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_tci = old->vlan_tci;
712 skb_copy_secmark(new, old);
716 * You should not add any new code to this function. Add it to
717 * __copy_skb_header above instead.
719 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
721 #define C(x) n->x = skb->x
723 n->next = n->prev = NULL;
725 __copy_skb_header(n, skb);
730 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
733 n->destructor = NULL;
740 atomic_set(&n->users, 1);
742 atomic_inc(&(skb_shinfo(skb)->dataref));
750 * skb_morph - morph one skb into another
751 * @dst: the skb to receive the contents
752 * @src: the skb to supply the contents
754 * This is identical to skb_clone except that the target skb is
755 * supplied by the user.
757 * The target skb is returned upon exit.
759 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
761 skb_release_all(dst);
762 return __skb_clone(dst, src);
764 EXPORT_SYMBOL_GPL(skb_morph);
767 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
768 * @skb: the skb to modify
769 * @gfp_mask: allocation priority
771 * This must be called on SKBTX_DEV_ZEROCOPY skb.
772 * It will copy all frags into kernel and drop the reference
773 * to userspace pages.
775 * If this function is called from an interrupt gfp_mask() must be
778 * Returns 0 on success or a negative error code on failure
779 * to allocate kernel memory to copy to.
781 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
784 int num_frags = skb_shinfo(skb)->nr_frags;
785 struct page *page, *head = NULL;
786 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
788 for (i = 0; i < num_frags; i++) {
790 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
792 page = alloc_page(gfp_mask);
795 struct page *next = (struct page *)head->private;
801 vaddr = kmap_atomic(skb_frag_page(f));
802 memcpy(page_address(page),
803 vaddr + f->page_offset, skb_frag_size(f));
804 kunmap_atomic(vaddr);
805 page->private = (unsigned long)head;
809 /* skb frags release userspace buffers */
810 for (i = 0; i < num_frags; i++)
811 skb_frag_unref(skb, i);
813 uarg->callback(uarg, false);
815 /* skb frags point to kernel buffers */
816 for (i = num_frags - 1; i >= 0; i--) {
817 __skb_fill_page_desc(skb, i, head, 0,
818 skb_shinfo(skb)->frags[i].size);
819 head = (struct page *)head->private;
822 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
825 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
828 * skb_clone - duplicate an sk_buff
829 * @skb: buffer to clone
830 * @gfp_mask: allocation priority
832 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
833 * copies share the same packet data but not structure. The new
834 * buffer has a reference count of 1. If the allocation fails the
835 * function returns %NULL otherwise the new buffer is returned.
837 * If this function is called from an interrupt gfp_mask() must be
841 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
845 if (skb_orphan_frags(skb, gfp_mask))
849 if (skb->fclone == SKB_FCLONE_ORIG &&
850 n->fclone == SKB_FCLONE_UNAVAILABLE) {
851 atomic_t *fclone_ref = (atomic_t *) (n + 1);
852 n->fclone = SKB_FCLONE_CLONE;
853 atomic_inc(fclone_ref);
855 if (skb_pfmemalloc(skb))
856 gfp_mask |= __GFP_MEMALLOC;
858 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
862 kmemcheck_annotate_bitfield(n, flags1);
863 kmemcheck_annotate_bitfield(n, flags2);
864 n->fclone = SKB_FCLONE_UNAVAILABLE;
867 return __skb_clone(n, skb);
869 EXPORT_SYMBOL(skb_clone);
871 static void skb_headers_offset_update(struct sk_buff *skb, int off)
873 /* {transport,network,mac}_header and tail are relative to skb->head */
874 skb->transport_header += off;
875 skb->network_header += off;
876 if (skb_mac_header_was_set(skb))
877 skb->mac_header += off;
878 skb->inner_transport_header += off;
879 skb->inner_network_header += off;
880 skb->inner_mac_header += off;
883 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
885 #ifndef NET_SKBUFF_DATA_USES_OFFSET
887 * Shift between the two data areas in bytes
889 unsigned long offset = new->data - old->data;
892 __copy_skb_header(new, old);
894 #ifndef NET_SKBUFF_DATA_USES_OFFSET
895 skb_headers_offset_update(new, offset);
897 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
898 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
899 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
902 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
904 if (skb_pfmemalloc(skb))
910 * skb_copy - create private copy of an sk_buff
911 * @skb: buffer to copy
912 * @gfp_mask: allocation priority
914 * Make a copy of both an &sk_buff and its data. This is used when the
915 * caller wishes to modify the data and needs a private copy of the
916 * data to alter. Returns %NULL on failure or the pointer to the buffer
917 * on success. The returned buffer has a reference count of 1.
919 * As by-product this function converts non-linear &sk_buff to linear
920 * one, so that &sk_buff becomes completely private and caller is allowed
921 * to modify all the data of returned buffer. This means that this
922 * function is not recommended for use in circumstances when only
923 * header is going to be modified. Use pskb_copy() instead.
