1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * Definitions for the 'struct sk_buff' memory handlers.
10 #ifndef _LINUX_SKBUFF_H
11 #define _LINUX_SKBUFF_H
13 #include <linux/kernel.h>
14 #include <linux/compiler.h>
15 #include <linux/time.h>
16 #include <linux/bug.h>
17 #include <linux/bvec.h>
18 #include <linux/cache.h>
19 #include <linux/rbtree.h>
20 #include <linux/socket.h>
21 #include <linux/refcount.h>
23 #include <linux/atomic.h>
24 #include <asm/types.h>
25 #include <linux/spinlock.h>
26 #include <net/checksum.h>
27 #include <linux/rcupdate.h>
28 #include <linux/dma-mapping.h>
29 #include <linux/netdev_features.h>
30 #include <net/flow_dissector.h>
31 #include <linux/in6.h>
32 #include <linux/if_packet.h>
33 #include <linux/llist.h>
35 #include <net/page_pool.h>
36 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
37 #include <linux/netfilter/nf_conntrack_common.h>
39 #include <net/net_debug.h>
40 #include <net/dropreason-core.h>
45 * The interface for checksum offload between the stack and networking drivers
48 * IP checksum related features
49 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
51 * Drivers advertise checksum offload capabilities in the features of a device.
52 * From the stack's point of view these are capabilities offered by the driver.
53 * A driver typically only advertises features that it is capable of offloading
56 * .. flat-table:: Checksum related device features
59 * * - %NETIF_F_HW_CSUM
60 * - The driver (or its device) is able to compute one
61 * IP (one's complement) checksum for any combination
62 * of protocols or protocol layering. The checksum is
63 * computed and set in a packet per the CHECKSUM_PARTIAL
64 * interface (see below).
66 * * - %NETIF_F_IP_CSUM
67 * - Driver (device) is only able to checksum plain
68 * TCP or UDP packets over IPv4. These are specifically
69 * unencapsulated packets of the form IPv4|TCP or
70 * IPv4|UDP where the Protocol field in the IPv4 header
71 * is TCP or UDP. The IPv4 header may contain IP options.
72 * This feature cannot be set in features for a device
73 * with NETIF_F_HW_CSUM also set. This feature is being
74 * DEPRECATED (see below).
76 * * - %NETIF_F_IPV6_CSUM
77 * - Driver (device) is only able to checksum plain
78 * TCP or UDP packets over IPv6. These are specifically
79 * unencapsulated packets of the form IPv6|TCP or
80 * IPv6|UDP where the Next Header field in the IPv6
81 * header is either TCP or UDP. IPv6 extension headers
82 * are not supported with this feature. This feature
83 * cannot be set in features for a device with
84 * NETIF_F_HW_CSUM also set. This feature is being
85 * DEPRECATED (see below).
88 * - Driver (device) performs receive checksum offload.
89 * This flag is only used to disable the RX checksum
90 * feature for a device. The stack will accept receive
91 * checksum indication in packets received on a device
92 * regardless of whether NETIF_F_RXCSUM is set.
94 * Checksumming of received packets by device
95 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
97 * Indication of checksum verification is set in &sk_buff.ip_summed.
98 * Possible values are:
102 * Device did not checksum this packet e.g. due to lack of capabilities.
103 * The packet contains full (though not verified) checksum in packet but
104 * not in skb->csum. Thus, skb->csum is undefined in this case.
106 * - %CHECKSUM_UNNECESSARY
108 * The hardware you're dealing with doesn't calculate the full checksum
109 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums
110 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY
111 * if their checksums are okay. &sk_buff.csum is still undefined in this case
112 * though. A driver or device must never modify the checksum field in the
113 * packet even if checksum is verified.
115 * %CHECKSUM_UNNECESSARY is applicable to following protocols:
117 * - TCP: IPv6 and IPv4.
118 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
119 * zero UDP checksum for either IPv4 or IPv6, the networking stack
120 * may perform further validation in this case.
121 * - GRE: only if the checksum is present in the header.
122 * - SCTP: indicates the CRC in SCTP header has been validated.
123 * - FCOE: indicates the CRC in FC frame has been validated.
125 * &sk_buff.csum_level indicates the number of consecutive checksums found in
126 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY.
127 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
128 * and a device is able to verify the checksums for UDP (possibly zero),
129 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to
130 * two. If the device were only able to verify the UDP checksum and not
131 * GRE, either because it doesn't support GRE checksum or because GRE
132 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
133 * not considered in this case).
135 * - %CHECKSUM_COMPLETE
137 * This is the most generic way. The device supplied checksum of the _whole_
138 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the
139 * hardware doesn't need to parse L3/L4 headers to implement this.
143 * - Even if device supports only some protocols, but is able to produce
144 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
145 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
147 * - %CHECKSUM_PARTIAL
149 * A checksum is set up to be offloaded to a device as described in the
150 * output description for CHECKSUM_PARTIAL. This may occur on a packet
151 * received directly from another Linux OS, e.g., a virtualized Linux kernel
152 * on the same host, or it may be set in the input path in GRO or remote
153 * checksum offload. For the purposes of checksum verification, the checksum
154 * referred to by skb->csum_start + skb->csum_offset and any preceding
155 * checksums in the packet are considered verified. Any checksums in the
156 * packet that are after the checksum being offloaded are not considered to
159 * Checksumming on transmit for non-GSO
160 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
162 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet.
165 * - %CHECKSUM_PARTIAL
167 * The driver is required to checksum the packet as seen by hard_start_xmit()
168 * from &sk_buff.csum_start up to the end, and to record/write the checksum at
169 * offset &sk_buff.csum_start + &sk_buff.csum_offset.
170 * A driver may verify that the
171 * csum_start and csum_offset values are valid values given the length and
172 * offset of the packet, but it should not attempt to validate that the
173 * checksum refers to a legitimate transport layer checksum -- it is the
174 * purview of the stack to validate that csum_start and csum_offset are set
177 * When the stack requests checksum offload for a packet, the driver MUST
178 * ensure that the checksum is set correctly. A driver can either offload the
179 * checksum calculation to the device, or call skb_checksum_help (in the case
180 * that the device does not support offload for a particular checksum).
182 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of
183 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate
184 * checksum offload capability.
185 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based
186 * on network device checksumming capabilities: if a packet does not match
187 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of
188 * &sk_buff.csum_not_inet, see :ref:`crc`)
189 * is called to resolve the checksum.
193 * The skb was already checksummed by the protocol, or a checksum is not
196 * - %CHECKSUM_UNNECESSARY
198 * This has the same meaning as CHECKSUM_NONE for checksum offload on
201 * - %CHECKSUM_COMPLETE
203 * Not used in checksum output. If a driver observes a packet with this value
204 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set.
208 * Non-IP checksum (CRC) offloads
209 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
214 * * - %NETIF_F_SCTP_CRC
215 * - This feature indicates that a device is capable of
216 * offloading the SCTP CRC in a packet. To perform this offload the stack
217 * will set csum_start and csum_offset accordingly, set ip_summed to
218 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication
219 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c.
220 * A driver that supports both IP checksum offload and SCTP CRC32c offload
221 * must verify which offload is configured for a packet by testing the
222 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to
223 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
225 * * - %NETIF_F_FCOE_CRC
226 * - This feature indicates that a device is capable of offloading the FCOE
227 * CRC in a packet. To perform this offload the stack will set ip_summed
228 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset
229 * accordingly. Note that there is no indication in the skbuff that the
230 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports
231 * both IP checksum offload and FCOE CRC offload must verify which offload
232 * is configured for a packet, presumably by inspecting packet headers.
234 * Checksumming on output with GSO
235 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
237 * In the case of a GSO packet (skb_is_gso() is true), checksum offload
238 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
239 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as
240 * part of the GSO operation is implied. If a checksum is being offloaded
241 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and
242 * csum_offset are set to refer to the outermost checksum being offloaded
243 * (two offloaded checksums are possible with UDP encapsulation).
246 /* Don't change this without changing skb_csum_unnecessary! */
247 #define CHECKSUM_NONE 0
248 #define CHECKSUM_UNNECESSARY 1
249 #define CHECKSUM_COMPLETE 2
250 #define CHECKSUM_PARTIAL 3
252 /* Maximum value in skb->csum_level */
253 #define SKB_MAX_CSUM_LEVEL 3
255 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
256 #define SKB_WITH_OVERHEAD(X) \
257 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
259 /* For X bytes available in skb->head, what is the minimal
260 * allocation needed, knowing struct skb_shared_info needs
263 #define SKB_HEAD_ALIGN(X) (SKB_DATA_ALIGN(X) + \
264 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
266 #define SKB_MAX_ORDER(X, ORDER) \
267 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
268 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
269 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
271 /* return minimum truesize of one skb containing X bytes of data */
272 #define SKB_TRUESIZE(X) ((X) + \
273 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
274 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
276 struct ahash_request;
279 struct pipe_inode_info;
287 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
288 struct nf_bridge_info {
290 BRNF_PROTO_UNCHANGED,
297 u8 sabotage_in_done:1;
299 struct net_device *physindev;
301 /* always valid & non-NULL from FORWARD on, for physdev match */
302 struct net_device *physoutdev;
304 /* prerouting: detect dnat in orig/reply direction */
306 struct in6_addr ipv6_daddr;
308 /* after prerouting + nat detected: store original source
309 * mac since neigh resolution overwrites it, only used while
310 * skb is out in neigh layer.
312 char neigh_header[8];
317 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
318 /* Chain in tc_skb_ext will be used to share the tc chain with
319 * ovs recirc_id. It will be set to the current chain by tc
320 * and read by ovs to recirc_id.
332 u8 act_miss:1; /* Set if act_miss_cookie is used */
336 struct sk_buff_head {
337 /* These two members must be first to match sk_buff. */
338 struct_group_tagged(sk_buff_list, list,
339 struct sk_buff *next;
340 struct sk_buff *prev;
349 #ifndef CONFIG_MAX_SKB_FRAGS
350 # define CONFIG_MAX_SKB_FRAGS 17
353 #define MAX_SKB_FRAGS CONFIG_MAX_SKB_FRAGS
355 extern int sysctl_max_skb_frags;
357 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
358 * segment using its current segmentation instead.
360 #define GSO_BY_FRAGS 0xFFFF
362 typedef struct bio_vec skb_frag_t;
365 * skb_frag_size() - Returns the size of a skb fragment
366 * @frag: skb fragment
368 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
374 * skb_frag_size_set() - Sets the size of a skb fragment
375 * @frag: skb fragment
376 * @size: size of fragment
378 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
384 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
385 * @frag: skb fragment
386 * @delta: value to add
388 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
390 frag->bv_len += delta;
394 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
395 * @frag: skb fragment
396 * @delta: value to subtract
398 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
400 frag->bv_len -= delta;
404 * skb_frag_must_loop - Test if %p is a high memory page
405 * @p: fragment's page
407 static inline bool skb_frag_must_loop(struct page *p)
409 #if defined(CONFIG_HIGHMEM)
410 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
417 * skb_frag_foreach_page - loop over pages in a fragment
419 * @f: skb frag to operate on
420 * @f_off: offset from start of f->bv_page
421 * @f_len: length from f_off to loop over
422 * @p: (temp var) current page
423 * @p_off: (temp var) offset from start of current page,
424 * non-zero only on first page.
425 * @p_len: (temp var) length in current page,
426 * < PAGE_SIZE only on first and last page.
427 * @copied: (temp var) length so far, excluding current p_len.
429 * A fragment can hold a compound page, in which case per-page
430 * operations, notably kmap_atomic, must be called for each
433 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
434 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
435 p_off = (f_off) & (PAGE_SIZE - 1), \
436 p_len = skb_frag_must_loop(p) ? \
437 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
440 copied += p_len, p++, p_off = 0, \
441 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
443 #define HAVE_HW_TIME_STAMP
446 * struct skb_shared_hwtstamps - hardware time stamps
447 * @hwtstamp: hardware time stamp transformed into duration
448 * since arbitrary point in time
449 * @netdev_data: address/cookie of network device driver used as
450 * reference to actual hardware time stamp
452 * Software time stamps generated by ktime_get_real() are stored in
455 * hwtstamps can only be compared against other hwtstamps from
458 * This structure is attached to packets as part of the
459 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
461 struct skb_shared_hwtstamps {
468 /* Definitions for tx_flags in struct skb_shared_info */
470 /* generate hardware time stamp */
471 SKBTX_HW_TSTAMP = 1 << 0,
473 /* generate software time stamp when queueing packet to NIC */
474 SKBTX_SW_TSTAMP = 1 << 1,
476 /* device driver is going to provide hardware time stamp */
477 SKBTX_IN_PROGRESS = 1 << 2,
479 /* generate hardware time stamp based on cycles if supported */
480 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3,
482 /* generate wifi status information (where possible) */
483 SKBTX_WIFI_STATUS = 1 << 4,
485 /* determine hardware time stamp based on time or cycles */
486 SKBTX_HW_TSTAMP_NETDEV = 1 << 5,
488 /* generate software time stamp when entering packet scheduling */
489 SKBTX_SCHED_TSTAMP = 1 << 6,
492 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
494 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \
495 SKBTX_HW_TSTAMP_USE_CYCLES | \
498 /* Definitions for flags in struct skb_shared_info */
500 /* use zcopy routines */
501 SKBFL_ZEROCOPY_ENABLE = BIT(0),
503 /* This indicates at least one fragment might be overwritten
504 * (as in vmsplice(), sendfile() ...)
505 * If we need to compute a TX checksum, we'll need to copy
506 * all frags to avoid possible bad checksum
508 SKBFL_SHARED_FRAG = BIT(1),
510 /* segment contains only zerocopy data and should not be
511 * charged to the kernel memory.
