1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * Definitions for the 'struct sk_buff' memory handlers.
6 * Alan Cox, <gw4pts@gw4pts.ampr.org>
7 * Florian La Roche, <rzsfl@rz.uni-sb.de>
10 #ifndef _LINUX_SKBUFF_H
11 #define _LINUX_SKBUFF_H
13 #include <linux/kernel.h>
14 #include <linux/compiler.h>
15 #include <linux/time.h>
16 #include <linux/bug.h>
17 #include <linux/bvec.h>
18 #include <linux/cache.h>
19 #include <linux/rbtree.h>
20 #include <linux/socket.h>
21 #include <linux/refcount.h>
23 #include <linux/atomic.h>
24 #include <asm/types.h>
25 #include <linux/spinlock.h>
26 #include <net/checksum.h>
27 #include <linux/rcupdate.h>
28 #include <linux/dma-mapping.h>
29 #include <linux/netdev_features.h>
30 #include <net/flow_dissector.h>
31 #include <linux/in6.h>
32 #include <linux/if_packet.h>
33 #include <linux/llist.h>
35 #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)))
258 #define SKB_MAX_ORDER(X, ORDER) \
259 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
260 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
261 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
263 /* return minimum truesize of one skb containing X bytes of data */
264 #define SKB_TRUESIZE(X) ((X) + \
265 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
266 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
268 struct ahash_request;
271 struct pipe_inode_info;
279 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
280 struct nf_bridge_info {
282 BRNF_PROTO_UNCHANGED,
290 struct net_device *physindev;
292 /* always valid & non-NULL from FORWARD on, for physdev match */
293 struct net_device *physoutdev;
295 /* prerouting: detect dnat in orig/reply direction */
297 struct in6_addr ipv6_daddr;
299 /* after prerouting + nat detected: store original source
300 * mac since neigh resolution overwrites it, only used while
301 * skb is out in neigh layer.
303 char neigh_header[8];
308 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
309 /* Chain in tc_skb_ext will be used to share the tc chain with
310 * ovs recirc_id. It will be set to the current chain by tc
311 * and read by ovs to recirc_id.
323 struct sk_buff_head {
324 /* These two members must be first to match sk_buff. */
325 struct_group_tagged(sk_buff_list, list,
326 struct sk_buff *next;
327 struct sk_buff *prev;
336 /* To allow 64K frame to be packed as single skb without frag_list we
337 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
338 * buffers which do not start on a page boundary.
340 * Since GRO uses frags we allocate at least 16 regardless of page
343 #if (65536/PAGE_SIZE + 1) < 16
344 #define MAX_SKB_FRAGS 16UL
346 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
348 extern int sysctl_max_skb_frags;
350 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
351 * segment using its current segmentation instead.
353 #define GSO_BY_FRAGS 0xFFFF
355 typedef struct bio_vec skb_frag_t;
358 * skb_frag_size() - Returns the size of a skb fragment
359 * @frag: skb fragment
361 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
367 * skb_frag_size_set() - Sets the size of a skb fragment
368 * @frag: skb fragment
369 * @size: size of fragment
371 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
377 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
378 * @frag: skb fragment
379 * @delta: value to add
381 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
383 frag->bv_len += delta;
387 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
388 * @frag: skb fragment
389 * @delta: value to subtract
391 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
393 frag->bv_len -= delta;
397 * skb_frag_must_loop - Test if %p is a high memory page
398 * @p: fragment's page
400 static inline bool skb_frag_must_loop(struct page *p)
402 #if defined(CONFIG_HIGHMEM)
403 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
410 * skb_frag_foreach_page - loop over pages in a fragment
412 * @f: skb frag to operate on
413 * @f_off: offset from start of f->bv_page
414 * @f_len: length from f_off to loop over
415 * @p: (temp var) current page
416 * @p_off: (temp var) offset from start of current page,
417 * non-zero only on first page.
418 * @p_len: (temp var) length in current page,
419 * < PAGE_SIZE only on first and last page.
420 * @copied: (temp var) length so far, excluding current p_len.
422 * A fragment can hold a compound page, in which case per-page
423 * operations, notably kmap_atomic, must be called for each
426 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
427 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
428 p_off = (f_off) & (PAGE_SIZE - 1), \
429 p_len = skb_frag_must_loop(p) ? \
430 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
433 copied += p_len, p++, p_off = 0, \
434 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
436 #define HAVE_HW_TIME_STAMP
439 * struct skb_shared_hwtstamps - hardware time stamps
440 * @hwtstamp: hardware time stamp transformed into duration
441 * since arbitrary point in time
442 * @netdev_data: address/cookie of network device driver used as
443 * reference to actual hardware time stamp
445 * Software time stamps generated by ktime_get_real() are stored in
448 * hwtstamps can only be compared against other hwtstamps from
451 * This structure is attached to packets as part of the
452 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
454 struct skb_shared_hwtstamps {
461 /* Definitions for tx_flags in struct skb_shared_info */
463 /* generate hardware time stamp */
464 SKBTX_HW_TSTAMP = 1 << 0,
466 /* generate software time stamp when queueing packet to NIC */
467 SKBTX_SW_TSTAMP = 1 << 1,
469 /* device driver is going to provide hardware time stamp */
470 SKBTX_IN_PROGRESS = 1 << 2,
472 /* generate hardware time stamp based on cycles if supported */
473 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3,
475 /* generate wifi status information (where possible) */
476 SKBTX_WIFI_STATUS = 1 << 4,
478 /* determine hardware time stamp based on time or cycles */
479 SKBTX_HW_TSTAMP_NETDEV = 1 << 5,
481 /* generate software time stamp when entering packet scheduling */
482 SKBTX_SCHED_TSTAMP = 1 << 6,
485 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
487 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \
488 SKBTX_HW_TSTAMP_USE_CYCLES | \
491 /* Definitions for flags in struct skb_shared_info */
493 /* use zcopy routines */
494 SKBFL_ZEROCOPY_ENABLE = BIT(0),
496 /* This indicates at least one fragment might be overwritten
497 * (as in vmsplice(), sendfile() ...)
498 * If we need to compute a TX checksum, we'll need to copy
499 * all frags to avoid possible bad checksum
501 SKBFL_SHARED_FRAG = BIT(1),
503 /* segment contains only zerocopy data and should not be
504 * charged to the kernel memory.
506 SKBFL_PURE_ZEROCOPY = BIT(2),
508 SKBFL_DONT_ORPHAN = BIT(3),
510 /* page references are managed by the ubuf_info, so it's safe to
511 * use frags only up until ubuf_info is released
513 SKBFL_MANAGED_FRAG_REFS = BIT(4),
516 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
517 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
518 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
521 * The callback notifies userspace to release buffers when skb DMA is done in
522 * lower device, the skb last reference should be 0 when calling this.
523 * The zerocopy_success argument is true if zero copy transmit occurred,
524 * false on data copy or out of memory error caused by data copy attempt.
525 * The ctx field is used to track device context.
526 * The desc field is used to track userspace buffer index.
529 void (*callback)(struct sk_buff *, struct ubuf_info *,
530 bool zerocopy_success);
535 struct ubuf_info_msgzc {
536 struct ubuf_info ubuf;
552 struct user_struct *user;
557 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
558 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \
561 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
562 void mm_unaccount_pinned_pages(struct mmpin *mmp);
564 /* This data is invariant across clones and lives at
565 * the end of the header data, ie. at skb->end.
567 struct skb_shared_info {
572 unsigned short gso_size;
573 /* Warning: this field is not always filled in (UFO)! */
574 unsigned short gso_segs;
575 struct sk_buff *frag_list;
576 struct skb_shared_hwtstamps hwtstamps;
577 unsigned int gso_type;
581 * Warning : all fields before dataref are cleared in __alloc_skb()
584 unsigned int xdp_frags_size;
586 /* Intermediate layers must ensure that destructor_arg
587 * remains valid until skb destructor */
588 void * destructor_arg;
590 /* must be last field, see pskb_expand_head() */
591 skb_frag_t frags[MAX_SKB_FRAGS];
595 * DOC: dataref and headerless skbs
597 * Transport layers send out clones of payload skbs they hold for
598 * retransmissions. To allow lower layers of the stack to prepend their headers
599 * we split &skb_shared_info.dataref into two halves.
600 * The lower 16 bits count the overall number of references.
601 * The higher 16 bits indicate how many of the references are payload-only.
602 * skb_header_cloned() checks if skb is allowed to add / write the headers.
604 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
605 * (via __skb_header_release()). Any clone created from marked skb will get
606 * &sk_buff.hdr_len populated with the available headroom.
607 * If there's the only clone in existence it's able to modify the headroom
608 * at will. The sequence of calls inside the transport layer is::
612 * __skb_header_release()
614 * // send the clone down the stack
616 * This is not a very generic construct and it depends on the transport layers
617 * doing the right thing. In practice there's usually only one payload-only skb.
618 * Having multiple payload-only skbs with different lengths of hdr_len is not
619 * possible. The payload-only skbs should never leave their owner.
621 #define SKB_DATAREF_SHIFT 16
622 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
626 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
627 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
628 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
632 SKB_GSO_TCPV4 = 1 << 0,
634 /* This indicates the skb is from an untrusted source. */
635 SKB_GSO_DODGY = 1 << 1,
637 /* This indicates the tcp segment has CWR set. */
638 SKB_GSO_TCP_ECN = 1 << 2,
640 SKB_GSO_TCP_FIXEDID = 1 << 3,
642 SKB_GSO_TCPV6 = 1 << 4,
644 SKB_GSO_FCOE = 1 << 5,
646 SKB_GSO_GRE = 1 << 6,
648 SKB_GSO_GRE_CSUM = 1 << 7,
650 SKB_GSO_IPXIP4 = 1 << 8,
652 SKB_GSO_IPXIP6 = 1 << 9,
654 SKB_GSO_UDP_TUNNEL = 1 << 10,
656 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
658 SKB_GSO_PARTIAL = 1 << 12,
660 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
662 SKB_GSO_SCTP = 1 << 14,
664 SKB_GSO_ESP = 1 << 15,
666 SKB_GSO_UDP = 1 << 16,
668 SKB_GSO_UDP_L4 = 1 << 17,
670 SKB_GSO_FRAGLIST = 1 << 18,
673 #if BITS_PER_LONG > 32
674 #define NET_SKBUFF_DATA_USES_OFFSET 1
677 #ifdef NET_SKBUFF_DATA_USES_OFFSET
678 typedef unsigned int sk_buff_data_t;
680 typedef unsigned char *sk_buff_data_t;
684 * DOC: Basic sk_buff geometry
686 * struct sk_buff itself is a metadata structure and does not hold any packet
687 * data. All the data is held in associated buffers.
689 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
692 * - data buffer, containing headers and sometimes payload;
693 * this is the part of the skb operated on by the common helpers
694 * such as skb_put() or skb_pull();
695 * - shared info (struct skb_shared_info) which holds an array of pointers
696 * to read-only data in the (page, offset, length) format.
698 * Optionally &skb_shared_info.frag_list may point to another skb.
700 * Basic diagram may look like this::
705 * ,--------------------------- + head
706 * / ,----------------- + data
707 * / / ,----------- + tail
711 * -----------------------------------------------
712 * | headroom | data | tailroom | skb_shared_info |
713 * -----------------------------------------------
717 * + [page frag] ---------
718 * + frag_list --> | sk_buff |
724 * struct sk_buff - socket buffer
725 * @next: Next buffer in list
726 * @prev: Previous buffer in list
727 * @tstamp: Time we arrived/left
728 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
729 * for retransmit timer
730 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
732 * @ll_node: anchor in an llist (eg socket defer_list)
733 * @sk: Socket we are owned by
734 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
735 * fragmentation management
736 * @dev: Device we arrived on/are leaving by
737 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
738 * @cb: Control buffer. Free for use by every layer. Put private vars here
739 * @_skb_refdst: destination entry (with norefcount bit)
740 * @sp: the security path, used for xfrm
741 * @len: Length of actual data
742 * @data_len: Data length
743 * @mac_len: Length of link layer header
744 * @hdr_len: writable header length of cloned skb
745 * @csum: Checksum (must include start/offset pair)
746 * @csum_start: Offset from skb->head where checksumming should start
747 * @csum_offset: Offset from csum_start where checksum should be stored
748 * @priority: Packet queueing priority
749 * @ignore_df: allow local fragmentation
750 * @cloned: Head may be cloned (check refcnt to be sure)
751 * @ip_summed: Driver fed us an IP checksum
752 * @nohdr: Payload reference only, must not modify header
753 * @pkt_type: Packet class
754 * @fclone: skbuff clone status
755 * @ipvs_property: skbuff is owned by ipvs
756 * @inner_protocol_type: whether the inner protocol is
757 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
758 * @remcsum_offload: remote checksum offload is enabled
759 * @offload_fwd_mark: Packet was L2-forwarded in hardware
760 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
761 * @tc_skip_classify: do not classify packet. set by IFB device
762 * @tc_at_ingress: used within tc_classify to distinguish in/egress
763 * @redirected: packet was redirected by packet classifier
764 * @from_ingress: packet was redirected from the ingress path
765 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
766 * @peeked: this packet has been seen already, so stats have been
767 * done for it, don't do them again
768 * @nf_trace: netfilter packet trace flag
769 * @protocol: Packet protocol from driver
770 * @destructor: Destruct function
771 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
772 * @_sk_redir: socket redirection information for skmsg
773 * @_nfct: Associated connection, if any (with nfctinfo bits)
774 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
775 * @skb_iif: ifindex of device we arrived on
776 * @tc_index: Traffic control index
777 * @hash: the packet hash
778 * @queue_mapping: Queue mapping for multiqueue devices
779 * @head_frag: skb was allocated from page fragments,
780 * not allocated by kmalloc() or vmalloc().
