net: skb_frag_t can be smaller on small arches
[linux-2.6/btrfs-unstable.git] / include / linux / skbuff.h
blob0b53c43ac92e0a2db2d202ba28071989332eff4c
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/cache.h>
23 #include <asm/atomic.h>
24 #include <asm/types.h>
25 #include <linux/spinlock.h>
26 #include <linux/net.h>
27 #include <linux/textsearch.h>
28 #include <net/checksum.h>
29 #include <linux/rcupdate.h>
30 #include <linux/dmaengine.h>
31 #include <linux/hrtimer.h>
33 /* Don't change this without changing skb_csum_unnecessary! */
34 #define CHECKSUM_NONE 0
35 #define CHECKSUM_UNNECESSARY 1
36 #define CHECKSUM_COMPLETE 2
37 #define CHECKSUM_PARTIAL 3
39 #define SKB_DATA_ALIGN(X) (((X) + (SMP_CACHE_BYTES - 1)) & \
40 ~(SMP_CACHE_BYTES - 1))
41 #define SKB_WITH_OVERHEAD(X) \
42 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
43 #define SKB_MAX_ORDER(X, ORDER) \
44 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
45 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
46 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
48 /* A. Checksumming of received packets by device.
50 * NONE: device failed to checksum this packet.
51 * skb->csum is undefined.
53 * UNNECESSARY: device parsed packet and wouldbe verified checksum.
54 * skb->csum is undefined.
55 * It is bad option, but, unfortunately, many of vendors do this.
56 * Apparently with secret goal to sell you new device, when you
57 * will add new protocol to your host. F.e. IPv6. 8)
59 * COMPLETE: the most generic way. Device supplied checksum of _all_
60 * the packet as seen by netif_rx in skb->csum.
61 * NOTE: Even if device supports only some protocols, but
62 * is able to produce some skb->csum, it MUST use COMPLETE,
63 * not UNNECESSARY.
65 * PARTIAL: identical to the case for output below. This may occur
66 * on a packet received directly from another Linux OS, e.g.,
67 * a virtualised Linux kernel on the same host. The packet can
68 * be treated in the same way as UNNECESSARY except that on
69 * output (i.e., forwarding) the checksum must be filled in
70 * by the OS or the hardware.
72 * B. Checksumming on output.
74 * NONE: skb is checksummed by protocol or csum is not required.
76 * PARTIAL: device is required to csum packet as seen by hard_start_xmit
77 * from skb->csum_start to the end and to record the checksum
78 * at skb->csum_start + skb->csum_offset.
80 * Device must show its capabilities in dev->features, set
81 * at device setup time.
82 * NETIF_F_HW_CSUM - it is clever device, it is able to checksum
83 * everything.
84 * NETIF_F_NO_CSUM - loopback or reliable single hop media.
85 * NETIF_F_IP_CSUM - device is dumb. It is able to csum only
86 * TCP/UDP over IPv4. Sigh. Vendors like this
87 * way by an unknown reason. Though, see comment above
88 * about CHECKSUM_UNNECESSARY. 8)
89 * NETIF_F_IPV6_CSUM about as dumb as the last one but does IPv6 instead.
91 * Any questions? No questions, good. --ANK
94 struct net_device;
95 struct scatterlist;
96 struct pipe_inode_info;
98 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
99 struct nf_conntrack {
100 atomic_t use;
102 #endif
104 #ifdef CONFIG_BRIDGE_NETFILTER
105 struct nf_bridge_info {
106 atomic_t use;
107 struct net_device *physindev;
108 struct net_device *physoutdev;
109 unsigned int mask;
110 unsigned long data[32 / sizeof(unsigned long)];
112 #endif
114 struct sk_buff_head {
115 /* These two members must be first. */
116 struct sk_buff *next;
117 struct sk_buff *prev;
119 __u32 qlen;
120 spinlock_t lock;
123 struct sk_buff;
125 /* To allow 64K frame to be packed as single skb without frag_list */
126 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 2)
128 typedef struct skb_frag_struct skb_frag_t;
130 struct skb_frag_struct {
131 struct page *page;
132 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
133 __u32 page_offset;
134 __u32 size;
135 #else
136 __u16 page_offset;
137 __u16 size;
138 #endif
141 #define HAVE_HW_TIME_STAMP
144 * struct skb_shared_hwtstamps - hardware time stamps
145 * @hwtstamp: hardware time stamp transformed into duration
146 * since arbitrary point in time
147 * @syststamp: hwtstamp transformed to system time base
149 * Software time stamps generated by ktime_get_real() are stored in
150 * skb->tstamp. The relation between the different kinds of time
151 * stamps is as follows:
153 * syststamp and tstamp can be compared against each other in
154 * arbitrary combinations. The accuracy of a
155 * syststamp/tstamp/"syststamp from other device" comparison is
156 * limited by the accuracy of the transformation into system time
157 * base. This depends on the device driver and its underlying
158 * hardware.
160 * hwtstamps can only be compared against other hwtstamps from
161 * the same device.
163 * This structure is attached to packets as part of the
164 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
166 struct skb_shared_hwtstamps {
167 ktime_t hwtstamp;
168 ktime_t syststamp;
171 /* Definitions for tx_flags in struct skb_shared_info */
172 enum {
173 /* generate hardware time stamp */
174 SKBTX_HW_TSTAMP = 1 << 0,
176 /* generate software time stamp */
177 SKBTX_SW_TSTAMP = 1 << 1,
179 /* device driver is going to provide hardware time stamp */
180 SKBTX_IN_PROGRESS = 1 << 2,
182 /* ensure the originating sk reference is available on driver level */
183 SKBTX_DRV_NEEDS_SK_REF = 1 << 3,
186 /* This data is invariant across clones and lives at
187 * the end of the header data, ie. at skb->end.
189 struct skb_shared_info {
190 unsigned short nr_frags;
191 unsigned short gso_size;
192 /* Warning: this field is not always filled in (UFO)! */
193 unsigned short gso_segs;
194 unsigned short gso_type;
195 __be32 ip6_frag_id;
196 __u8 tx_flags;
197 struct sk_buff *frag_list;
198 struct skb_shared_hwtstamps hwtstamps;
201 * Warning : all fields before dataref are cleared in __alloc_skb()
203 atomic_t dataref;
205 /* Intermediate layers must ensure that destructor_arg
206 * remains valid until skb destructor */
207 void * destructor_arg;
208 /* must be last field, see pskb_expand_head() */
209 skb_frag_t frags[MAX_SKB_FRAGS];
212 /* We divide dataref into two halves. The higher 16 bits hold references
213 * to the payload part of skb->data. The lower 16 bits hold references to
214 * the entire skb->data. A clone of a headerless skb holds the length of
215 * the header in skb->hdr_len.
217 * All users must obey the rule that the skb->data reference count must be
218 * greater than or equal to the payload reference count.
220 * Holding a reference to the payload part means that the user does not
221 * care about modifications to the header part of skb->data.
