;
2866 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2868 return skb->mac_header != (typeof(skb->mac_header))~0U;
2871 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2873 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2874 return skb->head + skb->mac_header;
2877 static inline int skb_mac_offset(const struct sk_buff *skb)
2879 return skb_mac_header(skb) - skb->data;
2882 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2884 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2885 return skb->network_header - skb->mac_header;
2888 static inline void skb_unset_mac_header(struct sk_buff *skb)
2890 skb->mac_header = (typeof(skb->mac_header))~0U;
2893 static inline void skb_reset_mac_header(struct sk_buff *skb)
2895 skb->mac_header = skb->data - skb->head;
2898 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2900 skb_reset_mac_header(skb);
2901 skb->mac_header += offset;
2904 static inline void skb_pop_mac_header(struct sk_buff *skb)
2906 skb->mac_header = skb->network_header;
2909 static inline void skb_probe_transport_header(struct sk_buff *skb)
2911 struct flow_keys_basic keys;
2913 if (skb_transport_header_was_set(skb))
2916 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2918 skb_set_transport_header(skb, keys.control.thoff);
2921 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2923 if (skb_mac_header_was_set(skb)) {
2924 const unsigned char *old_mac = skb_mac_header(skb);
2926 skb_set_mac_header(skb, -skb->mac_len);
2927 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2931 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2933 return skb->csum_start - skb_headroom(skb);
2936 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2938 return skb->head + skb->csum_start;
2941 static inline int skb_transport_offset(const struct sk_buff *skb)
2943 return skb_transport_header(skb) - skb->data;
2946 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2948 return skb->transport_header - skb->network_header;
2951 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2953 return skb->inner_transport_header - skb->inner_network_header;
2956 static inline int skb_network_offset(const struct sk_buff *skb)
2958 return skb_network_header(skb) - skb->data;
2961 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2963 return skb_inner_network_header(skb) - skb->data;
2966 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2968 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2972 * CPUs often take a performance hit when accessing unaligned memory
2973 * locations. The actual performance hit varies, it can be small if the
2974 * hardware handles it or large if we have to take an exception and fix it
2977 * Since an ethernet header is 14 bytes network drivers often end up with
2978 * the IP header at an unaligned offset. The IP header can be aligned by
2979 * shifting the start of the packet by 2 bytes. Drivers should do this
2982 * skb_reserve(skb, NET_IP_ALIGN);
2984 * The downside to this alignment of the IP header is that the DMA is now
2985 * unaligned. On some architectures the cost of an unaligned DMA is high
2986 * and this cost outweighs the gains made by aligning the IP header.
2988 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2991 #ifndef NET_IP_ALIGN
2992 #define NET_IP_ALIGN 2
2996 * The networking layer reserves some headroom in skb data (via
2997 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2998 * the header has to grow. In the default case, if the header has to grow
2999 * 32 bytes or less we avoid the reallocation.
3001 * Unfortunately this headroom changes the DMA alignment of the resulting
3002 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
3003 * on some architectures. An architecture can override this value,
3004 * perhaps setting it to a cacheline in size (since that will maintain
3005 * cacheline alignment of the DMA). It must be a power of 2.
3007 * Various parts of the networking layer expect at least 32 bytes of
3008 * headroom, you should not reduce this.
3010 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
3011 * to reduce average number of cache lines per packet.
3012 * get_rps_cpu() for example only access one 64 bytes aligned block :
3013 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
3016 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
3019 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
3021 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
3023 if (WARN_ON(skb_is_nonlinear(skb)))
3026 skb_set_tail_pointer(skb, len);
3029 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
3031 __skb_set_length(skb, len);
3034 void skb_trim(struct sk_buff *skb, unsigned int len);
3036 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
3039 return ___pskb_trim(skb, len);
3040 __skb_trim(skb, len);
3044 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
3046 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
3050 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
3051 * @skb: buffer to alter
3054 * This is identical to pskb_trim except that the caller knows that
3055 * the skb is not cloned so we should never get an error due to out-
3058 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3060 int err = pskb_trim(skb, len);
3064 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3066 unsigned int diff = len - skb->len;
3068 if (skb_tailroom(skb) < diff) {
3069 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3074 __skb_set_length(skb, len);
3079 * skb_orphan - orphan a buffer
3080 * @skb: buffer to orphan
3082 * If a buffer currently has an owner then we call the owner's
3083 * destructor function and make the @skb unowned. The buffer continues
3084 * to exist but is no longer charged to its former owner.
3086 static inline void skb_orphan(struct sk_buff *skb)
3088 if (skb->destructor) {
3089 skb->destructor(skb);
3090 skb->destructor = NULL;
3098 * skb_orphan_frags - orphan the frags contained in a buffer
3099 * @skb: buffer to orphan frags from
3100 * @gfp_mask: allocation mask for replacement pages
3102 * For each frag in the SKB which needs a destructor (i.e. has an
3103 * owner) create a copy of that frag and release the original
3104 * page by calling the destructor.
3106 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3108 if (likely(!skb_zcopy(skb)))
3110 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN)
3112 return skb_copy_ubufs(skb, gfp_mask);
3115 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3116 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3118 if (likely(!skb_zcopy(skb)))
3120 return skb_copy_ubufs(skb, gfp_mask);
3124 * __skb_queue_purge - empty a list
3125 * @list: list to empty
3127 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3128 * the list and one reference dropped. This function does not take the
3129 * list lock and the caller must hold the relevant locks to use it.
3131 static inline void __skb_queue_purge(struct sk_buff_head *list)
3133 struct sk_buff *skb;
3134 while ((skb = __skb_dequeue(list)) != NULL)
3137 void skb_queue_purge(struct sk_buff_head *list);
3139 unsigned int skb_rbtree_purge(struct rb_root *root);
3141 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3144 * netdev_alloc_frag - allocate a page fragment
3145 * @fragsz: fragment size
3147 * Allocates a frag from a page for receive buffer.
