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;
78 * skb_panic - private function for out-of-line support
82 * @msg: skb_over_panic or skb_under_panic
84 * Out-of-line support for skb_put() and skb_push().
85 * Called via the wrapper skb_over_panic() or skb_under_panic().
86 * Keep out of line to prevent kernel bloat.
87 * __builtin_return_address is not used because it is not always reliable.
89 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
92 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
93 msg, addr, skb->len, sz, skb->head, skb->data,
94 (unsigned long)skb->tail, (unsigned long)skb->end,
95 skb->dev ? skb->dev->name : "<NULL>");
99 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
101 skb_panic(skb, sz, addr, __func__);
104 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
106 skb_panic(skb, sz, addr, __func__);
110 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
111 * the caller if emergency pfmemalloc reserves are being used. If it is and
112 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
113 * may be used. Otherwise, the packet data may be discarded until enough
116 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
117 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
119 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
120 unsigned long ip, bool *pfmemalloc)
123 bool ret_pfmemalloc = false;
126 * Try a regular allocation, when that fails and we're not entitled
127 * to the reserves, fail.
129 obj = kmalloc_node_track_caller(size,
130 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
132 if (obj || !(gfp_pfmemalloc_allowed(flags)))
135 /* Try again but now we are using pfmemalloc reserves */
136 ret_pfmemalloc = true;
137 obj = kmalloc_node_track_caller(size, flags, node);
141 *pfmemalloc = ret_pfmemalloc;
146 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
147 * 'private' fields and also do memory statistics to find all the
152 struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
157 skb = kmem_cache_alloc_node(skbuff_head_cache,
158 gfp_mask & ~__GFP_DMA, node);
163 * Only clear those fields we need to clear, not those that we will
164 * actually initialise below. Hence, don't put any more fields after
165 * the tail pointer in struct sk_buff!
167 memset(skb, 0, offsetof(struct sk_buff, tail));
169 skb->truesize = sizeof(struct sk_buff);
170 atomic_set(&skb->users, 1);
172 skb->mac_header = (typeof(skb->mac_header))~0U;
178 * __alloc_skb - allocate a network buffer
179 * @size: size to allocate
180 * @gfp_mask: allocation mask
181 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
182 * instead of head cache and allocate a cloned (child) skb.
183 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
184 * allocations in case the data is required for writeback
185 * @node: numa node to allocate memory on
187 * Allocate a new &sk_buff. The returned buffer has no headroom and a
188 * tail room of at least size bytes. The object has a reference count
189 * of one. The return is the buffer. On a failure the return is %NULL.
191 * Buffers may only be allocated from interrupts using a @gfp_mask of
194 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
197 struct kmem_cache *cache;
198 struct skb_shared_info *shinfo;
203 cache = (flags & SKB_ALLOC_FCLONE)
204 ? skbuff_fclone_cache : skbuff_head_cache;
206 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
207 gfp_mask |= __GFP_MEMALLOC;
210 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
215 /* We do our best to align skb_shared_info on a separate cache
216 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
217 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
218 * Both skb->head and skb_shared_info are cache line aligned.
220 size = SKB_DATA_ALIGN(size);
221 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
222 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
225 /* kmalloc(size) might give us more room than requested.
226 * Put skb_shared_info exactly at the end of allocated zone,
227 * to allow max possible filling before reallocation.
229 size = SKB_WITH_OVERHEAD(ksize(data));
230 prefetchw(data + size);
233 * Only clear those fields we need to clear, not those that we will
234 * actually initialise below. Hence, don't put any more fields after
235 * the tail pointer in struct sk_buff!
237 memset(skb, 0, offsetof(struct sk_buff, tail));
238 /* Account for allocated memory : skb + skb->head */
239 skb->truesize = SKB_TRUESIZE(size);
240 skb->pfmemalloc = pfmemalloc;
241 atomic_set(&skb->users, 1);
244 skb_reset_tail_pointer(skb);
245 skb->end = skb->tail + size;
246 skb->mac_header = (typeof(skb->mac_header))~0U;
247 skb->transport_header = (typeof(skb->transport_header))~0U;
249 /* make sure we initialize shinfo sequentially */
250 shinfo = skb_shinfo(skb);
251 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
252 atomic_set(&shinfo->dataref, 1);
253 kmemcheck_annotate_variable(shinfo->destructor_arg);
255 if (flags & SKB_ALLOC_FCLONE) {
256 struct sk_buff *child = skb + 1;
257 atomic_t *fclone_ref = (atomic_t *) (child + 1);
259 kmemcheck_annotate_bitfield(child, flags1);
260 kmemcheck_annotate_bitfield(child, flags2);
261 skb->fclone = SKB_FCLONE_ORIG;
262 atomic_set(fclone_ref, 1);
264 child->fclone = SKB_FCLONE_UNAVAILABLE;
265 child->pfmemalloc = pfmemalloc;
270 kmem_cache_free(cache, skb);
274 EXPORT_SYMBOL(__alloc_skb);
277 * build_skb - build a network buffer
278 * @data: data buffer provided by caller
279 * @frag_size: size of fragment, or 0 if head was kmalloced
281 * Allocate a new &sk_buff. Caller provides space holding head and
282 * skb_shared_info. @data must have been allocated by kmalloc() only if
283 * @frag_size is 0, otherwise data should come from the page allocator.
284 * The return is the new skb buffer.
285 * On a failure the return is %NULL, and @data is not freed.
287 * Before IO, driver allocates only data buffer where NIC put incoming frame
288 * Driver should add room at head (NET_SKB_PAD) and
289 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
290 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
291 * before giving packet to stack.
292 * RX rings only contains data buffers, not full skbs.
294 struct sk_buff *build_skb(void *data, unsigned int frag_size)
296 struct skb_shared_info *shinfo;
298 unsigned int size = frag_size ? : ksize(data);
300 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
304 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
306 memset(skb, 0, offsetof(struct sk_buff, tail));
307 skb->truesize = SKB_TRUESIZE(size);
308 skb->head_frag = frag_size != 0;
309 atomic_set(&skb->users, 1);
312 skb_reset_tail_pointer(skb);
313 skb->end = skb->tail + size;
314 skb->mac_header = (typeof(skb->mac_header))~0U;
315 skb->transport_header = (typeof(skb->transport_header))~0U;
317 /* make sure we initialize shinfo sequentially */
318 shinfo = skb_shinfo(skb);
319 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
320 atomic_set(&shinfo->dataref, 1);
321 kmemcheck_annotate_variable(shinfo->destructor_arg);
325 EXPORT_SYMBOL(build_skb);
327 struct netdev_alloc_cache {
328 struct page_frag frag;
329 /* we maintain a pagecount bias, so that we dont dirty cache line
330 * containing page->_count every time we allocate a fragment.
332 unsigned int pagecnt_bias;
334 static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
336 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
338 struct netdev_alloc_cache *nc;
343 local_irq_save(flags);
344 nc = &__get_cpu_var(netdev_alloc_cache);
345 if (unlikely(!nc->frag.page)) {
347 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
348 gfp_t gfp = gfp_mask;
351 gfp |= __GFP_COMP | __GFP_NOWARN;
352 nc->frag.page = alloc_pages(gfp, order);
353 if (likely(nc->frag.page))
358 nc->frag.size = PAGE_SIZE << order;
360 atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS);
361 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
365 if (nc->frag.offset + fragsz > nc->frag.size) {
366 /* avoid unnecessary locked operations if possible */
367 if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) ||
368 atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count))
373 data = page_address(nc->frag.page) + nc->frag.offset;
374 nc->frag.offset += fragsz;
377 local_irq_restore(flags);
382 * netdev_alloc_frag - allocate a page fragment
383 * @fragsz: fragment size
385 * Allocates a frag from a page for receive buffer.
386 * Uses GFP_ATOMIC allocations.
388 void *netdev_alloc_frag(unsigned int fragsz)
390 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
392 EXPORT_SYMBOL(netdev_alloc_frag);
395 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
396 * @dev: network device to receive on
397 * @length: length to allocate
398 * @gfp_mask: get_free_pages mask, passed to alloc_skb
400 * Allocate a new &sk_buff and assign it a usage count of one. The
401 * buffer has unspecified headroom built in. Users should allocate
402 * the headroom they think they need without accounting for the
403 * built in space. The built in space is used for optimisations.
405 * %NULL is returned if there is no free memory.
407 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
408 unsigned int length, gfp_t gfp_mask)
410 struct sk_buff *skb = NULL;
411 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
412 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
414 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
417 if (sk_memalloc_socks())
418 gfp_mask |= __GFP_MEMALLOC;
420 data = __netdev_alloc_frag(fragsz, gfp_mask);
423 skb = build_skb(data, fragsz);
425 put_page(virt_to_head_page(data));
428 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
429 SKB_ALLOC_RX, NUMA_NO_NODE);
432 skb_reserve(skb, NET_SKB_PAD);
437 EXPORT_SYMBOL(__netdev_alloc_skb);
439 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
440 int size, unsigned int truesize)
442 skb_fill_page_desc(skb, i, page, off, size);
444 skb->data_len += size;
445 skb->truesize += truesize;
447 EXPORT_SYMBOL(skb_add_rx_frag);
449 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
450 unsigned int truesize)
452 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
454 skb_frag_size_add(frag, size);
456 skb->data_len += size;
457 skb->truesize += truesize;
459 EXPORT_SYMBOL(skb_coalesce_rx_frag);
461 static void skb_drop_list(struct sk_buff **listp)
463 kfree_skb_list(*listp);
467 static inline void skb_drop_fraglist(struct sk_buff *skb)
469 skb_drop_list(&skb_shinfo(skb)->frag_list);
472 static void skb_clone_fraglist(struct sk_buff *skb)
474 struct sk_buff *list;
476 skb_walk_frags(skb, list)
480 static void skb_free_head(struct sk_buff *skb)
483 put_page(virt_to_head_page(skb->head));
488 static void skb_release_data(struct sk_buff *skb)
491 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
492 &skb_shinfo(skb)->dataref)) {
493 if (skb_shinfo(skb)->nr_frags) {
495 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
496 skb_frag_unref(skb, i);
500 * If skb buf is from userspace, we need to notify the caller
501 * the lower device DMA has done;
503 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
504 struct ubuf_info *uarg;
506 uarg = skb_shinfo(skb)->destructor_arg;
508 uarg->callback(uarg, true);
511 if (skb_has_frag_list(skb))
512 skb_drop_fraglist(skb);
519 * Free an skbuff by memory without cleaning the state.