223 #define SKB_DATAREF_SHIFT 16
224 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
227 enum {
228 SKB_FCLONE_UNAVAILABLE,
229 SKB_FCLONE_ORIG,
230 SKB_FCLONE_CLONE,
233 enum {
234 SKB_GSO_TCPV4 = 1 << 0,
235 SKB_GSO_UDP = 1 << 1,
237 /* This indicates the skb is from an untrusted source. */
238 SKB_GSO_DODGY = 1 << 2,
240 /* This indicates the tcp segment has CWR set. */
241 SKB_GSO_TCP_ECN = 1 << 3,
243 SKB_GSO_TCPV6 = 1 << 4,
245 SKB_GSO_FCOE = 1 << 5,
248 #if BITS_PER_LONG > 32
249 #define NET_SKBUFF_DATA_USES_OFFSET 1
250 #endif
252 #ifdef NET_SKBUFF_DATA_USES_OFFSET
253 typedef unsigned int sk_buff_data_t;
254 #else
255 typedef unsigned char *sk_buff_data_t;
256 #endif
258 /**
259 * struct sk_buff - socket buffer
260 * @next: Next buffer in list
261 * @prev: Previous buffer in list
262 * @sk: Socket we are owned by
263 * @tstamp: Time we arrived
264 * @dev: Device we arrived on/are leaving by
265 * @transport_header: Transport layer header
266 * @network_header: Network layer header
267 * @mac_header: Link layer header
268 * @_skb_refdst: destination entry (with norefcount bit)
269 * @sp: the security path, used for xfrm
270 * @cb: Control buffer. Free for use by every layer. Put private vars here
271 * @len: Length of actual data
272 * @data_len: Data length
273 * @mac_len: Length of link layer header
274 * @hdr_len: writable header length of cloned skb
275 * @csum: Checksum (must include start/offset pair)
276 * @csum_start: Offset from skb->head where checksumming should start
277 * @csum_offset: Offset from csum_start where checksum should be stored
278 * @local_df: allow local fragmentation
279 * @cloned: Head may be cloned (check refcnt to be sure)
280 * @nohdr: Payload reference only, must not modify header
281 * @pkt_type: Packet class
282 * @fclone: skbuff clone status
283 * @ip_summed: Driver fed us an IP checksum
284 * @priority: Packet queueing priority
285 * @users: User count - see {datagram,tcp}.c
286 * @protocol: Packet protocol from driver
287 * @truesize: Buffer size
288 * @head: Head of buffer
289 * @data: Data head pointer
290 * @tail: Tail pointer
291 * @end: End pointer
292 * @destructor: Destruct function
293 * @mark: Generic packet mark
294 * @nfct: Associated connection, if any
295 * @ipvs_property: skbuff is owned by ipvs
296 * @peeked: this packet has been seen already, so stats have been
297 * done for it, don't do them again
298 * @nf_trace: netfilter packet trace flag
299 * @nfctinfo: Relationship of this skb to the connection
300 * @nfct_reasm: netfilter conntrack re-assembly pointer
301 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
302 * @skb_iif: ifindex of device we arrived on
303 * @rxhash: the packet hash computed on receive
304 * @queue_mapping: Queue mapping for multiqueue devices
305 * @tc_index: Traffic control index
306 * @tc_verd: traffic control verdict
307 * @ndisc_nodetype: router type (from link layer)
308 * @dma_cookie: a cookie to one of several possible DMA operations
309 * done by skb DMA functions
310 * @secmark: security marking
311 * @vlan_tci: vlan tag control information
314 struct sk_buff {
315 /* These two members must be first. */
316 struct sk_buff *next;
317 struct sk_buff *prev;
319 ktime_t tstamp;
321 struct sock *sk;
322 struct net_device *dev;
325 * This is the control buffer. It is free to use for every
326 * layer. Please put your private variables there. If you
327 * want to keep them across layers you have to do a skb_clone()
328 * first. This is owned by whoever has the skb queued ATM.
330 char cb[48] __aligned(8);
332 unsigned long _skb_refdst;
333 #ifdef CONFIG_XFRM
334 struct sec_path *sp;
335 #endif
336 unsigned int len,
337 data_len;
338 __u16 mac_len,
339 hdr_len;
340 union {
341 __wsum csum;
342 struct {
343 __u16 csum_start;
344 __u16 csum_offset;
347 __u32 priority;
348 kmemcheck_bitfield_begin(flags1);
349 __u8 local_df:1,
350 cloned:1,
351 ip_summed:2,
352 nohdr:1,
353 nfctinfo:3;
354 __u8 pkt_type:3,
355 fclone:2,
356 ipvs_property:1,
357 peeked:1,
358 nf_trace:1;
359 kmemcheck_bitfield_end(flags1);
360 __be16 protocol;
362 void (*destructor)(struct sk_buff *skb);
363 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
364 struct nf_conntrack *nfct;
365 struct sk_buff *nfct_reasm;
366 #endif
367 #ifdef CONFIG_BRIDGE_NETFILTER
368 struct nf_bridge_info *nf_bridge;
369 #endif
371 int skb_iif;
372 #ifdef CONFIG_NET_SCHED
373 __u16 tc_index; /* traffic control index */
374 #ifdef CONFIG_NET_CLS_ACT
375 __u16 tc_verd; /* traffic control verdict */
376 #endif
377 #endif
379 __u32 rxhash;
381 kmemcheck_bitfield_begin(flags2);
382 __u16 queue_mapping:16;
383 #ifdef CONFIG_IPV6_NDISC_NODETYPE
384 __u8 ndisc_nodetype:2,
385 deliver_no_wcard:1;
386 #else
387 __u8 deliver_no_wcard:1;
388 #endif
389 kmemcheck_bitfield_end(flags2);
391 /* 0/14 bit hole */
393 #ifdef CONFIG_NET_DMA
394 dma_cookie_t dma_cookie;
395 #endif
396 #ifdef CONFIG_NETWORK_SECMARK
397 __u32 secmark;
398 #endif
399 union {
400 __u32 mark;
401 __u32 dropcount;
404 __u16 vlan_tci;
406 sk_buff_data_t transport_header;
407 sk_buff_data_t network_header;
408 sk_buff_data_t mac_header;
409 /* These elements must be at the end, see alloc_skb() for details. */
410 sk_buff_data_t tail;
411 sk_buff_data_t end;
412 unsigned char *head,
413 *data;
414 unsigned int truesize;
415 atomic_t users;
418 #ifdef __KERNEL__
420 * Handling routines are only of interest to the kernel
422 #include <linux/slab.h>
424 #include <asm/system.h>
427 * skb might have a dst pointer attached, refcounted or not.
428 * _skb_refdst low order bit is set if refcount was _not_ taken
430 #define SKB_DST_NOREF 1UL
431 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
434 * skb_dst - returns skb dst_entry
435 * @skb: buffer
437 * Returns skb dst_entry, regardless of reference taken or not.
