blob: 76c68851474c3f549cd966c932ffa000c48490e4 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Definitions for the 'struct sk_buff' memory handlers.
3 *
4 * Authors:
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
7 *
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.
12 */
13
14#ifndef _LINUX_SKBUFF_H
15#define _LINUX_SKBUFF_H
16
17#include <linux/config.h>
18#include <linux/kernel.h>
19#include <linux/compiler.h>
20#include <linux/time.h>
21#include <linux/cache.h>
22
23#include <asm/atomic.h>
24#include <asm/types.h>
25#include <linux/spinlock.h>
26#include <linux/mm.h>
27#include <linux/highmem.h>
28#include <linux/poll.h>
29#include <linux/net.h>
Thomas Graf3fc7e8a2005-06-23 21:00:17 -070030#include <linux/textsearch.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <net/checksum.h>
32
33#define HAVE_ALLOC_SKB /* For the drivers to know */
34#define HAVE_ALIGNABLE_SKB /* Ditto 8) */
35#define SLAB_SKB /* Slabified skbuffs */
36
37#define CHECKSUM_NONE 0
38#define CHECKSUM_HW 1
39#define CHECKSUM_UNNECESSARY 2
40
41#define SKB_DATA_ALIGN(X) (((X) + (SMP_CACHE_BYTES - 1)) & \
42 ~(SMP_CACHE_BYTES - 1))
43#define SKB_MAX_ORDER(X, ORDER) (((PAGE_SIZE << (ORDER)) - (X) - \
44 sizeof(struct skb_shared_info)) & \
45 ~(SMP_CACHE_BYTES - 1))
46#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
47#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
48
49/* A. Checksumming of received packets by device.
50 *
51 * NONE: device failed to checksum this packet.
52 * skb->csum is undefined.
53 *
54 * UNNECESSARY: device parsed packet and wouldbe verified checksum.
55 * skb->csum is undefined.
56 * It is bad option, but, unfortunately, many of vendors do this.
57 * Apparently with secret goal to sell you new device, when you
58 * will add new protocol to your host. F.e. IPv6. 8)
59 *
60 * HW: the most generic way. Device supplied checksum of _all_
61 * the packet as seen by netif_rx in skb->csum.
62 * NOTE: Even if device supports only some protocols, but
63 * is able to produce some skb->csum, it MUST use HW,
64 * not UNNECESSARY.
65 *
66 * B. Checksumming on output.
67 *
68 * NONE: skb is checksummed by protocol or csum is not required.
69 *
70 * HW: device is required to csum packet as seen by hard_start_xmit
71 * from skb->h.raw to the end and to record the checksum
72 * at skb->h.raw+skb->csum.
73 *
74 * Device must show its capabilities in dev->features, set
75 * at device setup time.
76 * NETIF_F_HW_CSUM - it is clever device, it is able to checksum
77 * everything.
78 * NETIF_F_NO_CSUM - loopback or reliable single hop media.
79 * NETIF_F_IP_CSUM - device is dumb. It is able to csum only
80 * TCP/UDP over IPv4. Sigh. Vendors like this
81 * way by an unknown reason. Though, see comment above
82 * about CHECKSUM_UNNECESSARY. 8)
83 *
84 * Any questions? No questions, good. --ANK
85 */
86
Linus Torvalds1da177e2005-04-16 15:20:36 -070087struct net_device;
88
89#ifdef CONFIG_NETFILTER
90struct nf_conntrack {
91 atomic_t use;
92 void (*destroy)(struct nf_conntrack *);
93};
94
95#ifdef CONFIG_BRIDGE_NETFILTER
96struct nf_bridge_info {
97 atomic_t use;
98 struct net_device *physindev;
99 struct net_device *physoutdev;
100#if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
101 struct net_device *netoutdev;
102#endif
103 unsigned int mask;
104 unsigned long data[32 / sizeof(unsigned long)];
105};
106#endif
107
108#endif
109
110struct sk_buff_head {
111 /* These two members must be first. */
112 struct sk_buff *next;
113 struct sk_buff *prev;
114
115 __u32 qlen;
116 spinlock_t lock;
117};
118
119struct sk_buff;
120
121/* To allow 64K frame to be packed as single skb without frag_list */
122#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 2)
123
124typedef struct skb_frag_struct skb_frag_t;
125
126struct skb_frag_struct {
127 struct page *page;
128 __u16 page_offset;
129 __u16 size;
130};
131
132/* This data is invariant across clones and lives at
133 * the end of the header data, ie. at skb->end.
134 */
135struct skb_shared_info {
136 atomic_t dataref;
137 unsigned int nr_frags;
138 unsigned short tso_size;
139 unsigned short tso_segs;
140 struct sk_buff *frag_list;
141 skb_frag_t frags[MAX_SKB_FRAGS];
142};
143
144/* We divide dataref into two halves. The higher 16 bits hold references
145 * to the payload part of skb->data. The lower 16 bits hold references to
146 * the entire skb->data. It is up to the users of the skb to agree on
147 * where the payload starts.
148 *
149 * All users must obey the rule that the skb->data reference count must be
150 * greater than or equal to the payload reference count.
