2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
8 * Alan Cox : Fixed the worst of the load
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
36 * The functions in this file will not compile correctly with gcc 2.4.x
39 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
44 #include <linux/kmemcheck.h>
46 #include <linux/interrupt.h>
48 #include <linux/inet.h>
49 #include <linux/slab.h>
50 #include <linux/netdevice.h>
51 #ifdef CONFIG_NET_CLS_ACT
52 #include <net/pkt_sched.h>
54 #include <linux/string.h>
55 #include <linux/skbuff.h>
56 #include <linux/splice.h>
57 #include <linux/cache.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/init.h>
60 #include <linux/scatterlist.h>
61 #include <linux/errqueue.h>
62 #include <linux/prefetch.h>
64 #include <net/protocol.h>
67 #include <net/checksum.h>
70 #include <asm/uaccess.h>
71 #include <trace/events/skb.h>
72 #include <linux/highmem.h>
74 struct kmem_cache *skbuff_head_cache __read_mostly;
75 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
77 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
78 struct pipe_buffer *buf)
83 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
84 struct pipe_buffer *buf)
89 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
90 struct pipe_buffer *buf)
96 /* Pipe buffer operations for a socket. */
97 static const struct pipe_buf_operations sock_pipe_buf_ops = {
99 .map = generic_pipe_buf_map,
100 .unmap = generic_pipe_buf_unmap,
101 .confirm = generic_pipe_buf_confirm,
102 .release = sock_pipe_buf_release,
103 .steal = sock_pipe_buf_steal,
104 .get = sock_pipe_buf_get,
108 * skb_panic - private function for out-of-line support
112 * @msg: skb_over_panic or skb_under_panic
114 * Out-of-line support for skb_put() and skb_push().
115 * Called via the wrapper skb_over_panic() or skb_under_panic().
116 * Keep out of line to prevent kernel bloat.
117 * __builtin_return_address is not used because it is not always reliable.
119 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
122 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
123 msg, addr, skb->len, sz, skb->head, skb->data,
124 (unsigned long)skb->tail, (unsigned long)skb->end,
125 skb->dev ? skb->dev->name : "<NULL>");
129 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
131 skb_panic(skb, sz, addr, __func__);
134 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
136 skb_panic(skb, sz, addr, __func__);
140 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
141 * the caller if emergency pfmemalloc reserves are being used. If it is and
142 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
143 * may be used. Otherwise, the packet data may be discarded until enough
146 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
147 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
149 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
150 unsigned long ip, bool *pfmemalloc)
153 bool ret_pfmemalloc = false;
156 * Try a regular allocation, when that fails and we're not entitled
157 * to the reserves, fail.
159 obj = kmalloc_node_track_caller(size,
160 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
162 if (obj || !(gfp_pfmemalloc_allowed(flags)))
165 /* Try again but now we are using pfmemalloc reserves */
166 ret_pfmemalloc = true;
167 obj = kmalloc_node_track_caller(size, flags, node);
171 *pfmemalloc = ret_pfmemalloc;
176 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
177 * 'private' fields and also do memory statistics to find all the
183 * __alloc_skb - allocate a network buffer
184 * @size: size to allocate
185 * @gfp_mask: allocation mask
186 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
187 * instead of head cache and allocate a cloned (child) skb.
188 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
189 * allocations in case the data is required for writeback
190 * @node: numa node to allocate memory on
192 * Allocate a new &sk_buff. The returned buffer has no headroom and a
193 * tail room of at least size bytes. The object has a reference count
194 * of one. The return is the buffer. On a failure the return is %NULL.
196 * Buffers may only be allocated from interrupts using a @gfp_mask of
199 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
202 struct kmem_cache *cache;
203 struct skb_shared_info *shinfo;
208 cache = (flags & SKB_ALLOC_FCLONE)
209 ? skbuff_fclone_cache : skbuff_head_cache;
211 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
212 gfp_mask |= __GFP_MEMALLOC;
215 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
220 /* We do our best to align skb_shared_info on a separate cache
221 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
222 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
223 * Both skb->head and skb_shared_info are cache line aligned.
225 size = SKB_DATA_ALIGN(size);
226 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
227 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
230 /* kmalloc(size) might give us more room than requested.
231 * Put skb_shared_info exactly at the end of allocated zone,
232 * to allow max possible filling before reallocation.
234 size = SKB_WITH_OVERHEAD(ksize(data));
235 prefetchw(data + size);
238 * Only clear those fields we need to clear, not those that we will
239 * actually initialise below. Hence, don't put any more fields after
240 * the tail pointer in struct sk_buff!
242 memset(skb, 0, offsetof(struct sk_buff, tail));
243 /* Account for allocated memory : skb + skb->head */
244 skb->truesize = SKB_TRUESIZE(size);
245 skb->pfmemalloc = pfmemalloc;
246 atomic_set(&skb->users, 1);
249 skb_reset_tail_pointer(skb);
250 skb->end = skb->tail + size;
251 #ifdef NET_SKBUFF_DATA_USES_OFFSET
252 skb->mac_header = ~0U;
253 skb->transport_header = ~0U;
256 /* make sure we initialize shinfo sequentially */
257 shinfo = skb_shinfo(skb);
258 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
259 atomic_set(&shinfo->dataref, 1);
260 kmemcheck_annotate_variable(shinfo->destructor_arg);
262 if (flags & SKB_ALLOC_FCLONE) {
263 struct sk_buff *child = skb + 1;
264 atomic_t *fclone_ref = (atomic_t *) (child + 1);
266 kmemcheck_annotate_bitfield(child, flags1);
267 kmemcheck_annotate_bitfield(child, flags2);
268 skb->fclone = SKB_FCLONE_ORIG;
269 atomic_set(fclone_ref, 1);
271 child->fclone = SKB_FCLONE_UNAVAILABLE;
272 child->pfmemalloc = pfmemalloc;
277 kmem_cache_free(cache, skb);
281 EXPORT_SYMBOL(__alloc_skb);
284 * build_skb - build a network buffer
285 * @data: data buffer provided by caller
286 * @frag_size: size of fragment, or 0 if head was kmalloced
288 * Allocate a new &sk_buff. Caller provides space holding head and
289 * skb_shared_info. @data must have been allocated by kmalloc()
290 * The return is the new skb buffer.
291 * On a failure the return is %NULL, and @data is not freed.
293 * Before IO, driver allocates only data buffer where NIC put incoming frame
294 * Driver should add room at head (NET_SKB_PAD) and
295 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
296 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
297 * before giving packet to stack.
298 * RX rings only contains data buffers, not full skbs.
300 struct sk_buff *build_skb(void *data, unsigned int frag_size)
302 struct skb_shared_info *shinfo;
304 unsigned int size = frag_size ? : ksize(data);
306 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
310 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
312 memset(skb, 0, offsetof(struct sk_buff, tail));
313 skb->truesize = SKB_TRUESIZE(size);
314 skb->head_frag = frag_size != 0;
315 atomic_set(&skb->users, 1);
318 skb_reset_tail_pointer(skb);
319 skb->end = skb->tail + size;
320 #ifdef NET_SKBUFF_DATA_USES_OFFSET
321 skb->mac_header = ~0U;
322 skb->transport_header = ~0U;
325 /* make sure we initialize shinfo sequentially */
326 shinfo = skb_shinfo(skb);
327 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
328 atomic_set(&shinfo->dataref, 1);
329 kmemcheck_annotate_variable(shinfo->destructor_arg);
333 EXPORT_SYMBOL(build_skb);
335 struct netdev_alloc_cache {
336 struct page_frag frag;
337 /* we maintain a pagecount bias, so that we dont dirty cache line
338 * containing page->_count every time we allocate a fragment.
340 unsigned int pagecnt_bias;
342 static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
344 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
346 struct netdev_alloc_cache *nc;
351 local_irq_save(flags);
352 nc = &__get_cpu_var(netdev_alloc_cache);
353 if (unlikely(!nc->frag.page)) {
355 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
356 gfp_t gfp = gfp_mask;
359 gfp |= __GFP_COMP | __GFP_NOWARN;
360 nc->frag.page = alloc_pages(gfp, order);
361 if (likely(nc->frag.page))
366 nc->frag.size = PAGE_SIZE << order;
368 atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS);
369 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
373 if (nc->frag.offset + fragsz > nc->frag.size) {
374 /* avoid unnecessary locked operations if possible */
375 if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) ||
376 atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count))
381 data = page_address(nc->frag.page) + nc->frag.offset;
382 nc->frag.offset += fragsz;
385 local_irq_restore(flags);
390 * netdev_alloc_frag - allocate a page fragment
391 * @fragsz: fragment size
393 * Allocates a frag from a page for receive buffer.
394 * Uses GFP_ATOMIC allocations.
396 void *netdev_alloc_frag(unsigned int fragsz)
398 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
400 EXPORT_SYMBOL(netdev_alloc_frag);
403 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
404 * @dev: network device to receive on
405 * @length: length to allocate
406 * @gfp_mask: get_free_pages mask, passed to alloc_skb
408 * Allocate a new &sk_buff and assign it a usage count of one. The
409 * buffer has unspecified headroom built in. Users should allocate
410 * the headroom they think they need without accounting for the
411 * built in space. The built in space is used for optimisations.
