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 * Keep out-of-line to prevent kernel bloat.
109 * __builtin_return_address is not used because it is not always
114 * skb_over_panic - private function
119 * Out of line support code for skb_put(). Not user callable.
121 static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
123 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
124 __func__, here, skb->len, sz, skb->head, skb->data,
125 (unsigned long)skb->tail, (unsigned long)skb->end,
126 skb->dev ? skb->dev->name : "<NULL>");
131 * skb_under_panic - private function
136 * Out of line support code for skb_push(). Not user callable.
139 static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
141 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
142 __func__, here, skb->len, sz, skb->head, skb->data,
143 (unsigned long)skb->tail, (unsigned long)skb->end,
144 skb->dev ? skb->dev->name : "<NULL>");
150 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
151 * the caller if emergency pfmemalloc reserves are being used. If it is and
152 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
153 * may be used. Otherwise, the packet data may be discarded until enough
156 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
157 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
159 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
160 unsigned long ip, bool *pfmemalloc)
163 bool ret_pfmemalloc = false;
166 * Try a regular allocation, when that fails and we're not entitled
167 * to the reserves, fail.
169 obj = kmalloc_node_track_caller(size,
170 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
172 if (obj || !(gfp_pfmemalloc_allowed(flags)))
175 /* Try again but now we are using pfmemalloc reserves */
176 ret_pfmemalloc = true;
177 obj = kmalloc_node_track_caller(size, flags, node);
181 *pfmemalloc = ret_pfmemalloc;
186 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
187 * 'private' fields and also do memory statistics to find all the
193 * __alloc_skb - allocate a network buffer
194 * @size: size to allocate
195 * @gfp_mask: allocation mask
196 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
197 * instead of head cache and allocate a cloned (child) skb.
198 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
199 * allocations in case the data is required for writeback
200 * @node: numa node to allocate memory on
202 * Allocate a new &sk_buff. The returned buffer has no headroom and a
203 * tail room of at least size bytes. The object has a reference count
204 * of one. The return is the buffer. On a failure the return is %NULL.
206 * Buffers may only be allocated from interrupts using a @gfp_mask of
209 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
212 struct kmem_cache *cache;
213 struct skb_shared_info *shinfo;
218 cache = (flags & SKB_ALLOC_FCLONE)
219 ? skbuff_fclone_cache : skbuff_head_cache;
221 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
222 gfp_mask |= __GFP_MEMALLOC;
225 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
230 /* We do our best to align skb_shared_info on a separate cache
231 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
232 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
233 * Both skb->head and skb_shared_info are cache line aligned.
235 size = SKB_DATA_ALIGN(size);
236 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
237 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
240 /* kmalloc(size) might give us more room than requested.
241 * Put skb_shared_info exactly at the end of allocated zone,
242 * to allow max possible filling before reallocation.
244 size = SKB_WITH_OVERHEAD(ksize(data));
245 prefetchw(data + size);
248 * Only clear those fields we need to clear, not those that we will
249 * actually initialise below. Hence, don't put any more fields after
250 * the tail pointer in struct sk_buff!
252 memset(skb, 0, offsetof(struct sk_buff, tail));
253 /* Account for allocated memory : skb + skb->head */
254 skb->truesize = SKB_TRUESIZE(size);
255 skb->pfmemalloc = pfmemalloc;
256 atomic_set(&skb->users, 1);
259 skb_reset_tail_pointer(skb);
260 skb->end = skb->tail + size;
261 #ifdef NET_SKBUFF_DATA_USES_OFFSET
262 skb->mac_header = ~0U;
263 skb->transport_header = ~0U;
266 /* make sure we initialize shinfo sequentially */
267 shinfo = skb_shinfo(skb);
268 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
269 atomic_set(&shinfo->dataref, 1);
270 kmemcheck_annotate_variable(shinfo->destructor_arg);
272 if (flags & SKB_ALLOC_FCLONE) {
273 struct sk_buff *child = skb + 1;
274 atomic_t *fclone_ref = (atomic_t *) (child + 1);
276 kmemcheck_annotate_bitfield(child, flags1);
277 kmemcheck_annotate_bitfield(child, flags2);
278 skb->fclone = SKB_FCLONE_ORIG;
279 atomic_set(fclone_ref, 1);
281 child->fclone = SKB_FCLONE_UNAVAILABLE;
282 child->pfmemalloc = pfmemalloc;
287 kmem_cache_free(cache, skb);
291 EXPORT_SYMBOL(__alloc_skb);
294 * build_skb - build a network buffer
295 * @data: data buffer provided by caller
296 * @frag_size: size of fragment, or 0 if head was kmalloced
298 * Allocate a new &sk_buff. Caller provides space holding head and
299 * skb_shared_info. @data must have been allocated by kmalloc()
300 * The return is the new skb buffer.
301 * On a failure the return is %NULL, and @data is not freed.
303 * Before IO, driver allocates only data buffer where NIC put incoming frame
304 * Driver should add room at head (NET_SKB_PAD) and
305 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
306 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
307 * before giving packet to stack.
308 * RX rings only contains data buffers, not full skbs.
310 struct sk_buff *build_skb(void *data, unsigned int frag_size)
312 struct skb_shared_info *shinfo;
314 unsigned int size = frag_size ? : ksize(data);
316 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
320 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
322 memset(skb, 0, offsetof(struct sk_buff, tail));
323 skb->truesize = SKB_TRUESIZE(size);
324 skb->head_frag = frag_size != 0;
325 atomic_set(&skb->users, 1);
328 skb_reset_tail_pointer(skb);
329 skb->end = skb->tail + size;
330 #ifdef NET_SKBUFF_DATA_USES_OFFSET
331 skb->mac_header = ~0U;
332 skb->transport_header = ~0U;
335 /* make sure we initialize shinfo sequentially */
336 shinfo = skb_shinfo(skb);
337 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
338 atomic_set(&shinfo->dataref, 1);
339 kmemcheck_annotate_variable(shinfo->destructor_arg);
343 EXPORT_SYMBOL(build_skb);
345 struct netdev_alloc_cache {
346 struct page_frag frag;
347 /* we maintain a pagecount bias, so that we dont dirty cache line
348 * containing page->_count every time we allocate a fragment.
350 unsigned int pagecnt_bias;
352 static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
354 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
355 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
356 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
358 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
360 struct netdev_alloc_cache *nc;
365 local_irq_save(flags);
366 nc = &__get_cpu_var(netdev_alloc_cache);
367 if (unlikely(!nc->frag.page)) {
369 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
370 gfp_t gfp = gfp_mask;
373 gfp |= __GFP_COMP | __GFP_NOWARN;
374 nc->frag.page = alloc_pages(gfp, order);
375 if (likely(nc->frag.page))
380 nc->frag.size = PAGE_SIZE << order;
382 atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS);
383 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
387 if (nc->frag.offset + fragsz > nc->frag.size) {
388 /* avoid unnecessary locked operations if possible */
389 if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) ||
390 atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count))
395 data = page_address(nc->frag.page) + nc->frag.offset;
396 nc->frag.offset += fragsz;
399 local_irq_restore(flags);
404 * netdev_alloc_frag - allocate a page fragment
405 * @fragsz: fragment size
407 * Allocates a frag from a page for receive buffer.
408 * Uses GFP_ATOMIC allocations.
410 void *netdev_alloc_frag(unsigned int fragsz)
412 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
414 EXPORT_SYMBOL(netdev_alloc_frag);
417 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
418 * @dev: network device to receive on
419 * @length: length to allocate
420 * @gfp_mask: get_free_pages mask, passed to alloc_skb
422 * Allocate a new &sk_buff and assign it a usage count of one. The
423 * buffer has unspecified headroom built in. Users should allocate
424 * the headroom they think they need without accounting for the
425 * built in space. The built in space is used for optimisations.
427 * %NULL is returned if there is no free memory.
429 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
430 unsigned int length, gfp_t gfp_mask)
432 struct sk_buff *skb = NULL;
433 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
434 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
436 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
439 if (sk_memalloc_socks())
440 gfp_mask |= __GFP_MEMALLOC;
442 data = __netdev_alloc_frag(fragsz, gfp_mask);
445 skb = build_skb(data, fragsz);
447 put_page(virt_to_head_page(data));
450 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
451 SKB_ALLOC_RX, NUMA_NO_NODE);
454 skb_reserve(skb, NET_SKB_PAD);
459 EXPORT_SYMBOL(__netdev_alloc_skb);
461 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
462 int size, unsigned int truesize)
464 skb_fill_page_desc(skb, i, page, off, size);
466 skb->data_len += size;
467 skb->truesize += truesize;
469 EXPORT_SYMBOL(skb_add_rx_frag);
471 static void skb_drop_list(struct sk_buff **listp)
473 struct sk_buff *list = *listp;
478 struct sk_buff *this = list;
484 static inline void skb_drop_fraglist(struct sk_buff *skb)
486 skb_drop_list(&skb_shinfo(skb)->frag_list);
489 static void skb_clone_fraglist(struct sk_buff *skb)
491 struct sk_buff *list;
493 skb_walk_frags(skb, list)
497 static void skb_free_head(struct sk_buff *skb)
500 put_page(virt_to_head_page(skb->head));
505 static void skb_release_data(struct sk_buff *skb)
508 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
509 &skb_shinfo(skb)->dataref)) {
510 if (skb_shinfo(skb)->nr_frags) {
512 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
513 skb_frag_unref(skb, i);
517 * If skb buf is from userspace, we need to notify the caller
518 * the lower device DMA has done;
520 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
521 struct ubuf_info *uarg;
523 uarg = skb_shinfo(skb)->destructor_arg;
525 uarg->callback(uarg, true);
528 if (skb_has_frag_list(skb))
529 skb_drop_fraglist(skb);
536 * Free an skbuff by memory without cleaning the state.
