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),
691 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
692 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY)
695 * The callback notifies userspace to release buffers when skb DMA is done in
696 * lower device, the skb last reference should be 0 when calling this.
697 * The zerocopy_success argument is true if zero copy transmit occurred,
698 * false on data copy or out of memory error caused by data copy attempt.
699 * The ctx field is used to track device context.
700 * The desc field is used to track userspace buffer index.
703 void (*callback)(struct sk_buff *, struct ubuf_info *,
704 bool zerocopy_success);
721 struct user_struct *user;
726 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
728 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
729 void mm_unaccount_pinned_pages(struct mmpin *mmp);
731 /* This data is invariant across clones and lives at
732 * the end of the header data, ie. at skb->end.
734 struct skb_shared_info {
739 unsigned short gso_size;
740 /* Warning: this field is not always filled in (UFO)! */
741 unsigned short gso_segs;
742 struct sk_buff *frag_list;
743 struct skb_shared_hwtstamps hwtstamps;
744 unsigned int gso_type;
748 * Warning : all fields before dataref are cleared in __alloc_skb()
751 unsigned int xdp_frags_size;
753 /* Intermediate layers must ensure that destructor_arg
754 * remains valid until skb destructor */
755 void * destructor_arg;
757 /* must be last field, see pskb_expand_head() */
758 skb_frag_t frags[MAX_SKB_FRAGS];
762 * DOC: dataref and headerless skbs
764 * Transport layers send out clones of payload skbs they hold for
765 * retransmissions. To allow lower layers of the stack to prepend their headers
766 * we split &skb_shared_info.dataref into two halves.
767 * The lower 16 bits count the overall number of references.
768 * The higher 16 bits indicate how many of the references are payload-only.
769 * skb_header_cloned() checks if skb is allowed to add / write the headers.
771 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
772 * (via __skb_header_release()). Any clone created from marked skb will get
773 * &sk_buff.hdr_len populated with the available headroom.
774 * If there's the only clone in existence it's able to modify the headroom
775 * at will. The sequence of calls inside the transport layer is::
779 * __skb_header_release()
781 * // send the clone down the stack
783 * This is not a very generic construct and it depends on the transport layers
784 * doing the right thing. In practice there's usually only one payload-only skb.
785 * Having multiple payload-only skbs with different lengths of hdr_len is not
786 * possible. The payload-only skbs should never leave their owner.
788 #define SKB_DATAREF_SHIFT 16
789 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
793 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
794 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
795 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
799 SKB_GSO_TCPV4 = 1 << 0,
801 /* This indicates the skb is from an untrusted source. */
802 SKB_GSO_DODGY = 1 << 1,
804 /* This indicates the tcp segment has CWR set. */
805 SKB_GSO_TCP_ECN = 1 << 2,
807 SKB_GSO_TCP_FIXEDID = 1 << 3,
809 SKB_GSO_TCPV6 = 1 << 4,
811 SKB_GSO_FCOE = 1 << 5,
813 SKB_GSO_GRE = 1 << 6,
815 SKB_GSO_GRE_CSUM = 1 << 7,
817 SKB_GSO_IPXIP4 = 1 << 8,
819 SKB_GSO_IPXIP6 = 1 << 9,
821 SKB_GSO_UDP_TUNNEL = 1 << 10,
823 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
825 SKB_GSO_PARTIAL = 1 << 12,
827 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
829 SKB_GSO_SCTP = 1 << 14,
831 SKB_GSO_ESP = 1 << 15,
833 SKB_GSO_UDP = 1 << 16,
835 SKB_GSO_UDP_L4 = 1 << 17,
837 SKB_GSO_FRAGLIST = 1 << 18,
840 #if BITS_PER_LONG > 32
841 #define NET_SKBUFF_DATA_USES_OFFSET 1
844 #ifdef NET_SKBUFF_DATA_USES_OFFSET
845 typedef unsigned int sk_buff_data_t;
847 typedef unsigned char *sk_buff_data_t;
851 * DOC: Basic sk_buff geometry
853 * struct sk_buff itself is a metadata structure and does not hold any packet
854 * data. All the data is held in associated buffers.
856 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
859 * - data buffer, containing headers and sometimes payload;
860 * this is the part of the skb operated on by the common helpers
861 * such as skb_put() or skb_pull();
862 * - shared info (struct skb_shared_info) which holds an array of pointers
863 * to read-only data in the (page, offset, length) format.
865 * Optionally &skb_shared_info.frag_list may point to another skb.
867 * Basic diagram may look like this::
872 * ,--------------------------- + head
873 * / ,----------------- + data
874 * / / ,----------- + tail
878 * -----------------------------------------------
879 * | headroom | data | tailroom | skb_shared_info |
880 * -----------------------------------------------
884 * + [page frag] ---------
885 * + frag_list --> | sk_buff |
891 * struct sk_buff - socket buffer
892 * @next: Next buffer in list
893 * @prev: Previous buffer in list
894 * @tstamp: Time we arrived/left
895 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
896 * for retransmit timer
897 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
899 * @ll_node: anchor in an llist (eg socket defer_list)
900 * @sk: Socket we are owned by
901 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
902 * fragmentation management
903 * @dev: Device we arrived on/are leaving by
904 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
905 * @cb: Control buffer. Free for use by every layer. Put private vars here
906 * @_skb_refdst: destination entry (with norefcount bit)
907 * @sp: the security path, used for xfrm
908 * @len: Length of actual data
909 * @data_len: Data length
910 * @mac_len: Length of link layer header
911 * @hdr_len: writable header length of cloned skb
912 * @csum: Checksum (must include start/offset pair)
913 * @csum_start: Offset from skb->head where checksumming should start
914 * @csum_offset: Offset from csum_start where checksum should be stored
915 * @priority: Packet queueing priority
916 * @ignore_df: allow local fragmentation
917 * @cloned: Head may be cloned (check refcnt to be sure)
918 * @ip_summed: Driver fed us an IP checksum
919 * @nohdr: Payload reference only, must not modify header
920 * @pkt_type: Packet class
921 * @fclone: skbuff clone status
922 * @ipvs_property: skbuff is owned by ipvs
923 * @inner_protocol_type: whether the inner protocol is
924 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
925 * @remcsum_offload: remote checksum offload is enabled
926 * @offload_fwd_mark: Packet was L2-forwarded in hardware
927 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
928 * @tc_skip_classify: do not classify packet. set by IFB device
929 * @tc_at_ingress: used within tc_classify to distinguish in/egress
930 * @redirected: packet was redirected by packet classifier
931 * @from_ingress: packet was redirected from the ingress path
932 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
933 * @peeked: this packet has been seen already, so stats have been
934 * done for it, don't do them again
935 * @nf_trace: netfilter packet trace flag
936 * @protocol: Packet protocol from driver
937 * @destructor: Destruct function
938 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
939 * @_sk_redir: socket redirection information for skmsg
940 * @_nfct: Associated connection, if any (with nfctinfo bits)
941 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
942 * @skb_iif: ifindex of device we arrived on
943 * @tc_index: Traffic control index
944 * @hash: the packet hash
945 * @queue_mapping: Queue mapping for multiqueue devices
946 * @head_frag: skb was allocated from page fragments,
947 * not allocated by kmalloc() or vmalloc().
948 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
949 * @pp_recycle: mark the packet for recycling instead of freeing (implies
950 * page_pool support on driver)
951 * @active_extensions: active extensions (skb_ext_id types)
952 * @ndisc_nodetype: router type (from link layer)
953 * @ooo_okay: allow the mapping of a socket to a queue to be changed
954 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
956 * @sw_hash: indicates hash was computed in software stack
957 * @wifi_acked_valid: wifi_acked was set
958 * @wifi_acked: whether frame was acked on wifi or not
959 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
960 * @encapsulation: indicates the inner headers in the skbuff are valid
961 * @encap_hdr_csum: software checksum is needed
962 * @csum_valid: checksum is already valid
963 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
964 * @csum_complete_sw: checksum was completed by software
965 * @csum_level: indicates the number of consecutive checksums found in
966 * the packet minus one that have been verified as
967 * CHECKSUM_UNNECESSARY (max 3)
968 * @dst_pending_confirm: need to confirm neighbour
969 * @decrypted: Decrypted SKB
970 * @slow_gro: state present at GRO time, slower prepare step required
971 * @mono_delivery_time: When set, skb->tstamp has the
972 * delivery_time in mono clock base (i.e. EDT). Otherwise, the
973 * skb->tstamp has the (rcv) timestamp at ingress and
974 * delivery_time at egress.
975 * @napi_id: id of the NAPI struct this skb came from
976 * @sender_cpu: (aka @napi_id) source CPU in XPS
977 * @alloc_cpu: CPU which did the skb allocation.
978 * @secmark: security marking
979 * @mark: Generic packet mark
980 * @reserved_tailroom: (aka @mark) number of bytes of free space available
981 * at the tail of an sk_buff
982 * @vlan_present: VLAN tag is present
983 * @vlan_proto: vlan encapsulation protocol
984 * @vlan_tci: vlan tag control information
985 * @inner_protocol: Protocol (encapsulation)
986 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
987 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
988 * @inner_transport_header: Inner transport layer header (encapsulation)
989 * @inner_network_header: Network layer header (encapsulation)
990 * @inner_mac_header: Link layer header (encapsulation)
991 * @transport_header: Transport layer header
992 * @network_header: Network layer header
993 * @mac_header: Link layer header
994 * @kcov_handle: KCOV remote handle for remote coverage collection
995 * @tail: Tail pointer
997 * @head: Head of buffer
998 * @data: Data head pointer
999 * @truesize: Buffer size
1000 * @users: User count - see {datagram,tcp}.c
1001 * @extensions: allocated extensions, valid if active_extensions is nonzero
1007 /* These two members must be first to match sk_buff_head. */
1008 struct sk_buff *next;
1009 struct sk_buff *prev;
1012 struct net_device *dev;
1013 /* Some protocols might use this space to store information,
1014 * while device pointer would be NULL.
1015 * UDP receive path is one user.
1017 unsigned long dev_scratch;
1020 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
1021 struct list_head list;
1022 struct llist_node ll_node;
1027 int ip_defrag_offset;
1032 u64 skb_mstamp_ns; /* earliest departure time */
1035 * This is the control buffer. It is free to use for every
1036 * layer. Please put your private variables there. If you
1037 * want to keep them across layers you have to do a skb_clone()
1038 * first. This is owned by whoever has the skb queued ATM.
1040 char cb[48] __aligned(8);
1044 unsigned long _skb_refdst;
1045 void (*destructor)(struct sk_buff *skb);
1047 struct list_head tcp_tsorted_anchor;
1048 #ifdef CONFIG_NET_SOCK_MSG
1049 unsigned long _sk_redir;
1053 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1054 unsigned long _nfct;
1061 /* Following fields are _not_ copied in __copy_skb_header()
1062 * Note that queue_mapping is here mostly to fill a hole.
