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,
143 skb = kmem_cache_alloc(skbuff_fclone_cache,
144 gfp_mask & ~__GFP_DMA);
146 skb = kmem_cache_alloc(skbuff_head_cache,
147 gfp_mask & ~__GFP_DMA);
152 /* Get the DATA. Size must match skb_add_mtu(). */
153 size = SKB_DATA_ALIGN(size);
154 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
158 memset(skb, 0, offsetof(struct sk_buff, truesize));
159 skb->truesize = size + sizeof(struct sk_buff);
160 atomic_set(&skb->users, 1);
164 skb->end = data + size;
166 struct sk_buff *child = skb + 1;
167 atomic_t *fclone_ref = (atomic_t *) (child + 1);
169 skb->fclone = SKB_FCLONE_ORIG;
170 atomic_set(fclone_ref, 1);
172 child->fclone = SKB_FCLONE_UNAVAILABLE;
174 atomic_set(&(skb_shinfo(skb)->dataref), 1);
175 skb_shinfo(skb)->nr_frags = 0;
176 skb_shinfo(skb)->tso_size = 0;
177 skb_shinfo(skb)->tso_segs = 0;
178 skb_shinfo(skb)->frag_list = NULL;
182 kmem_cache_free(skbuff_head_cache, skb);
188 * alloc_skb_from_cache - allocate a network buffer
189 * @cp: kmem_cache from which to allocate the data area
190 * (object size must be big enough for @size bytes + skb overheads)
191 * @size: size to allocate
192 * @gfp_mask: allocation mask
194 * Allocate a new &sk_buff. The returned buffer has no headroom and
195 * tail room of size bytes. The object has a reference count of one.
196 * The return is the buffer. On a failure the return is %NULL.
198 * Buffers may only be allocated from interrupts using a @gfp_mask of
201 struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp,
209 skb = kmem_cache_alloc(skbuff_head_cache,
210 gfp_mask & ~__GFP_DMA);
215 size = SKB_DATA_ALIGN(size);
216 data = kmem_cache_alloc(cp, gfp_mask);
220 memset(skb, 0, offsetof(struct sk_buff, truesize));
221 skb->truesize = size + sizeof(struct sk_buff);
222 atomic_set(&skb->users, 1);
226 skb->end = data + size;
228 atomic_set(&(skb_shinfo(skb)->dataref), 1);
229 skb_shinfo(skb)->nr_frags = 0;
230 skb_shinfo(skb)->tso_size = 0;
231 skb_shinfo(skb)->tso_segs = 0;
232 skb_shinfo(skb)->frag_list = NULL;
236 kmem_cache_free(skbuff_head_cache, skb);
242 static void skb_drop_fraglist(struct sk_buff *skb)
244 struct sk_buff *list = skb_shinfo(skb)->frag_list;
246 skb_shinfo(skb)->frag_list = NULL;
249 struct sk_buff *this = list;
255 static void skb_clone_fraglist(struct sk_buff *skb)
257 struct sk_buff *list;
259 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
263 void skb_release_data(struct sk_buff *skb)
266 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
267 &skb_shinfo(skb)->dataref)) {
268 if (skb_shinfo(skb)->nr_frags) {
270 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
271 put_page(skb_shinfo(skb)->frags[i].page);
274 if (skb_shinfo(skb)->frag_list)
275 skb_drop_fraglist(skb);
282 * Free an skbuff by memory without cleaning the state.
284 void kfree_skbmem(struct sk_buff *skb)
286 struct sk_buff *other;
287 atomic_t *fclone_ref;
289 skb_release_data(skb);
290 switch (skb->fclone) {
291 case SKB_FCLONE_UNAVAILABLE:
292 kmem_cache_free(skbuff_head_cache, skb);
295 case SKB_FCLONE_ORIG:
296 fclone_ref = (atomic_t *) (skb + 2);
297 if (atomic_dec_and_test(fclone_ref))
298 kmem_cache_free(skbuff_fclone_cache, skb);
301 case SKB_FCLONE_CLONE:
302 fclone_ref = (atomic_t *) (skb + 1);
305 /* The clone portion is available for
306 * fast-cloning again.
308 skb->fclone = SKB_FCLONE_UNAVAILABLE;
310 if (atomic_dec_and_test(fclone_ref))
311 kmem_cache_free(skbuff_fclone_cache, other);
317 * __kfree_skb - private function
320 * Free an sk_buff. Release anything attached to the buffer.
