rhashtable: Move seed init into bucket_table_alloc
[linux-2.6/btrfs-unstable.git] / include / linux / skbuff.h
blobbba1330757c025fc783cf10c2d6e0cfd349f2f5c
1 /*
2 * Definitions for the 'struct sk_buff' memory handlers.
4 * Authors:
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
17 #include <linux/kernel.h>
18 #include <linux/kmemcheck.h>
19 #include <linux/compiler.h>
20 #include <linux/time.h>
21 #include <linux/bug.h>
22 #include <linux/cache.h>
23 #include <linux/rbtree.h>
24 #include <linux/socket.h>
26 #include <linux/atomic.h>
27 #include <asm/types.h>
28 #include <linux/spinlock.h>
29 #include <linux/net.h>
30 #include <linux/textsearch.h>
31 #include <net/checksum.h>
32 #include <linux/rcupdate.h>
33 #include <linux/hrtimer.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/netdev_features.h>
36 #include <linux/sched.h>
37 #include <net/flow_keys.h>
39 /* A. Checksumming of received packets by device.
41 * CHECKSUM_NONE:
43 * Device failed to checksum this packet e.g. due to lack of capabilities.
44 * The packet contains full (though not verified) checksum in packet but
45 * not in skb->csum. Thus, skb->csum is undefined in this case.
47 * CHECKSUM_UNNECESSARY:
49 * The hardware you're dealing with doesn't calculate the full checksum
50 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
51 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
52 * if their checksums are okay. skb->csum is still undefined in this case
53 * though. It is a bad option, but, unfortunately, nowadays most vendors do
54 * this. Apparently with the secret goal to sell you new devices, when you
55 * will add new protocol to your host, f.e. IPv6 8)
57 * CHECKSUM_UNNECESSARY is applicable to following protocols:
58 * TCP: IPv6 and IPv4.
59 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
60 * zero UDP checksum for either IPv4 or IPv6, the networking stack
61 * may perform further validation in this case.
62 * GRE: only if the checksum is present in the header.
63 * SCTP: indicates the CRC in SCTP header has been validated.
65 * skb->csum_level indicates the number of consecutive checksums found in
66 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
67 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
68 * and a device is able to verify the checksums for UDP (possibly zero),
69 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
70 * two. If the device were only able to verify the UDP checksum and not
71 * GRE, either because it doesn't support GRE checksum of because GRE
72 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
73 * not considered in this case).
75 * CHECKSUM_COMPLETE:
77 * This is the most generic way. The device supplied checksum of the _whole_
78 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
79 * hardware doesn't need to parse L3/L4 headers to implement this.
81 * Note: Even if device supports only some protocols, but is able to produce
82 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
84 * CHECKSUM_PARTIAL:
86 * A checksum is set up to be offloaded to a device as described in the
87 * output description for CHECKSUM_PARTIAL. This may occur on a packet
88 * received directly from another Linux OS, e.g., a virtualized Linux kernel
89 * on the same host, or it may be set in the input path in GRO or remote
90 * checksum offload. For the purposes of checksum verification, the checksum
91 * referred to by skb->csum_start + skb->csum_offset and any preceding
92 * checksums in the packet are considered verified. Any checksums in the
93 * packet that are after the checksum being offloaded are not considered to
94 * be verified.
96 * B. Checksumming on output.
98 * CHECKSUM_NONE:
100 * The skb was already checksummed by the protocol, or a checksum is not
101 * required.
103 * CHECKSUM_PARTIAL:
105 * The device is required to checksum the packet as seen by hard_start_xmit()
106 * from skb->csum_start up to the end, and to record/write the checksum at
107 * offset skb->csum_start + skb->csum_offset.
109 * The device must show its capabilities in dev->features, set up at device
110 * setup time, e.g. netdev_features.h:
112 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything.
113 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over
114 * IPv4. Sigh. Vendors like this way for an unknown reason.
115 * Though, see comment above about CHECKSUM_UNNECESSARY. 8)
116 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead.
117 * NETIF_F_... - Well, you get the picture.
119 * CHECKSUM_UNNECESSARY:
121 * Normally, the device will do per protocol specific checksumming. Protocol
122 * implementations that do not want the NIC to perform the checksum
123 * calculation should use this flag in their outgoing skbs.
125 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC
126 * offload. Correspondingly, the FCoE protocol driver
127 * stack should use CHECKSUM_UNNECESSARY.
129 * Any questions? No questions, good. --ANK
132 /* Don't change this without changing skb_csum_unnecessary! */
133 #define CHECKSUM_NONE 0
134 #define CHECKSUM_UNNECESSARY 1
135 #define CHECKSUM_COMPLETE 2
136 #define CHECKSUM_PARTIAL 3
138 /* Maximum value in skb->csum_level */
139 #define SKB_MAX_CSUM_LEVEL 3
141 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
142 #define SKB_WITH_OVERHEAD(X) \
143 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
144 #define SKB_MAX_ORDER(X, ORDER) \
145 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
146 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
147 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
149 /* return minimum truesize of one skb containing X bytes of data */
150 #define SKB_TRUESIZE(X) ((X) + \
151 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
152 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
154 struct net_device;
155 struct scatterlist;
156 struct pipe_inode_info;
157 struct iov_iter;
158 struct napi_struct;
160 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
161 struct nf_conntrack {
162 atomic_t use;
164 #endif
166 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
167 struct nf_bridge_info {
168 atomic_t use;
169 unsigned int mask;
170 struct net_device *physindev;
171 struct net_device *physoutdev;
172 unsigned long data[32 / sizeof(unsigned long)];
174 #endif
176 struct sk_buff_head {
177 /* These two members must be first. */
178 struct sk_buff *next;
179 struct sk_buff *prev;
181 __u32 qlen;
182 spinlock_t lock;
185 struct sk_buff;
187 /* To allow 64K frame to be packed as single skb without frag_list we
188 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
189 * buffers which do not start on a page boundary.
191 * Since GRO uses frags we allocate at least 16 regardless of page
192 * size.
194 #if (65536/PAGE_SIZE + 1) < 16
195 #define MAX_SKB_FRAGS 16UL
196 #else
197 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
198 #endif
200 typedef struct skb_frag_struct skb_frag_t;
202 struct skb_frag_struct {
203 struct {
204 struct page *p;
205 } page;
206 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
207 __u32 page_offset;
208 __u32 size;
209 #else
210 __u16 page_offset;
211 __u16 size;
212 #endif
215 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
217 return frag->size;
220 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
222 frag->size = size;
225 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
227 frag->size += delta;
230 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
232 frag->size -= delta;
235 #define HAVE_HW_TIME_STAMP
238 * struct skb_shared_hwtstamps - hardware time stamps
239 * @hwtstamp: hardware time stamp transformed into duration
240 * since arbitrary point in time
242 * Software time stamps generated by ktime_get_real() are stored in
243 * skb->tstamp.
245 * hwtstamps can only be compared against other hwtstamps from
246 * the same device.
248 * This structure is attached to packets as part of the
249 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
251 struct skb_shared_hwtstamps {
252 ktime_t hwtstamp;
255 /* Definitions for tx_flags in struct skb_shared_info */
256 enum {
257 /* generate hardware time stamp */
258 SKBTX_HW_TSTAMP = 1 << 0,
260 /* generate software time stamp when queueing packet to NIC */
261 SKBTX_SW_TSTAMP = 1 << 1,
263 /* device driver is going to provide hardware time stamp */
264 SKBTX_IN_PROGRESS = 1 << 2,
266 /* device driver supports TX zero-copy buffers */
267 SKBTX_DEV_ZEROCOPY = 1 << 3,
269 /* generate wifi status information (where possible) */
270 SKBTX_WIFI_STATUS = 1 << 4,
272 /* This indicates at least one fragment might be overwritten
273 * (as in vmsplice(), sendfile() ...)
274 * If we need to compute a TX checksum, we'll need to copy
275 * all frags to avoid possible bad checksum
277 SKBTX_SHARED_FRAG = 1 << 5,
279 /* generate software time stamp when entering packet scheduling */
280 SKBTX_SCHED_TSTAMP = 1 << 6,
282 /* generate software timestamp on peer data acknowledgment */
283 SKBTX_ACK_TSTAMP = 1 << 7,
286 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
287 SKBTX_SCHED_TSTAMP | \
288 SKBTX_ACK_TSTAMP)
289 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
292 * The callback notifies userspace to release buffers when skb DMA is done in
293 * lower device, the skb last reference should be 0 when calling this.
294 * The zerocopy_success argument is true if zero copy transmit occurred,
295 * false on data copy or out of memory error caused by data copy attempt.
296 * The ctx field is used to track device context.
297 * The desc field is used to track userspace buffer index.
299 struct ubuf_info {
300 void (*callback)(struct ubuf_info *, bool zerocopy_success);
301 void *ctx;
302 unsigned long desc;
305 /* This data is invariant across clones and lives at
306 * the end of the header data, ie. at skb->end.
308 struct skb_shared_info {
309 unsigned char nr_frags;
310 __u8 tx_flags;
311 unsigned short gso_size;
312 /* Warning: this field is not always filled in (UFO)! */
313 unsigned short gso_segs;
314 unsigned short gso_type;
315 struct sk_buff *frag_list;
316 struct skb_shared_hwtstamps hwtstamps;
317 u32 tskey;
318 __be32 ip6_frag_id;
321 * Warning : all fields before dataref are cleared in __alloc_skb()
323 atomic_t dataref;
325 /* Intermediate layers must ensure that destructor_arg
326 * remains valid until skb destructor */
327 void * destructor_arg;
329 /* must be last field, see pskb_expand_head() */
330 skb_frag_t frags[MAX_SKB_FRAGS];
333 /* We divide dataref into two halves. The higher 16 bits hold references
334 * to the payload part of skb->data. The lower 16 bits hold references to
335 * the entire skb->data. A clone of a headerless skb holds the length of
336 * the header in skb->hdr_len.
338 * All users must obey the rule that the skb->data reference count must be
339 * greater than or equal to the payload reference count.
341 * Holding a reference to the payload part means that the user does not
342 * care about modifications to the header part of skb->data.