538 static void kfree_skbmem(struct sk_buff *skb)
540 struct sk_buff *other;
541 atomic_t *fclone_ref;
543 switch (skb->fclone) {
544 case SKB_FCLONE_UNAVAILABLE:
545 kmem_cache_free(skbuff_head_cache, skb);
548 case SKB_FCLONE_ORIG:
549 fclone_ref = (atomic_t *) (skb + 2);
550 if (atomic_dec_and_test(fclone_ref))
551 kmem_cache_free(skbuff_fclone_cache, skb);
554 case SKB_FCLONE_CLONE:
555 fclone_ref = (atomic_t *) (skb + 1);
558 /* The clone portion is available for
559 * fast-cloning again.
561 skb->fclone = SKB_FCLONE_UNAVAILABLE;
563 if (atomic_dec_and_test(fclone_ref))
564 kmem_cache_free(skbuff_fclone_cache, other);
569 static void skb_release_head_state(struct sk_buff *skb)
573 secpath_put(skb->sp);
575 if (skb->destructor) {
577 skb->destructor(skb);
579 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
580 nf_conntrack_put(skb->nfct);
582 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
583 nf_conntrack_put_reasm(skb->nfct_reasm);
585 #ifdef CONFIG_BRIDGE_NETFILTER
586 nf_bridge_put(skb->nf_bridge);
588 /* XXX: IS this still necessary? - JHS */
589 #ifdef CONFIG_NET_SCHED
591 #ifdef CONFIG_NET_CLS_ACT
597 /* Free everything but the sk_buff shell. */
598 static void skb_release_all(struct sk_buff *skb)
600 skb_release_head_state(skb);
601 skb_release_data(skb);
605 * __kfree_skb - private function
608 * Free an sk_buff. Release anything attached to the buffer.
609 * Clean the state. This is an internal helper function. Users should
610 * always call kfree_skb
613 void __kfree_skb(struct sk_buff *skb)
615 skb_release_all(skb);
618 EXPORT_SYMBOL(__kfree_skb);
621 * kfree_skb - free an sk_buff
622 * @skb: buffer to free
624 * Drop a reference to the buffer and free it if the usage count has
627 void kfree_skb(struct sk_buff *skb)
631 if (likely(atomic_read(&skb->users) == 1))
633 else if (likely(!atomic_dec_and_test(&skb->users)))
635 trace_kfree_skb(skb, __builtin_return_address(0));
638 EXPORT_SYMBOL(kfree_skb);
641 * skb_tx_error - report an sk_buff xmit error
642 * @skb: buffer that triggered an error
644 * Report xmit error if a device callback is tracking this skb.
645 * skb must be freed afterwards.
647 void skb_tx_error(struct sk_buff *skb)
649 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
650 struct ubuf_info *uarg;
652 uarg = skb_shinfo(skb)->destructor_arg;
654 uarg->callback(uarg, false);
655 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
658 EXPORT_SYMBOL(skb_tx_error);
661 * consume_skb - free an skbuff
662 * @skb: buffer to free
664 * Drop a ref to the buffer and free it if the usage count has hit zero
665 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
666 * is being dropped after a failure and notes that
668 void consume_skb(struct sk_buff *skb)
672 if (likely(atomic_read(&skb->users) == 1))
674 else if (likely(!atomic_dec_and_test(&skb->users)))
676 trace_consume_skb(skb);
679 EXPORT_SYMBOL(consume_skb);
681 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
683 new->tstamp = old->tstamp;
685 new->transport_header = old->transport_header;
686 new->network_header = old->network_header;
687 new->mac_header = old->mac_header;
688 new->inner_transport_header = old->inner_transport_header;
689 new->inner_network_header = old->inner_network_header;
690 skb_dst_copy(new, old);
691 new->rxhash = old->rxhash;
692 new->ooo_okay = old->ooo_okay;
693 new->l4_rxhash = old->l4_rxhash;
694 new->no_fcs = old->no_fcs;
695 new->encapsulation = old->encapsulation;
697 new->sp = secpath_get(old->sp);
699 memcpy(new->cb, old->cb, sizeof(old->cb));
700 new->csum = old->csum;
701 new->local_df = old->local_df;
702 new->pkt_type = old->pkt_type;
703 new->ip_summed = old->ip_summed;
704 skb_copy_queue_mapping(new, old);
705 new->priority = old->priority;
706 #if IS_ENABLED(CONFIG_IP_VS)
707 new->ipvs_property = old->ipvs_property;
709 new->pfmemalloc = old->pfmemalloc;
710 new->protocol = old->protocol;
711 new->mark = old->mark;
712 new->skb_iif = old->skb_iif;
714 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
715 new->nf_trace = old->nf_trace;
717 #ifdef CONFIG_NET_SCHED
718 new->tc_index = old->tc_index;
719 #ifdef CONFIG_NET_CLS_ACT
720 new->tc_verd = old->tc_verd;
723 new->vlan_tci = old->vlan_tci;
725 skb_copy_secmark(new, old);
729 * You should not add any new code to this function. Add it to
730 * __copy_skb_header above instead.
732 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
734 #define C(x) n->x = skb->x
736 n->next = n->prev = NULL;
738 __copy_skb_header(n, skb);
743 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
746 n->destructor = NULL;
753 atomic_set(&n->users, 1);
755 atomic_inc(&(skb_shinfo(skb)->dataref));
763 * skb_morph - morph one skb into another
764 * @dst: the skb to receive the contents
765 * @src: the skb to supply the contents
767 * This is identical to skb_clone except that the target skb is
768 * supplied by the user.
770 * The target skb is returned upon exit.
772 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
774 skb_release_all(dst);
775 return __skb_clone(dst, src);
777 EXPORT_SYMBOL_GPL(skb_morph);
780 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
781 * @skb: the skb to modify
782 * @gfp_mask: allocation priority
784 * This must be called on SKBTX_DEV_ZEROCOPY skb.
785 * It will copy all frags into kernel and drop the reference
786 * to userspace pages.
788 * If this function is called from an interrupt gfp_mask() must be
791 * Returns 0 on success or a negative error code on failure
792 * to allocate kernel memory to copy to.
794 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
797 int num_frags = skb_shinfo(skb)->nr_frags;
798 struct page *page, *head = NULL;
799 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
801 for (i = 0; i < num_frags; i++) {
803 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
805 page = alloc_page(gfp_mask);
808 struct page *next = (struct page *)head->private;
814 vaddr = kmap_atomic(skb_frag_page(f));
815 memcpy(page_address(page),
816 vaddr + f->page_offset, skb_frag_size(f));
817 kunmap_atomic(vaddr);
818 page->private = (unsigned long)head;
822 /* skb frags release userspace buffers */
823 for (i = 0; i < num_frags; i++)
824 skb_frag_unref(skb, i);
826 uarg->callback(uarg, false);
828 /* skb frags point to kernel buffers */
829 for (i = num_frags - 1; i >= 0; i--) {
830 __skb_fill_page_desc(skb, i, head, 0,
831 skb_shinfo(skb)->frags[i].size);
832 head = (struct page *)head->private;
835 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
838 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
841 * skb_clone - duplicate an sk_buff
842 * @skb: buffer to clone
843 * @gfp_mask: allocation priority
845 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
846 * copies share the same packet data but not structure. The new
847 * buffer has a reference count of 1. If the allocation fails the
848 * function returns %NULL otherwise the new buffer is returned.
850 * If this function is called from an interrupt gfp_mask() must be
854 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
858 if (skb_orphan_frags(skb, gfp_mask))
862 if (skb->fclone == SKB_FCLONE_ORIG &&
863 n->fclone == SKB_FCLONE_UNAVAILABLE) {
864 atomic_t *fclone_ref = (atomic_t *) (n + 1);
865 n->fclone = SKB_FCLONE_CLONE;
866 atomic_inc(fclone_ref);
868 if (skb_pfmemalloc(skb))
869 gfp_mask |= __GFP_MEMALLOC;
871 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
875 kmemcheck_annotate_bitfield(n, flags1);
876 kmemcheck_annotate_bitfield(n, flags2);
877 n->fclone = SKB_FCLONE_UNAVAILABLE;
880 return __skb_clone(n, skb);
882 EXPORT_SYMBOL(skb_clone);
884 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
886 #ifndef NET_SKBUFF_DATA_USES_OFFSET
888 * Shift between the two data areas in bytes
890 unsigned long offset = new->data - old->data;
893 __copy_skb_header(new, old);
895 #ifndef NET_SKBUFF_DATA_USES_OFFSET
896 /* {transport,network,mac}_header are relative to skb->head */
897 new->transport_header += offset;
898 new->network_header += offset;
899 if (skb_mac_header_was_set(new))
900 new->mac_header += offset;
901 new->inner_transport_header += offset;
902 new->inner_network_header += offset;
904 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
905 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
906 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
909 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
911 if (skb_pfmemalloc(skb))
917 * skb_copy - create private copy of an sk_buff
918 * @skb: buffer to copy
919 * @gfp_mask: allocation priority
921 * Make a copy of both an &sk_buff and its data. This is used when the
922 * caller wishes to modify the data and needs a private copy of the
923 * data to alter. Returns %NULL on failure or the pointer to the buffer
924 * on success. The returned buffer has a reference count of 1.
