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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
44 #include <linux/kmemcheck.h>
46 #include <linux/interrupt.h>
48 #include <linux/inet.h>
49 #include <linux/slab.h>
50 #include <linux/netdevice.h>
51 #ifdef CONFIG_NET_CLS_ACT
52 #include <net/pkt_sched.h>
54 #include <linux/string.h>
55 #include <linux/skbuff.h>
56 #include <linux/splice.h>
57 #include <linux/cache.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/init.h>
60 #include <linux/scatterlist.h>
61 #include <linux/errqueue.h>
62 #include <linux/prefetch.h>
64 #include <net/protocol.h>
67 #include <net/checksum.h>
70 #include <asm/uaccess.h>
71 #include <trace/events/skb.h>
72 #include <linux/highmem.h>
74 struct kmem_cache *skbuff_head_cache __read_mostly;
75 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
77 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
78 struct pipe_buffer *buf)
83 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
84 struct pipe_buffer *buf)
89 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
90 struct pipe_buffer *buf)
96 /* Pipe buffer operations for a socket. */
97 static const struct pipe_buf_operations sock_pipe_buf_ops = {
99 .map = generic_pipe_buf_map,
100 .unmap = generic_pipe_buf_unmap,
101 .confirm = generic_pipe_buf_confirm,
102 .release = sock_pipe_buf_release,
103 .steal = sock_pipe_buf_steal,
104 .get = sock_pipe_buf_get,
108 * skb_panic - private function for out-of-line support
112 * @msg: skb_over_panic or skb_under_panic
114 * Out-of-line support for skb_put() and skb_push().
115 * Called via the wrapper skb_over_panic() or skb_under_panic().
116 * Keep out of line to prevent kernel bloat.
117 * __builtin_return_address is not used because it is not always reliable.
119 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
122 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
123 msg, addr, skb->len, sz, skb->head, skb->data,
124 (unsigned long)skb->tail, (unsigned long)skb->end,
125 skb->dev ? skb->dev->name : "<NULL>");
129 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
131 skb_panic(skb, sz, addr, __func__);
134 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
136 skb_panic(skb, sz, addr, __func__);
140 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
141 * the caller if emergency pfmemalloc reserves are being used. If it is and
142 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
143 * may be used. Otherwise, the packet data may be discarded until enough
146 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
147 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
149 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
150 unsigned long ip, bool *pfmemalloc)
153 bool ret_pfmemalloc = false;
156 * Try a regular allocation, when that fails and we're not entitled
157 * to the reserves, fail.
159 obj = kmalloc_node_track_caller(size,
160 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
162 if (obj || !(gfp_pfmemalloc_allowed(flags)))
165 /* Try again but now we are using pfmemalloc reserves */
166 ret_pfmemalloc = true;
167 obj = kmalloc_node_track_caller(size, flags, node);
171 *pfmemalloc = ret_pfmemalloc;
176 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
177 * 'private' fields and also do memory statistics to find all the
182 struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
187 skb = kmem_cache_alloc_node(skbuff_head_cache,
188 gfp_mask & ~__GFP_DMA, node);
193 * Only clear those fields we need to clear, not those that we will
194 * actually initialise below. Hence, don't put any more fields after
195 * the tail pointer in struct sk_buff!
197 memset(skb, 0, offsetof(struct sk_buff, tail));
199 skb->truesize = sizeof(struct sk_buff);
200 atomic_set(&skb->users, 1);
202 skb->mac_header = (typeof(skb->mac_header))~0U;
208 * __alloc_skb - allocate a network buffer
209 * @size: size to allocate
210 * @gfp_mask: allocation mask
211 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
212 * instead of head cache and allocate a cloned (child) skb.
213 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
214 * allocations in case the data is required for writeback
215 * @node: numa node to allocate memory on
217 * Allocate a new &sk_buff. The returned buffer has no headroom and a
218 * tail room of at least size bytes. The object has a reference count
219 * of one. The return is the buffer. On a failure the return is %NULL.
221 * Buffers may only be allocated from interrupts using a @gfp_mask of
224 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
227 struct kmem_cache *cache;
228 struct skb_shared_info *shinfo;
233 cache = (flags & SKB_ALLOC_FCLONE)
234 ? skbuff_fclone_cache : skbuff_head_cache;
236 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
237 gfp_mask |= __GFP_MEMALLOC;
240 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
245 /* We do our best to align skb_shared_info on a separate cache
246 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
247 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
248 * Both skb->head and skb_shared_info are cache line aligned.
250 size = SKB_DATA_ALIGN(size);
251 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
252 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
255 /* kmalloc(size) might give us more room than requested.
256 * Put skb_shared_info exactly at the end of allocated zone,
257 * to allow max possible filling before reallocation.
259 size = SKB_WITH_OVERHEAD(ksize(data));
260 prefetchw(data + size);
263 * Only clear those fields we need to clear, not those that we will
264 * actually initialise below. Hence, don't put any more fields after
265 * the tail pointer in struct sk_buff!
267 memset(skb, 0, offsetof(struct sk_buff, tail));
268 /* Account for allocated memory : skb + skb->head */
269 skb->truesize = SKB_TRUESIZE(size);
270 skb->pfmemalloc = pfmemalloc;
271 atomic_set(&skb->users, 1);
274 skb_reset_tail_pointer(skb);
275 skb->end = skb->tail + size;
276 skb->mac_header = (typeof(skb->mac_header))~0U;
277 skb->transport_header = (typeof(skb->transport_header))~0U;
279 /* make sure we initialize shinfo sequentially */
280 shinfo = skb_shinfo(skb);
281 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
282 atomic_set(&shinfo->dataref, 1);
283 kmemcheck_annotate_variable(shinfo->destructor_arg);
285 if (flags & SKB_ALLOC_FCLONE) {
286 struct sk_buff *child = skb + 1;
287 atomic_t *fclone_ref = (atomic_t *) (child + 1);
289 kmemcheck_annotate_bitfield(child, flags1);
290 kmemcheck_annotate_bitfield(child, flags2);
291 skb->fclone = SKB_FCLONE_ORIG;
292 atomic_set(fclone_ref, 1);
294 child->fclone = SKB_FCLONE_UNAVAILABLE;
295 child->pfmemalloc = pfmemalloc;
300 kmem_cache_free(cache, skb);
304 EXPORT_SYMBOL(__alloc_skb);
307 * build_skb - build a network buffer
308 * @data: data buffer provided by caller
309 * @frag_size: size of fragment, or 0 if head was kmalloced
311 * Allocate a new &sk_buff. Caller provides space holding head and
312 * skb_shared_info. @data must have been allocated by kmalloc() only if
313 * @frag_size is 0, otherwise data should come from the page allocator.
314 * The return is the new skb buffer.
315 * On a failure the return is %NULL, and @data is not freed.
317 * Before IO, driver allocates only data buffer where NIC put incoming frame
318 * Driver should add room at head (NET_SKB_PAD) and
319 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
320 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
321 * before giving packet to stack.
322 * RX rings only contains data buffers, not full skbs.
324 struct sk_buff *build_skb(void *data, unsigned int frag_size)
326 struct skb_shared_info *shinfo;
328 unsigned int size = frag_size ? : ksize(data);
330 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
334 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
336 memset(skb, 0, offsetof(struct sk_buff, tail));
337 skb->truesize = SKB_TRUESIZE(size);
338 skb->head_frag = frag_size != 0;
339 atomic_set(&skb->users, 1);
342 skb_reset_tail_pointer(skb);
343 skb->end = skb->tail + size;
344 skb->mac_header = (typeof(skb->mac_header))~0U;
345 skb->transport_header = (typeof(skb->transport_header))~0U;
347 /* make sure we initialize shinfo sequentially */
348 shinfo = skb_shinfo(skb);
349 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
350 atomic_set(&shinfo->dataref, 1);
351 kmemcheck_annotate_variable(shinfo->destructor_arg);
355 EXPORT_SYMBOL(build_skb);
357 struct netdev_alloc_cache {
358 struct page_frag frag;
359 /* we maintain a pagecount bias, so that we dont dirty cache line
360 * containing page->_count every time we allocate a fragment.
362 unsigned int pagecnt_bias;
364 static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
366 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
368 struct netdev_alloc_cache *nc;
373 local_irq_save(flags);
374 nc = &__get_cpu_var(netdev_alloc_cache);
375 if (unlikely(!nc->frag.page)) {
377 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
378 gfp_t gfp = gfp_mask;
381 gfp |= __GFP_COMP | __GFP_NOWARN;
382 nc->frag.page = alloc_pages(gfp, order);
383 if (likely(nc->frag.page))
388 nc->frag.size = PAGE_SIZE << order;
390 atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS);
391 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
395 if (nc->frag.offset + fragsz > nc->frag.size) {
396 /* avoid unnecessary locked operations if possible */
397 if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) ||
398 atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count))
403 data = page_address(nc->frag.page) + nc->frag.offset;
404 nc->frag.offset += fragsz;
407 local_irq_restore(flags);
412 * netdev_alloc_frag - allocate a page fragment
413 * @fragsz: fragment size
415 * Allocates a frag from a page for receive buffer.
416 * Uses GFP_ATOMIC allocations.
418 void *netdev_alloc_frag(unsigned int fragsz)
420 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
422 EXPORT_SYMBOL(netdev_alloc_frag);
425 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
426 * @dev: network device to receive on
427 * @length: length to allocate
428 * @gfp_mask: get_free_pages mask, passed to alloc_skb
430 * Allocate a new &sk_buff and assign it a usage count of one. The
431 * buffer has unspecified headroom built in. Users should allocate
432 * the headroom they think they need without accounting for the
433 * built in space. The built in space is used for optimisations.
435 * %NULL is returned if there is no free memory.
