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)->frag_list = NULL;
239 kmem_cache_free(skbuff_head_cache, skb);
245 static void skb_drop_fraglist(struct sk_buff *skb)
247 struct sk_buff *list = skb_shinfo(skb)->frag_list;
249 skb_shinfo(skb)->frag_list = NULL;
252 struct sk_buff *this = list;
258 static void skb_clone_fraglist(struct sk_buff *skb)
260 struct sk_buff *list;
262 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
266 void skb_release_data(struct sk_buff *skb)
269 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
270 &skb_shinfo(skb)->dataref)) {
271 if (skb_shinfo(skb)->nr_frags) {
273 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
274 put_page(skb_shinfo(skb)->frags[i].page);
277 if (skb_shinfo(skb)->frag_list)
278 skb_drop_fraglist(skb);
285 * Free an skbuff by memory without cleaning the state.
287 void kfree_skbmem(struct sk_buff *skb)
289 struct sk_buff *other;
290 atomic_t *fclone_ref;
292 skb_release_data(skb);
293 switch (skb->fclone) {
294 case SKB_FCLONE_UNAVAILABLE:
295 kmem_cache_free(skbuff_head_cache, skb);
298 case SKB_FCLONE_ORIG:
299 fclone_ref = (atomic_t *) (skb + 2);
300 if (atomic_dec_and_test(fclone_ref))
301 kmem_cache_free(skbuff_fclone_cache, skb);
304 case SKB_FCLONE_CLONE:
305 fclone_ref = (atomic_t *) (skb + 1);
308 /* The clone portion is available for
309 * fast-cloning again.
311 skb->fclone = SKB_FCLONE_UNAVAILABLE;
313 if (atomic_dec_and_test(fclone_ref))
314 kmem_cache_free(skbuff_fclone_cache, other);
320 * __kfree_skb - private function
323 * Free an sk_buff. Release anything attached to the buffer.
324 * Clean the state. This is an internal helper function. Users should
325 * always call kfree_skb
328 void __kfree_skb(struct sk_buff *skb)
330 dst_release(skb->dst);
332 secpath_put(skb->sp);
334 if (skb->destructor) {
336 skb->destructor(skb);
338 #ifdef CONFIG_NETFILTER
339 nf_conntrack_put(skb->nfct);
340 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
341 nf_conntrack_put_reasm(skb->nfct_reasm);
343 #ifdef CONFIG_BRIDGE_NETFILTER
344 nf_bridge_put(skb->nf_bridge);
347 /* XXX: IS this still necessary? - JHS */
348 #ifdef CONFIG_NET_SCHED
350 #ifdef CONFIG_NET_CLS_ACT
359 * skb_clone - duplicate an sk_buff
360 * @skb: buffer to clone
361 * @gfp_mask: allocation priority
363 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
364 * copies share the same packet data but not structure. The new
365 * buffer has a reference count of 1. If the allocation fails the
366 * function returns %NULL otherwise the new buffer is returned.
368 * If this function is called from an interrupt gfp_mask() must be
372 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
377 if (skb->fclone == SKB_FCLONE_ORIG &&
378 n->fclone == SKB_FCLONE_UNAVAILABLE) {
379 atomic_t *fclone_ref = (atomic_t *) (n + 1);
380 n->fclone = SKB_FCLONE_CLONE;
381 atomic_inc(fclone_ref);
383 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
386 n->fclone = SKB_FCLONE_UNAVAILABLE;
389 #define C(x) n->x = skb->x
391 n->next = n->prev = NULL;
402 secpath_get(skb->sp);
404 memcpy(n->cb, skb->cb, sizeof(skb->cb));
415 n->destructor = NULL;
416 #ifdef CONFIG_NETFILTER
419 nf_conntrack_get(skb->nfct);
421 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
423 nf_conntrack_get_reasm(skb->nfct_reasm);
425 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
428 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
430 nf_conntrack_get_reasm(skb->nfct_reasm);
432 #ifdef CONFIG_BRIDGE_NETFILTER
434 nf_bridge_get(skb->nf_bridge);
436 #endif /*CONFIG_NETFILTER*/
437 #ifdef CONFIG_NET_SCHED
439 #ifdef CONFIG_NET_CLS_ACT
440 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
441 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
442 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
448 atomic_set(&n->users, 1);
454 atomic_inc(&(skb_shinfo(skb)->dataref));
460 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
463 * Shift between the two data areas in bytes
465 unsigned long offset = new->data - old->data;
469 new->priority = old->priority;
470 new->protocol = old->protocol;
471 new->dst = dst_clone(old->dst);
473 new->sp = secpath_get(old->sp);