151 *
152 * Holding a reference to the payload part means that the user does not
153 * care about modifications to the header part of skb->data.
154 */
155#define SKB_DATAREF_SHIFT 16
156#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
157
158/**
159 * struct sk_buff - socket buffer
160 * @next: Next buffer in list
161 * @prev: Previous buffer in list
162 * @list: List we are on
163 * @sk: Socket we are owned by
164 * @stamp: Time we arrived
165 * @dev: Device we arrived on/are leaving by
166 * @input_dev: Device we arrived on
167 * @real_dev: The real device we are using
168 * @h: Transport layer header
169 * @nh: Network layer header
170 * @mac: Link layer header
Martin Waitz67be2dd2005-05-01 08:59:26 -0700171 * @dst: destination entry
172 * @sp: the security path, used for xfrm
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 * @cb: Control buffer. Free for use by every layer. Put private vars here
174 * @len: Length of actual data
175 * @data_len: Data length
176 * @mac_len: Length of link layer header
177 * @csum: Checksum
Martin Waitz67be2dd2005-05-01 08:59:26 -0700178 * @local_df: allow local fragmentation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 * @cloned: Head may be cloned (check refcnt to be sure)
180 * @nohdr: Payload reference only, must not modify header
181 * @pkt_type: Packet class
182 * @ip_summed: Driver fed us an IP checksum
183 * @priority: Packet queueing priority
184 * @users: User count - see {datagram,tcp}.c
185 * @protocol: Packet protocol from driver
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 * @truesize: Buffer size
187 * @head: Head of buffer
188 * @data: Data head pointer
189 * @tail: Tail pointer
190 * @end: End pointer
191 * @destructor: Destruct function
192 * @nfmark: Can be used for communication between hooks
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 * @nfct: Associated connection, if any
194 * @nfctinfo: Relationship of this skb to the connection
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
196 * @private: Data which is private to the HIPPI implementation
197 * @tc_index: Traffic control index
198 * @tc_verd: traffic control verdict
199 * @tc_classid: traffic control classid
200 */
201
202struct sk_buff {
203 /* These two members must be first. */
204 struct sk_buff *next;
205 struct sk_buff *prev;
206
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 struct sock *sk;
208 struct timeval stamp;
209 struct net_device *dev;
210 struct net_device *input_dev;
211 struct net_device *real_dev;
212
213 union {
214 struct tcphdr *th;
215 struct udphdr *uh;
216 struct icmphdr *icmph;
217 struct igmphdr *igmph;
218 struct iphdr *ipiph;
219 struct ipv6hdr *ipv6h;
220 unsigned char *raw;
221 } h;
222
223 union {
224 struct iphdr *iph;
225 struct ipv6hdr *ipv6h;
226 struct arphdr *arph;
227 unsigned char *raw;
228 } nh;
229
230 union {
231 unsigned char *raw;
232 } mac;
233
234 struct dst_entry *dst;
235 struct sec_path *sp;
236
237 /*
238 * This is the control buffer. It is free to use for every
239 * layer. Please put your private variables there. If you
240 * want to keep them across layers you have to do a skb_clone()
241 * first. This is owned by whoever has the skb queued ATM.
242 */
243 char cb[40];
244
245 unsigned int len,
246 data_len,
247 mac_len,
248 csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 __u32 priority;
Thomas Graf1cbb3382005-07-05 14:13:41 -0700250 __u8 local_df:1,
251 cloned:1,
252 ip_summed:2,
Harald Welte6869c4d2005-08-09 19:24:19 -0700253 nohdr:1,
254 nfctinfo:3;
Thomas Graf1cbb3382005-07-05 14:13:41 -0700255 __u8 pkt_type;
Alexey Dobriyana0d3bea2005-08-11 16:05:50 -0700256 __be16 protocol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
258 void (*destructor)(struct sk_buff *skb);
259#ifdef CONFIG_NETFILTER
Harald Weltebf3a46a2005-08-09 19:22:01 -0700260 __u32 nfmark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 struct nf_conntrack *nfct;
Harald Welte6869c4d2005-08-09 19:24:19 -0700262#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
263 __u8 ipvs_property:1;
264#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265#ifdef CONFIG_BRIDGE_NETFILTER
266 struct nf_bridge_info *nf_bridge;
267#endif
268#endif /* CONFIG_NETFILTER */
269#if defined(CONFIG_HIPPI)
270 union {
271 __u32 ifield;
272 } private;
273#endif
274#ifdef CONFIG_NET_SCHED
275 __u32 tc_index; /* traffic control index */
276#ifdef CONFIG_NET_CLS_ACT
277 __u32 tc_verd; /* traffic control verdict */
278 __u32 tc_classid; /* traffic control classid */
279#endif
280
281#endif
282
283
284 /* These elements must be at the end, see alloc_skb() for details. */
285 unsigned int truesize;
286 atomic_t users;
287 unsigned char *head,
288 *data,
289 *tail,
290 *end;
291};
292
293#ifdef __KERNEL__
294/*
295 * Handling routines are only of interest to the kernel
296 */
297#include <linux/slab.h>
298