1064 __u16 queue_mapping;
1066 /* if you move cloned around you also must adapt those constants */
1067 #ifdef __BIG_ENDIAN_BITFIELD
1068 #define CLONED_MASK (1 << 7)
1070 #define CLONED_MASK 1
1072 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
1075 __u8 __cloned_offset[0];
1083 pp_recycle:1; /* page_pool recycle indicator */
1084 #ifdef CONFIG_SKB_EXTENSIONS
1085 __u8 active_extensions;
1088 /* Fields enclosed in headers group are copied
1089 * using a single memcpy() in __copy_skb_header()
1091 struct_group(headers,
1094 __u8 __pkt_type_offset[0];
1096 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
1104 __u8 wifi_acked_valid:1;
1107 /* Indicates the inner headers are valid in the skbuff. */
1108 __u8 encapsulation:1;
1109 __u8 encap_hdr_csum:1;
1113 __u8 __pkt_vlan_present_offset[0];
1115 __u8 vlan_present:1; /* See PKT_VLAN_PRESENT_BIT */
1116 __u8 csum_complete_sw:1;
1118 __u8 dst_pending_confirm:1;
1119 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */
1120 #ifdef CONFIG_NET_CLS_ACT
1121 __u8 tc_skip_classify:1;
1122 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
1124 #ifdef CONFIG_IPV6_NDISC_NODETYPE
1125 __u8 ndisc_nodetype:2;
1128 __u8 ipvs_property:1;
1129 __u8 inner_protocol_type:1;
1130 __u8 remcsum_offload:1;
1131 #ifdef CONFIG_NET_SWITCHDEV
1132 __u8 offload_fwd_mark:1;
1133 __u8 offload_l3_fwd_mark:1;
1136 #ifdef CONFIG_NET_REDIRECT
1137 __u8 from_ingress:1;
1139 #ifdef CONFIG_NETFILTER_SKIP_EGRESS
1140 __u8 nf_skip_egress:1;
1142 #ifdef CONFIG_TLS_DEVICE
1146 __u8 csum_not_inet:1;
1148 #ifdef CONFIG_NET_SCHED
1149 __u16 tc_index; /* traffic control index */
1164 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
1166 unsigned int napi_id;
1167 unsigned int sender_cpu;
1171 #ifdef CONFIG_NETWORK_SECMARK
1177 __u32 reserved_tailroom;
1181 __be16 inner_protocol;
1185 __u16 inner_transport_header;
1186 __u16 inner_network_header;
1187 __u16 inner_mac_header;
1190 __u16 transport_header;
1191 __u16 network_header;
1198 ); /* end headers group */
1200 /* These elements must be at the end, see alloc_skb() for details. */
1201 sk_buff_data_t tail;
1203 unsigned char *head,
1205 unsigned int truesize;
1208 #ifdef CONFIG_SKB_EXTENSIONS
1209 /* only useable after checking ->active_extensions != 0 */
1210 struct skb_ext *extensions;
1214 /* if you move pkt_type around you also must adapt those constants */
1215 #ifdef __BIG_ENDIAN_BITFIELD
1216 #define PKT_TYPE_MAX (7 << 5)
1218 #define PKT_TYPE_MAX 7
1220 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1222 /* if you move pkt_vlan_present, tc_at_ingress, or mono_delivery_time
1223 * around, you also must adapt these constants.
1225 #ifdef __BIG_ENDIAN_BITFIELD
1226 #define PKT_VLAN_PRESENT_BIT 7
1227 #define TC_AT_INGRESS_MASK (1 << 0)
1228 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 2)
1230 #define PKT_VLAN_PRESENT_BIT 0
1231 #define TC_AT_INGRESS_MASK (1 << 7)
1232 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 5)
1234 #define PKT_VLAN_PRESENT_OFFSET offsetof(struct sk_buff, __pkt_vlan_present_offset)
1238 * Handling routines are only of interest to the kernel
1241 #define SKB_ALLOC_FCLONE 0x01
1242 #define SKB_ALLOC_RX 0x02
1243 #define SKB_ALLOC_NAPI 0x04
1246 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1249 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1251 return unlikely(skb->pfmemalloc);
1255 * skb might have a dst pointer attached, refcounted or not.
1256 * _skb_refdst low order bit is set if refcount was _not_ taken
1258 #define SKB_DST_NOREF 1UL
1259 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1262 * skb_dst - returns skb dst_entry
1265 * Returns skb dst_entry, regardless of reference taken or not.
1267 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1269 /* If refdst was not refcounted, check we still are in a
1270 * rcu_read_lock section
1272 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1273 !rcu_read_lock_held() &&
1274 !rcu_read_lock_bh_held());
1275 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1279 * skb_dst_set - sets skb dst
1283 * Sets skb dst, assuming a reference was taken on dst and should
1284 * be released by skb_dst_drop()
1286 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1288 skb->slow_gro |= !!dst;
1289 skb->_skb_refdst = (unsigned long)dst;
1293 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1297 * Sets skb dst, assuming a reference was not taken on dst.
1298 * If dst entry is cached, we do not take reference and dst_release
1299 * will be avoided by refdst_drop. If dst entry is not cached, we take
1300 * reference, so that last dst_release can destroy the dst immediately.
1302 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1304 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1305 skb->slow_gro |= !!dst;
1306 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1310 * skb_dst_is_noref - Test if skb dst isn't refcounted
1313 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1315 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1319 * skb_rtable - Returns the skb &rtable
1322 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1324 return (struct rtable *)skb_dst(skb);
1327 /* For mangling skb->pkt_type from user space side from applications
1328 * such as nft, tc, etc, we only allow a conservative subset of
1329 * possible pkt_types to be set.
1331 static inline bool skb_pkt_type_ok(u32 ptype)
1333 return ptype <= PACKET_OTHERHOST;
1337 * skb_napi_id - Returns the skb's NAPI id
1340 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1342 #ifdef CONFIG_NET_RX_BUSY_POLL
1343 return skb->napi_id;
1350 * skb_unref - decrement the skb's reference count
1353 * Returns true if we can free the skb.
1355 static inline bool skb_unref(struct sk_buff *skb)
1359 if (likely(refcount_read(&skb->users) == 1))
1361 else if (likely(!refcount_dec_and_test(&skb->users)))
1367 void kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason);
1370 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1371 * @skb: buffer to free
1373 static inline void kfree_skb(struct sk_buff *skb)
1375 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1378 void skb_release_head_state(struct sk_buff *skb);
1379 void kfree_skb_list_reason(struct sk_buff *segs,
1380 enum skb_drop_reason reason);
1381 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1382 void skb_tx_error(struct sk_buff *skb);
1384 static inline void kfree_skb_list(struct sk_buff *segs)
1386 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1389 #ifdef CONFIG_TRACEPOINTS
1390 void consume_skb(struct sk_buff *skb);
1392 static inline void consume_skb(struct sk_buff *skb)
1394 return kfree_skb(skb);
1398 void __consume_stateless_skb(struct sk_buff *skb);
1399 void __kfree_skb(struct sk_buff *skb);
1400 extern struct kmem_cache *skbuff_head_cache;
1402 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1403 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1404 bool *fragstolen, int *delta_truesize);
1406 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1408 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1409 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1410 struct sk_buff *build_skb_around(struct sk_buff *skb,
1411 void *data, unsigned int frag_size);
1412 void skb_attempt_defer_free(struct sk_buff *skb);
1414 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1417 * alloc_skb - allocate a network buffer
1418 * @size: size to allocate
1419 * @priority: allocation mask
1421 * This function is a convenient wrapper around __alloc_skb().
1423 static inline struct sk_buff *alloc_skb(unsigned int size,
1426 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1429 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1430 unsigned long data_len,
1434 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1436 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1437 struct sk_buff_fclones {
1438 struct sk_buff skb1;
1440 struct sk_buff skb2;
1442 refcount_t fclone_ref;
1446 * skb_fclone_busy - check if fclone is busy
1450 * Returns true if skb is a fast clone, and its clone is not freed.
1451 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1452 * so we also check that this didnt happen.
1454 static inline bool skb_fclone_busy(const struct sock *sk,
1455 const struct sk_buff *skb)
1457 const struct sk_buff_fclones *fclones;
1459 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1461 return skb->fclone == SKB_FCLONE_ORIG &&
1462 refcount_read(&fclones->fclone_ref) > 1 &&
1463 READ_ONCE(fclones->skb2.sk) == sk;
1467 * alloc_skb_fclone - allocate a network buffer from fclone cache
1468 * @size: size to allocate
1469 * @priority: allocation mask
1471 * This function is a convenient wrapper around __alloc_skb().
1473 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1476 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1479 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1480 void skb_headers_offset_update(struct sk_buff *skb, int off);
1481 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1482 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1483 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1484 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1485 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1486 gfp_t gfp_mask, bool fclone);
1487 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1490 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1493 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1494 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1495 unsigned int headroom);
1496 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1497 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1498 int newtailroom, gfp_t priority);
1499 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1500 int offset, int len);
1501 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1502 int offset, int len);
1503 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1504 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1507 * skb_pad - zero pad the tail of an skb
1508 * @skb: buffer to pad
1509 * @pad: space to pad
1511 * Ensure that a buffer is followed by a padding area that is zero
1512 * filled. Used by network drivers which may DMA or transfer data
1513 * beyond the buffer end onto the wire.
1515 * May return error in out of memory cases. The skb is freed on error.
1517 static inline int skb_pad(struct sk_buff *skb, int pad)
1519 return __skb_pad(skb, pad, true);
1521 #define dev_kfree_skb(a) consume_skb(a)
1523 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1524 int offset, size_t size);
1526 struct skb_seq_state {
1530 __u32 stepped_offset;
1531 struct sk_buff *root_skb;
1532 struct sk_buff *cur_skb;
1537 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1538 unsigned int to, struct skb_seq_state *st);
1539 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1540 struct skb_seq_state *st);
1541 void skb_abort_seq_read(struct skb_seq_state *st);
1543 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1544 unsigned int to, struct ts_config *config);
1547 * Packet hash types specify the type of hash in skb_set_hash.
1549 * Hash types refer to the protocol layer addresses which are used to
1550 * construct a packet's hash. The hashes are used to differentiate or identify
1551 * flows of the protocol layer for the hash type. Hash types are either
1552 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1554 * Properties of hashes:
1556 * 1) Two packets in different flows have different hash values
1557 * 2) Two packets in the same flow should have the same hash value
1559 * A hash at a higher layer is considered to be more specific. A driver should
1560 * set the most specific hash possible.
