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,
138 struct skb_shared_info *shinfo;
143 skb = kmem_cache_alloc(fclone ? skbuff_fclone_cache : skbuff_head_cache,
144 gfp_mask & ~__GFP_DMA);
148 /* Get the DATA. Size must match skb_add_mtu(). */
149 size = SKB_DATA_ALIGN(size);
150 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
154 memset(skb, 0, offsetof(struct sk_buff, truesize));
155 skb->truesize = size + sizeof(struct sk_buff);
156 atomic_set(&skb->users, 1);
160 skb->end = data + size;
161 /* make sure we initialize shinfo sequentially */
162 shinfo = skb_shinfo(skb);
163 atomic_set(&shinfo->dataref, 1);
164 shinfo->nr_frags = 0;
165 shinfo->tso_size = 0;
166 shinfo->tso_segs = 0;
167 shinfo->ufo_size = 0;
168 shinfo->ip6_frag_id = 0;
169 shinfo->frag_list = NULL;
172 struct sk_buff *child = skb + 1;
173 atomic_t *fclone_ref = (atomic_t *) (child + 1);
175 skb->fclone = SKB_FCLONE_ORIG;
176 atomic_set(fclone_ref, 1);
178 child->fclone = SKB_FCLONE_UNAVAILABLE;
183 kmem_cache_free(skbuff_head_cache, skb);
189 * alloc_skb_from_cache - allocate a network buffer
190 * @cp: kmem_cache from which to allocate the data area
191 * (object size must be big enough for @size bytes + skb overheads)
192 * @size: size to allocate
193 * @gfp_mask: allocation mask
195 * Allocate a new &sk_buff. The returned buffer has no headroom and
196 * tail room of size bytes. The object has a reference count of one.
197 * The return is the buffer. On a failure the return is %NULL.
199 * Buffers may only be allocated from interrupts using a @gfp_mask of
202 struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp,
210 skb = kmem_cache_alloc(skbuff_head_cache,
211 gfp_mask & ~__GFP_DMA);
216 size = SKB_DATA_ALIGN(size);
217 data = kmem_cache_alloc(cp, gfp_mask);
221 memset(skb, 0, offsetof(struct sk_buff, truesize));
222 skb->truesize = size + sizeof(struct sk_buff);
223 atomic_set(&skb->users, 1);
227 skb->end = data + size;
229 atomic_set(&(skb_shinfo(skb)->dataref), 1);
230 skb_shinfo(skb)->nr_frags = 0;
231 skb_shinfo(skb)->tso_size = 0;
232 skb_shinfo(skb)->tso_segs = 0;
233 skb_shinfo(skb)->frag_list = NULL;
237 kmem_cache_free(skbuff_head_cache, skb);
243 static void skb_drop_fraglist(struct sk_buff *skb)
245 struct sk_buff *list = skb_shinfo(skb)->frag_list;
247 skb_shinfo(skb)->frag_list = NULL;
250 struct sk_buff *this = list;
256 static void skb_clone_fraglist(struct sk_buff *skb)
258 struct sk_buff *list;
260 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
264 void skb_release_data(struct sk_buff *skb)
267 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
268 &skb_shinfo(skb)->dataref)) {
269 if (skb_shinfo(skb)->nr_frags) {
271 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
272 put_page(skb_shinfo(skb)->frags[i].page);
275 if (skb_shinfo(skb)->frag_list)
276 skb_drop_fraglist(skb);
283 * Free an skbuff by memory without cleaning the state.
285 void kfree_skbmem(struct sk_buff *skb)
287 struct sk_buff *other;
288 atomic_t *fclone_ref;
290 skb_release_data(skb);
291 switch (skb->fclone) {
292 case SKB_FCLONE_UNAVAILABLE:
293 kmem_cache_free(skbuff_head_cache, skb);
296 case SKB_FCLONE_ORIG:
297 fclone_ref = (atomic_t *) (skb + 2);
298 if (atomic_dec_and_test(fclone_ref))
299 kmem_cache_free(skbuff_fclone_cache, skb);
302 case SKB_FCLONE_CLONE:
303 fclone_ref = (atomic_t *) (skb + 1);
306 /* The clone portion is available for
307 * fast-cloning again.
309 skb->fclone = SKB_FCLONE_UNAVAILABLE;
311 if (atomic_dec_and_test(fclone_ref))
312 kmem_cache_free(skbuff_fclone_cache, other);
318 * __kfree_skb - private function
321 * Free an sk_buff. Release anything attached to the buffer.