781 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
782 * @pp_recycle: mark the packet for recycling instead of freeing (implies
783 * page_pool support on driver)
784 * @active_extensions: active extensions (skb_ext_id types)
785 * @ndisc_nodetype: router type (from link layer)
786 * @ooo_okay: allow the mapping of a socket to a queue to be changed
787 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
789 * @sw_hash: indicates hash was computed in software stack
790 * @wifi_acked_valid: wifi_acked was set
791 * @wifi_acked: whether frame was acked on wifi or not
792 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
793 * @encapsulation: indicates the inner headers in the skbuff are valid
794 * @encap_hdr_csum: software checksum is needed
795 * @csum_valid: checksum is already valid
796 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
797 * @csum_complete_sw: checksum was completed by software
798 * @csum_level: indicates the number of consecutive checksums found in
799 * the packet minus one that have been verified as
800 * CHECKSUM_UNNECESSARY (max 3)
801 * @scm_io_uring: SKB holds io_uring registered files
802 * @dst_pending_confirm: need to confirm neighbour
803 * @decrypted: Decrypted SKB
804 * @slow_gro: state present at GRO time, slower prepare step required
805 * @mono_delivery_time: When set, skb->tstamp has the
806 * delivery_time in mono clock base (i.e. EDT). Otherwise, the
807 * skb->tstamp has the (rcv) timestamp at ingress and
808 * delivery_time at egress.
809 * @napi_id: id of the NAPI struct this skb came from
810 * @sender_cpu: (aka @napi_id) source CPU in XPS
811 * @alloc_cpu: CPU which did the skb allocation.
812 * @secmark: security marking
813 * @mark: Generic packet mark
814 * @reserved_tailroom: (aka @mark) number of bytes of free space available
815 * at the tail of an sk_buff
816 * @vlan_all: vlan fields (proto & tci)
817 * @vlan_proto: vlan encapsulation protocol
818 * @vlan_tci: vlan tag control information
819 * @inner_protocol: Protocol (encapsulation)
820 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
821 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
822 * @inner_transport_header: Inner transport layer header (encapsulation)
823 * @inner_network_header: Network layer header (encapsulation)
824 * @inner_mac_header: Link layer header (encapsulation)
825 * @transport_header: Transport layer header
826 * @network_header: Network layer header
827 * @mac_header: Link layer header
828 * @kcov_handle: KCOV remote handle for remote coverage collection
829 * @tail: Tail pointer
831 * @head: Head of buffer
832 * @data: Data head pointer
833 * @truesize: Buffer size
834 * @users: User count - see {datagram,tcp}.c
835 * @extensions: allocated extensions, valid if active_extensions is nonzero
841 /* These two members must be first to match sk_buff_head. */
842 struct sk_buff *next;
843 struct sk_buff *prev;
846 struct net_device *dev;
847 /* Some protocols might use this space to store information,
848 * while device pointer would be NULL.
849 * UDP receive path is one user.
851 unsigned long dev_scratch;
854 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
855 struct list_head list;
856 struct llist_node ll_node;
861 int ip_defrag_offset;
866 u64 skb_mstamp_ns; /* earliest departure time */
869 * This is the control buffer. It is free to use for every
870 * layer. Please put your private variables there. If you
871 * want to keep them across layers you have to do a skb_clone()
872 * first. This is owned by whoever has the skb queued ATM.
874 char cb[48] __aligned(8);
878 unsigned long _skb_refdst;
879 void (*destructor)(struct sk_buff *skb);
881 struct list_head tcp_tsorted_anchor;
882 #ifdef CONFIG_NET_SOCK_MSG
883 unsigned long _sk_redir;
887 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
895 /* Following fields are _not_ copied in __copy_skb_header()
896 * Note that queue_mapping is here mostly to fill a hole.
900 /* if you move cloned around you also must adapt those constants */
901 #ifdef __BIG_ENDIAN_BITFIELD
902 #define CLONED_MASK (1 << 7)
904 #define CLONED_MASK 1
906 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
909 __u8 __cloned_offset[0];
917 pp_recycle:1; /* page_pool recycle indicator */
918 #ifdef CONFIG_SKB_EXTENSIONS
919 __u8 active_extensions;
922 /* Fields enclosed in headers group are copied
923 * using a single memcpy() in __copy_skb_header()
925 struct_group(headers,
928 __u8 __pkt_type_offset[0];
930 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
938 __u8 wifi_acked_valid:1;
941 /* Indicates the inner headers are valid in the skbuff. */
942 __u8 encapsulation:1;
943 __u8 encap_hdr_csum:1;
947 __u8 __pkt_vlan_present_offset[0];
949 __u8 remcsum_offload:1;
950 __u8 csum_complete_sw:1;
952 __u8 dst_pending_confirm:1;
953 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */
954 #ifdef CONFIG_NET_CLS_ACT
955 __u8 tc_skip_classify:1;
956 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
958 #ifdef CONFIG_IPV6_NDISC_NODETYPE
959 __u8 ndisc_nodetype:2;
962 __u8 ipvs_property:1;
963 __u8 inner_protocol_type:1;
964 #ifdef CONFIG_NET_SWITCHDEV
965 __u8 offload_fwd_mark:1;
966 __u8 offload_l3_fwd_mark:1;
969 #ifdef CONFIG_NET_REDIRECT
972 #ifdef CONFIG_NETFILTER_SKIP_EGRESS
973 __u8 nf_skip_egress:1;
975 #ifdef CONFIG_TLS_DEVICE
979 __u8 csum_not_inet:1;
982 #ifdef CONFIG_NET_SCHED
983 __u16 tc_index; /* traffic control index */
1003 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
1005 unsigned int napi_id;
1006 unsigned int sender_cpu;
1010 #ifdef CONFIG_NETWORK_SECMARK
1016 __u32 reserved_tailroom;
1020 __be16 inner_protocol;
1024 __u16 inner_transport_header;
1025 __u16 inner_network_header;
1026 __u16 inner_mac_header;
1029 __u16 transport_header;
1030 __u16 network_header;
1037 ); /* end headers group */
1039 /* These elements must be at the end, see alloc_skb() for details. */
1040 sk_buff_data_t tail;
1042 unsigned char *head,
1044 unsigned int truesize;
1047 #ifdef CONFIG_SKB_EXTENSIONS
1048 /* only useable after checking ->active_extensions != 0 */
1049 struct skb_ext *extensions;
1053 /* if you move pkt_type around you also must adapt those constants */
1054 #ifdef __BIG_ENDIAN_BITFIELD
1055 #define PKT_TYPE_MAX (7 << 5)
1057 #define PKT_TYPE_MAX 7
1059 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1061 /* if you move tc_at_ingress or mono_delivery_time
1062 * around, you also must adapt these constants.
1064 #ifdef __BIG_ENDIAN_BITFIELD
1065 #define TC_AT_INGRESS_MASK (1 << 0)
1066 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 2)
1068 #define TC_AT_INGRESS_MASK (1 << 7)
1069 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 5)
1071 #define PKT_VLAN_PRESENT_OFFSET offsetof(struct sk_buff, __pkt_vlan_present_offset)
1075 * Handling routines are only of interest to the kernel
1078 #define SKB_ALLOC_FCLONE 0x01
1079 #define SKB_ALLOC_RX 0x02
1080 #define SKB_ALLOC_NAPI 0x04
1083 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1086 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1088 return unlikely(skb->pfmemalloc);
1092 * skb might have a dst pointer attached, refcounted or not.
1093 * _skb_refdst low order bit is set if refcount was _not_ taken
1095 #define SKB_DST_NOREF 1UL
1096 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1099 * skb_dst - returns skb dst_entry
1102 * Returns skb dst_entry, regardless of reference taken or not.
1104 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1106 /* If refdst was not refcounted, check we still are in a
1107 * rcu_read_lock section
1109 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1110 !rcu_read_lock_held() &&
1111 !rcu_read_lock_bh_held());
1112 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1116 * skb_dst_set - sets skb dst
1120 * Sets skb dst, assuming a reference was taken on dst and should
1121 * be released by skb_dst_drop()
1123 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1125 skb->slow_gro |= !!dst;
1126 skb->_skb_refdst = (unsigned long)dst;
1130 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1134 * Sets skb dst, assuming a reference was not taken on dst.
1135 * If dst entry is cached, we do not take reference and dst_release
1136 * will be avoided by refdst_drop. If dst entry is not cached, we take
1137 * reference, so that last dst_release can destroy the dst immediately.
1139 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1141 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1142 skb->slow_gro |= !!dst;
1143 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1147 * skb_dst_is_noref - Test if skb dst isn't refcounted
1150 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1152 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1156 * skb_rtable - Returns the skb &rtable
1159 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1161 return (struct rtable *)skb_dst(skb);
1164 /* For mangling skb->pkt_type from user space side from applications
1165 * such as nft, tc, etc, we only allow a conservative subset of
1166 * possible pkt_types to be set.
1168 static inline bool skb_pkt_type_ok(u32 ptype)
1170 return ptype <= PACKET_OTHERHOST;
1174 * skb_napi_id - Returns the skb's NAPI id
1177 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1179 #ifdef CONFIG_NET_RX_BUSY_POLL
1180 return skb->napi_id;
1187 * skb_unref - decrement the skb's reference count
1190 * Returns true if we can free the skb.
1192 static inline bool skb_unref(struct sk_buff *skb)
1196 if (likely(refcount_read(&skb->users) == 1))
1198 else if (likely(!refcount_dec_and_test(&skb->users)))
1205 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason);
1208 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1209 * @skb: buffer to free
1211 static inline void kfree_skb(struct sk_buff *skb)
1213 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1216 void skb_release_head_state(struct sk_buff *skb);
1217 void kfree_skb_list_reason(struct sk_buff *segs,
1218 enum skb_drop_reason reason);
1219 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1220 void skb_tx_error(struct sk_buff *skb);
1222 static inline void kfree_skb_list(struct sk_buff *segs)
1224 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1227 #ifdef CONFIG_TRACEPOINTS
1228 void consume_skb(struct sk_buff *skb);
1230 static inline void consume_skb(struct sk_buff *skb)
1232 return kfree_skb(skb);
1236 void __consume_stateless_skb(struct sk_buff *skb);
1237 void __kfree_skb(struct sk_buff *skb);
1238 extern struct kmem_cache *skbuff_head_cache;
1240 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1241 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1242 bool *fragstolen, int *delta_truesize);
1244 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1246 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1247 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1248 struct sk_buff *build_skb_around(struct sk_buff *skb,
1249 void *data, unsigned int frag_size);
1250 void skb_attempt_defer_free(struct sk_buff *skb);
1252 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1253 struct sk_buff *slab_build_skb(void *data);
1256 * alloc_skb - allocate a network buffer
1257 * @size: size to allocate
1258 * @priority: allocation mask
1260 * This function is a convenient wrapper around __alloc_skb().
1262 static inline struct sk_buff *alloc_skb(unsigned int size,
1265 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1268 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1269 unsigned long data_len,
1273 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1275 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1276 struct sk_buff_fclones {
1277 struct sk_buff skb1;
1279 struct sk_buff skb2;
1281 refcount_t fclone_ref;
1285 * skb_fclone_busy - check if fclone is busy
1289 * Returns true if skb is a fast clone, and its clone is not freed.
1290 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1291 * so we also check that this didnt happen.
1293 static inline bool skb_fclone_busy(const struct sock *sk,
1294 const struct sk_buff *skb)
1296 const struct sk_buff_fclones *fclones;
1298 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1300 return skb->fclone == SKB_FCLONE_ORIG &&
1301 refcount_read(&fclones->fclone_ref) > 1 &&
1302 READ_ONCE(fclones->skb2.sk) == sk;
1306 * alloc_skb_fclone - allocate a network buffer from fclone cache
1307 * @size: size to allocate
1308 * @priority: allocation mask
1310 * This function is a convenient wrapper around __alloc_skb().