299#include <asm/system.h>
300
301extern void __kfree_skb(struct sk_buff *skb);
Victor Fusco86a76ca2005-07-08 14:57:47 -0700302extern struct sk_buff *alloc_skb(unsigned int size,
303 unsigned int __nocast priority);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304extern struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp,
Victor Fusco86a76ca2005-07-08 14:57:47 -0700305 unsigned int size,
306 unsigned int __nocast priority);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307extern void kfree_skbmem(struct sk_buff *skb);
Victor Fusco86a76ca2005-07-08 14:57:47 -0700308extern struct sk_buff *skb_clone(struct sk_buff *skb,
309 unsigned int __nocast priority);
310extern struct sk_buff *skb_copy(const struct sk_buff *skb,
311 unsigned int __nocast priority);
312extern struct sk_buff *pskb_copy(struct sk_buff *skb,
313 unsigned int __nocast gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314extern int pskb_expand_head(struct sk_buff *skb,
Victor Fusco86a76ca2005-07-08 14:57:47 -0700315 int nhead, int ntail,
316 unsigned int __nocast gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317extern struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
318 unsigned int headroom);
319extern struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
320 int newheadroom, int newtailroom,
Victor Fusco86a76ca2005-07-08 14:57:47 -0700321 unsigned int __nocast priority);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322extern struct sk_buff * skb_pad(struct sk_buff *skb, int pad);
323#define dev_kfree_skb(a) kfree_skb(a)
324extern void skb_over_panic(struct sk_buff *skb, int len,
325 void *here);
326extern void skb_under_panic(struct sk_buff *skb, int len,
327 void *here);
328
Thomas Graf677e90e2005-06-23 20:59:51 -0700329struct skb_seq_state
330{
331 __u32 lower_offset;
332 __u32 upper_offset;
333 __u32 frag_idx;
334 __u32 stepped_offset;
335 struct sk_buff *root_skb;
336 struct sk_buff *cur_skb;
337 __u8 *frag_data;
338};
339
340extern void skb_prepare_seq_read(struct sk_buff *skb,
341 unsigned int from, unsigned int to,
342 struct skb_seq_state *st);
343extern unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
344 struct skb_seq_state *st);
345extern void skb_abort_seq_read(struct skb_seq_state *st);
346
Thomas Graf3fc7e8a2005-06-23 21:00:17 -0700347extern unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
348 unsigned int to, struct ts_config *config,
349 struct ts_state *state);
350
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351/* Internal */
352#define skb_shinfo(SKB) ((struct skb_shared_info *)((SKB)->end))
353
354/**
355 * skb_queue_empty - check if a queue is empty
356 * @list: queue head
357 *
358 * Returns true if the queue is empty, false otherwise.
359 */
360static inline int skb_queue_empty(const struct sk_buff_head *list)
361{
362 return list->next == (struct sk_buff *)list;
363}
364
365/**
366 * skb_get - reference buffer
367 * @skb: buffer to reference
368 *
369 * Makes another reference to a socket buffer and returns a pointer
370 * to the buffer.
371 */
372static inline struct sk_buff *skb_get(struct sk_buff *skb)
373{
374 atomic_inc(&skb->users);
375 return skb;
376}
377
378/*
379 * If users == 1, we are the only owner and are can avoid redundant
380 * atomic change.
381 */
382
383/**
384 * kfree_skb - free an sk_buff
385 * @skb: buffer to free
386 *
387 * Drop a reference to the buffer and free it if the usage count has
388 * hit zero.
389 */
390static inline void kfree_skb(struct sk_buff *skb)
391{
392 if (likely(atomic_read(&skb->users) == 1))
393 smp_rmb();
394 else if (likely(!atomic_dec_and_test(&skb->users)))
395 return;
396 __kfree_skb(skb);
397}
398
399/**
400 * skb_cloned - is the buffer a clone
401 * @skb: buffer to check
402 *
403 * Returns true if the buffer was generated with skb_clone() and is
404 * one of multiple shared copies of the buffer. Cloned buffers are
405 * shared data so must not be written to under normal circumstances.
406 */
407static inline int skb_cloned(const struct sk_buff *skb)
408{
409 return skb->cloned &&
410 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
411}
412
413/**
414 * skb_header_cloned - is the header a clone
415 * @skb: buffer to check
416 *
417 * Returns true if modifying the header part of the buffer requires
418 * the data to be copied.
419 */
420static inline int skb_header_cloned(const struct sk_buff *skb)
421{
422 int dataref;
423
424 if (!skb->cloned)
425 return 0;
426
427 dataref = atomic_read(&skb_shinfo(skb)->dataref);
428 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
429 return dataref != 1;
430}
431
432/**
433 * skb_header_release - release reference to header
434 * @skb: buffer to operate on
435 *
436 * Drop a reference to the header part of the buffer. This is done
437 * by acquiring a payload reference. You must not read from the header
438 * part of skb->data after this.
