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_INFO "skput:over: %p:%d put:%d dev:%s",
90 here, skb->len, sz, skb->dev ? skb->dev->name : "<NULL>");
95 * skb_under_panic - private function
100 * Out of line support code for skb_push(). Not user callable.
103 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
105 printk(KERN_INFO "skput:under: %p:%d put:%d dev:%s",
106 here, skb->len, sz, skb->dev ? skb->dev->name : "<NULL>");
110 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
111 * 'private' fields and also do memory statistics to find all the
117 * alloc_skb - allocate a network buffer
118 * @size: size to allocate
119 * @gfp_mask: allocation mask
121 * Allocate a new &sk_buff. The returned buffer has no headroom and a
122 * tail room of size bytes. The object has a reference count of one.
123 * The return is the buffer. On a failure the return is %NULL.
125 * Buffers may only be allocated from interrupts using a @gfp_mask of
128 struct sk_buff *alloc_skb(unsigned int size, int gfp_mask)
134 skb = kmem_cache_alloc(skbuff_head_cache,
135 gfp_mask & ~__GFP_DMA);
139 /* Get the DATA. Size must match skb_add_mtu(). */
140 size = SKB_DATA_ALIGN(size);
141 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
145 memset(skb, 0, offsetof(struct sk_buff, truesize));
146 skb->truesize = size + sizeof(struct sk_buff);
147 atomic_set(&skb->users, 1);
151 skb->end = data + size;
153 atomic_set(&(skb_shinfo(skb)->dataref), 1);
154 skb_shinfo(skb)->nr_frags = 0;
155 skb_shinfo(skb)->tso_size = 0;
156 skb_shinfo(skb)->tso_segs = 0;
157 skb_shinfo(skb)->frag_list = NULL;
161 kmem_cache_free(skbuff_head_cache, skb);
167 static void skb_drop_fraglist(struct sk_buff *skb)
169 struct sk_buff *list = skb_shinfo(skb)->frag_list;
171 skb_shinfo(skb)->frag_list = NULL;
174 struct sk_buff *this = list;
180 static void skb_clone_fraglist(struct sk_buff *skb)
182 struct sk_buff *list;
184 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
188 void skb_release_data(struct sk_buff *skb)
191 atomic_dec_and_test(&(skb_shinfo(skb)->dataref))) {
192 if (skb_shinfo(skb)->nr_frags) {
194 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
195 put_page(skb_shinfo(skb)->frags[i].page);
198 if (skb_shinfo(skb)->frag_list)
199 skb_drop_fraglist(skb);
206 * Free an skbuff by memory without cleaning the state.
208 void kfree_skbmem(struct sk_buff *skb)
210 skb_release_data(skb);
211 kmem_cache_free(skbuff_head_cache, skb);
215 * __kfree_skb - private function
218 * Free an sk_buff. Release anything attached to the buffer.
219 * Clean the state. This is an internal helper function. Users should
220 * always call kfree_skb
223 void __kfree_skb(struct sk_buff *skb)
226 printk(KERN_WARNING "Warning: kfree_skb passed an skb still "
227 "on a list (from %p).\n", NET_CALLER(skb));
231 dst_release(skb->dst);
233 secpath_put(skb->sp);
235 if(skb->destructor) {
237 printk(KERN_WARNING "Warning: kfree_skb on "
238 "hard IRQ %p\n", NET_CALLER(skb));
239 skb->destructor(skb);
241 #ifdef CONFIG_NETFILTER
242 nf_conntrack_put(skb->nfct);
243 #ifdef CONFIG_BRIDGE_NETFILTER
244 nf_bridge_put(skb->nf_bridge);
247 /* XXX: IS this still necessary? - JHS */
248 #ifdef CONFIG_NET_SCHED
250 #ifdef CONFIG_NET_CLS_ACT
260 * skb_clone - duplicate an sk_buff
261 * @skb: buffer to clone
262 * @gfp_mask: allocation priority
264 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
265 * copies share the same packet data but not structure. The new
266 * buffer has a reference count of 1. If the allocation fails the
267 * function returns %NULL otherwise the new buffer is returned.