322 * Clean the state. This is an internal helper function. Users should
323 * always call kfree_skb
326 void __kfree_skb(struct sk_buff *skb)
328 dst_release(skb->dst);
330 secpath_put(skb->sp);
332 if (skb->destructor) {
334 skb->destructor(skb);
336 #ifdef CONFIG_NETFILTER
337 nf_conntrack_put(skb->nfct);
338 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
339 nf_conntrack_put_reasm(skb->nfct_reasm);
341 #ifdef CONFIG_BRIDGE_NETFILTER
342 nf_bridge_put(skb->nf_bridge);
345 /* XXX: IS this still necessary? - JHS */
346 #ifdef CONFIG_NET_SCHED
348 #ifdef CONFIG_NET_CLS_ACT
357 * skb_clone - duplicate an sk_buff
358 * @skb: buffer to clone
359 * @gfp_mask: allocation priority
361 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
362 * copies share the same packet data but not structure. The new
363 * buffer has a reference count of 1. If the allocation fails the
364 * function returns %NULL otherwise the new buffer is returned.
366 * If this function is called from an interrupt gfp_mask() must be
370 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
375 if (skb->fclone == SKB_FCLONE_ORIG &&
376 n->fclone == SKB_FCLONE_UNAVAILABLE) {
377 atomic_t *fclone_ref = (atomic_t *) (n + 1);
378 n->fclone = SKB_FCLONE_CLONE;
379 atomic_inc(fclone_ref);
381 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
384 n->fclone = SKB_FCLONE_UNAVAILABLE;
387 #define C(x) n->x = skb->x
389 n->next = n->prev = NULL;
400 secpath_get(skb->sp);
402 memcpy(n->cb, skb->cb, sizeof(skb->cb));
413 n->destructor = NULL;
414 #ifdef CONFIG_NETFILTER
417 nf_conntrack_get(skb->nfct);
419 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
421 nf_conntrack_get_reasm(skb->nfct_reasm);
423 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
426 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
428 nf_conntrack_get_reasm(skb->nfct_reasm);
430 #ifdef CONFIG_BRIDGE_NETFILTER
432 nf_bridge_get(skb->nf_bridge);
434 #endif /*CONFIG_NETFILTER*/
435 #ifdef CONFIG_NET_SCHED
437 #ifdef CONFIG_NET_CLS_ACT
438 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
439 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
440 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
446 atomic_set(&n->users, 1);
452 atomic_inc(&(skb_shinfo(skb)->dataref));
458 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
461 * Shift between the two data areas in bytes
463 unsigned long offset = new->data - old->data;
467 new->priority = old->priority;
468 new->protocol = old->protocol;
469 new->dst = dst_clone(old->dst);
471 new->sp = secpath_get(old->sp);
473 new->h.raw = old->h.raw + offset;
474 new->nh.raw = old->nh.raw + offset;
475 new->mac.raw = old->mac.raw + offset;
476 memcpy(new->cb, old->cb, sizeof(old->cb));
477 new->local_df = old->local_df;
478 new->fclone = SKB_FCLONE_UNAVAILABLE;
479 new->pkt_type = old->pkt_type;
480 new->tstamp = old->tstamp;
481 new->destructor = NULL;
482 #ifdef CONFIG_NETFILTER
483 new->nfmark = old->nfmark;
484 new->nfct = old->nfct;
485 nf_conntrack_get(old->nfct);
486 new->nfctinfo = old->nfctinfo;
487 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
488 new->nfct_reasm = old->nfct_reasm;
489 nf_conntrack_get_reasm(old->nfct_reasm);
491 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
492 new->ipvs_property = old->ipvs_property;
494 #ifdef CONFIG_BRIDGE_NETFILTER
495 new->nf_bridge = old->nf_bridge;
496 nf_bridge_get(old->nf_bridge);
499 #ifdef CONFIG_NET_SCHED
500 #ifdef CONFIG_NET_CLS_ACT
501 new->tc_verd = old->tc_verd;
503 new->tc_index = old->tc_index;
505 atomic_set(&new->users, 1);
506 skb_shinfo(new)->tso_size = skb_shinfo(old)->tso_size;
507 skb_shinfo(new)->tso_segs = skb_shinfo(old)->tso_segs;
511 * skb_copy - create private copy of an sk_buff
512 * @skb: buffer to copy
513 * @gfp_mask: allocation priority
515 * Make a copy of both an &sk_buff and its data. This is used when the
516 * caller wishes to modify the data and needs a private copy of the
517 * data to alter. Returns %NULL on failure or the pointer to the buffer
518 * on success. The returned buffer has a reference count of 1.
520 * As by-product this function converts non-linear &sk_buff to linear
521 * one, so that &sk_buff becomes completely private and caller is allowed
522 * to modify all the data of returned buffer. This means that this
523 * function is not recommended for use in circumstances when only
524 * header is going to be modified. Use pskb_copy() instead.
