2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
8 * Alan Cox : Fixed the worst of the load
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
36 * The functions in this file will not compile correctly with gcc 2.4.x
39 #include <linux/module.h>
40 #include <linux/types.h>
41 #include <linux/kernel.h>
42 #include <linux/kmemcheck.h>
44 #include <linux/interrupt.h>
46 #include <linux/inet.h>
47 #include <linux/slab.h>
48 #include <linux/netdevice.h>
49 #ifdef CONFIG_NET_CLS_ACT
50 #include <net/pkt_sched.h>
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/splice.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
59 #include <linux/errqueue.h>
60 #include <linux/prefetch.h>
62 #include <net/protocol.h>
65 #include <net/checksum.h>
68 #include <asm/uaccess.h>
69 #include <asm/system.h>
70 #include <trace/events/skb.h>
74 static struct kmem_cache *skbuff_head_cache __read_mostly;
75 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
78 * Keep out-of-line to prevent kernel bloat.
79 * __builtin_return_address is not used because it is not always
84 * skb_over_panic - private function
89 * Out of line support code for skb_put(). Not user callable.
91 static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
93 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
94 "data:%p tail:%#lx end:%#lx dev:%s\n",
95 here, skb->len, sz, skb->head, skb->data,
96 (unsigned long)skb->tail, (unsigned long)skb->end,
97 skb->dev ? skb->dev->name : "<NULL>");
102 * skb_under_panic - private function
107 * Out of line support code for skb_push(). Not user callable.
110 static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
112 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
113 "data:%p tail:%#lx end:%#lx dev:%s\n",
114 here, skb->len, sz, skb->head, skb->data,
115 (unsigned long)skb->tail, (unsigned long)skb->end,
116 skb->dev ? skb->dev->name : "<NULL>");
120 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
121 * 'private' fields and also do memory statistics to find all the
127 * __alloc_skb - allocate a network buffer
128 * @size: size to allocate
129 * @gfp_mask: allocation mask
130 * @fclone: allocate from fclone cache instead of head cache
131 * and allocate a cloned (child) skb
132 * @node: numa node to allocate memory on
134 * Allocate a new &sk_buff. The returned buffer has no headroom and a
135 * tail room of size bytes. The object has a reference count of one.
136 * The return is the buffer. On a failure the return is %NULL.
138 * Buffers may only be allocated from interrupts using a @gfp_mask of
141 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
142 int fclone, int node)
144 struct kmem_cache *cache;
145 struct skb_shared_info *shinfo;
149 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
152 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
157 /* We do our best to align skb_shared_info on a separate cache
158 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
159 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
160 * Both skb->head and skb_shared_info are cache line aligned.
162 size = SKB_DATA_ALIGN(size);
163 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
164 data = kmalloc_node_track_caller(size, gfp_mask, node);
167 /* kmalloc(size) might give us more room than requested.
168 * Put skb_shared_info exactly at the end of allocated zone,
169 * to allow max possible filling before reallocation.
171 size = SKB_WITH_OVERHEAD(ksize(data));
172 prefetchw(data + size);
175 * Only clear those fields we need to clear, not those that we will
176 * actually initialise below. Hence, don't put any more fields after
177 * the tail pointer in struct sk_buff!
179 memset(skb, 0, offsetof(struct sk_buff, tail));
180 /* Account for allocated memory : skb + skb->head */
181 skb->truesize = SKB_TRUESIZE(size);
182 atomic_set(&skb->users, 1);
185 skb_reset_tail_pointer(skb);
186 skb->end = skb->tail + size;
187 #ifdef NET_SKBUFF_DATA_USES_OFFSET
188 skb->mac_header = ~0U;
191 /* make sure we initialize shinfo sequentially */
192 shinfo = skb_shinfo(skb);
193 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
194 atomic_set(&shinfo->dataref, 1);
195 kmemcheck_annotate_variable(shinfo->destructor_arg);
198 struct sk_buff *child = skb + 1;
199 atomic_t *fclone_ref = (atomic_t *) (child + 1);
201 kmemcheck_annotate_bitfield(child, flags1);
202 kmemcheck_annotate_bitfield(child, flags2);
203 skb->fclone = SKB_FCLONE_ORIG;
204 atomic_set(fclone_ref, 1);
206 child->fclone = SKB_FCLONE_UNAVAILABLE;
211 kmem_cache_free(cache, skb);
215 EXPORT_SYMBOL(__alloc_skb);
218 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
219 * @dev: network device to receive on
220 * @length: length to allocate
221 * @gfp_mask: get_free_pages mask, passed to alloc_skb
223 * Allocate a new &sk_buff and assign it a usage count of one. The
224 * buffer has unspecified headroom built in. Users should allocate
225 * the headroom they think they need without accounting for the
226 * built in space. The built in space is used for optimisations.
228 * %NULL is returned if there is no free memory.
230 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
231 unsigned int length, gfp_t gfp_mask)
235 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
237 skb_reserve(skb, NET_SKB_PAD);
242 EXPORT_SYMBOL(__netdev_alloc_skb);
244 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
247 skb_fill_page_desc(skb, i, page, off, size);
249 skb->data_len += size;
250 skb->truesize += size;
252 EXPORT_SYMBOL(skb_add_rx_frag);
255 * dev_alloc_skb - allocate an skbuff for receiving
256 * @length: length to allocate
258 * Allocate a new &sk_buff and assign it a usage count of one. The
259 * buffer has unspecified headroom built in. Users should allocate
260 * the headroom they think they need without accounting for the
261 * built in space. The built in space is used for optimisations.
263 * %NULL is returned if there is no free memory. Although this function
264 * allocates memory it can be called from an interrupt.
266 struct sk_buff *dev_alloc_skb(unsigned int length)
269 * There is more code here than it seems:
270 * __dev_alloc_skb is an inline
272 return __dev_alloc_skb(length, GFP_ATOMIC);
274 EXPORT_SYMBOL(dev_alloc_skb);
276 static void skb_drop_list(struct sk_buff **listp)
278 struct sk_buff *list = *listp;
283 struct sk_buff *this = list;
289 static inline void skb_drop_fraglist(struct sk_buff *skb)
291 skb_drop_list(&skb_shinfo(skb)->frag_list);
294 static void skb_clone_fraglist(struct sk_buff *skb)
296 struct sk_buff *list;
298 skb_walk_frags(skb, list)
302 static void skb_release_data(struct sk_buff *skb)
305 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
306 &skb_shinfo(skb)->dataref)) {
307 if (skb_shinfo(skb)->nr_frags) {
309 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
310 skb_frag_unref(skb, i);
314 * If skb buf is from userspace, we need to notify the caller
315 * the lower device DMA has done;
317 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
318 struct ubuf_info *uarg;
320 uarg = skb_shinfo(skb)->destructor_arg;
322 uarg->callback(uarg);
325 if (skb_has_frag_list(skb))
326 skb_drop_fraglist(skb);
333 * Free an skbuff by memory without cleaning the state.
335 static void kfree_skbmem(struct sk_buff *skb)
337 struct sk_buff *other;
338 atomic_t *fclone_ref;
340 switch (skb->fclone) {
341 case SKB_FCLONE_UNAVAILABLE:
342 kmem_cache_free(skbuff_head_cache, skb);
345 case SKB_FCLONE_ORIG:
346 fclone_ref = (atomic_t *) (skb + 2);
347 if (atomic_dec_and_test(fclone_ref))
348 kmem_cache_free(skbuff_fclone_cache, skb);
351 case SKB_FCLONE_CLONE:
352 fclone_ref = (atomic_t *) (skb + 1);
355 /* The clone portion is available for
356 * fast-cloning again.
358 skb->fclone = SKB_FCLONE_UNAVAILABLE;
360 if (atomic_dec_and_test(fclone_ref))
361 kmem_cache_free(skbuff_fclone_cache, other);
366 static void skb_release_head_state(struct sk_buff *skb)
370 secpath_put(skb->sp);
372 if (skb->destructor) {
374 skb->destructor(skb);
376 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
377 nf_conntrack_put(skb->nfct);
379 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
380 nf_conntrack_put_reasm(skb->nfct_reasm);
382 #ifdef CONFIG_BRIDGE_NETFILTER
383 nf_bridge_put(skb->nf_bridge);
385 /* XXX: IS this still necessary? - JHS */
386 #ifdef CONFIG_NET_SCHED
388 #ifdef CONFIG_NET_CLS_ACT
394 /* Free everything but the sk_buff shell. */
395 static void skb_release_all(struct sk_buff *skb)
397 skb_release_head_state(skb);
398 skb_release_data(skb);
402 * __kfree_skb - private function
405 * Free an sk_buff. Release anything attached to the buffer.