439 */
440static inline void skb_header_release(struct sk_buff *skb)
441{
442 BUG_ON(skb->nohdr);
443 skb->nohdr = 1;
444 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
445}
446
447/**
448 * skb_shared - is the buffer shared
449 * @skb: buffer to check
450 *
451 * Returns true if more than one person has a reference to this
452 * buffer.
453 */
454static inline int skb_shared(const struct sk_buff *skb)
455{
456 return atomic_read(&skb->users) != 1;
457}
458
459/**
460 * skb_share_check - check if buffer is shared and if so clone it
461 * @skb: buffer to check
462 * @pri: priority for memory allocation
463 *
464 * If the buffer is shared the buffer is cloned and the old copy
465 * drops a reference. A new clone with a single reference is returned.
466 * If the buffer is not shared the original buffer is returned. When
467 * being called from interrupt status or with spinlocks held pri must
468 * be GFP_ATOMIC.
469 *
470 * NULL is returned on a memory allocation failure.
471 */
Victor Fusco86a76ca2005-07-08 14:57:47 -0700472static inline struct sk_buff *skb_share_check(struct sk_buff *skb,
473 unsigned int __nocast pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474{
475 might_sleep_if(pri & __GFP_WAIT);
476 if (skb_shared(skb)) {
477 struct sk_buff *nskb = skb_clone(skb, pri);
478 kfree_skb(skb);
479 skb = nskb;
480 }
481 return skb;
482}
483
484/*
485 * Copy shared buffers into a new sk_buff. We effectively do COW on
486 * packets to handle cases where we have a local reader and forward
487 * and a couple of other messy ones. The normal one is tcpdumping
488 * a packet thats being forwarded.
489 */
490
491/**
492 * skb_unshare - make a copy of a shared buffer
493 * @skb: buffer to check
494 * @pri: priority for memory allocation
495 *
496 * If the socket buffer is a clone then this function creates a new
497 * copy of the data, drops a reference count on the old copy and returns
498 * the new copy with the reference count at 1. If the buffer is not a clone
499 * the original buffer is returned. When called with a spinlock held or
500 * from interrupt state @pri must be %GFP_ATOMIC
501 *
502 * %NULL is returned on a memory allocation failure.
503 */
Victor Fuscoe2bf5212005-07-18 13:36:38 -0700504static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
505 unsigned int __nocast pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506{
507 might_sleep_if(pri & __GFP_WAIT);
508 if (skb_cloned(skb)) {
509 struct sk_buff *nskb = skb_copy(skb, pri);
510 kfree_skb(skb); /* Free our shared copy */
511 skb = nskb;
512 }
513 return skb;
514}
515
516/**
517 * skb_peek
518 * @list_: list to peek at
519 *
520 * Peek an &sk_buff. Unlike most other operations you _MUST_
521 * be careful with this one. A peek leaves the buffer on the
522 * list and someone else may run off with it. You must hold
523 * the appropriate locks or have a private queue to do this.
524 *
525 * Returns %NULL for an empty list or a pointer to the head element.
526 * The reference count is not incremented and the reference is therefore
527 * volatile. Use with caution.
528 */
529static inline struct sk_buff *skb_peek(struct sk_buff_head *list_)
530{
531 struct sk_buff *list = ((struct sk_buff *)list_)->next;
532 if (list == (struct sk_buff *)list_)
533 list = NULL;
534 return list;
535}
536
537/**
538 * skb_peek_tail
539 * @list_: list to peek at
540 *
541 * Peek an &sk_buff. Unlike most other operations you _MUST_
542 * be careful with this one. A peek leaves the buffer on the
543 * list and someone else may run off with it. You must hold
544 * the appropriate locks or have a private queue to do this.
545 *
546 * Returns %NULL for an empty list or a pointer to the tail element.
547 * The reference count is not incremented and the reference is therefore
548 * volatile. Use with caution.
549 */
550static inline struct sk_buff *skb_peek_tail(struct sk_buff_head *list_)
551{
552 struct sk_buff *list = ((struct sk_buff *)list_)->prev;
553 if (list == (struct sk_buff *)list_)
554 list = NULL;
555 return list;
556}
557
558/**
559 * skb_queue_len - get queue length
560 * @list_: list to measure
561 *
562 * Return the length of an &sk_buff queue.
563 */
564static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
565{
566 return list_->qlen;
567}
568
569static inline void skb_queue_head_init(struct sk_buff_head *list)
570{
571 spin_lock_init(&list->lock);
572 list->prev = list->next = (struct sk_buff *)list;
573 list->qlen = 0;
574}
575
576/*
577 * Insert an sk_buff at the start of a list.
578 *
579 * The "__skb_xxxx()" functions are the non-atomic ones that
580 * can only be called with interrupts disabled.
581 */
582
583/**
584 * __skb_queue_head - queue a buffer at the list head
585 * @list: list to use
586 * @newsk: buffer to queue
587 *
588 * Queue a buffer at the start of a list. This function takes no locks
589 * and you must therefore hold required locks before calling it.
590 *
591 * A buffer cannot be placed on two lists at the same time.