439 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
441 /* If refdst was not refcounted, check we still are in a
442 * rcu_read_lock section
444 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
445 !rcu_read_lock_held() &&
446 !rcu_read_lock_bh_held());
447 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
451 * skb_dst_set - sets skb dst
452 * @skb: buffer
453 * @dst: dst entry
455 * Sets skb dst, assuming a reference was taken on dst and should
456 * be released by skb_dst_drop()
458 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
460 skb->_skb_refdst = (unsigned long)dst;
464 * skb_dst_set_noref - sets skb dst, without a reference
465 * @skb: buffer
466 * @dst: dst entry
468 * Sets skb dst, assuming a reference was not taken on dst
469 * skb_dst_drop() should not dst_release() this dst
471 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
473 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
474 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
478 * skb_dst_is_noref - Test if skb dst isnt refcounted
479 * @skb: buffer
481 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
483 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
486 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
488 return (struct rtable *)skb_dst(skb);
491 extern void kfree_skb(struct sk_buff *skb);
492 extern void consume_skb(struct sk_buff *skb);
493 extern void __kfree_skb(struct sk_buff *skb);
494 extern struct sk_buff *__alloc_skb(unsigned int size,
495 gfp_t priority, int fclone, int node);
496 static inline struct sk_buff *alloc_skb(unsigned int size,
497 gfp_t priority)
499 return __alloc_skb(size, priority, 0, -1);
502 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
503 gfp_t priority)
505 return __alloc_skb(size, priority, 1, -1);
508 extern bool skb_recycle_check(struct sk_buff *skb, int skb_size);
510 extern struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
511 extern struct sk_buff *skb_clone(struct sk_buff *skb,
512 gfp_t priority);
513 extern struct sk_buff *skb_copy(const struct sk_buff *skb,
514 gfp_t priority);
515 extern struct sk_buff *pskb_copy(struct sk_buff *skb,
516 gfp_t gfp_mask);
517 extern int pskb_expand_head(struct sk_buff *skb,
518 int nhead, int ntail,
519 gfp_t gfp_mask);
520 extern struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
521 unsigned int headroom);
522 extern struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
523 int newheadroom, int newtailroom,
524 gfp_t priority);
525 extern int skb_to_sgvec(struct sk_buff *skb,
526 struct scatterlist *sg, int offset,
527 int len);
528 extern int skb_cow_data(struct sk_buff *skb, int tailbits,
529 struct sk_buff **trailer);
530 extern int skb_pad(struct sk_buff *skb, int pad);
531 #define dev_kfree_skb(a) consume_skb(a)
533 extern int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
534 int getfrag(void *from, char *to, int offset,
535 int len,int odd, struct sk_buff *skb),
536 void *from, int length);
538 struct skb_seq_state {
539 __u32 lower_offset;
540 __u32 upper_offset;
541 __u32 frag_idx;
542 __u32 stepped_offset;
543 struct sk_buff *root_skb;
544 struct sk_buff *cur_skb;
545 __u8 *frag_data;
548 extern void skb_prepare_seq_read(struct sk_buff *skb,
549 unsigned int from, unsigned int to,
550 struct skb_seq_state *st);
551 extern unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
552 struct skb_seq_state *st);
553 extern void skb_abort_seq_read(struct skb_seq_state *st);
555 extern unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
556 unsigned int to, struct ts_config *config,
557 struct ts_state *state);
559 extern __u32 __skb_get_rxhash(struct sk_buff *skb);
560 static inline __u32 skb_get_rxhash(struct sk_buff *skb)
562 if (!skb->rxhash)
563 skb->rxhash = __skb_get_rxhash(skb);
565 return skb->rxhash;
568 #ifdef NET_SKBUFF_DATA_USES_OFFSET
569 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
571 return skb->head + skb->end;
573 #else
574 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
576 return skb->end;
578 #endif
580 /* Internal */
581 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
583 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
585 return &skb_shinfo(skb)->hwtstamps;
589 * skb_queue_empty - check if a queue is empty
590 * @list: queue head
592 * Returns true if the queue is empty, false otherwise.
594 static inline int skb_queue_empty(const struct sk_buff_head *list)
596 return list->next == (struct sk_buff *)list;
600 * skb_queue_is_last - check if skb is the last entry in the queue
601 * @list: queue head
602 * @skb: buffer
604 * Returns true if @skb is the last buffer on the list.
606 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
607 const struct sk_buff *skb)
609 return skb->next == (struct sk_buff *)list;
613 * skb_queue_is_first - check if skb is the first entry in the queue
614 * @list: queue head
615 * @skb: buffer
617 * Returns true if @skb is the first buffer on the list.
619 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
620 const struct sk_buff *skb)
622 return skb->prev == (struct sk_buff *)list;
626 * skb_queue_next - return the next packet in the queue
627 * @list: queue head
628 * @skb: current buffer
630 * Return the next packet in @list after @skb. It is only valid to
631 * call this if skb_queue_is_last() evaluates to false.
633 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
634 const struct sk_buff *skb)
636 /* This BUG_ON may seem severe, but if we just return then we
637 * are going to dereference garbage.
639 BUG_ON(skb_queue_is_last(list, skb));
640 return skb->next;
644 * skb_queue_prev - return the prev packet in the queue
645 * @list: queue head
646 * @skb: current buffer
648 * Return the prev packet in @list before @skb. It is only valid to
649 * call this if skb_queue_is_first() evaluates to false.
651 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
652 const struct sk_buff *skb)
654 /* This BUG_ON may seem severe, but if we just return then we
655 * are going to dereference garbage.
657 BUG_ON(skb_queue_is_first(list, skb));
658 return skb->prev;
662 * skb_get - reference buffer
663 * @skb: buffer to reference
665 * Makes another reference to a socket buffer and returns a pointer
666 * to the buffer.
668 static inline struct sk_buff *skb_get(struct sk_buff *skb)
670 atomic_inc(&skb->users);
671 return skb;
675 * If users == 1, we are the only owner and are can avoid redundant
676 * atomic change.
680 * skb_cloned - is the buffer a clone
681 * @skb: buffer to check
683 * Returns true if the buffer was generated with skb_clone() and is
684 * one of multiple shared copies of the buffer. Cloned buffers are
685 * shared data so must not be written to under normal circumstances.
687 static inline int skb_cloned(const struct sk_buff *skb)
689 return skb->cloned &&
690 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
694 * skb_header_cloned - is the header a clone
695 * @skb: buffer to check
697 * Returns true if modifying the header part of the buffer requires
698 * the data to be copied.
700 static inline int skb_header_cloned(const struct sk_buff *skb)
702 int dataref;
704 if (!skb->cloned)
705 return 0;
707 dataref = atomic_read(&skb_shinfo(skb)->dataref);
708 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
709 return dataref != 1;
713 * skb_header_release - release reference to header
714 * @skb: buffer to operate on
716 * Drop a reference to the header part of the buffer. This is done
717 * by acquiring a payload reference. You must not read from the header
718 * part of skb->data after this.
720 static inline void skb_header_release(struct sk_buff *skb)
722 BUG_ON(skb->nohdr);
723 skb->nohdr = 1;
724 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
728 * skb_shared - is the buffer shared
729 * @skb: buffer to check
731 * Returns true if more than one person has a reference to this
732 * buffer.
734 static inline int skb_shared(const struct sk_buff *skb)
736 return atomic_read(&skb->users) != 1;
740 * skb_share_check - check if buffer is shared and if so clone it
741 * @skb: buffer to check
742 * @pri: priority for memory allocation
744 * If the buffer is shared the buffer is cloned and the old copy
745 * drops a reference. A new clone with a single reference is returned.
746 * If the buffer is not shared the original buffer is returned. When
747 * being called from interrupt status or with spinlocks held pri must
748 * be GFP_ATOMIC.
750 * NULL is returned on a memory allocation failure.
752 static inline struct sk_buff *skb_share_check(struct sk_buff *skb,
753 gfp_t pri)
755 might_sleep_if(pri & __GFP_WAIT);
756 if (skb_shared(skb)) {
757 struct sk_buff *nskb = skb_clone(skb, pri);
758 kfree_skb(skb);
759 skb = nskb;
761 return skb;
765 * Copy shared buffers into a new sk_buff. We effectively do COW on
766 * packets to handle cases where we have a local reader and forward
767 * and a couple of other messy ones. The normal one is tcpdumping
768 * a packet thats being forwarded.