513 SKBFL_PURE_ZEROCOPY = BIT(2),
515 SKBFL_DONT_ORPHAN = BIT(3),
517 /* page references are managed by the ubuf_info, so it's safe to
518 * use frags only up until ubuf_info is released
520 SKBFL_MANAGED_FRAG_REFS = BIT(4),
523 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
524 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
525 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
528 * The callback notifies userspace to release buffers when skb DMA is done in
529 * lower device, the skb last reference should be 0 when calling this.
530 * The zerocopy_success argument is true if zero copy transmit occurred,
531 * false on data copy or out of memory error caused by data copy attempt.
532 * The ctx field is used to track device context.
533 * The desc field is used to track userspace buffer index.
536 void (*callback)(struct sk_buff *, struct ubuf_info *,
537 bool zerocopy_success);
542 struct ubuf_info_msgzc {
543 struct ubuf_info ubuf;
559 struct user_struct *user;
564 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
565 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \
568 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
569 void mm_unaccount_pinned_pages(struct mmpin *mmp);
571 /* This data is invariant across clones and lives at
572 * the end of the header data, ie. at skb->end.
574 struct skb_shared_info {
579 unsigned short gso_size;
580 /* Warning: this field is not always filled in (UFO)! */
581 unsigned short gso_segs;
582 struct sk_buff *frag_list;
583 struct skb_shared_hwtstamps hwtstamps;
584 unsigned int gso_type;
588 * Warning : all fields before dataref are cleared in __alloc_skb()
591 unsigned int xdp_frags_size;
593 /* Intermediate layers must ensure that destructor_arg
594 * remains valid until skb destructor */
595 void * destructor_arg;
597 /* must be last field, see pskb_expand_head() */
598 skb_frag_t frags[MAX_SKB_FRAGS];
602 * DOC: dataref and headerless skbs
604 * Transport layers send out clones of payload skbs they hold for
605 * retransmissions. To allow lower layers of the stack to prepend their headers
606 * we split &skb_shared_info.dataref into two halves.
607 * The lower 16 bits count the overall number of references.
608 * The higher 16 bits indicate how many of the references are payload-only.
609 * skb_header_cloned() checks if skb is allowed to add / write the headers.
611 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
612 * (via __skb_header_release()). Any clone created from marked skb will get
613 * &sk_buff.hdr_len populated with the available headroom.
614 * If there's the only clone in existence it's able to modify the headroom
615 * at will. The sequence of calls inside the transport layer is::
619 * __skb_header_release()
621 * // send the clone down the stack
623 * This is not a very generic construct and it depends on the transport layers
624 * doing the right thing. In practice there's usually only one payload-only skb.
625 * Having multiple payload-only skbs with different lengths of hdr_len is not
626 * possible. The payload-only skbs should never leave their owner.
628 #define SKB_DATAREF_SHIFT 16
629 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
633 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
634 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
635 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
639 SKB_GSO_TCPV4 = 1 << 0,
641 /* This indicates the skb is from an untrusted source. */
642 SKB_GSO_DODGY = 1 << 1,
644 /* This indicates the tcp segment has CWR set. */
645 SKB_GSO_TCP_ECN = 1 << 2,
647 SKB_GSO_TCP_FIXEDID = 1 << 3,
649 SKB_GSO_TCPV6 = 1 << 4,
651 SKB_GSO_FCOE = 1 << 5,
653 SKB_GSO_GRE = 1 << 6,
655 SKB_GSO_GRE_CSUM = 1 << 7,
657 SKB_GSO_IPXIP4 = 1 << 8,
659 SKB_GSO_IPXIP6 = 1 << 9,
661 SKB_GSO_UDP_TUNNEL = 1 << 10,
663 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
665 SKB_GSO_PARTIAL = 1 << 12,
667 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
669 SKB_GSO_SCTP = 1 << 14,
671 SKB_GSO_ESP = 1 << 15,
673 SKB_GSO_UDP = 1 << 16,
675 SKB_GSO_UDP_L4 = 1 << 17,
677 SKB_GSO_FRAGLIST = 1 << 18,
680 #if BITS_PER_LONG > 32
681 #define NET_SKBUFF_DATA_USES_OFFSET 1
684 #ifdef NET_SKBUFF_DATA_USES_OFFSET
685 typedef unsigned int sk_buff_data_t;
687 typedef unsigned char *sk_buff_data_t;
691 * DOC: Basic sk_buff geometry
693 * struct sk_buff itself is a metadata structure and does not hold any packet
694 * data. All the data is held in associated buffers.
696 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
699 * - data buffer, containing headers and sometimes payload;
700 * this is the part of the skb operated on by the common helpers
701 * such as skb_put() or skb_pull();
702 * - shared info (struct skb_shared_info) which holds an array of pointers
703 * to read-only data in the (page, offset, length) format.
705 * Optionally &skb_shared_info.frag_list may point to another skb.
707 * Basic diagram may look like this::
712 * ,--------------------------- + head
713 * / ,----------------- + data
714 * / / ,----------- + tail
718 * -----------------------------------------------
719 * | headroom | data | tailroom | skb_shared_info |
720 * -----------------------------------------------
724 * + [page frag] ---------
725 * + frag_list --> | sk_buff |
731 * struct sk_buff - socket buffer
732 * @next: Next buffer in list
733 * @prev: Previous buffer in list
734 * @tstamp: Time we arrived/left
735 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
736 * for retransmit timer
737 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
739 * @ll_node: anchor in an llist (eg socket defer_list)
740 * @sk: Socket we are owned by
741 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
742 * fragmentation management
743 * @dev: Device we arrived on/are leaving by
744 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
745 * @cb: Control buffer. Free for use by every layer. Put private vars here
746 * @_skb_refdst: destination entry (with norefcount bit)
747 * @sp: the security path, used for xfrm
748 * @len: Length of actual data
749 * @data_len: Data length
750 * @mac_len: Length of link layer header
751 * @hdr_len: writable header length of cloned skb
752 * @csum: Checksum (must include start/offset pair)
753 * @csum_start: Offset from skb->head where checksumming should start
754 * @csum_offset: Offset from csum_start where checksum should be stored
755 * @priority: Packet queueing priority
756 * @ignore_df: allow local fragmentation
757 * @cloned: Head may be cloned (check refcnt to be sure)
758 * @ip_summed: Driver fed us an IP checksum
759 * @nohdr: Payload reference only, must not modify header
760 * @pkt_type: Packet class
761 * @fclone: skbuff clone status
762 * @ipvs_property: skbuff is owned by ipvs
763 * @inner_protocol_type: whether the inner protocol is
764 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
765 * @remcsum_offload: remote checksum offload is enabled
766 * @offload_fwd_mark: Packet was L2-forwarded in hardware
767 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
768 * @tc_skip_classify: do not classify packet. set by IFB device
769 * @tc_at_ingress: used within tc_classify to distinguish in/egress
770 * @redirected: packet was redirected by packet classifier
771 * @from_ingress: packet was redirected from the ingress path
772 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
773 * @peeked: this packet has been seen already, so stats have been
774 * done for it, don't do them again
775 * @nf_trace: netfilter packet trace flag
776 * @protocol: Packet protocol from driver
777 * @destructor: Destruct function
778 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
779 * @_sk_redir: socket redirection information for skmsg
780 * @_nfct: Associated connection, if any (with nfctinfo bits)
781 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
782 * @skb_iif: ifindex of device we arrived on
783 * @tc_index: Traffic control index
784 * @hash: the packet hash
785 * @queue_mapping: Queue mapping for multiqueue devices
786 * @head_frag: skb was allocated from page fragments,
787 * not allocated by kmalloc() or vmalloc().
788 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
789 * @pp_recycle: mark the packet for recycling instead of freeing (implies
790 * page_pool support on driver)
791 * @active_extensions: active extensions (skb_ext_id types)
792 * @ndisc_nodetype: router type (from link layer)
793 * @ooo_okay: allow the mapping of a socket to a queue to be changed
794 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
796 * @sw_hash: indicates hash was computed in software stack
797 * @wifi_acked_valid: wifi_acked was set
798 * @wifi_acked: whether frame was acked on wifi or not
799 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
800 * @encapsulation: indicates the inner headers in the skbuff are valid
801 * @encap_hdr_csum: software checksum is needed
802 * @csum_valid: checksum is already valid
803 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
804 * @csum_complete_sw: checksum was completed by software
805 * @csum_level: indicates the number of consecutive checksums found in
806 * the packet minus one that have been verified as
807 * CHECKSUM_UNNECESSARY (max 3)
808 * @dst_pending_confirm: need to confirm neighbour
809 * @decrypted: Decrypted SKB
810 * @slow_gro: state present at GRO time, slower prepare step required
811 * @mono_delivery_time: When set, skb->tstamp has the
812 * delivery_time in mono clock base (i.e. EDT). Otherwise, the
813 * skb->tstamp has the (rcv) timestamp at ingress and
814 * delivery_time at egress.
815 * @napi_id: id of the NAPI struct this skb came from
816 * @sender_cpu: (aka @napi_id) source CPU in XPS
817 * @alloc_cpu: CPU which did the skb allocation.
818 * @secmark: security marking
819 * @mark: Generic packet mark
820 * @reserved_tailroom: (aka @mark) number of bytes of free space available
821 * at the tail of an sk_buff
822 * @vlan_all: vlan fields (proto & tci)
823 * @vlan_proto: vlan encapsulation protocol
824 * @vlan_tci: vlan tag control information
825 * @inner_protocol: Protocol (encapsulation)
826 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
827 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
828 * @inner_transport_header: Inner transport layer header (encapsulation)
829 * @inner_network_header: Network layer header (encapsulation)
830 * @inner_mac_header: Link layer header (encapsulation)
831 * @transport_header: Transport layer header
832 * @network_header: Network layer header
833 * @mac_header: Link layer header
834 * @kcov_handle: KCOV remote handle for remote coverage collection
835 * @tail: Tail pointer
837 * @head: Head of buffer
838 * @data: Data head pointer
839 * @truesize: Buffer size
840 * @users: User count - see {datagram,tcp}.c
841 * @extensions: allocated extensions, valid if active_extensions is nonzero
847 /* These two members must be first to match sk_buff_head. */
848 struct sk_buff *next;
849 struct sk_buff *prev;
852 struct net_device *dev;
853 /* Some protocols might use this space to store information,
854 * while device pointer would be NULL.
855 * UDP receive path is one user.
857 unsigned long dev_scratch;
860 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
861 struct list_head list;
862 struct llist_node ll_node;
867 int ip_defrag_offset;
872 u64 skb_mstamp_ns; /* earliest departure time */
875 * This is the control buffer. It is free to use for every
876 * layer. Please put your private variables there. If you
877 * want to keep them across layers you have to do a skb_clone()
878 * first. This is owned by whoever has the skb queued ATM.
880 char cb[48] __aligned(8);
884 unsigned long _skb_refdst;
885 void (*destructor)(struct sk_buff *skb);
887 struct list_head tcp_tsorted_anchor;
888 #ifdef CONFIG_NET_SOCK_MSG
889 unsigned long _sk_redir;
893 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
901 /* Following fields are _not_ copied in __copy_skb_header()
902 * Note that queue_mapping is here mostly to fill a hole.
906 /* if you move cloned around you also must adapt those constants */
907 #ifdef __BIG_ENDIAN_BITFIELD
908 #define CLONED_MASK (1 << 7)
910 #define CLONED_MASK 1
912 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
915 __u8 __cloned_offset[0];
923 pp_recycle:1; /* page_pool recycle indicator */
924 #ifdef CONFIG_SKB_EXTENSIONS
925 __u8 active_extensions;
928 /* Fields enclosed in headers group are copied
929 * using a single memcpy() in __copy_skb_header()
931 struct_group(headers,
934 __u8 __pkt_type_offset[0];
936 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
938 __u8 dst_pending_confirm:1;
943 __u8 __mono_tc_offset[0];
945 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */
946 #ifdef CONFIG_NET_CLS_ACT
947 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
948 __u8 tc_skip_classify:1;
950 __u8 remcsum_offload:1;
951 __u8 csum_complete_sw:1;
953 __u8 inner_protocol_type:1;
957 #ifdef CONFIG_WIRELESS
958 __u8 wifi_acked_valid:1;
962 /* Indicates the inner headers are valid in the skbuff. */
963 __u8 encapsulation:1;
964 __u8 encap_hdr_csum:1;
966 #ifdef CONFIG_IPV6_NDISC_NODETYPE
967 __u8 ndisc_nodetype:2;
970 #if IS_ENABLED(CONFIG_IP_VS)
971 __u8 ipvs_property:1;
973 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
976 #ifdef CONFIG_NET_SWITCHDEV
977 __u8 offload_fwd_mark:1;
978 __u8 offload_l3_fwd_mark:1;
981 #ifdef CONFIG_NET_REDIRECT
984 #ifdef CONFIG_NETFILTER_SKIP_EGRESS
985 __u8 nf_skip_egress:1;
987 #ifdef CONFIG_TLS_DEVICE
991 #if IS_ENABLED(CONFIG_IP_SCTP)
992 __u8 csum_not_inet:1;
995 #ifdef CONFIG_NET_SCHED
996 __u16 tc_index; /* traffic control index */
1018 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
1020 unsigned int napi_id;
1021 unsigned int sender_cpu;
1024 #ifdef CONFIG_NETWORK_SECMARK
1030 __u32 reserved_tailroom;
1034 __be16 inner_protocol;
1038 __u16 inner_transport_header;
1039 __u16 inner_network_header;
1040 __u16 inner_mac_header;
1043 __u16 transport_header;
1044 __u16 network_header;
1051 ); /* end headers group */
1053 /* These elements must be at the end, see alloc_skb() for details. */
1054 sk_buff_data_t tail;
1056 unsigned char *head,
1058 unsigned int truesize;
1061 #ifdef CONFIG_SKB_EXTENSIONS
1062 /* only useable after checking ->active_extensions != 0 */
1063 struct skb_ext *extensions;
1067 /* if you move pkt_type around you also must adapt those constants */
1068 #ifdef __BIG_ENDIAN_BITFIELD
1069 #define PKT_TYPE_MAX (7 << 5)
1071 #define PKT_TYPE_MAX 7
1073 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1075 /* if you move tc_at_ingress or mono_delivery_time
1076 * around, you also must adapt these constants.
