2 * Definitions for the 'struct sk_buff' memory handlers.
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/compiler.h>
19 #include <linux/time.h>
20 #include <linux/cache.h>
22 #include <asm/atomic.h>
23 #include <asm/types.h>
24 #include <linux/spinlock.h>
26 #include <linux/highmem.h>
27 #include <linux/poll.h>
28 #include <linux/net.h>
29 #include <linux/textsearch.h>
30 #include <net/checksum.h>
31 #include <linux/dmaengine.h>
33 #define HAVE_ALLOC_SKB /* For the drivers to know */
34 #define HAVE_ALIGNABLE_SKB /* Ditto 8) */
36 #define CHECKSUM_NONE 0
38 #define CHECKSUM_UNNECESSARY 2
40 #define SKB_DATA_ALIGN(X) (((X) + (SMP_CACHE_BYTES - 1)) & \
41 ~(SMP_CACHE_BYTES - 1))
42 #define SKB_MAX_ORDER(X, ORDER) (((PAGE_SIZE << (ORDER)) - (X) - \
43 sizeof(struct skb_shared_info)) & \
44 ~(SMP_CACHE_BYTES - 1))
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 * HW: 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 HW,
65 * B. Checksumming on output.
67 * NONE: skb is checksummed by protocol or csum is not required.
69 * HW: device is required to csum packet as seen by hard_start_xmit
70 * from skb->h.raw to the end and to record the checksum
71 * at skb->h.raw+skb->csum.
73 * Device must show its capabilities in dev->features, set
74 * at device setup time.
75 * NETIF_F_HW_CSUM - it is clever device, it is able to checksum
77 * NETIF_F_NO_CSUM - loopback or reliable single hop media.
78 * NETIF_F_IP_CSUM - device is dumb. It is able to csum only
79 * TCP/UDP over IPv4. Sigh. Vendors like this
80 * way by an unknown reason. Though, see comment above
81 * about CHECKSUM_UNNECESSARY. 8)
83 * Any questions? No questions, good. --ANK
88 #ifdef CONFIG_NETFILTER
91 void (*destroy)(struct nf_conntrack *);
94 #ifdef CONFIG_BRIDGE_NETFILTER
95 struct nf_bridge_info {
97 struct net_device *physindev;
98 struct net_device *physoutdev;
99 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
100 struct net_device *netoutdev;
103 unsigned long data[32 / sizeof(unsigned long)];
109 struct sk_buff_head {
110 /* These two members must be first. */
111 struct sk_buff *next;
112 struct sk_buff *prev;
120 /* To allow 64K frame to be packed as single skb without frag_list */
121 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 2)
123 typedef struct skb_frag_struct skb_frag_t;
125 struct skb_frag_struct {
131 /* This data is invariant across clones and lives at
132 * the end of the header data, ie. at skb->end.
134 struct skb_shared_info {
136 unsigned short nr_frags;
137 unsigned short tso_size;
138 unsigned short tso_segs;
139 unsigned short ufo_size;
140 unsigned int ip6_frag_id;
141 struct sk_buff *frag_list;
142 skb_frag_t frags[MAX_SKB_FRAGS];
145 /* We divide dataref into two halves. The higher 16 bits hold references
146 * to the payload part of skb->data. The lower 16 bits hold references to
147 * the entire skb->data. It is up to the users of the skb to agree on
148 * where the payload starts.
150 * All users must obey the rule that the skb->data reference count must be
151 * greater than or equal to the payload reference count.
153 * Holding a reference to the payload part means that the user does not
154 * care about modifications to the header part of skb->data.