344 #define SKB_DATAREF_SHIFT 16
345 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
348 enum {
349 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
350 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
351 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
354 enum {
355 SKB_GSO_TCPV4 = 1 << 0,
356 SKB_GSO_UDP = 1 << 1,
358 /* This indicates the skb is from an untrusted source. */
359 SKB_GSO_DODGY = 1 << 2,
361 /* This indicates the tcp segment has CWR set. */
362 SKB_GSO_TCP_ECN = 1 << 3,
364 SKB_GSO_TCPV6 = 1 << 4,
366 SKB_GSO_FCOE = 1 << 5,
368 SKB_GSO_GRE = 1 << 6,
370 SKB_GSO_GRE_CSUM = 1 << 7,
372 SKB_GSO_IPIP = 1 << 8,
374 SKB_GSO_SIT = 1 << 9,
376 SKB_GSO_UDP_TUNNEL = 1 << 10,
378 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
380 SKB_GSO_TUNNEL_REMCSUM = 1 << 12,
383 #if BITS_PER_LONG > 32
384 #define NET_SKBUFF_DATA_USES_OFFSET 1
385 #endif
387 #ifdef NET_SKBUFF_DATA_USES_OFFSET
388 typedef unsigned int sk_buff_data_t;
389 #else
390 typedef unsigned char *sk_buff_data_t;
391 #endif
394 * struct skb_mstamp - multi resolution time stamps
395 * @stamp_us: timestamp in us resolution
396 * @stamp_jiffies: timestamp in jiffies
398 struct skb_mstamp {
399 union {
400 u64 v64;
401 struct {
402 u32 stamp_us;
403 u32 stamp_jiffies;
409 * skb_mstamp_get - get current timestamp
410 * @cl: place to store timestamps
412 static inline void skb_mstamp_get(struct skb_mstamp *cl)
414 u64 val = local_clock();
416 do_div(val, NSEC_PER_USEC);
417 cl->stamp_us = (u32)val;
418 cl->stamp_jiffies = (u32)jiffies;
422 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
423 * @t1: pointer to newest sample
424 * @t0: pointer to oldest sample
426 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
427 const struct skb_mstamp *t0)
429 s32 delta_us = t1->stamp_us - t0->stamp_us;
430 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
432 /* If delta_us is negative, this might be because interval is too big,
433 * or local_clock() drift is too big : fallback using jiffies.
435 if (delta_us <= 0 ||
436 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
438 delta_us = jiffies_to_usecs(delta_jiffies);
440 return delta_us;
444 /**
445 * struct sk_buff - socket buffer
446 * @next: Next buffer in list
447 * @prev: Previous buffer in list
448 * @tstamp: Time we arrived/left
449 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
450 * @sk: Socket we are owned by
451 * @dev: Device we arrived on/are leaving by
452 * @cb: Control buffer. Free for use by every layer. Put private vars here
453 * @_skb_refdst: destination entry (with norefcount bit)
454 * @sp: the security path, used for xfrm
455 * @len: Length of actual data
456 * @data_len: Data length
457 * @mac_len: Length of link layer header
458 * @hdr_len: writable header length of cloned skb
459 * @csum: Checksum (must include start/offset pair)
460 * @csum_start: Offset from skb->head where checksumming should start
461 * @csum_offset: Offset from csum_start where checksum should be stored
462 * @priority: Packet queueing priority
463 * @ignore_df: allow local fragmentation
464 * @cloned: Head may be cloned (check refcnt to be sure)
465 * @ip_summed: Driver fed us an IP checksum
466 * @nohdr: Payload reference only, must not modify header
467 * @nfctinfo: Relationship of this skb to the connection
468 * @pkt_type: Packet class
469 * @fclone: skbuff clone status
470 * @ipvs_property: skbuff is owned by ipvs
471 * @peeked: this packet has been seen already, so stats have been
472 * done for it, don't do them again
473 * @nf_trace: netfilter packet trace flag
474 * @protocol: Packet protocol from driver
475 * @destructor: Destruct function
476 * @nfct: Associated connection, if any
477 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
478 * @skb_iif: ifindex of device we arrived on
479 * @tc_index: Traffic control index
480 * @tc_verd: traffic control verdict
481 * @hash: the packet hash
482 * @queue_mapping: Queue mapping for multiqueue devices
483 * @xmit_more: More SKBs are pending for this queue
484 * @ndisc_nodetype: router type (from link layer)
485 * @ooo_okay: allow the mapping of a socket to a queue to be changed
486 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
487 * ports.
488 * @sw_hash: indicates hash was computed in software stack
489 * @wifi_acked_valid: wifi_acked was set
490 * @wifi_acked: whether frame was acked on wifi or not
491 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
492 * @napi_id: id of the NAPI struct this skb came from
493 * @secmark: security marking
494 * @mark: Generic packet mark
495 * @vlan_proto: vlan encapsulation protocol
496 * @vlan_tci: vlan tag control information
497 * @inner_protocol: Protocol (encapsulation)
498 * @inner_transport_header: Inner transport layer header (encapsulation)
499 * @inner_network_header: Network layer header (encapsulation)
500 * @inner_mac_header: Link layer header (encapsulation)
501 * @transport_header: Transport layer header
502 * @network_header: Network layer header
503 * @mac_header: Link layer header
504 * @tail: Tail pointer
505 * @end: End pointer
506 * @head: Head of buffer
507 * @data: Data head pointer
508 * @truesize: Buffer size
509 * @users: User count - see {datagram,tcp}.c
512 struct sk_buff {
513 union {
514 struct {
515 /* These two members must be first. */
516 struct sk_buff *next;
517 struct sk_buff *prev;
519 union {
520 ktime_t tstamp;
521 struct skb_mstamp skb_mstamp;
524 struct rb_node rbnode; /* used in netem & tcp stack */
526 struct sock *sk;
527 struct net_device *dev;
530 * This is the control buffer. It is free to use for every
531 * layer. Please put your private variables there. If you
532 * want to keep them across layers you have to do a skb_clone()
533 * first. This is owned by whoever has the skb queued ATM.
535 char cb[48] __aligned(8);
537 unsigned long _skb_refdst;
538 void (*destructor)(struct sk_buff *skb);
539 #ifdef CONFIG_XFRM
540 struct sec_path *sp;
541 #endif
542 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
543 struct nf_conntrack *nfct;
544 #endif
545 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
546 struct nf_bridge_info *nf_bridge;
547 #endif
548 unsigned int len,
549 data_len;
550 __u16 mac_len,
551 hdr_len;
553 /* Following fields are _not_ copied in __copy_skb_header()
554 * Note that queue_mapping is here mostly to fill a hole.
556 kmemcheck_bitfield_begin(flags1);
557 __u16 queue_mapping;
558 __u8 cloned:1,
559 nohdr:1,
560 fclone:2,
561 peeked:1,
562 head_frag:1,
563 xmit_more:1;
564 /* one bit hole */
565 kmemcheck_bitfield_end(flags1);
567 /* fields enclosed in headers_start/headers_end are copied
568 * using a single memcpy() in __copy_skb_header()
570 /* private: */
571 __u32 headers_start[0];
572 /* public: */
574 /* if you move pkt_type around you also must adapt those constants */
575 #ifdef __BIG_ENDIAN_BITFIELD
576 #define PKT_TYPE_MAX (7 << 5)
577 #else
578 #define PKT_TYPE_MAX 7
579 #endif
580 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
582 __u8 __pkt_type_offset[0];
583 __u8 pkt_type:3;
584 __u8 pfmemalloc:1;
585 __u8 ignore_df:1;
586 __u8 nfctinfo:3;
588 __u8 nf_trace:1;
589 __u8 ip_summed:2;
590 __u8 ooo_okay:1;
591 __u8 l4_hash:1;
592 __u8 sw_hash:1;
593 __u8 wifi_acked_valid:1;
594 __u8 wifi_acked:1;
596 __u8 no_fcs:1;
597 /* Indicates the inner headers are valid in the skbuff. */
598 __u8 encapsulation:1;
599 __u8 encap_hdr_csum:1;
600 __u8 csum_valid:1;
601 __u8 csum_complete_sw:1;
602 __u8 csum_level:2;
603 __u8 csum_bad:1;
605 #ifdef CONFIG_IPV6_NDISC_NODETYPE
606 __u8 ndisc_nodetype:2;
607 #endif
608 __u8 ipvs_property:1;
609 __u8 inner_protocol_type:1;
610 __u8 remcsum_offload:1;
611 /* 3 or 5 bit hole */
613 #ifdef CONFIG_NET_SCHED
614 __u16 tc_index; /* traffic control index */
615 #ifdef CONFIG_NET_CLS_ACT
616 __u16 tc_verd; /* traffic control verdict */
617 #endif
618 #endif
620 union {
621 __wsum csum;
622 struct {
623 __u16 csum_start;
624 __u16 csum_offset;
627 __u32 priority;
628 int skb_iif;
629 __u32 hash;
630 __be16 vlan_proto;
631 __u16 vlan_tci;
632 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
633 union {
634 unsigned int napi_id;
635 unsigned int sender_cpu;
637 #endif
638 #ifdef CONFIG_NETWORK_SECMARK
639 __u32 secmark;
640 #endif
641 union {
642 __u32 mark;
643 __u32 reserved_tailroom;
646 union {
647 __be16 inner_protocol;
648 __u8 inner_ipproto;
651 __u16 inner_transport_header;
652 __u16 inner_network_header;
653 __u16 inner_mac_header;
655 __be16 protocol;
656 __u16 transport_header;
657 __u16 network_header;
658 __u16 mac_header;
660 /* private: */
661 __u32 headers_end[0];
662 /* public: */
664 /* These elements must be at the end, see alloc_skb() for details. */
665 sk_buff_data_t tail;
666 sk_buff_data_t end;
667 unsigned char *head,
668 *data;
669 unsigned int truesize;
670 atomic_t users;
673 #ifdef __KERNEL__
675 * Handling routines are only of interest to the kernel
677 #include <linux/slab.h>
680 #define SKB_ALLOC_FCLONE 0x01
681 #define SKB_ALLOC_RX 0x02
682 #define SKB_ALLOC_NAPI 0x04
684 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
685 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
687 return unlikely(skb->pfmemalloc);
691 * skb might have a dst pointer attached, refcounted or not.
692 * _skb_refdst low order bit is set if refcount was _not_ taken
694 #define SKB_DST_NOREF 1UL
695 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
698 * skb_dst - returns skb dst_entry
699 * @skb: buffer
701 * Returns skb dst_entry, regardless of reference taken or not.
