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_NETFILTER_DEBUG
371 #ifdef CONFIG_BRIDGE_NETFILTER
373 nf_bridge_get(skb->nf_bridge);
375 #endif /*CONFIG_NETFILTER*/
376 #if defined(CONFIG_HIPPI)
379 #ifdef CONFIG_NET_SCHED
381 #ifdef CONFIG_NET_CLS_ACT
382 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
383 n->tc_verd = CLR_TC_OK2MUNGE(skb->tc_verd);
384 n->tc_verd = CLR_TC_MUNGED(skb->tc_verd);
390 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
394 atomic_set(&n->users, 1);
400 atomic_inc(&(skb_shinfo(skb)->dataref));
406 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
409 * Shift between the two data areas in bytes
411 unsigned long offset = new->data - old->data;
416 new->real_dev = old->real_dev;
417 new->priority = old->priority;
418 new->protocol = old->protocol;
419 new->dst = dst_clone(old->dst);
421 new->sp = secpath_get(old->sp);
423 new->h.raw = old->h.raw + offset;
424 new->nh.raw = old->nh.raw + offset;
425 new->mac.raw = old->mac.raw + offset;
426 memcpy(new->cb, old->cb, sizeof(old->cb));
427 new->local_df = old->local_df;
428 new->pkt_type = old->pkt_type;
429 new->stamp = old->stamp;
430 new->destructor = NULL;
431 new->security = old->security;
432 #ifdef CONFIG_NETFILTER
433 new->nfmark = old->nfmark;
434 new->nfcache = old->nfcache;
435 new->nfct = old->nfct;
436 nf_conntrack_get(old->nfct);
437 new->nfctinfo = old->nfctinfo;
438 #ifdef CONFIG_NETFILTER_DEBUG
439 new->nf_debug = old->nf_debug;
441 #ifdef CONFIG_BRIDGE_NETFILTER
442 new->nf_bridge = old->nf_bridge;
443 nf_bridge_get(old->nf_bridge);
446 #ifdef CONFIG_NET_SCHED
447 #ifdef CONFIG_NET_CLS_ACT
448 new->tc_verd = old->tc_verd;
450 new->tc_index = old->tc_index;
452 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
455 atomic_set(&new->users, 1);
456 skb_shinfo(new)->tso_size = skb_shinfo(old)->tso_size;
457 skb_shinfo(new)->tso_segs = skb_shinfo(old)->tso_segs;
461 * skb_copy - create private copy of an sk_buff
462 * @skb: buffer to copy
463 * @gfp_mask: allocation priority
465 * Make a copy of both an &sk_buff and its data. This is used when the
466 * caller wishes to modify the data and needs a private copy of the
467 * data to alter. Returns %NULL on failure or the pointer to the buffer
468 * on success. The returned buffer has a reference count of 1.
470 * As by-product this function converts non-linear &sk_buff to linear
471 * one, so that &sk_buff becomes completely private and caller is allowed
472 * to modify all the data of returned buffer. This means that this
473 * function is not recommended for use in circumstances when only
474 * header is going to be modified. Use pskb_copy() instead.
477 struct sk_buff *skb_copy(const struct sk_buff *skb, int gfp_mask)
479 int headerlen = skb->data - skb->head;
481 * Allocate the copy buffer
483 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
488 /* Set the data pointer */
489 skb_reserve(n, headerlen);
490 /* Set the tail pointer and length */
491 skb_put(n, skb->len);
493 n->ip_summed = skb->ip_summed;
495 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
498 copy_skb_header(n, skb);
504 * pskb_copy - create copy of an sk_buff with private head.
505 * @skb: buffer to copy
506 * @gfp_mask: allocation priority
508 * Make a copy of both an &sk_buff and part of its data, located
509 * in header. Fragmented data remain shared. This is used when
510 * the caller wishes to modify only header of &sk_buff and needs
511 * private copy of the header to alter. Returns %NULL on failure
512 * or the pointer to the buffer on success.
513 * The returned buffer has a reference count of 1.