592 */
593extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
594static inline void __skb_queue_head(struct sk_buff_head *list,
595 struct sk_buff *newsk)
596{
597 struct sk_buff *prev, *next;
598
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 list->qlen++;
600 prev = (struct sk_buff *)list;
601 next = prev->next;
602 newsk->next = next;
603 newsk->prev = prev;
604 next->prev = prev->next = newsk;
605}
606
607/**
608 * __skb_queue_tail - queue a buffer at the list tail
609 * @list: list to use
610 * @newsk: buffer to queue
611 *
612 * Queue a buffer at the end of a list. This function takes no locks
613 * and you must therefore hold required locks before calling it.
614 *
615 * A buffer cannot be placed on two lists at the same time.
616 */
617extern void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
618static inline void __skb_queue_tail(struct sk_buff_head *list,
619 struct sk_buff *newsk)
620{
621 struct sk_buff *prev, *next;
622
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 list->qlen++;
624 next = (struct sk_buff *)list;
625 prev = next->prev;
626 newsk->next = next;
627 newsk->prev = prev;
628 next->prev = prev->next = newsk;
629}
630
631
632/**
633 * __skb_dequeue - remove from the head of the queue
634 * @list: list to dequeue from
635 *
636 * Remove the head of the list. This function does not take any locks
637 * so must be used with appropriate locks held only. The head item is
638 * returned or %NULL if the list is empty.
639 */
640extern struct sk_buff *skb_dequeue(struct sk_buff_head *list);
641static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
642{
643 struct sk_buff *next, *prev, *result;
644
645 prev = (struct sk_buff *) list;
646 next = prev->next;
647 result = NULL;
648 if (next != prev) {
649 result = next;
650 next = next->next;
651 list->qlen--;
652 next->prev = prev;
653 prev->next = next;
654 result->next = result->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 }
656 return result;
657}
658
659
660/*
661 * Insert a packet on a list.
662 */
David S. Miller8728b832005-08-09 19:25:21 -0700663extern void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664static inline void __skb_insert(struct sk_buff *newsk,
665 struct sk_buff *prev, struct sk_buff *next,
666 struct sk_buff_head *list)
667{
668 newsk->next = next;
669 newsk->prev = prev;
670 next->prev = prev->next = newsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 list->qlen++;
672}
673
674/*
675 * Place a packet after a given packet in a list.
676 */
David S. Miller8728b832005-08-09 19:25:21 -0700677extern void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list);
678static inline void __skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
David S. Miller8728b832005-08-09 19:25:21 -0700680 __skb_insert(newsk, old, old->next, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681}
682
683/*
684 * remove sk_buff from list. _Must_ be called atomically, and with
685 * the list known..
686 */
David S. Miller8728b832005-08-09 19:25:21 -0700687extern void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
689{
690 struct sk_buff *next, *prev;
691
692 list->qlen--;
693 next = skb->next;
694 prev = skb->prev;
695 skb->next = skb->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 next->prev = prev;
697 prev->next = next;
698}
699
700
701/* XXX: more streamlined implementation */
702
703/**
704 * __skb_dequeue_tail - remove from the tail of the queue
705 * @list: list to dequeue from
706 *
707 * Remove the tail of the list. This function does not take any locks
708 * so must be used with appropriate locks held only. The tail item is
709 * returned or %NULL if the list is empty.
710 */
711extern struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
712static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
713{
714 struct sk_buff *skb = skb_peek_tail(list);
715 if (skb)
716 __skb_unlink(skb, list);
717 return skb;
718}
719
720
721static inline int skb_is_nonlinear(const struct sk_buff *skb)
722{
723 return skb->data_len;
724}
725
726static inline unsigned int skb_headlen(const struct sk_buff *skb)
727{
728 return skb->len - skb->data_len;
729}
730
731static inline int skb_pagelen(const struct sk_buff *skb)
732{
733 int i, len = 0;
734
735 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
736 len += skb_shinfo(skb)->frags[i].size;
737 return len + skb_headlen(skb);
738}
739
740static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
741 struct page *page, int off, int size)
742{
743 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
744
745 frag->page = page;
746 frag->page_offset = off;
747 frag->size = size;
748 skb_shinfo(skb)->nr_frags = i + 1;
749}
750
751#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
752#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_shinfo(skb)->frag_list)
753#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
754
755/*
756 * Add data to an sk_buff
757 */
758static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
759{
760 unsigned char *tmp = skb->tail;
761 SKB_LINEAR_ASSERT(skb);
762 skb->tail += len;
763 skb->len += len;
764 return tmp;
765}
766
767/**
768 * skb_put - add data to a buffer
769 * @skb: buffer to use
770 * @len: amount of data to add
771 *
772 * This function extends the used data area of the buffer. If this would
773 * exceed the total buffer size the kernel will panic. A pointer to the
774 * first byte of the extra data is returned.