269 * If this function is called from an interrupt gfp_mask() must be
273 struct sk_buff *skb_clone(struct sk_buff *skb, int gfp_mask)
275 struct sk_buff *n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
280 #define C(x) n->x = skb->x
282 n->next = n->prev = NULL;
295 secpath_get(skb->sp);
297 memcpy(n->cb, skb->cb, sizeof(skb->cb));
308 n->destructor = NULL;
309 #ifdef CONFIG_NETFILTER
313 nf_conntrack_get(skb->nfct);
314 #ifdef CONFIG_NETFILTER_DEBUG
317 #ifdef CONFIG_BRIDGE_NETFILTER
319 nf_bridge_get(skb->nf_bridge);
321 #endif /*CONFIG_NETFILTER*/
322 #if defined(CONFIG_HIPPI)
325 #ifdef CONFIG_NET_SCHED
327 #ifdef CONFIG_NET_CLS_ACT
328 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
329 n->tc_verd = CLR_TC_OK2MUNGE(skb->tc_verd);
330 n->tc_verd = CLR_TC_MUNGED(skb->tc_verd);
337 atomic_set(&n->users, 1);
343 atomic_inc(&(skb_shinfo(skb)->dataref));
349 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
352 * Shift between the two data areas in bytes
354 unsigned long offset = new->data - old->data;
359 new->real_dev = old->real_dev;
360 new->priority = old->priority;
361 new->protocol = old->protocol;
362 new->dst = dst_clone(old->dst);
364 new->sp = secpath_get(old->sp);
366 new->h.raw = old->h.raw + offset;
367 new->nh.raw = old->nh.raw + offset;
368 new->mac.raw = old->mac.raw + offset;
369 memcpy(new->cb, old->cb, sizeof(old->cb));
370 new->local_df = old->local_df;
371 new->pkt_type = old->pkt_type;
372 new->stamp = old->stamp;
373 new->destructor = NULL;
374 new->security = old->security;
375 #ifdef CONFIG_NETFILTER
376 new->nfmark = old->nfmark;
377 new->nfcache = old->nfcache;
378 new->nfct = old->nfct;
379 nf_conntrack_get(old->nfct);
380 #ifdef CONFIG_NETFILTER_DEBUG
381 new->nf_debug = old->nf_debug;
383 #ifdef CONFIG_BRIDGE_NETFILTER
384 new->nf_bridge = old->nf_bridge;
385 nf_bridge_get(old->nf_bridge);
388 #ifdef CONFIG_NET_SCHED
389 #ifdef CONFIG_NET_CLS_ACT
390 new->tc_verd = old->tc_verd;
392 new->tc_index = old->tc_index;
394 atomic_set(&new->users, 1);
398 * skb_copy - create private copy of an sk_buff
399 * @skb: buffer to copy
400 * @gfp_mask: allocation priority
402 * Make a copy of both an &sk_buff and its data. This is used when the
403 * caller wishes to modify the data and needs a private copy of the
404 * data to alter. Returns %NULL on failure or the pointer to the buffer
405 * on success. The returned buffer has a reference count of 1.
407 * As by-product this function converts non-linear &sk_buff to linear
408 * one, so that &sk_buff becomes completely private and caller is allowed
409 * to modify all the data of returned buffer. This means that this
410 * function is not recommended for use in circumstances when only
411 * header is going to be modified. Use pskb_copy() instead.
414 struct sk_buff *skb_copy(const struct sk_buff *skb, int gfp_mask)
416 int headerlen = skb->data - skb->head;
418 * Allocate the copy buffer
420 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
425 /* Set the data pointer */
426 skb_reserve(n, headerlen);
427 /* Set the tail pointer and length */
428 skb_put(n, skb->len);
430 n->ip_summed = skb->ip_summed;
432 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
435 copy_skb_header(n, skb);
441 * pskb_copy - create copy of an sk_buff with private head.
