2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <iiitac@pyr.swan.ac.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
7 * Version: $Id: skbuff.c,v 1.90 2001/11/07 05:56:19 davem Exp $
10 * Alan Cox : Fixed the worst of the load
12 * Dave Platt : Interrupt stacking fix.
13 * Richard Kooijman : Timestamp fixes.
14 * Alan Cox : Changed buffer format.
15 * Alan Cox : destructor hook for AF_UNIX etc.
16 * Linus Torvalds : Better skb_clone.
17 * Alan Cox : Added skb_copy.
18 * Alan Cox : Added all the changed routines Linus
19 * only put in the headers
20 * Ray VanTassle : Fixed --skb->lock in free
21 * Alan Cox : skb_copy copy arp field
22 * Andi Kleen : slabified it.
23 * Robert Olsson : Removed skb_head_pool
26 * The __skb_ routines should be called with interrupts
27 * disabled, or you better be *real* sure that the operation is atomic
28 * with respect to whatever list is being frobbed (e.g. via lock_sock()
29 * or via disabling bottom half handlers, etc).
31 * This program is free software; you can redistribute it and/or
32 * modify it under the terms of the GNU General Public License
33 * as published by the Free Software Foundation; either version
34 * 2 of the License, or (at your option) any later version.
38 * The functions in this file will not compile correctly with gcc 2.4.x
41 #include <linux/config.h>
42 #include <linux/module.h>
43 #include <linux/types.h>
44 #include <linux/kernel.h>
45 #include <linux/sched.h>
47 #include <linux/interrupt.h>
49 #include <linux/inet.h>
50 #include <linux/slab.h>
51 #include <linux/netdevice.h>
52 #ifdef CONFIG_NET_CLS_ACT
53 #include <net/pkt_sched.h>
55 #include <linux/string.h>
56 #include <linux/skbuff.h>
57 #include <linux/cache.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/init.h>
60 #include <linux/highmem.h>
62 #include <net/protocol.h>
65 #include <net/checksum.h>
68 #include <asm/uaccess.h>
69 #include <asm/system.h>
71 static kmem_cache_t *skbuff_head_cache;
74 * Keep out-of-line to prevent kernel bloat.
75 * __builtin_return_address is not used because it is not always
80 * skb_over_panic - private function
85 * Out of line support code for skb_put(). Not user callable.
87 void skb_over_panic(struct sk_buff *skb, int sz, void *here)
89 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
90 "data:%p tail:%p end:%p dev:%s\n",
91 here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
92 skb->dev ? skb->dev->name : "<NULL>");
97 * skb_under_panic - private function
102 * Out of line support code for skb_push(). Not user callable.
105 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
107 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
108 "data:%p tail:%p end:%p dev:%s\n",
109 here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
110 skb->dev ? skb->dev->name : "<NULL>");
114 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
115 * 'private' fields and also do memory statistics to find all the
121 * alloc_skb - allocate a network buffer
122 * @size: size to allocate
123 * @gfp_mask: allocation mask
125 * Allocate a new &sk_buff. The returned buffer has no headroom and a
126 * tail room of size bytes. The object has a reference count of one.
127 * The return is the buffer. On a failure the return is %NULL.
129 * Buffers may only be allocated from interrupts using a @gfp_mask of
132 struct sk_buff *alloc_skb(unsigned int size, int gfp_mask)
138 skb = kmem_cache_alloc(skbuff_head_cache,
139 gfp_mask & ~__GFP_DMA);
143 /* Get the DATA. Size must match skb_add_mtu(). */
144 size = SKB_DATA_ALIGN(size);
145 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
149 memset(skb, 0, offsetof(struct sk_buff, truesize));
150 skb->truesize = size + sizeof(struct sk_buff);
151 atomic_set(&skb->users, 1);
155 skb->end = data + size;
157 atomic_set(&(skb_shinfo(skb)->dataref), 1);
158 skb_shinfo(skb)->nr_frags = 0;
159 skb_shinfo(skb)->tso_size = 0;
160 skb_shinfo(skb)->tso_segs = 0;
161 skb_shinfo(skb)->frag_list = NULL;
165 kmem_cache_free(skbuff_head_cache, skb);
171 * alloc_skb_from_cache - allocate a network buffer
172 * @cp: kmem_cache from which to allocate the data area
173 * (object size must be big enough for @size bytes + skb overheads)
174 * @size: size to allocate
175 * @gfp_mask: allocation mask
177 * Allocate a new &sk_buff. The returned buffer has no headroom and
178 * tail room of size bytes. The object has a reference count of one.