926 * As by-product this function converts non-linear &sk_buff to linear
927 * one, so that &sk_buff becomes completely private and caller is allowed
928 * to modify all the data of returned buffer. This means that this
929 * function is not recommended for use in circumstances when only
930 * header is going to be modified. Use pskb_copy() instead.
933 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
935 int headerlen = skb_headroom(skb);
936 unsigned int size = skb_end_offset(skb) + skb->data_len;
937 struct sk_buff *n = __alloc_skb(size, gfp_mask,
938 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
943 /* Set the data pointer */
944 skb_reserve(n, headerlen);
945 /* Set the tail pointer and length */
946 skb_put(n, skb->len);
948 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
951 copy_skb_header(n, skb);
954 EXPORT_SYMBOL(skb_copy);
957 * __pskb_copy - create copy of an sk_buff with private head.
958 * @skb: buffer to copy
959 * @headroom: headroom of new skb
960 * @gfp_mask: allocation priority
962 * Make a copy of both an &sk_buff and part of its data, located
963 * in header. Fragmented data remain shared. This is used when
964 * the caller wishes to modify only header of &sk_buff and needs
965 * private copy of the header to alter. Returns %NULL on failure
966 * or the pointer to the buffer on success.
967 * The returned buffer has a reference count of 1.
970 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
972 unsigned int size = skb_headlen(skb) + headroom;
973 struct sk_buff *n = __alloc_skb(size, gfp_mask,
974 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
979 /* Set the data pointer */
980 skb_reserve(n, headroom);
981 /* Set the tail pointer and length */
982 skb_put(n, skb_headlen(skb));
984 skb_copy_from_linear_data(skb, n->data, n->len);
986 n->truesize += skb->data_len;
987 n->data_len = skb->data_len;
990 if (skb_shinfo(skb)->nr_frags) {
993 if (skb_orphan_frags(skb, gfp_mask)) {
998 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
999 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1000 skb_frag_ref(skb, i);
1002 skb_shinfo(n)->nr_frags = i;
1005 if (skb_has_frag_list(skb)) {
1006 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1007 skb_clone_fraglist(n);
1010 copy_skb_header(n, skb);
1014 EXPORT_SYMBOL(__pskb_copy);
1017 * pskb_expand_head - reallocate header of &sk_buff
1018 * @skb: buffer to reallocate
1019 * @nhead: room to add at head
1020 * @ntail: room to add at tail
1021 * @gfp_mask: allocation priority
1023 * Expands (or creates identical copy, if &nhead and &ntail are zero)
1024 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
1025 * reference count of 1. Returns zero in the case of success or error,
1026 * if expansion failed. In the last case, &sk_buff is not changed.
1028 * All the pointers pointing into skb header may change and must be
1029 * reloaded after call to this function.
1032 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1037 int size = nhead + skb_end_offset(skb) + ntail;
1042 if (skb_shared(skb))
1045 size = SKB_DATA_ALIGN(size);
1047 if (skb_pfmemalloc(skb))
1048 gfp_mask |= __GFP_MEMALLOC;
1049 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1050 gfp_mask, NUMA_NO_NODE, NULL);
1053 size = SKB_WITH_OVERHEAD(ksize(data));
1055 /* Copy only real data... and, alas, header. This should be
1056 * optimized for the cases when header is void.
1058 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1060 memcpy((struct skb_shared_info *)(data + size),
1062 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1065 * if shinfo is shared we must drop the old head gracefully, but if it
1066 * is not we can just drop the old head and let the existing refcount
1067 * be since all we did is relocate the values
1069 if (skb_cloned(skb)) {
1070 /* copy this zero copy skb frags */
1071 if (skb_orphan_frags(skb, gfp_mask))
1073 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1074 skb_frag_ref(skb, i);
1076 if (skb_has_frag_list(skb))
1077 skb_clone_fraglist(skb);
1079 skb_release_data(skb);
1083 off = (data + nhead) - skb->head;
1088 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1092 skb->end = skb->head + size;
1094 /* {transport,network,mac}_header and tail are relative to skb->head */
1096 skb->transport_header += off;
1097 skb->network_header += off;
1098 if (skb_mac_header_was_set(skb))
1099 skb->mac_header += off;
1100 skb->inner_transport_header += off;
1101 skb->inner_network_header += off;
1102 /* Only adjust this if it actually is csum_start rather than csum */
1103 if (skb->ip_summed == CHECKSUM_PARTIAL)
1104 skb->csum_start += nhead;
1108 atomic_set(&skb_shinfo(skb)->dataref, 1);
1116 EXPORT_SYMBOL(pskb_expand_head);
1118 /* Make private copy of skb with writable head and some headroom */
1120 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1122 struct sk_buff *skb2;
1123 int delta = headroom - skb_headroom(skb);
1126 skb2 = pskb_copy(skb, GFP_ATOMIC);
1128 skb2 = skb_clone(skb, GFP_ATOMIC);
1129 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1137 EXPORT_SYMBOL(skb_realloc_headroom);
1140 * skb_copy_expand - copy and expand sk_buff
1141 * @skb: buffer to copy
1142 * @newheadroom: new free bytes at head
1143 * @newtailroom: new free bytes at tail
1144 * @gfp_mask: allocation priority
1146 * Make a copy of both an &sk_buff and its data and while doing so
1147 * allocate additional space.
1149 * This is used when the caller wishes to modify the data and needs a
1150 * private copy of the data to alter as well as more space for new fields.
1151 * Returns %NULL on failure or the pointer to the buffer
1152 * on success. The returned buffer has a reference count of 1.
1154 * You must pass %GFP_ATOMIC as the allocation priority if this function
1155 * is called from an interrupt.
1157 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1158 int newheadroom, int newtailroom,
1162 * Allocate the copy buffer
1164 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1165 gfp_mask, skb_alloc_rx_flag(skb),
1167 int oldheadroom = skb_headroom(skb);
1168 int head_copy_len, head_copy_off;
1174 skb_reserve(n, newheadroom);
1176 /* Set the tail pointer and length */
1177 skb_put(n, skb->len);
1179 head_copy_len = oldheadroom;
1181 if (newheadroom <= head_copy_len)
1182 head_copy_len = newheadroom;
1184 head_copy_off = newheadroom - head_copy_len;
1186 /* Copy the linear header and data. */
1187 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1188 skb->len + head_copy_len))
1191 copy_skb_header(n, skb);
1193 off = newheadroom - oldheadroom;
1194 if (n->ip_summed == CHECKSUM_PARTIAL)
1195 n->csum_start += off;
1196 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1197 n->transport_header += off;
1198 n->network_header += off;
1199 if (skb_mac_header_was_set(skb))
1200 n->mac_header += off;
1201 n->inner_transport_header += off;
1202 n->inner_network_header += off;
1207 EXPORT_SYMBOL(skb_copy_expand);
1210 * skb_pad - zero pad the tail of an skb
1211 * @skb: buffer to pad
1212 * @pad: space to pad
1214 * Ensure that a buffer is followed by a padding area that is zero
1215 * filled. Used by network drivers which may DMA or transfer data
1216 * beyond the buffer end onto the wire.
1218 * May return error in out of memory cases. The skb is freed on error.
1221 int skb_pad(struct sk_buff *skb, int pad)
1226 /* If the skbuff is non linear tailroom is always zero.. */
1227 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1228 memset(skb->data+skb->len, 0, pad);
1232 ntail = skb->data_len + pad - (skb->end - skb->tail);
1233 if (likely(skb_cloned(skb) || ntail > 0)) {
1234 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1239 /* FIXME: The use of this function with non-linear skb's really needs
1242 err = skb_linearize(skb);
1246 memset(skb->data + skb->len, 0, pad);
1253 EXPORT_SYMBOL(skb_pad);
1256 * skb_put - add data to a buffer
1257 * @skb: buffer to use
1258 * @len: amount of data to add
1260 * This function extends the used data area of the buffer. If this would
1261 * exceed the total buffer size the kernel will panic. A pointer to the
1262 * first byte of the extra data is returned.
1264 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1266 unsigned char *tmp = skb_tail_pointer(skb);
1267 SKB_LINEAR_ASSERT(skb);
1270 if (unlikely(skb->tail > skb->end))
1271 skb_over_panic(skb, len, __builtin_return_address(0));
1274 EXPORT_SYMBOL(skb_put);
1277 * skb_push - add data to the start of a buffer
1278 * @skb: buffer to use
1279 * @len: amount of data to add
1281 * This function extends the used data area of the buffer at the buffer
1282 * start. If this would exceed the total buffer headroom the kernel will
1283 * panic. A pointer to the first byte of the extra data is returned.