437 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
438 unsigned int length, gfp_t gfp_mask)
440 struct sk_buff *skb = NULL;
441 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
442 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
444 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
447 if (sk_memalloc_socks())
448 gfp_mask |= __GFP_MEMALLOC;
450 data = __netdev_alloc_frag(fragsz, gfp_mask);
453 skb = build_skb(data, fragsz);
455 put_page(virt_to_head_page(data));
458 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
459 SKB_ALLOC_RX, NUMA_NO_NODE);
462 skb_reserve(skb, NET_SKB_PAD);
467 EXPORT_SYMBOL(__netdev_alloc_skb);
469 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
470 int size, unsigned int truesize)
472 skb_fill_page_desc(skb, i, page, off, size);
474 skb->data_len += size;
475 skb->truesize += truesize;
477 EXPORT_SYMBOL(skb_add_rx_frag);
479 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
480 unsigned int truesize)
482 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
484 skb_frag_size_add(frag, size);
486 skb->data_len += size;
487 skb->truesize += truesize;
489 EXPORT_SYMBOL(skb_coalesce_rx_frag);
491 static void skb_drop_list(struct sk_buff **listp)
493 kfree_skb_list(*listp);
497 static inline void skb_drop_fraglist(struct sk_buff *skb)
499 skb_drop_list(&skb_shinfo(skb)->frag_list);
502 static void skb_clone_fraglist(struct sk_buff *skb)
504 struct sk_buff *list;
506 skb_walk_frags(skb, list)
510 static void skb_free_head(struct sk_buff *skb)
513 put_page(virt_to_head_page(skb->head));
518 static void skb_release_data(struct sk_buff *skb)
521 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
522 &skb_shinfo(skb)->dataref)) {
523 if (skb_shinfo(skb)->nr_frags) {
525 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
526 skb_frag_unref(skb, i);
530 * If skb buf is from userspace, we need to notify the caller
531 * the lower device DMA has done;
533 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
534 struct ubuf_info *uarg;
536 uarg = skb_shinfo(skb)->destructor_arg;
538 uarg->callback(uarg, true);
541 if (skb_has_frag_list(skb))
542 skb_drop_fraglist(skb);
549 * Free an skbuff by memory without cleaning the state.
551 static void kfree_skbmem(struct sk_buff *skb)
553 struct sk_buff *other;
554 atomic_t *fclone_ref;
556 switch (skb->fclone) {
557 case SKB_FCLONE_UNAVAILABLE:
558 kmem_cache_free(skbuff_head_cache, skb);
561 case SKB_FCLONE_ORIG:
562 fclone_ref = (atomic_t *) (skb + 2);
563 if (atomic_dec_and_test(fclone_ref))
564 kmem_cache_free(skbuff_fclone_cache, skb);
567 case SKB_FCLONE_CLONE:
568 fclone_ref = (atomic_t *) (skb + 1);
571 /* The clone portion is available for
572 * fast-cloning again.
574 skb->fclone = SKB_FCLONE_UNAVAILABLE;
576 if (atomic_dec_and_test(fclone_ref))
577 kmem_cache_free(skbuff_fclone_cache, other);
582 static void skb_release_head_state(struct sk_buff *skb)
586 secpath_put(skb->sp);
588 if (skb->destructor) {
590 skb->destructor(skb);
592 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
593 nf_conntrack_put(skb->nfct);
595 #ifdef CONFIG_BRIDGE_NETFILTER
596 nf_bridge_put(skb->nf_bridge);
598 /* XXX: IS this still necessary? - JHS */
599 #ifdef CONFIG_NET_SCHED
601 #ifdef CONFIG_NET_CLS_ACT
607 /* Free everything but the sk_buff shell. */
608 static void skb_release_all(struct sk_buff *skb)
610 skb_release_head_state(skb);
611 if (likely(skb->head))
612 skb_release_data(skb);
616 * __kfree_skb - private function
619 * Free an sk_buff. Release anything attached to the buffer.
620 * Clean the state. This is an internal helper function. Users should
621 * always call kfree_skb
624 void __kfree_skb(struct sk_buff *skb)
626 skb_release_all(skb);
629 EXPORT_SYMBOL(__kfree_skb);
632 * kfree_skb - free an sk_buff
633 * @skb: buffer to free
635 * Drop a reference to the buffer and free it if the usage count has
638 void kfree_skb(struct sk_buff *skb)
642 if (likely(atomic_read(&skb->users) == 1))
644 else if (likely(!atomic_dec_and_test(&skb->users)))
646 trace_kfree_skb(skb, __builtin_return_address(0));
649 EXPORT_SYMBOL(kfree_skb);
651 void kfree_skb_list(struct sk_buff *segs)
654 struct sk_buff *next = segs->next;
660 EXPORT_SYMBOL(kfree_skb_list);
663 * skb_tx_error - report an sk_buff xmit error
664 * @skb: buffer that triggered an error
666 * Report xmit error if a device callback is tracking this skb.
667 * skb must be freed afterwards.
669 void skb_tx_error(struct sk_buff *skb)
671 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
672 struct ubuf_info *uarg;
674 uarg = skb_shinfo(skb)->destructor_arg;
676 uarg->callback(uarg, false);
677 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
680 EXPORT_SYMBOL(skb_tx_error);
683 * consume_skb - free an skbuff
684 * @skb: buffer to free
686 * Drop a ref to the buffer and free it if the usage count has hit zero
687 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
688 * is being dropped after a failure and notes that
690 void consume_skb(struct sk_buff *skb)
694 if (likely(atomic_read(&skb->users) == 1))
696 else if (likely(!atomic_dec_and_test(&skb->users)))
698 trace_consume_skb(skb);
701 EXPORT_SYMBOL(consume_skb);
703 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
705 new->tstamp = old->tstamp;
707 new->transport_header = old->transport_header;
708 new->network_header = old->network_header;
709 new->mac_header = old->mac_header;
710 new->inner_protocol = old->inner_protocol;
711 new->inner_transport_header = old->inner_transport_header;
712 new->inner_network_header = old->inner_network_header;
713 new->inner_mac_header = old->inner_mac_header;
714 skb_dst_copy(new, old);
715 skb_copy_hash(new, old);
716 new->ooo_okay = old->ooo_okay;
717 new->no_fcs = old->no_fcs;
718 new->encapsulation = old->encapsulation;
720 new->sp = secpath_get(old->sp);
722 memcpy(new->cb, old->cb, sizeof(old->cb));
723 new->csum = old->csum;
724 new->local_df = old->local_df;
725 new->pkt_type = old->pkt_type;
726 new->ip_summed = old->ip_summed;
727 skb_copy_queue_mapping(new, old);
728 new->priority = old->priority;
729 #if IS_ENABLED(CONFIG_IP_VS)
730 new->ipvs_property = old->ipvs_property;
732 new->pfmemalloc = old->pfmemalloc;
733 new->protocol = old->protocol;
734 new->mark = old->mark;
735 new->skb_iif = old->skb_iif;
737 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
738 new->nf_trace = old->nf_trace;
740 #ifdef CONFIG_NET_SCHED
741 new->tc_index = old->tc_index;
742 #ifdef CONFIG_NET_CLS_ACT
743 new->tc_verd = old->tc_verd;
746 new->vlan_proto = old->vlan_proto;
747 new->vlan_tci = old->vlan_tci;
749 skb_copy_secmark(new, old);
751 #ifdef CONFIG_NET_RX_BUSY_POLL
752 new->napi_id = old->napi_id;
757 * You should not add any new code to this function. Add it to
758 * __copy_skb_header above instead.
760 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
762 #define C(x) n->x = skb->x
764 n->next = n->prev = NULL;
766 __copy_skb_header(n, skb);
771 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
774 n->destructor = NULL;
781 atomic_set(&n->users, 1);
783 atomic_inc(&(skb_shinfo(skb)->dataref));
791 * skb_morph - morph one skb into another
792 * @dst: the skb to receive the contents
793 * @src: the skb to supply the contents
795 * This is identical to skb_clone except that the target skb is
796 * supplied by the user.
798 * The target skb is returned upon exit.
800 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
802 skb_release_all(dst);
803 return __skb_clone(dst, src);
805 EXPORT_SYMBOL_GPL(skb_morph);
808 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
809 * @skb: the skb to modify
810 * @gfp_mask: allocation priority
812 * This must be called on SKBTX_DEV_ZEROCOPY skb.
813 * It will copy all frags into kernel and drop the reference
814 * to userspace pages.
816 * If this function is called from an interrupt gfp_mask() must be
819 * Returns 0 on success or a negative error code on failure
820 * to allocate kernel memory to copy to.
822 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
825 int num_frags = skb_shinfo(skb)->nr_frags;
826 struct page *page, *head = NULL;
827 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
829 for (i = 0; i < num_frags; i++) {
831 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
833 page = alloc_page(gfp_mask);
836 struct page *next = (struct page *)page_private(head);
842 vaddr = kmap_atomic(skb_frag_page(f));
843 memcpy(page_address(page),
844 vaddr + f->page_offset, skb_frag_size(f));
845 kunmap_atomic(vaddr);
846 set_page_private(page, (unsigned long)head);
850 /* skb frags release userspace buffers */
851 for (i = 0; i < num_frags; i++)
852 skb_frag_unref(skb, i);
854 uarg->callback(uarg, false);
856 /* skb frags point to kernel buffers */
857 for (i = num_frags - 1; i >= 0; i--) {
858 __skb_fill_page_desc(skb, i, head, 0,
859 skb_shinfo(skb)->frags[i].size);
860 head = (struct page *)page_private(head);
863 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
866 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
869 * skb_clone - duplicate an sk_buff
870 * @skb: buffer to clone
871 * @gfp_mask: allocation priority
873 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
874 * copies share the same packet data but not structure. The new
875 * buffer has a reference count of 1. If the allocation fails the
876 * function returns %NULL otherwise the new buffer is returned.