413 * %NULL is returned if there is no free memory.
415 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
416 unsigned int length, gfp_t gfp_mask)
418 struct sk_buff *skb = NULL;
419 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
420 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
422 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
425 if (sk_memalloc_socks())
426 gfp_mask |= __GFP_MEMALLOC;
428 data = __netdev_alloc_frag(fragsz, gfp_mask);
431 skb = build_skb(data, fragsz);
433 put_page(virt_to_head_page(data));
436 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
437 SKB_ALLOC_RX, NUMA_NO_NODE);
440 skb_reserve(skb, NET_SKB_PAD);
445 EXPORT_SYMBOL(__netdev_alloc_skb);
447 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
448 int size, unsigned int truesize)
450 skb_fill_page_desc(skb, i, page, off, size);
452 skb->data_len += size;
453 skb->truesize += truesize;
455 EXPORT_SYMBOL(skb_add_rx_frag);
457 static void skb_drop_list(struct sk_buff **listp)
459 struct sk_buff *list = *listp;
464 struct sk_buff *this = list;
470 static inline void skb_drop_fraglist(struct sk_buff *skb)
472 skb_drop_list(&skb_shinfo(skb)->frag_list);
475 static void skb_clone_fraglist(struct sk_buff *skb)
477 struct sk_buff *list;
479 skb_walk_frags(skb, list)
483 static void skb_free_head(struct sk_buff *skb)
486 put_page(virt_to_head_page(skb->head));
491 static void skb_release_data(struct sk_buff *skb)
494 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
495 &skb_shinfo(skb)->dataref)) {
496 if (skb_shinfo(skb)->nr_frags) {
498 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
499 skb_frag_unref(skb, i);
503 * If skb buf is from userspace, we need to notify the caller
504 * the lower device DMA has done;
506 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
507 struct ubuf_info *uarg;
509 uarg = skb_shinfo(skb)->destructor_arg;
511 uarg->callback(uarg, true);
514 if (skb_has_frag_list(skb))
515 skb_drop_fraglist(skb);
522 * Free an skbuff by memory without cleaning the state.
524 static void kfree_skbmem(struct sk_buff *skb)
526 struct sk_buff *other;
527 atomic_t *fclone_ref;
529 switch (skb->fclone) {
530 case SKB_FCLONE_UNAVAILABLE:
531 kmem_cache_free(skbuff_head_cache, skb);
534 case SKB_FCLONE_ORIG:
535 fclone_ref = (atomic_t *) (skb + 2);
536 if (atomic_dec_and_test(fclone_ref))
537 kmem_cache_free(skbuff_fclone_cache, skb);
540 case SKB_FCLONE_CLONE:
541 fclone_ref = (atomic_t *) (skb + 1);
544 /* The clone portion is available for
545 * fast-cloning again.
547 skb->fclone = SKB_FCLONE_UNAVAILABLE;
549 if (atomic_dec_and_test(fclone_ref))
550 kmem_cache_free(skbuff_fclone_cache, other);
555 static void skb_release_head_state(struct sk_buff *skb)
559 secpath_put(skb->sp);
561 if (skb->destructor) {
563 skb->destructor(skb);
565 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
566 nf_conntrack_put(skb->nfct);
568 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
569 nf_conntrack_put_reasm(skb->nfct_reasm);
571 #ifdef CONFIG_BRIDGE_NETFILTER
572 nf_bridge_put(skb->nf_bridge);
574 /* XXX: IS this still necessary? - JHS */
575 #ifdef CONFIG_NET_SCHED
577 #ifdef CONFIG_NET_CLS_ACT
583 /* Free everything but the sk_buff shell. */
584 static void skb_release_all(struct sk_buff *skb)
586 skb_release_head_state(skb);
587 skb_release_data(skb);
591 * __kfree_skb - private function
594 * Free an sk_buff. Release anything attached to the buffer.
595 * Clean the state. This is an internal helper function. Users should
596 * always call kfree_skb
599 void __kfree_skb(struct sk_buff *skb)
601 skb_release_all(skb);
604 EXPORT_SYMBOL(__kfree_skb);
607 * kfree_skb - free an sk_buff
608 * @skb: buffer to free
610 * Drop a reference to the buffer and free it if the usage count has
613 void kfree_skb(struct sk_buff *skb)
617 if (likely(atomic_read(&skb->users) == 1))
619 else if (likely(!atomic_dec_and_test(&skb->users)))
621 trace_kfree_skb(skb, __builtin_return_address(0));
624 EXPORT_SYMBOL(kfree_skb);
627 * skb_tx_error - report an sk_buff xmit error
628 * @skb: buffer that triggered an error
630 * Report xmit error if a device callback is tracking this skb.
631 * skb must be freed afterwards.
633 void skb_tx_error(struct sk_buff *skb)
635 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
636 struct ubuf_info *uarg;
638 uarg = skb_shinfo(skb)->destructor_arg;
640 uarg->callback(uarg, false);
641 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
644 EXPORT_SYMBOL(skb_tx_error);
647 * consume_skb - free an skbuff
648 * @skb: buffer to free
650 * Drop a ref to the buffer and free it if the usage count has hit zero
651 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
652 * is being dropped after a failure and notes that
654 void consume_skb(struct sk_buff *skb)
658 if (likely(atomic_read(&skb->users) == 1))
660 else if (likely(!atomic_dec_and_test(&skb->users)))
662 trace_consume_skb(skb);
665 EXPORT_SYMBOL(consume_skb);
667 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
669 new->tstamp = old->tstamp;
671 new->transport_header = old->transport_header;
672 new->network_header = old->network_header;
673 new->mac_header = old->mac_header;
674 new->inner_transport_header = old->inner_transport_header;
675 new->inner_network_header = old->inner_network_header;
676 new->inner_mac_header = old->inner_mac_header;
677 skb_dst_copy(new, old);
678 new->rxhash = old->rxhash;
679 new->ooo_okay = old->ooo_okay;
680 new->l4_rxhash = old->l4_rxhash;
681 new->no_fcs = old->no_fcs;
682 new->encapsulation = old->encapsulation;
684 new->sp = secpath_get(old->sp);
686 memcpy(new->cb, old->cb, sizeof(old->cb));
687 new->csum = old->csum;
688 new->local_df = old->local_df;
689 new->pkt_type = old->pkt_type;
690 new->ip_summed = old->ip_summed;
691 skb_copy_queue_mapping(new, old);
692 new->priority = old->priority;
693 #if IS_ENABLED(CONFIG_IP_VS)
694 new->ipvs_property = old->ipvs_property;
696 new->pfmemalloc = old->pfmemalloc;
697 new->protocol = old->protocol;
698 new->mark = old->mark;
699 new->skb_iif = old->skb_iif;
701 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
702 new->nf_trace = old->nf_trace;
704 #ifdef CONFIG_NET_SCHED
705 new->tc_index = old->tc_index;
706 #ifdef CONFIG_NET_CLS_ACT
707 new->tc_verd = old->tc_verd;
710 new->vlan_proto = old->vlan_proto;
711 new->vlan_tci = old->vlan_tci;
713 skb_copy_secmark(new, old);
717 * You should not add any new code to this function. Add it to
718 * __copy_skb_header above instead.
720 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
722 #define C(x) n->x = skb->x
724 n->next = n->prev = NULL;
726 __copy_skb_header(n, skb);
731 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
734 n->destructor = NULL;
741 atomic_set(&n->users, 1);
743 atomic_inc(&(skb_shinfo(skb)->dataref));
751 * skb_morph - morph one skb into another
752 * @dst: the skb to receive the contents
753 * @src: the skb to supply the contents
755 * This is identical to skb_clone except that the target skb is
756 * supplied by the user.
758 * The target skb is returned upon exit.
760 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
762 skb_release_all(dst);
763 return __skb_clone(dst, src);
765 EXPORT_SYMBOL_GPL(skb_morph);
768 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
769 * @skb: the skb to modify
770 * @gfp_mask: allocation priority
772 * This must be called on SKBTX_DEV_ZEROCOPY skb.
773 * It will copy all frags into kernel and drop the reference
774 * to userspace pages.
776 * If this function is called from an interrupt gfp_mask() must be
779 * Returns 0 on success or a negative error code on failure
780 * to allocate kernel memory to copy to.
782 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
785 int num_frags = skb_shinfo(skb)->nr_frags;
786 struct page *page, *head = NULL;
787 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
789 for (i = 0; i < num_frags; i++) {
791 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
793 page = alloc_page(gfp_mask);
796 struct page *next = (struct page *)head->private;
802 vaddr = kmap_atomic(skb_frag_page(f));
803 memcpy(page_address(page),
804 vaddr + f->page_offset, skb_frag_size(f));
805 kunmap_atomic(vaddr);
806 page->private = (unsigned long)head;
810 /* skb frags release userspace buffers */
811 for (i = 0; i < num_frags; i++)
812 skb_frag_unref(skb, i);
814 uarg->callback(uarg, false);
816 /* skb frags point to kernel buffers */
817 for (i = num_frags - 1; i >= 0; i--) {
818 __skb_fill_page_desc(skb, i, head, 0,
819 skb_shinfo(skb)->frags[i].size);
820 head = (struct page *)head->private;
823 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
826 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
829 * skb_clone - duplicate an sk_buff
830 * @skb: buffer to clone
831 * @gfp_mask: allocation priority
833 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
834 * copies share the same packet data but not structure. The new
835 * buffer has a reference count of 1. If the allocation fails the
836 * function returns %NULL otherwise the new buffer is returned.