1285 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1289 if (unlikely(skb->data<skb->head))
1290 skb_under_panic(skb, len, __builtin_return_address(0));
1293 EXPORT_SYMBOL(skb_push);
1296 * skb_pull - remove data from the start of a buffer
1297 * @skb: buffer to use
1298 * @len: amount of data to remove
1300 * This function removes data from the start of a buffer, returning
1301 * the memory to the headroom. A pointer to the next data in the buffer
1302 * is returned. Once the data has been pulled future pushes will overwrite
1305 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1307 return skb_pull_inline(skb, len);
1309 EXPORT_SYMBOL(skb_pull);
1312 * skb_trim - remove end from a buffer
1313 * @skb: buffer to alter
1316 * Cut the length of a buffer down by removing data from the tail. If
1317 * the buffer is already under the length specified it is not modified.
1318 * The skb must be linear.
1320 void skb_trim(struct sk_buff *skb, unsigned int len)
1323 __skb_trim(skb, len);
1325 EXPORT_SYMBOL(skb_trim);
1327 /* Trims skb to length len. It can change skb pointers.
1330 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1332 struct sk_buff **fragp;
1333 struct sk_buff *frag;
1334 int offset = skb_headlen(skb);
1335 int nfrags = skb_shinfo(skb)->nr_frags;
1339 if (skb_cloned(skb) &&
1340 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1347 for (; i < nfrags; i++) {
1348 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1355 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1358 skb_shinfo(skb)->nr_frags = i;
1360 for (; i < nfrags; i++)
1361 skb_frag_unref(skb, i);
1363 if (skb_has_frag_list(skb))
1364 skb_drop_fraglist(skb);
1368 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1369 fragp = &frag->next) {
1370 int end = offset + frag->len;
1372 if (skb_shared(frag)) {
1373 struct sk_buff *nfrag;
1375 nfrag = skb_clone(frag, GFP_ATOMIC);
1376 if (unlikely(!nfrag))
1379 nfrag->next = frag->next;
1391 unlikely((err = pskb_trim(frag, len - offset))))
1395 skb_drop_list(&frag->next);
1400 if (len > skb_headlen(skb)) {
1401 skb->data_len -= skb->len - len;
1406 skb_set_tail_pointer(skb, len);
1411 EXPORT_SYMBOL(___pskb_trim);
1414 * __pskb_pull_tail - advance tail of skb header
1415 * @skb: buffer to reallocate
1416 * @delta: number of bytes to advance tail
1418 * The function makes a sense only on a fragmented &sk_buff,
1419 * it expands header moving its tail forward and copying necessary
1420 * data from fragmented part.
1422 * &sk_buff MUST have reference count of 1.
1424 * Returns %NULL (and &sk_buff does not change) if pull failed
1425 * or value of new tail of skb in the case of success.
1427 * All the pointers pointing into skb header may change and must be
1428 * reloaded after call to this function.
1431 /* Moves tail of skb head forward, copying data from fragmented part,
1432 * when it is necessary.
1433 * 1. It may fail due to malloc failure.
1434 * 2. It may change skb pointers.
1436 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1438 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1440 /* If skb has not enough free space at tail, get new one
1441 * plus 128 bytes for future expansions. If we have enough
1442 * room at tail, reallocate without expansion only if skb is cloned.
1444 int i, k, eat = (skb->tail + delta) - skb->end;
1446 if (eat > 0 || skb_cloned(skb)) {
1447 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1452 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1455 /* Optimization: no fragments, no reasons to preestimate
1456 * size of pulled pages. Superb.
1458 if (!skb_has_frag_list(skb))
1461 /* Estimate size of pulled pages. */
1463 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1464 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1471 /* If we need update frag list, we are in troubles.
1472 * Certainly, it possible to add an offset to skb data,
1473 * but taking into account that pulling is expected to
1474 * be very rare operation, it is worth to fight against
1475 * further bloating skb head and crucify ourselves here instead.
1476 * Pure masohism, indeed. 8)8)
1479 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1480 struct sk_buff *clone = NULL;
1481 struct sk_buff *insp = NULL;
1486 if (list->len <= eat) {
1487 /* Eaten as whole. */
1492 /* Eaten partially. */
1494 if (skb_shared(list)) {
1495 /* Sucks! We need to fork list. :-( */
1496 clone = skb_clone(list, GFP_ATOMIC);
1502 /* This may be pulled without
1506 if (!pskb_pull(list, eat)) {
1514 /* Free pulled out fragments. */
1515 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1516 skb_shinfo(skb)->frag_list = list->next;
1519 /* And insert new clone at head. */
1522 skb_shinfo(skb)->frag_list = clone;
1525 /* Success! Now we may commit changes to skb data. */
1530 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1531 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1534 skb_frag_unref(skb, i);
1537 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1539 skb_shinfo(skb)->frags[k].page_offset += eat;
1540 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1546 skb_shinfo(skb)->nr_frags = k;
1549 skb->data_len -= delta;
1551 return skb_tail_pointer(skb);
1553 EXPORT_SYMBOL(__pskb_pull_tail);
1556 * skb_copy_bits - copy bits from skb to kernel buffer
1558 * @offset: offset in source
1559 * @to: destination buffer
1560 * @len: number of bytes to copy
1562 * Copy the specified number of bytes from the source skb to the
1563 * destination buffer.
1566 * If its prototype is ever changed,
1567 * check arch/{*}/net/{*}.S files,
1568 * since it is called from BPF assembly code.
1570 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1572 int start = skb_headlen(skb);
1573 struct sk_buff *frag_iter;
1576 if (offset > (int)skb->len - len)
1580 if ((copy = start - offset) > 0) {
1583 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1584 if ((len -= copy) == 0)
1590 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1592 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1594 WARN_ON(start > offset + len);
1596 end = start + skb_frag_size(f);
1597 if ((copy = end - offset) > 0) {
1603 vaddr = kmap_atomic(skb_frag_page(f));
1605 vaddr + f->page_offset + offset - start,
1607 kunmap_atomic(vaddr);
1609 if ((len -= copy) == 0)
1617 skb_walk_frags(skb, frag_iter) {
1620 WARN_ON(start > offset + len);
1622 end = start + frag_iter->len;
1623 if ((copy = end - offset) > 0) {
1626 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1628 if ((len -= copy) == 0)
1642 EXPORT_SYMBOL(skb_copy_bits);
1645 * Callback from splice_to_pipe(), if we need to release some pages
1646 * at the end of the spd in case we error'ed out in filling the pipe.
1648 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1650 put_page(spd->pages[i]);
1653 static struct page *linear_to_page(struct page *page, unsigned int *len,
1654 unsigned int *offset,
1657 struct page_frag *pfrag = sk_page_frag(sk);
1659 if (!sk_page_frag_refill(sk, pfrag))
1662 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1664 memcpy(page_address(pfrag->page) + pfrag->offset,
1665 page_address(page) + *offset, *len);
1666 *offset = pfrag->offset;
1667 pfrag->offset += *len;
1672 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1674 unsigned int offset)
1676 return spd->nr_pages &&
1677 spd->pages[spd->nr_pages - 1] == page &&
1678 (spd->partial[spd->nr_pages - 1].offset +
1679 spd->partial[spd->nr_pages - 1].len == offset);
1683 * Fill page/offset/length into spd, if it can hold more pages.
1685 static bool spd_fill_page(struct splice_pipe_desc *spd,
1686 struct pipe_inode_info *pipe, struct page *page,
1687 unsigned int *len, unsigned int offset,
1691 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1695 page = linear_to_page(page, len, &offset, sk);
1699 if (spd_can_coalesce(spd, page, offset)) {
1700 spd->partial[spd->nr_pages - 1].len += *len;
1704 spd->pages[spd->nr_pages] = page;
1705 spd->partial[spd->nr_pages].len = *len;
1706 spd->partial[spd->nr_pages].offset = offset;
1712 static bool __splice_segment(struct page *page, unsigned int poff,
1713 unsigned int plen, unsigned int *off,
1715 struct splice_pipe_desc *spd, bool linear,
1717 struct pipe_inode_info *pipe)
1722 /* skip this segment if already processed */
1728 /* ignore any bits we already processed */
1734 unsigned int flen = min(*len, plen);
1736 if (spd_fill_page(spd, pipe, page, &flen, poff,
1742 } while (*len && plen);
1748 * Map linear and fragment data from the skb to spd. It reports true if the
1749 * pipe is full or if we already spliced the requested length.
1751 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1752 unsigned int *offset, unsigned int *len,
1753 struct splice_pipe_desc *spd, struct sock *sk)
1757 /* map the linear part :
1758 * If skb->head_frag is set, this 'linear' part is backed by a
1759 * fragment, and if the head is not shared with any clones then
1760 * we can avoid a copy since we own the head portion of this page.
1762 if (__splice_segment(virt_to_page(skb->data),
1763 (unsigned long) skb->data & (PAGE_SIZE - 1),
1766 skb_head_is_locked(skb),
1771 * then map the fragments
1773 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1774 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1776 if (__splice_segment(skb_frag_page(f),
1777 f->page_offset, skb_frag_size(f),
1778 offset, len, spd, false, sk, pipe))
1786 * Map data from the skb to a pipe. Should handle both the linear part,
1787 * the fragments, and the frag list. It does NOT handle frag lists within
1788 * the frag list, if such a thing exists. We'd probably need to recurse to
1789 * handle that cleanly.