475 new->h.raw = old->h.raw + offset;
476 new->nh.raw = old->nh.raw + offset;
477 new->mac.raw = old->mac.raw + offset;
478 memcpy(new->cb, old->cb, sizeof(old->cb));
479 new->local_df = old->local_df;
480 new->fclone = SKB_FCLONE_UNAVAILABLE;
481 new->pkt_type = old->pkt_type;
482 new->tstamp = old->tstamp;
483 new->destructor = NULL;
484 #ifdef CONFIG_NETFILTER
485 new->nfmark = old->nfmark;
486 new->nfct = old->nfct;
487 nf_conntrack_get(old->nfct);
488 new->nfctinfo = old->nfctinfo;
489 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
490 new->nfct_reasm = old->nfct_reasm;
491 nf_conntrack_get_reasm(old->nfct_reasm);
493 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
494 new->ipvs_property = old->ipvs_property;
496 #ifdef CONFIG_BRIDGE_NETFILTER
497 new->nf_bridge = old->nf_bridge;
498 nf_bridge_get(old->nf_bridge);
501 #ifdef CONFIG_NET_SCHED
502 #ifdef CONFIG_NET_CLS_ACT
503 new->tc_verd = old->tc_verd;
505 new->tc_index = old->tc_index;
507 atomic_set(&new->users, 1);
508 skb_shinfo(new)->tso_size = skb_shinfo(old)->tso_size;
509 skb_shinfo(new)->tso_segs = skb_shinfo(old)->tso_segs;
513 * skb_copy - create private copy of an sk_buff
514 * @skb: buffer to copy
515 * @gfp_mask: allocation priority
517 * Make a copy of both an &sk_buff and its data. This is used when the
518 * caller wishes to modify the data and needs a private copy of the
519 * data to alter. Returns %NULL on failure or the pointer to the buffer
520 * on success. The returned buffer has a reference count of 1.
522 * As by-product this function converts non-linear &sk_buff to linear
523 * one, so that &sk_buff becomes completely private and caller is allowed
524 * to modify all the data of returned buffer. This means that this
525 * function is not recommended for use in circumstances when only
526 * header is going to be modified. Use pskb_copy() instead.
529 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
531 int headerlen = skb->data - skb->head;
533 * Allocate the copy buffer
535 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
540 /* Set the data pointer */
541 skb_reserve(n, headerlen);
542 /* Set the tail pointer and length */
543 skb_put(n, skb->len);
545 n->ip_summed = skb->ip_summed;
547 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
550 copy_skb_header(n, skb);
556 * pskb_copy - create copy of an sk_buff with private head.
557 * @skb: buffer to copy
558 * @gfp_mask: allocation priority
560 * Make a copy of both an &sk_buff and part of its data, located
561 * in header. Fragmented data remain shared. This is used when
562 * the caller wishes to modify only header of &sk_buff and needs
563 * private copy of the header to alter. Returns %NULL on failure
564 * or the pointer to the buffer on success.
565 * The returned buffer has a reference count of 1.
568 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
571 * Allocate the copy buffer
573 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
578 /* Set the data pointer */
579 skb_reserve(n, skb->data - skb->head);
580 /* Set the tail pointer and length */
581 skb_put(n, skb_headlen(skb));
583 memcpy(n->data, skb->data, n->len);
585 n->ip_summed = skb->ip_summed;
587 n->data_len = skb->data_len;
590 if (skb_shinfo(skb)->nr_frags) {
593 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
594 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
595 get_page(skb_shinfo(n)->frags[i].page);
597 skb_shinfo(n)->nr_frags = i;
600 if (skb_shinfo(skb)->frag_list) {
601 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
602 skb_clone_fraglist(n);
605 copy_skb_header(n, skb);
611 * pskb_expand_head - reallocate header of &sk_buff
612 * @skb: buffer to reallocate
613 * @nhead: room to add at head
614 * @ntail: room to add at tail
615 * @gfp_mask: allocation priority
617 * Expands (or creates identical copy, if &nhead and &ntail are zero)
618 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
619 * reference count of 1. Returns zero in the case of success or error,
620 * if expansion failed. In the last case, &sk_buff is not changed.
622 * All the pointers pointing into skb header may change and must be
623 * reloaded after call to this function.