838 * If this function is called from an interrupt gfp_mask() must be
842 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
846 if (skb_orphan_frags(skb, gfp_mask))
850 if (skb->fclone == SKB_FCLONE_ORIG &&
851 n->fclone == SKB_FCLONE_UNAVAILABLE) {
852 atomic_t *fclone_ref = (atomic_t *) (n + 1);
853 n->fclone = SKB_FCLONE_CLONE;
854 atomic_inc(fclone_ref);
856 if (skb_pfmemalloc(skb))
857 gfp_mask |= __GFP_MEMALLOC;
859 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
863 kmemcheck_annotate_bitfield(n, flags1);
864 kmemcheck_annotate_bitfield(n, flags2);
865 n->fclone = SKB_FCLONE_UNAVAILABLE;
868 return __skb_clone(n, skb);
870 EXPORT_SYMBOL(skb_clone);
872 static void skb_headers_offset_update(struct sk_buff *skb, int off)
874 /* {transport,network,mac}_header and tail are relative to skb->head */
875 skb->transport_header += off;
876 skb->network_header += off;
877 if (skb_mac_header_was_set(skb))
878 skb->mac_header += off;
879 skb->inner_transport_header += off;
880 skb->inner_network_header += off;
881 skb->inner_mac_header += off;
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 skb_headers_offset_update(new, offset);
898 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
899 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
900 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
903 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
905 if (skb_pfmemalloc(skb))
911 * skb_copy - create private copy of an sk_buff
912 * @skb: buffer to copy
913 * @gfp_mask: allocation priority
915 * Make a copy of both an &sk_buff and its data. This is used when the
916 * caller wishes to modify the data and needs a private copy of the
917 * data to alter. Returns %NULL on failure or the pointer to the buffer
918 * on success. The returned buffer has a reference count of 1.
920 * As by-product this function converts non-linear &sk_buff to linear
921 * one, so that &sk_buff becomes completely private and caller is allowed
922 * to modify all the data of returned buffer. This means that this
923 * function is not recommended for use in circumstances when only
924 * header is going to be modified. Use pskb_copy() instead.
927 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
929 int headerlen = skb_headroom(skb);
930 unsigned int size = skb_end_offset(skb) + skb->data_len;
931 struct sk_buff *n = __alloc_skb(size, gfp_mask,
932 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
937 /* Set the data pointer */
938 skb_reserve(n, headerlen);
939 /* Set the tail pointer and length */
940 skb_put(n, skb->len);
942 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
945 copy_skb_header(n, skb);
948 EXPORT_SYMBOL(skb_copy);
951 * __pskb_copy - create copy of an sk_buff with private head.
952 * @skb: buffer to copy
953 * @headroom: headroom of new skb
954 * @gfp_mask: allocation priority
956 * Make a copy of both an &sk_buff and part of its data, located
957 * in header. Fragmented data remain shared. This is used when
958 * the caller wishes to modify only header of &sk_buff and needs
959 * private copy of the header to alter. Returns %NULL on failure
960 * or the pointer to the buffer on success.
961 * The returned buffer has a reference count of 1.
964 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
966 unsigned int size = skb_headlen(skb) + headroom;
967 struct sk_buff *n = __alloc_skb(size, gfp_mask,
968 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
973 /* Set the data pointer */
974 skb_reserve(n, headroom);
975 /* Set the tail pointer and length */
976 skb_put(n, skb_headlen(skb));
978 skb_copy_from_linear_data(skb, n->data, n->len);
980 n->truesize += skb->data_len;
981 n->data_len = skb->data_len;
984 if (skb_shinfo(skb)->nr_frags) {
987 if (skb_orphan_frags(skb, gfp_mask)) {
992 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
993 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
994 skb_frag_ref(skb, i);
996 skb_shinfo(n)->nr_frags = i;
999 if (skb_has_frag_list(skb)) {
1000 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1001 skb_clone_fraglist(n);
1004 copy_skb_header(n, skb);
1008 EXPORT_SYMBOL(__pskb_copy);
1011 * pskb_expand_head - reallocate header of &sk_buff
1012 * @skb: buffer to reallocate
1013 * @nhead: room to add at head
1014 * @ntail: room to add at tail
1015 * @gfp_mask: allocation priority
1017 * Expands (or creates identical copy, if &nhead and &ntail are zero)
1018 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
1019 * reference count of 1. Returns zero in the case of success or error,
1020 * if expansion failed. In the last case, &sk_buff is not changed.
1022 * All the pointers pointing into skb header may change and must be
1023 * reloaded after call to this function.
1026 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1031 int size = nhead + skb_end_offset(skb) + ntail;
1036 if (skb_shared(skb))
1039 size = SKB_DATA_ALIGN(size);
1041 if (skb_pfmemalloc(skb))
1042 gfp_mask |= __GFP_MEMALLOC;
1043 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1044 gfp_mask, NUMA_NO_NODE, NULL);
1047 size = SKB_WITH_OVERHEAD(ksize(data));
1049 /* Copy only real data... and, alas, header. This should be
1050 * optimized for the cases when header is void.
1052 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1054 memcpy((struct skb_shared_info *)(data + size),
1056 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1059 * if shinfo is shared we must drop the old head gracefully, but if it
1060 * is not we can just drop the old head and let the existing refcount
1061 * be since all we did is relocate the values
1063 if (skb_cloned(skb)) {
1064 /* copy this zero copy skb frags */
1065 if (skb_orphan_frags(skb, gfp_mask))
1067 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1068 skb_frag_ref(skb, i);
1070 if (skb_has_frag_list(skb))
1071 skb_clone_fraglist(skb);
1073 skb_release_data(skb);
1077 off = (data + nhead) - skb->head;
1082 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1086 skb->end = skb->head + size;
1089 skb_headers_offset_update(skb, off);
1090 /* Only adjust this if it actually is csum_start rather than csum */
1091 if (skb->ip_summed == CHECKSUM_PARTIAL)
1092 skb->csum_start += nhead;
1096 atomic_set(&skb_shinfo(skb)->dataref, 1);
1104 EXPORT_SYMBOL(pskb_expand_head);
1106 /* Make private copy of skb with writable head and some headroom */
1108 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1110 struct sk_buff *skb2;
1111 int delta = headroom - skb_headroom(skb);
1114 skb2 = pskb_copy(skb, GFP_ATOMIC);
1116 skb2 = skb_clone(skb, GFP_ATOMIC);
1117 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1125 EXPORT_SYMBOL(skb_realloc_headroom);
1128 * skb_copy_expand - copy and expand sk_buff
1129 * @skb: buffer to copy
1130 * @newheadroom: new free bytes at head
1131 * @newtailroom: new free bytes at tail
1132 * @gfp_mask: allocation priority
1134 * Make a copy of both an &sk_buff and its data and while doing so
1135 * allocate additional space.
1137 * This is used when the caller wishes to modify the data and needs a
1138 * private copy of the data to alter as well as more space for new fields.
1139 * Returns %NULL on failure or the pointer to the buffer
1140 * on success. The returned buffer has a reference count of 1.
1142 * You must pass %GFP_ATOMIC as the allocation priority if this function
1143 * is called from an interrupt.
1145 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1146 int newheadroom, int newtailroom,
1150 * Allocate the copy buffer
1152 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1153 gfp_mask, skb_alloc_rx_flag(skb),
1155 int oldheadroom = skb_headroom(skb);
1156 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 off = newheadroom - oldheadroom;
1182 if (n->ip_summed == CHECKSUM_PARTIAL)
1183 n->csum_start += off;
1184 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1185 skb_headers_offset_update(n, off);
1190 EXPORT_SYMBOL(skb_copy_expand);
1193 * skb_pad - zero pad the tail of an skb
1194 * @skb: buffer to pad
1195 * @pad: space to pad
1197 * Ensure that a buffer is followed by a padding area that is zero
1198 * filled. Used by network drivers which may DMA or transfer data
1199 * beyond the buffer end onto the wire.
1201 * May return error in out of memory cases. The skb is freed on error.
1204 int skb_pad(struct sk_buff *skb, int pad)
1209 /* If the skbuff is non linear tailroom is always zero.. */
1210 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1211 memset(skb->data+skb->len, 0, pad);
1215 ntail = skb->data_len + pad - (skb->end - skb->tail);
1216 if (likely(skb_cloned(skb) || ntail > 0)) {
1217 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1222 /* FIXME: The use of this function with non-linear skb's really needs
1225 err = skb_linearize(skb);
1229 memset(skb->data + skb->len, 0, pad);
1236 EXPORT_SYMBOL(skb_pad);
1239 * skb_put - add data to a buffer
1240 * @skb: buffer to use
1241 * @len: amount of data to add
1243 * This function extends the used data area of the buffer. If this would
1244 * exceed the total buffer size the kernel will panic. A pointer to the
1245 * first byte of the extra data is returned.