1791 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1792 struct pipe_inode_info *pipe, unsigned int tlen,
1795 struct partial_page partial[MAX_SKB_FRAGS];
1796 struct page *pages[MAX_SKB_FRAGS];
1797 struct splice_pipe_desc spd = {
1800 .nr_pages_max = MAX_SKB_FRAGS,
1802 .ops = &sock_pipe_buf_ops,
1803 .spd_release = sock_spd_release,
1805 struct sk_buff *frag_iter;
1806 struct sock *sk = skb->sk;
1810 * __skb_splice_bits() only fails if the output has no room left,
1811 * so no point in going over the frag_list for the error case.
1813 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1819 * now see if we have a frag_list to map
1821 skb_walk_frags(skb, frag_iter) {
1824 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1831 * Drop the socket lock, otherwise we have reverse
1832 * locking dependencies between sk_lock and i_mutex
1833 * here as compared to sendfile(). We enter here
1834 * with the socket lock held, and splice_to_pipe() will
1835 * grab the pipe inode lock. For sendfile() emulation,
1836 * we call into ->sendpage() with the i_mutex lock held
1837 * and networking will grab the socket lock.
1840 ret = splice_to_pipe(pipe, &spd);
1848 * skb_store_bits - store bits from kernel buffer to skb
1849 * @skb: destination buffer
1850 * @offset: offset in destination
1851 * @from: source buffer
1852 * @len: number of bytes to copy
1854 * Copy the specified number of bytes from the source buffer to the
1855 * destination skb. This function handles all the messy bits of
1856 * traversing fragment lists and such.
1859 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1861 int start = skb_headlen(skb);
1862 struct sk_buff *frag_iter;
1865 if (offset > (int)skb->len - len)
1868 if ((copy = start - offset) > 0) {
1871 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1872 if ((len -= copy) == 0)
1878 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1879 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1882 WARN_ON(start > offset + len);
1884 end = start + skb_frag_size(frag);
1885 if ((copy = end - offset) > 0) {
1891 vaddr = kmap_atomic(skb_frag_page(frag));
1892 memcpy(vaddr + frag->page_offset + offset - start,
1894 kunmap_atomic(vaddr);
1896 if ((len -= copy) == 0)
1904 skb_walk_frags(skb, frag_iter) {
1907 WARN_ON(start > offset + len);
1909 end = start + frag_iter->len;
1910 if ((copy = end - offset) > 0) {
1913 if (skb_store_bits(frag_iter, offset - start,
1916 if ((len -= copy) == 0)
1929 EXPORT_SYMBOL(skb_store_bits);
1931 /* Checksum skb data. */
1933 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1934 int len, __wsum csum)
1936 int start = skb_headlen(skb);
1937 int i, copy = start - offset;
1938 struct sk_buff *frag_iter;
1941 /* Checksum header. */
1945 csum = csum_partial(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 = csum_partial(vaddr + frag->page_offset +
1967 offset - start, copy, 0);
1968 kunmap_atomic(vaddr);
1969 csum = csum_block_add(csum, csum2, pos);
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 = csum_block_add(csum, csum2, pos);
1991 if ((len -= copy) == 0)
2002 EXPORT_SYMBOL(skb_checksum);
2004 /* Both of above in one bottle. */
2006 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2007 u8 *to, int len, __wsum csum)
2009 int start = skb_headlen(skb);
2010 int i, copy = start - offset;
2011 struct sk_buff *frag_iter;
2018 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2020 if ((len -= copy) == 0)
2027 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2030 WARN_ON(start > offset + len);
2032 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2033 if ((copy = end - offset) > 0) {
2036 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2040 vaddr = kmap_atomic(skb_frag_page(frag));
2041 csum2 = csum_partial_copy_nocheck(vaddr +
2045 kunmap_atomic(vaddr);
2046 csum = csum_block_add(csum, csum2, pos);
2056 skb_walk_frags(skb, frag_iter) {
2060 WARN_ON(start > offset + len);
2062 end = start + frag_iter->len;
2063 if ((copy = end - offset) > 0) {
2066 csum2 = skb_copy_and_csum_bits(frag_iter,
2069 csum = csum_block_add(csum, csum2, pos);
2070 if ((len -= copy) == 0)
2081 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2083 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2088 if (skb->ip_summed == CHECKSUM_PARTIAL)
2089 csstart = skb_checksum_start_offset(skb);
2091 csstart = skb_headlen(skb);
2093 BUG_ON(csstart > skb_headlen(skb));
2095 skb_copy_from_linear_data(skb, to, csstart);
2098 if (csstart != skb->len)
2099 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2100 skb->len - csstart, 0);
2102 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2103 long csstuff = csstart + skb->csum_offset;
2105 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2108 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2111 * skb_dequeue - remove from the head of the queue
2112 * @list: list to dequeue from
2114 * Remove the head of the list. The list lock is taken so the function
2115 * may be used safely with other locking list functions. The head item is
2116 * returned or %NULL if the list is empty.
2119 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2121 unsigned long flags;
2122 struct sk_buff *result;
2124 spin_lock_irqsave(&list->lock, flags);
2125 result = __skb_dequeue(list);
2126 spin_unlock_irqrestore(&list->lock, flags);
2129 EXPORT_SYMBOL(skb_dequeue);
2132 * skb_dequeue_tail - remove from the tail of the queue
2133 * @list: list to dequeue from
2135 * Remove the tail of the list. The list lock is taken so the function
2136 * may be used safely with other locking list functions. The tail item is
2137 * returned or %NULL if the list is empty.
2139 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2141 unsigned long flags;
2142 struct sk_buff *result;
2144 spin_lock_irqsave(&list->lock, flags);
2145 result = __skb_dequeue_tail(list);
2146 spin_unlock_irqrestore(&list->lock, flags);
2149 EXPORT_SYMBOL(skb_dequeue_tail);
2152 * skb_queue_purge - empty a list
2153 * @list: list to empty
2155 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2156 * the list and one reference dropped. This function takes the list
2157 * lock and is atomic with respect to other list locking functions.
2159 void skb_queue_purge(struct sk_buff_head *list)
2161 struct sk_buff *skb;
2162 while ((skb = skb_dequeue(list)) != NULL)
2165 EXPORT_SYMBOL(skb_queue_purge);
2168 * skb_queue_head - queue a buffer at the list head
2169 * @list: list to use
2170 * @newsk: buffer to queue
2172 * Queue a buffer at the start of the list. This function takes the
2173 * list lock and can be used safely with other locking &sk_buff functions
2176 * A buffer cannot be placed on two lists at the same time.
2178 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2180 unsigned long flags;
2182 spin_lock_irqsave(&list->lock, flags);
2183 __skb_queue_head(list, newsk);
2184 spin_unlock_irqrestore(&list->lock, flags);
2186 EXPORT_SYMBOL(skb_queue_head);
2189 * skb_queue_tail - queue a buffer at the list tail
2190 * @list: list to use
2191 * @newsk: buffer to queue
2193 * Queue a buffer at the tail of the list. This function takes the
2194 * list lock and can be used safely with other locking &sk_buff functions
2197 * A buffer cannot be placed on two lists at the same time.
2199 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2201 unsigned long flags;
2203 spin_lock_irqsave(&list->lock, flags);
2204 __skb_queue_tail(list, newsk);
2205 spin_unlock_irqrestore(&list->lock, flags);
2207 EXPORT_SYMBOL(skb_queue_tail);
2210 * skb_unlink - remove a buffer from a list
2211 * @skb: buffer to remove
2212 * @list: list to use
2214 * Remove a packet from a list. The list locks are taken and this
2215 * function is atomic with respect to other list locked calls
2217 * You must know what list the SKB is on.
2219 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2221 unsigned long flags;
2223 spin_lock_irqsave(&list->lock, flags);
2224 __skb_unlink(skb, list);
2225 spin_unlock_irqrestore(&list->lock, flags);
2227 EXPORT_SYMBOL(skb_unlink);
2230 * skb_append - append a buffer
2231 * @old: buffer to insert after
2232 * @newsk: buffer to insert
2233 * @list: list to use
2235 * Place a packet after a given packet in a list. The list locks are taken
2236 * and this function is atomic with respect to other list locked calls.
2237 * A buffer cannot be placed on two lists at the same time.
2239 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2241 unsigned long flags;
2243 spin_lock_irqsave(&list->lock, flags);
2244 __skb_queue_after(list, old, newsk);
2245 spin_unlock_irqrestore(&list->lock, flags);
2247 EXPORT_SYMBOL(skb_append);
2250 * skb_insert - insert a buffer
2251 * @old: buffer to insert before
2252 * @newsk: buffer to insert
2253 * @list: list to use
2255 * Place a packet before a given packet in a list. The list locks are
2256 * taken and this function is atomic with respect to other list locked
2259 * A buffer cannot be placed on two lists at the same time.