521 static void kfree_skbmem(struct sk_buff *skb)
523 struct sk_buff *other;
524 atomic_t *fclone_ref;
526 switch (skb->fclone) {
527 case SKB_FCLONE_UNAVAILABLE:
528 kmem_cache_free(skbuff_head_cache, skb);
531 case SKB_FCLONE_ORIG:
532 fclone_ref = (atomic_t *) (skb + 2);
533 if (atomic_dec_and_test(fclone_ref))
534 kmem_cache_free(skbuff_fclone_cache, skb);
537 case SKB_FCLONE_CLONE:
538 fclone_ref = (atomic_t *) (skb + 1);
541 /* The clone portion is available for
542 * fast-cloning again.
544 skb->fclone = SKB_FCLONE_UNAVAILABLE;
546 if (atomic_dec_and_test(fclone_ref))
547 kmem_cache_free(skbuff_fclone_cache, other);
552 static void skb_release_head_state(struct sk_buff *skb)
556 secpath_put(skb->sp);
558 if (skb->destructor) {
560 skb->destructor(skb);
562 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
563 nf_conntrack_put(skb->nfct);
565 #ifdef CONFIG_BRIDGE_NETFILTER
566 nf_bridge_put(skb->nf_bridge);
568 /* XXX: IS this still necessary? - JHS */
569 #ifdef CONFIG_NET_SCHED
571 #ifdef CONFIG_NET_CLS_ACT
577 /* Free everything but the sk_buff shell. */
578 static void skb_release_all(struct sk_buff *skb)
580 skb_release_head_state(skb);
581 if (likely(skb->head))
582 skb_release_data(skb);
586 * __kfree_skb - private function
589 * Free an sk_buff. Release anything attached to the buffer.
590 * Clean the state. This is an internal helper function. Users should
591 * always call kfree_skb
594 void __kfree_skb(struct sk_buff *skb)
596 skb_release_all(skb);
599 EXPORT_SYMBOL(__kfree_skb);
602 * kfree_skb - free an sk_buff
603 * @skb: buffer to free
605 * Drop a reference to the buffer and free it if the usage count has
608 void kfree_skb(struct sk_buff *skb)
612 if (likely(atomic_read(&skb->users) == 1))
614 else if (likely(!atomic_dec_and_test(&skb->users)))
616 trace_kfree_skb(skb, __builtin_return_address(0));
619 EXPORT_SYMBOL(kfree_skb);
621 void kfree_skb_list(struct sk_buff *segs)
624 struct sk_buff *next = segs->next;
630 EXPORT_SYMBOL(kfree_skb_list);
633 * skb_tx_error - report an sk_buff xmit error
634 * @skb: buffer that triggered an error
636 * Report xmit error if a device callback is tracking this skb.
637 * skb must be freed afterwards.
639 void skb_tx_error(struct sk_buff *skb)
641 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
642 struct ubuf_info *uarg;
644 uarg = skb_shinfo(skb)->destructor_arg;
646 uarg->callback(uarg, false);
647 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
650 EXPORT_SYMBOL(skb_tx_error);
653 * consume_skb - free an skbuff
654 * @skb: buffer to free
656 * Drop a ref to the buffer and free it if the usage count has hit zero
657 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
658 * is being dropped after a failure and notes that
660 void consume_skb(struct sk_buff *skb)
664 if (likely(atomic_read(&skb->users) == 1))
666 else if (likely(!atomic_dec_and_test(&skb->users)))
668 trace_consume_skb(skb);
671 EXPORT_SYMBOL(consume_skb);
673 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
675 new->tstamp = old->tstamp;
677 new->transport_header = old->transport_header;
678 new->network_header = old->network_header;
679 new->mac_header = old->mac_header;
680 new->inner_protocol = old->inner_protocol;
681 new->inner_transport_header = old->inner_transport_header;
682 new->inner_network_header = old->inner_network_header;
683 new->inner_mac_header = old->inner_mac_header;
684 skb_dst_copy(new, old);
685 new->rxhash = old->rxhash;
686 new->ooo_okay = old->ooo_okay;
687 new->l4_rxhash = old->l4_rxhash;
688 new->no_fcs = old->no_fcs;
689 new->encapsulation = old->encapsulation;
691 new->sp = secpath_get(old->sp);
693 memcpy(new->cb, old->cb, sizeof(old->cb));
694 new->csum = old->csum;
695 new->local_df = old->local_df;
696 new->pkt_type = old->pkt_type;
697 new->ip_summed = old->ip_summed;
698 skb_copy_queue_mapping(new, old);
699 new->priority = old->priority;
700 #if IS_ENABLED(CONFIG_IP_VS)
701 new->ipvs_property = old->ipvs_property;
703 new->pfmemalloc = old->pfmemalloc;
704 new->protocol = old->protocol;
705 new->mark = old->mark;
706 new->skb_iif = old->skb_iif;
708 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
709 new->nf_trace = old->nf_trace;
711 #ifdef CONFIG_NET_SCHED
712 new->tc_index = old->tc_index;
713 #ifdef CONFIG_NET_CLS_ACT
714 new->tc_verd = old->tc_verd;
717 new->vlan_proto = old->vlan_proto;
718 new->vlan_tci = old->vlan_tci;
720 skb_copy_secmark(new, old);
722 #ifdef CONFIG_NET_RX_BUSY_POLL
723 new->napi_id = old->napi_id;
728 * You should not add any new code to this function. Add it to
729 * __copy_skb_header above instead.
731 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
733 #define C(x) n->x = skb->x
735 n->next = n->prev = NULL;
737 __copy_skb_header(n, skb);
742 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
745 n->destructor = NULL;
752 atomic_set(&n->users, 1);
754 atomic_inc(&(skb_shinfo(skb)->dataref));
762 * skb_morph - morph one skb into another
763 * @dst: the skb to receive the contents
764 * @src: the skb to supply the contents
766 * This is identical to skb_clone except that the target skb is
767 * supplied by the user.
769 * The target skb is returned upon exit.
771 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
773 skb_release_all(dst);
774 return __skb_clone(dst, src);
776 EXPORT_SYMBOL_GPL(skb_morph);
779 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
780 * @skb: the skb to modify
781 * @gfp_mask: allocation priority
783 * This must be called on SKBTX_DEV_ZEROCOPY skb.
784 * It will copy all frags into kernel and drop the reference
785 * to userspace pages.
787 * If this function is called from an interrupt gfp_mask() must be
790 * Returns 0 on success or a negative error code on failure
791 * to allocate kernel memory to copy to.
793 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
796 int num_frags = skb_shinfo(skb)->nr_frags;
797 struct page *page, *head = NULL;
798 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
800 for (i = 0; i < num_frags; i++) {
802 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
804 page = alloc_page(gfp_mask);
807 struct page *next = (struct page *)page_private(head);
813 vaddr = kmap_atomic(skb_frag_page(f));
814 memcpy(page_address(page),
815 vaddr + f->page_offset, skb_frag_size(f));
816 kunmap_atomic(vaddr);
817 set_page_private(page, (unsigned long)head);
821 /* skb frags release userspace buffers */
822 for (i = 0; i < num_frags; i++)
823 skb_frag_unref(skb, i);
825 uarg->callback(uarg, false);
827 /* skb frags point to kernel buffers */
828 for (i = num_frags - 1; i >= 0; i--) {
829 __skb_fill_page_desc(skb, i, head, 0,
830 skb_shinfo(skb)->frags[i].size);
831 head = (struct page *)page_private(head);
834 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
837 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
840 * skb_clone - duplicate an sk_buff
841 * @skb: buffer to clone
842 * @gfp_mask: allocation priority
844 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
845 * copies share the same packet data but not structure. The new
846 * buffer has a reference count of 1. If the allocation fails the
847 * function returns %NULL otherwise the new buffer is returned.
849 * If this function is called from an interrupt gfp_mask() must be
853 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
857 if (skb_orphan_frags(skb, gfp_mask))
861 if (skb->fclone == SKB_FCLONE_ORIG &&
862 n->fclone == SKB_FCLONE_UNAVAILABLE) {
863 atomic_t *fclone_ref = (atomic_t *) (n + 1);
864 n->fclone = SKB_FCLONE_CLONE;
865 atomic_inc(fclone_ref);
867 if (skb_pfmemalloc(skb))
868 gfp_mask |= __GFP_MEMALLOC;
870 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
874 kmemcheck_annotate_bitfield(n, flags1);
875 kmemcheck_annotate_bitfield(n, flags2);
876 n->fclone = SKB_FCLONE_UNAVAILABLE;
879 return __skb_clone(n, skb);
881 EXPORT_SYMBOL(skb_clone);
883 static void skb_headers_offset_update(struct sk_buff *skb, int off)
885 /* Only adjust this if it actually is csum_start rather than csum */
886 if (skb->ip_summed == CHECKSUM_PARTIAL)
887 skb->csum_start += off;
888 /* {transport,network,mac}_header and tail are relative to skb->head */
889 skb->transport_header += off;
890 skb->network_header += off;
891 if (skb_mac_header_was_set(skb))
892 skb->mac_header += off;
893 skb->inner_transport_header += off;
894 skb->inner_network_header += off;
895 skb->inner_mac_header += off;
898 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
900 __copy_skb_header(new, old);
902 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
903 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
904 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
907 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
909 if (skb_pfmemalloc(skb))
915 * skb_copy - create private copy of an sk_buff
916 * @skb: buffer to copy
917 * @gfp_mask: allocation priority
919 * Make a copy of both an &sk_buff and its data. This is used when the
920 * caller wishes to modify the data and needs a private copy of the
921 * data to alter. Returns %NULL on failure or the pointer to the buffer
922 * on success. The returned buffer has a reference count of 1.