1247 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1249 unsigned char *tmp = skb_tail_pointer(skb);
1250 SKB_LINEAR_ASSERT(skb);
1253 if (unlikely(skb->tail > skb->end))
1254 skb_over_panic(skb, len, __builtin_return_address(0));
1257 EXPORT_SYMBOL(skb_put);
1260 * skb_push - add data to the start of a buffer
1261 * @skb: buffer to use
1262 * @len: amount of data to add
1264 * This function extends the used data area of the buffer at the buffer
1265 * start. If this would exceed the total buffer headroom the kernel will
1266 * panic. A pointer to the first byte of the extra data is returned.
1268 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1272 if (unlikely(skb->data<skb->head))
1273 skb_under_panic(skb, len, __builtin_return_address(0));
1276 EXPORT_SYMBOL(skb_push);
1279 * skb_pull - remove data from the start of a buffer
1280 * @skb: buffer to use
1281 * @len: amount of data to remove
1283 * This function removes data from the start of a buffer, returning
1284 * the memory to the headroom. A pointer to the next data in the buffer
1285 * is returned. Once the data has been pulled future pushes will overwrite
1288 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1290 return skb_pull_inline(skb, len);
1292 EXPORT_SYMBOL(skb_pull);
1295 * skb_trim - remove end from a buffer
1296 * @skb: buffer to alter
1299 * Cut the length of a buffer down by removing data from the tail. If
1300 * the buffer is already under the length specified it is not modified.
1301 * The skb must be linear.
1303 void skb_trim(struct sk_buff *skb, unsigned int len)
1306 __skb_trim(skb, len);
1308 EXPORT_SYMBOL(skb_trim);
1310 /* Trims skb to length len. It can change skb pointers.
1313 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1315 struct sk_buff **fragp;
1316 struct sk_buff *frag;
1317 int offset = skb_headlen(skb);
1318 int nfrags = skb_shinfo(skb)->nr_frags;
1322 if (skb_cloned(skb) &&
1323 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1330 for (; i < nfrags; i++) {
1331 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1338 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1341 skb_shinfo(skb)->nr_frags = i;
1343 for (; i < nfrags; i++)
1344 skb_frag_unref(skb, i);
1346 if (skb_has_frag_list(skb))
1347 skb_drop_fraglist(skb);
1351 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1352 fragp = &frag->next) {
1353 int end = offset + frag->len;
1355 if (skb_shared(frag)) {
1356 struct sk_buff *nfrag;
1358 nfrag = skb_clone(frag, GFP_ATOMIC);
1359 if (unlikely(!nfrag))
1362 nfrag->next = frag->next;
1374 unlikely((err = pskb_trim(frag, len - offset))))
1378 skb_drop_list(&frag->next);
1383 if (len > skb_headlen(skb)) {
1384 skb->data_len -= skb->len - len;
1389 skb_set_tail_pointer(skb, len);
1394 EXPORT_SYMBOL(___pskb_trim);
1397 * __pskb_pull_tail - advance tail of skb header
1398 * @skb: buffer to reallocate
1399 * @delta: number of bytes to advance tail
1401 * The function makes a sense only on a fragmented &sk_buff,
1402 * it expands header moving its tail forward and copying necessary
1403 * data from fragmented part.
1405 * &sk_buff MUST have reference count of 1.
1407 * Returns %NULL (and &sk_buff does not change) if pull failed
1408 * or value of new tail of skb in the case of success.
1410 * All the pointers pointing into skb header may change and must be
1411 * reloaded after call to this function.
1414 /* Moves tail of skb head forward, copying data from fragmented part,
1415 * when it is necessary.
1416 * 1. It may fail due to malloc failure.
1417 * 2. It may change skb pointers.
1419 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1421 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1423 /* If skb has not enough free space at tail, get new one
1424 * plus 128 bytes for future expansions. If we have enough
1425 * room at tail, reallocate without expansion only if skb is cloned.
1427 int i, k, eat = (skb->tail + delta) - skb->end;
1429 if (eat > 0 || skb_cloned(skb)) {
1430 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1435 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1438 /* Optimization: no fragments, no reasons to preestimate
1439 * size of pulled pages. Superb.
1441 if (!skb_has_frag_list(skb))
1444 /* Estimate size of pulled pages. */
1446 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1447 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1454 /* If we need update frag list, we are in troubles.
1455 * Certainly, it possible to add an offset to skb data,
1456 * but taking into account that pulling is expected to
1457 * be very rare operation, it is worth to fight against
1458 * further bloating skb head and crucify ourselves here instead.
1459 * Pure masohism, indeed. 8)8)
1462 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1463 struct sk_buff *clone = NULL;
1464 struct sk_buff *insp = NULL;
1469 if (list->len <= eat) {
1470 /* Eaten as whole. */
1475 /* Eaten partially. */
1477 if (skb_shared(list)) {
1478 /* Sucks! We need to fork list. :-( */
1479 clone = skb_clone(list, GFP_ATOMIC);
1485 /* This may be pulled without
1489 if (!pskb_pull(list, eat)) {
1497 /* Free pulled out fragments. */
1498 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1499 skb_shinfo(skb)->frag_list = list->next;
1502 /* And insert new clone at head. */
1505 skb_shinfo(skb)->frag_list = clone;
1508 /* Success! Now we may commit changes to skb data. */
1513 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1514 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1517 skb_frag_unref(skb, i);
1520 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1522 skb_shinfo(skb)->frags[k].page_offset += eat;
1523 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1529 skb_shinfo(skb)->nr_frags = k;
1532 skb->data_len -= delta;
1534 return skb_tail_pointer(skb);
1536 EXPORT_SYMBOL(__pskb_pull_tail);
1539 * skb_copy_bits - copy bits from skb to kernel buffer
1541 * @offset: offset in source
1542 * @to: destination buffer
1543 * @len: number of bytes to copy
1545 * Copy the specified number of bytes from the source skb to the
1546 * destination buffer.
1549 * If its prototype is ever changed,
1550 * check arch/{*}/net/{*}.S files,
1551 * since it is called from BPF assembly code.
1553 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1555 int start = skb_headlen(skb);
1556 struct sk_buff *frag_iter;
1559 if (offset > (int)skb->len - len)
1563 if ((copy = start - offset) > 0) {
1566 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1567 if ((len -= copy) == 0)
1573 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1575 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1577 WARN_ON(start > offset + len);
1579 end = start + skb_frag_size(f);
1580 if ((copy = end - offset) > 0) {
1586 vaddr = kmap_atomic(skb_frag_page(f));
1588 vaddr + f->page_offset + offset - start,
1590 kunmap_atomic(vaddr);
1592 if ((len -= copy) == 0)
1600 skb_walk_frags(skb, frag_iter) {
1603 WARN_ON(start > offset + len);
1605 end = start + frag_iter->len;
1606 if ((copy = end - offset) > 0) {
1609 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1611 if ((len -= copy) == 0)
1625 EXPORT_SYMBOL(skb_copy_bits);
1628 * Callback from splice_to_pipe(), if we need to release some pages
1629 * at the end of the spd in case we error'ed out in filling the pipe.
1631 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1633 put_page(spd->pages[i]);
1636 static struct page *linear_to_page(struct page *page, unsigned int *len,
1637 unsigned int *offset,
1640 struct page_frag *pfrag = sk_page_frag(sk);
1642 if (!sk_page_frag_refill(sk, pfrag))
1645 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1647 memcpy(page_address(pfrag->page) + pfrag->offset,
1648 page_address(page) + *offset, *len);
1649 *offset = pfrag->offset;
1650 pfrag->offset += *len;
1655 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1657 unsigned int offset)
1659 return spd->nr_pages &&
1660 spd->pages[spd->nr_pages - 1] == page &&
1661 (spd->partial[spd->nr_pages - 1].offset +
1662 spd->partial[spd->nr_pages - 1].len == offset);
1666 * Fill page/offset/length into spd, if it can hold more pages.
1668 static bool spd_fill_page(struct splice_pipe_desc *spd,
1669 struct pipe_inode_info *pipe, struct page *page,
1670 unsigned int *len, unsigned int offset,
1674 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1678 page = linear_to_page(page, len, &offset, sk);
1682 if (spd_can_coalesce(spd, page, offset)) {
1683 spd->partial[spd->nr_pages - 1].len += *len;
1687 spd->pages[spd->nr_pages] = page;
1688 spd->partial[spd->nr_pages].len = *len;
1689 spd->partial[spd->nr_pages].offset = offset;
1695 static bool __splice_segment(struct page *page, unsigned int poff,
1696 unsigned int plen, unsigned int *off,
1698 struct splice_pipe_desc *spd, bool linear,
1700 struct pipe_inode_info *pipe)
1705 /* skip this segment if already processed */
1711 /* ignore any bits we already processed */
1717 unsigned int flen = min(*len, plen);
1719 if (spd_fill_page(spd, pipe, page, &flen, poff,
1725 } while (*len && plen);
1731 * Map linear and fragment data from the skb to spd. It reports true if the
1732 * pipe is full or if we already spliced the requested length.
1734 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1735 unsigned int *offset, unsigned int *len,
1736 struct splice_pipe_desc *spd, struct sock *sk)
1740 /* map the linear part :
1741 * If skb->head_frag is set, this 'linear' part is backed by a
1742 * fragment, and if the head is not shared with any clones then
1743 * we can avoid a copy since we own the head portion of this page.