703 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
705 /* If refdst was not refcounted, check we still are in a
706 * rcu_read_lock section
708 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
709 !rcu_read_lock_held() &&
710 !rcu_read_lock_bh_held());
711 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
715 * skb_dst_set - sets skb dst
716 * @skb: buffer
717 * @dst: dst entry
719 * Sets skb dst, assuming a reference was taken on dst and should
720 * be released by skb_dst_drop()
722 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
724 skb->_skb_refdst = (unsigned long)dst;
728 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
729 * @skb: buffer
730 * @dst: dst entry
732 * Sets skb dst, assuming a reference was not taken on dst.
733 * If dst entry is cached, we do not take reference and dst_release
734 * will be avoided by refdst_drop. If dst entry is not cached, we take
735 * reference, so that last dst_release can destroy the dst immediately.
737 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
739 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
740 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
744 * skb_dst_is_noref - Test if skb dst isn't refcounted
745 * @skb: buffer
747 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
749 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
752 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
754 return (struct rtable *)skb_dst(skb);
757 void kfree_skb(struct sk_buff *skb);
758 void kfree_skb_list(struct sk_buff *segs);
759 void skb_tx_error(struct sk_buff *skb);
760 void consume_skb(struct sk_buff *skb);
761 void __kfree_skb(struct sk_buff *skb);
762 extern struct kmem_cache *skbuff_head_cache;
764 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
765 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
766 bool *fragstolen, int *delta_truesize);
768 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
769 int node);
770 struct sk_buff *build_skb(void *data, unsigned int frag_size);
771 static inline struct sk_buff *alloc_skb(unsigned int size,
772 gfp_t priority)
774 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
777 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
778 unsigned long data_len,
779 int max_page_order,
780 int *errcode,
781 gfp_t gfp_mask);
783 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
784 struct sk_buff_fclones {
785 struct sk_buff skb1;
787 struct sk_buff skb2;
789 atomic_t fclone_ref;
793 * skb_fclone_busy - check if fclone is busy
794 * @skb: buffer
796 * Returns true is skb is a fast clone, and its clone is not freed.
797 * Some drivers call skb_orphan() in their ndo_start_xmit(),
798 * so we also check that this didnt happen.
800 static inline bool skb_fclone_busy(const struct sock *sk,
801 const struct sk_buff *skb)
803 const struct sk_buff_fclones *fclones;
805 fclones = container_of(skb, struct sk_buff_fclones, skb1);
807 return skb->fclone == SKB_FCLONE_ORIG &&
808 atomic_read(&fclones->fclone_ref) > 1 &&
809 fclones->skb2.sk == sk;
812 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
813 gfp_t priority)
815 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
818 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
819 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
821 return __alloc_skb_head(priority, -1);
824 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
825 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
826 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
827 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
828 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
829 gfp_t gfp_mask, bool fclone);
830 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
831 gfp_t gfp_mask)
833 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
836 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
837 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
838 unsigned int headroom);
839 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
840 int newtailroom, gfp_t priority);
841 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
842 int offset, int len);
843 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
844 int len);
845 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
846 int skb_pad(struct sk_buff *skb, int pad);
847 #define dev_kfree_skb(a) consume_skb(a)
849 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
850 int getfrag(void *from, char *to, int offset,
851 int len, int odd, struct sk_buff *skb),
852 void *from, int length);
854 struct skb_seq_state {
855 __u32 lower_offset;
856 __u32 upper_offset;
857 __u32 frag_idx;
858 __u32 stepped_offset;
859 struct sk_buff *root_skb;
860 struct sk_buff *cur_skb;
861 __u8 *frag_data;
864 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
865 unsigned int to, struct skb_seq_state *st);
866 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
867 struct skb_seq_state *st);
868 void skb_abort_seq_read(struct skb_seq_state *st);
870 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
871 unsigned int to, struct ts_config *config);
874 * Packet hash types specify the type of hash in skb_set_hash.
876 * Hash types refer to the protocol layer addresses which are used to
877 * construct a packet's hash. The hashes are used to differentiate or identify
878 * flows of the protocol layer for the hash type. Hash types are either
879 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
881 * Properties of hashes:
883 * 1) Two packets in different flows have different hash values
884 * 2) Two packets in the same flow should have the same hash value
886 * A hash at a higher layer is considered to be more specific. A driver should
887 * set the most specific hash possible.
889 * A driver cannot indicate a more specific hash than the layer at which a hash
890 * was computed. For instance an L3 hash cannot be set as an L4 hash.
892 * A driver may indicate a hash level which is less specific than the
893 * actual layer the hash was computed on. For instance, a hash computed
894 * at L4 may be considered an L3 hash. This should only be done if the
895 * driver can't unambiguously determine that the HW computed the hash at
896 * the higher layer. Note that the "should" in the second property above
897 * permits this.
899 enum pkt_hash_types {
900 PKT_HASH_TYPE_NONE, /* Undefined type */
901 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
902 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
903 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
906 static inline void
907 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
909 skb->l4_hash = (type == PKT_HASH_TYPE_L4);
910 skb->sw_hash = 0;
911 skb->hash = hash;
914 void __skb_get_hash(struct sk_buff *skb);
915 static inline __u32 skb_get_hash(struct sk_buff *skb)
917 if (!skb->l4_hash && !skb->sw_hash)
918 __skb_get_hash(skb);
920 return skb->hash;
923 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
925 return skb->hash;
928 static inline void skb_clear_hash(struct sk_buff *skb)
930 skb->hash = 0;
931 skb->sw_hash = 0;
932 skb->l4_hash = 0;
935 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
937 if (!skb->l4_hash)
938 skb_clear_hash(skb);
941 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
943 to->hash = from->hash;
944 to->sw_hash = from->sw_hash;
945 to->l4_hash = from->l4_hash;
948 #ifdef NET_SKBUFF_DATA_USES_OFFSET
949 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
951 return skb->head + skb->end;
954 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
956 return skb->end;
958 #else
959 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
961 return skb->end;
964 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
966 return skb->end - skb->head;
968 #endif
970 /* Internal */
971 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
973 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
975 return &skb_shinfo(skb)->hwtstamps;
979 * skb_queue_empty - check if a queue is empty
980 * @list: queue head
982 * Returns true if the queue is empty, false otherwise.
984 static inline int skb_queue_empty(const struct sk_buff_head *list)
986 return list->next == (const struct sk_buff *) list;
990 * skb_queue_is_last - check if skb is the last entry in the queue
991 * @list: queue head
992 * @skb: buffer
994 * Returns true if @skb is the last buffer on the list.
996 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
997 const struct sk_buff *skb)
999 return skb->next == (const struct sk_buff *) list;
1003 * skb_queue_is_first - check if skb is the first entry in the queue
1004 * @list: queue head
1005 * @skb: buffer
1007 * Returns true if @skb is the first buffer on the list.
1009 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1010 const struct sk_buff *skb)
1012 return skb->prev == (const struct sk_buff *) list;
1016 * skb_queue_next - return the next packet in the queue
1017 * @list: queue head
1018 * @skb: current buffer
1020 * Return the next packet in @list after @skb. It is only valid to
1021 * call this if skb_queue_is_last() evaluates to false.
1023 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1024 const struct sk_buff *skb)
1026 /* This BUG_ON may seem severe, but if we just return then we
1027 * are going to dereference garbage.
1029 BUG_ON(skb_queue_is_last(list, skb));
1030 return skb->next;
1034 * skb_queue_prev - return the prev packet in the queue
1035 * @list: queue head
1036 * @skb: current buffer
1038 * Return the prev packet in @list before @skb. It is only valid to
1039 * call this if skb_queue_is_first() evaluates to false.
1041 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1042 const struct sk_buff *skb)
1044 /* This BUG_ON may seem severe, but if we just return then we
1045 * are going to dereference garbage.
1047 BUG_ON(skb_queue_is_first(list, skb));
1048 return skb->prev;
1052 * skb_get - reference buffer
1053 * @skb: buffer to reference
1055 * Makes another reference to a socket buffer and returns a pointer
1056 * to the buffer.
1058 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1060 atomic_inc(&skb->users);
1061 return skb;
1065 * If users == 1, we are the only owner and are can avoid redundant
1066 * atomic change.
1070 * skb_cloned - is the buffer a clone
1071 * @skb: buffer to check
1073 * Returns true if the buffer was generated with skb_clone() and is
1074 * one of multiple shared copies of the buffer. Cloned buffers are
1075 * shared data so must not be written to under normal circumstances.
1077 static inline int skb_cloned(const struct sk_buff *skb)
1079 return skb->cloned &&
1080 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1083 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1085 might_sleep_if(pri & __GFP_WAIT);
1087 if (skb_cloned(skb))
1088 return pskb_expand_head(skb, 0, 0, pri);
1090 return 0;
1094 * skb_header_cloned - is the header a clone
1095 * @skb: buffer to check
1097 * Returns true if modifying the header part of the buffer requires
1098 * the data to be copied.
1100 static inline int skb_header_cloned(const struct sk_buff *skb)
1102 int dataref;
1104 if (!skb->cloned)
1105 return 0;
1107 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1108 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1109 return dataref != 1;
1113 * skb_header_release - release reference to header
1114 * @skb: buffer to operate on
1116 * Drop a reference to the header part of the buffer. This is done
1117 * by acquiring a payload reference. You must not read from the header
1118 * part of skb->data after this.
1119 * Note : Check if you can use __skb_header_release() instead.
1121 static inline void skb_header_release(struct sk_buff *skb)
1123 BUG_ON(skb->nohdr);
1124 skb->nohdr = 1;
1125 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1129 * __skb_header_release - release reference to header
1130 * @skb: buffer to operate on
1132 * Variant of skb_header_release() assuming skb is private to caller.
1133 * We can avoid one atomic operation.
1135 static inline void __skb_header_release(struct sk_buff *skb)
1137 skb->nohdr = 1;
1138 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1143 * skb_shared - is the buffer shared
1144 * @skb: buffer to check
1146 * Returns true if more than one person has a reference to this
1147 * buffer.
1149 static inline int skb_shared(const struct sk_buff *skb)
1151 return atomic_read(&skb->users) != 1;
1155 * skb_share_check - check if buffer is shared and if so clone it
1156 * @skb: buffer to check
1157 * @pri: priority for memory allocation
1159 * If the buffer is shared the buffer is cloned and the old copy
1160 * drops a reference. A new clone with a single reference is returned.
1161 * If the buffer is not shared the original buffer is returned. When
1162 * being called from interrupt status or with spinlocks held pri must
1163 * be GFP_ATOMIC.
1165 * NULL is returned on a memory allocation failure.