926 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
928 int headerlen = skb_headroom(skb);
929 unsigned int size = skb_end_offset(skb) + skb->data_len;
930 struct sk_buff *n = __alloc_skb(size, gfp_mask,
931 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
936 /* Set the data pointer */
937 skb_reserve(n, headerlen);
938 /* Set the tail pointer and length */
939 skb_put(n, skb->len);
941 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
944 copy_skb_header(n, skb);
947 EXPORT_SYMBOL(skb_copy);
950 * __pskb_copy - create copy of an sk_buff with private head.
951 * @skb: buffer to copy
952 * @headroom: headroom of new skb
953 * @gfp_mask: allocation priority
955 * Make a copy of both an &sk_buff and part of its data, located
956 * in header. Fragmented data remain shared. This is used when
957 * the caller wishes to modify only header of &sk_buff and needs
958 * private copy of the header to alter. Returns %NULL on failure
959 * or the pointer to the buffer on success.
960 * The returned buffer has a reference count of 1.
963 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
965 unsigned int size = skb_headlen(skb) + headroom;
966 struct sk_buff *n = __alloc_skb(size, gfp_mask,
967 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
972 /* Set the data pointer */
973 skb_reserve(n, headroom);
974 /* Set the tail pointer and length */
975 skb_put(n, skb_headlen(skb));
977 skb_copy_from_linear_data(skb, n->data, n->len);
979 n->truesize += skb->data_len;
980 n->data_len = skb->data_len;
983 if (skb_shinfo(skb)->nr_frags) {
986 if (skb_orphan_frags(skb, gfp_mask)) {
991 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
992 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
993 skb_frag_ref(skb, i);
995 skb_shinfo(n)->nr_frags = i;
998 if (skb_has_frag_list(skb)) {
999 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1000 skb_clone_fraglist(n);
1003 copy_skb_header(n, skb);
1007 EXPORT_SYMBOL(__pskb_copy);
1010 * pskb_expand_head - reallocate header of &sk_buff
1011 * @skb: buffer to reallocate
1012 * @nhead: room to add at head
1013 * @ntail: room to add at tail
1014 * @gfp_mask: allocation priority
1016 * Expands (or creates identical copy, if &nhead and &ntail are zero)
1017 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
1018 * reference count of 1. Returns zero in the case of success or error,
1019 * if expansion failed. In the last case, &sk_buff is not changed.
1021 * All the pointers pointing into skb header may change and must be
1022 * reloaded after call to this function.
1025 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1030 int size = nhead + skb_end_offset(skb) + ntail;
1035 if (skb_shared(skb))
1038 size = SKB_DATA_ALIGN(size);
1040 if (skb_pfmemalloc(skb))
1041 gfp_mask |= __GFP_MEMALLOC;
1042 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1043 gfp_mask, NUMA_NO_NODE, NULL);
1046 size = SKB_WITH_OVERHEAD(ksize(data));
1048 /* Copy only real data... and, alas, header. This should be
1049 * optimized for the cases when header is void.
1051 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1053 memcpy((struct skb_shared_info *)(data + size),
1055 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1058 * if shinfo is shared we must drop the old head gracefully, but if it
1059 * is not we can just drop the old head and let the existing refcount
1060 * be since all we did is relocate the values
1062 if (skb_cloned(skb)) {
1063 /* copy this zero copy skb frags */
1064 if (skb_orphan_frags(skb, gfp_mask))
1066 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1067 skb_frag_ref(skb, i);
1069 if (skb_has_frag_list(skb))
1070 skb_clone_fraglist(skb);
1072 skb_release_data(skb);
1076 off = (data + nhead) - skb->head;
1081 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1085 skb->end = skb->head + size;
1088 skb_headers_offset_update(skb, off);
1089 /* Only adjust this if it actually is csum_start rather than csum */
1090 if (skb->ip_summed == CHECKSUM_PARTIAL)
1091 skb->csum_start += nhead;
1095 atomic_set(&skb_shinfo(skb)->dataref, 1);
1103 EXPORT_SYMBOL(pskb_expand_head);
1105 /* Make private copy of skb with writable head and some headroom */
1107 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1109 struct sk_buff *skb2;
1110 int delta = headroom - skb_headroom(skb);
1113 skb2 = pskb_copy(skb, GFP_ATOMIC);
1115 skb2 = skb_clone(skb, GFP_ATOMIC);
1116 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1124 EXPORT_SYMBOL(skb_realloc_headroom);
1127 * skb_copy_expand - copy and expand sk_buff
1128 * @skb: buffer to copy
1129 * @newheadroom: new free bytes at head
1130 * @newtailroom: new free bytes at tail
1131 * @gfp_mask: allocation priority
1133 * Make a copy of both an &sk_buff and its data and while doing so
1134 * allocate additional space.
1136 * This is used when the caller wishes to modify the data and needs a
1137 * private copy of the data to alter as well as more space for new fields.
1138 * Returns %NULL on failure or the pointer to the buffer
1139 * on success. The returned buffer has a reference count of 1.
1141 * You must pass %GFP_ATOMIC as the allocation priority if this function
1142 * is called from an interrupt.