772 * skb_unshare - make a copy of a shared buffer
773 * @skb: buffer to check
774 * @pri: priority for memory allocation
776 * If the socket buffer is a clone then this function creates a new
777 * copy of the data, drops a reference count on the old copy and returns
778 * the new copy with the reference count at 1. If the buffer is not a clone
779 * the original buffer is returned. When called with a spinlock held or
780 * from interrupt state @pri must be %GFP_ATOMIC
782 * %NULL is returned on a memory allocation failure.
784 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
785 gfp_t pri)
787 might_sleep_if(pri & __GFP_WAIT);
788 if (skb_cloned(skb)) {
789 struct sk_buff *nskb = skb_copy(skb, pri);
790 kfree_skb(skb); /* Free our shared copy */
791 skb = nskb;
793 return skb;
797 * skb_peek - peek at the head of an &sk_buff_head
798 * @list_: list to peek at
800 * Peek an &sk_buff. Unlike most other operations you _MUST_
801 * be careful with this one. A peek leaves the buffer on the
802 * list and someone else may run off with it. You must hold
803 * the appropriate locks or have a private queue to do this.
805 * Returns %NULL for an empty list or a pointer to the head element.
806 * The reference count is not incremented and the reference is therefore
807 * volatile. Use with caution.
809 static inline struct sk_buff *skb_peek(struct sk_buff_head *list_)
811 struct sk_buff *list = ((struct sk_buff *)list_)->next;
812 if (list == (struct sk_buff *)list_)
813 list = NULL;
814 return list;
818 * skb_peek_tail - peek at the tail of an &sk_buff_head
819 * @list_: list to peek at
821 * Peek an &sk_buff. Unlike most other operations you _MUST_
822 * be careful with this one. A peek leaves the buffer on the
823 * list and someone else may run off with it. You must hold
824 * the appropriate locks or have a private queue to do this.
826 * Returns %NULL for an empty list or a pointer to the tail element.
827 * The reference count is not incremented and the reference is therefore
828 * volatile. Use with caution.
830 static inline struct sk_buff *skb_peek_tail(struct sk_buff_head *list_)
832 struct sk_buff *list = ((struct sk_buff *)list_)->prev;
833 if (list == (struct sk_buff *)list_)
834 list = NULL;
835 return list;
839 * skb_queue_len - get queue length
840 * @list_: list to measure
842 * Return the length of an &sk_buff queue.
844 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
846 return list_->qlen;
850 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
851 * @list: queue to initialize
853 * This initializes only the list and queue length aspects of
854 * an sk_buff_head object. This allows to initialize the list
855 * aspects of an sk_buff_head without reinitializing things like
856 * the spinlock. It can also be used for on-stack sk_buff_head
857 * objects where the spinlock is known to not be used.
859 static inline void __skb_queue_head_init(struct sk_buff_head *list)
861 list->prev = list->next = (struct sk_buff *)list;
862 list->qlen = 0;
866 * This function creates a split out lock class for each invocation;
867 * this is needed for now since a whole lot of users of the skb-queue
868 * infrastructure in drivers have different locking usage (in hardirq)
869 * than the networking core (in softirq only). In the long run either the
870 * network layer or drivers should need annotation to consolidate the
871 * main types of usage into 3 classes.
873 static inline void skb_queue_head_init(struct sk_buff_head *list)
875 spin_lock_init(&list->lock);
876 __skb_queue_head_init(list);
879 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
880 struct lock_class_key *class)
882 skb_queue_head_init(list);
883 lockdep_set_class(&list->lock, class);
887 * Insert an sk_buff on a list.
889 * The "__skb_xxxx()" functions are the non-atomic ones that
890 * can only be called with interrupts disabled.
892 extern void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list);
893 static inline void __skb_insert(struct sk_buff *newsk,
894 struct sk_buff *prev, struct sk_buff *next,
895 struct sk_buff_head *list)
897 newsk->next = next;
898 newsk->prev = prev;
899 next->prev = prev->next = newsk;
900 list->qlen++;
903 static inline void __skb_queue_splice(const struct sk_buff_head *list,
904 struct sk_buff *prev,
905 struct sk_buff *next)
907 struct sk_buff *first = list->next;
908 struct sk_buff *last = list->prev;
910 first->prev = prev;
911 prev->next = first;
913 last->next = next;
914 next->prev = last;
918 * skb_queue_splice - join two skb lists, this is designed for stacks
919 * @list: the new list to add
920 * @head: the place to add it in the first list
922 static inline void skb_queue_splice(const struct sk_buff_head *list,
923 struct sk_buff_head *head)
925 if (!skb_queue_empty(list)) {
926 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
927 head->qlen += list->qlen;
932 * skb_queue_splice - join two skb lists and reinitialise the emptied list
933 * @list: the new list to add
934 * @head: the place to add it in the first list
936 * The list at @list is reinitialised
938 static inline void skb_queue_splice_init(struct sk_buff_head *list,
939 struct sk_buff_head *head)
941 if (!skb_queue_empty(list)) {
942 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
943 head->qlen += list->qlen;
944 __skb_queue_head_init(list);
949 * skb_queue_splice_tail - join two skb lists, each list being a queue
950 * @list: the new list to add
951 * @head: the place to add it in the first list
953 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
954 struct sk_buff_head *head)
956 if (!skb_queue_empty(list)) {
957 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
958 head->qlen += list->qlen;
963 * skb_queue_splice_tail - join two skb lists and reinitialise the emptied list
964 * @list: the new list to add
965 * @head: the place to add it in the first list
967 * Each of the lists is a queue.
968 * The list at @list is reinitialised
970 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
971 struct sk_buff_head *head)
973 if (!skb_queue_empty(list)) {
974 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
975 head->qlen += list->qlen;
976 __skb_queue_head_init(list);
981 * __skb_queue_after - queue a buffer at the list head
982 * @list: list to use
983 * @prev: place after this buffer
984 * @newsk: buffer to queue
986 * Queue a buffer int the middle of a list. This function takes no locks
987 * and you must therefore hold required locks before calling it.
989 * A buffer cannot be placed on two lists at the same time.
991 static inline void __skb_queue_after(struct sk_buff_head *list,
992 struct sk_buff *prev,
993 struct sk_buff *newsk)
995 __skb_insert(newsk, prev, prev->next, list);
998 extern void skb_append(struct sk_buff *old, struct sk_buff *newsk,
999 struct sk_buff_head *list);
1001 static inline void __skb_queue_before(struct sk_buff_head *list,
1002 struct sk_buff *next,
1003 struct sk_buff *newsk)
1005 __skb_insert(newsk, next->prev, next, list);
1009 * __skb_queue_head - queue a buffer at the list head
1010 * @list: list to use
1011 * @newsk: buffer to queue
1013 * Queue a buffer at the start of a list. This function takes no locks
1014 * and you must therefore hold required locks before calling it.
1016 * A buffer cannot be placed on two lists at the same time.
1018 extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1019 static inline void __skb_queue_head(struct sk_buff_head *list,
1020 struct sk_buff *newsk)
1022 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1026 * __skb_queue_tail - queue a buffer at the list tail
1027 * @list: list to use
1028 * @newsk: buffer to queue
1030 * Queue a buffer at the end of a list. This function takes no locks
1031 * and you must therefore hold required locks before calling it.