878 * If this function is called from an interrupt gfp_mask() must be
882 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
886 if (skb_orphan_frags(skb, gfp_mask))
890 if (skb->fclone == SKB_FCLONE_ORIG &&
891 n->fclone == SKB_FCLONE_UNAVAILABLE) {
892 atomic_t *fclone_ref = (atomic_t *) (n + 1);
893 n->fclone = SKB_FCLONE_CLONE;
894 atomic_inc(fclone_ref);
896 if (skb_pfmemalloc(skb))
897 gfp_mask |= __GFP_MEMALLOC;
899 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
903 kmemcheck_annotate_bitfield(n, flags1);
904 kmemcheck_annotate_bitfield(n, flags2);
905 n->fclone = SKB_FCLONE_UNAVAILABLE;
908 return __skb_clone(n, skb);
910 EXPORT_SYMBOL(skb_clone);
912 static void skb_headers_offset_update(struct sk_buff *skb, int off)
914 /* Only adjust this if it actually is csum_start rather than csum */
915 if (skb->ip_summed == CHECKSUM_PARTIAL)
916 skb->csum_start += off;
917 /* {transport,network,mac}_header and tail are relative to skb->head */
918 skb->transport_header += off;
919 skb->network_header += off;
920 if (skb_mac_header_was_set(skb))
921 skb->mac_header += off;
922 skb->inner_transport_header += off;
923 skb->inner_network_header += off;
924 skb->inner_mac_header += off;
927 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
929 __copy_skb_header(new, old);
931 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
932 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
933 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
936 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
938 if (skb_pfmemalloc(skb))
944 * skb_copy - create private copy of an sk_buff
945 * @skb: buffer to copy
946 * @gfp_mask: allocation priority
948 * Make a copy of both an &sk_buff and its data. This is used when the
949 * caller wishes to modify the data and needs a private copy of the
950 * data to alter. Returns %NULL on failure or the pointer to the buffer
951 * on success. The returned buffer has a reference count of 1.
953 * As by-product this function converts non-linear &sk_buff to linear
954 * one, so that &sk_buff becomes completely private and caller is allowed
955 * to modify all the data of returned buffer. This means that this
956 * function is not recommended for use in circumstances when only
957 * header is going to be modified. Use pskb_copy() instead.
960 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
962 int headerlen = skb_headroom(skb);
963 unsigned int size = skb_end_offset(skb) + skb->data_len;
964 struct sk_buff *n = __alloc_skb(size, gfp_mask,
965 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
970 /* Set the data pointer */
971 skb_reserve(n, headerlen);
972 /* Set the tail pointer and length */
973 skb_put(n, skb->len);
975 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
978 copy_skb_header(n, skb);
981 EXPORT_SYMBOL(skb_copy);
984 * __pskb_copy - create copy of an sk_buff with private head.
985 * @skb: buffer to copy
986 * @headroom: headroom of new skb
987 * @gfp_mask: allocation priority
989 * Make a copy of both an &sk_buff and part of its data, located
990 * in header. Fragmented data remain shared. This is used when
991 * the caller wishes to modify only header of &sk_buff and needs
992 * private copy of the header to alter. Returns %NULL on failure
993 * or the pointer to the buffer on success.
994 * The returned buffer has a reference count of 1.
997 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
999 unsigned int size = skb_headlen(skb) + headroom;
1000 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1001 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1006 /* Set the data pointer */
1007 skb_reserve(n, headroom);
1008 /* Set the tail pointer and length */
1009 skb_put(n, skb_headlen(skb));
1010 /* Copy the bytes */
1011 skb_copy_from_linear_data(skb, n->data, n->len);
1013 n->truesize += skb->data_len;
1014 n->data_len = skb->data_len;
1017 if (skb_shinfo(skb)->nr_frags) {
1020 if (skb_orphan_frags(skb, gfp_mask)) {
1025 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1026 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1027 skb_frag_ref(skb, i);
1029 skb_shinfo(n)->nr_frags = i;
1032 if (skb_has_frag_list(skb)) {
1033 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1034 skb_clone_fraglist(n);
1037 copy_skb_header(n, skb);
1041 EXPORT_SYMBOL(__pskb_copy);
1044 * pskb_expand_head - reallocate header of &sk_buff
1045 * @skb: buffer to reallocate
1046 * @nhead: room to add at head
1047 * @ntail: room to add at tail
1048 * @gfp_mask: allocation priority
1050 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1051 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1052 * reference count of 1. Returns zero in the case of success or error,
1053 * if expansion failed. In the last case, &sk_buff is not changed.
1055 * All the pointers pointing into skb header may change and must be
1056 * reloaded after call to this function.
1059 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1064 int size = nhead + skb_end_offset(skb) + ntail;
1069 if (skb_shared(skb))
1072 size = SKB_DATA_ALIGN(size);
1074 if (skb_pfmemalloc(skb))
1075 gfp_mask |= __GFP_MEMALLOC;
1076 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1077 gfp_mask, NUMA_NO_NODE, NULL);
1080 size = SKB_WITH_OVERHEAD(ksize(data));
1082 /* Copy only real data... and, alas, header. This should be
1083 * optimized for the cases when header is void.
1085 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1087 memcpy((struct skb_shared_info *)(data + size),
1089 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1092 * if shinfo is shared we must drop the old head gracefully, but if it
1093 * is not we can just drop the old head and let the existing refcount
1094 * be since all we did is relocate the values
1096 if (skb_cloned(skb)) {
1097 /* copy this zero copy skb frags */
1098 if (skb_orphan_frags(skb, gfp_mask))
1100 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1101 skb_frag_ref(skb, i);
1103 if (skb_has_frag_list(skb))
1104 skb_clone_fraglist(skb);
1106 skb_release_data(skb);
1110 off = (data + nhead) - skb->head;
1115 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1119 skb->end = skb->head + size;
1122 skb_headers_offset_update(skb, nhead);
1126 atomic_set(&skb_shinfo(skb)->dataref, 1);
1134 EXPORT_SYMBOL(pskb_expand_head);
1136 /* Make private copy of skb with writable head and some headroom */
1138 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1140 struct sk_buff *skb2;
1141 int delta = headroom - skb_headroom(skb);
1144 skb2 = pskb_copy(skb, GFP_ATOMIC);
1146 skb2 = skb_clone(skb, GFP_ATOMIC);
1147 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1155 EXPORT_SYMBOL(skb_realloc_headroom);
1158 * skb_copy_expand - copy and expand sk_buff
1159 * @skb: buffer to copy
1160 * @newheadroom: new free bytes at head
1161 * @newtailroom: new free bytes at tail
1162 * @gfp_mask: allocation priority
1164 * Make a copy of both an &sk_buff and its data and while doing so
1165 * allocate additional space.
1167 * This is used when the caller wishes to modify the data and needs a
1168 * private copy of the data to alter as well as more space for new fields.
1169 * Returns %NULL on failure or the pointer to the buffer
1170 * on success. The returned buffer has a reference count of 1.
1172 * You must pass %GFP_ATOMIC as the allocation priority if this function
1173 * is called from an interrupt.
1175 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1176 int newheadroom, int newtailroom,
1180 * Allocate the copy buffer
1182 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1183 gfp_mask, skb_alloc_rx_flag(skb),
1185 int oldheadroom = skb_headroom(skb);
1186 int head_copy_len, head_copy_off;
1191 skb_reserve(n, newheadroom);
1193 /* Set the tail pointer and length */
1194 skb_put(n, skb->len);
1196 head_copy_len = oldheadroom;
1198 if (newheadroom <= head_copy_len)
1199 head_copy_len = newheadroom;
1201 head_copy_off = newheadroom - head_copy_len;
1203 /* Copy the linear header and data. */
1204 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1205 skb->len + head_copy_len))
1208 copy_skb_header(n, skb);
1210 skb_headers_offset_update(n, newheadroom - oldheadroom);
1214 EXPORT_SYMBOL(skb_copy_expand);
1217 * skb_pad - zero pad the tail of an skb
1218 * @skb: buffer to pad
1219 * @pad: space to pad
1221 * Ensure that a buffer is followed by a padding area that is zero
1222 * filled. Used by network drivers which may DMA or transfer data
1223 * beyond the buffer end onto the wire.
1225 * May return error in out of memory cases. The skb is freed on error.
1228 int skb_pad(struct sk_buff *skb, int pad)
1233 /* If the skbuff is non linear tailroom is always zero.. */
1234 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1235 memset(skb->data+skb->len, 0, pad);
1239 ntail = skb->data_len + pad - (skb->end - skb->tail);
1240 if (likely(skb_cloned(skb) || ntail > 0)) {
1241 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1246 /* FIXME: The use of this function with non-linear skb's really needs
1249 err = skb_linearize(skb);
1253 memset(skb->data + skb->len, 0, pad);
1260 EXPORT_SYMBOL(skb_pad);
1263 * pskb_put - add data to the tail of a potentially fragmented buffer
1264 * @skb: start of the buffer to use
1265 * @tail: tail fragment of the buffer to use
1266 * @len: amount of data to add
1268 * This function extends the used data area of the potentially
1269 * fragmented buffer. @tail must be the last fragment of @skb -- or
1270 * @skb itself. If this would exceed the total buffer size the kernel
1271 * will panic. A pointer to the first byte of the extra data is
1275 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1278 skb->data_len += len;
1281 return skb_put(tail, len);
1283 EXPORT_SYMBOL_GPL(pskb_put);
1286 * skb_put - add data to a buffer
1287 * @skb: buffer to use
1288 * @len: amount of data to add
1290 * This function extends the used data area of the buffer. If this would
1291 * exceed the total buffer size the kernel will panic. A pointer to the
1292 * first byte of the extra data is returned.