321 * Clean the state. This is an internal helper function. Users should
322 * always call kfree_skb
325 void __kfree_skb(struct sk_buff *skb)
327 dst_release(skb->dst);
329 secpath_put(skb->sp);
331 if (skb->destructor) {
333 skb->destructor(skb);
335 #ifdef CONFIG_NETFILTER
336 nf_conntrack_put(skb->nfct);
337 #ifdef CONFIG_BRIDGE_NETFILTER
338 nf_bridge_put(skb->nf_bridge);
341 /* XXX: IS this still necessary? - JHS */
342 #ifdef CONFIG_NET_SCHED
344 #ifdef CONFIG_NET_CLS_ACT
353 * skb_clone - duplicate an sk_buff
354 * @skb: buffer to clone
355 * @gfp_mask: allocation priority
357 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
358 * copies share the same packet data but not structure. The new
359 * buffer has a reference count of 1. If the allocation fails the
360 * function returns %NULL otherwise the new buffer is returned.
362 * If this function is called from an interrupt gfp_mask() must be
366 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
371 if (skb->fclone == SKB_FCLONE_ORIG &&
372 n->fclone == SKB_FCLONE_UNAVAILABLE) {
373 atomic_t *fclone_ref = (atomic_t *) (n + 1);
374 n->fclone = SKB_FCLONE_CLONE;
375 atomic_inc(fclone_ref);
377 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
380 n->fclone = SKB_FCLONE_UNAVAILABLE;
383 #define C(x) n->x = skb->x
385 n->next = n->prev = NULL;
396 secpath_get(skb->sp);
398 memcpy(n->cb, skb->cb, sizeof(skb->cb));
409 n->destructor = NULL;
410 #ifdef CONFIG_NETFILTER
413 nf_conntrack_get(skb->nfct);
415 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
418 #ifdef CONFIG_BRIDGE_NETFILTER
420 nf_bridge_get(skb->nf_bridge);
422 #endif /*CONFIG_NETFILTER*/
423 #ifdef CONFIG_NET_SCHED
425 #ifdef CONFIG_NET_CLS_ACT
426 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
427 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
428 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
434 atomic_set(&n->users, 1);
440 atomic_inc(&(skb_shinfo(skb)->dataref));
446 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
449 * Shift between the two data areas in bytes
451 unsigned long offset = new->data - old->data;
455 new->priority = old->priority;
456 new->protocol = old->protocol;
457 new->dst = dst_clone(old->dst);
459 new->sp = secpath_get(old->sp);
461 new->h.raw = old->h.raw + offset;
462 new->nh.raw = old->nh.raw + offset;
463 new->mac.raw = old->mac.raw + offset;
464 memcpy(new->cb, old->cb, sizeof(old->cb));
465 new->local_df = old->local_df;
466 new->fclone = SKB_FCLONE_UNAVAILABLE;
467 new->pkt_type = old->pkt_type;
468 new->tstamp = old->tstamp;
469 new->destructor = NULL;
470 #ifdef CONFIG_NETFILTER
471 new->nfmark = old->nfmark;
472 new->nfct = old->nfct;
473 nf_conntrack_get(old->nfct);
474 new->nfctinfo = old->nfctinfo;
475 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
476 new->ipvs_property = old->ipvs_property;
478 #ifdef CONFIG_BRIDGE_NETFILTER
479 new->nf_bridge = old->nf_bridge;
480 nf_bridge_get(old->nf_bridge);
483 #ifdef CONFIG_NET_SCHED
484 #ifdef CONFIG_NET_CLS_ACT
485 new->tc_verd = old->tc_verd;
487 new->tc_index = old->tc_index;
489 atomic_set(&new->users, 1);
490 skb_shinfo(new)->tso_size = skb_shinfo(old)->tso_size;
491 skb_shinfo(new)->tso_segs = skb_shinfo(old)->tso_segs;
495 * skb_copy - create private copy of an sk_buff
496 * @skb: buffer to copy
497 * @gfp_mask: allocation priority
499 * Make a copy of both an &sk_buff and its data. This is used when the
500 * caller wishes to modify the data and needs a private copy of the
501 * data to alter. Returns %NULL on failure or the pointer to the buffer
502 * on success. The returned buffer has a reference count of 1.
504 * As by-product this function converts non-linear &sk_buff to linear
505 * one, so that &sk_buff becomes completely private and caller is allowed
506 * to modify all the data of returned buffer. This means that this
507 * function is not recommended for use in circumstances when only
508 * header is going to be modified. Use pskb_copy() instead.
