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 __read_mostly;
72 static kmem_cache_t *skbuff_fclone_cache __read_mostly;
75 * Keep out-of-line to prevent kernel bloat.
76 * __builtin_return_address is not used because it is not always
81 * skb_over_panic - private function
86 * Out of line support code for skb_put(). Not user callable.
88 void skb_over_panic(struct sk_buff *skb, int sz, void *here)
90 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
91 "data:%p tail:%p end:%p dev:%s\n",
92 here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
93 skb->dev ? skb->dev->name : "<NULL>");
98 * skb_under_panic - private function
103 * Out of line support code for skb_push(). Not user callable.
106 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
108 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
109 "data:%p tail:%p end:%p dev:%s\n",
110 here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
111 skb->dev ? skb->dev->name : "<NULL>");
115 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
116 * 'private' fields and also do memory statistics to find all the
122 * __alloc_skb - allocate a network buffer
123 * @size: size to allocate
124 * @gfp_mask: allocation mask
125 * @fclone: allocate from fclone cache instead of head cache
126 * and allocate a cloned (child) skb
128 * Allocate a new &sk_buff. The returned buffer has no headroom and a
129 * tail room of size bytes. The object has a reference count of one.
130 * The return is the buffer. On a failure the return is %NULL.
132 * Buffers may only be allocated from interrupts using a @gfp_mask of
135 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
139 struct skb_shared_info *shinfo;
143 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
146 skb = kmem_cache_alloc(cache, gfp_mask & ~__GFP_DMA);
150 /* Get the DATA. Size must match skb_add_mtu(). */
151 size = SKB_DATA_ALIGN(size);
152 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
156 memset(skb, 0, offsetof(struct sk_buff, truesize));
157 skb->truesize = size + sizeof(struct sk_buff);
158 atomic_set(&skb->users, 1);
162 skb->end = data + size;
163 /* make sure we initialize shinfo sequentially */
164 shinfo = skb_shinfo(skb);
165 atomic_set(&shinfo->dataref, 1);
166 shinfo->nr_frags = 0;
167 shinfo->tso_size = 0;
168 shinfo->tso_segs = 0;
169 shinfo->ufo_size = 0;
170 shinfo->ip6_frag_id = 0;
171 shinfo->frag_list = NULL;
174 struct sk_buff *child = skb + 1;
175 atomic_t *fclone_ref = (atomic_t *) (child + 1);
177 skb->fclone = SKB_FCLONE_ORIG;
178 atomic_set(fclone_ref, 1);
180 child->fclone = SKB_FCLONE_UNAVAILABLE;
185 kmem_cache_free(cache, skb);
191 * alloc_skb_from_cache - allocate a network buffer
192 * @cp: kmem_cache from which to allocate the data area
193 * (object size must be big enough for @size bytes + skb overheads)
194 * @size: size to allocate
195 * @gfp_mask: allocation mask
197 * Allocate a new &sk_buff. The returned buffer has no headroom and
198 * tail room of size bytes. The object has a reference count of one.
199 * The return is the buffer. On a failure the return is %NULL.
201 * Buffers may only be allocated from interrupts using a @gfp_mask of
204 struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp,
212 skb = kmem_cache_alloc(skbuff_head_cache,
213 gfp_mask & ~__GFP_DMA);
218 size = SKB_DATA_ALIGN(size);
219 data = kmem_cache_alloc(cp, gfp_mask);
223 memset(skb, 0, offsetof(struct sk_buff, truesize));
224 skb->truesize = size + sizeof(struct sk_buff);
225 atomic_set(&skb->users, 1);
229 skb->end = data + size;
231 atomic_set(&(skb_shinfo(skb)->dataref), 1);
232 skb_shinfo(skb)->nr_frags = 0;
233 skb_shinfo(skb)->tso_size = 0;
234 skb_shinfo(skb)->tso_segs = 0;
235 skb_shinfo(skb)->ufo_size = 0;
236 skb_shinfo(skb)->frag_list = NULL;
240 kmem_cache_free(skbuff_head_cache, skb);
246 static void skb_drop_list(struct sk_buff **listp)
248 struct sk_buff *list = *listp;
253 struct sk_buff *this = list;
259 static inline void skb_drop_fraglist(struct sk_buff *skb)
261 skb_drop_list(&skb_shinfo(skb)->frag_list);
264 static void skb_clone_fraglist(struct sk_buff *skb)
266 struct sk_buff *list;
268 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
272 void skb_release_data(struct sk_buff *skb)
275 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
276 &skb_shinfo(skb)->dataref)) {
277 if (skb_shinfo(skb)->nr_frags) {
279 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
280 put_page(skb_shinfo(skb)->frags[i].page);
283 if (skb_shinfo(skb)->frag_list)
284 skb_drop_fraglist(skb);
291 * Free an skbuff by memory without cleaning the state.