1078 #ifdef __BIG_ENDIAN_BITFIELD
1079 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 7)
1080 #define TC_AT_INGRESS_MASK (1 << 6)
1082 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 0)
1083 #define TC_AT_INGRESS_MASK (1 << 1)
1085 #define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset)
1089 * Handling routines are only of interest to the kernel
1092 #define SKB_ALLOC_FCLONE 0x01
1093 #define SKB_ALLOC_RX 0x02
1094 #define SKB_ALLOC_NAPI 0x04
1097 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1100 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1102 return unlikely(skb->pfmemalloc);
1106 * skb might have a dst pointer attached, refcounted or not.
1107 * _skb_refdst low order bit is set if refcount was _not_ taken
1109 #define SKB_DST_NOREF 1UL
1110 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1113 * skb_dst - returns skb dst_entry
1116 * Returns skb dst_entry, regardless of reference taken or not.
1118 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1120 /* If refdst was not refcounted, check we still are in a
1121 * rcu_read_lock section
1123 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1124 !rcu_read_lock_held() &&
1125 !rcu_read_lock_bh_held());
1126 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1130 * skb_dst_set - sets skb dst
1134 * Sets skb dst, assuming a reference was taken on dst and should
1135 * be released by skb_dst_drop()
1137 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1139 skb->slow_gro |= !!dst;
1140 skb->_skb_refdst = (unsigned long)dst;
1144 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1148 * Sets skb dst, assuming a reference was not taken on dst.
1149 * If dst entry is cached, we do not take reference and dst_release
1150 * will be avoided by refdst_drop. If dst entry is not cached, we take
1151 * reference, so that last dst_release can destroy the dst immediately.
1153 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1155 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1156 skb->slow_gro |= !!dst;
1157 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1161 * skb_dst_is_noref - Test if skb dst isn't refcounted
1164 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1166 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1170 * skb_rtable - Returns the skb &rtable
1173 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1175 return (struct rtable *)skb_dst(skb);
1178 /* For mangling skb->pkt_type from user space side from applications
1179 * such as nft, tc, etc, we only allow a conservative subset of
1180 * possible pkt_types to be set.
1182 static inline bool skb_pkt_type_ok(u32 ptype)
1184 return ptype <= PACKET_OTHERHOST;
1188 * skb_napi_id - Returns the skb's NAPI id
1191 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1193 #ifdef CONFIG_NET_RX_BUSY_POLL
1194 return skb->napi_id;
1200 static inline bool skb_wifi_acked_valid(const struct sk_buff *skb)
1202 #ifdef CONFIG_WIRELESS
1203 return skb->wifi_acked_valid;
1210 * skb_unref - decrement the skb's reference count
1213 * Returns true if we can free the skb.
1215 static inline bool skb_unref(struct sk_buff *skb)
1219 if (likely(refcount_read(&skb->users) == 1))
1221 else if (likely(!refcount_dec_and_test(&skb->users)))
1228 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason);
1231 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1232 * @skb: buffer to free
1234 static inline void kfree_skb(struct sk_buff *skb)
1236 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1239 void skb_release_head_state(struct sk_buff *skb);
1240 void kfree_skb_list_reason(struct sk_buff *segs,
1241 enum skb_drop_reason reason);
1242 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1243 void skb_tx_error(struct sk_buff *skb);
1245 static inline void kfree_skb_list(struct sk_buff *segs)
1247 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1250 #ifdef CONFIG_TRACEPOINTS
1251 void consume_skb(struct sk_buff *skb);
1253 static inline void consume_skb(struct sk_buff *skb)
1255 return kfree_skb(skb);
1259 void __consume_stateless_skb(struct sk_buff *skb);
1260 void __kfree_skb(struct sk_buff *skb);
1261 extern struct kmem_cache *skbuff_cache;
1263 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1264 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1265 bool *fragstolen, int *delta_truesize);
1267 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1269 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1270 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1271 struct sk_buff *build_skb_around(struct sk_buff *skb,
1272 void *data, unsigned int frag_size);
1273 void skb_attempt_defer_free(struct sk_buff *skb);
1275 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1276 struct sk_buff *slab_build_skb(void *data);
1279 * alloc_skb - allocate a network buffer
1280 * @size: size to allocate
1281 * @priority: allocation mask
1283 * This function is a convenient wrapper around __alloc_skb().
1285 static inline struct sk_buff *alloc_skb(unsigned int size,
1288 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1291 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1292 unsigned long data_len,
1296 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1298 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1299 struct sk_buff_fclones {
1300 struct sk_buff skb1;
1302 struct sk_buff skb2;
1304 refcount_t fclone_ref;
1308 * skb_fclone_busy - check if fclone is busy
1312 * Returns true if skb is a fast clone, and its clone is not freed.
1313 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1314 * so we also check that this didnt happen.
1316 static inline bool skb_fclone_busy(const struct sock *sk,
1317 const struct sk_buff *skb)
1319 const struct sk_buff_fclones *fclones;
1321 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1323 return skb->fclone == SKB_FCLONE_ORIG &&
1324 refcount_read(&fclones->fclone_ref) > 1 &&
1325 READ_ONCE(fclones->skb2.sk) == sk;
1329 * alloc_skb_fclone - allocate a network buffer from fclone cache
1330 * @size: size to allocate
1331 * @priority: allocation mask
1333 * This function is a convenient wrapper around __alloc_skb().
1335 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1338 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1341 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1342 void skb_headers_offset_update(struct sk_buff *skb, int off);
1343 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1344 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1345 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1346 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1347 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1348 gfp_t gfp_mask, bool fclone);
1349 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1352 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1355 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1356 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1357 unsigned int headroom);
1358 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1359 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1360 int newtailroom, gfp_t priority);
1361 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1362 int offset, int len);
1363 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1364 int offset, int len);
1365 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1366 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1369 * skb_pad - zero pad the tail of an skb
1370 * @skb: buffer to pad
1371 * @pad: space to pad
1373 * Ensure that a buffer is followed by a padding area that is zero
1374 * filled. Used by network drivers which may DMA or transfer data
1375 * beyond the buffer end onto the wire.
1377 * May return error in out of memory cases. The skb is freed on error.
1379 static inline int skb_pad(struct sk_buff *skb, int pad)
1381 return __skb_pad(skb, pad, true);
1383 #define dev_kfree_skb(a) consume_skb(a)
1385 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1386 int offset, size_t size);
1388 struct skb_seq_state {
1392 __u32 stepped_offset;
1393 struct sk_buff *root_skb;
1394 struct sk_buff *cur_skb;
1399 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1400 unsigned int to, struct skb_seq_state *st);
1401 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1402 struct skb_seq_state *st);
1403 void skb_abort_seq_read(struct skb_seq_state *st);
1405 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1406 unsigned int to, struct ts_config *config);
1409 * Packet hash types specify the type of hash in skb_set_hash.
1411 * Hash types refer to the protocol layer addresses which are used to
1412 * construct a packet's hash. The hashes are used to differentiate or identify
1413 * flows of the protocol layer for the hash type. Hash types are either
1414 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1416 * Properties of hashes:
1418 * 1) Two packets in different flows have different hash values
1419 * 2) Two packets in the same flow should have the same hash value
1421 * A hash at a higher layer is considered to be more specific. A driver should
1422 * set the most specific hash possible.
1424 * A driver cannot indicate a more specific hash than the layer at which a hash
1425 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1427 * A driver may indicate a hash level which is less specific than the
1428 * actual layer the hash was computed on. For instance, a hash computed
1429 * at L4 may be considered an L3 hash. This should only be done if the
1430 * driver can't unambiguously determine that the HW computed the hash at
1431 * the higher layer. Note that the "should" in the second property above
1434 enum pkt_hash_types {
1435 PKT_HASH_TYPE_NONE, /* Undefined type */
1436 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1437 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1438 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1441 static inline void skb_clear_hash(struct sk_buff *skb)
1448 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1451 skb_clear_hash(skb);
1455 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1457 skb->l4_hash = is_l4;
1458 skb->sw_hash = is_sw;
1463 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1465 /* Used by drivers to set hash from HW */
1466 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1470 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1472 __skb_set_hash(skb, hash, true, is_l4);
1475 void __skb_get_hash(struct sk_buff *skb);
1476 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1477 u32 skb_get_poff(const struct sk_buff *skb);
1478 u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1479 const struct flow_keys_basic *keys, int hlen);
1480 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1481 const void *data, int hlen_proto);
1483 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1484 int thoff, u8 ip_proto)
1486 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1489 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1490 const struct flow_dissector_key *key,
1491 unsigned int key_count);
1493 struct bpf_flow_dissector;
1494 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1495 __be16 proto, int nhoff, int hlen, unsigned int flags);
1497 bool __skb_flow_dissect(const struct net *net,
1498 const struct sk_buff *skb,
1499 struct flow_dissector *flow_dissector,
1500 void *target_container, const void *data,
1501 __be16 proto, int nhoff, int hlen, unsigned int flags);
1503 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1504 struct flow_dissector *flow_dissector,
1505 void *target_container, unsigned int flags)
1507 return __skb_flow_dissect(NULL, skb, flow_dissector,
1508 target_container, NULL, 0, 0, 0, flags);
1511 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1512 struct flow_keys *flow,
1515 memset(flow, 0, sizeof(*flow));
1516 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1517 flow, NULL, 0, 0, 0, flags);
1521 skb_flow_dissect_flow_keys_basic(const struct net *net,
1522 const struct sk_buff *skb,
1523 struct flow_keys_basic *flow,
1524 const void *data, __be16 proto,
1525 int nhoff, int hlen, unsigned int flags)
1527 memset(flow, 0, sizeof(*flow));
1528 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1529 data, proto, nhoff, hlen, flags);
1532 void skb_flow_dissect_meta(const struct sk_buff *skb,
1533 struct flow_dissector *flow_dissector,
1534 void *target_container);
1536 /* Gets a skb connection tracking info, ctinfo map should be a
1537 * map of mapsize to translate enum ip_conntrack_info states
1541 skb_flow_dissect_ct(const struct sk_buff *skb,
1542 struct flow_dissector *flow_dissector,
1543 void *target_container,
1544 u16 *ctinfo_map, size_t mapsize,
1545 bool post_ct, u16 zone);
1547 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1548 struct flow_dissector *flow_dissector,
1549 void *target_container);
1551 void skb_flow_dissect_hash(const struct sk_buff *skb,
1552 struct flow_dissector *flow_dissector,
1553 void *target_container);
1555 static inline __u32 skb_get_hash(struct sk_buff *skb)
1557 if (!skb->l4_hash && !skb->sw_hash)
1558 __skb_get_hash(skb);
1563 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1565 if (!skb->l4_hash && !skb->sw_hash) {
1566 struct flow_keys keys;
1567 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1569 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1575 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1576 const siphash_key_t *perturb);
1578 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1583 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1585 to->hash = from->hash;
1586 to->sw_hash = from->sw_hash;
1587 to->l4_hash = from->l4_hash;
1590 static inline void skb_copy_decrypted(struct sk_buff *to,
1591 const struct sk_buff *from)
1593 #ifdef CONFIG_TLS_DEVICE
1594 to->decrypted = from->decrypted;
1598 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1599 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1601 return skb->head + skb->end;
1604 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1609 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1614 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1619 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1621 return skb->end - skb->head;
1624 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1626 skb->end = skb->head + offset;
1630 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1631 struct ubuf_info *uarg);
1633 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1635 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1638 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
1639 struct sk_buff *skb, struct iov_iter *from,
1642 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1643 struct msghdr *msg, int len)
1645 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len);
1648 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1649 struct msghdr *msg, int len,
1650 struct ubuf_info *uarg);
1653 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1655 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1657 return &skb_shinfo(skb)->hwtstamps;
1660 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1662 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1664 return is_zcopy ? skb_uarg(skb) : NULL;
1667 static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1669 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1672 static inline bool skb_zcopy_managed(const struct sk_buff *skb)
1674 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS;
1677 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1678 const struct sk_buff *skb2)
1680 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1683 static inline void net_zcopy_get(struct ubuf_info *uarg)
1685 refcount_inc(&uarg->refcnt);
1688 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1690 skb_shinfo(skb)->destructor_arg = uarg;
1691 skb_shinfo(skb)->flags |= uarg->flags;
1694 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1697 if (skb && uarg && !skb_zcopy(skb)) {
1698 if (unlikely(have_ref && *have_ref))
1701 net_zcopy_get(uarg);
1702 skb_zcopy_init(skb, uarg);
1706 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1708 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1709 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1712 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1714 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1717 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1719 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1722 static inline void net_zcopy_put(struct ubuf_info *uarg)
1725 uarg->callback(NULL, uarg, true);
1728 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1731 if (uarg->callback == msg_zerocopy_callback)
1732 msg_zerocopy_put_abort(uarg, have_uref);
1734 net_zcopy_put(uarg);
1738 /* Release a reference on a zerocopy structure */
1739 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1741 struct ubuf_info *uarg = skb_zcopy(skb);
1744 if (!skb_zcopy_is_nouarg(skb))
1745 uarg->callback(skb, uarg, zerocopy_success);
1747 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1751 void __skb_zcopy_downgrade_managed(struct sk_buff *skb);
1753 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb)
1755 if (unlikely(skb_zcopy_managed(skb)))
1756 __skb_zcopy_downgrade_managed(skb);
1759 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1764 static inline void skb_poison_list(struct sk_buff *skb)
1766 #ifdef CONFIG_DEBUG_NET
1767 skb->next = SKB_LIST_POISON_NEXT;
1771 /* Iterate through singly-linked GSO fragments of an skb. */
1772 #define skb_list_walk_safe(first, skb, next_skb) \
1773 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1774 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1776 static inline void skb_list_del_init(struct sk_buff *skb)
1778 __list_del_entry(&skb->list);
1779 skb_mark_not_on_list(skb);
1783 * skb_queue_empty - check if a queue is empty
1786 * Returns true if the queue is empty, false otherwise.