924 * As by-product this function converts non-linear &sk_buff to linear
925 * one, so that &sk_buff becomes completely private and caller is allowed
926 * to modify all the data of returned buffer. This means that this
927 * function is not recommended for use in circumstances when only
928 * header is going to be modified. Use pskb_copy() instead.
931 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
933 int headerlen = skb_headroom(skb);
934 unsigned int size = skb_end_offset(skb) + skb->data_len;
935 struct sk_buff *n = __alloc_skb(size, gfp_mask,
936 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
941 /* Set the data pointer */
942 skb_reserve(n, headerlen);
943 /* Set the tail pointer and length */
944 skb_put(n, skb->len);
946 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
949 copy_skb_header(n, skb);
952 EXPORT_SYMBOL(skb_copy);
955 * __pskb_copy - create copy of an sk_buff with private head.
956 * @skb: buffer to copy
957 * @headroom: headroom of new skb
958 * @gfp_mask: allocation priority
960 * Make a copy of both an &sk_buff and part of its data, located
961 * in header. Fragmented data remain shared. This is used when
962 * the caller wishes to modify only header of &sk_buff and needs
963 * private copy of the header to alter. Returns %NULL on failure
964 * or the pointer to the buffer on success.
965 * The returned buffer has a reference count of 1.
968 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
970 unsigned int size = skb_headlen(skb) + headroom;
971 struct sk_buff *n = __alloc_skb(size, gfp_mask,
972 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
977 /* Set the data pointer */
978 skb_reserve(n, headroom);
979 /* Set the tail pointer and length */
980 skb_put(n, skb_headlen(skb));
982 skb_copy_from_linear_data(skb, n->data, n->len);
984 n->truesize += skb->data_len;
985 n->data_len = skb->data_len;
988 if (skb_shinfo(skb)->nr_frags) {
991 if (skb_orphan_frags(skb, gfp_mask)) {
996 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
997 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
998 skb_frag_ref(skb, i);
1000 skb_shinfo(n)->nr_frags = i;
1003 if (skb_has_frag_list(skb)) {
1004 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1005 skb_clone_fraglist(n);
1008 copy_skb_header(n, skb);
1012 EXPORT_SYMBOL(__pskb_copy);
1015 * pskb_expand_head - reallocate header of &sk_buff
1016 * @skb: buffer to reallocate
1017 * @nhead: room to add at head
1018 * @ntail: room to add at tail
1019 * @gfp_mask: allocation priority
1021 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1022 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1023 * reference count of 1. Returns zero in the case of success or error,
1024 * if expansion failed. In the last case, &sk_buff is not changed.
1026 * All the pointers pointing into skb header may change and must be
1027 * reloaded after call to this function.
1030 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1035 int size = nhead + skb_end_offset(skb) + ntail;
1040 if (skb_shared(skb))
1043 size = SKB_DATA_ALIGN(size);
1045 if (skb_pfmemalloc(skb))
1046 gfp_mask |= __GFP_MEMALLOC;
1047 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1048 gfp_mask, NUMA_NO_NODE, NULL);
1051 size = SKB_WITH_OVERHEAD(ksize(data));
1053 /* Copy only real data... and, alas, header. This should be
1054 * optimized for the cases when header is void.
1056 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1058 memcpy((struct skb_shared_info *)(data + size),
1060 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1063 * if shinfo is shared we must drop the old head gracefully, but if it
1064 * is not we can just drop the old head and let the existing refcount
1065 * be since all we did is relocate the values
1067 if (skb_cloned(skb)) {
1068 /* copy this zero copy skb frags */
1069 if (skb_orphan_frags(skb, gfp_mask))
1071 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1072 skb_frag_ref(skb, i);
1074 if (skb_has_frag_list(skb))
1075 skb_clone_fraglist(skb);
1077 skb_release_data(skb);
1081 off = (data + nhead) - skb->head;
1086 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1090 skb->end = skb->head + size;
1093 skb_headers_offset_update(skb, nhead);
1097 atomic_set(&skb_shinfo(skb)->dataref, 1);
1105 EXPORT_SYMBOL(pskb_expand_head);
1107 /* Make private copy of skb with writable head and some headroom */
1109 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1111 struct sk_buff *skb2;
1112 int delta = headroom - skb_headroom(skb);
1115 skb2 = pskb_copy(skb, GFP_ATOMIC);
1117 skb2 = skb_clone(skb, GFP_ATOMIC);
1118 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1126 EXPORT_SYMBOL(skb_realloc_headroom);
1129 * skb_copy_expand - copy and expand sk_buff
1130 * @skb: buffer to copy
1131 * @newheadroom: new free bytes at head
1132 * @newtailroom: new free bytes at tail
1133 * @gfp_mask: allocation priority
1135 * Make a copy of both an &sk_buff and its data and while doing so
1136 * allocate additional space.
1138 * This is used when the caller wishes to modify the data and needs a
1139 * private copy of the data to alter as well as more space for new fields.
1140 * Returns %NULL on failure or the pointer to the buffer
1141 * on success. The returned buffer has a reference count of 1.
1143 * You must pass %GFP_ATOMIC as the allocation priority if this function
1144 * is called from an interrupt.
1146 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1147 int newheadroom, int newtailroom,
1151 * Allocate the copy buffer
1153 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1154 gfp_mask, skb_alloc_rx_flag(skb),
1156 int oldheadroom = skb_headroom(skb);
1157 int head_copy_len, head_copy_off;
1162 skb_reserve(n, newheadroom);
1164 /* Set the tail pointer and length */
1165 skb_put(n, skb->len);
1167 head_copy_len = oldheadroom;
1169 if (newheadroom <= head_copy_len)
1170 head_copy_len = newheadroom;
1172 head_copy_off = newheadroom - head_copy_len;
1174 /* Copy the linear header and data. */
1175 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1176 skb->len + head_copy_len))
1179 copy_skb_header(n, skb);
1181 skb_headers_offset_update(n, newheadroom - oldheadroom);
1185 EXPORT_SYMBOL(skb_copy_expand);
1188 * skb_pad - zero pad the tail of an skb
1189 * @skb: buffer to pad
1190 * @pad: space to pad
1192 * Ensure that a buffer is followed by a padding area that is zero
1193 * filled. Used by network drivers which may DMA or transfer data
1194 * beyond the buffer end onto the wire.
1196 * May return error in out of memory cases. The skb is freed on error.
1199 int skb_pad(struct sk_buff *skb, int pad)
1204 /* If the skbuff is non linear tailroom is always zero.. */
1205 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1206 memset(skb->data+skb->len, 0, pad);
1210 ntail = skb->data_len + pad - (skb->end - skb->tail);
1211 if (likely(skb_cloned(skb) || ntail > 0)) {
1212 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1217 /* FIXME: The use of this function with non-linear skb's really needs
1220 err = skb_linearize(skb);
1224 memset(skb->data + skb->len, 0, pad);
1231 EXPORT_SYMBOL(skb_pad);
1234 * pskb_put - add data to the tail of a potentially fragmented buffer
1235 * @skb: start of the buffer to use
1236 * @tail: tail fragment of the buffer to use
1237 * @len: amount of data to add
1239 * This function extends the used data area of the potentially
1240 * fragmented buffer. @tail must be the last fragment of @skb -- or
1241 * @skb itself. If this would exceed the total buffer size the kernel
1242 * will panic. A pointer to the first byte of the extra data is
1246 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1249 skb->data_len += len;
1252 return skb_put(tail, len);
1254 EXPORT_SYMBOL_GPL(pskb_put);
1257 * skb_put - add data to a buffer
1258 * @skb: buffer to use
1259 * @len: amount of data to add
1261 * This function extends the used data area of the buffer. If this would
1262 * exceed the total buffer size the kernel will panic. A pointer to the
1263 * first byte of the extra data is returned.
1265 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1267 unsigned char *tmp = skb_tail_pointer(skb);
1268 SKB_LINEAR_ASSERT(skb);
1271 if (unlikely(skb->tail > skb->end))
1272 skb_over_panic(skb, len, __builtin_return_address(0));
1275 EXPORT_SYMBOL(skb_put);
1278 * skb_push - add data to the start of a buffer
1279 * @skb: buffer to use
1280 * @len: amount of data to add
1282 * This function extends the used data area of the buffer at the buffer
1283 * start. If this would exceed the total buffer headroom the kernel will
1284 * panic. A pointer to the first byte of the extra data is returned.
