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 <linux/net.h>
27 #include <linux/textsearch.h>
28 #include <net/checksum.h>
29 #include <linux/rcupdate.h>
30 #include <linux/hrtimer.h>
31 #include <linux/dma-mapping.h>
32 #include <linux/netdev_features.h>
33 #include <linux/sched.h>
34 #include <linux/sched/clock.h>
35 #include <net/flow_dissector.h>
36 #include <linux/splice.h>
37 #include <linux/in6.h>
38 #include <linux/if_packet.h>
39 #include <linux/llist.h>
41 #include <net/page_pool.h>
42 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
43 #include <linux/netfilter/nf_conntrack_common.h>
45 #include <net/net_debug.h>
50 * The interface for checksum offload between the stack and networking drivers
53 * IP checksum related features
54 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
56 * Drivers advertise checksum offload capabilities in the features of a device.
57 * From the stack's point of view these are capabilities offered by the driver.
58 * A driver typically only advertises features that it is capable of offloading
61 * .. flat-table:: Checksum related device features
64 * * - %NETIF_F_HW_CSUM
65 * - The driver (or its device) is able to compute one
66 * IP (one's complement) checksum for any combination
67 * of protocols or protocol layering. The checksum is
68 * computed and set in a packet per the CHECKSUM_PARTIAL
69 * interface (see below).
71 * * - %NETIF_F_IP_CSUM
72 * - Driver (device) is only able to checksum plain
73 * TCP or UDP packets over IPv4. These are specifically
74 * unencapsulated packets of the form IPv4|TCP or
75 * IPv4|UDP where the Protocol field in the IPv4 header
76 * is TCP or UDP. The IPv4 header may contain IP options.
77 * This feature cannot be set in features for a device
78 * with NETIF_F_HW_CSUM also set. This feature is being
79 * DEPRECATED (see below).
81 * * - %NETIF_F_IPV6_CSUM
82 * - Driver (device) is only able to checksum plain
83 * TCP or UDP packets over IPv6. These are specifically
84 * unencapsulated packets of the form IPv6|TCP or
85 * IPv6|UDP where the Next Header field in the IPv6
86 * header is either TCP or UDP. IPv6 extension headers
87 * are not supported with this feature. This feature
88 * cannot be set in features for a device with
89 * NETIF_F_HW_CSUM also set. This feature is being
90 * DEPRECATED (see below).
93 * - Driver (device) performs receive checksum offload.
94 * This flag is only used to disable the RX checksum
95 * feature for a device. The stack will accept receive
96 * checksum indication in packets received on a device
97 * regardless of whether NETIF_F_RXCSUM is set.
99 * Checksumming of received packets by device
100 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
102 * Indication of checksum verification is set in &sk_buff.ip_summed.
103 * Possible values are:
107 * Device did not checksum this packet e.g. due to lack of capabilities.
108 * The packet contains full (though not verified) checksum in packet but
109 * not in skb->csum. Thus, skb->csum is undefined in this case.
111 * - %CHECKSUM_UNNECESSARY
113 * The hardware you're dealing with doesn't calculate the full checksum
114 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums
115 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY
116 * if their checksums are okay. &sk_buff.csum is still undefined in this case
117 * though. A driver or device must never modify the checksum field in the
118 * packet even if checksum is verified.
120 * %CHECKSUM_UNNECESSARY is applicable to following protocols:
122 * - TCP: IPv6 and IPv4.
123 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
124 * zero UDP checksum for either IPv4 or IPv6, the networking stack
125 * may perform further validation in this case.
126 * - GRE: only if the checksum is present in the header.
127 * - SCTP: indicates the CRC in SCTP header has been validated.
128 * - FCOE: indicates the CRC in FC frame has been validated.
130 * &sk_buff.csum_level indicates the number of consecutive checksums found in
131 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY.
132 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
133 * and a device is able to verify the checksums for UDP (possibly zero),
134 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to
135 * two. If the device were only able to verify the UDP checksum and not
136 * GRE, either because it doesn't support GRE checksum or because GRE
137 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
138 * not considered in this case).
140 * - %CHECKSUM_COMPLETE
142 * This is the most generic way. The device supplied checksum of the _whole_
143 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the
144 * hardware doesn't need to parse L3/L4 headers to implement this.
148 * - Even if device supports only some protocols, but is able to produce
149 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
150 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
152 * - %CHECKSUM_PARTIAL
154 * A checksum is set up to be offloaded to a device as described in the
155 * output description for CHECKSUM_PARTIAL. This may occur on a packet
156 * received directly from another Linux OS, e.g., a virtualized Linux kernel
157 * on the same host, or it may be set in the input path in GRO or remote
158 * checksum offload. For the purposes of checksum verification, the checksum
159 * referred to by skb->csum_start + skb->csum_offset and any preceding
160 * checksums in the packet are considered verified. Any checksums in the
161 * packet that are after the checksum being offloaded are not considered to
164 * Checksumming on transmit for non-GSO
165 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
167 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet.
170 * - %CHECKSUM_PARTIAL
172 * The driver is required to checksum the packet as seen by hard_start_xmit()
173 * from &sk_buff.csum_start up to the end, and to record/write the checksum at
174 * offset &sk_buff.csum_start + &sk_buff.csum_offset.
175 * A driver may verify that the
176 * csum_start and csum_offset values are valid values given the length and
177 * offset of the packet, but it should not attempt to validate that the
178 * checksum refers to a legitimate transport layer checksum -- it is the
179 * purview of the stack to validate that csum_start and csum_offset are set
182 * When the stack requests checksum offload for a packet, the driver MUST
183 * ensure that the checksum is set correctly. A driver can either offload the
184 * checksum calculation to the device, or call skb_checksum_help (in the case
185 * that the device does not support offload for a particular checksum).
187 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of
188 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate
189 * checksum offload capability.
190 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based
191 * on network device checksumming capabilities: if a packet does not match
192 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of
193 * &sk_buff.csum_not_inet, see :ref:`crc`)
194 * is called to resolve the checksum.
198 * The skb was already checksummed by the protocol, or a checksum is not
201 * - %CHECKSUM_UNNECESSARY
203 * This has the same meaning as CHECKSUM_NONE for checksum offload on
206 * - %CHECKSUM_COMPLETE
208 * Not used in checksum output. If a driver observes a packet with this value
209 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set.
213 * Non-IP checksum (CRC) offloads
214 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
219 * * - %NETIF_F_SCTP_CRC
220 * - This feature indicates that a device is capable of
221 * offloading the SCTP CRC in a packet. To perform this offload the stack
222 * will set csum_start and csum_offset accordingly, set ip_summed to
223 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication
224 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c.
225 * A driver that supports both IP checksum offload and SCTP CRC32c offload
226 * must verify which offload is configured for a packet by testing the
227 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to
228 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
230 * * - %NETIF_F_FCOE_CRC
231 * - This feature indicates that a device is capable of offloading the FCOE
232 * CRC in a packet. To perform this offload the stack will set ip_summed
233 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset
234 * accordingly. Note that there is no indication in the skbuff that the
235 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports
236 * both IP checksum offload and FCOE CRC offload must verify which offload
237 * is configured for a packet, presumably by inspecting packet headers.
239 * Checksumming on output with GSO
240 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
242 * In the case of a GSO packet (skb_is_gso() is true), checksum offload
243 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
244 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as
245 * part of the GSO operation is implied. If a checksum is being offloaded
246 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and
247 * csum_offset are set to refer to the outermost checksum being offloaded
248 * (two offloaded checksums are possible with UDP encapsulation).
251 /* Don't change this without changing skb_csum_unnecessary! */
252 #define CHECKSUM_NONE 0
253 #define CHECKSUM_UNNECESSARY 1
254 #define CHECKSUM_COMPLETE 2
255 #define CHECKSUM_PARTIAL 3
257 /* Maximum value in skb->csum_level */
258 #define SKB_MAX_CSUM_LEVEL 3
260 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
261 #define SKB_WITH_OVERHEAD(X) \
262 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
263 #define SKB_MAX_ORDER(X, ORDER) \
264 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
265 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
266 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
268 /* return minimum truesize of one skb containing X bytes of data */
269 #define SKB_TRUESIZE(X) ((X) + \
270 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
271 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
273 struct ahash_request;
276 struct pipe_inode_info;
283 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
284 struct nf_bridge_info {
286 BRNF_PROTO_UNCHANGED,
294 struct net_device *physindev;
296 /* always valid & non-NULL from FORWARD on, for physdev match */
297 struct net_device *physoutdev;
299 /* prerouting: detect dnat in orig/reply direction */
301 struct in6_addr ipv6_daddr;
303 /* after prerouting + nat detected: store original source
304 * mac since neigh resolution overwrites it, only used while
305 * skb is out in neigh layer.
307 char neigh_header[8];
312 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
313 /* Chain in tc_skb_ext will be used to share the tc chain with
314 * ovs recirc_id. It will be set to the current chain by tc
315 * and read by ovs to recirc_id.
327 struct sk_buff_head {
328 /* These two members must be first to match sk_buff. */
329 struct_group_tagged(sk_buff_list, list,
330 struct sk_buff *next;
331 struct sk_buff *prev;
340 /* The reason of skb drop, which is used in kfree_skb_reason().
341 * en...maybe they should be splited by group?
343 * Each item here should also be in 'TRACE_SKB_DROP_REASON', which is
344 * used to translate the reason to string.
346 enum skb_drop_reason {
347 SKB_NOT_DROPPED_YET = 0,
348 SKB_DROP_REASON_NOT_SPECIFIED, /* drop reason is not specified */
349 SKB_DROP_REASON_NO_SOCKET, /* socket not found */
350 SKB_DROP_REASON_PKT_TOO_SMALL, /* packet size is too small */
351 SKB_DROP_REASON_TCP_CSUM, /* TCP checksum error */
352 SKB_DROP_REASON_SOCKET_FILTER, /* dropped by socket filter */
353 SKB_DROP_REASON_UDP_CSUM, /* UDP checksum error */
354 SKB_DROP_REASON_NETFILTER_DROP, /* dropped by netfilter */
355 SKB_DROP_REASON_OTHERHOST, /* packet don't belong to current
356 * host (interface is in promisc
359 SKB_DROP_REASON_IP_CSUM, /* IP checksum error */
360 SKB_DROP_REASON_IP_INHDR, /* there is something wrong with
362 * IPSTATS_MIB_INHDRERRORS)
364 SKB_DROP_REASON_IP_RPFILTER, /* IP rpfilter validate failed.
365 * see the document for rp_filter
366 * in ip-sysctl.rst for more
369 SKB_DROP_REASON_UNICAST_IN_L2_MULTICAST, /* destination address of L2
370 * is multicast, but L3 is
373 SKB_DROP_REASON_XFRM_POLICY, /* xfrm policy check failed */
374 SKB_DROP_REASON_IP_NOPROTO, /* no support for IP protocol */
375 SKB_DROP_REASON_SOCKET_RCVBUFF, /* socket receive buff is full */
376 SKB_DROP_REASON_PROTO_MEM, /* proto memory limition, such as
377 * udp packet drop out of
378 * udp_memory_allocated.
380 SKB_DROP_REASON_TCP_MD5NOTFOUND, /* no MD5 hash and one
381 * expected, corresponding
382 * to LINUX_MIB_TCPMD5NOTFOUND
384 SKB_DROP_REASON_TCP_MD5UNEXPECTED, /* MD5 hash and we're not
385 * expecting one, corresponding
386 * to LINUX_MIB_TCPMD5UNEXPECTED
388 SKB_DROP_REASON_TCP_MD5FAILURE, /* MD5 hash and its wrong,
390 * LINUX_MIB_TCPMD5FAILURE
392 SKB_DROP_REASON_SOCKET_BACKLOG, /* failed to add skb to socket
394 * LINUX_MIB_TCPBACKLOGDROP)
396 SKB_DROP_REASON_TCP_FLAGS, /* TCP flags invalid */
397 SKB_DROP_REASON_TCP_ZEROWINDOW, /* TCP receive window size is zero,
398 * see LINUX_MIB_TCPZEROWINDOWDROP
400 SKB_DROP_REASON_TCP_OLD_DATA, /* the TCP data reveived is already
401 * received before (spurious retrans
403 * LINUX_MIB_DELAYEDACKLOST
405 SKB_DROP_REASON_TCP_OVERWINDOW, /* the TCP data is out of window,
406 * the seq of the first byte exceed
407 * the right edges of receive
410 SKB_DROP_REASON_TCP_OFOMERGE, /* the data of skb is already in
411 * the ofo queue, corresponding to
412 * LINUX_MIB_TCPOFOMERGE
414 SKB_DROP_REASON_TCP_RFC7323_PAWS, /* PAWS check, corresponding to
415 * LINUX_MIB_PAWSESTABREJECTED
417 SKB_DROP_REASON_TCP_INVALID_SEQUENCE, /* Not acceptable SEQ field */
418 SKB_DROP_REASON_TCP_RESET, /* Invalid RST packet */
419 SKB_DROP_REASON_TCP_INVALID_SYN, /* Incoming packet has unexpected SYN flag */
420 SKB_DROP_REASON_TCP_CLOSE, /* TCP socket in CLOSE state */
421 SKB_DROP_REASON_TCP_FASTOPEN, /* dropped by FASTOPEN request socket */
422 SKB_DROP_REASON_TCP_OLD_ACK, /* TCP ACK is old, but in window */
423 SKB_DROP_REASON_TCP_TOO_OLD_ACK, /* TCP ACK is too old */
424 SKB_DROP_REASON_TCP_ACK_UNSENT_DATA, /* TCP ACK for data we haven't sent yet */
425 SKB_DROP_REASON_TCP_OFO_QUEUE_PRUNE, /* pruned from TCP OFO queue */
426 SKB_DROP_REASON_TCP_OFO_DROP, /* data already in receive queue */
427 SKB_DROP_REASON_IP_OUTNOROUTES, /* route lookup failed */
428 SKB_DROP_REASON_BPF_CGROUP_EGRESS, /* dropped by
429 * BPF_PROG_TYPE_CGROUP_SKB
432 SKB_DROP_REASON_IPV6DISABLED, /* IPv6 is disabled on the device */
433 SKB_DROP_REASON_NEIGH_CREATEFAIL, /* failed to create neigh
436 SKB_DROP_REASON_NEIGH_FAILED, /* neigh entry in failed state */
437 SKB_DROP_REASON_NEIGH_QUEUEFULL, /* arp_queue for neigh
440 SKB_DROP_REASON_NEIGH_DEAD, /* neigh entry is dead */
441 SKB_DROP_REASON_TC_EGRESS, /* dropped in TC egress HOOK */
442 SKB_DROP_REASON_QDISC_DROP, /* dropped by qdisc when packet
443 * outputting (failed to enqueue to
446 SKB_DROP_REASON_CPU_BACKLOG, /* failed to enqueue the skb to
447 * the per CPU backlog queue. This
448 * can be caused by backlog queue
449 * full (see netdev_max_backlog in
450 * net.rst) or RPS flow limit
452 SKB_DROP_REASON_XDP, /* dropped by XDP in input path */
453 SKB_DROP_REASON_TC_INGRESS, /* dropped in TC ingress HOOK */
454 SKB_DROP_REASON_UNHANDLED_PROTO, /* protocol not implemented
457 SKB_DROP_REASON_SKB_CSUM, /* sk_buff checksum computation
460 SKB_DROP_REASON_SKB_GSO_SEG, /* gso segmentation error */
461 SKB_DROP_REASON_SKB_UCOPY_FAULT, /* failed to copy data from
462 * user space, e.g., via
463 * zerocopy_sg_from_iter()
464 * or skb_orphan_frags_rx()
466 SKB_DROP_REASON_DEV_HDR, /* device driver specific
467 * header/metadata is invalid
469 /* the device is not ready to xmit/recv due to any of its data
470 * structure that is not up/ready/initialized, e.g., the IFF_UP is
471 * not set, or driver specific tun->tfiles[txq] is not initialized
473 SKB_DROP_REASON_DEV_READY,
474 SKB_DROP_REASON_FULL_RING, /* ring buffer is full */
475 SKB_DROP_REASON_NOMEM, /* error due to OOM */
476 SKB_DROP_REASON_HDR_TRUNC, /* failed to trunc/extract the header
477 * from networking data, e.g., failed
478 * to pull the protocol header from
479 * frags via pskb_may_pull()
481 SKB_DROP_REASON_TAP_FILTER, /* dropped by (ebpf) filter directly
482 * attached to tun/tap, e.g., via
485 SKB_DROP_REASON_TAP_TXFILTER, /* dropped by tx filter implemented
486 * at tun/tap, e.g., check_filter()
488 SKB_DROP_REASON_ICMP_CSUM, /* ICMP checksum error */
489 SKB_DROP_REASON_INVALID_PROTO, /* the packet doesn't follow RFC
490 * 2211, such as a broadcasts
493 SKB_DROP_REASON_IP_INADDRERRORS, /* host unreachable, corresponding
494 * to IPSTATS_MIB_INADDRERRORS
496 SKB_DROP_REASON_IP_INNOROUTES, /* network unreachable, corresponding
497 * to IPSTATS_MIB_INADDRERRORS
499 SKB_DROP_REASON_PKT_TOO_BIG, /* packet size is too big (maybe exceed
505 #define SKB_DR_INIT(name, reason) \
506 enum skb_drop_reason name = SKB_DROP_REASON_##reason
507 #define SKB_DR(name) \
508 SKB_DR_INIT(name, NOT_SPECIFIED)
509 #define SKB_DR_SET(name, reason) \
510 (name = SKB_DROP_REASON_##reason)
511 #define SKB_DR_OR(name, reason) \
513 if (name == SKB_DROP_REASON_NOT_SPECIFIED || \
514 name == SKB_NOT_DROPPED_YET) \
515 SKB_DR_SET(name, reason); \
518 /* To allow 64K frame to be packed as single skb without frag_list we
519 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
520 * buffers which do not start on a page boundary.