511 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
513 int headerlen = skb->data - skb->head;
515 * Allocate the copy buffer
517 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
522 /* Set the data pointer */
523 skb_reserve(n, headerlen);
524 /* Set the tail pointer and length */
525 skb_put(n, skb->len);
527 n->ip_summed = skb->ip_summed;
529 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
532 copy_skb_header(n, skb);
538 * pskb_copy - create copy of an sk_buff with private head.
539 * @skb: buffer to copy
540 * @gfp_mask: allocation priority
542 * Make a copy of both an &sk_buff and part of its data, located
543 * in header. Fragmented data remain shared. This is used when
544 * the caller wishes to modify only header of &sk_buff and needs
545 * private copy of the header to alter. Returns %NULL on failure
546 * or the pointer to the buffer on success.
547 * The returned buffer has a reference count of 1.
550 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
553 * Allocate the copy buffer
555 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
560 /* Set the data pointer */
561 skb_reserve(n, skb->data - skb->head);
562 /* Set the tail pointer and length */
563 skb_put(n, skb_headlen(skb));
565 memcpy(n->data, skb->data, n->len);
567 n->ip_summed = skb->ip_summed;
569 n->data_len = skb->data_len;
572 if (skb_shinfo(skb)->nr_frags) {
575 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
576 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
577 get_page(skb_shinfo(n)->frags[i].page);
579 skb_shinfo(n)->nr_frags = i;
582 if (skb_shinfo(skb)->frag_list) {
583 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
584 skb_clone_fraglist(n);
587 copy_skb_header(n, skb);
593 * pskb_expand_head - reallocate header of &sk_buff
594 * @skb: buffer to reallocate
595 * @nhead: room to add at head
596 * @ntail: room to add at tail
597 * @gfp_mask: allocation priority
599 * Expands (or creates identical copy, if &nhead and &ntail are zero)
600 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
601 * reference count of 1. Returns zero in the case of success or error,
602 * if expansion failed. In the last case, &sk_buff is not changed.
604 * All the pointers pointing into skb header may change and must be
605 * reloaded after call to this function.
608 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
613 int size = nhead + (skb->end - skb->head) + ntail;
619 size = SKB_DATA_ALIGN(size);
621 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
625 /* Copy only real data... and, alas, header. This should be
626 * optimized for the cases when header is void. */
627 memcpy(data + nhead, skb->head, skb->tail - skb->head);
628 memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
630 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
631 get_page(skb_shinfo(skb)->frags[i].page);
633 if (skb_shinfo(skb)->frag_list)
634 skb_clone_fraglist(skb);
636 skb_release_data(skb);
638 off = (data + nhead) - skb->head;
641 skb->end = data + size;
649 atomic_set(&skb_shinfo(skb)->dataref, 1);
656 /* Make private copy of skb with writable head and some headroom */
658 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
660 struct sk_buff *skb2;
661 int delta = headroom - skb_headroom(skb);
664 skb2 = pskb_copy(skb, GFP_ATOMIC);
666 skb2 = skb_clone(skb, GFP_ATOMIC);
667 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
678 * skb_copy_expand - copy and expand sk_buff
679 * @skb: buffer to copy
680 * @newheadroom: new free bytes at head
681 * @newtailroom: new free bytes at tail
682 * @gfp_mask: allocation priority
684 * Make a copy of both an &sk_buff and its data and while doing so
685 * allocate additional space.
687 * This is used when the caller wishes to modify the data and needs a
688 * private copy of the data to alter as well as more space for new fields.
689 * Returns %NULL on failure or the pointer to the buffer
690 * on success. The returned buffer has a reference count of 1.
692 * You must pass %GFP_ATOMIC as the allocation priority if this function
693 * is called from an interrupt.
695 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
696 * only by netfilter in the cases when checksum is recalculated? --ANK
698 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
699 int newheadroom, int newtailroom,
703 * Allocate the copy buffer
705 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
707 int head_copy_len, head_copy_off;
712 skb_reserve(n, newheadroom);
714 /* Set the tail pointer and length */
715 skb_put(n, skb->len);
717 head_copy_len = skb_headroom(skb);
719 if (newheadroom <= head_copy_len)
720 head_copy_len = newheadroom;
722 head_copy_off = newheadroom - head_copy_len;
724 /* Copy the linear header and data. */
725 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
726 skb->len + head_copy_len))
729 copy_skb_header(n, skb);
735 * skb_pad - zero pad the tail of an skb
736 * @skb: buffer to pad
739 * Ensure that a buffer is followed by a padding area that is zero
740 * filled. Used by network drivers which may DMA or transfer data
741 * beyond the buffer end onto the wire.