293 void kfree_skbmem(struct sk_buff *skb)
295 struct sk_buff *other;
296 atomic_t *fclone_ref;
298 skb_release_data(skb);
299 switch (skb->fclone) {
300 case SKB_FCLONE_UNAVAILABLE:
301 kmem_cache_free(skbuff_head_cache, skb);
304 case SKB_FCLONE_ORIG:
305 fclone_ref = (atomic_t *) (skb + 2);
306 if (atomic_dec_and_test(fclone_ref))
307 kmem_cache_free(skbuff_fclone_cache, skb);
310 case SKB_FCLONE_CLONE:
311 fclone_ref = (atomic_t *) (skb + 1);
314 /* The clone portion is available for
315 * fast-cloning again.
317 skb->fclone = SKB_FCLONE_UNAVAILABLE;
319 if (atomic_dec_and_test(fclone_ref))
320 kmem_cache_free(skbuff_fclone_cache, other);
326 * __kfree_skb - private function
329 * Free an sk_buff. Release anything attached to the buffer.
330 * Clean the state. This is an internal helper function. Users should
331 * always call kfree_skb
334 void __kfree_skb(struct sk_buff *skb)
336 dst_release(skb->dst);
338 secpath_put(skb->sp);
340 if (skb->destructor) {
342 skb->destructor(skb);
344 #ifdef CONFIG_NETFILTER
345 nf_conntrack_put(skb->nfct);
346 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
347 nf_conntrack_put_reasm(skb->nfct_reasm);
349 #ifdef CONFIG_BRIDGE_NETFILTER
350 nf_bridge_put(skb->nf_bridge);
353 /* XXX: IS this still necessary? - JHS */
354 #ifdef CONFIG_NET_SCHED
356 #ifdef CONFIG_NET_CLS_ACT
365 * skb_clone - duplicate an sk_buff
366 * @skb: buffer to clone
367 * @gfp_mask: allocation priority
369 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
370 * copies share the same packet data but not structure. The new
371 * buffer has a reference count of 1. If the allocation fails the
372 * function returns %NULL otherwise the new buffer is returned.
374 * If this function is called from an interrupt gfp_mask() must be
378 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
383 if (skb->fclone == SKB_FCLONE_ORIG &&
384 n->fclone == SKB_FCLONE_UNAVAILABLE) {
385 atomic_t *fclone_ref = (atomic_t *) (n + 1);
386 n->fclone = SKB_FCLONE_CLONE;
387 atomic_inc(fclone_ref);
389 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
392 n->fclone = SKB_FCLONE_UNAVAILABLE;
395 #define C(x) n->x = skb->x
397 n->next = n->prev = NULL;
408 secpath_get(skb->sp);
410 memcpy(n->cb, skb->cb, sizeof(skb->cb));
420 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
424 n->destructor = NULL;
425 #ifdef CONFIG_NETFILTER
428 nf_conntrack_get(skb->nfct);
430 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
432 nf_conntrack_get_reasm(skb->nfct_reasm);
434 #ifdef CONFIG_BRIDGE_NETFILTER
436 nf_bridge_get(skb->nf_bridge);
438 #endif /*CONFIG_NETFILTER*/
439 #ifdef CONFIG_NET_SCHED
441 #ifdef CONFIG_NET_CLS_ACT
442 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
443 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
444 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
450 atomic_set(&n->users, 1);
456 atomic_inc(&(skb_shinfo(skb)->dataref));
462 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
465 * Shift between the two data areas in bytes
467 unsigned long offset = new->data - old->data;
471 new->priority = old->priority;
472 new->protocol = old->protocol;
473 new->dst = dst_clone(old->dst);
475 new->sp = secpath_get(old->sp);
477 new->h.raw = old->h.raw + offset;
478 new->nh.raw = old->nh.raw + offset;
479 new->mac.raw = old->mac.raw + offset;
480 memcpy(new->cb, old->cb, sizeof(old->cb));
481 new->local_df = old->local_df;
482 new->fclone = SKB_FCLONE_UNAVAILABLE;
483 new->pkt_type = old->pkt_type;
484 new->tstamp = old->tstamp;
485 new->destructor = NULL;
486 #ifdef CONFIG_NETFILTER
487 new->nfmark = old->nfmark;
488 new->nfct = old->nfct;
489 nf_conntrack_get(old->nfct);
490 new->nfctinfo = old->nfctinfo;
491 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
492 new->nfct_reasm = old->nfct_reasm;
493 nf_conntrack_get_reasm(old->nfct_reasm);
495 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
496 new->ipvs_property = old->ipvs_property;
498 #ifdef CONFIG_BRIDGE_NETFILTER
499 new->nf_bridge = old->nf_bridge;
500 nf_bridge_get(old->nf_bridge);
503 #ifdef CONFIG_NET_SCHED
504 #ifdef CONFIG_NET_CLS_ACT
505 new->tc_verd = old->tc_verd;
507 new->tc_index = old->tc_index;
509 atomic_set(&new->users, 1);
510 skb_shinfo(new)->tso_size = skb_shinfo(old)->tso_size;
511 skb_shinfo(new)->tso_segs = skb_shinfo(old)->tso_segs;
512 skb_shinfo(new)->ufo_size = skb_shinfo(old)->ufo_size;
516 * skb_copy - create private copy of an sk_buff
517 * @skb: buffer to copy
518 * @gfp_mask: allocation priority
520 * Make a copy of both an &sk_buff and its data. This is used when the
521 * caller wishes to modify the data and needs a private copy of the
522 * data to alter. Returns %NULL on failure or the pointer to the buffer
523 * on success. The returned buffer has a reference count of 1.