156 #define SKB_DATAREF_SHIFT 16
157 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
166 SKB_FCLONE_UNAVAILABLE,
172 * struct sk_buff - socket buffer
173 * @next: Next buffer in list
174 * @prev: Previous buffer in list
175 * @sk: Socket we are owned by
176 * @tstamp: Time we arrived
177 * @dev: Device we arrived on/are leaving by
178 * @input_dev: Device we arrived on
179 * @h: Transport layer header
180 * @nh: Network layer header
181 * @mac: Link layer header
182 * @dst: destination entry
183 * @sp: the security path, used for xfrm
184 * @cb: Control buffer. Free for use by every layer. Put private vars here
185 * @len: Length of actual data
186 * @data_len: Data length
187 * @mac_len: Length of link layer header
189 * @local_df: allow local fragmentation
190 * @cloned: Head may be cloned (check refcnt to be sure)
191 * @nohdr: Payload reference only, must not modify header
192 * @pkt_type: Packet class
193 * @fclone: skbuff clone status
194 * @ip_summed: Driver fed us an IP checksum
195 * @priority: Packet queueing priority
196 * @users: User count - see {datagram,tcp}.c
197 * @protocol: Packet protocol from driver
198 * @truesize: Buffer size
199 * @head: Head of buffer
200 * @data: Data head pointer
201 * @tail: Tail pointer
203 * @destructor: Destruct function
204 * @nfmark: Can be used for communication between hooks
205 * @nfct: Associated connection, if any
206 * @ipvs_property: skbuff is owned by ipvs
207 * @nfctinfo: Relationship of this skb to the connection
208 * @nfct_reasm: netfilter conntrack re-assembly pointer
209 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
210 * @tc_index: Traffic control index
211 * @tc_verd: traffic control verdict
212 * @secmark: security marking
216 /* These two members must be first. */
217 struct sk_buff *next;
218 struct sk_buff *prev;
221 struct skb_timeval tstamp;
222 struct net_device *dev;
223 struct net_device *input_dev;
228 struct icmphdr *icmph;
229 struct igmphdr *igmph;
231 struct ipv6hdr *ipv6h;
237 struct ipv6hdr *ipv6h;
246 struct dst_entry *dst;
250 * This is the control buffer. It is free to use for every
251 * layer. Please put your private variables there. If you
252 * want to keep them across layers you have to do a skb_clone()
253 * first. This is owned by whoever has the skb queued ATM.
272 void (*destructor)(struct sk_buff *skb);
273 #ifdef CONFIG_NETFILTER
274 struct nf_conntrack *nfct;
275 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
276 struct sk_buff *nfct_reasm;
278 #ifdef CONFIG_BRIDGE_NETFILTER
279 struct nf_bridge_info *nf_bridge;
282 #endif /* CONFIG_NETFILTER */
283 #ifdef CONFIG_NET_SCHED
284 __u16 tc_index; /* traffic control index */
285 #ifdef CONFIG_NET_CLS_ACT
286 __u16 tc_verd; /* traffic control verdict */
289 #ifdef CONFIG_NET_DMA
290 dma_cookie_t dma_cookie;
292 #ifdef CONFIG_NETWORK_SECMARK
297 /* These elements must be at the end, see alloc_skb() for details. */
298 unsigned int truesize;
308 * Handling routines are only of interest to the kernel
310 #include <linux/slab.h>
312 #include <asm/system.h>
314 extern void kfree_skb(struct sk_buff *skb);
315 extern void __kfree_skb(struct sk_buff *skb);
316 extern struct sk_buff *__alloc_skb(unsigned int size,
317 gfp_t priority, int fclone);
318 static inline struct sk_buff *alloc_skb(unsigned int size,
321 return __alloc_skb(size, priority, 0);
324 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
327 return __alloc_skb(size, priority, 1);
330 extern struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp,
333 extern void kfree_skbmem(struct sk_buff *skb);
334 extern struct sk_buff *skb_clone(struct sk_buff *skb,
336 extern struct sk_buff *skb_copy(const struct sk_buff *skb,
338 extern struct sk_buff *pskb_copy(struct sk_buff *skb,
340 extern int pskb_expand_head(struct sk_buff *skb,
341 int nhead, int ntail,
343 extern struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
344 unsigned int headroom);
345 extern struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
346 int newheadroom, int newtailroom,
348 extern struct sk_buff * skb_pad(struct sk_buff *skb, int pad);
349 #define dev_kfree_skb(a) kfree_skb(a)
350 extern void skb_over_panic(struct sk_buff *skb, int len,
352 extern void skb_under_panic(struct sk_buff *skb, int len,
354 extern void skb_truesize_bug(struct sk_buff *skb);
356 static inline void skb_truesize_check(struct sk_buff *skb)
358 if (unlikely((int)skb->truesize < sizeof(struct sk_buff) + skb->len))
359 skb_truesize_bug(skb);
362 extern int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
363 int getfrag(void *from, char *to, int offset,
364 int len,int odd, struct sk_buff *skb),
365 void *from, int length);
372 __u32 stepped_offset;
373 struct sk_buff *root_skb;
374 struct sk_buff *cur_skb;
378 extern void skb_prepare_seq_read(struct sk_buff *skb,
379 unsigned int from, unsigned int to,
380 struct skb_seq_state *st);
381 extern unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
382 struct skb_seq_state *st);
383 extern void skb_abort_seq_read(struct skb_seq_state *st);
385 extern unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
386 unsigned int to, struct ts_config *config,
387 struct ts_state *state);
390 #define skb_shinfo(SKB) ((struct skb_shared_info *)((SKB)->end))
393 * skb_queue_empty - check if a queue is empty
396 * Returns true if the queue is empty, false otherwise.