442 * @skb: buffer to copy
443 * @gfp_mask: allocation priority
445 * Make a copy of both an &sk_buff and part of its data, located
446 * in header. Fragmented data remain shared. This is used when
447 * the caller wishes to modify only header of &sk_buff and needs
448 * private copy of the header to alter. Returns %NULL on failure
449 * or the pointer to the buffer on success.
450 * The returned buffer has a reference count of 1.
453 struct sk_buff *pskb_copy(struct sk_buff *skb, int gfp_mask)
456 * Allocate the copy buffer
458 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
463 /* Set the data pointer */
464 skb_reserve(n, skb->data - skb->head);
465 /* Set the tail pointer and length */
466 skb_put(n, skb_headlen(skb));
468 memcpy(n->data, skb->data, n->len);
470 n->ip_summed = skb->ip_summed;
472 n->data_len = skb->data_len;
475 if (skb_shinfo(skb)->nr_frags) {
478 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
479 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
480 get_page(skb_shinfo(n)->frags[i].page);
482 skb_shinfo(n)->nr_frags = i;
484 skb_shinfo(n)->tso_size = skb_shinfo(skb)->tso_size;
485 skb_shinfo(n)->tso_segs = skb_shinfo(skb)->tso_segs;
487 if (skb_shinfo(skb)->frag_list) {
488 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
489 skb_clone_fraglist(n);
492 copy_skb_header(n, skb);
498 * pskb_expand_head - reallocate header of &sk_buff
499 * @skb: buffer to reallocate
500 * @nhead: room to add at head
501 * @ntail: room to add at tail
502 * @gfp_mask: allocation priority
504 * Expands (or creates identical copy, if &nhead and &ntail are zero)
505 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
506 * reference count of 1. Returns zero in the case of success or error,
507 * if expansion failed. In the last case, &sk_buff is not changed.
509 * All the pointers pointing into skb header may change and must be
510 * reloaded after call to this function.
513 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, int gfp_mask)
517 int size = nhead + (skb->end - skb->head) + ntail;
523 size = SKB_DATA_ALIGN(size);
525 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
529 /* Copy only real data... and, alas, header. This should be
530 * optimized for the cases when header is void. */
531 memcpy(data + nhead, skb->head, skb->tail - skb->head);
532 memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
534 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
535 get_page(skb_shinfo(skb)->frags[i].page);
537 if (skb_shinfo(skb)->frag_list)
538 skb_clone_fraglist(skb);
540 skb_release_data(skb);
542 off = (data + nhead) - skb->head;
545 skb->end = data + size;
552 atomic_set(&skb_shinfo(skb)->dataref, 1);
559 /* Make private copy of skb with writable head and some headroom */
561 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
563 struct sk_buff *skb2;
564 int delta = headroom - skb_headroom(skb);
567 skb2 = pskb_copy(skb, GFP_ATOMIC);
569 skb2 = skb_clone(skb, GFP_ATOMIC);
570 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
581 * skb_copy_expand - copy and expand sk_buff
582 * @skb: buffer to copy
583 * @newheadroom: new free bytes at head
584 * @newtailroom: new free bytes at tail
585 * @gfp_mask: allocation priority
587 * Make a copy of both an &sk_buff and its data and while doing so
588 * allocate additional space.
590 * This is used when the caller wishes to modify the data and needs a
591 * private copy of the data to alter as well as more space for new fields.
592 * Returns %NULL on failure or the pointer to the buffer
593 * on success. The returned buffer has a reference count of 1.
595 * You must pass %GFP_ATOMIC as the allocation priority if this function
596 * is called from an interrupt.