3148 * Uses GFP_ATOMIC allocations.
3150 static inline void *netdev_alloc_frag(unsigned int fragsz)
3152 return __netdev_alloc_frag_align(fragsz, ~0u);
3155 static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3158 WARN_ON_ONCE(!is_power_of_2(align));
3159 return __netdev_alloc_frag_align(fragsz, -align);
3162 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3166 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3167 * @dev: network device to receive on
3168 * @length: length to allocate
3170 * Allocate a new &sk_buff and assign it a usage count of one. The
3171 * buffer has unspecified headroom built in. Users should allocate
3172 * the headroom they think they need without accounting for the
3173 * built in space. The built in space is used for optimisations.
3175 * %NULL is returned if there is no free memory. Although this function
3176 * allocates memory it can be called from an interrupt.
3178 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3179 unsigned int length)
3181 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3184 /* legacy helper around __netdev_alloc_skb() */
3185 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3188 return __netdev_alloc_skb(NULL, length, gfp_mask);
3191 /* legacy helper around netdev_alloc_skb() */
3192 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3194 return netdev_alloc_skb(NULL, length);
3198 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3199 unsigned int length, gfp_t gfp)
3201 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3203 if (NET_IP_ALIGN && skb)
3204 skb_reserve(skb, NET_IP_ALIGN);
3208 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3209 unsigned int length)
3211 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3214 static inline void skb_free_frag(void *addr)
3216 page_frag_free(addr);
3219 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3221 static inline void *napi_alloc_frag(unsigned int fragsz)
3223 return __napi_alloc_frag_align(fragsz, ~0u);
3226 static inline void *napi_alloc_frag_align(unsigned int fragsz,
3229 WARN_ON_ONCE(!is_power_of_2(align));
3230 return __napi_alloc_frag_align(fragsz, -align);
3233 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
3234 unsigned int length, gfp_t gfp_mask);
3235 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
3236 unsigned int length)
3238 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
3240 void napi_consume_skb(struct sk_buff *skb, int budget);
3242 void napi_skb_free_stolen_head(struct sk_buff *skb);
3243 void __kfree_skb_defer(struct sk_buff *skb);
3246 * __dev_alloc_pages - allocate page for network Rx
3247 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3248 * @order: size of the allocation
3250 * Allocate a new page.
3252 * %NULL is returned if there is no free memory.
3254 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
3257 /* This piece of code contains several assumptions.
3258 * 1. This is for device Rx, therefor a cold page is preferred.
3259 * 2. The expectation is the user wants a compound page.
3260 * 3. If requesting a order 0 page it will not be compound
3261 * due to the check to see if order has a value in prep_new_page
3262 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3263 * code in gfp_to_alloc_flags that should be enforcing this.
3265 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3267 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
3270 static inline struct page *dev_alloc_pages(unsigned int order)
3272 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
3276 * __dev_alloc_page - allocate a page for network Rx
3277 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3279 * Allocate a new page.
3281 * %NULL is returned if there is no free memory.
3283 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
3285 return __dev_alloc_pages(gfp_mask, 0);
3288 static inline struct page *dev_alloc_page(void)
3290 return dev_alloc_pages(0);
3294 * dev_page_is_reusable - check whether a page can be reused for network Rx
3295 * @page: the page to test
3297 * A page shouldn't be considered for reusing/recycling if it was allocated
3298 * under memory pressure or at a distant memory node.
3300 * Returns false if this page should be returned to page allocator, true
3303 static inline bool dev_page_is_reusable(const struct page *page)
3305 return likely(page_to_nid(page) == numa_mem_id() &&
3306 !page_is_pfmemalloc(page));
3310 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3311 * @page: The page that was allocated from skb_alloc_page
3312 * @skb: The skb that may need pfmemalloc set
3314 static inline void skb_propagate_pfmemalloc(const struct page *page,
3315 struct sk_buff *skb)
3317 if (page_is_pfmemalloc(page))
3318 skb->pfmemalloc = true;
3322 * skb_frag_off() - Returns the offset of a skb fragment
3323 * @frag: the paged fragment
3325 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3327 return frag->bv_offset;
3331 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3332 * @frag: skb fragment
3333 * @delta: value to add
3335 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3337 frag->bv_offset += delta;
3341 * skb_frag_off_set() - Sets the offset of a skb fragment
3342 * @frag: skb fragment
3343 * @offset: offset of fragment
3345 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3347 frag->bv_offset = offset;
3351 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3352 * @fragto: skb fragment where offset is set
3353 * @fragfrom: skb fragment offset is copied from
3355 static inline void skb_frag_off_copy(skb_frag_t *fragto,
3356 const skb_frag_t *fragfrom)
3358 fragto->bv_offset = fragfrom->bv_offset;
3362 * skb_frag_page - retrieve the page referred to by a paged fragment
3363 * @frag: the paged fragment
3365 * Returns the &struct page associated with @frag.
3367 static inline struct page *skb_frag_page(const skb_frag_t *frag)
3369 return frag->bv_page;
3373 * __skb_frag_ref - take an addition reference on a paged fragment.
3374 * @frag: the paged fragment
3376 * Takes an additional reference on the paged fragment @frag.
3378 static inline void __skb_frag_ref(skb_frag_t *frag)
3380 get_page(skb_frag_page(frag));
3384 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
3386 * @f: the fragment offset.
3388 * Takes an additional reference on the @f'th paged fragment of @skb.
3390 static inline void skb_frag_ref(struct sk_buff *skb, int f)
3392 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
3396 * __skb_frag_unref - release a reference on a paged fragment.
3397 * @frag: the paged fragment
3398 * @recycle: recycle the page if allocated via page_pool
3400 * Releases a reference on the paged fragment @frag
3401 * or recycles the page via the page_pool API.