1144 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1145 int newheadroom, int newtailroom,
1149 * Allocate the copy buffer
1151 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1152 gfp_mask, skb_alloc_rx_flag(skb),
1154 int oldheadroom = skb_headroom(skb);
1155 int head_copy_len, head_copy_off;
1161 skb_reserve(n, newheadroom);
1163 /* Set the tail pointer and length */
1164 skb_put(n, skb->len);
1166 head_copy_len = oldheadroom;
1168 if (newheadroom <= head_copy_len)
1169 head_copy_len = newheadroom;
1171 head_copy_off = newheadroom - head_copy_len;
1173 /* Copy the linear header and data. */
1174 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1175 skb->len + head_copy_len))
1178 copy_skb_header(n, skb);
1180 off = newheadroom - oldheadroom;
1181 if (n->ip_summed == CHECKSUM_PARTIAL)
1182 n->csum_start += off;
1183 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1184 skb_headers_offset_update(n, off);
1189 EXPORT_SYMBOL(skb_copy_expand);
1192 * skb_pad - zero pad the tail of an skb
1193 * @skb: buffer to pad
1194 * @pad: space to pad
1196 * Ensure that a buffer is followed by a padding area that is zero
1197 * filled. Used by network drivers which may DMA or transfer data
1198 * beyond the buffer end onto the wire.
1200 * May return error in out of memory cases. The skb is freed on error.
1203 int skb_pad(struct sk_buff *skb, int pad)
1208 /* If the skbuff is non linear tailroom is always zero.. */
1209 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1210 memset(skb->data+skb->len, 0, pad);
1214 ntail = skb->data_len + pad - (skb->end - skb->tail);
1215 if (likely(skb_cloned(skb) || ntail > 0)) {
1216 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1221 /* FIXME: The use of this function with non-linear skb's really needs
1224 err = skb_linearize(skb);
1228 memset(skb->data + skb->len, 0, pad);
1235 EXPORT_SYMBOL(skb_pad);
1238 * skb_put - add data to a buffer
1239 * @skb: buffer to use
1240 * @len: amount of data to add
1242 * This function extends the used data area of the buffer. If this would
1243 * exceed the total buffer size the kernel will panic. A pointer to the
1244 * first byte of the extra data is returned.
1246 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1248 unsigned char *tmp = skb_tail_pointer(skb);
1249 SKB_LINEAR_ASSERT(skb);
1252 if (unlikely(skb->tail > skb->end))
1253 skb_over_panic(skb, len, __builtin_return_address(0));
1256 EXPORT_SYMBOL(skb_put);
1259 * skb_push - add data to the start of a buffer
1260 * @skb: buffer to use
1261 * @len: amount of data to add
1263 * This function extends the used data area of the buffer at the buffer
1264 * start. If this would exceed the total buffer headroom the kernel will
1265 * panic. A pointer to the first byte of the extra data is returned.
1267 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1271 if (unlikely(skb->data<skb->head))
1272 skb_under_panic(skb, len, __builtin_return_address(0));
1275 EXPORT_SYMBOL(skb_push);
1278 * skb_pull - remove data from the start of a buffer
1279 * @skb: buffer to use
1280 * @len: amount of data to remove
1282 * This function removes data from the start of a buffer, returning
1283 * the memory to the headroom. A pointer to the next data in the buffer
1284 * is returned. Once the data has been pulled future pushes will overwrite
1287 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1289 return skb_pull_inline(skb, len);
1291 EXPORT_SYMBOL(skb_pull);
1294 * skb_trim - remove end from a buffer
1295 * @skb: buffer to alter
1298 * Cut the length of a buffer down by removing data from the tail. If
1299 * the buffer is already under the length specified it is not modified.
1300 * The skb must be linear.
1302 void skb_trim(struct sk_buff *skb, unsigned int len)
1305 __skb_trim(skb, len);
1307 EXPORT_SYMBOL(skb_trim);
1309 /* Trims skb to length len. It can change skb pointers.
1312 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1314 struct sk_buff **fragp;
1315 struct sk_buff *frag;
1316 int offset = skb_headlen(skb);
1317 int nfrags = skb_shinfo(skb)->nr_frags;
1321 if (skb_cloned(skb) &&
1322 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1329 for (; i < nfrags; i++) {
1330 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1337 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1340 skb_shinfo(skb)->nr_frags = i;
1342 for (; i < nfrags; i++)
1343 skb_frag_unref(skb, i);
1345 if (skb_has_frag_list(skb))
1346 skb_drop_fraglist(skb);
1350 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1351 fragp = &frag->next) {
1352 int end = offset + frag->len;
1354 if (skb_shared(frag)) {
1355 struct sk_buff *nfrag;
1357 nfrag = skb_clone(frag, GFP_ATOMIC);
1358 if (unlikely(!nfrag))
1361 nfrag->next = frag->next;
1373 unlikely((err = pskb_trim(frag, len - offset))))
1377 skb_drop_list(&frag->next);
1382 if (len > skb_headlen(skb)) {
1383 skb->data_len -= skb->len - len;
1388 skb_set_tail_pointer(skb, len);
1393 EXPORT_SYMBOL(___pskb_trim);
1396 * __pskb_pull_tail - advance tail of skb header
1397 * @skb: buffer to reallocate
1398 * @delta: number of bytes to advance tail
1400 * The function makes a sense only on a fragmented &sk_buff,
1401 * it expands header moving its tail forward and copying necessary
1402 * data from fragmented part.