598 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
599 * only by netfilter in the cases when checksum is recalculated? --ANK
601 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
602 int newheadroom, int newtailroom, int gfp_mask)
605 * Allocate the copy buffer
607 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
609 int head_copy_len, head_copy_off;
614 skb_reserve(n, newheadroom);
616 /* Set the tail pointer and length */
617 skb_put(n, skb->len);
619 head_copy_len = skb_headroom(skb);
621 if (newheadroom <= head_copy_len)
622 head_copy_len = newheadroom;
624 head_copy_off = newheadroom - head_copy_len;
626 /* Copy the linear header and data. */
627 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
628 skb->len + head_copy_len))
631 copy_skb_header(n, skb);
632 skb_shinfo(n)->tso_size = skb_shinfo(skb)->tso_size;
633 skb_shinfo(n)->tso_segs = skb_shinfo(skb)->tso_segs;
639 * skb_pad - zero pad the tail of an skb
640 * @skb: buffer to pad
643 * Ensure that a buffer is followed by a padding area that is zero
644 * filled. Used by network drivers which may DMA or transfer data
645 * beyond the buffer end onto the wire.
647 * May return NULL in out of memory cases.
650 struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
652 struct sk_buff *nskb;
654 /* If the skbuff is non linear tailroom is always zero.. */
655 if (skb_tailroom(skb) >= pad) {
656 memset(skb->data+skb->len, 0, pad);
660 nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
663 memset(nskb->data+nskb->len, 0, pad);
667 /* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
668 * If realloc==0 and trimming is impossible without change of data,
672 int ___pskb_trim(struct sk_buff *skb, unsigned int len, int realloc)
674 int offset = skb_headlen(skb);
675 int nfrags = skb_shinfo(skb)->nr_frags;
678 for (i = 0; i < nfrags; i++) {
679 int end = offset + skb_shinfo(skb)->frags[i].size;
681 if (skb_cloned(skb)) {
684 if (pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
688 put_page(skb_shinfo(skb)->frags[i].page);
689 skb_shinfo(skb)->nr_frags--;
691 skb_shinfo(skb)->frags[i].size = len - offset;
698 skb->data_len -= skb->len - len;
701 if (len <= skb_headlen(skb)) {
704 skb->tail = skb->data + len;
705 if (skb_shinfo(skb)->frag_list && !skb_cloned(skb))
706 skb_drop_fraglist(skb);
708 skb->data_len -= skb->len - len;
717 * __pskb_pull_tail - advance tail of skb header
718 * @skb: buffer to reallocate
719 * @delta: number of bytes to advance tail
721 * The function makes a sense only on a fragmented &sk_buff,
722 * it expands header moving its tail forward and copying necessary
723 * data from fragmented part.
725 * &sk_buff MUST have reference count of 1.
727 * Returns %NULL (and &sk_buff does not change) if pull failed
728 * or value of new tail of skb in the case of success.
730 * All the pointers pointing into skb header may change and must be
731 * reloaded after call to this function.
734 /* Moves tail of skb head forward, copying data from fragmented part,
735 * when it is necessary.
736 * 1. It may fail due to malloc failure.
737 * 2. It may change skb pointers.
739 * It is pretty complicated. Luckily, it is called only in exceptional cases.
741 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
743 /* If skb has not enough free space at tail, get new one
744 * plus 128 bytes for future expansions. If we have enough
745 * room at tail, reallocate without expansion only if skb is cloned.
747 int i, k, eat = (skb->tail + delta) - skb->end;
749 if (eat > 0 || skb_cloned(skb)) {
750 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
755 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
758 /* Optimization: no fragments, no reasons to preestimate
759 * size of pulled pages. Superb.
761 if (!skb_shinfo(skb)->frag_list)
764 /* Estimate size of pulled pages. */
766 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
767 if (skb_shinfo(skb)->frags[i].size >= eat)
769 eat -= skb_shinfo(skb)->frags[i].size;
772 /* If we need update frag list, we are in troubles.