406 * Clean the state. This is an internal helper function. Users should
407 * always call kfree_skb
410 void __kfree_skb(struct sk_buff *skb)
412 skb_release_all(skb);
415 EXPORT_SYMBOL(__kfree_skb);
418 * kfree_skb - free an sk_buff
419 * @skb: buffer to free
421 * Drop a reference to the buffer and free it if the usage count has
424 void kfree_skb(struct sk_buff *skb)
428 if (likely(atomic_read(&skb->users) == 1))
430 else if (likely(!atomic_dec_and_test(&skb->users)))
432 trace_kfree_skb(skb, __builtin_return_address(0));
435 EXPORT_SYMBOL(kfree_skb);
438 * consume_skb - free an skbuff
439 * @skb: buffer to free
441 * Drop a ref to the buffer and free it if the usage count has hit zero
442 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
443 * is being dropped after a failure and notes that
445 void consume_skb(struct sk_buff *skb)
449 if (likely(atomic_read(&skb->users) == 1))
451 else if (likely(!atomic_dec_and_test(&skb->users)))
453 trace_consume_skb(skb);
456 EXPORT_SYMBOL(consume_skb);
459 * skb_recycle - clean up an skb for reuse
462 * Recycles the skb to be reused as a receive buffer. This
463 * function does any necessary reference count dropping, and
464 * cleans up the skbuff as if it just came from __alloc_skb().
466 void skb_recycle(struct sk_buff *skb)
468 struct skb_shared_info *shinfo;
470 skb_release_head_state(skb);
472 shinfo = skb_shinfo(skb);
473 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
474 atomic_set(&shinfo->dataref, 1);
476 memset(skb, 0, offsetof(struct sk_buff, tail));
477 skb->data = skb->head + NET_SKB_PAD;
478 skb_reset_tail_pointer(skb);
480 EXPORT_SYMBOL(skb_recycle);
483 * skb_recycle_check - check if skb can be reused for receive
485 * @skb_size: minimum receive buffer size
487 * Checks that the skb passed in is not shared or cloned, and
488 * that it is linear and its head portion at least as large as
489 * skb_size so that it can be recycled as a receive buffer.
490 * If these conditions are met, this function does any necessary
491 * reference count dropping and cleans up the skbuff as if it
492 * just came from __alloc_skb().
494 bool skb_recycle_check(struct sk_buff *skb, int skb_size)
496 if (!skb_is_recycleable(skb, skb_size))
503 EXPORT_SYMBOL(skb_recycle_check);
505 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
507 new->tstamp = old->tstamp;
509 new->transport_header = old->transport_header;
510 new->network_header = old->network_header;
511 new->mac_header = old->mac_header;
512 skb_dst_copy(new, old);
513 new->rxhash = old->rxhash;
514 new->ooo_okay = old->ooo_okay;
515 new->l4_rxhash = old->l4_rxhash;
517 new->sp = secpath_get(old->sp);
519 memcpy(new->cb, old->cb, sizeof(old->cb));
520 new->csum = old->csum;
521 new->local_df = old->local_df;
522 new->pkt_type = old->pkt_type;
523 new->ip_summed = old->ip_summed;
524 skb_copy_queue_mapping(new, old);
525 new->priority = old->priority;
526 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
527 new->ipvs_property = old->ipvs_property;
529 new->protocol = old->protocol;
530 new->mark = old->mark;
531 new->skb_iif = old->skb_iif;
533 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
534 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
535 new->nf_trace = old->nf_trace;
537 #ifdef CONFIG_NET_SCHED
538 new->tc_index = old->tc_index;
539 #ifdef CONFIG_NET_CLS_ACT
540 new->tc_verd = old->tc_verd;
543 new->vlan_tci = old->vlan_tci;
545 skb_copy_secmark(new, old);
549 * You should not add any new code to this function. Add it to
550 * __copy_skb_header above instead.
552 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
554 #define C(x) n->x = skb->x
556 n->next = n->prev = NULL;
558 __copy_skb_header(n, skb);
563 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
566 n->destructor = NULL;
572 atomic_set(&n->users, 1);
574 atomic_inc(&(skb_shinfo(skb)->dataref));
582 * skb_morph - morph one skb into another
583 * @dst: the skb to receive the contents
584 * @src: the skb to supply the contents
586 * This is identical to skb_clone except that the target skb is
587 * supplied by the user.
589 * The target skb is returned upon exit.
591 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
593 skb_release_all(dst);
594 return __skb_clone(dst, src);
596 EXPORT_SYMBOL_GPL(skb_morph);
598 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
599 * @skb: the skb to modify
600 * @gfp_mask: allocation priority
602 * This must be called on SKBTX_DEV_ZEROCOPY skb.
603 * It will copy all frags into kernel and drop the reference
604 * to userspace pages.
606 * If this function is called from an interrupt gfp_mask() must be
609 * Returns 0 on success or a negative error code on failure
610 * to allocate kernel memory to copy to.
612 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
615 int num_frags = skb_shinfo(skb)->nr_frags;
616 struct page *page, *head = NULL;
617 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
619 for (i = 0; i < num_frags; i++) {
621 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
623 page = alloc_page(GFP_ATOMIC);
626 struct page *next = (struct page *)head->private;
632 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
633 memcpy(page_address(page),
634 vaddr + f->page_offset, skb_frag_size(f));
635 kunmap_skb_frag(vaddr);
636 page->private = (unsigned long)head;
640 /* skb frags release userspace buffers */
641 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
642 skb_frag_unref(skb, i);
644 uarg->callback(uarg);
646 /* skb frags point to kernel buffers */
647 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
648 __skb_fill_page_desc(skb, i-1, head, 0,
649 skb_shinfo(skb)->frags[i - 1].size);
650 head = (struct page *)head->private;
653 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
659 * skb_clone - duplicate an sk_buff
660 * @skb: buffer to clone
661 * @gfp_mask: allocation priority
663 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
664 * copies share the same packet data but not structure. The new
665 * buffer has a reference count of 1. If the allocation fails the
666 * function returns %NULL otherwise the new buffer is returned.
668 * If this function is called from an interrupt gfp_mask() must be
672 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
676 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
677 if (skb_copy_ubufs(skb, gfp_mask))
682 if (skb->fclone == SKB_FCLONE_ORIG &&
683 n->fclone == SKB_FCLONE_UNAVAILABLE) {
684 atomic_t *fclone_ref = (atomic_t *) (n + 1);
685 n->fclone = SKB_FCLONE_CLONE;
686 atomic_inc(fclone_ref);
688 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
692 kmemcheck_annotate_bitfield(n, flags1);
693 kmemcheck_annotate_bitfield(n, flags2);
694 n->fclone = SKB_FCLONE_UNAVAILABLE;
697 return __skb_clone(n, skb);
699 EXPORT_SYMBOL(skb_clone);
701 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
703 #ifndef NET_SKBUFF_DATA_USES_OFFSET
705 * Shift between the two data areas in bytes
707 unsigned long offset = new->data - old->data;
710 __copy_skb_header(new, old);
712 #ifndef NET_SKBUFF_DATA_USES_OFFSET
713 /* {transport,network,mac}_header are relative to skb->head */
714 new->transport_header += offset;
715 new->network_header += offset;
716 if (skb_mac_header_was_set(new))
717 new->mac_header += offset;
719 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
720 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
721 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
725 * skb_copy - create private copy of an sk_buff
726 * @skb: buffer to copy
727 * @gfp_mask: allocation priority
729 * Make a copy of both an &sk_buff and its data. This is used when the
730 * caller wishes to modify the data and needs a private copy of the
731 * data to alter. Returns %NULL on failure or the pointer to the buffer
732 * on success. The returned buffer has a reference count of 1.
734 * As by-product this function converts non-linear &sk_buff to linear
735 * one, so that &sk_buff becomes completely private and caller is allowed
736 * to modify all the data of returned buffer. This means that this
737 * function is not recommended for use in circumstances when only
738 * header is going to be modified. Use pskb_copy() instead.
741 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
743 int headerlen = skb_headroom(skb);
744 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
745 struct sk_buff *n = alloc_skb(size, gfp_mask);
750 /* Set the data pointer */
751 skb_reserve(n, headerlen);
752 /* Set the tail pointer and length */
753 skb_put(n, skb->len);
755 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
758 copy_skb_header(n, skb);
761 EXPORT_SYMBOL(skb_copy);
764 * pskb_copy - create copy of an sk_buff with private head.
