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);
391 atomic_set(&n->users, 1);
397 atomic_inc(&(skb_shinfo(skb)->dataref));
403 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
406 * Shift between the two data areas in bytes
408 unsigned long offset = new->data - old->data;
413 new->real_dev = old->real_dev;
414 new->priority = old->priority;
415 new->protocol = old->protocol;
416 new->dst = dst_clone(old->dst);
418 new->sp = secpath_get(old->sp);
420 new->h.raw = old->h.raw + offset;
421 new->nh.raw = old->nh.raw + offset;
422 new->mac.raw = old->mac.raw + offset;
423 memcpy(new->cb, old->cb, sizeof(old->cb));
424 new->local_df = old->local_df;
425 new->pkt_type = old->pkt_type;
426 new->stamp = old->stamp;
427 new->destructor = NULL;
428 new->security = old->security;
429 #ifdef CONFIG_NETFILTER
430 new->nfmark = old->nfmark;
431 new->nfcache = old->nfcache;
432 new->nfct = old->nfct;
433 nf_conntrack_get(old->nfct);
434 new->nfctinfo = old->nfctinfo;
435 #ifdef CONFIG_NETFILTER_DEBUG
436 new->nf_debug = old->nf_debug;
438 #ifdef CONFIG_BRIDGE_NETFILTER
439 new->nf_bridge = old->nf_bridge;
440 nf_bridge_get(old->nf_bridge);
443 #ifdef CONFIG_NET_SCHED
444 #ifdef CONFIG_NET_CLS_ACT
445 new->tc_verd = old->tc_verd;
447 new->tc_index = old->tc_index;
449 atomic_set(&new->users, 1);
450 skb_shinfo(new)->tso_size = skb_shinfo(old)->tso_size;
451 skb_shinfo(new)->tso_segs = skb_shinfo(old)->tso_segs;
455 * skb_copy - create private copy of an sk_buff
456 * @skb: buffer to copy
457 * @gfp_mask: allocation priority
459 * Make a copy of both an &sk_buff and its data. This is used when the
460 * caller wishes to modify the data and needs a private copy of the
461 * data to alter. Returns %NULL on failure or the pointer to the buffer
462 * on success. The returned buffer has a reference count of 1.
464 * As by-product this function converts non-linear &sk_buff to linear
465 * one, so that &sk_buff becomes completely private and caller is allowed
466 * to modify all the data of returned buffer. This means that this
467 * function is not recommended for use in circumstances when only
468 * header is going to be modified. Use pskb_copy() instead.
471 struct sk_buff *skb_copy(const struct sk_buff *skb, int gfp_mask)
473 int headerlen = skb->data - skb->head;
475 * Allocate the copy buffer
477 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
482 /* Set the data pointer */
483 skb_reserve(n, headerlen);
484 /* Set the tail pointer and length */
485 skb_put(n, skb->len);
487 n->ip_summed = skb->ip_summed;
489 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
492 copy_skb_header(n, skb);
498 * pskb_copy - create copy of an sk_buff with private head.
499 * @skb: buffer to copy
500 * @gfp_mask: allocation priority
502 * Make a copy of both an &sk_buff and part of its data, located
503 * in header. Fragmented data remain shared. This is used when
504 * the caller wishes to modify only header of &sk_buff and needs
505 * private copy of the header to alter. Returns %NULL on failure
506 * or the pointer to the buffer on success.
507 * The returned buffer has a reference count of 1.
510 struct sk_buff *pskb_copy(struct sk_buff *skb, int gfp_mask)
513 * Allocate the copy buffer
515 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
520 /* Set the data pointer */
521 skb_reserve(n, skb->data - skb->head);
522 /* Set the tail pointer and length */
523 skb_put(n, skb_headlen(skb));
525 memcpy(n->data, skb->data, n->len);
527 n->ip_summed = skb->ip_summed;
529 n->data_len = skb->data_len;
532 if (skb_shinfo(skb)->nr_frags) {
535 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
536 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
537 get_page(skb_shinfo(n)->frags[i].page);
539 skb_shinfo(n)->nr_frags = i;
542 if (skb_shinfo(skb)->frag_list) {
543 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
544 skb_clone_fraglist(n);
547 copy_skb_header(n, skb);
553 * pskb_expand_head - reallocate header of &sk_buff
554 * @skb: buffer to reallocate
555 * @nhead: room to add at head
556 * @ntail: room to add at tail
557 * @gfp_mask: allocation priority
559 * Expands (or creates identical copy, if &nhead and &ntail are zero)
560 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
561 * reference count of 1. Returns zero in the case of success or error,
562 * if expansion failed. In the last case, &sk_buff is not changed.