773 * Certainly, it possible to add an offset to skb data,
774 * but taking into account that pulling is expected to
775 * be very rare operation, it is worth to fight against
776 * further bloating skb head and crucify ourselves here instead.
777 * Pure masohism, indeed. 8)8)
780 struct sk_buff *list = skb_shinfo(skb)->frag_list;
781 struct sk_buff *clone = NULL;
782 struct sk_buff *insp = NULL;
788 if (list->len <= eat) {
789 /* Eaten as whole. */
794 /* Eaten partially. */
796 if (skb_shared(list)) {
797 /* Sucks! We need to fork list. :-( */
798 clone = skb_clone(list, GFP_ATOMIC);
804 /* This may be pulled without
808 if (!pskb_pull(list, eat)) {
817 /* Free pulled out fragments. */
818 while ((list = skb_shinfo(skb)->frag_list) != insp) {
819 skb_shinfo(skb)->frag_list = list->next;
822 /* And insert new clone at head. */
825 skb_shinfo(skb)->frag_list = clone;
828 /* Success! Now we may commit changes to skb data. */
833 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
834 if (skb_shinfo(skb)->frags[i].size <= eat) {
835 put_page(skb_shinfo(skb)->frags[i].page);
836 eat -= skb_shinfo(skb)->frags[i].size;
838 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
840 skb_shinfo(skb)->frags[k].page_offset += eat;
841 skb_shinfo(skb)->frags[k].size -= eat;
847 skb_shinfo(skb)->nr_frags = k;
850 skb->data_len -= delta;
855 /* Copy some data bits from skb to kernel buffer. */
857 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
860 int start = skb_headlen(skb);
862 if (offset > (int)skb->len - len)
866 if ((copy = start - offset) > 0) {
869 memcpy(to, skb->data + offset, copy);
870 if ((len -= copy) == 0)
876 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
879 BUG_TRAP(start <= offset + len);
881 end = start + skb_shinfo(skb)->frags[i].size;
882 if ((copy = end - offset) > 0) {
888 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
890 vaddr + skb_shinfo(skb)->frags[i].page_offset+
891 offset - start, copy);
892 kunmap_skb_frag(vaddr);
894 if ((len -= copy) == 0)
902 if (skb_shinfo(skb)->frag_list) {
903 struct sk_buff *list = skb_shinfo(skb)->frag_list;
905 for (; list; list = list->next) {
908 BUG_TRAP(start <= offset + len);
910 end = start + list->len;
911 if ((copy = end - offset) > 0) {
914 if (skb_copy_bits(list, offset - start,
917 if ((len -= copy) == 0)
932 /* Keep iterating until skb_iter_next returns false. */
933 void skb_iter_first(const struct sk_buff *skb, struct skb_iter *i)
935 i->len = skb_headlen(skb);
936 i->data = (unsigned char *)skb->data;
941 int skb_iter_next(const struct sk_buff *skb, struct skb_iter *i)
943 /* Unmap previous, if not head fragment. */
945 kunmap_skb_frag(i->data);
949 /* We're iterating through fraglist. */
950 if (i->nextfrag < skb_shinfo(i->fraglist)->nr_frags) {
951 i->data = kmap_skb_frag(&skb_shinfo(i->fraglist)
952 ->frags[i->nextfrag]);
953 i->len = skb_shinfo(i->fraglist)->frags[i->nextfrag]
958 /* Fragments with fragments? Too hard! */
959 BUG_ON(skb_shinfo(i->fraglist)->frag_list);
960 i->fraglist = i->fraglist->next;
964 i->len = skb_headlen(i->fraglist);
965 i->data = i->fraglist->data;
970 if (i->nextfrag < skb_shinfo(skb)->nr_frags) {
971 i->data = kmap_skb_frag(&skb_shinfo(skb)->frags[i->nextfrag]);
972 i->len = skb_shinfo(skb)->frags[i->nextfrag].size;
977 i->fraglist = skb_shinfo(skb)->frag_list;
982 /* Bug trap for callers */
987 void skb_iter_abort(const struct sk_buff *skb, struct skb_iter *i)
989 /* Unmap previous, if not head fragment. */
990 if (i->data && i->nextfrag)
991 kunmap_skb_frag(i->data);