522 * Since GRO uses frags we allocate at least 16 regardless of page
525 #if (65536/PAGE_SIZE + 1) < 16
526 #define MAX_SKB_FRAGS 16UL
528 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
530 extern int sysctl_max_skb_frags;
532 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
533 * segment using its current segmentation instead.
535 #define GSO_BY_FRAGS 0xFFFF
537 typedef struct bio_vec skb_frag_t;
540 * skb_frag_size() - Returns the size of a skb fragment
541 * @frag: skb fragment
543 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
549 * skb_frag_size_set() - Sets the size of a skb fragment
550 * @frag: skb fragment
551 * @size: size of fragment
553 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
559 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
560 * @frag: skb fragment
561 * @delta: value to add
563 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
565 frag->bv_len += delta;
569 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
570 * @frag: skb fragment
571 * @delta: value to subtract
573 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
575 frag->bv_len -= delta;
579 * skb_frag_must_loop - Test if %p is a high memory page
580 * @p: fragment's page
582 static inline bool skb_frag_must_loop(struct page *p)
584 #if defined(CONFIG_HIGHMEM)
585 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
592 * skb_frag_foreach_page - loop over pages in a fragment
594 * @f: skb frag to operate on
595 * @f_off: offset from start of f->bv_page
596 * @f_len: length from f_off to loop over
597 * @p: (temp var) current page
598 * @p_off: (temp var) offset from start of current page,
599 * non-zero only on first page.
600 * @p_len: (temp var) length in current page,
601 * < PAGE_SIZE only on first and last page.
602 * @copied: (temp var) length so far, excluding current p_len.
604 * A fragment can hold a compound page, in which case per-page
605 * operations, notably kmap_atomic, must be called for each
608 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
609 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
610 p_off = (f_off) & (PAGE_SIZE - 1), \
611 p_len = skb_frag_must_loop(p) ? \
612 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
615 copied += p_len, p++, p_off = 0, \
616 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
618 #define HAVE_HW_TIME_STAMP
621 * struct skb_shared_hwtstamps - hardware time stamps
622 * @hwtstamp: hardware time stamp transformed into duration
623 * since arbitrary point in time
624 * @netdev_data: address/cookie of network device driver used as
625 * reference to actual hardware time stamp
627 * Software time stamps generated by ktime_get_real() are stored in
630 * hwtstamps can only be compared against other hwtstamps from
633 * This structure is attached to packets as part of the
634 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
636 struct skb_shared_hwtstamps {
643 /* Definitions for tx_flags in struct skb_shared_info */
645 /* generate hardware time stamp */
646 SKBTX_HW_TSTAMP = 1 << 0,
648 /* generate software time stamp when queueing packet to NIC */
649 SKBTX_SW_TSTAMP = 1 << 1,
651 /* device driver is going to provide hardware time stamp */
652 SKBTX_IN_PROGRESS = 1 << 2,
654 /* generate hardware time stamp based on cycles if supported */
655 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3,
657 /* generate wifi status information (where possible) */
658 SKBTX_WIFI_STATUS = 1 << 4,
660 /* determine hardware time stamp based on time or cycles */
661 SKBTX_HW_TSTAMP_NETDEV = 1 << 5,
663 /* generate software time stamp when entering packet scheduling */
664 SKBTX_SCHED_TSTAMP = 1 << 6,
667 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
669 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \
670 SKBTX_HW_TSTAMP_USE_CYCLES | \
673 /* Definitions for flags in struct skb_shared_info */
675 /* use zcopy routines */
676 SKBFL_ZEROCOPY_ENABLE = BIT(0),
678 /* This indicates at least one fragment might be overwritten
679 * (as in vmsplice(), sendfile() ...)
680 * If we need to compute a TX checksum, we'll need to copy
681 * all frags to avoid possible bad checksum
683 SKBFL_SHARED_FRAG = BIT(1),
685 /* segment contains only zerocopy data and should not be
686 * charged to the kernel memory.
688 SKBFL_PURE_ZEROCOPY = BIT(2),
690 SKBFL_DONT_ORPHAN = BIT(3),
692 /* page references are managed by the ubuf_info, so it's safe to
693 * use frags only up until ubuf_info is released
695 SKBFL_MANAGED_FRAG_REFS = BIT(4),
698 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
699 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
700 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
703 * The callback notifies userspace to release buffers when skb DMA is done in
704 * lower device, the skb last reference should be 0 when calling this.
705 * The zerocopy_success argument is true if zero copy transmit occurred,
706 * false on data copy or out of memory error caused by data copy attempt.
707 * The ctx field is used to track device context.
708 * The desc field is used to track userspace buffer index.
711 void (*callback)(struct sk_buff *, struct ubuf_info *,
712 bool zerocopy_success);
729 struct user_struct *user;
734 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
736 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
737 void mm_unaccount_pinned_pages(struct mmpin *mmp);
739 /* This data is invariant across clones and lives at
740 * the end of the header data, ie. at skb->end.
742 struct skb_shared_info {
747 unsigned short gso_size;
748 /* Warning: this field is not always filled in (UFO)! */
749 unsigned short gso_segs;
750 struct sk_buff *frag_list;
751 struct skb_shared_hwtstamps hwtstamps;
752 unsigned int gso_type;
756 * Warning : all fields before dataref are cleared in __alloc_skb()
759 unsigned int xdp_frags_size;
761 /* Intermediate layers must ensure that destructor_arg
762 * remains valid until skb destructor */
763 void * destructor_arg;
765 /* must be last field, see pskb_expand_head() */
766 skb_frag_t frags[MAX_SKB_FRAGS];
770 * DOC: dataref and headerless skbs
772 * Transport layers send out clones of payload skbs they hold for
773 * retransmissions. To allow lower layers of the stack to prepend their headers
774 * we split &skb_shared_info.dataref into two halves.
775 * The lower 16 bits count the overall number of references.
776 * The higher 16 bits indicate how many of the references are payload-only.
777 * skb_header_cloned() checks if skb is allowed to add / write the headers.
779 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
780 * (via __skb_header_release()). Any clone created from marked skb will get
781 * &sk_buff.hdr_len populated with the available headroom.
782 * If there's the only clone in existence it's able to modify the headroom
783 * at will. The sequence of calls inside the transport layer is::
787 * __skb_header_release()
789 * // send the clone down the stack
791 * This is not a very generic construct and it depends on the transport layers
792 * doing the right thing. In practice there's usually only one payload-only skb.
793 * Having multiple payload-only skbs with different lengths of hdr_len is not
794 * possible. The payload-only skbs should never leave their owner.
796 #define SKB_DATAREF_SHIFT 16
797 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
801 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
802 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
803 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
807 SKB_GSO_TCPV4 = 1 << 0,
809 /* This indicates the skb is from an untrusted source. */
810 SKB_GSO_DODGY = 1 << 1,
812 /* This indicates the tcp segment has CWR set. */
813 SKB_GSO_TCP_ECN = 1 << 2,
815 SKB_GSO_TCP_FIXEDID = 1 << 3,
817 SKB_GSO_TCPV6 = 1 << 4,
819 SKB_GSO_FCOE = 1 << 5,
821 SKB_GSO_GRE = 1 << 6,
823 SKB_GSO_GRE_CSUM = 1 << 7,
825 SKB_GSO_IPXIP4 = 1 << 8,
827 SKB_GSO_IPXIP6 = 1 << 9,
829 SKB_GSO_UDP_TUNNEL = 1 << 10,
831 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
833 SKB_GSO_PARTIAL = 1 << 12,
835 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
837 SKB_GSO_SCTP = 1 << 14,
839 SKB_GSO_ESP = 1 << 15,
841 SKB_GSO_UDP = 1 << 16,
843 SKB_GSO_UDP_L4 = 1 << 17,
845 SKB_GSO_FRAGLIST = 1 << 18,
848 #if BITS_PER_LONG > 32
849 #define NET_SKBUFF_DATA_USES_OFFSET 1
852 #ifdef NET_SKBUFF_DATA_USES_OFFSET
853 typedef unsigned int sk_buff_data_t;
855 typedef unsigned char *sk_buff_data_t;
859 * DOC: Basic sk_buff geometry
861 * struct sk_buff itself is a metadata structure and does not hold any packet
862 * data. All the data is held in associated buffers.
864 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
867 * - data buffer, containing headers and sometimes payload;
868 * this is the part of the skb operated on by the common helpers
869 * such as skb_put() or skb_pull();
870 * - shared info (struct skb_shared_info) which holds an array of pointers
871 * to read-only data in the (page, offset, length) format.
873 * Optionally &skb_shared_info.frag_list may point to another skb.
875 * Basic diagram may look like this::
880 * ,--------------------------- + head
881 * / ,----------------- + data
882 * / / ,----------- + tail
886 * -----------------------------------------------
887 * | headroom | data | tailroom | skb_shared_info |
888 * -----------------------------------------------
892 * + [page frag] ---------
893 * + frag_list --> | sk_buff |
899 * struct sk_buff - socket buffer
900 * @next: Next buffer in list
901 * @prev: Previous buffer in list
902 * @tstamp: Time we arrived/left
903 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
904 * for retransmit timer
905 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
907 * @ll_node: anchor in an llist (eg socket defer_list)
908 * @sk: Socket we are owned by
909 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
910 * fragmentation management
911 * @dev: Device we arrived on/are leaving by
912 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
913 * @cb: Control buffer. Free for use by every layer. Put private vars here
914 * @_skb_refdst: destination entry (with norefcount bit)
915 * @sp: the security path, used for xfrm
916 * @len: Length of actual data
917 * @data_len: Data length
918 * @mac_len: Length of link layer header
919 * @hdr_len: writable header length of cloned skb
920 * @csum: Checksum (must include start/offset pair)
921 * @csum_start: Offset from skb->head where checksumming should start
922 * @csum_offset: Offset from csum_start where checksum should be stored
923 * @priority: Packet queueing priority
924 * @ignore_df: allow local fragmentation
925 * @cloned: Head may be cloned (check refcnt to be sure)
926 * @ip_summed: Driver fed us an IP checksum
927 * @nohdr: Payload reference only, must not modify header
928 * @pkt_type: Packet class
929 * @fclone: skbuff clone status
930 * @ipvs_property: skbuff is owned by ipvs
931 * @inner_protocol_type: whether the inner protocol is
932 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
933 * @remcsum_offload: remote checksum offload is enabled
934 * @offload_fwd_mark: Packet was L2-forwarded in hardware
935 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
936 * @tc_skip_classify: do not classify packet. set by IFB device
937 * @tc_at_ingress: used within tc_classify to distinguish in/egress
938 * @redirected: packet was redirected by packet classifier
939 * @from_ingress: packet was redirected from the ingress path
940 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
941 * @peeked: this packet has been seen already, so stats have been
942 * done for it, don't do them again
943 * @nf_trace: netfilter packet trace flag
944 * @protocol: Packet protocol from driver
945 * @destructor: Destruct function
946 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
947 * @_sk_redir: socket redirection information for skmsg
948 * @_nfct: Associated connection, if any (with nfctinfo bits)
949 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
950 * @skb_iif: ifindex of device we arrived on
951 * @tc_index: Traffic control index
952 * @hash: the packet hash
953 * @queue_mapping: Queue mapping for multiqueue devices
954 * @head_frag: skb was allocated from page fragments,
955 * not allocated by kmalloc() or vmalloc().