765 * @skb: buffer to copy
766 * @gfp_mask: allocation priority
768 * Make a copy of both an &sk_buff and part of its data, located
769 * in header. Fragmented data remain shared. This is used when
770 * the caller wishes to modify only header of &sk_buff and needs
771 * private copy of the header to alter. Returns %NULL on failure
772 * or the pointer to the buffer on success.
773 * The returned buffer has a reference count of 1.
776 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
778 unsigned int size = skb_end_pointer(skb) - skb->head;
779 struct sk_buff *n = alloc_skb(size, gfp_mask);
784 /* Set the data pointer */
785 skb_reserve(n, skb_headroom(skb));
786 /* Set the tail pointer and length */
787 skb_put(n, skb_headlen(skb));
789 skb_copy_from_linear_data(skb, n->data, n->len);
791 n->truesize += skb->data_len;
792 n->data_len = skb->data_len;
795 if (skb_shinfo(skb)->nr_frags) {
798 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
799 if (skb_copy_ubufs(skb, gfp_mask)) {
805 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
806 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
807 skb_frag_ref(skb, i);
809 skb_shinfo(n)->nr_frags = i;
812 if (skb_has_frag_list(skb)) {
813 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
814 skb_clone_fraglist(n);
817 copy_skb_header(n, skb);
821 EXPORT_SYMBOL(pskb_copy);
824 * pskb_expand_head - reallocate header of &sk_buff
825 * @skb: buffer to reallocate
826 * @nhead: room to add at head
827 * @ntail: room to add at tail
828 * @gfp_mask: allocation priority
830 * Expands (or creates identical copy, if &nhead and &ntail are zero)
831 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
832 * reference count of 1. Returns zero in the case of success or error,
833 * if expansion failed. In the last case, &sk_buff is not changed.
835 * All the pointers pointing into skb header may change and must be
836 * reloaded after call to this function.
839 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
844 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
853 size = SKB_DATA_ALIGN(size);
855 /* Check if we can avoid taking references on fragments if we own
856 * the last reference on skb->head. (see skb_release_data())
861 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
862 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
866 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
867 memmove(skb->head + size, skb_shinfo(skb),
868 offsetof(struct skb_shared_info,
869 frags[skb_shinfo(skb)->nr_frags]));
870 memmove(skb->head + nhead, skb->head,
871 skb_tail_pointer(skb) - skb->head);
876 data = kmalloc(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
880 size = SKB_WITH_OVERHEAD(ksize(data));
882 /* Copy only real data... and, alas, header. This should be
883 * optimized for the cases when header is void.
885 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
887 memcpy((struct skb_shared_info *)(data + size),
889 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
894 /* copy this zero copy skb frags */
895 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
896 if (skb_copy_ubufs(skb, gfp_mask))
899 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
900 skb_frag_ref(skb, i);
902 if (skb_has_frag_list(skb))
903 skb_clone_fraglist(skb);
905 skb_release_data(skb);
907 off = (data + nhead) - skb->head;
912 #ifdef NET_SKBUFF_DATA_USES_OFFSET
916 skb->end = skb->head + size;
918 /* {transport,network,mac}_header and tail are relative to skb->head */
920 skb->transport_header += off;
921 skb->network_header += off;
922 if (skb_mac_header_was_set(skb))
923 skb->mac_header += off;
924 /* Only adjust this if it actually is csum_start rather than csum */
925 if (skb->ip_summed == CHECKSUM_PARTIAL)
926 skb->csum_start += nhead;
930 atomic_set(&skb_shinfo(skb)->dataref, 1);
938 EXPORT_SYMBOL(pskb_expand_head);
940 /* Make private copy of skb with writable head and some headroom */
942 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
944 struct sk_buff *skb2;
945 int delta = headroom - skb_headroom(skb);
948 skb2 = pskb_copy(skb, GFP_ATOMIC);
950 skb2 = skb_clone(skb, GFP_ATOMIC);
951 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
959 EXPORT_SYMBOL(skb_realloc_headroom);
962 * skb_copy_expand - copy and expand sk_buff
963 * @skb: buffer to copy
964 * @newheadroom: new free bytes at head
965 * @newtailroom: new free bytes at tail
966 * @gfp_mask: allocation priority
968 * Make a copy of both an &sk_buff and its data and while doing so
969 * allocate additional space.
971 * This is used when the caller wishes to modify the data and needs a
972 * private copy of the data to alter as well as more space for new fields.
973 * Returns %NULL on failure or the pointer to the buffer
974 * on success. The returned buffer has a reference count of 1.
976 * You must pass %GFP_ATOMIC as the allocation priority if this function
977 * is called from an interrupt.
979 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
980 int newheadroom, int newtailroom,
984 * Allocate the copy buffer
986 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
988 int oldheadroom = skb_headroom(skb);
989 int head_copy_len, head_copy_off;
995 skb_reserve(n, newheadroom);
997 /* Set the tail pointer and length */
998 skb_put(n, skb->len);
1000 head_copy_len = oldheadroom;
1002 if (newheadroom <= head_copy_len)
1003 head_copy_len = newheadroom;
1005 head_copy_off = newheadroom - head_copy_len;
1007 /* Copy the linear header and data. */
1008 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1009 skb->len + head_copy_len))
1012 copy_skb_header(n, skb);
1014 off = newheadroom - oldheadroom;
1015 if (n->ip_summed == CHECKSUM_PARTIAL)
1016 n->csum_start += off;
1017 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1018 n->transport_header += off;
1019 n->network_header += off;
1020 if (skb_mac_header_was_set(skb))
1021 n->mac_header += off;
1026 EXPORT_SYMBOL(skb_copy_expand);
1029 * skb_pad - zero pad the tail of an skb
1030 * @skb: buffer to pad
1031 * @pad: space to pad
1033 * Ensure that a buffer is followed by a padding area that is zero
1034 * filled. Used by network drivers which may DMA or transfer data
1035 * beyond the buffer end onto the wire.
1037 * May return error in out of memory cases. The skb is freed on error.
1040 int skb_pad(struct sk_buff *skb, int pad)
1045 /* If the skbuff is non linear tailroom is always zero.. */
1046 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1047 memset(skb->data+skb->len, 0, pad);
1051 ntail = skb->data_len + pad - (skb->end - skb->tail);
1052 if (likely(skb_cloned(skb) || ntail > 0)) {
1053 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1058 /* FIXME: The use of this function with non-linear skb's really needs
1061 err = skb_linearize(skb);
1065 memset(skb->data + skb->len, 0, pad);
1072 EXPORT_SYMBOL(skb_pad);
1075 * skb_put - add data to a buffer
1076 * @skb: buffer to use
1077 * @len: amount of data to add
1079 * This function extends the used data area of the buffer. If this would
1080 * exceed the total buffer size the kernel will panic. A pointer to the
1081 * first byte of the extra data is returned.
1083 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1085 unsigned char *tmp = skb_tail_pointer(skb);
1086 SKB_LINEAR_ASSERT(skb);
1089 if (unlikely(skb->tail > skb->end))
1090 skb_over_panic(skb, len, __builtin_return_address(0));
1093 EXPORT_SYMBOL(skb_put);
1096 * skb_push - add data to the start of a buffer
1097 * @skb: buffer to use
1098 * @len: amount of data to add
1100 * This function extends the used data area of the buffer at the buffer
1101 * start. If this would exceed the total buffer headroom the kernel will
1102 * panic. A pointer to the first byte of the extra data is returned.
1104 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1108 if (unlikely(skb->data<skb->head))
1109 skb_under_panic(skb, len, __builtin_return_address(0));
1112 EXPORT_SYMBOL(skb_push);
1115 * skb_pull - remove data from the start of a buffer
1116 * @skb: buffer to use
1117 * @len: amount of data to remove
1119 * This function removes data from the start of a buffer, returning
1120 * the memory to the headroom. A pointer to the next data in the buffer
1121 * is returned. Once the data has been pulled future pushes will overwrite
1124 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1126 return skb_pull_inline(skb, len);
1128 EXPORT_SYMBOL(skb_pull);
1131 * skb_trim - remove end from a buffer
1132 * @skb: buffer to alter
1135 * Cut the length of a buffer down by removing data from the tail. If
1136 * the buffer is already under the length specified it is not modified.
1137 * The skb must be linear.
1139 void skb_trim(struct sk_buff *skb, unsigned int len)
1142 __skb_trim(skb, len);
1144 EXPORT_SYMBOL(skb_trim);
1146 /* Trims skb to length len. It can change skb pointers.