1167 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1169 might_sleep_if(pri & __GFP_WAIT);
1170 if (skb_shared(skb)) {
1171 struct sk_buff *nskb = skb_clone(skb, pri);
1173 if (likely(nskb))
1174 consume_skb(skb);
1175 else
1176 kfree_skb(skb);
1177 skb = nskb;
1179 return skb;
1183 * Copy shared buffers into a new sk_buff. We effectively do COW on
1184 * packets to handle cases where we have a local reader and forward
1185 * and a couple of other messy ones. The normal one is tcpdumping
1186 * a packet thats being forwarded.
1190 * skb_unshare - make a copy of a shared buffer
1191 * @skb: buffer to check
1192 * @pri: priority for memory allocation
1194 * If the socket buffer is a clone then this function creates a new
1195 * copy of the data, drops a reference count on the old copy and returns
1196 * the new copy with the reference count at 1. If the buffer is not a clone
1197 * the original buffer is returned. When called with a spinlock held or
1198 * from interrupt state @pri must be %GFP_ATOMIC
1200 * %NULL is returned on a memory allocation failure.
1202 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1203 gfp_t pri)
1205 might_sleep_if(pri & __GFP_WAIT);
1206 if (skb_cloned(skb)) {
1207 struct sk_buff *nskb = skb_copy(skb, pri);
1209 /* Free our shared copy */
1210 if (likely(nskb))
1211 consume_skb(skb);
1212 else
1213 kfree_skb(skb);
1214 skb = nskb;
1216 return skb;
1220 * skb_peek - peek at the head of an &sk_buff_head
1221 * @list_: list to peek at
1223 * Peek an &sk_buff. Unlike most other operations you _MUST_
1224 * be careful with this one. A peek leaves the buffer on the
1225 * list and someone else may run off with it. You must hold
1226 * the appropriate locks or have a private queue to do this.
1228 * Returns %NULL for an empty list or a pointer to the head element.
1229 * The reference count is not incremented and the reference is therefore
1230 * volatile. Use with caution.
1232 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1234 struct sk_buff *skb = list_->next;
1236 if (skb == (struct sk_buff *)list_)
1237 skb = NULL;
1238 return skb;
1242 * skb_peek_next - peek skb following the given one from a queue
1243 * @skb: skb to start from
1244 * @list_: list to peek at
1246 * Returns %NULL when the end of the list is met or a pointer to the
1247 * next element. The reference count is not incremented and the
1248 * reference is therefore volatile. Use with caution.
1250 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1251 const struct sk_buff_head *list_)
1253 struct sk_buff *next = skb->next;
1255 if (next == (struct sk_buff *)list_)
1256 next = NULL;
1257 return next;
1261 * skb_peek_tail - peek at the tail of an &sk_buff_head
1262 * @list_: list to peek at
1264 * Peek an &sk_buff. Unlike most other operations you _MUST_
1265 * be careful with this one. A peek leaves the buffer on the
1266 * list and someone else may run off with it. You must hold
1267 * the appropriate locks or have a private queue to do this.
1269 * Returns %NULL for an empty list or a pointer to the tail element.
1270 * The reference count is not incremented and the reference is therefore
1271 * volatile. Use with caution.
1273 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1275 struct sk_buff *skb = list_->prev;
1277 if (skb == (struct sk_buff *)list_)
1278 skb = NULL;
1279 return skb;
1284 * skb_queue_len - get queue length
1285 * @list_: list to measure
1287 * Return the length of an &sk_buff queue.
1289 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1291 return list_->qlen;
1295 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1296 * @list: queue to initialize
1298 * This initializes only the list and queue length aspects of
1299 * an sk_buff_head object. This allows to initialize the list
1300 * aspects of an sk_buff_head without reinitializing things like
1301 * the spinlock. It can also be used for on-stack sk_buff_head
1302 * objects where the spinlock is known to not be used.
1304 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1306 list->prev = list->next = (struct sk_buff *)list;
1307 list->qlen = 0;
1311 * This function creates a split out lock class for each invocation;
1312 * this is needed for now since a whole lot of users of the skb-queue
1313 * infrastructure in drivers have different locking usage (in hardirq)
1314 * than the networking core (in softirq only). In the long run either the
1315 * network layer or drivers should need annotation to consolidate the
1316 * main types of usage into 3 classes.
1318 static inline void skb_queue_head_init(struct sk_buff_head *list)
1320 spin_lock_init(&list->lock);
1321 __skb_queue_head_init(list);
1324 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1325 struct lock_class_key *class)
1327 skb_queue_head_init(list);
1328 lockdep_set_class(&list->lock, class);
1332 * Insert an sk_buff on a list.
1334 * The "__skb_xxxx()" functions are the non-atomic ones that
1335 * can only be called with interrupts disabled.
1337 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1338 struct sk_buff_head *list);
1339 static inline void __skb_insert(struct sk_buff *newsk,
1340 struct sk_buff *prev, struct sk_buff *next,
1341 struct sk_buff_head *list)
1343 newsk->next = next;
1344 newsk->prev = prev;
1345 next->prev = prev->next = newsk;
1346 list->qlen++;
1349 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1350 struct sk_buff *prev,
1351 struct sk_buff *next)
1353 struct sk_buff *first = list->next;
1354 struct sk_buff *last = list->prev;
1356 first->prev = prev;
1357 prev->next = first;
1359 last->next = next;
1360 next->prev = last;
1364 * skb_queue_splice - join two skb lists, this is designed for stacks
1365 * @list: the new list to add
1366 * @head: the place to add it in the first list
1368 static inline void skb_queue_splice(const struct sk_buff_head *list,
1369 struct sk_buff_head *head)
1371 if (!skb_queue_empty(list)) {
1372 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1373 head->qlen += list->qlen;
1378 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1379 * @list: the new list to add
1380 * @head: the place to add it in the first list
1382 * The list at @list is reinitialised
1384 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1385 struct sk_buff_head *head)
1387 if (!skb_queue_empty(list)) {
1388 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1389 head->qlen += list->qlen;
1390 __skb_queue_head_init(list);
1395 * skb_queue_splice_tail - join two skb lists, each list being a queue
1396 * @list: the new list to add
1397 * @head: the place to add it in the first list
1399 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1400 struct sk_buff_head *head)
1402 if (!skb_queue_empty(list)) {
1403 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1404 head->qlen += list->qlen;
1409 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1410 * @list: the new list to add
1411 * @head: the place to add it in the first list
1413 * Each of the lists is a queue.
1414 * The list at @list is reinitialised
1416 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1417 struct sk_buff_head *head)
1419 if (!skb_queue_empty(list)) {
1420 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1421 head->qlen += list->qlen;
1422 __skb_queue_head_init(list);
1427 * __skb_queue_after - queue a buffer at the list head
1428 * @list: list to use
1429 * @prev: place after this buffer
1430 * @newsk: buffer to queue
1432 * Queue a buffer int the middle of a list. This function takes no locks
1433 * and you must therefore hold required locks before calling it.
1435 * A buffer cannot be placed on two lists at the same time.
1437 static inline void __skb_queue_after(struct sk_buff_head *list,
1438 struct sk_buff *prev,
1439 struct sk_buff *newsk)
1441 __skb_insert(newsk, prev, prev->next, list);
1444 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1445 struct sk_buff_head *list);
1447 static inline void __skb_queue_before(struct sk_buff_head *list,
1448 struct sk_buff *next,
1449 struct sk_buff *newsk)
1451 __skb_insert(newsk, next->prev, next, list);
1455 * __skb_queue_head - queue a buffer at the list head
1456 * @list: list to use
1457 * @newsk: buffer to queue
1459 * Queue a buffer at the start of a list. This function takes no locks
1460 * and you must therefore hold required locks before calling it.
1462 * A buffer cannot be placed on two lists at the same time.
1464 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1465 static inline void __skb_queue_head(struct sk_buff_head *list,
1466 struct sk_buff *newsk)
1468 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1472 * __skb_queue_tail - queue a buffer at the list tail
1473 * @list: list to use
1474 * @newsk: buffer to queue
1476 * Queue a buffer at the end of a list. This function takes no locks
1477 * and you must therefore hold required locks before calling it.
1479 * A buffer cannot be placed on two lists at the same time.
1481 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1482 static inline void __skb_queue_tail(struct sk_buff_head *list,
1483 struct sk_buff *newsk)
1485 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1489 * remove sk_buff from list. _Must_ be called atomically, and with
1490 * the list known..
1492 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1493 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1495 struct sk_buff *next, *prev;
1497 list->qlen--;
1498 next = skb->next;
1499 prev = skb->prev;
1500 skb->next = skb->prev = NULL;
1501 next->prev = prev;
1502 prev->next = next;
1506 * __skb_dequeue - remove from the head of the queue
1507 * @list: list to dequeue from
1509 * Remove the head of the list. This function does not take any locks
1510 * so must be used with appropriate locks held only. The head item is
1511 * returned or %NULL if the list is empty.
1513 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1514 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1516 struct sk_buff *skb = skb_peek(list);
1517 if (skb)
1518 __skb_unlink(skb, list);
1519 return skb;
1523 * __skb_dequeue_tail - remove from the tail of the queue
1524 * @list: list to dequeue from
1526 * Remove the tail of the list. This function does not take any locks
1527 * so must be used with appropriate locks held only. The tail item is
1528 * returned or %NULL if the list is empty.
1530 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1531 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1533 struct sk_buff *skb = skb_peek_tail(list);
1534 if (skb)
1535 __skb_unlink(skb, list);
1536 return skb;
1540 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1542 return skb->data_len;
1545 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1547 return skb->len - skb->data_len;
1550 static inline int skb_pagelen(const struct sk_buff *skb)
1552 int i, len = 0;
1554 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1555 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1556 return len + skb_headlen(skb);
1560 * __skb_fill_page_desc - initialise a paged fragment in an skb
1561 * @skb: buffer containing fragment to be initialised
1562 * @i: paged fragment index to initialise
1563 * @page: the page to use for this fragment
1564 * @off: the offset to the data with @page
1565 * @size: the length of the data
1567 * Initialises the @i'th fragment of @skb to point to &size bytes at
1568 * offset @off within @page.
1570 * Does not take any additional reference on the fragment.
1572 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1573 struct page *page, int off, int size)
1575 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1578 * Propagate page->pfmemalloc to the skb if we can. The problem is
1579 * that not all callers have unique ownership of the page. If
1580 * pfmemalloc is set, we check the mapping as a mapping implies
1581 * page->index is set (index and pfmemalloc share space).