1562 * A driver cannot indicate a more specific hash than the layer at which a hash
1563 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1565 * A driver may indicate a hash level which is less specific than the
1566 * actual layer the hash was computed on. For instance, a hash computed
1567 * at L4 may be considered an L3 hash. This should only be done if the
1568 * driver can't unambiguously determine that the HW computed the hash at
1569 * the higher layer. Note that the "should" in the second property above
1572 enum pkt_hash_types {
1573 PKT_HASH_TYPE_NONE, /* Undefined type */
1574 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1575 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1576 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1579 static inline void skb_clear_hash(struct sk_buff *skb)
1586 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1589 skb_clear_hash(skb);
1593 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1595 skb->l4_hash = is_l4;
1596 skb->sw_hash = is_sw;
1601 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1603 /* Used by drivers to set hash from HW */
1604 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1608 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1610 __skb_set_hash(skb, hash, true, is_l4);
1613 void __skb_get_hash(struct sk_buff *skb);
1614 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1615 u32 skb_get_poff(const struct sk_buff *skb);
1616 u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1617 const struct flow_keys_basic *keys, int hlen);
1618 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1619 const void *data, int hlen_proto);
1621 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1622 int thoff, u8 ip_proto)
1624 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1627 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1628 const struct flow_dissector_key *key,
1629 unsigned int key_count);
1631 struct bpf_flow_dissector;
1632 bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1633 __be16 proto, int nhoff, int hlen, unsigned int flags);
1635 bool __skb_flow_dissect(const struct net *net,
1636 const struct sk_buff *skb,
1637 struct flow_dissector *flow_dissector,
1638 void *target_container, const void *data,
1639 __be16 proto, int nhoff, int hlen, unsigned int flags);
1641 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1642 struct flow_dissector *flow_dissector,
1643 void *target_container, unsigned int flags)
1645 return __skb_flow_dissect(NULL, skb, flow_dissector,
1646 target_container, NULL, 0, 0, 0, flags);
1649 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1650 struct flow_keys *flow,
1653 memset(flow, 0, sizeof(*flow));
1654 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1655 flow, NULL, 0, 0, 0, flags);
1659 skb_flow_dissect_flow_keys_basic(const struct net *net,
1660 const struct sk_buff *skb,
1661 struct flow_keys_basic *flow,
1662 const void *data, __be16 proto,
1663 int nhoff, int hlen, unsigned int flags)
1665 memset(flow, 0, sizeof(*flow));
1666 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1667 data, proto, nhoff, hlen, flags);
1670 void skb_flow_dissect_meta(const struct sk_buff *skb,
1671 struct flow_dissector *flow_dissector,
1672 void *target_container);
1674 /* Gets a skb connection tracking info, ctinfo map should be a
1675 * map of mapsize to translate enum ip_conntrack_info states
1679 skb_flow_dissect_ct(const struct sk_buff *skb,
1680 struct flow_dissector *flow_dissector,
1681 void *target_container,
1682 u16 *ctinfo_map, size_t mapsize,
1683 bool post_ct, u16 zone);
1685 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1686 struct flow_dissector *flow_dissector,
1687 void *target_container);
1689 void skb_flow_dissect_hash(const struct sk_buff *skb,
1690 struct flow_dissector *flow_dissector,
1691 void *target_container);
1693 static inline __u32 skb_get_hash(struct sk_buff *skb)
1695 if (!skb->l4_hash && !skb->sw_hash)
1696 __skb_get_hash(skb);
1701 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1703 if (!skb->l4_hash && !skb->sw_hash) {
1704 struct flow_keys keys;
1705 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1707 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1713 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1714 const siphash_key_t *perturb);
1716 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1721 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1723 to->hash = from->hash;
1724 to->sw_hash = from->sw_hash;
1725 to->l4_hash = from->l4_hash;
1728 static inline void skb_copy_decrypted(struct sk_buff *to,
1729 const struct sk_buff *from)
1731 #ifdef CONFIG_TLS_DEVICE
1732 to->decrypted = from->decrypted;
1736 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1737 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1739 return skb->head + skb->end;
1742 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1747 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1752 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1757 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1759 return skb->end - skb->head;
1762 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1764 skb->end = skb->head + offset;
1768 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1769 struct ubuf_info *uarg);
1771 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1773 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1776 int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb,
1777 struct iov_iter *from, size_t length);
1779 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1780 struct msghdr *msg, int len)
1782 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1785 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1786 struct msghdr *msg, int len,
1787 struct ubuf_info *uarg);
1790 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1792 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1794 return &skb_shinfo(skb)->hwtstamps;
1797 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1799 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1801 return is_zcopy ? skb_uarg(skb) : NULL;
1804 static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1806 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1809 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1810 const struct sk_buff *skb2)
1812 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1815 static inline void net_zcopy_get(struct ubuf_info *uarg)
1817 refcount_inc(&uarg->refcnt);
1820 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1822 skb_shinfo(skb)->destructor_arg = uarg;
1823 skb_shinfo(skb)->flags |= uarg->flags;
1826 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1829 if (skb && uarg && !skb_zcopy(skb)) {
1830 if (unlikely(have_ref && *have_ref))
1833 net_zcopy_get(uarg);
1834 skb_zcopy_init(skb, uarg);
1838 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1840 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1841 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1844 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1846 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1849 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1851 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1854 static inline void net_zcopy_put(struct ubuf_info *uarg)
1857 uarg->callback(NULL, uarg, true);
1860 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1863 if (uarg->callback == msg_zerocopy_callback)
1864 msg_zerocopy_put_abort(uarg, have_uref);
1866 net_zcopy_put(uarg);
1870 /* Release a reference on a zerocopy structure */
1871 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1873 struct ubuf_info *uarg = skb_zcopy(skb);
1876 if (!skb_zcopy_is_nouarg(skb))
1877 uarg->callback(skb, uarg, zerocopy_success);
1879 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1883 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1888 /* Iterate through singly-linked GSO fragments of an skb. */
1889 #define skb_list_walk_safe(first, skb, next_skb) \
1890 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1891 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1893 static inline void skb_list_del_init(struct sk_buff *skb)
1895 __list_del_entry(&skb->list);
1896 skb_mark_not_on_list(skb);
1900 * skb_queue_empty - check if a queue is empty
1903 * Returns true if the queue is empty, false otherwise.
1905 static inline int skb_queue_empty(const struct sk_buff_head *list)
1907 return list->next == (const struct sk_buff *) list;
1911 * skb_queue_empty_lockless - check if a queue is empty
1914 * Returns true if the queue is empty, false otherwise.
1915 * This variant can be used in lockless contexts.
1917 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1919 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1924 * skb_queue_is_last - check if skb is the last entry in the queue
1928 * Returns true if @skb is the last buffer on the list.
1930 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1931 const struct sk_buff *skb)
1933 return skb->next == (const struct sk_buff *) list;
1937 * skb_queue_is_first - check if skb is the first entry in the queue
1941 * Returns true if @skb is the first buffer on the list.
1943 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1944 const struct sk_buff *skb)
1946 return skb->prev == (const struct sk_buff *) list;
1950 * skb_queue_next - return the next packet in the queue
1952 * @skb: current buffer
1954 * Return the next packet in @list after @skb. It is only valid to
1955 * call this if skb_queue_is_last() evaluates to false.
1957 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1958 const struct sk_buff *skb)
1960 /* This BUG_ON may seem severe, but if we just return then we
1961 * are going to dereference garbage.
1963 BUG_ON(skb_queue_is_last(list, skb));
1968 * skb_queue_prev - return the prev packet in the queue
1970 * @skb: current buffer
1972 * Return the prev packet in @list before @skb. It is only valid to
1973 * call this if skb_queue_is_first() evaluates to false.
1975 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1976 const struct sk_buff *skb)
1978 /* This BUG_ON may seem severe, but if we just return then we
1979 * are going to dereference garbage.
1981 BUG_ON(skb_queue_is_first(list, skb));
1986 * skb_get - reference buffer
1987 * @skb: buffer to reference
1989 * Makes another reference to a socket buffer and returns a pointer
1992 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1994 refcount_inc(&skb->users);
1999 * If users == 1, we are the only owner and can avoid redundant atomic changes.
2003 * skb_cloned - is the buffer a clone
2004 * @skb: buffer to check
2006 * Returns true if the buffer was generated with skb_clone() and is
2007 * one of multiple shared copies of the buffer. Cloned buffers are
2008 * shared data so must not be written to under normal circumstances.
2010 static inline int skb_cloned(const struct sk_buff *skb)
2012 return skb->cloned &&
2013 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
2016 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
2018 might_sleep_if(gfpflags_allow_blocking(pri));
2020 if (skb_cloned(skb))
2021 return pskb_expand_head(skb, 0, 0, pri);
2026 /* This variant of skb_unclone() makes sure skb->truesize
2027 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
2029 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
2030 * when various debugging features are in place.
2032 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
2033 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
2035 might_sleep_if(gfpflags_allow_blocking(pri));
2037 if (skb_cloned(skb))
2038 return __skb_unclone_keeptruesize(skb, pri);
2043 * skb_header_cloned - is the header a clone
2044 * @skb: buffer to check
2046 * Returns true if modifying the header part of the buffer requires
2047 * the data to be copied.
2049 static inline int skb_header_cloned(const struct sk_buff *skb)
2056 dataref = atomic_read(&skb_shinfo(skb)->dataref);
2057 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
2058 return dataref != 1;
2061 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
2063 might_sleep_if(gfpflags_allow_blocking(pri));
2065 if (skb_header_cloned(skb))
2066 return pskb_expand_head(skb, 0, 0, pri);
2072 * __skb_header_release() - allow clones to use the headroom
2073 * @skb: buffer to operate on
2075 * See "DOC: dataref and headerless skbs".
2077 static inline void __skb_header_release(struct sk_buff *skb)
2080 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
2085 * skb_shared - is the buffer shared
2086 * @skb: buffer to check
2088 * Returns true if more than one person has a reference to this
2091 static inline int skb_shared(const struct sk_buff *skb)
2093 return refcount_read(&skb->users) != 1;
2097 * skb_share_check - check if buffer is shared and if so clone it
2098 * @skb: buffer to check
2099 * @pri: priority for memory allocation
2101 * If the buffer is shared the buffer is cloned and the old copy
2102 * drops a reference. A new clone with a single reference is returned.
2103 * If the buffer is not shared the original buffer is returned. When
2104 * being called from interrupt status or with spinlocks held pri must
2107 * NULL is returned on a memory allocation failure.
2109 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
2111 might_sleep_if(gfpflags_allow_blocking(pri));
2112 if (skb_shared(skb)) {
2113 struct sk_buff *nskb = skb_clone(skb, pri);
2125 * Copy shared buffers into a new sk_buff. We effectively do COW on
2126 * packets to handle cases where we have a local reader and forward
2127 * and a couple of other messy ones. The normal one is tcpdumping
2128 * a packet thats being forwarded.
2132 * skb_unshare - make a copy of a shared buffer
2133 * @skb: buffer to check
2134 * @pri: priority for memory allocation
2136 * If the socket buffer is a clone then this function creates a new
2137 * copy of the data, drops a reference count on the old copy and returns
2138 * the new copy with the reference count at 1. If the buffer is not a clone
2139 * the original buffer is returned. When called with a spinlock held or
2140 * from interrupt state @pri must be %GFP_ATOMIC
2142 * %NULL is returned on a memory allocation failure.
2144 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
2147 might_sleep_if(gfpflags_allow_blocking(pri));
2148 if (skb_cloned(skb)) {
2149 struct sk_buff *nskb = skb_copy(skb, pri);
2151 /* Free our shared copy */
2162 * skb_peek - peek at the head of an &sk_buff_head
2163 * @list_: list to peek at
2165 * Peek an &sk_buff. Unlike most other operations you _MUST_
2166 * be careful with this one. A peek leaves the buffer on the
2167 * list and someone else may run off with it. You must hold
2168 * the appropriate locks or have a private queue to do this.
2170 * Returns %NULL for an empty list or a pointer to the head element.
2171 * The reference count is not incremented and the reference is therefore
2172 * volatile. Use with caution.
2174 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2176 struct sk_buff *skb = list_->next;
2178 if (skb == (struct sk_buff *)list_)
2184 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2185 * @list_: list to peek at
2187 * Like skb_peek(), but the caller knows that the list is not empty.
2189 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2195 * skb_peek_next - peek skb following the given one from a queue
2196 * @skb: skb to start from
2197 * @list_: list to peek at
2199 * Returns %NULL when the end of the list is met or a pointer to the
2200 * next element. The reference count is not incremented and the
2201 * reference is therefore volatile. Use with caution.
2203 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2204 const struct sk_buff_head *list_)
2206 struct sk_buff *next = skb->next;
2208 if (next == (struct sk_buff *)list_)
2214 * skb_peek_tail - peek at the tail of an &sk_buff_head
2215 * @list_: list to peek at
2217 * Peek an &sk_buff. Unlike most other operations you _MUST_
2218 * be careful with this one. A peek leaves the buffer on the
2219 * list and someone else may run off with it. You must hold
2220 * the appropriate locks or have a private queue to do this.
2222 * Returns %NULL for an empty list or a pointer to the tail element.
2223 * The reference count is not incremented and the reference is therefore
2224 * volatile. Use with caution.
2226 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2228 struct sk_buff *skb = READ_ONCE(list_->prev);
2230 if (skb == (struct sk_buff *)list_)
2237 * skb_queue_len - get queue length
2238 * @list_: list to measure
2240 * Return the length of an &sk_buff queue.
2242 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2248 * skb_queue_len_lockless - get queue length
2249 * @list_: list to measure
2251 * Return the length of an &sk_buff queue.
2252 * This variant can be used in lockless contexts.
2254 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2256 return READ_ONCE(list_->qlen);
2260 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2261 * @list: queue to initialize
2263 * This initializes only the list and queue length aspects of
2264 * an sk_buff_head object. This allows to initialize the list
2265 * aspects of an sk_buff_head without reinitializing things like
2266 * the spinlock. It can also be used for on-stack sk_buff_head
2267 * objects where the spinlock is known to not be used.
2269 static inline void __skb_queue_head_init(struct sk_buff_head *list)
2271 list->prev = list->next = (struct sk_buff *)list;
2276 * This function creates a split out lock class for each invocation;
2277 * this is needed for now since a whole lot of users of the skb-queue
2278 * infrastructure in drivers have different locking usage (in hardirq)
2279 * than the networking core (in softirq only). In the long run either the
2280 * network layer or drivers should need annotation to consolidate the
2281 * main types of usage into 3 classes.