1312 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1315 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1318 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1319 void skb_headers_offset_update(struct sk_buff *skb, int off);
1320 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1321 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1322 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1323 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1324 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1325 gfp_t gfp_mask, bool fclone);
1326 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1329 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1332 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1333 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1334 unsigned int headroom);
1335 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1336 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1337 int newtailroom, gfp_t priority);
1338 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1339 int offset, int len);
1340 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1341 int offset, int len);
1342 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1343 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1346 * skb_pad - zero pad the tail of an skb
1347 * @skb: buffer to pad
1348 * @pad: space to pad
1350 * Ensure that a buffer is followed by a padding area that is zero
1351 * filled. Used by network drivers which may DMA or transfer data
1352 * beyond the buffer end onto the wire.
1354 * May return error in out of memory cases. The skb is freed on error.
1356 static inline int skb_pad(struct sk_buff *skb, int pad)
1358 return __skb_pad(skb, pad, true);
1360 #define dev_kfree_skb(a) consume_skb(a)
1362 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1363 int offset, size_t size);
1365 struct skb_seq_state {
1369 __u32 stepped_offset;
1370 struct sk_buff *root_skb;
1371 struct sk_buff *cur_skb;
1376 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1377 unsigned int to, struct skb_seq_state *st);
1378 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1379 struct skb_seq_state *st);
1380 void skb_abort_seq_read(struct skb_seq_state *st);
1382 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1383 unsigned int to, struct ts_config *config);
1386 * Packet hash types specify the type of hash in skb_set_hash.
1388 * Hash types refer to the protocol layer addresses which are used to
1389 * construct a packet's hash. The hashes are used to differentiate or identify
1390 * flows of the protocol layer for the hash type. Hash types are either
1391 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1393 * Properties of hashes:
1395 * 1) Two packets in different flows have different hash values
1396 * 2) Two packets in the same flow should have the same hash value
1398 * A hash at a higher layer is considered to be more specific. A driver should
1399 * set the most specific hash possible.
1401 * A driver cannot indicate a more specific hash than the layer at which a hash
1402 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1404 * A driver may indicate a hash level which is less specific than the
1405 * actual layer the hash was computed on. For instance, a hash computed
1406 * at L4 may be considered an L3 hash. This should only be done if the
1407 * driver can't unambiguously determine that the HW computed the hash at
1408 * the higher layer. Note that the "should" in the second property above
1411 enum pkt_hash_types {
1412 PKT_HASH_TYPE_NONE, /* Undefined type */
1413 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1414 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1415 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1418 static inline void skb_clear_hash(struct sk_buff *skb)
1425 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1428 skb_clear_hash(skb);
1432 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1434 skb->l4_hash = is_l4;
1435 skb->sw_hash = is_sw;
1440 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1442 /* Used by drivers to set hash from HW */
1443 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1447 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1449 __skb_set_hash(skb, hash, true, is_l4);
1452 void __skb_get_hash(struct sk_buff *skb);
1453 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1454 u32 skb_get_poff(const struct sk_buff *skb);
1455 u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1456 const struct flow_keys_basic *keys, int hlen);
1457 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1458 const void *data, int hlen_proto);
1460 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1461 int thoff, u8 ip_proto)
1463 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1466 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1467 const struct flow_dissector_key *key,
1468 unsigned int key_count);
1470 struct bpf_flow_dissector;
1471 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1472 __be16 proto, int nhoff, int hlen, unsigned int flags);
1474 bool __skb_flow_dissect(const struct net *net,
1475 const struct sk_buff *skb,
1476 struct flow_dissector *flow_dissector,
1477 void *target_container, const void *data,
1478 __be16 proto, int nhoff, int hlen, unsigned int flags);
1480 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1481 struct flow_dissector *flow_dissector,
1482 void *target_container, unsigned int flags)
1484 return __skb_flow_dissect(NULL, skb, flow_dissector,
1485 target_container, NULL, 0, 0, 0, flags);
1488 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1489 struct flow_keys *flow,
1492 memset(flow, 0, sizeof(*flow));
1493 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1494 flow, NULL, 0, 0, 0, flags);
1498 skb_flow_dissect_flow_keys_basic(const struct net *net,
1499 const struct sk_buff *skb,
1500 struct flow_keys_basic *flow,
1501 const void *data, __be16 proto,
1502 int nhoff, int hlen, unsigned int flags)
1504 memset(flow, 0, sizeof(*flow));
1505 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1506 data, proto, nhoff, hlen, flags);
1509 void skb_flow_dissect_meta(const struct sk_buff *skb,
1510 struct flow_dissector *flow_dissector,
1511 void *target_container);
1513 /* Gets a skb connection tracking info, ctinfo map should be a
1514 * map of mapsize to translate enum ip_conntrack_info states
1518 skb_flow_dissect_ct(const struct sk_buff *skb,
1519 struct flow_dissector *flow_dissector,
1520 void *target_container,
1521 u16 *ctinfo_map, size_t mapsize,
1522 bool post_ct, u16 zone);
1524 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1525 struct flow_dissector *flow_dissector,
1526 void *target_container);
1528 void skb_flow_dissect_hash(const struct sk_buff *skb,
1529 struct flow_dissector *flow_dissector,
1530 void *target_container);
1532 static inline __u32 skb_get_hash(struct sk_buff *skb)
1534 if (!skb->l4_hash && !skb->sw_hash)
1535 __skb_get_hash(skb);
1540 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1542 if (!skb->l4_hash && !skb->sw_hash) {
1543 struct flow_keys keys;
1544 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1546 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1552 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1553 const siphash_key_t *perturb);
1555 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1560 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1562 to->hash = from->hash;
1563 to->sw_hash = from->sw_hash;
1564 to->l4_hash = from->l4_hash;
1567 static inline void skb_copy_decrypted(struct sk_buff *to,
1568 const struct sk_buff *from)
1570 #ifdef CONFIG_TLS_DEVICE
1571 to->decrypted = from->decrypted;
1575 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1576 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1578 return skb->head + skb->end;
1581 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1586 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1591 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1596 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1598 return skb->end - skb->head;
1601 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1603 skb->end = skb->head + offset;
1607 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1608 struct ubuf_info *uarg);
1610 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1612 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1615 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
1616 struct sk_buff *skb, struct iov_iter *from,
1619 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1620 struct msghdr *msg, int len)
1622 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len);
1625 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1626 struct msghdr *msg, int len,
1627 struct ubuf_info *uarg);
1630 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1632 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1634 return &skb_shinfo(skb)->hwtstamps;
1637 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1639 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1641 return is_zcopy ? skb_uarg(skb) : NULL;
1644 static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1646 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1649 static inline bool skb_zcopy_managed(const struct sk_buff *skb)
1651 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS;
1654 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1655 const struct sk_buff *skb2)
1657 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1660 static inline void net_zcopy_get(struct ubuf_info *uarg)
1662 refcount_inc(&uarg->refcnt);
1665 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1667 skb_shinfo(skb)->destructor_arg = uarg;
1668 skb_shinfo(skb)->flags |= uarg->flags;
1671 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1674 if (skb && uarg && !skb_zcopy(skb)) {
1675 if (unlikely(have_ref && *have_ref))
1678 net_zcopy_get(uarg);
1679 skb_zcopy_init(skb, uarg);
1683 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1685 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1686 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1689 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1691 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1694 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1696 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1699 static inline void net_zcopy_put(struct ubuf_info *uarg)
1702 uarg->callback(NULL, uarg, true);
1705 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1708 if (uarg->callback == msg_zerocopy_callback)
1709 msg_zerocopy_put_abort(uarg, have_uref);
1711 net_zcopy_put(uarg);
1715 /* Release a reference on a zerocopy structure */
1716 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1718 struct ubuf_info *uarg = skb_zcopy(skb);
1721 if (!skb_zcopy_is_nouarg(skb))
1722 uarg->callback(skb, uarg, zerocopy_success);
1724 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1728 void __skb_zcopy_downgrade_managed(struct sk_buff *skb);
1730 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb)
1732 if (unlikely(skb_zcopy_managed(skb)))
1733 __skb_zcopy_downgrade_managed(skb);
1736 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1741 /* Iterate through singly-linked GSO fragments of an skb. */
1742 #define skb_list_walk_safe(first, skb, next_skb) \
1743 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1744 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1746 static inline void skb_list_del_init(struct sk_buff *skb)
1748 __list_del_entry(&skb->list);
1749 skb_mark_not_on_list(skb);
1753 * skb_queue_empty - check if a queue is empty
1756 * Returns true if the queue is empty, false otherwise.
1758 static inline int skb_queue_empty(const struct sk_buff_head *list)
1760 return list->next == (const struct sk_buff *) list;
1764 * skb_queue_empty_lockless - check if a queue is empty
1767 * Returns true if the queue is empty, false otherwise.
1768 * This variant can be used in lockless contexts.
1770 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1772 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1777 * skb_queue_is_last - check if skb is the last entry in the queue
1781 * Returns true if @skb is the last buffer on the list.
1783 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1784 const struct sk_buff *skb)
1786 return skb->next == (const struct sk_buff *) list;
1790 * skb_queue_is_first - check if skb is the first entry in the queue
1794 * Returns true if @skb is the first buffer on the list.
1796 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1797 const struct sk_buff *skb)
1799 return skb->prev == (const struct sk_buff *) list;
1803 * skb_queue_next - return the next packet in the queue
1805 * @skb: current buffer
1807 * Return the next packet in @list after @skb. It is only valid to
1808 * call this if skb_queue_is_last() evaluates to false.
1810 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1811 const struct sk_buff *skb)
1813 /* This BUG_ON may seem severe, but if we just return then we
1814 * are going to dereference garbage.
1816 BUG_ON(skb_queue_is_last(list, skb));
1821 * skb_queue_prev - return the prev packet in the queue
1823 * @skb: current buffer
1825 * Return the prev packet in @list before @skb. It is only valid to
1826 * call this if skb_queue_is_first() evaluates to false.
1828 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1829 const struct sk_buff *skb)
1831 /* This BUG_ON may seem severe, but if we just return then we
1832 * are going to dereference garbage.
1834 BUG_ON(skb_queue_is_first(list, skb));
1839 * skb_get - reference buffer
1840 * @skb: buffer to reference
1842 * Makes another reference to a socket buffer and returns a pointer
1845 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1847 refcount_inc(&skb->users);
1852 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1856 * skb_cloned - is the buffer a clone
1857 * @skb: buffer to check
1859 * Returns true if the buffer was generated with skb_clone() and is
1860 * one of multiple shared copies of the buffer. Cloned buffers are
1861 * shared data so must not be written to under normal circumstances.
1863 static inline int skb_cloned(const struct sk_buff *skb)
1865 return skb->cloned &&
1866 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1869 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1871 might_sleep_if(gfpflags_allow_blocking(pri));
1873 if (skb_cloned(skb))
1874 return pskb_expand_head(skb, 0, 0, pri);
1879 /* This variant of skb_unclone() makes sure skb->truesize
1880 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
1882 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
1883 * when various debugging features are in place.
1885 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
1886 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1888 might_sleep_if(gfpflags_allow_blocking(pri));
1890 if (skb_cloned(skb))
1891 return __skb_unclone_keeptruesize(skb, pri);
1896 * skb_header_cloned - is the header a clone
1897 * @skb: buffer to check
1899 * Returns true if modifying the header part of the buffer requires
1900 * the data to be copied.
1902 static inline int skb_header_cloned(const struct sk_buff *skb)
1909 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1910 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1911 return dataref != 1;
1914 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1916 might_sleep_if(gfpflags_allow_blocking(pri));
1918 if (skb_header_cloned(skb))
1919 return pskb_expand_head(skb, 0, 0, pri);
1925 * __skb_header_release() - allow clones to use the headroom
1926 * @skb: buffer to operate on
1928 * See "DOC: dataref and headerless skbs".
1930 static inline void __skb_header_release(struct sk_buff *skb)
1933 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1938 * skb_shared - is the buffer shared
1939 * @skb: buffer to check
1941 * Returns true if more than one person has a reference to this
1944 static inline int skb_shared(const struct sk_buff *skb)
1946 return refcount_read(&skb->users) != 1;
1950 * skb_share_check - check if buffer is shared and if so clone it
1951 * @skb: buffer to check
1952 * @pri: priority for memory allocation
1954 * If the buffer is shared the buffer is cloned and the old copy
1955 * drops a reference. A new clone with a single reference is returned.
1956 * If the buffer is not shared the original buffer is returned. When
1957 * being called from interrupt status or with spinlocks held pri must
1960 * NULL is returned on a memory allocation failure.