1582 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1583 * do not lose pfmemalloc information as the pages would not be
1584 * allocated using __GFP_MEMALLOC.
1586 frag->page.p = page;
1587 frag->page_offset = off;
1588 skb_frag_size_set(frag, size);
1590 page = compound_head(page);
1591 if (page->pfmemalloc && !page->mapping)
1592 skb->pfmemalloc = true;
1596 * skb_fill_page_desc - initialise a paged fragment in an skb
1597 * @skb: buffer containing fragment to be initialised
1598 * @i: paged fragment index to initialise
1599 * @page: the page to use for this fragment
1600 * @off: the offset to the data with @page
1601 * @size: the length of the data
1603 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1604 * @skb to point to @size bytes at offset @off within @page. In
1605 * addition updates @skb such that @i is the last fragment.
1607 * Does not take any additional reference on the fragment.
1609 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1610 struct page *page, int off, int size)
1612 __skb_fill_page_desc(skb, i, page, off, size);
1613 skb_shinfo(skb)->nr_frags = i + 1;
1616 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1617 int size, unsigned int truesize);
1619 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1620 unsigned int truesize);
1622 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1623 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1624 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1626 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1627 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1629 return skb->head + skb->tail;
1632 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1634 skb->tail = skb->data - skb->head;
1637 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1639 skb_reset_tail_pointer(skb);
1640 skb->tail += offset;
1643 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1644 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1646 return skb->tail;
1649 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1651 skb->tail = skb->data;
1654 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1656 skb->tail = skb->data + offset;
1659 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1662 * Add data to an sk_buff
1664 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1665 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1666 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1668 unsigned char *tmp = skb_tail_pointer(skb);
1669 SKB_LINEAR_ASSERT(skb);
1670 skb->tail += len;
1671 skb->len += len;
1672 return tmp;
1675 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1676 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1678 skb->data -= len;
1679 skb->len += len;
1680 return skb->data;
1683 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1684 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1686 skb->len -= len;
1687 BUG_ON(skb->len < skb->data_len);
1688 return skb->data += len;
1691 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1693 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1696 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1698 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1700 if (len > skb_headlen(skb) &&
1701 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1702 return NULL;
1703 skb->len -= len;
1704 return skb->data += len;
1707 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1709 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1712 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1714 if (likely(len <= skb_headlen(skb)))
1715 return 1;
1716 if (unlikely(len > skb->len))
1717 return 0;
1718 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1722 * skb_headroom - bytes at buffer head
1723 * @skb: buffer to check
1725 * Return the number of bytes of free space at the head of an &sk_buff.
1727 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1729 return skb->data - skb->head;
1733 * skb_tailroom - bytes at buffer end
1734 * @skb: buffer to check
1736 * Return the number of bytes of free space at the tail of an sk_buff
1738 static inline int skb_tailroom(const struct sk_buff *skb)
1740 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1744 * skb_availroom - bytes at buffer end
1745 * @skb: buffer to check
1747 * Return the number of bytes of free space at the tail of an sk_buff
1748 * allocated by sk_stream_alloc()
1750 static inline int skb_availroom(const struct sk_buff *skb)
1752 if (skb_is_nonlinear(skb))
1753 return 0;
1755 return skb->end - skb->tail - skb->reserved_tailroom;
1759 * skb_reserve - adjust headroom
1760 * @skb: buffer to alter
1761 * @len: bytes to move
1763 * Increase the headroom of an empty &sk_buff by reducing the tail
1764 * room. This is only allowed for an empty buffer.
1766 static inline void skb_reserve(struct sk_buff *skb, int len)
1768 skb->data += len;
1769 skb->tail += len;
1772 #define ENCAP_TYPE_ETHER 0
1773 #define ENCAP_TYPE_IPPROTO 1
1775 static inline void skb_set_inner_protocol(struct sk_buff *skb,
1776 __be16 protocol)
1778 skb->inner_protocol = protocol;
1779 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
1782 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
1783 __u8 ipproto)
1785 skb->inner_ipproto = ipproto;
1786 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
1789 static inline void skb_reset_inner_headers(struct sk_buff *skb)
1791 skb->inner_mac_header = skb->mac_header;
1792 skb->inner_network_header = skb->network_header;
1793 skb->inner_transport_header = skb->transport_header;
1796 static inline void skb_reset_mac_len(struct sk_buff *skb)
1798 skb->mac_len = skb->network_header - skb->mac_header;
1801 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1802 *skb)
1804 return skb->head + skb->inner_transport_header;
1807 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1809 skb->inner_transport_header = skb->data - skb->head;
1812 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1813 const int offset)
1815 skb_reset_inner_transport_header(skb);
1816 skb->inner_transport_header += offset;
1819 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1821 return skb->head + skb->inner_network_header;
1824 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1826 skb->inner_network_header = skb->data - skb->head;
1829 static inline void skb_set_inner_network_header(struct sk_buff *skb,
1830 const int offset)
1832 skb_reset_inner_network_header(skb);
1833 skb->inner_network_header += offset;
1836 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1838 return skb->head + skb->inner_mac_header;
1841 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1843 skb->inner_mac_header = skb->data - skb->head;
1846 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1847 const int offset)
1849 skb_reset_inner_mac_header(skb);
1850 skb->inner_mac_header += offset;
1852 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1854 return skb->transport_header != (typeof(skb->transport_header))~0U;
1857 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1859 return skb->head + skb->transport_header;
1862 static inline void skb_reset_transport_header(struct sk_buff *skb)
1864 skb->transport_header = skb->data - skb->head;
1867 static inline void skb_set_transport_header(struct sk_buff *skb,
1868 const int offset)
1870 skb_reset_transport_header(skb);
1871 skb->transport_header += offset;
1874 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1876 return skb->head + skb->network_header;
1879 static inline void skb_reset_network_header(struct sk_buff *skb)
1881 skb->network_header = skb->data - skb->head;
1884 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1886 skb_reset_network_header(skb);
1887 skb->network_header += offset;
1890 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1892 return skb->head + skb->mac_header;
1895 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1897 return skb->mac_header != (typeof(skb->mac_header))~0U;
1900 static inline void skb_reset_mac_header(struct sk_buff *skb)
1902 skb->mac_header = skb->data - skb->head;
1905 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1907 skb_reset_mac_header(skb);
1908 skb->mac_header += offset;
1911 static inline void skb_pop_mac_header(struct sk_buff *skb)
1913 skb->mac_header = skb->network_header;
1916 static inline void skb_probe_transport_header(struct sk_buff *skb,
1917 const int offset_hint)
1919 struct flow_keys keys;
1921 if (skb_transport_header_was_set(skb))
1922 return;
1923 else if (skb_flow_dissect(skb, &keys))
1924 skb_set_transport_header(skb, keys.thoff);
1925 else
1926 skb_set_transport_header(skb, offset_hint);
1929 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1931 if (skb_mac_header_was_set(skb)) {
1932 const unsigned char *old_mac = skb_mac_header(skb);
1934 skb_set_mac_header(skb, -skb->mac_len);
1935 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
1939 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
1941 return skb->csum_start - skb_headroom(skb);
1944 static inline int skb_transport_offset(const struct sk_buff *skb)
1946 return skb_transport_header(skb) - skb->data;
1949 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1951 return skb->transport_header - skb->network_header;
1954 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
1956 return skb->inner_transport_header - skb->inner_network_header;
1959 static inline int skb_network_offset(const struct sk_buff *skb)
1961 return skb_network_header(skb) - skb->data;
1964 static inline int skb_inner_network_offset(const struct sk_buff *skb)
1966 return skb_inner_network_header(skb) - skb->data;
1969 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1971 return pskb_may_pull(skb, skb_network_offset(skb) + len);
1975 * CPUs often take a performance hit when accessing unaligned memory
1976 * locations. The actual performance hit varies, it can be small if the
1977 * hardware handles it or large if we have to take an exception and fix it
1978 * in software.
1980 * Since an ethernet header is 14 bytes network drivers often end up with
1981 * the IP header at an unaligned offset. The IP header can be aligned by
1982 * shifting the start of the packet by 2 bytes. Drivers should do this
1983 * with:
1985 * skb_reserve(skb, NET_IP_ALIGN);
1987 * The downside to this alignment of the IP header is that the DMA is now
1988 * unaligned. On some architectures the cost of an unaligned DMA is high
1989 * and this cost outweighs the gains made by aligning the IP header.
1991 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1992 * to be overridden.
1994 #ifndef NET_IP_ALIGN
1995 #define NET_IP_ALIGN 2
1996 #endif
1999 * The networking layer reserves some headroom in skb data (via
2000 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2001 * the header has to grow. In the default case, if the header has to grow
2002 * 32 bytes or less we avoid the reallocation.
2004 * Unfortunately this headroom changes the DMA alignment of the resulting
2005 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2006 * on some architectures. An architecture can override this value,
2007 * perhaps setting it to a cacheline in size (since that will maintain
2008 * cacheline alignment of the DMA). It must be a power of 2.
2010 * Various parts of the networking layer expect at least 32 bytes of
2011 * headroom, you should not reduce this.
2013 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2014 * to reduce average number of cache lines per packet.
2015 * get_rps_cpus() for example only access one 64 bytes aligned block :
2016 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2018 #ifndef NET_SKB_PAD
2019 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2020 #endif
2022 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2024 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2026 if (unlikely(skb_is_nonlinear(skb))) {
2027 WARN_ON(1);
2028 return;
2030 skb->len = len;
2031 skb_set_tail_pointer(skb, len);
2034 void skb_trim(struct sk_buff *skb, unsigned int len);
2036 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2038 if (skb->data_len)
2039 return ___pskb_trim(skb, len);
2040 __skb_trim(skb, len);
2041 return 0;
2044 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2046 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2050 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2051 * @skb: buffer to alter
2052 * @len: new length
2054 * This is identical to pskb_trim except that the caller knows that
2055 * the skb is not cloned so we should never get an error due to out-
2056 * of-memory.
2058 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2060 int err = pskb_trim(skb, len);
2061 BUG_ON(err);
2065 * skb_orphan - orphan a buffer
2066 * @skb: buffer to orphan
2068 * If a buffer currently has an owner then we call the owner's
2069 * destructor function and make the @skb unowned. The buffer continues
2070 * to exist but is no longer charged to its former owner.