2261 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2263 unsigned long flags;
2265 spin_lock_irqsave(&list->lock, flags);
2266 __skb_insert(newsk, old->prev, old, list);
2267 spin_unlock_irqrestore(&list->lock, flags);
2269 EXPORT_SYMBOL(skb_insert);
2271 static inline void skb_split_inside_header(struct sk_buff *skb,
2272 struct sk_buff* skb1,
2273 const u32 len, const int pos)
2277 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2279 /* And move data appendix as is. */
2280 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2281 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2283 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2284 skb_shinfo(skb)->nr_frags = 0;
2285 skb1->data_len = skb->data_len;
2286 skb1->len += skb1->data_len;
2289 skb_set_tail_pointer(skb, len);
2292 static inline void skb_split_no_header(struct sk_buff *skb,
2293 struct sk_buff* skb1,
2294 const u32 len, int pos)
2297 const int nfrags = skb_shinfo(skb)->nr_frags;
2299 skb_shinfo(skb)->nr_frags = 0;
2300 skb1->len = skb1->data_len = skb->len - len;
2302 skb->data_len = len - pos;
2304 for (i = 0; i < nfrags; i++) {
2305 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2307 if (pos + size > len) {
2308 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2312 * We have two variants in this case:
2313 * 1. Move all the frag to the second
2314 * part, if it is possible. F.e.
2315 * this approach is mandatory for TUX,
2316 * where splitting is expensive.
2317 * 2. Split is accurately. We make this.
2319 skb_frag_ref(skb, i);
2320 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2321 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2322 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2323 skb_shinfo(skb)->nr_frags++;
2327 skb_shinfo(skb)->nr_frags++;
2330 skb_shinfo(skb1)->nr_frags = k;
2334 * skb_split - Split fragmented skb to two parts at length len.
2335 * @skb: the buffer to split
2336 * @skb1: the buffer to receive the second part
2337 * @len: new length for skb
2339 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2341 int pos = skb_headlen(skb);
2343 skb_shinfo(skb1)->gso_type = skb_shinfo(skb)->gso_type;
2345 if (len < pos) /* Split line is inside header. */
2346 skb_split_inside_header(skb, skb1, len, pos);
2347 else /* Second chunk has no header, nothing to copy. */
2348 skb_split_no_header(skb, skb1, len, pos);
2350 EXPORT_SYMBOL(skb_split);
2352 /* Shifting from/to a cloned skb is a no-go.
2354 * Caller cannot keep skb_shinfo related pointers past calling here!
2356 static int skb_prepare_for_shift(struct sk_buff *skb)
2358 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2362 * skb_shift - Shifts paged data partially from skb to another
2363 * @tgt: buffer into which tail data gets added
2364 * @skb: buffer from which the paged data comes from
2365 * @shiftlen: shift up to this many bytes
2367 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2368 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2369 * It's up to caller to free skb if everything was shifted.
2371 * If @tgt runs out of frags, the whole operation is aborted.
2373 * Skb cannot include anything else but paged data while tgt is allowed
2374 * to have non-paged data as well.
2376 * TODO: full sized shift could be optimized but that would need
2377 * specialized skb free'er to handle frags without up-to-date nr_frags.
2379 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2381 int from, to, merge, todo;
2382 struct skb_frag_struct *fragfrom, *fragto;
2384 BUG_ON(shiftlen > skb->len);
2385 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2389 to = skb_shinfo(tgt)->nr_frags;
2390 fragfrom = &skb_shinfo(skb)->frags[from];
2392 /* Actual merge is delayed until the point when we know we can
2393 * commit all, so that we don't have to undo partial changes
2396 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2397 fragfrom->page_offset)) {
2402 todo -= skb_frag_size(fragfrom);
2404 if (skb_prepare_for_shift(skb) ||
2405 skb_prepare_for_shift(tgt))
2408 /* All previous frag pointers might be stale! */
2409 fragfrom = &skb_shinfo(skb)->frags[from];
2410 fragto = &skb_shinfo(tgt)->frags[merge];
2412 skb_frag_size_add(fragto, shiftlen);
2413 skb_frag_size_sub(fragfrom, shiftlen);
2414 fragfrom->page_offset += shiftlen;
2422 /* Skip full, not-fitting skb to avoid expensive operations */
2423 if ((shiftlen == skb->len) &&
2424 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2427 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2430 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2431 if (to == MAX_SKB_FRAGS)
2434 fragfrom = &skb_shinfo(skb)->frags[from];
2435 fragto = &skb_shinfo(tgt)->frags[to];
2437 if (todo >= skb_frag_size(fragfrom)) {
2438 *fragto = *fragfrom;
2439 todo -= skb_frag_size(fragfrom);
2444 __skb_frag_ref(fragfrom);
2445 fragto->page = fragfrom->page;
2446 fragto->page_offset = fragfrom->page_offset;
2447 skb_frag_size_set(fragto, todo);
2449 fragfrom->page_offset += todo;
2450 skb_frag_size_sub(fragfrom, todo);
2458 /* Ready to "commit" this state change to tgt */
2459 skb_shinfo(tgt)->nr_frags = to;
2462 fragfrom = &skb_shinfo(skb)->frags[0];
2463 fragto = &skb_shinfo(tgt)->frags[merge];
2465 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2466 __skb_frag_unref(fragfrom);
2469 /* Reposition in the original skb */
2471 while (from < skb_shinfo(skb)->nr_frags)
2472 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2473 skb_shinfo(skb)->nr_frags = to;
2475 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2478 /* Most likely the tgt won't ever need its checksum anymore, skb on
2479 * the other hand might need it if it needs to be resent
2481 tgt->ip_summed = CHECKSUM_PARTIAL;
2482 skb->ip_summed = CHECKSUM_PARTIAL;
2484 /* Yak, is it really working this way? Some helper please? */
2485 skb->len -= shiftlen;
2486 skb->data_len -= shiftlen;
2487 skb->truesize -= shiftlen;
2488 tgt->len += shiftlen;
2489 tgt->data_len += shiftlen;
2490 tgt->truesize += shiftlen;
2496 * skb_prepare_seq_read - Prepare a sequential read of skb data
2497 * @skb: the buffer to read
2498 * @from: lower offset of data to be read
2499 * @to: upper offset of data to be read
2500 * @st: state variable
2502 * Initializes the specified state variable. Must be called before
2503 * invoking skb_seq_read() for the first time.
2505 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2506 unsigned int to, struct skb_seq_state *st)
2508 st->lower_offset = from;
2509 st->upper_offset = to;
2510 st->root_skb = st->cur_skb = skb;
2511 st->frag_idx = st->stepped_offset = 0;
2512 st->frag_data = NULL;
2514 EXPORT_SYMBOL(skb_prepare_seq_read);
2517 * skb_seq_read - Sequentially read skb data
2518 * @consumed: number of bytes consumed by the caller so far
2519 * @data: destination pointer for data to be returned
2520 * @st: state variable
2522 * Reads a block of skb data at &consumed relative to the
2523 * lower offset specified to skb_prepare_seq_read(). Assigns
2524 * the head of the data block to &data and returns the length
2525 * of the block or 0 if the end of the skb data or the upper
2526 * offset has been reached.
2528 * The caller is not required to consume all of the data
2529 * returned, i.e. &consumed is typically set to the number
2530 * of bytes already consumed and the next call to
2531 * skb_seq_read() will return the remaining part of the block.
2533 * Note 1: The size of each block of data returned can be arbitrary,
2534 * this limitation is the cost for zerocopy seqeuental
2535 * reads of potentially non linear data.
2537 * Note 2: Fragment lists within fragments are not implemented
2538 * at the moment, state->root_skb could be replaced with
2539 * a stack for this purpose.
2541 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2542 struct skb_seq_state *st)
2544 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2547 if (unlikely(abs_offset >= st->upper_offset))
2551 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2553 if (abs_offset < block_limit && !st->frag_data) {
2554 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2555 return block_limit - abs_offset;
2558 if (st->frag_idx == 0 && !st->frag_data)
2559 st->stepped_offset += skb_headlen(st->cur_skb);
2561 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2562 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2563 block_limit = skb_frag_size(frag) + st->stepped_offset;
2565 if (abs_offset < block_limit) {
2567 st->frag_data = kmap_atomic(skb_frag_page(frag));
2569 *data = (u8 *) st->frag_data + frag->page_offset +
2570 (abs_offset - st->stepped_offset);
2572 return block_limit - abs_offset;
2575 if (st->frag_data) {
2576 kunmap_atomic(st->frag_data);
2577 st->frag_data = NULL;
2581 st->stepped_offset += skb_frag_size(frag);
2584 if (st->frag_data) {
2585 kunmap_atomic(st->frag_data);
2586 st->frag_data = NULL;
2589 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2590 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2593 } else if (st->cur_skb->next) {
2594 st->cur_skb = st->cur_skb->next;
2601 EXPORT_SYMBOL(skb_seq_read);
2604 * skb_abort_seq_read - Abort a sequential read of skb data
2605 * @st: state variable
2607 * Must be called if skb_seq_read() was not called until it
2610 void skb_abort_seq_read(struct skb_seq_state *st)
2613 kunmap_atomic(st->frag_data);
2615 EXPORT_SYMBOL(skb_abort_seq_read);
2617 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2619 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2620 struct ts_config *conf,
2621 struct ts_state *state)
2623 return skb_seq_read(offset, text, TS_SKB_CB(state));
2626 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2628 skb_abort_seq_read(TS_SKB_CB(state));
2632 * skb_find_text - Find a text pattern in skb data
2633 * @skb: the buffer to look in
2634 * @from: search offset
2636 * @config: textsearch configuration
2637 * @state: uninitialized textsearch state variable
2639 * Finds a pattern in the skb data according to the specified
2640 * textsearch configuration. Use textsearch_next() to retrieve
2641 * subsequent occurrences of the pattern. Returns the offset
2642 * to the first occurrence or UINT_MAX if no match was found.