1745 if (__splice_segment(virt_to_page(skb->data),
1746 (unsigned long) skb->data & (PAGE_SIZE - 1),
1749 skb_head_is_locked(skb),
1754 * then map the fragments
1756 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1757 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1759 if (__splice_segment(skb_frag_page(f),
1760 f->page_offset, skb_frag_size(f),
1761 offset, len, spd, false, sk, pipe))
1769 * Map data from the skb to a pipe. Should handle both the linear part,
1770 * the fragments, and the frag list. It does NOT handle frag lists within
1771 * the frag list, if such a thing exists. We'd probably need to recurse to
1772 * handle that cleanly.
1774 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1775 struct pipe_inode_info *pipe, unsigned int tlen,
1778 struct partial_page partial[MAX_SKB_FRAGS];
1779 struct page *pages[MAX_SKB_FRAGS];
1780 struct splice_pipe_desc spd = {
1783 .nr_pages_max = MAX_SKB_FRAGS,
1785 .ops = &sock_pipe_buf_ops,
1786 .spd_release = sock_spd_release,
1788 struct sk_buff *frag_iter;
1789 struct sock *sk = skb->sk;
1793 * __skb_splice_bits() only fails if the output has no room left,
1794 * so no point in going over the frag_list for the error case.
1796 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1802 * now see if we have a frag_list to map
1804 skb_walk_frags(skb, frag_iter) {
1807 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1814 * Drop the socket lock, otherwise we have reverse
1815 * locking dependencies between sk_lock and i_mutex
1816 * here as compared to sendfile(). We enter here
1817 * with the socket lock held, and splice_to_pipe() will
1818 * grab the pipe inode lock. For sendfile() emulation,
1819 * we call into ->sendpage() with the i_mutex lock held
1820 * and networking will grab the socket lock.
1823 ret = splice_to_pipe(pipe, &spd);
1831 * skb_store_bits - store bits from kernel buffer to skb
1832 * @skb: destination buffer
1833 * @offset: offset in destination
1834 * @from: source buffer
1835 * @len: number of bytes to copy
1837 * Copy the specified number of bytes from the source buffer to the
1838 * destination skb. This function handles all the messy bits of
1839 * traversing fragment lists and such.
1842 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1844 int start = skb_headlen(skb);
1845 struct sk_buff *frag_iter;
1848 if (offset > (int)skb->len - len)
1851 if ((copy = start - offset) > 0) {
1854 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1855 if ((len -= copy) == 0)
1861 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1862 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1865 WARN_ON(start > offset + len);
1867 end = start + skb_frag_size(frag);
1868 if ((copy = end - offset) > 0) {
1874 vaddr = kmap_atomic(skb_frag_page(frag));
1875 memcpy(vaddr + frag->page_offset + offset - start,
1877 kunmap_atomic(vaddr);
1879 if ((len -= copy) == 0)
1887 skb_walk_frags(skb, frag_iter) {
1890 WARN_ON(start > offset + len);
1892 end = start + frag_iter->len;
1893 if ((copy = end - offset) > 0) {
1896 if (skb_store_bits(frag_iter, offset - start,
1899 if ((len -= copy) == 0)
1912 EXPORT_SYMBOL(skb_store_bits);
1914 /* Checksum skb data. */
1916 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1917 int len, __wsum csum)
1919 int start = skb_headlen(skb);
1920 int i, copy = start - offset;
1921 struct sk_buff *frag_iter;
1924 /* Checksum header. */
1928 csum = csum_partial(skb->data + offset, copy, csum);
1929 if ((len -= copy) == 0)
1935 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1937 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1939 WARN_ON(start > offset + len);
1941 end = start + skb_frag_size(frag);
1942 if ((copy = end - offset) > 0) {
1948 vaddr = kmap_atomic(skb_frag_page(frag));
1949 csum2 = csum_partial(vaddr + frag->page_offset +
1950 offset - start, copy, 0);
1951 kunmap_atomic(vaddr);
1952 csum = csum_block_add(csum, csum2, pos);
1961 skb_walk_frags(skb, frag_iter) {
1964 WARN_ON(start > offset + len);
1966 end = start + frag_iter->len;
1967 if ((copy = end - offset) > 0) {
1971 csum2 = skb_checksum(frag_iter, offset - start,
1973 csum = csum_block_add(csum, csum2, pos);
1974 if ((len -= copy) == 0)
1985 EXPORT_SYMBOL(skb_checksum);
1987 /* Both of above in one bottle. */
1989 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1990 u8 *to, int len, __wsum csum)
1992 int start = skb_headlen(skb);
1993 int i, copy = start - offset;
1994 struct sk_buff *frag_iter;
2001 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2003 if ((len -= copy) == 0)
2010 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2013 WARN_ON(start > offset + len);
2015 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2016 if ((copy = end - offset) > 0) {
2019 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2023 vaddr = kmap_atomic(skb_frag_page(frag));
2024 csum2 = csum_partial_copy_nocheck(vaddr +
2028 kunmap_atomic(vaddr);
2029 csum = csum_block_add(csum, csum2, pos);
2039 skb_walk_frags(skb, frag_iter) {
2043 WARN_ON(start > offset + len);
2045 end = start + frag_iter->len;
2046 if ((copy = end - offset) > 0) {
2049 csum2 = skb_copy_and_csum_bits(frag_iter,
2052 csum = csum_block_add(csum, csum2, pos);
2053 if ((len -= copy) == 0)
2064 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2066 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2071 if (skb->ip_summed == CHECKSUM_PARTIAL)
2072 csstart = skb_checksum_start_offset(skb);
2074 csstart = skb_headlen(skb);
2076 BUG_ON(csstart > skb_headlen(skb));
2078 skb_copy_from_linear_data(skb, to, csstart);
2081 if (csstart != skb->len)
2082 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2083 skb->len - csstart, 0);
2085 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2086 long csstuff = csstart + skb->csum_offset;
2088 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2091 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2094 * skb_dequeue - remove from the head of the queue
2095 * @list: list to dequeue from
2097 * Remove the head of the list. The list lock is taken so the function
2098 * may be used safely with other locking list functions. The head item is
2099 * returned or %NULL if the list is empty.
2102 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2104 unsigned long flags;
2105 struct sk_buff *result;
2107 spin_lock_irqsave(&list->lock, flags);
2108 result = __skb_dequeue(list);
2109 spin_unlock_irqrestore(&list->lock, flags);
2112 EXPORT_SYMBOL(skb_dequeue);
2115 * skb_dequeue_tail - remove from the tail of the queue
2116 * @list: list to dequeue from
2118 * Remove the tail of the list. The list lock is taken so the function
2119 * may be used safely with other locking list functions. The tail item is
2120 * returned or %NULL if the list is empty.
2122 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2124 unsigned long flags;
2125 struct sk_buff *result;
2127 spin_lock_irqsave(&list->lock, flags);
2128 result = __skb_dequeue_tail(list);
2129 spin_unlock_irqrestore(&list->lock, flags);
2132 EXPORT_SYMBOL(skb_dequeue_tail);
2135 * skb_queue_purge - empty a list
2136 * @list: list to empty
2138 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2139 * the list and one reference dropped. This function takes the list
2140 * lock and is atomic with respect to other list locking functions.
2142 void skb_queue_purge(struct sk_buff_head *list)
2144 struct sk_buff *skb;
2145 while ((skb = skb_dequeue(list)) != NULL)
2148 EXPORT_SYMBOL(skb_queue_purge);
2151 * skb_queue_head - queue a buffer at the list head
2152 * @list: list to use
2153 * @newsk: buffer to queue
2155 * Queue a buffer at the start of the list. This function takes the
2156 * list lock and can be used safely with other locking &sk_buff functions
2159 * A buffer cannot be placed on two lists at the same time.
2161 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2163 unsigned long flags;
2165 spin_lock_irqsave(&list->lock, flags);
2166 __skb_queue_head(list, newsk);
2167 spin_unlock_irqrestore(&list->lock, flags);
2169 EXPORT_SYMBOL(skb_queue_head);
2172 * skb_queue_tail - queue a buffer at the list tail
2173 * @list: list to use
2174 * @newsk: buffer to queue
2176 * Queue a buffer at the tail of the list. This function takes the
2177 * list lock and can be used safely with other locking &sk_buff functions
2180 * A buffer cannot be placed on two lists at the same time.
2182 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2184 unsigned long flags;
2186 spin_lock_irqsave(&list->lock, flags);
2187 __skb_queue_tail(list, newsk);
2188 spin_unlock_irqrestore(&list->lock, flags);
2190 EXPORT_SYMBOL(skb_queue_tail);
2193 * skb_unlink - remove a buffer from a list
2194 * @skb: buffer to remove
2195 * @list: list to use
2197 * Remove a packet from a list. The list locks are taken and this
2198 * function is atomic with respect to other list locked calls
2200 * You must know what list the SKB is on.