527 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
529 int headerlen = skb->data - skb->head;
531 * Allocate the copy buffer
533 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
538 /* Set the data pointer */
539 skb_reserve(n, headerlen);
540 /* Set the tail pointer and length */
541 skb_put(n, skb->len);
543 n->ip_summed = skb->ip_summed;
545 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
548 copy_skb_header(n, skb);
554 * pskb_copy - create copy of an sk_buff with private head.
555 * @skb: buffer to copy
556 * @gfp_mask: allocation priority
558 * Make a copy of both an &sk_buff and part of its data, located
559 * in header. Fragmented data remain shared. This is used when
560 * the caller wishes to modify only header of &sk_buff and needs
561 * private copy of the header to alter. Returns %NULL on failure
562 * or the pointer to the buffer on success.
563 * The returned buffer has a reference count of 1.
566 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
569 * Allocate the copy buffer
571 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
576 /* Set the data pointer */
577 skb_reserve(n, skb->data - skb->head);
578 /* Set the tail pointer and length */
579 skb_put(n, skb_headlen(skb));
581 memcpy(n->data, skb->data, n->len);
583 n->ip_summed = skb->ip_summed;
585 n->data_len = skb->data_len;
588 if (skb_shinfo(skb)->nr_frags) {
591 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
592 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
593 get_page(skb_shinfo(n)->frags[i].page);
595 skb_shinfo(n)->nr_frags = i;
598 if (skb_shinfo(skb)->frag_list) {
599 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
600 skb_clone_fraglist(n);
603 copy_skb_header(n, skb);
609 * pskb_expand_head - reallocate header of &sk_buff
610 * @skb: buffer to reallocate
611 * @nhead: room to add at head
612 * @ntail: room to add at tail
613 * @gfp_mask: allocation priority
615 * Expands (or creates identical copy, if &nhead and &ntail are zero)
616 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
617 * reference count of 1. Returns zero in the case of success or error,
618 * if expansion failed. In the last case, &sk_buff is not changed.
620 * All the pointers pointing into skb header may change and must be
621 * reloaded after call to this function.
624 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
629 int size = nhead + (skb->end - skb->head) + ntail;
635 size = SKB_DATA_ALIGN(size);
637 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
641 /* Copy only real data... and, alas, header. This should be
642 * optimized for the cases when header is void. */
643 memcpy(data + nhead, skb->head, skb->tail - skb->head);
644 memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
646 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
647 get_page(skb_shinfo(skb)->frags[i].page);
649 if (skb_shinfo(skb)->frag_list)
650 skb_clone_fraglist(skb);
652 skb_release_data(skb);
654 off = (data + nhead) - skb->head;
657 skb->end = data + size;
665 atomic_set(&skb_shinfo(skb)->dataref, 1);
672 /* Make private copy of skb with writable head and some headroom */
674 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
676 struct sk_buff *skb2;
677 int delta = headroom - skb_headroom(skb);
680 skb2 = pskb_copy(skb, GFP_ATOMIC);
682 skb2 = skb_clone(skb, GFP_ATOMIC);
683 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
694 * skb_copy_expand - copy and expand sk_buff
695 * @skb: buffer to copy
696 * @newheadroom: new free bytes at head
697 * @newtailroom: new free bytes at tail
698 * @gfp_mask: allocation priority
700 * Make a copy of both an &sk_buff and its data and while doing so
701 * allocate additional space.
703 * This is used when the caller wishes to modify the data and needs a
704 * private copy of the data to alter as well as more space for new fields.
705 * Returns %NULL on failure or the pointer to the buffer
706 * on success. The returned buffer has a reference count of 1.
708 * You must pass %GFP_ATOMIC as the allocation priority if this function
709 * is called from an interrupt.
711 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
712 * only by netfilter in the cases when checksum is recalculated? --ANK
714 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
715 int newheadroom, int newtailroom,
719 * Allocate the copy buffer
721 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
723 int head_copy_len, head_copy_off;
728 skb_reserve(n, newheadroom);
730 /* Set the tail pointer and length */
731 skb_put(n, skb->len);
733 head_copy_len = skb_headroom(skb);
735 if (newheadroom <= head_copy_len)
736 head_copy_len = newheadroom;
738 head_copy_off = newheadroom - head_copy_len;
740 /* Copy the linear header and data. */
741 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
742 skb->len + head_copy_len))
745 copy_skb_header(n, skb);
751 * skb_pad - zero pad the tail of an skb
752 * @skb: buffer to pad
755 * Ensure that a buffer is followed by a padding area that is zero
756 * filled. Used by network drivers which may DMA or transfer data
757 * beyond the buffer end onto the wire.
759 * May return NULL in out of memory cases.