1149 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1151 struct sk_buff **fragp;
1152 struct sk_buff *frag;
1153 int offset = skb_headlen(skb);
1154 int nfrags = skb_shinfo(skb)->nr_frags;
1158 if (skb_cloned(skb) &&
1159 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1166 for (; i < nfrags; i++) {
1167 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1174 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1177 skb_shinfo(skb)->nr_frags = i;
1179 for (; i < nfrags; i++)
1180 skb_frag_unref(skb, i);
1182 if (skb_has_frag_list(skb))
1183 skb_drop_fraglist(skb);
1187 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1188 fragp = &frag->next) {
1189 int end = offset + frag->len;
1191 if (skb_shared(frag)) {
1192 struct sk_buff *nfrag;
1194 nfrag = skb_clone(frag, GFP_ATOMIC);
1195 if (unlikely(!nfrag))
1198 nfrag->next = frag->next;
1210 unlikely((err = pskb_trim(frag, len - offset))))
1214 skb_drop_list(&frag->next);
1219 if (len > skb_headlen(skb)) {
1220 skb->data_len -= skb->len - len;
1225 skb_set_tail_pointer(skb, len);
1230 EXPORT_SYMBOL(___pskb_trim);
1233 * __pskb_pull_tail - advance tail of skb header
1234 * @skb: buffer to reallocate
1235 * @delta: number of bytes to advance tail
1237 * The function makes a sense only on a fragmented &sk_buff,
1238 * it expands header moving its tail forward and copying necessary
1239 * data from fragmented part.
1241 * &sk_buff MUST have reference count of 1.
1243 * Returns %NULL (and &sk_buff does not change) if pull failed
1244 * or value of new tail of skb in the case of success.
1246 * All the pointers pointing into skb header may change and must be
1247 * reloaded after call to this function.
1250 /* Moves tail of skb head forward, copying data from fragmented part,
1251 * when it is necessary.
1252 * 1. It may fail due to malloc failure.
1253 * 2. It may change skb pointers.
1255 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1257 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1259 /* If skb has not enough free space at tail, get new one
1260 * plus 128 bytes for future expansions. If we have enough
1261 * room at tail, reallocate without expansion only if skb is cloned.
1263 int i, k, eat = (skb->tail + delta) - skb->end;
1265 if (eat > 0 || skb_cloned(skb)) {
1266 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1271 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1274 /* Optimization: no fragments, no reasons to preestimate
1275 * size of pulled pages. Superb.
1277 if (!skb_has_frag_list(skb))
1280 /* Estimate size of pulled pages. */
1282 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1283 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1290 /* If we need update frag list, we are in troubles.
1291 * Certainly, it possible to add an offset to skb data,
1292 * but taking into account that pulling is expected to
1293 * be very rare operation, it is worth to fight against
1294 * further bloating skb head and crucify ourselves here instead.
1295 * Pure masohism, indeed. 8)8)
1298 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1299 struct sk_buff *clone = NULL;
1300 struct sk_buff *insp = NULL;
1305 if (list->len <= eat) {
1306 /* Eaten as whole. */
1311 /* Eaten partially. */
1313 if (skb_shared(list)) {
1314 /* Sucks! We need to fork list. :-( */
1315 clone = skb_clone(list, GFP_ATOMIC);
1321 /* This may be pulled without
1325 if (!pskb_pull(list, eat)) {
1333 /* Free pulled out fragments. */
1334 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1335 skb_shinfo(skb)->frag_list = list->next;
1338 /* And insert new clone at head. */
1341 skb_shinfo(skb)->frag_list = clone;
1344 /* Success! Now we may commit changes to skb data. */
1349 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1350 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1353 skb_frag_unref(skb, i);
1356 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1358 skb_shinfo(skb)->frags[k].page_offset += eat;
1359 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1365 skb_shinfo(skb)->nr_frags = k;
1368 skb->data_len -= delta;
1370 return skb_tail_pointer(skb);
1372 EXPORT_SYMBOL(__pskb_pull_tail);
1375 * skb_copy_bits - copy bits from skb to kernel buffer
1377 * @offset: offset in source
1378 * @to: destination buffer
1379 * @len: number of bytes to copy
1381 * Copy the specified number of bytes from the source skb to the
1382 * destination buffer.
1385 * If its prototype is ever changed,
1386 * check arch/{*}/net/{*}.S files,
1387 * since it is called from BPF assembly code.
1389 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1391 int start = skb_headlen(skb);
1392 struct sk_buff *frag_iter;
1395 if (offset > (int)skb->len - len)
1399 if ((copy = start - offset) > 0) {
1402 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1403 if ((len -= copy) == 0)
1409 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1412 WARN_ON(start > offset + len);
1414 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1415 if ((copy = end - offset) > 0) {
1421 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1423 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1424 offset - start, copy);
1425 kunmap_skb_frag(vaddr);
1427 if ((len -= copy) == 0)
1435 skb_walk_frags(skb, frag_iter) {
1438 WARN_ON(start > offset + len);
1440 end = start + frag_iter->len;
1441 if ((copy = end - offset) > 0) {
1444 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1446 if ((len -= copy) == 0)
1460 EXPORT_SYMBOL(skb_copy_bits);
1463 * Callback from splice_to_pipe(), if we need to release some pages
1464 * at the end of the spd in case we error'ed out in filling the pipe.
1466 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1468 put_page(spd->pages[i]);
1471 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1472 unsigned int *offset,
1473 struct sk_buff *skb, struct sock *sk)
1475 struct page *p = sk->sk_sndmsg_page;
1480 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1484 off = sk->sk_sndmsg_off = 0;
1485 /* hold one ref to this page until it's full */
1489 off = sk->sk_sndmsg_off;
1490 mlen = PAGE_SIZE - off;
1491 if (mlen < 64 && mlen < *len) {
1496 *len = min_t(unsigned int, *len, mlen);
1499 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1500 sk->sk_sndmsg_off += *len;
1508 * Fill page/offset/length into spd, if it can hold more pages.
1510 static inline int spd_fill_page(struct splice_pipe_desc *spd,
1511 struct pipe_inode_info *pipe, struct page *page,
1512 unsigned int *len, unsigned int offset,
1513 struct sk_buff *skb, int linear,
1516 if (unlikely(spd->nr_pages == pipe->buffers))
1520 page = linear_to_page(page, len, &offset, skb, sk);
1526 spd->pages[spd->nr_pages] = page;
1527 spd->partial[spd->nr_pages].len = *len;
1528 spd->partial[spd->nr_pages].offset = offset;
1534 static inline void __segment_seek(struct page **page, unsigned int *poff,
1535 unsigned int *plen, unsigned int off)
1540 n = *poff / PAGE_SIZE;
1542 *page = nth_page(*page, n);
1544 *poff = *poff % PAGE_SIZE;
1548 static inline int __splice_segment(struct page *page, unsigned int poff,
1549 unsigned int plen, unsigned int *off,
1550 unsigned int *len, struct sk_buff *skb,
1551 struct splice_pipe_desc *spd, int linear,
1553 struct pipe_inode_info *pipe)
1558 /* skip this segment if already processed */
1564 /* ignore any bits we already processed */
1566 __segment_seek(&page, &poff, &plen, *off);
1571 unsigned int flen = min(*len, plen);
1573 /* the linear region may spread across several pages */
1574 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1576 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1579 __segment_seek(&page, &poff, &plen, flen);
1582 } while (*len && plen);
1588 * Map linear and fragment data from the skb to spd. It reports failure if the
1589 * pipe is full or if we already spliced the requested length.
1591 static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1592 unsigned int *offset, unsigned int *len,
1593 struct splice_pipe_desc *spd, struct sock *sk)
1598 * map the linear part
1600 if (__splice_segment(virt_to_page(skb->data),
1601 (unsigned long) skb->data & (PAGE_SIZE - 1),
1603 offset, len, skb, spd, 1, sk, pipe))
1607 * then map the fragments
1609 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1610 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1612 if (__splice_segment(skb_frag_page(f),
1613 f->page_offset, skb_frag_size(f),
1614 offset, len, skb, spd, 0, sk, pipe))
1622 * Map data from the skb to a pipe. Should handle both the linear part,
1623 * the fragments, and the frag list. It does NOT handle frag lists within
1624 * the frag list, if such a thing exists. We'd probably need to recurse to
1625 * handle that cleanly.
1627 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1628 struct pipe_inode_info *pipe, unsigned int tlen,
1631 struct partial_page partial[PIPE_DEF_BUFFERS];
1632 struct page *pages[PIPE_DEF_BUFFERS];
1633 struct splice_pipe_desc spd = {
1636 .nr_pages_max = MAX_SKB_FRAGS,
1638 .ops = &nosteal_pipe_buf_ops,
1639 .spd_release = sock_spd_release,
1641 struct sk_buff *frag_iter;
1642 struct sock *sk = skb->sk;
1645 if (splice_grow_spd(pipe, &spd))
1649 * __skb_splice_bits() only fails if the output has no room left,
1650 * so no point in going over the frag_list for the error case.
