1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * Definitions for the 'struct sk_buff' memory handlers.
6 * Alan Cox, <gw4pts@gw4pts.ampr.org>
7 * Florian La Roche, <rzsfl@rz.uni-sb.de>
10 #ifndef _LINUX_SKBUFF_H
11 #define _LINUX_SKBUFF_H
13 #include <linux/kernel.h>
14 #include <linux/compiler.h>
15 #include <linux/time.h>
16 #include <linux/bug.h>
17 #include <linux/bvec.h>
18 #include <linux/cache.h>
19 #include <linux/rbtree.h>
20 #include <linux/socket.h>
21 #include <linux/refcount.h>
23 #include <linux/atomic.h>
24 #include <asm/types.h>
25 #include <linux/spinlock.h>
26 #include <net/checksum.h>
27 #include <linux/rcupdate.h>
28 #include <linux/dma-mapping.h>
29 #include <linux/netdev_features.h>
30 #include <net/flow_dissector.h>
31 #include <linux/in6.h>
32 #include <linux/if_packet.h>
33 #include <linux/llist.h>
35 #include <net/page_pool.h>
36 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
37 #include <linux/netfilter/nf_conntrack_common.h>
39 #include <net/net_debug.h>
40 #include <net/dropreason.h>
45 * The interface for checksum offload between the stack and networking drivers
48 * IP checksum related features
49 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
51 * Drivers advertise checksum offload capabilities in the features of a device.
52 * From the stack's point of view these are capabilities offered by the driver.
53 * A driver typically only advertises features that it is capable of offloading
56 * .. flat-table:: Checksum related device features
59 * * - %NETIF_F_HW_CSUM
60 * - The driver (or its device) is able to compute one
61 * IP (one's complement) checksum for any combination
62 * of protocols or protocol layering. The checksum is
63 * computed and set in a packet per the CHECKSUM_PARTIAL
64 * interface (see below).
66 * * - %NETIF_F_IP_CSUM
67 * - Driver (device) is only able to checksum plain
68 * TCP or UDP packets over IPv4. These are specifically
69 * unencapsulated packets of the form IPv4|TCP or
70 * IPv4|UDP where the Protocol field in the IPv4 header
71 * is TCP or UDP. The IPv4 header may contain IP options.
72 * This feature cannot be set in features for a device
73 * with NETIF_F_HW_CSUM also set. This feature is being
74 * DEPRECATED (see below).
76 * * - %NETIF_F_IPV6_CSUM
77 * - Driver (device) is only able to checksum plain
78 * TCP or UDP packets over IPv6. These are specifically
79 * unencapsulated packets of the form IPv6|TCP or
80 * IPv6|UDP where the Next Header field in the IPv6
81 * header is either TCP or UDP. IPv6 extension headers
82 * are not supported with this feature. This feature
83 * cannot be set in features for a device with
84 * NETIF_F_HW_CSUM also set. This feature is being
85 * DEPRECATED (see below).
88 * - Driver (device) performs receive checksum offload.
89 * This flag is only used to disable the RX checksum
90 * feature for a device. The stack will accept receive
91 * checksum indication in packets received on a device
92 * regardless of whether NETIF_F_RXCSUM is set.
94 * Checksumming of received packets by device
95 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
97 * Indication of checksum verification is set in &sk_buff.ip_summed.
98 * Possible values are:
102 * Device did not checksum this packet e.g. due to lack of capabilities.
103 * The packet contains full (though not verified) checksum in packet but
104 * not in skb->csum. Thus, skb->csum is undefined in this case.
106 * - %CHECKSUM_UNNECESSARY
108 * The hardware you're dealing with doesn't calculate the full checksum
109 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums
110 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY
111 * if their checksums are okay. &sk_buff.csum is still undefined in this case
112 * though. A driver or device must never modify the checksum field in the
113 * packet even if checksum is verified.
115 * %CHECKSUM_UNNECESSARY is applicable to following protocols:
117 * - TCP: IPv6 and IPv4.
118 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
119 * zero UDP checksum for either IPv4 or IPv6, the networking stack
120 * may perform further validation in this case.
121 * - GRE: only if the checksum is present in the header.
122 * - SCTP: indicates the CRC in SCTP header has been validated.
123 * - FCOE: indicates the CRC in FC frame has been validated.
125 * &sk_buff.csum_level indicates the number of consecutive checksums found in
126 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY.
127 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
128 * and a device is able to verify the checksums for UDP (possibly zero),
129 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to
130 * two. If the device were only able to verify the UDP checksum and not
131 * GRE, either because it doesn't support GRE checksum or because GRE
132 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
133 * not considered in this case).
135 * - %CHECKSUM_COMPLETE
137 * This is the most generic way. The device supplied checksum of the _whole_
138 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the
139 * hardware doesn't need to parse L3/L4 headers to implement this.
143 * - Even if device supports only some protocols, but is able to produce
144 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
145 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
147 * - %CHECKSUM_PARTIAL
149 * A checksum is set up to be offloaded to a device as described in the
150 * output description for CHECKSUM_PARTIAL. This may occur on a packet
151 * received directly from another Linux OS, e.g., a virtualized Linux kernel
152 * on the same host, or it may be set in the input path in GRO or remote
153 * checksum offload. For the purposes of checksum verification, the checksum
154 * referred to by skb->csum_start + skb->csum_offset and any preceding
155 * checksums in the packet are considered verified. Any checksums in the
156 * packet that are after the checksum being offloaded are not considered to
159 * Checksumming on transmit for non-GSO
160 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
162 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet.
165 * - %CHECKSUM_PARTIAL
167 * The driver is required to checksum the packet as seen by hard_start_xmit()
168 * from &sk_buff.csum_start up to the end, and to record/write the checksum at
169 * offset &sk_buff.csum_start + &sk_buff.csum_offset.
170 * A driver may verify that the
171 * csum_start and csum_offset values are valid values given the length and
172 * offset of the packet, but it should not attempt to validate that the
173 * checksum refers to a legitimate transport layer checksum -- it is the
174 * purview of the stack to validate that csum_start and csum_offset are set
177 * When the stack requests checksum offload for a packet, the driver MUST
178 * ensure that the checksum is set correctly. A driver can either offload the
179 * checksum calculation to the device, or call skb_checksum_help (in the case
180 * that the device does not support offload for a particular checksum).
182 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of
183 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate
184 * checksum offload capability.
185 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based
186 * on network device checksumming capabilities: if a packet does not match
187 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of
188 * &sk_buff.csum_not_inet, see :ref:`crc`)
189 * is called to resolve the checksum.
193 * The skb was already checksummed by the protocol, or a checksum is not
196 * - %CHECKSUM_UNNECESSARY
198 * This has the same meaning as CHECKSUM_NONE for checksum offload on
201 * - %CHECKSUM_COMPLETE
203 * Not used in checksum output. If a driver observes a packet with this value
204 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set.
208 * Non-IP checksum (CRC) offloads
209 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
214 * * - %NETIF_F_SCTP_CRC
215 * - This feature indicates that a device is capable of
216 * offloading the SCTP CRC in a packet. To perform this offload the stack
217 * will set csum_start and csum_offset accordingly, set ip_summed to
218 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication
219 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c.
220 * A driver that supports both IP checksum offload and SCTP CRC32c offload
221 * must verify which offload is configured for a packet by testing the
222 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to
223 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
225 * * - %NETIF_F_FCOE_CRC
226 * - This feature indicates that a device is capable of offloading the FCOE
227 * CRC in a packet. To perform this offload the stack will set ip_summed
228 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset
229 * accordingly. Note that there is no indication in the skbuff that the
230 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports
231 * both IP checksum offload and FCOE CRC offload must verify which offload
232 * is configured for a packet, presumably by inspecting packet headers.
234 * Checksumming on output with GSO
235 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
237 * In the case of a GSO packet (skb_is_gso() is true), checksum offload
238 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
239 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as
240 * part of the GSO operation is implied. If a checksum is being offloaded
241 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and
242 * csum_offset are set to refer to the outermost checksum being offloaded
243 * (two offloaded checksums are possible with UDP encapsulation).
246 /* Don't change this without changing skb_csum_unnecessary! */
247 #define CHECKSUM_NONE 0
248 #define CHECKSUM_UNNECESSARY 1
249 #define CHECKSUM_COMPLETE 2
250 #define CHECKSUM_PARTIAL 3
252 /* Maximum value in skb->csum_level */
253 #define SKB_MAX_CSUM_LEVEL 3
255 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
256 #define SKB_WITH_OVERHEAD(X) \
257 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
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 /* To allow 64K frame to be packed as single skb without frag_list we
350 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
351 * buffers which do not start on a page boundary.
353 * Since GRO uses frags we allocate at least 16 regardless of page
356 #if (65536/PAGE_SIZE + 1) < 16
357 #define MAX_SKB_FRAGS 16UL
359 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
361 extern int sysctl_max_skb_frags;
363 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
364 * segment using its current segmentation instead.
366 #define GSO_BY_FRAGS 0xFFFF
368 typedef struct bio_vec skb_frag_t;
371 * skb_frag_size() - Returns the size of a skb fragment
372 * @frag: skb fragment
374 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
380 * skb_frag_size_set() - Sets the size of a skb fragment
381 * @frag: skb fragment
382 * @size: size of fragment
384 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
390 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
391 * @frag: skb fragment
392 * @delta: value to add
394 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
396 frag->bv_len += delta;
400 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
401 * @frag: skb fragment
402 * @delta: value to subtract
404 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
406 frag->bv_len -= delta;
410 * skb_frag_must_loop - Test if %p is a high memory page
411 * @p: fragment's page
413 static inline bool skb_frag_must_loop(struct page *p)
415 #if defined(CONFIG_HIGHMEM)
416 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
423 * skb_frag_foreach_page - loop over pages in a fragment
425 * @f: skb frag to operate on
426 * @f_off: offset from start of f->bv_page
427 * @f_len: length from f_off to loop over
428 * @p: (temp var) current page
429 * @p_off: (temp var) offset from start of current page,
430 * non-zero only on first page.
431 * @p_len: (temp var) length in current page,
432 * < PAGE_SIZE only on first and last page.
433 * @copied: (temp var) length so far, excluding current p_len.
435 * A fragment can hold a compound page, in which case per-page
436 * operations, notably kmap_atomic, must be called for each
439 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
440 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
441 p_off = (f_off) & (PAGE_SIZE - 1), \
442 p_len = skb_frag_must_loop(p) ? \
443 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
446 copied += p_len, p++, p_off = 0, \
447 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
449 #define HAVE_HW_TIME_STAMP
452 * struct skb_shared_hwtstamps - hardware time stamps
453 * @hwtstamp: hardware time stamp transformed into duration
454 * since arbitrary point in time
455 * @netdev_data: address/cookie of network device driver used as
456 * reference to actual hardware time stamp
458 * Software time stamps generated by ktime_get_real() are stored in
461 * hwtstamps can only be compared against other hwtstamps from
464 * This structure is attached to packets as part of the
465 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
467 struct skb_shared_hwtstamps {
474 /* Definitions for tx_flags in struct skb_shared_info */
476 /* generate hardware time stamp */
477 SKBTX_HW_TSTAMP = 1 << 0,
479 /* generate software time stamp when queueing packet to NIC */
480 SKBTX_SW_TSTAMP = 1 << 1,
482 /* device driver is going to provide hardware time stamp */
483 SKBTX_IN_PROGRESS = 1 << 2,
485 /* generate hardware time stamp based on cycles if supported */
486 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3,
488 /* generate wifi status information (where possible) */
489 SKBTX_WIFI_STATUS = 1 << 4,
491 /* determine hardware time stamp based on time or cycles */
492 SKBTX_HW_TSTAMP_NETDEV = 1 << 5,
494 /* generate software time stamp when entering packet scheduling */
495 SKBTX_SCHED_TSTAMP = 1 << 6,
498 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
500 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \
501 SKBTX_HW_TSTAMP_USE_CYCLES | \
504 /* Definitions for flags in struct skb_shared_info */
506 /* use zcopy routines */
507 SKBFL_ZEROCOPY_ENABLE = BIT(0),
509 /* This indicates at least one fragment might be overwritten
510 * (as in vmsplice(), sendfile() ...)
511 * If we need to compute a TX checksum, we'll need to copy
512 * all frags to avoid possible bad checksum
514 SKBFL_SHARED_FRAG = BIT(1),
516 /* segment contains only zerocopy data and should not be
517 * charged to the kernel memory.
519 SKBFL_PURE_ZEROCOPY = BIT(2),
521 SKBFL_DONT_ORPHAN = BIT(3),
523 /* page references are managed by the ubuf_info, so it's safe to
524 * use frags only up until ubuf_info is released
526 SKBFL_MANAGED_FRAG_REFS = BIT(4),
529 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
530 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
531 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
534 * The callback notifies userspace to release buffers when skb DMA is done in
535 * lower device, the skb last reference should be 0 when calling this.
536 * The zerocopy_success argument is true if zero copy transmit occurred,
537 * false on data copy or out of memory error caused by data copy attempt.
538 * The ctx field is used to track device context.
539 * The desc field is used to track userspace buffer index.
542 void (*callback)(struct sk_buff *, struct ubuf_info *,
543 bool zerocopy_success);
548 struct ubuf_info_msgzc {
549 struct ubuf_info ubuf;
565 struct user_struct *user;
570 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
571 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \
574 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
575 void mm_unaccount_pinned_pages(struct mmpin *mmp);
577 /* This data is invariant across clones and lives at
578 * the end of the header data, ie. at skb->end.