626 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
631 int size = nhead + (skb->end - skb->head) + ntail;
637 size = SKB_DATA_ALIGN(size);
639 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
643 /* Copy only real data... and, alas, header. This should be
644 * optimized for the cases when header is void. */
645 memcpy(data + nhead, skb->head, skb->tail - skb->head);
646 memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
648 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
649 get_page(skb_shinfo(skb)->frags[i].page);
651 if (skb_shinfo(skb)->frag_list)
652 skb_clone_fraglist(skb);
654 skb_release_data(skb);
656 off = (data + nhead) - skb->head;
659 skb->end = data + size;
667 atomic_set(&skb_shinfo(skb)->dataref, 1);
674 /* Make private copy of skb with writable head and some headroom */
676 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
678 struct sk_buff *skb2;
679 int delta = headroom - skb_headroom(skb);
682 skb2 = pskb_copy(skb, GFP_ATOMIC);
684 skb2 = skb_clone(skb, GFP_ATOMIC);
685 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
696 * skb_copy_expand - copy and expand sk_buff
697 * @skb: buffer to copy
698 * @newheadroom: new free bytes at head
699 * @newtailroom: new free bytes at tail
700 * @gfp_mask: allocation priority
702 * Make a copy of both an &sk_buff and its data and while doing so
703 * allocate additional space.
705 * This is used when the caller wishes to modify the data and needs a
706 * private copy of the data to alter as well as more space for new fields.
707 * Returns %NULL on failure or the pointer to the buffer
708 * on success. The returned buffer has a reference count of 1.
710 * You must pass %GFP_ATOMIC as the allocation priority if this function
711 * is called from an interrupt.
713 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
714 * only by netfilter in the cases when checksum is recalculated? --ANK
716 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
717 int newheadroom, int newtailroom,
721 * Allocate the copy buffer
723 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
725 int head_copy_len, head_copy_off;
730 skb_reserve(n, newheadroom);
732 /* Set the tail pointer and length */
733 skb_put(n, skb->len);
735 head_copy_len = skb_headroom(skb);
737 if (newheadroom <= head_copy_len)
738 head_copy_len = newheadroom;
740 head_copy_off = newheadroom - head_copy_len;
742 /* Copy the linear header and data. */
743 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
744 skb->len + head_copy_len))
747 copy_skb_header(n, skb);
753 * skb_pad - zero pad the tail of an skb
754 * @skb: buffer to pad
757 * Ensure that a buffer is followed by a padding area that is zero
758 * filled. Used by network drivers which may DMA or transfer data
759 * beyond the buffer end onto the wire.
761 * May return NULL in out of memory cases.
764 struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
766 struct sk_buff *nskb;
768 /* If the skbuff is non linear tailroom is always zero.. */
769 if (skb_tailroom(skb) >= pad) {
770 memset(skb->data+skb->len, 0, pad);
774 nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
777 memset(nskb->data+nskb->len, 0, pad);
781 /* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
782 * If realloc==0 and trimming is impossible without change of data,
786 int ___pskb_trim(struct sk_buff *skb, unsigned int len, int realloc)
788 int offset = skb_headlen(skb);
789 int nfrags = skb_shinfo(skb)->nr_frags;
792 for (i = 0; i < nfrags; i++) {
793 int end = offset + skb_shinfo(skb)->frags[i].size;
795 if (skb_cloned(skb)) {
797 if (pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
801 put_page(skb_shinfo(skb)->frags[i].page);
802 skb_shinfo(skb)->nr_frags--;
804 skb_shinfo(skb)->frags[i].size = len - offset;
811 skb->data_len -= skb->len - len;
814 if (len <= skb_headlen(skb)) {
817 skb->tail = skb->data + len;
818 if (skb_shinfo(skb)->frag_list && !skb_cloned(skb))
819 skb_drop_fraglist(skb);
821 skb->data_len -= skb->len - len;
830 * __pskb_pull_tail - advance tail of skb header
831 * @skb: buffer to reallocate
832 * @delta: number of bytes to advance tail
834 * The function makes a sense only on a fragmented &sk_buff,
835 * it expands header moving its tail forward and copying necessary
836 * data from fragmented part.
838 * &sk_buff MUST have reference count of 1.
840 * Returns %NULL (and &sk_buff does not change) if pull failed
841 * or value of new tail of skb in the case of success.
843 * All the pointers pointing into skb header may change and must be
844 * reloaded after call to this function.