2072 static inline void skb_orphan(struct sk_buff *skb)
2074 if (skb->destructor) {
2075 skb->destructor(skb);
2076 skb->destructor = NULL;
2077 skb->sk = NULL;
2078 } else {
2079 BUG_ON(skb->sk);
2084 * skb_orphan_frags - orphan the frags contained in a buffer
2085 * @skb: buffer to orphan frags from
2086 * @gfp_mask: allocation mask for replacement pages
2088 * For each frag in the SKB which needs a destructor (i.e. has an
2089 * owner) create a copy of that frag and release the original
2090 * page by calling the destructor.
2092 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2094 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2095 return 0;
2096 return skb_copy_ubufs(skb, gfp_mask);
2100 * __skb_queue_purge - empty a list
2101 * @list: list to empty
2103 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2104 * the list and one reference dropped. This function does not take the
2105 * list lock and the caller must hold the relevant locks to use it.
2107 void skb_queue_purge(struct sk_buff_head *list);
2108 static inline void __skb_queue_purge(struct sk_buff_head *list)
2110 struct sk_buff *skb;
2111 while ((skb = __skb_dequeue(list)) != NULL)
2112 kfree_skb(skb);
2115 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
2116 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
2117 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
2119 void *netdev_alloc_frag(unsigned int fragsz);
2121 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2122 gfp_t gfp_mask);
2125 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2126 * @dev: network device to receive on
2127 * @length: length to allocate
2129 * Allocate a new &sk_buff and assign it a usage count of one. The
2130 * buffer has unspecified headroom built in. Users should allocate
2131 * the headroom they think they need without accounting for the
2132 * built in space. The built in space is used for optimisations.
2134 * %NULL is returned if there is no free memory. Although this function
2135 * allocates memory it can be called from an interrupt.
2137 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2138 unsigned int length)
2140 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2143 /* legacy helper around __netdev_alloc_skb() */
2144 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2145 gfp_t gfp_mask)
2147 return __netdev_alloc_skb(NULL, length, gfp_mask);
2150 /* legacy helper around netdev_alloc_skb() */
2151 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2153 return netdev_alloc_skb(NULL, length);
2157 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2158 unsigned int length, gfp_t gfp)
2160 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2162 if (NET_IP_ALIGN && skb)
2163 skb_reserve(skb, NET_IP_ALIGN);
2164 return skb;
2167 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2168 unsigned int length)
2170 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2173 void *napi_alloc_frag(unsigned int fragsz);
2174 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2175 unsigned int length, gfp_t gfp_mask);
2176 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2177 unsigned int length)
2179 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2183 * __dev_alloc_pages - allocate page for network Rx
2184 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2185 * @order: size of the allocation
2187 * Allocate a new page.
2189 * %NULL is returned if there is no free memory.
2191 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2192 unsigned int order)
2194 /* This piece of code contains several assumptions.
2195 * 1. This is for device Rx, therefor a cold page is preferred.
2196 * 2. The expectation is the user wants a compound page.
2197 * 3. If requesting a order 0 page it will not be compound
2198 * due to the check to see if order has a value in prep_new_page
2199 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2200 * code in gfp_to_alloc_flags that should be enforcing this.
2202 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2204 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2207 static inline struct page *dev_alloc_pages(unsigned int order)
2209 return __dev_alloc_pages(GFP_ATOMIC, order);
2213 * __dev_alloc_page - allocate a page for network Rx
2214 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2216 * Allocate a new page.
2218 * %NULL is returned if there is no free memory.
2220 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2222 return __dev_alloc_pages(gfp_mask, 0);
2225 static inline struct page *dev_alloc_page(void)
2227 return __dev_alloc_page(GFP_ATOMIC);
2231 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2232 * @page: The page that was allocated from skb_alloc_page
2233 * @skb: The skb that may need pfmemalloc set
2235 static inline void skb_propagate_pfmemalloc(struct page *page,
2236 struct sk_buff *skb)
2238 if (page && page->pfmemalloc)
2239 skb->pfmemalloc = true;
2243 * skb_frag_page - retrieve the page referred to by a paged fragment
2244 * @frag: the paged fragment
2246 * Returns the &struct page associated with @frag.
2248 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2250 return frag->page.p;
2254 * __skb_frag_ref - take an addition reference on a paged fragment.
2255 * @frag: the paged fragment
2257 * Takes an additional reference on the paged fragment @frag.
2259 static inline void __skb_frag_ref(skb_frag_t *frag)
2261 get_page(skb_frag_page(frag));
2265 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2266 * @skb: the buffer
2267 * @f: the fragment offset.
2269 * Takes an additional reference on the @f'th paged fragment of @skb.
2271 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2273 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2277 * __skb_frag_unref - release a reference on a paged fragment.
2278 * @frag: the paged fragment
2280 * Releases a reference on the paged fragment @frag.
2282 static inline void __skb_frag_unref(skb_frag_t *frag)
2284 put_page(skb_frag_page(frag));
2288 * skb_frag_unref - release a reference on a paged fragment of an skb.
2289 * @skb: the buffer
2290 * @f: the fragment offset
2292 * Releases a reference on the @f'th paged fragment of @skb.
2294 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2296 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2300 * skb_frag_address - gets the address of the data contained in a paged fragment
2301 * @frag: the paged fragment buffer
2303 * Returns the address of the data within @frag. The page must already
2304 * be mapped.
2306 static inline void *skb_frag_address(const skb_frag_t *frag)
2308 return page_address(skb_frag_page(frag)) + frag->page_offset;
2312 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2313 * @frag: the paged fragment buffer
2315 * Returns the address of the data within @frag. Checks that the page
2316 * is mapped and returns %NULL otherwise.
2318 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2320 void *ptr = page_address(skb_frag_page(frag));
2321 if (unlikely(!ptr))
2322 return NULL;
2324 return ptr + frag->page_offset;
2328 * __skb_frag_set_page - sets the page contained in a paged fragment
2329 * @frag: the paged fragment
2330 * @page: the page to set
2332 * Sets the fragment @frag to contain @page.
2334 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2336 frag->page.p = page;
2340 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2341 * @skb: the buffer
2342 * @f: the fragment offset
2343 * @page: the page to set
2345 * Sets the @f'th fragment of @skb to contain @page.
2347 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2348 struct page *page)
2350 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2353 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2356 * skb_frag_dma_map - maps a paged fragment via the DMA API
2357 * @dev: the device to map the fragment to
2358 * @frag: the paged fragment to map
2359 * @offset: the offset within the fragment (starting at the
2360 * fragment's own offset)
2361 * @size: the number of bytes to map
2362 * @dir: the direction of the mapping (%PCI_DMA_*)
2364 * Maps the page associated with @frag to @device.
2366 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2367 const skb_frag_t *frag,
2368 size_t offset, size_t size,
2369 enum dma_data_direction dir)
2371 return dma_map_page(dev, skb_frag_page(frag),
2372 frag->page_offset + offset, size, dir);
2375 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2376 gfp_t gfp_mask)
2378 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2382 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2383 gfp_t gfp_mask)
2385 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2390 * skb_clone_writable - is the header of a clone writable
2391 * @skb: buffer to check
2392 * @len: length up to which to write
2394 * Returns true if modifying the header part of the cloned buffer
2395 * does not requires the data to be copied.
2397 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2399 return !skb_header_cloned(skb) &&
2400 skb_headroom(skb) + len <= skb->hdr_len;
2403 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2404 int cloned)
2406 int delta = 0;
2408 if (headroom > skb_headroom(skb))
2409 delta = headroom - skb_headroom(skb);
2411 if (delta || cloned)
2412 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2413 GFP_ATOMIC);
2414 return 0;
2418 * skb_cow - copy header of skb when it is required
2419 * @skb: buffer to cow
2420 * @headroom: needed headroom
2422 * If the skb passed lacks sufficient headroom or its data part
2423 * is shared, data is reallocated. If reallocation fails, an error
2424 * is returned and original skb is not changed.
2426 * The result is skb with writable area skb->head...skb->tail
2427 * and at least @headroom of space at head.
2429 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2431 return __skb_cow(skb, headroom, skb_cloned(skb));
2435 * skb_cow_head - skb_cow but only making the head writable
2436 * @skb: buffer to cow
2437 * @headroom: needed headroom
2439 * This function is identical to skb_cow except that we replace the
2440 * skb_cloned check by skb_header_cloned. It should be used when
2441 * you only need to push on some header and do not need to modify
2442 * the data.
2444 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2446 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2450 * skb_padto - pad an skbuff up to a minimal size
2451 * @skb: buffer to pad
2452 * @len: minimal length
2454 * Pads up a buffer to ensure the trailing bytes exist and are
2455 * blanked. If the buffer already contains sufficient data it
2456 * is untouched. Otherwise it is extended. Returns zero on
2457 * success. The skb is freed on error.
2459 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2461 unsigned int size = skb->len;
2462 if (likely(size >= len))
2463 return 0;
2464 return skb_pad(skb, len - size);
2468 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2469 * @skb: buffer to pad
2470 * @len: minimal length
2472 * Pads up a buffer to ensure the trailing bytes exist and are
2473 * blanked. If the buffer already contains sufficient data it
2474 * is untouched. Otherwise it is extended. Returns zero on
2475 * success. The skb is freed on error.
2477 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2479 unsigned int size = skb->len;
2481 if (unlikely(size < len)) {
2482 len -= size;
2483 if (skb_pad(skb, len))
2484 return -ENOMEM;
2485 __skb_put(skb, len);
2487 return 0;
2490 static inline int skb_add_data(struct sk_buff *skb,
2491 struct iov_iter *from, int copy)
2493 const int off = skb->len;
2495 if (skb->ip_summed == CHECKSUM_NONE) {
2496 __wsum csum = 0;
2497 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2498 &csum, from) == copy) {
2499 skb->csum = csum_block_add(skb->csum, csum, off);
2500 return 0;
2502 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
2503 return 0;
2505 __skb_trim(skb, off);
2506 return -EFAULT;
2509 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2510 const struct page *page, int off)
2512 if (i) {
2513 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2515 return page == skb_frag_page(frag) &&
2516 off == frag->page_offset + skb_frag_size(frag);
2518 return false;
2521 static inline int __skb_linearize(struct sk_buff *skb)
2523 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2527 * skb_linearize - convert paged skb to linear one
2528 * @skb: buffer to linarize
2530 * If there is no free memory -ENOMEM is returned, otherwise zero
2531 * is returned and the old skb data released.