580 struct skb_shared_info {
585 unsigned short gso_size;
586 /* Warning: this field is not always filled in (UFO)! */
587 unsigned short gso_segs;
588 struct sk_buff *frag_list;
589 struct skb_shared_hwtstamps hwtstamps;
590 unsigned int gso_type;
594 * Warning : all fields before dataref are cleared in __alloc_skb()
597 unsigned int xdp_frags_size;
599 /* Intermediate layers must ensure that destructor_arg
600 * remains valid until skb destructor */
601 void * destructor_arg;
603 /* must be last field, see pskb_expand_head() */
604 skb_frag_t frags[MAX_SKB_FRAGS];
608 * DOC: dataref and headerless skbs
610 * Transport layers send out clones of payload skbs they hold for
611 * retransmissions. To allow lower layers of the stack to prepend their headers
612 * we split &skb_shared_info.dataref into two halves.
613 * The lower 16 bits count the overall number of references.
614 * The higher 16 bits indicate how many of the references are payload-only.
615 * skb_header_cloned() checks if skb is allowed to add / write the headers.
617 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
618 * (via __skb_header_release()). Any clone created from marked skb will get
619 * &sk_buff.hdr_len populated with the available headroom.
620 * If there's the only clone in existence it's able to modify the headroom
621 * at will. The sequence of calls inside the transport layer is::
625 * __skb_header_release()
627 * // send the clone down the stack
629 * This is not a very generic construct and it depends on the transport layers
630 * doing the right thing. In practice there's usually only one payload-only skb.
631 * Having multiple payload-only skbs with different lengths of hdr_len is not
632 * possible. The payload-only skbs should never leave their owner.
634 #define SKB_DATAREF_SHIFT 16
635 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
639 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
640 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
641 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
645 SKB_GSO_TCPV4 = 1 << 0,
647 /* This indicates the skb is from an untrusted source. */
648 SKB_GSO_DODGY = 1 << 1,
650 /* This indicates the tcp segment has CWR set. */
651 SKB_GSO_TCP_ECN = 1 << 2,
653 SKB_GSO_TCP_FIXEDID = 1 << 3,
655 SKB_GSO_TCPV6 = 1 << 4,
657 SKB_GSO_FCOE = 1 << 5,
659 SKB_GSO_GRE = 1 << 6,
661 SKB_GSO_GRE_CSUM = 1 << 7,
663 SKB_GSO_IPXIP4 = 1 << 8,
665 SKB_GSO_IPXIP6 = 1 << 9,
667 SKB_GSO_UDP_TUNNEL = 1 << 10,
669 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
671 SKB_GSO_PARTIAL = 1 << 12,
673 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
675 SKB_GSO_SCTP = 1 << 14,
677 SKB_GSO_ESP = 1 << 15,
679 SKB_GSO_UDP = 1 << 16,
681 SKB_GSO_UDP_L4 = 1 << 17,
683 SKB_GSO_FRAGLIST = 1 << 18,
686 #if BITS_PER_LONG > 32
687 #define NET_SKBUFF_DATA_USES_OFFSET 1
690 #ifdef NET_SKBUFF_DATA_USES_OFFSET
691 typedef unsigned int sk_buff_data_t;
693 typedef unsigned char *sk_buff_data_t;
697 * DOC: Basic sk_buff geometry
699 * struct sk_buff itself is a metadata structure and does not hold any packet
700 * data. All the data is held in associated buffers.
702 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
705 * - data buffer, containing headers and sometimes payload;
706 * this is the part of the skb operated on by the common helpers
707 * such as skb_put() or skb_pull();
708 * - shared info (struct skb_shared_info) which holds an array of pointers
709 * to read-only data in the (page, offset, length) format.
711 * Optionally &skb_shared_info.frag_list may point to another skb.
713 * Basic diagram may look like this::
718 * ,--------------------------- + head
719 * / ,----------------- + data
720 * / / ,----------- + tail
724 * -----------------------------------------------
725 * | headroom | data | tailroom | skb_shared_info |
726 * -----------------------------------------------
730 * + [page frag] ---------
731 * + frag_list --> | sk_buff |
737 * struct sk_buff - socket buffer
738 * @next: Next buffer in list
739 * @prev: Previous buffer in list
740 * @tstamp: Time we arrived/left
741 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
742 * for retransmit timer
743 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
745 * @ll_node: anchor in an llist (eg socket defer_list)
746 * @sk: Socket we are owned by
747 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
748 * fragmentation management
749 * @dev: Device we arrived on/are leaving by
750 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
751 * @cb: Control buffer. Free for use by every layer. Put private vars here
752 * @_skb_refdst: destination entry (with norefcount bit)
753 * @sp: the security path, used for xfrm
754 * @len: Length of actual data
755 * @data_len: Data length
756 * @mac_len: Length of link layer header
757 * @hdr_len: writable header length of cloned skb
758 * @csum: Checksum (must include start/offset pair)
759 * @csum_start: Offset from skb->head where checksumming should start
760 * @csum_offset: Offset from csum_start where checksum should be stored
761 * @priority: Packet queueing priority
762 * @ignore_df: allow local fragmentation
763 * @cloned: Head may be cloned (check refcnt to be sure)
764 * @ip_summed: Driver fed us an IP checksum
765 * @nohdr: Payload reference only, must not modify header
766 * @pkt_type: Packet class
767 * @fclone: skbuff clone status
768 * @ipvs_property: skbuff is owned by ipvs
769 * @inner_protocol_type: whether the inner protocol is
770 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
771 * @remcsum_offload: remote checksum offload is enabled
772 * @offload_fwd_mark: Packet was L2-forwarded in hardware
773 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
774 * @tc_skip_classify: do not classify packet. set by IFB device
775 * @tc_at_ingress: used within tc_classify to distinguish in/egress
776 * @redirected: packet was redirected by packet classifier
777 * @from_ingress: packet was redirected from the ingress path
778 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
779 * @peeked: this packet has been seen already, so stats have been
780 * done for it, don't do them again
781 * @nf_trace: netfilter packet trace flag
782 * @protocol: Packet protocol from driver
783 * @destructor: Destruct function
784 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
785 * @_sk_redir: socket redirection information for skmsg
786 * @_nfct: Associated connection, if any (with nfctinfo bits)
787 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
788 * @skb_iif: ifindex of device we arrived on
789 * @tc_index: Traffic control index
790 * @hash: the packet hash
791 * @queue_mapping: Queue mapping for multiqueue devices
792 * @head_frag: skb was allocated from page fragments,
793 * not allocated by kmalloc() or vmalloc().
794 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
795 * @pp_recycle: mark the packet for recycling instead of freeing (implies
796 * page_pool support on driver)
797 * @active_extensions: active extensions (skb_ext_id types)
798 * @ndisc_nodetype: router type (from link layer)
799 * @ooo_okay: allow the mapping of a socket to a queue to be changed
800 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
802 * @sw_hash: indicates hash was computed in software stack
803 * @wifi_acked_valid: wifi_acked was set
804 * @wifi_acked: whether frame was acked on wifi or not
805 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
806 * @encapsulation: indicates the inner headers in the skbuff are valid
807 * @encap_hdr_csum: software checksum is needed
808 * @csum_valid: checksum is already valid
809 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
810 * @csum_complete_sw: checksum was completed by software
811 * @csum_level: indicates the number of consecutive checksums found in
812 * the packet minus one that have been verified as
813 * CHECKSUM_UNNECESSARY (max 3)
814 * @scm_io_uring: SKB holds io_uring registered files
815 * @dst_pending_confirm: need to confirm neighbour
816 * @decrypted: Decrypted SKB
817 * @slow_gro: state present at GRO time, slower prepare step required
818 * @mono_delivery_time: When set, skb->tstamp has the
819 * delivery_time in mono clock base (i.e. EDT). Otherwise, the
820 * skb->tstamp has the (rcv) timestamp at ingress and
821 * delivery_time at egress.
822 * @napi_id: id of the NAPI struct this skb came from
823 * @sender_cpu: (aka @napi_id) source CPU in XPS
824 * @alloc_cpu: CPU which did the skb allocation.
825 * @secmark: security marking
826 * @mark: Generic packet mark
827 * @reserved_tailroom: (aka @mark) number of bytes of free space available
828 * at the tail of an sk_buff
829 * @vlan_all: vlan fields (proto & tci)
830 * @vlan_proto: vlan encapsulation protocol
831 * @vlan_tci: vlan tag control information
832 * @inner_protocol: Protocol (encapsulation)
833 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
834 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
835 * @inner_transport_header: Inner transport layer header (encapsulation)
836 * @inner_network_header: Network layer header (encapsulation)
837 * @inner_mac_header: Link layer header (encapsulation)
838 * @transport_header: Transport layer header
839 * @network_header: Network layer header
840 * @mac_header: Link layer header
841 * @kcov_handle: KCOV remote handle for remote coverage collection
842 * @tail: Tail pointer
844 * @head: Head of buffer
845 * @data: Data head pointer
846 * @truesize: Buffer size
847 * @users: User count - see {datagram,tcp}.c
848 * @extensions: allocated extensions, valid if active_extensions is nonzero
854 /* These two members must be first to match sk_buff_head. */
855 struct sk_buff *next;
856 struct sk_buff *prev;
859 struct net_device *dev;
860 /* Some protocols might use this space to store information,
861 * while device pointer would be NULL.
862 * UDP receive path is one user.
864 unsigned long dev_scratch;
867 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
868 struct list_head list;
869 struct llist_node ll_node;
874 int ip_defrag_offset;
879 u64 skb_mstamp_ns; /* earliest departure time */
882 * This is the control buffer. It is free to use for every
883 * layer. Please put your private variables there. If you
884 * want to keep them across layers you have to do a skb_clone()
885 * first. This is owned by whoever has the skb queued ATM.
887 char cb[48] __aligned(8);
891 unsigned long _skb_refdst;
892 void (*destructor)(struct sk_buff *skb);
894 struct list_head tcp_tsorted_anchor;
895 #ifdef CONFIG_NET_SOCK_MSG
896 unsigned long _sk_redir;
900 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
908 /* Following fields are _not_ copied in __copy_skb_header()
909 * Note that queue_mapping is here mostly to fill a hole.
913 /* if you move cloned around you also must adapt those constants */
914 #ifdef __BIG_ENDIAN_BITFIELD
915 #define CLONED_MASK (1 << 7)
917 #define CLONED_MASK 1
919 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
922 __u8 __cloned_offset[0];
930 pp_recycle:1; /* page_pool recycle indicator */
931 #ifdef CONFIG_SKB_EXTENSIONS
932 __u8 active_extensions;
935 /* Fields enclosed in headers group are copied
936 * using a single memcpy() in __copy_skb_header()
938 struct_group(headers,
941 __u8 __pkt_type_offset[0];
943 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
951 __u8 wifi_acked_valid:1;
954 /* Indicates the inner headers are valid in the skbuff. */
955 __u8 encapsulation:1;
956 __u8 encap_hdr_csum:1;
960 __u8 __pkt_vlan_present_offset[0];
962 __u8 remcsum_offload:1;
963 __u8 csum_complete_sw:1;
965 __u8 dst_pending_confirm:1;
966 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */
967 #ifdef CONFIG_NET_CLS_ACT
968 __u8 tc_skip_classify:1;
969 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
971 #ifdef CONFIG_IPV6_NDISC_NODETYPE
972 __u8 ndisc_nodetype:2;
975 __u8 ipvs_property:1;
976 __u8 inner_protocol_type:1;
977 #ifdef CONFIG_NET_SWITCHDEV
978 __u8 offload_fwd_mark:1;
979 __u8 offload_l3_fwd_mark:1;
982 #ifdef CONFIG_NET_REDIRECT
985 #ifdef CONFIG_NETFILTER_SKIP_EGRESS
986 __u8 nf_skip_egress:1;
988 #ifdef CONFIG_TLS_DEVICE
992 __u8 csum_not_inet:1;
995 #ifdef CONFIG_NET_SCHED
996 __u16 tc_index; /* traffic control index */
1016 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
1018 unsigned int napi_id;
1019 unsigned int sender_cpu;
1023 #ifdef CONFIG_NETWORK_SECMARK
1029 __u32 reserved_tailroom;
1033 __be16 inner_protocol;
1037 __u16 inner_transport_header;
1038 __u16 inner_network_header;
1039 __u16 inner_mac_header;
1042 __u16 transport_header;
1043 __u16 network_header;
1050 ); /* end headers group */
1052 /* These elements must be at the end, see alloc_skb() for details. */
1053 sk_buff_data_t tail;
1055 unsigned char *head,
1057 unsigned int truesize;
1060 #ifdef CONFIG_SKB_EXTENSIONS
1061 /* only useable after checking ->active_extensions != 0 */
1062 struct skb_ext *extensions;
1066 /* if you move pkt_type around you also must adapt those constants */
1067 #ifdef __BIG_ENDIAN_BITFIELD
1068 #define PKT_TYPE_MAX (7 << 5)
1070 #define PKT_TYPE_MAX 7
1072 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1074 /* if you move tc_at_ingress or mono_delivery_time
1075 * around, you also must adapt these constants.
1077 #ifdef __BIG_ENDIAN_BITFIELD
1078 #define TC_AT_INGRESS_MASK (1 << 0)
1079 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 2)
1081 #define TC_AT_INGRESS_MASK (1 << 7)
1082 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 5)
1084 #define PKT_VLAN_PRESENT_OFFSET offsetof(struct sk_buff, __pkt_vlan_present_offset)
1088 * Handling routines are only of interest to the kernel
1091 #define SKB_ALLOC_FCLONE 0x01
1092 #define SKB_ALLOC_RX 0x02
1093 #define SKB_ALLOC_NAPI 0x04
1096 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1099 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1101 return unlikely(skb->pfmemalloc);
1105 * skb might have a dst pointer attached, refcounted or not.