847 /* Moves tail of skb head forward, copying data from fragmented part,
848 * when it is necessary.
849 * 1. It may fail due to malloc failure.
850 * 2. It may change skb pointers.
852 * It is pretty complicated. Luckily, it is called only in exceptional cases.
854 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
856 /* If skb has not enough free space at tail, get new one
857 * plus 128 bytes for future expansions. If we have enough
858 * room at tail, reallocate without expansion only if skb is cloned.
860 int i, k, eat = (skb->tail + delta) - skb->end;
862 if (eat > 0 || skb_cloned(skb)) {
863 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
868 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
871 /* Optimization: no fragments, no reasons to preestimate
872 * size of pulled pages. Superb.
874 if (!skb_shinfo(skb)->frag_list)
877 /* Estimate size of pulled pages. */
879 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
880 if (skb_shinfo(skb)->frags[i].size >= eat)
882 eat -= skb_shinfo(skb)->frags[i].size;
885 /* If we need update frag list, we are in troubles.
886 * Certainly, it possible to add an offset to skb data,
887 * but taking into account that pulling is expected to
888 * be very rare operation, it is worth to fight against
889 * further bloating skb head and crucify ourselves here instead.
890 * Pure masohism, indeed. 8)8)
893 struct sk_buff *list = skb_shinfo(skb)->frag_list;
894 struct sk_buff *clone = NULL;
895 struct sk_buff *insp = NULL;
900 if (list->len <= eat) {
901 /* Eaten as whole. */
906 /* Eaten partially. */
908 if (skb_shared(list)) {
909 /* Sucks! We need to fork list. :-( */
910 clone = skb_clone(list, GFP_ATOMIC);
916 /* This may be pulled without
920 if (!pskb_pull(list, eat)) {
929 /* Free pulled out fragments. */
930 while ((list = skb_shinfo(skb)->frag_list) != insp) {
931 skb_shinfo(skb)->frag_list = list->next;
934 /* And insert new clone at head. */
937 skb_shinfo(skb)->frag_list = clone;
940 /* Success! Now we may commit changes to skb data. */
945 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
946 if (skb_shinfo(skb)->frags[i].size <= eat) {
947 put_page(skb_shinfo(skb)->frags[i].page);
948 eat -= skb_shinfo(skb)->frags[i].size;
950 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
952 skb_shinfo(skb)->frags[k].page_offset += eat;
953 skb_shinfo(skb)->frags[k].size -= eat;
959 skb_shinfo(skb)->nr_frags = k;
962 skb->data_len -= delta;
967 /* Copy some data bits from skb to kernel buffer. */
969 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
972 int start = skb_headlen(skb);
974 if (offset > (int)skb->len - len)
978 if ((copy = start - offset) > 0) {
981 memcpy(to, skb->data + offset, copy);
982 if ((len -= copy) == 0)
988 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
991 BUG_TRAP(start <= offset + len);
993 end = start + skb_shinfo(skb)->frags[i].size;
994 if ((copy = end - offset) > 0) {
1000 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1002 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1003 offset - start, copy);
1004 kunmap_skb_frag(vaddr);
1006 if ((len -= copy) == 0)
1014 if (skb_shinfo(skb)->frag_list) {
1015 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1017 for (; list; list = list->next) {
1020 BUG_TRAP(start <= offset + len);
1022 end = start + list->len;
1023 if ((copy = end - offset) > 0) {
1026 if (skb_copy_bits(list, offset - start,
1029 if ((len -= copy) == 0)
1045 * skb_store_bits - store bits from kernel buffer to skb
1046 * @skb: destination buffer
1047 * @offset: offset in destination
1048 * @from: source buffer
1049 * @len: number of bytes to copy
1051 * Copy the specified number of bytes from the source buffer to the
1052 * destination skb. This function handles all the messy bits of
1053 * traversing fragment lists and such.