516 struct sk_buff *pskb_copy(struct sk_buff *skb, int gfp_mask)
519 * Allocate the copy buffer
521 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
526 /* Set the data pointer */
527 skb_reserve(n, skb->data - skb->head);
528 /* Set the tail pointer and length */
529 skb_put(n, skb_headlen(skb));
531 memcpy(n->data, skb->data, n->len);
533 n->ip_summed = skb->ip_summed;
535 n->data_len = skb->data_len;
538 if (skb_shinfo(skb)->nr_frags) {
541 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
542 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
543 get_page(skb_shinfo(n)->frags[i].page);
545 skb_shinfo(n)->nr_frags = i;
548 if (skb_shinfo(skb)->frag_list) {
549 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
550 skb_clone_fraglist(n);
553 copy_skb_header(n, skb);
559 * pskb_expand_head - reallocate header of &sk_buff
560 * @skb: buffer to reallocate
561 * @nhead: room to add at head
562 * @ntail: room to add at tail
563 * @gfp_mask: allocation priority
565 * Expands (or creates identical copy, if &nhead and &ntail are zero)
566 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
567 * reference count of 1. Returns zero in the case of success or error,
568 * if expansion failed. In the last case, &sk_buff is not changed.
570 * All the pointers pointing into skb header may change and must be
571 * reloaded after call to this function.
574 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, int gfp_mask)
578 int size = nhead + (skb->end - skb->head) + ntail;
584 size = SKB_DATA_ALIGN(size);
586 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
590 /* Copy only real data... and, alas, header. This should be
591 * optimized for the cases when header is void. */
592 memcpy(data + nhead, skb->head, skb->tail - skb->head);
593 memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
595 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
596 get_page(skb_shinfo(skb)->frags[i].page);
598 if (skb_shinfo(skb)->frag_list)
599 skb_clone_fraglist(skb);
601 skb_release_data(skb);
603 off = (data + nhead) - skb->head;
606 skb->end = data + size;
614 atomic_set(&skb_shinfo(skb)->dataref, 1);
621 /* Make private copy of skb with writable head and some headroom */
623 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
625 struct sk_buff *skb2;
626 int delta = headroom - skb_headroom(skb);
629 skb2 = pskb_copy(skb, GFP_ATOMIC);
631 skb2 = skb_clone(skb, GFP_ATOMIC);
632 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
643 * skb_copy_expand - copy and expand sk_buff
644 * @skb: buffer to copy
645 * @newheadroom: new free bytes at head
646 * @newtailroom: new free bytes at tail
647 * @gfp_mask: allocation priority
649 * Make a copy of both an &sk_buff and its data and while doing so
650 * allocate additional space.
652 * This is used when the caller wishes to modify the data and needs a
653 * private copy of the data to alter as well as more space for new fields.
654 * Returns %NULL on failure or the pointer to the buffer
655 * on success. The returned buffer has a reference count of 1.
657 * You must pass %GFP_ATOMIC as the allocation priority if this function
658 * is called from an interrupt.
660 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
661 * only by netfilter in the cases when checksum is recalculated? --ANK
663 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
664 int newheadroom, int newtailroom, int gfp_mask)
667 * Allocate the copy buffer
669 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
671 int head_copy_len, head_copy_off;
676 skb_reserve(n, newheadroom);
678 /* Set the tail pointer and length */
679 skb_put(n, skb->len);
681 head_copy_len = skb_headroom(skb);
683 if (newheadroom <= head_copy_len)
684 head_copy_len = newheadroom;
686 head_copy_off = newheadroom - head_copy_len;
688 /* Copy the linear header and data. */
689 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
690 skb->len + head_copy_len))
693 copy_skb_header(n, skb);
699 * skb_pad - zero pad the tail of an skb
700 * @skb: buffer to pad
703 * Ensure that a buffer is followed by a padding area that is zero
704 * filled. Used by network drivers which may DMA or transfer data
705 * beyond the buffer end onto the wire.
707 * May return NULL in out of memory cases.