564 * All the pointers pointing into skb header may change and must be
565 * reloaded after call to this function.
568 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, int gfp_mask)
572 int size = nhead + (skb->end - skb->head) + ntail;
578 size = SKB_DATA_ALIGN(size);
580 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
584 /* Copy only real data... and, alas, header. This should be
585 * optimized for the cases when header is void. */
586 memcpy(data + nhead, skb->head, skb->tail - skb->head);
587 memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
589 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
590 get_page(skb_shinfo(skb)->frags[i].page);
592 if (skb_shinfo(skb)->frag_list)
593 skb_clone_fraglist(skb);
595 skb_release_data(skb);
597 off = (data + nhead) - skb->head;
600 skb->end = data + size;
608 atomic_set(&skb_shinfo(skb)->dataref, 1);
615 /* Make private copy of skb with writable head and some headroom */
617 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
619 struct sk_buff *skb2;
620 int delta = headroom - skb_headroom(skb);
623 skb2 = pskb_copy(skb, GFP_ATOMIC);
625 skb2 = skb_clone(skb, GFP_ATOMIC);
626 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
637 * skb_copy_expand - copy and expand sk_buff
638 * @skb: buffer to copy
639 * @newheadroom: new free bytes at head
640 * @newtailroom: new free bytes at tail
641 * @gfp_mask: allocation priority
643 * Make a copy of both an &sk_buff and its data and while doing so
644 * allocate additional space.
646 * This is used when the caller wishes to modify the data and needs a
647 * private copy of the data to alter as well as more space for new fields.
648 * Returns %NULL on failure or the pointer to the buffer
649 * on success. The returned buffer has a reference count of 1.
651 * You must pass %GFP_ATOMIC as the allocation priority if this function
652 * is called from an interrupt.
654 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
655 * only by netfilter in the cases when checksum is recalculated? --ANK
657 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
658 int newheadroom, int newtailroom, int gfp_mask)
661 * Allocate the copy buffer
663 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
665 int head_copy_len, head_copy_off;
670 skb_reserve(n, newheadroom);
672 /* Set the tail pointer and length */
673 skb_put(n, skb->len);
675 head_copy_len = skb_headroom(skb);
677 if (newheadroom <= head_copy_len)
678 head_copy_len = newheadroom;
680 head_copy_off = newheadroom - head_copy_len;
682 /* Copy the linear header and data. */
683 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
684 skb->len + head_copy_len))
687 copy_skb_header(n, skb);
693 * skb_pad - zero pad the tail of an skb
694 * @skb: buffer to pad
697 * Ensure that a buffer is followed by a padding area that is zero
698 * filled. Used by network drivers which may DMA or transfer data
699 * beyond the buffer end onto the wire.
701 * May return NULL in out of memory cases.
704 struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
706 struct sk_buff *nskb;
708 /* If the skbuff is non linear tailroom is always zero.. */
709 if (skb_tailroom(skb) >= pad) {
710 memset(skb->data+skb->len, 0, pad);
714 nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
717 memset(nskb->data+nskb->len, 0, pad);
721 /* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
722 * If realloc==0 and trimming is impossible without change of data,
726 int ___pskb_trim(struct sk_buff *skb, unsigned int len, int realloc)
728 int offset = skb_headlen(skb);
729 int nfrags = skb_shinfo(skb)->nr_frags;
732 for (i = 0; i < nfrags; i++) {
733 int end = offset + skb_shinfo(skb)->frags[i].size;
735 if (skb_cloned(skb)) {
738 if (pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
742 put_page(skb_shinfo(skb)->frags[i].page);
743 skb_shinfo(skb)->nr_frags--;
745 skb_shinfo(skb)->frags[i].size = len - offset;
752 skb->data_len -= skb->len - len;
755 if (len <= skb_headlen(skb)) {
758 skb->tail = skb->data + len;
759 if (skb_shinfo(skb)->frag_list && !skb_cloned(skb))
760 skb_drop_fraglist(skb);
762 skb->data_len -= skb->len - len;
771 * __pskb_pull_tail - advance tail of skb header
772 * @skb: buffer to reallocate
773 * @delta: number of bytes to advance tail
775 * The function makes a sense only on a fragmented &sk_buff,
776 * it expands header moving its tail forward and copying necessary
777 * data from fragmented part.