1788 static inline int skb_queue_empty(const struct sk_buff_head *list)
1790 return list->next == (const struct sk_buff *) list;
1794 * skb_queue_empty_lockless - check if a queue is empty
1797 * Returns true if the queue is empty, false otherwise.
1798 * This variant can be used in lockless contexts.
1800 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1802 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1807 * skb_queue_is_last - check if skb is the last entry in the queue
1811 * Returns true if @skb is the last buffer on the list.
1813 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1814 const struct sk_buff *skb)
1816 return skb->next == (const struct sk_buff *) list;
1820 * skb_queue_is_first - check if skb is the first entry in the queue
1824 * Returns true if @skb is the first buffer on the list.
1826 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1827 const struct sk_buff *skb)
1829 return skb->prev == (const struct sk_buff *) list;
1833 * skb_queue_next - return the next packet in the queue
1835 * @skb: current buffer
1837 * Return the next packet in @list after @skb. It is only valid to
1838 * call this if skb_queue_is_last() evaluates to false.
1840 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1841 const struct sk_buff *skb)
1843 /* This BUG_ON may seem severe, but if we just return then we
1844 * are going to dereference garbage.
1846 BUG_ON(skb_queue_is_last(list, skb));
1851 * skb_queue_prev - return the prev packet in the queue
1853 * @skb: current buffer
1855 * Return the prev packet in @list before @skb. It is only valid to
1856 * call this if skb_queue_is_first() evaluates to false.
1858 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1859 const struct sk_buff *skb)
1861 /* This BUG_ON may seem severe, but if we just return then we
1862 * are going to dereference garbage.
1864 BUG_ON(skb_queue_is_first(list, skb));
1869 * skb_get - reference buffer
1870 * @skb: buffer to reference
1872 * Makes another reference to a socket buffer and returns a pointer
1875 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1877 refcount_inc(&skb->users);
1882 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1886 * skb_cloned - is the buffer a clone
1887 * @skb: buffer to check
1889 * Returns true if the buffer was generated with skb_clone() and is
1890 * one of multiple shared copies of the buffer. Cloned buffers are
1891 * shared data so must not be written to under normal circumstances.
1893 static inline int skb_cloned(const struct sk_buff *skb)
1895 return skb->cloned &&
1896 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1899 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1901 might_sleep_if(gfpflags_allow_blocking(pri));
1903 if (skb_cloned(skb))
1904 return pskb_expand_head(skb, 0, 0, pri);
1909 /* This variant of skb_unclone() makes sure skb->truesize
1910 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
1912 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
1913 * when various debugging features are in place.
1915 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
1916 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1918 might_sleep_if(gfpflags_allow_blocking(pri));
1920 if (skb_cloned(skb))
1921 return __skb_unclone_keeptruesize(skb, pri);
1926 * skb_header_cloned - is the header a clone
1927 * @skb: buffer to check
1929 * Returns true if modifying the header part of the buffer requires
1930 * the data to be copied.
1932 static inline int skb_header_cloned(const struct sk_buff *skb)
1939 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1940 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1941 return dataref != 1;
1944 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1946 might_sleep_if(gfpflags_allow_blocking(pri));
1948 if (skb_header_cloned(skb))
1949 return pskb_expand_head(skb, 0, 0, pri);
1955 * __skb_header_release() - allow clones to use the headroom
1956 * @skb: buffer to operate on
1958 * See "DOC: dataref and headerless skbs".
1960 static inline void __skb_header_release(struct sk_buff *skb)
1963 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1968 * skb_shared - is the buffer shared
1969 * @skb: buffer to check
1971 * Returns true if more than one person has a reference to this
1974 static inline int skb_shared(const struct sk_buff *skb)
1976 return refcount_read(&skb->users) != 1;
1980 * skb_share_check - check if buffer is shared and if so clone it
1981 * @skb: buffer to check
1982 * @pri: priority for memory allocation
1984 * If the buffer is shared the buffer is cloned and the old copy
1985 * drops a reference. A new clone with a single reference is returned.
1986 * If the buffer is not shared the original buffer is returned. When
1987 * being called from interrupt status or with spinlocks held pri must
1990 * NULL is returned on a memory allocation failure.
1992 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1994 might_sleep_if(gfpflags_allow_blocking(pri));
1995 if (skb_shared(skb)) {
1996 struct sk_buff *nskb = skb_clone(skb, pri);
2008 * Copy shared buffers into a new sk_buff. We effectively do COW on
2009 * packets to handle cases where we have a local reader and forward
2010 * and a couple of other messy ones. The normal one is tcpdumping
2011 * a packet thats being forwarded.
2015 * skb_unshare - make a copy of a shared buffer
2016 * @skb: buffer to check
2017 * @pri: priority for memory allocation
2019 * If the socket buffer is a clone then this function creates a new
2020 * copy of the data, drops a reference count on the old copy and returns
2021 * the new copy with the reference count at 1. If the buffer is not a clone
2022 * the original buffer is returned. When called with a spinlock held or
2023 * from interrupt state @pri must be %GFP_ATOMIC
2025 * %NULL is returned on a memory allocation failure.
2027 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
2030 might_sleep_if(gfpflags_allow_blocking(pri));
2031 if (skb_cloned(skb)) {
2032 struct sk_buff *nskb = skb_copy(skb, pri);
2034 /* Free our shared copy */
2045 * skb_peek - peek at the head of an &sk_buff_head
2046 * @list_: list to peek at
2048 * Peek an &sk_buff. Unlike most other operations you _MUST_
2049 * be careful with this one. A peek leaves the buffer on the
2050 * list and someone else may run off with it. You must hold
2051 * the appropriate locks or have a private queue to do this.
2053 * Returns %NULL for an empty list or a pointer to the head element.
2054 * The reference count is not incremented and the reference is therefore
2055 * volatile. Use with caution.
2057 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2059 struct sk_buff *skb = list_->next;
2061 if (skb == (struct sk_buff *)list_)
2067 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2068 * @list_: list to peek at
2070 * Like skb_peek(), but the caller knows that the list is not empty.
2072 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2078 * skb_peek_next - peek skb following the given one from a queue
2079 * @skb: skb to start from
2080 * @list_: list to peek at
2082 * Returns %NULL when the end of the list is met or a pointer to the
2083 * next element. The reference count is not incremented and the
2084 * reference is therefore volatile. Use with caution.
2086 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2087 const struct sk_buff_head *list_)
2089 struct sk_buff *next = skb->next;
2091 if (next == (struct sk_buff *)list_)
2097 * skb_peek_tail - peek at the tail of an &sk_buff_head
2098 * @list_: list to peek at
2100 * Peek an &sk_buff. Unlike most other operations you _MUST_
2101 * be careful with this one. A peek leaves the buffer on the
2102 * list and someone else may run off with it. You must hold
2103 * the appropriate locks or have a private queue to do this.
2105 * Returns %NULL for an empty list or a pointer to the tail element.
2106 * The reference count is not incremented and the reference is therefore
2107 * volatile. Use with caution.
2109 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2111 struct sk_buff *skb = READ_ONCE(list_->prev);
2113 if (skb == (struct sk_buff *)list_)
2120 * skb_queue_len - get queue length
2121 * @list_: list to measure
2123 * Return the length of an &sk_buff queue.
2125 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2131 * skb_queue_len_lockless - get queue length
2132 * @list_: list to measure
2134 * Return the length of an &sk_buff queue.
2135 * This variant can be used in lockless contexts.
2137 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2139 return READ_ONCE(list_->qlen);
2143 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2144 * @list: queue to initialize
2146 * This initializes only the list and queue length aspects of
2147 * an sk_buff_head object. This allows to initialize the list
2148 * aspects of an sk_buff_head without reinitializing things like
2149 * the spinlock. It can also be used for on-stack sk_buff_head
2150 * objects where the spinlock is known to not be used.
2152 static inline void __skb_queue_head_init(struct sk_buff_head *list)
2154 list->prev = list->next = (struct sk_buff *)list;
2159 * This function creates a split out lock class for each invocation;
2160 * this is needed for now since a whole lot of users of the skb-queue
2161 * infrastructure in drivers have different locking usage (in hardirq)
2162 * than the networking core (in softirq only). In the long run either the
2163 * network layer or drivers should need annotation to consolidate the
2164 * main types of usage into 3 classes.
2166 static inline void skb_queue_head_init(struct sk_buff_head *list)
2168 spin_lock_init(&list->lock);
2169 __skb_queue_head_init(list);
2172 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2173 struct lock_class_key *class)
2175 skb_queue_head_init(list);
2176 lockdep_set_class(&list->lock, class);
2180 * Insert an sk_buff on a list.
2182 * The "__skb_xxxx()" functions are the non-atomic ones that
2183 * can only be called with interrupts disabled.
2185 static inline void __skb_insert(struct sk_buff *newsk,
2186 struct sk_buff *prev, struct sk_buff *next,
2187 struct sk_buff_head *list)
2189 /* See skb_queue_empty_lockless() and skb_peek_tail()
2190 * for the opposite READ_ONCE()
2192 WRITE_ONCE(newsk->next, next);
2193 WRITE_ONCE(newsk->prev, prev);
2194 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2195 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2196 WRITE_ONCE(list->qlen, list->qlen + 1);
2199 static inline void __skb_queue_splice(const struct sk_buff_head *list,
2200 struct sk_buff *prev,
2201 struct sk_buff *next)
2203 struct sk_buff *first = list->next;
2204 struct sk_buff *last = list->prev;
2206 WRITE_ONCE(first->prev, prev);
2207 WRITE_ONCE(prev->next, first);
2209 WRITE_ONCE(last->next, next);
2210 WRITE_ONCE(next->prev, last);
2214 * skb_queue_splice - join two skb lists, this is designed for stacks
2215 * @list: the new list to add
2216 * @head: the place to add it in the first list
2218 static inline void skb_queue_splice(const struct sk_buff_head *list,
2219 struct sk_buff_head *head)
2221 if (!skb_queue_empty(list)) {
2222 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2223 head->qlen += list->qlen;
2228 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2229 * @list: the new list to add
2230 * @head: the place to add it in the first list
2232 * The list at @list is reinitialised
2234 static inline void skb_queue_splice_init(struct sk_buff_head *list,
2235 struct sk_buff_head *head)
2237 if (!skb_queue_empty(list)) {
2238 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2239 head->qlen += list->qlen;
2240 __skb_queue_head_init(list);
2245 * skb_queue_splice_tail - join two skb lists, each list being a queue
2246 * @list: the new list to add
2247 * @head: the place to add it in the first list
2249 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2250 struct sk_buff_head *head)
2252 if (!skb_queue_empty(list)) {
2253 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2254 head->qlen += list->qlen;
2259 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2260 * @list: the new list to add
2261 * @head: the place to add it in the first list
2263 * Each of the lists is a queue.
2264 * The list at @list is reinitialised
2266 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2267 struct sk_buff_head *head)
2269 if (!skb_queue_empty(list)) {
2270 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2271 head->qlen += list->qlen;
2272 __skb_queue_head_init(list);
2277 * __skb_queue_after - queue a buffer at the list head
2278 * @list: list to use
2279 * @prev: place after this buffer
2280 * @newsk: buffer to queue
2282 * Queue a buffer int the middle of a list. This function takes no locks
2283 * and you must therefore hold required locks before calling it.
2285 * A buffer cannot be placed on two lists at the same time.
2287 static inline void __skb_queue_after(struct sk_buff_head *list,
2288 struct sk_buff *prev,
2289 struct sk_buff *newsk)
2291 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2294 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2295 struct sk_buff_head *list);
2297 static inline void __skb_queue_before(struct sk_buff_head *list,
2298 struct sk_buff *next,
2299 struct sk_buff *newsk)
2301 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2305 * __skb_queue_head - queue a buffer at the list head
2306 * @list: list to use
2307 * @newsk: buffer to queue
2309 * Queue a buffer at the start of a list. This function takes no locks
2310 * and you must therefore hold required locks before calling it.
2312 * A buffer cannot be placed on two lists at the same time.
2314 static inline void __skb_queue_head(struct sk_buff_head *list,
2315 struct sk_buff *newsk)
2317 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2319 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2322 * __skb_queue_tail - queue a buffer at the list tail
2323 * @list: list to use
2324 * @newsk: buffer to queue
2326 * Queue a buffer at the end of a list. This function takes no locks
2327 * and you must therefore hold required locks before calling it.
2329 * A buffer cannot be placed on two lists at the same time.
2331 static inline void __skb_queue_tail(struct sk_buff_head *list,
2332 struct sk_buff *newsk)
2334 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2336 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2339 * remove sk_buff from list. _Must_ be called atomically, and with
2342 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2343 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2345 struct sk_buff *next, *prev;
2347 WRITE_ONCE(list->qlen, list->qlen - 1);
2350 skb->next = skb->prev = NULL;
2351 WRITE_ONCE(next->prev, prev);
2352 WRITE_ONCE(prev->next, next);
2356 * __skb_dequeue - remove from the head of the queue
2357 * @list: list to dequeue from
2359 * Remove the head of the list. This function does not take any locks
2360 * so must be used with appropriate locks held only. The head item is
2361 * returned or %NULL if the list is empty.
2363 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2365 struct sk_buff *skb = skb_peek(list);
2367 __skb_unlink(skb, list);
2370 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2373 * __skb_dequeue_tail - remove from the tail of the queue
2374 * @list: list to dequeue from
2376 * Remove the tail of the list. This function does not take any locks
2377 * so must be used with appropriate locks held only. The tail item is
2378 * returned or %NULL if the list is empty.