525 * As by-product this function converts non-linear &sk_buff to linear
526 * one, so that &sk_buff becomes completely private and caller is allowed
527 * to modify all the data of returned buffer. This means that this
528 * function is not recommended for use in circumstances when only
529 * header is going to be modified. Use pskb_copy() instead.
532 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
534 int headerlen = skb->data - skb->head;
536 * Allocate the copy buffer
538 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
543 /* Set the data pointer */
544 skb_reserve(n, headerlen);
545 /* Set the tail pointer and length */
546 skb_put(n, skb->len);
548 n->ip_summed = skb->ip_summed;
550 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
553 copy_skb_header(n, skb);
559 * pskb_copy - create copy of an sk_buff with private head.
560 * @skb: buffer to copy
561 * @gfp_mask: allocation priority
563 * Make a copy of both an &sk_buff and part of its data, located
564 * in header. Fragmented data remain shared. This is used when
565 * the caller wishes to modify only header of &sk_buff and needs
566 * private copy of the header to alter. Returns %NULL on failure
567 * or the pointer to the buffer on success.
568 * The returned buffer has a reference count of 1.
571 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
574 * Allocate the copy buffer
576 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
581 /* Set the data pointer */
582 skb_reserve(n, skb->data - skb->head);
583 /* Set the tail pointer and length */
584 skb_put(n, skb_headlen(skb));
586 memcpy(n->data, skb->data, n->len);
588 n->ip_summed = skb->ip_summed;
590 n->truesize += skb->data_len;
591 n->data_len = skb->data_len;
594 if (skb_shinfo(skb)->nr_frags) {
597 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
598 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
599 get_page(skb_shinfo(n)->frags[i].page);
601 skb_shinfo(n)->nr_frags = i;
604 if (skb_shinfo(skb)->frag_list) {
605 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
606 skb_clone_fraglist(n);
609 copy_skb_header(n, skb);
615 * pskb_expand_head - reallocate header of &sk_buff
616 * @skb: buffer to reallocate
617 * @nhead: room to add at head
618 * @ntail: room to add at tail
619 * @gfp_mask: allocation priority
621 * Expands (or creates identical copy, if &nhead and &ntail are zero)
622 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
623 * reference count of 1. Returns zero in the case of success or error,
624 * if expansion failed. In the last case, &sk_buff is not changed.
626 * All the pointers pointing into skb header may change and must be
627 * reloaded after call to this function.
630 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
635 int size = nhead + (skb->end - skb->head) + ntail;
641 size = SKB_DATA_ALIGN(size);
643 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
647 /* Copy only real data... and, alas, header. This should be
648 * optimized for the cases when header is void. */
649 memcpy(data + nhead, skb->head, skb->tail - skb->head);
650 memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
652 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
653 get_page(skb_shinfo(skb)->frags[i].page);
655 if (skb_shinfo(skb)->frag_list)
656 skb_clone_fraglist(skb);
658 skb_release_data(skb);
660 off = (data + nhead) - skb->head;
663 skb->end = data + size;
671 atomic_set(&skb_shinfo(skb)->dataref, 1);
678 /* Make private copy of skb with writable head and some headroom */
680 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
682 struct sk_buff *skb2;
683 int delta = headroom - skb_headroom(skb);
686 skb2 = pskb_copy(skb, GFP_ATOMIC);
688 skb2 = skb_clone(skb, GFP_ATOMIC);
689 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
700 * skb_copy_expand - copy and expand sk_buff
701 * @skb: buffer to copy
702 * @newheadroom: new free bytes at head
703 * @newtailroom: new free bytes at tail
704 * @gfp_mask: allocation priority
706 * Make a copy of both an &sk_buff and its data and while doing so
707 * allocate additional space.
709 * This is used when the caller wishes to modify the data and needs a
710 * private copy of the data to alter as well as more space for new fields.
711 * Returns %NULL on failure or the pointer to the buffer
712 * on success. The returned buffer has a reference count of 1.
714 * You must pass %GFP_ATOMIC as the allocation priority if this function
715 * is called from an interrupt.