398 static inline int skb_queue_empty(const struct sk_buff_head *list)
400 return list->next == (struct sk_buff *)list;
404 * skb_get - reference buffer
405 * @skb: buffer to reference
407 * Makes another reference to a socket buffer and returns a pointer
410 static inline struct sk_buff *skb_get(struct sk_buff *skb)
412 atomic_inc(&skb->users);
417 * If users == 1, we are the only owner and are can avoid redundant
422 * skb_cloned - is the buffer a clone
423 * @skb: buffer to check
425 * Returns true if the buffer was generated with skb_clone() and is
426 * one of multiple shared copies of the buffer. Cloned buffers are
427 * shared data so must not be written to under normal circumstances.
429 static inline int skb_cloned(const struct sk_buff *skb)
431 return skb->cloned &&
432 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
436 * skb_header_cloned - is the header a clone
437 * @skb: buffer to check
439 * Returns true if modifying the header part of the buffer requires
440 * the data to be copied.
442 static inline int skb_header_cloned(const struct sk_buff *skb)
449 dataref = atomic_read(&skb_shinfo(skb)->dataref);
450 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
455 * skb_header_release - release reference to header
456 * @skb: buffer to operate on
458 * Drop a reference to the header part of the buffer. This is done
459 * by acquiring a payload reference. You must not read from the header
460 * part of skb->data after this.
462 static inline void skb_header_release(struct sk_buff *skb)
466 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
470 * skb_shared - is the buffer shared
471 * @skb: buffer to check
473 * Returns true if more than one person has a reference to this
476 static inline int skb_shared(const struct sk_buff *skb)
478 return atomic_read(&skb->users) != 1;
482 * skb_share_check - check if buffer is shared and if so clone it
483 * @skb: buffer to check
484 * @pri: priority for memory allocation
486 * If the buffer is shared the buffer is cloned and the old copy
487 * drops a reference. A new clone with a single reference is returned.
488 * If the buffer is not shared the original buffer is returned. When
489 * being called from interrupt status or with spinlocks held pri must
492 * NULL is returned on a memory allocation failure.
494 static inline struct sk_buff *skb_share_check(struct sk_buff *skb,
497 might_sleep_if(pri & __GFP_WAIT);
498 if (skb_shared(skb)) {
499 struct sk_buff *nskb = skb_clone(skb, pri);
507 * Copy shared buffers into a new sk_buff. We effectively do COW on
508 * packets to handle cases where we have a local reader and forward
509 * and a couple of other messy ones. The normal one is tcpdumping
510 * a packet thats being forwarded.
514 * skb_unshare - make a copy of a shared buffer
515 * @skb: buffer to check
516 * @pri: priority for memory allocation
518 * If the socket buffer is a clone then this function creates a new
519 * copy of the data, drops a reference count on the old copy and returns
520 * the new copy with the reference count at 1. If the buffer is not a clone
521 * the original buffer is returned. When called with a spinlock held or
522 * from interrupt state @pri must be %GFP_ATOMIC
524 * %NULL is returned on a memory allocation failure.
526 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
529 might_sleep_if(pri & __GFP_WAIT);
530 if (skb_cloned(skb)) {
531 struct sk_buff *nskb = skb_copy(skb, pri);
532 kfree_skb(skb); /* Free our shared copy */
540 * @list_: list to peek at
542 * Peek an &sk_buff. Unlike most other operations you _MUST_
543 * be careful with this one. A peek leaves the buffer on the
544 * list and someone else may run off with it. You must hold
545 * the appropriate locks or have a private queue to do this.