743 * May return NULL in out of memory cases.
746 struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
748 struct sk_buff *nskb;
750 /* If the skbuff is non linear tailroom is always zero.. */
751 if (skb_tailroom(skb) >= pad) {
752 memset(skb->data+skb->len, 0, pad);
756 nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
759 memset(nskb->data+nskb->len, 0, pad);
763 /* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
764 * If realloc==0 and trimming is impossible without change of data,
768 int ___pskb_trim(struct sk_buff *skb, unsigned int len, int realloc)
770 int offset = skb_headlen(skb);
771 int nfrags = skb_shinfo(skb)->nr_frags;
774 for (i = 0; i < nfrags; i++) {
775 int end = offset + skb_shinfo(skb)->frags[i].size;
777 if (skb_cloned(skb)) {
780 if (pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
784 put_page(skb_shinfo(skb)->frags[i].page);
785 skb_shinfo(skb)->nr_frags--;
787 skb_shinfo(skb)->frags[i].size = len - offset;
794 skb->data_len -= skb->len - len;
797 if (len <= skb_headlen(skb)) {
800 skb->tail = skb->data + len;
801 if (skb_shinfo(skb)->frag_list && !skb_cloned(skb))
802 skb_drop_fraglist(skb);
804 skb->data_len -= skb->len - len;
813 * __pskb_pull_tail - advance tail of skb header
814 * @skb: buffer to reallocate
815 * @delta: number of bytes to advance tail
817 * The function makes a sense only on a fragmented &sk_buff,
818 * it expands header moving its tail forward and copying necessary
819 * data from fragmented part.
821 * &sk_buff MUST have reference count of 1.
823 * Returns %NULL (and &sk_buff does not change) if pull failed
824 * or value of new tail of skb in the case of success.
826 * All the pointers pointing into skb header may change and must be
827 * reloaded after call to this function.
830 /* Moves tail of skb head forward, copying data from fragmented part,
831 * when it is necessary.
832 * 1. It may fail due to malloc failure.
833 * 2. It may change skb pointers.
835 * It is pretty complicated. Luckily, it is called only in exceptional cases.
837 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
839 /* If skb has not enough free space at tail, get new one
840 * plus 128 bytes for future expansions. If we have enough
841 * room at tail, reallocate without expansion only if skb is cloned.
843 int i, k, eat = (skb->tail + delta) - skb->end;
845 if (eat > 0 || skb_cloned(skb)) {
846 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
851 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
854 /* Optimization: no fragments, no reasons to preestimate
855 * size of pulled pages. Superb.
857 if (!skb_shinfo(skb)->frag_list)
860 /* Estimate size of pulled pages. */
862 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
863 if (skb_shinfo(skb)->frags[i].size >= eat)
865 eat -= skb_shinfo(skb)->frags[i].size;
868 /* If we need update frag list, we are in troubles.
869 * Certainly, it possible to add an offset to skb data,
870 * but taking into account that pulling is expected to
871 * be very rare operation, it is worth to fight against
872 * further bloating skb head and crucify ourselves here instead.