1404 * &sk_buff MUST have reference count of 1.
1406 * Returns %NULL (and &sk_buff does not change) if pull failed
1407 * or value of new tail of skb in the case of success.
1409 * All the pointers pointing into skb header may change and must be
1410 * reloaded after call to this function.
1413 /* Moves tail of skb head forward, copying data from fragmented part,
1414 * when it is necessary.
1415 * 1. It may fail due to malloc failure.
1416 * 2. It may change skb pointers.
1418 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1420 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1422 /* If skb has not enough free space at tail, get new one
1423 * plus 128 bytes for future expansions. If we have enough
1424 * room at tail, reallocate without expansion only if skb is cloned.
1426 int i, k, eat = (skb->tail + delta) - skb->end;
1428 if (eat > 0 || skb_cloned(skb)) {
1429 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1434 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1437 /* Optimization: no fragments, no reasons to preestimate
1438 * size of pulled pages. Superb.
1440 if (!skb_has_frag_list(skb))
1443 /* Estimate size of pulled pages. */
1445 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1446 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1453 /* If we need update frag list, we are in troubles.
1454 * Certainly, it possible to add an offset to skb data,
1455 * but taking into account that pulling is expected to
1456 * be very rare operation, it is worth to fight against
1457 * further bloating skb head and crucify ourselves here instead.
1458 * Pure masohism, indeed. 8)8)
1461 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1462 struct sk_buff *clone = NULL;
1463 struct sk_buff *insp = NULL;
1468 if (list->len <= eat) {
1469 /* Eaten as whole. */
1474 /* Eaten partially. */
1476 if (skb_shared(list)) {
1477 /* Sucks! We need to fork list. :-( */
1478 clone = skb_clone(list, GFP_ATOMIC);
1484 /* This may be pulled without
1488 if (!pskb_pull(list, eat)) {
1496 /* Free pulled out fragments. */
1497 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1498 skb_shinfo(skb)->frag_list = list->next;
1501 /* And insert new clone at head. */
1504 skb_shinfo(skb)->frag_list = clone;
1507 /* Success! Now we may commit changes to skb data. */
1512 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1513 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1516 skb_frag_unref(skb, i);
1519 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1521 skb_shinfo(skb)->frags[k].page_offset += eat;
1522 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1528 skb_shinfo(skb)->nr_frags = k;
1531 skb->data_len -= delta;
1533 return skb_tail_pointer(skb);
1535 EXPORT_SYMBOL(__pskb_pull_tail);
1538 * skb_copy_bits - copy bits from skb to kernel buffer
1540 * @offset: offset in source
1541 * @to: destination buffer
1542 * @len: number of bytes to copy
1544 * Copy the specified number of bytes from the source skb to the
1545 * destination buffer.
1548 * If its prototype is ever changed,
1549 * check arch/{*}/net/{*}.S files,
1550 * since it is called from BPF assembly code.
1552 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1554 int start = skb_headlen(skb);
1555 struct sk_buff *frag_iter;
1558 if (offset > (int)skb->len - len)
1562 if ((copy = start - offset) > 0) {
1565 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1566 if ((len -= copy) == 0)
1572 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1574 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1576 WARN_ON(start > offset + len);
1578 end = start + skb_frag_size(f);
1579 if ((copy = end - offset) > 0) {
1585 vaddr = kmap_atomic(skb_frag_page(f));
1587 vaddr + f->page_offset + offset - start,
1589 kunmap_atomic(vaddr);
1591 if ((len -= copy) == 0)
1599 skb_walk_frags(skb, frag_iter) {
1602 WARN_ON(start > offset + len);
1604 end = start + frag_iter->len;
1605 if ((copy = end - offset) > 0) {
1608 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1610 if ((len -= copy) == 0)
1624 EXPORT_SYMBOL(skb_copy_bits);
1627 * Callback from splice_to_pipe(), if we need to release some pages
1628 * at the end of the spd in case we error'ed out in filling the pipe.
1630 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1632 put_page(spd->pages[i]);
1635 static struct page *linear_to_page(struct page *page, unsigned int *len,
1636 unsigned int *offset,
1639 struct page_frag *pfrag = sk_page_frag(sk);
1641 if (!sk_page_frag_refill(sk, pfrag))
1644 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1646 memcpy(page_address(pfrag->page) + pfrag->offset,
1647 page_address(page) + *offset, *len);
1648 *offset = pfrag->offset;
1649 pfrag->offset += *len;
1654 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1656 unsigned int offset)
1658 return spd->nr_pages &&
1659 spd->pages[spd->nr_pages - 1] == page &&
1660 (spd->partial[spd->nr_pages - 1].offset +
1661 spd->partial[spd->nr_pages - 1].len == offset);
1665 * Fill page/offset/length into spd, if it can hold more pages.