2202 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2204 unsigned long flags;
2206 spin_lock_irqsave(&list->lock, flags);
2207 __skb_unlink(skb, list);
2208 spin_unlock_irqrestore(&list->lock, flags);
2210 EXPORT_SYMBOL(skb_unlink);
2213 * skb_append - append a buffer
2214 * @old: buffer to insert after
2215 * @newsk: buffer to insert
2216 * @list: list to use
2218 * Place a packet after a given packet in a list. The list locks are taken
2219 * and this function is atomic with respect to other list locked calls.
2220 * A buffer cannot be placed on two lists at the same time.
2222 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2224 unsigned long flags;
2226 spin_lock_irqsave(&list->lock, flags);
2227 __skb_queue_after(list, old, newsk);
2228 spin_unlock_irqrestore(&list->lock, flags);
2230 EXPORT_SYMBOL(skb_append);
2233 * skb_insert - insert a buffer
2234 * @old: buffer to insert before
2235 * @newsk: buffer to insert
2236 * @list: list to use
2238 * Place a packet before a given packet in a list. The list locks are
2239 * taken and this function is atomic with respect to other list locked
2242 * A buffer cannot be placed on two lists at the same time.
2244 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2246 unsigned long flags;
2248 spin_lock_irqsave(&list->lock, flags);
2249 __skb_insert(newsk, old->prev, old, list);
2250 spin_unlock_irqrestore(&list->lock, flags);
2252 EXPORT_SYMBOL(skb_insert);
2254 static inline void skb_split_inside_header(struct sk_buff *skb,
2255 struct sk_buff* skb1,
2256 const u32 len, const int pos)
2260 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2262 /* And move data appendix as is. */
2263 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2264 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2266 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2267 skb_shinfo(skb)->nr_frags = 0;
2268 skb1->data_len = skb->data_len;
2269 skb1->len += skb1->data_len;
2272 skb_set_tail_pointer(skb, len);
2275 static inline void skb_split_no_header(struct sk_buff *skb,
2276 struct sk_buff* skb1,
2277 const u32 len, int pos)
2280 const int nfrags = skb_shinfo(skb)->nr_frags;
2282 skb_shinfo(skb)->nr_frags = 0;
2283 skb1->len = skb1->data_len = skb->len - len;
2285 skb->data_len = len - pos;
2287 for (i = 0; i < nfrags; i++) {
2288 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2290 if (pos + size > len) {
2291 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2295 * We have two variants in this case:
2296 * 1. Move all the frag to the second
2297 * part, if it is possible. F.e.
2298 * this approach is mandatory for TUX,
2299 * where splitting is expensive.
2300 * 2. Split is accurately. We make this.
2302 skb_frag_ref(skb, i);
2303 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2304 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2305 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2306 skb_shinfo(skb)->nr_frags++;
2310 skb_shinfo(skb)->nr_frags++;
2313 skb_shinfo(skb1)->nr_frags = k;
2317 * skb_split - Split fragmented skb to two parts at length len.
2318 * @skb: the buffer to split
2319 * @skb1: the buffer to receive the second part
2320 * @len: new length for skb
2322 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2324 int pos = skb_headlen(skb);
2326 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2327 if (len < pos) /* Split line is inside header. */
2328 skb_split_inside_header(skb, skb1, len, pos);
2329 else /* Second chunk has no header, nothing to copy. */
2330 skb_split_no_header(skb, skb1, len, pos);
2332 EXPORT_SYMBOL(skb_split);
2334 /* Shifting from/to a cloned skb is a no-go.
2336 * Caller cannot keep skb_shinfo related pointers past calling here!
2338 static int skb_prepare_for_shift(struct sk_buff *skb)
2340 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2344 * skb_shift - Shifts paged data partially from skb to another
2345 * @tgt: buffer into which tail data gets added
2346 * @skb: buffer from which the paged data comes from
2347 * @shiftlen: shift up to this many bytes
2349 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2350 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2351 * It's up to caller to free skb if everything was shifted.
2353 * If @tgt runs out of frags, the whole operation is aborted.
2355 * Skb cannot include anything else but paged data while tgt is allowed
2356 * to have non-paged data as well.
2358 * TODO: full sized shift could be optimized but that would need
2359 * specialized skb free'er to handle frags without up-to-date nr_frags.
2361 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2363 int from, to, merge, todo;
2364 struct skb_frag_struct *fragfrom, *fragto;
2366 BUG_ON(shiftlen > skb->len);
2367 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2371 to = skb_shinfo(tgt)->nr_frags;
2372 fragfrom = &skb_shinfo(skb)->frags[from];
2374 /* Actual merge is delayed until the point when we know we can
2375 * commit all, so that we don't have to undo partial changes
2378 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2379 fragfrom->page_offset)) {
2384 todo -= skb_frag_size(fragfrom);
2386 if (skb_prepare_for_shift(skb) ||
2387 skb_prepare_for_shift(tgt))
2390 /* All previous frag pointers might be stale! */
2391 fragfrom = &skb_shinfo(skb)->frags[from];
2392 fragto = &skb_shinfo(tgt)->frags[merge];
2394 skb_frag_size_add(fragto, shiftlen);
2395 skb_frag_size_sub(fragfrom, shiftlen);
2396 fragfrom->page_offset += shiftlen;
2404 /* Skip full, not-fitting skb to avoid expensive operations */
2405 if ((shiftlen == skb->len) &&
2406 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2409 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2412 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2413 if (to == MAX_SKB_FRAGS)
2416 fragfrom = &skb_shinfo(skb)->frags[from];
2417 fragto = &skb_shinfo(tgt)->frags[to];
2419 if (todo >= skb_frag_size(fragfrom)) {
2420 *fragto = *fragfrom;
2421 todo -= skb_frag_size(fragfrom);
2426 __skb_frag_ref(fragfrom);
2427 fragto->page = fragfrom->page;
2428 fragto->page_offset = fragfrom->page_offset;
2429 skb_frag_size_set(fragto, todo);
2431 fragfrom->page_offset += todo;
2432 skb_frag_size_sub(fragfrom, todo);
2440 /* Ready to "commit" this state change to tgt */
2441 skb_shinfo(tgt)->nr_frags = to;
2444 fragfrom = &skb_shinfo(skb)->frags[0];
2445 fragto = &skb_shinfo(tgt)->frags[merge];
2447 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2448 __skb_frag_unref(fragfrom);
2451 /* Reposition in the original skb */
2453 while (from < skb_shinfo(skb)->nr_frags)
2454 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2455 skb_shinfo(skb)->nr_frags = to;
2457 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2460 /* Most likely the tgt won't ever need its checksum anymore, skb on
2461 * the other hand might need it if it needs to be resent
2463 tgt->ip_summed = CHECKSUM_PARTIAL;
2464 skb->ip_summed = CHECKSUM_PARTIAL;
2466 /* Yak, is it really working this way? Some helper please? */
2467 skb->len -= shiftlen;
2468 skb->data_len -= shiftlen;
2469 skb->truesize -= shiftlen;
2470 tgt->len += shiftlen;
2471 tgt->data_len += shiftlen;
2472 tgt->truesize += shiftlen;
2478 * skb_prepare_seq_read - Prepare a sequential read of skb data
2479 * @skb: the buffer to read
2480 * @from: lower offset of data to be read
2481 * @to: upper offset of data to be read
2482 * @st: state variable
2484 * Initializes the specified state variable. Must be called before
2485 * invoking skb_seq_read() for the first time.
2487 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2488 unsigned int to, struct skb_seq_state *st)
2490 st->lower_offset = from;
2491 st->upper_offset = to;
2492 st->root_skb = st->cur_skb = skb;
2493 st->frag_idx = st->stepped_offset = 0;
2494 st->frag_data = NULL;
2496 EXPORT_SYMBOL(skb_prepare_seq_read);
2499 * skb_seq_read - Sequentially read skb data
2500 * @consumed: number of bytes consumed by the caller so far
2501 * @data: destination pointer for data to be returned
2502 * @st: state variable
2504 * Reads a block of skb data at &consumed relative to the
2505 * lower offset specified to skb_prepare_seq_read(). Assigns
2506 * the head of the data block to &data and returns the length
2507 * of the block or 0 if the end of the skb data or the upper
2508 * offset has been reached.
2510 * The caller is not required to consume all of the data
2511 * returned, i.e. &consumed is typically set to the number
2512 * of bytes already consumed and the next call to
2513 * skb_seq_read() will return the remaining part of the block.
2515 * Note 1: The size of each block of data returned can be arbitrary,
2516 * this limitation is the cost for zerocopy seqeuental
2517 * reads of potentially non linear data.
2519 * Note 2: Fragment lists within fragments are not implemented
2520 * at the moment, state->root_skb could be replaced with
2521 * a stack for this purpose.