873 * Pure masohism, indeed. 8)8)
876 struct sk_buff *list = skb_shinfo(skb)->frag_list;
877 struct sk_buff *clone = NULL;
878 struct sk_buff *insp = NULL;
884 if (list->len <= eat) {
885 /* Eaten as whole. */
890 /* Eaten partially. */
892 if (skb_shared(list)) {
893 /* Sucks! We need to fork list. :-( */
894 clone = skb_clone(list, GFP_ATOMIC);
900 /* This may be pulled without
904 if (!pskb_pull(list, eat)) {
913 /* Free pulled out fragments. */
914 while ((list = skb_shinfo(skb)->frag_list) != insp) {
915 skb_shinfo(skb)->frag_list = list->next;
918 /* And insert new clone at head. */
921 skb_shinfo(skb)->frag_list = clone;
924 /* Success! Now we may commit changes to skb data. */
929 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
930 if (skb_shinfo(skb)->frags[i].size <= eat) {
931 put_page(skb_shinfo(skb)->frags[i].page);
932 eat -= skb_shinfo(skb)->frags[i].size;
934 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
936 skb_shinfo(skb)->frags[k].page_offset += eat;
937 skb_shinfo(skb)->frags[k].size -= eat;
943 skb_shinfo(skb)->nr_frags = k;
946 skb->data_len -= delta;
951 /* Copy some data bits from skb to kernel buffer. */
953 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
956 int start = skb_headlen(skb);
958 if (offset > (int)skb->len - len)
962 if ((copy = start - offset) > 0) {
965 memcpy(to, skb->data + offset, copy);
966 if ((len -= copy) == 0)
972 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
975 BUG_TRAP(start <= offset + len);
977 end = start + skb_shinfo(skb)->frags[i].size;
978 if ((copy = end - offset) > 0) {
984 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
986 vaddr + skb_shinfo(skb)->frags[i].page_offset+
987 offset - start, copy);
988 kunmap_skb_frag(vaddr);
990 if ((len -= copy) == 0)
998 if (skb_shinfo(skb)->frag_list) {
999 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1001 for (; list; list = list->next) {
1004 BUG_TRAP(start <= offset + len);
1006 end = start + list->len;
1007 if ((copy = end - offset) > 0) {
1010 if (skb_copy_bits(list, offset - start,
1013 if ((len -= copy) == 0)
1029 * skb_store_bits - store bits from kernel buffer to skb
1030 * @skb: destination buffer
1031 * @offset: offset in destination
1032 * @from: source buffer
1033 * @len: number of bytes to copy
1035 * Copy the specified number of bytes from the source buffer to the
1036 * destination skb. This function handles all the messy bits of
1037 * traversing fragment lists and such.
1040 int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
1043 int start = skb_headlen(skb);
1045 if (offset > (int)skb->len - len)
1048 if ((copy = start - offset) > 0) {
1051 memcpy(skb->data + offset, from, copy);
1052 if ((len -= copy) == 0)
1058 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1059 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1062 BUG_TRAP(start <= offset + len);
1064 end = start + frag->size;
1065 if ((copy = end - offset) > 0) {
1071 vaddr = kmap_skb_frag(frag);
1072 memcpy(vaddr + frag->page_offset + offset - start,
1074 kunmap_skb_frag(vaddr);
1076 if ((len -= copy) == 0)
1084 if (skb_shinfo(skb)->frag_list) {
1085 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1087 for (; list; list = list->next) {
1090 BUG_TRAP(start <= offset + len);
1092 end = start + list->len;
1093 if ((copy = end - offset) > 0) {
1096 if (skb_store_bits(list, offset - start,
1099 if ((len -= copy) == 0)
1114 EXPORT_SYMBOL(skb_store_bits);
1116 /* Checksum skb data. */
1118 unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1119 int len, unsigned int csum)
1121 int start = skb_headlen(skb);
1122 int i, copy = start - offset;
1125 /* Checksum header. */
1129 csum = csum_partial(skb->data + offset, copy, csum);
1130 if ((len -= copy) == 0)
1136 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1139 BUG_TRAP(start <= offset + len);
1141 end = start + skb_shinfo(skb)->frags[i].size;
1142 if ((copy = end - offset) > 0) {
1145 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1149 vaddr = kmap_skb_frag(frag);
1150 csum2 = csum_partial(vaddr + frag->page_offset +
1151 offset - start, copy, 0);
1152 kunmap_skb_frag(vaddr);
1153 csum = csum_block_add(csum, csum2, pos);
1162 if (skb_shinfo(skb)->frag_list) {
1163 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1165 for (; list; list = list->next) {
1168 BUG_TRAP(start <= offset + len);
1170 end = start + list->len;
1171 if ((copy = end - offset) > 0) {
1175 csum2 = skb_checksum(list, offset - start,
1177 csum = csum_block_add(csum, csum2, pos);
1178 if ((len -= copy) == 0)
1192 /* Both of above in one bottle. */
1194 unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1195 u8 *to, int len, unsigned int csum)
1197 int start = skb_headlen(skb);
1198 int i, copy = start - offset;
1205 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1207 if ((len -= copy) == 0)
1214 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1217 BUG_TRAP(start <= offset + len);
1219 end = start + skb_shinfo(skb)->frags[i].size;
1220 if ((copy = end - offset) > 0) {
1223 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1227 vaddr = kmap_skb_frag(frag);
1228 csum2 = csum_partial_copy_nocheck(vaddr +
1232 kunmap_skb_frag(vaddr);
1233 csum = csum_block_add(csum, csum2, pos);
1243 if (skb_shinfo(skb)->frag_list) {
1244 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1246 for (; list; list = list->next) {
1250 BUG_TRAP(start <= offset + len);
1252 end = start + list->len;
1253 if ((copy = end - offset) > 0) {
1256 csum2 = skb_copy_and_csum_bits(list,
1259 csum = csum_block_add(csum, csum2, pos);
1260 if ((len -= copy) == 0)
1274 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1279 if (skb->ip_summed == CHECKSUM_HW)
1280 csstart = skb->h.raw - skb->data;
1282 csstart = skb_headlen(skb);
1284 if (csstart > skb_headlen(skb))
1287 memcpy(to, skb->data, csstart);
1290 if (csstart != skb->len)
1291 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1292 skb->len - csstart, 0);
1294 if (skb->ip_summed == CHECKSUM_HW) {
1295 long csstuff = csstart + skb->csum;
1297 *((unsigned short *)(to + csstuff)) = csum_fold(csum);
1302 * skb_dequeue - remove from the head of the queue
1303 * @list: list to dequeue from
1305 * Remove the head of the list. The list lock is taken so the function
1306 * may be used safely with other locking list functions. The head item is
1307 * returned or %NULL if the list is empty.