1106 * _skb_refdst low order bit is set if refcount was _not_ taken
1108 #define SKB_DST_NOREF 1UL
1109 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1112 * skb_dst - returns skb dst_entry
1115 * Returns skb dst_entry, regardless of reference taken or not.
1117 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1119 /* If refdst was not refcounted, check we still are in a
1120 * rcu_read_lock section
1122 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1123 !rcu_read_lock_held() &&
1124 !rcu_read_lock_bh_held());
1125 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1129 * skb_dst_set - sets skb dst
1133 * Sets skb dst, assuming a reference was taken on dst and should
1134 * be released by skb_dst_drop()
1136 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1138 skb->slow_gro |= !!dst;
1139 skb->_skb_refdst = (unsigned long)dst;
1143 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1147 * Sets skb dst, assuming a reference was not taken on dst.
1148 * If dst entry is cached, we do not take reference and dst_release
1149 * will be avoided by refdst_drop. If dst entry is not cached, we take
1150 * reference, so that last dst_release can destroy the dst immediately.
1152 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1154 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1155 skb->slow_gro |= !!dst;
1156 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1160 * skb_dst_is_noref - Test if skb dst isn't refcounted
1163 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1165 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1169 * skb_rtable - Returns the skb &rtable
1172 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1174 return (struct rtable *)skb_dst(skb);
1177 /* For mangling skb->pkt_type from user space side from applications
1178 * such as nft, tc, etc, we only allow a conservative subset of
1179 * possible pkt_types to be set.
1181 static inline bool skb_pkt_type_ok(u32 ptype)
1183 return ptype <= PACKET_OTHERHOST;
1187 * skb_napi_id - Returns the skb's NAPI id
1190 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1192 #ifdef CONFIG_NET_RX_BUSY_POLL
1193 return skb->napi_id;
1200 * skb_unref - decrement the skb's reference count
1203 * Returns true if we can free the skb.
1205 static inline bool skb_unref(struct sk_buff *skb)
1209 if (likely(refcount_read(&skb->users) == 1))
1211 else if (likely(!refcount_dec_and_test(&skb->users)))
1218 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason);
1221 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1222 * @skb: buffer to free
1224 static inline void kfree_skb(struct sk_buff *skb)
1226 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1229 void skb_release_head_state(struct sk_buff *skb);
1230 void kfree_skb_list_reason(struct sk_buff *segs,
1231 enum skb_drop_reason reason);
1232 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1233 void skb_tx_error(struct sk_buff *skb);
1235 static inline void kfree_skb_list(struct sk_buff *segs)
1237 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1240 #ifdef CONFIG_TRACEPOINTS
1241 void consume_skb(struct sk_buff *skb);
1243 static inline void consume_skb(struct sk_buff *skb)
1245 return kfree_skb(skb);
1249 void __consume_stateless_skb(struct sk_buff *skb);
1250 void __kfree_skb(struct sk_buff *skb);
1251 extern struct kmem_cache *skbuff_cache;
1253 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1254 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1255 bool *fragstolen, int *delta_truesize);
1257 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1259 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1260 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1261 struct sk_buff *build_skb_around(struct sk_buff *skb,
1262 void *data, unsigned int frag_size);
1263 void skb_attempt_defer_free(struct sk_buff *skb);
1265 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1266 struct sk_buff *slab_build_skb(void *data);
1269 * alloc_skb - allocate a network buffer
1270 * @size: size to allocate
1271 * @priority: allocation mask
1273 * This function is a convenient wrapper around __alloc_skb().
1275 static inline struct sk_buff *alloc_skb(unsigned int size,
1278 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1281 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1282 unsigned long data_len,
1286 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1288 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1289 struct sk_buff_fclones {
1290 struct sk_buff skb1;
1292 struct sk_buff skb2;
1294 refcount_t fclone_ref;
1298 * skb_fclone_busy - check if fclone is busy
1302 * Returns true if skb is a fast clone, and its clone is not freed.
1303 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1304 * so we also check that this didnt happen.
1306 static inline bool skb_fclone_busy(const struct sock *sk,
1307 const struct sk_buff *skb)
1309 const struct sk_buff_fclones *fclones;
1311 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1313 return skb->fclone == SKB_FCLONE_ORIG &&
1314 refcount_read(&fclones->fclone_ref) > 1 &&
1315 READ_ONCE(fclones->skb2.sk) == sk;
1319 * alloc_skb_fclone - allocate a network buffer from fclone cache
1320 * @size: size to allocate
1321 * @priority: allocation mask
1323 * This function is a convenient wrapper around __alloc_skb().
1325 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1328 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1331 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1332 void skb_headers_offset_update(struct sk_buff *skb, int off);
1333 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1334 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1335 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1336 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1337 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1338 gfp_t gfp_mask, bool fclone);
1339 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1342 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1345 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1346 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1347 unsigned int headroom);
1348 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1349 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1350 int newtailroom, gfp_t priority);
1351 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1352 int offset, int len);
1353 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1354 int offset, int len);
1355 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1356 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1359 * skb_pad - zero pad the tail of an skb
1360 * @skb: buffer to pad
1361 * @pad: space to pad
1363 * Ensure that a buffer is followed by a padding area that is zero
1364 * filled. Used by network drivers which may DMA or transfer data
1365 * beyond the buffer end onto the wire.
1367 * May return error in out of memory cases. The skb is freed on error.
1369 static inline int skb_pad(struct sk_buff *skb, int pad)
1371 return __skb_pad(skb, pad, true);
1373 #define dev_kfree_skb(a) consume_skb(a)
1375 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1376 int offset, size_t size);
1378 struct skb_seq_state {
1382 __u32 stepped_offset;
1383 struct sk_buff *root_skb;
1384 struct sk_buff *cur_skb;
1389 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1390 unsigned int to, struct skb_seq_state *st);
1391 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1392 struct skb_seq_state *st);
1393 void skb_abort_seq_read(struct skb_seq_state *st);
1395 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1396 unsigned int to, struct ts_config *config);
1399 * Packet hash types specify the type of hash in skb_set_hash.
1401 * Hash types refer to the protocol layer addresses which are used to
1402 * construct a packet's hash. The hashes are used to differentiate or identify
1403 * flows of the protocol layer for the hash type. Hash types are either
1404 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1406 * Properties of hashes:
1408 * 1) Two packets in different flows have different hash values
1409 * 2) Two packets in the same flow should have the same hash value
1411 * A hash at a higher layer is considered to be more specific. A driver should
1412 * set the most specific hash possible.
1414 * A driver cannot indicate a more specific hash than the layer at which a hash
1415 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1417 * A driver may indicate a hash level which is less specific than the
1418 * actual layer the hash was computed on. For instance, a hash computed
1419 * at L4 may be considered an L3 hash. This should only be done if the
1420 * driver can't unambiguously determine that the HW computed the hash at
1421 * the higher layer. Note that the "should" in the second property above
1424 enum pkt_hash_types {
1425 PKT_HASH_TYPE_NONE, /* Undefined type */
1426 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1427 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1428 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1431 static inline void skb_clear_hash(struct sk_buff *skb)
1438 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1441 skb_clear_hash(skb);
1445 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1447 skb->l4_hash = is_l4;
1448 skb->sw_hash = is_sw;
1453 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1455 /* Used by drivers to set hash from HW */
1456 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1460 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1462 __skb_set_hash(skb, hash, true, is_l4);
1465 void __skb_get_hash(struct sk_buff *skb);
1466 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1467 u32 skb_get_poff(const struct sk_buff *skb);
1468 u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1469 const struct flow_keys_basic *keys, int hlen);
1470 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1471 const void *data, int hlen_proto);
1473 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1474 int thoff, u8 ip_proto)
1476 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1479 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1480 const struct flow_dissector_key *key,
1481 unsigned int key_count);
1483 struct bpf_flow_dissector;
1484 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1485 __be16 proto, int nhoff, int hlen, unsigned int flags);
1487 bool __skb_flow_dissect(const struct net *net,
1488 const struct sk_buff *skb,
1489 struct flow_dissector *flow_dissector,
1490 void *target_container, const void *data,
1491 __be16 proto, int nhoff, int hlen, unsigned int flags);
1493 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1494 struct flow_dissector *flow_dissector,
1495 void *target_container, unsigned int flags)
1497 return __skb_flow_dissect(NULL, skb, flow_dissector,
1498 target_container, NULL, 0, 0, 0, flags);
1501 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1502 struct flow_keys *flow,
1505 memset(flow, 0, sizeof(*flow));
1506 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1507 flow, NULL, 0, 0, 0, flags);
1511 skb_flow_dissect_flow_keys_basic(const struct net *net,
1512 const struct sk_buff *skb,
1513 struct flow_keys_basic *flow,
1514 const void *data, __be16 proto,
1515 int nhoff, int hlen, unsigned int flags)
1517 memset(flow, 0, sizeof(*flow));
1518 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1519 data, proto, nhoff, hlen, flags);
1522 void skb_flow_dissect_meta(const struct sk_buff *skb,
1523 struct flow_dissector *flow_dissector,
1524 void *target_container);
1526 /* Gets a skb connection tracking info, ctinfo map should be a
1527 * map of mapsize to translate enum ip_conntrack_info states
1531 skb_flow_dissect_ct(const struct sk_buff *skb,
1532 struct flow_dissector *flow_dissector,
1533 void *target_container,
1534 u16 *ctinfo_map, size_t mapsize,
1535 bool post_ct, u16 zone);
1537 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1538 struct flow_dissector *flow_dissector,
1539 void *target_container);
1541 void skb_flow_dissect_hash(const struct sk_buff *skb,
1542 struct flow_dissector *flow_dissector,
1543 void *target_container);
1545 static inline __u32 skb_get_hash(struct sk_buff *skb)
1547 if (!skb->l4_hash && !skb->sw_hash)
1548 __skb_get_hash(skb);
1553 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1555 if (!skb->l4_hash && !skb->sw_hash) {
1556 struct flow_keys keys;
1557 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1559 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1565 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1566 const siphash_key_t *perturb);
1568 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1573 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1575 to->hash = from->hash;
1576 to->sw_hash = from->sw_hash;
1577 to->l4_hash = from->l4_hash;
1580 static inline void skb_copy_decrypted(struct sk_buff *to,
1581 const struct sk_buff *from)
1583 #ifdef CONFIG_TLS_DEVICE
1584 to->decrypted = from->decrypted;
1588 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1589 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1591 return skb->head + skb->end;
1594 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1599 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1604 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1609 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1611 return skb->end - skb->head;
1614 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1616 skb->end = skb->head + offset;
1620 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1621 struct ubuf_info *uarg);
1623 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1625 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1628 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
1629 struct sk_buff *skb, struct iov_iter *from,
1632 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1633 struct msghdr *msg, int len)
1635 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len);
1638 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1639 struct msghdr *msg, int len,
1640 struct ubuf_info *uarg);
1643 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1645 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1647 return &skb_shinfo(skb)->hwtstamps;
1650 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1652 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1654 return is_zcopy ? skb_uarg(skb) : NULL;
1657 static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1659 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1662 static inline bool skb_zcopy_managed(const struct sk_buff *skb)
1664 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS;
1667 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1668 const struct sk_buff *skb2)
1670 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1673 static inline void net_zcopy_get(struct ubuf_info *uarg)
1675 refcount_inc(&uarg->refcnt);
1678 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1680 skb_shinfo(skb)->destructor_arg = uarg;
1681 skb_shinfo(skb)->flags |= uarg->flags;
1684 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1687 if (skb && uarg && !skb_zcopy(skb)) {
1688 if (unlikely(have_ref && *have_ref))
1691 net_zcopy_get(uarg);
1692 skb_zcopy_init(skb, uarg);
1696 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1698 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1699 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1702 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1704 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1707 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1709 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1712 static inline void net_zcopy_put(struct ubuf_info *uarg)
1715 uarg->callback(NULL, uarg, true);
1718 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1721 if (uarg->callback == msg_zerocopy_callback)
1722 msg_zerocopy_put_abort(uarg, have_uref);
1724 net_zcopy_put(uarg);
1728 /* Release a reference on a zerocopy structure */
1729 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1731 struct ubuf_info *uarg = skb_zcopy(skb);
1734 if (!skb_zcopy_is_nouarg(skb))
1735 uarg->callback(skb, uarg, zerocopy_success);
1737 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1741 void __skb_zcopy_downgrade_managed(struct sk_buff *skb);
1743 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb)
1745 if (unlikely(skb_zcopy_managed(skb)))
1746 __skb_zcopy_downgrade_managed(skb);
1749 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1754 static inline void skb_poison_list(struct sk_buff *skb)
1756 #ifdef CONFIG_DEBUG_NET
1757 skb->next = SKB_LIST_POISON_NEXT;
1761 /* Iterate through singly-linked GSO fragments of an skb. */
1762 #define skb_list_walk_safe(first, skb, next_skb) \
1763 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1764 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1766 static inline void skb_list_del_init(struct sk_buff *skb)
1768 __list_del_entry(&skb->list);
1769 skb_mark_not_on_list(skb);
1773 * skb_queue_empty - check if a queue is empty
1776 * Returns true if the queue is empty, false otherwise.
1778 static inline int skb_queue_empty(const struct sk_buff_head *list)
1780 return list->next == (const struct sk_buff *) list;
1784 * skb_queue_empty_lockless - check if a queue is empty
1787 * Returns true if the queue is empty, false otherwise.
1788 * This variant can be used in lockless contexts.
1790 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1792 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1797 * skb_queue_is_last - check if skb is the last entry in the queue
1801 * Returns true if @skb is the last buffer on the list.