1056 int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
1059 int start = skb_headlen(skb);
1061 if (offset > (int)skb->len - len)
1064 if ((copy = start - offset) > 0) {
1067 memcpy(skb->data + offset, from, copy);
1068 if ((len -= copy) == 0)
1074 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1075 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1078 BUG_TRAP(start <= offset + len);
1080 end = start + frag->size;
1081 if ((copy = end - offset) > 0) {
1087 vaddr = kmap_skb_frag(frag);
1088 memcpy(vaddr + frag->page_offset + offset - start,
1090 kunmap_skb_frag(vaddr);
1092 if ((len -= copy) == 0)
1100 if (skb_shinfo(skb)->frag_list) {
1101 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1103 for (; list; list = list->next) {
1106 BUG_TRAP(start <= offset + len);
1108 end = start + list->len;
1109 if ((copy = end - offset) > 0) {
1112 if (skb_store_bits(list, offset - start,
1115 if ((len -= copy) == 0)
1130 EXPORT_SYMBOL(skb_store_bits);
1132 /* Checksum skb data. */
1134 unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1135 int len, unsigned int csum)
1137 int start = skb_headlen(skb);
1138 int i, copy = start - offset;
1141 /* Checksum header. */
1145 csum = csum_partial(skb->data + offset, copy, csum);
1146 if ((len -= copy) == 0)
1152 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1155 BUG_TRAP(start <= offset + len);
1157 end = start + skb_shinfo(skb)->frags[i].size;
1158 if ((copy = end - offset) > 0) {
1161 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1165 vaddr = kmap_skb_frag(frag);
1166 csum2 = csum_partial(vaddr + frag->page_offset +
1167 offset - start, copy, 0);
1168 kunmap_skb_frag(vaddr);
1169 csum = csum_block_add(csum, csum2, pos);
1178 if (skb_shinfo(skb)->frag_list) {
1179 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1181 for (; list; list = list->next) {
1184 BUG_TRAP(start <= offset + len);
1186 end = start + list->len;
1187 if ((copy = end - offset) > 0) {
1191 csum2 = skb_checksum(list, offset - start,
1193 csum = csum_block_add(csum, csum2, pos);
1194 if ((len -= copy) == 0)
1207 /* Both of above in one bottle. */
1209 unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1210 u8 *to, int len, unsigned int csum)
1212 int start = skb_headlen(skb);
1213 int i, copy = start - offset;
1220 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1222 if ((len -= copy) == 0)
1229 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1232 BUG_TRAP(start <= offset + len);
1234 end = start + skb_shinfo(skb)->frags[i].size;
1235 if ((copy = end - offset) > 0) {
1238 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1242 vaddr = kmap_skb_frag(frag);
1243 csum2 = csum_partial_copy_nocheck(vaddr +
1247 kunmap_skb_frag(vaddr);
1248 csum = csum_block_add(csum, csum2, pos);
1258 if (skb_shinfo(skb)->frag_list) {
1259 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1261 for (; list; list = list->next) {
1265 BUG_TRAP(start <= offset + len);
1267 end = start + list->len;
1268 if ((copy = end - offset) > 0) {
1271 csum2 = skb_copy_and_csum_bits(list,
1274 csum = csum_block_add(csum, csum2, pos);
1275 if ((len -= copy) == 0)
1288 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1293 if (skb->ip_summed == CHECKSUM_HW)
1294 csstart = skb->h.raw - skb->data;
1296 csstart = skb_headlen(skb);
1298 BUG_ON(csstart > skb_headlen(skb));
1300 memcpy(to, skb->data, csstart);
1303 if (csstart != skb->len)
1304 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1305 skb->len - csstart, 0);
1307 if (skb->ip_summed == CHECKSUM_HW) {
1308 long csstuff = csstart + skb->csum;
1310 *((unsigned short *)(to + csstuff)) = csum_fold(csum);
1315 * skb_dequeue - remove from the head of the queue
1316 * @list: list to dequeue from
1318 * Remove the head of the list. The list lock is taken so the function
1319 * may be used safely with other locking list functions. The head item is
1320 * returned or %NULL if the list is empty.
1323 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1325 unsigned long flags;
1326 struct sk_buff *result;
1328 spin_lock_irqsave(&list->lock, flags);
1329 result = __skb_dequeue(list);
1330 spin_unlock_irqrestore(&list->lock, flags);
1335 * skb_dequeue_tail - remove from the tail of the queue
1336 * @list: list to dequeue from
1338 * Remove the tail of the list. The list lock is taken so the function
1339 * may be used safely with other locking list functions. The tail item is
1340 * returned or %NULL if the list is empty.
1342 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1344 unsigned long flags;
1345 struct sk_buff *result;
1347 spin_lock_irqsave(&list->lock, flags);
1348 result = __skb_dequeue_tail(list);
1349 spin_unlock_irqrestore(&list->lock, flags);
1354 * skb_queue_purge - empty a list
1355 * @list: list to empty
1357 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1358 * the list and one reference dropped. This function takes the list
1359 * lock and is atomic with respect to other list locking functions.