1310 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1312 unsigned long flags;
1313 struct sk_buff *result;
1315 spin_lock_irqsave(&list->lock, flags);
1316 result = __skb_dequeue(list);
1317 spin_unlock_irqrestore(&list->lock, flags);
1322 * skb_dequeue_tail - remove from the tail of the queue
1323 * @list: list to dequeue from
1325 * Remove the tail of the list. The list lock is taken so the function
1326 * may be used safely with other locking list functions. The tail item is
1327 * returned or %NULL if the list is empty.
1329 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1331 unsigned long flags;
1332 struct sk_buff *result;
1334 spin_lock_irqsave(&list->lock, flags);
1335 result = __skb_dequeue_tail(list);
1336 spin_unlock_irqrestore(&list->lock, flags);
1341 * skb_queue_purge - empty a list
1342 * @list: list to empty
1344 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1345 * the list and one reference dropped. This function takes the list
1346 * lock and is atomic with respect to other list locking functions.
1348 void skb_queue_purge(struct sk_buff_head *list)
1350 struct sk_buff *skb;
1351 while ((skb = skb_dequeue(list)) != NULL)
1356 * skb_queue_head - queue a buffer at the list head
1357 * @list: list to use
1358 * @newsk: buffer to queue
1360 * Queue a buffer at the start of the list. This function takes the
1361 * list lock and can be used safely with other locking &sk_buff functions
1364 * A buffer cannot be placed on two lists at the same time.
1366 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1368 unsigned long flags;
1370 spin_lock_irqsave(&list->lock, flags);
1371 __skb_queue_head(list, newsk);
1372 spin_unlock_irqrestore(&list->lock, flags);
1376 * skb_queue_tail - queue a buffer at the list tail
1377 * @list: list to use
1378 * @newsk: buffer to queue
1380 * Queue a buffer at the tail of the list. This function takes the
1381 * list lock and can be used safely with other locking &sk_buff functions
1384 * A buffer cannot be placed on two lists at the same time.
1386 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1388 unsigned long flags;
1390 spin_lock_irqsave(&list->lock, flags);
1391 __skb_queue_tail(list, newsk);
1392 spin_unlock_irqrestore(&list->lock, flags);
1396 * skb_unlink - remove a buffer from a list
1397 * @skb: buffer to remove
1398 * @list: list to use
1400 * Remove a packet from a list. The list locks are taken and this
1401 * function is atomic with respect to other list locked calls
1403 * You must know what list the SKB is on.
1405 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1407 unsigned long flags;
1409 spin_lock_irqsave(&list->lock, flags);
1410 __skb_unlink(skb, list);
1411 spin_unlock_irqrestore(&list->lock, flags);
1415 * skb_append - append a buffer
1416 * @old: buffer to insert after
1417 * @newsk: buffer to insert
1418 * @list: list to use
1420 * Place a packet after a given packet in a list. The list locks are taken
1421 * and this function is atomic with respect to other list locked calls.
1422 * A buffer cannot be placed on two lists at the same time.
1424 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1426 unsigned long flags;
1428 spin_lock_irqsave(&list->lock, flags);
1429 __skb_append(old, newsk, list);
1430 spin_unlock_irqrestore(&list->lock, flags);
1435 * skb_insert - insert a buffer
1436 * @old: buffer to insert before
1437 * @newsk: buffer to insert
1438 * @list: list to use
1440 * Place a packet before a given packet in a list. The list locks are
1441 * taken and this function is atomic with respect to other list locked
1444 * A buffer cannot be placed on two lists at the same time.