2283 static inline void skb_queue_head_init(struct sk_buff_head *list)
2285 spin_lock_init(&list->lock);
2286 __skb_queue_head_init(list);
2289 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2290 struct lock_class_key *class)
2292 skb_queue_head_init(list);
2293 lockdep_set_class(&list->lock, class);
2297 * Insert an sk_buff on a list.
2299 * The "__skb_xxxx()" functions are the non-atomic ones that
2300 * can only be called with interrupts disabled.
2302 static inline void __skb_insert(struct sk_buff *newsk,
2303 struct sk_buff *prev, struct sk_buff *next,
2304 struct sk_buff_head *list)
2306 /* See skb_queue_empty_lockless() and skb_peek_tail()
2307 * for the opposite READ_ONCE()
2309 WRITE_ONCE(newsk->next, next);
2310 WRITE_ONCE(newsk->prev, prev);
2311 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2312 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2313 WRITE_ONCE(list->qlen, list->qlen + 1);
2316 static inline void __skb_queue_splice(const struct sk_buff_head *list,
2317 struct sk_buff *prev,
2318 struct sk_buff *next)
2320 struct sk_buff *first = list->next;
2321 struct sk_buff *last = list->prev;
2323 WRITE_ONCE(first->prev, prev);
2324 WRITE_ONCE(prev->next, first);
2326 WRITE_ONCE(last->next, next);
2327 WRITE_ONCE(next->prev, last);
2331 * skb_queue_splice - join two skb lists, this is designed for stacks
2332 * @list: the new list to add
2333 * @head: the place to add it in the first list
2335 static inline void skb_queue_splice(const struct sk_buff_head *list,
2336 struct sk_buff_head *head)
2338 if (!skb_queue_empty(list)) {
2339 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2340 head->qlen += list->qlen;
2345 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2346 * @list: the new list to add
2347 * @head: the place to add it in the first list
2349 * The list at @list is reinitialised
2351 static inline void skb_queue_splice_init(struct sk_buff_head *list,
2352 struct sk_buff_head *head)
2354 if (!skb_queue_empty(list)) {
2355 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2356 head->qlen += list->qlen;
2357 __skb_queue_head_init(list);
2362 * skb_queue_splice_tail - join two skb lists, each list being a queue
2363 * @list: the new list to add
2364 * @head: the place to add it in the first list
2366 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2367 struct sk_buff_head *head)
2369 if (!skb_queue_empty(list)) {
2370 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2371 head->qlen += list->qlen;
2376 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2377 * @list: the new list to add
2378 * @head: the place to add it in the first list
2380 * Each of the lists is a queue.
2381 * The list at @list is reinitialised
2383 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2384 struct sk_buff_head *head)
2386 if (!skb_queue_empty(list)) {
2387 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2388 head->qlen += list->qlen;
2389 __skb_queue_head_init(list);
2394 * __skb_queue_after - queue a buffer at the list head
2395 * @list: list to use
2396 * @prev: place after this buffer
2397 * @newsk: buffer to queue
2399 * Queue a buffer int the middle of a list. This function takes no locks
2400 * and you must therefore hold required locks before calling it.
2402 * A buffer cannot be placed on two lists at the same time.
2404 static inline void __skb_queue_after(struct sk_buff_head *list,
2405 struct sk_buff *prev,
2406 struct sk_buff *newsk)
2408 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2411 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2412 struct sk_buff_head *list);
2414 static inline void __skb_queue_before(struct sk_buff_head *list,
2415 struct sk_buff *next,
2416 struct sk_buff *newsk)
2418 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2422 * __skb_queue_head - queue a buffer at the list head
2423 * @list: list to use
2424 * @newsk: buffer to queue
2426 * Queue a buffer at the start of a list. This function takes no locks
2427 * and you must therefore hold required locks before calling it.
2429 * A buffer cannot be placed on two lists at the same time.
2431 static inline void __skb_queue_head(struct sk_buff_head *list,
2432 struct sk_buff *newsk)
2434 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2436 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2439 * __skb_queue_tail - queue a buffer at the list tail
2440 * @list: list to use
2441 * @newsk: buffer to queue
2443 * Queue a buffer at the end of a list. This function takes no locks
2444 * and you must therefore hold required locks before calling it.
2446 * A buffer cannot be placed on two lists at the same time.
2448 static inline void __skb_queue_tail(struct sk_buff_head *list,
2449 struct sk_buff *newsk)
2451 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2453 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2456 * remove sk_buff from list. _Must_ be called atomically, and with
2459 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2460 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2462 struct sk_buff *next, *prev;
2464 WRITE_ONCE(list->qlen, list->qlen - 1);
2467 skb->next = skb->prev = NULL;
2468 WRITE_ONCE(next->prev, prev);
2469 WRITE_ONCE(prev->next, next);
2473 * __skb_dequeue - remove from the head of the queue
2474 * @list: list to dequeue from
2476 * Remove the head of the list. This function does not take any locks
2477 * so must be used with appropriate locks held only. The head item is
2478 * returned or %NULL if the list is empty.
2480 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2482 struct sk_buff *skb = skb_peek(list);
2484 __skb_unlink(skb, list);
2487 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2490 * __skb_dequeue_tail - remove from the tail of the queue
2491 * @list: list to dequeue from
2493 * Remove the tail of the list. This function does not take any locks
2494 * so must be used with appropriate locks held only. The tail item is
2495 * returned or %NULL if the list is empty.
2497 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2499 struct sk_buff *skb = skb_peek_tail(list);
2501 __skb_unlink(skb, list);
2504 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2507 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2509 return skb->data_len;
2512 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2514 return skb->len - skb->data_len;
2517 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2519 unsigned int i, len = 0;
2521 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2522 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2526 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2528 return skb_headlen(skb) + __skb_pagelen(skb);
2532 * __skb_fill_page_desc - initialise a paged fragment in an skb
2533 * @skb: buffer containing fragment to be initialised
2534 * @i: paged fragment index to initialise
2535 * @page: the page to use for this fragment
2536 * @off: the offset to the data with @page
2537 * @size: the length of the data
2539 * Initialises the @i'th fragment of @skb to point to &size bytes at
2540 * offset @off within @page.
2542 * Does not take any additional reference on the fragment.
2544 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2545 struct page *page, int off, int size)
2547 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2550 * Propagate page pfmemalloc to the skb if we can. The problem is
2551 * that not all callers have unique ownership of the page but rely
2552 * on page_is_pfmemalloc doing the right thing(tm).
2554 frag->bv_page = page;
2555 frag->bv_offset = off;
2556 skb_frag_size_set(frag, size);
2558 page = compound_head(page);
2559 if (page_is_pfmemalloc(page))
2560 skb->pfmemalloc = true;
2564 * skb_fill_page_desc - initialise a paged fragment in an skb
2565 * @skb: buffer containing fragment to be initialised
2566 * @i: paged fragment index to initialise
2567 * @page: the page to use for this fragment
2568 * @off: the offset to the data with @page
2569 * @size: the length of the data
2571 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2572 * @skb to point to @size bytes at offset @off within @page. In
2573 * addition updates @skb such that @i is the last fragment.
2575 * Does not take any additional reference on the fragment.
2577 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2578 struct page *page, int off, int size)
2580 __skb_fill_page_desc(skb, i, page, off, size);
2581 skb_shinfo(skb)->nr_frags = i + 1;
2584 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2585 int size, unsigned int truesize);
2587 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2588 unsigned int truesize);
2590 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2592 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2593 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2595 return skb->head + skb->tail;
2598 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2600 skb->tail = skb->data - skb->head;
2603 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2605 skb_reset_tail_pointer(skb);
2606 skb->tail += offset;
2609 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2610 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2615 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2617 skb->tail = skb->data;
2620 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2622 skb->tail = skb->data + offset;
2625 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2628 * Add data to an sk_buff
2630 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2631 void *skb_put(struct sk_buff *skb, unsigned int len);
2632 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2634 void *tmp = skb_tail_pointer(skb);
2635 SKB_LINEAR_ASSERT(skb);
2641 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2643 void *tmp = __skb_put(skb, len);
2645 memset(tmp, 0, len);
2649 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2652 void *tmp = __skb_put(skb, len);
2654 memcpy(tmp, data, len);
2658 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2660 *(u8 *)__skb_put(skb, 1) = val;
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);
2672 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2675 void *tmp = skb_put(skb, len);
2677 memcpy(tmp, data, len);
2682 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2684 *(u8 *)skb_put(skb, 1) = val;
2687 void *skb_push(struct sk_buff *skb, unsigned int len);
2688 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2695 void *skb_pull(struct sk_buff *skb, unsigned int len);
2696 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2699 if (unlikely(skb->len < skb->data_len)) {
2700 #if defined(CONFIG_DEBUG_NET)
2702 pr_err("__skb_pull(len=%u)\n", len);
2703 skb_dump(KERN_ERR, skb, false);
2707 return skb->data += len;
2710 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2712 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2715 void *skb_pull_data(struct sk_buff *skb, size_t len);
2717 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2719 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2721 if (len > skb_headlen(skb) &&
2722 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2725 return skb->data += len;
2728 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2730 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2733 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2735 if (likely(len <= skb_headlen(skb)))
2737 if (unlikely(len > skb->len))
2739 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2742 void skb_condense(struct sk_buff *skb);
2745 * skb_headroom - bytes at buffer head
2746 * @skb: buffer to check
2748 * Return the number of bytes of free space at the head of an &sk_buff.
2750 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2752 return skb->data - skb->head;
2756 * skb_tailroom - bytes at buffer end
2757 * @skb: buffer to check
2759 * Return the number of bytes of free space at the tail of an sk_buff
2761 static inline int skb_tailroom(const struct sk_buff *skb)
2763 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2767 * skb_availroom - bytes at buffer end
2768 * @skb: buffer to check
2770 * Return the number of bytes of free space at the tail of an sk_buff
2771 * allocated by sk_stream_alloc()
2773 static inline int skb_availroom(const struct sk_buff *skb)
2775 if (skb_is_nonlinear(skb))
2778 return skb->end - skb->tail - skb->reserved_tailroom;
2782 * skb_reserve - adjust headroom
2783 * @skb: buffer to alter
2784 * @len: bytes to move
2786 * Increase the headroom of an empty &sk_buff by reducing the tail
2787 * room. This is only allowed for an empty buffer.