547 * Returns %NULL for an empty list or a pointer to the head element.
548 * The reference count is not incremented and the reference is therefore
549 * volatile. Use with caution.
551 static inline struct sk_buff *skb_peek(struct sk_buff_head *list_)
553 struct sk_buff *list = ((struct sk_buff *)list_)->next;
554 if (list == (struct sk_buff *)list_)
561 * @list_: list to peek at
563 * Peek an &sk_buff. Unlike most other operations you _MUST_
564 * be careful with this one. A peek leaves the buffer on the
565 * list and someone else may run off with it. You must hold
566 * the appropriate locks or have a private queue to do this.
568 * Returns %NULL for an empty list or a pointer to the tail element.
569 * The reference count is not incremented and the reference is therefore
570 * volatile. Use with caution.
572 static inline struct sk_buff *skb_peek_tail(struct sk_buff_head *list_)
574 struct sk_buff *list = ((struct sk_buff *)list_)->prev;
575 if (list == (struct sk_buff *)list_)
581 * skb_queue_len - get queue length
582 * @list_: list to measure
584 * Return the length of an &sk_buff queue.
586 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
591 static inline void skb_queue_head_init(struct sk_buff_head *list)
593 spin_lock_init(&list->lock);
594 list->prev = list->next = (struct sk_buff *)list;
599 * Insert an sk_buff at the start of a list.
601 * The "__skb_xxxx()" functions are the non-atomic ones that
602 * can only be called with interrupts disabled.
606 * __skb_queue_after - queue a buffer at the list head
608 * @prev: place after this buffer
609 * @newsk: buffer to queue
611 * Queue a buffer int the middle of a list. This function takes no locks
612 * and you must therefore hold required locks before calling it.
614 * A buffer cannot be placed on two lists at the same time.
616 static inline void __skb_queue_after(struct sk_buff_head *list,
617 struct sk_buff *prev,
618 struct sk_buff *newsk)
620 struct sk_buff *next;
626 next->prev = prev->next = newsk;
630 * __skb_queue_head - queue a buffer at the list head
632 * @newsk: buffer to queue
634 * Queue a buffer at the start of a list. This function takes no locks
635 * and you must therefore hold required locks before calling it.
637 * A buffer cannot be placed on two lists at the same time.
639 extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
640 static inline void __skb_queue_head(struct sk_buff_head *list,
641 struct sk_buff *newsk)
643 __skb_queue_after(list, (struct sk_buff *)list, newsk);
647 * __skb_queue_tail - queue a buffer at the list tail
649 * @newsk: buffer to queue
651 * Queue a buffer at the end of a list. This function takes no locks
652 * and you must therefore hold required locks before calling it.
654 * A buffer cannot be placed on two lists at the same time.
656 extern void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
657 static inline void __skb_queue_tail(struct sk_buff_head *list,
658 struct sk_buff *newsk)
660 struct sk_buff *prev, *next;
663 next = (struct sk_buff *)list;
667 next->prev = prev->next = newsk;
672 * __skb_dequeue - remove from the head of the queue
673 * @list: list to dequeue from
675 * Remove the head of the list. This function does not take any locks
676 * so must be used with appropriate locks held only. The head item is
677 * returned or %NULL if the list is empty.
679 extern struct sk_buff *skb_dequeue(struct sk_buff_head *list);
680 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
682 struct sk_buff *next, *prev, *result;
684 prev = (struct sk_buff *) list;
693 result->next = result->prev = NULL;
700 * Insert a packet on a list.
702 extern void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list);
703 static inline void __skb_insert(struct sk_buff *newsk,
704 struct sk_buff *prev, struct sk_buff *next,
705 struct sk_buff_head *list)
709 next->prev = prev->next = newsk;
714 * Place a packet after a given packet in a list.
716 extern void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list);
717 static inline void __skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
719 __skb_insert(newsk, old, old->next, list);
723 * remove sk_buff from list. _Must_ be called atomically, and with
726 extern void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
727 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
729 struct sk_buff *next, *prev;
734 skb->next = skb->prev = NULL;
740 /* XXX: more streamlined implementation */
743 * __skb_dequeue_tail - remove from the tail of the queue
744 * @list: list to dequeue from
746 * Remove the tail of the list. This function does not take any locks
747 * so must be used with appropriate locks held only. The tail item is
748 * returned or %NULL if the list is empty.