1286 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1290 if (unlikely(skb->data<skb->head))
1291 skb_under_panic(skb, len, __builtin_return_address(0));
1294 EXPORT_SYMBOL(skb_push);
1297 * skb_pull - remove data from the start of a buffer
1298 * @skb: buffer to use
1299 * @len: amount of data to remove
1301 * This function removes data from the start of a buffer, returning
1302 * the memory to the headroom. A pointer to the next data in the buffer
1303 * is returned. Once the data has been pulled future pushes will overwrite
1306 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1308 return skb_pull_inline(skb, len);
1310 EXPORT_SYMBOL(skb_pull);
1313 * skb_trim - remove end from a buffer
1314 * @skb: buffer to alter
1317 * Cut the length of a buffer down by removing data from the tail. If
1318 * the buffer is already under the length specified it is not modified.
1319 * The skb must be linear.
1321 void skb_trim(struct sk_buff *skb, unsigned int len)
1324 __skb_trim(skb, len);
1326 EXPORT_SYMBOL(skb_trim);
1328 /* Trims skb to length len. It can change skb pointers.
1331 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1333 struct sk_buff **fragp;
1334 struct sk_buff *frag;
1335 int offset = skb_headlen(skb);
1336 int nfrags = skb_shinfo(skb)->nr_frags;
1340 if (skb_cloned(skb) &&
1341 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1348 for (; i < nfrags; i++) {
1349 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1356 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1359 skb_shinfo(skb)->nr_frags = i;
1361 for (; i < nfrags; i++)
1362 skb_frag_unref(skb, i);
1364 if (skb_has_frag_list(skb))
1365 skb_drop_fraglist(skb);
1369 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1370 fragp = &frag->next) {
1371 int end = offset + frag->len;
1373 if (skb_shared(frag)) {
1374 struct sk_buff *nfrag;
1376 nfrag = skb_clone(frag, GFP_ATOMIC);
1377 if (unlikely(!nfrag))
1380 nfrag->next = frag->next;
1392 unlikely((err = pskb_trim(frag, len - offset))))
1396 skb_drop_list(&frag->next);
1401 if (len > skb_headlen(skb)) {
1402 skb->data_len -= skb->len - len;
1407 skb_set_tail_pointer(skb, len);
1412 EXPORT_SYMBOL(___pskb_trim);
1415 * __pskb_pull_tail - advance tail of skb header
1416 * @skb: buffer to reallocate
1417 * @delta: number of bytes to advance tail
1419 * The function makes a sense only on a fragmented &sk_buff,
1420 * it expands header moving its tail forward and copying necessary
1421 * data from fragmented part.
1423 * &sk_buff MUST have reference count of 1.
1425 * Returns %NULL (and &sk_buff does not change) if pull failed
1426 * or value of new tail of skb in the case of success.
1428 * All the pointers pointing into skb header may change and must be
1429 * reloaded after call to this function.
1432 /* Moves tail of skb head forward, copying data from fragmented part,
1433 * when it is necessary.
1434 * 1. It may fail due to malloc failure.
1435 * 2. It may change skb pointers.
1437 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1439 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1441 /* If skb has not enough free space at tail, get new one
1442 * plus 128 bytes for future expansions. If we have enough
1443 * room at tail, reallocate without expansion only if skb is cloned.
1445 int i, k, eat = (skb->tail + delta) - skb->end;
1447 if (eat > 0 || skb_cloned(skb)) {
1448 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1453 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1456 /* Optimization: no fragments, no reasons to preestimate
1457 * size of pulled pages. Superb.
1459 if (!skb_has_frag_list(skb))
1462 /* Estimate size of pulled pages. */
1464 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1465 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1472 /* If we need update frag list, we are in troubles.
1473 * Certainly, it possible to add an offset to skb data,
1474 * but taking into account that pulling is expected to
1475 * be very rare operation, it is worth to fight against
1476 * further bloating skb head and crucify ourselves here instead.
1477 * Pure masohism, indeed. 8)8)
1480 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1481 struct sk_buff *clone = NULL;
1482 struct sk_buff *insp = NULL;
1487 if (list->len <= eat) {
1488 /* Eaten as whole. */
1493 /* Eaten partially. */
1495 if (skb_shared(list)) {
1496 /* Sucks! We need to fork list. :-( */
1497 clone = skb_clone(list, GFP_ATOMIC);
1503 /* This may be pulled without
1507 if (!pskb_pull(list, eat)) {
1515 /* Free pulled out fragments. */
1516 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1517 skb_shinfo(skb)->frag_list = list->next;
1520 /* And insert new clone at head. */
1523 skb_shinfo(skb)->frag_list = clone;
1526 /* Success! Now we may commit changes to skb data. */
1531 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1532 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1535 skb_frag_unref(skb, i);
1538 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1540 skb_shinfo(skb)->frags[k].page_offset += eat;
1541 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1547 skb_shinfo(skb)->nr_frags = k;
1550 skb->data_len -= delta;
1552 return skb_tail_pointer(skb);
1554 EXPORT_SYMBOL(__pskb_pull_tail);
1557 * skb_copy_bits - copy bits from skb to kernel buffer
1559 * @offset: offset in source
1560 * @to: destination buffer
1561 * @len: number of bytes to copy
1563 * Copy the specified number of bytes from the source skb to the
1564 * destination buffer.
1567 * If its prototype is ever changed,
1568 * check arch/{*}/net/{*}.S files,
1569 * since it is called from BPF assembly code.
1571 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1573 int start = skb_headlen(skb);
1574 struct sk_buff *frag_iter;
1577 if (offset > (int)skb->len - len)
1581 if ((copy = start - offset) > 0) {
1584 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1585 if ((len -= copy) == 0)
1591 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1593 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1595 WARN_ON(start > offset + len);
1597 end = start + skb_frag_size(f);
1598 if ((copy = end - offset) > 0) {
1604 vaddr = kmap_atomic(skb_frag_page(f));
1606 vaddr + f->page_offset + offset - start,
1608 kunmap_atomic(vaddr);
1610 if ((len -= copy) == 0)
1618 skb_walk_frags(skb, frag_iter) {
1621 WARN_ON(start > offset + len);
1623 end = start + frag_iter->len;
1624 if ((copy = end - offset) > 0) {
1627 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1629 if ((len -= copy) == 0)
1643 EXPORT_SYMBOL(skb_copy_bits);
1646 * Callback from splice_to_pipe(), if we need to release some pages
1647 * at the end of the spd in case we error'ed out in filling the pipe.
1649 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1651 put_page(spd->pages[i]);
1654 static struct page *linear_to_page(struct page *page, unsigned int *len,
1655 unsigned int *offset,
1658 struct page_frag *pfrag = sk_page_frag(sk);
1660 if (!sk_page_frag_refill(sk, pfrag))
1663 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1665 memcpy(page_address(pfrag->page) + pfrag->offset,
1666 page_address(page) + *offset, *len);
1667 *offset = pfrag->offset;
1668 pfrag->offset += *len;
1673 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1675 unsigned int offset)
1677 return spd->nr_pages &&
1678 spd->pages[spd->nr_pages - 1] == page &&
1679 (spd->partial[spd->nr_pages - 1].offset +
1680 spd->partial[spd->nr_pages - 1].len == offset);
1684 * Fill page/offset/length into spd, if it can hold more pages.
1686 static bool spd_fill_page(struct splice_pipe_desc *spd,
1687 struct pipe_inode_info *pipe, struct page *page,
1688 unsigned int *len, unsigned int offset,
1692 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1696 page = linear_to_page(page, len, &offset, sk);
1700 if (spd_can_coalesce(spd, page, offset)) {
1701 spd->partial[spd->nr_pages - 1].len += *len;
1705 spd->pages[spd->nr_pages] = page;
1706 spd->partial[spd->nr_pages].len = *len;
1707 spd->partial[spd->nr_pages].offset = offset;
1713 static bool __splice_segment(struct page *page, unsigned int poff,
1714 unsigned int plen, unsigned int *off,
1716 struct splice_pipe_desc *spd, bool linear,
1718 struct pipe_inode_info *pipe)
1723 /* skip this segment if already processed */
1729 /* ignore any bits we already processed */
1735 unsigned int flen = min(*len, plen);
1737 if (spd_fill_page(spd, pipe, page, &flen, poff,
1743 } while (*len && plen);
1749 * Map linear and fragment data from the skb to spd. It reports true if the
1750 * pipe is full or if we already spliced the requested length.
1752 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1753 unsigned int *offset, unsigned int *len,
1754 struct splice_pipe_desc *spd, struct sock *sk)
1758 /* map the linear part :
1759 * If skb->head_frag is set, this 'linear' part is backed by a
1760 * fragment, and if the head is not shared with any clones then
1761 * we can avoid a copy since we own the head portion of this page.
1763 if (__splice_segment(virt_to_page(skb->data),
1764 (unsigned long) skb->data & (PAGE_SIZE - 1),
1767 skb_head_is_locked(skb),
1772 * then map the fragments
1774 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1775 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1777 if (__splice_segment(skb_frag_page(f),
1778 f->page_offset, skb_frag_size(f),
1779 offset, len, spd, false, sk, pipe))
1787 * Map data from the skb to a pipe. Should handle both the linear part,
1788 * the fragments, and the frag list. It does NOT handle frag lists within
1789 * the frag list, if such a thing exists. We'd probably need to recurse to
1790 * handle that cleanly.
1792 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1793 struct pipe_inode_info *pipe, unsigned int tlen,
1796 struct partial_page partial[MAX_SKB_FRAGS];
1797 struct page *pages[MAX_SKB_FRAGS];
1798 struct splice_pipe_desc spd = {
1801 .nr_pages_max = MAX_SKB_FRAGS,
1803 .ops = &nosteal_pipe_buf_ops,
1804 .spd_release = sock_spd_release,
1806 struct sk_buff *frag_iter;
1807 struct sock *sk = skb->sk;
1811 * __skb_splice_bits() only fails if the output has no room left,
1812 * so no point in going over the frag_list for the error case.