179 * The return is the buffer. On a failure the return is %NULL.
181 * Buffers may only be allocated from interrupts using a @gfp_mask of
184 struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp,
185 unsigned int size, int gfp_mask)
191 skb = kmem_cache_alloc(skbuff_head_cache,
192 gfp_mask & ~__GFP_DMA);
197 size = SKB_DATA_ALIGN(size);
198 data = kmem_cache_alloc(cp, gfp_mask);
202 memset(skb, 0, offsetof(struct sk_buff, truesize));
203 skb->truesize = size + sizeof(struct sk_buff);
204 atomic_set(&skb->users, 1);
208 skb->end = data + size;
210 atomic_set(&(skb_shinfo(skb)->dataref), 1);
211 skb_shinfo(skb)->nr_frags = 0;
212 skb_shinfo(skb)->tso_size = 0;
213 skb_shinfo(skb)->tso_segs = 0;
214 skb_shinfo(skb)->frag_list = NULL;
218 kmem_cache_free(skbuff_head_cache, skb);
224 static void skb_drop_fraglist(struct sk_buff *skb)
226 struct sk_buff *list = skb_shinfo(skb)->frag_list;
228 skb_shinfo(skb)->frag_list = NULL;
231 struct sk_buff *this = list;
237 static void skb_clone_fraglist(struct sk_buff *skb)
239 struct sk_buff *list;
241 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
245 void skb_release_data(struct sk_buff *skb)
248 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
249 &skb_shinfo(skb)->dataref)) {
250 if (skb_shinfo(skb)->nr_frags) {
252 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
253 put_page(skb_shinfo(skb)->frags[i].page);
256 if (skb_shinfo(skb)->frag_list)
257 skb_drop_fraglist(skb);
264 * Free an skbuff by memory without cleaning the state.
266 void kfree_skbmem(struct sk_buff *skb)
268 skb_release_data(skb);
269 kmem_cache_free(skbuff_head_cache, skb);
273 * __kfree_skb - private function
276 * Free an sk_buff. Release anything attached to the buffer.
277 * Clean the state. This is an internal helper function. Users should
278 * always call kfree_skb
281 void __kfree_skb(struct sk_buff *skb)
283 BUG_ON(skb->list != NULL);
285 dst_release(skb->dst);
287 secpath_put(skb->sp);
289 if (skb->destructor) {
291 skb->destructor(skb);
293 #ifdef CONFIG_NETFILTER
294 nf_conntrack_put(skb->nfct);
295 #ifdef CONFIG_BRIDGE_NETFILTER
296 nf_bridge_put(skb->nf_bridge);
299 /* XXX: IS this still necessary? - JHS */
300 #ifdef CONFIG_NET_SCHED
302 #ifdef CONFIG_NET_CLS_ACT
312 * skb_clone - duplicate an sk_buff
313 * @skb: buffer to clone
314 * @gfp_mask: allocation priority
316 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
317 * copies share the same packet data but not structure. The new
318 * buffer has a reference count of 1. If the allocation fails the
319 * function returns %NULL otherwise the new buffer is returned.