1446 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1448 unsigned long flags;
1450 spin_lock_irqsave(&list->lock, flags);
1451 __skb_insert(newsk, old->prev, old, list);
1452 spin_unlock_irqrestore(&list->lock, flags);
1457 * Tune the memory allocator for a new MTU size.
1459 void skb_add_mtu(int mtu)
1461 /* Must match allocation in alloc_skb */
1462 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1464 kmem_add_cache_size(mtu);
1468 static inline void skb_split_inside_header(struct sk_buff *skb,
1469 struct sk_buff* skb1,
1470 const u32 len, const int pos)
1474 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1476 /* And move data appendix as is. */
1477 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1478 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1480 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1481 skb_shinfo(skb)->nr_frags = 0;
1482 skb1->data_len = skb->data_len;
1483 skb1->len += skb1->data_len;
1486 skb->tail = skb->data + len;
1489 static inline void skb_split_no_header(struct sk_buff *skb,
1490 struct sk_buff* skb1,
1491 const u32 len, int pos)
1494 const int nfrags = skb_shinfo(skb)->nr_frags;
1496 skb_shinfo(skb)->nr_frags = 0;
1497 skb1->len = skb1->data_len = skb->len - len;
1499 skb->data_len = len - pos;
1501 for (i = 0; i < nfrags; i++) {
1502 int size = skb_shinfo(skb)->frags[i].size;
1504 if (pos + size > len) {
1505 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1509 * We have two variants in this case:
1510 * 1. Move all the frag to the second
1511 * part, if it is possible. F.e.
1512 * this approach is mandatory for TUX,
1513 * where splitting is expensive.
1514 * 2. Split is accurately. We make this.
1516 get_page(skb_shinfo(skb)->frags[i].page);
1517 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1518 skb_shinfo(skb1)->frags[0].size -= len - pos;
1519 skb_shinfo(skb)->frags[i].size = len - pos;
1520 skb_shinfo(skb)->nr_frags++;
1524 skb_shinfo(skb)->nr_frags++;
1527 skb_shinfo(skb1)->nr_frags = k;
1531 * skb_split - Split fragmented skb to two parts at length len.
1532 * @skb: the buffer to split
1533 * @skb1: the buffer to receive the second part
1534 * @len: new length for skb
1536 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1538 int pos = skb_headlen(skb);
1540 if (len < pos) /* Split line is inside header. */
1541 skb_split_inside_header(skb, skb1, len, pos);
1542 else /* Second chunk has no header, nothing to copy. */
1543 skb_split_no_header(skb, skb1, len, pos);
1547 * skb_prepare_seq_read - Prepare a sequential read of skb data
1548 * @skb: the buffer to read
1549 * @from: lower offset of data to be read
1550 * @to: upper offset of data to be read
1551 * @st: state variable
1553 * Initializes the specified state variable. Must be called before
1554 * invoking skb_seq_read() for the first time.
1556 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1557 unsigned int to, struct skb_seq_state *st)
1559 st->lower_offset = from;
1560 st->upper_offset = to;
1561 st->root_skb = st->cur_skb = skb;
1562 st->frag_idx = st->stepped_offset = 0;
1563 st->frag_data = NULL;
1567 * skb_seq_read - Sequentially read skb data
1568 * @consumed: number of bytes consumed by the caller so far
1569 * @data: destination pointer for data to be returned
1570 * @st: state variable
1572 * Reads a block of skb data at &consumed relative to the
1573 * lower offset specified to skb_prepare_seq_read(). Assigns
1574 * the head of the data block to &data and returns the length
1575 * of the block or 0 if the end of the skb data or the upper
1576 * offset has been reached.
1578 * The caller is not required to consume all of the data
1579 * returned, i.e. &consumed is typically set to the number
1580 * of bytes already consumed and the next call to
1581 * skb_seq_read() will return the remaining part of the block.
1583 * Note: The size of each block of data returned can be arbitary,
1584 * this limitation is the cost for zerocopy seqeuental
1585 * reads of potentially non linear data.
1587 * Note: Fragment lists within fragments are not implemented
1588 * at the moment, state->root_skb could be replaced with
1589 * a stack for this purpose.