3403 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle)
3405 struct page *page = skb_frag_page(frag);
3407 #ifdef CONFIG_PAGE_POOL
3408 if (recycle && page_pool_return_skb_page(page))
3415 * skb_frag_unref - release a reference on a paged fragment of an skb.
3417 * @f: the fragment offset
3419 * Releases a reference on the @f'th paged fragment of @skb.
3421 static inline void skb_frag_unref(struct sk_buff *skb, int f)
3423 struct skb_shared_info *shinfo = skb_shinfo(skb);
3425 if (!skb_zcopy_managed(skb))
3426 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle);
3430 * skb_frag_address - gets the address of the data contained in a paged fragment
3431 * @frag: the paged fragment buffer
3433 * Returns the address of the data within @frag. The page must already
3436 static inline void *skb_frag_address(const skb_frag_t *frag)
3438 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3442 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3443 * @frag: the paged fragment buffer
3445 * Returns the address of the data within @frag. Checks that the page
3446 * is mapped and returns %NULL otherwise.
3448 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3450 void *ptr = page_address(skb_frag_page(frag));
3454 return ptr + skb_frag_off(frag);
3458 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3459 * @fragto: skb fragment where page is set
3460 * @fragfrom: skb fragment page is copied from
3462 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3463 const skb_frag_t *fragfrom)
3465 fragto->bv_page = fragfrom->bv_page;
3469 * __skb_frag_set_page - sets the page contained in a paged fragment
3470 * @frag: the paged fragment
3471 * @page: the page to set
3473 * Sets the fragment @frag to contain @page.
3475 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3477 frag->bv_page = page;
3481 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3483 * @f: the fragment offset
3484 * @page: the page to set
3486 * Sets the @f'th fragment of @skb to contain @page.
3488 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3491 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3494 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3497 * skb_frag_dma_map - maps a paged fragment via the DMA API
3498 * @dev: the device to map the fragment to
3499 * @frag: the paged fragment to map
3500 * @offset: the offset within the fragment (starting at the
3501 * fragment's own offset)
3502 * @size: the number of bytes to map
3503 * @dir: the direction of the mapping (``PCI_DMA_*``)
3505 * Maps the page associated with @frag to @device.
3507 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3508 const skb_frag_t *frag,
3509 size_t offset, size_t size,
3510 enum dma_data_direction dir)
3512 return dma_map_page(dev, skb_frag_page(frag),
3513 skb_frag_off(frag) + offset, size, dir);
3516 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3519 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3523 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3526 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3531 * skb_clone_writable - is the header of a clone writable
3532 * @skb: buffer to check
3533 * @len: length up to which to write
3535 * Returns true if modifying the header part of the cloned buffer
3536 * does not requires the data to be copied.
3538 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3540 return !skb_header_cloned(skb) &&
3541 skb_headroom(skb) + len <= skb->hdr_len;
3544 static inline int skb_try_make_writable(struct sk_buff *skb,
3545 unsigned int write_len)
3547 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3548 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3551 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3556 if (headroom > skb_headroom(skb))
3557 delta = headroom - skb_headroom(skb);
3559 if (delta || cloned)
3560 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3566 * skb_cow - copy header of skb when it is required
3567 * @skb: buffer to cow
3568 * @headroom: needed headroom
3570 * If the skb passed lacks sufficient headroom or its data part
3571 * is shared, data is reallocated. If reallocation fails, an error
3572 * is returned and original skb is not changed.
3574 * The result is skb with writable area skb->head...skb->tail
3575 * and at least @headroom of space at head.
3577 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3579 return __skb_cow(skb, headroom, skb_cloned(skb));
3583 * skb_cow_head - skb_cow but only making the head writable
3584 * @skb: buffer to cow
3585 * @headroom: needed headroom
3587 * This function is identical to skb_cow except that we replace the
3588 * skb_cloned check by skb_header_cloned. It should be used when
3589 * you only need to push on some header and do not need to modify
3592 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3594 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3598 * skb_padto - pad an skbuff up to a minimal size
3599 * @skb: buffer to pad
3600 * @len: minimal length
3602 * Pads up a buffer to ensure the trailing bytes exist and are
3603 * blanked. If the buffer already contains sufficient data it
3604 * is untouched. Otherwise it is extended. Returns zero on
3605 * success. The skb is freed on error.
3607 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3609 unsigned int size = skb->len;
3610 if (likely(size >= len))
3612 return skb_pad(skb, len - size);
3616 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3617 * @skb: buffer to pad
3618 * @len: minimal length
3619 * @free_on_error: free buffer on error
3621 * Pads up a buffer to ensure the trailing bytes exist and are
3622 * blanked. If the buffer already contains sufficient data it
3623 * is untouched. Otherwise it is extended. Returns zero on
3624 * success. The skb is freed on error if @free_on_error is true.
3626 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3630 unsigned int size = skb->len;
3632 if (unlikely(size < len)) {
3634 if (__skb_pad(skb, len, free_on_error))
3636 __skb_put(skb, len);
3642 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3643 * @skb: buffer to pad
3644 * @len: minimal length
3646 * Pads up a buffer to ensure the trailing bytes exist and are
3647 * blanked. If the buffer already contains sufficient data it
3648 * is untouched. Otherwise it is extended. Returns zero on
3649 * success. The skb is freed on error.
3651 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3653 return __skb_put_padto(skb, len, true);
3656 static inline int skb_add_data(struct sk_buff *skb,
3657 struct iov_iter *from, int copy)
3659 const int off = skb->len;
3661 if (skb->ip_summed == CHECKSUM_NONE) {
3663 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3665 skb->csum = csum_block_add(skb->csum, csum, off);
3668 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3671 __skb_trim(skb, off);
3675 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3676 const struct page *page, int off)
3681 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3683 return page == skb_frag_page(frag) &&
3684 off == skb_frag_off(frag) + skb_frag_size(frag);
3689 static inline int __skb_linearize(struct sk_buff *skb)
3691 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3695 * skb_linearize - convert paged skb to linear one
3696 * @skb: buffer to linarize
3698 * If there is no free memory -ENOMEM is returned, otherwise zero
3699 * is returned and the old skb data released.