2789 static inline void skb_reserve(struct sk_buff *skb, int len)
2796 * skb_tailroom_reserve - adjust reserved_tailroom
2797 * @skb: buffer to alter
2798 * @mtu: maximum amount of headlen permitted
2799 * @needed_tailroom: minimum amount of reserved_tailroom
2801 * Set reserved_tailroom so that headlen can be as large as possible but
2802 * not larger than mtu and tailroom cannot be smaller than
2804 * The required headroom should already have been reserved before using
2807 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2808 unsigned int needed_tailroom)
2810 SKB_LINEAR_ASSERT(skb);
2811 if (mtu < skb_tailroom(skb) - needed_tailroom)
2812 /* use at most mtu */
2813 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2815 /* use up to all available space */
2816 skb->reserved_tailroom = needed_tailroom;
2819 #define ENCAP_TYPE_ETHER 0
2820 #define ENCAP_TYPE_IPPROTO 1
2822 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2825 skb->inner_protocol = protocol;
2826 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2829 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2832 skb->inner_ipproto = ipproto;
2833 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2836 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2838 skb->inner_mac_header = skb->mac_header;
2839 skb->inner_network_header = skb->network_header;
2840 skb->inner_transport_header = skb->transport_header;
2843 static inline void skb_reset_mac_len(struct sk_buff *skb)
2845 skb->mac_len = skb->network_header - skb->mac_header;
2848 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2851 return skb->head + skb->inner_transport_header;
2854 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2856 return skb_inner_transport_header(skb) - skb->data;
2859 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2861 skb->inner_transport_header = skb->data - skb->head;
2864 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2867 skb_reset_inner_transport_header(skb);
2868 skb->inner_transport_header += offset;
2871 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2873 return skb->head + skb->inner_network_header;
2876 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2878 skb->inner_network_header = skb->data - skb->head;
2881 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2884 skb_reset_inner_network_header(skb);
2885 skb->inner_network_header += offset;
2888 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2890 return skb->head + skb->inner_mac_header;
2893 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2895 skb->inner_mac_header = skb->data - skb->head;
2898 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2901 skb_reset_inner_mac_header(skb);
2902 skb->inner_mac_header += offset;
2904 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2906 return skb->transport_header != (typeof(skb->transport_header))~0U;
2909 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2911 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
2912 return skb->head + skb->transport_header;
2915 static inline void skb_reset_transport_header(struct sk_buff *skb)
2917 skb->transport_header = skb->data - skb->head;
2920 static inline void skb_set_transport_header(struct sk_buff *skb,
2923 skb_reset_transport_header(skb);
2924 skb->transport_header += offset;
2927 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2929 return skb->head + skb->network_header;
2932 static inline void skb_reset_network_header(struct sk_buff *skb)
2934 skb->network_header = skb->data - skb->head;
2937 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2939 skb_reset_network_header(skb);
2940 skb->network_header += offset;
2943 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2945 return skb->head + skb->mac_header;
2948 static inline int skb_mac_offset(const struct sk_buff *skb)
2950 return skb_mac_header(skb) - skb->data;
2953 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2955 return skb->network_header - skb->mac_header;
2958 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2960 return skb->mac_header != (typeof(skb->mac_header))~0U;
2963 static inline void skb_unset_mac_header(struct sk_buff *skb)
2965 skb->mac_header = (typeof(skb->mac_header))~0U;
2968 static inline void skb_reset_mac_header(struct sk_buff *skb)
2970 skb->mac_header = skb->data - skb->head;
2973 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2975 skb_reset_mac_header(skb);
2976 skb->mac_header += offset;
2979 static inline void skb_pop_mac_header(struct sk_buff *skb)
2981 skb->mac_header = skb->network_header;
2984 static inline void skb_probe_transport_header(struct sk_buff *skb)
2986 struct flow_keys_basic keys;
2988 if (skb_transport_header_was_set(skb))
2991 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2993 skb_set_transport_header(skb, keys.control.thoff);
2996 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2998 if (skb_mac_header_was_set(skb)) {
2999 const unsigned char *old_mac = skb_mac_header(skb);
3001 skb_set_mac_header(skb, -skb->mac_len);
3002 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
3006 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
3008 return skb->csum_start - skb_headroom(skb);
3011 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
3013 return skb->head + skb->csum_start;
3016 static inline int skb_transport_offset(const struct sk_buff *skb)
3018 return skb_transport_header(skb) - skb->data;
3021 static inline u32 skb_network_header_len(const struct sk_buff *skb)
3023 return skb->transport_header - skb->network_header;
3026 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
3028 return skb->inner_transport_header - skb->inner_network_header;
3031 static inline int skb_network_offset(const struct sk_buff *skb)
3033 return skb_network_header(skb) - skb->data;
3036 static inline int skb_inner_network_offset(const struct sk_buff *skb)
3038 return skb_inner_network_header(skb) - skb->data;
3041 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
3043 return pskb_may_pull(skb, skb_network_offset(skb) + len);
3047 * CPUs often take a performance hit when accessing unaligned memory
3048 * locations. The actual performance hit varies, it can be small if the
3049 * hardware handles it or large if we have to take an exception and fix it
3052 * Since an ethernet header is 14 bytes network drivers often end up with
3053 * the IP header at an unaligned offset. The IP header can be aligned by
3054 * shifting the start of the packet by 2 bytes. Drivers should do this
3057 * skb_reserve(skb, NET_IP_ALIGN);
3059 * The downside to this alignment of the IP header is that the DMA is now
3060 * unaligned. On some architectures the cost of an unaligned DMA is high
3061 * and this cost outweighs the gains made by aligning the IP header.
3063 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
3066 #ifndef NET_IP_ALIGN
3067 #define NET_IP_ALIGN 2
3071 * The networking layer reserves some headroom in skb data (via
3072 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
3073 * the header has to grow. In the default case, if the header has to grow
3074 * 32 bytes or less we avoid the reallocation.
3076 * Unfortunately this headroom changes the DMA alignment of the resulting
3077 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
3078 * on some architectures. An architecture can override this value,
3079 * perhaps setting it to a cacheline in size (since that will maintain
3080 * cacheline alignment of the DMA). It must be a power of 2.
3082 * Various parts of the networking layer expect at least 32 bytes of
3083 * headroom, you should not reduce this.
3085 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
3086 * to reduce average number of cache lines per packet.
3087 * get_rps_cpu() for example only access one 64 bytes aligned block :
3088 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
3091 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
3094 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
3096 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
3098 if (WARN_ON(skb_is_nonlinear(skb)))
3101 skb_set_tail_pointer(skb, len);
3104 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
3106 __skb_set_length(skb, len);
3109 void skb_trim(struct sk_buff *skb, unsigned int len);
3111 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
3114 return ___pskb_trim(skb, len);
3115 __skb_trim(skb, len);
3119 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
3121 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
3125 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
3126 * @skb: buffer to alter
3129 * This is identical to pskb_trim except that the caller knows that
3130 * the skb is not cloned so we should never get an error due to out-
3133 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3135 int err = pskb_trim(skb, len);
3139 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3141 unsigned int diff = len - skb->len;
3143 if (skb_tailroom(skb) < diff) {
3144 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3149 __skb_set_length(skb, len);
3154 * skb_orphan - orphan a buffer
3155 * @skb: buffer to orphan
3157 * If a buffer currently has an owner then we call the owner's
3158 * destructor function and make the @skb unowned. The buffer continues
3159 * to exist but is no longer charged to its former owner.
3161 static inline void skb_orphan(struct sk_buff *skb)
3163 if (skb->destructor) {
3164 skb->destructor(skb);
3165 skb->destructor = NULL;
3173 * skb_orphan_frags - orphan the frags contained in a buffer
3174 * @skb: buffer to orphan frags from
3175 * @gfp_mask: allocation mask for replacement pages
3177 * For each frag in the SKB which needs a destructor (i.e. has an
3178 * owner) create a copy of that frag and release the original
3179 * page by calling the destructor.
3181 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3183 if (likely(!skb_zcopy(skb)))
3185 if (!skb_zcopy_is_nouarg(skb) &&
3186 skb_uarg(skb)->callback == msg_zerocopy_callback)
3188 return skb_copy_ubufs(skb, gfp_mask);
3191 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3192 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3194 if (likely(!skb_zcopy(skb)))
3196 return skb_copy_ubufs(skb, gfp_mask);
3200 * __skb_queue_purge - empty a list
3201 * @list: list to empty
3203 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3204 * the list and one reference dropped. This function does not take the
3205 * list lock and the caller must hold the relevant locks to use it.
3207 static inline void __skb_queue_purge(struct sk_buff_head *list)
3209 struct sk_buff *skb;
3210 while ((skb = __skb_dequeue(list)) != NULL)
3213 void skb_queue_purge(struct sk_buff_head *list);
3215 unsigned int skb_rbtree_purge(struct rb_root *root);
3217 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3220 * netdev_alloc_frag - allocate a page fragment
3221 * @fragsz: fragment size
3223 * Allocates a frag from a page for receive buffer.
3224 * Uses GFP_ATOMIC allocations.
3226 static inline void *netdev_alloc_frag(unsigned int fragsz)
3228 return __netdev_alloc_frag_align(fragsz, ~0u);
3231 static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3234 WARN_ON_ONCE(!is_power_of_2(align));
3235 return __netdev_alloc_frag_align(fragsz, -align);
3238 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3242 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3243 * @dev: network device to receive on
3244 * @length: length to allocate
3246 * Allocate a new &sk_buff and assign it a usage count of one. The
3247 * buffer has unspecified headroom built in. Users should allocate
3248 * the headroom they think they need without accounting for the
3249 * built in space. The built in space is used for optimisations.
3251 * %NULL is returned if there is no free memory. Although this function
3252 * allocates memory it can be called from an interrupt.
3254 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3255 unsigned int length)
3257 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3260 /* legacy helper around __netdev_alloc_skb() */
3261 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3264 return __netdev_alloc_skb(NULL, length, gfp_mask);
3267 /* legacy helper around netdev_alloc_skb() */
3268 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3270 return netdev_alloc_skb(NULL, length);
3274 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3275 unsigned int length, gfp_t gfp)
3277 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3279 if (NET_IP_ALIGN && skb)
3280 skb_reserve(skb, NET_IP_ALIGN);
3284 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3285 unsigned int length)
3287 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3290 static inline void skb_free_frag(void *addr)
3292 page_frag_free(addr);
3295 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3297 static inline void *napi_alloc_frag(unsigned int fragsz)
3299 return __napi_alloc_frag_align(fragsz, ~0u);
3302 static inline void *napi_alloc_frag_align(unsigned int fragsz,
3305 WARN_ON_ONCE(!is_power_of_2(align));
3306 return __napi_alloc_frag_align(fragsz, -align);
3309 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
3310 unsigned int length, gfp_t gfp_mask);
3311 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
3312 unsigned int length)
3314 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
3316 void napi_consume_skb(struct sk_buff *skb, int budget);
3318 void napi_skb_free_stolen_head(struct sk_buff *skb);
3319 void __kfree_skb_defer(struct sk_buff *skb);
3322 * __dev_alloc_pages - allocate page for network Rx
3323 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3324 * @order: size of the allocation
3326 * Allocate a new page.
3328 * %NULL is returned if there is no free memory.
3330 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
3333 /* This piece of code contains several assumptions.
3334 * 1. This is for device Rx, therefor a cold page is preferred.
3335 * 2. The expectation is the user wants a compound page.
3336 * 3. If requesting a order 0 page it will not be compound
3337 * due to the check to see if order has a value in prep_new_page
3338 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3339 * code in gfp_to_alloc_flags that should be enforcing this.
3341 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3343 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
3346 static inline struct page *dev_alloc_pages(unsigned int order)
3348 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
3352 * __dev_alloc_page - allocate a page for network Rx
3353 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3355 * Allocate a new page.
3357 * %NULL is returned if there is no free memory.
3359 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
3361 return __dev_alloc_pages(gfp_mask, 0);
3364 static inline struct page *dev_alloc_page(void)
3366 return dev_alloc_pages(0);
3370 * dev_page_is_reusable - check whether a page can be reused for network Rx
3371 * @page: the page to test
3373 * A page shouldn't be considered for reusing/recycling if it was allocated
3374 * under memory pressure or at a distant memory node.
3376 * Returns false if this page should be returned to page allocator, true
3379 static inline bool dev_page_is_reusable(const struct page *page)
3381 return likely(page_to_nid(page) == numa_mem_id() &&
3382 !page_is_pfmemalloc(page));
3386 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3387 * @page: The page that was allocated from skb_alloc_page
3388 * @skb: The skb that may need pfmemalloc set
3390 static inline void skb_propagate_pfmemalloc(const struct page *page,
3391 struct sk_buff *skb)
3393 if (page_is_pfmemalloc(page))
3394 skb->pfmemalloc = true;
3398 * skb_frag_off() - Returns the offset of a skb fragment
3399 * @frag: the paged fragment
3401 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3403 return frag->bv_offset;
3407 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3408 * @frag: skb fragment
3409 * @delta: value to add
3411 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3413 frag->bv_offset += delta;
3417 * skb_frag_off_set() - Sets the offset of a skb fragment
3418 * @frag: skb fragment
3419 * @offset: offset of fragment
3421 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3423 frag->bv_offset = offset;
3427 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3428 * @fragto: skb fragment where offset is set
3429 * @fragfrom: skb fragment offset is copied from
3431 static inline void skb_frag_off_copy(skb_frag_t *fragto,
3432 const skb_frag_t *fragfrom)
3434 fragto->bv_offset = fragfrom->bv_offset;
3438 * skb_frag_page - retrieve the page referred to by a paged fragment
3439 * @frag: the paged fragment
3441 * Returns the &struct page associated with @frag.