1361 void skb_queue_purge(struct sk_buff_head *list)
1363 struct sk_buff *skb;
1364 while ((skb = skb_dequeue(list)) != NULL)
1369 * skb_queue_head - queue a buffer at the list head
1370 * @list: list to use
1371 * @newsk: buffer to queue
1373 * Queue a buffer at the start of the list. This function takes the
1374 * list lock and can be used safely with other locking &sk_buff functions
1377 * A buffer cannot be placed on two lists at the same time.
1379 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1381 unsigned long flags;
1383 spin_lock_irqsave(&list->lock, flags);
1384 __skb_queue_head(list, newsk);
1385 spin_unlock_irqrestore(&list->lock, flags);
1389 * skb_queue_tail - queue a buffer at the list tail
1390 * @list: list to use
1391 * @newsk: buffer to queue
1393 * Queue a buffer at the tail of the list. This function takes the
1394 * list lock and can be used safely with other locking &sk_buff functions
1397 * A buffer cannot be placed on two lists at the same time.
1399 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1401 unsigned long flags;
1403 spin_lock_irqsave(&list->lock, flags);
1404 __skb_queue_tail(list, newsk);
1405 spin_unlock_irqrestore(&list->lock, flags);
1409 * skb_unlink - remove a buffer from a list
1410 * @skb: buffer to remove
1411 * @list: list to use
1413 * Remove a packet from a list. The list locks are taken and this
1414 * function is atomic with respect to other list locked calls
1416 * You must know what list the SKB is on.
1418 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1420 unsigned long flags;
1422 spin_lock_irqsave(&list->lock, flags);
1423 __skb_unlink(skb, list);
1424 spin_unlock_irqrestore(&list->lock, flags);
1428 * skb_append - append a buffer
1429 * @old: buffer to insert after
1430 * @newsk: buffer to insert
1431 * @list: list to use
1433 * Place a packet after a given packet in a list. The list locks are taken
1434 * and this function is atomic with respect to other list locked calls.
1435 * A buffer cannot be placed on two lists at the same time.
1437 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1439 unsigned long flags;
1441 spin_lock_irqsave(&list->lock, flags);
1442 __skb_append(old, newsk, list);
1443 spin_unlock_irqrestore(&list->lock, flags);
1448 * skb_insert - insert a buffer
1449 * @old: buffer to insert before
1450 * @newsk: buffer to insert
1451 * @list: list to use
1453 * Place a packet before a given packet in a list. The list locks are
1454 * taken and this function is atomic with respect to other list locked
1457 * A buffer cannot be placed on two lists at the same time.
1459 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1461 unsigned long flags;
1463 spin_lock_irqsave(&list->lock, flags);
1464 __skb_insert(newsk, old->prev, old, list);
1465 spin_unlock_irqrestore(&list->lock, flags);
1470 * Tune the memory allocator for a new MTU size.
1472 void skb_add_mtu(int mtu)
1474 /* Must match allocation in alloc_skb */
1475 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1477 kmem_add_cache_size(mtu);
1481 static inline void skb_split_inside_header(struct sk_buff *skb,
1482 struct sk_buff* skb1,
1483 const u32 len, const int pos)
1487 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1489 /* And move data appendix as is. */
1490 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1491 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1493 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1494 skb_shinfo(skb)->nr_frags = 0;
1495 skb1->data_len = skb->data_len;
1496 skb1->len += skb1->data_len;
1499 skb->tail = skb->data + len;
1502 static inline void skb_split_no_header(struct sk_buff *skb,
1503 struct sk_buff* skb1,
1504 const u32 len, int pos)
1507 const int nfrags = skb_shinfo(skb)->nr_frags;
1509 skb_shinfo(skb)->nr_frags = 0;
1510 skb1->len = skb1->data_len = skb->len - len;
1512 skb->data_len = len - pos;
1514 for (i = 0; i < nfrags; i++) {
1515 int size = skb_shinfo(skb)->frags[i].size;
1517 if (pos + size > len) {
1518 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1522 * We have two variants in this case:
1523 * 1. Move all the frag to the second
1524 * part, if it is possible. F.e.
1525 * this approach is mandatory for TUX,
1526 * where splitting is expensive.
1527 * 2. Split is accurately. We make this.