717 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
718 * only by netfilter in the cases when checksum is recalculated? --ANK
720 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
721 int newheadroom, int newtailroom,
725 * Allocate the copy buffer
727 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
729 int head_copy_len, head_copy_off;
734 skb_reserve(n, newheadroom);
736 /* Set the tail pointer and length */
737 skb_put(n, skb->len);
739 head_copy_len = skb_headroom(skb);
741 if (newheadroom <= head_copy_len)
742 head_copy_len = newheadroom;
744 head_copy_off = newheadroom - head_copy_len;
746 /* Copy the linear header and data. */
747 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
748 skb->len + head_copy_len))
751 copy_skb_header(n, skb);
757 * skb_pad - zero pad the tail of an skb
758 * @skb: buffer to pad
761 * Ensure that a buffer is followed by a padding area that is zero
762 * filled. Used by network drivers which may DMA or transfer data
763 * beyond the buffer end onto the wire.
765 * May return NULL in out of memory cases.
768 struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
770 struct sk_buff *nskb;
772 /* If the skbuff is non linear tailroom is always zero.. */
773 if (skb_tailroom(skb) >= pad) {
774 memset(skb->data+skb->len, 0, pad);
778 nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
781 memset(nskb->data+nskb->len, 0, pad);
785 /* Trims skb to length len. It can change skb pointers.
788 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
790 struct sk_buff **fragp;
791 struct sk_buff *frag;
792 int offset = skb_headlen(skb);
793 int nfrags = skb_shinfo(skb)->nr_frags;
797 if (skb_cloned(skb) &&
798 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
805 for (; i < nfrags; i++) {
806 int end = offset + skb_shinfo(skb)->frags[i].size;
813 skb_shinfo(skb)->frags[i++].size = len - offset;
816 skb_shinfo(skb)->nr_frags = i;
818 for (; i < nfrags; i++)
819 put_page(skb_shinfo(skb)->frags[i].page);
821 if (skb_shinfo(skb)->frag_list)
822 skb_drop_fraglist(skb);
826 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
827 fragp = &frag->next) {
828 int end = offset + frag->len;
830 if (skb_shared(frag)) {
831 struct sk_buff *nfrag;
833 nfrag = skb_clone(frag, GFP_ATOMIC);
834 if (unlikely(!nfrag))
837 nfrag->next = frag->next;
849 unlikely((err = pskb_trim(frag, len - offset))))
853 skb_drop_list(&frag->next);
858 if (len > skb_headlen(skb)) {
859 skb->data_len -= skb->len - len;
864 skb->tail = skb->data + len;
871 * __pskb_pull_tail - advance tail of skb header
872 * @skb: buffer to reallocate
873 * @delta: number of bytes to advance tail
875 * The function makes a sense only on a fragmented &sk_buff,
876 * it expands header moving its tail forward and copying necessary
877 * data from fragmented part.
879 * &sk_buff MUST have reference count of 1.
881 * Returns %NULL (and &sk_buff does not change) if pull failed
882 * or value of new tail of skb in the case of success.
884 * All the pointers pointing into skb header may change and must be
885 * reloaded after call to this function.
888 /* Moves tail of skb head forward, copying data from fragmented part,
889 * when it is necessary.
890 * 1. It may fail due to malloc failure.
891 * 2. It may change skb pointers.
893 * It is pretty complicated. Luckily, it is called only in exceptional cases.
895 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
897 /* If skb has not enough free space at tail, get new one
898 * plus 128 bytes for future expansions. If we have enough
899 * room at tail, reallocate without expansion only if skb is cloned.
901 int i, k, eat = (skb->tail + delta) - skb->end;
903 if (eat > 0 || skb_cloned(skb)) {
904 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
909 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
912 /* Optimization: no fragments, no reasons to preestimate
913 * size of pulled pages. Superb.
915 if (!skb_shinfo(skb)->frag_list)
918 /* Estimate size of pulled pages. */
920 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
921 if (skb_shinfo(skb)->frags[i].size >= eat)
923 eat -= skb_shinfo(skb)->frags[i].size;
926 /* If we need update frag list, we are in troubles.
927 * Certainly, it possible to add an offset to skb data,
928 * but taking into account that pulling is expected to
929 * be very rare operation, it is worth to fight against
930 * further bloating skb head and crucify ourselves here instead.