750 extern struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
751 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
753 struct sk_buff *skb = skb_peek_tail(list);
755 __skb_unlink(skb, list);
760 static inline int skb_is_nonlinear(const struct sk_buff *skb)
762 return skb->data_len;
765 static inline unsigned int skb_headlen(const struct sk_buff *skb)
767 return skb->len - skb->data_len;
770 static inline int skb_pagelen(const struct sk_buff *skb)
774 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
775 len += skb_shinfo(skb)->frags[i].size;
776 return len + skb_headlen(skb);
779 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
780 struct page *page, int off, int size)
782 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
785 frag->page_offset = off;
787 skb_shinfo(skb)->nr_frags = i + 1;
790 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
791 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_shinfo(skb)->frag_list)
792 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
795 * Add data to an sk_buff
797 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
799 unsigned char *tmp = skb->tail;
800 SKB_LINEAR_ASSERT(skb);
807 * skb_put - add data to a buffer
808 * @skb: buffer to use
809 * @len: amount of data to add
811 * This function extends the used data area of the buffer. If this would
812 * exceed the total buffer size the kernel will panic. A pointer to the
813 * first byte of the extra data is returned.
815 static inline unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
817 unsigned char *tmp = skb->tail;
818 SKB_LINEAR_ASSERT(skb);
821 if (unlikely(skb->tail>skb->end))
822 skb_over_panic(skb, len, current_text_addr());
826 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
834 * skb_push - add data to the start of a buffer
835 * @skb: buffer to use
836 * @len: amount of data to add
838 * This function extends the used data area of the buffer at the buffer
839 * start. If this would exceed the total buffer headroom the kernel will
840 * panic. A pointer to the first byte of the extra data is returned.
842 static inline unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
846 if (unlikely(skb->data<skb->head))
847 skb_under_panic(skb, len, current_text_addr());
851 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
854 BUG_ON(skb->len < skb->data_len);
855 return skb->data += len;
859 * skb_pull - remove data from the start of a buffer
860 * @skb: buffer to use
861 * @len: amount of data to remove
863 * This function removes data from the start of a buffer, returning
864 * the memory to the headroom. A pointer to the next data in the buffer
865 * is returned. Once the data has been pulled future pushes will overwrite
868 static inline unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
870 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
873 extern unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
875 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
877 if (len > skb_headlen(skb) &&
878 !__pskb_pull_tail(skb, len-skb_headlen(skb)))
881 return skb->data += len;
884 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
886 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
889 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
891 if (likely(len <= skb_headlen(skb)))
893 if (unlikely(len > skb->len))
895 return __pskb_pull_tail(skb, len-skb_headlen(skb)) != NULL;
899 * skb_headroom - bytes at buffer head
900 * @skb: buffer to check
902 * Return the number of bytes of free space at the head of an &sk_buff.
904 static inline int skb_headroom(const struct sk_buff *skb)
906 return skb->data - skb->head;
910 * skb_tailroom - bytes at buffer end
911 * @skb: buffer to check
913 * Return the number of bytes of free space at the tail of an sk_buff
915 static inline int skb_tailroom(const struct sk_buff *skb)
917 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
921 * skb_reserve - adjust headroom
922 * @skb: buffer to alter
923 * @len: bytes to move
925 * Increase the headroom of an empty &sk_buff by reducing the tail
926 * room. This is only allowed for an empty buffer.
928 static inline void skb_reserve(struct sk_buff *skb, int len)
935 * CPUs often take a performance hit when accessing unaligned memory
936 * locations. The actual performance hit varies, it can be small if the
937 * hardware handles it or large if we have to take an exception and fix it
940 * Since an ethernet header is 14 bytes network drivers often end up with
941 * the IP header at an unaligned offset. The IP header can be aligned by
942 * shifting the start of the packet by 2 bytes. Drivers should do this
945 * skb_reserve(NET_IP_ALIGN);
947 * The downside to this alignment of the IP header is that the DMA is now
948 * unaligned. On some architectures the cost of an unaligned DMA is high
949 * and this cost outweighs the gains made by aligning the IP header.