1814 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1820 * now see if we have a frag_list to map
1822 skb_walk_frags(skb, frag_iter) {
1825 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1832 * Drop the socket lock, otherwise we have reverse
1833 * locking dependencies between sk_lock and i_mutex
1834 * here as compared to sendfile(). We enter here
1835 * with the socket lock held, and splice_to_pipe() will
1836 * grab the pipe inode lock. For sendfile() emulation,
1837 * we call into ->sendpage() with the i_mutex lock held
1838 * and networking will grab the socket lock.
1841 ret = splice_to_pipe(pipe, &spd);
1849 * skb_store_bits - store bits from kernel buffer to skb
1850 * @skb: destination buffer
1851 * @offset: offset in destination
1852 * @from: source buffer
1853 * @len: number of bytes to copy
1855 * Copy the specified number of bytes from the source buffer to the
1856 * destination skb. This function handles all the messy bits of
1857 * traversing fragment lists and such.
1860 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1862 int start = skb_headlen(skb);
1863 struct sk_buff *frag_iter;
1866 if (offset > (int)skb->len - len)
1869 if ((copy = start - offset) > 0) {
1872 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1873 if ((len -= copy) == 0)
1879 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1880 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1883 WARN_ON(start > offset + len);
1885 end = start + skb_frag_size(frag);
1886 if ((copy = end - offset) > 0) {
1892 vaddr = kmap_atomic(skb_frag_page(frag));
1893 memcpy(vaddr + frag->page_offset + offset - start,
1895 kunmap_atomic(vaddr);
1897 if ((len -= copy) == 0)
1905 skb_walk_frags(skb, frag_iter) {
1908 WARN_ON(start > offset + len);
1910 end = start + frag_iter->len;
1911 if ((copy = end - offset) > 0) {
1914 if (skb_store_bits(frag_iter, offset - start,
1917 if ((len -= copy) == 0)
1930 EXPORT_SYMBOL(skb_store_bits);
1932 /* Checksum skb data. */
1933 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
1934 __wsum csum, const struct skb_checksum_ops *ops)
1936 int start = skb_headlen(skb);
1937 int i, copy = start - offset;
1938 struct sk_buff *frag_iter;
1941 /* Checksum header. */
1945 csum = ops->update(skb->data + offset, copy, csum);
1946 if ((len -= copy) == 0)
1952 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1954 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1956 WARN_ON(start > offset + len);
1958 end = start + skb_frag_size(frag);
1959 if ((copy = end - offset) > 0) {
1965 vaddr = kmap_atomic(skb_frag_page(frag));
1966 csum2 = ops->update(vaddr + frag->page_offset +
1967 offset - start, copy, 0);
1968 kunmap_atomic(vaddr);
1969 csum = ops->combine(csum, csum2, pos, copy);
1978 skb_walk_frags(skb, frag_iter) {
1981 WARN_ON(start > offset + len);
1983 end = start + frag_iter->len;
1984 if ((copy = end - offset) > 0) {
1988 csum2 = __skb_checksum(frag_iter, offset - start,
1990 csum = ops->combine(csum, csum2, pos, copy);
1991 if ((len -= copy) == 0)
2002 EXPORT_SYMBOL(__skb_checksum);
2004 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2005 int len, __wsum csum)
2007 const struct skb_checksum_ops ops = {
2008 .update = csum_partial_ext,
2009 .combine = csum_block_add_ext,
2012 return __skb_checksum(skb, offset, len, csum, &ops);
2014 EXPORT_SYMBOL(skb_checksum);
2016 /* Both of above in one bottle. */
2018 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2019 u8 *to, int len, __wsum csum)
2021 int start = skb_headlen(skb);
2022 int i, copy = start - offset;
2023 struct sk_buff *frag_iter;
2030 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2032 if ((len -= copy) == 0)
2039 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2042 WARN_ON(start > offset + len);
2044 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2045 if ((copy = end - offset) > 0) {
2048 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2052 vaddr = kmap_atomic(skb_frag_page(frag));
2053 csum2 = csum_partial_copy_nocheck(vaddr +
2057 kunmap_atomic(vaddr);
2058 csum = csum_block_add(csum, csum2, pos);
2068 skb_walk_frags(skb, frag_iter) {
2072 WARN_ON(start > offset + len);
2074 end = start + frag_iter->len;
2075 if ((copy = end - offset) > 0) {
2078 csum2 = skb_copy_and_csum_bits(frag_iter,
2081 csum = csum_block_add(csum, csum2, pos);
2082 if ((len -= copy) == 0)
2093 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2095 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2100 if (skb->ip_summed == CHECKSUM_PARTIAL)
2101 csstart = skb_checksum_start_offset(skb);
2103 csstart = skb_headlen(skb);
2105 BUG_ON(csstart > skb_headlen(skb));
2107 skb_copy_from_linear_data(skb, to, csstart);
2110 if (csstart != skb->len)
2111 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2112 skb->len - csstart, 0);
2114 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2115 long csstuff = csstart + skb->csum_offset;
2117 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2120 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2123 * skb_dequeue - remove from the head of the queue
2124 * @list: list to dequeue from
2126 * Remove the head of the list. The list lock is taken so the function
2127 * may be used safely with other locking list functions. The head item is
2128 * returned or %NULL if the list is empty.
2131 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2133 unsigned long flags;
2134 struct sk_buff *result;
2136 spin_lock_irqsave(&list->lock, flags);
2137 result = __skb_dequeue(list);
2138 spin_unlock_irqrestore(&list->lock, flags);
2141 EXPORT_SYMBOL(skb_dequeue);
2144 * skb_dequeue_tail - remove from the tail of the queue
2145 * @list: list to dequeue from
2147 * Remove the tail of the list. The list lock is taken so the function
2148 * may be used safely with other locking list functions. The tail item is
2149 * returned or %NULL if the list is empty.
2151 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2153 unsigned long flags;
2154 struct sk_buff *result;
2156 spin_lock_irqsave(&list->lock, flags);
2157 result = __skb_dequeue_tail(list);
2158 spin_unlock_irqrestore(&list->lock, flags);
2161 EXPORT_SYMBOL(skb_dequeue_tail);
2164 * skb_queue_purge - empty a list
2165 * @list: list to empty
2167 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2168 * the list and one reference dropped. This function takes the list
2169 * lock and is atomic with respect to other list locking functions.
2171 void skb_queue_purge(struct sk_buff_head *list)
2173 struct sk_buff *skb;
2174 while ((skb = skb_dequeue(list)) != NULL)
2177 EXPORT_SYMBOL(skb_queue_purge);
2180 * skb_queue_head - queue a buffer at the list head
2181 * @list: list to use
2182 * @newsk: buffer to queue
2184 * Queue a buffer at the start of the list. This function takes the
2185 * list lock and can be used safely with other locking &sk_buff functions
2188 * A buffer cannot be placed on two lists at the same time.
2190 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2192 unsigned long flags;
2194 spin_lock_irqsave(&list->lock, flags);
2195 __skb_queue_head(list, newsk);
2196 spin_unlock_irqrestore(&list->lock, flags);
2198 EXPORT_SYMBOL(skb_queue_head);
2201 * skb_queue_tail - queue a buffer at the list tail
2202 * @list: list to use
2203 * @newsk: buffer to queue
2205 * Queue a buffer at the tail of the list. This function takes the
2206 * list lock and can be used safely with other locking &sk_buff functions
2209 * A buffer cannot be placed on two lists at the same time.
2211 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2213 unsigned long flags;
2215 spin_lock_irqsave(&list->lock, flags);
2216 __skb_queue_tail(list, newsk);
2217 spin_unlock_irqrestore(&list->lock, flags);
2219 EXPORT_SYMBOL(skb_queue_tail);
2222 * skb_unlink - remove a buffer from a list
2223 * @skb: buffer to remove
2224 * @list: list to use
2226 * Remove a packet from a list. The list locks are taken and this
2227 * function is atomic with respect to other list locked calls
2229 * You must know what list the SKB is on.
2231 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2233 unsigned long flags;
2235 spin_lock_irqsave(&list->lock, flags);
2236 __skb_unlink(skb, list);
2237 spin_unlock_irqrestore(&list->lock, flags);
2239 EXPORT_SYMBOL(skb_unlink);
2242 * skb_append - append a buffer
2243 * @old: buffer to insert after
2244 * @newsk: buffer to insert
2245 * @list: list to use
2247 * Place a packet after a given packet in a list. The list locks are taken
2248 * and this function is atomic with respect to other list locked calls.
2249 * A buffer cannot be placed on two lists at the same time.
2251 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2253 unsigned long flags;
2255 spin_lock_irqsave(&list->lock, flags);
2256 __skb_queue_after(list, old, newsk);
2257 spin_unlock_irqrestore(&list->lock, flags);
2259 EXPORT_SYMBOL(skb_append);
2262 * skb_insert - insert a buffer
2263 * @old: buffer to insert before
2264 * @newsk: buffer to insert
2265 * @list: list to use
2267 * Place a packet before a given packet in a list. The list locks are
2268 * taken and this function is atomic with respect to other list locked
2271 * A buffer cannot be placed on two lists at the same time.