2380 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2382 struct sk_buff *skb = skb_peek_tail(list);
2384 __skb_unlink(skb, list);
2387 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2390 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2392 return skb->data_len;
2395 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2397 return skb->len - skb->data_len;
2400 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2402 unsigned int i, len = 0;
2404 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2405 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2409 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2411 return skb_headlen(skb) + __skb_pagelen(skb);
2414 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo,
2415 int i, struct page *page,
2418 skb_frag_t *frag = &shinfo->frags[i];
2421 * Propagate page pfmemalloc to the skb if we can. The problem is
2422 * that not all callers have unique ownership of the page but rely
2423 * on page_is_pfmemalloc doing the right thing(tm).
2425 frag->bv_page = page;
2426 frag->bv_offset = off;
2427 skb_frag_size_set(frag, size);
2431 * skb_len_add - adds a number to len fields of skb
2432 * @skb: buffer to add len to
2433 * @delta: number of bytes to add
2435 static inline void skb_len_add(struct sk_buff *skb, int delta)
2438 skb->data_len += delta;
2439 skb->truesize += delta;
2443 * __skb_fill_page_desc - initialise a paged fragment in an skb
2444 * @skb: buffer containing fragment to be initialised
2445 * @i: paged fragment index to initialise
2446 * @page: the page to use for this fragment
2447 * @off: the offset to the data with @page
2448 * @size: the length of the data
2450 * Initialises the @i'th fragment of @skb to point to &size bytes at
2451 * offset @off within @page.
2453 * Does not take any additional reference on the fragment.
2455 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2456 struct page *page, int off, int size)
2458 __skb_fill_page_desc_noacc(skb_shinfo(skb), i, page, off, size);
2459 page = compound_head(page);
2460 if (page_is_pfmemalloc(page))
2461 skb->pfmemalloc = true;
2465 * skb_fill_page_desc - initialise a paged fragment in an skb
2466 * @skb: buffer containing fragment to be initialised
2467 * @i: paged fragment index to initialise
2468 * @page: the page to use for this fragment
2469 * @off: the offset to the data with @page
2470 * @size: the length of the data
2472 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2473 * @skb to point to @size bytes at offset @off within @page. In
2474 * addition updates @skb such that @i is the last fragment.
2476 * Does not take any additional reference on the fragment.
2478 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2479 struct page *page, int off, int size)
2481 __skb_fill_page_desc(skb, i, page, off, size);
2482 skb_shinfo(skb)->nr_frags = i + 1;
2486 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb
2487 * @skb: buffer containing fragment to be initialised
2488 * @i: paged fragment index to initialise
2489 * @page: the page to use for this fragment
2490 * @off: the offset to the data with @page
2491 * @size: the length of the data
2493 * Variant of skb_fill_page_desc() which does not deal with
2494 * pfmemalloc, if page is not owned by us.
2496 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i,
2497 struct page *page, int off,
2500 struct skb_shared_info *shinfo = skb_shinfo(skb);
2502 __skb_fill_page_desc_noacc(shinfo, i, page, off, size);
2503 shinfo->nr_frags = i + 1;
2506 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2507 int size, unsigned int truesize);
2509 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2510 unsigned int truesize);
2512 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2514 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2515 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2517 return skb->head + skb->tail;
2520 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2522 skb->tail = skb->data - skb->head;
2525 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2527 skb_reset_tail_pointer(skb);
2528 skb->tail += offset;
2531 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2532 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2537 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2539 skb->tail = skb->data;
2542 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2544 skb->tail = skb->data + offset;
2547 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2549 static inline void skb_assert_len(struct sk_buff *skb)
2551 #ifdef CONFIG_DEBUG_NET
2552 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2553 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2554 #endif /* CONFIG_DEBUG_NET */
2558 * Add data to an sk_buff
2560 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2561 void *skb_put(struct sk_buff *skb, unsigned int len);
2562 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2564 void *tmp = skb_tail_pointer(skb);
2565 SKB_LINEAR_ASSERT(skb);
2571 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2573 void *tmp = __skb_put(skb, len);
2575 memset(tmp, 0, len);
2579 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2582 void *tmp = __skb_put(skb, len);
2584 memcpy(tmp, data, len);
2588 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2590 *(u8 *)__skb_put(skb, 1) = val;
2593 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2595 void *tmp = skb_put(skb, len);
2597 memset(tmp, 0, len);
2602 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2605 void *tmp = skb_put(skb, len);
2607 memcpy(tmp, data, len);
2612 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2614 *(u8 *)skb_put(skb, 1) = val;
2617 void *skb_push(struct sk_buff *skb, unsigned int len);
2618 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2625 void *skb_pull(struct sk_buff *skb, unsigned int len);
2626 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2629 if (unlikely(skb->len < skb->data_len)) {
2630 #if defined(CONFIG_DEBUG_NET)
2632 pr_err("__skb_pull(len=%u)\n", len);
2633 skb_dump(KERN_ERR, skb, false);
2637 return skb->data += len;
2640 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2642 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2645 void *skb_pull_data(struct sk_buff *skb, size_t len);
2647 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2649 static inline enum skb_drop_reason
2650 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len)
2652 if (likely(len <= skb_headlen(skb)))
2653 return SKB_NOT_DROPPED_YET;
2655 if (unlikely(len > skb->len))
2656 return SKB_DROP_REASON_PKT_TOO_SMALL;
2658 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb))))
2659 return SKB_DROP_REASON_NOMEM;
2661 return SKB_NOT_DROPPED_YET;
2664 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2666 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET;
2669 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2671 if (!pskb_may_pull(skb, len))
2675 return skb->data += len;
2678 void skb_condense(struct sk_buff *skb);
2681 * skb_headroom - bytes at buffer head
2682 * @skb: buffer to check
2684 * Return the number of bytes of free space at the head of an &sk_buff.
2686 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2688 return skb->data - skb->head;
2692 * skb_tailroom - bytes at buffer end
2693 * @skb: buffer to check
2695 * Return the number of bytes of free space at the tail of an sk_buff
2697 static inline int skb_tailroom(const struct sk_buff *skb)
2699 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2703 * skb_availroom - bytes at buffer end
2704 * @skb: buffer to check
2706 * Return the number of bytes of free space at the tail of an sk_buff
2707 * allocated by sk_stream_alloc()
2709 static inline int skb_availroom(const struct sk_buff *skb)
2711 if (skb_is_nonlinear(skb))
2714 return skb->end - skb->tail - skb->reserved_tailroom;
2718 * skb_reserve - adjust headroom
2719 * @skb: buffer to alter
2720 * @len: bytes to move
2722 * Increase the headroom of an empty &sk_buff by reducing the tail
2723 * room. This is only allowed for an empty buffer.
2725 static inline void skb_reserve(struct sk_buff *skb, int len)
2732 * skb_tailroom_reserve - adjust reserved_tailroom
2733 * @skb: buffer to alter
2734 * @mtu: maximum amount of headlen permitted
2735 * @needed_tailroom: minimum amount of reserved_tailroom
2737 * Set reserved_tailroom so that headlen can be as large as possible but
2738 * not larger than mtu and tailroom cannot be smaller than
2740 * The required headroom should already have been reserved before using
2743 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2744 unsigned int needed_tailroom)
2746 SKB_LINEAR_ASSERT(skb);
2747 if (mtu < skb_tailroom(skb) - needed_tailroom)
2748 /* use at most mtu */
2749 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2751 /* use up to all available space */
2752 skb->reserved_tailroom = needed_tailroom;
2755 #define ENCAP_TYPE_ETHER 0
2756 #define ENCAP_TYPE_IPPROTO 1
2758 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2761 skb->inner_protocol = protocol;
2762 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2765 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2768 skb->inner_ipproto = ipproto;
2769 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2772 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2774 skb->inner_mac_header = skb->mac_header;
2775 skb->inner_network_header = skb->network_header;
2776 skb->inner_transport_header = skb->transport_header;
2779 static inline void skb_reset_mac_len(struct sk_buff *skb)
2781 skb->mac_len = skb->network_header - skb->mac_header;
2784 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2787 return skb->head + skb->inner_transport_header;
2790 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2792 return skb_inner_transport_header(skb) - skb->data;
2795 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2797 skb->inner_transport_header = skb->data - skb->head;
2800 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2803 skb_reset_inner_transport_header(skb);
2804 skb->inner_transport_header += offset;
2807 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2809 return skb->head + skb->inner_network_header;
2812 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2814 skb->inner_network_header = skb->data - skb->head;
2817 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2820 skb_reset_inner_network_header(skb);
2821 skb->inner_network_header += offset;
2824 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2826 return skb->head + skb->inner_mac_header;
2829 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2831 skb->inner_mac_header = skb->data - skb->head;
2834 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2837 skb_reset_inner_mac_header(skb);
2838 skb->inner_mac_header += offset;
2840 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2842 return skb->transport_header != (typeof(skb->transport_header))~0U;
2845 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2847 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
2848 return skb->head + skb->transport_header;
2851 static inline void skb_reset_transport_header(struct sk_buff *skb)
2853 skb->transport_header = skb->data - skb->head;
2856 static inline void skb_set_transport_header(struct sk_buff *skb,
2859 skb_reset_transport_header(skb);
2860 skb->transport_header += offset;
2863 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2865 return skb->head + skb->network_header;
2868 static inline void skb_reset_network_header(struct sk_buff *skb)
2870 skb->network_header = skb->data - skb->head;
2873 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2875 skb_reset_network_header(skb);
2876 skb->network_header += offset;
2879 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2881 return skb->mac_header != (typeof(skb->mac_header))~0U;
2884 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2886 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2887 return skb->head + skb->mac_header;
2890 static inline int skb_mac_offset(const struct sk_buff *skb)
2892 return skb_mac_header(skb) - skb->data;
2895 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2897 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2898 return skb->network_header - skb->mac_header;
2901 static inline void skb_unset_mac_header(struct sk_buff *skb)
2903 skb->mac_header = (typeof(skb->mac_header))~0U;
2906 static inline void skb_reset_mac_header(struct sk_buff *skb)
2908 skb->mac_header = skb->data - skb->head;
2911 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2913 skb_reset_mac_header(skb);
2914 skb->mac_header += offset;
2917 static inline void skb_pop_mac_header(struct sk_buff *skb)
2919 skb->mac_header = skb->network_header;
2922 static inline void skb_probe_transport_header(struct sk_buff *skb)
2924 struct flow_keys_basic keys;
2926 if (skb_transport_header_was_set(skb))
2929 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2931 skb_set_transport_header(skb, keys.control.thoff);
2934 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2936 if (skb_mac_header_was_set(skb)) {
2937 const unsigned char *old_mac = skb_mac_header(skb);
2939 skb_set_mac_header(skb, -skb->mac_len);
2940 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2944 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2946 return skb->csum_start - skb_headroom(skb);
2949 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2951 return skb->head + skb->csum_start;
2954 static inline int skb_transport_offset(const struct sk_buff *skb)
2956 return skb_transport_header(skb) - skb->data;
2959 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2961 return skb->transport_header - skb->network_header;
2964 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2966 return skb->inner_transport_header - skb->inner_network_header;
2969 static inline int skb_network_offset(const struct sk_buff *skb)
2971 return skb_network_header(skb) - skb->data;
2974 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2976 return skb_inner_network_header(skb) - skb->data;
2979 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2981 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2985 * CPUs often take a performance hit when accessing unaligned memory
2986 * locations. The actual performance hit varies, it can be small if the
2987 * hardware handles it or large if we have to take an exception and fix it
2990 * Since an ethernet header is 14 bytes network drivers often end up with
2991 * the IP header at an unaligned offset. The IP header can be aligned by
2992 * shifting the start of the packet by 2 bytes. Drivers should do this
2995 * skb_reserve(skb, NET_IP_ALIGN);
2997 * The downside to this alignment of the IP header is that the DMA is now
2998 * unaligned. On some architectures the cost of an unaligned DMA is high
2999 * and this cost outweighs the gains made by aligning the IP header.
3001 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
3004 #ifndef NET_IP_ALIGN
3005 #define NET_IP_ALIGN 2
3009 * The networking layer reserves some headroom in skb data (via
3010 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
3011 * the header has to grow. In the default case, if the header has to grow
3012 * 32 bytes or less we avoid the reallocation.
3014 * Unfortunately this headroom changes the DMA alignment of the resulting
3015 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
3016 * on some architectures. An architecture can override this value,
3017 * perhaps setting it to a cacheline in size (since that will maintain
3018 * cacheline alignment of the DMA). It must be a power of 2.
3020 * Various parts of the networking layer expect at least 32 bytes of
3021 * headroom, you should not reduce this.
3023 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
3024 * to reduce average number of cache lines per packet.
3025 * get_rps_cpu() for example only access one 64 bytes aligned block :
3026 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
3029 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
3032 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
3034 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
3036 if (WARN_ON(skb_is_nonlinear(skb)))
3039 skb_set_tail_pointer(skb, len);
3042 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
3044 __skb_set_length(skb, len);
3047 void skb_trim(struct sk_buff *skb, unsigned int len);
3049 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
3052 return ___pskb_trim(skb, len);
3053 __skb_trim(skb, len);
3057 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
3059 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
3063 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
3064 * @skb: buffer to alter
3067 * This is identical to pskb_trim except that the caller knows that
3068 * the skb is not cloned so we should never get an error due to out-
3071 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3073 int err = pskb_trim(skb, len);
3077 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3079 unsigned int diff = len - skb->len;
3081 if (skb_tailroom(skb) < diff) {
3082 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3087 __skb_set_length(skb, len);
3092 * skb_orphan - orphan a buffer
3093 * @skb: buffer to orphan
3095 * If a buffer currently has an owner then we call the owner's
3096 * destructor function and make the @skb unowned. The buffer continues
3097 * to exist but is no longer charged to its former owner.
3099 static inline void skb_orphan(struct sk_buff *skb)
3101 if (skb->destructor) {
3102 skb->destructor(skb);
3103 skb->destructor = NULL;
3111 * skb_orphan_frags - orphan the frags contained in a buffer
3112 * @skb: buffer to orphan frags from
3113 * @gfp_mask: allocation mask for replacement pages
3115 * For each frag in the SKB which needs a destructor (i.e. has an
3116 * owner) create a copy of that frag and release the original
3117 * page by calling the destructor.