1667 static bool spd_fill_page(struct splice_pipe_desc *spd,
1668 struct pipe_inode_info *pipe, struct page *page,
1669 unsigned int *len, unsigned int offset,
1673 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1677 page = linear_to_page(page, len, &offset, sk);
1681 if (spd_can_coalesce(spd, page, offset)) {
1682 spd->partial[spd->nr_pages - 1].len += *len;
1686 spd->pages[spd->nr_pages] = page;
1687 spd->partial[spd->nr_pages].len = *len;
1688 spd->partial[spd->nr_pages].offset = offset;
1694 static bool __splice_segment(struct page *page, unsigned int poff,
1695 unsigned int plen, unsigned int *off,
1697 struct splice_pipe_desc *spd, bool linear,
1699 struct pipe_inode_info *pipe)
1704 /* skip this segment if already processed */
1710 /* ignore any bits we already processed */
1716 unsigned int flen = min(*len, plen);
1718 if (spd_fill_page(spd, pipe, page, &flen, poff,
1724 } while (*len && plen);
1730 * Map linear and fragment data from the skb to spd. It reports true if the
1731 * pipe is full or if we already spliced the requested length.
1733 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1734 unsigned int *offset, unsigned int *len,
1735 struct splice_pipe_desc *spd, struct sock *sk)
1739 /* map the linear part :
1740 * If skb->head_frag is set, this 'linear' part is backed by a
1741 * fragment, and if the head is not shared with any clones then
1742 * we can avoid a copy since we own the head portion of this page.
1744 if (__splice_segment(virt_to_page(skb->data),
1745 (unsigned long) skb->data & (PAGE_SIZE - 1),
1748 skb_head_is_locked(skb),
1753 * then map the fragments
1755 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1756 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1758 if (__splice_segment(skb_frag_page(f),
1759 f->page_offset, skb_frag_size(f),
1760 offset, len, spd, false, sk, pipe))
1768 * Map data from the skb to a pipe. Should handle both the linear part,
1769 * the fragments, and the frag list. It does NOT handle frag lists within
1770 * the frag list, if such a thing exists. We'd probably need to recurse to
1771 * handle that cleanly.
1773 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1774 struct pipe_inode_info *pipe, unsigned int tlen,
1777 struct partial_page partial[MAX_SKB_FRAGS];
1778 struct page *pages[MAX_SKB_FRAGS];
1779 struct splice_pipe_desc spd = {
1782 .nr_pages_max = MAX_SKB_FRAGS,
1784 .ops = &sock_pipe_buf_ops,
1785 .spd_release = sock_spd_release,
1787 struct sk_buff *frag_iter;
1788 struct sock *sk = skb->sk;
1792 * __skb_splice_bits() only fails if the output has no room left,
1793 * so no point in going over the frag_list for the error case.
1795 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1801 * now see if we have a frag_list to map
1803 skb_walk_frags(skb, frag_iter) {
1806 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1813 * Drop the socket lock, otherwise we have reverse
1814 * locking dependencies between sk_lock and i_mutex
1815 * here as compared to sendfile(). We enter here
1816 * with the socket lock held, and splice_to_pipe() will
1817 * grab the pipe inode lock. For sendfile() emulation,
1818 * we call into ->sendpage() with the i_mutex lock held
1819 * and networking will grab the socket lock.
1822 ret = splice_to_pipe(pipe, &spd);
1830 * skb_store_bits - store bits from kernel buffer to skb
1831 * @skb: destination buffer
1832 * @offset: offset in destination
1833 * @from: source buffer
1834 * @len: number of bytes to copy
1836 * Copy the specified number of bytes from the source buffer to the
1837 * destination skb. This function handles all the messy bits of
1838 * traversing fragment lists and such.
1841 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1843 int start = skb_headlen(skb);
1844 struct sk_buff *frag_iter;
1847 if (offset > (int)skb->len - len)
1850 if ((copy = start - offset) > 0) {
1853 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1854 if ((len -= copy) == 0)
1860 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1861 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1864 WARN_ON(start > offset + len);
1866 end = start + skb_frag_size(frag);
1867 if ((copy = end - offset) > 0) {
1873 vaddr = kmap_atomic(skb_frag_page(frag));
1874 memcpy(vaddr + frag->page_offset + offset - start,
1876 kunmap_atomic(vaddr);
1878 if ((len -= copy) == 0)
1886 skb_walk_frags(skb, frag_iter) {
1889 WARN_ON(start > offset + len);
1891 end = start + frag_iter->len;
1892 if ((copy = end - offset) > 0) {
1895 if (skb_store_bits(frag_iter, offset - start,
1898 if ((len -= copy) == 0)
1911 EXPORT_SYMBOL(skb_store_bits);
1913 /* Checksum skb data. */
1915 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1916 int len, __wsum csum)
1918 int start = skb_headlen(skb);
1919 int i, copy = start - offset;
1920 struct sk_buff *frag_iter;
1923 /* Checksum header. */
1927 csum = csum_partial(skb->data + offset, copy, csum);
1928 if ((len -= copy) == 0)
1934 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1936 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1938 WARN_ON(start > offset + len);
1940 end = start + skb_frag_size(frag);
1941 if ((copy = end - offset) > 0) {
1947 vaddr = kmap_atomic(skb_frag_page(frag));
1948 csum2 = csum_partial(vaddr + frag->page_offset +
1949 offset - start, copy, 0);
1950 kunmap_atomic(vaddr);
1951 csum = csum_block_add(csum, csum2, pos);
1960 skb_walk_frags(skb, frag_iter) {
1963 WARN_ON(start > offset + len);