1294 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1296 unsigned char *tmp = skb_tail_pointer(skb);
1297 SKB_LINEAR_ASSERT(skb);
1300 if (unlikely(skb->tail > skb->end))
1301 skb_over_panic(skb, len, __builtin_return_address(0));
1304 EXPORT_SYMBOL(skb_put);
1307 * skb_push - add data to the start of a buffer
1308 * @skb: buffer to use
1309 * @len: amount of data to add
1311 * This function extends the used data area of the buffer at the buffer
1312 * start. If this would exceed the total buffer headroom the kernel will
1313 * panic. A pointer to the first byte of the extra data is returned.
1315 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1319 if (unlikely(skb->data<skb->head))
1320 skb_under_panic(skb, len, __builtin_return_address(0));
1323 EXPORT_SYMBOL(skb_push);
1326 * skb_pull - remove data from the start of a buffer
1327 * @skb: buffer to use
1328 * @len: amount of data to remove
1330 * This function removes data from the start of a buffer, returning
1331 * the memory to the headroom. A pointer to the next data in the buffer
1332 * is returned. Once the data has been pulled future pushes will overwrite
1335 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1337 return skb_pull_inline(skb, len);
1339 EXPORT_SYMBOL(skb_pull);
1342 * skb_trim - remove end from a buffer
1343 * @skb: buffer to alter
1346 * Cut the length of a buffer down by removing data from the tail. If
1347 * the buffer is already under the length specified it is not modified.
1348 * The skb must be linear.
1350 void skb_trim(struct sk_buff *skb, unsigned int len)
1353 __skb_trim(skb, len);
1355 EXPORT_SYMBOL(skb_trim);
1357 /* Trims skb to length len. It can change skb pointers.
1360 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1362 struct sk_buff **fragp;
1363 struct sk_buff *frag;
1364 int offset = skb_headlen(skb);
1365 int nfrags = skb_shinfo(skb)->nr_frags;
1369 if (skb_cloned(skb) &&
1370 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1377 for (; i < nfrags; i++) {
1378 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1385 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1388 skb_shinfo(skb)->nr_frags = i;
1390 for (; i < nfrags; i++)
1391 skb_frag_unref(skb, i);
1393 if (skb_has_frag_list(skb))
1394 skb_drop_fraglist(skb);
1398 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1399 fragp = &frag->next) {
1400 int end = offset + frag->len;
1402 if (skb_shared(frag)) {
1403 struct sk_buff *nfrag;
1405 nfrag = skb_clone(frag, GFP_ATOMIC);
1406 if (unlikely(!nfrag))
1409 nfrag->next = frag->next;
1421 unlikely((err = pskb_trim(frag, len - offset))))
1425 skb_drop_list(&frag->next);
1430 if (len > skb_headlen(skb)) {
1431 skb->data_len -= skb->len - len;
1436 skb_set_tail_pointer(skb, len);
1441 EXPORT_SYMBOL(___pskb_trim);
1444 * __pskb_pull_tail - advance tail of skb header
1445 * @skb: buffer to reallocate
1446 * @delta: number of bytes to advance tail
1448 * The function makes a sense only on a fragmented &sk_buff,
1449 * it expands header moving its tail forward and copying necessary
1450 * data from fragmented part.
1452 * &sk_buff MUST have reference count of 1.
1454 * Returns %NULL (and &sk_buff does not change) if pull failed
1455 * or value of new tail of skb in the case of success.
1457 * All the pointers pointing into skb header may change and must be
1458 * reloaded after call to this function.
1461 /* Moves tail of skb head forward, copying data from fragmented part,
1462 * when it is necessary.
1463 * 1. It may fail due to malloc failure.
1464 * 2. It may change skb pointers.
1466 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1468 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1470 /* If skb has not enough free space at tail, get new one
1471 * plus 128 bytes for future expansions. If we have enough
1472 * room at tail, reallocate without expansion only if skb is cloned.
1474 int i, k, eat = (skb->tail + delta) - skb->end;
1476 if (eat > 0 || skb_cloned(skb)) {
1477 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1482 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1485 /* Optimization: no fragments, no reasons to preestimate
1486 * size of pulled pages. Superb.
1488 if (!skb_has_frag_list(skb))
1491 /* Estimate size of pulled pages. */
1493 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1494 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1501 /* If we need update frag list, we are in troubles.
1502 * Certainly, it possible to add an offset to skb data,
1503 * but taking into account that pulling is expected to
1504 * be very rare operation, it is worth to fight against
1505 * further bloating skb head and crucify ourselves here instead.
1506 * Pure masohism, indeed. 8)8)
1509 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1510 struct sk_buff *clone = NULL;
1511 struct sk_buff *insp = NULL;
1516 if (list->len <= eat) {
1517 /* Eaten as whole. */
1522 /* Eaten partially. */
1524 if (skb_shared(list)) {
1525 /* Sucks! We need to fork list. :-( */
1526 clone = skb_clone(list, GFP_ATOMIC);
1532 /* This may be pulled without
1536 if (!pskb_pull(list, eat)) {
1544 /* Free pulled out fragments. */
1545 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1546 skb_shinfo(skb)->frag_list = list->next;
1549 /* And insert new clone at head. */
1552 skb_shinfo(skb)->frag_list = clone;
1555 /* Success! Now we may commit changes to skb data. */
1560 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1561 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1564 skb_frag_unref(skb, i);
1567 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1569 skb_shinfo(skb)->frags[k].page_offset += eat;
1570 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1576 skb_shinfo(skb)->nr_frags = k;
1579 skb->data_len -= delta;
1581 return skb_tail_pointer(skb);
1583 EXPORT_SYMBOL(__pskb_pull_tail);
1586 * skb_copy_bits - copy bits from skb to kernel buffer
1588 * @offset: offset in source
1589 * @to: destination buffer
1590 * @len: number of bytes to copy
1592 * Copy the specified number of bytes from the source skb to the
1593 * destination buffer.
1596 * If its prototype is ever changed,
1597 * check arch/{*}/net/{*}.S files,
1598 * since it is called from BPF assembly code.
1600 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1602 int start = skb_headlen(skb);
1603 struct sk_buff *frag_iter;
1606 if (offset > (int)skb->len - len)
1610 if ((copy = start - offset) > 0) {
1613 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1614 if ((len -= copy) == 0)
1620 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1622 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1624 WARN_ON(start > offset + len);
1626 end = start + skb_frag_size(f);
1627 if ((copy = end - offset) > 0) {
1633 vaddr = kmap_atomic(skb_frag_page(f));
1635 vaddr + f->page_offset + offset - start,
1637 kunmap_atomic(vaddr);
1639 if ((len -= copy) == 0)
1647 skb_walk_frags(skb, frag_iter) {
1650 WARN_ON(start > offset + len);
1652 end = start + frag_iter->len;
1653 if ((copy = end - offset) > 0) {
1656 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1658 if ((len -= copy) == 0)
1672 EXPORT_SYMBOL(skb_copy_bits);
1675 * Callback from splice_to_pipe(), if we need to release some pages
1676 * at the end of the spd in case we error'ed out in filling the pipe.
1678 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1680 put_page(spd->pages[i]);
1683 static struct page *linear_to_page(struct page *page, unsigned int *len,
1684 unsigned int *offset,
1687 struct page_frag *pfrag = sk_page_frag(sk);
1689 if (!sk_page_frag_refill(sk, pfrag))
1692 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1694 memcpy(page_address(pfrag->page) + pfrag->offset,
1695 page_address(page) + *offset, *len);
1696 *offset = pfrag->offset;
1697 pfrag->offset += *len;
1702 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1704 unsigned int offset)
1706 return spd->nr_pages &&
1707 spd->pages[spd->nr_pages - 1] == page &&
1708 (spd->partial[spd->nr_pages - 1].offset +
1709 spd->partial[spd->nr_pages - 1].len == offset);
1713 * Fill page/offset/length into spd, if it can hold more pages.
1715 static bool spd_fill_page(struct splice_pipe_desc *spd,
1716 struct pipe_inode_info *pipe, struct page *page,
1717 unsigned int *len, unsigned int offset,
1721 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1725 page = linear_to_page(page, len, &offset, sk);
1729 if (spd_can_coalesce(spd, page, offset)) {
1730 spd->partial[spd->nr_pages - 1].len += *len;
1734 spd->pages[spd->nr_pages] = page;
1735 spd->partial[spd->nr_pages].len = *len;
1736 spd->partial[spd->nr_pages].offset = offset;
1742 static bool __splice_segment(struct page *page, unsigned int poff,
1743 unsigned int plen, unsigned int *off,
1745 struct splice_pipe_desc *spd, bool linear,
1747 struct pipe_inode_info *pipe)
1752 /* skip this segment if already processed */
1758 /* ignore any bits we already processed */
1764 unsigned int flen = min(*len, plen);
1766 if (spd_fill_page(spd, pipe, page, &flen, poff,
1772 } while (*len && plen);
1778 * Map linear and fragment data from the skb to spd. It reports true if the
1779 * pipe is full or if we already spliced the requested length.
1781 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1782 unsigned int *offset, unsigned int *len,
1783 struct splice_pipe_desc *spd, struct sock *sk)
1787 /* map the linear part :
1788 * If skb->head_frag is set, this 'linear' part is backed by a
1789 * fragment, and if the head is not shared with any clones then
1790 * we can avoid a copy since we own the head portion of this page.
1792 if (__splice_segment(virt_to_page(skb->data),
1793 (unsigned long) skb->data & (PAGE_SIZE - 1),
1796 skb_head_is_locked(skb),
1801 * then map the fragments
1803 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1804 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1806 if (__splice_segment(skb_frag_page(f),
1807 f->page_offset, skb_frag_size(f),
1808 offset, len, spd, false, sk, pipe))
1816 * Map data from the skb to a pipe. Should handle both the linear part,
1817 * the fragments, and the frag list. It does NOT handle frag lists within
1818 * the frag list, if such a thing exists. We'd probably need to recurse to
1819 * handle that cleanly.