956 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
957 * @pp_recycle: mark the packet for recycling instead of freeing (implies
958 * page_pool support on driver)
959 * @active_extensions: active extensions (skb_ext_id types)
960 * @ndisc_nodetype: router type (from link layer)
961 * @ooo_okay: allow the mapping of a socket to a queue to be changed
962 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
964 * @sw_hash: indicates hash was computed in software stack
965 * @wifi_acked_valid: wifi_acked was set
966 * @wifi_acked: whether frame was acked on wifi or not
967 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
968 * @encapsulation: indicates the inner headers in the skbuff are valid
969 * @encap_hdr_csum: software checksum is needed
970 * @csum_valid: checksum is already valid
971 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
972 * @csum_complete_sw: checksum was completed by software
973 * @csum_level: indicates the number of consecutive checksums found in
974 * the packet minus one that have been verified as
975 * CHECKSUM_UNNECESSARY (max 3)
976 * @dst_pending_confirm: need to confirm neighbour
977 * @decrypted: Decrypted SKB
978 * @slow_gro: state present at GRO time, slower prepare step required
979 * @mono_delivery_time: When set, skb->tstamp has the
980 * delivery_time in mono clock base (i.e. EDT). Otherwise, the
981 * skb->tstamp has the (rcv) timestamp at ingress and
982 * delivery_time at egress.
983 * @napi_id: id of the NAPI struct this skb came from
984 * @sender_cpu: (aka @napi_id) source CPU in XPS
985 * @alloc_cpu: CPU which did the skb allocation.
986 * @secmark: security marking
987 * @mark: Generic packet mark
988 * @reserved_tailroom: (aka @mark) number of bytes of free space available
989 * at the tail of an sk_buff
990 * @vlan_present: VLAN tag is present
991 * @vlan_proto: vlan encapsulation protocol
992 * @vlan_tci: vlan tag control information
993 * @inner_protocol: Protocol (encapsulation)
994 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
995 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
996 * @inner_transport_header: Inner transport layer header (encapsulation)
997 * @inner_network_header: Network layer header (encapsulation)
998 * @inner_mac_header: Link layer header (encapsulation)
999 * @transport_header: Transport layer header
1000 * @network_header: Network layer header
1001 * @mac_header: Link layer header
1002 * @kcov_handle: KCOV remote handle for remote coverage collection
1003 * @tail: Tail pointer
1005 * @head: Head of buffer
1006 * @data: Data head pointer
1007 * @truesize: Buffer size
1008 * @users: User count - see {datagram,tcp}.c
1009 * @extensions: allocated extensions, valid if active_extensions is nonzero
1015 /* These two members must be first to match sk_buff_head. */
1016 struct sk_buff *next;
1017 struct sk_buff *prev;
1020 struct net_device *dev;
1021 /* Some protocols might use this space to store information,
1022 * while device pointer would be NULL.
1023 * UDP receive path is one user.
1025 unsigned long dev_scratch;
1028 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
1029 struct list_head list;
1030 struct llist_node ll_node;
1035 int ip_defrag_offset;
1040 u64 skb_mstamp_ns; /* earliest departure time */
1043 * This is the control buffer. It is free to use for every
1044 * layer. Please put your private variables there. If you
1045 * want to keep them across layers you have to do a skb_clone()
1046 * first. This is owned by whoever has the skb queued ATM.
1048 char cb[48] __aligned(8);
1052 unsigned long _skb_refdst;
1053 void (*destructor)(struct sk_buff *skb);
1055 struct list_head tcp_tsorted_anchor;
1056 #ifdef CONFIG_NET_SOCK_MSG
1057 unsigned long _sk_redir;
1061 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1062 unsigned long _nfct;
1069 /* Following fields are _not_ copied in __copy_skb_header()
1070 * Note that queue_mapping is here mostly to fill a hole.
1072 __u16 queue_mapping;
1074 /* if you move cloned around you also must adapt those constants */
1075 #ifdef __BIG_ENDIAN_BITFIELD
1076 #define CLONED_MASK (1 << 7)
1078 #define CLONED_MASK 1
1080 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
1083 __u8 __cloned_offset[0];
1091 pp_recycle:1; /* page_pool recycle indicator */
1092 #ifdef CONFIG_SKB_EXTENSIONS
1093 __u8 active_extensions;
1096 /* Fields enclosed in headers group are copied
1097 * using a single memcpy() in __copy_skb_header()
1099 struct_group(headers,
1102 __u8 __pkt_type_offset[0];
1104 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
1112 __u8 wifi_acked_valid:1;
1115 /* Indicates the inner headers are valid in the skbuff. */
1116 __u8 encapsulation:1;
1117 __u8 encap_hdr_csum:1;
1121 __u8 __pkt_vlan_present_offset[0];
1123 __u8 vlan_present:1; /* See PKT_VLAN_PRESENT_BIT */
1124 __u8 csum_complete_sw:1;
1126 __u8 dst_pending_confirm:1;
1127 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */
1128 #ifdef CONFIG_NET_CLS_ACT
1129 __u8 tc_skip_classify:1;
1130 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
1132 #ifdef CONFIG_IPV6_NDISC_NODETYPE
1133 __u8 ndisc_nodetype:2;
1136 __u8 ipvs_property:1;
1137 __u8 inner_protocol_type:1;
1138 __u8 remcsum_offload:1;
1139 #ifdef CONFIG_NET_SWITCHDEV
1140 __u8 offload_fwd_mark:1;
1141 __u8 offload_l3_fwd_mark:1;
1144 #ifdef CONFIG_NET_REDIRECT
1145 __u8 from_ingress:1;
1147 #ifdef CONFIG_NETFILTER_SKIP_EGRESS
1148 __u8 nf_skip_egress:1;
1150 #ifdef CONFIG_TLS_DEVICE
1154 __u8 csum_not_inet:1;
1156 #ifdef CONFIG_NET_SCHED
1157 __u16 tc_index; /* traffic control index */
1172 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
1174 unsigned int napi_id;
1175 unsigned int sender_cpu;
1179 #ifdef CONFIG_NETWORK_SECMARK
1185 __u32 reserved_tailroom;
1189 __be16 inner_protocol;
1193 __u16 inner_transport_header;
1194 __u16 inner_network_header;
1195 __u16 inner_mac_header;
1198 __u16 transport_header;
1199 __u16 network_header;
1206 ); /* end headers group */
1208 /* These elements must be at the end, see alloc_skb() for details. */
1209 sk_buff_data_t tail;
1211 unsigned char *head,
1213 unsigned int truesize;
1216 #ifdef CONFIG_SKB_EXTENSIONS
1217 /* only useable after checking ->active_extensions != 0 */
1218 struct skb_ext *extensions;
1222 /* if you move pkt_type around you also must adapt those constants */
1223 #ifdef __BIG_ENDIAN_BITFIELD
1224 #define PKT_TYPE_MAX (7 << 5)
1226 #define PKT_TYPE_MAX 7
1228 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1230 /* if you move pkt_vlan_present, tc_at_ingress, or mono_delivery_time
1231 * around, you also must adapt these constants.
1233 #ifdef __BIG_ENDIAN_BITFIELD
1234 #define PKT_VLAN_PRESENT_BIT 7
1235 #define TC_AT_INGRESS_MASK (1 << 0)
1236 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 2)
1238 #define PKT_VLAN_PRESENT_BIT 0
1239 #define TC_AT_INGRESS_MASK (1 << 7)
1240 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 5)
1242 #define PKT_VLAN_PRESENT_OFFSET offsetof(struct sk_buff, __pkt_vlan_present_offset)
1246 * Handling routines are only of interest to the kernel
1249 #define SKB_ALLOC_FCLONE 0x01
1250 #define SKB_ALLOC_RX 0x02
1251 #define SKB_ALLOC_NAPI 0x04
1254 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1257 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1259 return unlikely(skb->pfmemalloc);
1263 * skb might have a dst pointer attached, refcounted or not.
1264 * _skb_refdst low order bit is set if refcount was _not_ taken
1266 #define SKB_DST_NOREF 1UL
1267 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1270 * skb_dst - returns skb dst_entry
1273 * Returns skb dst_entry, regardless of reference taken or not.
1275 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1277 /* If refdst was not refcounted, check we still are in a
1278 * rcu_read_lock section
1280 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1281 !rcu_read_lock_held() &&
1282 !rcu_read_lock_bh_held());
1283 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1287 * skb_dst_set - sets skb dst
1291 * Sets skb dst, assuming a reference was taken on dst and should
1292 * be released by skb_dst_drop()
1294 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1296 skb->slow_gro |= !!dst;
1297 skb->_skb_refdst = (unsigned long)dst;
1301 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1305 * Sets skb dst, assuming a reference was not taken on dst.
1306 * If dst entry is cached, we do not take reference and dst_release
1307 * will be avoided by refdst_drop. If dst entry is not cached, we take
1308 * reference, so that last dst_release can destroy the dst immediately.
1310 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1312 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1313 skb->slow_gro |= !!dst;
1314 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1318 * skb_dst_is_noref - Test if skb dst isn't refcounted
1321 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1323 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1327 * skb_rtable - Returns the skb &rtable
1330 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1332 return (struct rtable *)skb_dst(skb);
1335 /* For mangling skb->pkt_type from user space side from applications
1336 * such as nft, tc, etc, we only allow a conservative subset of
1337 * possible pkt_types to be set.
1339 static inline bool skb_pkt_type_ok(u32 ptype)
1341 return ptype <= PACKET_OTHERHOST;
1345 * skb_napi_id - Returns the skb's NAPI id
1348 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1350 #ifdef CONFIG_NET_RX_BUSY_POLL
1351 return skb->napi_id;
1358 * skb_unref - decrement the skb's reference count
1361 * Returns true if we can free the skb.
1363 static inline bool skb_unref(struct sk_buff *skb)
1367 if (likely(refcount_read(&skb->users) == 1))
1369 else if (likely(!refcount_dec_and_test(&skb->users)))
1375 void kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason);
1378 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1379 * @skb: buffer to free
1381 static inline void kfree_skb(struct sk_buff *skb)
1383 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1386 void skb_release_head_state(struct sk_buff *skb);
1387 void kfree_skb_list_reason(struct sk_buff *segs,
1388 enum skb_drop_reason reason);
1389 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1390 void skb_tx_error(struct sk_buff *skb);
1392 static inline void kfree_skb_list(struct sk_buff *segs)
1394 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1397 #ifdef CONFIG_TRACEPOINTS
1398 void consume_skb(struct sk_buff *skb);
1400 static inline void consume_skb(struct sk_buff *skb)
1402 return kfree_skb(skb);
1406 void __consume_stateless_skb(struct sk_buff *skb);
1407 void __kfree_skb(struct sk_buff *skb);
1408 extern struct kmem_cache *skbuff_head_cache;
1410 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1411 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1412 bool *fragstolen, int *delta_truesize);
1414 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1416 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1417 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1418 struct sk_buff *build_skb_around(struct sk_buff *skb,
1419 void *data, unsigned int frag_size);
1420 void skb_attempt_defer_free(struct sk_buff *skb);
1422 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1425 * alloc_skb - allocate a network buffer
1426 * @size: size to allocate
1427 * @priority: allocation mask
1429 * This function is a convenient wrapper around __alloc_skb().
1431 static inline struct sk_buff *alloc_skb(unsigned int size,
1434 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1437 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1438 unsigned long data_len,
1442 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1444 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1445 struct sk_buff_fclones {
1446 struct sk_buff skb1;
1448 struct sk_buff skb2;
1450 refcount_t fclone_ref;
1454 * skb_fclone_busy - check if fclone is busy
1458 * Returns true if skb is a fast clone, and its clone is not freed.
1459 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1460 * so we also check that this didnt happen.
1462 static inline bool skb_fclone_busy(const struct sock *sk,
1463 const struct sk_buff *skb)
1465 const struct sk_buff_fclones *fclones;
1467 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1469 return skb->fclone == SKB_FCLONE_ORIG &&
1470 refcount_read(&fclones->fclone_ref) > 1 &&
1471 READ_ONCE(fclones->skb2.sk) == sk;
1475 * alloc_skb_fclone - allocate a network buffer from fclone cache
1476 * @size: size to allocate
1477 * @priority: allocation mask
1479 * This function is a convenient wrapper around __alloc_skb().
1481 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1484 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1487 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1488 void skb_headers_offset_update(struct sk_buff *skb, int off);
1489 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1490 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1491 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1492 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1493 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1494 gfp_t gfp_mask, bool fclone);
1495 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1498 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1501 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1502 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1503 unsigned int headroom);
1504 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1505 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1506 int newtailroom, gfp_t priority);
1507 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1508 int offset, int len);
1509 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1510 int offset, int len);
1511 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1512 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1515 * skb_pad - zero pad the tail of an skb
1516 * @skb: buffer to pad
1517 * @pad: space to pad
1519 * Ensure that a buffer is followed by a padding area that is zero
1520 * filled. Used by network drivers which may DMA or transfer data
1521 * beyond the buffer end onto the wire.
1523 * May return error in out of memory cases. The skb is freed on error.
1525 static inline int skb_pad(struct sk_buff *skb, int pad)
1527 return __skb_pad(skb, pad, true);
1529 #define dev_kfree_skb(a) consume_skb(a)
1531 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1532 int offset, size_t size);
1534 struct skb_seq_state {
1538 __u32 stepped_offset;
1539 struct sk_buff *root_skb;
1540 struct sk_buff *cur_skb;
1545 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1546 unsigned int to, struct skb_seq_state *st);
1547 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1548 struct skb_seq_state *st);
1549 void skb_abort_seq_read(struct skb_seq_state *st);
1551 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1552 unsigned int to, struct ts_config *config);
1555 * Packet hash types specify the type of hash in skb_set_hash.
1557 * Hash types refer to the protocol layer addresses which are used to
1558 * construct a packet's hash. The hashes are used to differentiate or identify
1559 * flows of the protocol layer for the hash type. Hash types are either
1560 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1562 * Properties of hashes:
1564 * 1) Two packets in different flows have different hash values
1565 * 2) Two packets in the same flow should have the same hash value
1567 * A hash at a higher layer is considered to be more specific. A driver should
1568 * set the most specific hash possible.