1652 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1658 * now see if we have a frag_list to map
1660 skb_walk_frags(skb, frag_iter) {
1663 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1670 * Drop the socket lock, otherwise we have reverse
1671 * locking dependencies between sk_lock and i_mutex
1672 * here as compared to sendfile(). We enter here
1673 * with the socket lock held, and splice_to_pipe() will
1674 * grab the pipe inode lock. For sendfile() emulation,
1675 * we call into ->sendpage() with the i_mutex lock held
1676 * and networking will grab the socket lock.
1679 ret = splice_to_pipe(pipe, &spd);
1683 splice_shrink_spd(&spd);
1688 * skb_store_bits - store bits from kernel buffer to skb
1689 * @skb: destination buffer
1690 * @offset: offset in destination
1691 * @from: source buffer
1692 * @len: number of bytes to copy
1694 * Copy the specified number of bytes from the source buffer to the
1695 * destination skb. This function handles all the messy bits of
1696 * traversing fragment lists and such.
1699 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1701 int start = skb_headlen(skb);
1702 struct sk_buff *frag_iter;
1705 if (offset > (int)skb->len - len)
1708 if ((copy = start - offset) > 0) {
1711 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1712 if ((len -= copy) == 0)
1718 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1719 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1722 WARN_ON(start > offset + len);
1724 end = start + skb_frag_size(frag);
1725 if ((copy = end - offset) > 0) {
1731 vaddr = kmap_skb_frag(frag);
1732 memcpy(vaddr + frag->page_offset + offset - start,
1734 kunmap_skb_frag(vaddr);
1736 if ((len -= copy) == 0)
1744 skb_walk_frags(skb, frag_iter) {
1747 WARN_ON(start > offset + len);
1749 end = start + frag_iter->len;
1750 if ((copy = end - offset) > 0) {
1753 if (skb_store_bits(frag_iter, offset - start,
1756 if ((len -= copy) == 0)
1769 EXPORT_SYMBOL(skb_store_bits);
1771 /* Checksum skb data. */
1773 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1774 int len, __wsum csum)
1776 int start = skb_headlen(skb);
1777 int i, copy = start - offset;
1778 struct sk_buff *frag_iter;
1781 /* Checksum header. */
1785 csum = csum_partial(skb->data + offset, copy, csum);
1786 if ((len -= copy) == 0)
1792 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1795 WARN_ON(start > offset + len);
1797 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1798 if ((copy = end - offset) > 0) {
1801 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1805 vaddr = kmap_skb_frag(frag);
1806 csum2 = csum_partial(vaddr + frag->page_offset +
1807 offset - start, copy, 0);
1808 kunmap_skb_frag(vaddr);
1809 csum = csum_block_add(csum, csum2, pos);
1818 skb_walk_frags(skb, frag_iter) {
1821 WARN_ON(start > offset + len);
1823 end = start + frag_iter->len;
1824 if ((copy = end - offset) > 0) {
1828 csum2 = skb_checksum(frag_iter, offset - start,
1830 csum = csum_block_add(csum, csum2, pos);
1831 if ((len -= copy) == 0)
1842 EXPORT_SYMBOL(skb_checksum);
1844 /* Both of above in one bottle. */
1846 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1847 u8 *to, int len, __wsum csum)
1849 int start = skb_headlen(skb);
1850 int i, copy = start - offset;
1851 struct sk_buff *frag_iter;
1858 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1860 if ((len -= copy) == 0)
1867 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1870 WARN_ON(start > offset + len);
1872 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1873 if ((copy = end - offset) > 0) {
1876 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1880 vaddr = kmap_skb_frag(frag);
1881 csum2 = csum_partial_copy_nocheck(vaddr +
1885 kunmap_skb_frag(vaddr);
1886 csum = csum_block_add(csum, csum2, pos);
1896 skb_walk_frags(skb, frag_iter) {
1900 WARN_ON(start > offset + len);
1902 end = start + frag_iter->len;
1903 if ((copy = end - offset) > 0) {
1906 csum2 = skb_copy_and_csum_bits(frag_iter,
1909 csum = csum_block_add(csum, csum2, pos);
1910 if ((len -= copy) == 0)
1921 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1923 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1928 if (skb->ip_summed == CHECKSUM_PARTIAL)
1929 csstart = skb_checksum_start_offset(skb);
1931 csstart = skb_headlen(skb);
1933 BUG_ON(csstart > skb_headlen(skb));
1935 skb_copy_from_linear_data(skb, to, csstart);
1938 if (csstart != skb->len)
1939 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1940 skb->len - csstart, 0);
1942 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1943 long csstuff = csstart + skb->csum_offset;
1945 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1948 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1951 * skb_dequeue - remove from the head of the queue
1952 * @list: list to dequeue from
1954 * Remove the head of the list. The list lock is taken so the function
1955 * may be used safely with other locking list functions. The head item is
1956 * returned or %NULL if the list is empty.
1959 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1961 unsigned long flags;
1962 struct sk_buff *result;
1964 spin_lock_irqsave(&list->lock, flags);
1965 result = __skb_dequeue(list);
1966 spin_unlock_irqrestore(&list->lock, flags);
1969 EXPORT_SYMBOL(skb_dequeue);
1972 * skb_dequeue_tail - remove from the tail of the queue
1973 * @list: list to dequeue from
1975 * Remove the tail of the list. The list lock is taken so the function
1976 * may be used safely with other locking list functions. The tail item is
1977 * returned or %NULL if the list is empty.
1979 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1981 unsigned long flags;
1982 struct sk_buff *result;
1984 spin_lock_irqsave(&list->lock, flags);
1985 result = __skb_dequeue_tail(list);
1986 spin_unlock_irqrestore(&list->lock, flags);
1989 EXPORT_SYMBOL(skb_dequeue_tail);
1992 * skb_queue_purge - empty a list
1993 * @list: list to empty
1995 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1996 * the list and one reference dropped. This function takes the list
1997 * lock and is atomic with respect to other list locking functions.
1999 void skb_queue_purge(struct sk_buff_head *list)
2001 struct sk_buff *skb;
2002 while ((skb = skb_dequeue(list)) != NULL)
2005 EXPORT_SYMBOL(skb_queue_purge);
2008 * skb_queue_head - queue a buffer at the list head
2009 * @list: list to use
2010 * @newsk: buffer to queue
2012 * Queue a buffer at the start of the list. This function takes the
2013 * list lock and can be used safely with other locking &sk_buff functions
2016 * A buffer cannot be placed on two lists at the same time.
2018 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2020 unsigned long flags;
2022 spin_lock_irqsave(&list->lock, flags);
2023 __skb_queue_head(list, newsk);
2024 spin_unlock_irqrestore(&list->lock, flags);
2026 EXPORT_SYMBOL(skb_queue_head);
2029 * skb_queue_tail - queue a buffer at the list tail
2030 * @list: list to use
2031 * @newsk: buffer to queue
2033 * Queue a buffer at the tail of the list. This function takes the
2034 * list lock and can be used safely with other locking &sk_buff functions
2037 * A buffer cannot be placed on two lists at the same time.
2039 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2041 unsigned long flags;
2043 spin_lock_irqsave(&list->lock, flags);
2044 __skb_queue_tail(list, newsk);
2045 spin_unlock_irqrestore(&list->lock, flags);
2047 EXPORT_SYMBOL(skb_queue_tail);
2050 * skb_unlink - remove a buffer from a list
2051 * @skb: buffer to remove
2052 * @list: list to use
2054 * Remove a packet from a list. The list locks are taken and this
2055 * function is atomic with respect to other list locked calls
2057 * You must know what list the SKB is on.
2059 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2061 unsigned long flags;
2063 spin_lock_irqsave(&list->lock, flags);
2064 __skb_unlink(skb, list);
2065 spin_unlock_irqrestore(&list->lock, flags);
2067 EXPORT_SYMBOL(skb_unlink);
2070 * skb_append - append a buffer
2071 * @old: buffer to insert after
2072 * @newsk: buffer to insert
2073 * @list: list to use
2075 * Place a packet after a given packet in a list. The list locks are taken
2076 * and this function is atomic with respect to other list locked calls.
2077 * A buffer cannot be placed on two lists at the same time.