3119 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3121 if (likely(!skb_zcopy(skb)))
3123 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN)
3125 return skb_copy_ubufs(skb, gfp_mask);
3128 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3129 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3131 if (likely(!skb_zcopy(skb)))
3133 return skb_copy_ubufs(skb, gfp_mask);
3137 * __skb_queue_purge - empty a list
3138 * @list: list to empty
3140 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3141 * the list and one reference dropped. This function does not take the
3142 * list lock and the caller must hold the relevant locks to use it.
3144 static inline void __skb_queue_purge(struct sk_buff_head *list)
3146 struct sk_buff *skb;
3147 while ((skb = __skb_dequeue(list)) != NULL)
3150 void skb_queue_purge(struct sk_buff_head *list);
3152 unsigned int skb_rbtree_purge(struct rb_root *root);
3154 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3157 * netdev_alloc_frag - allocate a page fragment
3158 * @fragsz: fragment size
3160 * Allocates a frag from a page for receive buffer.
3161 * Uses GFP_ATOMIC allocations.
3163 static inline void *netdev_alloc_frag(unsigned int fragsz)
3165 return __netdev_alloc_frag_align(fragsz, ~0u);
3168 static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3171 WARN_ON_ONCE(!is_power_of_2(align));
3172 return __netdev_alloc_frag_align(fragsz, -align);
3175 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3179 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3180 * @dev: network device to receive on
3181 * @length: length to allocate
3183 * Allocate a new &sk_buff and assign it a usage count of one. The
3184 * buffer has unspecified headroom built in. Users should allocate
3185 * the headroom they think they need without accounting for the
3186 * built in space. The built in space is used for optimisations.
3188 * %NULL is returned if there is no free memory. Although this function
3189 * allocates memory it can be called from an interrupt.
3191 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3192 unsigned int length)
3194 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3197 /* legacy helper around __netdev_alloc_skb() */
3198 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3201 return __netdev_alloc_skb(NULL, length, gfp_mask);
3204 /* legacy helper around netdev_alloc_skb() */
3205 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3207 return netdev_alloc_skb(NULL, length);
3211 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3212 unsigned int length, gfp_t gfp)
3214 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3216 if (NET_IP_ALIGN && skb)
3217 skb_reserve(skb, NET_IP_ALIGN);
3221 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3222 unsigned int length)
3224 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3227 static inline void skb_free_frag(void *addr)
3229 page_frag_free(addr);
3232 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3234 static inline void *napi_alloc_frag(unsigned int fragsz)
3236 return __napi_alloc_frag_align(fragsz, ~0u);
3239 static inline void *napi_alloc_frag_align(unsigned int fragsz,
3242 WARN_ON_ONCE(!is_power_of_2(align));
3243 return __napi_alloc_frag_align(fragsz, -align);
3246 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
3247 unsigned int length, gfp_t gfp_mask);
3248 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
3249 unsigned int length)
3251 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
3253 void napi_consume_skb(struct sk_buff *skb, int budget);
3255 void napi_skb_free_stolen_head(struct sk_buff *skb);
3256 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason);
3259 * __dev_alloc_pages - allocate page for network Rx
3260 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3261 * @order: size of the allocation
3263 * Allocate a new page.
3265 * %NULL is returned if there is no free memory.
3267 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
3270 /* This piece of code contains several assumptions.
3271 * 1. This is for device Rx, therefor a cold page is preferred.
3272 * 2. The expectation is the user wants a compound page.
3273 * 3. If requesting a order 0 page it will not be compound
3274 * due to the check to see if order has a value in prep_new_page
3275 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3276 * code in gfp_to_alloc_flags that should be enforcing this.
3278 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3280 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
3283 static inline struct page *dev_alloc_pages(unsigned int order)
3285 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
3289 * __dev_alloc_page - allocate a page for network Rx
3290 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3292 * Allocate a new page.
3294 * %NULL is returned if there is no free memory.
3296 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
3298 return __dev_alloc_pages(gfp_mask, 0);
3301 static inline struct page *dev_alloc_page(void)
3303 return dev_alloc_pages(0);
3307 * dev_page_is_reusable - check whether a page can be reused for network Rx
3308 * @page: the page to test
3310 * A page shouldn't be considered for reusing/recycling if it was allocated
3311 * under memory pressure or at a distant memory node.
3313 * Returns false if this page should be returned to page allocator, true
3316 static inline bool dev_page_is_reusable(const struct page *page)
3318 return likely(page_to_nid(page) == numa_mem_id() &&
3319 !page_is_pfmemalloc(page));
3323 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3324 * @page: The page that was allocated from skb_alloc_page
3325 * @skb: The skb that may need pfmemalloc set
3327 static inline void skb_propagate_pfmemalloc(const struct page *page,
3328 struct sk_buff *skb)
3330 if (page_is_pfmemalloc(page))
3331 skb->pfmemalloc = true;
3335 * skb_frag_off() - Returns the offset of a skb fragment
3336 * @frag: the paged fragment
3338 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3340 return frag->bv_offset;
3344 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3345 * @frag: skb fragment
3346 * @delta: value to add
3348 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3350 frag->bv_offset += delta;
3354 * skb_frag_off_set() - Sets the offset of a skb fragment
3355 * @frag: skb fragment
3356 * @offset: offset of fragment
3358 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3360 frag->bv_offset = offset;
3364 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3365 * @fragto: skb fragment where offset is set
3366 * @fragfrom: skb fragment offset is copied from
3368 static inline void skb_frag_off_copy(skb_frag_t *fragto,
3369 const skb_frag_t *fragfrom)
3371 fragto->bv_offset = fragfrom->bv_offset;
3375 * skb_frag_page - retrieve the page referred to by a paged fragment
3376 * @frag: the paged fragment
3378 * Returns the &struct page associated with @frag.
3380 static inline struct page *skb_frag_page(const skb_frag_t *frag)
3382 return frag->bv_page;
3386 * __skb_frag_ref - take an addition reference on a paged fragment.
3387 * @frag: the paged fragment
3389 * Takes an additional reference on the paged fragment @frag.
3391 static inline void __skb_frag_ref(skb_frag_t *frag)
3393 get_page(skb_frag_page(frag));
3397 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
3399 * @f: the fragment offset.
3401 * Takes an additional reference on the @f'th paged fragment of @skb.
3403 static inline void skb_frag_ref(struct sk_buff *skb, int f)
3405 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
3409 napi_frag_unref(skb_frag_t *frag, bool recycle, bool napi_safe)
3411 struct page *page = skb_frag_page(frag);
3413 #ifdef CONFIG_PAGE_POOL
3414 if (recycle && page_pool_return_skb_page(page, napi_safe))
3421 * __skb_frag_unref - release a reference on a paged fragment.
3422 * @frag: the paged fragment
3423 * @recycle: recycle the page if allocated via page_pool
3425 * Releases a reference on the paged fragment @frag
3426 * or recycles the page via the page_pool API.
3428 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle)
3430 napi_frag_unref(frag, recycle, false);
3434 * skb_frag_unref - release a reference on a paged fragment of an skb.
3436 * @f: the fragment offset
3438 * Releases a reference on the @f'th paged fragment of @skb.
3440 static inline void skb_frag_unref(struct sk_buff *skb, int f)
3442 struct skb_shared_info *shinfo = skb_shinfo(skb);
3444 if (!skb_zcopy_managed(skb))
3445 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle);
3449 * skb_frag_address - gets the address of the data contained in a paged fragment
3450 * @frag: the paged fragment buffer
3452 * Returns the address of the data within @frag. The page must already
3455 static inline void *skb_frag_address(const skb_frag_t *frag)
3457 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3461 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3462 * @frag: the paged fragment buffer
3464 * Returns the address of the data within @frag. Checks that the page
3465 * is mapped and returns %NULL otherwise.
3467 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3469 void *ptr = page_address(skb_frag_page(frag));
3473 return ptr + skb_frag_off(frag);
3477 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3478 * @fragto: skb fragment where page is set
3479 * @fragfrom: skb fragment page is copied from
3481 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3482 const skb_frag_t *fragfrom)
3484 fragto->bv_page = fragfrom->bv_page;
3488 * __skb_frag_set_page - sets the page contained in a paged fragment
3489 * @frag: the paged fragment
3490 * @page: the page to set
3492 * Sets the fragment @frag to contain @page.
3494 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3496 frag->bv_page = page;
3500 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3502 * @f: the fragment offset
3503 * @page: the page to set
3505 * Sets the @f'th fragment of @skb to contain @page.
3507 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3510 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3513 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3516 * skb_frag_dma_map - maps a paged fragment via the DMA API
3517 * @dev: the device to map the fragment to
3518 * @frag: the paged fragment to map
3519 * @offset: the offset within the fragment (starting at the
3520 * fragment's own offset)
3521 * @size: the number of bytes to map
3522 * @dir: the direction of the mapping (``PCI_DMA_*``)
3524 * Maps the page associated with @frag to @device.
3526 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3527 const skb_frag_t *frag,
3528 size_t offset, size_t size,
3529 enum dma_data_direction dir)
3531 return dma_map_page(dev, skb_frag_page(frag),
3532 skb_frag_off(frag) + offset, size, dir);
3535 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3538 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3542 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3545 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3550 * skb_clone_writable - is the header of a clone writable
3551 * @skb: buffer to check
3552 * @len: length up to which to write
3554 * Returns true if modifying the header part of the cloned buffer
3555 * does not requires the data to be copied.
3557 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3559 return !skb_header_cloned(skb) &&
3560 skb_headroom(skb) + len <= skb->hdr_len;
3563 static inline int skb_try_make_writable(struct sk_buff *skb,
3564 unsigned int write_len)
3566 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3567 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3570 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3575 if (headroom > skb_headroom(skb))
3576 delta = headroom - skb_headroom(skb);
3578 if (delta || cloned)
3579 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3585 * skb_cow - copy header of skb when it is required
3586 * @skb: buffer to cow
3587 * @headroom: needed headroom
3589 * If the skb passed lacks sufficient headroom or its data part
3590 * is shared, data is reallocated. If reallocation fails, an error
3591 * is returned and original skb is not changed.
3593 * The result is skb with writable area skb->head...skb->tail
3594 * and at least @headroom of space at head.
3596 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3598 return __skb_cow(skb, headroom, skb_cloned(skb));
3602 * skb_cow_head - skb_cow but only making the head writable
3603 * @skb: buffer to cow
3604 * @headroom: needed headroom
3606 * This function is identical to skb_cow except that we replace the
3607 * skb_cloned check by skb_header_cloned. It should be used when
3608 * you only need to push on some header and do not need to modify
3611 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3613 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3617 * skb_padto - pad an skbuff up to a minimal size
3618 * @skb: buffer to pad
3619 * @len: minimal length
3621 * Pads up a buffer to ensure the trailing bytes exist and are
3622 * blanked. If the buffer already contains sufficient data it
3623 * is untouched. Otherwise it is extended. Returns zero on
3624 * success. The skb is freed on error.
3626 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3628 unsigned int size = skb->len;
3629 if (likely(size >= len))
3631 return skb_pad(skb, len - size);
3635 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3636 * @skb: buffer to pad
3637 * @len: minimal length
3638 * @free_on_error: free buffer on error
3640 * Pads up a buffer to ensure the trailing bytes exist and are
3641 * blanked. If the buffer already contains sufficient data it
3642 * is untouched. Otherwise it is extended. Returns zero on
3643 * success. The skb is freed on error if @free_on_error is true.
3645 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3649 unsigned int size = skb->len;
3651 if (unlikely(size < len)) {
3653 if (__skb_pad(skb, len, free_on_error))
3655 __skb_put(skb, len);
3661 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3662 * @skb: buffer to pad
3663 * @len: minimal length
3665 * Pads up a buffer to ensure the trailing bytes exist and are
3666 * blanked. If the buffer already contains sufficient data it
3667 * is untouched. Otherwise it is extended. Returns zero on
3668 * success. The skb is freed on error.
3670 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3672 return __skb_put_padto(skb, len, true);
3675 static inline int skb_add_data(struct sk_buff *skb,
3676 struct iov_iter *from, int copy)
3678 const int off = skb->len;
3680 if (skb->ip_summed == CHECKSUM_NONE) {
3682 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3684 skb->csum = csum_block_add(skb->csum, csum, off);
3687 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3690 __skb_trim(skb, off);
3694 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3695 const struct page *page, int off)
3700 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3702 return page == skb_frag_page(frag) &&
3703 off == skb_frag_off(frag) + skb_frag_size(frag);
3708 static inline int __skb_linearize(struct sk_buff *skb)
3710 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3714 * skb_linearize - convert paged skb to linear one
3715 * @skb: buffer to linarize
3717 * If there is no free memory -ENOMEM is returned, otherwise zero
3718 * is returned and the old skb data released.
3720 static inline int skb_linearize(struct sk_buff *skb)
3722 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3726 * skb_has_shared_frag - can any frag be overwritten
3727 * @skb: buffer to test
3729 * Return true if the skb has at least one frag that might be modified
3730 * by an external entity (as in vmsplice()/sendfile())
3732 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3734 return skb_is_nonlinear(skb) &&
3735 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3739 * skb_linearize_cow - make sure skb is linear and writable
3740 * @skb: buffer to process
3742 * If there is no free memory -ENOMEM is returned, otherwise zero
3743 * is returned and the old skb data released.
3745 static inline int skb_linearize_cow(struct sk_buff *skb)
3747 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3748 __skb_linearize(skb) : 0;
3751 static __always_inline void
3752 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3755 if (skb->ip_summed == CHECKSUM_COMPLETE)
3756 skb->csum = csum_block_sub(skb->csum,
3757 csum_partial(start, len, 0), off);
3758 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3759 skb_checksum_start_offset(skb) < 0)
3760 skb->ip_summed = CHECKSUM_NONE;
3764 * skb_postpull_rcsum - update checksum for received skb after pull
3765 * @skb: buffer to update
3766 * @start: start of data before pull
3767 * @len: length of data pulled
3769 * After doing a pull on a received packet, you need to call this to
3770 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3771 * CHECKSUM_NONE so that it can be recomputed from scratch.