1570 * A driver cannot indicate a more specific hash than the layer at which a hash
1571 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1573 * A driver may indicate a hash level which is less specific than the
1574 * actual layer the hash was computed on. For instance, a hash computed
1575 * at L4 may be considered an L3 hash. This should only be done if the
1576 * driver can't unambiguously determine that the HW computed the hash at
1577 * the higher layer. Note that the "should" in the second property above
1580 enum pkt_hash_types {
1581 PKT_HASH_TYPE_NONE, /* Undefined type */
1582 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1583 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1584 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1587 static inline void skb_clear_hash(struct sk_buff *skb)
1594 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1597 skb_clear_hash(skb);
1601 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1603 skb->l4_hash = is_l4;
1604 skb->sw_hash = is_sw;
1609 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1611 /* Used by drivers to set hash from HW */
1612 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1616 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1618 __skb_set_hash(skb, hash, true, is_l4);
1621 void __skb_get_hash(struct sk_buff *skb);
1622 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1623 u32 skb_get_poff(const struct sk_buff *skb);
1624 u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1625 const struct flow_keys_basic *keys, int hlen);
1626 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1627 const void *data, int hlen_proto);
1629 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1630 int thoff, u8 ip_proto)
1632 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1635 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1636 const struct flow_dissector_key *key,
1637 unsigned int key_count);
1639 struct bpf_flow_dissector;
1640 bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1641 __be16 proto, int nhoff, int hlen, unsigned int flags);
1643 bool __skb_flow_dissect(const struct net *net,
1644 const struct sk_buff *skb,
1645 struct flow_dissector *flow_dissector,
1646 void *target_container, const void *data,
1647 __be16 proto, int nhoff, int hlen, unsigned int flags);
1649 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1650 struct flow_dissector *flow_dissector,
1651 void *target_container, unsigned int flags)
1653 return __skb_flow_dissect(NULL, skb, flow_dissector,
1654 target_container, NULL, 0, 0, 0, flags);
1657 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1658 struct flow_keys *flow,
1661 memset(flow, 0, sizeof(*flow));
1662 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1663 flow, NULL, 0, 0, 0, flags);
1667 skb_flow_dissect_flow_keys_basic(const struct net *net,
1668 const struct sk_buff *skb,
1669 struct flow_keys_basic *flow,
1670 const void *data, __be16 proto,
1671 int nhoff, int hlen, unsigned int flags)
1673 memset(flow, 0, sizeof(*flow));
1674 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1675 data, proto, nhoff, hlen, flags);
1678 void skb_flow_dissect_meta(const struct sk_buff *skb,
1679 struct flow_dissector *flow_dissector,
1680 void *target_container);
1682 /* Gets a skb connection tracking info, ctinfo map should be a
1683 * map of mapsize to translate enum ip_conntrack_info states
1687 skb_flow_dissect_ct(const struct sk_buff *skb,
1688 struct flow_dissector *flow_dissector,
1689 void *target_container,
1690 u16 *ctinfo_map, size_t mapsize,
1691 bool post_ct, u16 zone);
1693 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1694 struct flow_dissector *flow_dissector,
1695 void *target_container);
1697 void skb_flow_dissect_hash(const struct sk_buff *skb,
1698 struct flow_dissector *flow_dissector,
1699 void *target_container);
1701 static inline __u32 skb_get_hash(struct sk_buff *skb)
1703 if (!skb->l4_hash && !skb->sw_hash)
1704 __skb_get_hash(skb);
1709 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1711 if (!skb->l4_hash && !skb->sw_hash) {
1712 struct flow_keys keys;
1713 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1715 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1721 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1722 const siphash_key_t *perturb);
1724 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1729 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1731 to->hash = from->hash;
1732 to->sw_hash = from->sw_hash;
1733 to->l4_hash = from->l4_hash;
1736 static inline void skb_copy_decrypted(struct sk_buff *to,
1737 const struct sk_buff *from)
1739 #ifdef CONFIG_TLS_DEVICE
1740 to->decrypted = from->decrypted;
1744 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1745 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1747 return skb->head + skb->end;
1750 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1755 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1760 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1765 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1767 return skb->end - skb->head;
1770 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1772 skb->end = skb->head + offset;
1776 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1777 struct ubuf_info *uarg);
1779 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1781 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1784 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
1785 struct sk_buff *skb, struct iov_iter *from,
1788 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1789 struct msghdr *msg, int len)
1791 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len);
1794 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1795 struct msghdr *msg, int len,
1796 struct ubuf_info *uarg);
1799 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1801 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1803 return &skb_shinfo(skb)->hwtstamps;
1806 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1808 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1810 return is_zcopy ? skb_uarg(skb) : NULL;
1813 static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1815 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1818 static inline bool skb_zcopy_managed(const struct sk_buff *skb)
1820 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS;
1823 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1824 const struct sk_buff *skb2)
1826 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1829 static inline void net_zcopy_get(struct ubuf_info *uarg)
1831 refcount_inc(&uarg->refcnt);
1834 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1836 skb_shinfo(skb)->destructor_arg = uarg;
1837 skb_shinfo(skb)->flags |= uarg->flags;
1840 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1843 if (skb && uarg && !skb_zcopy(skb)) {
1844 if (unlikely(have_ref && *have_ref))
1847 net_zcopy_get(uarg);
1848 skb_zcopy_init(skb, uarg);
1852 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1854 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1855 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1858 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1860 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1863 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1865 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1868 static inline void net_zcopy_put(struct ubuf_info *uarg)
1871 uarg->callback(NULL, uarg, true);
1874 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1877 if (uarg->callback == msg_zerocopy_callback)
1878 msg_zerocopy_put_abort(uarg, have_uref);
1880 net_zcopy_put(uarg);
1884 /* Release a reference on a zerocopy structure */
1885 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1887 struct ubuf_info *uarg = skb_zcopy(skb);
1890 if (!skb_zcopy_is_nouarg(skb))
1891 uarg->callback(skb, uarg, zerocopy_success);
1893 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1897 void __skb_zcopy_downgrade_managed(struct sk_buff *skb);
1899 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb)
1901 if (unlikely(skb_zcopy_managed(skb)))
1902 __skb_zcopy_downgrade_managed(skb);
1905 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1910 /* Iterate through singly-linked GSO fragments of an skb. */
1911 #define skb_list_walk_safe(first, skb, next_skb) \
1912 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1913 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1915 static inline void skb_list_del_init(struct sk_buff *skb)
1917 __list_del_entry(&skb->list);
1918 skb_mark_not_on_list(skb);
1922 * skb_queue_empty - check if a queue is empty
1925 * Returns true if the queue is empty, false otherwise.
1927 static inline int skb_queue_empty(const struct sk_buff_head *list)
1929 return list->next == (const struct sk_buff *) list;
1933 * skb_queue_empty_lockless - check if a queue is empty
1936 * Returns true if the queue is empty, false otherwise.
1937 * This variant can be used in lockless contexts.
1939 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1941 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1946 * skb_queue_is_last - check if skb is the last entry in the queue
1950 * Returns true if @skb is the last buffer on the list.
1952 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1953 const struct sk_buff *skb)
1955 return skb->next == (const struct sk_buff *) list;
1959 * skb_queue_is_first - check if skb is the first entry in the queue
1963 * Returns true if @skb is the first buffer on the list.
1965 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1966 const struct sk_buff *skb)
1968 return skb->prev == (const struct sk_buff *) list;
1972 * skb_queue_next - return the next packet in the queue
1974 * @skb: current buffer
1976 * Return the next packet in @list after @skb. It is only valid to
1977 * call this if skb_queue_is_last() evaluates to false.
1979 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1980 const struct sk_buff *skb)
1982 /* This BUG_ON may seem severe, but if we just return then we
1983 * are going to dereference garbage.
1985 BUG_ON(skb_queue_is_last(list, skb));
1990 * skb_queue_prev - return the prev packet in the queue
1992 * @skb: current buffer
1994 * Return the prev packet in @list before @skb. It is only valid to
1995 * call this if skb_queue_is_first() evaluates to false.
1997 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1998 const struct sk_buff *skb)
2000 /* This BUG_ON may seem severe, but if we just return then we
2001 * are going to dereference garbage.
2003 BUG_ON(skb_queue_is_first(list, skb));
2008 * skb_get - reference buffer
2009 * @skb: buffer to reference
2011 * Makes another reference to a socket buffer and returns a pointer
2014 static inline struct sk_buff *skb_get(struct sk_buff *skb)
2016 refcount_inc(&skb->users);
2021 * If users == 1, we are the only owner and can avoid redundant atomic changes.
2025 * skb_cloned - is the buffer a clone
2026 * @skb: buffer to check
2028 * Returns true if the buffer was generated with skb_clone() and is
2029 * one of multiple shared copies of the buffer. Cloned buffers are
2030 * shared data so must not be written to under normal circumstances.
2032 static inline int skb_cloned(const struct sk_buff *skb)
2034 return skb->cloned &&
2035 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
2038 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
2040 might_sleep_if(gfpflags_allow_blocking(pri));
2042 if (skb_cloned(skb))
2043 return pskb_expand_head(skb, 0, 0, pri);
2048 /* This variant of skb_unclone() makes sure skb->truesize
2049 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
2051 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
2052 * when various debugging features are in place.
2054 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
2055 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
2057 might_sleep_if(gfpflags_allow_blocking(pri));
2059 if (skb_cloned(skb))
2060 return __skb_unclone_keeptruesize(skb, pri);
2065 * skb_header_cloned - is the header a clone
2066 * @skb: buffer to check
2068 * Returns true if modifying the header part of the buffer requires
2069 * the data to be copied.
2071 static inline int skb_header_cloned(const struct sk_buff *skb)
2078 dataref = atomic_read(&skb_shinfo(skb)->dataref);
2079 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
2080 return dataref != 1;
2083 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
2085 might_sleep_if(gfpflags_allow_blocking(pri));
2087 if (skb_header_cloned(skb))
2088 return pskb_expand_head(skb, 0, 0, pri);
2094 * __skb_header_release() - allow clones to use the headroom
2095 * @skb: buffer to operate on
2097 * See "DOC: dataref and headerless skbs".
2099 static inline void __skb_header_release(struct sk_buff *skb)
2102 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
2107 * skb_shared - is the buffer shared
2108 * @skb: buffer to check
2110 * Returns true if more than one person has a reference to this
2113 static inline int skb_shared(const struct sk_buff *skb)
2115 return refcount_read(&skb->users) != 1;
2119 * skb_share_check - check if buffer is shared and if so clone it
2120 * @skb: buffer to check
2121 * @pri: priority for memory allocation
2123 * If the buffer is shared the buffer is cloned and the old copy
2124 * drops a reference. A new clone with a single reference is returned.
2125 * If the buffer is not shared the original buffer is returned. When
2126 * being called from interrupt status or with spinlocks held pri must
2129 * NULL is returned on a memory allocation failure.
2131 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
2133 might_sleep_if(gfpflags_allow_blocking(pri));
2134 if (skb_shared(skb)) {
2135 struct sk_buff *nskb = skb_clone(skb, pri);
2147 * Copy shared buffers into a new sk_buff. We effectively do COW on
2148 * packets to handle cases where we have a local reader and forward
2149 * and a couple of other messy ones. The normal one is tcpdumping
2150 * a packet thats being forwarded.
2154 * skb_unshare - make a copy of a shared buffer
2155 * @skb: buffer to check
2156 * @pri: priority for memory allocation
2158 * If the socket buffer is a clone then this function creates a new
2159 * copy of the data, drops a reference count on the old copy and returns
2160 * the new copy with the reference count at 1. If the buffer is not a clone
2161 * the original buffer is returned. When called with a spinlock held or
2162 * from interrupt state @pri must be %GFP_ATOMIC
2164 * %NULL is returned on a memory allocation failure.
2166 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
2169 might_sleep_if(gfpflags_allow_blocking(pri));
2170 if (skb_cloned(skb)) {
2171 struct sk_buff *nskb = skb_copy(skb, pri);
2173 /* Free our shared copy */
2184 * skb_peek - peek at the head of an &sk_buff_head
2185 * @list_: list to peek at
2187 * Peek an &sk_buff. Unlike most other operations you _MUST_
2188 * be careful with this one. A peek leaves the buffer on the
2189 * list and someone else may run off with it. You must hold
2190 * the appropriate locks or have a private queue to do this.
2192 * Returns %NULL for an empty list or a pointer to the head element.
2193 * The reference count is not incremented and the reference is therefore
2194 * volatile. Use with caution.
2196 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2198 struct sk_buff *skb = list_->next;
2200 if (skb == (struct sk_buff *)list_)
2206 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2207 * @list_: list to peek at
2209 * Like skb_peek(), but the caller knows that the list is not empty.
2211 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2217 * skb_peek_next - peek skb following the given one from a queue
2218 * @skb: skb to start from
2219 * @list_: list to peek at
2221 * Returns %NULL when the end of the list is met or a pointer to the
2222 * next element. The reference count is not incremented and the
2223 * reference is therefore volatile. Use with caution.
2225 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2226 const struct sk_buff_head *list_)
2228 struct sk_buff *next = skb->next;
2230 if (next == (struct sk_buff *)list_)
2236 * skb_peek_tail - peek at the tail of an &sk_buff_head
2237 * @list_: list to peek at
2239 * Peek an &sk_buff. Unlike most other operations you _MUST_
2240 * be careful with this one. A peek leaves the buffer on the
2241 * list and someone else may run off with it. You must hold
2242 * the appropriate locks or have a private queue to do this.
2244 * Returns %NULL for an empty list or a pointer to the tail element.
2245 * The reference count is not incremented and the reference is therefore
2246 * volatile. Use with caution.
2248 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2250 struct sk_buff *skb = READ_ONCE(list_->prev);
2252 if (skb == (struct sk_buff *)list_)
2259 * skb_queue_len - get queue length
2260 * @list_: list to measure
2262 * Return the length of an &sk_buff queue.
2264 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2270 * skb_queue_len_lockless - get queue length
2271 * @list_: list to measure
2273 * Return the length of an &sk_buff queue.
2274 * This variant can be used in lockless contexts.
2276 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2278 return READ_ONCE(list_->qlen);
2282 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2283 * @list: queue to initialize
2285 * This initializes only the list and queue length aspects of
2286 * an sk_buff_head object. This allows to initialize the list
2287 * aspects of an sk_buff_head without reinitializing things like
2288 * the spinlock. It can also be used for on-stack sk_buff_head
2289 * objects where the spinlock is known to not be used.
2291 static inline void __skb_queue_head_init(struct sk_buff_head *list)
2293 list->prev = list->next = (struct sk_buff *)list;
2298 * This function creates a split out lock class for each invocation;
2299 * this is needed for now since a whole lot of users of the skb-queue
2300 * infrastructure in drivers have different locking usage (in hardirq)
2301 * than the networking core (in softirq only). In the long run either the
2302 * network layer or drivers should need annotation to consolidate the
2303 * main types of usage into 3 classes.
2305 static inline void skb_queue_head_init(struct sk_buff_head *list)
2307 spin_lock_init(&list->lock);
2308 __skb_queue_head_init(list);
2311 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2312 struct lock_class_key *class)
2314 skb_queue_head_init(list);
2315 lockdep_set_class(&list->lock, class);
2319 * Insert an sk_buff on a list.
2321 * The "__skb_xxxx()" functions are the non-atomic ones that
2322 * can only be called with interrupts disabled.