2523 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2524 struct skb_seq_state *st)
2526 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2529 if (unlikely(abs_offset >= st->upper_offset))
2533 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2535 if (abs_offset < block_limit && !st->frag_data) {
2536 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2537 return block_limit - abs_offset;
2540 if (st->frag_idx == 0 && !st->frag_data)
2541 st->stepped_offset += skb_headlen(st->cur_skb);
2543 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2544 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2545 block_limit = skb_frag_size(frag) + st->stepped_offset;
2547 if (abs_offset < block_limit) {
2549 st->frag_data = kmap_atomic(skb_frag_page(frag));
2551 *data = (u8 *) st->frag_data + frag->page_offset +
2552 (abs_offset - st->stepped_offset);
2554 return block_limit - abs_offset;
2557 if (st->frag_data) {
2558 kunmap_atomic(st->frag_data);
2559 st->frag_data = NULL;
2563 st->stepped_offset += skb_frag_size(frag);
2566 if (st->frag_data) {
2567 kunmap_atomic(st->frag_data);
2568 st->frag_data = NULL;
2571 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2572 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2575 } else if (st->cur_skb->next) {
2576 st->cur_skb = st->cur_skb->next;
2583 EXPORT_SYMBOL(skb_seq_read);
2586 * skb_abort_seq_read - Abort a sequential read of skb data
2587 * @st: state variable
2589 * Must be called if skb_seq_read() was not called until it
2592 void skb_abort_seq_read(struct skb_seq_state *st)
2595 kunmap_atomic(st->frag_data);
2597 EXPORT_SYMBOL(skb_abort_seq_read);
2599 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2601 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2602 struct ts_config *conf,
2603 struct ts_state *state)
2605 return skb_seq_read(offset, text, TS_SKB_CB(state));
2608 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2610 skb_abort_seq_read(TS_SKB_CB(state));
2614 * skb_find_text - Find a text pattern in skb data
2615 * @skb: the buffer to look in
2616 * @from: search offset
2618 * @config: textsearch configuration
2619 * @state: uninitialized textsearch state variable
2621 * Finds a pattern in the skb data according to the specified
2622 * textsearch configuration. Use textsearch_next() to retrieve
2623 * subsequent occurrences of the pattern. Returns the offset
2624 * to the first occurrence or UINT_MAX if no match was found.
2626 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2627 unsigned int to, struct ts_config *config,
2628 struct ts_state *state)
2632 config->get_next_block = skb_ts_get_next_block;
2633 config->finish = skb_ts_finish;
2635 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2637 ret = textsearch_find(config, state);
2638 return (ret <= to - from ? ret : UINT_MAX);
2640 EXPORT_SYMBOL(skb_find_text);
2643 * skb_append_datato_frags - append the user data to a skb
2644 * @sk: sock structure
2645 * @skb: skb structure to be appened with user data.
2646 * @getfrag: call back function to be used for getting the user data
2647 * @from: pointer to user message iov
2648 * @length: length of the iov message
2650 * Description: This procedure append the user data in the fragment part
2651 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2653 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2654 int (*getfrag)(void *from, char *to, int offset,
2655 int len, int odd, struct sk_buff *skb),
2656 void *from, int length)
2658 int frg_cnt = skb_shinfo(skb)->nr_frags;
2662 struct page_frag *pfrag = ¤t->task_frag;
2665 /* Return error if we don't have space for new frag */
2666 if (frg_cnt >= MAX_SKB_FRAGS)
2669 if (!sk_page_frag_refill(sk, pfrag))
2672 /* copy the user data to page */
2673 copy = min_t(int, length, pfrag->size - pfrag->offset);
2675 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2676 offset, copy, 0, skb);
2680 /* copy was successful so update the size parameters */
2681 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2684 pfrag->offset += copy;
2685 get_page(pfrag->page);
2687 skb->truesize += copy;
2688 atomic_add(copy, &sk->sk_wmem_alloc);
2690 skb->data_len += copy;
2694 } while (length > 0);
2698 EXPORT_SYMBOL(skb_append_datato_frags);
2701 * skb_pull_rcsum - pull skb and update receive checksum
2702 * @skb: buffer to update
2703 * @len: length of data pulled
2705 * This function performs an skb_pull on the packet and updates
2706 * the CHECKSUM_COMPLETE checksum. It should be used on
2707 * receive path processing instead of skb_pull unless you know
2708 * that the checksum difference is zero (e.g., a valid IP header)
2709 * or you are setting ip_summed to CHECKSUM_NONE.
2711 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2713 BUG_ON(len > skb->len);
2715 BUG_ON(skb->len < skb->data_len);
2716 skb_postpull_rcsum(skb, skb->data, len);
2717 return skb->data += len;
2719 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2722 * skb_segment - Perform protocol segmentation on skb.
2723 * @skb: buffer to segment
2724 * @features: features for the output path (see dev->features)
2726 * This function performs segmentation on the given skb. It returns
2727 * a pointer to the first in a list of new skbs for the segments.
2728 * In case of error it returns ERR_PTR(err).
2730 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2732 struct sk_buff *segs = NULL;
2733 struct sk_buff *tail = NULL;
2734 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2735 unsigned int mss = skb_shinfo(skb)->gso_size;
2736 unsigned int doffset = skb->data - skb_mac_header(skb);
2737 unsigned int offset = doffset;
2738 unsigned int tnl_hlen = skb_tnl_header_len(skb);
2739 unsigned int headroom;
2743 int sg = !!(features & NETIF_F_SG);
2744 int nfrags = skb_shinfo(skb)->nr_frags;
2749 proto = skb_network_protocol(skb);
2750 if (unlikely(!proto))
2751 return ERR_PTR(-EINVAL);
2753 csum = !!can_checksum_protocol(features, proto);
2754 __skb_push(skb, doffset);
2755 headroom = skb_headroom(skb);
2756 pos = skb_headlen(skb);
2759 struct sk_buff *nskb;
2764 len = skb->len - offset;
2768 hsize = skb_headlen(skb) - offset;
2771 if (hsize > len || !sg)
2774 if (!hsize && i >= nfrags) {
2775 BUG_ON(fskb->len != len);
2778 nskb = skb_clone(fskb, GFP_ATOMIC);
2781 if (unlikely(!nskb))
2784 hsize = skb_end_offset(nskb);
2785 if (skb_cow_head(nskb, doffset + headroom)) {
2790 nskb->truesize += skb_end_offset(nskb) - hsize;
2791 skb_release_head_state(nskb);
2792 __skb_push(nskb, doffset);
2794 nskb = __alloc_skb(hsize + doffset + headroom,
2795 GFP_ATOMIC, skb_alloc_rx_flag(skb),
2798 if (unlikely(!nskb))
2801 skb_reserve(nskb, headroom);
2802 __skb_put(nskb, doffset);
2811 __copy_skb_header(nskb, skb);
2812 nskb->mac_len = skb->mac_len;
2814 /* nskb and skb might have different headroom */
2815 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2816 nskb->csum_start += skb_headroom(nskb) - headroom;
2818 skb_reset_mac_header(nskb);
2819 skb_set_network_header(nskb, skb->mac_len);
2820 nskb->transport_header = (nskb->network_header +
2821 skb_network_header_len(skb));
2823 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
2824 nskb->data - tnl_hlen,
2825 doffset + tnl_hlen);
2827 if (fskb != skb_shinfo(skb)->frag_list)
2831 nskb->ip_summed = CHECKSUM_NONE;
2832 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2838 frag = skb_shinfo(nskb)->frags;
2840 skb_copy_from_linear_data_offset(skb, offset,
2841 skb_put(nskb, hsize), hsize);
2843 skb_shinfo(nskb)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2845 while (pos < offset + len && i < nfrags) {
2846 *frag = skb_shinfo(skb)->frags[i];
2847 __skb_frag_ref(frag);
2848 size = skb_frag_size(frag);
2851 frag->page_offset += offset - pos;
2852 skb_frag_size_sub(frag, offset - pos);
2855 skb_shinfo(nskb)->nr_frags++;
2857 if (pos + size <= offset + len) {
2861 skb_frag_size_sub(frag, pos + size - (offset + len));
2868 if (pos < offset + len) {
2869 struct sk_buff *fskb2 = fskb;
2871 BUG_ON(pos + fskb->len != offset + len);
2877 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2883 SKB_FRAG_ASSERT(nskb);
2884 skb_shinfo(nskb)->frag_list = fskb2;
2888 nskb->data_len = len - hsize;
2889 nskb->len += nskb->data_len;
2890 nskb->truesize += nskb->data_len;
2893 nskb->csum = skb_checksum(nskb, doffset,
2894 nskb->len - doffset, 0);
2895 nskb->ip_summed = CHECKSUM_NONE;
2897 } while ((offset += len) < skb->len);
2902 while ((skb = segs)) {
2906 return ERR_PTR(err);
2908 EXPORT_SYMBOL_GPL(skb_segment);
2910 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2912 struct sk_buff *p = *head;
2913 struct sk_buff *nskb;
2914 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2915 struct skb_shared_info *pinfo = skb_shinfo(p);
2916 unsigned int headroom;
2917 unsigned int len = skb_gro_len(skb);
2918 unsigned int offset = skb_gro_offset(skb);
2919 unsigned int headlen = skb_headlen(skb);
2920 unsigned int delta_truesize;
2922 if (p->len + len >= 65536)
2925 if (pinfo->frag_list)
2927 else if (headlen <= offset) {
2930 int i = skbinfo->nr_frags;
2931 int nr_frags = pinfo->nr_frags + i;
2935 if (nr_frags > MAX_SKB_FRAGS)
2938 pinfo->nr_frags = nr_frags;
2939 skbinfo->nr_frags = 0;
2941 frag = pinfo->frags + nr_frags;
2942 frag2 = skbinfo->frags + i;
2947 frag->page_offset += offset;
2948 skb_frag_size_sub(frag, offset);
2950 /* all fragments truesize : remove (head size + sk_buff) */
2951 delta_truesize = skb->truesize -
2952 SKB_TRUESIZE(skb_end_offset(skb));