931 * Pure masohism, indeed. 8)8)
934 struct sk_buff *list = skb_shinfo(skb)->frag_list;
935 struct sk_buff *clone = NULL;
936 struct sk_buff *insp = NULL;
941 if (list->len <= eat) {
942 /* Eaten as whole. */
947 /* Eaten partially. */
949 if (skb_shared(list)) {
950 /* Sucks! We need to fork list. :-( */
951 clone = skb_clone(list, GFP_ATOMIC);
957 /* This may be pulled without
961 if (!pskb_pull(list, eat)) {
970 /* Free pulled out fragments. */
971 while ((list = skb_shinfo(skb)->frag_list) != insp) {
972 skb_shinfo(skb)->frag_list = list->next;
975 /* And insert new clone at head. */
978 skb_shinfo(skb)->frag_list = clone;
981 /* Success! Now we may commit changes to skb data. */
986 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
987 if (skb_shinfo(skb)->frags[i].size <= eat) {
988 put_page(skb_shinfo(skb)->frags[i].page);
989 eat -= skb_shinfo(skb)->frags[i].size;
991 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
993 skb_shinfo(skb)->frags[k].page_offset += eat;
994 skb_shinfo(skb)->frags[k].size -= eat;
1000 skb_shinfo(skb)->nr_frags = k;
1003 skb->data_len -= delta;
1008 /* Copy some data bits from skb to kernel buffer. */
1010 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1013 int start = skb_headlen(skb);
1015 if (offset > (int)skb->len - len)
1019 if ((copy = start - offset) > 0) {
1022 memcpy(to, skb->data + offset, copy);
1023 if ((len -= copy) == 0)
1029 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1032 BUG_TRAP(start <= offset + len);
1034 end = start + skb_shinfo(skb)->frags[i].size;
1035 if ((copy = end - offset) > 0) {
1041 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1043 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1044 offset - start, copy);
1045 kunmap_skb_frag(vaddr);
1047 if ((len -= copy) == 0)
1055 if (skb_shinfo(skb)->frag_list) {
1056 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1058 for (; list; list = list->next) {
1061 BUG_TRAP(start <= offset + len);
1063 end = start + list->len;
1064 if ((copy = end - offset) > 0) {
1067 if (skb_copy_bits(list, offset - start,
1070 if ((len -= copy) == 0)
1086 * skb_store_bits - store bits from kernel buffer to skb
1087 * @skb: destination buffer
1088 * @offset: offset in destination
1089 * @from: source buffer
1090 * @len: number of bytes to copy
1092 * Copy the specified number of bytes from the source buffer to the
1093 * destination skb. This function handles all the messy bits of
1094 * traversing fragment lists and such.
1097 int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
1100 int start = skb_headlen(skb);
1102 if (offset > (int)skb->len - len)
1105 if ((copy = start - offset) > 0) {
1108 memcpy(skb->data + offset, from, copy);
1109 if ((len -= copy) == 0)
1115 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1116 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1119 BUG_TRAP(start <= offset + len);
1121 end = start + frag->size;
1122 if ((copy = end - offset) > 0) {
1128 vaddr = kmap_skb_frag(frag);
1129 memcpy(vaddr + frag->page_offset + offset - start,
1131 kunmap_skb_frag(vaddr);
1133 if ((len -= copy) == 0)
1141 if (skb_shinfo(skb)->frag_list) {
1142 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1144 for (; list; list = list->next) {
1147 BUG_TRAP(start <= offset + len);
1149 end = start + list->len;
1150 if ((copy = end - offset) > 0) {
1153 if (skb_store_bits(list, offset - start,
1156 if ((len -= copy) == 0)
1171 EXPORT_SYMBOL(skb_store_bits);
1173 /* Checksum skb data. */
1175 unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1176 int len, unsigned int csum)
1178 int start = skb_headlen(skb);
1179 int i, copy = start - offset;
1182 /* Checksum header. */
1186 csum = csum_partial(skb->data + offset, copy, csum);
1187 if ((len -= copy) == 0)
1193 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1196 BUG_TRAP(start <= offset + len);
1198 end = start + skb_shinfo(skb)->frags[i].size;
1199 if ((copy = end - offset) > 0) {
1202 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1206 vaddr = kmap_skb_frag(frag);
1207 csum2 = csum_partial(vaddr + frag->page_offset +
1208 offset - start, copy, 0);
1209 kunmap_skb_frag(vaddr);
1210 csum = csum_block_add(csum, csum2, pos);
1219 if (skb_shinfo(skb)->frag_list) {
1220 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1222 for (; list; list = list->next) {
1225 BUG_TRAP(start <= offset + len);
1227 end = start + list->len;
1228 if ((copy = end - offset) > 0) {
1232 csum2 = skb_checksum(list, offset - start,
1234 csum = csum_block_add(csum, csum2, pos);
1235 if ((len -= copy) == 0)
1248 /* Both of above in one bottle. */
1250 unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1251 u8 *to, int len, unsigned int csum)
1253 int start = skb_headlen(skb);
1254 int i, copy = start - offset;
1261 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1263 if ((len -= copy) == 0)
1270 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1273 BUG_TRAP(start <= offset + len);
1275 end = start + skb_shinfo(skb)->frags[i].size;
1276 if ((copy = end - offset) > 0) {
1279 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1283 vaddr = kmap_skb_frag(frag);
1284 csum2 = csum_partial_copy_nocheck(vaddr +
1288 kunmap_skb_frag(vaddr);
1289 csum = csum_block_add(csum, csum2, pos);
1299 if (skb_shinfo(skb)->frag_list) {
1300 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1302 for (; list; list = list->next) {
1306 BUG_TRAP(start <= offset + len);
1308 end = start + list->len;
1309 if ((copy = end - offset) > 0) {
1312 csum2 = skb_copy_and_csum_bits(list,
1315 csum = csum_block_add(csum, csum2, pos);
1316 if ((len -= copy) == 0)
1329 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1334 if (skb->ip_summed == CHECKSUM_HW)
1335 csstart = skb->h.raw - skb->data;
1337 csstart = skb_headlen(skb);
1339 BUG_ON(csstart > skb_headlen(skb));
1341 memcpy(to, skb->data, csstart);
1344 if (csstart != skb->len)
1345 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1346 skb->len - csstart, 0);
1348 if (skb->ip_summed == CHECKSUM_HW) {
1349 long csstuff = csstart + skb->csum;
1351 *((unsigned short *)(to + csstuff)) = csum_fold(csum);
1356 * skb_dequeue - remove from the head of the queue
1357 * @list: list to dequeue from
1359 * Remove the head of the list. The list lock is taken so the function
1360 * may be used safely with other locking list functions. The head item is
1361 * returned or %NULL if the list is empty.