2644 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2645 unsigned int to, struct ts_config *config,
2646 struct ts_state *state)
2650 config->get_next_block = skb_ts_get_next_block;
2651 config->finish = skb_ts_finish;
2653 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2655 ret = textsearch_find(config, state);
2656 return (ret <= to - from ? ret : UINT_MAX);
2658 EXPORT_SYMBOL(skb_find_text);
2661 * skb_append_datato_frags - append the user data to a skb
2662 * @sk: sock structure
2663 * @skb: skb structure to be appened with user data.
2664 * @getfrag: call back function to be used for getting the user data
2665 * @from: pointer to user message iov
2666 * @length: length of the iov message
2668 * Description: This procedure append the user data in the fragment part
2669 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2671 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2672 int (*getfrag)(void *from, char *to, int offset,
2673 int len, int odd, struct sk_buff *skb),
2674 void *from, int length)
2676 int frg_cnt = skb_shinfo(skb)->nr_frags;
2680 struct page_frag *pfrag = ¤t->task_frag;
2683 /* Return error if we don't have space for new frag */
2684 if (frg_cnt >= MAX_SKB_FRAGS)
2687 if (!sk_page_frag_refill(sk, pfrag))
2690 /* copy the user data to page */
2691 copy = min_t(int, length, pfrag->size - pfrag->offset);
2693 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2694 offset, copy, 0, skb);
2698 /* copy was successful so update the size parameters */
2699 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2702 pfrag->offset += copy;
2703 get_page(pfrag->page);
2705 skb->truesize += copy;
2706 atomic_add(copy, &sk->sk_wmem_alloc);
2708 skb->data_len += copy;
2712 } while (length > 0);
2716 EXPORT_SYMBOL(skb_append_datato_frags);
2719 * skb_pull_rcsum - pull skb and update receive checksum
2720 * @skb: buffer to update
2721 * @len: length of data pulled
2723 * This function performs an skb_pull on the packet and updates
2724 * the CHECKSUM_COMPLETE checksum. It should be used on
2725 * receive path processing instead of skb_pull unless you know
2726 * that the checksum difference is zero (e.g., a valid IP header)
2727 * or you are setting ip_summed to CHECKSUM_NONE.
2729 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2731 BUG_ON(len > skb->len);
2733 BUG_ON(skb->len < skb->data_len);
2734 skb_postpull_rcsum(skb, skb->data, len);
2735 return skb->data += len;
2737 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2740 * skb_segment - Perform protocol segmentation on skb.
2741 * @skb: buffer to segment
2742 * @features: features for the output path (see dev->features)
2744 * This function performs segmentation on the given skb. It returns
2745 * a pointer to the first in a list of new skbs for the segments.
2746 * In case of error it returns ERR_PTR(err).
2748 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2750 struct sk_buff *segs = NULL;
2751 struct sk_buff *tail = NULL;
2752 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2753 unsigned int mss = skb_shinfo(skb)->gso_size;
2754 unsigned int doffset = skb->data - skb_mac_header(skb);
2755 unsigned int offset = doffset;
2756 unsigned int headroom;
2758 int sg = !!(features & NETIF_F_SG);
2759 int nfrags = skb_shinfo(skb)->nr_frags;
2764 __skb_push(skb, doffset);
2765 headroom = skb_headroom(skb);
2766 pos = skb_headlen(skb);
2769 struct sk_buff *nskb;
2774 len = skb->len - offset;
2778 hsize = skb_headlen(skb) - offset;
2781 if (hsize > len || !sg)
2784 if (!hsize && i >= nfrags) {
2785 BUG_ON(fskb->len != len);
2788 nskb = skb_clone(fskb, GFP_ATOMIC);
2791 if (unlikely(!nskb))
2794 hsize = skb_end_offset(nskb);
2795 if (skb_cow_head(nskb, doffset + headroom)) {
2800 nskb->truesize += skb_end_offset(nskb) - hsize;
2801 skb_release_head_state(nskb);
2802 __skb_push(nskb, doffset);
2804 nskb = __alloc_skb(hsize + doffset + headroom,
2805 GFP_ATOMIC, skb_alloc_rx_flag(skb),
2808 if (unlikely(!nskb))
2811 skb_reserve(nskb, headroom);
2812 __skb_put(nskb, doffset);
2821 __copy_skb_header(nskb, skb);
2822 nskb->mac_len = skb->mac_len;
2824 /* nskb and skb might have different headroom */
2825 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2826 nskb->csum_start += skb_headroom(nskb) - headroom;
2828 skb_reset_mac_header(nskb);
2829 skb_set_network_header(nskb, skb->mac_len);
2830 nskb->transport_header = (nskb->network_header +
2831 skb_network_header_len(skb));
2832 skb_copy_from_linear_data(skb, nskb->data, doffset);
2834 if (fskb != skb_shinfo(skb)->frag_list)
2838 nskb->ip_summed = CHECKSUM_NONE;
2839 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2845 frag = skb_shinfo(nskb)->frags;
2847 skb_copy_from_linear_data_offset(skb, offset,
2848 skb_put(nskb, hsize), hsize);
2850 skb_shinfo(nskb)->gso_type = skb_shinfo(skb)->gso_type;
2852 while (pos < offset + len && i < nfrags) {
2853 *frag = skb_shinfo(skb)->frags[i];
2854 __skb_frag_ref(frag);
2855 size = skb_frag_size(frag);
2858 frag->page_offset += offset - pos;
2859 skb_frag_size_sub(frag, offset - pos);
2862 skb_shinfo(nskb)->nr_frags++;
2864 if (pos + size <= offset + len) {
2868 skb_frag_size_sub(frag, pos + size - (offset + len));
2875 if (pos < offset + len) {
2876 struct sk_buff *fskb2 = fskb;
2878 BUG_ON(pos + fskb->len != offset + len);
2884 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2890 SKB_FRAG_ASSERT(nskb);
2891 skb_shinfo(nskb)->frag_list = fskb2;
2895 nskb->data_len = len - hsize;
2896 nskb->len += nskb->data_len;
2897 nskb->truesize += nskb->data_len;
2898 } while ((offset += len) < skb->len);
2903 while ((skb = segs)) {
2907 return ERR_PTR(err);
2909 EXPORT_SYMBOL_GPL(skb_segment);
2911 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2913 struct sk_buff *p = *head;
2914 struct sk_buff *nskb;
2915 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2916 struct skb_shared_info *pinfo = skb_shinfo(p);
2917 unsigned int headroom;
2918 unsigned int len = skb_gro_len(skb);
2919 unsigned int offset = skb_gro_offset(skb);
2920 unsigned int headlen = skb_headlen(skb);
2921 unsigned int delta_truesize;
2923 if (p->len + len >= 65536)
2926 if (pinfo->frag_list)
2928 else if (headlen <= offset) {
2931 int i = skbinfo->nr_frags;
2932 int nr_frags = pinfo->nr_frags + i;
2936 if (nr_frags > MAX_SKB_FRAGS)
2939 pinfo->nr_frags = nr_frags;
2940 skbinfo->nr_frags = 0;
2942 frag = pinfo->frags + nr_frags;
2943 frag2 = skbinfo->frags + i;
2948 frag->page_offset += offset;
2949 skb_frag_size_sub(frag, offset);
2951 /* all fragments truesize : remove (head size + sk_buff) */
2952 delta_truesize = skb->truesize -
2953 SKB_TRUESIZE(skb_end_offset(skb));
2955 skb->truesize -= skb->data_len;
2956 skb->len -= skb->data_len;
2959 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
2961 } else if (skb->head_frag) {
2962 int nr_frags = pinfo->nr_frags;
2963 skb_frag_t *frag = pinfo->frags + nr_frags;
2964 struct page *page = virt_to_head_page(skb->head);
2965 unsigned int first_size = headlen - offset;
2966 unsigned int first_offset;
2968 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
2971 first_offset = skb->data -
2972 (unsigned char *)page_address(page) +
2975 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
2977 frag->page.p = page;
2978 frag->page_offset = first_offset;
2979 skb_frag_size_set(frag, first_size);
2981 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
2982 /* We dont need to clear skbinfo->nr_frags here */
2984 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
2985 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
2987 } else if (skb_gro_len(p) != pinfo->gso_size)
2990 headroom = skb_headroom(p);
2991 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2992 if (unlikely(!nskb))
2995 __copy_skb_header(nskb, p);
2996 nskb->mac_len = p->mac_len;
2998 skb_reserve(nskb, headroom);
2999 __skb_put(nskb, skb_gro_offset(p));
3001 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3002 skb_set_network_header(nskb, skb_network_offset(p));
3003 skb_set_transport_header(nskb, skb_transport_offset(p));
3005 __skb_pull(p, skb_gro_offset(p));
3006 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3007 p->data - skb_mac_header(p));
3009 skb_shinfo(nskb)->frag_list = p;
3010 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3011 pinfo->gso_size = 0;
3012 skb_header_release(p);
3013 NAPI_GRO_CB(nskb)->last = p;
3015 nskb->data_len += p->len;
3016 nskb->truesize += p->truesize;
3017 nskb->len += p->len;
3020 nskb->next = p->next;
3026 delta_truesize = skb->truesize;
3027 if (offset > headlen) {
3028 unsigned int eat = offset - headlen;
3030 skbinfo->frags[0].page_offset += eat;
3031 skb_frag_size_sub(&skbinfo->frags[0], eat);
3032 skb->data_len -= eat;
3037 __skb_pull(skb, offset);
3039 NAPI_GRO_CB(p)->last->next = skb;
3040 NAPI_GRO_CB(p)->last = skb;
3041 skb_header_release(skb);
3044 NAPI_GRO_CB(p)->count++;
3046 p->truesize += delta_truesize;
3049 NAPI_GRO_CB(skb)->same_flow = 1;
3052 EXPORT_SYMBOL_GPL(skb_gro_receive);
3054 void __init skb_init(void)
3056 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3057 sizeof(struct sk_buff),
3059 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3061 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3062 (2*sizeof(struct sk_buff)) +
3065 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3070 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3071 * @skb: Socket buffer containing the buffers to be mapped
3072 * @sg: The scatter-gather list to map into
3073 * @offset: The offset into the buffer's contents to start mapping
3074 * @len: Length of buffer space to be mapped
3076 * Fill the specified scatter-gather list with mappings/pointers into a
3077 * region of the buffer space attached to a socket buffer.