1803 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1804 const struct sk_buff *skb)
1806 return skb->next == (const struct sk_buff *) list;
1810 * skb_queue_is_first - check if skb is the first entry in the queue
1814 * Returns true if @skb is the first buffer on the list.
1816 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1817 const struct sk_buff *skb)
1819 return skb->prev == (const struct sk_buff *) list;
1823 * skb_queue_next - return the next packet in the queue
1825 * @skb: current buffer
1827 * Return the next packet in @list after @skb. It is only valid to
1828 * call this if skb_queue_is_last() evaluates to false.
1830 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1831 const struct sk_buff *skb)
1833 /* This BUG_ON may seem severe, but if we just return then we
1834 * are going to dereference garbage.
1836 BUG_ON(skb_queue_is_last(list, skb));
1841 * skb_queue_prev - return the prev packet in the queue
1843 * @skb: current buffer
1845 * Return the prev packet in @list before @skb. It is only valid to
1846 * call this if skb_queue_is_first() evaluates to false.
1848 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1849 const struct sk_buff *skb)
1851 /* This BUG_ON may seem severe, but if we just return then we
1852 * are going to dereference garbage.
1854 BUG_ON(skb_queue_is_first(list, skb));
1859 * skb_get - reference buffer
1860 * @skb: buffer to reference
1862 * Makes another reference to a socket buffer and returns a pointer
1865 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1867 refcount_inc(&skb->users);
1872 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1876 * skb_cloned - is the buffer a clone
1877 * @skb: buffer to check
1879 * Returns true if the buffer was generated with skb_clone() and is
1880 * one of multiple shared copies of the buffer. Cloned buffers are
1881 * shared data so must not be written to under normal circumstances.
1883 static inline int skb_cloned(const struct sk_buff *skb)
1885 return skb->cloned &&
1886 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1889 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1891 might_sleep_if(gfpflags_allow_blocking(pri));
1893 if (skb_cloned(skb))
1894 return pskb_expand_head(skb, 0, 0, pri);
1899 /* This variant of skb_unclone() makes sure skb->truesize
1900 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
1902 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
1903 * when various debugging features are in place.
1905 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
1906 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1908 might_sleep_if(gfpflags_allow_blocking(pri));
1910 if (skb_cloned(skb))
1911 return __skb_unclone_keeptruesize(skb, pri);
1916 * skb_header_cloned - is the header a clone
1917 * @skb: buffer to check
1919 * Returns true if modifying the header part of the buffer requires
1920 * the data to be copied.
1922 static inline int skb_header_cloned(const struct sk_buff *skb)
1929 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1930 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1931 return dataref != 1;
1934 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1936 might_sleep_if(gfpflags_allow_blocking(pri));
1938 if (skb_header_cloned(skb))
1939 return pskb_expand_head(skb, 0, 0, pri);
1945 * __skb_header_release() - allow clones to use the headroom
1946 * @skb: buffer to operate on
1948 * See "DOC: dataref and headerless skbs".
1950 static inline void __skb_header_release(struct sk_buff *skb)
1953 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1958 * skb_shared - is the buffer shared
1959 * @skb: buffer to check
1961 * Returns true if more than one person has a reference to this
1964 static inline int skb_shared(const struct sk_buff *skb)
1966 return refcount_read(&skb->users) != 1;
1970 * skb_share_check - check if buffer is shared and if so clone it
1971 * @skb: buffer to check
1972 * @pri: priority for memory allocation
1974 * If the buffer is shared the buffer is cloned and the old copy
1975 * drops a reference. A new clone with a single reference is returned.
1976 * If the buffer is not shared the original buffer is returned. When
1977 * being called from interrupt status or with spinlocks held pri must
1980 * NULL is returned on a memory allocation failure.
1982 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1984 might_sleep_if(gfpflags_allow_blocking(pri));
1985 if (skb_shared(skb)) {
1986 struct sk_buff *nskb = skb_clone(skb, pri);
1998 * Copy shared buffers into a new sk_buff. We effectively do COW on
1999 * packets to handle cases where we have a local reader and forward
2000 * and a couple of other messy ones. The normal one is tcpdumping
2001 * a packet thats being forwarded.
2005 * skb_unshare - make a copy of a shared buffer
2006 * @skb: buffer to check
2007 * @pri: priority for memory allocation
2009 * If the socket buffer is a clone then this function creates a new
2010 * copy of the data, drops a reference count on the old copy and returns
2011 * the new copy with the reference count at 1. If the buffer is not a clone
2012 * the original buffer is returned. When called with a spinlock held or
2013 * from interrupt state @pri must be %GFP_ATOMIC
2015 * %NULL is returned on a memory allocation failure.
2017 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
2020 might_sleep_if(gfpflags_allow_blocking(pri));
2021 if (skb_cloned(skb)) {
2022 struct sk_buff *nskb = skb_copy(skb, pri);
2024 /* Free our shared copy */
2035 * skb_peek - peek at the head of an &sk_buff_head
2036 * @list_: list to peek at
2038 * Peek an &sk_buff. Unlike most other operations you _MUST_
2039 * be careful with this one. A peek leaves the buffer on the
2040 * list and someone else may run off with it. You must hold
2041 * the appropriate locks or have a private queue to do this.
2043 * Returns %NULL for an empty list or a pointer to the head element.
2044 * The reference count is not incremented and the reference is therefore
2045 * volatile. Use with caution.
2047 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2049 struct sk_buff *skb = list_->next;
2051 if (skb == (struct sk_buff *)list_)
2057 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2058 * @list_: list to peek at
2060 * Like skb_peek(), but the caller knows that the list is not empty.
2062 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2068 * skb_peek_next - peek skb following the given one from a queue
2069 * @skb: skb to start from
2070 * @list_: list to peek at
2072 * Returns %NULL when the end of the list is met or a pointer to the
2073 * next element. The reference count is not incremented and the
2074 * reference is therefore volatile. Use with caution.
2076 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2077 const struct sk_buff_head *list_)
2079 struct sk_buff *next = skb->next;
2081 if (next == (struct sk_buff *)list_)
2087 * skb_peek_tail - peek at the tail of an &sk_buff_head
2088 * @list_: list to peek at
2090 * Peek an &sk_buff. Unlike most other operations you _MUST_
2091 * be careful with this one. A peek leaves the buffer on the
2092 * list and someone else may run off with it. You must hold
2093 * the appropriate locks or have a private queue to do this.
2095 * Returns %NULL for an empty list or a pointer to the tail element.
2096 * The reference count is not incremented and the reference is therefore
2097 * volatile. Use with caution.
2099 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2101 struct sk_buff *skb = READ_ONCE(list_->prev);
2103 if (skb == (struct sk_buff *)list_)
2110 * skb_queue_len - get queue length
2111 * @list_: list to measure
2113 * Return the length of an &sk_buff queue.
2115 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2121 * skb_queue_len_lockless - get queue length
2122 * @list_: list to measure
2124 * Return the length of an &sk_buff queue.
2125 * This variant can be used in lockless contexts.
2127 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2129 return READ_ONCE(list_->qlen);
2133 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2134 * @list: queue to initialize
2136 * This initializes only the list and queue length aspects of
2137 * an sk_buff_head object. This allows to initialize the list
2138 * aspects of an sk_buff_head without reinitializing things like
2139 * the spinlock. It can also be used for on-stack sk_buff_head
2140 * objects where the spinlock is known to not be used.
2142 static inline void __skb_queue_head_init(struct sk_buff_head *list)
2144 list->prev = list->next = (struct sk_buff *)list;
2149 * This function creates a split out lock class for each invocation;
2150 * this is needed for now since a whole lot of users of the skb-queue
2151 * infrastructure in drivers have different locking usage (in hardirq)
2152 * than the networking core (in softirq only). In the long run either the
2153 * network layer or drivers should need annotation to consolidate the
2154 * main types of usage into 3 classes.
2156 static inline void skb_queue_head_init(struct sk_buff_head *list)
2158 spin_lock_init(&list->lock);
2159 __skb_queue_head_init(list);
2162 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2163 struct lock_class_key *class)
2165 skb_queue_head_init(list);
2166 lockdep_set_class(&list->lock, class);
2170 * Insert an sk_buff on a list.
2172 * The "__skb_xxxx()" functions are the non-atomic ones that
2173 * can only be called with interrupts disabled.
2175 static inline void __skb_insert(struct sk_buff *newsk,
2176 struct sk_buff *prev, struct sk_buff *next,
2177 struct sk_buff_head *list)
2179 /* See skb_queue_empty_lockless() and skb_peek_tail()
2180 * for the opposite READ_ONCE()
2182 WRITE_ONCE(newsk->next, next);
2183 WRITE_ONCE(newsk->prev, prev);
2184 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2185 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2186 WRITE_ONCE(list->qlen, list->qlen + 1);
2189 static inline void __skb_queue_splice(const struct sk_buff_head *list,
2190 struct sk_buff *prev,
2191 struct sk_buff *next)
2193 struct sk_buff *first = list->next;
2194 struct sk_buff *last = list->prev;
2196 WRITE_ONCE(first->prev, prev);
2197 WRITE_ONCE(prev->next, first);
2199 WRITE_ONCE(last->next, next);
2200 WRITE_ONCE(next->prev, last);
2204 * skb_queue_splice - join two skb lists, this is designed for stacks
2205 * @list: the new list to add
2206 * @head: the place to add it in the first list
2208 static inline void skb_queue_splice(const struct sk_buff_head *list,
2209 struct sk_buff_head *head)
2211 if (!skb_queue_empty(list)) {
2212 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2213 head->qlen += list->qlen;
2218 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2219 * @list: the new list to add
2220 * @head: the place to add it in the first list
2222 * The list at @list is reinitialised
2224 static inline void skb_queue_splice_init(struct sk_buff_head *list,
2225 struct sk_buff_head *head)
2227 if (!skb_queue_empty(list)) {
2228 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2229 head->qlen += list->qlen;
2230 __skb_queue_head_init(list);
2235 * skb_queue_splice_tail - join two skb lists, each list being a queue
2236 * @list: the new list to add
2237 * @head: the place to add it in the first list
2239 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2240 struct sk_buff_head *head)
2242 if (!skb_queue_empty(list)) {
2243 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2244 head->qlen += list->qlen;
2249 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2250 * @list: the new list to add
2251 * @head: the place to add it in the first list
2253 * Each of the lists is a queue.
2254 * The list at @list is reinitialised
2256 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2257 struct sk_buff_head *head)
2259 if (!skb_queue_empty(list)) {
2260 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2261 head->qlen += list->qlen;
2262 __skb_queue_head_init(list);
2267 * __skb_queue_after - queue a buffer at the list head
2268 * @list: list to use
2269 * @prev: place after this buffer
2270 * @newsk: buffer to queue
2272 * Queue a buffer int the middle of a list. This function takes no locks
2273 * and you must therefore hold required locks before calling it.
2275 * A buffer cannot be placed on two lists at the same time.
2277 static inline void __skb_queue_after(struct sk_buff_head *list,
2278 struct sk_buff *prev,
2279 struct sk_buff *newsk)
2281 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2284 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2285 struct sk_buff_head *list);
2287 static inline void __skb_queue_before(struct sk_buff_head *list,
2288 struct sk_buff *next,
2289 struct sk_buff *newsk)
2291 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2295 * __skb_queue_head - queue a buffer at the list head
2296 * @list: list to use
2297 * @newsk: buffer to queue
2299 * Queue a buffer at the start of a list. This function takes no locks
2300 * and you must therefore hold required locks before calling it.
2302 * A buffer cannot be placed on two lists at the same time.
2304 static inline void __skb_queue_head(struct sk_buff_head *list,
2305 struct sk_buff *newsk)
2307 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2309 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2312 * __skb_queue_tail - queue a buffer at the list tail
2313 * @list: list to use
2314 * @newsk: buffer to queue
2316 * Queue a buffer at the end of a list. This function takes no locks
2317 * and you must therefore hold required locks before calling it.
2319 * A buffer cannot be placed on two lists at the same time.
2321 static inline void __skb_queue_tail(struct sk_buff_head *list,
2322 struct sk_buff *newsk)
2324 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2326 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2329 * remove sk_buff from list. _Must_ be called atomically, and with
2332 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2333 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2335 struct sk_buff *next, *prev;
2337 WRITE_ONCE(list->qlen, list->qlen - 1);
2340 skb->next = skb->prev = NULL;
2341 WRITE_ONCE(next->prev, prev);
2342 WRITE_ONCE(prev->next, next);
2346 * __skb_dequeue - remove from the head of the queue
2347 * @list: list to dequeue from
2349 * Remove the head of the list. This function does not take any locks
2350 * so must be used with appropriate locks held only. The head item is
2351 * returned or %NULL if the list is empty.
2353 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2355 struct sk_buff *skb = skb_peek(list);
2357 __skb_unlink(skb, list);
2360 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2363 * __skb_dequeue_tail - remove from the tail of the queue
2364 * @list: list to dequeue from
2366 * Remove the tail of the list. This function does not take any locks
2367 * so must be used with appropriate locks held only. The tail item is
2368 * returned or %NULL if the list is empty.
2370 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2372 struct sk_buff *skb = skb_peek_tail(list);
2374 __skb_unlink(skb, list);
2377 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2380 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2382 return skb->data_len;
2385 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2387 return skb->len - skb->data_len;
2390 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2392 unsigned int i, len = 0;
2394 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2395 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2399 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2401 return skb_headlen(skb) + __skb_pagelen(skb);
2404 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo,
2405 int i, struct page *page,
2408 skb_frag_t *frag = &shinfo->frags[i];
2411 * Propagate page pfmemalloc to the skb if we can. The problem is
2412 * that not all callers have unique ownership of the page but rely
2413 * on page_is_pfmemalloc doing the right thing(tm).