1033 * A buffer cannot be placed on two lists at the same time.
1035 extern void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1036 static inline void __skb_queue_tail(struct sk_buff_head *list,
1037 struct sk_buff *newsk)
1039 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1043 * remove sk_buff from list. _Must_ be called atomically, and with
1044 * the list known..
1046 extern void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1047 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1049 struct sk_buff *next, *prev;
1051 list->qlen--;
1052 next = skb->next;
1053 prev = skb->prev;
1054 skb->next = skb->prev = NULL;
1055 next->prev = prev;
1056 prev->next = next;
1060 * __skb_dequeue - remove from the head of the queue
1061 * @list: list to dequeue from
1063 * Remove the head of the list. This function does not take any locks
1064 * so must be used with appropriate locks held only. The head item is
1065 * returned or %NULL if the list is empty.
1067 extern struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1068 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1070 struct sk_buff *skb = skb_peek(list);
1071 if (skb)
1072 __skb_unlink(skb, list);
1073 return skb;
1077 * __skb_dequeue_tail - remove from the tail of the queue
1078 * @list: list to dequeue from
1080 * Remove the tail of the list. This function does not take any locks
1081 * so must be used with appropriate locks held only. The tail item is
1082 * returned or %NULL if the list is empty.
1084 extern struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1085 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1087 struct sk_buff *skb = skb_peek_tail(list);
1088 if (skb)
1089 __skb_unlink(skb, list);
1090 return skb;
1094 static inline int skb_is_nonlinear(const struct sk_buff *skb)
1096 return skb->data_len;
1099 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1101 return skb->len - skb->data_len;
1104 static inline int skb_pagelen(const struct sk_buff *skb)
1106 int i, len = 0;
1108 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1109 len += skb_shinfo(skb)->frags[i].size;
1110 return len + skb_headlen(skb);
1113 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1114 struct page *page, int off, int size)
1116 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1118 frag->page = page;
1119 frag->page_offset = off;
1120 frag->size = size;
1121 skb_shinfo(skb)->nr_frags = i + 1;
1124 extern void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page,
1125 int off, int size);
1127 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1128 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1129 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1131 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1132 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1134 return skb->head + skb->tail;
1137 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1139 skb->tail = skb->data - skb->head;
1142 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1144 skb_reset_tail_pointer(skb);
1145 skb->tail += offset;
1147 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1148 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1150 return skb->tail;
1153 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1155 skb->tail = skb->data;
1158 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1160 skb->tail = skb->data + offset;
1163 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1166 * Add data to an sk_buff
1168 extern unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1169 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1171 unsigned char *tmp = skb_tail_pointer(skb);
1172 SKB_LINEAR_ASSERT(skb);
1173 skb->tail += len;
1174 skb->len += len;
1175 return tmp;
1178 extern unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1179 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1181 skb->data -= len;
1182 skb->len += len;
1183 return skb->data;
1186 extern unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1187 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1189 skb->len -= len;
1190 BUG_ON(skb->len < skb->data_len);
1191 return skb->data += len;
1194 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1196 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1199 extern unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1201 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1203 if (len > skb_headlen(skb) &&
1204 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1205 return NULL;
1206 skb->len -= len;
1207 return skb->data += len;
1210 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1212 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1215 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1217 if (likely(len <= skb_headlen(skb)))
1218 return 1;
1219 if (unlikely(len > skb->len))
1220 return 0;
1221 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1225 * skb_headroom - bytes at buffer head
1226 * @skb: buffer to check
1228 * Return the number of bytes of free space at the head of an &sk_buff.
1230 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1232 return skb->data - skb->head;
1236 * skb_tailroom - bytes at buffer end
1237 * @skb: buffer to check
1239 * Return the number of bytes of free space at the tail of an sk_buff
1241 static inline int skb_tailroom(const struct sk_buff *skb)
1243 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1247 * skb_reserve - adjust headroom
1248 * @skb: buffer to alter
1249 * @len: bytes to move
1251 * Increase the headroom of an empty &sk_buff by reducing the tail
1252 * room. This is only allowed for an empty buffer.
1254 static inline void skb_reserve(struct sk_buff *skb, int len)
1256 skb->data += len;
1257 skb->tail += len;
1260 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1261 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1263 return skb->head + skb->transport_header;
1266 static inline void skb_reset_transport_header(struct sk_buff *skb)
1268 skb->transport_header = skb->data - skb->head;
1271 static inline void skb_set_transport_header(struct sk_buff *skb,
1272 const int offset)
1274 skb_reset_transport_header(skb);
1275 skb->transport_header += offset;
1278 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1280 return skb->head + skb->network_header;
1283 static inline void skb_reset_network_header(struct sk_buff *skb)
1285 skb->network_header = skb->data - skb->head;
1288 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1290 skb_reset_network_header(skb);
1291 skb->network_header += offset;
1294 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1296 return skb->head + skb->mac_header;
1299 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1301 return skb->mac_header != ~0U;
1304 static inline void skb_reset_mac_header(struct sk_buff *skb)
1306 skb->mac_header = skb->data - skb->head;
1309 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1311 skb_reset_mac_header(skb);
1312 skb->mac_header += offset;
1315 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1317 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1319 return skb->transport_header;
1322 static inline void skb_reset_transport_header(struct sk_buff *skb)
1324 skb->transport_header = skb->data;
1327 static inline void skb_set_transport_header(struct sk_buff *skb,
1328 const int offset)
1330 skb->transport_header = skb->data + offset;
1333 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1335 return skb->network_header;
1338 static inline void skb_reset_network_header(struct sk_buff *skb)
1340 skb->network_header = skb->data;
1343 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1345 skb->network_header = skb->data + offset;
1348 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1350 return skb->mac_header;
1353 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1355 return skb->mac_header != NULL;
1358 static inline void skb_reset_mac_header(struct sk_buff *skb)
1360 skb->mac_header = skb->data;
1363 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1365 skb->mac_header = skb->data + offset;
1367 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1369 static inline int skb_transport_offset(const struct sk_buff *skb)
1371 return skb_transport_header(skb) - skb->data;
1374 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1376 return skb->transport_header - skb->network_header;
1379 static inline int skb_network_offset(const struct sk_buff *skb)
1381 return skb_network_header(skb) - skb->data;
1384 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1386 return pskb_may_pull(skb, skb_network_offset(skb) + len);
1390 * CPUs often take a performance hit when accessing unaligned memory
1391 * locations. The actual performance hit varies, it can be small if the
1392 * hardware handles it or large if we have to take an exception and fix it
1393 * in software.
1395 * Since an ethernet header is 14 bytes network drivers often end up with
1396 * the IP header at an unaligned offset. The IP header can be aligned by
1397 * shifting the start of the packet by 2 bytes. Drivers should do this
1398 * with:
1400 * skb_reserve(skb, NET_IP_ALIGN);
1402 * The downside to this alignment of the IP header is that the DMA is now
1403 * unaligned. On some architectures the cost of an unaligned DMA is high
1404 * and this cost outweighs the gains made by aligning the IP header.
1406 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1407 * to be overridden.
1409 #ifndef NET_IP_ALIGN
1410 #define NET_IP_ALIGN 2
1411 #endif
1414 * The networking layer reserves some headroom in skb data (via
1415 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
1416 * the header has to grow. In the default case, if the header has to grow
1417 * 32 bytes or less we avoid the reallocation.