992 /* Bug trap for callers */
996 /* Checksum skb data. */
998 unsigned int skb_checksum(const struct sk_buff *skb, int offset,
999 int len, unsigned int csum)
1001 int start = skb_headlen(skb);
1002 int i, copy = start - offset;
1005 /* Checksum header. */
1009 csum = csum_partial(skb->data + offset, copy, csum);
1010 if ((len -= copy) == 0)
1016 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1019 BUG_TRAP(start <= offset + len);
1021 end = start + skb_shinfo(skb)->frags[i].size;
1022 if ((copy = end - offset) > 0) {
1025 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1029 vaddr = kmap_skb_frag(frag);
1030 csum2 = csum_partial(vaddr + frag->page_offset +
1031 offset - start, copy, 0);
1032 kunmap_skb_frag(vaddr);
1033 csum = csum_block_add(csum, csum2, pos);
1042 if (skb_shinfo(skb)->frag_list) {
1043 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1045 for (; list; list = list->next) {
1048 BUG_TRAP(start <= offset + len);
1050 end = start + list->len;
1051 if ((copy = end - offset) > 0) {
1055 csum2 = skb_checksum(list, offset - start,
1057 csum = csum_block_add(csum, csum2, pos);
1058 if ((len -= copy) == 0)
1072 /* Both of above in one bottle. */
1074 unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1075 u8 *to, int len, unsigned int csum)
1077 int start = skb_headlen(skb);
1078 int i, copy = start - offset;
1085 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1087 if ((len -= copy) == 0)
1094 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1097 BUG_TRAP(start <= offset + len);
1099 end = start + skb_shinfo(skb)->frags[i].size;
1100 if ((copy = end - offset) > 0) {
1103 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1107 vaddr = kmap_skb_frag(frag);
1108 csum2 = csum_partial_copy_nocheck(vaddr +
1112 kunmap_skb_frag(vaddr);
1113 csum = csum_block_add(csum, csum2, pos);
1123 if (skb_shinfo(skb)->frag_list) {
1124 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1126 for (; list; list = list->next) {
1130 BUG_TRAP(start <= offset + len);
1132 end = start + list->len;
1133 if ((copy = end - offset) > 0) {
1136 csum2 = skb_copy_and_csum_bits(list,
1139 csum = csum_block_add(csum, csum2, pos);
1140 if ((len -= copy) == 0)
1154 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1159 if (skb->ip_summed == CHECKSUM_HW)
1160 csstart = skb->h.raw - skb->data;
1162 csstart = skb_headlen(skb);
1164 if (csstart > skb_headlen(skb))
1167 memcpy(to, skb->data, csstart);
1170 if (csstart != skb->len)
1171 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1172 skb->len - csstart, 0);
1174 if (skb->ip_summed == CHECKSUM_HW) {
1175 long csstuff = csstart + skb->csum;
1177 *((unsigned short *)(to + csstuff)) = csum_fold(csum);
1182 * skb_dequeue - remove from the head of the queue
1183 * @list: list to dequeue from
1185 * Remove the head of the list. The list lock is taken so the function
1186 * may be used safely with other locking list functions. The head item is
1187 * returned or %NULL if the list is empty.
1190 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1192 unsigned long flags;
1193 struct sk_buff *result;
1195 spin_lock_irqsave(&list->lock, flags);
1196 result = __skb_dequeue(list);
1197 spin_unlock_irqrestore(&list->lock, flags);
1202 * skb_dequeue_tail - remove from the tail of the queue
1203 * @list: list to dequeue from
1205 * Remove the tail of the list. The list lock is taken so the function
1206 * may be used safely with other locking list functions. The tail item is
1207 * returned or %NULL if the list is empty.