762 struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
764 struct sk_buff *nskb;
766 /* If the skbuff is non linear tailroom is always zero.. */
767 if (skb_tailroom(skb) >= pad) {
768 memset(skb->data+skb->len, 0, pad);
772 nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
775 memset(nskb->data+nskb->len, 0, pad);
779 /* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
780 * If realloc==0 and trimming is impossible without change of data,
784 int ___pskb_trim(struct sk_buff *skb, unsigned int len, int realloc)
786 int offset = skb_headlen(skb);
787 int nfrags = skb_shinfo(skb)->nr_frags;
790 for (i = 0; i < nfrags; i++) {
791 int end = offset + skb_shinfo(skb)->frags[i].size;
793 if (skb_cloned(skb)) {
795 if (pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
799 put_page(skb_shinfo(skb)->frags[i].page);
800 skb_shinfo(skb)->nr_frags--;
802 skb_shinfo(skb)->frags[i].size = len - offset;
809 skb->data_len -= skb->len - len;
812 if (len <= skb_headlen(skb)) {
815 skb->tail = skb->data + len;
816 if (skb_shinfo(skb)->frag_list && !skb_cloned(skb))
817 skb_drop_fraglist(skb);
819 skb->data_len -= skb->len - len;
828 * __pskb_pull_tail - advance tail of skb header
829 * @skb: buffer to reallocate
830 * @delta: number of bytes to advance tail
832 * The function makes a sense only on a fragmented &sk_buff,
833 * it expands header moving its tail forward and copying necessary
834 * data from fragmented part.
836 * &sk_buff MUST have reference count of 1.
838 * Returns %NULL (and &sk_buff does not change) if pull failed
839 * or value of new tail of skb in the case of success.
841 * All the pointers pointing into skb header may change and must be
842 * reloaded after call to this function.
845 /* Moves tail of skb head forward, copying data from fragmented part,
846 * when it is necessary.
847 * 1. It may fail due to malloc failure.
848 * 2. It may change skb pointers.
850 * It is pretty complicated. Luckily, it is called only in exceptional cases.
852 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
854 /* If skb has not enough free space at tail, get new one
855 * plus 128 bytes for future expansions. If we have enough
856 * room at tail, reallocate without expansion only if skb is cloned.
858 int i, k, eat = (skb->tail + delta) - skb->end;
860 if (eat > 0 || skb_cloned(skb)) {
861 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
866 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
869 /* Optimization: no fragments, no reasons to preestimate
870 * size of pulled pages. Superb.
872 if (!skb_shinfo(skb)->frag_list)
875 /* Estimate size of pulled pages. */
877 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
878 if (skb_shinfo(skb)->frags[i].size >= eat)
880 eat -= skb_shinfo(skb)->frags[i].size;
883 /* If we need update frag list, we are in troubles.
884 * Certainly, it possible to add an offset to skb data,
885 * but taking into account that pulling is expected to
886 * be very rare operation, it is worth to fight against
887 * further bloating skb head and crucify ourselves here instead.
888 * Pure masohism, indeed. 8)8)
891 struct sk_buff *list = skb_shinfo(skb)->frag_list;
892 struct sk_buff *clone = NULL;
893 struct sk_buff *insp = NULL;
898 if (list->len <= eat) {
899 /* Eaten as whole. */
904 /* Eaten partially. */
906 if (skb_shared(list)) {
907 /* Sucks! We need to fork list. :-( */
908 clone = skb_clone(list, GFP_ATOMIC);
914 /* This may be pulled without
918 if (!pskb_pull(list, eat)) {
927 /* Free pulled out fragments. */
928 while ((list = skb_shinfo(skb)->frag_list) != insp) {
929 skb_shinfo(skb)->frag_list = list->next;
932 /* And insert new clone at head. */
935 skb_shinfo(skb)->frag_list = clone;
938 /* Success! Now we may commit changes to skb data. */
943 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
944 if (skb_shinfo(skb)->frags[i].size <= eat) {
945 put_page(skb_shinfo(skb)->frags[i].page);
946 eat -= skb_shinfo(skb)->frags[i].size;
948 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
950 skb_shinfo(skb)->frags[k].page_offset += eat;
951 skb_shinfo(skb)->frags[k].size -= eat;
957 skb_shinfo(skb)->nr_frags = k;
960 skb->data_len -= delta;
965 /* Copy some data bits from skb to kernel buffer. */
967 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
970 int start = skb_headlen(skb);
972 if (offset > (int)skb->len - len)
976 if ((copy = start - offset) > 0) {
979 memcpy(to, skb->data + offset, copy);
980 if ((len -= copy) == 0)
986 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
989 BUG_TRAP(start <= offset + len);
991 end = start + skb_shinfo(skb)->frags[i].size;
992 if ((copy = end - offset) > 0) {
998 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1000 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1001 offset - start, copy);
1002 kunmap_skb_frag(vaddr);
1004 if ((len -= copy) == 0)
1012 if (skb_shinfo(skb)->frag_list) {
1013 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1015 for (; list; list = list->next) {
1018 BUG_TRAP(start <= offset + len);
1020 end = start + list->len;
1021 if ((copy = end - offset) > 0) {
1024 if (skb_copy_bits(list, offset - start,
1027 if ((len -= copy) == 0)
1043 * skb_store_bits - store bits from kernel buffer to skb
1044 * @skb: destination buffer
1045 * @offset: offset in destination
1046 * @from: source buffer
1047 * @len: number of bytes to copy
1049 * Copy the specified number of bytes from the source buffer to the
1050 * destination skb. This function handles all the messy bits of
1051 * traversing fragment lists and such.