710 struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
712 struct sk_buff *nskb;
714 /* If the skbuff is non linear tailroom is always zero.. */
715 if (skb_tailroom(skb) >= pad) {
716 memset(skb->data+skb->len, 0, pad);
720 nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
723 memset(nskb->data+nskb->len, 0, pad);
727 /* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
728 * If realloc==0 and trimming is impossible without change of data,
732 int ___pskb_trim(struct sk_buff *skb, unsigned int len, int realloc)
734 int offset = skb_headlen(skb);
735 int nfrags = skb_shinfo(skb)->nr_frags;
738 for (i = 0; i < nfrags; i++) {
739 int end = offset + skb_shinfo(skb)->frags[i].size;
741 if (skb_cloned(skb)) {
744 if (pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
748 put_page(skb_shinfo(skb)->frags[i].page);
749 skb_shinfo(skb)->nr_frags--;
751 skb_shinfo(skb)->frags[i].size = len - offset;
758 skb->data_len -= skb->len - len;
761 if (len <= skb_headlen(skb)) {
764 skb->tail = skb->data + len;
765 if (skb_shinfo(skb)->frag_list && !skb_cloned(skb))
766 skb_drop_fraglist(skb);
768 skb->data_len -= skb->len - len;
777 * __pskb_pull_tail - advance tail of skb header
778 * @skb: buffer to reallocate
779 * @delta: number of bytes to advance tail
781 * The function makes a sense only on a fragmented &sk_buff,
782 * it expands header moving its tail forward and copying necessary
783 * data from fragmented part.
785 * &sk_buff MUST have reference count of 1.
787 * Returns %NULL (and &sk_buff does not change) if pull failed
788 * or value of new tail of skb in the case of success.
790 * All the pointers pointing into skb header may change and must be
791 * reloaded after call to this function.
794 /* Moves tail of skb head forward, copying data from fragmented part,
795 * when it is necessary.
796 * 1. It may fail due to malloc failure.
797 * 2. It may change skb pointers.
799 * It is pretty complicated. Luckily, it is called only in exceptional cases.
801 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
803 /* If skb has not enough free space at tail, get new one
804 * plus 128 bytes for future expansions. If we have enough
805 * room at tail, reallocate without expansion only if skb is cloned.
807 int i, k, eat = (skb->tail + delta) - skb->end;
809 if (eat > 0 || skb_cloned(skb)) {
810 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
815 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
818 /* Optimization: no fragments, no reasons to preestimate
819 * size of pulled pages. Superb.
821 if (!skb_shinfo(skb)->frag_list)
824 /* Estimate size of pulled pages. */
826 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
827 if (skb_shinfo(skb)->frags[i].size >= eat)
829 eat -= skb_shinfo(skb)->frags[i].size;
832 /* If we need update frag list, we are in troubles.
833 * Certainly, it possible to add an offset to skb data,
834 * but taking into account that pulling is expected to
835 * be very rare operation, it is worth to fight against
836 * further bloating skb head and crucify ourselves here instead.
837 * Pure masohism, indeed. 8)8)
840 struct sk_buff *list = skb_shinfo(skb)->frag_list;
841 struct sk_buff *clone = NULL;
842 struct sk_buff *insp = NULL;
848 if (list->len <= eat) {
849 /* Eaten as whole. */
854 /* Eaten partially. */
856 if (skb_shared(list)) {
857 /* Sucks! We need to fork list. :-( */
858 clone = skb_clone(list, GFP_ATOMIC);
864 /* This may be pulled without
868 if (!pskb_pull(list, eat)) {
877 /* Free pulled out fragments. */
878 while ((list = skb_shinfo(skb)->frag_list) != insp) {
879 skb_shinfo(skb)->frag_list = list->next;
882 /* And insert new clone at head. */
885 skb_shinfo(skb)->frag_list = clone;
888 /* Success! Now we may commit changes to skb data. */
893 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
894 if (skb_shinfo(skb)->frags[i].size <= eat) {
895 put_page(skb_shinfo(skb)->frags[i].page);
896 eat -= skb_shinfo(skb)->frags[i].size;
898 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
900 skb_shinfo(skb)->frags[k].page_offset += eat;
901 skb_shinfo(skb)->frags[k].size -= eat;
907 skb_shinfo(skb)->nr_frags = k;
910 skb->data_len -= delta;
915 /* Copy some data bits from skb to kernel buffer. */
917 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
920 int start = skb_headlen(skb);
922 if (offset > (int)skb->len - len)
926 if ((copy = start - offset) > 0) {
929 memcpy(to, skb->data + offset, copy);
930 if ((len -= copy) == 0)
936 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
939 BUG_TRAP(start <= offset + len);
941 end = start + skb_shinfo(skb)->frags[i].size;
942 if ((copy = end - offset) > 0) {
948 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
950 vaddr + skb_shinfo(skb)->frags[i].page_offset+
951 offset - start, copy);
952 kunmap_skb_frag(vaddr);
954 if ((len -= copy) == 0)
962 if (skb_shinfo(skb)->frag_list) {
963 struct sk_buff *list = skb_shinfo(skb)->frag_list;
965 for (; list; list = list->next) {
968 BUG_TRAP(start <= offset + len);
970 end = start + list->len;
971 if ((copy = end - offset) > 0) {
974 if (skb_copy_bits(list, offset - start,
977 if ((len -= copy) == 0)
993 * skb_store_bits - store bits from kernel buffer to skb
994 * @skb: destination buffer
995 * @offset: offset in destination
996 * @from: source buffer
997 * @len: number of bytes to copy
999 * Copy the specified number of bytes from the source buffer to the
1000 * destination skb. This function handles all the messy bits of
1001 * traversing fragment lists and such.