1821 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1822 struct pipe_inode_info *pipe, unsigned int tlen,
1825 struct partial_page partial[MAX_SKB_FRAGS];
1826 struct page *pages[MAX_SKB_FRAGS];
1827 struct splice_pipe_desc spd = {
1830 .nr_pages_max = MAX_SKB_FRAGS,
1832 .ops = &sock_pipe_buf_ops,
1833 .spd_release = sock_spd_release,
1835 struct sk_buff *frag_iter;
1836 struct sock *sk = skb->sk;
1840 * __skb_splice_bits() only fails if the output has no room left,
1841 * so no point in going over the frag_list for the error case.
1843 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1849 * now see if we have a frag_list to map
1851 skb_walk_frags(skb, frag_iter) {
1854 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1861 * Drop the socket lock, otherwise we have reverse
1862 * locking dependencies between sk_lock and i_mutex
1863 * here as compared to sendfile(). We enter here
1864 * with the socket lock held, and splice_to_pipe() will
1865 * grab the pipe inode lock. For sendfile() emulation,
1866 * we call into ->sendpage() with the i_mutex lock held
1867 * and networking will grab the socket lock.
1870 ret = splice_to_pipe(pipe, &spd);
1878 * skb_store_bits - store bits from kernel buffer to skb
1879 * @skb: destination buffer
1880 * @offset: offset in destination
1881 * @from: source buffer
1882 * @len: number of bytes to copy
1884 * Copy the specified number of bytes from the source buffer to the
1885 * destination skb. This function handles all the messy bits of
1886 * traversing fragment lists and such.
1889 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1891 int start = skb_headlen(skb);
1892 struct sk_buff *frag_iter;
1895 if (offset > (int)skb->len - len)
1898 if ((copy = start - offset) > 0) {
1901 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1902 if ((len -= copy) == 0)
1908 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1909 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1912 WARN_ON(start > offset + len);
1914 end = start + skb_frag_size(frag);
1915 if ((copy = end - offset) > 0) {
1921 vaddr = kmap_atomic(skb_frag_page(frag));
1922 memcpy(vaddr + frag->page_offset + offset - start,
1924 kunmap_atomic(vaddr);
1926 if ((len -= copy) == 0)
1934 skb_walk_frags(skb, frag_iter) {
1937 WARN_ON(start > offset + len);
1939 end = start + frag_iter->len;
1940 if ((copy = end - offset) > 0) {
1943 if (skb_store_bits(frag_iter, offset - start,
1946 if ((len -= copy) == 0)
1959 EXPORT_SYMBOL(skb_store_bits);
1961 /* Checksum skb data. */
1962 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
1963 __wsum csum, const struct skb_checksum_ops *ops)
1965 int start = skb_headlen(skb);
1966 int i, copy = start - offset;
1967 struct sk_buff *frag_iter;
1970 /* Checksum header. */
1974 csum = ops->update(skb->data + offset, copy, csum);
1975 if ((len -= copy) == 0)
1981 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1983 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1985 WARN_ON(start > offset + len);
1987 end = start + skb_frag_size(frag);
1988 if ((copy = end - offset) > 0) {
1994 vaddr = kmap_atomic(skb_frag_page(frag));
1995 csum2 = ops->update(vaddr + frag->page_offset +
1996 offset - start, copy, 0);
1997 kunmap_atomic(vaddr);
1998 csum = ops->combine(csum, csum2, pos, copy);
2007 skb_walk_frags(skb, frag_iter) {
2010 WARN_ON(start > offset + len);
2012 end = start + frag_iter->len;
2013 if ((copy = end - offset) > 0) {
2017 csum2 = __skb_checksum(frag_iter, offset - start,
2019 csum = ops->combine(csum, csum2, pos, copy);
2020 if ((len -= copy) == 0)
2031 EXPORT_SYMBOL(__skb_checksum);
2033 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2034 int len, __wsum csum)
2036 const struct skb_checksum_ops ops = {
2037 .update = csum_partial_ext,
2038 .combine = csum_block_add_ext,
2041 return __skb_checksum(skb, offset, len, csum, &ops);
2043 EXPORT_SYMBOL(skb_checksum);
2045 /* Both of above in one bottle. */
2047 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2048 u8 *to, int len, __wsum csum)
2050 int start = skb_headlen(skb);
2051 int i, copy = start - offset;
2052 struct sk_buff *frag_iter;
2059 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2061 if ((len -= copy) == 0)
2068 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2071 WARN_ON(start > offset + len);
2073 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2074 if ((copy = end - offset) > 0) {
2077 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2081 vaddr = kmap_atomic(skb_frag_page(frag));
2082 csum2 = csum_partial_copy_nocheck(vaddr +
2086 kunmap_atomic(vaddr);
2087 csum = csum_block_add(csum, csum2, pos);
2097 skb_walk_frags(skb, frag_iter) {
2101 WARN_ON(start > offset + len);
2103 end = start + frag_iter->len;
2104 if ((copy = end - offset) > 0) {
2107 csum2 = skb_copy_and_csum_bits(frag_iter,
2110 csum = csum_block_add(csum, csum2, pos);
2111 if ((len -= copy) == 0)
2122 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2124 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2129 if (skb->ip_summed == CHECKSUM_PARTIAL)
2130 csstart = skb_checksum_start_offset(skb);
2132 csstart = skb_headlen(skb);
2134 BUG_ON(csstart > skb_headlen(skb));
2136 skb_copy_from_linear_data(skb, to, csstart);
2139 if (csstart != skb->len)
2140 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2141 skb->len - csstart, 0);
2143 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2144 long csstuff = csstart + skb->csum_offset;
2146 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2149 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2152 * skb_dequeue - remove from the head of the queue
2153 * @list: list to dequeue from
2155 * Remove the head of the list. The list lock is taken so the function
2156 * may be used safely with other locking list functions. The head item is
2157 * returned or %NULL if the list is empty.
2160 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2162 unsigned long flags;
2163 struct sk_buff *result;
2165 spin_lock_irqsave(&list->lock, flags);
2166 result = __skb_dequeue(list);
2167 spin_unlock_irqrestore(&list->lock, flags);
2170 EXPORT_SYMBOL(skb_dequeue);
2173 * skb_dequeue_tail - remove from the tail of the queue
2174 * @list: list to dequeue from
2176 * Remove the tail of the list. The list lock is taken so the function
2177 * may be used safely with other locking list functions. The tail item is
2178 * returned or %NULL if the list is empty.
2180 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2182 unsigned long flags;
2183 struct sk_buff *result;
2185 spin_lock_irqsave(&list->lock, flags);
2186 result = __skb_dequeue_tail(list);
2187 spin_unlock_irqrestore(&list->lock, flags);
2190 EXPORT_SYMBOL(skb_dequeue_tail);
2193 * skb_queue_purge - empty a list
2194 * @list: list to empty
2196 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2197 * the list and one reference dropped. This function takes the list
2198 * lock and is atomic with respect to other list locking functions.
2200 void skb_queue_purge(struct sk_buff_head *list)
2202 struct sk_buff *skb;
2203 while ((skb = skb_dequeue(list)) != NULL)
2206 EXPORT_SYMBOL(skb_queue_purge);
2209 * skb_queue_head - queue a buffer at the list head
2210 * @list: list to use
2211 * @newsk: buffer to queue
2213 * Queue a buffer at the start of the list. This function takes the
2214 * list lock and can be used safely with other locking &sk_buff functions
2217 * A buffer cannot be placed on two lists at the same time.
2219 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2221 unsigned long flags;
2223 spin_lock_irqsave(&list->lock, flags);
2224 __skb_queue_head(list, newsk);
2225 spin_unlock_irqrestore(&list->lock, flags);
2227 EXPORT_SYMBOL(skb_queue_head);
2230 * skb_queue_tail - queue a buffer at the list tail
2231 * @list: list to use
2232 * @newsk: buffer to queue
2234 * Queue a buffer at the tail of the list. This function takes the
2235 * list lock and can be used safely with other locking &sk_buff functions
2238 * A buffer cannot be placed on two lists at the same time.
2240 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2242 unsigned long flags;
2244 spin_lock_irqsave(&list->lock, flags);
2245 __skb_queue_tail(list, newsk);
2246 spin_unlock_irqrestore(&list->lock, flags);
2248 EXPORT_SYMBOL(skb_queue_tail);
2251 * skb_unlink - remove a buffer from a list
2252 * @skb: buffer to remove
2253 * @list: list to use
2255 * Remove a packet from a list. The list locks are taken and this
2256 * function is atomic with respect to other list locked calls
2258 * You must know what list the SKB is on.
2260 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2262 unsigned long flags;
2264 spin_lock_irqsave(&list->lock, flags);
2265 __skb_unlink(skb, list);
2266 spin_unlock_irqrestore(&list->lock, flags);
2268 EXPORT_SYMBOL(skb_unlink);
2271 * skb_append - append a buffer
2272 * @old: buffer to insert after
2273 * @newsk: buffer to insert
2274 * @list: list to use
2276 * Place a packet after a given packet in a list. The list locks are taken
2277 * and this function is atomic with respect to other list locked calls.
2278 * A buffer cannot be placed on two lists at the same time.
2280 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2282 unsigned long flags;
2284 spin_lock_irqsave(&list->lock, flags);
2285 __skb_queue_after(list, old, newsk);
2286 spin_unlock_irqrestore(&list->lock, flags);
2288 EXPORT_SYMBOL(skb_append);
2291 * skb_insert - insert a buffer
2292 * @old: buffer to insert before
2293 * @newsk: buffer to insert
2294 * @list: list to use
2296 * Place a packet before a given packet in a list. The list locks are
2297 * taken and this function is atomic with respect to other list locked
2300 * A buffer cannot be placed on two lists at the same time.