3701 static inline int skb_linearize(struct sk_buff *skb)
3703 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3707 * skb_has_shared_frag - can any frag be overwritten
3708 * @skb: buffer to test
3710 * Return true if the skb has at least one frag that might be modified
3711 * by an external entity (as in vmsplice()/sendfile())
3713 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3715 return skb_is_nonlinear(skb) &&
3716 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3720 * skb_linearize_cow - make sure skb is linear and writable
3721 * @skb: buffer to process
3723 * If there is no free memory -ENOMEM is returned, otherwise zero
3724 * is returned and the old skb data released.
3726 static inline int skb_linearize_cow(struct sk_buff *skb)
3728 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3729 __skb_linearize(skb) : 0;
3732 static __always_inline void
3733 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3736 if (skb->ip_summed == CHECKSUM_COMPLETE)
3737 skb->csum = csum_block_sub(skb->csum,
3738 csum_partial(start, len, 0), off);
3739 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3740 skb_checksum_start_offset(skb) < 0)
3741 skb->ip_summed = CHECKSUM_NONE;
3745 * skb_postpull_rcsum - update checksum for received skb after pull
3746 * @skb: buffer to update
3747 * @start: start of data before pull
3748 * @len: length of data pulled
3750 * After doing a pull on a received packet, you need to call this to
3751 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3752 * CHECKSUM_NONE so that it can be recomputed from scratch.
3754 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3755 const void *start, unsigned int len)
3757 if (skb->ip_summed == CHECKSUM_COMPLETE)
3758 skb->csum = wsum_negate(csum_partial(start, len,
3759 wsum_negate(skb->csum)));
3760 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3761 skb_checksum_start_offset(skb) < 0)
3762 skb->ip_summed = CHECKSUM_NONE;
3765 static __always_inline void
3766 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3769 if (skb->ip_summed == CHECKSUM_COMPLETE)
3770 skb->csum = csum_block_add(skb->csum,
3771 csum_partial(start, len, 0), off);
3775 * skb_postpush_rcsum - update checksum for received skb after push
3776 * @skb: buffer to update
3777 * @start: start of data after push
3778 * @len: length of data pushed
3780 * After doing a push on a received packet, you need to call this to
3781 * update the CHECKSUM_COMPLETE checksum.
3783 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3784 const void *start, unsigned int len)
3786 __skb_postpush_rcsum(skb, start, len, 0);
3789 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3792 * skb_push_rcsum - push skb and update receive checksum
3793 * @skb: buffer to update
3794 * @len: length of data pulled
3796 * This function performs an skb_push on the packet and updates
3797 * the CHECKSUM_COMPLETE checksum. It should be used on
3798 * receive path processing instead of skb_push unless you know
3799 * that the checksum difference is zero (e.g., a valid IP header)
3800 * or you are setting ip_summed to CHECKSUM_NONE.
3802 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3805 skb_postpush_rcsum(skb, skb->data, len);
3809 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3811 * pskb_trim_rcsum - trim received skb and update checksum
3812 * @skb: buffer to trim
3815 * This is exactly the same as pskb_trim except that it ensures the
3816 * checksum of received packets are still valid after the operation.
3817 * It can change skb pointers.
3820 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3822 if (likely(len >= skb->len))
3824 return pskb_trim_rcsum_slow(skb, len);
3827 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3829 if (skb->ip_summed == CHECKSUM_COMPLETE)
3830 skb->ip_summed = CHECKSUM_NONE;
3831 __skb_trim(skb, len);
3835 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3837 if (skb->ip_summed == CHECKSUM_COMPLETE)
3838 skb->ip_summed = CHECKSUM_NONE;
3839 return __skb_grow(skb, len);
3842 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3843 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3844 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3845 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3846 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3848 #define skb_queue_walk(queue, skb) \
3849 for (skb = (queue)->next; \
3850 skb != (struct sk_buff *)(queue); \
3853 #define skb_queue_walk_safe(queue, skb, tmp) \
3854 for (skb = (queue)->next, tmp = skb->next; \
3855 skb != (struct sk_buff *)(queue); \
3856 skb = tmp, tmp = skb->next)
3858 #define skb_queue_walk_from(queue, skb) \
3859 for (; skb != (struct sk_buff *)(queue); \
3862 #define skb_rbtree_walk(skb, root) \
3863 for (skb = skb_rb_first(root); skb != NULL; \
3864 skb = skb_rb_next(skb))
3866 #define skb_rbtree_walk_from(skb) \
3867 for (; skb != NULL; \
3868 skb = skb_rb_next(skb))
3870 #define skb_rbtree_walk_from_safe(skb, tmp) \
3871 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3874 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3875 for (tmp = skb->next; \
3876 skb != (struct sk_buff *)(queue); \
3877 skb = tmp, tmp = skb->next)
3879 #define skb_queue_reverse_walk(queue, skb) \
3880 for (skb = (queue)->prev; \
3881 skb != (struct sk_buff *)(queue); \
3884 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3885 for (skb = (queue)->prev, tmp = skb->prev; \
3886 skb != (struct sk_buff *)(queue); \
3887 skb = tmp, tmp = skb->prev)
3889 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3890 for (tmp = skb->prev; \
3891 skb != (struct sk_buff *)(queue); \
3892 skb = tmp, tmp = skb->prev)
3894 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3896 return skb_shinfo(skb)->frag_list != NULL;
3899 static inline void skb_frag_list_init(struct sk_buff *skb)
3901 skb_shinfo(skb)->frag_list = NULL;
3904 #define skb_walk_frags(skb, iter) \
3905 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3908 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3909 int *err, long *timeo_p,
3910 const struct sk_buff *skb);
3911 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3912 struct sk_buff_head *queue,
3915 struct sk_buff **last);
3916 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
3917 struct sk_buff_head *queue,
3918 unsigned int flags, int *off, int *err,
3919 struct sk_buff **last);
3920 struct sk_buff *__skb_recv_datagram(struct sock *sk,
3921 struct sk_buff_head *sk_queue,
3922 unsigned int flags, int *off, int *err);
3923 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
3924 __poll_t datagram_poll(struct file *file, struct socket *sock,
3925 struct poll_table_struct *wait);