2324 static inline void __skb_insert(struct sk_buff *newsk,
2325 struct sk_buff *prev, struct sk_buff *next,
2326 struct sk_buff_head *list)
2328 /* See skb_queue_empty_lockless() and skb_peek_tail()
2329 * for the opposite READ_ONCE()
2331 WRITE_ONCE(newsk->next, next);
2332 WRITE_ONCE(newsk->prev, prev);
2333 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2334 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2335 WRITE_ONCE(list->qlen, list->qlen + 1);
2338 static inline void __skb_queue_splice(const struct sk_buff_head *list,
2339 struct sk_buff *prev,
2340 struct sk_buff *next)
2342 struct sk_buff *first = list->next;
2343 struct sk_buff *last = list->prev;
2345 WRITE_ONCE(first->prev, prev);
2346 WRITE_ONCE(prev->next, first);
2348 WRITE_ONCE(last->next, next);
2349 WRITE_ONCE(next->prev, last);
2353 * skb_queue_splice - join two skb lists, this is designed for stacks
2354 * @list: the new list to add
2355 * @head: the place to add it in the first list
2357 static inline void skb_queue_splice(const struct sk_buff_head *list,
2358 struct sk_buff_head *head)
2360 if (!skb_queue_empty(list)) {
2361 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2362 head->qlen += list->qlen;
2367 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2368 * @list: the new list to add
2369 * @head: the place to add it in the first list
2371 * The list at @list is reinitialised
2373 static inline void skb_queue_splice_init(struct sk_buff_head *list,
2374 struct sk_buff_head *head)
2376 if (!skb_queue_empty(list)) {
2377 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2378 head->qlen += list->qlen;
2379 __skb_queue_head_init(list);
2384 * skb_queue_splice_tail - join two skb lists, each list being a queue
2385 * @list: the new list to add
2386 * @head: the place to add it in the first list
2388 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2389 struct sk_buff_head *head)
2391 if (!skb_queue_empty(list)) {
2392 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2393 head->qlen += list->qlen;
2398 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2399 * @list: the new list to add
2400 * @head: the place to add it in the first list
2402 * Each of the lists is a queue.
2403 * The list at @list is reinitialised
2405 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2406 struct sk_buff_head *head)
2408 if (!skb_queue_empty(list)) {
2409 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2410 head->qlen += list->qlen;
2411 __skb_queue_head_init(list);
2416 * __skb_queue_after - queue a buffer at the list head
2417 * @list: list to use
2418 * @prev: place after this buffer
2419 * @newsk: buffer to queue
2421 * Queue a buffer int the middle of a list. This function takes no locks
2422 * and you must therefore hold required locks before calling it.
2424 * A buffer cannot be placed on two lists at the same time.
2426 static inline void __skb_queue_after(struct sk_buff_head *list,
2427 struct sk_buff *prev,
2428 struct sk_buff *newsk)
2430 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2433 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2434 struct sk_buff_head *list);
2436 static inline void __skb_queue_before(struct sk_buff_head *list,
2437 struct sk_buff *next,
2438 struct sk_buff *newsk)
2440 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2444 * __skb_queue_head - queue a buffer at the list head
2445 * @list: list to use
2446 * @newsk: buffer to queue
2448 * Queue a buffer at the start of a list. This function takes no locks
2449 * and you must therefore hold required locks before calling it.
2451 * A buffer cannot be placed on two lists at the same time.
2453 static inline void __skb_queue_head(struct sk_buff_head *list,
2454 struct sk_buff *newsk)
2456 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2458 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2461 * __skb_queue_tail - queue a buffer at the list tail
2462 * @list: list to use
2463 * @newsk: buffer to queue
2465 * Queue a buffer at the end of a list. This function takes no locks
2466 * and you must therefore hold required locks before calling it.
2468 * A buffer cannot be placed on two lists at the same time.
2470 static inline void __skb_queue_tail(struct sk_buff_head *list,
2471 struct sk_buff *newsk)
2473 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2475 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2478 * remove sk_buff from list. _Must_ be called atomically, and with
2481 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2482 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2484 struct sk_buff *next, *prev;
2486 WRITE_ONCE(list->qlen, list->qlen - 1);
2489 skb->next = skb->prev = NULL;
2490 WRITE_ONCE(next->prev, prev);
2491 WRITE_ONCE(prev->next, next);
2495 * __skb_dequeue - remove from the head of the queue
2496 * @list: list to dequeue from
2498 * Remove the head of the list. This function does not take any locks
2499 * so must be used with appropriate locks held only. The head item is
2500 * returned or %NULL if the list is empty.
2502 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2504 struct sk_buff *skb = skb_peek(list);
2506 __skb_unlink(skb, list);
2509 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2512 * __skb_dequeue_tail - remove from the tail of the queue
2513 * @list: list to dequeue from
2515 * Remove the tail of the list. This function does not take any locks
2516 * so must be used with appropriate locks held only. The tail item is
2517 * returned or %NULL if the list is empty.
2519 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2521 struct sk_buff *skb = skb_peek_tail(list);
2523 __skb_unlink(skb, list);
2526 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2529 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2531 return skb->data_len;
2534 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2536 return skb->len - skb->data_len;
2539 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2541 unsigned int i, len = 0;
2543 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2544 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2548 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2550 return skb_headlen(skb) + __skb_pagelen(skb);
2554 * __skb_fill_page_desc - initialise a paged fragment in an skb
2555 * @skb: buffer containing fragment to be initialised
2556 * @i: paged fragment index to initialise
2557 * @page: the page to use for this fragment
2558 * @off: the offset to the data with @page
2559 * @size: the length of the data
2561 * Initialises the @i'th fragment of @skb to point to &size bytes at
2562 * offset @off within @page.
2564 * Does not take any additional reference on the fragment.
2566 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2567 struct page *page, int off, int size)
2569 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2572 * Propagate page pfmemalloc to the skb if we can. The problem is
2573 * that not all callers have unique ownership of the page but rely
2574 * on page_is_pfmemalloc doing the right thing(tm).
2576 frag->bv_page = page;
2577 frag->bv_offset = off;
2578 skb_frag_size_set(frag, size);
2580 page = compound_head(page);
2581 if (page_is_pfmemalloc(page))
2582 skb->pfmemalloc = true;
2586 * skb_fill_page_desc - initialise a paged fragment in an skb
2587 * @skb: buffer containing fragment to be initialised
2588 * @i: paged fragment index to initialise
2589 * @page: the page to use for this fragment
2590 * @off: the offset to the data with @page
2591 * @size: the length of the data
2593 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2594 * @skb to point to @size bytes at offset @off within @page. In
2595 * addition updates @skb such that @i is the last fragment.
2597 * Does not take any additional reference on the fragment.
2599 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2600 struct page *page, int off, int size)
2602 __skb_fill_page_desc(skb, i, page, off, size);
2603 skb_shinfo(skb)->nr_frags = i + 1;
2606 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2607 int size, unsigned int truesize);
2609 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2610 unsigned int truesize);
2612 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2614 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2615 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2617 return skb->head + skb->tail;
2620 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2622 skb->tail = skb->data - skb->head;
2625 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2627 skb_reset_tail_pointer(skb);
2628 skb->tail += offset;
2631 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2632 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2637 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2639 skb->tail = skb->data;
2642 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2644 skb->tail = skb->data + offset;
2647 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2650 * Add data to an sk_buff
2652 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2653 void *skb_put(struct sk_buff *skb, unsigned int len);
2654 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2656 void *tmp = skb_tail_pointer(skb);
2657 SKB_LINEAR_ASSERT(skb);
2663 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2665 void *tmp = __skb_put(skb, len);
2667 memset(tmp, 0, len);
2671 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2674 void *tmp = __skb_put(skb, len);
2676 memcpy(tmp, data, len);
2680 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2682 *(u8 *)__skb_put(skb, 1) = val;
2685 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2687 void *tmp = skb_put(skb, len);
2689 memset(tmp, 0, len);
2694 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2697 void *tmp = skb_put(skb, len);
2699 memcpy(tmp, data, len);
2704 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2706 *(u8 *)skb_put(skb, 1) = val;
2709 void *skb_push(struct sk_buff *skb, unsigned int len);
2710 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2717 void *skb_pull(struct sk_buff *skb, unsigned int len);
2718 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2721 if (unlikely(skb->len < skb->data_len)) {
2722 #if defined(CONFIG_DEBUG_NET)
2724 pr_err("__skb_pull(len=%u)\n", len);
2725 skb_dump(KERN_ERR, skb, false);
2729 return skb->data += len;
2732 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2734 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2737 void *skb_pull_data(struct sk_buff *skb, size_t len);
2739 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2741 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2743 if (len > skb_headlen(skb) &&
2744 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2747 return skb->data += len;
2750 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2752 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2755 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2757 if (likely(len <= skb_headlen(skb)))
2759 if (unlikely(len > skb->len))
2761 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2764 void skb_condense(struct sk_buff *skb);
2767 * skb_headroom - bytes at buffer head
2768 * @skb: buffer to check
2770 * Return the number of bytes of free space at the head of an &sk_buff.
2772 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2774 return skb->data - skb->head;
2778 * skb_tailroom - bytes at buffer end
2779 * @skb: buffer to check
2781 * Return the number of bytes of free space at the tail of an sk_buff
2783 static inline int skb_tailroom(const struct sk_buff *skb)
2785 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2789 * skb_availroom - bytes at buffer end
2790 * @skb: buffer to check
2792 * Return the number of bytes of free space at the tail of an sk_buff
2793 * allocated by sk_stream_alloc()
2795 static inline int skb_availroom(const struct sk_buff *skb)
2797 if (skb_is_nonlinear(skb))
2800 return skb->end - skb->tail - skb->reserved_tailroom;
2804 * skb_reserve - adjust headroom
2805 * @skb: buffer to alter
2806 * @len: bytes to move
2808 * Increase the headroom of an empty &sk_buff by reducing the tail
2809 * room. This is only allowed for an empty buffer.
2811 static inline void skb_reserve(struct sk_buff *skb, int len)
2818 * skb_tailroom_reserve - adjust reserved_tailroom
2819 * @skb: buffer to alter
2820 * @mtu: maximum amount of headlen permitted
2821 * @needed_tailroom: minimum amount of reserved_tailroom
2823 * Set reserved_tailroom so that headlen can be as large as possible but
2824 * not larger than mtu and tailroom cannot be smaller than
2826 * The required headroom should already have been reserved before using
2829 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2830 unsigned int needed_tailroom)
2832 SKB_LINEAR_ASSERT(skb);
2833 if (mtu < skb_tailroom(skb) - needed_tailroom)
2834 /* use at most mtu */
2835 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2837 /* use up to all available space */
2838 skb->reserved_tailroom = needed_tailroom;
2841 #define ENCAP_TYPE_ETHER 0
2842 #define ENCAP_TYPE_IPPROTO 1
2844 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2847 skb->inner_protocol = protocol;
2848 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2851 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2854 skb->inner_ipproto = ipproto;
2855 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2858 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2860 skb->inner_mac_header = skb->mac_header;
2861 skb->inner_network_header = skb->network_header;
2862 skb->inner_transport_header = skb->transport_header;
2865 static inline void skb_reset_mac_len(struct sk_buff *skb)
2867 skb->mac_len = skb->network_header - skb->mac_header;
2870 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2873 return skb->head + skb->inner_transport_header;
2876 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2878 return skb_inner_transport_header(skb) - skb->data;
2881 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2883 skb->inner_transport_header = skb->data - skb->head;
2886 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2889 skb_reset_inner_transport_header(skb);
2890 skb->inner_transport_header += offset;
2893 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2895 return skb->head + skb->inner_network_header;
2898 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2900 skb->inner_network_header = skb->data - skb->head;
2903 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2906 skb_reset_inner_network_header(skb);
2907 skb->inner_network_header += offset;
2910 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2912 return skb->head + skb->inner_mac_header;
2915 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2917 skb->inner_mac_header = skb->data - skb->head;
2920 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2923 skb_reset_inner_mac_header(skb);
2924 skb->inner_mac_header += offset;
2926 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2928 return skb->transport_header != (typeof(skb->transport_header))~0U;
2931 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2933 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
2934 return skb->head + skb->transport_header;
2937 static inline void skb_reset_transport_header(struct sk_buff *skb)
2939 skb->transport_header = skb->data - skb->head;
2942 static inline void skb_set_transport_header(struct sk_buff *skb,
2945 skb_reset_transport_header(skb);
2946 skb->transport_header += offset;
2949 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2951 return skb->head + skb->network_header;
2954 static inline void skb_reset_network_header(struct sk_buff *skb)
2956 skb->network_header = skb->data - skb->head;
2959 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2961 skb_reset_network_header(skb);
2962 skb->network_header += offset;
2965 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2967 return skb->head + skb->mac_header;
2970 static inline int skb_mac_offset(const struct sk_buff *skb)
2972 return skb_mac_header(skb) - skb->data;
2975 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2977 return skb->network_header - skb->mac_header;
2980 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2982 return skb->mac_header != (typeof(skb->mac_header))~0U;
2985 static inline void skb_unset_mac_header(struct sk_buff *skb)
2987 skb->mac_header = (typeof(skb->mac_header))~0U;
2990 static inline void skb_reset_mac_header(struct sk_buff *skb)
2992 skb->mac_header = skb->data - skb->head;
2995 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2997 skb_reset_mac_header(skb);
2998 skb->mac_header += offset;
3001 static inline void skb_pop_mac_header(struct sk_buff *skb)
3003 skb->mac_header = skb->network_header;
3006 static inline void skb_probe_transport_header(struct sk_buff *skb)
3008 struct flow_keys_basic keys;
3010 if (skb_transport_header_was_set(skb))
3013 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
3015 skb_set_transport_header(skb, keys.control.thoff);
3018 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
3020 if (skb_mac_header_was_set(skb)) {
3021 const unsigned char *old_mac = skb_mac_header(skb);
3023 skb_set_mac_header(skb, -skb->mac_len);
3024 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
3028 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
3030 return skb->csum_start - skb_headroom(skb);
3033 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
3035 return skb->head + skb->csum_start;
3038 static inline int skb_transport_offset(const struct sk_buff *skb)
3040 return skb_transport_header(skb) - skb->data;
3043 static inline u32 skb_network_header_len(const struct sk_buff *skb)
3045 return skb->transport_header - skb->network_header;
3048 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
3050 return skb->inner_transport_header - skb->inner_network_header;
3053 static inline int skb_network_offset(const struct sk_buff *skb)
3055 return skb_network_header(skb) - skb->data;
3058 static inline int skb_inner_network_offset(const struct sk_buff *skb)
3060 return skb_inner_network_header(skb) - skb->data;
3063 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
3065 return pskb_may_pull(skb, skb_network_offset(skb) + len);
3069 * CPUs often take a performance hit when accessing unaligned memory
3070 * locations. The actual performance hit varies, it can be small if the
3071 * hardware handles it or large if we have to take an exception and fix it
3074 * Since an ethernet header is 14 bytes network drivers often end up with
3075 * the IP header at an unaligned offset. The IP header can be aligned by
3076 * shifting the start of the packet by 2 bytes. Drivers should do this
3079 * skb_reserve(skb, NET_IP_ALIGN);
3081 * The downside to this alignment of the IP header is that the DMA is now
3082 * unaligned. On some architectures the cost of an unaligned DMA is high
3083 * and this cost outweighs the gains made by aligning the IP header.