2302 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2304 unsigned long flags;
2306 spin_lock_irqsave(&list->lock, flags);
2307 __skb_insert(newsk, old->prev, old, list);
2308 spin_unlock_irqrestore(&list->lock, flags);
2310 EXPORT_SYMBOL(skb_insert);
2312 static inline void skb_split_inside_header(struct sk_buff *skb,
2313 struct sk_buff* skb1,
2314 const u32 len, const int pos)
2318 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2320 /* And move data appendix as is. */
2321 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2322 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2324 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2325 skb_shinfo(skb)->nr_frags = 0;
2326 skb1->data_len = skb->data_len;
2327 skb1->len += skb1->data_len;
2330 skb_set_tail_pointer(skb, len);
2333 static inline void skb_split_no_header(struct sk_buff *skb,
2334 struct sk_buff* skb1,
2335 const u32 len, int pos)
2338 const int nfrags = skb_shinfo(skb)->nr_frags;
2340 skb_shinfo(skb)->nr_frags = 0;
2341 skb1->len = skb1->data_len = skb->len - len;
2343 skb->data_len = len - pos;
2345 for (i = 0; i < nfrags; i++) {
2346 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2348 if (pos + size > len) {
2349 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2353 * We have two variants in this case:
2354 * 1. Move all the frag to the second
2355 * part, if it is possible. F.e.
2356 * this approach is mandatory for TUX,
2357 * where splitting is expensive.
2358 * 2. Split is accurately. We make this.
2360 skb_frag_ref(skb, i);
2361 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2362 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2363 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2364 skb_shinfo(skb)->nr_frags++;
2368 skb_shinfo(skb)->nr_frags++;
2371 skb_shinfo(skb1)->nr_frags = k;
2375 * skb_split - Split fragmented skb to two parts at length len.
2376 * @skb: the buffer to split
2377 * @skb1: the buffer to receive the second part
2378 * @len: new length for skb
2380 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2382 int pos = skb_headlen(skb);
2384 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2385 if (len < pos) /* Split line is inside header. */
2386 skb_split_inside_header(skb, skb1, len, pos);
2387 else /* Second chunk has no header, nothing to copy. */
2388 skb_split_no_header(skb, skb1, len, pos);
2390 EXPORT_SYMBOL(skb_split);
2392 /* Shifting from/to a cloned skb is a no-go.
2394 * Caller cannot keep skb_shinfo related pointers past calling here!
2396 static int skb_prepare_for_shift(struct sk_buff *skb)
2398 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2402 * skb_shift - Shifts paged data partially from skb to another
2403 * @tgt: buffer into which tail data gets added
2404 * @skb: buffer from which the paged data comes from
2405 * @shiftlen: shift up to this many bytes
2407 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2408 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2409 * It's up to caller to free skb if everything was shifted.
2411 * If @tgt runs out of frags, the whole operation is aborted.
2413 * Skb cannot include anything else but paged data while tgt is allowed
2414 * to have non-paged data as well.
2416 * TODO: full sized shift could be optimized but that would need
2417 * specialized skb free'er to handle frags without up-to-date nr_frags.
2419 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2421 int from, to, merge, todo;
2422 struct skb_frag_struct *fragfrom, *fragto;
2424 BUG_ON(shiftlen > skb->len);
2425 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2429 to = skb_shinfo(tgt)->nr_frags;
2430 fragfrom = &skb_shinfo(skb)->frags[from];
2432 /* Actual merge is delayed until the point when we know we can
2433 * commit all, so that we don't have to undo partial changes
2436 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2437 fragfrom->page_offset)) {
2442 todo -= skb_frag_size(fragfrom);
2444 if (skb_prepare_for_shift(skb) ||
2445 skb_prepare_for_shift(tgt))
2448 /* All previous frag pointers might be stale! */
2449 fragfrom = &skb_shinfo(skb)->frags[from];
2450 fragto = &skb_shinfo(tgt)->frags[merge];
2452 skb_frag_size_add(fragto, shiftlen);
2453 skb_frag_size_sub(fragfrom, shiftlen);
2454 fragfrom->page_offset += shiftlen;
2462 /* Skip full, not-fitting skb to avoid expensive operations */
2463 if ((shiftlen == skb->len) &&
2464 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2467 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2470 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2471 if (to == MAX_SKB_FRAGS)
2474 fragfrom = &skb_shinfo(skb)->frags[from];
2475 fragto = &skb_shinfo(tgt)->frags[to];
2477 if (todo >= skb_frag_size(fragfrom)) {
2478 *fragto = *fragfrom;
2479 todo -= skb_frag_size(fragfrom);
2484 __skb_frag_ref(fragfrom);
2485 fragto->page = fragfrom->page;
2486 fragto->page_offset = fragfrom->page_offset;
2487 skb_frag_size_set(fragto, todo);
2489 fragfrom->page_offset += todo;
2490 skb_frag_size_sub(fragfrom, todo);
2498 /* Ready to "commit" this state change to tgt */
2499 skb_shinfo(tgt)->nr_frags = to;
2502 fragfrom = &skb_shinfo(skb)->frags[0];
2503 fragto = &skb_shinfo(tgt)->frags[merge];
2505 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2506 __skb_frag_unref(fragfrom);
2509 /* Reposition in the original skb */
2511 while (from < skb_shinfo(skb)->nr_frags)
2512 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2513 skb_shinfo(skb)->nr_frags = to;
2515 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2518 /* Most likely the tgt won't ever need its checksum anymore, skb on
2519 * the other hand might need it if it needs to be resent
2521 tgt->ip_summed = CHECKSUM_PARTIAL;
2522 skb->ip_summed = CHECKSUM_PARTIAL;
2524 /* Yak, is it really working this way? Some helper please? */
2525 skb->len -= shiftlen;
2526 skb->data_len -= shiftlen;
2527 skb->truesize -= shiftlen;
2528 tgt->len += shiftlen;
2529 tgt->data_len += shiftlen;
2530 tgt->truesize += shiftlen;
2536 * skb_prepare_seq_read - Prepare a sequential read of skb data
2537 * @skb: the buffer to read
2538 * @from: lower offset of data to be read
2539 * @to: upper offset of data to be read
2540 * @st: state variable
2542 * Initializes the specified state variable. Must be called before
2543 * invoking skb_seq_read() for the first time.
2545 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2546 unsigned int to, struct skb_seq_state *st)
2548 st->lower_offset = from;
2549 st->upper_offset = to;
2550 st->root_skb = st->cur_skb = skb;
2551 st->frag_idx = st->stepped_offset = 0;
2552 st->frag_data = NULL;
2554 EXPORT_SYMBOL(skb_prepare_seq_read);
2557 * skb_seq_read - Sequentially read skb data
2558 * @consumed: number of bytes consumed by the caller so far
2559 * @data: destination pointer for data to be returned
2560 * @st: state variable
2562 * Reads a block of skb data at @consumed relative to the
2563 * lower offset specified to skb_prepare_seq_read(). Assigns
2564 * the head of the data block to @data and returns the length
2565 * of the block or 0 if the end of the skb data or the upper
2566 * offset has been reached.
2568 * The caller is not required to consume all of the data
2569 * returned, i.e. @consumed is typically set to the number
2570 * of bytes already consumed and the next call to
2571 * skb_seq_read() will return the remaining part of the block.
2573 * Note 1: The size of each block of data returned can be arbitrary,
2574 * this limitation is the cost for zerocopy seqeuental
2575 * reads of potentially non linear data.
2577 * Note 2: Fragment lists within fragments are not implemented
2578 * at the moment, state->root_skb could be replaced with
2579 * a stack for this purpose.
2581 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2582 struct skb_seq_state *st)
2584 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2587 if (unlikely(abs_offset >= st->upper_offset)) {
2588 if (st->frag_data) {
2589 kunmap_atomic(st->frag_data);
2590 st->frag_data = NULL;
2596 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2598 if (abs_offset < block_limit && !st->frag_data) {
2599 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2600 return block_limit - abs_offset;
2603 if (st->frag_idx == 0 && !st->frag_data)
2604 st->stepped_offset += skb_headlen(st->cur_skb);
2606 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2607 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2608 block_limit = skb_frag_size(frag) + st->stepped_offset;
2610 if (abs_offset < block_limit) {
2612 st->frag_data = kmap_atomic(skb_frag_page(frag));
2614 *data = (u8 *) st->frag_data + frag->page_offset +
2615 (abs_offset - st->stepped_offset);
2617 return block_limit - abs_offset;
2620 if (st->frag_data) {
2621 kunmap_atomic(st->frag_data);
2622 st->frag_data = NULL;
2626 st->stepped_offset += skb_frag_size(frag);
2629 if (st->frag_data) {
2630 kunmap_atomic(st->frag_data);
2631 st->frag_data = NULL;
2634 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2635 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2638 } else if (st->cur_skb->next) {
2639 st->cur_skb = st->cur_skb->next;
2646 EXPORT_SYMBOL(skb_seq_read);
2649 * skb_abort_seq_read - Abort a sequential read of skb data
2650 * @st: state variable
2652 * Must be called if skb_seq_read() was not called until it
2655 void skb_abort_seq_read(struct skb_seq_state *st)
2658 kunmap_atomic(st->frag_data);
2660 EXPORT_SYMBOL(skb_abort_seq_read);
2662 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2664 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2665 struct ts_config *conf,
2666 struct ts_state *state)
2668 return skb_seq_read(offset, text, TS_SKB_CB(state));
2671 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2673 skb_abort_seq_read(TS_SKB_CB(state));
2677 * skb_find_text - Find a text pattern in skb data
2678 * @skb: the buffer to look in
2679 * @from: search offset
2681 * @config: textsearch configuration
2682 * @state: uninitialized textsearch state variable
2684 * Finds a pattern in the skb data according to the specified
2685 * textsearch configuration. Use textsearch_next() to retrieve
2686 * subsequent occurrences of the pattern. Returns the offset
2687 * to the first occurrence or UINT_MAX if no match was found.