1529 get_page(skb_shinfo(skb)->frags[i].page);
1530 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1531 skb_shinfo(skb1)->frags[0].size -= len - pos;
1532 skb_shinfo(skb)->frags[i].size = len - pos;
1533 skb_shinfo(skb)->nr_frags++;
1537 skb_shinfo(skb)->nr_frags++;
1540 skb_shinfo(skb1)->nr_frags = k;
1544 * skb_split - Split fragmented skb to two parts at length len.
1545 * @skb: the buffer to split
1546 * @skb1: the buffer to receive the second part
1547 * @len: new length for skb
1549 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1551 int pos = skb_headlen(skb);
1553 if (len < pos) /* Split line is inside header. */
1554 skb_split_inside_header(skb, skb1, len, pos);
1555 else /* Second chunk has no header, nothing to copy. */
1556 skb_split_no_header(skb, skb1, len, pos);
1560 * skb_prepare_seq_read - Prepare a sequential read of skb data
1561 * @skb: the buffer to read
1562 * @from: lower offset of data to be read
1563 * @to: upper offset of data to be read
1564 * @st: state variable
1566 * Initializes the specified state variable. Must be called before
1567 * invoking skb_seq_read() for the first time.
1569 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1570 unsigned int to, struct skb_seq_state *st)
1572 st->lower_offset = from;
1573 st->upper_offset = to;
1574 st->root_skb = st->cur_skb = skb;
1575 st->frag_idx = st->stepped_offset = 0;
1576 st->frag_data = NULL;
1580 * skb_seq_read - Sequentially read skb data
1581 * @consumed: number of bytes consumed by the caller so far
1582 * @data: destination pointer for data to be returned
1583 * @st: state variable
1585 * Reads a block of skb data at &consumed relative to the
1586 * lower offset specified to skb_prepare_seq_read(). Assigns
1587 * the head of the data block to &data and returns the length
1588 * of the block or 0 if the end of the skb data or the upper
1589 * offset has been reached.
1591 * The caller is not required to consume all of the data
1592 * returned, i.e. &consumed is typically set to the number
1593 * of bytes already consumed and the next call to
1594 * skb_seq_read() will return the remaining part of the block.
1596 * Note: The size of each block of data returned can be arbitary,
1597 * this limitation is the cost for zerocopy seqeuental
1598 * reads of potentially non linear data.
1600 * Note: Fragment lists within fragments are not implemented
1601 * at the moment, state->root_skb could be replaced with
1602 * a stack for this purpose.
1604 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1605 struct skb_seq_state *st)
1607 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1610 if (unlikely(abs_offset >= st->upper_offset))
1614 block_limit = skb_headlen(st->cur_skb);
1616 if (abs_offset < block_limit) {
1617 *data = st->cur_skb->data + abs_offset;
1618 return block_limit - abs_offset;
1621 if (st->frag_idx == 0 && !st->frag_data)
1622 st->stepped_offset += skb_headlen(st->cur_skb);
1624 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1625 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1626 block_limit = frag->size + st->stepped_offset;
1628 if (abs_offset < block_limit) {
1630 st->frag_data = kmap_skb_frag(frag);
1632 *data = (u8 *) st->frag_data + frag->page_offset +
1633 (abs_offset - st->stepped_offset);
1635 return block_limit - abs_offset;
1638 if (st->frag_data) {
1639 kunmap_skb_frag(st->frag_data);
1640 st->frag_data = NULL;
1644 st->stepped_offset += frag->size;
1647 if (st->cur_skb->next) {
1648 st->cur_skb = st->cur_skb->next;
1651 } else if (st->root_skb == st->cur_skb &&
1652 skb_shinfo(st->root_skb)->frag_list) {
1653 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1661 * skb_abort_seq_read - Abort a sequential read of skb data
1662 * @st: state variable
1664 * Must be called if skb_seq_read() was not called until it
1667 void skb_abort_seq_read(struct skb_seq_state *st)
1670 kunmap_skb_frag(st->frag_data);
1673 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1675 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1676 struct ts_config *conf,
1677 struct ts_state *state)
1679 return skb_seq_read(offset, text, TS_SKB_CB(state));
1682 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1684 skb_abort_seq_read(TS_SKB_CB(state));
1688 * skb_find_text - Find a text pattern in skb data
1689 * @skb: the buffer to look in
1690 * @from: search offset
1692 * @config: textsearch configuration
1693 * @state: uninitialized textsearch state variable
1695 * Finds a pattern in the skb data according to the specified
1696 * textsearch configuration. Use textsearch_next() to retrieve
1697 * subsequent occurrences of the pattern. Returns the offset
1698 * to the first occurrence or UINT_MAX if no match was found.