1962 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1964 might_sleep_if(gfpflags_allow_blocking(pri));
1965 if (skb_shared(skb)) {
1966 struct sk_buff *nskb = skb_clone(skb, pri);
1978 * Copy shared buffers into a new sk_buff. We effectively do COW on
1979 * packets to handle cases where we have a local reader and forward
1980 * and a couple of other messy ones. The normal one is tcpdumping
1981 * a packet thats being forwarded.
1985 * skb_unshare - make a copy of a shared buffer
1986 * @skb: buffer to check
1987 * @pri: priority for memory allocation
1989 * If the socket buffer is a clone then this function creates a new
1990 * copy of the data, drops a reference count on the old copy and returns
1991 * the new copy with the reference count at 1. If the buffer is not a clone
1992 * the original buffer is returned. When called with a spinlock held or
1993 * from interrupt state @pri must be %GFP_ATOMIC
1995 * %NULL is returned on a memory allocation failure.
1997 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
2000 might_sleep_if(gfpflags_allow_blocking(pri));
2001 if (skb_cloned(skb)) {
2002 struct sk_buff *nskb = skb_copy(skb, pri);
2004 /* Free our shared copy */
2015 * skb_peek - peek at the head of an &sk_buff_head
2016 * @list_: list to peek at
2018 * Peek an &sk_buff. Unlike most other operations you _MUST_
2019 * be careful with this one. A peek leaves the buffer on the
2020 * list and someone else may run off with it. You must hold
2021 * the appropriate locks or have a private queue to do this.
2023 * Returns %NULL for an empty list or a pointer to the head element.
2024 * The reference count is not incremented and the reference is therefore
2025 * volatile. Use with caution.
2027 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2029 struct sk_buff *skb = list_->next;
2031 if (skb == (struct sk_buff *)list_)
2037 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2038 * @list_: list to peek at
2040 * Like skb_peek(), but the caller knows that the list is not empty.
2042 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2048 * skb_peek_next - peek skb following the given one from a queue
2049 * @skb: skb to start from
2050 * @list_: list to peek at
2052 * Returns %NULL when the end of the list is met or a pointer to the
2053 * next element. The reference count is not incremented and the
2054 * reference is therefore volatile. Use with caution.
2056 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2057 const struct sk_buff_head *list_)
2059 struct sk_buff *next = skb->next;
2061 if (next == (struct sk_buff *)list_)
2067 * skb_peek_tail - peek at the tail of an &sk_buff_head
2068 * @list_: list to peek at
2070 * Peek an &sk_buff. Unlike most other operations you _MUST_
2071 * be careful with this one. A peek leaves the buffer on the
2072 * list and someone else may run off with it. You must hold
2073 * the appropriate locks or have a private queue to do this.
2075 * Returns %NULL for an empty list or a pointer to the tail element.
2076 * The reference count is not incremented and the reference is therefore
2077 * volatile. Use with caution.
2079 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2081 struct sk_buff *skb = READ_ONCE(list_->prev);
2083 if (skb == (struct sk_buff *)list_)
2090 * skb_queue_len - get queue length
2091 * @list_: list to measure
2093 * Return the length of an &sk_buff queue.
2095 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2101 * skb_queue_len_lockless - get queue length
2102 * @list_: list to measure
2104 * Return the length of an &sk_buff queue.
2105 * This variant can be used in lockless contexts.
2107 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2109 return READ_ONCE(list_->qlen);
2113 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2114 * @list: queue to initialize
2116 * This initializes only the list and queue length aspects of
2117 * an sk_buff_head object. This allows to initialize the list
2118 * aspects of an sk_buff_head without reinitializing things like
2119 * the spinlock. It can also be used for on-stack sk_buff_head
2120 * objects where the spinlock is known to not be used.
2122 static inline void __skb_queue_head_init(struct sk_buff_head *list)
2124 list->prev = list->next = (struct sk_buff *)list;
2129 * This function creates a split out lock class for each invocation;
2130 * this is needed for now since a whole lot of users of the skb-queue
2131 * infrastructure in drivers have different locking usage (in hardirq)
2132 * than the networking core (in softirq only). In the long run either the
2133 * network layer or drivers should need annotation to consolidate the
2134 * main types of usage into 3 classes.
2136 static inline void skb_queue_head_init(struct sk_buff_head *list)
2138 spin_lock_init(&list->lock);
2139 __skb_queue_head_init(list);
2142 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2143 struct lock_class_key *class)
2145 skb_queue_head_init(list);
2146 lockdep_set_class(&list->lock, class);
2150 * Insert an sk_buff on a list.
2152 * The "__skb_xxxx()" functions are the non-atomic ones that
2153 * can only be called with interrupts disabled.
2155 static inline void __skb_insert(struct sk_buff *newsk,
2156 struct sk_buff *prev, struct sk_buff *next,
2157 struct sk_buff_head *list)
2159 /* See skb_queue_empty_lockless() and skb_peek_tail()
2160 * for the opposite READ_ONCE()
2162 WRITE_ONCE(newsk->next, next);
2163 WRITE_ONCE(newsk->prev, prev);
2164 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2165 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2166 WRITE_ONCE(list->qlen, list->qlen + 1);
2169 static inline void __skb_queue_splice(const struct sk_buff_head *list,
2170 struct sk_buff *prev,
2171 struct sk_buff *next)
2173 struct sk_buff *first = list->next;
2174 struct sk_buff *last = list->prev;
2176 WRITE_ONCE(first->prev, prev);
2177 WRITE_ONCE(prev->next, first);
2179 WRITE_ONCE(last->next, next);
2180 WRITE_ONCE(next->prev, last);
2184 * skb_queue_splice - join two skb lists, this is designed for stacks
2185 * @list: the new list to add
2186 * @head: the place to add it in the first list
2188 static inline void skb_queue_splice(const struct sk_buff_head *list,
2189 struct sk_buff_head *head)
2191 if (!skb_queue_empty(list)) {
2192 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2193 head->qlen += list->qlen;
2198 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2199 * @list: the new list to add
2200 * @head: the place to add it in the first list
2202 * The list at @list is reinitialised
2204 static inline void skb_queue_splice_init(struct sk_buff_head *list,
2205 struct sk_buff_head *head)
2207 if (!skb_queue_empty(list)) {
2208 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2209 head->qlen += list->qlen;
2210 __skb_queue_head_init(list);
2215 * skb_queue_splice_tail - join two skb lists, each list being a queue
2216 * @list: the new list to add
2217 * @head: the place to add it in the first list
2219 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2220 struct sk_buff_head *head)
2222 if (!skb_queue_empty(list)) {
2223 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2224 head->qlen += list->qlen;
2229 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2230 * @list: the new list to add
2231 * @head: the place to add it in the first list
2233 * Each of the lists is a queue.
2234 * The list at @list is reinitialised
2236 static inline void skb_queue_splice_tail_init(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;
2242 __skb_queue_head_init(list);
2247 * __skb_queue_after - queue a buffer at the list head
2248 * @list: list to use
2249 * @prev: place after this buffer
2250 * @newsk: buffer to queue
2252 * Queue a buffer int the middle of a list. This function takes no locks
2253 * and you must therefore hold required locks before calling it.
2255 * A buffer cannot be placed on two lists at the same time.
2257 static inline void __skb_queue_after(struct sk_buff_head *list,
2258 struct sk_buff *prev,
2259 struct sk_buff *newsk)
2261 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2264 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2265 struct sk_buff_head *list);
2267 static inline void __skb_queue_before(struct sk_buff_head *list,
2268 struct sk_buff *next,
2269 struct sk_buff *newsk)
2271 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2275 * __skb_queue_head - queue a buffer at the list head
2276 * @list: list to use
2277 * @newsk: buffer to queue
2279 * Queue a buffer at the start of a list. This function takes no locks
2280 * and you must therefore hold required locks before calling it.
2282 * A buffer cannot be placed on two lists at the same time.
2284 static inline void __skb_queue_head(struct sk_buff_head *list,
2285 struct sk_buff *newsk)
2287 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2289 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2292 * __skb_queue_tail - queue a buffer at the list tail
2293 * @list: list to use
2294 * @newsk: buffer to queue
2296 * Queue a buffer at the end 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_tail(struct sk_buff_head *list,
2302 struct sk_buff *newsk)
2304 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2306 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2309 * remove sk_buff from list. _Must_ be called atomically, and with
2312 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2313 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2315 struct sk_buff *next, *prev;
2317 WRITE_ONCE(list->qlen, list->qlen - 1);
2320 skb->next = skb->prev = NULL;
2321 WRITE_ONCE(next->prev, prev);
2322 WRITE_ONCE(prev->next, next);
2326 * __skb_dequeue - remove from the head of the queue
2327 * @list: list to dequeue from
2329 * Remove the head of the list. This function does not take any locks
2330 * so must be used with appropriate locks held only. The head item is
2331 * returned or %NULL if the list is empty.
2333 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2335 struct sk_buff *skb = skb_peek(list);
2337 __skb_unlink(skb, list);
2340 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2343 * __skb_dequeue_tail - remove from the tail of the queue
2344 * @list: list to dequeue from
2346 * Remove the tail of the list. This function does not take any locks
2347 * so must be used with appropriate locks held only. The tail item is
2348 * returned or %NULL if the list is empty.
2350 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2352 struct sk_buff *skb = skb_peek_tail(list);
2354 __skb_unlink(skb, list);
2357 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2360 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2362 return skb->data_len;
2365 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2367 return skb->len - skb->data_len;
2370 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2372 unsigned int i, len = 0;
2374 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2375 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2379 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2381 return skb_headlen(skb) + __skb_pagelen(skb);
2384 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo,
2385 int i, struct page *page,
2388 skb_frag_t *frag = &shinfo->frags[i];
2391 * Propagate page pfmemalloc to the skb if we can. The problem is
2392 * that not all callers have unique ownership of the page but rely
2393 * on page_is_pfmemalloc doing the right thing(tm).
2395 frag->bv_page = page;
2396 frag->bv_offset = off;
2397 skb_frag_size_set(frag, size);
2401 * skb_len_add - adds a number to len fields of skb
2402 * @skb: buffer to add len to
2403 * @delta: number of bytes to add
2405 static inline void skb_len_add(struct sk_buff *skb, int delta)
2408 skb->data_len += delta;
2409 skb->truesize += delta;
2413 * __skb_fill_page_desc - initialise a paged fragment in an skb
2414 * @skb: buffer containing fragment to be initialised
2415 * @i: paged fragment index to initialise
2416 * @page: the page to use for this fragment
2417 * @off: the offset to the data with @page
2418 * @size: the length of the data
2420 * Initialises the @i'th fragment of @skb to point to &size bytes at
2421 * offset @off within @page.
2423 * Does not take any additional reference on the fragment.
2425 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2426 struct page *page, int off, int size)
2428 __skb_fill_page_desc_noacc(skb_shinfo(skb), i, page, off, size);
2429 page = compound_head(page);
2430 if (page_is_pfmemalloc(page))
2431 skb->pfmemalloc = true;
2435 * skb_fill_page_desc - initialise a paged fragment in an skb
2436 * @skb: buffer containing fragment to be initialised
2437 * @i: paged fragment index to initialise
2438 * @page: the page to use for this fragment
2439 * @off: the offset to the data with @page
2440 * @size: the length of the data
2442 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2443 * @skb to point to @size bytes at offset @off within @page. In
2444 * addition updates @skb such that @i is the last fragment.
2446 * Does not take any additional reference on the fragment.
2448 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2449 struct page *page, int off, int size)
2451 __skb_fill_page_desc(skb, i, page, off, size);
2452 skb_shinfo(skb)->nr_frags = i + 1;
2456 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb
2457 * @skb: buffer containing fragment to be initialised
2458 * @i: paged fragment index to initialise
2459 * @page: the page to use for this fragment
2460 * @off: the offset to the data with @page
2461 * @size: the length of the data
2463 * Variant of skb_fill_page_desc() which does not deal with
2464 * pfmemalloc, if page is not owned by us.