3443 static inline struct page *skb_frag_page(const skb_frag_t *frag)
3445 return frag->bv_page;
3449 * __skb_frag_ref - take an addition reference on a paged fragment.
3450 * @frag: the paged fragment
3452 * Takes an additional reference on the paged fragment @frag.
3454 static inline void __skb_frag_ref(skb_frag_t *frag)
3456 get_page(skb_frag_page(frag));
3460 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
3462 * @f: the fragment offset.
3464 * Takes an additional reference on the @f'th paged fragment of @skb.
3466 static inline void skb_frag_ref(struct sk_buff *skb, int f)
3468 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
3472 * __skb_frag_unref - release a reference on a paged fragment.
3473 * @frag: the paged fragment
3474 * @recycle: recycle the page if allocated via page_pool
3476 * Releases a reference on the paged fragment @frag
3477 * or recycles the page via the page_pool API.
3479 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle)
3481 struct page *page = skb_frag_page(frag);
3483 #ifdef CONFIG_PAGE_POOL
3484 if (recycle && page_pool_return_skb_page(page))
3491 * skb_frag_unref - release a reference on a paged fragment of an skb.
3493 * @f: the fragment offset
3495 * Releases a reference on the @f'th paged fragment of @skb.
3497 static inline void skb_frag_unref(struct sk_buff *skb, int f)
3499 __skb_frag_unref(&skb_shinfo(skb)->frags[f], skb->pp_recycle);
3503 * skb_frag_address - gets the address of the data contained in a paged fragment
3504 * @frag: the paged fragment buffer
3506 * Returns the address of the data within @frag. The page must already
3509 static inline void *skb_frag_address(const skb_frag_t *frag)
3511 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3515 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3516 * @frag: the paged fragment buffer
3518 * Returns the address of the data within @frag. Checks that the page
3519 * is mapped and returns %NULL otherwise.
3521 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3523 void *ptr = page_address(skb_frag_page(frag));
3527 return ptr + skb_frag_off(frag);
3531 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3532 * @fragto: skb fragment where page is set
3533 * @fragfrom: skb fragment page is copied from
3535 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3536 const skb_frag_t *fragfrom)
3538 fragto->bv_page = fragfrom->bv_page;
3542 * __skb_frag_set_page - sets the page contained in a paged fragment
3543 * @frag: the paged fragment
3544 * @page: the page to set
3546 * Sets the fragment @frag to contain @page.
3548 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
3550 frag->bv_page = page;
3554 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
3556 * @f: the fragment offset
3557 * @page: the page to set
3559 * Sets the @f'th fragment of @skb to contain @page.
3561 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
3564 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
3567 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3570 * skb_frag_dma_map - maps a paged fragment via the DMA API
3571 * @dev: the device to map the fragment to
3572 * @frag: the paged fragment to map
3573 * @offset: the offset within the fragment (starting at the
3574 * fragment's own offset)
3575 * @size: the number of bytes to map
3576 * @dir: the direction of the mapping (``PCI_DMA_*``)
3578 * Maps the page associated with @frag to @device.
3580 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3581 const skb_frag_t *frag,
3582 size_t offset, size_t size,
3583 enum dma_data_direction dir)
3585 return dma_map_page(dev, skb_frag_page(frag),
3586 skb_frag_off(frag) + offset, size, dir);
3589 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3592 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3596 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3599 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3604 * skb_clone_writable - is the header of a clone writable
3605 * @skb: buffer to check
3606 * @len: length up to which to write
3608 * Returns true if modifying the header part of the cloned buffer
3609 * does not requires the data to be copied.
3611 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3613 return !skb_header_cloned(skb) &&
3614 skb_headroom(skb) + len <= skb->hdr_len;
3617 static inline int skb_try_make_writable(struct sk_buff *skb,
3618 unsigned int write_len)
3620 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3621 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3624 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3629 if (headroom > skb_headroom(skb))
3630 delta = headroom - skb_headroom(skb);
3632 if (delta || cloned)
3633 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3639 * skb_cow - copy header of skb when it is required
3640 * @skb: buffer to cow
3641 * @headroom: needed headroom
3643 * If the skb passed lacks sufficient headroom or its data part
3644 * is shared, data is reallocated. If reallocation fails, an error
3645 * is returned and original skb is not changed.
3647 * The result is skb with writable area skb->head...skb->tail
3648 * and at least @headroom of space at head.
3650 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3652 return __skb_cow(skb, headroom, skb_cloned(skb));
3656 * skb_cow_head - skb_cow but only making the head writable
3657 * @skb: buffer to cow
3658 * @headroom: needed headroom
3660 * This function is identical to skb_cow except that we replace the
3661 * skb_cloned check by skb_header_cloned. It should be used when
3662 * you only need to push on some header and do not need to modify
3665 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3667 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3671 * skb_padto - pad an skbuff up to a minimal size
3672 * @skb: buffer to pad
3673 * @len: minimal length
3675 * Pads up a buffer to ensure the trailing bytes exist and are
3676 * blanked. If the buffer already contains sufficient data it
3677 * is untouched. Otherwise it is extended. Returns zero on
3678 * success. The skb is freed on error.
3680 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3682 unsigned int size = skb->len;
3683 if (likely(size >= len))
3685 return skb_pad(skb, len - size);
3689 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3690 * @skb: buffer to pad
3691 * @len: minimal length
3692 * @free_on_error: free buffer on error
3694 * Pads up a buffer to ensure the trailing bytes exist and are
3695 * blanked. If the buffer already contains sufficient data it
3696 * is untouched. Otherwise it is extended. Returns zero on
3697 * success. The skb is freed on error if @free_on_error is true.
3699 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3703 unsigned int size = skb->len;
3705 if (unlikely(size < len)) {
3707 if (__skb_pad(skb, len, free_on_error))
3709 __skb_put(skb, len);
3715 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3716 * @skb: buffer to pad
3717 * @len: minimal length
3719 * Pads up a buffer to ensure the trailing bytes exist and are
3720 * blanked. If the buffer already contains sufficient data it
3721 * is untouched. Otherwise it is extended. Returns zero on
3722 * success. The skb is freed on error.
3724 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3726 return __skb_put_padto(skb, len, true);
3729 static inline int skb_add_data(struct sk_buff *skb,
3730 struct iov_iter *from, int copy)
3732 const int off = skb->len;
3734 if (skb->ip_summed == CHECKSUM_NONE) {
3736 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3738 skb->csum = csum_block_add(skb->csum, csum, off);
3741 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3744 __skb_trim(skb, off);
3748 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3749 const struct page *page, int off)
3754 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3756 return page == skb_frag_page(frag) &&
3757 off == skb_frag_off(frag) + skb_frag_size(frag);
3762 static inline int __skb_linearize(struct sk_buff *skb)
3764 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3768 * skb_linearize - convert paged skb to linear one
3769 * @skb: buffer to linarize
3771 * If there is no free memory -ENOMEM is returned, otherwise zero
3772 * is returned and the old skb data released.
3774 static inline int skb_linearize(struct sk_buff *skb)
3776 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3780 * skb_has_shared_frag - can any frag be overwritten
3781 * @skb: buffer to test
3783 * Return true if the skb has at least one frag that might be modified
3784 * by an external entity (as in vmsplice()/sendfile())
3786 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3788 return skb_is_nonlinear(skb) &&
3789 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3793 * skb_linearize_cow - make sure skb is linear and writable
3794 * @skb: buffer to process
3796 * If there is no free memory -ENOMEM is returned, otherwise zero
3797 * is returned and the old skb data released.
3799 static inline int skb_linearize_cow(struct sk_buff *skb)
3801 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3802 __skb_linearize(skb) : 0;
3805 static __always_inline void
3806 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3809 if (skb->ip_summed == CHECKSUM_COMPLETE)
3810 skb->csum = csum_block_sub(skb->csum,
3811 csum_partial(start, len, 0), off);
3812 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3813 skb_checksum_start_offset(skb) < 0)
3814 skb->ip_summed = CHECKSUM_NONE;
3818 * skb_postpull_rcsum - update checksum for received skb after pull
3819 * @skb: buffer to update
3820 * @start: start of data before pull
3821 * @len: length of data pulled
3823 * After doing a pull on a received packet, you need to call this to
3824 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3825 * CHECKSUM_NONE so that it can be recomputed from scratch.
3827 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3828 const void *start, unsigned int len)
3830 if (skb->ip_summed == CHECKSUM_COMPLETE)
3831 skb->csum = wsum_negate(csum_partial(start, len,
3832 wsum_negate(skb->csum)));
3833 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3834 skb_checksum_start_offset(skb) < 0)
3835 skb->ip_summed = CHECKSUM_NONE;
3838 static __always_inline void
3839 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3842 if (skb->ip_summed == CHECKSUM_COMPLETE)
3843 skb->csum = csum_block_add(skb->csum,
3844 csum_partial(start, len, 0), off);
3848 * skb_postpush_rcsum - update checksum for received skb after push
3849 * @skb: buffer to update
3850 * @start: start of data after push
3851 * @len: length of data pushed
3853 * After doing a push on a received packet, you need to call this to
3854 * update the CHECKSUM_COMPLETE checksum.
3856 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3857 const void *start, unsigned int len)
3859 __skb_postpush_rcsum(skb, start, len, 0);
3862 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3865 * skb_push_rcsum - push skb and update receive checksum
3866 * @skb: buffer to update
3867 * @len: length of data pulled
3869 * This function performs an skb_push on the packet and updates
3870 * the CHECKSUM_COMPLETE checksum. It should be used on
3871 * receive path processing instead of skb_push unless you know
3872 * that the checksum difference is zero (e.g., a valid IP header)
3873 * or you are setting ip_summed to CHECKSUM_NONE.
3875 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3878 skb_postpush_rcsum(skb, skb->data, len);
3882 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3884 * pskb_trim_rcsum - trim received skb and update checksum
3885 * @skb: buffer to trim
3888 * This is exactly the same as pskb_trim except that it ensures the
3889 * checksum of received packets are still valid after the operation.
3890 * It can change skb pointers.