3926 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3927 struct iov_iter *to, int size);
3928 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3929 struct msghdr *msg, int size)
3931 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3933 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3934 struct msghdr *msg);
3935 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3936 struct iov_iter *to, int len,
3937 struct ahash_request *hash);
3938 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3939 struct iov_iter *from, int len);
3940 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3941 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3942 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3943 static inline void skb_free_datagram_locked(struct sock *sk,
3944 struct sk_buff *skb)
3946 __skb_free_datagram_locked(sk, skb, 0);
3948 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3949 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3950 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3951 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3953 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3954 struct pipe_inode_info *pipe, unsigned int len,
3955 unsigned int flags);
3956 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3958 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3959 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3960 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3961 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3963 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3964 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3965 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3966 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3967 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3968 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3969 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
3970 unsigned int offset);
3971 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3972 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
3973 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3974 int skb_vlan_pop(struct sk_buff *skb);
3975 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3976 int skb_eth_pop(struct sk_buff *skb);
3977 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
3978 const unsigned char *src);
3979 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
3980 int mac_len, bool ethernet);
3981 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
3983 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3984 int skb_mpls_dec_ttl(struct sk_buff *skb);
3985 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3988 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3990 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3993 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3995 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3998 struct skb_checksum_ops {
3999 __wsum (*update)(const void *mem, int len, __wsum wsum);
4000 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
4003 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
4005 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
4006 __wsum csum, const struct skb_checksum_ops *ops);
4007 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
4010 static inline void * __must_check
4011 __skb_header_pointer(const struct sk_buff *skb, int offset, int len,
4012 const void *data, int hlen, void *buffer)
4014 if (likely(hlen - offset >= len))
4015 return (void *)data + offset;
4017 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
4023 static inline void * __must_check
4024 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
4026 return __skb_header_pointer(skb, offset, len, skb->data,
4027 skb_headlen(skb), buffer);
4031 * skb_needs_linearize - check if we need to linearize a given skb
4032 * depending on the given device features.
4033 * @skb: socket buffer to check
4034 * @features: net device features
4036 * Returns true if either:
4037 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
4038 * 2. skb is fragmented and the device does not support SG.
4040 static inline bool skb_needs_linearize(struct sk_buff *skb,
4041 netdev_features_t features)
4043 return skb_is_nonlinear(skb) &&
4044 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
4045 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
4048 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
4050 const unsigned int len)
4052 memcpy(to, skb->data, len);
4055 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4056 const int offset, void *to,
4057 const unsigned int len)
4059 memcpy(to, skb->data + offset, len);
4062 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4064 const unsigned int len)
4066 memcpy(skb->data, from, len);
4069 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4072 const unsigned int len)
4074 memcpy(skb->data + offset, from, len);
4077 void skb_init(void);
4079 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4085 * skb_get_timestamp - get timestamp from a skb
4086 * @skb: skb to get stamp from
4087 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4089 * Timestamps are stored in the skb as offsets to a base timestamp.
4090 * This function converts the offset back to a struct timeval and stores
4093 static inline void skb_get_timestamp(const struct sk_buff *skb,
4094 struct __kernel_old_timeval *stamp)
4096 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4099 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4100 struct __kernel_sock_timeval *stamp)
4102 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4104 stamp->tv_sec = ts.tv_sec;
4105 stamp->tv_usec = ts.tv_nsec / 1000;
4108 static inline void skb_get_timestampns(const struct sk_buff *skb,
4109 struct __kernel_old_timespec *stamp)
4111 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4113 stamp->tv_sec = ts.tv_sec;
4114 stamp->tv_nsec = ts.tv_nsec;
4117 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4118 struct __kernel_timespec *stamp)
4120 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4122 stamp->tv_sec = ts.tv_sec;
4123 stamp->tv_nsec = ts.tv_nsec;
4126 static inline void __net_timestamp(struct sk_buff *skb)
4128 skb->tstamp = ktime_get_real();
4129 skb->mono_delivery_time = 0;
4132 static inline ktime_t net_timedelta(ktime_t t)
4134 return ktime_sub(ktime_get_real(), t);
4137 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4141 skb->mono_delivery_time = kt && mono;
4144 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4146 /* It is used in the ingress path to clear the delivery_time.
4147 * If needed, set the skb->tstamp to the (rcv) timestamp.