1364 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1366 unsigned long flags;
1367 struct sk_buff *result;
1369 spin_lock_irqsave(&list->lock, flags);
1370 result = __skb_dequeue(list);
1371 spin_unlock_irqrestore(&list->lock, flags);
1376 * skb_dequeue_tail - remove from the tail of the queue
1377 * @list: list to dequeue from
1379 * Remove the tail of the list. The list lock is taken so the function
1380 * may be used safely with other locking list functions. The tail item is
1381 * returned or %NULL if the list is empty.
1383 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1385 unsigned long flags;
1386 struct sk_buff *result;
1388 spin_lock_irqsave(&list->lock, flags);
1389 result = __skb_dequeue_tail(list);
1390 spin_unlock_irqrestore(&list->lock, flags);
1395 * skb_queue_purge - empty a list
1396 * @list: list to empty
1398 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1399 * the list and one reference dropped. This function takes the list
1400 * lock and is atomic with respect to other list locking functions.
1402 void skb_queue_purge(struct sk_buff_head *list)
1404 struct sk_buff *skb;
1405 while ((skb = skb_dequeue(list)) != NULL)
1410 * skb_queue_head - queue a buffer at the list head
1411 * @list: list to use
1412 * @newsk: buffer to queue
1414 * Queue a buffer at the start of the list. This function takes the
1415 * list lock and can be used safely with other locking &sk_buff functions
1418 * A buffer cannot be placed on two lists at the same time.
1420 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1422 unsigned long flags;
1424 spin_lock_irqsave(&list->lock, flags);
1425 __skb_queue_head(list, newsk);
1426 spin_unlock_irqrestore(&list->lock, flags);
1430 * skb_queue_tail - queue a buffer at the list tail
1431 * @list: list to use
1432 * @newsk: buffer to queue
1434 * Queue a buffer at the tail of the list. This function takes the
1435 * list lock and can be used safely with other locking &sk_buff functions
1438 * A buffer cannot be placed on two lists at the same time.
1440 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1442 unsigned long flags;
1444 spin_lock_irqsave(&list->lock, flags);
1445 __skb_queue_tail(list, newsk);
1446 spin_unlock_irqrestore(&list->lock, flags);
1450 * skb_unlink - remove a buffer from a list
1451 * @skb: buffer to remove
1452 * @list: list to use
1454 * Remove a packet from a list. The list locks are taken and this
1455 * function is atomic with respect to other list locked calls
1457 * You must know what list the SKB is on.
1459 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1461 unsigned long flags;
1463 spin_lock_irqsave(&list->lock, flags);
1464 __skb_unlink(skb, list);
1465 spin_unlock_irqrestore(&list->lock, flags);
1469 * skb_append - append a buffer
1470 * @old: buffer to insert after
1471 * @newsk: buffer to insert
1472 * @list: list to use
1474 * Place a packet after a given packet in a list. The list locks are taken
1475 * and this function is atomic with respect to other list locked calls.
1476 * A buffer cannot be placed on two lists at the same time.
1478 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1480 unsigned long flags;
1482 spin_lock_irqsave(&list->lock, flags);
1483 __skb_append(old, newsk, list);
1484 spin_unlock_irqrestore(&list->lock, flags);
1489 * skb_insert - insert a buffer
1490 * @old: buffer to insert before
1491 * @newsk: buffer to insert
1492 * @list: list to use
1494 * Place a packet before a given packet in a list. The list locks are
1495 * taken and this function is atomic with respect to other list locked
1498 * A buffer cannot be placed on two lists at the same time.
1500 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1502 unsigned long flags;
1504 spin_lock_irqsave(&list->lock, flags);
1505 __skb_insert(newsk, old->prev, old, list);
1506 spin_unlock_irqrestore(&list->lock, flags);
1511 * Tune the memory allocator for a new MTU size.