3893 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3895 if (likely(len >= skb->len))
3897 return pskb_trim_rcsum_slow(skb, len);
3900 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3902 if (skb->ip_summed == CHECKSUM_COMPLETE)
3903 skb->ip_summed = CHECKSUM_NONE;
3904 __skb_trim(skb, len);
3908 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3910 if (skb->ip_summed == CHECKSUM_COMPLETE)
3911 skb->ip_summed = CHECKSUM_NONE;
3912 return __skb_grow(skb, len);
3915 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3916 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3917 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3918 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3919 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3921 #define skb_queue_walk(queue, skb) \
3922 for (skb = (queue)->next; \
3923 skb != (struct sk_buff *)(queue); \
3926 #define skb_queue_walk_safe(queue, skb, tmp) \
3927 for (skb = (queue)->next, tmp = skb->next; \
3928 skb != (struct sk_buff *)(queue); \
3929 skb = tmp, tmp = skb->next)
3931 #define skb_queue_walk_from(queue, skb) \
3932 for (; skb != (struct sk_buff *)(queue); \
3935 #define skb_rbtree_walk(skb, root) \
3936 for (skb = skb_rb_first(root); skb != NULL; \
3937 skb = skb_rb_next(skb))
3939 #define skb_rbtree_walk_from(skb) \
3940 for (; skb != NULL; \
3941 skb = skb_rb_next(skb))
3943 #define skb_rbtree_walk_from_safe(skb, tmp) \
3944 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3947 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3948 for (tmp = skb->next; \
3949 skb != (struct sk_buff *)(queue); \
3950 skb = tmp, tmp = skb->next)
3952 #define skb_queue_reverse_walk(queue, skb) \
3953 for (skb = (queue)->prev; \
3954 skb != (struct sk_buff *)(queue); \
3957 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3958 for (skb = (queue)->prev, tmp = skb->prev; \
3959 skb != (struct sk_buff *)(queue); \
3960 skb = tmp, tmp = skb->prev)
3962 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3963 for (tmp = skb->prev; \
3964 skb != (struct sk_buff *)(queue); \
3965 skb = tmp, tmp = skb->prev)
3967 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3969 return skb_shinfo(skb)->frag_list != NULL;
3972 static inline void skb_frag_list_init(struct sk_buff *skb)
3974 skb_shinfo(skb)->frag_list = NULL;
3977 #define skb_walk_frags(skb, iter) \
3978 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3981 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3982 int *err, long *timeo_p,
3983 const struct sk_buff *skb);
3984 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3985 struct sk_buff_head *queue,
3988 struct sk_buff **last);
3989 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
3990 struct sk_buff_head *queue,
3991 unsigned int flags, int *off, int *err,
3992 struct sk_buff **last);
3993 struct sk_buff *__skb_recv_datagram(struct sock *sk,
3994 struct sk_buff_head *sk_queue,
3995 unsigned int flags, int *off, int *err);
3996 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
3997 __poll_t datagram_poll(struct file *file, struct socket *sock,
3998 struct poll_table_struct *wait);
3999 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
4000 struct iov_iter *to, int size);
4001 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
4002 struct msghdr *msg, int size)
4004 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
4006 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
4007 struct msghdr *msg);
4008 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
4009 struct iov_iter *to, int len,
4010 struct ahash_request *hash);
4011 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
4012 struct iov_iter *from, int len);
4013 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
4014 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
4015 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
4016 static inline void skb_free_datagram_locked(struct sock *sk,
4017 struct sk_buff *skb)
4019 __skb_free_datagram_locked(sk, skb, 0);
4021 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
4022 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
4023 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
4024 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
4026 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
4027 struct pipe_inode_info *pipe, unsigned int len,
4028 unsigned int flags);
4029 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
4031 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
4032 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
4033 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
4034 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
4036 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
4037 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
4038 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
4039 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
4040 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
4041 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
4042 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
4043 unsigned int offset);
4044 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
4045 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
4046 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
4047 int skb_vlan_pop(struct sk_buff *skb);
4048 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
4049 int skb_eth_pop(struct sk_buff *skb);
4050 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
4051 const unsigned char *src);
4052 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
4053 int mac_len, bool ethernet);
4054 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
4056 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
4057 int skb_mpls_dec_ttl(struct sk_buff *skb);
4058 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
4061 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
4063 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
4066 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
4068 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
4071 struct skb_checksum_ops {
4072 __wsum (*update)(const void *mem, int len, __wsum wsum);
4073 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
4076 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
4078 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
4079 __wsum csum, const struct skb_checksum_ops *ops);
4080 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
4083 static inline void * __must_check
4084 __skb_header_pointer(const struct sk_buff *skb, int offset, int len,
4085 const void *data, int hlen, void *buffer)
4087 if (likely(hlen - offset >= len))
4088 return (void *)data + offset;
4090 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
4096 static inline void * __must_check
4097 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
4099 return __skb_header_pointer(skb, offset, len, skb->data,
4100 skb_headlen(skb), buffer);
4104 * skb_needs_linearize - check if we need to linearize a given skb
4105 * depending on the given device features.
4106 * @skb: socket buffer to check
4107 * @features: net device features
4109 * Returns true if either:
4110 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
4111 * 2. skb is fragmented and the device does not support SG.
4113 static inline bool skb_needs_linearize(struct sk_buff *skb,
4114 netdev_features_t features)
4116 return skb_is_nonlinear(skb) &&
4117 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
4118 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
4121 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
4123 const unsigned int len)
4125 memcpy(to, skb->data, len);
4128 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4129 const int offset, void *to,
4130 const unsigned int len)
4132 memcpy(to, skb->data + offset, len);
4135 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4137 const unsigned int len)
4139 memcpy(skb->data, from, len);
4142 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4145 const unsigned int len)
4147 memcpy(skb->data + offset, from, len);
4150 void skb_init(void);
4152 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4158 * skb_get_timestamp - get timestamp from a skb
4159 * @skb: skb to get stamp from
4160 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4162 * Timestamps are stored in the skb as offsets to a base timestamp.
4163 * This function converts the offset back to a struct timeval and stores
4166 static inline void skb_get_timestamp(const struct sk_buff *skb,
4167 struct __kernel_old_timeval *stamp)
4169 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4172 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4173 struct __kernel_sock_timeval *stamp)
4175 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4177 stamp->tv_sec = ts.tv_sec;
4178 stamp->tv_usec = ts.tv_nsec / 1000;
4181 static inline void skb_get_timestampns(const struct sk_buff *skb,
4182 struct __kernel_old_timespec *stamp)
4184 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4186 stamp->tv_sec = ts.tv_sec;
4187 stamp->tv_nsec = ts.tv_nsec;
4190 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4191 struct __kernel_timespec *stamp)
4193 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4195 stamp->tv_sec = ts.tv_sec;
4196 stamp->tv_nsec = ts.tv_nsec;
4199 static inline void __net_timestamp(struct sk_buff *skb)
4201 skb->tstamp = ktime_get_real();
4202 skb->mono_delivery_time = 0;
4205 static inline ktime_t net_timedelta(ktime_t t)
4207 return ktime_sub(ktime_get_real(), t);
4210 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4214 skb->mono_delivery_time = kt && mono;
4217 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4219 /* It is used in the ingress path to clear the delivery_time.
4220 * If needed, set the skb->tstamp to the (rcv) timestamp.
4222 static inline void skb_clear_delivery_time(struct sk_buff *skb)
4224 if (skb->mono_delivery_time) {
4225 skb->mono_delivery_time = 0;
4226 if (static_branch_unlikely(&netstamp_needed_key))
4227 skb->tstamp = ktime_get_real();
4233 static inline void skb_clear_tstamp(struct sk_buff *skb)
4235 if (skb->mono_delivery_time)
4241 static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4243 if (skb->mono_delivery_time)
4249 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4251 if (!skb->mono_delivery_time && skb->tstamp)
4254 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4255 return ktime_get_real();
4260 static inline u8 skb_metadata_len(const struct sk_buff *skb)
4262 return skb_shinfo(skb)->meta_len;
4265 static inline void *skb_metadata_end(const struct sk_buff *skb)
4267 return skb_mac_header(skb);
4270 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4271 const struct sk_buff *skb_b,
4274 const void *a = skb_metadata_end(skb_a);
4275 const void *b = skb_metadata_end(skb_b);
4276 /* Using more efficient varaiant than plain call to memcmp(). */
4277 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
4281 #define __it(x, op) (x -= sizeof(u##op))
4282 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4283 case 32: diffs |= __it_diff(a, b, 64);
4285 case 24: diffs |= __it_diff(a, b, 64);
4287 case 16: diffs |= __it_diff(a, b, 64);
4289 case 8: diffs |= __it_diff(a, b, 64);
4291 case 28: diffs |= __it_diff(a, b, 64);
4293 case 20: diffs |= __it_diff(a, b, 64);
4295 case 12: diffs |= __it_diff(a, b, 64);
4297 case 4: diffs |= __it_diff(a, b, 32);
4302 return memcmp(a - meta_len, b - meta_len, meta_len);
4306 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4307 const struct sk_buff *skb_b)
4309 u8 len_a = skb_metadata_len(skb_a);
4310 u8 len_b = skb_metadata_len(skb_b);
4312 if (!(len_a | len_b))
4315 return len_a != len_b ?
4316 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4319 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4321 skb_shinfo(skb)->meta_len = meta_len;
4324 static inline void skb_metadata_clear(struct sk_buff *skb)
4326 skb_metadata_set(skb, 0);
4329 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4331 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4333 void skb_clone_tx_timestamp(struct sk_buff *skb);
4334 bool skb_defer_rx_timestamp(struct sk_buff *skb);
4336 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4338 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4342 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4347 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4350 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4352 * PHY drivers may accept clones of transmitted packets for
4353 * timestamping via their phy_driver.txtstamp method. These drivers
4354 * must call this function to return the skb back to the stack with a
4357 * @skb: clone of the original outgoing packet
4358 * @hwtstamps: hardware time stamps
4361 void skb_complete_tx_timestamp(struct sk_buff *skb,
4362 struct skb_shared_hwtstamps *hwtstamps);
4364 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4365 struct skb_shared_hwtstamps *hwtstamps,
4366 struct sock *sk, int tstype);
4369 * skb_tstamp_tx - queue clone of skb with send time stamps
4370 * @orig_skb: the original outgoing packet
4371 * @hwtstamps: hardware time stamps, may be NULL if not available
4373 * If the skb has a socket associated, then this function clones the
4374 * skb (thus sharing the actual data and optional structures), stores
4375 * the optional hardware time stamping information (if non NULL) or
4376 * generates a software time stamp (otherwise), then queues the clone
4377 * to the error queue of the socket. Errors are silently ignored.
4379 void skb_tstamp_tx(struct sk_buff *orig_skb,
4380 struct skb_shared_hwtstamps *hwtstamps);
4383 * skb_tx_timestamp() - Driver hook for transmit timestamping
4385 * Ethernet MAC Drivers should call this function in their hard_xmit()
4386 * function immediately before giving the sk_buff to the MAC hardware.
4388 * Specifically, one should make absolutely sure that this function is
4389 * called before TX completion of this packet can trigger. Otherwise
4390 * the packet could potentially already be freed.
4392 * @skb: A socket buffer.
4394 static inline void skb_tx_timestamp(struct sk_buff *skb)
4396 skb_clone_tx_timestamp(skb);
4397 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4398 skb_tstamp_tx(skb, NULL);
4402 * skb_complete_wifi_ack - deliver skb with wifi status
4404 * @skb: the original outgoing packet
4405 * @acked: ack status
4408 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4410 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4411 __sum16 __skb_checksum_complete(struct sk_buff *skb);
4413 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4415 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4417 (skb->ip_summed == CHECKSUM_PARTIAL &&
4418 skb_checksum_start_offset(skb) >= 0));
4422 * skb_checksum_complete - Calculate checksum of an entire packet
4423 * @skb: packet to process
4425 * This function calculates the checksum over the entire packet plus
4426 * the value of skb->csum. The latter can be used to supply the
4427 * checksum of a pseudo header as used by TCP/UDP. It returns the
4430 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4431 * this function can be used to verify that checksum on received
4432 * packets. In that case the function should return zero if the
4433 * checksum is correct. In particular, this function will return zero
4434 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4435 * hardware has already verified the correctness of the checksum.
4437 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4439 return skb_csum_unnecessary(skb) ?
4440 0 : __skb_checksum_complete(skb);
4443 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4445 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4446 if (skb->csum_level == 0)
4447 skb->ip_summed = CHECKSUM_NONE;
4453 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4455 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4456 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4458 } else if (skb->ip_summed == CHECKSUM_NONE) {
4459 skb->ip_summed = CHECKSUM_UNNECESSARY;
4460 skb->csum_level = 0;
4464 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4466 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4467 skb->ip_summed = CHECKSUM_NONE;
4468 skb->csum_level = 0;
4472 /* Check if we need to perform checksum complete validation.
4474 * Returns true if checksum complete is needed, false otherwise
4475 * (either checksum is unnecessary or zero checksum is allowed).
4477 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4481 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4482 skb->csum_valid = 1;
4483 __skb_decr_checksum_unnecessary(skb);
4490 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4493 #define CHECKSUM_BREAK 76
4495 /* Unset checksum-complete
4497 * Unset checksum complete can be done when packet is being modified
4498 * (uncompressed for instance) and checksum-complete value is
4501 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4503 if (skb->ip_summed == CHECKSUM_COMPLETE)
4504 skb->ip_summed = CHECKSUM_NONE;
4507 /* Validate (init) checksum based on checksum complete.