2466 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i,
2467 struct page *page, int off,
2470 struct skb_shared_info *shinfo = skb_shinfo(skb);
2472 __skb_fill_page_desc_noacc(shinfo, i, page, off, size);
2473 shinfo->nr_frags = i + 1;
2476 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2477 int size, unsigned int truesize);
2479 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2480 unsigned int truesize);
2482 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2484 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2485 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2487 return skb->head + skb->tail;
2490 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2492 skb->tail = skb->data - skb->head;
2495 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2497 skb_reset_tail_pointer(skb);
2498 skb->tail += offset;
2501 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2502 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2507 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2509 skb->tail = skb->data;
2512 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2514 skb->tail = skb->data + offset;
2517 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2519 static inline void skb_assert_len(struct sk_buff *skb)
2521 #ifdef CONFIG_DEBUG_NET
2522 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2523 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2524 #endif /* CONFIG_DEBUG_NET */
2528 * Add data to an sk_buff
2530 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2531 void *skb_put(struct sk_buff *skb, unsigned int len);
2532 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2534 void *tmp = skb_tail_pointer(skb);
2535 SKB_LINEAR_ASSERT(skb);
2541 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2543 void *tmp = __skb_put(skb, len);
2545 memset(tmp, 0, len);
2549 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2552 void *tmp = __skb_put(skb, len);
2554 memcpy(tmp, data, len);
2558 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2560 *(u8 *)__skb_put(skb, 1) = val;
2563 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2565 void *tmp = skb_put(skb, len);
2567 memset(tmp, 0, len);
2572 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2575 void *tmp = skb_put(skb, len);
2577 memcpy(tmp, data, len);
2582 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2584 *(u8 *)skb_put(skb, 1) = val;
2587 void *skb_push(struct sk_buff *skb, unsigned int len);
2588 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2595 void *skb_pull(struct sk_buff *skb, unsigned int len);
2596 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2599 if (unlikely(skb->len < skb->data_len)) {
2600 #if defined(CONFIG_DEBUG_NET)
2602 pr_err("__skb_pull(len=%u)\n", len);
2603 skb_dump(KERN_ERR, skb, false);
2607 return skb->data += len;
2610 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2612 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2615 void *skb_pull_data(struct sk_buff *skb, size_t len);
2617 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2619 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2621 if (likely(len <= skb_headlen(skb)))
2623 if (unlikely(len > skb->len))
2625 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2628 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2630 if (!pskb_may_pull(skb, len))
2634 return skb->data += len;
2637 void skb_condense(struct sk_buff *skb);
2640 * skb_headroom - bytes at buffer head
2641 * @skb: buffer to check
2643 * Return the number of bytes of free space at the head of an &sk_buff.
2645 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2647 return skb->data - skb->head;
2651 * skb_tailroom - bytes at buffer end
2652 * @skb: buffer to check
2654 * Return the number of bytes of free space at the tail of an sk_buff
2656 static inline int skb_tailroom(const struct sk_buff *skb)
2658 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2662 * skb_availroom - bytes at buffer end
2663 * @skb: buffer to check
2665 * Return the number of bytes of free space at the tail of an sk_buff
2666 * allocated by sk_stream_alloc()
2668 static inline int skb_availroom(const struct sk_buff *skb)
2670 if (skb_is_nonlinear(skb))
2673 return skb->end - skb->tail - skb->reserved_tailroom;
2677 * skb_reserve - adjust headroom
2678 * @skb: buffer to alter
2679 * @len: bytes to move
2681 * Increase the headroom of an empty &sk_buff by reducing the tail
2682 * room. This is only allowed for an empty buffer.
2684 static inline void skb_reserve(struct sk_buff *skb, int len)
2691 * skb_tailroom_reserve - adjust reserved_tailroom
2692 * @skb: buffer to alter
2693 * @mtu: maximum amount of headlen permitted
2694 * @needed_tailroom: minimum amount of reserved_tailroom
2696 * Set reserved_tailroom so that headlen can be as large as possible but
2697 * not larger than mtu and tailroom cannot be smaller than
2699 * The required headroom should already have been reserved before using
2702 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2703 unsigned int needed_tailroom)
2705 SKB_LINEAR_ASSERT(skb);
2706 if (mtu < skb_tailroom(skb) - needed_tailroom)
2707 /* use at most mtu */
2708 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2710 /* use up to all available space */
2711 skb->reserved_tailroom = needed_tailroom;
2714 #define ENCAP_TYPE_ETHER 0
2715 #define ENCAP_TYPE_IPPROTO 1
2717 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2720 skb->inner_protocol = protocol;
2721 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2724 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2727 skb->inner_ipproto = ipproto;
2728 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2731 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2733 skb->inner_mac_header = skb->mac_header;
2734 skb->inner_network_header = skb->network_header;
2735 skb->inner_transport_header = skb->transport_header;
2738 static inline void skb_reset_mac_len(struct sk_buff *skb)
2740 skb->mac_len = skb->network_header - skb->mac_header;
2743 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2746 return skb->head + skb->inner_transport_header;
2749 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2751 return skb_inner_transport_header(skb) - skb->data;
2754 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2756 skb->inner_transport_header = skb->data - skb->head;
2759 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2762 skb_reset_inner_transport_header(skb);
2763 skb->inner_transport_header += offset;
2766 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2768 return skb->head + skb->inner_network_header;
2771 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2773 skb->inner_network_header = skb->data - skb->head;
2776 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2779 skb_reset_inner_network_header(skb);
2780 skb->inner_network_header += offset;
2783 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2785 return skb->head + skb->inner_mac_header;
2788 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2790 skb->inner_mac_header = skb->data - skb->head;
2793 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2796 skb_reset_inner_mac_header(skb);
2797 skb->inner_mac_header += offset;
2799 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2801 return skb->transport_header != (typeof(skb->transport_header))~0U;
2804 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2806 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
2807 return skb->head + skb->transport_header;
2810 static inline void skb_reset_transport_header(struct sk_buff *skb)
2812 skb->transport_header = skb->data - skb->head;
2815 static inline void skb_set_transport_header(struct sk_buff *skb,
2818 skb_reset_transport_header(skb);
2819 skb->transport_header += offset;
2822 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2824 return skb->head + skb->network_header;
2827 static inline void skb_reset_network_header(struct sk_buff *skb)
2829 skb->network_header = skb->data - skb->head;
2832 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2834 skb_reset_network_header(skb);
2835 skb->network_header += offset;
2838 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2840 return skb->mac_header != (typeof(skb->mac_header))~0U;
2843 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2845 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2846 return skb->head + skb->mac_header;
2849 static inline int skb_mac_offset(const struct sk_buff *skb)
2851 return skb_mac_header(skb) - skb->data;
2854 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2856 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2857 return skb->network_header - skb->mac_header;
2860 static inline void skb_unset_mac_header(struct sk_buff *skb)
2862 skb->mac_header = (typeof(skb->mac_header))~0U;
2865 static inline void skb_reset_mac_header(struct sk_buff *skb)
2867 skb->mac_header = skb->data - skb->head;
2870 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2872 skb_reset_mac_header(skb);
2873 skb->mac_header += offset;
2876 static inline void skb_pop_mac_header(struct sk_buff *skb)
2878 skb->mac_header = skb->network_header;
2881 static inline void skb_probe_transport_header(struct sk_buff *skb)
2883 struct flow_keys_basic keys;
2885 if (skb_transport_header_was_set(skb))
2888 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2890 skb_set_transport_header(skb, keys.control.thoff);
2893 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2895 if (skb_mac_header_was_set(skb)) {
2896 const unsigned char *old_mac = skb_mac_header(skb);
2898 skb_set_mac_header(skb, -skb->mac_len);
2899 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2903 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2905 return skb->csum_start - skb_headroom(skb);
2908 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2910 return skb->head + skb->csum_start;
2913 static inline int skb_transport_offset(const struct sk_buff *skb)
2915 return skb_transport_header(skb) - skb->data;
2918 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2920 return skb->transport_header - skb->network_header;
2923 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2925 return skb->inner_transport_header - skb->inner_network_header;
2928 static inline int skb_network_offset(const struct sk_buff *skb)
2930 return skb_network_header(skb) - skb->data;
2933 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2935 return skb_inner_network_header(skb) - skb->data;
2938 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2940 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2944 * CPUs often take a performance hit when accessing unaligned memory
2945 * locations. The actual performance hit varies, it can be small if the
2946 * hardware handles it or large if we have to take an exception and fix it
2949 * Since an ethernet header is 14 bytes network drivers often end up with
2950 * the IP header at an unaligned offset. The IP header can be aligned by
2951 * shifting the start of the packet by 2 bytes. Drivers should do this
2954 * skb_reserve(skb, NET_IP_ALIGN);
2956 * The downside to this alignment of the IP header is that the DMA is now
2957 * unaligned. On some architectures the cost of an unaligned DMA is high
2958 * and this cost outweighs the gains made by aligning the IP header.
2960 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2963 #ifndef NET_IP_ALIGN
2964 #define NET_IP_ALIGN 2
2968 * The networking layer reserves some headroom in skb data (via
2969 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2970 * the header has to grow. In the default case, if the header has to grow
2971 * 32 bytes or less we avoid the reallocation.
2973 * Unfortunately this headroom changes the DMA alignment of the resulting
2974 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2975 * on some architectures. An architecture can override this value,
2976 * perhaps setting it to a cacheline in size (since that will maintain
2977 * cacheline alignment of the DMA). It must be a power of 2.
2979 * Various parts of the networking layer expect at least 32 bytes of
2980 * headroom, you should not reduce this.
2982 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2983 * to reduce average number of cache lines per packet.
2984 * get_rps_cpu() for example only access one 64 bytes aligned block :
2985 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2988 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2991 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2993 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2995 if (WARN_ON(skb_is_nonlinear(skb)))
2998 skb_set_tail_pointer(skb, len);
3001 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
3003 __skb_set_length(skb, len);
3006 void skb_trim(struct sk_buff *skb, unsigned int len);
3008 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
3011 return ___pskb_trim(skb, len);
3012 __skb_trim(skb, len);
3016 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
3018 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
3022 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
3023 * @skb: buffer to alter
3026 * This is identical to pskb_trim except that the caller knows that
3027 * the skb is not cloned so we should never get an error due to out-
3030 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3032 int err = pskb_trim(skb, len);
3036 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3038 unsigned int diff = len - skb->len;
3040 if (skb_tailroom(skb) < diff) {
3041 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3046 __skb_set_length(skb, len);
3051 * skb_orphan - orphan a buffer
3052 * @skb: buffer to orphan
3054 * If a buffer currently has an owner then we call the owner's
3055 * destructor function and make the @skb unowned. The buffer continues
3056 * to exist but is no longer charged to its former owner.
3058 static inline void skb_orphan(struct sk_buff *skb)
3060 if (skb->destructor) {
3061 skb->destructor(skb);
3062 skb->destructor = NULL;
3070 * skb_orphan_frags - orphan the frags contained in a buffer
3071 * @skb: buffer to orphan frags from
3072 * @gfp_mask: allocation mask for replacement pages
3074 * For each frag in the SKB which needs a destructor (i.e. has an
3075 * owner) create a copy of that frag and release the original
3076 * page by calling the destructor.
3078 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3080 if (likely(!skb_zcopy(skb)))
3082 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN)
3084 return skb_copy_ubufs(skb, gfp_mask);
3087 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3088 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3090 if (likely(!skb_zcopy(skb)))
3092 return skb_copy_ubufs(skb, gfp_mask);
3096 * __skb_queue_purge - empty a list
3097 * @list: list to empty
3099 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3100 * the list and one reference dropped. This function does not take the
3101 * list lock and the caller must hold the relevant locks to use it.
3103 static inline void __skb_queue_purge(struct sk_buff_head *list)
3105 struct sk_buff *skb;
3106 while ((skb = __skb_dequeue(list)) != NULL)
3109 void skb_queue_purge(struct sk_buff_head *list);
3111 unsigned int skb_rbtree_purge(struct rb_root *root);
3113 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3116 * netdev_alloc_frag - allocate a page fragment
3117 * @fragsz: fragment size
3119 * Allocates a frag from a page for receive buffer.
3120 * Uses GFP_ATOMIC allocations.
3122 static inline void *netdev_alloc_frag(unsigned int fragsz)
3124 return __netdev_alloc_frag_align(fragsz, ~0u);
3127 static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3130 WARN_ON_ONCE(!is_power_of_2(align));
3131 return __netdev_alloc_frag_align(fragsz, -align);
3134 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3138 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3139 * @dev: network device to receive on
3140 * @length: length to allocate
3142 * Allocate a new &sk_buff and assign it a usage count of one. The
3143 * buffer has unspecified headroom built in. Users should allocate
3144 * the headroom they think they need without accounting for the
3145 * built in space. The built in space is used for optimisations.
3147 * %NULL is returned if there is no free memory. Although this function
3148 * allocates memory it can be called from an interrupt.
3150 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3151 unsigned int length)
3153 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3156 /* legacy helper around __netdev_alloc_skb() */
3157 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3160 return __netdev_alloc_skb(NULL, length, gfp_mask);
3163 /* legacy helper around netdev_alloc_skb() */
3164 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3166 return netdev_alloc_skb(NULL, length);
3170 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3171 unsigned int length, gfp_t gfp)
3173 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3175 if (NET_IP_ALIGN && skb)
3176 skb_reserve(skb, NET_IP_ALIGN);
3180 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3181 unsigned int length)
3183 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3186 static inline void skb_free_frag(void *addr)
3188 page_frag_free(addr);
3191 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3193 static inline void *napi_alloc_frag(unsigned int fragsz)
3195 return __napi_alloc_frag_align(fragsz, ~0u);
3198 static inline void *napi_alloc_frag_align(unsigned int fragsz,
3201 WARN_ON_ONCE(!is_power_of_2(align));
3202 return __napi_alloc_frag_align(fragsz, -align);
3205 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
3206 unsigned int length, gfp_t gfp_mask);
3207 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
3208 unsigned int length)
3210 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
3212 void napi_consume_skb(struct sk_buff *skb, int budget);
3214 void napi_skb_free_stolen_head(struct sk_buff *skb);
3215 void __kfree_skb_defer(struct sk_buff *skb);
3218 * __dev_alloc_pages - allocate page for network Rx
3219 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3220 * @order: size of the allocation
3222 * Allocate a new page.