4149 static inline void skb_clear_delivery_time(struct sk_buff *skb)
4151 if (skb->mono_delivery_time) {
4152 skb->mono_delivery_time = 0;
4153 if (static_branch_unlikely(&netstamp_needed_key))
4154 skb->tstamp = ktime_get_real();
4160 static inline void skb_clear_tstamp(struct sk_buff *skb)
4162 if (skb->mono_delivery_time)
4168 static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4170 if (skb->mono_delivery_time)
4176 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4178 if (!skb->mono_delivery_time && skb->tstamp)
4181 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4182 return ktime_get_real();
4187 static inline u8 skb_metadata_len(const struct sk_buff *skb)
4189 return skb_shinfo(skb)->meta_len;
4192 static inline void *skb_metadata_end(const struct sk_buff *skb)
4194 return skb_mac_header(skb);
4197 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4198 const struct sk_buff *skb_b,
4201 const void *a = skb_metadata_end(skb_a);
4202 const void *b = skb_metadata_end(skb_b);
4203 /* Using more efficient varaiant than plain call to memcmp(). */
4204 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
4208 #define __it(x, op) (x -= sizeof(u##op))
4209 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4210 case 32: diffs |= __it_diff(a, b, 64);
4212 case 24: diffs |= __it_diff(a, b, 64);
4214 case 16: diffs |= __it_diff(a, b, 64);
4216 case 8: diffs |= __it_diff(a, b, 64);
4218 case 28: diffs |= __it_diff(a, b, 64);
4220 case 20: diffs |= __it_diff(a, b, 64);
4222 case 12: diffs |= __it_diff(a, b, 64);
4224 case 4: diffs |= __it_diff(a, b, 32);
4229 return memcmp(a - meta_len, b - meta_len, meta_len);
4233 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4234 const struct sk_buff *skb_b)
4236 u8 len_a = skb_metadata_len(skb_a);
4237 u8 len_b = skb_metadata_len(skb_b);
4239 if (!(len_a | len_b))
4242 return len_a != len_b ?
4243 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4246 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4248 skb_shinfo(skb)->meta_len = meta_len;
4251 static inline void skb_metadata_clear(struct sk_buff *skb)
4253 skb_metadata_set(skb, 0);
4256 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4258 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4260 void skb_clone_tx_timestamp(struct sk_buff *skb);
4261 bool skb_defer_rx_timestamp(struct sk_buff *skb);
4263 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4265 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4269 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4274 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4277 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4279 * PHY drivers may accept clones of transmitted packets for
4280 * timestamping via their phy_driver.txtstamp method. These drivers
4281 * must call this function to return the skb back to the stack with a
4284 * @skb: clone of the original outgoing packet
4285 * @hwtstamps: hardware time stamps
4288 void skb_complete_tx_timestamp(struct sk_buff *skb,
4289 struct skb_shared_hwtstamps *hwtstamps);
4291 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4292 struct skb_shared_hwtstamps *hwtstamps,
4293 struct sock *sk, int tstype);
4296 * skb_tstamp_tx - queue clone of skb with send time stamps
4297 * @orig_skb: the original outgoing packet
4298 * @hwtstamps: hardware time stamps, may be NULL if not available
4300 * If the skb has a socket associated, then this function clones the
4301 * skb (thus sharing the actual data and optional structures), stores
4302 * the optional hardware time stamping information (if non NULL) or
4303 * generates a software time stamp (otherwise), then queues the clone
4304 * to the error queue of the socket. Errors are silently ignored.
4306 void skb_tstamp_tx(struct sk_buff *orig_skb,
4307 struct skb_shared_hwtstamps *hwtstamps);
4310 * skb_tx_timestamp() - Driver hook for transmit timestamping
4312 * Ethernet MAC Drivers should call this function in their hard_xmit()
4313 * function immediately before giving the sk_buff to the MAC hardware.
4315 * Specifically, one should make absolutely sure that this function is
4316 * called before TX completion of this packet can trigger. Otherwise
4317 * the packet could potentially already be freed.
4319 * @skb: A socket buffer.
4321 static inline void skb_tx_timestamp(struct sk_buff *skb)
4323 skb_clone_tx_timestamp(skb);
4324 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4325 skb_tstamp_tx(skb, NULL);
4329 * skb_complete_wifi_ack - deliver skb with wifi status
4331 * @skb: the original outgoing packet
4332 * @acked: ack status
4335 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4337 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4338 __sum16 __skb_checksum_complete(struct sk_buff *skb);
4340 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4342 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4344 (skb->ip_summed == CHECKSUM_PARTIAL &&
4345 skb_checksum_start_offset(skb) >= 0));
4349 * skb_checksum_complete - Calculate checksum of an entire packet
4350 * @skb: packet to process
4352 * This function calculates the checksum over the entire packet plus
4353 * the value of skb->csum. The latter can be used to supply the
4354 * checksum of a pseudo header as used by TCP/UDP. It returns the
4357 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4358 * this function can be used to verify that checksum on received
4359 * packets. In that case the function should return zero if the
4360 * checksum is correct. In particular, this function will return zero
4361 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4362 * hardware has already verified the correctness of the checksum.
4364 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4366 return skb_csum_unnecessary(skb) ?
4367 0 : __skb_checksum_complete(skb);
4370 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4372 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4373 if (skb->csum_level == 0)
4374 skb->ip_summed = CHECKSUM_NONE;
4380 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4382 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4383 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4385 } else if (skb->ip_summed == CHECKSUM_NONE) {
4386 skb->ip_summed = CHECKSUM_UNNECESSARY;
4387 skb->csum_level = 0;
4391 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4393 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4394 skb->ip_summed = CHECKSUM_NONE;
4395 skb->csum_level = 0;
4399 /* Check if we need to perform checksum complete validation.
4401 * Returns true if checksum complete is needed, false otherwise
4402 * (either checksum is unnecessary or zero checksum is allowed).
4404 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4408 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4409 skb->csum_valid = 1;
4410 __skb_decr_checksum_unnecessary(skb);
4417 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4420 #define CHECKSUM_BREAK 76
4422 /* Unset checksum-complete
4424 * Unset checksum complete can be done when packet is being modified
4425 * (uncompressed for instance) and checksum-complete value is
4428 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4430 if (skb->ip_summed == CHECKSUM_COMPLETE)
4431 skb->ip_summed = CHECKSUM_NONE;
4434 /* Validate (init) checksum based on checksum complete.