4510 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4511 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4512 * checksum is stored in skb->csum for use in __skb_checksum_complete
4513 * non-zero: value of invalid checksum
4516 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4520 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4521 if (!csum_fold(csum_add(psum, skb->csum))) {
4522 skb->csum_valid = 1;
4529 if (complete || skb->len <= CHECKSUM_BREAK) {
4532 csum = __skb_checksum_complete(skb);
4533 skb->csum_valid = !csum;
4540 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4545 /* Perform checksum validate (init). Note that this is a macro since we only
4546 * want to calculate the pseudo header which is an input function if necessary.
4547 * First we try to validate without any computation (checksum unnecessary) and
4548 * then calculate based on checksum complete calling the function to compute
4552 * 0: checksum is validated or try to in skb_checksum_complete
4553 * non-zero: value of invalid checksum
4555 #define __skb_checksum_validate(skb, proto, complete, \
4556 zero_okay, check, compute_pseudo) \
4558 __sum16 __ret = 0; \
4559 skb->csum_valid = 0; \
4560 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4561 __ret = __skb_checksum_validate_complete(skb, \
4562 complete, compute_pseudo(skb, proto)); \
4566 #define skb_checksum_init(skb, proto, compute_pseudo) \
4567 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4569 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4570 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4572 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4573 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4575 #define skb_checksum_validate_zero_check(skb, proto, check, \
4577 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4579 #define skb_checksum_simple_validate(skb) \
4580 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4582 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4584 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4587 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4589 skb->csum = ~pseudo;
4590 skb->ip_summed = CHECKSUM_COMPLETE;
4593 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4595 if (__skb_checksum_convert_check(skb)) \
4596 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4599 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4600 u16 start, u16 offset)
4602 skb->ip_summed = CHECKSUM_PARTIAL;
4603 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4604 skb->csum_offset = offset - start;
4607 /* Update skbuf and packet to reflect the remote checksum offload operation.
4608 * When called, ptr indicates the starting point for skb->csum when
4609 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4610 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4612 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4613 int start, int offset, bool nopartial)
4618 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4622 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4623 __skb_checksum_complete(skb);
4624 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4627 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4629 /* Adjust skb->csum since we changed the packet */
4630 skb->csum = csum_add(skb->csum, delta);
4633 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4635 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4636 return (void *)(skb->_nfct & NFCT_PTRMASK);
4642 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4644 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4651 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4653 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4654 skb->slow_gro |= !!nfct;
4659 #ifdef CONFIG_SKB_EXTENSIONS
4661 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4667 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4670 #if IS_ENABLED(CONFIG_MPTCP)
4673 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4676 SKB_EXT_NUM, /* must be last */
4680 * struct skb_ext - sk_buff extensions
4681 * @refcnt: 1 on allocation, deallocated on 0
4682 * @offset: offset to add to @data to obtain extension address
4683 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4684 * @data: start of extension data, variable sized
4686 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4687 * to use 'u8' types while allowing up to 2kb worth of extension data.
4691 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4692 u8 chunks; /* same */
4693 char data[] __aligned(8);
4696 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4697 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4698 struct skb_ext *ext);
4699 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4700 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4701 void __skb_ext_put(struct skb_ext *ext);
4703 static inline void skb_ext_put(struct sk_buff *skb)
4705 if (skb->active_extensions)
4706 __skb_ext_put(skb->extensions);
4709 static inline void __skb_ext_copy(struct sk_buff *dst,
4710 const struct sk_buff *src)
4712 dst->active_extensions = src->active_extensions;
4714 if (src->active_extensions) {
4715 struct skb_ext *ext = src->extensions;
4717 refcount_inc(&ext->refcnt);
4718 dst->extensions = ext;
4722 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4725 __skb_ext_copy(dst, src);
4728 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4730 return !!ext->offset[i];
4733 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4735 return skb->active_extensions & (1 << id);
4738 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4740 if (skb_ext_exist(skb, id))
4741 __skb_ext_del(skb, id);
4744 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4746 if (skb_ext_exist(skb, id)) {
4747 struct skb_ext *ext = skb->extensions;
4749 return (void *)ext + (ext->offset[id] << 3);
4755 static inline void skb_ext_reset(struct sk_buff *skb)
4757 if (unlikely(skb->active_extensions)) {
4758 __skb_ext_put(skb->extensions);
4759 skb->active_extensions = 0;
4763 static inline bool skb_has_extensions(struct sk_buff *skb)
4765 return unlikely(skb->active_extensions);
4768 static inline void skb_ext_put(struct sk_buff *skb) {}
4769 static inline void skb_ext_reset(struct sk_buff *skb) {}
4770 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4771 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4772 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4773 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4774 #endif /* CONFIG_SKB_EXTENSIONS */
4776 static inline void nf_reset_ct(struct sk_buff *skb)
4778 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4779 nf_conntrack_put(skb_nfct(skb));
4784 static inline void nf_reset_trace(struct sk_buff *skb)
4786 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4791 static inline void ipvs_reset(struct sk_buff *skb)
4793 #if IS_ENABLED(CONFIG_IP_VS)
4794 skb->ipvs_property = 0;
4798 /* Note: This doesn't put any conntrack info in dst. */
4799 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4802 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4803 dst->_nfct = src->_nfct;
4804 nf_conntrack_get(skb_nfct(src));
4806 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4808 dst->nf_trace = src->nf_trace;
4812 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4814 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4815 nf_conntrack_put(skb_nfct(dst));
4817 dst->slow_gro = src->slow_gro;
4818 __nf_copy(dst, src, true);
4821 #ifdef CONFIG_NETWORK_SECMARK
4822 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4824 to->secmark = from->secmark;
4827 static inline void skb_init_secmark(struct sk_buff *skb)
4832 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4835 static inline void skb_init_secmark(struct sk_buff *skb)
4839 static inline int secpath_exists(const struct sk_buff *skb)
4842 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4848 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4850 return !skb->destructor &&
4851 !secpath_exists(skb) &&
4853 !skb->_skb_refdst &&
4854 !skb_has_frag_list(skb);
4857 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4859 skb->queue_mapping = queue_mapping;
4862 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4864 return skb->queue_mapping;
4867 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4869 to->queue_mapping = from->queue_mapping;
4872 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4874 skb->queue_mapping = rx_queue + 1;
4877 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4879 return skb->queue_mapping - 1;
4882 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4884 return skb->queue_mapping != 0;
4887 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4889 skb->dst_pending_confirm = val;
4892 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4894 return skb->dst_pending_confirm != 0;
4897 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4900 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4906 /* Keeps track of mac header offset relative to skb->head.
4907 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4908 * For non-tunnel skb it points to skb_mac_header() and for
4909 * tunnel skb it points to outer mac header.
4910 * Keeps track of level of encapsulation of network headers.
4921 #define SKB_GSO_CB_OFFSET 32
4922 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET))
4924 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4926 return (skb_mac_header(inner_skb) - inner_skb->head) -
4927 SKB_GSO_CB(inner_skb)->mac_offset;
4930 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4932 int new_headroom, headroom;
4935 headroom = skb_headroom(skb);
4936 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4940 new_headroom = skb_headroom(skb);
4941 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4945 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4947 /* Do not update partial checksums if remote checksum is enabled. */
4948 if (skb->remcsum_offload)
4951 SKB_GSO_CB(skb)->csum = res;
4952 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4955 /* Compute the checksum for a gso segment. First compute the checksum value
4956 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4957 * then add in skb->csum (checksum from csum_start to end of packet).
4958 * skb->csum and csum_start are then updated to reflect the checksum of the
4959 * resultant packet starting from the transport header-- the resultant checksum
4960 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4963 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4965 unsigned char *csum_start = skb_transport_header(skb);
4966 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4967 __wsum partial = SKB_GSO_CB(skb)->csum;
4969 SKB_GSO_CB(skb)->csum = res;
4970 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4972 return csum_fold(csum_partial(csum_start, plen, partial));
4975 static inline bool skb_is_gso(const struct sk_buff *skb)
4977 return skb_shinfo(skb)->gso_size;
4980 /* Note: Should be called only if skb_is_gso(skb) is true */
4981 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4983 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4986 /* Note: Should be called only if skb_is_gso(skb) is true */
4987 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4989 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4992 /* Note: Should be called only if skb_is_gso(skb) is true */
4993 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4995 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4998 static inline void skb_gso_reset(struct sk_buff *skb)
5000 skb_shinfo(skb)->gso_size = 0;
5001 skb_shinfo(skb)->gso_segs = 0;
5002 skb_shinfo(skb)->gso_type = 0;
5005 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
5008 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
5010 shinfo->gso_size += increment;
5013 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
5016 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
5018 shinfo->gso_size -= decrement;
5021 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
5023 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
5025 /* LRO sets gso_size but not gso_type, whereas if GSO is really
5026 * wanted then gso_type will be set. */
5027 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5029 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
5030 unlikely(shinfo->gso_type == 0)) {
5031 __skb_warn_lro_forwarding(skb);
5037 static inline void skb_forward_csum(struct sk_buff *skb)
5039 /* Unfortunately we don't support this one. Any brave souls? */
5040 if (skb->ip_summed == CHECKSUM_COMPLETE)
5041 skb->ip_summed = CHECKSUM_NONE;
5045 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
5046 * @skb: skb to check
5048 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
5049 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
5050 * use this helper, to document places where we make this assertion.
5052 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
5054 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE);
5057 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
5059 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
5060 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5061 unsigned int transport_len,
5062 __sum16(*skb_chkf)(struct sk_buff *skb));
5065 * skb_head_is_locked - Determine if the skb->head is locked down
5066 * @skb: skb to check
5068 * The head on skbs build around a head frag can be removed if they are
5069 * not cloned. This function returns true if the skb head is locked down
5070 * due to either being allocated via kmalloc, or by being a clone with
5071 * multiple references to the head.
5073 static inline bool skb_head_is_locked(const struct sk_buff *skb)
5075 return !skb->head_frag || skb_cloned(skb);
5078 /* Local Checksum Offload.
5079 * Compute outer checksum based on the assumption that the
5080 * inner checksum will be offloaded later.
5081 * See Documentation/networking/checksum-offloads.rst for
5082 * explanation of how this works.
5083 * Fill in outer checksum adjustment (e.g. with sum of outer
5084 * pseudo-header) before calling.
5085 * Also ensure that inner checksum is in linear data area.
5087 static inline __wsum lco_csum(struct sk_buff *skb)
5089 unsigned char *csum_start = skb_checksum_start(skb);
5090 unsigned char *l4_hdr = skb_transport_header(skb);
5093 /* Start with complement of inner checksum adjustment */
5094 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
5097 /* Add in checksum of our headers (incl. outer checksum
5098 * adjustment filled in by caller) and return result.
5100 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
5103 static inline bool skb_is_redirected(const struct sk_buff *skb)
5105 return skb->redirected;
5108 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
5110 skb->redirected = 1;
5111 #ifdef CONFIG_NET_REDIRECT
5112 skb->from_ingress = from_ingress;
5113 if (skb->from_ingress)
5114 skb_clear_tstamp(skb);
5118 static inline void skb_reset_redirect(struct sk_buff *skb)
5120 skb->redirected = 0;
5123 static inline bool skb_csum_is_sctp(struct sk_buff *skb)
5125 return skb->csum_not_inet;
5128 static inline void skb_set_kcov_handle(struct sk_buff *skb,
5129 const u64 kcov_handle)
5132 skb->kcov_handle = kcov_handle;
5136 static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
5139 return skb->kcov_handle;
5145 #ifdef CONFIG_PAGE_POOL
5146 static inline void skb_mark_for_recycle(struct sk_buff *skb)
5148 skb->pp_recycle = 1;
5152 static inline bool skb_pp_recycle(struct sk_buff *skb, void *data)
5154 if (!IS_ENABLED(CONFIG_PAGE_POOL) || !skb->pp_recycle)
5156 return page_pool_return_skb_page(virt_to_page(data));
5159 #endif /* __KERNEL__ */
5160 #endif /* _LINUX_SKBUFF_H */