3080 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3082 int start = skb_headlen(skb);
3083 int i, copy = start - offset;
3084 struct sk_buff *frag_iter;
3090 sg_set_buf(sg, skb->data + offset, copy);
3092 if ((len -= copy) == 0)
3097 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3100 WARN_ON(start > offset + len);
3102 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3103 if ((copy = end - offset) > 0) {
3104 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3108 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3109 frag->page_offset+offset-start);
3118 skb_walk_frags(skb, frag_iter) {
3121 WARN_ON(start > offset + len);
3123 end = start + frag_iter->len;
3124 if ((copy = end - offset) > 0) {
3127 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3129 if ((len -= copy) == 0)
3139 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3141 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3143 sg_mark_end(&sg[nsg - 1]);
3147 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3150 * skb_cow_data - Check that a socket buffer's data buffers are writable
3151 * @skb: The socket buffer to check.
3152 * @tailbits: Amount of trailing space to be added
3153 * @trailer: Returned pointer to the skb where the @tailbits space begins
3155 * Make sure that the data buffers attached to a socket buffer are
3156 * writable. If they are not, private copies are made of the data buffers
3157 * and the socket buffer is set to use these instead.
3159 * If @tailbits is given, make sure that there is space to write @tailbits
3160 * bytes of data beyond current end of socket buffer. @trailer will be
3161 * set to point to the skb in which this space begins.
3163 * The number of scatterlist elements required to completely map the
3164 * COW'd and extended socket buffer will be returned.
3166 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3170 struct sk_buff *skb1, **skb_p;
3172 /* If skb is cloned or its head is paged, reallocate
3173 * head pulling out all the pages (pages are considered not writable
3174 * at the moment even if they are anonymous).
3176 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3177 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3180 /* Easy case. Most of packets will go this way. */
3181 if (!skb_has_frag_list(skb)) {
3182 /* A little of trouble, not enough of space for trailer.
3183 * This should not happen, when stack is tuned to generate
3184 * good frames. OK, on miss we reallocate and reserve even more
3185 * space, 128 bytes is fair. */
3187 if (skb_tailroom(skb) < tailbits &&
3188 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3196 /* Misery. We are in troubles, going to mincer fragments... */
3199 skb_p = &skb_shinfo(skb)->frag_list;
3202 while ((skb1 = *skb_p) != NULL) {
3205 /* The fragment is partially pulled by someone,
3206 * this can happen on input. Copy it and everything
3209 if (skb_shared(skb1))
3212 /* If the skb is the last, worry about trailer. */
3214 if (skb1->next == NULL && tailbits) {
3215 if (skb_shinfo(skb1)->nr_frags ||
3216 skb_has_frag_list(skb1) ||
3217 skb_tailroom(skb1) < tailbits)
3218 ntail = tailbits + 128;
3224 skb_shinfo(skb1)->nr_frags ||
3225 skb_has_frag_list(skb1)) {
3226 struct sk_buff *skb2;
3228 /* Fuck, we are miserable poor guys... */
3230 skb2 = skb_copy(skb1, GFP_ATOMIC);
3232 skb2 = skb_copy_expand(skb1,
3236 if (unlikely(skb2 == NULL))
3240 skb_set_owner_w(skb2, skb1->sk);
3242 /* Looking around. Are we still alive?
3243 * OK, link new skb, drop old one */
3245 skb2->next = skb1->next;
3252 skb_p = &skb1->next;
3257 EXPORT_SYMBOL_GPL(skb_cow_data);
3259 static void sock_rmem_free(struct sk_buff *skb)
3261 struct sock *sk = skb->sk;
3263 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3267 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3269 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3273 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3274 (unsigned int)sk->sk_rcvbuf)
3279 skb->destructor = sock_rmem_free;
3280 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3282 /* before exiting rcu section, make sure dst is refcounted */
3285 skb_queue_tail(&sk->sk_error_queue, skb);
3286 if (!sock_flag(sk, SOCK_DEAD))
3287 sk->sk_data_ready(sk, len);
3290 EXPORT_SYMBOL(sock_queue_err_skb);
3292 void skb_tstamp_tx(struct sk_buff *orig_skb,
3293 struct skb_shared_hwtstamps *hwtstamps)
3295 struct sock *sk = orig_skb->sk;
3296 struct sock_exterr_skb *serr;
3297 struct sk_buff *skb;
3303 skb = skb_clone(orig_skb, GFP_ATOMIC);
3308 *skb_hwtstamps(skb) =
3312 * no hardware time stamps available,
3313 * so keep the shared tx_flags and only
3314 * store software time stamp
3316 skb->tstamp = ktime_get_real();
3319 serr = SKB_EXT_ERR(skb);
3320 memset(serr, 0, sizeof(*serr));
3321 serr->ee.ee_errno = ENOMSG;
3322 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3324 err = sock_queue_err_skb(sk, skb);
3329 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3331 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3333 struct sock *sk = skb->sk;
3334 struct sock_exterr_skb *serr;
3337 skb->wifi_acked_valid = 1;
3338 skb->wifi_acked = acked;
3340 serr = SKB_EXT_ERR(skb);
3341 memset(serr, 0, sizeof(*serr));
3342 serr->ee.ee_errno = ENOMSG;
3343 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3345 err = sock_queue_err_skb(sk, skb);
3349 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3353 * skb_partial_csum_set - set up and verify partial csum values for packet
3354 * @skb: the skb to set
3355 * @start: the number of bytes after skb->data to start checksumming.
3356 * @off: the offset from start to place the checksum.
3358 * For untrusted partially-checksummed packets, we need to make sure the values
3359 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3361 * This function checks and sets those values and skb->ip_summed: if this
3362 * returns false you should drop the packet.
3364 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3366 if (unlikely(start > skb_headlen(skb)) ||
3367 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3368 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3369 start, off, skb_headlen(skb));
3372 skb->ip_summed = CHECKSUM_PARTIAL;
3373 skb->csum_start = skb_headroom(skb) + start;
3374 skb->csum_offset = off;
3377 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3379 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3381 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3384 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3386 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3389 skb_release_head_state(skb);
3390 kmem_cache_free(skbuff_head_cache, skb);
3395 EXPORT_SYMBOL(kfree_skb_partial);
3398 * skb_try_coalesce - try to merge skb to prior one
3400 * @from: buffer to add
3401 * @fragstolen: pointer to boolean
3402 * @delta_truesize: how much more was allocated than was requested
3404 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3405 bool *fragstolen, int *delta_truesize)
3407 int i, delta, len = from->len;
3409 *fragstolen = false;
3414 if (len <= skb_tailroom(to)) {
3415 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3416 *delta_truesize = 0;
3420 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3423 if (skb_headlen(from) != 0) {
3425 unsigned int offset;
3427 if (skb_shinfo(to)->nr_frags +
3428 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3431 if (skb_head_is_locked(from))
3434 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3436 page = virt_to_head_page(from->head);
3437 offset = from->data - (unsigned char *)page_address(page);
3439 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3440 page, offset, skb_headlen(from));
3443 if (skb_shinfo(to)->nr_frags +
3444 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3447 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
3450 WARN_ON_ONCE(delta < len);
3452 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3453 skb_shinfo(from)->frags,
3454 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3455 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3457 if (!skb_cloned(from))
3458 skb_shinfo(from)->nr_frags = 0;
3460 /* if the skb is not cloned this does nothing
3461 * since we set nr_frags to 0.
3463 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3464 skb_frag_ref(from, i);
3466 to->truesize += delta;
3468 to->data_len += len;
3470 *delta_truesize = delta;
3473 EXPORT_SYMBOL(skb_try_coalesce);