1054 int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
1057 int start = skb_headlen(skb);
1059 if (offset > (int)skb->len - len)
1062 if ((copy = start - offset) > 0) {
1065 memcpy(skb->data + offset, from, copy);
1066 if ((len -= copy) == 0)
1072 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1073 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1076 BUG_TRAP(start <= offset + len);
1078 end = start + frag->size;
1079 if ((copy = end - offset) > 0) {
1085 vaddr = kmap_skb_frag(frag);
1086 memcpy(vaddr + frag->page_offset + offset - start,
1088 kunmap_skb_frag(vaddr);
1090 if ((len -= copy) == 0)
1098 if (skb_shinfo(skb)->frag_list) {
1099 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1101 for (; list; list = list->next) {
1104 BUG_TRAP(start <= offset + len);
1106 end = start + list->len;
1107 if ((copy = end - offset) > 0) {
1110 if (skb_store_bits(list, offset - start,
1113 if ((len -= copy) == 0)
1128 EXPORT_SYMBOL(skb_store_bits);
1130 /* Checksum skb data. */
1132 unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1133 int len, unsigned int csum)
1135 int start = skb_headlen(skb);
1136 int i, copy = start - offset;
1139 /* Checksum header. */
1143 csum = csum_partial(skb->data + offset, copy, csum);
1144 if ((len -= copy) == 0)
1150 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1153 BUG_TRAP(start <= offset + len);
1155 end = start + skb_shinfo(skb)->frags[i].size;
1156 if ((copy = end - offset) > 0) {
1159 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1163 vaddr = kmap_skb_frag(frag);
1164 csum2 = csum_partial(vaddr + frag->page_offset +
1165 offset - start, copy, 0);
1166 kunmap_skb_frag(vaddr);
1167 csum = csum_block_add(csum, csum2, pos);
1176 if (skb_shinfo(skb)->frag_list) {
1177 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1179 for (; list; list = list->next) {
1182 BUG_TRAP(start <= offset + len);
1184 end = start + list->len;
1185 if ((copy = end - offset) > 0) {
1189 csum2 = skb_checksum(list, offset - start,
1191 csum = csum_block_add(csum, csum2, pos);
1192 if ((len -= copy) == 0)
1205 /* Both of above in one bottle. */
1207 unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1208 u8 *to, int len, unsigned int csum)
1210 int start = skb_headlen(skb);
1211 int i, copy = start - offset;
1218 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1220 if ((len -= copy) == 0)
1227 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1230 BUG_TRAP(start <= offset + len);
1232 end = start + skb_shinfo(skb)->frags[i].size;
1233 if ((copy = end - offset) > 0) {
1236 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1240 vaddr = kmap_skb_frag(frag);
1241 csum2 = csum_partial_copy_nocheck(vaddr +
1245 kunmap_skb_frag(vaddr);
1246 csum = csum_block_add(csum, csum2, pos);
1256 if (skb_shinfo(skb)->frag_list) {
1257 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1259 for (; list; list = list->next) {
1263 BUG_TRAP(start <= offset + len);
1265 end = start + list->len;
1266 if ((copy = end - offset) > 0) {
1269 csum2 = skb_copy_and_csum_bits(list,
1272 csum = csum_block_add(csum, csum2, pos);
1273 if ((len -= copy) == 0)
1286 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1291 if (skb->ip_summed == CHECKSUM_HW)
1292 csstart = skb->h.raw - skb->data;
1294 csstart = skb_headlen(skb);
1296 BUG_ON(csstart > skb_headlen(skb));
1298 memcpy(to, skb->data, csstart);
1301 if (csstart != skb->len)
1302 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1303 skb->len - csstart, 0);
1305 if (skb->ip_summed == CHECKSUM_HW) {
1306 long csstuff = csstart + skb->csum;
1308 *((unsigned short *)(to + csstuff)) = csum_fold(csum);
1313 * skb_dequeue - remove from the head of the queue
1314 * @list: list to dequeue from
1316 * Remove the head of the list. The list lock is taken so the function
1317 * may be used safely with other locking list functions. The head item is
1318 * returned or %NULL if the list is empty.