1965 end = start + frag_iter->len;
1966 if ((copy = end - offset) > 0) {
1970 csum2 = skb_checksum(frag_iter, offset - start,
1972 csum = csum_block_add(csum, csum2, pos);
1973 if ((len -= copy) == 0)
1984 EXPORT_SYMBOL(skb_checksum);
1986 /* Both of above in one bottle. */
1988 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1989 u8 *to, int len, __wsum csum)
1991 int start = skb_headlen(skb);
1992 int i, copy = start - offset;
1993 struct sk_buff *frag_iter;
2000 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2002 if ((len -= copy) == 0)
2009 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2012 WARN_ON(start > offset + len);
2014 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2015 if ((copy = end - offset) > 0) {
2018 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2022 vaddr = kmap_atomic(skb_frag_page(frag));
2023 csum2 = csum_partial_copy_nocheck(vaddr +
2027 kunmap_atomic(vaddr);
2028 csum = csum_block_add(csum, csum2, pos);
2038 skb_walk_frags(skb, frag_iter) {
2042 WARN_ON(start > offset + len);
2044 end = start + frag_iter->len;
2045 if ((copy = end - offset) > 0) {
2048 csum2 = skb_copy_and_csum_bits(frag_iter,
2051 csum = csum_block_add(csum, csum2, pos);
2052 if ((len -= copy) == 0)
2063 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2065 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2070 if (skb->ip_summed == CHECKSUM_PARTIAL)
2071 csstart = skb_checksum_start_offset(skb);
2073 csstart = skb_headlen(skb);
2075 BUG_ON(csstart > skb_headlen(skb));
2077 skb_copy_from_linear_data(skb, to, csstart);
2080 if (csstart != skb->len)
2081 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2082 skb->len - csstart, 0);
2084 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2085 long csstuff = csstart + skb->csum_offset;
2087 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2090 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2093 * skb_dequeue - remove from the head of the queue
2094 * @list: list to dequeue from
2096 * Remove the head of the list. The list lock is taken so the function
2097 * may be used safely with other locking list functions. The head item is
2098 * returned or %NULL if the list is empty.
2101 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2103 unsigned long flags;
2104 struct sk_buff *result;
2106 spin_lock_irqsave(&list->lock, flags);
2107 result = __skb_dequeue(list);
2108 spin_unlock_irqrestore(&list->lock, flags);
2111 EXPORT_SYMBOL(skb_dequeue);
2114 * skb_dequeue_tail - remove from the tail of the queue
2115 * @list: list to dequeue from
2117 * Remove the tail of the list. The list lock is taken so the function
2118 * may be used safely with other locking list functions. The tail item is
2119 * returned or %NULL if the list is empty.
2121 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2123 unsigned long flags;
2124 struct sk_buff *result;
2126 spin_lock_irqsave(&list->lock, flags);
2127 result = __skb_dequeue_tail(list);
2128 spin_unlock_irqrestore(&list->lock, flags);
2131 EXPORT_SYMBOL(skb_dequeue_tail);
2134 * skb_queue_purge - empty a list
2135 * @list: list to empty
2137 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2138 * the list and one reference dropped. This function takes the list
2139 * lock and is atomic with respect to other list locking functions.
2141 void skb_queue_purge(struct sk_buff_head *list)
2143 struct sk_buff *skb;
2144 while ((skb = skb_dequeue(list)) != NULL)
2147 EXPORT_SYMBOL(skb_queue_purge);
2150 * skb_queue_head - queue a buffer at the list head
2151 * @list: list to use
2152 * @newsk: buffer to queue
2154 * Queue a buffer at the start of the list. This function takes the
2155 * list lock and can be used safely with other locking &sk_buff functions
2158 * A buffer cannot be placed on two lists at the same time.
2160 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2162 unsigned long flags;
2164 spin_lock_irqsave(&list->lock, flags);
2165 __skb_queue_head(list, newsk);
2166 spin_unlock_irqrestore(&list->lock, flags);
2168 EXPORT_SYMBOL(skb_queue_head);
2171 * skb_queue_tail - queue a buffer at the list tail
2172 * @list: list to use
2173 * @newsk: buffer to queue
2175 * Queue a buffer at the tail of the list. This function takes the
2176 * list lock and can be used safely with other locking &sk_buff functions
2179 * A buffer cannot be placed on two lists at the same time.
2181 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2183 unsigned long flags;
2185 spin_lock_irqsave(&list->lock, flags);
2186 __skb_queue_tail(list, newsk);
2187 spin_unlock_irqrestore(&list->lock, flags);
2189 EXPORT_SYMBOL(skb_queue_tail);
2192 * skb_unlink - remove a buffer from a list
2193 * @skb: buffer to remove
2194 * @list: list to use
2196 * Remove a packet from a list. The list locks are taken and this
2197 * function is atomic with respect to other list locked calls
2199 * You must know what list the SKB is on.
2201 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2203 unsigned long flags;
2205 spin_lock_irqsave(&list->lock, flags);
2206 __skb_unlink(skb, list);
2207 spin_unlock_irqrestore(&list->lock, flags);
2209 EXPORT_SYMBOL(skb_unlink);
2212 * skb_append - append a buffer
2213 * @old: buffer to insert after
2214 * @newsk: buffer to insert
2215 * @list: list to use
2217 * Place a packet after a given packet in a list. The list locks are taken
2218 * and this function is atomic with respect to other list locked calls.