1513 void skb_add_mtu(int mtu)
1515 /* Must match allocation in alloc_skb */
1516 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1518 kmem_add_cache_size(mtu);
1522 static inline void skb_split_inside_header(struct sk_buff *skb,
1523 struct sk_buff* skb1,
1524 const u32 len, const int pos)
1528 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1530 /* And move data appendix as is. */
1531 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1532 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1534 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1535 skb_shinfo(skb)->nr_frags = 0;
1536 skb1->data_len = skb->data_len;
1537 skb1->len += skb1->data_len;
1540 skb->tail = skb->data + len;
1543 static inline void skb_split_no_header(struct sk_buff *skb,
1544 struct sk_buff* skb1,
1545 const u32 len, int pos)
1548 const int nfrags = skb_shinfo(skb)->nr_frags;
1550 skb_shinfo(skb)->nr_frags = 0;
1551 skb1->len = skb1->data_len = skb->len - len;
1553 skb->data_len = len - pos;
1555 for (i = 0; i < nfrags; i++) {
1556 int size = skb_shinfo(skb)->frags[i].size;
1558 if (pos + size > len) {
1559 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1563 * We have two variants in this case:
1564 * 1. Move all the frag to the second
1565 * part, if it is possible. F.e.
1566 * this approach is mandatory for TUX,
1567 * where splitting is expensive.
1568 * 2. Split is accurately. We make this.
1570 get_page(skb_shinfo(skb)->frags[i].page);
1571 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1572 skb_shinfo(skb1)->frags[0].size -= len - pos;
1573 skb_shinfo(skb)->frags[i].size = len - pos;
1574 skb_shinfo(skb)->nr_frags++;
1578 skb_shinfo(skb)->nr_frags++;
1581 skb_shinfo(skb1)->nr_frags = k;
1585 * skb_split - Split fragmented skb to two parts at length len.
1586 * @skb: the buffer to split
1587 * @skb1: the buffer to receive the second part
1588 * @len: new length for skb
1590 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1592 int pos = skb_headlen(skb);
1594 if (len < pos) /* Split line is inside header. */
1595 skb_split_inside_header(skb, skb1, len, pos);
1596 else /* Second chunk has no header, nothing to copy. */
1597 skb_split_no_header(skb, skb1, len, pos);
1601 * skb_prepare_seq_read - Prepare a sequential read of skb data
1602 * @skb: the buffer to read
1603 * @from: lower offset of data to be read
1604 * @to: upper offset of data to be read
1605 * @st: state variable
1607 * Initializes the specified state variable. Must be called before
1608 * invoking skb_seq_read() for the first time.
1610 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1611 unsigned int to, struct skb_seq_state *st)
1613 st->lower_offset = from;
1614 st->upper_offset = to;
1615 st->root_skb = st->cur_skb = skb;
1616 st->frag_idx = st->stepped_offset = 0;
1617 st->frag_data = NULL;
1621 * skb_seq_read - Sequentially read skb data
1622 * @consumed: number of bytes consumed by the caller so far
1623 * @data: destination pointer for data to be returned
1624 * @st: state variable
1626 * Reads a block of skb data at &consumed relative to the
1627 * lower offset specified to skb_prepare_seq_read(). Assigns
1628 * the head of the data block to &data and returns the length
1629 * of the block or 0 if the end of the skb data or the upper
1630 * offset has been reached.
1632 * The caller is not required to consume all of the data
1633 * returned, i.e. &consumed is typically set to the number
1634 * of bytes already consumed and the next call to
1635 * skb_seq_read() will return the remaining part of the block.
1637 * Note: The size of each block of data returned can be arbitary,
1638 * this limitation is the cost for zerocopy seqeuental
1639 * reads of potentially non linear data.
1641 * Note: Fragment lists within fragments are not implemented
1642 * at the moment, state->root_skb could be replaced with
1643 * a stack for this purpose.
1645 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1646 struct skb_seq_state *st)
1648 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1651 if (unlikely(abs_offset >= st->upper_offset))
1655 block_limit = skb_headlen(st->cur_skb);
1657 if (abs_offset < block_limit) {
1658 *data = st->cur_skb->data + abs_offset;
1659 return block_limit - abs_offset;
1662 if (st->frag_idx == 0 && !st->frag_data)
1663 st->stepped_offset += skb_headlen(st->cur_skb);
1665 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1666 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1667 block_limit = frag->size + st->stepped_offset;
1669 if (abs_offset < block_limit) {
1671 st->frag_data = kmap_skb_frag(frag);
1673 *data = (u8 *) st->frag_data + frag->page_offset +
1674 (abs_offset - st->stepped_offset);
1676 return block_limit - abs_offset;
1679 if (st->frag_data) {
1680 kunmap_skb_frag(st->frag_data);
1681 st->frag_data = NULL;
1685 st->stepped_offset += frag->size;
1688 if (st->cur_skb->next) {
1689 st->cur_skb = st->cur_skb->next;
1692 } else if (st->root_skb == st->cur_skb &&
1693 skb_shinfo(st->root_skb)->frag_list) {
1694 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1702 * skb_abort_seq_read - Abort a sequential read of skb data
1703 * @st: state variable
1705 * Must be called if skb_seq_read() was not called until it
1708 void skb_abort_seq_read(struct skb_seq_state *st)
1711 kunmap_skb_frag(st->frag_data);
1714 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1716 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1717 struct ts_config *conf,
1718 struct ts_state *state)
1720 return skb_seq_read(offset, text, TS_SKB_CB(state));
1723 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1725 skb_abort_seq_read(TS_SKB_CB(state));
1729 * skb_find_text - Find a text pattern in skb data
1730 * @skb: the buffer to look in
1731 * @from: search offset
1733 * @config: textsearch configuration
1734 * @state: uninitialized textsearch state variable
1736 * Finds a pattern in the skb data according to the specified
1737 * textsearch configuration. Use textsearch_next() to retrieve
1738 * subsequent occurrences of the pattern. Returns the offset
1739 * to the first occurrence or UINT_MAX if no match was found.