4437 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4438 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4439 * checksum is stored in skb->csum for use in __skb_checksum_complete
4440 * non-zero: value of invalid checksum
4443 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4447 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4448 if (!csum_fold(csum_add(psum, skb->csum))) {
4449 skb->csum_valid = 1;
4456 if (complete || skb->len <= CHECKSUM_BREAK) {
4459 csum = __skb_checksum_complete(skb);
4460 skb->csum_valid = !csum;
4467 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4472 /* Perform checksum validate (init). Note that this is a macro since we only
4473 * want to calculate the pseudo header which is an input function if necessary.
4474 * First we try to validate without any computation (checksum unnecessary) and
4475 * then calculate based on checksum complete calling the function to compute
4479 * 0: checksum is validated or try to in skb_checksum_complete
4480 * non-zero: value of invalid checksum
4482 #define __skb_checksum_validate(skb, proto, complete, \
4483 zero_okay, check, compute_pseudo) \
4485 __sum16 __ret = 0; \
4486 skb->csum_valid = 0; \
4487 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4488 __ret = __skb_checksum_validate_complete(skb, \
4489 complete, compute_pseudo(skb, proto)); \
4493 #define skb_checksum_init(skb, proto, compute_pseudo) \
4494 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4496 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4497 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4499 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4500 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4502 #define skb_checksum_validate_zero_check(skb, proto, check, \
4504 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4506 #define skb_checksum_simple_validate(skb) \
4507 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4509 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4511 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4514 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4516 skb->csum = ~pseudo;
4517 skb->ip_summed = CHECKSUM_COMPLETE;
4520 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4522 if (__skb_checksum_convert_check(skb)) \
4523 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4526 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4527 u16 start, u16 offset)
4529 skb->ip_summed = CHECKSUM_PARTIAL;
4530 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4531 skb->csum_offset = offset - start;
4534 /* Update skbuf and packet to reflect the remote checksum offload operation.
4535 * When called, ptr indicates the starting point for skb->csum when
4536 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4537 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4539 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4540 int start, int offset, bool nopartial)
4545 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4549 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4550 __skb_checksum_complete(skb);
4551 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4554 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4556 /* Adjust skb->csum since we changed the packet */
4557 skb->csum = csum_add(skb->csum, delta);
4560 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4562 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4563 return (void *)(skb->_nfct & NFCT_PTRMASK);
4569 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4571 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4578 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4580 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4581 skb->slow_gro |= !!nfct;
4586 #ifdef CONFIG_SKB_EXTENSIONS
4588 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4594 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4597 #if IS_ENABLED(CONFIG_MPTCP)
4600 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4603 SKB_EXT_NUM, /* must be last */
4607 * struct skb_ext - sk_buff extensions
4608 * @refcnt: 1 on allocation, deallocated on 0
4609 * @offset: offset to add to @data to obtain extension address
4610 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4611 * @data: start of extension data, variable sized
4613 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4614 * to use 'u8' types while allowing up to 2kb worth of extension data.
4618 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4619 u8 chunks; /* same */
4620 char data[] __aligned(8);
4623 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4624 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4625 struct skb_ext *ext);
4626 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4627 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4628 void __skb_ext_put(struct skb_ext *ext);
4630 static inline void skb_ext_put(struct sk_buff *skb)
4632 if (skb->active_extensions)
4633 __skb_ext_put(skb->extensions);
4636 static inline void __skb_ext_copy(struct sk_buff *dst,
4637 const struct sk_buff *src)
4639 dst->active_extensions = src->active_extensions;
4641 if (src->active_extensions) {
4642 struct skb_ext *ext = src->extensions;
4644 refcount_inc(&ext->refcnt);
4645 dst->extensions = ext;
4649 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4652 __skb_ext_copy(dst, src);
4655 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4657 return !!ext->offset[i];
4660 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4662 return skb->active_extensions & (1 << id);
4665 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4667 if (skb_ext_exist(skb, id))
4668 __skb_ext_del(skb, id);
4671 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4673 if (skb_ext_exist(skb, id)) {
4674 struct skb_ext *ext = skb->extensions;
4676 return (void *)ext + (ext->offset[id] << 3);
4682 static inline void skb_ext_reset(struct sk_buff *skb)
4684 if (unlikely(skb->active_extensions)) {
4685 __skb_ext_put(skb->extensions);
4686 skb->active_extensions = 0;
4690 static inline bool skb_has_extensions(struct sk_buff *skb)
4692 return unlikely(skb->active_extensions);
4695 static inline void skb_ext_put(struct sk_buff *skb) {}
4696 static inline void skb_ext_reset(struct sk_buff *skb) {}
4697 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4698 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4699 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4700 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4701 #endif /* CONFIG_SKB_EXTENSIONS */
4703 static inline void nf_reset_ct(struct sk_buff *skb)
4705 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4706 nf_conntrack_put(skb_nfct(skb));
4711 static inline void nf_reset_trace(struct sk_buff *skb)
4713 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4718 static inline void ipvs_reset(struct sk_buff *skb)
4720 #if IS_ENABLED(CONFIG_IP_VS)
4721 skb->ipvs_property = 0;
4725 /* Note: This doesn't put any conntrack info in dst. */
4726 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4729 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4730 dst->_nfct = src->_nfct;
4731 nf_conntrack_get(skb_nfct(src));
4733 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4735 dst->nf_trace = src->nf_trace;
4739 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4741 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4742 nf_conntrack_put(skb_nfct(dst));
4744 dst->slow_gro = src->slow_gro;
4745 __nf_copy(dst, src, true);
4748 #ifdef CONFIG_NETWORK_SECMARK
4749 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4751 to->secmark = from->secmark;
4754 static inline void skb_init_secmark(struct sk_buff *skb)
4759 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4762 static inline void skb_init_secmark(struct sk_buff *skb)
4766 static inline int secpath_exists(const struct sk_buff *skb)
4769 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4775 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4777 return !skb->destructor &&
4778 !secpath_exists(skb) &&
4780 !skb->_skb_refdst &&
4781 !skb_has_frag_list(skb);
4784 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4786 skb->queue_mapping = queue_mapping;
4789 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4791 return skb->queue_mapping;
4794 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4796 to->queue_mapping = from->queue_mapping;
4799 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4801 skb->queue_mapping = rx_queue + 1;
4804 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4806 return skb->queue_mapping - 1;
4809 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4811 return skb->queue_mapping != 0;
4814 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4816 skb->dst_pending_confirm = val;
4819 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4821 return skb->dst_pending_confirm != 0;
4824 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4827 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4833 /* Keeps track of mac header offset relative to skb->head.