775 */
776static inline unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
777{
778 unsigned char *tmp = skb->tail;
779 SKB_LINEAR_ASSERT(skb);
780 skb->tail += len;
781 skb->len += len;
782 if (unlikely(skb->tail>skb->end))
783 skb_over_panic(skb, len, current_text_addr());
784 return tmp;
785}
786
787static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
788{
789 skb->data -= len;
790 skb->len += len;
791 return skb->data;
792}
793
794/**
795 * skb_push - add data to the start of a buffer
796 * @skb: buffer to use
797 * @len: amount of data to add
798 *
799 * This function extends the used data area of the buffer at the buffer
800 * start. If this would exceed the total buffer headroom the kernel will
801 * panic. A pointer to the first byte of the extra data is returned.
802 */
803static inline unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
804{
805 skb->data -= len;
806 skb->len += len;
807 if (unlikely(skb->data<skb->head))
808 skb_under_panic(skb, len, current_text_addr());
809 return skb->data;
810}
811
812static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
813{
814 skb->len -= len;
815 BUG_ON(skb->len < skb->data_len);
816 return skb->data += len;
817}
818
819/**
820 * skb_pull - remove data from the start of a buffer
821 * @skb: buffer to use
822 * @len: amount of data to remove
823 *
824 * This function removes data from the start of a buffer, returning
825 * the memory to the headroom. A pointer to the next data in the buffer
826 * is returned. Once the data has been pulled future pushes will overwrite
827 * the old data.
828 */
829static inline unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
830{
831 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
832}
833
834extern unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
835
836static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
837{
838 if (len > skb_headlen(skb) &&
839 !__pskb_pull_tail(skb, len-skb_headlen(skb)))
840 return NULL;
841 skb->len -= len;
842 return skb->data += len;
843}
844
845static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
846{
847 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
848}
849
850static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
851{
852 if (likely(len <= skb_headlen(skb)))
853 return 1;
854 if (unlikely(len > skb->len))
855 return 0;
856 return __pskb_pull_tail(skb, len-skb_headlen(skb)) != NULL;
857}
858
859/**
860 * skb_headroom - bytes at buffer head
861 * @skb: buffer to check
862 *
863 * Return the number of bytes of free space at the head of an &sk_buff.
864 */
865static inline int skb_headroom(const struct sk_buff *skb)
866{
867 return skb->data - skb->head;
868}
869
870/**
871 * skb_tailroom - bytes at buffer end
872 * @skb: buffer to check
873 *
874 * Return the number of bytes of free space at the tail of an sk_buff
875 */
876static inline int skb_tailroom(const struct sk_buff *skb)
877{
878 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
879}
880
881/**
882 * skb_reserve - adjust headroom
883 * @skb: buffer to alter
884 * @len: bytes to move
885 *
886 * Increase the headroom of an empty &sk_buff by reducing the tail
887 * room. This is only allowed for an empty buffer.
888 */
889static inline void skb_reserve(struct sk_buff *skb, unsigned int len)
890{
891 skb->data += len;
892 skb->tail += len;
893}
894
895/*
896 * CPUs often take a performance hit when accessing unaligned memory
897 * locations. The actual performance hit varies, it can be small if the
898 * hardware handles it or large if we have to take an exception and fix it
899 * in software.
900 *
901 * Since an ethernet header is 14 bytes network drivers often end up with
902 * the IP header at an unaligned offset. The IP header can be aligned by
903 * shifting the start of the packet by 2 bytes. Drivers should do this
904 * with:
905 *
906 * skb_reserve(NET_IP_ALIGN);
907 *
908 * The downside to this alignment of the IP header is that the DMA is now
909 * unaligned. On some architectures the cost of an unaligned DMA is high
910 * and this cost outweighs the gains made by aligning the IP header.
911 *
912 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
913 * to be overridden.
914 */
915#ifndef NET_IP_ALIGN
916#define NET_IP_ALIGN 2
917#endif
918
919extern int ___pskb_trim(struct sk_buff *skb, unsigned int len, int realloc);
920
921static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
922{
923 if (!skb->data_len) {
924 skb->len = len;
925 skb->tail = skb->data + len;
926 } else
927 ___pskb_trim(skb, len, 0);
928}
929
930/**
931 * skb_trim - remove end from a buffer
932 * @skb: buffer to alter
933 * @len: new length
934 *
935 * Cut the length of a buffer down by removing data from the tail. If
936 * the buffer is already under the length specified it is not modified.
937 */
938static inline void skb_trim(struct sk_buff *skb, unsigned int len)
939{
940 if (skb->len > len)
941 __skb_trim(skb, len);
942}
943
944
945static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
946{
947 if (!skb->data_len) {
948 skb->len = len;
949 skb->tail = skb->data+len;
950 return 0;
951 }
952 return ___pskb_trim(skb, len, 1);
953}
954
955static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
956{
957 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
958}
959
960/**
961 * skb_orphan - orphan a buffer
962 * @skb: buffer to orphan
963 *
964 * If a buffer currently has an owner then we call the owner's
965 * destructor function and make the @skb unowned. The buffer continues
966 * to exist but is no longer charged to its former owner.
967 */
968static inline void skb_orphan(struct sk_buff *skb)
969{
970 if (skb->destructor)
971 skb->destructor(skb);
972 skb->destructor = NULL;
973 skb->sk = NULL;
974}
975
976/**
977 * __skb_queue_purge - empty a list
978 * @list: list to empty
979 *
980 * Delete all buffers on an &sk_buff list. Each buffer is removed from
981 * the list and one reference dropped. This function does not take the
982 * list lock and the caller must hold the relevant locks to use it.