779 * &sk_buff MUST have reference count of 1.
781 * Returns %NULL (and &sk_buff does not change) if pull failed
782 * or value of new tail of skb in the case of success.
784 * All the pointers pointing into skb header may change and must be
785 * reloaded after call to this function.
788 /* Moves tail of skb head forward, copying data from fragmented part,
789 * when it is necessary.
790 * 1. It may fail due to malloc failure.
791 * 2. It may change skb pointers.
793 * It is pretty complicated. Luckily, it is called only in exceptional cases.
795 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
797 /* If skb has not enough free space at tail, get new one
798 * plus 128 bytes for future expansions. If we have enough
799 * room at tail, reallocate without expansion only if skb is cloned.
801 int i, k, eat = (skb->tail + delta) - skb->end;
803 if (eat > 0 || skb_cloned(skb)) {
804 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
809 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
812 /* Optimization: no fragments, no reasons to preestimate
813 * size of pulled pages. Superb.
815 if (!skb_shinfo(skb)->frag_list)
818 /* Estimate size of pulled pages. */
820 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
821 if (skb_shinfo(skb)->frags[i].size >= eat)
823 eat -= skb_shinfo(skb)->frags[i].size;
826 /* If we need update frag list, we are in troubles.
827 * Certainly, it possible to add an offset to skb data,
828 * but taking into account that pulling is expected to
829 * be very rare operation, it is worth to fight against
830 * further bloating skb head and crucify ourselves here instead.
831 * Pure masohism, indeed. 8)8)
834 struct sk_buff *list = skb_shinfo(skb)->frag_list;
835 struct sk_buff *clone = NULL;
836 struct sk_buff *insp = NULL;
842 if (list->len <= eat) {
843 /* Eaten as whole. */
848 /* Eaten partially. */
850 if (skb_shared(list)) {
851 /* Sucks! We need to fork list. :-( */
852 clone = skb_clone(list, GFP_ATOMIC);
858 /* This may be pulled without
862 if (!pskb_pull(list, eat)) {
871 /* Free pulled out fragments. */
872 while ((list = skb_shinfo(skb)->frag_list) != insp) {
873 skb_shinfo(skb)->frag_list = list->next;
876 /* And insert new clone at head. */
879 skb_shinfo(skb)->frag_list = clone;
882 /* Success! Now we may commit changes to skb data. */
887 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
888 if (skb_shinfo(skb)->frags[i].size <= eat) {
889 put_page(skb_shinfo(skb)->frags[i].page);
890 eat -= skb_shinfo(skb)->frags[i].size;
892 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
894 skb_shinfo(skb)->frags[k].page_offset += eat;
895 skb_shinfo(skb)->frags[k].size -= eat;
901 skb_shinfo(skb)->nr_frags = k;
904 skb->data_len -= delta;
909 /* Copy some data bits from skb to kernel buffer. */
911 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
914 int start = skb_headlen(skb);
916 if (offset > (int)skb->len - len)
920 if ((copy = start - offset) > 0) {
923 memcpy(to, skb->data + offset, copy);
924 if ((len -= copy) == 0)
930 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
933 BUG_TRAP(start <= offset + len);
935 end = start + skb_shinfo(skb)->frags[i].size;
936 if ((copy = end - offset) > 0) {
942 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
944 vaddr + skb_shinfo(skb)->frags[i].page_offset+
945 offset - start, copy);
946 kunmap_skb_frag(vaddr);
948 if ((len -= copy) == 0)
956 if (skb_shinfo(skb)->frag_list) {
957 struct sk_buff *list = skb_shinfo(skb)->frag_list;
959 for (; list; list = list->next) {
962 BUG_TRAP(start <= offset + len);
964 end = start + list->len;
965 if ((copy = end - offset) > 0) {
968 if (skb_copy_bits(list, offset - start,
971 if ((len -= copy) == 0)
987 * skb_store_bits - store bits from kernel buffer to skb
988 * @skb: destination buffer
989 * @offset: offset in destination
990 * @from: source buffer
991 * @len: number of bytes to copy
993 * Copy the specified number of bytes from the source buffer to the
994 * destination skb. This function handles all the messy bits of
995 * traversing fragment lists and such.