1004 int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
1007 int start = skb_headlen(skb);
1009 if (offset > (int)skb->len - len)
1012 if ((copy = start - offset) > 0) {
1015 memcpy(skb->data + offset, from, copy);
1016 if ((len -= copy) == 0)
1022 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1023 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1026 BUG_TRAP(start <= offset + len);
1028 end = start + frag->size;
1029 if ((copy = end - offset) > 0) {
1035 vaddr = kmap_skb_frag(frag);
1036 memcpy(vaddr + frag->page_offset + offset - start,
1038 kunmap_skb_frag(vaddr);
1040 if ((len -= copy) == 0)
1048 if (skb_shinfo(skb)->frag_list) {
1049 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1051 for (; list; list = list->next) {
1054 BUG_TRAP(start <= offset + len);
1056 end = start + list->len;
1057 if ((copy = end - offset) > 0) {
1060 if (skb_store_bits(list, offset - start,
1063 if ((len -= copy) == 0)
1078 EXPORT_SYMBOL(skb_store_bits);
1080 /* Checksum skb data. */
1082 unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1083 int len, unsigned int csum)
1085 int start = skb_headlen(skb);
1086 int i, copy = start - offset;
1089 /* Checksum header. */
1093 csum = csum_partial(skb->data + offset, copy, csum);
1094 if ((len -= copy) == 0)
1100 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1103 BUG_TRAP(start <= offset + len);
1105 end = start + skb_shinfo(skb)->frags[i].size;
1106 if ((copy = end - offset) > 0) {
1109 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1113 vaddr = kmap_skb_frag(frag);
1114 csum2 = csum_partial(vaddr + frag->page_offset +
1115 offset - start, copy, 0);
1116 kunmap_skb_frag(vaddr);
1117 csum = csum_block_add(csum, csum2, pos);
1126 if (skb_shinfo(skb)->frag_list) {
1127 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1129 for (; list; list = list->next) {
1132 BUG_TRAP(start <= offset + len);
1134 end = start + list->len;
1135 if ((copy = end - offset) > 0) {
1139 csum2 = skb_checksum(list, offset - start,
1141 csum = csum_block_add(csum, csum2, pos);
1142 if ((len -= copy) == 0)
1156 /* Both of above in one bottle. */
1158 unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1159 u8 *to, int len, unsigned int csum)
1161 int start = skb_headlen(skb);
1162 int i, copy = start - offset;
1169 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1171 if ((len -= copy) == 0)
1178 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1181 BUG_TRAP(start <= offset + len);
1183 end = start + skb_shinfo(skb)->frags[i].size;
1184 if ((copy = end - offset) > 0) {
1187 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1191 vaddr = kmap_skb_frag(frag);
1192 csum2 = csum_partial_copy_nocheck(vaddr +
1196 kunmap_skb_frag(vaddr);
1197 csum = csum_block_add(csum, csum2, pos);
1207 if (skb_shinfo(skb)->frag_list) {
1208 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1210 for (; list; list = list->next) {
1214 BUG_TRAP(start <= offset + len);
1216 end = start + list->len;
1217 if ((copy = end - offset) > 0) {
1220 csum2 = skb_copy_and_csum_bits(list,
1223 csum = csum_block_add(csum, csum2, pos);
1224 if ((len -= copy) == 0)
1238 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1243 if (skb->ip_summed == CHECKSUM_HW)
1244 csstart = skb->h.raw - skb->data;
1246 csstart = skb_headlen(skb);
1248 if (csstart > skb_headlen(skb))
1251 memcpy(to, skb->data, csstart);
1254 if (csstart != skb->len)
1255 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1256 skb->len - csstart, 0);
1258 if (skb->ip_summed == CHECKSUM_HW) {
1259 long csstuff = csstart + skb->csum;
1261 *((unsigned short *)(to + csstuff)) = csum_fold(csum);
1266 * skb_dequeue - remove from the head of the queue
1267 * @list: list to dequeue from
1269 * Remove the head of the list. The list lock is taken so the function
1270 * may be used safely with other locking list functions. The head item is
1271 * returned or %NULL if the list is empty.