2273 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2275 unsigned long flags;
2277 spin_lock_irqsave(&list->lock, flags);
2278 __skb_insert(newsk, old->prev, old, list);
2279 spin_unlock_irqrestore(&list->lock, flags);
2281 EXPORT_SYMBOL(skb_insert);
2283 static inline void skb_split_inside_header(struct sk_buff *skb,
2284 struct sk_buff* skb1,
2285 const u32 len, const int pos)
2289 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2291 /* And move data appendix as is. */
2292 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2293 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2295 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2296 skb_shinfo(skb)->nr_frags = 0;
2297 skb1->data_len = skb->data_len;
2298 skb1->len += skb1->data_len;
2301 skb_set_tail_pointer(skb, len);
2304 static inline void skb_split_no_header(struct sk_buff *skb,
2305 struct sk_buff* skb1,
2306 const u32 len, int pos)
2309 const int nfrags = skb_shinfo(skb)->nr_frags;
2311 skb_shinfo(skb)->nr_frags = 0;
2312 skb1->len = skb1->data_len = skb->len - len;
2314 skb->data_len = len - pos;
2316 for (i = 0; i < nfrags; i++) {
2317 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2319 if (pos + size > len) {
2320 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2324 * We have two variants in this case:
2325 * 1. Move all the frag to the second
2326 * part, if it is possible. F.e.
2327 * this approach is mandatory for TUX,
2328 * where splitting is expensive.
2329 * 2. Split is accurately. We make this.
2331 skb_frag_ref(skb, i);
2332 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2333 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2334 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2335 skb_shinfo(skb)->nr_frags++;
2339 skb_shinfo(skb)->nr_frags++;
2342 skb_shinfo(skb1)->nr_frags = k;
2346 * skb_split - Split fragmented skb to two parts at length len.
2347 * @skb: the buffer to split
2348 * @skb1: the buffer to receive the second part
2349 * @len: new length for skb
2351 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2353 int pos = skb_headlen(skb);
2355 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2356 if (len < pos) /* Split line is inside header. */
2357 skb_split_inside_header(skb, skb1, len, pos);
2358 else /* Second chunk has no header, nothing to copy. */
2359 skb_split_no_header(skb, skb1, len, pos);
2361 EXPORT_SYMBOL(skb_split);
2363 /* Shifting from/to a cloned skb is a no-go.
2365 * Caller cannot keep skb_shinfo related pointers past calling here!
2367 static int skb_prepare_for_shift(struct sk_buff *skb)
2369 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2373 * skb_shift - Shifts paged data partially from skb to another
2374 * @tgt: buffer into which tail data gets added
2375 * @skb: buffer from which the paged data comes from
2376 * @shiftlen: shift up to this many bytes
2378 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2379 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2380 * It's up to caller to free skb if everything was shifted.
2382 * If @tgt runs out of frags, the whole operation is aborted.
2384 * Skb cannot include anything else but paged data while tgt is allowed
2385 * to have non-paged data as well.
2387 * TODO: full sized shift could be optimized but that would need
2388 * specialized skb free'er to handle frags without up-to-date nr_frags.
2390 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2392 int from, to, merge, todo;
2393 struct skb_frag_struct *fragfrom, *fragto;
2395 BUG_ON(shiftlen > skb->len);
2396 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2400 to = skb_shinfo(tgt)->nr_frags;
2401 fragfrom = &skb_shinfo(skb)->frags[from];
2403 /* Actual merge is delayed until the point when we know we can
2404 * commit all, so that we don't have to undo partial changes
2407 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2408 fragfrom->page_offset)) {
2413 todo -= skb_frag_size(fragfrom);
2415 if (skb_prepare_for_shift(skb) ||
2416 skb_prepare_for_shift(tgt))
2419 /* All previous frag pointers might be stale! */
2420 fragfrom = &skb_shinfo(skb)->frags[from];
2421 fragto = &skb_shinfo(tgt)->frags[merge];
2423 skb_frag_size_add(fragto, shiftlen);
2424 skb_frag_size_sub(fragfrom, shiftlen);
2425 fragfrom->page_offset += shiftlen;
2433 /* Skip full, not-fitting skb to avoid expensive operations */
2434 if ((shiftlen == skb->len) &&
2435 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2438 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2441 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2442 if (to == MAX_SKB_FRAGS)
2445 fragfrom = &skb_shinfo(skb)->frags[from];
2446 fragto = &skb_shinfo(tgt)->frags[to];
2448 if (todo >= skb_frag_size(fragfrom)) {
2449 *fragto = *fragfrom;
2450 todo -= skb_frag_size(fragfrom);
2455 __skb_frag_ref(fragfrom);
2456 fragto->page = fragfrom->page;
2457 fragto->page_offset = fragfrom->page_offset;
2458 skb_frag_size_set(fragto, todo);
2460 fragfrom->page_offset += todo;
2461 skb_frag_size_sub(fragfrom, todo);
2469 /* Ready to "commit" this state change to tgt */
2470 skb_shinfo(tgt)->nr_frags = to;
2473 fragfrom = &skb_shinfo(skb)->frags[0];
2474 fragto = &skb_shinfo(tgt)->frags[merge];
2476 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2477 __skb_frag_unref(fragfrom);
2480 /* Reposition in the original skb */
2482 while (from < skb_shinfo(skb)->nr_frags)
2483 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2484 skb_shinfo(skb)->nr_frags = to;
2486 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2489 /* Most likely the tgt won't ever need its checksum anymore, skb on
2490 * the other hand might need it if it needs to be resent
2492 tgt->ip_summed = CHECKSUM_PARTIAL;
2493 skb->ip_summed = CHECKSUM_PARTIAL;
2495 /* Yak, is it really working this way? Some helper please? */
2496 skb->len -= shiftlen;
2497 skb->data_len -= shiftlen;
2498 skb->truesize -= shiftlen;
2499 tgt->len += shiftlen;
2500 tgt->data_len += shiftlen;
2501 tgt->truesize += shiftlen;
2507 * skb_prepare_seq_read - Prepare a sequential read of skb data
2508 * @skb: the buffer to read
2509 * @from: lower offset of data to be read
2510 * @to: upper offset of data to be read
2511 * @st: state variable
2513 * Initializes the specified state variable. Must be called before
2514 * invoking skb_seq_read() for the first time.
2516 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2517 unsigned int to, struct skb_seq_state *st)
2519 st->lower_offset = from;
2520 st->upper_offset = to;
2521 st->root_skb = st->cur_skb = skb;
2522 st->frag_idx = st->stepped_offset = 0;
2523 st->frag_data = NULL;
2525 EXPORT_SYMBOL(skb_prepare_seq_read);
2528 * skb_seq_read - Sequentially read skb data
2529 * @consumed: number of bytes consumed by the caller so far
2530 * @data: destination pointer for data to be returned
2531 * @st: state variable
2533 * Reads a block of skb data at @consumed relative to the
2534 * lower offset specified to skb_prepare_seq_read(). Assigns
2535 * the head of the data block to @data and returns the length
2536 * of the block or 0 if the end of the skb data or the upper
2537 * offset has been reached.
2539 * The caller is not required to consume all of the data
2540 * returned, i.e. @consumed is typically set to the number
2541 * of bytes already consumed and the next call to
2542 * skb_seq_read() will return the remaining part of the block.
2544 * Note 1: The size of each block of data returned can be arbitrary,
2545 * this limitation is the cost for zerocopy seqeuental
2546 * reads of potentially non linear data.
2548 * Note 2: Fragment lists within fragments are not implemented
2549 * at the moment, state->root_skb could be replaced with
2550 * a stack for this purpose.
2552 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2553 struct skb_seq_state *st)
2555 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2558 if (unlikely(abs_offset >= st->upper_offset)) {
2559 if (st->frag_data) {
2560 kunmap_atomic(st->frag_data);
2561 st->frag_data = NULL;
2567 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2569 if (abs_offset < block_limit && !st->frag_data) {
2570 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2571 return block_limit - abs_offset;
2574 if (st->frag_idx == 0 && !st->frag_data)
2575 st->stepped_offset += skb_headlen(st->cur_skb);
2577 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2578 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2579 block_limit = skb_frag_size(frag) + st->stepped_offset;
2581 if (abs_offset < block_limit) {
2583 st->frag_data = kmap_atomic(skb_frag_page(frag));
2585 *data = (u8 *) st->frag_data + frag->page_offset +
2586 (abs_offset - st->stepped_offset);
2588 return block_limit - abs_offset;
2591 if (st->frag_data) {
2592 kunmap_atomic(st->frag_data);
2593 st->frag_data = NULL;
2597 st->stepped_offset += skb_frag_size(frag);
2600 if (st->frag_data) {
2601 kunmap_atomic(st->frag_data);
2602 st->frag_data = NULL;
2605 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2606 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2609 } else if (st->cur_skb->next) {
2610 st->cur_skb = st->cur_skb->next;
2617 EXPORT_SYMBOL(skb_seq_read);
2620 * skb_abort_seq_read - Abort a sequential read of skb data
2621 * @st: state variable
2623 * Must be called if skb_seq_read() was not called until it
2626 void skb_abort_seq_read(struct skb_seq_state *st)
2629 kunmap_atomic(st->frag_data);
2631 EXPORT_SYMBOL(skb_abort_seq_read);
2633 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2635 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2636 struct ts_config *conf,
2637 struct ts_state *state)
2639 return skb_seq_read(offset, text, TS_SKB_CB(state));
2642 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2644 skb_abort_seq_read(TS_SKB_CB(state));
2648 * skb_find_text - Find a text pattern in skb data
2649 * @skb: the buffer to look in
2650 * @from: search offset
2652 * @config: textsearch configuration
2653 * @state: uninitialized textsearch state variable
2655 * Finds a pattern in the skb data according to the specified
2656 * textsearch configuration. Use textsearch_next() to retrieve
2657 * subsequent occurrences of the pattern. Returns the offset
2658 * to the first occurrence or UINT_MAX if no match was found.