321 * If this function is called from an interrupt gfp_mask() must be
325 struct sk_buff *skb_clone(struct sk_buff *skb, int gfp_mask)
327 struct sk_buff *n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
332 #define C(x) n->x = skb->x
334 n->next = n->prev = NULL;
347 secpath_get(skb->sp);
349 memcpy(n->cb, skb->cb, sizeof(skb->cb));
361 n->destructor = NULL;
362 #ifdef CONFIG_NETFILTER
366 nf_conntrack_get(skb->nfct);
368 #ifdef CONFIG_BRIDGE_NETFILTER
370 nf_bridge_get(skb->nf_bridge);
372 #endif /*CONFIG_NETFILTER*/
373 #if defined(CONFIG_HIPPI)
376 #ifdef CONFIG_NET_SCHED
378 #ifdef CONFIG_NET_CLS_ACT
379 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
380 n->tc_verd = CLR_TC_OK2MUNGE(skb->tc_verd);
381 n->tc_verd = CLR_TC_MUNGED(skb->tc_verd);
388 atomic_set(&n->users, 1);
394 atomic_inc(&(skb_shinfo(skb)->dataref));
400 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
403 * Shift between the two data areas in bytes
405 unsigned long offset = new->data - old->data;
410 new->real_dev = old->real_dev;
411 new->priority = old->priority;
412 new->protocol = old->protocol;
413 new->dst = dst_clone(old->dst);
415 new->sp = secpath_get(old->sp);
417 new->h.raw = old->h.raw + offset;
418 new->nh.raw = old->nh.raw + offset;
419 new->mac.raw = old->mac.raw + offset;
420 memcpy(new->cb, old->cb, sizeof(old->cb));
421 new->local_df = old->local_df;
422 new->pkt_type = old->pkt_type;
423 new->stamp = old->stamp;
424 new->destructor = NULL;
425 new->security = old->security;
426 #ifdef CONFIG_NETFILTER
427 new->nfmark = old->nfmark;
428 new->nfcache = old->nfcache;
429 new->nfct = old->nfct;
430 nf_conntrack_get(old->nfct);
431 new->nfctinfo = old->nfctinfo;
432 #ifdef CONFIG_BRIDGE_NETFILTER
433 new->nf_bridge = old->nf_bridge;
434 nf_bridge_get(old->nf_bridge);
437 #ifdef CONFIG_NET_SCHED
438 #ifdef CONFIG_NET_CLS_ACT
439 new->tc_verd = old->tc_verd;
441 new->tc_index = old->tc_index;
443 atomic_set(&new->users, 1);
444 skb_shinfo(new)->tso_size = skb_shinfo(old)->tso_size;
445 skb_shinfo(new)->tso_segs = skb_shinfo(old)->tso_segs;
449 * skb_copy - create private copy of an sk_buff
450 * @skb: buffer to copy
451 * @gfp_mask: allocation priority
453 * Make a copy of both an &sk_buff and its data. This is used when the
454 * caller wishes to modify the data and needs a private copy of the
455 * data to alter. Returns %NULL on failure or the pointer to the buffer
456 * on success. The returned buffer has a reference count of 1.
458 * As by-product this function converts non-linear &sk_buff to linear
459 * one, so that &sk_buff becomes completely private and caller is allowed
460 * to modify all the data of returned buffer. This means that this
461 * function is not recommended for use in circumstances when only
462 * header is going to be modified. Use pskb_copy() instead.
465 struct sk_buff *skb_copy(const struct sk_buff *skb, int gfp_mask)
467 int headerlen = skb->data - skb->head;
469 * Allocate the copy buffer
471 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
476 /* Set the data pointer */
477 skb_reserve(n, headerlen);
478 /* Set the tail pointer and length */
479 skb_put(n, skb->len);
481 n->ip_summed = skb->ip_summed;
483 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
486 copy_skb_header(n, skb);
492 * pskb_copy - create copy of an sk_buff with private head.
493 * @skb: buffer to copy
494 * @gfp_mask: allocation priority
496 * Make a copy of both an &sk_buff and part of its data, located
497 * in header. Fragmented data remain shared. This is used when
498 * the caller wishes to modify only header of &sk_buff and needs
499 * private copy of the header to alter. Returns %NULL on failure
500 * or the pointer to the buffer on success.
501 * The returned buffer has a reference count of 1.
504 struct sk_buff *pskb_copy(struct sk_buff *skb, int gfp_mask)
507 * Allocate the copy buffer
509 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
514 /* Set the data pointer */
515 skb_reserve(n, skb->data - skb->head);
516 /* Set the tail pointer and length */
517 skb_put(n, skb_headlen(skb));
519 memcpy(n->data, skb->data, n->len);
521 n->ip_summed = skb->ip_summed;
523 n->data_len = skb->data_len;
526 if (skb_shinfo(skb)->nr_frags) {
529 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
530 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
531 get_page(skb_shinfo(n)->frags[i].page);
533 skb_shinfo(n)->nr_frags = i;
536 if (skb_shinfo(skb)->frag_list) {
537 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
538 skb_clone_fraglist(n);
541 copy_skb_header(n, skb);
547 * pskb_expand_head - reallocate header of &sk_buff
548 * @skb: buffer to reallocate
549 * @nhead: room to add at head
550 * @ntail: room to add at tail
551 * @gfp_mask: allocation priority
553 * Expands (or creates identical copy, if &nhead and &ntail are zero)
554 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
555 * reference count of 1. Returns zero in the case of success or error,
556 * if expansion failed. In the last case, &sk_buff is not changed.