998 int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
1001 int start = skb_headlen(skb);
1003 if (offset > (int)skb->len - len)
1006 if ((copy = start - offset) > 0) {
1009 memcpy(skb->data + offset, from, copy);
1010 if ((len -= copy) == 0)
1016 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1017 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1020 BUG_TRAP(start <= offset + len);
1022 end = start + frag->size;
1023 if ((copy = end - offset) > 0) {
1029 vaddr = kmap_skb_frag(frag);
1030 memcpy(vaddr + frag->page_offset + offset - start,
1032 kunmap_skb_frag(vaddr);
1034 if ((len -= copy) == 0)
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) {
1054 if (skb_store_bits(list, offset - start,
1057 if ((len -= copy) == 0)
1072 EXPORT_SYMBOL(skb_store_bits);
1074 /* Checksum skb data. */
1076 unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1077 int len, unsigned int csum)
1079 int start = skb_headlen(skb);
1080 int i, copy = start - offset;
1083 /* Checksum header. */
1087 csum = csum_partial(skb->data + offset, copy, csum);
1088 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(vaddr + frag->page_offset +
1109 offset - start, copy, 0);
1110 kunmap_skb_frag(vaddr);
1111 csum = csum_block_add(csum, csum2, pos);
1120 if (skb_shinfo(skb)->frag_list) {
1121 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1123 for (; list; list = list->next) {
1126 BUG_TRAP(start <= offset + len);
1128 end = start + list->len;
1129 if ((copy = end - offset) > 0) {
1133 csum2 = skb_checksum(list, offset - start,
1135 csum = csum_block_add(csum, csum2, pos);
1136 if ((len -= copy) == 0)
1150 /* Both of above in one bottle. */
1152 unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1153 u8 *to, int len, unsigned int csum)
1155 int start = skb_headlen(skb);
1156 int i, copy = start - offset;
1163 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1165 if ((len -= copy) == 0)
1172 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1175 BUG_TRAP(start <= offset + len);
1177 end = start + skb_shinfo(skb)->frags[i].size;
1178 if ((copy = end - offset) > 0) {
1181 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1185 vaddr = kmap_skb_frag(frag);
1186 csum2 = csum_partial_copy_nocheck(vaddr +
1190 kunmap_skb_frag(vaddr);
1191 csum = csum_block_add(csum, csum2, pos);
1201 if (skb_shinfo(skb)->frag_list) {
1202 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1204 for (; list; list = list->next) {
1208 BUG_TRAP(start <= offset + len);
1210 end = start + list->len;
1211 if ((copy = end - offset) > 0) {
1214 csum2 = skb_copy_and_csum_bits(list,
1217 csum = csum_block_add(csum, csum2, pos);
1218 if ((len -= copy) == 0)
1232 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1237 if (skb->ip_summed == CHECKSUM_HW)
1238 csstart = skb->h.raw - skb->data;
1240 csstart = skb_headlen(skb);
1242 if (csstart > skb_headlen(skb))
1245 memcpy(to, skb->data, csstart);
1248 if (csstart != skb->len)
1249 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1250 skb->len - csstart, 0);
1252 if (skb->ip_summed == CHECKSUM_HW) {
1253 long csstuff = csstart + skb->csum;
1255 *((unsigned short *)(to + csstuff)) = csum_fold(csum);
1260 * skb_dequeue - remove from the head of the queue
1261 * @list: list to dequeue from
1263 * Remove the head of the list. The list lock is taken so the function
1264 * may be used safely with other locking list functions. The head item is
1265 * returned or %NULL if the list is empty.