2689 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2690 unsigned int to, struct ts_config *config,
2691 struct ts_state *state)
2695 config->get_next_block = skb_ts_get_next_block;
2696 config->finish = skb_ts_finish;
2698 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2700 ret = textsearch_find(config, state);
2701 return (ret <= to - from ? ret : UINT_MAX);
2703 EXPORT_SYMBOL(skb_find_text);
2706 * skb_append_datato_frags - append the user data to a skb
2707 * @sk: sock structure
2708 * @skb: skb structure to be appened with user data.
2709 * @getfrag: call back function to be used for getting the user data
2710 * @from: pointer to user message iov
2711 * @length: length of the iov message
2713 * Description: This procedure append the user data in the fragment part
2714 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2716 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2717 int (*getfrag)(void *from, char *to, int offset,
2718 int len, int odd, struct sk_buff *skb),
2719 void *from, int length)
2721 int frg_cnt = skb_shinfo(skb)->nr_frags;
2725 struct page_frag *pfrag = ¤t->task_frag;
2728 /* Return error if we don't have space for new frag */
2729 if (frg_cnt >= MAX_SKB_FRAGS)
2732 if (!sk_page_frag_refill(sk, pfrag))
2735 /* copy the user data to page */
2736 copy = min_t(int, length, pfrag->size - pfrag->offset);
2738 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2739 offset, copy, 0, skb);
2743 /* copy was successful so update the size parameters */
2744 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2747 pfrag->offset += copy;
2748 get_page(pfrag->page);
2750 skb->truesize += copy;
2751 atomic_add(copy, &sk->sk_wmem_alloc);
2753 skb->data_len += copy;
2757 } while (length > 0);
2761 EXPORT_SYMBOL(skb_append_datato_frags);
2764 * skb_pull_rcsum - pull skb and update receive checksum
2765 * @skb: buffer to update
2766 * @len: length of data pulled
2768 * This function performs an skb_pull on the packet and updates
2769 * the CHECKSUM_COMPLETE checksum. It should be used on
2770 * receive path processing instead of skb_pull unless you know
2771 * that the checksum difference is zero (e.g., a valid IP header)
2772 * or you are setting ip_summed to CHECKSUM_NONE.
2774 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2776 BUG_ON(len > skb->len);
2778 BUG_ON(skb->len < skb->data_len);
2779 skb_postpull_rcsum(skb, skb->data, len);
2780 return skb->data += len;
2782 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2785 * skb_segment - Perform protocol segmentation on skb.
2786 * @skb: buffer to segment
2787 * @features: features for the output path (see dev->features)
2789 * This function performs segmentation on the given skb. It returns
2790 * a pointer to the first in a list of new skbs for the segments.
2791 * In case of error it returns ERR_PTR(err).
2793 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2795 struct sk_buff *segs = NULL;
2796 struct sk_buff *tail = NULL;
2797 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2798 skb_frag_t *skb_frag = skb_shinfo(skb)->frags;
2799 unsigned int mss = skb_shinfo(skb)->gso_size;
2800 unsigned int doffset = skb->data - skb_mac_header(skb);
2801 unsigned int offset = doffset;
2802 unsigned int tnl_hlen = skb_tnl_header_len(skb);
2803 unsigned int headroom;
2807 int sg = !!(features & NETIF_F_SG);
2808 int nfrags = skb_shinfo(skb)->nr_frags;
2813 proto = skb_network_protocol(skb);
2814 if (unlikely(!proto))
2815 return ERR_PTR(-EINVAL);
2817 csum = !!can_checksum_protocol(features, proto);
2818 __skb_push(skb, doffset);
2819 headroom = skb_headroom(skb);
2820 pos = skb_headlen(skb);
2823 struct sk_buff *nskb;
2828 len = skb->len - offset;
2832 hsize = skb_headlen(skb) - offset;
2835 if (hsize > len || !sg)
2838 if (!hsize && i >= nfrags && skb_headlen(fskb) &&
2839 (skb_headlen(fskb) == len || sg)) {
2840 BUG_ON(skb_headlen(fskb) > len);
2843 nfrags = skb_shinfo(fskb)->nr_frags;
2844 skb_frag = skb_shinfo(fskb)->frags;
2845 pos += skb_headlen(fskb);
2847 while (pos < offset + len) {
2848 BUG_ON(i >= nfrags);
2850 size = skb_frag_size(skb_frag);
2851 if (pos + size > offset + len)
2859 nskb = skb_clone(fskb, GFP_ATOMIC);
2862 if (unlikely(!nskb))
2865 if (unlikely(pskb_trim(nskb, len))) {
2870 hsize = skb_end_offset(nskb);
2871 if (skb_cow_head(nskb, doffset + headroom)) {
2876 nskb->truesize += skb_end_offset(nskb) - hsize;
2877 skb_release_head_state(nskb);
2878 __skb_push(nskb, doffset);
2880 nskb = __alloc_skb(hsize + doffset + headroom,
2881 GFP_ATOMIC, skb_alloc_rx_flag(skb),
2884 if (unlikely(!nskb))
2887 skb_reserve(nskb, headroom);
2888 __skb_put(nskb, doffset);
2897 __copy_skb_header(nskb, skb);
2898 nskb->mac_len = skb->mac_len;
2900 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
2902 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
2903 nskb->data - tnl_hlen,
2904 doffset + tnl_hlen);
2906 if (nskb->len == len + doffset)
2907 goto perform_csum_check;
2910 nskb->ip_summed = CHECKSUM_NONE;
2911 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2917 frag = skb_shinfo(nskb)->frags;
2919 skb_copy_from_linear_data_offset(skb, offset,
2920 skb_put(nskb, hsize), hsize);
2922 skb_shinfo(nskb)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2924 while (pos < offset + len) {
2926 BUG_ON(skb_headlen(fskb));
2929 nfrags = skb_shinfo(fskb)->nr_frags;
2930 skb_frag = skb_shinfo(fskb)->frags;
2937 if (unlikely(skb_shinfo(nskb)->nr_frags >=
2939 net_warn_ratelimited(
2940 "skb_segment: too many frags: %u %u\n",
2946 __skb_frag_ref(frag);
2947 size = skb_frag_size(frag);
2950 frag->page_offset += offset - pos;
2951 skb_frag_size_sub(frag, offset - pos);
2954 skb_shinfo(nskb)->nr_frags++;
2956 if (pos + size <= offset + len) {
2961 skb_frag_size_sub(frag, pos + size - (offset + len));
2969 nskb->data_len = len - hsize;
2970 nskb->len += nskb->data_len;
2971 nskb->truesize += nskb->data_len;
2975 nskb->csum = skb_checksum(nskb, doffset,
2976 nskb->len - doffset, 0);
2977 nskb->ip_summed = CHECKSUM_NONE;
2979 } while ((offset += len) < skb->len);
2984 kfree_skb_list(segs);
2985 return ERR_PTR(err);
2987 EXPORT_SYMBOL_GPL(skb_segment);
2989 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2991 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
2992 unsigned int offset = skb_gro_offset(skb);
2993 unsigned int headlen = skb_headlen(skb);
2994 struct sk_buff *nskb, *lp, *p = *head;
2995 unsigned int len = skb_gro_len(skb);
2996 unsigned int delta_truesize;
2997 unsigned int headroom;
2999 if (unlikely(p->len + len >= 65536))
3002 lp = NAPI_GRO_CB(p)->last ?: p;
3003 pinfo = skb_shinfo(lp);
3005 if (headlen <= offset) {
3008 int i = skbinfo->nr_frags;
3009 int nr_frags = pinfo->nr_frags + i;
3011 if (nr_frags > MAX_SKB_FRAGS)
3015 pinfo->nr_frags = nr_frags;
3016 skbinfo->nr_frags = 0;
3018 frag = pinfo->frags + nr_frags;
3019 frag2 = skbinfo->frags + i;
3024 frag->page_offset += offset;
3025 skb_frag_size_sub(frag, offset);
3027 /* all fragments truesize : remove (head size + sk_buff) */
3028 delta_truesize = skb->truesize -
3029 SKB_TRUESIZE(skb_end_offset(skb));
3031 skb->truesize -= skb->data_len;
3032 skb->len -= skb->data_len;
3035 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3037 } else if (skb->head_frag) {
3038 int nr_frags = pinfo->nr_frags;
3039 skb_frag_t *frag = pinfo->frags + nr_frags;
3040 struct page *page = virt_to_head_page(skb->head);
3041 unsigned int first_size = headlen - offset;
3042 unsigned int first_offset;
3044 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3047 first_offset = skb->data -
3048 (unsigned char *)page_address(page) +
3051 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3053 frag->page.p = page;
3054 frag->page_offset = first_offset;
3055 skb_frag_size_set(frag, first_size);
3057 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3058 /* We dont need to clear skbinfo->nr_frags here */
3060 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3061 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3064 if (pinfo->frag_list)
3066 if (skb_gro_len(p) != pinfo->gso_size)
3069 headroom = skb_headroom(p);
3070 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
3071 if (unlikely(!nskb))
3074 __copy_skb_header(nskb, p);
3075 nskb->mac_len = p->mac_len;
3077 skb_reserve(nskb, headroom);
3078 __skb_put(nskb, skb_gro_offset(p));
3080 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3081 skb_set_network_header(nskb, skb_network_offset(p));
3082 skb_set_transport_header(nskb, skb_transport_offset(p));
3084 __skb_pull(p, skb_gro_offset(p));
3085 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3086 p->data - skb_mac_header(p));
3088 skb_shinfo(nskb)->frag_list = p;
3089 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3090 pinfo->gso_size = 0;
3091 skb_header_release(p);
3092 NAPI_GRO_CB(nskb)->last = p;
3094 nskb->data_len += p->len;
3095 nskb->truesize += p->truesize;
3096 nskb->len += p->len;
3099 nskb->next = p->next;
3105 delta_truesize = skb->truesize;
3106 if (offset > headlen) {
3107 unsigned int eat = offset - headlen;
3109 skbinfo->frags[0].page_offset += eat;
3110 skb_frag_size_sub(&skbinfo->frags[0], eat);
3111 skb->data_len -= eat;
3116 __skb_pull(skb, offset);
3118 if (!NAPI_GRO_CB(p)->last)
3119 skb_shinfo(p)->frag_list = skb;
3121 NAPI_GRO_CB(p)->last->next = skb;
3122 NAPI_GRO_CB(p)->last = skb;
3123 skb_header_release(skb);
3127 NAPI_GRO_CB(p)->count++;
3129 p->truesize += delta_truesize;
3132 lp->data_len += len;
3133 lp->truesize += delta_truesize;
3136 NAPI_GRO_CB(skb)->same_flow = 1;
3139 EXPORT_SYMBOL_GPL(skb_gro_receive);
3141 void __init skb_init(void)
3143 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3144 sizeof(struct sk_buff),
3146 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3148 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3149 (2*sizeof(struct sk_buff)) +
3152 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3157 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3158 * @skb: Socket buffer containing the buffers to be mapped
3159 * @sg: The scatter-gather list to map into
3160 * @offset: The offset into the buffer's contents to start mapping
3161 * @len: Length of buffer space to be mapped
3163 * Fill the specified scatter-gather list with mappings/pointers into a
3164 * region of the buffer space attached to a socket buffer.