3085 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
3088 #ifndef NET_IP_ALIGN
3089 #define NET_IP_ALIGN 2
3093 * The networking layer reserves some headroom in skb data (via
3094 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
3095 * the header has to grow. In the default case, if the header has to grow
3096 * 32 bytes or less we avoid the reallocation.
3098 * Unfortunately this headroom changes the DMA alignment of the resulting
3099 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
3100 * on some architectures. An architecture can override this value,
3101 * perhaps setting it to a cacheline in size (since that will maintain
3102 * cacheline alignment of the DMA). It must be a power of 2.
3104 * Various parts of the networking layer expect at least 32 bytes of
3105 * headroom, you should not reduce this.
3107 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
3108 * to reduce average number of cache lines per packet.
3109 * get_rps_cpu() for example only access one 64 bytes aligned block :
3110 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
3113 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
3116 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
3118 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
3120 if (WARN_ON(skb_is_nonlinear(skb)))
3123 skb_set_tail_pointer(skb, len);
3126 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
3128 __skb_set_length(skb, len);
3131 void skb_trim(struct sk_buff *skb, unsigned int len);
3133 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
3136 return ___pskb_trim(skb, len);
3137 __skb_trim(skb, len);
3141 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
3143 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
3147 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
3148 * @skb: buffer to alter
3151 * This is identical to pskb_trim except that the caller knows that
3152 * the skb is not cloned so we should never get an error due to out-
3155 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3157 int err = pskb_trim(skb, len);
3161 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3163 unsigned int diff = len - skb->len;
3165 if (skb_tailroom(skb) < diff) {
3166 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3171 __skb_set_length(skb, len);
3176 * skb_orphan - orphan a buffer
3177 * @skb: buffer to orphan
3179 * If a buffer currently has an owner then we call the owner's
3180 * destructor function and make the @skb unowned. The buffer continues
3181 * to exist but is no longer charged to its former owner.
3183 static inline void skb_orphan(struct sk_buff *skb)
3185 if (skb->destructor) {
3186 skb->destructor(skb);
3187 skb->destructor = NULL;
3195 * skb_orphan_frags - orphan the frags contained in a buffer
3196 * @skb: buffer to orphan frags from
3197 * @gfp_mask: allocation mask for replacement pages
3199 * For each frag in the SKB which needs a destructor (i.e. has an
3200 * owner) create a copy of that frag and release the original
3201 * page by calling the destructor.
3203 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3205 if (likely(!skb_zcopy(skb)))
3207 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN)
3209 return skb_copy_ubufs(skb, gfp_mask);
3212 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3213 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3215 if (likely(!skb_zcopy(skb)))
3217 return skb_copy_ubufs(skb, gfp_mask);
3221 * __skb_queue_purge - empty a list
3222 * @list: list to empty
3224 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3225 * the list and one reference dropped. This function does not take the
3226 * list lock and the caller must hold the relevant locks to use it.
3228 static inline void __skb_queue_purge(struct sk_buff_head *list)
3230 struct sk_buff *skb;
3231 while ((skb = __skb_dequeue(list)) != NULL)
3234 void skb_queue_purge(struct sk_buff_head *list);
3236 unsigned int skb_rbtree_purge(struct rb_root *root);
3238 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3241 * netdev_alloc_frag - allocate a page fragment
3242 * @fragsz: fragment size
3244 * Allocates a frag from a page for receive buffer.
3245 * Uses GFP_ATOMIC allocations.
3247 static inline void *netdev_alloc_frag(unsigned int fragsz)
3249 return __netdev_alloc_frag_align(fragsz, ~0u);
3252 static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3255 WARN_ON_ONCE(!is_power_of_2(align));
3256 return __netdev_alloc_frag_align(fragsz, -align);
3259 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3263 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3264 * @dev: network device to receive on
3265 * @length: length to allocate
3267 * Allocate a new &sk_buff and assign it a usage count of one. The
3268 * buffer has unspecified headroom built in. Users should allocate
3269 * the headroom they think they need without accounting for the
3270 * built in space. The built in space is used for optimisations.
3272 * %NULL is returned if there is no free memory. Although this function
3273 * allocates memory it can be called from an interrupt.
3275 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3276 unsigned int length)
3278 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3281 /* legacy helper around __netdev_alloc_skb() */
3282 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3285 return __netdev_alloc_skb(NULL, length, gfp_mask);
3288 /* legacy helper around netdev_alloc_skb() */
3289 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3291 return netdev_alloc_skb(NULL, length);
3295 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3296 unsigned int length, gfp_t gfp)
3298 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3300 if (NET_IP_ALIGN && skb)
3301 skb_reserve(skb, NET_IP_ALIGN);
3305 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3306 unsigned int length)
3308 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3311 static inline void skb_free_frag(void *addr)
3313 page_frag_free(addr);
3316 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3318 static inline void *napi_alloc_frag(unsigned int fragsz)
3320 return __napi_alloc_frag_align(fragsz, ~0u);
3323 static inline void *napi_alloc_frag_align(unsigned int fragsz,
3326 WARN_ON_ONCE(!is_power_of_2(align));
3327 return __napi_alloc_frag_align(fragsz, -align);
3330 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
3331 unsigned int length, gfp_t gfp_mask);
3332 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
3333 unsigned int length)
3335 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
3337 void napi_consume_skb(struct sk_buff *skb, int budget);
3339 void napi_skb_free_stolen_head(struct sk_buff *skb);
3340 void __kfree_skb_defer(struct sk_buff *skb);
3343 * __dev_alloc_pages - allocate page for network Rx
3344 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3345 * @order: size of the allocation
3347 * Allocate a new page.
3349 * %NULL is returned if there is no free memory.
3351 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
3354 /* This piece of code contains several assumptions.
3355 * 1. This is for device Rx, therefor a cold page is preferred.
3356 * 2. The expectation is the user wants a compound page.
3357 * 3. If requesting a order 0 page it will not be compound
3358 * due to the check to see if order has a value in prep_new_page
3359 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3360 * code in gfp_to_alloc_flags that should be enforcing this.
3362 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3364 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
3367 static inline struct page *dev_alloc_pages(unsigned int order)
3369 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
3373 * __dev_alloc_page - allocate a page for network Rx
3374 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3376 * Allocate a new page.
3378 * %NULL is returned if there is no free memory.
3380 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
3382 return __dev_alloc_pages(gfp_mask, 0);
3385 static inline struct page *dev_alloc_page(void)
3387 return dev_alloc_pages(0);
3391 * dev_page_is_reusable - check whether a page can be reused for network Rx
3392 * @page: the page to test
3394 * A page shouldn't be considered for reusing/recycling if it was allocated
3395 * under memory pressure or at a distant memory node.
3397 * Returns false if this page should be returned to page allocator, true
3400 static inline bool dev_page_is_reusable(const struct page *page)
3402 return likely(page_to_nid(page) == numa_mem_id() &&
3403 !page_is_pfmemalloc(page));
3407 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3408 * @page: The page that was allocated from skb_alloc_page
3409 * @skb: The skb that may need pfmemalloc set
3411 static inline void skb_propagate_pfmemalloc(const struct page *page,
3412 struct sk_buff *skb)
3414 if (page_is_pfmemalloc(page))
3415 skb->pfmemalloc = true;
3419 * skb_frag_off() - Returns the offset of a skb fragment
3420 * @frag: the paged fragment
3422 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3424 return frag->bv_offset;
3428 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3429 * @frag: skb fragment
3430 * @delta: value to add
3432 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3434 frag->bv_offset += delta;
3438 * skb_frag_off_set() - Sets the offset of a skb fragment
3439 * @frag: skb fragment
3440 * @offset: offset of fragment
3442 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3444 frag->bv_offset = offset;
3448 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3449 * @fragto: skb fragment where offset is set
3450 * @fragfrom: skb fragment offset is copied from
3452 static inline void skb_frag_off_copy(skb_frag_t *fragto,
3453 const skb_frag_t *fragfrom)
3455 fragto->bv_offset = fragfrom->bv_offset;
3459 * skb_frag_page - retrieve the page referred to by a paged fragment
3460 * @frag: the paged fragment
3462 * Returns the &struct page associated with @frag.
3464 static inline struct page *skb_frag_page(const skb_frag_t *frag)
3466 return frag->bv_page;
3470 * __skb_frag_ref - take an addition reference on a paged fragment.
3471 * @frag: the paged fragment
3473 * Takes an additional reference on the paged fragment @frag.
3475 static inline void __skb_frag_ref(skb_frag_t *frag)
3477 get_page(skb_frag_page(frag));
3481 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
3483 * @f: the fragment offset.
3485 * Takes an additional reference on the @f'th paged fragment of @skb.
3487 static inline void skb_frag_ref(struct sk_buff *skb, int f)
3489 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
3493 * __skb_frag_unref - release a reference on a paged fragment.
3494 * @frag: the paged fragment
3495 * @recycle: recycle the page if allocated via page_pool
3497 * Releases a reference on the paged fragment @frag
3498 * or recycles the page via the page_pool API.
3500 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle)
3502 struct page *page = skb_frag_page(frag);
3504 #ifdef CONFIG_PAGE_POOL
3505 if (recycle && page_pool_return_skb_page(page))
3512 * skb_frag_unref - release a reference on a paged fragment of an skb.
3514 * @f: the fragment offset
3516 * Releases a reference on the @f'th paged fragment of @skb.
3518 static inline void skb_frag_unref(struct sk_buff *skb, int f)
3520 struct skb_shared_info *shinfo = skb_shinfo(skb);
3522 if (!skb_zcopy_managed(skb))
3523 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle);
3527 * skb_frag_address - gets the address of the data contained in a paged fragment
3528 * @frag: the paged fragment buffer
3530 * Returns the address of the data within @frag. The page must already
3533 static inline void *skb_frag_address(const skb_frag_t *frag)
3535 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3539 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3540 * @frag: the paged fragment buffer
3542 * Returns the address of the data within @frag. Checks that the page
3543 * is mapped and returns %NULL otherwise.
3545 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3547 void *ptr = page_address(skb_frag_page(frag));
3551 return ptr + skb_frag_off(frag);
3555 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3556 * @fragto: skb fragment where page is set
3557 * @fragfrom: skb fragment page is copied from
3559 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3560 const skb_frag_t *fragfrom)
3562 fragto->bv_page = fragfrom->bv_page;
3566 * __skb_frag_set_page - sets the page contained in a paged fragment
3567 * @frag: the paged fragment
3568 * @page: the page to set
3570 * Sets the fragment @frag to contain @page.
3572 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3574 frag->bv_page = page;
3578 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3580 * @f: the fragment offset
3581 * @page: the page to set
3583 * Sets the @f'th fragment of @skb to contain @page.
3585 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3588 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3591 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3594 * skb_frag_dma_map - maps a paged fragment via the DMA API
3595 * @dev: the device to map the fragment to
3596 * @frag: the paged fragment to map
3597 * @offset: the offset within the fragment (starting at the
3598 * fragment's own offset)
3599 * @size: the number of bytes to map
3600 * @dir: the direction of the mapping (``PCI_DMA_*``)
3602 * Maps the page associated with @frag to @device.
3604 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3605 const skb_frag_t *frag,
3606 size_t offset, size_t size,
3607 enum dma_data_direction dir)
3609 return dma_map_page(dev, skb_frag_page(frag),
3610 skb_frag_off(frag) + offset, size, dir);
3613 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3616 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3620 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3623 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3628 * skb_clone_writable - is the header of a clone writable
3629 * @skb: buffer to check
3630 * @len: length up to which to write
3632 * Returns true if modifying the header part of the cloned buffer
3633 * does not requires the data to be copied.
3635 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3637 return !skb_header_cloned(skb) &&
3638 skb_headroom(skb) + len <= skb->hdr_len;
3641 static inline int skb_try_make_writable(struct sk_buff *skb,
3642 unsigned int write_len)
3644 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3645 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3648 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3653 if (headroom > skb_headroom(skb))
3654 delta = headroom - skb_headroom(skb);
3656 if (delta || cloned)
3657 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3663 * skb_cow - copy header of skb when it is required
3664 * @skb: buffer to cow
3665 * @headroom: needed headroom
3667 * If the skb passed lacks sufficient headroom or its data part
3668 * is shared, data is reallocated. If reallocation fails, an error
3669 * is returned and original skb is not changed.
3671 * The result is skb with writable area skb->head...skb->tail
3672 * and at least @headroom of space at head.
3674 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3676 return __skb_cow(skb, headroom, skb_cloned(skb));
3680 * skb_cow_head - skb_cow but only making the head writable
3681 * @skb: buffer to cow
3682 * @headroom: needed headroom
3684 * This function is identical to skb_cow except that we replace the
3685 * skb_cloned check by skb_header_cloned. It should be used when
3686 * you only need to push on some header and do not need to modify
3689 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3691 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3695 * skb_padto - pad an skbuff up to a minimal size
3696 * @skb: buffer to pad
3697 * @len: minimal length
3699 * Pads up a buffer to ensure the trailing bytes exist and are
3700 * blanked. If the buffer already contains sufficient data it
3701 * is untouched. Otherwise it is extended. Returns zero on
3702 * success. The skb is freed on error.
3704 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3706 unsigned int size = skb->len;
3707 if (likely(size >= len))
3709 return skb_pad(skb, len - size);
3713 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3714 * @skb: buffer to pad
3715 * @len: minimal length
3716 * @free_on_error: free buffer on error
3718 * Pads up a buffer to ensure the trailing bytes exist and are
3719 * blanked. If the buffer already contains sufficient data it
3720 * is untouched. Otherwise it is extended. Returns zero on
3721 * success. The skb is freed on error if @free_on_error is true.
3723 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3727 unsigned int size = skb->len;
3729 if (unlikely(size < len)) {
3731 if (__skb_pad(skb, len, free_on_error))
3733 __skb_put(skb, len);
3739 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3740 * @skb: buffer to pad
3741 * @len: minimal length
3743 * Pads up a buffer to ensure the trailing bytes exist and are
3744 * blanked. If the buffer already contains sufficient data it
3745 * is untouched. Otherwise it is extended. Returns zero on
3746 * success. The skb is freed on error.