1268 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1270 unsigned long flags;
1271 struct sk_buff *result;
1273 spin_lock_irqsave(&list->lock, flags);
1274 result = __skb_dequeue(list);
1275 spin_unlock_irqrestore(&list->lock, flags);
1280 * skb_dequeue_tail - remove from the tail of the queue
1281 * @list: list to dequeue from
1283 * Remove the tail of the list. The list lock is taken so the function
1284 * may be used safely with other locking list functions. The tail item is
1285 * returned or %NULL if the list is empty.
1287 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1289 unsigned long flags;
1290 struct sk_buff *result;
1292 spin_lock_irqsave(&list->lock, flags);
1293 result = __skb_dequeue_tail(list);
1294 spin_unlock_irqrestore(&list->lock, flags);
1299 * skb_queue_purge - empty a list
1300 * @list: list to empty
1302 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1303 * the list and one reference dropped. This function takes the list
1304 * lock and is atomic with respect to other list locking functions.
1306 void skb_queue_purge(struct sk_buff_head *list)
1308 struct sk_buff *skb;
1309 while ((skb = skb_dequeue(list)) != NULL)
1314 * skb_queue_head - queue a buffer at the list head
1315 * @list: list to use
1316 * @newsk: buffer to queue
1318 * Queue a buffer at the start of the list. This function takes the
1319 * list lock and can be used safely with other locking &sk_buff functions
1322 * A buffer cannot be placed on two lists at the same time.
1324 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1326 unsigned long flags;
1328 spin_lock_irqsave(&list->lock, flags);
1329 __skb_queue_head(list, newsk);
1330 spin_unlock_irqrestore(&list->lock, flags);
1334 * skb_queue_tail - queue a buffer at the list tail
1335 * @list: list to use
1336 * @newsk: buffer to queue
1338 * Queue a buffer at the tail of the list. This function takes the
1339 * list lock and can be used safely with other locking &sk_buff functions
1342 * A buffer cannot be placed on two lists at the same time.
1344 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1346 unsigned long flags;
1348 spin_lock_irqsave(&list->lock, flags);
1349 __skb_queue_tail(list, newsk);
1350 spin_unlock_irqrestore(&list->lock, flags);
1353 * skb_unlink - remove a buffer from a list
1354 * @skb: buffer to remove
1356 * Place a packet after a given packet in a list. The list locks are taken
1357 * and this function is atomic with respect to other list locked calls
1359 * Works even without knowing the list it is sitting on, which can be
1360 * handy at times. It also means that THE LIST MUST EXIST when you
1361 * unlink. Thus a list must have its contents unlinked before it is
1364 void skb_unlink(struct sk_buff *skb)
1366 struct sk_buff_head *list = skb->list;
1369 unsigned long flags;
1371 spin_lock_irqsave(&list->lock, flags);
1372 if (skb->list == list)
1373 __skb_unlink(skb, skb->list);
1374 spin_unlock_irqrestore(&list->lock, flags);
1380 * skb_append - append a buffer
1381 * @old: buffer to insert after
1382 * @newsk: buffer to insert
1384 * Place a packet after a given packet in a list. The list locks are taken
1385 * and this function is atomic with respect to other list locked calls.
1386 * A buffer cannot be placed on two lists at the same time.
1389 void skb_append(struct sk_buff *old, struct sk_buff *newsk)
1391 unsigned long flags;
1393 spin_lock_irqsave(&old->list->lock, flags);
1394 __skb_append(old, newsk);
1395 spin_unlock_irqrestore(&old->list->lock, flags);
1400 * skb_insert - insert a buffer
1401 * @old: buffer to insert before
1402 * @newsk: buffer to insert
1404 * Place a packet before a given packet in a list. The list locks are taken
1405 * and this function is atomic with respect to other list locked calls
1406 * A buffer cannot be placed on two lists at the same time.
1409 void skb_insert(struct sk_buff *old, struct sk_buff *newsk)
1411 unsigned long flags;
1413 spin_lock_irqsave(&old->list->lock, flags);
1414 __skb_insert(newsk, old->prev, old, old->list);
1415 spin_unlock_irqrestore(&old->list->lock, flags);
1420 * Tune the memory allocator for a new MTU size.