951 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
955 #define NET_IP_ALIGN 2
959 * The networking layer reserves some headroom in skb data (via
960 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
961 * the header has to grow. In the default case, if the header has to grow
962 * 16 bytes or less we avoid the reallocation.
964 * Unfortunately this headroom changes the DMA alignment of the resulting
965 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
966 * on some architectures. An architecture can override this value,
967 * perhaps setting it to a cacheline in size (since that will maintain
968 * cacheline alignment of the DMA). It must be a power of 2.
970 * Various parts of the networking layer expect at least 16 bytes of
971 * headroom, you should not reduce this.
974 #define NET_SKB_PAD 16
977 extern int ___pskb_trim(struct sk_buff *skb, unsigned int len);
979 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
981 if (unlikely(skb->data_len)) {
986 skb->tail = skb->data + len;
990 * skb_trim - remove end from a buffer
991 * @skb: buffer to alter
994 * Cut the length of a buffer down by removing data from the tail. If
995 * the buffer is already under the length specified it is not modified.
996 * The skb must be linear.
998 static inline void skb_trim(struct sk_buff *skb, unsigned int len)
1001 __skb_trim(skb, len);
1005 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
1008 return ___pskb_trim(skb, len);
1009 __skb_trim(skb, len);
1013 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
1015 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
1019 * skb_orphan - orphan a buffer
1020 * @skb: buffer to orphan
1022 * If a buffer currently has an owner then we call the owner's
1023 * destructor function and make the @skb unowned. The buffer continues
1024 * to exist but is no longer charged to its former owner.
1026 static inline void skb_orphan(struct sk_buff *skb)
1028 if (skb->destructor)
1029 skb->destructor(skb);
1030 skb->destructor = NULL;
1035 * __skb_queue_purge - empty a list
1036 * @list: list to empty
1038 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1039 * the list and one reference dropped. This function does not take the
1040 * list lock and the caller must hold the relevant locks to use it.
1042 extern void skb_queue_purge(struct sk_buff_head *list);
1043 static inline void __skb_queue_purge(struct sk_buff_head *list)
1045 struct sk_buff *skb;
1046 while ((skb = __skb_dequeue(list)) != NULL)
1050 #ifndef CONFIG_HAVE_ARCH_DEV_ALLOC_SKB
1052 * __dev_alloc_skb - allocate an skbuff for sending
1053 * @length: length to allocate
1054 * @gfp_mask: get_free_pages mask, passed to alloc_skb
1056 * Allocate a new &sk_buff and assign it a usage count of one. The
1057 * buffer has unspecified headroom built in. Users should allocate
1058 * the headroom they think they need without accounting for the
1059 * built in space. The built in space is used for optimisations.
1061 * %NULL is returned in there is no free memory.
1063 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
1066 struct sk_buff *skb = alloc_skb(length + NET_SKB_PAD, gfp_mask);
1068 skb_reserve(skb, NET_SKB_PAD);
1072 extern struct sk_buff *__dev_alloc_skb(unsigned int length, int gfp_mask);
1076 * dev_alloc_skb - allocate an skbuff for sending
1077 * @length: length to allocate
1079 * Allocate a new &sk_buff and assign it a usage count of one. The
1080 * buffer has unspecified headroom built in. Users should allocate
1081 * the headroom they think they need without accounting for the
1082 * built in space. The built in space is used for optimisations.
1084 * %NULL is returned in there is no free memory. Although this function
1085 * allocates memory it can be called from an interrupt.
1087 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
1089 return __dev_alloc_skb(length, GFP_ATOMIC);
1093 * skb_cow - copy header of skb when it is required
1094 * @skb: buffer to cow
1095 * @headroom: needed headroom
1097 * If the skb passed lacks sufficient headroom or its data part
1098 * is shared, data is reallocated. If reallocation fails, an error
1099 * is returned and original skb is not changed.
1101 * The result is skb with writable area skb->head...skb->tail
1102 * and at least @headroom of space at head.
1104 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
1106 int delta = (headroom > NET_SKB_PAD ? headroom : NET_SKB_PAD) -
1112 if (delta || skb_cloned(skb))
1113 return pskb_expand_head(skb, (delta + (NET_SKB_PAD-1)) &
1114 ~(NET_SKB_PAD-1), 0, GFP_ATOMIC);
1119 * skb_padto - pad an skbuff up to a minimal size
1120 * @skb: buffer to pad
1121 * @len: minimal length
1123 * Pads up a buffer to ensure the trailing bytes exist and are
1124 * blanked. If the buffer already contains sufficient data it
1125 * is untouched. Returns the buffer, which may be a replacement
1126 * for the original, or NULL for out of memory - in which case
1127 * the original buffer is still freed.