1209 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1211 unsigned long flags;
1212 struct sk_buff *result;
1214 spin_lock_irqsave(&list->lock, flags);
1215 result = __skb_dequeue_tail(list);
1216 spin_unlock_irqrestore(&list->lock, flags);
1221 * skb_queue_purge - empty a list
1222 * @list: list to empty
1224 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1225 * the list and one reference dropped. This function takes the list
1226 * lock and is atomic with respect to other list locking functions.
1228 void skb_queue_purge(struct sk_buff_head *list)
1230 struct sk_buff *skb;
1231 while ((skb = skb_dequeue(list)) != NULL)
1236 * skb_queue_head - queue a buffer at the list head
1237 * @list: list to use
1238 * @newsk: buffer to queue
1240 * Queue a buffer at the start of the list. This function takes the
1241 * list lock and can be used safely with other locking &sk_buff functions
1244 * A buffer cannot be placed on two lists at the same time.
1246 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1248 unsigned long flags;
1250 spin_lock_irqsave(&list->lock, flags);
1251 __skb_queue_head(list, newsk);
1252 spin_unlock_irqrestore(&list->lock, flags);
1256 * skb_queue_tail - queue a buffer at the list tail
1257 * @list: list to use
1258 * @newsk: buffer to queue
1260 * Queue a buffer at the tail of the list. This function takes the
1261 * list lock and can be used safely with other locking &sk_buff functions
1264 * A buffer cannot be placed on two lists at the same time.
1266 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1268 unsigned long flags;
1270 spin_lock_irqsave(&list->lock, flags);
1271 __skb_queue_tail(list, newsk);
1272 spin_unlock_irqrestore(&list->lock, flags);
1275 * skb_unlink - remove a buffer from a list
1276 * @skb: buffer to remove
1278 * Place a packet after a given packet in a list. The list locks are taken
1279 * and this function is atomic with respect to other list locked calls
1281 * Works even without knowing the list it is sitting on, which can be
1282 * handy at times. It also means that THE LIST MUST EXIST when you
1283 * unlink. Thus a list must have its contents unlinked before it is
1286 void skb_unlink(struct sk_buff *skb)
1288 struct sk_buff_head *list = skb->list;
1291 unsigned long flags;
1293 spin_lock_irqsave(&list->lock, flags);
1294 if (skb->list == list)
1295 __skb_unlink(skb, skb->list);
1296 spin_unlock_irqrestore(&list->lock, flags);
1302 * skb_append - append a buffer
1303 * @old: buffer to insert after
1304 * @newsk: buffer to insert
1306 * Place a packet after a given packet in a list. The list locks are taken
1307 * and this function is atomic with respect to other list locked calls.
1308 * A buffer cannot be placed on two lists at the same time.
1311 void skb_append(struct sk_buff *old, struct sk_buff *newsk)
1313 unsigned long flags;
1315 spin_lock_irqsave(&old->list->lock, flags);
1316 __skb_append(old, newsk);
1317 spin_unlock_irqrestore(&old->list->lock, flags);
1322 * skb_insert - insert a buffer
1323 * @old: buffer to insert before
1324 * @newsk: buffer to insert
1326 * Place a packet before a given packet in a list. The list locks are taken
1327 * and this function is atomic with respect to other list locked calls
1328 * A buffer cannot be placed on two lists at the same time.
1331 void skb_insert(struct sk_buff *old, struct sk_buff *newsk)
1333 unsigned long flags;
1335 spin_lock_irqsave(&old->list->lock, flags);
1336 __skb_insert(newsk, old->prev, old, old->list);
1337 spin_unlock_irqrestore(&old->list->lock, flags);
1342 * Tune the memory allocator for a new MTU size.