983 */
984extern void skb_queue_purge(struct sk_buff_head *list);
985static inline void __skb_queue_purge(struct sk_buff_head *list)
986{
987 struct sk_buff *skb;
988 while ((skb = __skb_dequeue(list)) != NULL)
989 kfree_skb(skb);
990}
991
Pavel Pisa4dc3b162005-05-01 08:59:25 -0700992#ifndef CONFIG_HAVE_ARCH_DEV_ALLOC_SKB
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993/**
994 * __dev_alloc_skb - allocate an skbuff for sending
995 * @length: length to allocate
996 * @gfp_mask: get_free_pages mask, passed to alloc_skb
997 *
998 * Allocate a new &sk_buff and assign it a usage count of one. The
999 * buffer has unspecified headroom built in. Users should allocate
1000 * the headroom they think they need without accounting for the
1001 * built in space. The built in space is used for optimisations.
1002 *
1003 * %NULL is returned in there is no free memory.
1004 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
Victor Fusco86a76ca2005-07-08 14:57:47 -07001006 unsigned int __nocast gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007{
1008 struct sk_buff *skb = alloc_skb(length + 16, gfp_mask);
1009 if (likely(skb))
1010 skb_reserve(skb, 16);
1011 return skb;
1012}
1013#else
1014extern struct sk_buff *__dev_alloc_skb(unsigned int length, int gfp_mask);
1015#endif
1016
1017/**
1018 * dev_alloc_skb - allocate an skbuff for sending
1019 * @length: length to allocate
1020 *
1021 * Allocate a new &sk_buff and assign it a usage count of one. The
1022 * buffer has unspecified headroom built in. Users should allocate
1023 * the headroom they think they need without accounting for the
1024 * built in space. The built in space is used for optimisations.
1025 *
1026 * %NULL is returned in there is no free memory. Although this function
1027 * allocates memory it can be called from an interrupt.
1028 */
1029static inline struct sk_buff *dev_alloc_skb(unsigned int length)
1030{
1031 return __dev_alloc_skb(length, GFP_ATOMIC);
1032}
1033
1034/**
1035 * skb_cow - copy header of skb when it is required
1036 * @skb: buffer to cow
1037 * @headroom: needed headroom
1038 *
1039 * If the skb passed lacks sufficient headroom or its data part
1040 * is shared, data is reallocated. If reallocation fails, an error
1041 * is returned and original skb is not changed.
1042 *
1043 * The result is skb with writable area skb->head...skb->tail
1044 * and at least @headroom of space at head.
1045 */
1046static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
1047{
1048 int delta = (headroom > 16 ? headroom : 16) - skb_headroom(skb);
1049
1050 if (delta < 0)
1051 delta = 0;
1052
1053 if (delta || skb_cloned(skb))
1054 return pskb_expand_head(skb, (delta + 15) & ~15, 0, GFP_ATOMIC);
1055 return 0;
1056}
1057
1058/**
1059 * skb_padto - pad an skbuff up to a minimal size
1060 * @skb: buffer to pad
1061 * @len: minimal length
1062 *
1063 * Pads up a buffer to ensure the trailing bytes exist and are
1064 * blanked. If the buffer already contains sufficient data it
1065 * is untouched. Returns the buffer, which may be a replacement
1066 * for the original, or NULL for out of memory - in which case
1067 * the original buffer is still freed.
1068 */
1069
1070static inline struct sk_buff *skb_padto(struct sk_buff *skb, unsigned int len)
1071{
1072 unsigned int size = skb->len;
1073 if (likely(size >= len))
1074 return skb;
1075 return skb_pad(skb, len-size);
1076}
1077
1078static inline int skb_add_data(struct sk_buff *skb,
1079 char __user *from, int copy)
1080{
1081 const int off = skb->len;
1082
1083 if (skb->ip_summed == CHECKSUM_NONE) {
1084 int err = 0;
1085 unsigned int csum = csum_and_copy_from_user(from,
1086 skb_put(skb, copy),
1087 copy, 0, &err);
1088 if (!err) {
1089 skb->csum = csum_block_add(skb->csum, csum, off);
1090 return 0;
1091 }
1092 } else if (!copy_from_user(skb_put(skb, copy), from, copy))
1093 return 0;
1094
1095 __skb_trim(skb, off);
1096 return -EFAULT;
1097}
1098
1099static inline int skb_can_coalesce(struct sk_buff *skb, int i,
1100 struct page *page, int off)
1101{
1102 if (i) {
1103 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
1104
1105 return page == frag->page &&
1106 off == frag->page_offset + frag->size;
1107 }
1108 return 0;
1109}
1110
1111/**
1112 * skb_linearize - convert paged skb to linear one
1113 * @skb: buffer to linarize
1114 * @gfp: allocation mode
1115 *
1116 * If there is no free memory -ENOMEM is returned, otherwise zero
1117 * is returned and the old skb data released.