558 * All the pointers pointing into skb header may change and must be
559 * reloaded after call to this function.
562 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, int gfp_mask)
566 int size = nhead + (skb->end - skb->head) + ntail;
572 size = SKB_DATA_ALIGN(size);
574 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
578 /* Copy only real data... and, alas, header. This should be
579 * optimized for the cases when header is void. */
580 memcpy(data + nhead, skb->head, skb->tail - skb->head);
581 memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
583 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
584 get_page(skb_shinfo(skb)->frags[i].page);
586 if (skb_shinfo(skb)->frag_list)
587 skb_clone_fraglist(skb);
589 skb_release_data(skb);
591 off = (data + nhead) - skb->head;
594 skb->end = data + size;
602 atomic_set(&skb_shinfo(skb)->dataref, 1);
609 /* Make private copy of skb with writable head and some headroom */
611 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
613 struct sk_buff *skb2;
614 int delta = headroom - skb_headroom(skb);
617 skb2 = pskb_copy(skb, GFP_ATOMIC);
619 skb2 = skb_clone(skb, GFP_ATOMIC);
620 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
631 * skb_copy_expand - copy and expand sk_buff
632 * @skb: buffer to copy
633 * @newheadroom: new free bytes at head
634 * @newtailroom: new free bytes at tail
635 * @gfp_mask: allocation priority
637 * Make a copy of both an &sk_buff and its data and while doing so
638 * allocate additional space.
640 * This is used when the caller wishes to modify the data and needs a
641 * private copy of the data to alter as well as more space for new fields.
642 * Returns %NULL on failure or the pointer to the buffer
643 * on success. The returned buffer has a reference count of 1.
645 * You must pass %GFP_ATOMIC as the allocation priority if this function
646 * is called from an interrupt.
648 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
649 * only by netfilter in the cases when checksum is recalculated? --ANK
651 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
652 int newheadroom, int newtailroom, int gfp_mask)
655 * Allocate the copy buffer
657 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
659 int head_copy_len, head_copy_off;
664 skb_reserve(n, newheadroom);
666 /* Set the tail pointer and length */
667 skb_put(n, skb->len);
669 head_copy_len = skb_headroom(skb);
671 if (newheadroom <= head_copy_len)
672 head_copy_len = newheadroom;
674 head_copy_off = newheadroom - head_copy_len;
676 /* Copy the linear header and data. */
677 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
678 skb->len + head_copy_len))
681 copy_skb_header(n, skb);
687 * skb_pad - zero pad the tail of an skb
688 * @skb: buffer to pad
691 * Ensure that a buffer is followed by a padding area that is zero
692 * filled. Used by network drivers which may DMA or transfer data
693 * beyond the buffer end onto the wire.
695 * May return NULL in out of memory cases.
698 struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
700 struct sk_buff *nskb;
702 /* If the skbuff is non linear tailroom is always zero.. */
703 if (skb_tailroom(skb) >= pad) {
704 memset(skb->data+skb->len, 0, pad);
708 nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
711 memset(nskb->data+nskb->len, 0, pad);
715 /* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
716 * If realloc==0 and trimming is impossible without change of data,
720 int ___pskb_trim(struct sk_buff *skb, unsigned int len, int realloc)
722 int offset = skb_headlen(skb);
723 int nfrags = skb_shinfo(skb)->nr_frags;
726 for (i = 0; i < nfrags; i++) {
727 int end = offset + skb_shinfo(skb)->frags[i].size;
729 if (skb_cloned(skb)) {
732 if (pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
736 put_page(skb_shinfo(skb)->frags[i].page);
737 skb_shinfo(skb)->nr_frags--;
739 skb_shinfo(skb)->frags[i].size = len - offset;
746 skb->data_len -= skb->len - len;
749 if (len <= skb_headlen(skb)) {
752 skb->tail = skb->data + len;
753 if (skb_shinfo(skb)->frag_list && !skb_cloned(skb))
754 skb_drop_fraglist(skb);
756 skb->data_len -= skb->len - len;
765 * __pskb_pull_tail - advance tail of skb header
766 * @skb: buffer to reallocate
767 * @delta: number of bytes to advance tail
769 * The function makes a sense only on a fragmented &sk_buff,
770 * it expands header moving its tail forward and copying necessary
771 * data from fragmented part.