2533 static inline int skb_linearize(struct sk_buff *skb)
2535 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2539 * skb_has_shared_frag - can any frag be overwritten
2540 * @skb: buffer to test
2542 * Return true if the skb has at least one frag that might be modified
2543 * by an external entity (as in vmsplice()/sendfile())
2545 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2547 return skb_is_nonlinear(skb) &&
2548 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2552 * skb_linearize_cow - make sure skb is linear and writable
2553 * @skb: buffer to process
2555 * If there is no free memory -ENOMEM is returned, otherwise zero
2556 * is returned and the old skb data released.
2558 static inline int skb_linearize_cow(struct sk_buff *skb)
2560 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2561 __skb_linearize(skb) : 0;
2565 * skb_postpull_rcsum - update checksum for received skb after pull
2566 * @skb: buffer to update
2567 * @start: start of data before pull
2568 * @len: length of data pulled
2570 * After doing a pull on a received packet, you need to call this to
2571 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2572 * CHECKSUM_NONE so that it can be recomputed from scratch.
2575 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2576 const void *start, unsigned int len)
2578 if (skb->ip_summed == CHECKSUM_COMPLETE)
2579 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2582 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2585 * pskb_trim_rcsum - trim received skb and update checksum
2586 * @skb: buffer to trim
2587 * @len: new length
2589 * This is exactly the same as pskb_trim except that it ensures the
2590 * checksum of received packets are still valid after the operation.
2593 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2595 if (likely(len >= skb->len))
2596 return 0;
2597 if (skb->ip_summed == CHECKSUM_COMPLETE)
2598 skb->ip_summed = CHECKSUM_NONE;
2599 return __pskb_trim(skb, len);
2602 #define skb_queue_walk(queue, skb) \
2603 for (skb = (queue)->next; \
2604 skb != (struct sk_buff *)(queue); \
2605 skb = skb->next)
2607 #define skb_queue_walk_safe(queue, skb, tmp) \
2608 for (skb = (queue)->next, tmp = skb->next; \
2609 skb != (struct sk_buff *)(queue); \
2610 skb = tmp, tmp = skb->next)
2612 #define skb_queue_walk_from(queue, skb) \
2613 for (; skb != (struct sk_buff *)(queue); \
2614 skb = skb->next)
2616 #define skb_queue_walk_from_safe(queue, skb, tmp) \
2617 for (tmp = skb->next; \
2618 skb != (struct sk_buff *)(queue); \
2619 skb = tmp, tmp = skb->next)
2621 #define skb_queue_reverse_walk(queue, skb) \
2622 for (skb = (queue)->prev; \
2623 skb != (struct sk_buff *)(queue); \
2624 skb = skb->prev)
2626 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2627 for (skb = (queue)->prev, tmp = skb->prev; \
2628 skb != (struct sk_buff *)(queue); \
2629 skb = tmp, tmp = skb->prev)
2631 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2632 for (tmp = skb->prev; \
2633 skb != (struct sk_buff *)(queue); \
2634 skb = tmp, tmp = skb->prev)
2636 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2638 return skb_shinfo(skb)->frag_list != NULL;
2641 static inline void skb_frag_list_init(struct sk_buff *skb)
2643 skb_shinfo(skb)->frag_list = NULL;
2646 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
2648 frag->next = skb_shinfo(skb)->frag_list;
2649 skb_shinfo(skb)->frag_list = frag;
2652 #define skb_walk_frags(skb, iter) \
2653 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2655 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2656 int *peeked, int *off, int *err);
2657 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2658 int *err);
2659 unsigned int datagram_poll(struct file *file, struct socket *sock,
2660 struct poll_table_struct *wait);
2661 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2662 struct iov_iter *to, int size);
2663 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2664 struct msghdr *msg, int size)
2666 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
2668 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2669 struct msghdr *msg);
2670 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2671 struct iov_iter *from, int len);
2672 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
2673 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2674 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2675 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
2676 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2677 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2678 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2679 int len, __wsum csum);
2680 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
2681 struct pipe_inode_info *pipe, unsigned int len,
2682 unsigned int flags);
2683 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2684 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
2685 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2686 int len, int hlen);
2687 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2688 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2689 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
2690 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
2691 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
2692 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
2693 int skb_ensure_writable(struct sk_buff *skb, int write_len);
2694 int skb_vlan_pop(struct sk_buff *skb);
2695 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
2697 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2699 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
2702 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
2704 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
2707 struct skb_checksum_ops {
2708 __wsum (*update)(const void *mem, int len, __wsum wsum);
2709 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2712 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2713 __wsum csum, const struct skb_checksum_ops *ops);
2714 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2715 __wsum csum);
2717 static inline void *__skb_header_pointer(const struct sk_buff *skb, int offset,
2718 int len, void *data, int hlen, void *buffer)
2720 if (hlen - offset >= len)
2721 return data + offset;
2723 if (!skb ||
2724 skb_copy_bits(skb, offset, buffer, len) < 0)
2725 return NULL;
2727 return buffer;
2730 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
2731 int len, void *buffer)
2733 return __skb_header_pointer(skb, offset, len, skb->data,
2734 skb_headlen(skb), buffer);
2738 * skb_needs_linearize - check if we need to linearize a given skb
2739 * depending on the given device features.
2740 * @skb: socket buffer to check
2741 * @features: net device features
2743 * Returns true if either:
2744 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2745 * 2. skb is fragmented and the device does not support SG.
2747 static inline bool skb_needs_linearize(struct sk_buff *skb,
2748 netdev_features_t features)
2750 return skb_is_nonlinear(skb) &&
2751 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2752 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2755 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2756 void *to,
2757 const unsigned int len)
2759 memcpy(to, skb->data, len);
2762 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2763 const int offset, void *to,
2764 const unsigned int len)
2766 memcpy(to, skb->data + offset, len);
2769 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2770 const void *from,
2771 const unsigned int len)
2773 memcpy(skb->data, from, len);
2776 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2777 const int offset,
2778 const void *from,
2779 const unsigned int len)
2781 memcpy(skb->data + offset, from, len);
2784 void skb_init(void);
2786 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2788 return skb->tstamp;
2792 * skb_get_timestamp - get timestamp from a skb
2793 * @skb: skb to get stamp from
2794 * @stamp: pointer to struct timeval to store stamp in
2796 * Timestamps are stored in the skb as offsets to a base timestamp.
2797 * This function converts the offset back to a struct timeval and stores
2798 * it in stamp.
2800 static inline void skb_get_timestamp(const struct sk_buff *skb,
2801 struct timeval *stamp)
2803 *stamp = ktime_to_timeval(skb->tstamp);
2806 static inline void skb_get_timestampns(const struct sk_buff *skb,
2807 struct timespec *stamp)
2809 *stamp = ktime_to_timespec(skb->tstamp);
2812 static inline void __net_timestamp(struct sk_buff *skb)
2814 skb->tstamp = ktime_get_real();
2817 static inline ktime_t net_timedelta(ktime_t t)
2819 return ktime_sub(ktime_get_real(), t);
2822 static inline ktime_t net_invalid_timestamp(void)
2824 return ktime_set(0, 0);
2827 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
2829 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2831 void skb_clone_tx_timestamp(struct sk_buff *skb);
2832 bool skb_defer_rx_timestamp(struct sk_buff *skb);
2834 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2836 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2840 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2842 return false;
2845 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2848 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2850 * PHY drivers may accept clones of transmitted packets for
2851 * timestamping via their phy_driver.txtstamp method. These drivers
2852 * must call this function to return the skb back to the stack, with
2853 * or without a timestamp.
2855 * @skb: clone of the the original outgoing packet
2856 * @hwtstamps: hardware time stamps, may be NULL if not available
2859 void skb_complete_tx_timestamp(struct sk_buff *skb,
2860 struct skb_shared_hwtstamps *hwtstamps);
2862 void __skb_tstamp_tx(struct sk_buff *orig_skb,
2863 struct skb_shared_hwtstamps *hwtstamps,
2864 struct sock *sk, int tstype);
2867 * skb_tstamp_tx - queue clone of skb with send time stamps
2868 * @orig_skb: the original outgoing packet
2869 * @hwtstamps: hardware time stamps, may be NULL if not available
2871 * If the skb has a socket associated, then this function clones the
2872 * skb (thus sharing the actual data and optional structures), stores
2873 * the optional hardware time stamping information (if non NULL) or
2874 * generates a software time stamp (otherwise), then queues the clone
2875 * to the error queue of the socket. Errors are silently ignored.
2877 void skb_tstamp_tx(struct sk_buff *orig_skb,
2878 struct skb_shared_hwtstamps *hwtstamps);
2880 static inline void sw_tx_timestamp(struct sk_buff *skb)
2882 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2883 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
2884 skb_tstamp_tx(skb, NULL);
2888 * skb_tx_timestamp() - Driver hook for transmit timestamping
2890 * Ethernet MAC Drivers should call this function in their hard_xmit()
2891 * function immediately before giving the sk_buff to the MAC hardware.
2893 * Specifically, one should make absolutely sure that this function is
2894 * called before TX completion of this packet can trigger. Otherwise
2895 * the packet could potentially already be freed.
2897 * @skb: A socket buffer.
2899 static inline void skb_tx_timestamp(struct sk_buff *skb)
2901 skb_clone_tx_timestamp(skb);
2902 sw_tx_timestamp(skb);
2906 * skb_complete_wifi_ack - deliver skb with wifi status
2908 * @skb: the original outgoing packet
2909 * @acked: ack status
2912 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2914 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2915 __sum16 __skb_checksum_complete(struct sk_buff *skb);
2917 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2919 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
2920 skb->csum_valid ||
2921 (skb->ip_summed == CHECKSUM_PARTIAL &&
2922 skb_checksum_start_offset(skb) >= 0));
2926 * skb_checksum_complete - Calculate checksum of an entire packet
2927 * @skb: packet to process
2929 * This function calculates the checksum over the entire packet plus
2930 * the value of skb->csum. The latter can be used to supply the
2931 * checksum of a pseudo header as used by TCP/UDP. It returns the
2932 * checksum.
2934 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2935 * this function can be used to verify that checksum on received
2936 * packets. In that case the function should return zero if the
2937 * checksum is correct. In particular, this function will return zero
2938 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2939 * hardware has already verified the correctness of the checksum.
2941 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
2943 return skb_csum_unnecessary(skb) ?