1130 static inline struct sk_buff *skb_padto(struct sk_buff *skb, unsigned int len)
1132 unsigned int size = skb->len;
1133 if (likely(size >= len))
1135 return skb_pad(skb, len-size);
1138 static inline int skb_add_data(struct sk_buff *skb,
1139 char __user *from, int copy)
1141 const int off = skb->len;
1143 if (skb->ip_summed == CHECKSUM_NONE) {
1145 unsigned int csum = csum_and_copy_from_user(from,
1149 skb->csum = csum_block_add(skb->csum, csum, off);
1152 } else if (!copy_from_user(skb_put(skb, copy), from, copy))
1155 __skb_trim(skb, off);
1159 static inline int skb_can_coalesce(struct sk_buff *skb, int i,
1160 struct page *page, int off)
1163 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
1165 return page == frag->page &&
1166 off == frag->page_offset + frag->size;
1171 static inline int __skb_linearize(struct sk_buff *skb)
1173 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
1177 * skb_linearize - convert paged skb to linear one
1178 * @skb: buffer to linarize
1180 * If there is no free memory -ENOMEM is returned, otherwise zero
1181 * is returned and the old skb data released.
1183 static inline int skb_linearize(struct sk_buff *skb)
1185 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
1189 * skb_linearize_cow - make sure skb is linear and writable
1190 * @skb: buffer to process
1192 * If there is no free memory -ENOMEM is returned, otherwise zero
1193 * is returned and the old skb data released.
1195 static inline int skb_linearize_cow(struct sk_buff *skb)
1197 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
1198 __skb_linearize(skb) : 0;
1202 * skb_postpull_rcsum - update checksum for received skb after pull
1203 * @skb: buffer to update
1204 * @start: start of data before pull
1205 * @len: length of data pulled
1207 * After doing a pull on a received packet, you need to call this to
1208 * update the CHECKSUM_HW checksum, or set ip_summed to CHECKSUM_NONE
1209 * so that it can be recomputed from scratch.
1212 static inline void skb_postpull_rcsum(struct sk_buff *skb,
1213 const void *start, unsigned int len)
1215 if (skb->ip_summed == CHECKSUM_HW)
1216 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
1219 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
1222 * pskb_trim_rcsum - trim received skb and update checksum
1223 * @skb: buffer to trim
1226 * This is exactly the same as pskb_trim except that it ensures the
1227 * checksum of received packets are still valid after the operation.
1230 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
1232 if (likely(len >= skb->len))
1234 if (skb->ip_summed == CHECKSUM_HW)
1235 skb->ip_summed = CHECKSUM_NONE;
1236 return __pskb_trim(skb, len);
1239 static inline void *kmap_skb_frag(const skb_frag_t *frag)
1241 #ifdef CONFIG_HIGHMEM
1246 return kmap_atomic(frag->page, KM_SKB_DATA_SOFTIRQ);
1249 static inline void kunmap_skb_frag(void *vaddr)
1251 kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1252 #ifdef CONFIG_HIGHMEM
1257 #define skb_queue_walk(queue, skb) \
1258 for (skb = (queue)->next; \
1259 prefetch(skb->next), (skb != (struct sk_buff *)(queue)); \
1262 #define skb_queue_reverse_walk(queue, skb) \
1263 for (skb = (queue)->prev; \
1264 prefetch(skb->prev), (skb != (struct sk_buff *)(queue)); \
1268 extern struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
1269 int noblock, int *err);
1270 extern unsigned int datagram_poll(struct file *file, struct socket *sock,
1271 struct poll_table_struct *wait);
1272 extern int skb_copy_datagram_iovec(const struct sk_buff *from,
1273 int offset, struct iovec *to,
1275 extern int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
1278 extern void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
1279 extern void skb_kill_datagram(struct sock *sk, struct sk_buff *skb,
1280 unsigned int flags);
1281 extern unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1282 int len, unsigned int csum);
1283 extern int skb_copy_bits(const struct sk_buff *skb, int offset,
1285 extern int skb_store_bits(const struct sk_buff *skb, int offset,
1286 void *from, int len);
1287 extern unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb,
1288 int offset, u8 *to, int len,
1290 extern void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
1291 extern void skb_split(struct sk_buff *skb,
1292 struct sk_buff *skb1, const u32 len);
1294 extern void skb_release_data(struct sk_buff *skb);
1296 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
1297 int len, void *buffer)
1299 int hlen = skb_headlen(skb);
1301 if (hlen - offset >= len)
1302 return skb->data + offset;
1304 if (skb_copy_bits(skb, offset, buffer, len) < 0)
1310 extern void skb_init(void);
1311 extern void skb_add_mtu(int mtu);
1314 * skb_get_timestamp - get timestamp from a skb
1315 * @skb: skb to get stamp from
1316 * @stamp: pointer to struct timeval to store stamp in
1318 * Timestamps are stored in the skb as offsets to a base timestamp.