3748 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3750 return __skb_put_padto(skb, len, true);
3753 static inline int skb_add_data(struct sk_buff *skb,
3754 struct iov_iter *from, int copy)
3756 const int off = skb->len;
3758 if (skb->ip_summed == CHECKSUM_NONE) {
3760 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3762 skb->csum = csum_block_add(skb->csum, csum, off);
3765 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3768 __skb_trim(skb, off);
3772 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3773 const struct page *page, int off)
3778 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3780 return page == skb_frag_page(frag) &&
3781 off == skb_frag_off(frag) + skb_frag_size(frag);
3786 static inline int __skb_linearize(struct sk_buff *skb)
3788 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3792 * skb_linearize - convert paged skb to linear one
3793 * @skb: buffer to linarize
3795 * If there is no free memory -ENOMEM is returned, otherwise zero
3796 * is returned and the old skb data released.
3798 static inline int skb_linearize(struct sk_buff *skb)
3800 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3804 * skb_has_shared_frag - can any frag be overwritten
3805 * @skb: buffer to test
3807 * Return true if the skb has at least one frag that might be modified
3808 * by an external entity (as in vmsplice()/sendfile())
3810 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3812 return skb_is_nonlinear(skb) &&
3813 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3817 * skb_linearize_cow - make sure skb is linear and writable
3818 * @skb: buffer to process
3820 * If there is no free memory -ENOMEM is returned, otherwise zero
3821 * is returned and the old skb data released.
3823 static inline int skb_linearize_cow(struct sk_buff *skb)
3825 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3826 __skb_linearize(skb) : 0;
3829 static __always_inline void
3830 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3833 if (skb->ip_summed == CHECKSUM_COMPLETE)
3834 skb->csum = csum_block_sub(skb->csum,
3835 csum_partial(start, len, 0), off);
3836 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3837 skb_checksum_start_offset(skb) < 0)
3838 skb->ip_summed = CHECKSUM_NONE;
3842 * skb_postpull_rcsum - update checksum for received skb after pull
3843 * @skb: buffer to update
3844 * @start: start of data before pull
3845 * @len: length of data pulled
3847 * After doing a pull on a received packet, you need to call this to
3848 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3849 * CHECKSUM_NONE so that it can be recomputed from scratch.
3851 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3852 const void *start, unsigned int len)
3854 if (skb->ip_summed == CHECKSUM_COMPLETE)
3855 skb->csum = wsum_negate(csum_partial(start, len,
3856 wsum_negate(skb->csum)));
3857 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3858 skb_checksum_start_offset(skb) < 0)
3859 skb->ip_summed = CHECKSUM_NONE;
3862 static __always_inline void
3863 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3866 if (skb->ip_summed == CHECKSUM_COMPLETE)
3867 skb->csum = csum_block_add(skb->csum,
3868 csum_partial(start, len, 0), off);
3872 * skb_postpush_rcsum - update checksum for received skb after push
3873 * @skb: buffer to update
3874 * @start: start of data after push
3875 * @len: length of data pushed
3877 * After doing a push on a received packet, you need to call this to
3878 * update the CHECKSUM_COMPLETE checksum.
3880 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3881 const void *start, unsigned int len)
3883 __skb_postpush_rcsum(skb, start, len, 0);
3886 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3889 * skb_push_rcsum - push skb and update receive checksum
3890 * @skb: buffer to update
3891 * @len: length of data pulled
3893 * This function performs an skb_push on the packet and updates
3894 * the CHECKSUM_COMPLETE checksum. It should be used on
3895 * receive path processing instead of skb_push unless you know
3896 * that the checksum difference is zero (e.g., a valid IP header)
3897 * or you are setting ip_summed to CHECKSUM_NONE.
3899 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3902 skb_postpush_rcsum(skb, skb->data, len);
3906 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3908 * pskb_trim_rcsum - trim received skb and update checksum
3909 * @skb: buffer to trim
3912 * This is exactly the same as pskb_trim except that it ensures the
3913 * checksum of received packets are still valid after the operation.
3914 * It can change skb pointers.
3917 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3919 if (likely(len >= skb->len))
3921 return pskb_trim_rcsum_slow(skb, len);
3924 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3926 if (skb->ip_summed == CHECKSUM_COMPLETE)
3927 skb->ip_summed = CHECKSUM_NONE;
3928 __skb_trim(skb, len);
3932 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3934 if (skb->ip_summed == CHECKSUM_COMPLETE)
3935 skb->ip_summed = CHECKSUM_NONE;
3936 return __skb_grow(skb, len);
3939 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3940 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3941 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3942 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3943 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3945 #define skb_queue_walk(queue, skb) \
3946 for (skb = (queue)->next; \
3947 skb != (struct sk_buff *)(queue); \
3950 #define skb_queue_walk_safe(queue, skb, tmp) \
3951 for (skb = (queue)->next, tmp = skb->next; \
3952 skb != (struct sk_buff *)(queue); \
3953 skb = tmp, tmp = skb->next)
3955 #define skb_queue_walk_from(queue, skb) \
3956 for (; skb != (struct sk_buff *)(queue); \
3959 #define skb_rbtree_walk(skb, root) \
3960 for (skb = skb_rb_first(root); skb != NULL; \
3961 skb = skb_rb_next(skb))
3963 #define skb_rbtree_walk_from(skb) \
3964 for (; skb != NULL; \
3965 skb = skb_rb_next(skb))
3967 #define skb_rbtree_walk_from_safe(skb, tmp) \
3968 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3971 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3972 for (tmp = skb->next; \
3973 skb != (struct sk_buff *)(queue); \
3974 skb = tmp, tmp = skb->next)
3976 #define skb_queue_reverse_walk(queue, skb) \
3977 for (skb = (queue)->prev; \
3978 skb != (struct sk_buff *)(queue); \
3981 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3982 for (skb = (queue)->prev, tmp = skb->prev; \
3983 skb != (struct sk_buff *)(queue); \
3984 skb = tmp, tmp = skb->prev)
3986 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3987 for (tmp = skb->prev; \
3988 skb != (struct sk_buff *)(queue); \
3989 skb = tmp, tmp = skb->prev)
3991 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3993 return skb_shinfo(skb)->frag_list != NULL;
3996 static inline void skb_frag_list_init(struct sk_buff *skb)
3998 skb_shinfo(skb)->frag_list = NULL;
4001 #define skb_walk_frags(skb, iter) \
4002 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
4005 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
4006 int *err, long *timeo_p,
4007 const struct sk_buff *skb);
4008 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
4009 struct sk_buff_head *queue,
4012 struct sk_buff **last);
4013 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
4014 struct sk_buff_head *queue,
4015 unsigned int flags, int *off, int *err,
4016 struct sk_buff **last);
4017 struct sk_buff *__skb_recv_datagram(struct sock *sk,
4018 struct sk_buff_head *sk_queue,
4019 unsigned int flags, int *off, int *err);
4020 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
4021 __poll_t datagram_poll(struct file *file, struct socket *sock,
4022 struct poll_table_struct *wait);
4023 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
4024 struct iov_iter *to, int size);
4025 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
4026 struct msghdr *msg, int size)
4028 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
4030 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
4031 struct msghdr *msg);
4032 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
4033 struct iov_iter *to, int len,
4034 struct ahash_request *hash);
4035 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
4036 struct iov_iter *from, int len);
4037 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
4038 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
4039 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
4040 static inline void skb_free_datagram_locked(struct sock *sk,
4041 struct sk_buff *skb)
4043 __skb_free_datagram_locked(sk, skb, 0);
4045 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
4046 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
4047 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
4048 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
4050 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
4051 struct pipe_inode_info *pipe, unsigned int len,
4052 unsigned int flags);
4053 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
4055 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
4056 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
4057 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
4058 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
4060 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
4061 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
4062 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
4063 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
4064 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
4065 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
4066 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
4067 unsigned int offset);
4068 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
4069 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
4070 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
4071 int skb_vlan_pop(struct sk_buff *skb);
4072 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
4073 int skb_eth_pop(struct sk_buff *skb);
4074 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
4075 const unsigned char *src);
4076 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
4077 int mac_len, bool ethernet);
4078 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
4080 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
4081 int skb_mpls_dec_ttl(struct sk_buff *skb);
4082 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
4085 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
4087 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
4090 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
4092 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
4095 struct skb_checksum_ops {
4096 __wsum (*update)(const void *mem, int len, __wsum wsum);
4097 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
4100 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
4102 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
4103 __wsum csum, const struct skb_checksum_ops *ops);
4104 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
4107 static inline void * __must_check
4108 __skb_header_pointer(const struct sk_buff *skb, int offset, int len,
4109 const void *data, int hlen, void *buffer)
4111 if (likely(hlen - offset >= len))
4112 return (void *)data + offset;
4114 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
4120 static inline void * __must_check
4121 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
4123 return __skb_header_pointer(skb, offset, len, skb->data,
4124 skb_headlen(skb), buffer);
4128 * skb_needs_linearize - check if we need to linearize a given skb
4129 * depending on the given device features.
4130 * @skb: socket buffer to check
4131 * @features: net device features
4133 * Returns true if either:
4134 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
4135 * 2. skb is fragmented and the device does not support SG.
4137 static inline bool skb_needs_linearize(struct sk_buff *skb,
4138 netdev_features_t features)
4140 return skb_is_nonlinear(skb) &&
4141 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
4142 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
4145 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
4147 const unsigned int len)
4149 memcpy(to, skb->data, len);
4152 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4153 const int offset, void *to,
4154 const unsigned int len)
4156 memcpy(to, skb->data + offset, len);
4159 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4161 const unsigned int len)
4163 memcpy(skb->data, from, len);
4166 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4169 const unsigned int len)
4171 memcpy(skb->data + offset, from, len);
4174 void skb_init(void);
4176 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4182 * skb_get_timestamp - get timestamp from a skb
4183 * @skb: skb to get stamp from
4184 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4186 * Timestamps are stored in the skb as offsets to a base timestamp.
4187 * This function converts the offset back to a struct timeval and stores
4190 static inline void skb_get_timestamp(const struct sk_buff *skb,
4191 struct __kernel_old_timeval *stamp)
4193 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4196 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4197 struct __kernel_sock_timeval *stamp)
4199 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4201 stamp->tv_sec = ts.tv_sec;
4202 stamp->tv_usec = ts.tv_nsec / 1000;
4205 static inline void skb_get_timestampns(const struct sk_buff *skb,
4206 struct __kernel_old_timespec *stamp)
4208 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4210 stamp->tv_sec = ts.tv_sec;
4211 stamp->tv_nsec = ts.tv_nsec;
4214 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4215 struct __kernel_timespec *stamp)
4217 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4219 stamp->tv_sec = ts.tv_sec;
4220 stamp->tv_nsec = ts.tv_nsec;
4223 static inline void __net_timestamp(struct sk_buff *skb)
4225 skb->tstamp = ktime_get_real();
4226 skb->mono_delivery_time = 0;
4229 static inline ktime_t net_timedelta(ktime_t t)
4231 return ktime_sub(ktime_get_real(), t);
4234 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4238 skb->mono_delivery_time = kt && mono;
4241 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4243 /* It is used in the ingress path to clear the delivery_time.
4244 * If needed, set the skb->tstamp to the (rcv) timestamp.
4246 static inline void skb_clear_delivery_time(struct sk_buff *skb)
4248 if (skb->mono_delivery_time) {
4249 skb->mono_delivery_time = 0;
4250 if (static_branch_unlikely(&netstamp_needed_key))
4251 skb->tstamp = ktime_get_real();
4257 static inline void skb_clear_tstamp(struct sk_buff *skb)
4259 if (skb->mono_delivery_time)
4265 static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4267 if (skb->mono_delivery_time)
4273 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4275 if (!skb->mono_delivery_time && skb->tstamp)
4278 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4279 return ktime_get_real();
4284 static inline u8 skb_metadata_len(const struct sk_buff *skb)
4286 return skb_shinfo(skb)->meta_len;
4289 static inline void *skb_metadata_end(const struct sk_buff *skb)
4291 return skb_mac_header(skb);
4294 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4295 const struct sk_buff *skb_b,
4298 const void *a = skb_metadata_end(skb_a);
4299 const void *b = skb_metadata_end(skb_b);
4300 /* Using more efficient varaiant than plain call to memcmp(). */
4301 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
4305 #define __it(x, op) (x -= sizeof(u##op))
4306 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4307 case 32: diffs |= __it_diff(a, b, 64);
4309 case 24: diffs |= __it_diff(a, b, 64);
4311 case 16: diffs |= __it_diff(a, b, 64);
4313 case 8: diffs |= __it_diff(a, b, 64);
4315 case 28: diffs |= __it_diff(a, b, 64);
4317 case 20: diffs |= __it_diff(a, b, 64);
4319 case 12: diffs |= __it_diff(a, b, 64);
4321 case 4: diffs |= __it_diff(a, b, 32);
4326 return memcmp(a - meta_len, b - meta_len, meta_len);
4330 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4331 const struct sk_buff *skb_b)
4333 u8 len_a = skb_metadata_len(skb_a);
4334 u8 len_b = skb_metadata_len(skb_b);
4336 if (!(len_a | len_b))
4339 return len_a != len_b ?
4340 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4343 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4345 skb_shinfo(skb)->meta_len = meta_len;
4348 static inline void skb_metadata_clear(struct sk_buff *skb)
4350 skb_metadata_set(skb, 0);
4353 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4355 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4357 void skb_clone_tx_timestamp(struct sk_buff *skb);
4358 bool skb_defer_rx_timestamp(struct sk_buff *skb);
4360 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4362 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4366 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4371 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4374 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4376 * PHY drivers may accept clones of transmitted packets for
4377 * timestamping via their phy_driver.txtstamp method. These drivers
4378 * must call this function to return the skb back to the stack with a
4381 * @skb: clone of the original outgoing packet
4382 * @hwtstamps: hardware time stamps
4385 void skb_complete_tx_timestamp(struct sk_buff *skb,
4386 struct skb_shared_hwtstamps *hwtstamps);
4388 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4389 struct skb_shared_hwtstamps *hwtstamps,
4390 struct sock *sk, int tstype);
4393 * skb_tstamp_tx - queue clone of skb with send time stamps
4394 * @orig_skb: the original outgoing packet
4395 * @hwtstamps: hardware time stamps, may be NULL if not available
4397 * If the skb has a socket associated, then this function clones the
4398 * skb (thus sharing the actual data and optional structures), stores
4399 * the optional hardware time stamping information (if non NULL) or
4400 * generates a software time stamp (otherwise), then queues the clone
4401 * to the error queue of the socket. Errors are silently ignored.