773 * &sk_buff MUST have reference count of 1.
775 * Returns %NULL (and &sk_buff does not change) if pull failed
776 * or value of new tail of skb in the case of success.
778 * All the pointers pointing into skb header may change and must be
779 * reloaded after call to this function.
782 /* Moves tail of skb head forward, copying data from fragmented part,
783 * when it is necessary.
784 * 1. It may fail due to malloc failure.
785 * 2. It may change skb pointers.
787 * It is pretty complicated. Luckily, it is called only in exceptional cases.
789 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
791 /* If skb has not enough free space at tail, get new one
792 * plus 128 bytes for future expansions. If we have enough
793 * room at tail, reallocate without expansion only if skb is cloned.
795 int i, k, eat = (skb->tail + delta) - skb->end;
797 if (eat > 0 || skb_cloned(skb)) {
798 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
803 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
806 /* Optimization: no fragments, no reasons to preestimate
807 * size of pulled pages. Superb.
809 if (!skb_shinfo(skb)->frag_list)
812 /* Estimate size of pulled pages. */
814 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
815 if (skb_shinfo(skb)->frags[i].size >= eat)
817 eat -= skb_shinfo(skb)->frags[i].size;
820 /* If we need update frag list, we are in troubles.
821 * Certainly, it possible to add an offset to skb data,
822 * but taking into account that pulling is expected to
823 * be very rare operation, it is worth to fight against
824 * further bloating skb head and crucify ourselves here instead.
825 * Pure masohism, indeed. 8)8)
828 struct sk_buff *list = skb_shinfo(skb)->frag_list;
829 struct sk_buff *clone = NULL;
830 struct sk_buff *insp = NULL;
836 if (list->len <= eat) {
837 /* Eaten as whole. */
842 /* Eaten partially. */
844 if (skb_shared(list)) {
845 /* Sucks! We need to fork list. :-( */
846 clone = skb_clone(list, GFP_ATOMIC);
852 /* This may be pulled without
856 if (!pskb_pull(list, eat)) {
865 /* Free pulled out fragments. */
866 while ((list = skb_shinfo(skb)->frag_list) != insp) {
867 skb_shinfo(skb)->frag_list = list->next;
870 /* And insert new clone at head. */
873 skb_shinfo(skb)->frag_list = clone;
876 /* Success! Now we may commit changes to skb data. */
881 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
882 if (skb_shinfo(skb)->frags[i].size <= eat) {
883 put_page(skb_shinfo(skb)->frags[i].page);
884 eat -= skb_shinfo(skb)->frags[i].size;
886 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
888 skb_shinfo(skb)->frags[k].page_offset += eat;
889 skb_shinfo(skb)->frags[k].size -= eat;
895 skb_shinfo(skb)->nr_frags = k;
898 skb->data_len -= delta;
903 /* Copy some data bits from skb to kernel buffer. */
905 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
908 int start = skb_headlen(skb);
910 if (offset > (int)skb->len - len)
914 if ((copy = start - offset) > 0) {
917 memcpy(to, skb->data + offset, copy);
918 if ((len -= copy) == 0)
924 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
927 BUG_TRAP(start <= offset + len);
929 end = start + skb_shinfo(skb)->frags[i].size;
930 if ((copy = end - offset) > 0) {
936 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
938 vaddr + skb_shinfo(skb)->frags[i].page_offset+
939 offset - start, copy);
940 kunmap_skb_frag(vaddr);
942 if ((len -= copy) == 0)
950 if (skb_shinfo(skb)->frag_list) {
951 struct sk_buff *list = skb_shinfo(skb)->frag_list;
953 for (; list; list = list->next) {
956 BUG_TRAP(start <= offset + len);
958 end = start + list->len;
959 if ((copy = end - offset) > 0) {
962 if (skb_copy_bits(list, offset - start,
965 if ((len -= copy) == 0)
981 * skb_store_bits - store bits from kernel buffer to skb
982 * @skb: destination buffer
983 * @offset: offset in destination
984 * @from: source buffer
985 * @len: number of bytes to copy
987 * Copy the specified number of bytes from the source buffer to the
988 * destination skb. This function handles all the messy bits of
989 * traversing fragment lists and such.