2415 frag->bv_page = page;
2416 frag->bv_offset = off;
2417 skb_frag_size_set(frag, size);
2421 * skb_len_add - adds a number to len fields of skb
2422 * @skb: buffer to add len to
2423 * @delta: number of bytes to add
2425 static inline void skb_len_add(struct sk_buff *skb, int delta)
2428 skb->data_len += delta;
2429 skb->truesize += delta;
2433 * __skb_fill_page_desc - initialise a paged fragment in an skb
2434 * @skb: buffer containing fragment to be initialised
2435 * @i: paged fragment index to initialise
2436 * @page: the page to use for this fragment
2437 * @off: the offset to the data with @page
2438 * @size: the length of the data
2440 * Initialises the @i'th fragment of @skb to point to &size bytes at
2441 * offset @off within @page.
2443 * Does not take any additional reference on the fragment.
2445 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2446 struct page *page, int off, int size)
2448 __skb_fill_page_desc_noacc(skb_shinfo(skb), i, page, off, size);
2449 page = compound_head(page);
2450 if (page_is_pfmemalloc(page))
2451 skb->pfmemalloc = true;
2455 * skb_fill_page_desc - initialise a paged fragment in an skb
2456 * @skb: buffer containing fragment to be initialised
2457 * @i: paged fragment index to initialise
2458 * @page: the page to use for this fragment
2459 * @off: the offset to the data with @page
2460 * @size: the length of the data
2462 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2463 * @skb to point to @size bytes at offset @off within @page. In
2464 * addition updates @skb such that @i is the last fragment.
2466 * Does not take any additional reference on the fragment.
2468 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2469 struct page *page, int off, int size)
2471 __skb_fill_page_desc(skb, i, page, off, size);
2472 skb_shinfo(skb)->nr_frags = i + 1;
2476 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb
2477 * @skb: buffer containing fragment to be initialised
2478 * @i: paged fragment index to initialise
2479 * @page: the page to use for this fragment
2480 * @off: the offset to the data with @page
2481 * @size: the length of the data
2483 * Variant of skb_fill_page_desc() which does not deal with
2484 * pfmemalloc, if page is not owned by us.
2486 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i,
2487 struct page *page, int off,
2490 struct skb_shared_info *shinfo = skb_shinfo(skb);
2492 __skb_fill_page_desc_noacc(shinfo, i, page, off, size);
2493 shinfo->nr_frags = i + 1;
2496 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2497 int size, unsigned int truesize);
2499 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2500 unsigned int truesize);
2502 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2504 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2505 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2507 return skb->head + skb->tail;
2510 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2512 skb->tail = skb->data - skb->head;
2515 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2517 skb_reset_tail_pointer(skb);
2518 skb->tail += offset;
2521 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2522 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2527 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2529 skb->tail = skb->data;
2532 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2534 skb->tail = skb->data + offset;
2537 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2539 static inline void skb_assert_len(struct sk_buff *skb)
2541 #ifdef CONFIG_DEBUG_NET
2542 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2543 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2544 #endif /* CONFIG_DEBUG_NET */
2548 * Add data to an sk_buff
2550 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2551 void *skb_put(struct sk_buff *skb, unsigned int len);
2552 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2554 void *tmp = skb_tail_pointer(skb);
2555 SKB_LINEAR_ASSERT(skb);
2561 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2563 void *tmp = __skb_put(skb, len);
2565 memset(tmp, 0, len);
2569 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2572 void *tmp = __skb_put(skb, len);
2574 memcpy(tmp, data, len);
2578 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2580 *(u8 *)__skb_put(skb, 1) = val;
2583 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2585 void *tmp = skb_put(skb, len);
2587 memset(tmp, 0, len);
2592 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2595 void *tmp = skb_put(skb, len);
2597 memcpy(tmp, data, len);
2602 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2604 *(u8 *)skb_put(skb, 1) = val;
2607 void *skb_push(struct sk_buff *skb, unsigned int len);
2608 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2615 void *skb_pull(struct sk_buff *skb, unsigned int len);
2616 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2619 if (unlikely(skb->len < skb->data_len)) {
2620 #if defined(CONFIG_DEBUG_NET)
2622 pr_err("__skb_pull(len=%u)\n", len);
2623 skb_dump(KERN_ERR, skb, false);
2627 return skb->data += len;
2630 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2632 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2635 void *skb_pull_data(struct sk_buff *skb, size_t len);
2637 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2639 static inline enum skb_drop_reason
2640 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len)
2642 if (likely(len <= skb_headlen(skb)))
2643 return SKB_NOT_DROPPED_YET;
2645 if (unlikely(len > skb->len))
2646 return SKB_DROP_REASON_PKT_TOO_SMALL;
2648 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb))))
2649 return SKB_DROP_REASON_NOMEM;
2651 return SKB_NOT_DROPPED_YET;
2654 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2656 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET;
2659 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2661 if (!pskb_may_pull(skb, len))
2665 return skb->data += len;
2668 void skb_condense(struct sk_buff *skb);
2671 * skb_headroom - bytes at buffer head
2672 * @skb: buffer to check
2674 * Return the number of bytes of free space at the head of an &sk_buff.
2676 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2678 return skb->data - skb->head;
2682 * skb_tailroom - bytes at buffer end
2683 * @skb: buffer to check
2685 * Return the number of bytes of free space at the tail of an sk_buff
2687 static inline int skb_tailroom(const struct sk_buff *skb)
2689 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2693 * skb_availroom - bytes at buffer end
2694 * @skb: buffer to check
2696 * Return the number of bytes of free space at the tail of an sk_buff
2697 * allocated by sk_stream_alloc()
2699 static inline int skb_availroom(const struct sk_buff *skb)
2701 if (skb_is_nonlinear(skb))
2704 return skb->end - skb->tail - skb->reserved_tailroom;
2708 * skb_reserve - adjust headroom
2709 * @skb: buffer to alter
2710 * @len: bytes to move
2712 * Increase the headroom of an empty &sk_buff by reducing the tail
2713 * room. This is only allowed for an empty buffer.
2715 static inline void skb_reserve(struct sk_buff *skb, int len)
2722 * skb_tailroom_reserve - adjust reserved_tailroom
2723 * @skb: buffer to alter
2724 * @mtu: maximum amount of headlen permitted
2725 * @needed_tailroom: minimum amount of reserved_tailroom
2727 * Set reserved_tailroom so that headlen can be as large as possible but
2728 * not larger than mtu and tailroom cannot be smaller than
2730 * The required headroom should already have been reserved before using
2733 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2734 unsigned int needed_tailroom)
2736 SKB_LINEAR_ASSERT(skb);
2737 if (mtu < skb_tailroom(skb) - needed_tailroom)
2738 /* use at most mtu */
2739 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2741 /* use up to all available space */
2742 skb->reserved_tailroom = needed_tailroom;
2745 #define ENCAP_TYPE_ETHER 0
2746 #define ENCAP_TYPE_IPPROTO 1
2748 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2751 skb->inner_protocol = protocol;
2752 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2755 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2758 skb->inner_ipproto = ipproto;
2759 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2762 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2764 skb->inner_mac_header = skb->mac_header;
2765 skb->inner_network_header = skb->network_header;
2766 skb->inner_transport_header = skb->transport_header;
2769 static inline void skb_reset_mac_len(struct sk_buff *skb)
2771 skb->mac_len = skb->network_header - skb->mac_header;
2774 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2777 return skb->head + skb->inner_transport_header;
2780 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2782 return skb_inner_transport_header(skb) - skb->data;
2785 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2787 skb->inner_transport_header = skb->data - skb->head;
2790 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2793 skb_reset_inner_transport_header(skb);
2794 skb->inner_transport_header += offset;
2797 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2799 return skb->head + skb->inner_network_header;
2802 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2804 skb->inner_network_header = skb->data - skb->head;
2807 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2810 skb_reset_inner_network_header(skb);
2811 skb->inner_network_header += offset;
2814 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2816 return skb->head + skb->inner_mac_header;
2819 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2821 skb->inner_mac_header = skb->data - skb->head;
2824 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2827 skb_reset_inner_mac_header(skb);
2828 skb->inner_mac_header += offset;
2830 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2832 return skb->transport_header != (typeof(skb->transport_header))~0U;
2835 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2837 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
2838 return skb->head + skb->transport_header;
2841 static inline void skb_reset_transport_header(struct sk_buff *skb)
2843 skb->transport_header = skb->data - skb->head;
2846 static inline void skb_set_transport_header(struct sk_buff *skb,
2849 skb_reset_transport_header(skb);
2850 skb->transport_header += offset;
2853 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2855 return skb->head + skb->network_header;
2858 static inline void skb_reset_network_header(struct sk_buff *skb)
2860 skb->network_header = skb->data - skb->head;
2863 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2865 skb_reset_network_header(skb);
2866 skb->network_header += offset;
2869 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2871 return skb->mac_header != (typeof(skb->mac_header))~0U;
2874 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2876 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2877 return skb->head + skb->mac_header;
2880 static inline int skb_mac_offset(const struct sk_buff *skb)
2882 return skb_mac_header(skb) - skb->data;
2885 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2887 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2888 return skb->network_header - skb->mac_header;
2891 static inline void skb_unset_mac_header(struct sk_buff *skb)
2893 skb->mac_header = (typeof(skb->mac_header))~0U;
2896 static inline void skb_reset_mac_header(struct sk_buff *skb)
2898 skb->mac_header = skb->data - skb->head;
2901 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2903 skb_reset_mac_header(skb);
2904 skb->mac_header += offset;
2907 static inline void skb_pop_mac_header(struct sk_buff *skb)
2909 skb->mac_header = skb->network_header;
2912 static inline void skb_probe_transport_header(struct sk_buff *skb)
2914 struct flow_keys_basic keys;
2916 if (skb_transport_header_was_set(skb))
2919 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2921 skb_set_transport_header(skb, keys.control.thoff);
2924 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2926 if (skb_mac_header_was_set(skb)) {
2927 const unsigned char *old_mac = skb_mac_header(skb);
2929 skb_set_mac_header(skb, -skb->mac_len);
2930 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2934 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2936 return skb->csum_start - skb_headroom(skb);
2939 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2941 return skb->head + skb->csum_start;
2944 static inline int skb_transport_offset(const struct sk_buff *skb)
2946 return skb_transport_header(skb) - skb->data;
2949 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2951 return skb->transport_header - skb->network_header;
2954 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2956 return skb->inner_transport_header - skb->inner_network_header;
2959 static inline int skb_network_offset(const struct sk_buff *skb)
2961 return skb_network_header(skb) - skb->data;
2964 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2966 return skb_inner_network_header(skb) - skb->data;
2969 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2971 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2975 * CPUs often take a performance hit when accessing unaligned memory
2976 * locations. The actual performance hit varies, it can be small if the
2977 * hardware handles it or large if we have to take an exception and fix it
2980 * Since an ethernet header is 14 bytes network drivers often end up with
2981 * the IP header at an unaligned offset. The IP header can be aligned by
2982 * shifting the start of the packet by 2 bytes. Drivers should do this
2985 * skb_reserve(skb, NET_IP_ALIGN);
2987 * The downside to this alignment of the IP header is that the DMA is now
2988 * unaligned. On some architectures the cost of an unaligned DMA is high
2989 * and this cost outweighs the gains made by aligning the IP header.
2991 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2994 #ifndef NET_IP_ALIGN
2995 #define NET_IP_ALIGN 2
2999 * The networking layer reserves some headroom in skb data (via
3000 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
3001 * the header has to grow. In the default case, if the header has to grow
3002 * 32 bytes or less we avoid the reallocation.
3004 * Unfortunately this headroom changes the DMA alignment of the resulting
3005 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
3006 * on some architectures. An architecture can override this value,
3007 * perhaps setting it to a cacheline in size (since that will maintain
3008 * cacheline alignment of the DMA). It must be a power of 2.
3010 * Various parts of the networking layer expect at least 32 bytes of
3011 * headroom, you should not reduce this.
3013 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
3014 * to reduce average number of cache lines per packet.
3015 * get_rps_cpu() for example only access one 64 bytes aligned block :
3016 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
3019 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
3022 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
3024 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
3026 if (WARN_ON(skb_is_nonlinear(skb)))
3029 skb_set_tail_pointer(skb, len);
3032 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
3034 __skb_set_length(skb, len);
3037 void skb_trim(struct sk_buff *skb, unsigned int len);
3039 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
3042 return ___pskb_trim(skb, len);
3043 __skb_trim(skb, len);
3047 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
3049 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
3053 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
3054 * @skb: buffer to alter
3057 * This is identical to pskb_trim except that the caller knows that
3058 * the skb is not cloned so we should never get an error due to out-
3061 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3063 int err = pskb_trim(skb, len);
3067 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3069 unsigned int diff = len - skb->len;
3071 if (skb_tailroom(skb) < diff) {
3072 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3077 __skb_set_length(skb, len);
3082 * skb_orphan - orphan a buffer
3083 * @skb: buffer to orphan
3085 * If a buffer currently has an owner then we call the owner's
3086 * destructor function and make the @skb unowned. The buffer continues
3087 * to exist but is no longer charged to its former owner.
3089 static inline void skb_orphan(struct sk_buff *skb)
3091 if (skb->destructor) {
3092 skb->destructor(skb);
3093 skb->destructor = NULL;
3101 * skb_orphan_frags - orphan the frags contained in a buffer
3102 * @skb: buffer to orphan frags from
3103 * @gfp_mask: allocation mask for replacement pages
3105 * For each frag in the SKB which needs a destructor (i.e. has an
3106 * owner) create a copy of that frag and release the original
3107 * page by calling the destructor.