4403 void skb_tstamp_tx(struct sk_buff *orig_skb,
4404 struct skb_shared_hwtstamps *hwtstamps);
4407 * skb_tx_timestamp() - Driver hook for transmit timestamping
4409 * Ethernet MAC Drivers should call this function in their hard_xmit()
4410 * function immediately before giving the sk_buff to the MAC hardware.
4412 * Specifically, one should make absolutely sure that this function is
4413 * called before TX completion of this packet can trigger. Otherwise
4414 * the packet could potentially already be freed.
4416 * @skb: A socket buffer.
4418 static inline void skb_tx_timestamp(struct sk_buff *skb)
4420 skb_clone_tx_timestamp(skb);
4421 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4422 skb_tstamp_tx(skb, NULL);
4426 * skb_complete_wifi_ack - deliver skb with wifi status
4428 * @skb: the original outgoing packet
4429 * @acked: ack status
4432 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4434 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4435 __sum16 __skb_checksum_complete(struct sk_buff *skb);
4437 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4439 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4441 (skb->ip_summed == CHECKSUM_PARTIAL &&
4442 skb_checksum_start_offset(skb) >= 0));
4446 * skb_checksum_complete - Calculate checksum of an entire packet
4447 * @skb: packet to process
4449 * This function calculates the checksum over the entire packet plus
4450 * the value of skb->csum. The latter can be used to supply the
4451 * checksum of a pseudo header as used by TCP/UDP. It returns the
4454 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4455 * this function can be used to verify that checksum on received
4456 * packets. In that case the function should return zero if the
4457 * checksum is correct. In particular, this function will return zero
4458 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4459 * hardware has already verified the correctness of the checksum.
4461 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4463 return skb_csum_unnecessary(skb) ?
4464 0 : __skb_checksum_complete(skb);
4467 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4469 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4470 if (skb->csum_level == 0)
4471 skb->ip_summed = CHECKSUM_NONE;
4477 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4479 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4480 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4482 } else if (skb->ip_summed == CHECKSUM_NONE) {
4483 skb->ip_summed = CHECKSUM_UNNECESSARY;
4484 skb->csum_level = 0;
4488 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4490 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4491 skb->ip_summed = CHECKSUM_NONE;
4492 skb->csum_level = 0;
4496 /* Check if we need to perform checksum complete validation.
4498 * Returns true if checksum complete is needed, false otherwise
4499 * (either checksum is unnecessary or zero checksum is allowed).
4501 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4505 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4506 skb->csum_valid = 1;
4507 __skb_decr_checksum_unnecessary(skb);
4514 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4517 #define CHECKSUM_BREAK 76
4519 /* Unset checksum-complete
4521 * Unset checksum complete can be done when packet is being modified
4522 * (uncompressed for instance) and checksum-complete value is
4525 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4527 if (skb->ip_summed == CHECKSUM_COMPLETE)
4528 skb->ip_summed = CHECKSUM_NONE;
4531 /* Validate (init) checksum based on checksum complete.
4534 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4535 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4536 * checksum is stored in skb->csum for use in __skb_checksum_complete
4537 * non-zero: value of invalid checksum
4540 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4544 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4545 if (!csum_fold(csum_add(psum, skb->csum))) {
4546 skb->csum_valid = 1;
4553 if (complete || skb->len <= CHECKSUM_BREAK) {
4556 csum = __skb_checksum_complete(skb);
4557 skb->csum_valid = !csum;
4564 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4569 /* Perform checksum validate (init). Note that this is a macro since we only
4570 * want to calculate the pseudo header which is an input function if necessary.
4571 * First we try to validate without any computation (checksum unnecessary) and
4572 * then calculate based on checksum complete calling the function to compute
4576 * 0: checksum is validated or try to in skb_checksum_complete
4577 * non-zero: value of invalid checksum
4579 #define __skb_checksum_validate(skb, proto, complete, \
4580 zero_okay, check, compute_pseudo) \
4582 __sum16 __ret = 0; \
4583 skb->csum_valid = 0; \
4584 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4585 __ret = __skb_checksum_validate_complete(skb, \
4586 complete, compute_pseudo(skb, proto)); \
4590 #define skb_checksum_init(skb, proto, compute_pseudo) \
4591 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4593 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4594 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4596 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4597 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4599 #define skb_checksum_validate_zero_check(skb, proto, check, \
4601 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4603 #define skb_checksum_simple_validate(skb) \
4604 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4606 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4608 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4611 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4613 skb->csum = ~pseudo;
4614 skb->ip_summed = CHECKSUM_COMPLETE;
4617 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4619 if (__skb_checksum_convert_check(skb)) \
4620 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4623 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4624 u16 start, u16 offset)
4626 skb->ip_summed = CHECKSUM_PARTIAL;
4627 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4628 skb->csum_offset = offset - start;
4631 /* Update skbuf and packet to reflect the remote checksum offload operation.
4632 * When called, ptr indicates the starting point for skb->csum when
4633 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4634 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4636 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4637 int start, int offset, bool nopartial)
4642 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4646 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4647 __skb_checksum_complete(skb);
4648 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4651 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4653 /* Adjust skb->csum since we changed the packet */
4654 skb->csum = csum_add(skb->csum, delta);
4657 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4659 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4660 return (void *)(skb->_nfct & NFCT_PTRMASK);
4666 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4668 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4675 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4677 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4678 skb->slow_gro |= !!nfct;
4683 #ifdef CONFIG_SKB_EXTENSIONS
4685 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4691 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4694 #if IS_ENABLED(CONFIG_MPTCP)
4697 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4700 SKB_EXT_NUM, /* must be last */
4704 * struct skb_ext - sk_buff extensions
4705 * @refcnt: 1 on allocation, deallocated on 0
4706 * @offset: offset to add to @data to obtain extension address
4707 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4708 * @data: start of extension data, variable sized
4710 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4711 * to use 'u8' types while allowing up to 2kb worth of extension data.
4715 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4716 u8 chunks; /* same */
4717 char data[] __aligned(8);
4720 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4721 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4722 struct skb_ext *ext);
4723 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4724 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4725 void __skb_ext_put(struct skb_ext *ext);
4727 static inline void skb_ext_put(struct sk_buff *skb)
4729 if (skb->active_extensions)
4730 __skb_ext_put(skb->extensions);
4733 static inline void __skb_ext_copy(struct sk_buff *dst,
4734 const struct sk_buff *src)
4736 dst->active_extensions = src->active_extensions;
4738 if (src->active_extensions) {
4739 struct skb_ext *ext = src->extensions;
4741 refcount_inc(&ext->refcnt);
4742 dst->extensions = ext;
4746 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4749 __skb_ext_copy(dst, src);
4752 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4754 return !!ext->offset[i];
4757 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4759 return skb->active_extensions & (1 << id);
4762 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4764 if (skb_ext_exist(skb, id))
4765 __skb_ext_del(skb, id);
4768 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4770 if (skb_ext_exist(skb, id)) {
4771 struct skb_ext *ext = skb->extensions;
4773 return (void *)ext + (ext->offset[id] << 3);
4779 static inline void skb_ext_reset(struct sk_buff *skb)
4781 if (unlikely(skb->active_extensions)) {
4782 __skb_ext_put(skb->extensions);
4783 skb->active_extensions = 0;
4787 static inline bool skb_has_extensions(struct sk_buff *skb)
4789 return unlikely(skb->active_extensions);
4792 static inline void skb_ext_put(struct sk_buff *skb) {}
4793 static inline void skb_ext_reset(struct sk_buff *skb) {}
4794 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4795 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4796 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4797 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4798 #endif /* CONFIG_SKB_EXTENSIONS */
4800 static inline void nf_reset_ct(struct sk_buff *skb)
4802 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4803 nf_conntrack_put(skb_nfct(skb));
4808 static inline void nf_reset_trace(struct sk_buff *skb)
4810 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4815 static inline void ipvs_reset(struct sk_buff *skb)
4817 #if IS_ENABLED(CONFIG_IP_VS)
4818 skb->ipvs_property = 0;
4822 /* Note: This doesn't put any conntrack info in dst. */
4823 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4826 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4827 dst->_nfct = src->_nfct;
4828 nf_conntrack_get(skb_nfct(src));
4830 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4832 dst->nf_trace = src->nf_trace;
4836 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4838 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4839 nf_conntrack_put(skb_nfct(dst));
4841 dst->slow_gro = src->slow_gro;
4842 __nf_copy(dst, src, true);
4845 #ifdef CONFIG_NETWORK_SECMARK
4846 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4848 to->secmark = from->secmark;
4851 static inline void skb_init_secmark(struct sk_buff *skb)
4856 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4859 static inline void skb_init_secmark(struct sk_buff *skb)
4863 static inline int secpath_exists(const struct sk_buff *skb)
4866 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4872 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4874 return !skb->destructor &&
4875 !secpath_exists(skb) &&
4877 !skb->_skb_refdst &&
4878 !skb_has_frag_list(skb);
4881 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4883 skb->queue_mapping = queue_mapping;
4886 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4888 return skb->queue_mapping;
4891 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4893 to->queue_mapping = from->queue_mapping;
4896 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4898 skb->queue_mapping = rx_queue + 1;
4901 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4903 return skb->queue_mapping - 1;
4906 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4908 return skb->queue_mapping != 0;
4911 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4913 skb->dst_pending_confirm = val;
4916 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4918 return skb->dst_pending_confirm != 0;
4921 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4924 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4930 /* Keeps track of mac header offset relative to skb->head.
4931 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4932 * For non-tunnel skb it points to skb_mac_header() and for
4933 * tunnel skb it points to outer mac header.
4934 * Keeps track of level of encapsulation of network headers.
4945 #define SKB_GSO_CB_OFFSET 32
4946 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET))
4948 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4950 return (skb_mac_header(inner_skb) - inner_skb->head) -
4951 SKB_GSO_CB(inner_skb)->mac_offset;
4954 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4956 int new_headroom, headroom;
4959 headroom = skb_headroom(skb);
4960 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4964 new_headroom = skb_headroom(skb);
4965 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4969 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4971 /* Do not update partial checksums if remote checksum is enabled. */
4972 if (skb->remcsum_offload)
4975 SKB_GSO_CB(skb)->csum = res;
4976 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4979 /* Compute the checksum for a gso segment. First compute the checksum value
4980 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4981 * then add in skb->csum (checksum from csum_start to end of packet).
4982 * skb->csum and csum_start are then updated to reflect the checksum of the
4983 * resultant packet starting from the transport header-- the resultant checksum
4984 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4987 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4989 unsigned char *csum_start = skb_transport_header(skb);
4990 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4991 __wsum partial = SKB_GSO_CB(skb)->csum;
4993 SKB_GSO_CB(skb)->csum = res;
4994 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4996 return csum_fold(csum_partial(csum_start, plen, partial));
4999 static inline bool skb_is_gso(const struct sk_buff *skb)
5001 return skb_shinfo(skb)->gso_size;
5004 /* Note: Should be called only if skb_is_gso(skb) is true */
5005 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
5007 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
5010 /* Note: Should be called only if skb_is_gso(skb) is true */
5011 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
5013 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
5016 /* Note: Should be called only if skb_is_gso(skb) is true */
5017 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
5019 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
5022 static inline void skb_gso_reset(struct sk_buff *skb)
5024 skb_shinfo(skb)->gso_size = 0;
5025 skb_shinfo(skb)->gso_segs = 0;
5026 skb_shinfo(skb)->gso_type = 0;
5029 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
5032 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
5034 shinfo->gso_size += increment;
5037 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
5040 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
5042 shinfo->gso_size -= decrement;
5045 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
5047 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
5049 /* LRO sets gso_size but not gso_type, whereas if GSO is really
5050 * wanted then gso_type will be set. */
5051 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5053 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
5054 unlikely(shinfo->gso_type == 0)) {
5055 __skb_warn_lro_forwarding(skb);
5061 static inline void skb_forward_csum(struct sk_buff *skb)
5063 /* Unfortunately we don't support this one. Any brave souls? */
5064 if (skb->ip_summed == CHECKSUM_COMPLETE)
5065 skb->ip_summed = CHECKSUM_NONE;
5069 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
5070 * @skb: skb to check
5072 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
5073 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
5074 * use this helper, to document places where we make this assertion.
5076 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
5078 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE);
5081 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
5083 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
5084 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5085 unsigned int transport_len,
5086 __sum16(*skb_chkf)(struct sk_buff *skb));
5089 * skb_head_is_locked - Determine if the skb->head is locked down
5090 * @skb: skb to check
5092 * The head on skbs build around a head frag can be removed if they are
5093 * not cloned. This function returns true if the skb head is locked down
5094 * due to either being allocated via kmalloc, or by being a clone with
5095 * multiple references to the head.
5097 static inline bool skb_head_is_locked(const struct sk_buff *skb)
5099 return !skb->head_frag || skb_cloned(skb);
5102 /* Local Checksum Offload.
5103 * Compute outer checksum based on the assumption that the
5104 * inner checksum will be offloaded later.
5105 * See Documentation/networking/checksum-offloads.rst for
5106 * explanation of how this works.
5107 * Fill in outer checksum adjustment (e.g. with sum of outer
5108 * pseudo-header) before calling.
5109 * Also ensure that inner checksum is in linear data area.
5111 static inline __wsum lco_csum(struct sk_buff *skb)
5113 unsigned char *csum_start = skb_checksum_start(skb);
5114 unsigned char *l4_hdr = skb_transport_header(skb);
5117 /* Start with complement of inner checksum adjustment */
5118 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
5121 /* Add in checksum of our headers (incl. outer checksum
5122 * adjustment filled in by caller) and return result.
5124 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
5127 static inline bool skb_is_redirected(const struct sk_buff *skb)
5129 return skb->redirected;
5132 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
5134 skb->redirected = 1;
5135 #ifdef CONFIG_NET_REDIRECT
5136 skb->from_ingress = from_ingress;
5137 if (skb->from_ingress)
5138 skb_clear_tstamp(skb);
5142 static inline void skb_reset_redirect(struct sk_buff *skb)
5144 skb->redirected = 0;
5147 static inline bool skb_csum_is_sctp(struct sk_buff *skb)
5149 return skb->csum_not_inet;
5152 static inline void skb_set_kcov_handle(struct sk_buff *skb,
5153 const u64 kcov_handle)
5156 skb->kcov_handle = kcov_handle;
5160 static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
5163 return skb->kcov_handle;
5169 #ifdef CONFIG_PAGE_POOL
5170 static inline void skb_mark_for_recycle(struct sk_buff *skb)
5172 skb->pp_recycle = 1;
5176 static inline bool skb_pp_recycle(struct sk_buff *skb, void *data)
5178 if (!IS_ENABLED(CONFIG_PAGE_POOL) || !skb->pp_recycle)
5180 return page_pool_return_skb_page(virt_to_page(data));
5183 #endif /* __KERNEL__ */
5184 #endif /* _LINUX_SKBUFF_H */