2219 * A buffer cannot be placed on two lists at the same time.
2221 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2223 unsigned long flags;
2225 spin_lock_irqsave(&list->lock, flags);
2226 __skb_queue_after(list, old, newsk);
2227 spin_unlock_irqrestore(&list->lock, flags);
2229 EXPORT_SYMBOL(skb_append);
2232 * skb_insert - insert a buffer
2233 * @old: buffer to insert before
2234 * @newsk: buffer to insert
2235 * @list: list to use
2237 * Place a packet before a given packet in a list. The list locks are
2238 * taken and this function is atomic with respect to other list locked
2241 * A buffer cannot be placed on two lists at the same time.
2243 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2245 unsigned long flags;
2247 spin_lock_irqsave(&list->lock, flags);
2248 __skb_insert(newsk, old->prev, old, list);
2249 spin_unlock_irqrestore(&list->lock, flags);
2251 EXPORT_SYMBOL(skb_insert);
2253 static inline void skb_split_inside_header(struct sk_buff *skb,
2254 struct sk_buff* skb1,
2255 const u32 len, const int pos)
2259 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2261 /* And move data appendix as is. */
2262 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2263 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2265 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2266 skb_shinfo(skb)->nr_frags = 0;
2267 skb1->data_len = skb->data_len;
2268 skb1->len += skb1->data_len;
2271 skb_set_tail_pointer(skb, len);
2274 static inline void skb_split_no_header(struct sk_buff *skb,
2275 struct sk_buff* skb1,
2276 const u32 len, int pos)
2279 const int nfrags = skb_shinfo(skb)->nr_frags;
2281 skb_shinfo(skb)->nr_frags = 0;
2282 skb1->len = skb1->data_len = skb->len - len;
2284 skb->data_len = len - pos;
2286 for (i = 0; i < nfrags; i++) {
2287 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2289 if (pos + size > len) {
2290 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2294 * We have two variants in this case:
2295 * 1. Move all the frag to the second
2296 * part, if it is possible. F.e.
2297 * this approach is mandatory for TUX,
2298 * where splitting is expensive.
2299 * 2. Split is accurately. We make this.
2301 skb_frag_ref(skb, i);
2302 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2303 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2304 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2305 skb_shinfo(skb)->nr_frags++;
2309 skb_shinfo(skb)->nr_frags++;
2312 skb_shinfo(skb1)->nr_frags = k;
2316 * skb_split - Split fragmented skb to two parts at length len.
2317 * @skb: the buffer to split
2318 * @skb1: the buffer to receive the second part
2319 * @len: new length for skb
2321 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2323 int pos = skb_headlen(skb);
2325 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2326 if (len < pos) /* Split line is inside header. */
2327 skb_split_inside_header(skb, skb1, len, pos);
2328 else /* Second chunk has no header, nothing to copy. */
2329 skb_split_no_header(skb, skb1, len, pos);
2331 EXPORT_SYMBOL(skb_split);
2333 /* Shifting from/to a cloned skb is a no-go.
2335 * Caller cannot keep skb_shinfo related pointers past calling here!
2337 static int skb_prepare_for_shift(struct sk_buff *skb)
2339 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2343 * skb_shift - Shifts paged data partially from skb to another
2344 * @tgt: buffer into which tail data gets added
2345 * @skb: buffer from which the paged data comes from
2346 * @shiftlen: shift up to this many bytes
2348 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2349 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2350 * It's up to caller to free skb if everything was shifted.
2352 * If @tgt runs out of frags, the whole operation is aborted.
2354 * Skb cannot include anything else but paged data while tgt is allowed
2355 * to have non-paged data as well.
2357 * TODO: full sized shift could be optimized but that would need
2358 * specialized skb free'er to handle frags without up-to-date nr_frags.