3773 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3774 const void *start, unsigned int len)
3776 if (skb->ip_summed == CHECKSUM_COMPLETE)
3777 skb->csum = wsum_negate(csum_partial(start, len,
3778 wsum_negate(skb->csum)));
3779 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3780 skb_checksum_start_offset(skb) < 0)
3781 skb->ip_summed = CHECKSUM_NONE;
3784 static __always_inline void
3785 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3788 if (skb->ip_summed == CHECKSUM_COMPLETE)
3789 skb->csum = csum_block_add(skb->csum,
3790 csum_partial(start, len, 0), off);
3794 * skb_postpush_rcsum - update checksum for received skb after push
3795 * @skb: buffer to update
3796 * @start: start of data after push
3797 * @len: length of data pushed
3799 * After doing a push on a received packet, you need to call this to
3800 * update the CHECKSUM_COMPLETE checksum.
3802 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3803 const void *start, unsigned int len)
3805 __skb_postpush_rcsum(skb, start, len, 0);
3808 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3811 * skb_push_rcsum - push skb and update receive checksum
3812 * @skb: buffer to update
3813 * @len: length of data pulled
3815 * This function performs an skb_push on the packet and updates
3816 * the CHECKSUM_COMPLETE checksum. It should be used on
3817 * receive path processing instead of skb_push unless you know
3818 * that the checksum difference is zero (e.g., a valid IP header)
3819 * or you are setting ip_summed to CHECKSUM_NONE.
3821 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3824 skb_postpush_rcsum(skb, skb->data, len);
3828 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3830 * pskb_trim_rcsum - trim received skb and update checksum
3831 * @skb: buffer to trim
3834 * This is exactly the same as pskb_trim except that it ensures the
3835 * checksum of received packets are still valid after the operation.
3836 * It can change skb pointers.
3839 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3841 if (likely(len >= skb->len))
3843 return pskb_trim_rcsum_slow(skb, len);
3846 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3848 if (skb->ip_summed == CHECKSUM_COMPLETE)
3849 skb->ip_summed = CHECKSUM_NONE;
3850 __skb_trim(skb, len);
3854 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3856 if (skb->ip_summed == CHECKSUM_COMPLETE)
3857 skb->ip_summed = CHECKSUM_NONE;
3858 return __skb_grow(skb, len);
3861 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3862 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3863 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3864 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3865 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3867 #define skb_queue_walk(queue, skb) \
3868 for (skb = (queue)->next; \
3869 skb != (struct sk_buff *)(queue); \
3872 #define skb_queue_walk_safe(queue, skb, tmp) \
3873 for (skb = (queue)->next, tmp = skb->next; \
3874 skb != (struct sk_buff *)(queue); \
3875 skb = tmp, tmp = skb->next)
3877 #define skb_queue_walk_from(queue, skb) \
3878 for (; skb != (struct sk_buff *)(queue); \
3881 #define skb_rbtree_walk(skb, root) \
3882 for (skb = skb_rb_first(root); skb != NULL; \
3883 skb = skb_rb_next(skb))
3885 #define skb_rbtree_walk_from(skb) \
3886 for (; skb != NULL; \
3887 skb = skb_rb_next(skb))
3889 #define skb_rbtree_walk_from_safe(skb, tmp) \
3890 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3893 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3894 for (tmp = skb->next; \
3895 skb != (struct sk_buff *)(queue); \
3896 skb = tmp, tmp = skb->next)
3898 #define skb_queue_reverse_walk(queue, skb) \
3899 for (skb = (queue)->prev; \
3900 skb != (struct sk_buff *)(queue); \
3903 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3904 for (skb = (queue)->prev, tmp = skb->prev; \
3905 skb != (struct sk_buff *)(queue); \
3906 skb = tmp, tmp = skb->prev)
3908 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3909 for (tmp = skb->prev; \
3910 skb != (struct sk_buff *)(queue); \
3911 skb = tmp, tmp = skb->prev)
3913 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3915 return skb_shinfo(skb)->frag_list != NULL;
3918 static inline void skb_frag_list_init(struct sk_buff *skb)
3920 skb_shinfo(skb)->frag_list = NULL;
3923 #define skb_walk_frags(skb, iter) \
3924 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3927 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3928 int *err, long *timeo_p,
3929 const struct sk_buff *skb);
3930 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3931 struct sk_buff_head *queue,
3934 struct sk_buff **last);
3935 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
3936 struct sk_buff_head *queue,
3937 unsigned int flags, int *off, int *err,
3938 struct sk_buff **last);
3939 struct sk_buff *__skb_recv_datagram(struct sock *sk,
3940 struct sk_buff_head *sk_queue,
3941 unsigned int flags, int *off, int *err);
3942 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
3943 __poll_t datagram_poll(struct file *file, struct socket *sock,
3944 struct poll_table_struct *wait);
3945 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3946 struct iov_iter *to, int size);
3947 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3948 struct msghdr *msg, int size)
3950 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3952 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3953 struct msghdr *msg);
3954 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3955 struct iov_iter *to, int len,
3956 struct ahash_request *hash);
3957 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3958 struct iov_iter *from, int len);
3959 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3960 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3961 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3962 static inline void skb_free_datagram_locked(struct sock *sk,
3963 struct sk_buff *skb)
3965 __skb_free_datagram_locked(sk, skb, 0);
3967 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3968 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3969 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3970 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3972 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3973 struct pipe_inode_info *pipe, unsigned int len,
3974 unsigned int flags);
3975 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3977 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3978 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3979 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3980 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3982 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3983 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3984 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3985 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3986 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3987 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3988 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
3989 unsigned int offset);
3990 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3991 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
3992 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3993 int skb_vlan_pop(struct sk_buff *skb);
3994 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3995 int skb_eth_pop(struct sk_buff *skb);
3996 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
3997 const unsigned char *src);
3998 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
3999 int mac_len, bool ethernet);
4000 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
4002 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
4003 int skb_mpls_dec_ttl(struct sk_buff *skb);
4004 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
4007 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
4009 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
4012 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
4014 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
4017 struct skb_checksum_ops {
4018 __wsum (*update)(const void *mem, int len, __wsum wsum);
4019 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
4022 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
4024 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
4025 __wsum csum, const struct skb_checksum_ops *ops);
4026 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
4029 static inline void * __must_check
4030 __skb_header_pointer(const struct sk_buff *skb, int offset, int len,
4031 const void *data, int hlen, void *buffer)
4033 if (likely(hlen - offset >= len))
4034 return (void *)data + offset;
4036 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
4042 static inline void * __must_check
4043 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
4045 return __skb_header_pointer(skb, offset, len, skb->data,
4046 skb_headlen(skb), buffer);
4050 * skb_needs_linearize - check if we need to linearize a given skb
4051 * depending on the given device features.
4052 * @skb: socket buffer to check
4053 * @features: net device features
4055 * Returns true if either:
4056 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
4057 * 2. skb is fragmented and the device does not support SG.
4059 static inline bool skb_needs_linearize(struct sk_buff *skb,
4060 netdev_features_t features)
4062 return skb_is_nonlinear(skb) &&
4063 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
4064 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
4067 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
4069 const unsigned int len)
4071 memcpy(to, skb->data, len);
4074 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4075 const int offset, void *to,
4076 const unsigned int len)
4078 memcpy(to, skb->data + offset, len);
4081 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4083 const unsigned int len)
4085 memcpy(skb->data, from, len);
4088 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4091 const unsigned int len)
4093 memcpy(skb->data + offset, from, len);
4096 void skb_init(void);
4098 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4104 * skb_get_timestamp - get timestamp from a skb
4105 * @skb: skb to get stamp from
4106 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4108 * Timestamps are stored in the skb as offsets to a base timestamp.
4109 * This function converts the offset back to a struct timeval and stores
4112 static inline void skb_get_timestamp(const struct sk_buff *skb,
4113 struct __kernel_old_timeval *stamp)
4115 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4118 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4119 struct __kernel_sock_timeval *stamp)
4121 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4123 stamp->tv_sec = ts.tv_sec;
4124 stamp->tv_usec = ts.tv_nsec / 1000;
4127 static inline void skb_get_timestampns(const struct sk_buff *skb,
4128 struct __kernel_old_timespec *stamp)
4130 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4132 stamp->tv_sec = ts.tv_sec;
4133 stamp->tv_nsec = ts.tv_nsec;
4136 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4137 struct __kernel_timespec *stamp)
4139 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4141 stamp->tv_sec = ts.tv_sec;
4142 stamp->tv_nsec = ts.tv_nsec;
4145 static inline void __net_timestamp(struct sk_buff *skb)
4147 skb->tstamp = ktime_get_real();
4148 skb->mono_delivery_time = 0;
4151 static inline ktime_t net_timedelta(ktime_t t)
4153 return ktime_sub(ktime_get_real(), t);
4156 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4160 skb->mono_delivery_time = kt && mono;
4163 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4165 /* It is used in the ingress path to clear the delivery_time.
4166 * If needed, set the skb->tstamp to the (rcv) timestamp.
4168 static inline void skb_clear_delivery_time(struct sk_buff *skb)
4170 if (skb->mono_delivery_time) {
4171 skb->mono_delivery_time = 0;
4172 if (static_branch_unlikely(&netstamp_needed_key))
4173 skb->tstamp = ktime_get_real();
4179 static inline void skb_clear_tstamp(struct sk_buff *skb)
4181 if (skb->mono_delivery_time)
4187 static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4189 if (skb->mono_delivery_time)
4195 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4197 if (!skb->mono_delivery_time && skb->tstamp)
4200 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4201 return ktime_get_real();
4206 static inline u8 skb_metadata_len(const struct sk_buff *skb)
4208 return skb_shinfo(skb)->meta_len;
4211 static inline void *skb_metadata_end(const struct sk_buff *skb)
4213 return skb_mac_header(skb);
4216 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4217 const struct sk_buff *skb_b,
4220 const void *a = skb_metadata_end(skb_a);
4221 const void *b = skb_metadata_end(skb_b);
4222 /* Using more efficient varaiant than plain call to memcmp(). */
4223 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
4227 #define __it(x, op) (x -= sizeof(u##op))
4228 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4229 case 32: diffs |= __it_diff(a, b, 64);
4231 case 24: diffs |= __it_diff(a, b, 64);
4233 case 16: diffs |= __it_diff(a, b, 64);
4235 case 8: diffs |= __it_diff(a, b, 64);
4237 case 28: diffs |= __it_diff(a, b, 64);
4239 case 20: diffs |= __it_diff(a, b, 64);
4241 case 12: diffs |= __it_diff(a, b, 64);
4243 case 4: diffs |= __it_diff(a, b, 32);
4248 return memcmp(a - meta_len, b - meta_len, meta_len);
4252 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4253 const struct sk_buff *skb_b)
4255 u8 len_a = skb_metadata_len(skb_a);
4256 u8 len_b = skb_metadata_len(skb_b);
4258 if (!(len_a | len_b))
4261 return len_a != len_b ?
4262 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4265 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4267 skb_shinfo(skb)->meta_len = meta_len;
4270 static inline void skb_metadata_clear(struct sk_buff *skb)
4272 skb_metadata_set(skb, 0);
4275 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4277 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4279 void skb_clone_tx_timestamp(struct sk_buff *skb);
4280 bool skb_defer_rx_timestamp(struct sk_buff *skb);
4282 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4284 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4288 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4293 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4296 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4298 * PHY drivers may accept clones of transmitted packets for
4299 * timestamping via their phy_driver.txtstamp method. These drivers
4300 * must call this function to return the skb back to the stack with a
4303 * @skb: clone of the original outgoing packet
4304 * @hwtstamps: hardware time stamps
4307 void skb_complete_tx_timestamp(struct sk_buff *skb,
4308 struct skb_shared_hwtstamps *hwtstamps);
4310 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4311 struct skb_shared_hwtstamps *hwtstamps,
4312 struct sock *sk, int tstype);
4315 * skb_tstamp_tx - queue clone of skb with send time stamps
4316 * @orig_skb: the original outgoing packet
4317 * @hwtstamps: hardware time stamps, may be NULL if not available
4319 * If the skb has a socket associated, then this function clones the
4320 * skb (thus sharing the actual data and optional structures), stores
4321 * the optional hardware time stamping information (if non NULL) or
4322 * generates a software time stamp (otherwise), then queues the clone
4323 * to the error queue of the socket. Errors are silently ignored.
4325 void skb_tstamp_tx(struct sk_buff *orig_skb,
4326 struct skb_shared_hwtstamps *hwtstamps);
4329 * skb_tx_timestamp() - Driver hook for transmit timestamping
4331 * Ethernet MAC Drivers should call this function in their hard_xmit()
4332 * function immediately before giving the sk_buff to the MAC hardware.
4334 * Specifically, one should make absolutely sure that this function is
4335 * called before TX completion of this packet can trigger. Otherwise
4336 * the packet could potentially already be freed.
4338 * @skb: A socket buffer.
4340 static inline void skb_tx_timestamp(struct sk_buff *skb)
4342 skb_clone_tx_timestamp(skb);
4343 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4344 skb_tstamp_tx(skb, NULL);
4348 * skb_complete_wifi_ack - deliver skb with wifi status
4350 * @skb: the original outgoing packet
4351 * @acked: ack status
4354 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4356 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4357 __sum16 __skb_checksum_complete(struct sk_buff *skb);
4359 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4361 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4363 (skb->ip_summed == CHECKSUM_PARTIAL &&
4364 skb_checksum_start_offset(skb) >= 0));
4368 * skb_checksum_complete - Calculate checksum of an entire packet
4369 * @skb: packet to process
4371 * This function calculates the checksum over the entire packet plus
4372 * the value of skb->csum. The latter can be used to supply the
4373 * checksum of a pseudo header as used by TCP/UDP. It returns the
4376 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4377 * this function can be used to verify that checksum on received
4378 * packets. In that case the function should return zero if the
4379 * checksum is correct. In particular, this function will return zero
4380 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4381 * hardware has already verified the correctness of the checksum.