1274 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1276 unsigned long flags;
1277 struct sk_buff *result;
1279 spin_lock_irqsave(&list->lock, flags);
1280 result = __skb_dequeue(list);
1281 spin_unlock_irqrestore(&list->lock, flags);
1286 * skb_dequeue_tail - remove from the tail of the queue
1287 * @list: list to dequeue from
1289 * Remove the tail of the list. The list lock is taken so the function
1290 * may be used safely with other locking list functions. The tail item is
1291 * returned or %NULL if the list is empty.
1293 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1295 unsigned long flags;
1296 struct sk_buff *result;
1298 spin_lock_irqsave(&list->lock, flags);
1299 result = __skb_dequeue_tail(list);
1300 spin_unlock_irqrestore(&list->lock, flags);
1305 * skb_queue_purge - empty a list
1306 * @list: list to empty
1308 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1309 * the list and one reference dropped. This function takes the list
1310 * lock and is atomic with respect to other list locking functions.
1312 void skb_queue_purge(struct sk_buff_head *list)
1314 struct sk_buff *skb;
1315 while ((skb = skb_dequeue(list)) != NULL)
1320 * skb_queue_head - queue a buffer at the list head
1321 * @list: list to use
1322 * @newsk: buffer to queue
1324 * Queue a buffer at the start of the list. This function takes the
1325 * list lock and can be used safely with other locking &sk_buff functions
1328 * A buffer cannot be placed on two lists at the same time.
1330 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1332 unsigned long flags;
1334 spin_lock_irqsave(&list->lock, flags);
1335 __skb_queue_head(list, newsk);
1336 spin_unlock_irqrestore(&list->lock, flags);
1340 * skb_queue_tail - queue a buffer at the list tail
1341 * @list: list to use
1342 * @newsk: buffer to queue
1344 * Queue a buffer at the tail of the list. This function takes the
1345 * list lock and can be used safely with other locking &sk_buff functions
1348 * A buffer cannot be placed on two lists at the same time.
1350 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1352 unsigned long flags;
1354 spin_lock_irqsave(&list->lock, flags);
1355 __skb_queue_tail(list, newsk);
1356 spin_unlock_irqrestore(&list->lock, flags);
1359 * skb_unlink - remove a buffer from a list
1360 * @skb: buffer to remove
1362 * Place a packet after a given packet in a list. The list locks are taken
1363 * and this function is atomic with respect to other list locked calls
1365 * Works even without knowing the list it is sitting on, which can be
1366 * handy at times. It also means that THE LIST MUST EXIST when you
1367 * unlink. Thus a list must have its contents unlinked before it is
1370 void skb_unlink(struct sk_buff *skb)
1372 struct sk_buff_head *list = skb->list;
1375 unsigned long flags;
1377 spin_lock_irqsave(&list->lock, flags);
1378 if (skb->list == list)
1379 __skb_unlink(skb, skb->list);
1380 spin_unlock_irqrestore(&list->lock, flags);
1386 * skb_append - append a buffer
1387 * @old: buffer to insert after
1388 * @newsk: buffer to insert
1390 * Place a packet after a given packet in a list. The list locks are taken
1391 * and this function is atomic with respect to other list locked calls.
1392 * A buffer cannot be placed on two lists at the same time.
1395 void skb_append(struct sk_buff *old, struct sk_buff *newsk)
1397 unsigned long flags;
1399 spin_lock_irqsave(&old->list->lock, flags);
1400 __skb_append(old, newsk);
1401 spin_unlock_irqrestore(&old->list->lock, flags);
1406 * skb_insert - insert a buffer
1407 * @old: buffer to insert before
1408 * @newsk: buffer to insert
1410 * Place a packet before a given packet in a list. The list locks are taken
1411 * and this function is atomic with respect to other list locked calls
1412 * A buffer cannot be placed on two lists at the same time.