3224 * %NULL is returned if there is no free memory.
3226 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
3229 /* This piece of code contains several assumptions.
3230 * 1. This is for device Rx, therefor a cold page is preferred.
3231 * 2. The expectation is the user wants a compound page.
3232 * 3. If requesting a order 0 page it will not be compound
3233 * due to the check to see if order has a value in prep_new_page
3234 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3235 * code in gfp_to_alloc_flags that should be enforcing this.
3237 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3239 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
3242 static inline struct page *dev_alloc_pages(unsigned int order)
3244 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
3248 * __dev_alloc_page - allocate a page for network Rx
3249 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3251 * Allocate a new page.
3253 * %NULL is returned if there is no free memory.
3255 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
3257 return __dev_alloc_pages(gfp_mask, 0);
3260 static inline struct page *dev_alloc_page(void)
3262 return dev_alloc_pages(0);
3266 * dev_page_is_reusable - check whether a page can be reused for network Rx
3267 * @page: the page to test
3269 * A page shouldn't be considered for reusing/recycling if it was allocated
3270 * under memory pressure or at a distant memory node.
3272 * Returns false if this page should be returned to page allocator, true
3275 static inline bool dev_page_is_reusable(const struct page *page)
3277 return likely(page_to_nid(page) == numa_mem_id() &&
3278 !page_is_pfmemalloc(page));
3282 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3283 * @page: The page that was allocated from skb_alloc_page
3284 * @skb: The skb that may need pfmemalloc set
3286 static inline void skb_propagate_pfmemalloc(const struct page *page,
3287 struct sk_buff *skb)
3289 if (page_is_pfmemalloc(page))
3290 skb->pfmemalloc = true;
3294 * skb_frag_off() - Returns the offset of a skb fragment
3295 * @frag: the paged fragment
3297 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3299 return frag->bv_offset;
3303 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3304 * @frag: skb fragment
3305 * @delta: value to add
3307 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3309 frag->bv_offset += delta;
3313 * skb_frag_off_set() - Sets the offset of a skb fragment
3314 * @frag: skb fragment
3315 * @offset: offset of fragment
3317 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3319 frag->bv_offset = offset;
3323 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3324 * @fragto: skb fragment where offset is set
3325 * @fragfrom: skb fragment offset is copied from
3327 static inline void skb_frag_off_copy(skb_frag_t *fragto,
3328 const skb_frag_t *fragfrom)
3330 fragto->bv_offset = fragfrom->bv_offset;
3334 * skb_frag_page - retrieve the page referred to by a paged fragment
3335 * @frag: the paged fragment
3337 * Returns the &struct page associated with @frag.
3339 static inline struct page *skb_frag_page(const skb_frag_t *frag)
3341 return frag->bv_page;
3345 * __skb_frag_ref - take an addition reference on a paged fragment.
3346 * @frag: the paged fragment
3348 * Takes an additional reference on the paged fragment @frag.
3350 static inline void __skb_frag_ref(skb_frag_t *frag)
3352 get_page(skb_frag_page(frag));
3356 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
3358 * @f: the fragment offset.
3360 * Takes an additional reference on the @f'th paged fragment of @skb.
3362 static inline void skb_frag_ref(struct sk_buff *skb, int f)
3364 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
3368 * __skb_frag_unref - release a reference on a paged fragment.
3369 * @frag: the paged fragment
3370 * @recycle: recycle the page if allocated via page_pool
3372 * Releases a reference on the paged fragment @frag
3373 * or recycles the page via the page_pool API.
3375 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle)
3377 struct page *page = skb_frag_page(frag);
3379 #ifdef CONFIG_PAGE_POOL
3380 if (recycle && page_pool_return_skb_page(page))
3387 * skb_frag_unref - release a reference on a paged fragment of an skb.
3389 * @f: the fragment offset
3391 * Releases a reference on the @f'th paged fragment of @skb.
3393 static inline void skb_frag_unref(struct sk_buff *skb, int f)
3395 struct skb_shared_info *shinfo = skb_shinfo(skb);
3397 if (!skb_zcopy_managed(skb))
3398 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle);
3402 * skb_frag_address - gets the address of the data contained in a paged fragment
3403 * @frag: the paged fragment buffer
3405 * Returns the address of the data within @frag. The page must already
3408 static inline void *skb_frag_address(const skb_frag_t *frag)
3410 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3414 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3415 * @frag: the paged fragment buffer
3417 * Returns the address of the data within @frag. Checks that the page
3418 * is mapped and returns %NULL otherwise.
3420 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3422 void *ptr = page_address(skb_frag_page(frag));
3426 return ptr + skb_frag_off(frag);
3430 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3431 * @fragto: skb fragment where page is set
3432 * @fragfrom: skb fragment page is copied from
3434 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3435 const skb_frag_t *fragfrom)
3437 fragto->bv_page = fragfrom->bv_page;
3441 * __skb_frag_set_page - sets the page contained in a paged fragment
3442 * @frag: the paged fragment
3443 * @page: the page to set
3445 * Sets the fragment @frag to contain @page.
3447 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3449 frag->bv_page = page;
3453 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3455 * @f: the fragment offset
3456 * @page: the page to set
3458 * Sets the @f'th fragment of @skb to contain @page.
3460 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3463 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3466 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3469 * skb_frag_dma_map - maps a paged fragment via the DMA API
3470 * @dev: the device to map the fragment to
3471 * @frag: the paged fragment to map
3472 * @offset: the offset within the fragment (starting at the
3473 * fragment's own offset)
3474 * @size: the number of bytes to map
3475 * @dir: the direction of the mapping (``PCI_DMA_*``)
3477 * Maps the page associated with @frag to @device.
3479 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3480 const skb_frag_t *frag,
3481 size_t offset, size_t size,
3482 enum dma_data_direction dir)
3484 return dma_map_page(dev, skb_frag_page(frag),
3485 skb_frag_off(frag) + offset, size, dir);
3488 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3491 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3495 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3498 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3503 * skb_clone_writable - is the header of a clone writable
3504 * @skb: buffer to check
3505 * @len: length up to which to write
3507 * Returns true if modifying the header part of the cloned buffer
3508 * does not requires the data to be copied.
3510 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3512 return !skb_header_cloned(skb) &&
3513 skb_headroom(skb) + len <= skb->hdr_len;
3516 static inline int skb_try_make_writable(struct sk_buff *skb,
3517 unsigned int write_len)
3519 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3520 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3523 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3528 if (headroom > skb_headroom(skb))
3529 delta = headroom - skb_headroom(skb);
3531 if (delta || cloned)
3532 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3538 * skb_cow - copy header of skb when it is required
3539 * @skb: buffer to cow
3540 * @headroom: needed headroom
3542 * If the skb passed lacks sufficient headroom or its data part
3543 * is shared, data is reallocated. If reallocation fails, an error
3544 * is returned and original skb is not changed.
3546 * The result is skb with writable area skb->head...skb->tail
3547 * and at least @headroom of space at head.
3549 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3551 return __skb_cow(skb, headroom, skb_cloned(skb));
3555 * skb_cow_head - skb_cow but only making the head writable
3556 * @skb: buffer to cow
3557 * @headroom: needed headroom
3559 * This function is identical to skb_cow except that we replace the
3560 * skb_cloned check by skb_header_cloned. It should be used when
3561 * you only need to push on some header and do not need to modify
3564 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3566 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3570 * skb_padto - pad an skbuff up to a minimal size
3571 * @skb: buffer to pad
3572 * @len: minimal length
3574 * Pads up a buffer to ensure the trailing bytes exist and are
3575 * blanked. If the buffer already contains sufficient data it
3576 * is untouched. Otherwise it is extended. Returns zero on
3577 * success. The skb is freed on error.
3579 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3581 unsigned int size = skb->len;
3582 if (likely(size >= len))
3584 return skb_pad(skb, len - size);
3588 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3589 * @skb: buffer to pad
3590 * @len: minimal length
3591 * @free_on_error: free buffer on error
3593 * Pads up a buffer to ensure the trailing bytes exist and are
3594 * blanked. If the buffer already contains sufficient data it
3595 * is untouched. Otherwise it is extended. Returns zero on
3596 * success. The skb is freed on error if @free_on_error is true.
3598 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3602 unsigned int size = skb->len;
3604 if (unlikely(size < len)) {
3606 if (__skb_pad(skb, len, free_on_error))
3608 __skb_put(skb, len);
3614 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3615 * @skb: buffer to pad
3616 * @len: minimal length
3618 * Pads up a buffer to ensure the trailing bytes exist and are
3619 * blanked. If the buffer already contains sufficient data it
3620 * is untouched. Otherwise it is extended. Returns zero on
3621 * success. The skb is freed on error.
3623 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3625 return __skb_put_padto(skb, len, true);
3628 static inline int skb_add_data(struct sk_buff *skb,
3629 struct iov_iter *from, int copy)
3631 const int off = skb->len;
3633 if (skb->ip_summed == CHECKSUM_NONE) {
3635 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3637 skb->csum = csum_block_add(skb->csum, csum, off);
3640 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3643 __skb_trim(skb, off);
3647 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3648 const struct page *page, int off)
3653 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3655 return page == skb_frag_page(frag) &&
3656 off == skb_frag_off(frag) + skb_frag_size(frag);
3661 static inline int __skb_linearize(struct sk_buff *skb)
3663 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3667 * skb_linearize - convert paged skb to linear one
3668 * @skb: buffer to linarize
3670 * If there is no free memory -ENOMEM is returned, otherwise zero
3671 * is returned and the old skb data released.
3673 static inline int skb_linearize(struct sk_buff *skb)
3675 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3679 * skb_has_shared_frag - can any frag be overwritten
3680 * @skb: buffer to test
3682 * Return true if the skb has at least one frag that might be modified
3683 * by an external entity (as in vmsplice()/sendfile())
3685 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3687 return skb_is_nonlinear(skb) &&
3688 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3692 * skb_linearize_cow - make sure skb is linear and writable
3693 * @skb: buffer to process
3695 * If there is no free memory -ENOMEM is returned, otherwise zero
3696 * is returned and the old skb data released.
3698 static inline int skb_linearize_cow(struct sk_buff *skb)
3700 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3701 __skb_linearize(skb) : 0;
3704 static __always_inline void
3705 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3708 if (skb->ip_summed == CHECKSUM_COMPLETE)
3709 skb->csum = csum_block_sub(skb->csum,
3710 csum_partial(start, len, 0), off);
3711 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3712 skb_checksum_start_offset(skb) < 0)
3713 skb->ip_summed = CHECKSUM_NONE;
3717 * skb_postpull_rcsum - update checksum for received skb after pull
3718 * @skb: buffer to update
3719 * @start: start of data before pull
3720 * @len: length of data pulled
3722 * After doing a pull on a received packet, you need to call this to
3723 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3724 * CHECKSUM_NONE so that it can be recomputed from scratch.
3726 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3727 const void *start, unsigned int len)
3729 if (skb->ip_summed == CHECKSUM_COMPLETE)
3730 skb->csum = wsum_negate(csum_partial(start, len,
3731 wsum_negate(skb->csum)));
3732 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3733 skb_checksum_start_offset(skb) < 0)
3734 skb->ip_summed = CHECKSUM_NONE;
3737 static __always_inline void
3738 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3741 if (skb->ip_summed == CHECKSUM_COMPLETE)
3742 skb->csum = csum_block_add(skb->csum,
3743 csum_partial(start, len, 0), off);
3747 * skb_postpush_rcsum - update checksum for received skb after push
3748 * @skb: buffer to update
3749 * @start: start of data after push
3750 * @len: length of data pushed
3752 * After doing a push on a received packet, you need to call this to
3753 * update the CHECKSUM_COMPLETE checksum.
3755 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3756 const void *start, unsigned int len)
3758 __skb_postpush_rcsum(skb, start, len, 0);
3761 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3764 * skb_push_rcsum - push skb and update receive checksum
3765 * @skb: buffer to update
3766 * @len: length of data pulled
3768 * This function performs an skb_push on the packet and updates
3769 * the CHECKSUM_COMPLETE checksum. It should be used on
3770 * receive path processing instead of skb_push unless you know
3771 * that the checksum difference is zero (e.g., a valid IP header)
3772 * or you are setting ip_summed to CHECKSUM_NONE.