3109 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3111 if (likely(!skb_zcopy(skb)))
3113 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN)
3115 return skb_copy_ubufs(skb, gfp_mask);
3118 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3119 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3121 if (likely(!skb_zcopy(skb)))
3123 return skb_copy_ubufs(skb, gfp_mask);
3127 * __skb_queue_purge - empty a list
3128 * @list: list to empty
3130 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3131 * the list and one reference dropped. This function does not take the
3132 * list lock and the caller must hold the relevant locks to use it.
3134 static inline void __skb_queue_purge(struct sk_buff_head *list)
3136 struct sk_buff *skb;
3137 while ((skb = __skb_dequeue(list)) != NULL)
3140 void skb_queue_purge(struct sk_buff_head *list);
3142 unsigned int skb_rbtree_purge(struct rb_root *root);
3144 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3147 * netdev_alloc_frag - allocate a page fragment
3148 * @fragsz: fragment size
3150 * Allocates a frag from a page for receive buffer.
3151 * Uses GFP_ATOMIC allocations.
3153 static inline void *netdev_alloc_frag(unsigned int fragsz)
3155 return __netdev_alloc_frag_align(fragsz, ~0u);
3158 static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3161 WARN_ON_ONCE(!is_power_of_2(align));
3162 return __netdev_alloc_frag_align(fragsz, -align);
3165 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3169 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3170 * @dev: network device to receive on
3171 * @length: length to allocate
3173 * Allocate a new &sk_buff and assign it a usage count of one. The
3174 * buffer has unspecified headroom built in. Users should allocate
3175 * the headroom they think they need without accounting for the
3176 * built in space. The built in space is used for optimisations.
3178 * %NULL is returned if there is no free memory. Although this function
3179 * allocates memory it can be called from an interrupt.
3181 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3182 unsigned int length)
3184 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3187 /* legacy helper around __netdev_alloc_skb() */
3188 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3191 return __netdev_alloc_skb(NULL, length, gfp_mask);
3194 /* legacy helper around netdev_alloc_skb() */
3195 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3197 return netdev_alloc_skb(NULL, length);
3201 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3202 unsigned int length, gfp_t gfp)
3204 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3206 if (NET_IP_ALIGN && skb)
3207 skb_reserve(skb, NET_IP_ALIGN);
3211 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3212 unsigned int length)
3214 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3217 static inline void skb_free_frag(void *addr)
3219 page_frag_free(addr);
3222 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3224 static inline void *napi_alloc_frag(unsigned int fragsz)
3226 return __napi_alloc_frag_align(fragsz, ~0u);
3229 static inline void *napi_alloc_frag_align(unsigned int fragsz,
3232 WARN_ON_ONCE(!is_power_of_2(align));
3233 return __napi_alloc_frag_align(fragsz, -align);
3236 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
3237 unsigned int length, gfp_t gfp_mask);
3238 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
3239 unsigned int length)
3241 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
3243 void napi_consume_skb(struct sk_buff *skb, int budget);
3245 void napi_skb_free_stolen_head(struct sk_buff *skb);
3246 void __kfree_skb_defer(struct sk_buff *skb);
3249 * __dev_alloc_pages - allocate page for network Rx
3250 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3251 * @order: size of the allocation
3253 * Allocate a new page.
3255 * %NULL is returned if there is no free memory.
3257 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
3260 /* This piece of code contains several assumptions.
3261 * 1. This is for device Rx, therefor a cold page is preferred.
3262 * 2. The expectation is the user wants a compound page.
3263 * 3. If requesting a order 0 page it will not be compound
3264 * due to the check to see if order has a value in prep_new_page
3265 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3266 * code in gfp_to_alloc_flags that should be enforcing this.
3268 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3270 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
3273 static inline struct page *dev_alloc_pages(unsigned int order)
3275 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
3279 * __dev_alloc_page - allocate a page for network Rx
3280 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3282 * Allocate a new page.
3284 * %NULL is returned if there is no free memory.
3286 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
3288 return __dev_alloc_pages(gfp_mask, 0);
3291 static inline struct page *dev_alloc_page(void)
3293 return dev_alloc_pages(0);
3297 * dev_page_is_reusable - check whether a page can be reused for network Rx
3298 * @page: the page to test
3300 * A page shouldn't be considered for reusing/recycling if it was allocated
3301 * under memory pressure or at a distant memory node.
3303 * Returns false if this page should be returned to page allocator, true
3306 static inline bool dev_page_is_reusable(const struct page *page)
3308 return likely(page_to_nid(page) == numa_mem_id() &&
3309 !page_is_pfmemalloc(page));
3313 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3314 * @page: The page that was allocated from skb_alloc_page
3315 * @skb: The skb that may need pfmemalloc set
3317 static inline void skb_propagate_pfmemalloc(const struct page *page,
3318 struct sk_buff *skb)
3320 if (page_is_pfmemalloc(page))
3321 skb->pfmemalloc = true;
3325 * skb_frag_off() - Returns the offset of a skb fragment
3326 * @frag: the paged fragment
3328 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3330 return frag->bv_offset;
3334 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3335 * @frag: skb fragment
3336 * @delta: value to add
3338 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3340 frag->bv_offset += delta;
3344 * skb_frag_off_set() - Sets the offset of a skb fragment
3345 * @frag: skb fragment
3346 * @offset: offset of fragment
3348 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3350 frag->bv_offset = offset;
3354 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3355 * @fragto: skb fragment where offset is set
3356 * @fragfrom: skb fragment offset is copied from
3358 static inline void skb_frag_off_copy(skb_frag_t *fragto,
3359 const skb_frag_t *fragfrom)
3361 fragto->bv_offset = fragfrom->bv_offset;
3365 * skb_frag_page - retrieve the page referred to by a paged fragment
3366 * @frag: the paged fragment
3368 * Returns the &struct page associated with @frag.
3370 static inline struct page *skb_frag_page(const skb_frag_t *frag)
3372 return frag->bv_page;
3376 * __skb_frag_ref - take an addition reference on a paged fragment.
3377 * @frag: the paged fragment
3379 * Takes an additional reference on the paged fragment @frag.
3381 static inline void __skb_frag_ref(skb_frag_t *frag)
3383 get_page(skb_frag_page(frag));
3387 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
3389 * @f: the fragment offset.
3391 * Takes an additional reference on the @f'th paged fragment of @skb.
3393 static inline void skb_frag_ref(struct sk_buff *skb, int f)
3395 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
3399 * __skb_frag_unref - release a reference on a paged fragment.
3400 * @frag: the paged fragment
3401 * @recycle: recycle the page if allocated via page_pool
3403 * Releases a reference on the paged fragment @frag
3404 * or recycles the page via the page_pool API.
3406 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle)
3408 struct page *page = skb_frag_page(frag);
3410 #ifdef CONFIG_PAGE_POOL
3411 if (recycle && page_pool_return_skb_page(page))
3418 * skb_frag_unref - release a reference on a paged fragment of an skb.
3420 * @f: the fragment offset
3422 * Releases a reference on the @f'th paged fragment of @skb.
3424 static inline void skb_frag_unref(struct sk_buff *skb, int f)
3426 struct skb_shared_info *shinfo = skb_shinfo(skb);
3428 if (!skb_zcopy_managed(skb))
3429 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle);
3433 * skb_frag_address - gets the address of the data contained in a paged fragment
3434 * @frag: the paged fragment buffer
3436 * Returns the address of the data within @frag. The page must already
3439 static inline void *skb_frag_address(const skb_frag_t *frag)
3441 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3445 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3446 * @frag: the paged fragment buffer
3448 * Returns the address of the data within @frag. Checks that the page
3449 * is mapped and returns %NULL otherwise.
3451 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3453 void *ptr = page_address(skb_frag_page(frag));
3457 return ptr + skb_frag_off(frag);
3461 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3462 * @fragto: skb fragment where page is set
3463 * @fragfrom: skb fragment page is copied from
3465 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3466 const skb_frag_t *fragfrom)
3468 fragto->bv_page = fragfrom->bv_page;
3472 * __skb_frag_set_page - sets the page contained in a paged fragment
3473 * @frag: the paged fragment
3474 * @page: the page to set
3476 * Sets the fragment @frag to contain @page.
3478 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3480 frag->bv_page = page;
3484 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3486 * @f: the fragment offset
3487 * @page: the page to set
3489 * Sets the @f'th fragment of @skb to contain @page.
3491 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3494 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3497 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3500 * skb_frag_dma_map - maps a paged fragment via the DMA API
3501 * @dev: the device to map the fragment to
3502 * @frag: the paged fragment to map
3503 * @offset: the offset within the fragment (starting at the
3504 * fragment's own offset)
3505 * @size: the number of bytes to map
3506 * @dir: the direction of the mapping (``PCI_DMA_*``)
3508 * Maps the page associated with @frag to @device.
3510 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3511 const skb_frag_t *frag,
3512 size_t offset, size_t size,
3513 enum dma_data_direction dir)
3515 return dma_map_page(dev, skb_frag_page(frag),
3516 skb_frag_off(frag) + offset, size, dir);
3519 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3522 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3526 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3529 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3534 * skb_clone_writable - is the header of a clone writable
3535 * @skb: buffer to check
3536 * @len: length up to which to write
3538 * Returns true if modifying the header part of the cloned buffer
3539 * does not requires the data to be copied.
3541 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3543 return !skb_header_cloned(skb) &&
3544 skb_headroom(skb) + len <= skb->hdr_len;
3547 static inline int skb_try_make_writable(struct sk_buff *skb,
3548 unsigned int write_len)
3550 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3551 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3554 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3559 if (headroom > skb_headroom(skb))
3560 delta = headroom - skb_headroom(skb);
3562 if (delta || cloned)
3563 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3569 * skb_cow - copy header of skb when it is required
3570 * @skb: buffer to cow
3571 * @headroom: needed headroom
3573 * If the skb passed lacks sufficient headroom or its data part
3574 * is shared, data is reallocated. If reallocation fails, an error
3575 * is returned and original skb is not changed.
3577 * The result is skb with writable area skb->head...skb->tail
3578 * and at least @headroom of space at head.
3580 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3582 return __skb_cow(skb, headroom, skb_cloned(skb));
3586 * skb_cow_head - skb_cow but only making the head writable
3587 * @skb: buffer to cow
3588 * @headroom: needed headroom
3590 * This function is identical to skb_cow except that we replace the
3591 * skb_cloned check by skb_header_cloned. It should be used when
3592 * you only need to push on some header and do not need to modify
3595 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3597 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3601 * skb_padto - pad an skbuff up to a minimal size
3602 * @skb: buffer to pad
3603 * @len: minimal length
3605 * Pads up a buffer to ensure the trailing bytes exist and are
3606 * blanked. If the buffer already contains sufficient data it
3607 * is untouched. Otherwise it is extended. Returns zero on
3608 * success. The skb is freed on error.
3610 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3612 unsigned int size = skb->len;
3613 if (likely(size >= len))
3615 return skb_pad(skb, len - size);
3619 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3620 * @skb: buffer to pad
3621 * @len: minimal length
3622 * @free_on_error: free buffer on error
3624 * Pads up a buffer to ensure the trailing bytes exist and are
3625 * blanked. If the buffer already contains sufficient data it
3626 * is untouched. Otherwise it is extended. Returns zero on
3627 * success. The skb is freed on error if @free_on_error is true.
3629 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3633 unsigned int size = skb->len;
3635 if (unlikely(size < len)) {
3637 if (__skb_pad(skb, len, free_on_error))
3639 __skb_put(skb, len);
3645 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3646 * @skb: buffer to pad
3647 * @len: minimal length
3649 * Pads up a buffer to ensure the trailing bytes exist and are
3650 * blanked. If the buffer already contains sufficient data it
3651 * is untouched. Otherwise it is extended. Returns zero on
3652 * success. The skb is freed on error.
3654 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3656 return __skb_put_padto(skb, len, true);
3659 static inline int skb_add_data(struct sk_buff *skb,
3660 struct iov_iter *from, int copy)
3662 const int off = skb->len;
3664 if (skb->ip_summed == CHECKSUM_NONE) {
3666 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3668 skb->csum = csum_block_add(skb->csum, csum, off);
3671 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3674 __skb_trim(skb, off);
3678 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3679 const struct page *page, int off)
3684 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3686 return page == skb_frag_page(frag) &&
3687 off == skb_frag_off(frag) + skb_frag_size(frag);
3692 static inline int __skb_linearize(struct sk_buff *skb)
3694 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3698 * skb_linearize - convert paged skb to linear one
3699 * @skb: buffer to linarize
3701 * If there is no free memory -ENOMEM is returned, otherwise zero
3702 * is returned and the old skb data released.
3704 static inline int skb_linearize(struct sk_buff *skb)
3706 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3710 * skb_has_shared_frag - can any frag be overwritten
3711 * @skb: buffer to test
3713 * Return true if the skb has at least one frag that might be modified
3714 * by an external entity (as in vmsplice()/sendfile())
3716 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3718 return skb_is_nonlinear(skb) &&
3719 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3723 * skb_linearize_cow - make sure skb is linear and writable
3724 * @skb: buffer to process
3726 * If there is no free memory -ENOMEM is returned, otherwise zero
3727 * is returned and the old skb data released.