1321 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1323 unsigned long flags;
1324 struct sk_buff *result;
1326 spin_lock_irqsave(&list->lock, flags);
1327 result = __skb_dequeue(list);
1328 spin_unlock_irqrestore(&list->lock, flags);
1333 * skb_dequeue_tail - remove from the tail of the queue
1334 * @list: list to dequeue from
1336 * Remove the tail of the list. The list lock is taken so the function
1337 * may be used safely with other locking list functions. The tail item is
1338 * returned or %NULL if the list is empty.
1340 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1342 unsigned long flags;
1343 struct sk_buff *result;
1345 spin_lock_irqsave(&list->lock, flags);
1346 result = __skb_dequeue_tail(list);
1347 spin_unlock_irqrestore(&list->lock, flags);
1352 * skb_queue_purge - empty a list
1353 * @list: list to empty
1355 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1356 * the list and one reference dropped. This function takes the list
1357 * lock and is atomic with respect to other list locking functions.
1359 void skb_queue_purge(struct sk_buff_head *list)
1361 struct sk_buff *skb;
1362 while ((skb = skb_dequeue(list)) != NULL)
1367 * skb_queue_head - queue a buffer at the list head
1368 * @list: list to use
1369 * @newsk: buffer to queue
1371 * Queue a buffer at the start of the list. This function takes the
1372 * list lock and can be used safely with other locking &sk_buff functions
1375 * A buffer cannot be placed on two lists at the same time.
1377 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1379 unsigned long flags;
1381 spin_lock_irqsave(&list->lock, flags);
1382 __skb_queue_head(list, newsk);
1383 spin_unlock_irqrestore(&list->lock, flags);
1387 * skb_queue_tail - queue a buffer at the list tail
1388 * @list: list to use
1389 * @newsk: buffer to queue
1391 * Queue a buffer at the tail of the list. This function takes the
1392 * list lock and can be used safely with other locking &sk_buff functions
1395 * A buffer cannot be placed on two lists at the same time.
1397 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1399 unsigned long flags;
1401 spin_lock_irqsave(&list->lock, flags);
1402 __skb_queue_tail(list, newsk);
1403 spin_unlock_irqrestore(&list->lock, flags);
1407 * skb_unlink - remove a buffer from a list
1408 * @skb: buffer to remove
1409 * @list: list to use
1411 * Remove a packet from a list. The list locks are taken and this
1412 * function is atomic with respect to other list locked calls
1414 * You must know what list the SKB is on.
1416 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1418 unsigned long flags;
1420 spin_lock_irqsave(&list->lock, flags);
1421 __skb_unlink(skb, list);
1422 spin_unlock_irqrestore(&list->lock, flags);
1426 * skb_append - append a buffer
1427 * @old: buffer to insert after
1428 * @newsk: buffer to insert
1429 * @list: list to use
1431 * Place a packet after a given packet in a list. The list locks are taken
1432 * and this function is atomic with respect to other list locked calls.
1433 * A buffer cannot be placed on two lists at the same time.
1435 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1437 unsigned long flags;
1439 spin_lock_irqsave(&list->lock, flags);
1440 __skb_append(old, newsk, list);
1441 spin_unlock_irqrestore(&list->lock, flags);
1446 * skb_insert - insert a buffer
1447 * @old: buffer to insert before
1448 * @newsk: buffer to insert
1449 * @list: list to use
1451 * Place a packet before a given packet in a list. The list locks are
1452 * taken and this function is atomic with respect to other list locked
1455 * A buffer cannot be placed on two lists at the same time.
1457 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1459 unsigned long flags;
1461 spin_lock_irqsave(&list->lock, flags);
1462 __skb_insert(newsk, old->prev, old, list);
1463 spin_unlock_irqrestore(&list->lock, flags);
1468 * Tune the memory allocator for a new MTU size.
1470 void skb_add_mtu(int mtu)
1472 /* Must match allocation in alloc_skb */
1473 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1475 kmem_add_cache_size(mtu);
1479 static inline void skb_split_inside_header(struct sk_buff *skb,
1480 struct sk_buff* skb1,
1481 const u32 len, const int pos)
1485 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1487 /* And move data appendix as is. */
1488 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1489 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1491 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1492 skb_shinfo(skb)->nr_frags = 0;
1493 skb1->data_len = skb->data_len;
1494 skb1->len += skb1->data_len;
1497 skb->tail = skb->data + len;
1500 static inline void skb_split_no_header(struct sk_buff *skb,
1501 struct sk_buff* skb1,
1502 const u32 len, int pos)
1505 const int nfrags = skb_shinfo(skb)->nr_frags;
1507 skb_shinfo(skb)->nr_frags = 0;
1508 skb1->len = skb1->data_len = skb->len - len;
1510 skb->data_len = len - pos;
1512 for (i = 0; i < nfrags; i++) {
1513 int size = skb_shinfo(skb)->frags[i].size;
1515 if (pos + size > len) {
1516 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1520 * We have two variants in this case:
1521 * 1. Move all the frag to the second
1522 * part, if it is possible. F.e.