1700 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1701 unsigned int to, struct ts_config *config,
1702 struct ts_state *state)
1704 config->get_next_block = skb_ts_get_next_block;
1705 config->finish = skb_ts_finish;
1707 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1709 return textsearch_find(config, state);
1713 * skb_append_datato_frags: - append the user data to a skb
1714 * @sk: sock structure
1715 * @skb: skb structure to be appened with user data.
1716 * @getfrag: call back function to be used for getting the user data
1717 * @from: pointer to user message iov
1718 * @length: length of the iov message
1720 * Description: This procedure append the user data in the fragment part
1721 * of the skb if any page alloc fails user this procedure returns -ENOMEM
1723 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1724 int (*getfrag)(void *from, char *to, int offset,
1725 int len, int odd, struct sk_buff *skb),
1726 void *from, int length)
1729 skb_frag_t *frag = NULL;
1730 struct page *page = NULL;
1736 /* Return error if we don't have space for new frag */
1737 frg_cnt = skb_shinfo(skb)->nr_frags;
1738 if (frg_cnt >= MAX_SKB_FRAGS)
1741 /* allocate a new page for next frag */
1742 page = alloc_pages(sk->sk_allocation, 0);
1744 /* If alloc_page fails just return failure and caller will
1745 * free previous allocated pages by doing kfree_skb()
1750 /* initialize the next frag */
1751 sk->sk_sndmsg_page = page;
1752 sk->sk_sndmsg_off = 0;
1753 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
1754 skb->truesize += PAGE_SIZE;
1755 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1757 /* get the new initialized frag */
1758 frg_cnt = skb_shinfo(skb)->nr_frags;
1759 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
1761 /* copy the user data to page */
1762 left = PAGE_SIZE - frag->page_offset;
1763 copy = (length > left)? left : length;
1765 ret = getfrag(from, (page_address(frag->page) +
1766 frag->page_offset + frag->size),
1767 offset, copy, 0, skb);
1771 /* copy was successful so update the size parameters */
1772 sk->sk_sndmsg_off += copy;
1775 skb->data_len += copy;
1779 } while (length > 0);
1784 void __init skb_init(void)
1786 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1787 sizeof(struct sk_buff),
1791 if (!skbuff_head_cache)
1792 panic("cannot create skbuff cache");
1794 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
1795 (2*sizeof(struct sk_buff)) +
1800 if (!skbuff_fclone_cache)
1801 panic("cannot create skbuff cache");
1804 EXPORT_SYMBOL(___pskb_trim);
1805 EXPORT_SYMBOL(__kfree_skb);
1806 EXPORT_SYMBOL(__pskb_pull_tail);
1807 EXPORT_SYMBOL(__alloc_skb);
1808 EXPORT_SYMBOL(pskb_copy);
1809 EXPORT_SYMBOL(pskb_expand_head);
1810 EXPORT_SYMBOL(skb_checksum);
1811 EXPORT_SYMBOL(skb_clone);
1812 EXPORT_SYMBOL(skb_clone_fraglist);
1813 EXPORT_SYMBOL(skb_copy);
1814 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1815 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1816 EXPORT_SYMBOL(skb_copy_bits);
1817 EXPORT_SYMBOL(skb_copy_expand);
1818 EXPORT_SYMBOL(skb_over_panic);
1819 EXPORT_SYMBOL(skb_pad);
1820 EXPORT_SYMBOL(skb_realloc_headroom);
1821 EXPORT_SYMBOL(skb_under_panic);
1822 EXPORT_SYMBOL(skb_dequeue);
1823 EXPORT_SYMBOL(skb_dequeue_tail);
1824 EXPORT_SYMBOL(skb_insert);
1825 EXPORT_SYMBOL(skb_queue_purge);
1826 EXPORT_SYMBOL(skb_queue_head);
1827 EXPORT_SYMBOL(skb_queue_tail);
1828 EXPORT_SYMBOL(skb_unlink);
1829 EXPORT_SYMBOL(skb_append);
1830 EXPORT_SYMBOL(skb_split);
1831 EXPORT_SYMBOL(skb_prepare_seq_read);
1832 EXPORT_SYMBOL(skb_seq_read);
1833 EXPORT_SYMBOL(skb_abort_seq_read);
1834 EXPORT_SYMBOL(skb_find_text);
1835 EXPORT_SYMBOL(skb_append_datato_frags);