3729 static inline int skb_linearize_cow(struct sk_buff *skb)
3731 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3732 __skb_linearize(skb) : 0;
3735 static __always_inline void
3736 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3739 if (skb->ip_summed == CHECKSUM_COMPLETE)
3740 skb->csum = csum_block_sub(skb->csum,
3741 csum_partial(start, len, 0), off);
3742 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3743 skb_checksum_start_offset(skb) < 0)
3744 skb->ip_summed = CHECKSUM_NONE;
3748 * skb_postpull_rcsum - update checksum for received skb after pull
3749 * @skb: buffer to update
3750 * @start: start of data before pull
3751 * @len: length of data pulled
3753 * After doing a pull on a received packet, you need to call this to
3754 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3755 * CHECKSUM_NONE so that it can be recomputed from scratch.
3757 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3758 const void *start, unsigned int len)
3760 if (skb->ip_summed == CHECKSUM_COMPLETE)
3761 skb->csum = wsum_negate(csum_partial(start, len,
3762 wsum_negate(skb->csum)));
3763 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3764 skb_checksum_start_offset(skb) < 0)
3765 skb->ip_summed = CHECKSUM_NONE;
3768 static __always_inline void
3769 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3772 if (skb->ip_summed == CHECKSUM_COMPLETE)
3773 skb->csum = csum_block_add(skb->csum,
3774 csum_partial(start, len, 0), off);
3778 * skb_postpush_rcsum - update checksum for received skb after push
3779 * @skb: buffer to update
3780 * @start: start of data after push
3781 * @len: length of data pushed
3783 * After doing a push on a received packet, you need to call this to
3784 * update the CHECKSUM_COMPLETE checksum.
3786 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3787 const void *start, unsigned int len)
3789 __skb_postpush_rcsum(skb, start, len, 0);
3792 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3795 * skb_push_rcsum - push skb and update receive checksum
3796 * @skb: buffer to update
3797 * @len: length of data pulled
3799 * This function performs an skb_push on the packet and updates
3800 * the CHECKSUM_COMPLETE checksum. It should be used on
3801 * receive path processing instead of skb_push unless you know
3802 * that the checksum difference is zero (e.g., a valid IP header)
3803 * or you are setting ip_summed to CHECKSUM_NONE.
3805 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3808 skb_postpush_rcsum(skb, skb->data, len);
3812 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3814 * pskb_trim_rcsum - trim received skb and update checksum
3815 * @skb: buffer to trim
3818 * This is exactly the same as pskb_trim except that it ensures the
3819 * checksum of received packets are still valid after the operation.
3820 * It can change skb pointers.
3823 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3825 if (likely(len >= skb->len))
3827 return pskb_trim_rcsum_slow(skb, len);
3830 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3832 if (skb->ip_summed == CHECKSUM_COMPLETE)
3833 skb->ip_summed = CHECKSUM_NONE;
3834 __skb_trim(skb, len);
3838 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3840 if (skb->ip_summed == CHECKSUM_COMPLETE)
3841 skb->ip_summed = CHECKSUM_NONE;
3842 return __skb_grow(skb, len);
3845 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3846 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3847 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3848 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3849 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3851 #define skb_queue_walk(queue, skb) \
3852 for (skb = (queue)->next; \
3853 skb != (struct sk_buff *)(queue); \
3856 #define skb_queue_walk_safe(queue, skb, tmp) \
3857 for (skb = (queue)->next, tmp = skb->next; \
3858 skb != (struct sk_buff *)(queue); \
3859 skb = tmp, tmp = skb->next)
3861 #define skb_queue_walk_from(queue, skb) \
3862 for (; skb != (struct sk_buff *)(queue); \
3865 #define skb_rbtree_walk(skb, root) \
3866 for (skb = skb_rb_first(root); skb != NULL; \
3867 skb = skb_rb_next(skb))
3869 #define skb_rbtree_walk_from(skb) \
3870 for (; skb != NULL; \
3871 skb = skb_rb_next(skb))
3873 #define skb_rbtree_walk_from_safe(skb, tmp) \
3874 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3877 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3878 for (tmp = skb->next; \
3879 skb != (struct sk_buff *)(queue); \
3880 skb = tmp, tmp = skb->next)
3882 #define skb_queue_reverse_walk(queue, skb) \
3883 for (skb = (queue)->prev; \
3884 skb != (struct sk_buff *)(queue); \
3887 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3888 for (skb = (queue)->prev, tmp = skb->prev; \
3889 skb != (struct sk_buff *)(queue); \
3890 skb = tmp, tmp = skb->prev)
3892 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3893 for (tmp = skb->prev; \
3894 skb != (struct sk_buff *)(queue); \
3895 skb = tmp, tmp = skb->prev)
3897 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3899 return skb_shinfo(skb)->frag_list != NULL;
3902 static inline void skb_frag_list_init(struct sk_buff *skb)
3904 skb_shinfo(skb)->frag_list = NULL;
3907 #define skb_walk_frags(skb, iter) \
3908 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3911 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3912 int *err, long *timeo_p,
3913 const struct sk_buff *skb);
3914 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3915 struct sk_buff_head *queue,
3918 struct sk_buff **last);
3919 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
3920 struct sk_buff_head *queue,
3921 unsigned int flags, int *off, int *err,
3922 struct sk_buff **last);
3923 struct sk_buff *__skb_recv_datagram(struct sock *sk,
3924 struct sk_buff_head *sk_queue,
3925 unsigned int flags, int *off, int *err);
3926 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
3927 __poll_t datagram_poll(struct file *file, struct socket *sock,
3928 struct poll_table_struct *wait);
3929 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3930 struct iov_iter *to, int size);
3931 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3932 struct msghdr *msg, int size)
3934 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3936 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3937 struct msghdr *msg);
3938 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3939 struct iov_iter *to, int len,
3940 struct ahash_request *hash);
3941 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3942 struct iov_iter *from, int len);
3943 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3944 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3945 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3946 static inline void skb_free_datagram_locked(struct sock *sk,
3947 struct sk_buff *skb)
3949 __skb_free_datagram_locked(sk, skb, 0);
3951 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3952 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3953 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3954 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3956 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3957 struct pipe_inode_info *pipe, unsigned int len,
3958 unsigned int flags);
3959 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3961 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3962 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3963 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3964 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3966 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3967 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3968 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3969 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3970 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3971 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3972 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
3973 unsigned int offset);
3974 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3975 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
3976 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3977 int skb_vlan_pop(struct sk_buff *skb);
3978 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3979 int skb_eth_pop(struct sk_buff *skb);
3980 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
3981 const unsigned char *src);
3982 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
3983 int mac_len, bool ethernet);
3984 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
3986 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3987 int skb_mpls_dec_ttl(struct sk_buff *skb);
3988 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3991 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3993 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3996 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3998 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
4001 struct skb_checksum_ops {
4002 __wsum (*update)(const void *mem, int len, __wsum wsum);
4003 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
4006 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
4008 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
4009 __wsum csum, const struct skb_checksum_ops *ops);
4010 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
4013 static inline void * __must_check
4014 __skb_header_pointer(const struct sk_buff *skb, int offset, int len,
4015 const void *data, int hlen, void *buffer)
4017 if (likely(hlen - offset >= len))
4018 return (void *)data + offset;
4020 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
4026 static inline void * __must_check
4027 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
4029 return __skb_header_pointer(skb, offset, len, skb->data,
4030 skb_headlen(skb), buffer);
4034 * skb_needs_linearize - check if we need to linearize a given skb
4035 * depending on the given device features.
4036 * @skb: socket buffer to check
4037 * @features: net device features
4039 * Returns true if either:
4040 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
4041 * 2. skb is fragmented and the device does not support SG.
4043 static inline bool skb_needs_linearize(struct sk_buff *skb,
4044 netdev_features_t features)
4046 return skb_is_nonlinear(skb) &&
4047 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
4048 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
4051 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
4053 const unsigned int len)
4055 memcpy(to, skb->data, len);
4058 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4059 const int offset, void *to,
4060 const unsigned int len)
4062 memcpy(to, skb->data + offset, len);
4065 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4067 const unsigned int len)
4069 memcpy(skb->data, from, len);
4072 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4075 const unsigned int len)
4077 memcpy(skb->data + offset, from, len);
4080 void skb_init(void);
4082 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4088 * skb_get_timestamp - get timestamp from a skb
4089 * @skb: skb to get stamp from
4090 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4092 * Timestamps are stored in the skb as offsets to a base timestamp.
4093 * This function converts the offset back to a struct timeval and stores
4096 static inline void skb_get_timestamp(const struct sk_buff *skb,
4097 struct __kernel_old_timeval *stamp)
4099 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4102 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4103 struct __kernel_sock_timeval *stamp)
4105 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4107 stamp->tv_sec = ts.tv_sec;
4108 stamp->tv_usec = ts.tv_nsec / 1000;
4111 static inline void skb_get_timestampns(const struct sk_buff *skb,
4112 struct __kernel_old_timespec *stamp)
4114 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4116 stamp->tv_sec = ts.tv_sec;
4117 stamp->tv_nsec = ts.tv_nsec;
4120 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4121 struct __kernel_timespec *stamp)
4123 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4125 stamp->tv_sec = ts.tv_sec;
4126 stamp->tv_nsec = ts.tv_nsec;
4129 static inline void __net_timestamp(struct sk_buff *skb)
4131 skb->tstamp = ktime_get_real();
4132 skb->mono_delivery_time = 0;
4135 static inline ktime_t net_timedelta(ktime_t t)
4137 return ktime_sub(ktime_get_real(), t);
4140 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4144 skb->mono_delivery_time = kt && mono;
4147 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4149 /* It is used in the ingress path to clear the delivery_time.
4150 * If needed, set the skb->tstamp to the (rcv) timestamp.
4152 static inline void skb_clear_delivery_time(struct sk_buff *skb)
4154 if (skb->mono_delivery_time) {
4155 skb->mono_delivery_time = 0;
4156 if (static_branch_unlikely(&netstamp_needed_key))
4157 skb->tstamp = ktime_get_real();
4163 static inline void skb_clear_tstamp(struct sk_buff *skb)
4165 if (skb->mono_delivery_time)
4171 static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4173 if (skb->mono_delivery_time)
4179 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4181 if (!skb->mono_delivery_time && skb->tstamp)
4184 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4185 return ktime_get_real();
4190 static inline u8 skb_metadata_len(const struct sk_buff *skb)
4192 return skb_shinfo(skb)->meta_len;
4195 static inline void *skb_metadata_end(const struct sk_buff *skb)
4197 return skb_mac_header(skb);
4200 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4201 const struct sk_buff *skb_b,
4204 const void *a = skb_metadata_end(skb_a);
4205 const void *b = skb_metadata_end(skb_b);
4206 /* Using more efficient varaiant than plain call to memcmp(). */
4207 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
4211 #define __it(x, op) (x -= sizeof(u##op))
4212 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4213 case 32: diffs |= __it_diff(a, b, 64);
4215 case 24: diffs |= __it_diff(a, b, 64);
4217 case 16: diffs |= __it_diff(a, b, 64);
4219 case 8: diffs |= __it_diff(a, b, 64);
4221 case 28: diffs |= __it_diff(a, b, 64);
4223 case 20: diffs |= __it_diff(a, b, 64);
4225 case 12: diffs |= __it_diff(a, b, 64);
4227 case 4: diffs |= __it_diff(a, b, 32);
4232 return memcmp(a - meta_len, b - meta_len, meta_len);
4236 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4237 const struct sk_buff *skb_b)
4239 u8 len_a = skb_metadata_len(skb_a);
4240 u8 len_b = skb_metadata_len(skb_b);
4242 if (!(len_a | len_b))
4245 return len_a != len_b ?
4246 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4249 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4251 skb_shinfo(skb)->meta_len = meta_len;
4254 static inline void skb_metadata_clear(struct sk_buff *skb)
4256 skb_metadata_set(skb, 0);
4259 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4261 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4263 void skb_clone_tx_timestamp(struct sk_buff *skb);
4264 bool skb_defer_rx_timestamp(struct sk_buff *skb);
4266 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4268 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4272 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4277 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4280 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4282 * PHY drivers may accept clones of transmitted packets for
4283 * timestamping via their phy_driver.txtstamp method. These drivers
4284 * must call this function to return the skb back to the stack with a
4287 * @skb: clone of the original outgoing packet
4288 * @hwtstamps: hardware time stamps
4291 void skb_complete_tx_timestamp(struct sk_buff *skb,
4292 struct skb_shared_hwtstamps *hwtstamps);
4294 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4295 struct skb_shared_hwtstamps *hwtstamps,
4296 struct sock *sk, int tstype);
4299 * skb_tstamp_tx - queue clone of skb with send time stamps
4300 * @orig_skb: the original outgoing packet
4301 * @hwtstamps: hardware time stamps, may be NULL if not available
4303 * If the skb has a socket associated, then this function clones the
4304 * skb (thus sharing the actual data and optional structures), stores
4305 * the optional hardware time stamping information (if non NULL) or
4306 * generates a software time stamp (otherwise), then queues the clone
4307 * to the error queue of the socket. Errors are silently ignored.
4309 void skb_tstamp_tx(struct sk_buff *orig_skb,
4310 struct skb_shared_hwtstamps *hwtstamps);
4313 * skb_tx_timestamp() - Driver hook for transmit timestamping
4315 * Ethernet MAC Drivers should call this function in their hard_xmit()
4316 * function immediately before giving the sk_buff to the MAC hardware.
4318 * Specifically, one should make absolutely sure that this function is
4319 * called before TX completion of this packet can trigger. Otherwise
4320 * the packet could potentially already be freed.
4322 * @skb: A socket buffer.