1419 * Unfortunately this headroom changes the DMA alignment of the resulting
1420 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
1421 * on some architectures. An architecture can override this value,
1422 * perhaps setting it to a cacheline in size (since that will maintain
1423 * cacheline alignment of the DMA). It must be a power of 2.
1425 * Various parts of the networking layer expect at least 32 bytes of
1426 * headroom, you should not reduce this.
1428 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
1429 * to reduce average number of cache lines per packet.
1430 * get_rps_cpus() for example only access one 64 bytes aligned block :
1431 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
1433 #ifndef NET_SKB_PAD
1434 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
1435 #endif
1437 extern int ___pskb_trim(struct sk_buff *skb, unsigned int len);
1439 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
1441 if (unlikely(skb->data_len)) {
1442 WARN_ON(1);
1443 return;
1445 skb->len = len;
1446 skb_set_tail_pointer(skb, len);
1449 extern void skb_trim(struct sk_buff *skb, unsigned int len);
1451 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
1453 if (skb->data_len)
1454 return ___pskb_trim(skb, len);
1455 __skb_trim(skb, len);
1456 return 0;
1459 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
1461 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
1465 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
1466 * @skb: buffer to alter
1467 * @len: new length
1469 * This is identical to pskb_trim except that the caller knows that
1470 * the skb is not cloned so we should never get an error due to out-
1471 * of-memory.
1473 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
1475 int err = pskb_trim(skb, len);
1476 BUG_ON(err);
1480 * skb_orphan - orphan a buffer
1481 * @skb: buffer to orphan
1483 * If a buffer currently has an owner then we call the owner's
1484 * destructor function and make the @skb unowned. The buffer continues
1485 * to exist but is no longer charged to its former owner.
1487 static inline void skb_orphan(struct sk_buff *skb)
1489 if (skb->destructor)
1490 skb->destructor(skb);
1491 skb->destructor = NULL;
1492 skb->sk = NULL;
1496 * __skb_queue_purge - empty a list
1497 * @list: list to empty
1499 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1500 * the list and one reference dropped. This function does not take the
1501 * list lock and the caller must hold the relevant locks to use it.
1503 extern void skb_queue_purge(struct sk_buff_head *list);
1504 static inline void __skb_queue_purge(struct sk_buff_head *list)
1506 struct sk_buff *skb;
1507 while ((skb = __skb_dequeue(list)) != NULL)
1508 kfree_skb(skb);
1512 * __dev_alloc_skb - allocate an skbuff for receiving
1513 * @length: length to allocate
1514 * @gfp_mask: get_free_pages mask, passed to alloc_skb
1516 * Allocate a new &sk_buff and assign it a usage count of one. The
1517 * buffer has unspecified headroom built in. Users should allocate
1518 * the headroom they think they need without accounting for the
1519 * built in space. The built in space is used for optimisations.
1521 * %NULL is returned if there is no free memory.
1523 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
1524 gfp_t gfp_mask)
1526 struct sk_buff *skb = alloc_skb(length + NET_SKB_PAD, gfp_mask);
1527 if (likely(skb))
1528 skb_reserve(skb, NET_SKB_PAD);
1529 return skb;
1532 extern struct sk_buff *dev_alloc_skb(unsigned int length);
1534 extern struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
1535 unsigned int length, gfp_t gfp_mask);
1538 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
1539 * @dev: network device to receive on
1540 * @length: length to allocate
1542 * Allocate a new &sk_buff and assign it a usage count of one. The
1543 * buffer has unspecified headroom built in. Users should allocate
1544 * the headroom they think they need without accounting for the
1545 * built in space. The built in space is used for optimisations.
1547 * %NULL is returned if there is no free memory. Although this function
1548 * allocates memory it can be called from an interrupt.
1550 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
1551 unsigned int length)
1553 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
1556 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
1557 unsigned int length)
1559 struct sk_buff *skb = netdev_alloc_skb(dev, length + NET_IP_ALIGN);
1561 if (NET_IP_ALIGN && skb)
1562 skb_reserve(skb, NET_IP_ALIGN);
1563 return skb;
1566 extern struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask);
1569 * netdev_alloc_page - allocate a page for ps-rx on a specific device
1570 * @dev: network device to receive on
1572 * Allocate a new page node local to the specified device.
1574 * %NULL is returned if there is no free memory.
1576 static inline struct page *netdev_alloc_page(struct net_device *dev)
1578 return __netdev_alloc_page(dev, GFP_ATOMIC);
1581 static inline void netdev_free_page(struct net_device *dev, struct page *page)
1583 __free_page(page);
1587 * skb_clone_writable - is the header of a clone writable
1588 * @skb: buffer to check
1589 * @len: length up to which to write
1591 * Returns true if modifying the header part of the cloned buffer
1592 * does not requires the data to be copied.
1594 static inline int skb_clone_writable(struct sk_buff *skb, unsigned int len)
1596 return !skb_header_cloned(skb) &&
1597 skb_headroom(skb) + len <= skb->hdr_len;
1600 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
1601 int cloned)
1603 int delta = 0;
1605 if (headroom < NET_SKB_PAD)
1606 headroom = NET_SKB_PAD;
1607 if (headroom > skb_headroom(skb))
1608 delta = headroom - skb_headroom(skb);
1610 if (delta || cloned)
1611 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
1612 GFP_ATOMIC);
1613 return 0;
1617 * skb_cow - copy header of skb when it is required
1618 * @skb: buffer to cow
1619 * @headroom: needed headroom
1621 * If the skb passed lacks sufficient headroom or its data part
1622 * is shared, data is reallocated. If reallocation fails, an error
1623 * is returned and original skb is not changed.
1625 * The result is skb with writable area skb->head...skb->tail
1626 * and at least @headroom of space at head.
1628 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
1630 return __skb_cow(skb, headroom, skb_cloned(skb));
1634 * skb_cow_head - skb_cow but only making the head writable
1635 * @skb: buffer to cow
1636 * @headroom: needed headroom
1638 * This function is identical to skb_cow except that we replace the
1639 * skb_cloned check by skb_header_cloned. It should be used when
1640 * you only need to push on some header and do not need to modify
1641 * the data.
1643 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
1645 return __skb_cow(skb, headroom, skb_header_cloned(skb));
1649 * skb_padto - pad an skbuff up to a minimal size
1650 * @skb: buffer to pad
1651 * @len: minimal length
1653 * Pads up a buffer to ensure the trailing bytes exist and are
1654 * blanked. If the buffer already contains sufficient data it
1655 * is untouched. Otherwise it is extended. Returns zero on
1656 * success. The skb is freed on error.
1659 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
1661 unsigned int size = skb->len;
1662 if (likely(size >= len))
1663 return 0;
1664 return skb_pad(skb, len - size);
1667 static inline int skb_add_data(struct sk_buff *skb,
1668 char __user *from, int copy)
1670 const int off = skb->len;
1672 if (skb->ip_summed == CHECKSUM_NONE) {
1673 int err = 0;
1674 __wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
1675 copy, 0, &err);
1676 if (!err) {
1677 skb->csum = csum_block_add(skb->csum, csum, off);
1678 return 0;
1680 } else if (!copy_from_user(skb_put(skb, copy), from, copy))
1681 return 0;
1683 __skb_trim(skb, off);
1684 return -EFAULT;
1687 static inline int skb_can_coalesce(struct sk_buff *skb, int i,
1688 struct page *page, int off)
1690 if (i) {
1691 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
1693 return page == frag->page &&
1694 off == frag->page_offset + frag->size;
1696 return 0;
1699 static inline int __skb_linearize(struct sk_buff *skb)
1701 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
1705 * skb_linearize - convert paged skb to linear one
1706 * @skb: buffer to linarize
1708 * If there is no free memory -ENOMEM is returned, otherwise zero
1709 * is returned and the old skb data released.