1741 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1742 unsigned int to, struct ts_config *config,
1743 struct ts_state *state)
1745 config->get_next_block = skb_ts_get_next_block;
1746 config->finish = skb_ts_finish;
1748 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1750 return textsearch_find(config, state);
1754 * skb_append_datato_frags: - append the user data to a skb
1755 * @sk: sock structure
1756 * @skb: skb structure to be appened with user data.
1757 * @getfrag: call back function to be used for getting the user data
1758 * @from: pointer to user message iov
1759 * @length: length of the iov message
1761 * Description: This procedure append the user data in the fragment part
1762 * of the skb if any page alloc fails user this procedure returns -ENOMEM
1764 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1765 int (*getfrag)(void *from, char *to, int offset,
1766 int len, int odd, struct sk_buff *skb),
1767 void *from, int length)
1770 skb_frag_t *frag = NULL;
1771 struct page *page = NULL;
1777 /* Return error if we don't have space for new frag */
1778 frg_cnt = skb_shinfo(skb)->nr_frags;
1779 if (frg_cnt >= MAX_SKB_FRAGS)
1782 /* allocate a new page for next frag */
1783 page = alloc_pages(sk->sk_allocation, 0);
1785 /* If alloc_page fails just return failure and caller will
1786 * free previous allocated pages by doing kfree_skb()
1791 /* initialize the next frag */
1792 sk->sk_sndmsg_page = page;
1793 sk->sk_sndmsg_off = 0;
1794 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
1795 skb->truesize += PAGE_SIZE;
1796 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1798 /* get the new initialized frag */
1799 frg_cnt = skb_shinfo(skb)->nr_frags;
1800 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
1802 /* copy the user data to page */
1803 left = PAGE_SIZE - frag->page_offset;
1804 copy = (length > left)? left : length;
1806 ret = getfrag(from, (page_address(frag->page) +
1807 frag->page_offset + frag->size),
1808 offset, copy, 0, skb);
1812 /* copy was successful so update the size parameters */
1813 sk->sk_sndmsg_off += copy;
1816 skb->data_len += copy;
1820 } while (length > 0);
1825 void __init skb_init(void)
1827 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1828 sizeof(struct sk_buff),
1832 if (!skbuff_head_cache)
1833 panic("cannot create skbuff cache");
1835 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
1836 (2*sizeof(struct sk_buff)) +
1841 if (!skbuff_fclone_cache)
1842 panic("cannot create skbuff cache");
1845 EXPORT_SYMBOL(___pskb_trim);
1846 EXPORT_SYMBOL(__kfree_skb);
1847 EXPORT_SYMBOL(__pskb_pull_tail);
1848 EXPORT_SYMBOL(__alloc_skb);
1849 EXPORT_SYMBOL(pskb_copy);
1850 EXPORT_SYMBOL(pskb_expand_head);
1851 EXPORT_SYMBOL(skb_checksum);
1852 EXPORT_SYMBOL(skb_clone);
1853 EXPORT_SYMBOL(skb_clone_fraglist);
1854 EXPORT_SYMBOL(skb_copy);
1855 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1856 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1857 EXPORT_SYMBOL(skb_copy_bits);
1858 EXPORT_SYMBOL(skb_copy_expand);
1859 EXPORT_SYMBOL(skb_over_panic);
1860 EXPORT_SYMBOL(skb_pad);
1861 EXPORT_SYMBOL(skb_realloc_headroom);
1862 EXPORT_SYMBOL(skb_under_panic);
1863 EXPORT_SYMBOL(skb_dequeue);
1864 EXPORT_SYMBOL(skb_dequeue_tail);
1865 EXPORT_SYMBOL(skb_insert);
1866 EXPORT_SYMBOL(skb_queue_purge);
1867 EXPORT_SYMBOL(skb_queue_head);
1868 EXPORT_SYMBOL(skb_queue_tail);
1869 EXPORT_SYMBOL(skb_unlink);
1870 EXPORT_SYMBOL(skb_append);
1871 EXPORT_SYMBOL(skb_split);
1872 EXPORT_SYMBOL(skb_prepare_seq_read);
1873 EXPORT_SYMBOL(skb_seq_read);
1874 EXPORT_SYMBOL(skb_abort_seq_read);
1875 EXPORT_SYMBOL(skb_find_text);
1876 EXPORT_SYMBOL(skb_append_datato_frags);