992 int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
995 int start = skb_headlen(skb);
997 if (offset > (int)skb->len - len)
1000 if ((copy = start - offset) > 0) {
1003 memcpy(skb->data + offset, from, copy);
1004 if ((len -= copy) == 0)
1010 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1011 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1014 BUG_TRAP(start <= offset + len);
1016 end = start + frag->size;
1017 if ((copy = end - offset) > 0) {
1023 vaddr = kmap_skb_frag(frag);
1024 memcpy(vaddr + frag->page_offset + offset - start,
1026 kunmap_skb_frag(vaddr);
1028 if ((len -= copy) == 0)
1036 if (skb_shinfo(skb)->frag_list) {
1037 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1039 for (; list; list = list->next) {
1042 BUG_TRAP(start <= offset + len);
1044 end = start + list->len;
1045 if ((copy = end - offset) > 0) {
1048 if (skb_store_bits(list, offset - start,
1051 if ((len -= copy) == 0)
1066 EXPORT_SYMBOL(skb_store_bits);
1068 /* Checksum skb data. */
1070 unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1071 int len, unsigned int csum)
1073 int start = skb_headlen(skb);
1074 int i, copy = start - offset;
1077 /* Checksum header. */
1081 csum = csum_partial(skb->data + offset, copy, csum);
1082 if ((len -= copy) == 0)
1088 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1091 BUG_TRAP(start <= offset + len);
1093 end = start + skb_shinfo(skb)->frags[i].size;
1094 if ((copy = end - offset) > 0) {
1097 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1101 vaddr = kmap_skb_frag(frag);
1102 csum2 = csum_partial(vaddr + frag->page_offset +
1103 offset - start, copy, 0);
1104 kunmap_skb_frag(vaddr);
1105 csum = csum_block_add(csum, csum2, pos);
1114 if (skb_shinfo(skb)->frag_list) {
1115 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1117 for (; list; list = list->next) {
1120 BUG_TRAP(start <= offset + len);
1122 end = start + list->len;
1123 if ((copy = end - offset) > 0) {
1127 csum2 = skb_checksum(list, offset - start,
1129 csum = csum_block_add(csum, csum2, pos);
1130 if ((len -= copy) == 0)
1144 /* Both of above in one bottle. */
1146 unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1147 u8 *to, int len, unsigned int csum)
1149 int start = skb_headlen(skb);
1150 int i, copy = start - offset;
1157 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1159 if ((len -= copy) == 0)
1166 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1169 BUG_TRAP(start <= offset + len);
1171 end = start + skb_shinfo(skb)->frags[i].size;
1172 if ((copy = end - offset) > 0) {
1175 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1179 vaddr = kmap_skb_frag(frag);
1180 csum2 = csum_partial_copy_nocheck(vaddr +
1184 kunmap_skb_frag(vaddr);
1185 csum = csum_block_add(csum, csum2, pos);
1195 if (skb_shinfo(skb)->frag_list) {
1196 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1198 for (; list; list = list->next) {
1202 BUG_TRAP(start <= offset + len);
1204 end = start + list->len;
1205 if ((copy = end - offset) > 0) {
1208 csum2 = skb_copy_and_csum_bits(list,
1211 csum = csum_block_add(csum, csum2, pos);
1212 if ((len -= copy) == 0)
1226 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1231 if (skb->ip_summed == CHECKSUM_HW)
1232 csstart = skb->h.raw - skb->data;
1234 csstart = skb_headlen(skb);
1236 if (csstart > skb_headlen(skb))
1239 memcpy(to, skb->data, csstart);
1242 if (csstart != skb->len)
1243 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1244 skb->len - csstart, 0);
1246 if (skb->ip_summed == CHECKSUM_HW) {
1247 long csstuff = csstart + skb->csum;
1249 *((unsigned short *)(to + csstuff)) = csum_fold(csum);
1254 * skb_dequeue - remove from the head of the queue
1255 * @list: list to dequeue from
1257 * Remove the head of the list. The list lock is taken so the function
1258 * may be used safely with other locking list functions. The head item is
1259 * returned or %NULL if the list is empty.