4383 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4385 return skb_csum_unnecessary(skb) ?
4386 0 : __skb_checksum_complete(skb);
4389 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4391 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4392 if (skb->csum_level == 0)
4393 skb->ip_summed = CHECKSUM_NONE;
4399 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4401 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4402 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4404 } else if (skb->ip_summed == CHECKSUM_NONE) {
4405 skb->ip_summed = CHECKSUM_UNNECESSARY;
4406 skb->csum_level = 0;
4410 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4412 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4413 skb->ip_summed = CHECKSUM_NONE;
4414 skb->csum_level = 0;
4418 /* Check if we need to perform checksum complete validation.
4420 * Returns true if checksum complete is needed, false otherwise
4421 * (either checksum is unnecessary or zero checksum is allowed).
4423 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4427 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4428 skb->csum_valid = 1;
4429 __skb_decr_checksum_unnecessary(skb);
4436 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4439 #define CHECKSUM_BREAK 76
4441 /* Unset checksum-complete
4443 * Unset checksum complete can be done when packet is being modified
4444 * (uncompressed for instance) and checksum-complete value is
4447 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4449 if (skb->ip_summed == CHECKSUM_COMPLETE)
4450 skb->ip_summed = CHECKSUM_NONE;
4453 /* Validate (init) checksum based on checksum complete.
4456 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4457 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4458 * checksum is stored in skb->csum for use in __skb_checksum_complete
4459 * non-zero: value of invalid checksum
4462 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4466 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4467 if (!csum_fold(csum_add(psum, skb->csum))) {
4468 skb->csum_valid = 1;
4475 if (complete || skb->len <= CHECKSUM_BREAK) {
4478 csum = __skb_checksum_complete(skb);
4479 skb->csum_valid = !csum;
4486 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4491 /* Perform checksum validate (init). Note that this is a macro since we only
4492 * want to calculate the pseudo header which is an input function if necessary.
4493 * First we try to validate without any computation (checksum unnecessary) and
4494 * then calculate based on checksum complete calling the function to compute
4498 * 0: checksum is validated or try to in skb_checksum_complete
4499 * non-zero: value of invalid checksum
4501 #define __skb_checksum_validate(skb, proto, complete, \
4502 zero_okay, check, compute_pseudo) \
4504 __sum16 __ret = 0; \
4505 skb->csum_valid = 0; \
4506 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4507 __ret = __skb_checksum_validate_complete(skb, \
4508 complete, compute_pseudo(skb, proto)); \
4512 #define skb_checksum_init(skb, proto, compute_pseudo) \
4513 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4515 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4516 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4518 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4519 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4521 #define skb_checksum_validate_zero_check(skb, proto, check, \
4523 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4525 #define skb_checksum_simple_validate(skb) \
4526 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4528 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4530 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4533 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4535 skb->csum = ~pseudo;
4536 skb->ip_summed = CHECKSUM_COMPLETE;
4539 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4541 if (__skb_checksum_convert_check(skb)) \
4542 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4545 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4546 u16 start, u16 offset)
4548 skb->ip_summed = CHECKSUM_PARTIAL;
4549 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4550 skb->csum_offset = offset - start;
4553 /* Update skbuf and packet to reflect the remote checksum offload operation.
4554 * When called, ptr indicates the starting point for skb->csum when
4555 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4556 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4558 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4559 int start, int offset, bool nopartial)
4564 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4568 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4569 __skb_checksum_complete(skb);
4570 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4573 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4575 /* Adjust skb->csum since we changed the packet */
4576 skb->csum = csum_add(skb->csum, delta);
4579 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4581 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4582 return (void *)(skb->_nfct & NFCT_PTRMASK);
4588 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4590 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4597 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4599 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4600 skb->slow_gro |= !!nfct;
4605 #ifdef CONFIG_SKB_EXTENSIONS
4607 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4613 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4616 #if IS_ENABLED(CONFIG_MPTCP)
4619 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4622 SKB_EXT_NUM, /* must be last */
4626 * struct skb_ext - sk_buff extensions
4627 * @refcnt: 1 on allocation, deallocated on 0
4628 * @offset: offset to add to @data to obtain extension address
4629 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4630 * @data: start of extension data, variable sized
4632 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4633 * to use 'u8' types while allowing up to 2kb worth of extension data.
4637 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4638 u8 chunks; /* same */
4639 char data[] __aligned(8);
4642 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4643 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4644 struct skb_ext *ext);
4645 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4646 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4647 void __skb_ext_put(struct skb_ext *ext);
4649 static inline void skb_ext_put(struct sk_buff *skb)
4651 if (skb->active_extensions)
4652 __skb_ext_put(skb->extensions);
4655 static inline void __skb_ext_copy(struct sk_buff *dst,
4656 const struct sk_buff *src)
4658 dst->active_extensions = src->active_extensions;
4660 if (src->active_extensions) {
4661 struct skb_ext *ext = src->extensions;
4663 refcount_inc(&ext->refcnt);
4664 dst->extensions = ext;
4668 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4671 __skb_ext_copy(dst, src);
4674 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4676 return !!ext->offset[i];
4679 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4681 return skb->active_extensions & (1 << id);
4684 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4686 if (skb_ext_exist(skb, id))
4687 __skb_ext_del(skb, id);
4690 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4692 if (skb_ext_exist(skb, id)) {
4693 struct skb_ext *ext = skb->extensions;
4695 return (void *)ext + (ext->offset[id] << 3);
4701 static inline void skb_ext_reset(struct sk_buff *skb)
4703 if (unlikely(skb->active_extensions)) {
4704 __skb_ext_put(skb->extensions);
4705 skb->active_extensions = 0;
4709 static inline bool skb_has_extensions(struct sk_buff *skb)
4711 return unlikely(skb->active_extensions);
4714 static inline void skb_ext_put(struct sk_buff *skb) {}
4715 static inline void skb_ext_reset(struct sk_buff *skb) {}
4716 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4717 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4718 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4719 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4720 #endif /* CONFIG_SKB_EXTENSIONS */
4722 static inline void nf_reset_ct(struct sk_buff *skb)
4724 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4725 nf_conntrack_put(skb_nfct(skb));
4730 static inline void nf_reset_trace(struct sk_buff *skb)
4732 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4737 static inline void ipvs_reset(struct sk_buff *skb)
4739 #if IS_ENABLED(CONFIG_IP_VS)
4740 skb->ipvs_property = 0;
4744 /* Note: This doesn't put any conntrack info in dst. */
4745 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4748 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4749 dst->_nfct = src->_nfct;
4750 nf_conntrack_get(skb_nfct(src));
4752 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4754 dst->nf_trace = src->nf_trace;
4758 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4760 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4761 nf_conntrack_put(skb_nfct(dst));
4763 dst->slow_gro = src->slow_gro;
4764 __nf_copy(dst, src, true);
4767 #ifdef CONFIG_NETWORK_SECMARK
4768 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4770 to->secmark = from->secmark;
4773 static inline void skb_init_secmark(struct sk_buff *skb)
4778 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4781 static inline void skb_init_secmark(struct sk_buff *skb)
4785 static inline int secpath_exists(const struct sk_buff *skb)
4788 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4794 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4796 return !skb->destructor &&
4797 !secpath_exists(skb) &&
4799 !skb->_skb_refdst &&
4800 !skb_has_frag_list(skb);
4803 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4805 skb->queue_mapping = queue_mapping;
4808 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4810 return skb->queue_mapping;
4813 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4815 to->queue_mapping = from->queue_mapping;
4818 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4820 skb->queue_mapping = rx_queue + 1;
4823 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4825 return skb->queue_mapping - 1;
4828 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4830 return skb->queue_mapping != 0;
4833 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4835 skb->dst_pending_confirm = val;
4838 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4840 return skb->dst_pending_confirm != 0;
4843 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4846 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4852 /* Keeps track of mac header offset relative to skb->head.
4853 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4854 * For non-tunnel skb it points to skb_mac_header() and for
4855 * tunnel skb it points to outer mac header.
4856 * Keeps track of level of encapsulation of network headers.
4867 #define SKB_GSO_CB_OFFSET 32
4868 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET))
4870 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4872 return (skb_mac_header(inner_skb) - inner_skb->head) -
4873 SKB_GSO_CB(inner_skb)->mac_offset;
4876 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4878 int new_headroom, headroom;
4881 headroom = skb_headroom(skb);
4882 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4886 new_headroom = skb_headroom(skb);
4887 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4891 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4893 /* Do not update partial checksums if remote checksum is enabled. */
4894 if (skb->remcsum_offload)
4897 SKB_GSO_CB(skb)->csum = res;
4898 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4901 /* Compute the checksum for a gso segment. First compute the checksum value
4902 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4903 * then add in skb->csum (checksum from csum_start to end of packet).
4904 * skb->csum and csum_start are then updated to reflect the checksum of the
4905 * resultant packet starting from the transport header-- the resultant checksum
4906 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4909 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4911 unsigned char *csum_start = skb_transport_header(skb);
4912 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4913 __wsum partial = SKB_GSO_CB(skb)->csum;
4915 SKB_GSO_CB(skb)->csum = res;
4916 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4918 return csum_fold(csum_partial(csum_start, plen, partial));
4921 static inline bool skb_is_gso(const struct sk_buff *skb)
4923 return skb_shinfo(skb)->gso_size;
4926 /* Note: Should be called only if skb_is_gso(skb) is true */
4927 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4929 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4932 /* Note: Should be called only if skb_is_gso(skb) is true */
4933 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4935 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4938 /* Note: Should be called only if skb_is_gso(skb) is true */
4939 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4941 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4944 static inline void skb_gso_reset(struct sk_buff *skb)
4946 skb_shinfo(skb)->gso_size = 0;
4947 skb_shinfo(skb)->gso_segs = 0;
4948 skb_shinfo(skb)->gso_type = 0;
4951 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4954 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4956 shinfo->gso_size += increment;
4959 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4962 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4964 shinfo->gso_size -= decrement;
4967 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4969 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4971 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4972 * wanted then gso_type will be set. */
4973 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4975 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4976 unlikely(shinfo->gso_type == 0)) {
4977 __skb_warn_lro_forwarding(skb);
4983 static inline void skb_forward_csum(struct sk_buff *skb)
4985 /* Unfortunately we don't support this one. Any brave souls? */
4986 if (skb->ip_summed == CHECKSUM_COMPLETE)
4987 skb->ip_summed = CHECKSUM_NONE;
4991 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4992 * @skb: skb to check
4994 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4995 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4996 * use this helper, to document places where we make this assertion.
4998 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
5000 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE);
5003 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
5005 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
5006 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5007 unsigned int transport_len,
5008 __sum16(*skb_chkf)(struct sk_buff *skb));
5011 * skb_head_is_locked - Determine if the skb->head is locked down
5012 * @skb: skb to check
5014 * The head on skbs build around a head frag can be removed if they are
5015 * not cloned. This function returns true if the skb head is locked down
5016 * due to either being allocated via kmalloc, or by being a clone with
5017 * multiple references to the head.
5019 static inline bool skb_head_is_locked(const struct sk_buff *skb)
5021 return !skb->head_frag || skb_cloned(skb);
5024 /* Local Checksum Offload.
5025 * Compute outer checksum based on the assumption that the
5026 * inner checksum will be offloaded later.
5027 * See Documentation/networking/checksum-offloads.rst for
5028 * explanation of how this works.
5029 * Fill in outer checksum adjustment (e.g. with sum of outer
5030 * pseudo-header) before calling.
5031 * Also ensure that inner checksum is in linear data area.
5033 static inline __wsum lco_csum(struct sk_buff *skb)
5035 unsigned char *csum_start = skb_checksum_start(skb);
5036 unsigned char *l4_hdr = skb_transport_header(skb);
5039 /* Start with complement of inner checksum adjustment */
5040 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
5043 /* Add in checksum of our headers (incl. outer checksum
5044 * adjustment filled in by caller) and return result.
5046 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
5049 static inline bool skb_is_redirected(const struct sk_buff *skb)
5051 return skb->redirected;
5054 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
5056 skb->redirected = 1;
5057 #ifdef CONFIG_NET_REDIRECT
5058 skb->from_ingress = from_ingress;
5059 if (skb->from_ingress)
5060 skb_clear_tstamp(skb);
5064 static inline void skb_reset_redirect(struct sk_buff *skb)
5066 skb->redirected = 0;
5069 static inline void skb_set_redirected_noclear(struct sk_buff *skb,
5072 skb->redirected = 1;
5073 #ifdef CONFIG_NET_REDIRECT
5074 skb->from_ingress = from_ingress;
5078 static inline bool skb_csum_is_sctp(struct sk_buff *skb)
5080 #if IS_ENABLED(CONFIG_IP_SCTP)
5081 return skb->csum_not_inet;
5087 static inline void skb_reset_csum_not_inet(struct sk_buff *skb)
5089 skb->ip_summed = CHECKSUM_NONE;
5090 #if IS_ENABLED(CONFIG_IP_SCTP)
5091 skb->csum_not_inet = 0;
5095 static inline void skb_set_kcov_handle(struct sk_buff *skb,
5096 const u64 kcov_handle)
5099 skb->kcov_handle = kcov_handle;
5103 static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
5106 return skb->kcov_handle;
5112 static inline void skb_mark_for_recycle(struct sk_buff *skb)
5114 #ifdef CONFIG_PAGE_POOL
5115 skb->pp_recycle = 1;
5119 #endif /* __KERNEL__ */
5120 #endif /* _LINUX_SKBUFF_H */