1523 * this approach is mandatory for TUX,
1524 * where splitting is expensive.
1525 * 2. Split is accurately. We make this.
1527 get_page(skb_shinfo(skb)->frags[i].page);
1528 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1529 skb_shinfo(skb1)->frags[0].size -= len - pos;
1530 skb_shinfo(skb)->frags[i].size = len - pos;
1531 skb_shinfo(skb)->nr_frags++;
1535 skb_shinfo(skb)->nr_frags++;
1538 skb_shinfo(skb1)->nr_frags = k;
1542 * skb_split - Split fragmented skb to two parts at length len.
1543 * @skb: the buffer to split
1544 * @skb1: the buffer to receive the second part
1545 * @len: new length for skb
1547 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1549 int pos = skb_headlen(skb);
1551 if (len < pos) /* Split line is inside header. */
1552 skb_split_inside_header(skb, skb1, len, pos);
1553 else /* Second chunk has no header, nothing to copy. */
1554 skb_split_no_header(skb, skb1, len, pos);
1558 * skb_prepare_seq_read - Prepare a sequential read of skb data
1559 * @skb: the buffer to read
1560 * @from: lower offset of data to be read
1561 * @to: upper offset of data to be read
1562 * @st: state variable
1564 * Initializes the specified state variable. Must be called before
1565 * invoking skb_seq_read() for the first time.
1567 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1568 unsigned int to, struct skb_seq_state *st)
1570 st->lower_offset = from;
1571 st->upper_offset = to;
1572 st->root_skb = st->cur_skb = skb;
1573 st->frag_idx = st->stepped_offset = 0;
1574 st->frag_data = NULL;
1578 * skb_seq_read - Sequentially read skb data
1579 * @consumed: number of bytes consumed by the caller so far
1580 * @data: destination pointer for data to be returned
1581 * @st: state variable
1583 * Reads a block of skb data at &consumed relative to the
1584 * lower offset specified to skb_prepare_seq_read(). Assigns
1585 * the head of the data block to &data and returns the length
1586 * of the block or 0 if the end of the skb data or the upper
1587 * offset has been reached.
1589 * The caller is not required to consume all of the data
1590 * returned, i.e. &consumed is typically set to the number
1591 * of bytes already consumed and the next call to
1592 * skb_seq_read() will return the remaining part of the block.
1594 * Note: The size of each block of data returned can be arbitary,
1595 * this limitation is the cost for zerocopy seqeuental
1596 * reads of potentially non linear data.
1598 * Note: Fragment lists within fragments are not implemented
1599 * at the moment, state->root_skb could be replaced with
1600 * a stack for this purpose.
1602 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1603 struct skb_seq_state *st)
1605 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1608 if (unlikely(abs_offset >= st->upper_offset))
1612 block_limit = skb_headlen(st->cur_skb);
1614 if (abs_offset < block_limit) {
1615 *data = st->cur_skb->data + abs_offset;
1616 return block_limit - abs_offset;
1619 if (st->frag_idx == 0 && !st->frag_data)
1620 st->stepped_offset += skb_headlen(st->cur_skb);
1622 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1623 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1624 block_limit = frag->size + st->stepped_offset;
1626 if (abs_offset < block_limit) {
1628 st->frag_data = kmap_skb_frag(frag);
1630 *data = (u8 *) st->frag_data + frag->page_offset +
1631 (abs_offset - st->stepped_offset);
1633 return block_limit - abs_offset;
1636 if (st->frag_data) {
1637 kunmap_skb_frag(st->frag_data);
1638 st->frag_data = NULL;
1642 st->stepped_offset += frag->size;
1645 if (st->cur_skb->next) {
1646 st->cur_skb = st->cur_skb->next;
1649 } else if (st->root_skb == st->cur_skb &&
1650 skb_shinfo(st->root_skb)->frag_list) {
1651 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1659 * skb_abort_seq_read - Abort a sequential read of skb data
1660 * @st: state variable
1662 * Must be called if skb_seq_read() was not called until it
1665 void skb_abort_seq_read(struct skb_seq_state *st)
1668 kunmap_skb_frag(st->frag_data);
1671 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1673 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1674 struct ts_config *conf,
1675 struct ts_state *state)
1677 return skb_seq_read(offset, text, TS_SKB_CB(state));
1680 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1682 skb_abort_seq_read(TS_SKB_CB(state));
1686 * skb_find_text - Find a text pattern in skb data
1687 * @skb: the buffer to look in
1688 * @from: search offset
1690 * @config: textsearch configuration
1691 * @state: uninitialized textsearch state variable
1693 * Finds a pattern in the skb data according to the specified
1694 * textsearch configuration. Use textsearch_next() to retrieve
1695 * subsequent occurrences of the pattern. Returns the offset
1696 * to the first occurrence or UINT_MAX if no match was found.