1711 static inline int skb_linearize(struct sk_buff *skb)
1713 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
1717 * skb_linearize_cow - make sure skb is linear and writable
1718 * @skb: buffer to process
1720 * If there is no free memory -ENOMEM is returned, otherwise zero
1721 * is returned and the old skb data released.
1723 static inline int skb_linearize_cow(struct sk_buff *skb)
1725 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
1726 __skb_linearize(skb) : 0;
1730 * skb_postpull_rcsum - update checksum for received skb after pull
1731 * @skb: buffer to update
1732 * @start: start of data before pull
1733 * @len: length of data pulled
1735 * After doing a pull on a received packet, you need to call this to
1736 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
1737 * CHECKSUM_NONE so that it can be recomputed from scratch.
1740 static inline void skb_postpull_rcsum(struct sk_buff *skb,
1741 const void *start, unsigned int len)
1743 if (skb->ip_summed == CHECKSUM_COMPLETE)
1744 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
1747 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
1750 * pskb_trim_rcsum - trim received skb and update checksum
1751 * @skb: buffer to trim
1752 * @len: new length
1754 * This is exactly the same as pskb_trim except that it ensures the
1755 * checksum of received packets are still valid after the operation.
1758 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
1760 if (likely(len >= skb->len))
1761 return 0;
1762 if (skb->ip_summed == CHECKSUM_COMPLETE)
1763 skb->ip_summed = CHECKSUM_NONE;
1764 return __pskb_trim(skb, len);
1767 #define skb_queue_walk(queue, skb) \
1768 for (skb = (queue)->next; \
1769 prefetch(skb->next), (skb != (struct sk_buff *)(queue)); \
1770 skb = skb->next)
1772 #define skb_queue_walk_safe(queue, skb, tmp) \
1773 for (skb = (queue)->next, tmp = skb->next; \
1774 skb != (struct sk_buff *)(queue); \
1775 skb = tmp, tmp = skb->next)
1777 #define skb_queue_walk_from(queue, skb) \
1778 for (; prefetch(skb->next), (skb != (struct sk_buff *)(queue)); \
1779 skb = skb->next)
1781 #define skb_queue_walk_from_safe(queue, skb, tmp) \
1782 for (tmp = skb->next; \
1783 skb != (struct sk_buff *)(queue); \
1784 skb = tmp, tmp = skb->next)
1786 #define skb_queue_reverse_walk(queue, skb) \
1787 for (skb = (queue)->prev; \
1788 prefetch(skb->prev), (skb != (struct sk_buff *)(queue)); \
1789 skb = skb->prev)
1792 static inline bool skb_has_frag_list(const struct sk_buff *skb)
1794 return skb_shinfo(skb)->frag_list != NULL;
1797 static inline void skb_frag_list_init(struct sk_buff *skb)
1799 skb_shinfo(skb)->frag_list = NULL;
1802 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
1804 frag->next = skb_shinfo(skb)->frag_list;
1805 skb_shinfo(skb)->frag_list = frag;
1808 #define skb_walk_frags(skb, iter) \
1809 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
1811 extern struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
1812 int *peeked, int *err);
1813 extern struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
1814 int noblock, int *err);
1815 extern unsigned int datagram_poll(struct file *file, struct socket *sock,
1816 struct poll_table_struct *wait);
1817 extern int skb_copy_datagram_iovec(const struct sk_buff *from,
1818 int offset, struct iovec *to,
1819 int size);
1820 extern int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
1821 int hlen,
1822 struct iovec *iov);
1823 extern int skb_copy_datagram_from_iovec(struct sk_buff *skb,
1824 int offset,
1825 const struct iovec *from,
1826 int from_offset,
1827 int len);
1828 extern int skb_copy_datagram_const_iovec(const struct sk_buff *from,
1829 int offset,
1830 const struct iovec *to,
1831 int to_offset,
1832 int size);
1833 extern void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
1834 extern void skb_free_datagram_locked(struct sock *sk,
1835 struct sk_buff *skb);
1836 extern int skb_kill_datagram(struct sock *sk, struct sk_buff *skb,
1837 unsigned int flags);
1838 extern __wsum skb_checksum(const struct sk_buff *skb, int offset,
1839 int len, __wsum csum);
1840 extern int skb_copy_bits(const struct sk_buff *skb, int offset,
1841 void *to, int len);
1842 extern int skb_store_bits(struct sk_buff *skb, int offset,
1843 const void *from, int len);
1844 extern __wsum skb_copy_and_csum_bits(const struct sk_buff *skb,
1845 int offset, u8 *to, int len,
1846 __wsum csum);
1847 extern int skb_splice_bits(struct sk_buff *skb,
1848 unsigned int offset,
1849 struct pipe_inode_info *pipe,
1850 unsigned int len,
1851 unsigned int flags);
1852 extern void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
1853 extern void skb_split(struct sk_buff *skb,
1854 struct sk_buff *skb1, const u32 len);
1855 extern int skb_shift(struct sk_buff *tgt, struct sk_buff *skb,
1856 int shiftlen);
1858 extern struct sk_buff *skb_segment(struct sk_buff *skb, int features);
1860 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
1861 int len, void *buffer)
1863 int hlen = skb_headlen(skb);
1865 if (hlen - offset >= len)
1866 return skb->data + offset;
1868 if (skb_copy_bits(skb, offset, buffer, len) < 0)
1869 return NULL;
1871 return buffer;
1874 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
1875 void *to,
1876 const unsigned int len)
1878 memcpy(to, skb->data, len);
1881 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
1882 const int offset, void *to,
1883 const unsigned int len)
1885 memcpy(to, skb->data + offset, len);
1888 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
1889 const void *from,
1890 const unsigned int len)
1892 memcpy(skb->data, from, len);
1895 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
1896 const int offset,
1897 const void *from,
1898 const unsigned int len)
1900 memcpy(skb->data + offset, from, len);
1903 extern void skb_init(void);
1905 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
1907 return skb->tstamp;
1911 * skb_get_timestamp - get timestamp from a skb
1912 * @skb: skb to get stamp from
1913 * @stamp: pointer to struct timeval to store stamp in
1915 * Timestamps are stored in the skb as offsets to a base timestamp.
1916 * This function converts the offset back to a struct timeval and stores
1917 * it in stamp.