4834 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4835 * For non-tunnel skb it points to skb_mac_header() and for
4836 * tunnel skb it points to outer mac header.
4837 * Keeps track of level of encapsulation of network headers.
4848 #define SKB_GSO_CB_OFFSET 32
4849 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET))
4851 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4853 return (skb_mac_header(inner_skb) - inner_skb->head) -
4854 SKB_GSO_CB(inner_skb)->mac_offset;
4857 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4859 int new_headroom, headroom;
4862 headroom = skb_headroom(skb);
4863 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4867 new_headroom = skb_headroom(skb);
4868 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4872 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4874 /* Do not update partial checksums if remote checksum is enabled. */
4875 if (skb->remcsum_offload)
4878 SKB_GSO_CB(skb)->csum = res;
4879 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4882 /* Compute the checksum for a gso segment. First compute the checksum value
4883 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4884 * then add in skb->csum (checksum from csum_start to end of packet).
4885 * skb->csum and csum_start are then updated to reflect the checksum of the
4886 * resultant packet starting from the transport header-- the resultant checksum
4887 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4890 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4892 unsigned char *csum_start = skb_transport_header(skb);
4893 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4894 __wsum partial = SKB_GSO_CB(skb)->csum;
4896 SKB_GSO_CB(skb)->csum = res;
4897 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4899 return csum_fold(csum_partial(csum_start, plen, partial));
4902 static inline bool skb_is_gso(const struct sk_buff *skb)
4904 return skb_shinfo(skb)->gso_size;
4907 /* Note: Should be called only if skb_is_gso(skb) is true */
4908 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4910 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4913 /* Note: Should be called only if skb_is_gso(skb) is true */
4914 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4916 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4919 /* Note: Should be called only if skb_is_gso(skb) is true */
4920 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4922 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4925 static inline void skb_gso_reset(struct sk_buff *skb)
4927 skb_shinfo(skb)->gso_size = 0;
4928 skb_shinfo(skb)->gso_segs = 0;
4929 skb_shinfo(skb)->gso_type = 0;
4932 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4935 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4937 shinfo->gso_size += increment;
4940 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4943 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4945 shinfo->gso_size -= decrement;
4948 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4950 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4952 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4953 * wanted then gso_type will be set. */
4954 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4956 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4957 unlikely(shinfo->gso_type == 0)) {
4958 __skb_warn_lro_forwarding(skb);
4964 static inline void skb_forward_csum(struct sk_buff *skb)
4966 /* Unfortunately we don't support this one. Any brave souls? */
4967 if (skb->ip_summed == CHECKSUM_COMPLETE)
4968 skb->ip_summed = CHECKSUM_NONE;
4972 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4973 * @skb: skb to check
4975 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4976 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4977 * use this helper, to document places where we make this assertion.
4979 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4981 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE);
4984 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4986 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4987 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4988 unsigned int transport_len,
4989 __sum16(*skb_chkf)(struct sk_buff *skb));
4992 * skb_head_is_locked - Determine if the skb->head is locked down
4993 * @skb: skb to check
4995 * The head on skbs build around a head frag can be removed if they are
4996 * not cloned. This function returns true if the skb head is locked down
4997 * due to either being allocated via kmalloc, or by being a clone with
4998 * multiple references to the head.
5000 static inline bool skb_head_is_locked(const struct sk_buff *skb)
5002 return !skb->head_frag || skb_cloned(skb);
5005 /* Local Checksum Offload.
5006 * Compute outer checksum based on the assumption that the
5007 * inner checksum will be offloaded later.
5008 * See Documentation/networking/checksum-offloads.rst for
5009 * explanation of how this works.
5010 * Fill in outer checksum adjustment (e.g. with sum of outer
5011 * pseudo-header) before calling.
5012 * Also ensure that inner checksum is in linear data area.
5014 static inline __wsum lco_csum(struct sk_buff *skb)
5016 unsigned char *csum_start = skb_checksum_start(skb);
5017 unsigned char *l4_hdr = skb_transport_header(skb);
5020 /* Start with complement of inner checksum adjustment */
5021 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
5024 /* Add in checksum of our headers (incl. outer checksum
5025 * adjustment filled in by caller) and return result.
5027 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
5030 static inline bool skb_is_redirected(const struct sk_buff *skb)
5032 return skb->redirected;
5035 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
5037 skb->redirected = 1;
5038 #ifdef CONFIG_NET_REDIRECT
5039 skb->from_ingress = from_ingress;
5040 if (skb->from_ingress)
5041 skb_clear_tstamp(skb);
5045 static inline void skb_reset_redirect(struct sk_buff *skb)
5047 skb->redirected = 0;
5050 static inline bool skb_csum_is_sctp(struct sk_buff *skb)
5052 return skb->csum_not_inet;
5055 static inline void skb_set_kcov_handle(struct sk_buff *skb,
5056 const u64 kcov_handle)
5059 skb->kcov_handle = kcov_handle;
5063 static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
5066 return skb->kcov_handle;
5072 #ifdef CONFIG_PAGE_POOL
5073 static inline void skb_mark_for_recycle(struct sk_buff *skb)
5075 skb->pp_recycle = 1;
5079 #endif /* __KERNEL__ */
5080 #endif /* _LINUX_SKBUFF_H */