1698 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1699 unsigned int to, struct ts_config *config,
1700 struct ts_state *state)
1702 config->get_next_block = skb_ts_get_next_block;
1703 config->finish = skb_ts_finish;
1705 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1707 return textsearch_find(config, state);
1711 * skb_append_datato_frags: - append the user data to a skb
1712 * @sk: sock structure
1713 * @skb: skb structure to be appened with user data.
1714 * @getfrag: call back function to be used for getting the user data
1715 * @from: pointer to user message iov
1716 * @length: length of the iov message
1718 * Description: This procedure append the user data in the fragment part
1719 * of the skb if any page alloc fails user this procedure returns -ENOMEM
1721 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1722 int (*getfrag)(void *from, char *to, int offset,
1723 int len, int odd, struct sk_buff *skb),
1724 void *from, int length)
1727 skb_frag_t *frag = NULL;
1728 struct page *page = NULL;
1734 /* Return error if we don't have space for new frag */
1735 frg_cnt = skb_shinfo(skb)->nr_frags;
1736 if (frg_cnt >= MAX_SKB_FRAGS)
1739 /* allocate a new page for next frag */
1740 page = alloc_pages(sk->sk_allocation, 0);
1742 /* If alloc_page fails just return failure and caller will
1743 * free previous allocated pages by doing kfree_skb()
1748 /* initialize the next frag */
1749 sk->sk_sndmsg_page = page;
1750 sk->sk_sndmsg_off = 0;
1751 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
1752 skb->truesize += PAGE_SIZE;
1753 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1755 /* get the new initialized frag */
1756 frg_cnt = skb_shinfo(skb)->nr_frags;
1757 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
1759 /* copy the user data to page */
1760 left = PAGE_SIZE - frag->page_offset;
1761 copy = (length > left)? left : length;
1763 ret = getfrag(from, (page_address(frag->page) +
1764 frag->page_offset + frag->size),
1765 offset, copy, 0, skb);
1769 /* copy was successful so update the size parameters */
1770 sk->sk_sndmsg_off += copy;
1773 skb->data_len += copy;
1777 } while (length > 0);
1782 void __init skb_init(void)
1784 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1785 sizeof(struct sk_buff),
1789 if (!skbuff_head_cache)
1790 panic("cannot create skbuff cache");
1792 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
1793 (2*sizeof(struct sk_buff)) +
1798 if (!skbuff_fclone_cache)
1799 panic("cannot create skbuff cache");
1802 EXPORT_SYMBOL(___pskb_trim);
1803 EXPORT_SYMBOL(__kfree_skb);
1804 EXPORT_SYMBOL(__pskb_pull_tail);
1805 EXPORT_SYMBOL(__alloc_skb);
1806 EXPORT_SYMBOL(pskb_copy);
1807 EXPORT_SYMBOL(pskb_expand_head);
1808 EXPORT_SYMBOL(skb_checksum);
1809 EXPORT_SYMBOL(skb_clone);
1810 EXPORT_SYMBOL(skb_clone_fraglist);
1811 EXPORT_SYMBOL(skb_copy);
1812 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1813 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1814 EXPORT_SYMBOL(skb_copy_bits);
1815 EXPORT_SYMBOL(skb_copy_expand);
1816 EXPORT_SYMBOL(skb_over_panic);
1817 EXPORT_SYMBOL(skb_pad);
1818 EXPORT_SYMBOL(skb_realloc_headroom);
1819 EXPORT_SYMBOL(skb_under_panic);
1820 EXPORT_SYMBOL(skb_dequeue);
1821 EXPORT_SYMBOL(skb_dequeue_tail);
1822 EXPORT_SYMBOL(skb_insert);
1823 EXPORT_SYMBOL(skb_queue_purge);
1824 EXPORT_SYMBOL(skb_queue_head);
1825 EXPORT_SYMBOL(skb_queue_tail);
1826 EXPORT_SYMBOL(skb_unlink);
1827 EXPORT_SYMBOL(skb_append);
1828 EXPORT_SYMBOL(skb_split);
1829 EXPORT_SYMBOL(skb_prepare_seq_read);
1830 EXPORT_SYMBOL(skb_seq_read);
1831 EXPORT_SYMBOL(skb_abort_seq_read);
1832 EXPORT_SYMBOL(skb_find_text);
1833 EXPORT_SYMBOL(skb_append_datato_frags);