1591 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1592 struct skb_seq_state *st)
1594 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1597 if (unlikely(abs_offset >= st->upper_offset))
1601 block_limit = skb_headlen(st->cur_skb);
1603 if (abs_offset < block_limit) {
1604 *data = st->cur_skb->data + abs_offset;
1605 return block_limit - abs_offset;
1608 if (st->frag_idx == 0 && !st->frag_data)
1609 st->stepped_offset += skb_headlen(st->cur_skb);
1611 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1612 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1613 block_limit = frag->size + st->stepped_offset;
1615 if (abs_offset < block_limit) {
1617 st->frag_data = kmap_skb_frag(frag);
1619 *data = (u8 *) st->frag_data + frag->page_offset +
1620 (abs_offset - st->stepped_offset);
1622 return block_limit - abs_offset;
1625 if (st->frag_data) {
1626 kunmap_skb_frag(st->frag_data);
1627 st->frag_data = NULL;
1631 st->stepped_offset += frag->size;
1634 if (st->cur_skb->next) {
1635 st->cur_skb = st->cur_skb->next;
1638 } else if (st->root_skb == st->cur_skb &&
1639 skb_shinfo(st->root_skb)->frag_list) {
1640 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1648 * skb_abort_seq_read - Abort a sequential read of skb data
1649 * @st: state variable
1651 * Must be called if skb_seq_read() was not called until it
1654 void skb_abort_seq_read(struct skb_seq_state *st)
1657 kunmap_skb_frag(st->frag_data);
1660 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
1662 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1663 struct ts_config *conf,
1664 struct ts_state *state)
1666 return skb_seq_read(offset, text, TS_SKB_CB(state));
1669 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1671 skb_abort_seq_read(TS_SKB_CB(state));
1675 * skb_find_text - Find a text pattern in skb data
1676 * @skb: the buffer to look in
1677 * @from: search offset
1679 * @config: textsearch configuration
1680 * @state: uninitialized textsearch state variable
1682 * Finds a pattern in the skb data according to the specified
1683 * textsearch configuration. Use textsearch_next() to retrieve
1684 * subsequent occurrences of the pattern. Returns the offset
1685 * to the first occurrence or UINT_MAX if no match was found.
1687 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1688 unsigned int to, struct ts_config *config,
1689 struct ts_state *state)
1691 config->get_next_block = skb_ts_get_next_block;
1692 config->finish = skb_ts_finish;
1694 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1696 return textsearch_find(config, state);
1699 void __init skb_init(void)
1701 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1702 sizeof(struct sk_buff),
1706 if (!skbuff_head_cache)
1707 panic("cannot create skbuff cache");
1709 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
1710 (2*sizeof(struct sk_buff)) +
1715 if (!skbuff_fclone_cache)
1716 panic("cannot create skbuff cache");
1719 EXPORT_SYMBOL(___pskb_trim);
1720 EXPORT_SYMBOL(__kfree_skb);
1721 EXPORT_SYMBOL(__pskb_pull_tail);
1722 EXPORT_SYMBOL(__alloc_skb);
1723 EXPORT_SYMBOL(pskb_copy);
1724 EXPORT_SYMBOL(pskb_expand_head);
1725 EXPORT_SYMBOL(skb_checksum);
1726 EXPORT_SYMBOL(skb_clone);
1727 EXPORT_SYMBOL(skb_clone_fraglist);
1728 EXPORT_SYMBOL(skb_copy);
1729 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1730 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1731 EXPORT_SYMBOL(skb_copy_bits);
1732 EXPORT_SYMBOL(skb_copy_expand);
1733 EXPORT_SYMBOL(skb_over_panic);
1734 EXPORT_SYMBOL(skb_pad);
1735 EXPORT_SYMBOL(skb_realloc_headroom);
1736 EXPORT_SYMBOL(skb_under_panic);
1737 EXPORT_SYMBOL(skb_dequeue);
1738 EXPORT_SYMBOL(skb_dequeue_tail);
1739 EXPORT_SYMBOL(skb_insert);
1740 EXPORT_SYMBOL(skb_queue_purge);
1741 EXPORT_SYMBOL(skb_queue_head);
1742 EXPORT_SYMBOL(skb_queue_tail);
1743 EXPORT_SYMBOL(skb_unlink);
1744 EXPORT_SYMBOL(skb_append);
1745 EXPORT_SYMBOL(skb_split);
1746 EXPORT_SYMBOL(skb_prepare_seq_read);
1747 EXPORT_SYMBOL(skb_seq_read);
1748 EXPORT_SYMBOL(skb_abort_seq_read);
1749 EXPORT_SYMBOL(skb_find_text);