2660 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2661 unsigned int to, struct ts_config *config,
2662 struct ts_state *state)
2666 config->get_next_block = skb_ts_get_next_block;
2667 config->finish = skb_ts_finish;
2669 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2671 ret = textsearch_find(config, state);
2672 return (ret <= to - from ? ret : UINT_MAX);
2674 EXPORT_SYMBOL(skb_find_text);
2677 * skb_append_datato_frags - append the user data to a skb
2678 * @sk: sock structure
2679 * @skb: skb structure to be appened with user data.
2680 * @getfrag: call back function to be used for getting the user data
2681 * @from: pointer to user message iov
2682 * @length: length of the iov message
2684 * Description: This procedure append the user data in the fragment part
2685 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2687 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2688 int (*getfrag)(void *from, char *to, int offset,
2689 int len, int odd, struct sk_buff *skb),
2690 void *from, int length)
2692 int frg_cnt = skb_shinfo(skb)->nr_frags;
2696 struct page_frag *pfrag = ¤t->task_frag;
2699 /* Return error if we don't have space for new frag */
2700 if (frg_cnt >= MAX_SKB_FRAGS)
2703 if (!sk_page_frag_refill(sk, pfrag))
2706 /* copy the user data to page */
2707 copy = min_t(int, length, pfrag->size - pfrag->offset);
2709 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2710 offset, copy, 0, skb);
2714 /* copy was successful so update the size parameters */
2715 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2718 pfrag->offset += copy;
2719 get_page(pfrag->page);
2721 skb->truesize += copy;
2722 atomic_add(copy, &sk->sk_wmem_alloc);
2724 skb->data_len += copy;
2728 } while (length > 0);
2732 EXPORT_SYMBOL(skb_append_datato_frags);
2735 * skb_pull_rcsum - pull skb and update receive checksum
2736 * @skb: buffer to update
2737 * @len: length of data pulled
2739 * This function performs an skb_pull on the packet and updates
2740 * the CHECKSUM_COMPLETE checksum. It should be used on
2741 * receive path processing instead of skb_pull unless you know
2742 * that the checksum difference is zero (e.g., a valid IP header)
2743 * or you are setting ip_summed to CHECKSUM_NONE.
2745 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2747 BUG_ON(len > skb->len);
2749 BUG_ON(skb->len < skb->data_len);
2750 skb_postpull_rcsum(skb, skb->data, len);
2751 return skb->data += len;
2753 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2756 * skb_segment - Perform protocol segmentation on skb.
2757 * @skb: buffer to segment
2758 * @features: features for the output path (see dev->features)
2760 * This function performs segmentation on the given skb. It returns
2761 * a pointer to the first in a list of new skbs for the segments.
2762 * In case of error it returns ERR_PTR(err).
2764 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2766 struct sk_buff *segs = NULL;
2767 struct sk_buff *tail = NULL;
2768 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2769 skb_frag_t *skb_frag = skb_shinfo(skb)->frags;
2770 unsigned int mss = skb_shinfo(skb)->gso_size;
2771 unsigned int doffset = skb->data - skb_mac_header(skb);
2772 unsigned int offset = doffset;
2773 unsigned int tnl_hlen = skb_tnl_header_len(skb);
2774 unsigned int headroom;
2778 int sg = !!(features & NETIF_F_SG);
2779 int nfrags = skb_shinfo(skb)->nr_frags;
2784 proto = skb_network_protocol(skb);
2785 if (unlikely(!proto))
2786 return ERR_PTR(-EINVAL);
2788 csum = !!can_checksum_protocol(features, proto);
2789 __skb_push(skb, doffset);
2790 headroom = skb_headroom(skb);
2791 pos = skb_headlen(skb);
2794 struct sk_buff *nskb;
2799 len = skb->len - offset;
2803 hsize = skb_headlen(skb) - offset;
2806 if (hsize > len || !sg)
2809 if (!hsize && i >= nfrags && skb_headlen(fskb) &&
2810 (skb_headlen(fskb) == len || sg)) {
2811 BUG_ON(skb_headlen(fskb) > len);
2814 nfrags = skb_shinfo(fskb)->nr_frags;
2815 skb_frag = skb_shinfo(fskb)->frags;
2816 pos += skb_headlen(fskb);
2818 while (pos < offset + len) {
2819 BUG_ON(i >= nfrags);
2821 size = skb_frag_size(skb_frag);
2822 if (pos + size > offset + len)
2830 nskb = skb_clone(fskb, GFP_ATOMIC);
2833 if (unlikely(!nskb))
2836 if (unlikely(pskb_trim(nskb, len))) {
2841 hsize = skb_end_offset(nskb);
2842 if (skb_cow_head(nskb, doffset + headroom)) {
2847 nskb->truesize += skb_end_offset(nskb) - hsize;
2848 skb_release_head_state(nskb);
2849 __skb_push(nskb, doffset);
2851 nskb = __alloc_skb(hsize + doffset + headroom,
2852 GFP_ATOMIC, skb_alloc_rx_flag(skb),
2855 if (unlikely(!nskb))
2858 skb_reserve(nskb, headroom);
2859 __skb_put(nskb, doffset);
2868 __copy_skb_header(nskb, skb);
2869 nskb->mac_len = skb->mac_len;
2871 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
2873 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
2874 nskb->data - tnl_hlen,
2875 doffset + tnl_hlen);
2877 if (nskb->len == len + doffset)
2878 goto perform_csum_check;
2881 nskb->ip_summed = CHECKSUM_NONE;
2882 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2888 frag = skb_shinfo(nskb)->frags;
2890 skb_copy_from_linear_data_offset(skb, offset,
2891 skb_put(nskb, hsize), hsize);
2893 skb_shinfo(nskb)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2895 while (pos < offset + len) {
2897 BUG_ON(skb_headlen(fskb));
2900 nfrags = skb_shinfo(fskb)->nr_frags;
2901 skb_frag = skb_shinfo(fskb)->frags;
2908 if (unlikely(skb_shinfo(nskb)->nr_frags >=
2910 net_warn_ratelimited(
2911 "skb_segment: too many frags: %u %u\n",
2917 __skb_frag_ref(frag);
2918 size = skb_frag_size(frag);
2921 frag->page_offset += offset - pos;
2922 skb_frag_size_sub(frag, offset - pos);
2925 skb_shinfo(nskb)->nr_frags++;
2927 if (pos + size <= offset + len) {
2932 skb_frag_size_sub(frag, pos + size - (offset + len));
2940 nskb->data_len = len - hsize;
2941 nskb->len += nskb->data_len;
2942 nskb->truesize += nskb->data_len;
2946 nskb->csum = skb_checksum(nskb, doffset,
2947 nskb->len - doffset, 0);
2948 nskb->ip_summed = CHECKSUM_NONE;
2950 } while ((offset += len) < skb->len);
2955 while ((skb = segs)) {
2959 return ERR_PTR(err);
2961 EXPORT_SYMBOL_GPL(skb_segment);
2963 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2965 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
2966 unsigned int offset = skb_gro_offset(skb);
2967 unsigned int headlen = skb_headlen(skb);
2968 struct sk_buff *nskb, *lp, *p = *head;
2969 unsigned int len = skb_gro_len(skb);
2970 unsigned int delta_truesize;
2971 unsigned int headroom;
2973 if (unlikely(p->len + len >= 65536))
2976 lp = NAPI_GRO_CB(p)->last ?: p;
2977 pinfo = skb_shinfo(lp);
2979 if (headlen <= offset) {
2982 int i = skbinfo->nr_frags;
2983 int nr_frags = pinfo->nr_frags + i;
2985 if (nr_frags > MAX_SKB_FRAGS)
2989 pinfo->nr_frags = nr_frags;
2990 skbinfo->nr_frags = 0;
2992 frag = pinfo->frags + nr_frags;
2993 frag2 = skbinfo->frags + i;
2998 frag->page_offset += offset;
2999 skb_frag_size_sub(frag, offset);
3001 /* all fragments truesize : remove (head size + sk_buff) */
3002 delta_truesize = skb->truesize -
3003 SKB_TRUESIZE(skb_end_offset(skb));
3005 skb->truesize -= skb->data_len;
3006 skb->len -= skb->data_len;
3009 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3011 } else if (skb->head_frag) {
3012 int nr_frags = pinfo->nr_frags;
3013 skb_frag_t *frag = pinfo->frags + nr_frags;
3014 struct page *page = virt_to_head_page(skb->head);
3015 unsigned int first_size = headlen - offset;
3016 unsigned int first_offset;
3018 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3021 first_offset = skb->data -
3022 (unsigned char *)page_address(page) +
3025 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3027 frag->page.p = page;
3028 frag->page_offset = first_offset;
3029 skb_frag_size_set(frag, first_size);
3031 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3032 /* We dont need to clear skbinfo->nr_frags here */
3034 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3035 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3038 if (pinfo->frag_list)
3040 if (skb_gro_len(p) != pinfo->gso_size)
3043 headroom = skb_headroom(p);
3044 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
3045 if (unlikely(!nskb))
3048 __copy_skb_header(nskb, p);
3049 nskb->mac_len = p->mac_len;
3051 skb_reserve(nskb, headroom);
3052 __skb_put(nskb, skb_gro_offset(p));
3054 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3055 skb_set_network_header(nskb, skb_network_offset(p));
3056 skb_set_transport_header(nskb, skb_transport_offset(p));
3058 __skb_pull(p, skb_gro_offset(p));
3059 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3060 p->data - skb_mac_header(p));
3062 skb_shinfo(nskb)->frag_list = p;
3063 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3064 pinfo->gso_size = 0;
3065 skb_header_release(p);
3066 NAPI_GRO_CB(nskb)->last = p;
3068 nskb->data_len += p->len;
3069 nskb->truesize += p->truesize;
3070 nskb->len += p->len;
3073 nskb->next = p->next;
3079 delta_truesize = skb->truesize;
3080 if (offset > headlen) {
3081 unsigned int eat = offset - headlen;
3083 skbinfo->frags[0].page_offset += eat;
3084 skb_frag_size_sub(&skbinfo->frags[0], eat);
3085 skb->data_len -= eat;
3090 __skb_pull(skb, offset);
3092 if (!NAPI_GRO_CB(p)->last)
3093 skb_shinfo(p)->frag_list = skb;
3095 NAPI_GRO_CB(p)->last->next = skb;
3096 NAPI_GRO_CB(p)->last = skb;
3097 skb_header_release(skb);
3101 NAPI_GRO_CB(p)->count++;
3103 p->truesize += delta_truesize;
3106 lp->data_len += len;
3107 lp->truesize += delta_truesize;
3110 NAPI_GRO_CB(skb)->same_flow = 1;
3113 EXPORT_SYMBOL_GPL(skb_gro_receive);
3115 void __init skb_init(void)
3117 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3118 sizeof(struct sk_buff),
3120 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3122 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3123 (2*sizeof(struct sk_buff)) +
3126 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3131 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3132 * @skb: Socket buffer containing the buffers to be mapped
3133 * @sg: The scatter-gather list to map into
3134 * @offset: The offset into the buffer's contents to start mapping
3135 * @len: Length of buffer space to be mapped
3137 * Fill the specified scatter-gather list with mappings/pointers into a
3138 * region of the buffer space attached to a socket buffer.