3774 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3777 skb_postpush_rcsum(skb, skb->data, len);
3781 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3783 * pskb_trim_rcsum - trim received skb and update checksum
3784 * @skb: buffer to trim
3787 * This is exactly the same as pskb_trim except that it ensures the
3788 * checksum of received packets are still valid after the operation.
3789 * It can change skb pointers.
3792 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3794 if (likely(len >= skb->len))
3796 return pskb_trim_rcsum_slow(skb, len);
3799 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3801 if (skb->ip_summed == CHECKSUM_COMPLETE)
3802 skb->ip_summed = CHECKSUM_NONE;
3803 __skb_trim(skb, len);
3807 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3809 if (skb->ip_summed == CHECKSUM_COMPLETE)
3810 skb->ip_summed = CHECKSUM_NONE;
3811 return __skb_grow(skb, len);
3814 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3815 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3816 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3817 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3818 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3820 #define skb_queue_walk(queue, skb) \
3821 for (skb = (queue)->next; \
3822 skb != (struct sk_buff *)(queue); \
3825 #define skb_queue_walk_safe(queue, skb, tmp) \
3826 for (skb = (queue)->next, tmp = skb->next; \
3827 skb != (struct sk_buff *)(queue); \
3828 skb = tmp, tmp = skb->next)
3830 #define skb_queue_walk_from(queue, skb) \
3831 for (; skb != (struct sk_buff *)(queue); \
3834 #define skb_rbtree_walk(skb, root) \
3835 for (skb = skb_rb_first(root); skb != NULL; \
3836 skb = skb_rb_next(skb))
3838 #define skb_rbtree_walk_from(skb) \
3839 for (; skb != NULL; \
3840 skb = skb_rb_next(skb))
3842 #define skb_rbtree_walk_from_safe(skb, tmp) \
3843 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3846 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3847 for (tmp = skb->next; \
3848 skb != (struct sk_buff *)(queue); \
3849 skb = tmp, tmp = skb->next)
3851 #define skb_queue_reverse_walk(queue, skb) \
3852 for (skb = (queue)->prev; \
3853 skb != (struct sk_buff *)(queue); \
3856 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3857 for (skb = (queue)->prev, tmp = skb->prev; \
3858 skb != (struct sk_buff *)(queue); \
3859 skb = tmp, tmp = skb->prev)
3861 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3862 for (tmp = skb->prev; \
3863 skb != (struct sk_buff *)(queue); \
3864 skb = tmp, tmp = skb->prev)
3866 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3868 return skb_shinfo(skb)->frag_list != NULL;
3871 static inline void skb_frag_list_init(struct sk_buff *skb)
3873 skb_shinfo(skb)->frag_list = NULL;
3876 #define skb_walk_frags(skb, iter) \
3877 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3880 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3881 int *err, long *timeo_p,
3882 const struct sk_buff *skb);
3883 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3884 struct sk_buff_head *queue,
3887 struct sk_buff **last);
3888 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
3889 struct sk_buff_head *queue,
3890 unsigned int flags, int *off, int *err,
3891 struct sk_buff **last);
3892 struct sk_buff *__skb_recv_datagram(struct sock *sk,
3893 struct sk_buff_head *sk_queue,
3894 unsigned int flags, int *off, int *err);
3895 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
3896 __poll_t datagram_poll(struct file *file, struct socket *sock,
3897 struct poll_table_struct *wait);
3898 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3899 struct iov_iter *to, int size);
3900 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3901 struct msghdr *msg, int size)
3903 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3905 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3906 struct msghdr *msg);
3907 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3908 struct iov_iter *to, int len,
3909 struct ahash_request *hash);
3910 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3911 struct iov_iter *from, int len);
3912 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3913 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3914 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3915 static inline void skb_free_datagram_locked(struct sock *sk,
3916 struct sk_buff *skb)
3918 __skb_free_datagram_locked(sk, skb, 0);
3920 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3921 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3922 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3923 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3925 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3926 struct pipe_inode_info *pipe, unsigned int len,
3927 unsigned int flags);
3928 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3930 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3931 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3932 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3933 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3935 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3936 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3937 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3938 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3939 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3940 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3941 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
3942 unsigned int offset);
3943 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3944 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
3945 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3946 int skb_vlan_pop(struct sk_buff *skb);
3947 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3948 int skb_eth_pop(struct sk_buff *skb);
3949 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
3950 const unsigned char *src);
3951 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
3952 int mac_len, bool ethernet);
3953 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
3955 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3956 int skb_mpls_dec_ttl(struct sk_buff *skb);
3957 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3960 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3962 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3965 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3967 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3970 struct skb_checksum_ops {
3971 __wsum (*update)(const void *mem, int len, __wsum wsum);
3972 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3975 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3977 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3978 __wsum csum, const struct skb_checksum_ops *ops);
3979 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3982 static inline void * __must_check
3983 __skb_header_pointer(const struct sk_buff *skb, int offset, int len,
3984 const void *data, int hlen, void *buffer)
3986 if (likely(hlen - offset >= len))
3987 return (void *)data + offset;
3989 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
3995 static inline void * __must_check
3996 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3998 return __skb_header_pointer(skb, offset, len, skb->data,
3999 skb_headlen(skb), buffer);
4003 * skb_needs_linearize - check if we need to linearize a given skb
4004 * depending on the given device features.
4005 * @skb: socket buffer to check
4006 * @features: net device features
4008 * Returns true if either:
4009 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
4010 * 2. skb is fragmented and the device does not support SG.
4012 static inline bool skb_needs_linearize(struct sk_buff *skb,
4013 netdev_features_t features)
4015 return skb_is_nonlinear(skb) &&
4016 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
4017 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
4020 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
4022 const unsigned int len)
4024 memcpy(to, skb->data, len);
4027 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4028 const int offset, void *to,
4029 const unsigned int len)
4031 memcpy(to, skb->data + offset, len);
4034 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4036 const unsigned int len)
4038 memcpy(skb->data, from, len);
4041 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4044 const unsigned int len)
4046 memcpy(skb->data + offset, from, len);
4049 void skb_init(void);
4051 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4057 * skb_get_timestamp - get timestamp from a skb
4058 * @skb: skb to get stamp from
4059 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4061 * Timestamps are stored in the skb as offsets to a base timestamp.
4062 * This function converts the offset back to a struct timeval and stores
4065 static inline void skb_get_timestamp(const struct sk_buff *skb,
4066 struct __kernel_old_timeval *stamp)
4068 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4071 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4072 struct __kernel_sock_timeval *stamp)
4074 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4076 stamp->tv_sec = ts.tv_sec;
4077 stamp->tv_usec = ts.tv_nsec / 1000;
4080 static inline void skb_get_timestampns(const struct sk_buff *skb,
4081 struct __kernel_old_timespec *stamp)
4083 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4085 stamp->tv_sec = ts.tv_sec;
4086 stamp->tv_nsec = ts.tv_nsec;
4089 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4090 struct __kernel_timespec *stamp)
4092 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4094 stamp->tv_sec = ts.tv_sec;
4095 stamp->tv_nsec = ts.tv_nsec;
4098 static inline void __net_timestamp(struct sk_buff *skb)
4100 skb->tstamp = ktime_get_real();
4101 skb->mono_delivery_time = 0;
4104 static inline ktime_t net_timedelta(ktime_t t)
4106 return ktime_sub(ktime_get_real(), t);
4109 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4113 skb->mono_delivery_time = kt && mono;
4116 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4118 /* It is used in the ingress path to clear the delivery_time.
4119 * If needed, set the skb->tstamp to the (rcv) timestamp.
4121 static inline void skb_clear_delivery_time(struct sk_buff *skb)
4123 if (skb->mono_delivery_time) {
4124 skb->mono_delivery_time = 0;
4125 if (static_branch_unlikely(&netstamp_needed_key))
4126 skb->tstamp = ktime_get_real();
4132 static inline void skb_clear_tstamp(struct sk_buff *skb)
4134 if (skb->mono_delivery_time)
4140 static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4142 if (skb->mono_delivery_time)
4148 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4150 if (!skb->mono_delivery_time && skb->tstamp)
4153 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4154 return ktime_get_real();
4159 static inline u8 skb_metadata_len(const struct sk_buff *skb)
4161 return skb_shinfo(skb)->meta_len;
4164 static inline void *skb_metadata_end(const struct sk_buff *skb)
4166 return skb_mac_header(skb);
4169 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4170 const struct sk_buff *skb_b,
4173 const void *a = skb_metadata_end(skb_a);
4174 const void *b = skb_metadata_end(skb_b);
4175 /* Using more efficient varaiant than plain call to memcmp(). */
4176 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
4180 #define __it(x, op) (x -= sizeof(u##op))
4181 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4182 case 32: diffs |= __it_diff(a, b, 64);
4184 case 24: diffs |= __it_diff(a, b, 64);
4186 case 16: diffs |= __it_diff(a, b, 64);
4188 case 8: diffs |= __it_diff(a, b, 64);
4190 case 28: diffs |= __it_diff(a, b, 64);
4192 case 20: diffs |= __it_diff(a, b, 64);
4194 case 12: diffs |= __it_diff(a, b, 64);
4196 case 4: diffs |= __it_diff(a, b, 32);
4201 return memcmp(a - meta_len, b - meta_len, meta_len);
4205 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4206 const struct sk_buff *skb_b)
4208 u8 len_a = skb_metadata_len(skb_a);
4209 u8 len_b = skb_metadata_len(skb_b);
4211 if (!(len_a | len_b))
4214 return len_a != len_b ?
4215 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4218 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4220 skb_shinfo(skb)->meta_len = meta_len;
4223 static inline void skb_metadata_clear(struct sk_buff *skb)
4225 skb_metadata_set(skb, 0);
4228 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4230 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4232 void skb_clone_tx_timestamp(struct sk_buff *skb);
4233 bool skb_defer_rx_timestamp(struct sk_buff *skb);
4235 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4237 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4241 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4246 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4249 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4251 * PHY drivers may accept clones of transmitted packets for
4252 * timestamping via their phy_driver.txtstamp method. These drivers
4253 * must call this function to return the skb back to the stack with a
4256 * @skb: clone of the original outgoing packet
4257 * @hwtstamps: hardware time stamps
4260 void skb_complete_tx_timestamp(struct sk_buff *skb,
4261 struct skb_shared_hwtstamps *hwtstamps);
4263 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4264 struct skb_shared_hwtstamps *hwtstamps,
4265 struct sock *sk, int tstype);
4268 * skb_tstamp_tx - queue clone of skb with send time stamps
4269 * @orig_skb: the original outgoing packet
4270 * @hwtstamps: hardware time stamps, may be NULL if not available
4272 * If the skb has a socket associated, then this function clones the
4273 * skb (thus sharing the actual data and optional structures), stores
4274 * the optional hardware time stamping information (if non NULL) or
4275 * generates a software time stamp (otherwise), then queues the clone
4276 * to the error queue of the socket. Errors are silently ignored.
4278 void skb_tstamp_tx(struct sk_buff *orig_skb,
4279 struct skb_shared_hwtstamps *hwtstamps);
4282 * skb_tx_timestamp() - Driver hook for transmit timestamping
4284 * Ethernet MAC Drivers should call this function in their hard_xmit()
4285 * function immediately before giving the sk_buff to the MAC hardware.
4287 * Specifically, one should make absolutely sure that this function is
4288 * called before TX completion of this packet can trigger. Otherwise
4289 * the packet could potentially already be freed.
4291 * @skb: A socket buffer.
4293 static inline void skb_tx_timestamp(struct sk_buff *skb)
4295 skb_clone_tx_timestamp(skb);
4296 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4297 skb_tstamp_tx(skb, NULL);
4301 * skb_complete_wifi_ack - deliver skb with wifi status
4303 * @skb: the original outgoing packet
4304 * @acked: ack status
4307 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4309 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4310 __sum16 __skb_checksum_complete(struct sk_buff *skb);
4312 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4314 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4316 (skb->ip_summed == CHECKSUM_PARTIAL &&
4317 skb_checksum_start_offset(skb) >= 0));
4321 * skb_checksum_complete - Calculate checksum of an entire packet
4322 * @skb: packet to process
4324 * This function calculates the checksum over the entire packet plus
4325 * the value of skb->csum. The latter can be used to supply the
4326 * checksum of a pseudo header as used by TCP/UDP. It returns the
4329 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4330 * this function can be used to verify that checksum on received
4331 * packets. In that case the function should return zero if the
4332 * checksum is correct. In particular, this function will return zero
4333 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4334 * hardware has already verified the correctness of the checksum.
4336 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4338 return skb_csum_unnecessary(skb) ?