3167 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3169 int start = skb_headlen(skb);
3170 int i, copy = start - offset;
3171 struct sk_buff *frag_iter;
3177 sg_set_buf(sg, skb->data + offset, copy);
3179 if ((len -= copy) == 0)
3184 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3187 WARN_ON(start > offset + len);
3189 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3190 if ((copy = end - offset) > 0) {
3191 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3195 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3196 frag->page_offset+offset-start);
3205 skb_walk_frags(skb, frag_iter) {
3208 WARN_ON(start > offset + len);
3210 end = start + frag_iter->len;
3211 if ((copy = end - offset) > 0) {
3214 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3216 if ((len -= copy) == 0)
3226 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3228 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3230 sg_mark_end(&sg[nsg - 1]);
3234 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3237 * skb_cow_data - Check that a socket buffer's data buffers are writable
3238 * @skb: The socket buffer to check.
3239 * @tailbits: Amount of trailing space to be added
3240 * @trailer: Returned pointer to the skb where the @tailbits space begins
3242 * Make sure that the data buffers attached to a socket buffer are
3243 * writable. If they are not, private copies are made of the data buffers
3244 * and the socket buffer is set to use these instead.
3246 * If @tailbits is given, make sure that there is space to write @tailbits
3247 * bytes of data beyond current end of socket buffer. @trailer will be
3248 * set to point to the skb in which this space begins.
3250 * The number of scatterlist elements required to completely map the
3251 * COW'd and extended socket buffer will be returned.
3253 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3257 struct sk_buff *skb1, **skb_p;
3259 /* If skb is cloned or its head is paged, reallocate
3260 * head pulling out all the pages (pages are considered not writable
3261 * at the moment even if they are anonymous).
3263 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3264 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3267 /* Easy case. Most of packets will go this way. */
3268 if (!skb_has_frag_list(skb)) {
3269 /* A little of trouble, not enough of space for trailer.
3270 * This should not happen, when stack is tuned to generate
3271 * good frames. OK, on miss we reallocate and reserve even more
3272 * space, 128 bytes is fair. */
3274 if (skb_tailroom(skb) < tailbits &&
3275 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3283 /* Misery. We are in troubles, going to mincer fragments... */
3286 skb_p = &skb_shinfo(skb)->frag_list;
3289 while ((skb1 = *skb_p) != NULL) {
3292 /* The fragment is partially pulled by someone,
3293 * this can happen on input. Copy it and everything
3296 if (skb_shared(skb1))
3299 /* If the skb is the last, worry about trailer. */
3301 if (skb1->next == NULL && tailbits) {
3302 if (skb_shinfo(skb1)->nr_frags ||
3303 skb_has_frag_list(skb1) ||
3304 skb_tailroom(skb1) < tailbits)
3305 ntail = tailbits + 128;
3311 skb_shinfo(skb1)->nr_frags ||
3312 skb_has_frag_list(skb1)) {
3313 struct sk_buff *skb2;
3315 /* Fuck, we are miserable poor guys... */
3317 skb2 = skb_copy(skb1, GFP_ATOMIC);
3319 skb2 = skb_copy_expand(skb1,
3323 if (unlikely(skb2 == NULL))
3327 skb_set_owner_w(skb2, skb1->sk);
3329 /* Looking around. Are we still alive?
3330 * OK, link new skb, drop old one */
3332 skb2->next = skb1->next;
3339 skb_p = &skb1->next;
3344 EXPORT_SYMBOL_GPL(skb_cow_data);
3346 static void sock_rmem_free(struct sk_buff *skb)
3348 struct sock *sk = skb->sk;
3350 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3354 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3356 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3360 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3361 (unsigned int)sk->sk_rcvbuf)
3366 skb->destructor = sock_rmem_free;
3367 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3369 /* before exiting rcu section, make sure dst is refcounted */
3372 skb_queue_tail(&sk->sk_error_queue, skb);
3373 if (!sock_flag(sk, SOCK_DEAD))
3374 sk->sk_data_ready(sk, len);
3377 EXPORT_SYMBOL(sock_queue_err_skb);
3379 void skb_tstamp_tx(struct sk_buff *orig_skb,
3380 struct skb_shared_hwtstamps *hwtstamps)
3382 struct sock *sk = orig_skb->sk;
3383 struct sock_exterr_skb *serr;
3384 struct sk_buff *skb;
3391 *skb_hwtstamps(orig_skb) =
3395 * no hardware time stamps available,
3396 * so keep the shared tx_flags and only
3397 * store software time stamp
3399 orig_skb->tstamp = ktime_get_real();
3402 skb = skb_clone(orig_skb, GFP_ATOMIC);
3406 serr = SKB_EXT_ERR(skb);
3407 memset(serr, 0, sizeof(*serr));
3408 serr->ee.ee_errno = ENOMSG;
3409 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3411 err = sock_queue_err_skb(sk, skb);
3416 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3418 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3420 struct sock *sk = skb->sk;
3421 struct sock_exterr_skb *serr;
3424 skb->wifi_acked_valid = 1;
3425 skb->wifi_acked = acked;
3427 serr = SKB_EXT_ERR(skb);
3428 memset(serr, 0, sizeof(*serr));
3429 serr->ee.ee_errno = ENOMSG;
3430 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3432 err = sock_queue_err_skb(sk, skb);
3436 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3440 * skb_partial_csum_set - set up and verify partial csum values for packet
3441 * @skb: the skb to set
3442 * @start: the number of bytes after skb->data to start checksumming.
3443 * @off: the offset from start to place the checksum.
3445 * For untrusted partially-checksummed packets, we need to make sure the values
3446 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3448 * This function checks and sets those values and skb->ip_summed: if this
3449 * returns false you should drop the packet.
3451 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3453 if (unlikely(start > skb_headlen(skb)) ||
3454 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3455 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3456 start, off, skb_headlen(skb));
3459 skb->ip_summed = CHECKSUM_PARTIAL;
3460 skb->csum_start = skb_headroom(skb) + start;
3461 skb->csum_offset = off;
3462 skb_set_transport_header(skb, start);
3465 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3467 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3469 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3472 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3474 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3477 skb_release_head_state(skb);
3478 kmem_cache_free(skbuff_head_cache, skb);
3483 EXPORT_SYMBOL(kfree_skb_partial);
3486 * skb_try_coalesce - try to merge skb to prior one
3488 * @from: buffer to add
3489 * @fragstolen: pointer to boolean
3490 * @delta_truesize: how much more was allocated than was requested
3492 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3493 bool *fragstolen, int *delta_truesize)
3495 int i, delta, len = from->len;
3497 *fragstolen = false;
3502 if (len <= skb_tailroom(to)) {
3503 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3504 *delta_truesize = 0;
3508 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3511 if (skb_headlen(from) != 0) {
3513 unsigned int offset;
3515 if (skb_shinfo(to)->nr_frags +
3516 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3519 if (skb_head_is_locked(from))
3522 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3524 page = virt_to_head_page(from->head);
3525 offset = from->data - (unsigned char *)page_address(page);
3527 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3528 page, offset, skb_headlen(from));
3531 if (skb_shinfo(to)->nr_frags +
3532 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3535 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
3538 WARN_ON_ONCE(delta < len);
3540 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3541 skb_shinfo(from)->frags,
3542 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3543 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3545 if (!skb_cloned(from))
3546 skb_shinfo(from)->nr_frags = 0;
3548 /* if the skb is not cloned this does nothing
3549 * since we set nr_frags to 0.
3551 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3552 skb_frag_ref(from, i);
3554 to->truesize += delta;
3556 to->data_len += len;
3558 *delta_truesize = delta;
3561 EXPORT_SYMBOL(skb_try_coalesce);
3564 * skb_scrub_packet - scrub an skb
3566 * @skb: buffer to clean
3567 * @xnet: packet is crossing netns
3569 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
3570 * into/from a tunnel. Some information have to be cleared during these
3572 * skb_scrub_packet can also be used to clean a skb before injecting it in
3573 * another namespace (@xnet == true). We have to clear all information in the
3574 * skb that could impact namespace isolation.
3576 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
3580 skb->tstamp.tv64 = 0;
3581 skb->pkt_type = PACKET_HOST;
3588 nf_reset_trace(skb);
3590 EXPORT_SYMBOL_GPL(skb_scrub_packet);