4324 static inline void skb_tx_timestamp(struct sk_buff *skb)
4326 skb_clone_tx_timestamp(skb);
4327 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4328 skb_tstamp_tx(skb, NULL);
4332 * skb_complete_wifi_ack - deliver skb with wifi status
4334 * @skb: the original outgoing packet
4335 * @acked: ack status
4338 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4340 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4341 __sum16 __skb_checksum_complete(struct sk_buff *skb);
4343 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4345 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4347 (skb->ip_summed == CHECKSUM_PARTIAL &&
4348 skb_checksum_start_offset(skb) >= 0));
4352 * skb_checksum_complete - Calculate checksum of an entire packet
4353 * @skb: packet to process
4355 * This function calculates the checksum over the entire packet plus
4356 * the value of skb->csum. The latter can be used to supply the
4357 * checksum of a pseudo header as used by TCP/UDP. It returns the
4360 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4361 * this function can be used to verify that checksum on received
4362 * packets. In that case the function should return zero if the
4363 * checksum is correct. In particular, this function will return zero
4364 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4365 * hardware has already verified the correctness of the checksum.
4367 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4369 return skb_csum_unnecessary(skb) ?
4370 0 : __skb_checksum_complete(skb);
4373 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4375 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4376 if (skb->csum_level == 0)
4377 skb->ip_summed = CHECKSUM_NONE;
4383 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4385 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4386 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4388 } else if (skb->ip_summed == CHECKSUM_NONE) {
4389 skb->ip_summed = CHECKSUM_UNNECESSARY;
4390 skb->csum_level = 0;
4394 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4396 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4397 skb->ip_summed = CHECKSUM_NONE;
4398 skb->csum_level = 0;
4402 /* Check if we need to perform checksum complete validation.
4404 * Returns true if checksum complete is needed, false otherwise
4405 * (either checksum is unnecessary or zero checksum is allowed).
4407 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4411 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4412 skb->csum_valid = 1;
4413 __skb_decr_checksum_unnecessary(skb);
4420 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4423 #define CHECKSUM_BREAK 76
4425 /* Unset checksum-complete
4427 * Unset checksum complete can be done when packet is being modified
4428 * (uncompressed for instance) and checksum-complete value is
4431 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4433 if (skb->ip_summed == CHECKSUM_COMPLETE)
4434 skb->ip_summed = CHECKSUM_NONE;
4437 /* Validate (init) checksum based on checksum complete.
4440 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4441 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4442 * checksum is stored in skb->csum for use in __skb_checksum_complete
4443 * non-zero: value of invalid checksum
4446 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4450 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4451 if (!csum_fold(csum_add(psum, skb->csum))) {
4452 skb->csum_valid = 1;
4459 if (complete || skb->len <= CHECKSUM_BREAK) {
4462 csum = __skb_checksum_complete(skb);
4463 skb->csum_valid = !csum;
4470 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4475 /* Perform checksum validate (init). Note that this is a macro since we only
4476 * want to calculate the pseudo header which is an input function if necessary.
4477 * First we try to validate without any computation (checksum unnecessary) and
4478 * then calculate based on checksum complete calling the function to compute
4482 * 0: checksum is validated or try to in skb_checksum_complete
4483 * non-zero: value of invalid checksum
4485 #define __skb_checksum_validate(skb, proto, complete, \
4486 zero_okay, check, compute_pseudo) \
4488 __sum16 __ret = 0; \
4489 skb->csum_valid = 0; \
4490 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4491 __ret = __skb_checksum_validate_complete(skb, \
4492 complete, compute_pseudo(skb, proto)); \
4496 #define skb_checksum_init(skb, proto, compute_pseudo) \
4497 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4499 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4500 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4502 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4503 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4505 #define skb_checksum_validate_zero_check(skb, proto, check, \
4507 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4509 #define skb_checksum_simple_validate(skb) \
4510 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4512 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4514 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4517 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4519 skb->csum = ~pseudo;
4520 skb->ip_summed = CHECKSUM_COMPLETE;
4523 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4525 if (__skb_checksum_convert_check(skb)) \
4526 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4529 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4530 u16 start, u16 offset)
4532 skb->ip_summed = CHECKSUM_PARTIAL;
4533 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4534 skb->csum_offset = offset - start;
4537 /* Update skbuf and packet to reflect the remote checksum offload operation.
4538 * When called, ptr indicates the starting point for skb->csum when
4539 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4540 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4542 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4543 int start, int offset, bool nopartial)
4548 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4552 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4553 __skb_checksum_complete(skb);
4554 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4557 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4559 /* Adjust skb->csum since we changed the packet */
4560 skb->csum = csum_add(skb->csum, delta);
4563 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4565 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4566 return (void *)(skb->_nfct & NFCT_PTRMASK);
4572 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4574 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4581 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4583 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4584 skb->slow_gro |= !!nfct;
4589 #ifdef CONFIG_SKB_EXTENSIONS
4591 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4597 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4600 #if IS_ENABLED(CONFIG_MPTCP)
4603 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4606 SKB_EXT_NUM, /* must be last */
4610 * struct skb_ext - sk_buff extensions
4611 * @refcnt: 1 on allocation, deallocated on 0
4612 * @offset: offset to add to @data to obtain extension address
4613 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4614 * @data: start of extension data, variable sized
4616 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4617 * to use 'u8' types while allowing up to 2kb worth of extension data.
4621 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4622 u8 chunks; /* same */
4623 char data[] __aligned(8);
4626 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4627 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4628 struct skb_ext *ext);
4629 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4630 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4631 void __skb_ext_put(struct skb_ext *ext);
4633 static inline void skb_ext_put(struct sk_buff *skb)
4635 if (skb->active_extensions)
4636 __skb_ext_put(skb->extensions);
4639 static inline void __skb_ext_copy(struct sk_buff *dst,
4640 const struct sk_buff *src)
4642 dst->active_extensions = src->active_extensions;
4644 if (src->active_extensions) {
4645 struct skb_ext *ext = src->extensions;
4647 refcount_inc(&ext->refcnt);
4648 dst->extensions = ext;
4652 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4655 __skb_ext_copy(dst, src);
4658 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4660 return !!ext->offset[i];
4663 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4665 return skb->active_extensions & (1 << id);
4668 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4670 if (skb_ext_exist(skb, id))
4671 __skb_ext_del(skb, id);
4674 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4676 if (skb_ext_exist(skb, id)) {
4677 struct skb_ext *ext = skb->extensions;
4679 return (void *)ext + (ext->offset[id] << 3);
4685 static inline void skb_ext_reset(struct sk_buff *skb)
4687 if (unlikely(skb->active_extensions)) {
4688 __skb_ext_put(skb->extensions);
4689 skb->active_extensions = 0;
4693 static inline bool skb_has_extensions(struct sk_buff *skb)
4695 return unlikely(skb->active_extensions);
4698 static inline void skb_ext_put(struct sk_buff *skb) {}
4699 static inline void skb_ext_reset(struct sk_buff *skb) {}
4700 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4701 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4702 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4703 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4704 #endif /* CONFIG_SKB_EXTENSIONS */
4706 static inline void nf_reset_ct(struct sk_buff *skb)
4708 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4709 nf_conntrack_put(skb_nfct(skb));
4714 static inline void nf_reset_trace(struct sk_buff *skb)
4716 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4721 static inline void ipvs_reset(struct sk_buff *skb)
4723 #if IS_ENABLED(CONFIG_IP_VS)
4724 skb->ipvs_property = 0;
4728 /* Note: This doesn't put any conntrack info in dst. */
4729 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4732 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4733 dst->_nfct = src->_nfct;
4734 nf_conntrack_get(skb_nfct(src));
4736 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4738 dst->nf_trace = src->nf_trace;
4742 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4744 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4745 nf_conntrack_put(skb_nfct(dst));
4747 dst->slow_gro = src->slow_gro;
4748 __nf_copy(dst, src, true);
4751 #ifdef CONFIG_NETWORK_SECMARK
4752 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4754 to->secmark = from->secmark;
4757 static inline void skb_init_secmark(struct sk_buff *skb)
4762 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4765 static inline void skb_init_secmark(struct sk_buff *skb)
4769 static inline int secpath_exists(const struct sk_buff *skb)
4772 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4778 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4780 return !skb->destructor &&
4781 !secpath_exists(skb) &&
4783 !skb->_skb_refdst &&
4784 !skb_has_frag_list(skb);
4787 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4789 skb->queue_mapping = queue_mapping;
4792 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4794 return skb->queue_mapping;
4797 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4799 to->queue_mapping = from->queue_mapping;
4802 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4804 skb->queue_mapping = rx_queue + 1;
4807 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4809 return skb->queue_mapping - 1;
4812 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4814 return skb->queue_mapping != 0;
4817 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4819 skb->dst_pending_confirm = val;
4822 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4824 return skb->dst_pending_confirm != 0;
4827 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4830 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4836 /* Keeps track of mac header offset relative to skb->head.
4837 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4838 * For non-tunnel skb it points to skb_mac_header() and for
4839 * tunnel skb it points to outer mac header.
4840 * Keeps track of level of encapsulation of network headers.
4851 #define SKB_GSO_CB_OFFSET 32
4852 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET))
4854 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4856 return (skb_mac_header(inner_skb) - inner_skb->head) -
4857 SKB_GSO_CB(inner_skb)->mac_offset;
4860 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4862 int new_headroom, headroom;
4865 headroom = skb_headroom(skb);
4866 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4870 new_headroom = skb_headroom(skb);
4871 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4875 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4877 /* Do not update partial checksums if remote checksum is enabled. */
4878 if (skb->remcsum_offload)
4881 SKB_GSO_CB(skb)->csum = res;
4882 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4885 /* Compute the checksum for a gso segment. First compute the checksum value
4886 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4887 * then add in skb->csum (checksum from csum_start to end of packet).
4888 * skb->csum and csum_start are then updated to reflect the checksum of the
4889 * resultant packet starting from the transport header-- the resultant checksum
4890 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4893 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4895 unsigned char *csum_start = skb_transport_header(skb);
4896 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4897 __wsum partial = SKB_GSO_CB(skb)->csum;
4899 SKB_GSO_CB(skb)->csum = res;
4900 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4902 return csum_fold(csum_partial(csum_start, plen, partial));
4905 static inline bool skb_is_gso(const struct sk_buff *skb)
4907 return skb_shinfo(skb)->gso_size;
4910 /* Note: Should be called only if skb_is_gso(skb) is true */
4911 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4913 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4916 /* Note: Should be called only if skb_is_gso(skb) is true */
4917 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4919 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4922 /* Note: Should be called only if skb_is_gso(skb) is true */
4923 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4925 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4928 static inline void skb_gso_reset(struct sk_buff *skb)
4930 skb_shinfo(skb)->gso_size = 0;
4931 skb_shinfo(skb)->gso_segs = 0;
4932 skb_shinfo(skb)->gso_type = 0;
4935 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4938 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4940 shinfo->gso_size += increment;
4943 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4946 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4948 shinfo->gso_size -= decrement;
4951 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4953 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4955 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4956 * wanted then gso_type will be set. */
4957 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4959 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4960 unlikely(shinfo->gso_type == 0)) {
4961 __skb_warn_lro_forwarding(skb);
4967 static inline void skb_forward_csum(struct sk_buff *skb)
4969 /* Unfortunately we don't support this one. Any brave souls? */
4970 if (skb->ip_summed == CHECKSUM_COMPLETE)
4971 skb->ip_summed = CHECKSUM_NONE;
4975 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4976 * @skb: skb to check
4978 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4979 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4980 * use this helper, to document places where we make this assertion.
4982 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4984 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE);
4987 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4989 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4990 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4991 unsigned int transport_len,
4992 __sum16(*skb_chkf)(struct sk_buff *skb));
4995 * skb_head_is_locked - Determine if the skb->head is locked down
4996 * @skb: skb to check
4998 * The head on skbs build around a head frag can be removed if they are
4999 * not cloned. This function returns true if the skb head is locked down
5000 * due to either being allocated via kmalloc, or by being a clone with
5001 * multiple references to the head.
5003 static inline bool skb_head_is_locked(const struct sk_buff *skb)
5005 return !skb->head_frag || skb_cloned(skb);
5008 /* Local Checksum Offload.
5009 * Compute outer checksum based on the assumption that the
5010 * inner checksum will be offloaded later.
5011 * See Documentation/networking/checksum-offloads.rst for
5012 * explanation of how this works.
5013 * Fill in outer checksum adjustment (e.g. with sum of outer
5014 * pseudo-header) before calling.
5015 * Also ensure that inner checksum is in linear data area.
5017 static inline __wsum lco_csum(struct sk_buff *skb)
5019 unsigned char *csum_start = skb_checksum_start(skb);
5020 unsigned char *l4_hdr = skb_transport_header(skb);
5023 /* Start with complement of inner checksum adjustment */
5024 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
5027 /* Add in checksum of our headers (incl. outer checksum
5028 * adjustment filled in by caller) and return result.
5030 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
5033 static inline bool skb_is_redirected(const struct sk_buff *skb)
5035 return skb->redirected;
5038 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
5040 skb->redirected = 1;
5041 #ifdef CONFIG_NET_REDIRECT
5042 skb->from_ingress = from_ingress;
5043 if (skb->from_ingress)
5044 skb_clear_tstamp(skb);
5048 static inline void skb_reset_redirect(struct sk_buff *skb)
5050 skb->redirected = 0;
5053 static inline bool skb_csum_is_sctp(struct sk_buff *skb)
5055 return skb->csum_not_inet;
5058 static inline void skb_set_kcov_handle(struct sk_buff *skb,
5059 const u64 kcov_handle)
5062 skb->kcov_handle = kcov_handle;
5066 static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
5069 return skb->kcov_handle;
5075 #ifdef CONFIG_PAGE_POOL
5076 static inline void skb_mark_for_recycle(struct sk_buff *skb)
5078 skb->pp_recycle = 1;
5082 #endif /* __KERNEL__ */
5083 #endif /* _LINUX_SKBUFF_H */