2954 skb->truesize -= skb->data_len;
2955 skb->len -= skb->data_len;
2958 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
2960 } else if (skb->head_frag) {
2961 int nr_frags = pinfo->nr_frags;
2962 skb_frag_t *frag = pinfo->frags + nr_frags;
2963 struct page *page = virt_to_head_page(skb->head);
2964 unsigned int first_size = headlen - offset;
2965 unsigned int first_offset;
2967 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
2970 first_offset = skb->data -
2971 (unsigned char *)page_address(page) +
2974 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
2976 frag->page.p = page;
2977 frag->page_offset = first_offset;
2978 skb_frag_size_set(frag, first_size);
2980 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
2981 /* We dont need to clear skbinfo->nr_frags here */
2983 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
2984 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
2986 } else if (skb_gro_len(p) != pinfo->gso_size)
2989 headroom = skb_headroom(p);
2990 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2991 if (unlikely(!nskb))
2994 __copy_skb_header(nskb, p);
2995 nskb->mac_len = p->mac_len;
2997 skb_reserve(nskb, headroom);
2998 __skb_put(nskb, skb_gro_offset(p));
3000 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3001 skb_set_network_header(nskb, skb_network_offset(p));
3002 skb_set_transport_header(nskb, skb_transport_offset(p));
3004 __skb_pull(p, skb_gro_offset(p));
3005 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3006 p->data - skb_mac_header(p));
3008 skb_shinfo(nskb)->frag_list = p;
3009 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3010 pinfo->gso_size = 0;
3011 skb_header_release(p);
3012 NAPI_GRO_CB(nskb)->last = p;
3014 nskb->data_len += p->len;
3015 nskb->truesize += p->truesize;
3016 nskb->len += p->len;
3019 nskb->next = p->next;
3025 delta_truesize = skb->truesize;
3026 if (offset > headlen) {
3027 unsigned int eat = offset - headlen;
3029 skbinfo->frags[0].page_offset += eat;
3030 skb_frag_size_sub(&skbinfo->frags[0], eat);
3031 skb->data_len -= eat;
3036 __skb_pull(skb, offset);
3038 NAPI_GRO_CB(p)->last->next = skb;
3039 NAPI_GRO_CB(p)->last = skb;
3040 skb_header_release(skb);
3043 NAPI_GRO_CB(p)->count++;
3045 p->truesize += delta_truesize;
3048 NAPI_GRO_CB(skb)->same_flow = 1;
3051 EXPORT_SYMBOL_GPL(skb_gro_receive);
3053 void __init skb_init(void)
3055 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3056 sizeof(struct sk_buff),
3058 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3060 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3061 (2*sizeof(struct sk_buff)) +
3064 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3069 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3070 * @skb: Socket buffer containing the buffers to be mapped
3071 * @sg: The scatter-gather list to map into
3072 * @offset: The offset into the buffer's contents to start mapping
3073 * @len: Length of buffer space to be mapped
3075 * Fill the specified scatter-gather list with mappings/pointers into a
3076 * region of the buffer space attached to a socket buffer.
3079 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3081 int start = skb_headlen(skb);
3082 int i, copy = start - offset;
3083 struct sk_buff *frag_iter;
3089 sg_set_buf(sg, skb->data + offset, copy);
3091 if ((len -= copy) == 0)
3096 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3099 WARN_ON(start > offset + len);
3101 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3102 if ((copy = end - offset) > 0) {
3103 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3107 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3108 frag->page_offset+offset-start);
3117 skb_walk_frags(skb, frag_iter) {
3120 WARN_ON(start > offset + len);
3122 end = start + frag_iter->len;
3123 if ((copy = end - offset) > 0) {
3126 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3128 if ((len -= copy) == 0)
3138 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3140 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3142 sg_mark_end(&sg[nsg - 1]);
3146 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3149 * skb_cow_data - Check that a socket buffer's data buffers are writable
3150 * @skb: The socket buffer to check.
3151 * @tailbits: Amount of trailing space to be added
3152 * @trailer: Returned pointer to the skb where the @tailbits space begins
3154 * Make sure that the data buffers attached to a socket buffer are
3155 * writable. If they are not, private copies are made of the data buffers
3156 * and the socket buffer is set to use these instead.
3158 * If @tailbits is given, make sure that there is space to write @tailbits
3159 * bytes of data beyond current end of socket buffer. @trailer will be
3160 * set to point to the skb in which this space begins.
3162 * The number of scatterlist elements required to completely map the
3163 * COW'd and extended socket buffer will be returned.
3165 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3169 struct sk_buff *skb1, **skb_p;
3171 /* If skb is cloned or its head is paged, reallocate
3172 * head pulling out all the pages (pages are considered not writable
3173 * at the moment even if they are anonymous).
3175 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3176 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3179 /* Easy case. Most of packets will go this way. */
3180 if (!skb_has_frag_list(skb)) {
3181 /* A little of trouble, not enough of space for trailer.
3182 * This should not happen, when stack is tuned to generate
3183 * good frames. OK, on miss we reallocate and reserve even more
3184 * space, 128 bytes is fair. */
3186 if (skb_tailroom(skb) < tailbits &&
3187 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3195 /* Misery. We are in troubles, going to mincer fragments... */
3198 skb_p = &skb_shinfo(skb)->frag_list;
3201 while ((skb1 = *skb_p) != NULL) {
3204 /* The fragment is partially pulled by someone,
3205 * this can happen on input. Copy it and everything
3208 if (skb_shared(skb1))
3211 /* If the skb is the last, worry about trailer. */
3213 if (skb1->next == NULL && tailbits) {
3214 if (skb_shinfo(skb1)->nr_frags ||
3215 skb_has_frag_list(skb1) ||
3216 skb_tailroom(skb1) < tailbits)
3217 ntail = tailbits + 128;
3223 skb_shinfo(skb1)->nr_frags ||
3224 skb_has_frag_list(skb1)) {
3225 struct sk_buff *skb2;
3227 /* Fuck, we are miserable poor guys... */
3229 skb2 = skb_copy(skb1, GFP_ATOMIC);
3231 skb2 = skb_copy_expand(skb1,
3235 if (unlikely(skb2 == NULL))
3239 skb_set_owner_w(skb2, skb1->sk);
3241 /* Looking around. Are we still alive?
3242 * OK, link new skb, drop old one */
3244 skb2->next = skb1->next;
3251 skb_p = &skb1->next;
3256 EXPORT_SYMBOL_GPL(skb_cow_data);
3258 static void sock_rmem_free(struct sk_buff *skb)
3260 struct sock *sk = skb->sk;
3262 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3266 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3268 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3272 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3273 (unsigned int)sk->sk_rcvbuf)
3278 skb->destructor = sock_rmem_free;
3279 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3281 /* before exiting rcu section, make sure dst is refcounted */
3284 skb_queue_tail(&sk->sk_error_queue, skb);
3285 if (!sock_flag(sk, SOCK_DEAD))
3286 sk->sk_data_ready(sk, len);
3289 EXPORT_SYMBOL(sock_queue_err_skb);
3291 void skb_tstamp_tx(struct sk_buff *orig_skb,
3292 struct skb_shared_hwtstamps *hwtstamps)
3294 struct sock *sk = orig_skb->sk;
3295 struct sock_exterr_skb *serr;
3296 struct sk_buff *skb;
3302 skb = skb_clone(orig_skb, GFP_ATOMIC);
3307 *skb_hwtstamps(skb) =
3311 * no hardware time stamps available,
3312 * so keep the shared tx_flags and only
3313 * store software time stamp
3315 skb->tstamp = ktime_get_real();
3318 serr = SKB_EXT_ERR(skb);
3319 memset(serr, 0, sizeof(*serr));
3320 serr->ee.ee_errno = ENOMSG;
3321 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3323 err = sock_queue_err_skb(sk, skb);
3328 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3330 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3332 struct sock *sk = skb->sk;
3333 struct sock_exterr_skb *serr;
3336 skb->wifi_acked_valid = 1;
3337 skb->wifi_acked = acked;
3339 serr = SKB_EXT_ERR(skb);
3340 memset(serr, 0, sizeof(*serr));
3341 serr->ee.ee_errno = ENOMSG;
3342 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3344 err = sock_queue_err_skb(sk, skb);
3348 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3352 * skb_partial_csum_set - set up and verify partial csum values for packet
3353 * @skb: the skb to set
3354 * @start: the number of bytes after skb->data to start checksumming.
3355 * @off: the offset from start to place the checksum.
3357 * For untrusted partially-checksummed packets, we need to make sure the values
3358 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3360 * This function checks and sets those values and skb->ip_summed: if this
3361 * returns false you should drop the packet.
3363 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3365 if (unlikely(start > skb_headlen(skb)) ||
3366 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3367 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3368 start, off, skb_headlen(skb));
3371 skb->ip_summed = CHECKSUM_PARTIAL;
3372 skb->csum_start = skb_headroom(skb) + start;
3373 skb->csum_offset = off;
3374 skb_set_transport_header(skb, start);
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);