1319 * This function converts the offset back to a struct timeval and stores
1322 static inline void skb_get_timestamp(const struct sk_buff *skb, struct timeval *stamp)
1324 stamp->tv_sec = skb->tstamp.off_sec;
1325 stamp->tv_usec = skb->tstamp.off_usec;
1329 * skb_set_timestamp - set timestamp of a skb
1330 * @skb: skb to set stamp of
1331 * @stamp: pointer to struct timeval to get stamp from
1333 * Timestamps are stored in the skb as offsets to a base timestamp.
1334 * This function converts a struct timeval to an offset and stores
1337 static inline void skb_set_timestamp(struct sk_buff *skb, const struct timeval *stamp)
1339 skb->tstamp.off_sec = stamp->tv_sec;
1340 skb->tstamp.off_usec = stamp->tv_usec;
1343 extern void __net_timestamp(struct sk_buff *skb);
1345 extern unsigned int __skb_checksum_complete(struct sk_buff *skb);
1348 * skb_checksum_complete - Calculate checksum of an entire packet
1349 * @skb: packet to process
1351 * This function calculates the checksum over the entire packet plus
1352 * the value of skb->csum. The latter can be used to supply the
1353 * checksum of a pseudo header as used by TCP/UDP. It returns the
1356 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
1357 * this function can be used to verify that checksum on received
1358 * packets. In that case the function should return zero if the
1359 * checksum is correct. In particular, this function will return zero
1360 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
1361 * hardware has already verified the correctness of the checksum.
1363 static inline unsigned int skb_checksum_complete(struct sk_buff *skb)
1365 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
1366 __skb_checksum_complete(skb);
1369 #ifdef CONFIG_NETFILTER
1370 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
1372 if (nfct && atomic_dec_and_test(&nfct->use))
1373 nfct->destroy(nfct);
1375 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
1378 atomic_inc(&nfct->use);
1380 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1381 static inline void nf_conntrack_get_reasm(struct sk_buff *skb)
1384 atomic_inc(&skb->users);
1386 static inline void nf_conntrack_put_reasm(struct sk_buff *skb)
1392 #ifdef CONFIG_BRIDGE_NETFILTER
1393 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
1395 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
1398 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
1401 atomic_inc(&nf_bridge->use);
1403 #endif /* CONFIG_BRIDGE_NETFILTER */
1404 static inline void nf_reset(struct sk_buff *skb)
1406 nf_conntrack_put(skb->nfct);
1408 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1409 nf_conntrack_put_reasm(skb->nfct_reasm);
1410 skb->nfct_reasm = NULL;
1412 #ifdef CONFIG_BRIDGE_NETFILTER
1413 nf_bridge_put(skb->nf_bridge);
1414 skb->nf_bridge = NULL;
1418 #else /* CONFIG_NETFILTER */
1419 static inline void nf_reset(struct sk_buff *skb) {}
1420 #endif /* CONFIG_NETFILTER */
1422 #ifdef CONFIG_NETWORK_SECMARK
1423 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
1425 to->secmark = from->secmark;
1428 static inline void skb_init_secmark(struct sk_buff *skb)
1433 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
1436 static inline void skb_init_secmark(struct sk_buff *skb)
1440 #endif /* __KERNEL__ */
1441 #endif /* _LINUX_SKBUFF_H */