2944 0 : __skb_checksum_complete(skb);
2947 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
2949 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2950 if (skb->csum_level == 0)
2951 skb->ip_summed = CHECKSUM_NONE;
2952 else
2953 skb->csum_level--;
2957 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
2959 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2960 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
2961 skb->csum_level++;
2962 } else if (skb->ip_summed == CHECKSUM_NONE) {
2963 skb->ip_summed = CHECKSUM_UNNECESSARY;
2964 skb->csum_level = 0;
2968 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
2970 /* Mark current checksum as bad (typically called from GRO
2971 * path). In the case that ip_summed is CHECKSUM_NONE
2972 * this must be the first checksum encountered in the packet.
2973 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
2974 * checksum after the last one validated. For UDP, a zero
2975 * checksum can not be marked as bad.
2978 if (skb->ip_summed == CHECKSUM_NONE ||
2979 skb->ip_summed == CHECKSUM_UNNECESSARY)
2980 skb->csum_bad = 1;
2983 /* Check if we need to perform checksum complete validation.
2985 * Returns true if checksum complete is needed, false otherwise
2986 * (either checksum is unnecessary or zero checksum is allowed).
2988 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
2989 bool zero_okay,
2990 __sum16 check)
2992 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
2993 skb->csum_valid = 1;
2994 __skb_decr_checksum_unnecessary(skb);
2995 return false;
2998 return true;
3001 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3002 * in checksum_init.
3004 #define CHECKSUM_BREAK 76
3006 /* Validate (init) checksum based on checksum complete.
3008 * Return values:
3009 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3010 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3011 * checksum is stored in skb->csum for use in __skb_checksum_complete
3012 * non-zero: value of invalid checksum
3015 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3016 bool complete,
3017 __wsum psum)
3019 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3020 if (!csum_fold(csum_add(psum, skb->csum))) {
3021 skb->csum_valid = 1;
3022 return 0;
3024 } else if (skb->csum_bad) {
3025 /* ip_summed == CHECKSUM_NONE in this case */
3026 return 1;
3029 skb->csum = psum;
3031 if (complete || skb->len <= CHECKSUM_BREAK) {
3032 __sum16 csum;
3034 csum = __skb_checksum_complete(skb);
3035 skb->csum_valid = !csum;
3036 return csum;
3039 return 0;
3042 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3044 return 0;
3047 /* Perform checksum validate (init). Note that this is a macro since we only
3048 * want to calculate the pseudo header which is an input function if necessary.
3049 * First we try to validate without any computation (checksum unnecessary) and
3050 * then calculate based on checksum complete calling the function to compute
3051 * pseudo header.
3053 * Return values:
3054 * 0: checksum is validated or try to in skb_checksum_complete
3055 * non-zero: value of invalid checksum
3057 #define __skb_checksum_validate(skb, proto, complete, \
3058 zero_okay, check, compute_pseudo) \
3059 ({ \
3060 __sum16 __ret = 0; \
3061 skb->csum_valid = 0; \
3062 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3063 __ret = __skb_checksum_validate_complete(skb, \
3064 complete, compute_pseudo(skb, proto)); \
3065 __ret; \
3068 #define skb_checksum_init(skb, proto, compute_pseudo) \
3069 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3071 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3072 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3074 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3075 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3077 #define skb_checksum_validate_zero_check(skb, proto, check, \
3078 compute_pseudo) \
3079 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3081 #define skb_checksum_simple_validate(skb) \
3082 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3084 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3086 return (skb->ip_summed == CHECKSUM_NONE &&
3087 skb->csum_valid && !skb->csum_bad);
3090 static inline void __skb_checksum_convert(struct sk_buff *skb,
3091 __sum16 check, __wsum pseudo)
3093 skb->csum = ~pseudo;
3094 skb->ip_summed = CHECKSUM_COMPLETE;
3097 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3098 do { \
3099 if (__skb_checksum_convert_check(skb)) \
3100 __skb_checksum_convert(skb, check, \
3101 compute_pseudo(skb, proto)); \
3102 } while (0)
3104 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3105 u16 start, u16 offset)
3107 skb->ip_summed = CHECKSUM_PARTIAL;
3108 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3109 skb->csum_offset = offset - start;
3112 /* Update skbuf and packet to reflect the remote checksum offload operation.
3113 * When called, ptr indicates the starting point for skb->csum when
3114 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3115 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3117 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3118 int start, int offset, bool nopartial)
3120 __wsum delta;
3122 if (!nopartial) {
3123 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3124 return;
3127 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3128 __skb_checksum_complete(skb);
3129 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3132 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3134 /* Adjust skb->csum since we changed the packet */
3135 skb->csum = csum_add(skb->csum, delta);
3138 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3139 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3140 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3142 if (nfct && atomic_dec_and_test(&nfct->use))
3143 nf_conntrack_destroy(nfct);
3145 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3147 if (nfct)
3148 atomic_inc(&nfct->use);
3150 #endif
3151 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3152 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3154 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3155 kfree(nf_bridge);
3157 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3159 if (nf_bridge)
3160 atomic_inc(&nf_bridge->use);
3162 #endif /* CONFIG_BRIDGE_NETFILTER */
3163 static inline void nf_reset(struct sk_buff *skb)
3165 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3166 nf_conntrack_put(skb->nfct);
3167 skb->nfct = NULL;
3168 #endif
3169 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3170 nf_bridge_put(skb->nf_bridge);
3171 skb->nf_bridge = NULL;
3172 #endif
3175 static inline void nf_reset_trace(struct sk_buff *skb)
3177 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3178 skb->nf_trace = 0;
3179 #endif
3182 /* Note: This doesn't put any conntrack and bridge info in dst. */
3183 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3184 bool copy)
3186 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3187 dst->nfct = src->nfct;
3188 nf_conntrack_get(src->nfct);
3189 if (copy)
3190 dst->nfctinfo = src->nfctinfo;
3191 #endif
3192 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3193 dst->nf_bridge = src->nf_bridge;
3194 nf_bridge_get(src->nf_bridge);
3195 #endif
3196 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3197 if (copy)
3198 dst->nf_trace = src->nf_trace;
3199 #endif
3202 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3204 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3205 nf_conntrack_put(dst->nfct);
3206 #endif
3207 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3208 nf_bridge_put(dst->nf_bridge);
3209 #endif
3210 __nf_copy(dst, src, true);
3213 #ifdef CONFIG_NETWORK_SECMARK
3214 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3216 to->secmark = from->secmark;
3219 static inline void skb_init_secmark(struct sk_buff *skb)
3221 skb->secmark = 0;
3223 #else
3224 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3227 static inline void skb_init_secmark(struct sk_buff *skb)
3229 #endif
3231 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3233 return !skb->destructor &&
3234 #if IS_ENABLED(CONFIG_XFRM)
3235 !skb->sp &&
3236 #endif
3237 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3238 !skb->nfct &&
3239 #endif
3240 !skb->_skb_refdst &&
3241 !skb_has_frag_list(skb);
3244 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3246 skb->queue_mapping = queue_mapping;
3249 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3251 return skb->queue_mapping;
3254 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3256 to->queue_mapping = from->queue_mapping;
3259 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3261 skb->queue_mapping = rx_queue + 1;
3264 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3266 return skb->queue_mapping - 1;
3269 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3271 return skb->queue_mapping != 0;
3274 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
3275 unsigned int num_tx_queues);
3277 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3279 #ifdef CONFIG_XFRM
3280 return skb->sp;
3281 #else
3282 return NULL;
3283 #endif
3286 /* Keeps track of mac header offset relative to skb->head.
3287 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3288 * For non-tunnel skb it points to skb_mac_header() and for
3289 * tunnel skb it points to outer mac header.
3290 * Keeps track of level of encapsulation of network headers.
3292 struct skb_gso_cb {
3293 int mac_offset;
3294 int encap_level;
3295 __u16 csum_start;
3297 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3299 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3301 return (skb_mac_header(inner_skb) - inner_skb->head) -
3302 SKB_GSO_CB(inner_skb)->mac_offset;
3305 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3307 int new_headroom, headroom;
3308 int ret;
3310 headroom = skb_headroom(skb);
3311 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3312 if (ret)
3313 return ret;
3315 new_headroom = skb_headroom(skb);
3316 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3317 return 0;
3320 /* Compute the checksum for a gso segment. First compute the checksum value
3321 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3322 * then add in skb->csum (checksum from csum_start to end of packet).
3323 * skb->csum and csum_start are then updated to reflect the checksum of the
3324 * resultant packet starting from the transport header-- the resultant checksum
3325 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3326 * header.
3328 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3330 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3331 skb_transport_offset(skb);
3332 __u16 csum;
3334 csum = csum_fold(csum_partial(skb_transport_header(skb),
3335 plen, skb->csum));
3336 skb->csum = res;
3337 SKB_GSO_CB(skb)->csum_start -= plen;
3339 return csum;
3342 static inline bool skb_is_gso(const struct sk_buff *skb)
3344 return skb_shinfo(skb)->gso_size;
3347 /* Note: Should be called only if skb_is_gso(skb) is true */
3348 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3350 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3353 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3355 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3357 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3358 * wanted then gso_type will be set. */
3359 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3361 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3362 unlikely(shinfo->gso_type == 0)) {
3363 __skb_warn_lro_forwarding(skb);
3364 return true;
3366 return false;
3369 static inline void skb_forward_csum(struct sk_buff *skb)
3371 /* Unfortunately we don't support this one. Any brave souls? */
3372 if (skb->ip_summed == CHECKSUM_COMPLETE)
3373 skb->ip_summed = CHECKSUM_NONE;
3377 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3378 * @skb: skb to check
3380 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3381 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3382 * use this helper, to document places where we make this assertion.
3384 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3386 #ifdef DEBUG
3387 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3388 #endif
3391 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3393 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3395 u32 skb_get_poff(const struct sk_buff *skb);
3396 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
3397 const struct flow_keys *keys, int hlen);
3400 * skb_head_is_locked - Determine if the skb->head is locked down
3401 * @skb: skb to check
3403 * The head on skbs build around a head frag can be removed if they are
3404 * not cloned. This function returns true if the skb head is locked down
3405 * due to either being allocated via kmalloc, or by being a clone with
3406 * multiple references to the head.
3408 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3410 return !skb->head_frag || skb_cloned(skb);
3414 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3416 * @skb: GSO skb
3418 * skb_gso_network_seglen is used to determine the real size of the
3419 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3421 * The MAC/L2 header is not accounted for.
3423 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3425 unsigned int hdr_len = skb_transport_header(skb) -
3426 skb_network_header(skb);
3427 return hdr_len + skb_gso_transport_seglen(skb);
3429 #endif /* __KERNEL__ */
3430 #endif /* _LINUX_SKBUFF_H */