1919 static inline void skb_get_timestamp(const struct sk_buff *skb,
1920 struct timeval *stamp)
1922 *stamp = ktime_to_timeval(skb->tstamp);
1925 static inline void skb_get_timestampns(const struct sk_buff *skb,
1926 struct timespec *stamp)
1928 *stamp = ktime_to_timespec(skb->tstamp);
1931 static inline void __net_timestamp(struct sk_buff *skb)
1933 skb->tstamp = ktime_get_real();
1936 static inline ktime_t net_timedelta(ktime_t t)
1938 return ktime_sub(ktime_get_real(), t);
1941 static inline ktime_t net_invalid_timestamp(void)
1943 return ktime_set(0, 0);
1946 extern void skb_timestamping_init(void);
1948 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
1950 extern void skb_clone_tx_timestamp(struct sk_buff *skb);
1951 extern bool skb_defer_rx_timestamp(struct sk_buff *skb);
1953 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
1955 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
1959 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
1961 return false;
1964 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
1967 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
1969 * @skb: clone of the the original outgoing packet
1970 * @hwtstamps: hardware time stamps
1973 void skb_complete_tx_timestamp(struct sk_buff *skb,
1974 struct skb_shared_hwtstamps *hwtstamps);
1977 * skb_tstamp_tx - queue clone of skb with send time stamps
1978 * @orig_skb: the original outgoing packet
1979 * @hwtstamps: hardware time stamps, may be NULL if not available
1981 * If the skb has a socket associated, then this function clones the
1982 * skb (thus sharing the actual data and optional structures), stores
1983 * the optional hardware time stamping information (if non NULL) or
1984 * generates a software time stamp (otherwise), then queues the clone
1985 * to the error queue of the socket. Errors are silently ignored.
1987 extern void skb_tstamp_tx(struct sk_buff *orig_skb,
1988 struct skb_shared_hwtstamps *hwtstamps);
1990 static inline void sw_tx_timestamp(struct sk_buff *skb)
1992 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
1993 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
1994 skb_tstamp_tx(skb, NULL);
1998 * skb_tx_timestamp() - Driver hook for transmit timestamping
2000 * Ethernet MAC Drivers should call this function in their hard_xmit()
2001 * function as soon as possible after giving the sk_buff to the MAC
2002 * hardware, but before freeing the sk_buff.
2004 * @skb: A socket buffer.
2006 static inline void skb_tx_timestamp(struct sk_buff *skb)
2008 skb_clone_tx_timestamp(skb);
2009 sw_tx_timestamp(skb);
2012 extern __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2013 extern __sum16 __skb_checksum_complete(struct sk_buff *skb);
2015 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2017 return skb->ip_summed & CHECKSUM_UNNECESSARY;
2021 * skb_checksum_complete - Calculate checksum of an entire packet
2022 * @skb: packet to process
2024 * This function calculates the checksum over the entire packet plus
2025 * the value of skb->csum. The latter can be used to supply the
2026 * checksum of a pseudo header as used by TCP/UDP. It returns the
2027 * checksum.
2029 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2030 * this function can be used to verify that checksum on received
2031 * packets. In that case the function should return zero if the
2032 * checksum is correct. In particular, this function will return zero
2033 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2034 * hardware has already verified the correctness of the checksum.
2036 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
2038 return skb_csum_unnecessary(skb) ?
2039 0 : __skb_checksum_complete(skb);
2042 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2043 extern void nf_conntrack_destroy(struct nf_conntrack *nfct);
2044 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
2046 if (nfct && atomic_dec_and_test(&nfct->use))
2047 nf_conntrack_destroy(nfct);
2049 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
2051 if (nfct)
2052 atomic_inc(&nfct->use);
2054 static inline void nf_conntrack_get_reasm(struct sk_buff *skb)
2056 if (skb)
2057 atomic_inc(&skb->users);
2059 static inline void nf_conntrack_put_reasm(struct sk_buff *skb)
2061 if (skb)
2062 kfree_skb(skb);
2064 #endif
2065 #ifdef CONFIG_BRIDGE_NETFILTER
2066 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
2068 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
2069 kfree(nf_bridge);
2071 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
2073 if (nf_bridge)
2074 atomic_inc(&nf_bridge->use);
2076 #endif /* CONFIG_BRIDGE_NETFILTER */
2077 static inline void nf_reset(struct sk_buff *skb)
2079 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2080 nf_conntrack_put(skb->nfct);
2081 skb->nfct = NULL;
2082 nf_conntrack_put_reasm(skb->nfct_reasm);
2083 skb->nfct_reasm = NULL;
2084 #endif
2085 #ifdef CONFIG_BRIDGE_NETFILTER
2086 nf_bridge_put(skb->nf_bridge);
2087 skb->nf_bridge = NULL;
2088 #endif
2091 /* Note: This doesn't put any conntrack and bridge info in dst. */
2092 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2094 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2095 dst->nfct = src->nfct;
2096 nf_conntrack_get(src->nfct);
2097 dst->nfctinfo = src->nfctinfo;
2098 dst->nfct_reasm = src->nfct_reasm;
2099 nf_conntrack_get_reasm(src->nfct_reasm);
2100 #endif
2101 #ifdef CONFIG_BRIDGE_NETFILTER
2102 dst->nf_bridge = src->nf_bridge;
2103 nf_bridge_get(src->nf_bridge);
2104 #endif
2107 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2109 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2110 nf_conntrack_put(dst->nfct);
2111 nf_conntrack_put_reasm(dst->nfct_reasm);
2112 #endif
2113 #ifdef CONFIG_BRIDGE_NETFILTER
2114 nf_bridge_put(dst->nf_bridge);
2115 #endif
2116 __nf_copy(dst, src);
2119 #ifdef CONFIG_NETWORK_SECMARK
2120 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2122 to->secmark = from->secmark;
2125 static inline void skb_init_secmark(struct sk_buff *skb)
2127 skb->secmark = 0;
2129 #else
2130 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2133 static inline void skb_init_secmark(struct sk_buff *skb)
2135 #endif
2137 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
2139 skb->queue_mapping = queue_mapping;
2142 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
2144 return skb->queue_mapping;
2147 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
2149 to->queue_mapping = from->queue_mapping;
2152 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
2154 skb->queue_mapping = rx_queue + 1;
2157 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
2159 return skb->queue_mapping - 1;
2162 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
2164 return skb->queue_mapping != 0;
2167 extern u16 skb_tx_hash(const struct net_device *dev,
2168 const struct sk_buff *skb);
2170 #ifdef CONFIG_XFRM
2171 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
2173 return skb->sp;
2175 #else
2176 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
2178 return NULL;
2180 #endif
2182 static inline int skb_is_gso(const struct sk_buff *skb)
2184 return skb_shinfo(skb)->gso_size;
2187 static inline int skb_is_gso_v6(const struct sk_buff *skb)
2189 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
2192 extern void __skb_warn_lro_forwarding(const struct sk_buff *skb);
2194 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
2196 /* LRO sets gso_size but not gso_type, whereas if GSO is really
2197 * wanted then gso_type will be set. */
2198 struct skb_shared_info *shinfo = skb_shinfo(skb);
2199 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
2200 unlikely(shinfo->gso_type == 0)) {
2201 __skb_warn_lro_forwarding(skb);
2202 return true;
2204 return false;
2207 static inline void skb_forward_csum(struct sk_buff *skb)
2209 /* Unfortunately we don't support this one. Any brave souls? */
2210 if (skb->ip_summed == CHECKSUM_COMPLETE)
2211 skb->ip_summed = CHECKSUM_NONE;
2215 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
2216 * @skb: skb to check
2218 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
2219 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
2220 * use this helper, to document places where we make this assertion.
2222 static inline void skb_checksum_none_assert(struct sk_buff *skb)
2224 #ifdef DEBUG
2225 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
2226 #endif
2229 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
2230 #endif /* __KERNEL__ */
2231 #endif /* _LINUX_SKBUFF_H */