1118 */
Victor Fusco86a76ca2005-07-08 14:57:47 -07001119extern int __skb_linearize(struct sk_buff *skb, unsigned int __nocast gfp);
1120static inline int skb_linearize(struct sk_buff *skb, unsigned int __nocast gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121{
1122 return __skb_linearize(skb, gfp);
1123}
1124
1125/**
1126 * skb_postpull_rcsum - update checksum for received skb after pull
1127 * @skb: buffer to update
1128 * @start: start of data before pull
1129 * @len: length of data pulled
1130 *
1131 * After doing a pull on a received packet, you need to call this to
1132 * update the CHECKSUM_HW checksum, or set ip_summed to CHECKSUM_NONE
1133 * so that it can be recomputed from scratch.
1134 */
1135
1136static inline void skb_postpull_rcsum(struct sk_buff *skb,
1137 const void *start, int len)
1138{
1139 if (skb->ip_summed == CHECKSUM_HW)
1140 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
1141}
1142
1143/**
1144 * pskb_trim_rcsum - trim received skb and update checksum
1145 * @skb: buffer to trim
1146 * @len: new length
1147 *
1148 * This is exactly the same as pskb_trim except that it ensures the
1149 * checksum of received packets are still valid after the operation.
1150 */
1151
1152static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
1153{
1154 if (len >= skb->len)
1155 return 0;
1156 if (skb->ip_summed == CHECKSUM_HW)
1157 skb->ip_summed = CHECKSUM_NONE;
1158 return __pskb_trim(skb, len);
1159}
1160
1161static inline void *kmap_skb_frag(const skb_frag_t *frag)
1162{
1163#ifdef CONFIG_HIGHMEM
1164 BUG_ON(in_irq());
1165
1166 local_bh_disable();
1167#endif
1168 return kmap_atomic(frag->page, KM_SKB_DATA_SOFTIRQ);
1169}
1170
1171static inline void kunmap_skb_frag(void *vaddr)
1172{
1173 kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1174#ifdef CONFIG_HIGHMEM
1175 local_bh_enable();
1176#endif
1177}
1178
1179#define skb_queue_walk(queue, skb) \
1180 for (skb = (queue)->next; \
1181 prefetch(skb->next), (skb != (struct sk_buff *)(queue)); \
1182 skb = skb->next)
1183
1184
1185extern struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
1186 int noblock, int *err);
1187extern unsigned int datagram_poll(struct file *file, struct socket *sock,
1188 struct poll_table_struct *wait);
1189extern int skb_copy_datagram_iovec(const struct sk_buff *from,
1190 int offset, struct iovec *to,
1191 int size);
1192extern int skb_copy_and_csum_datagram_iovec(const
1193 struct sk_buff *skb,
1194 int hlen,
1195 struct iovec *iov);
1196extern void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
1197extern unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1198 int len, unsigned int csum);
1199extern int skb_copy_bits(const struct sk_buff *skb, int offset,
1200 void *to, int len);
Herbert Xu357b40a2005-04-19 22:30:14 -07001201extern int skb_store_bits(const struct sk_buff *skb, int offset,
1202 void *from, int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203extern unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb,
1204 int offset, u8 *to, int len,
1205 unsigned int csum);
1206extern void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
1207extern void skb_split(struct sk_buff *skb,
1208 struct sk_buff *skb1, const u32 len);
1209
1210static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
1211 int len, void *buffer)
1212{
1213 int hlen = skb_headlen(skb);
1214
Patrick McHardy55820ee2005-07-05 14:08:10 -07001215 if (hlen - offset >= len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 return skb->data + offset;
1217
1218 if (skb_copy_bits(skb, offset, buffer, len) < 0)
1219 return NULL;
1220
1221 return buffer;
1222}
1223
1224extern void skb_init(void);
1225extern void skb_add_mtu(int mtu);
1226
1227#ifdef CONFIG_NETFILTER
1228static inline void nf_conntrack_put(struct nf_conntrack *nfct)
1229{
1230 if (nfct && atomic_dec_and_test(&nfct->use))
1231 nfct->destroy(nfct);
1232}
1233static inline void nf_conntrack_get(struct nf_conntrack *nfct)
1234{
1235 if (nfct)
1236 atomic_inc(&nfct->use);
1237}
1238static inline void nf_reset(struct sk_buff *skb)
1239{
1240 nf_conntrack_put(skb->nfct);
1241 skb->nfct = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242}
1243
1244#ifdef CONFIG_BRIDGE_NETFILTER
1245static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
1246{
1247 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
1248 kfree(nf_bridge);
1249}
1250static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
1251{
1252 if (nf_bridge)
1253 atomic_inc(&nf_bridge->use);
1254}
1255#endif /* CONFIG_BRIDGE_NETFILTER */
1256#else /* CONFIG_NETFILTER */
1257static inline void nf_reset(struct sk_buff *skb) {}
1258#endif /* CONFIG_NETFILTER */
1259
1260#endif /* __KERNEL__ */
1261#endif /* _LINUX_SKBUFF_H */