1415 void skb_insert(struct sk_buff *old, struct sk_buff *newsk)
1417 unsigned long flags;
1419 spin_lock_irqsave(&old->list->lock, flags);
1420 __skb_insert(newsk, old->prev, old, old->list);
1421 spin_unlock_irqrestore(&old->list->lock, flags);
1426 * Tune the memory allocator for a new MTU size.
1428 void skb_add_mtu(int mtu)
1430 /* Must match allocation in alloc_skb */
1431 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1433 kmem_add_cache_size(mtu);
1437 static inline void skb_split_inside_header(struct sk_buff *skb,
1438 struct sk_buff* skb1,
1439 const u32 len, const int pos)
1443 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1445 /* And move data appendix as is. */
1446 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1447 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1449 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1450 skb_shinfo(skb)->nr_frags = 0;
1451 skb1->data_len = skb->data_len;
1452 skb1->len += skb1->data_len;
1455 skb->tail = skb->data + len;
1458 static inline void skb_split_no_header(struct sk_buff *skb,
1459 struct sk_buff* skb1,
1460 const u32 len, int pos)
1463 const int nfrags = skb_shinfo(skb)->nr_frags;
1465 skb_shinfo(skb)->nr_frags = 0;
1466 skb1->len = skb1->data_len = skb->len - len;
1468 skb->data_len = len - pos;
1470 for (i = 0; i < nfrags; i++) {
1471 int size = skb_shinfo(skb)->frags[i].size;
1473 if (pos + size > len) {
1474 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1478 * We have two variants in this case:
1479 * 1. Move all the frag to the second
1480 * part, if it is possible. F.e.
1481 * this approach is mandatory for TUX,
1482 * where splitting is expensive.
1483 * 2. Split is accurately. We make this.
1485 get_page(skb_shinfo(skb)->frags[i].page);
1486 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1487 skb_shinfo(skb1)->frags[0].size -= len - pos;
1488 skb_shinfo(skb)->frags[i].size = len - pos;
1489 skb_shinfo(skb)->nr_frags++;
1493 skb_shinfo(skb)->nr_frags++;
1496 skb_shinfo(skb1)->nr_frags = k;
1500 * skb_split - Split fragmented skb to two parts at length len.
1501 * @skb: the buffer to split
1502 * @skb1: the buffer to receive the second part
1503 * @len: new length for skb
1505 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1507 int pos = skb_headlen(skb);
1509 if (len < pos) /* Split line is inside header. */
1510 skb_split_inside_header(skb, skb1, len, pos);
1511 else /* Second chunk has no header, nothing to copy. */
1512 skb_split_no_header(skb, skb1, len, pos);
1515 void __init skb_init(void)
1517 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1518 sizeof(struct sk_buff),
1522 if (!skbuff_head_cache)
1523 panic("cannot create skbuff cache");
1526 EXPORT_SYMBOL(___pskb_trim);
1527 EXPORT_SYMBOL(__kfree_skb);
1528 EXPORT_SYMBOL(__pskb_pull_tail);
1529 EXPORT_SYMBOL(alloc_skb);
1530 EXPORT_SYMBOL(pskb_copy);
1531 EXPORT_SYMBOL(pskb_expand_head);
1532 EXPORT_SYMBOL(skb_checksum);
1533 EXPORT_SYMBOL(skb_clone);
1534 EXPORT_SYMBOL(skb_clone_fraglist);
1535 EXPORT_SYMBOL(skb_copy);
1536 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1537 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1538 EXPORT_SYMBOL(skb_copy_bits);
1539 EXPORT_SYMBOL(skb_copy_expand);
1540 EXPORT_SYMBOL(skb_over_panic);
1541 EXPORT_SYMBOL(skb_pad);
1542 EXPORT_SYMBOL(skb_realloc_headroom);
1543 EXPORT_SYMBOL(skb_under_panic);
1544 EXPORT_SYMBOL(skb_dequeue);
1545 EXPORT_SYMBOL(skb_dequeue_tail);
1546 EXPORT_SYMBOL(skb_insert);
1547 EXPORT_SYMBOL(skb_queue_purge);
1548 EXPORT_SYMBOL(skb_queue_head);
1549 EXPORT_SYMBOL(skb_queue_tail);
1550 EXPORT_SYMBOL(skb_unlink);
1551 EXPORT_SYMBOL(skb_append);
1552 EXPORT_SYMBOL(skb_split);