1344 void skb_add_mtu(int mtu)
1346 /* Must match allocation in alloc_skb */
1347 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1349 kmem_add_cache_size(mtu);
1353 static void inline skb_split_inside_header(struct sk_buff *skb,
1354 struct sk_buff* skb1,
1355 const u32 len, const int pos)
1359 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1361 /* And move data appendix as is. */
1362 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1363 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1365 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1366 skb_shinfo(skb)->nr_frags = 0;
1367 skb1->data_len = skb->data_len;
1368 skb1->len += skb1->data_len;
1371 skb->tail = skb->data + len;
1374 static void inline skb_split_no_header(struct sk_buff *skb,
1375 struct sk_buff* skb1,
1376 const u32 len, int pos)
1379 const int nfrags = skb_shinfo(skb)->nr_frags;
1381 skb_shinfo(skb)->nr_frags = 0;
1382 skb1->len = skb1->data_len = skb->len - len;
1384 skb->data_len = len - pos;
1386 for (i = 0; i < nfrags; i++) {
1387 int size = skb_shinfo(skb)->frags[i].size;
1389 if (pos + size > len) {
1390 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1394 * We have to variants in this case:
1395 * 1. Move all the frag to the second
1396 * part, if it is possible. F.e.
1397 * this approach is mandatory for TUX,
1398 * where splitting is expensive.
1399 * 2. Split is accurately. We make this.
1401 get_page(skb_shinfo(skb)->frags[i].page);
1402 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1403 skb_shinfo(skb1)->frags[0].size -= len - pos;
1404 skb_shinfo(skb)->frags[i].size = len - pos;
1405 skb_shinfo(skb)->nr_frags++;
1409 skb_shinfo(skb)->nr_frags++;
1412 skb_shinfo(skb1)->nr_frags = k;
1416 * skb_split - Split fragmented skb to two parts at length len.
1418 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1420 int pos = skb_headlen(skb);
1422 if (len < pos) /* Split line is inside header. */
1423 skb_split_inside_header(skb, skb1, len, pos);
1424 else /* Second chunk has no header, nothing to copy. */
1425 skb_split_no_header(skb, skb1, len, pos);
1428 void __init skb_init(void)
1430 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1431 sizeof(struct sk_buff),
1435 if (!skbuff_head_cache)
1436 panic("cannot create skbuff cache");
1439 EXPORT_SYMBOL(___pskb_trim);
1440 EXPORT_SYMBOL(__kfree_skb);
1441 EXPORT_SYMBOL(__pskb_pull_tail);
1442 EXPORT_SYMBOL(alloc_skb);
1443 EXPORT_SYMBOL(pskb_copy);
1444 EXPORT_SYMBOL(pskb_expand_head);
1445 EXPORT_SYMBOL(skb_checksum);
1446 EXPORT_SYMBOL(skb_clone);
1447 EXPORT_SYMBOL(skb_clone_fraglist);
1448 EXPORT_SYMBOL(skb_copy);
1449 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1450 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1451 EXPORT_SYMBOL(skb_copy_bits);
1452 EXPORT_SYMBOL(skb_copy_expand);
1453 EXPORT_SYMBOL(skb_over_panic);
1454 EXPORT_SYMBOL(skb_pad);
1455 EXPORT_SYMBOL(skb_realloc_headroom);
1456 EXPORT_SYMBOL(skb_under_panic);
1457 EXPORT_SYMBOL(skb_dequeue);
1458 EXPORT_SYMBOL(skb_dequeue_tail);
1459 EXPORT_SYMBOL(skb_insert);
1460 EXPORT_SYMBOL(skb_queue_purge);
1461 EXPORT_SYMBOL(skb_queue_head);
1462 EXPORT_SYMBOL(skb_queue_tail);
1463 EXPORT_SYMBOL(skb_unlink);
1464 EXPORT_SYMBOL(skb_append);
1465 EXPORT_SYMBOL(skb_split);
1466 EXPORT_SYMBOL(skb_iter_first);
1467 EXPORT_SYMBOL(skb_iter_next);
1468 EXPORT_SYMBOL(skb_iter_abort);