2079 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2081 unsigned long flags;
2083 spin_lock_irqsave(&list->lock, flags);
2084 __skb_queue_after(list, old, newsk);
2085 spin_unlock_irqrestore(&list->lock, flags);
2087 EXPORT_SYMBOL(skb_append);
2090 * skb_insert - insert a buffer
2091 * @old: buffer to insert before
2092 * @newsk: buffer to insert
2093 * @list: list to use
2095 * Place a packet before a given packet in a list. The list locks are
2096 * taken and this function is atomic with respect to other list locked
2099 * A buffer cannot be placed on two lists at the same time.
2101 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2103 unsigned long flags;
2105 spin_lock_irqsave(&list->lock, flags);
2106 __skb_insert(newsk, old->prev, old, list);
2107 spin_unlock_irqrestore(&list->lock, flags);
2109 EXPORT_SYMBOL(skb_insert);
2111 static inline void skb_split_inside_header(struct sk_buff *skb,
2112 struct sk_buff* skb1,
2113 const u32 len, const int pos)
2117 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2119 /* And move data appendix as is. */
2120 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2121 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2123 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2124 skb_shinfo(skb)->nr_frags = 0;
2125 skb1->data_len = skb->data_len;
2126 skb1->len += skb1->data_len;
2129 skb_set_tail_pointer(skb, len);
2132 static inline void skb_split_no_header(struct sk_buff *skb,
2133 struct sk_buff* skb1,
2134 const u32 len, int pos)
2137 const int nfrags = skb_shinfo(skb)->nr_frags;
2139 skb_shinfo(skb)->nr_frags = 0;
2140 skb1->len = skb1->data_len = skb->len - len;
2142 skb->data_len = len - pos;
2144 for (i = 0; i < nfrags; i++) {
2145 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2147 if (pos + size > len) {
2148 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2152 * We have two variants in this case:
2153 * 1. Move all the frag to the second
2154 * part, if it is possible. F.e.
2155 * this approach is mandatory for TUX,
2156 * where splitting is expensive.
2157 * 2. Split is accurately. We make this.
2159 skb_frag_ref(skb, i);
2160 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2161 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2162 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2163 skb_shinfo(skb)->nr_frags++;
2167 skb_shinfo(skb)->nr_frags++;
2170 skb_shinfo(skb1)->nr_frags = k;
2174 * skb_split - Split fragmented skb to two parts at length len.
2175 * @skb: the buffer to split
2176 * @skb1: the buffer to receive the second part
2177 * @len: new length for skb
2179 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2181 int pos = skb_headlen(skb);
2183 if (len < pos) /* Split line is inside header. */
2184 skb_split_inside_header(skb, skb1, len, pos);
2185 else /* Second chunk has no header, nothing to copy. */
2186 skb_split_no_header(skb, skb1, len, pos);
2188 EXPORT_SYMBOL(skb_split);
2190 /* Shifting from/to a cloned skb is a no-go.
2192 * Caller cannot keep skb_shinfo related pointers past calling here!
2194 static int skb_prepare_for_shift(struct sk_buff *skb)
2196 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2200 * skb_shift - Shifts paged data partially from skb to another
2201 * @tgt: buffer into which tail data gets added
2202 * @skb: buffer from which the paged data comes from
2203 * @shiftlen: shift up to this many bytes
2205 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2206 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2207 * It's up to caller to free skb if everything was shifted.
2209 * If @tgt runs out of frags, the whole operation is aborted.
2211 * Skb cannot include anything else but paged data while tgt is allowed
2212 * to have non-paged data as well.
2214 * TODO: full sized shift could be optimized but that would need
2215 * specialized skb free'er to handle frags without up-to-date nr_frags.
2217 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2219 int from, to, merge, todo;
2220 struct skb_frag_struct *fragfrom, *fragto;
2222 BUG_ON(shiftlen > skb->len);
2223 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2227 to = skb_shinfo(tgt)->nr_frags;
2228 fragfrom = &skb_shinfo(skb)->frags[from];
2230 /* Actual merge is delayed until the point when we know we can
2231 * commit all, so that we don't have to undo partial changes
2234 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2235 fragfrom->page_offset)) {
2240 todo -= skb_frag_size(fragfrom);
2242 if (skb_prepare_for_shift(skb) ||
2243 skb_prepare_for_shift(tgt))
2246 /* All previous frag pointers might be stale! */
2247 fragfrom = &skb_shinfo(skb)->frags[from];
2248 fragto = &skb_shinfo(tgt)->frags[merge];
2250 skb_frag_size_add(fragto, shiftlen);
2251 skb_frag_size_sub(fragfrom, shiftlen);
2252 fragfrom->page_offset += shiftlen;
2260 /* Skip full, not-fitting skb to avoid expensive operations */
2261 if ((shiftlen == skb->len) &&
2262 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2265 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2268 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2269 if (to == MAX_SKB_FRAGS)
2272 fragfrom = &skb_shinfo(skb)->frags[from];
2273 fragto = &skb_shinfo(tgt)->frags[to];
2275 if (todo >= skb_frag_size(fragfrom)) {
2276 *fragto = *fragfrom;
2277 todo -= skb_frag_size(fragfrom);
2282 __skb_frag_ref(fragfrom);
2283 fragto->page = fragfrom->page;
2284 fragto->page_offset = fragfrom->page_offset;
2285 skb_frag_size_set(fragto, todo);
2287 fragfrom->page_offset += todo;
2288 skb_frag_size_sub(fragfrom, todo);
2296 /* Ready to "commit" this state change to tgt */
2297 skb_shinfo(tgt)->nr_frags = to;
2300 fragfrom = &skb_shinfo(skb)->frags[0];
2301 fragto = &skb_shinfo(tgt)->frags[merge];
2303 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2304 __skb_frag_unref(fragfrom);
2307 /* Reposition in the original skb */
2309 while (from < skb_shinfo(skb)->nr_frags)
2310 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2311 skb_shinfo(skb)->nr_frags = to;
2313 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2316 /* Most likely the tgt won't ever need its checksum anymore, skb on
2317 * the other hand might need it if it needs to be resent
2319 tgt->ip_summed = CHECKSUM_PARTIAL;
2320 skb->ip_summed = CHECKSUM_PARTIAL;
2322 /* Yak, is it really working this way? Some helper please? */
2323 skb->len -= shiftlen;
2324 skb->data_len -= shiftlen;
2325 skb->truesize -= shiftlen;
2326 tgt->len += shiftlen;
2327 tgt->data_len += shiftlen;
2328 tgt->truesize += shiftlen;
2334 * skb_prepare_seq_read - Prepare a sequential read of skb data
2335 * @skb: the buffer to read
2336 * @from: lower offset of data to be read
2337 * @to: upper offset of data to be read
2338 * @st: state variable
2340 * Initializes the specified state variable. Must be called before
2341 * invoking skb_seq_read() for the first time.
2343 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2344 unsigned int to, struct skb_seq_state *st)
2346 st->lower_offset = from;
2347 st->upper_offset = to;
2348 st->root_skb = st->cur_skb = skb;
2349 st->frag_idx = st->stepped_offset = 0;
2350 st->frag_data = NULL;
2352 EXPORT_SYMBOL(skb_prepare_seq_read);
2355 * skb_seq_read - Sequentially read skb data
2356 * @consumed: number of bytes consumed by the caller so far
2357 * @data: destination pointer for data to be returned
2358 * @st: state variable
2360 * Reads a block of skb data at &consumed relative to the
2361 * lower offset specified to skb_prepare_seq_read(). Assigns
2362 * the head of the data block to &data and returns the length
2363 * of the block or 0 if the end of the skb data or the upper
2364 * offset has been reached.
2366 * The caller is not required to consume all of the data
2367 * returned, i.e. &consumed is typically set to the number
2368 * of bytes already consumed and the next call to
2369 * skb_seq_read() will return the remaining part of the block.
2371 * Note 1: The size of each block of data returned can be arbitrary,
2372 * this limitation is the cost for zerocopy seqeuental
2373 * reads of potentially non linear data.
2375 * Note 2: Fragment lists within fragments are not implemented
2376 * at the moment, state->root_skb could be replaced with
2377 * a stack for this purpose.