1422 void skb_add_mtu(int mtu)
1424 /* Must match allocation in alloc_skb */
1425 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1427 kmem_add_cache_size(mtu);
1431 static inline void skb_split_inside_header(struct sk_buff *skb,
1432 struct sk_buff* skb1,
1433 const u32 len, const int pos)
1437 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1439 /* And move data appendix as is. */
1440 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1441 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1443 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1444 skb_shinfo(skb)->nr_frags = 0;
1445 skb1->data_len = skb->data_len;
1446 skb1->len += skb1->data_len;
1449 skb->tail = skb->data + len;
1452 static inline void skb_split_no_header(struct sk_buff *skb,
1453 struct sk_buff* skb1,
1454 const u32 len, int pos)
1457 const int nfrags = skb_shinfo(skb)->nr_frags;
1459 skb_shinfo(skb)->nr_frags = 0;
1460 skb1->len = skb1->data_len = skb->len - len;
1462 skb->data_len = len - pos;
1464 for (i = 0; i < nfrags; i++) {
1465 int size = skb_shinfo(skb)->frags[i].size;
1467 if (pos + size > len) {
1468 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1472 * We have two variants in this case:
1473 * 1. Move all the frag to the second
1474 * part, if it is possible. F.e.
1475 * this approach is mandatory for TUX,
1476 * where splitting is expensive.
1477 * 2. Split is accurately. We make this.
1479 get_page(skb_shinfo(skb)->frags[i].page);
1480 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1481 skb_shinfo(skb1)->frags[0].size -= len - pos;
1482 skb_shinfo(skb)->frags[i].size = len - pos;
1483 skb_shinfo(skb)->nr_frags++;
1487 skb_shinfo(skb)->nr_frags++;
1490 skb_shinfo(skb1)->nr_frags = k;
1494 * skb_split - Split fragmented skb to two parts at length len.
1495 * @skb: the buffer to split
1496 * @skb1: the buffer to receive the second part
1497 * @len: new length for skb
1499 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1501 int pos = skb_headlen(skb);
1503 if (len < pos) /* Split line is inside header. */
1504 skb_split_inside_header(skb, skb1, len, pos);
1505 else /* Second chunk has no header, nothing to copy. */
1506 skb_split_no_header(skb, skb1, len, pos);
1509 void __init skb_init(void)
1511 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1512 sizeof(struct sk_buff),
1516 if (!skbuff_head_cache)
1517 panic("cannot create skbuff cache");
1520 EXPORT_SYMBOL(___pskb_trim);
1521 EXPORT_SYMBOL(__kfree_skb);
1522 EXPORT_SYMBOL(__pskb_pull_tail);
1523 EXPORT_SYMBOL(alloc_skb);
1524 EXPORT_SYMBOL(pskb_copy);
1525 EXPORT_SYMBOL(pskb_expand_head);
1526 EXPORT_SYMBOL(skb_checksum);
1527 EXPORT_SYMBOL(skb_clone);
1528 EXPORT_SYMBOL(skb_clone_fraglist);
1529 EXPORT_SYMBOL(skb_copy);
1530 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1531 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1532 EXPORT_SYMBOL(skb_copy_bits);
1533 EXPORT_SYMBOL(skb_copy_expand);
1534 EXPORT_SYMBOL(skb_over_panic);
1535 EXPORT_SYMBOL(skb_pad);
1536 EXPORT_SYMBOL(skb_realloc_headroom);
1537 EXPORT_SYMBOL(skb_under_panic);
1538 EXPORT_SYMBOL(skb_dequeue);
1539 EXPORT_SYMBOL(skb_dequeue_tail);
1540 EXPORT_SYMBOL(skb_insert);
1541 EXPORT_SYMBOL(skb_queue_purge);
1542 EXPORT_SYMBOL(skb_queue_head);
1543 EXPORT_SYMBOL(skb_queue_tail);
1544 EXPORT_SYMBOL(skb_unlink);
1545 EXPORT_SYMBOL(skb_append);
1546 EXPORT_SYMBOL(skb_split);