3141 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3143 int start = skb_headlen(skb);
3144 int i, copy = start - offset;
3145 struct sk_buff *frag_iter;
3151 sg_set_buf(sg, skb->data + offset, copy);
3153 if ((len -= copy) == 0)
3158 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3161 WARN_ON(start > offset + len);
3163 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3164 if ((copy = end - offset) > 0) {
3165 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3169 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3170 frag->page_offset+offset-start);
3179 skb_walk_frags(skb, frag_iter) {
3182 WARN_ON(start > offset + len);
3184 end = start + frag_iter->len;
3185 if ((copy = end - offset) > 0) {
3188 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3190 if ((len -= copy) == 0)
3200 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3202 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3204 sg_mark_end(&sg[nsg - 1]);
3208 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3211 * skb_cow_data - Check that a socket buffer's data buffers are writable
3212 * @skb: The socket buffer to check.
3213 * @tailbits: Amount of trailing space to be added
3214 * @trailer: Returned pointer to the skb where the @tailbits space begins
3216 * Make sure that the data buffers attached to a socket buffer are
3217 * writable. If they are not, private copies are made of the data buffers
3218 * and the socket buffer is set to use these instead.
3220 * If @tailbits is given, make sure that there is space to write @tailbits
3221 * bytes of data beyond current end of socket buffer. @trailer will be
3222 * set to point to the skb in which this space begins.
3224 * The number of scatterlist elements required to completely map the
3225 * COW'd and extended socket buffer will be returned.
3227 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3231 struct sk_buff *skb1, **skb_p;
3233 /* If skb is cloned or its head is paged, reallocate
3234 * head pulling out all the pages (pages are considered not writable
3235 * at the moment even if they are anonymous).
3237 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3238 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3241 /* Easy case. Most of packets will go this way. */
3242 if (!skb_has_frag_list(skb)) {
3243 /* A little of trouble, not enough of space for trailer.
3244 * This should not happen, when stack is tuned to generate
3245 * good frames. OK, on miss we reallocate and reserve even more
3246 * space, 128 bytes is fair. */
3248 if (skb_tailroom(skb) < tailbits &&
3249 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3257 /* Misery. We are in troubles, going to mincer fragments... */
3260 skb_p = &skb_shinfo(skb)->frag_list;
3263 while ((skb1 = *skb_p) != NULL) {
3266 /* The fragment is partially pulled by someone,
3267 * this can happen on input. Copy it and everything
3270 if (skb_shared(skb1))
3273 /* If the skb is the last, worry about trailer. */
3275 if (skb1->next == NULL && tailbits) {
3276 if (skb_shinfo(skb1)->nr_frags ||
3277 skb_has_frag_list(skb1) ||
3278 skb_tailroom(skb1) < tailbits)
3279 ntail = tailbits + 128;
3285 skb_shinfo(skb1)->nr_frags ||
3286 skb_has_frag_list(skb1)) {
3287 struct sk_buff *skb2;
3289 /* Fuck, we are miserable poor guys... */
3291 skb2 = skb_copy(skb1, GFP_ATOMIC);
3293 skb2 = skb_copy_expand(skb1,
3297 if (unlikely(skb2 == NULL))
3301 skb_set_owner_w(skb2, skb1->sk);
3303 /* Looking around. Are we still alive?
3304 * OK, link new skb, drop old one */
3306 skb2->next = skb1->next;
3313 skb_p = &skb1->next;
3318 EXPORT_SYMBOL_GPL(skb_cow_data);
3320 static void sock_rmem_free(struct sk_buff *skb)
3322 struct sock *sk = skb->sk;
3324 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3328 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3330 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3334 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3335 (unsigned int)sk->sk_rcvbuf)
3340 skb->destructor = sock_rmem_free;
3341 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3343 /* before exiting rcu section, make sure dst is refcounted */
3346 skb_queue_tail(&sk->sk_error_queue, skb);
3347 if (!sock_flag(sk, SOCK_DEAD))
3348 sk->sk_data_ready(sk, len);
3351 EXPORT_SYMBOL(sock_queue_err_skb);
3353 void skb_tstamp_tx(struct sk_buff *orig_skb,
3354 struct skb_shared_hwtstamps *hwtstamps)
3356 struct sock *sk = orig_skb->sk;
3357 struct sock_exterr_skb *serr;
3358 struct sk_buff *skb;
3365 *skb_hwtstamps(orig_skb) =
3369 * no hardware time stamps available,
3370 * so keep the shared tx_flags and only
3371 * store software time stamp
3373 orig_skb->tstamp = ktime_get_real();
3376 skb = skb_clone(orig_skb, GFP_ATOMIC);
3380 serr = SKB_EXT_ERR(skb);
3381 memset(serr, 0, sizeof(*serr));
3382 serr->ee.ee_errno = ENOMSG;
3383 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3385 err = sock_queue_err_skb(sk, skb);
3390 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3392 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3394 struct sock *sk = skb->sk;
3395 struct sock_exterr_skb *serr;
3398 skb->wifi_acked_valid = 1;
3399 skb->wifi_acked = acked;
3401 serr = SKB_EXT_ERR(skb);
3402 memset(serr, 0, sizeof(*serr));
3403 serr->ee.ee_errno = ENOMSG;
3404 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3406 err = sock_queue_err_skb(sk, skb);
3410 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3414 * skb_partial_csum_set - set up and verify partial csum values for packet
3415 * @skb: the skb to set
3416 * @start: the number of bytes after skb->data to start checksumming.
3417 * @off: the offset from start to place the checksum.
3419 * For untrusted partially-checksummed packets, we need to make sure the values
3420 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3422 * This function checks and sets those values and skb->ip_summed: if this
3423 * returns false you should drop the packet.
3425 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3427 if (unlikely(start > skb_headlen(skb)) ||
3428 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3429 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3430 start, off, skb_headlen(skb));
3433 skb->ip_summed = CHECKSUM_PARTIAL;
3434 skb->csum_start = skb_headroom(skb) + start;
3435 skb->csum_offset = off;
3436 skb_set_transport_header(skb, start);
3439 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3441 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3443 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3446 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3448 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3451 skb_release_head_state(skb);
3452 kmem_cache_free(skbuff_head_cache, skb);
3457 EXPORT_SYMBOL(kfree_skb_partial);
3460 * skb_try_coalesce - try to merge skb to prior one
3462 * @from: buffer to add
3463 * @fragstolen: pointer to boolean
3464 * @delta_truesize: how much more was allocated than was requested
3466 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3467 bool *fragstolen, int *delta_truesize)
3469 int i, delta, len = from->len;
3471 *fragstolen = false;
3476 if (len <= skb_tailroom(to)) {
3477 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3478 *delta_truesize = 0;
3482 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3485 if (skb_headlen(from) != 0) {
3487 unsigned int offset;
3489 if (skb_shinfo(to)->nr_frags +
3490 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3493 if (skb_head_is_locked(from))
3496 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3498 page = virt_to_head_page(from->head);
3499 offset = from->data - (unsigned char *)page_address(page);
3501 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3502 page, offset, skb_headlen(from));
3505 if (skb_shinfo(to)->nr_frags +
3506 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3509 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
3512 WARN_ON_ONCE(delta < len);
3514 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3515 skb_shinfo(from)->frags,
3516 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3517 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3519 if (!skb_cloned(from))
3520 skb_shinfo(from)->nr_frags = 0;
3522 /* if the skb is not cloned this does nothing
3523 * since we set nr_frags to 0.
3525 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3526 skb_frag_ref(from, i);
3528 to->truesize += delta;
3530 to->data_len += len;
3532 *delta_truesize = delta;
3535 EXPORT_SYMBOL(skb_try_coalesce);
3538 * skb_scrub_packet - scrub an skb
3540 * @skb: buffer to clean
3541 * @xnet: packet is crossing netns
3543 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
3544 * into/from a tunnel. Some information have to be cleared during these
3546 * skb_scrub_packet can also be used to clean a skb before injecting it in
3547 * another namespace (@xnet == true). We have to clear all information in the
3548 * skb that could impact namespace isolation.
3550 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
3554 skb->tstamp.tv64 = 0;
3555 skb->pkt_type = PACKET_HOST;
3562 nf_reset_trace(skb);
3564 EXPORT_SYMBOL_GPL(skb_scrub_packet);