1262 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1264 unsigned long flags;
1265 struct sk_buff *result;
1267 spin_lock_irqsave(&list->lock, flags);
1268 result = __skb_dequeue(list);
1269 spin_unlock_irqrestore(&list->lock, flags);
1274 * skb_dequeue_tail - remove from the tail of the queue
1275 * @list: list to dequeue from
1277 * Remove the tail of the list. The list lock is taken so the function
1278 * may be used safely with other locking list functions. The tail item is
1279 * returned or %NULL if the list is empty.
1281 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1283 unsigned long flags;
1284 struct sk_buff *result;
1286 spin_lock_irqsave(&list->lock, flags);
1287 result = __skb_dequeue_tail(list);
1288 spin_unlock_irqrestore(&list->lock, flags);
1293 * skb_queue_purge - empty a list
1294 * @list: list to empty
1296 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1297 * the list and one reference dropped. This function takes the list
1298 * lock and is atomic with respect to other list locking functions.
1300 void skb_queue_purge(struct sk_buff_head *list)
1302 struct sk_buff *skb;
1303 while ((skb = skb_dequeue(list)) != NULL)
1308 * skb_queue_head - queue a buffer at the list head
1309 * @list: list to use
1310 * @newsk: buffer to queue
1312 * Queue a buffer at the start of the list. This function takes the
1313 * list lock and can be used safely with other locking &sk_buff functions
1316 * A buffer cannot be placed on two lists at the same time.
1318 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1320 unsigned long flags;
1322 spin_lock_irqsave(&list->lock, flags);
1323 __skb_queue_head(list, newsk);
1324 spin_unlock_irqrestore(&list->lock, flags);
1328 * skb_queue_tail - queue a buffer at the list tail
1329 * @list: list to use
1330 * @newsk: buffer to queue
1332 * Queue a buffer at the tail of the list. This function takes the
1333 * list lock and can be used safely with other locking &sk_buff functions
1336 * A buffer cannot be placed on two lists at the same time.
1338 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1340 unsigned long flags;
1342 spin_lock_irqsave(&list->lock, flags);
1343 __skb_queue_tail(list, newsk);
1344 spin_unlock_irqrestore(&list->lock, flags);
1347 * skb_unlink - remove a buffer from a list
1348 * @skb: buffer to remove
1350 * Place a packet after a given packet in a list. The list locks are taken
1351 * and this function is atomic with respect to other list locked calls
1353 * Works even without knowing the list it is sitting on, which can be
1354 * handy at times. It also means that THE LIST MUST EXIST when you
1355 * unlink. Thus a list must have its contents unlinked before it is
1358 void skb_unlink(struct sk_buff *skb)
1360 struct sk_buff_head *list = skb->list;
1363 unsigned long flags;
1365 spin_lock_irqsave(&list->lock, flags);
1366 if (skb->list == list)
1367 __skb_unlink(skb, skb->list);
1368 spin_unlock_irqrestore(&list->lock, flags);
1374 * skb_append - append a buffer
1375 * @old: buffer to insert after
1376 * @newsk: buffer to insert
1378 * Place a packet after a given packet in a list. The list locks are taken
1379 * and this function is atomic with respect to other list locked calls.
1380 * A buffer cannot be placed on two lists at the same time.
1383 void skb_append(struct sk_buff *old, struct sk_buff *newsk)
1385 unsigned long flags;
1387 spin_lock_irqsave(&old->list->lock, flags);
1388 __skb_append(old, newsk);
1389 spin_unlock_irqrestore(&old->list->lock, flags);
1394 * skb_insert - insert a buffer
1395 * @old: buffer to insert before
1396 * @newsk: buffer to insert
1398 * Place a packet before a given packet in a list. The list locks are taken
1399 * and this function is atomic with respect to other list locked calls
1400 * A buffer cannot be placed on two lists at the same time.