2379 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2380 struct skb_seq_state *st)
2382 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2385 if (unlikely(abs_offset >= st->upper_offset))
2389 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2391 if (abs_offset < block_limit && !st->frag_data) {
2392 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2393 return block_limit - abs_offset;
2396 if (st->frag_idx == 0 && !st->frag_data)
2397 st->stepped_offset += skb_headlen(st->cur_skb);
2399 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2400 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2401 block_limit = skb_frag_size(frag) + st->stepped_offset;
2403 if (abs_offset < block_limit) {
2405 st->frag_data = kmap_skb_frag(frag);
2407 *data = (u8 *) st->frag_data + frag->page_offset +
2408 (abs_offset - st->stepped_offset);
2410 return block_limit - abs_offset;
2413 if (st->frag_data) {
2414 kunmap_skb_frag(st->frag_data);
2415 st->frag_data = NULL;
2419 st->stepped_offset += skb_frag_size(frag);
2422 if (st->frag_data) {
2423 kunmap_skb_frag(st->frag_data);
2424 st->frag_data = NULL;
2427 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2428 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2431 } else if (st->cur_skb->next) {
2432 st->cur_skb = st->cur_skb->next;
2439 EXPORT_SYMBOL(skb_seq_read);
2442 * skb_abort_seq_read - Abort a sequential read of skb data
2443 * @st: state variable
2445 * Must be called if skb_seq_read() was not called until it
2448 void skb_abort_seq_read(struct skb_seq_state *st)
2451 kunmap_skb_frag(st->frag_data);
2453 EXPORT_SYMBOL(skb_abort_seq_read);
2455 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2457 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2458 struct ts_config *conf,
2459 struct ts_state *state)
2461 return skb_seq_read(offset, text, TS_SKB_CB(state));
2464 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2466 skb_abort_seq_read(TS_SKB_CB(state));
2470 * skb_find_text - Find a text pattern in skb data
2471 * @skb: the buffer to look in
2472 * @from: search offset
2474 * @config: textsearch configuration
2475 * @state: uninitialized textsearch state variable
2477 * Finds a pattern in the skb data according to the specified
2478 * textsearch configuration. Use textsearch_next() to retrieve
2479 * subsequent occurrences of the pattern. Returns the offset
2480 * to the first occurrence or UINT_MAX if no match was found.
2482 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2483 unsigned int to, struct ts_config *config,
2484 struct ts_state *state)
2488 config->get_next_block = skb_ts_get_next_block;
2489 config->finish = skb_ts_finish;
2491 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2493 ret = textsearch_find(config, state);
2494 return (ret <= to - from ? ret : UINT_MAX);
2496 EXPORT_SYMBOL(skb_find_text);
2499 * skb_append_datato_frags: - append the user data to a skb
2500 * @sk: sock structure
2501 * @skb: skb structure to be appened with user data.
2502 * @getfrag: call back function to be used for getting the user data
2503 * @from: pointer to user message iov
2504 * @length: length of the iov message
2506 * Description: This procedure append the user data in the fragment part
2507 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2509 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2510 int (*getfrag)(void *from, char *to, int offset,
2511 int len, int odd, struct sk_buff *skb),
2512 void *from, int length)
2515 skb_frag_t *frag = NULL;
2516 struct page *page = NULL;
2522 /* Return error if we don't have space for new frag */
2523 frg_cnt = skb_shinfo(skb)->nr_frags;
2524 if (frg_cnt >= MAX_SKB_FRAGS)
2527 /* allocate a new page for next frag */
2528 page = alloc_pages(sk->sk_allocation, 0);
2530 /* If alloc_page fails just return failure and caller will
2531 * free previous allocated pages by doing kfree_skb()
2536 /* initialize the next frag */
2537 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2538 skb->truesize += PAGE_SIZE;
2539 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2541 /* get the new initialized frag */
2542 frg_cnt = skb_shinfo(skb)->nr_frags;
2543 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2545 /* copy the user data to page */
2546 left = PAGE_SIZE - frag->page_offset;
2547 copy = (length > left)? left : length;
2549 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
2550 offset, copy, 0, skb);
2554 /* copy was successful so update the size parameters */
2555 skb_frag_size_add(frag, copy);
2557 skb->data_len += copy;
2561 } while (length > 0);
2565 EXPORT_SYMBOL(skb_append_datato_frags);
2568 * skb_pull_rcsum - pull skb and update receive checksum
2569 * @skb: buffer to update
2570 * @len: length of data pulled
2572 * This function performs an skb_pull on the packet and updates
2573 * the CHECKSUM_COMPLETE checksum. It should be used on
2574 * receive path processing instead of skb_pull unless you know
2575 * that the checksum difference is zero (e.g., a valid IP header)
2576 * or you are setting ip_summed to CHECKSUM_NONE.
2578 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2580 BUG_ON(len > skb->len);
2582 BUG_ON(skb->len < skb->data_len);
2583 skb_postpull_rcsum(skb, skb->data, len);
2584 return skb->data += len;
2586 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2589 * skb_segment - Perform protocol segmentation on skb.
2590 * @skb: buffer to segment
2591 * @features: features for the output path (see dev->features)
2593 * This function performs segmentation on the given skb. It returns
2594 * a pointer to the first in a list of new skbs for the segments.
2595 * In case of error it returns ERR_PTR(err).
2597 struct sk_buff *skb_segment(struct sk_buff *skb, u32 features)
2599 struct sk_buff *segs = NULL;
2600 struct sk_buff *tail = NULL;
2601 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2602 unsigned int mss = skb_shinfo(skb)->gso_size;
2603 unsigned int doffset = skb->data - skb_mac_header(skb);
2604 unsigned int offset = doffset;
2605 unsigned int headroom;
2607 int sg = !!(features & NETIF_F_SG);
2608 int nfrags = skb_shinfo(skb)->nr_frags;
2613 __skb_push(skb, doffset);
2614 headroom = skb_headroom(skb);
2615 pos = skb_headlen(skb);
2618 struct sk_buff *nskb;
2623 len = skb->len - offset;
2627 hsize = skb_headlen(skb) - offset;
2630 if (hsize > len || !sg)
2633 if (!hsize && i >= nfrags) {
2634 BUG_ON(fskb->len != len);
2637 nskb = skb_clone(fskb, GFP_ATOMIC);
2640 if (unlikely(!nskb))
2643 hsize = skb_end_pointer(nskb) - nskb->head;
2644 if (skb_cow_head(nskb, doffset + headroom)) {
2649 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2651 skb_release_head_state(nskb);
2652 __skb_push(nskb, doffset);
2654 nskb = alloc_skb(hsize + doffset + headroom,
2657 if (unlikely(!nskb))
2660 skb_reserve(nskb, headroom);
2661 __skb_put(nskb, doffset);
2670 __copy_skb_header(nskb, skb);
2671 nskb->mac_len = skb->mac_len;
2673 /* nskb and skb might have different headroom */
2674 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2675 nskb->csum_start += skb_headroom(nskb) - headroom;
2677 skb_reset_mac_header(nskb);
2678 skb_set_network_header(nskb, skb->mac_len);
2679 nskb->transport_header = (nskb->network_header +
2680 skb_network_header_len(skb));
2681 skb_copy_from_linear_data(skb, nskb->data, doffset);
2683 if (fskb != skb_shinfo(skb)->frag_list)
2687 nskb->ip_summed = CHECKSUM_NONE;
2688 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2694 frag = skb_shinfo(nskb)->frags;
2696 skb_copy_from_linear_data_offset(skb, offset,
2697 skb_put(nskb, hsize), hsize);
2699 while (pos < offset + len && i < nfrags) {
2700 *frag = skb_shinfo(skb)->frags[i];
2701 __skb_frag_ref(frag);
2702 size = skb_frag_size(frag);
2705 frag->page_offset += offset - pos;
2706 skb_frag_size_sub(frag, offset - pos);
2709 skb_shinfo(nskb)->nr_frags++;
2711 if (pos + size <= offset + len) {
2715 skb_frag_size_sub(frag, pos + size - (offset + len));
2722 if (pos < offset + len) {
2723 struct sk_buff *fskb2 = fskb;
2725 BUG_ON(pos + fskb->len != offset + len);
2731 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2737 SKB_FRAG_ASSERT(nskb);
2738 skb_shinfo(nskb)->frag_list = fskb2;
2742 nskb->data_len = len - hsize;
2743 nskb->len += nskb->data_len;
2744 nskb->truesize += nskb->data_len;
2745 } while ((offset += len) < skb->len);
2750 while ((skb = segs)) {
2754 return ERR_PTR(err);
2756 EXPORT_SYMBOL_GPL(skb_segment);
2758 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2760 struct sk_buff *p = *head;
2761 struct sk_buff *nskb;
2762 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2763 struct skb_shared_info *pinfo = skb_shinfo(p);
2764 unsigned int headroom;
2765 unsigned int len = skb_gro_len(skb);
2766 unsigned int offset = skb_gro_offset(skb);