4339 0 : __skb_checksum_complete(skb);
4342 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4344 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4345 if (skb->csum_level == 0)
4346 skb->ip_summed = CHECKSUM_NONE;
4352 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4354 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4355 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4357 } else if (skb->ip_summed == CHECKSUM_NONE) {
4358 skb->ip_summed = CHECKSUM_UNNECESSARY;
4359 skb->csum_level = 0;
4363 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4365 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4366 skb->ip_summed = CHECKSUM_NONE;
4367 skb->csum_level = 0;
4371 /* Check if we need to perform checksum complete validation.
4373 * Returns true if checksum complete is needed, false otherwise
4374 * (either checksum is unnecessary or zero checksum is allowed).
4376 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4380 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4381 skb->csum_valid = 1;
4382 __skb_decr_checksum_unnecessary(skb);
4389 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4392 #define CHECKSUM_BREAK 76
4394 /* Unset checksum-complete
4396 * Unset checksum complete can be done when packet is being modified
4397 * (uncompressed for instance) and checksum-complete value is
4400 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4402 if (skb->ip_summed == CHECKSUM_COMPLETE)
4403 skb->ip_summed = CHECKSUM_NONE;
4406 /* Validate (init) checksum based on checksum complete.
4409 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4410 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4411 * checksum is stored in skb->csum for use in __skb_checksum_complete
4412 * non-zero: value of invalid checksum
4415 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4419 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4420 if (!csum_fold(csum_add(psum, skb->csum))) {
4421 skb->csum_valid = 1;
4428 if (complete || skb->len <= CHECKSUM_BREAK) {
4431 csum = __skb_checksum_complete(skb);
4432 skb->csum_valid = !csum;
4439 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4444 /* Perform checksum validate (init). Note that this is a macro since we only
4445 * want to calculate the pseudo header which is an input function if necessary.
4446 * First we try to validate without any computation (checksum unnecessary) and
4447 * then calculate based on checksum complete calling the function to compute
4451 * 0: checksum is validated or try to in skb_checksum_complete
4452 * non-zero: value of invalid checksum
4454 #define __skb_checksum_validate(skb, proto, complete, \
4455 zero_okay, check, compute_pseudo) \
4457 __sum16 __ret = 0; \
4458 skb->csum_valid = 0; \
4459 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4460 __ret = __skb_checksum_validate_complete(skb, \
4461 complete, compute_pseudo(skb, proto)); \
4465 #define skb_checksum_init(skb, proto, compute_pseudo) \
4466 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4468 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4469 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4471 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4472 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4474 #define skb_checksum_validate_zero_check(skb, proto, check, \
4476 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4478 #define skb_checksum_simple_validate(skb) \
4479 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4481 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4483 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4486 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4488 skb->csum = ~pseudo;
4489 skb->ip_summed = CHECKSUM_COMPLETE;
4492 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4494 if (__skb_checksum_convert_check(skb)) \
4495 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4498 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4499 u16 start, u16 offset)
4501 skb->ip_summed = CHECKSUM_PARTIAL;
4502 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4503 skb->csum_offset = offset - start;
4506 /* Update skbuf and packet to reflect the remote checksum offload operation.
4507 * When called, ptr indicates the starting point for skb->csum when
4508 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4509 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4511 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4512 int start, int offset, bool nopartial)
4517 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4521 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4522 __skb_checksum_complete(skb);
4523 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4526 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4528 /* Adjust skb->csum since we changed the packet */
4529 skb->csum = csum_add(skb->csum, delta);
4532 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4534 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4535 return (void *)(skb->_nfct & NFCT_PTRMASK);
4541 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4543 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4550 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4552 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4553 skb->slow_gro |= !!nfct;
4558 #ifdef CONFIG_SKB_EXTENSIONS
4560 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4566 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4569 #if IS_ENABLED(CONFIG_MPTCP)
4572 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4575 SKB_EXT_NUM, /* must be last */
4579 * struct skb_ext - sk_buff extensions
4580 * @refcnt: 1 on allocation, deallocated on 0
4581 * @offset: offset to add to @data to obtain extension address
4582 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4583 * @data: start of extension data, variable sized
4585 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4586 * to use 'u8' types while allowing up to 2kb worth of extension data.
4590 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4591 u8 chunks; /* same */
4592 char data[] __aligned(8);
4595 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4596 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4597 struct skb_ext *ext);
4598 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4599 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4600 void __skb_ext_put(struct skb_ext *ext);
4602 static inline void skb_ext_put(struct sk_buff *skb)
4604 if (skb->active_extensions)
4605 __skb_ext_put(skb->extensions);
4608 static inline void __skb_ext_copy(struct sk_buff *dst,
4609 const struct sk_buff *src)
4611 dst->active_extensions = src->active_extensions;
4613 if (src->active_extensions) {
4614 struct skb_ext *ext = src->extensions;
4616 refcount_inc(&ext->refcnt);
4617 dst->extensions = ext;
4621 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4624 __skb_ext_copy(dst, src);
4627 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4629 return !!ext->offset[i];
4632 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4634 return skb->active_extensions & (1 << id);
4637 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4639 if (skb_ext_exist(skb, id))
4640 __skb_ext_del(skb, id);
4643 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4645 if (skb_ext_exist(skb, id)) {
4646 struct skb_ext *ext = skb->extensions;
4648 return (void *)ext + (ext->offset[id] << 3);
4654 static inline void skb_ext_reset(struct sk_buff *skb)
4656 if (unlikely(skb->active_extensions)) {
4657 __skb_ext_put(skb->extensions);
4658 skb->active_extensions = 0;
4662 static inline bool skb_has_extensions(struct sk_buff *skb)
4664 return unlikely(skb->active_extensions);
4667 static inline void skb_ext_put(struct sk_buff *skb) {}
4668 static inline void skb_ext_reset(struct sk_buff *skb) {}
4669 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4670 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4671 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4672 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4673 #endif /* CONFIG_SKB_EXTENSIONS */
4675 static inline void nf_reset_ct(struct sk_buff *skb)
4677 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4678 nf_conntrack_put(skb_nfct(skb));
4683 static inline void nf_reset_trace(struct sk_buff *skb)
4685 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4690 static inline void ipvs_reset(struct sk_buff *skb)
4692 #if IS_ENABLED(CONFIG_IP_VS)
4693 skb->ipvs_property = 0;
4697 /* Note: This doesn't put any conntrack info in dst. */
4698 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4701 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4702 dst->_nfct = src->_nfct;
4703 nf_conntrack_get(skb_nfct(src));
4705 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4707 dst->nf_trace = src->nf_trace;
4711 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4713 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4714 nf_conntrack_put(skb_nfct(dst));
4716 dst->slow_gro = src->slow_gro;
4717 __nf_copy(dst, src, true);
4720 #ifdef CONFIG_NETWORK_SECMARK
4721 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4723 to->secmark = from->secmark;
4726 static inline void skb_init_secmark(struct sk_buff *skb)
4731 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4734 static inline void skb_init_secmark(struct sk_buff *skb)
4738 static inline int secpath_exists(const struct sk_buff *skb)
4741 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4747 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4749 return !skb->destructor &&
4750 !secpath_exists(skb) &&
4752 !skb->_skb_refdst &&
4753 !skb_has_frag_list(skb);
4756 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4758 skb->queue_mapping = queue_mapping;
4761 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4763 return skb->queue_mapping;
4766 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4768 to->queue_mapping = from->queue_mapping;
4771 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4773 skb->queue_mapping = rx_queue + 1;
4776 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4778 return skb->queue_mapping - 1;
4781 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4783 return skb->queue_mapping != 0;
4786 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4788 skb->dst_pending_confirm = val;
4791 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4793 return skb->dst_pending_confirm != 0;
4796 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4799 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4805 /* Keeps track of mac header offset relative to skb->head.
4806 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4807 * For non-tunnel skb it points to skb_mac_header() and for
4808 * tunnel skb it points to outer mac header.
4809 * Keeps track of level of encapsulation of network headers.
4820 #define SKB_GSO_CB_OFFSET 32
4821 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET))
4823 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4825 return (skb_mac_header(inner_skb) - inner_skb->head) -
4826 SKB_GSO_CB(inner_skb)->mac_offset;
4829 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4831 int new_headroom, headroom;
4834 headroom = skb_headroom(skb);
4835 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4839 new_headroom = skb_headroom(skb);
4840 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4844 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4846 /* Do not update partial checksums if remote checksum is enabled. */
4847 if (skb->remcsum_offload)
4850 SKB_GSO_CB(skb)->csum = res;
4851 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4854 /* Compute the checksum for a gso segment. First compute the checksum value
4855 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4856 * then add in skb->csum (checksum from csum_start to end of packet).
4857 * skb->csum and csum_start are then updated to reflect the checksum of the
4858 * resultant packet starting from the transport header-- the resultant checksum
4859 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4862 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4864 unsigned char *csum_start = skb_transport_header(skb);
4865 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4866 __wsum partial = SKB_GSO_CB(skb)->csum;
4868 SKB_GSO_CB(skb)->csum = res;
4869 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4871 return csum_fold(csum_partial(csum_start, plen, partial));
4874 static inline bool skb_is_gso(const struct sk_buff *skb)
4876 return skb_shinfo(skb)->gso_size;
4879 /* Note: Should be called only if skb_is_gso(skb) is true */
4880 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4882 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4885 /* Note: Should be called only if skb_is_gso(skb) is true */
4886 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4888 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4891 /* Note: Should be called only if skb_is_gso(skb) is true */
4892 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4894 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4897 static inline void skb_gso_reset(struct sk_buff *skb)
4899 skb_shinfo(skb)->gso_size = 0;
4900 skb_shinfo(skb)->gso_segs = 0;
4901 skb_shinfo(skb)->gso_type = 0;
4904 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4907 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4909 shinfo->gso_size += increment;
4912 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4915 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4917 shinfo->gso_size -= decrement;
4920 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4922 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4924 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4925 * wanted then gso_type will be set. */
4926 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4928 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4929 unlikely(shinfo->gso_type == 0)) {
4930 __skb_warn_lro_forwarding(skb);
4936 static inline void skb_forward_csum(struct sk_buff *skb)
4938 /* Unfortunately we don't support this one. Any brave souls? */
4939 if (skb->ip_summed == CHECKSUM_COMPLETE)
4940 skb->ip_summed = CHECKSUM_NONE;
4944 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4945 * @skb: skb to check
4947 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4948 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4949 * use this helper, to document places where we make this assertion.
4951 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4953 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE);
4956 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4958 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4959 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4960 unsigned int transport_len,
4961 __sum16(*skb_chkf)(struct sk_buff *skb));
4964 * skb_head_is_locked - Determine if the skb->head is locked down
4965 * @skb: skb to check
4967 * The head on skbs build around a head frag can be removed if they are
4968 * not cloned. This function returns true if the skb head is locked down
4969 * due to either being allocated via kmalloc, or by being a clone with
4970 * multiple references to the head.
4972 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4974 return !skb->head_frag || skb_cloned(skb);
4977 /* Local Checksum Offload.
4978 * Compute outer checksum based on the assumption that the
4979 * inner checksum will be offloaded later.
4980 * See Documentation/networking/checksum-offloads.rst for
4981 * explanation of how this works.
4982 * Fill in outer checksum adjustment (e.g. with sum of outer
4983 * pseudo-header) before calling.
4984 * Also ensure that inner checksum is in linear data area.
4986 static inline __wsum lco_csum(struct sk_buff *skb)
4988 unsigned char *csum_start = skb_checksum_start(skb);
4989 unsigned char *l4_hdr = skb_transport_header(skb);
4992 /* Start with complement of inner checksum adjustment */
4993 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4996 /* Add in checksum of our headers (incl. outer checksum
4997 * adjustment filled in by caller) and return result.
4999 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
5002 static inline bool skb_is_redirected(const struct sk_buff *skb)
5004 return skb->redirected;
5007 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
5009 skb->redirected = 1;
5010 #ifdef CONFIG_NET_REDIRECT
5011 skb->from_ingress = from_ingress;
5012 if (skb->from_ingress)
5013 skb_clear_tstamp(skb);
5017 static inline void skb_reset_redirect(struct sk_buff *skb)
5019 skb->redirected = 0;
5022 static inline bool skb_csum_is_sctp(struct sk_buff *skb)
5024 return skb->csum_not_inet;
5027 static inline void skb_set_kcov_handle(struct sk_buff *skb,
5028 const u64 kcov_handle)
5031 skb->kcov_handle = kcov_handle;
5035 static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
5038 return skb->kcov_handle;
5044 #ifdef CONFIG_PAGE_POOL
5045 static inline void skb_mark_for_recycle(struct sk_buff *skb)
5047 skb->pp_recycle = 1;
5051 #endif /* __KERNEL__ */
5052 #endif /* _LINUX_SKBUFF_H */