1403 void skb_insert(struct sk_buff *old, struct sk_buff *newsk)
1405 unsigned long flags;
1407 spin_lock_irqsave(&old->list->lock, flags);
1408 __skb_insert(newsk, old->prev, old, old->list);
1409 spin_unlock_irqrestore(&old->list->lock, flags);
1414 * Tune the memory allocator for a new MTU size.
1416 void skb_add_mtu(int mtu)
1418 /* Must match allocation in alloc_skb */
1419 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1421 kmem_add_cache_size(mtu);
1425 static inline void skb_split_inside_header(struct sk_buff *skb,
1426 struct sk_buff* skb1,
1427 const u32 len, const int pos)
1431 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1433 /* And move data appendix as is. */
1434 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1435 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1437 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1438 skb_shinfo(skb)->nr_frags = 0;
1439 skb1->data_len = skb->data_len;
1440 skb1->len += skb1->data_len;
1443 skb->tail = skb->data + len;
1446 static inline void skb_split_no_header(struct sk_buff *skb,
1447 struct sk_buff* skb1,
1448 const u32 len, int pos)
1451 const int nfrags = skb_shinfo(skb)->nr_frags;
1453 skb_shinfo(skb)->nr_frags = 0;
1454 skb1->len = skb1->data_len = skb->len - len;
1456 skb->data_len = len - pos;
1458 for (i = 0; i < nfrags; i++) {
1459 int size = skb_shinfo(skb)->frags[i].size;
1461 if (pos + size > len) {
1462 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1466 * We have two variants in this case:
1467 * 1. Move all the frag to the second
1468 * part, if it is possible. F.e.
1469 * this approach is mandatory for TUX,
1470 * where splitting is expensive.
1471 * 2. Split is accurately. We make this.
1473 get_page(skb_shinfo(skb)->frags[i].page);
1474 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1475 skb_shinfo(skb1)->frags[0].size -= len - pos;
1476 skb_shinfo(skb)->frags[i].size = len - pos;
1477 skb_shinfo(skb)->nr_frags++;
1481 skb_shinfo(skb)->nr_frags++;
1484 skb_shinfo(skb1)->nr_frags = k;
1488 * skb_split - Split fragmented skb to two parts at length len.
1489 * @skb: the buffer to split
1490 * @skb1: the buffer to receive the second part
1491 * @len: new length for skb
1493 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1495 int pos = skb_headlen(skb);
1497 if (len < pos) /* Split line is inside header. */
1498 skb_split_inside_header(skb, skb1, len, pos);
1499 else /* Second chunk has no header, nothing to copy. */
1500 skb_split_no_header(skb, skb1, len, pos);
1503 void __init skb_init(void)
1505 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1506 sizeof(struct sk_buff),
1510 if (!skbuff_head_cache)
1511 panic("cannot create skbuff cache");
1514 EXPORT_SYMBOL(___pskb_trim);
1515 EXPORT_SYMBOL(__kfree_skb);
1516 EXPORT_SYMBOL(__pskb_pull_tail);
1517 EXPORT_SYMBOL(alloc_skb);
1518 EXPORT_SYMBOL(pskb_copy);
1519 EXPORT_SYMBOL(pskb_expand_head);
1520 EXPORT_SYMBOL(skb_checksum);
1521 EXPORT_SYMBOL(skb_clone);
1522 EXPORT_SYMBOL(skb_clone_fraglist);
1523 EXPORT_SYMBOL(skb_copy);
1524 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1525 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1526 EXPORT_SYMBOL(skb_copy_bits);
1527 EXPORT_SYMBOL(skb_copy_expand);
1528 EXPORT_SYMBOL(skb_over_panic);
1529 EXPORT_SYMBOL(skb_pad);
1530 EXPORT_SYMBOL(skb_realloc_headroom);
1531 EXPORT_SYMBOL(skb_under_panic);
1532 EXPORT_SYMBOL(skb_dequeue);
1533 EXPORT_SYMBOL(skb_dequeue_tail);
1534 EXPORT_SYMBOL(skb_insert);
1535 EXPORT_SYMBOL(skb_queue_purge);
1536 EXPORT_SYMBOL(skb_queue_head);
1537 EXPORT_SYMBOL(skb_queue_tail);
1538 EXPORT_SYMBOL(skb_unlink);
1539 EXPORT_SYMBOL(skb_append);
1540 EXPORT_SYMBOL(skb_split);