2767 unsigned int headlen = skb_headlen(skb);
2769 if (p->len + len >= 65536)
2772 if (pinfo->frag_list)
2774 else if (headlen <= offset) {
2777 int i = skbinfo->nr_frags;
2778 int nr_frags = pinfo->nr_frags + i;
2782 if (nr_frags > MAX_SKB_FRAGS)
2785 pinfo->nr_frags = nr_frags;
2786 skbinfo->nr_frags = 0;
2788 frag = pinfo->frags + nr_frags;
2789 frag2 = skbinfo->frags + i;
2794 frag->page_offset += offset;
2795 skb_frag_size_sub(frag, offset);
2797 skb->truesize -= skb->data_len;
2798 skb->len -= skb->data_len;
2801 NAPI_GRO_CB(skb)->free = 1;
2803 } else if (skb_gro_len(p) != pinfo->gso_size)
2806 headroom = skb_headroom(p);
2807 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2808 if (unlikely(!nskb))
2811 __copy_skb_header(nskb, p);
2812 nskb->mac_len = p->mac_len;
2814 skb_reserve(nskb, headroom);
2815 __skb_put(nskb, skb_gro_offset(p));
2817 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2818 skb_set_network_header(nskb, skb_network_offset(p));
2819 skb_set_transport_header(nskb, skb_transport_offset(p));
2821 __skb_pull(p, skb_gro_offset(p));
2822 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2823 p->data - skb_mac_header(p));
2825 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2826 skb_shinfo(nskb)->frag_list = p;
2827 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2828 pinfo->gso_size = 0;
2829 skb_header_release(p);
2832 nskb->data_len += p->len;
2833 nskb->truesize += p->len;
2834 nskb->len += p->len;
2837 nskb->next = p->next;
2843 if (offset > headlen) {
2844 unsigned int eat = offset - headlen;
2846 skbinfo->frags[0].page_offset += eat;
2847 skb_frag_size_sub(&skbinfo->frags[0], eat);
2848 skb->data_len -= eat;
2853 __skb_pull(skb, offset);
2855 p->prev->next = skb;
2857 skb_header_release(skb);
2860 NAPI_GRO_CB(p)->count++;
2865 NAPI_GRO_CB(skb)->same_flow = 1;
2868 EXPORT_SYMBOL_GPL(skb_gro_receive);
2870 void __init skb_init(void)
2872 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2873 sizeof(struct sk_buff),
2875 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2877 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2878 (2*sizeof(struct sk_buff)) +
2881 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2886 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2887 * @skb: Socket buffer containing the buffers to be mapped
2888 * @sg: The scatter-gather list to map into
2889 * @offset: The offset into the buffer's contents to start mapping
2890 * @len: Length of buffer space to be mapped
2892 * Fill the specified scatter-gather list with mappings/pointers into a
2893 * region of the buffer space attached to a socket buffer.
2896 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2898 int start = skb_headlen(skb);
2899 int i, copy = start - offset;
2900 struct sk_buff *frag_iter;
2906 sg_set_buf(sg, skb->data + offset, copy);
2908 if ((len -= copy) == 0)
2913 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2916 WARN_ON(start > offset + len);
2918 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2919 if ((copy = end - offset) > 0) {
2920 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2924 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
2925 frag->page_offset+offset-start);
2934 skb_walk_frags(skb, frag_iter) {
2937 WARN_ON(start > offset + len);
2939 end = start + frag_iter->len;
2940 if ((copy = end - offset) > 0) {
2943 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2945 if ((len -= copy) == 0)
2955 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2957 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2959 sg_mark_end(&sg[nsg - 1]);
2963 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2966 * skb_cow_data - Check that a socket buffer's data buffers are writable
2967 * @skb: The socket buffer to check.
2968 * @tailbits: Amount of trailing space to be added
2969 * @trailer: Returned pointer to the skb where the @tailbits space begins
2971 * Make sure that the data buffers attached to a socket buffer are
2972 * writable. If they are not, private copies are made of the data buffers
2973 * and the socket buffer is set to use these instead.
2975 * If @tailbits is given, make sure that there is space to write @tailbits
2976 * bytes of data beyond current end of socket buffer. @trailer will be
2977 * set to point to the skb in which this space begins.
2979 * The number of scatterlist elements required to completely map the
2980 * COW'd and extended socket buffer will be returned.
2982 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2986 struct sk_buff *skb1, **skb_p;
2988 /* If skb is cloned or its head is paged, reallocate
2989 * head pulling out all the pages (pages are considered not writable
2990 * at the moment even if they are anonymous).
2992 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2993 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2996 /* Easy case. Most of packets will go this way. */
2997 if (!skb_has_frag_list(skb)) {
2998 /* A little of trouble, not enough of space for trailer.
2999 * This should not happen, when stack is tuned to generate
3000 * good frames. OK, on miss we reallocate and reserve even more
3001 * space, 128 bytes is fair. */
3003 if (skb_tailroom(skb) < tailbits &&
3004 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3012 /* Misery. We are in troubles, going to mincer fragments... */
3015 skb_p = &skb_shinfo(skb)->frag_list;
3018 while ((skb1 = *skb_p) != NULL) {
3021 /* The fragment is partially pulled by someone,
3022 * this can happen on input. Copy it and everything
3025 if (skb_shared(skb1))
3028 /* If the skb is the last, worry about trailer. */
3030 if (skb1->next == NULL && tailbits) {
3031 if (skb_shinfo(skb1)->nr_frags ||
3032 skb_has_frag_list(skb1) ||
3033 skb_tailroom(skb1) < tailbits)
3034 ntail = tailbits + 128;
3040 skb_shinfo(skb1)->nr_frags ||
3041 skb_has_frag_list(skb1)) {
3042 struct sk_buff *skb2;
3044 /* Fuck, we are miserable poor guys... */
3046 skb2 = skb_copy(skb1, GFP_ATOMIC);
3048 skb2 = skb_copy_expand(skb1,
3052 if (unlikely(skb2 == NULL))
3056 skb_set_owner_w(skb2, skb1->sk);
3058 /* Looking around. Are we still alive?
3059 * OK, link new skb, drop old one */
3061 skb2->next = skb1->next;
3068 skb_p = &skb1->next;
3073 EXPORT_SYMBOL_GPL(skb_cow_data);
3075 static void sock_rmem_free(struct sk_buff *skb)
3077 struct sock *sk = skb->sk;
3079 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3083 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3085 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3089 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3090 (unsigned)sk->sk_rcvbuf)
3095 skb->destructor = sock_rmem_free;
3096 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3098 /* before exiting rcu section, make sure dst is refcounted */
3101 skb_queue_tail(&sk->sk_error_queue, skb);
3102 if (!sock_flag(sk, SOCK_DEAD))
3103 sk->sk_data_ready(sk, len);
3106 EXPORT_SYMBOL(sock_queue_err_skb);
3108 void skb_tstamp_tx(struct sk_buff *orig_skb,
3109 struct skb_shared_hwtstamps *hwtstamps)
3111 struct sock *sk = orig_skb->sk;
3112 struct sock_exterr_skb *serr;
3113 struct sk_buff *skb;
3119 skb = skb_clone(orig_skb, GFP_ATOMIC);
3124 *skb_hwtstamps(skb) =
3128 * no hardware time stamps available,
3129 * so keep the shared tx_flags and only
3130 * store software time stamp
3132 skb->tstamp = ktime_get_real();
3135 serr = SKB_EXT_ERR(skb);
3136 memset(serr, 0, sizeof(*serr));
3137 serr->ee.ee_errno = ENOMSG;
3138 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3140 err = sock_queue_err_skb(sk, skb);
3145 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3149 * skb_partial_csum_set - set up and verify partial csum values for packet
3150 * @skb: the skb to set
3151 * @start: the number of bytes after skb->data to start checksumming.
3152 * @off: the offset from start to place the checksum.
3154 * For untrusted partially-checksummed packets, we need to make sure the values
3155 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3157 * This function checks and sets those values and skb->ip_summed: if this
3158 * returns false you should drop the packet.
3160 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3162 if (unlikely(start > skb_headlen(skb)) ||
3163 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3164 if (net_ratelimit())
3166 "bad partial csum: csum=%u/%u len=%u\n",
3167 start, off, skb_headlen(skb));
3170 skb->ip_summed = CHECKSUM_PARTIAL;
3171 skb->csum_start = skb_headroom(skb) + start;
3172 skb->csum_offset = off;
3175 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3177 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3179 if (net_ratelimit())
3180 pr_warning("%s: received packets cannot be forwarded"
3181 " while LRO is enabled\n", skb->dev->name);
3183 EXPORT_SYMBOL(__skb_warn_lro_forwarding);