2360 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2362 int from, to, merge, todo;
2363 struct skb_frag_struct *fragfrom, *fragto;
2365 BUG_ON(shiftlen > skb->len);
2366 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2370 to = skb_shinfo(tgt)->nr_frags;
2371 fragfrom = &skb_shinfo(skb)->frags[from];
2373 /* Actual merge is delayed until the point when we know we can
2374 * commit all, so that we don't have to undo partial changes
2377 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2378 fragfrom->page_offset)) {
2383 todo -= skb_frag_size(fragfrom);
2385 if (skb_prepare_for_shift(skb) ||
2386 skb_prepare_for_shift(tgt))
2389 /* All previous frag pointers might be stale! */
2390 fragfrom = &skb_shinfo(skb)->frags[from];
2391 fragto = &skb_shinfo(tgt)->frags[merge];
2393 skb_frag_size_add(fragto, shiftlen);
2394 skb_frag_size_sub(fragfrom, shiftlen);
2395 fragfrom->page_offset += shiftlen;
2403 /* Skip full, not-fitting skb to avoid expensive operations */
2404 if ((shiftlen == skb->len) &&
2405 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2408 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2411 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2412 if (to == MAX_SKB_FRAGS)
2415 fragfrom = &skb_shinfo(skb)->frags[from];
2416 fragto = &skb_shinfo(tgt)->frags[to];
2418 if (todo >= skb_frag_size(fragfrom)) {
2419 *fragto = *fragfrom;
2420 todo -= skb_frag_size(fragfrom);
2425 __skb_frag_ref(fragfrom);
2426 fragto->page = fragfrom->page;
2427 fragto->page_offset = fragfrom->page_offset;
2428 skb_frag_size_set(fragto, todo);
2430 fragfrom->page_offset += todo;
2431 skb_frag_size_sub(fragfrom, todo);
2439 /* Ready to "commit" this state change to tgt */
2440 skb_shinfo(tgt)->nr_frags = to;
2443 fragfrom = &skb_shinfo(skb)->frags[0];
2444 fragto = &skb_shinfo(tgt)->frags[merge];
2446 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2447 __skb_frag_unref(fragfrom);
2450 /* Reposition in the original skb */
2452 while (from < skb_shinfo(skb)->nr_frags)
2453 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2454 skb_shinfo(skb)->nr_frags = to;
2456 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2459 /* Most likely the tgt won't ever need its checksum anymore, skb on
2460 * the other hand might need it if it needs to be resent
2462 tgt->ip_summed = CHECKSUM_PARTIAL;
2463 skb->ip_summed = CHECKSUM_PARTIAL;
2465 /* Yak, is it really working this way? Some helper please? */
2466 skb->len -= shiftlen;
2467 skb->data_len -= shiftlen;
2468 skb->truesize -= shiftlen;
2469 tgt->len += shiftlen;
2470 tgt->data_len += shiftlen;
2471 tgt->truesize += shiftlen;
2477 * skb_prepare_seq_read - Prepare a sequential read of skb data
2478 * @skb: the buffer to read
2479 * @from: lower offset of data to be read
2480 * @to: upper offset of data to be read
2481 * @st: state variable
2483 * Initializes the specified state variable. Must be called before
2484 * invoking skb_seq_read() for the first time.
2486 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2487 unsigned int to, struct skb_seq_state *st)
2489 st->lower_offset = from;
2490 st->upper_offset = to;
2491 st->root_skb = st->cur_skb = skb;
2492 st->frag_idx = st->stepped_offset = 0;
2493 st->frag_data = NULL;
2495 EXPORT_SYMBOL(skb_prepare_seq_read);
2498 * skb_seq_read - Sequentially read skb data
2499 * @consumed: number of bytes consumed by the caller so far
2500 * @data: destination pointer for data to be returned
2501 * @st: state variable
2503 * Reads a block of skb data at &consumed relative to the
2504 * lower offset specified to skb_prepare_seq_read(). Assigns
2505 * the head of the data block to &data and returns the length
2506 * of the block or 0 if the end of the skb data or the upper
2507 * offset has been reached.
2509 * The caller is not required to consume all of the data
2510 * returned, i.e. &consumed is typically set to the number
2511 * of bytes already consumed and the next call to
2512 * skb_seq_read() will return the remaining part of the block.
2514 * Note 1: The size of each block of data returned can be arbitrary,
2515 * this limitation is the cost for zerocopy seqeuental
2516 * reads of potentially non linear data.
2518 * Note 2: Fragment lists within fragments are not implemented
2519 * at the moment, state->root_skb could be replaced with
2520 * a stack for this purpose.
2522 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2523 struct skb_seq_state *st)
2525 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2528 if (unlikely(abs_offset >= st->upper_offset))
2532 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2534 if (abs_offset < block_limit && !st->frag_data) {
2535 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2536 return block_limit - abs_offset;
2539 if (st->frag_idx == 0 && !st->frag_data)
2540 st->stepped_offset += skb_headlen(st->cur_skb);
2542 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2543 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2544 block_limit = skb_frag_size(frag) + st->stepped_offset;
2546 if (abs_offset < block_limit) {
2548 st->frag_data = kmap_atomic(skb_frag_page(frag));
2550 *data = (u8 *) st->frag_data + frag->page_offset +
2551 (abs_offset - st->stepped_offset);
2553 return block_limit - abs_offset;
2556 if (st->frag_data) {
2557 kunmap_atomic(st->frag_data);
2558 st->frag_data = NULL;
2562 st->stepped_offset += skb_frag_size(frag);
2565 if (st->frag_data) {
2566 kunmap_atomic(st->frag_data);
2567 st->frag_data = NULL;
2570 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2571 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2574 } else if (st->cur_skb->next) {
2575 st->cur_skb = st->cur_skb->next;
2582 EXPORT_SYMBOL(skb_seq_read);
2585 * skb_abort_seq_read - Abort a sequential read of skb data
2586 * @st: state variable
2588 * Must be called if skb_seq_read() was not called until it
2591 void skb_abort_seq_read(struct skb_seq_state *st)
2594 kunmap_atomic(st->frag_data);
2596 EXPORT_SYMBOL(skb_abort_seq_read);
2598 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2600 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2601 struct ts_config *conf,
2602 struct ts_state *state)
2604 return skb_seq_read(offset, text, TS_SKB_CB(state));
2607 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2609 skb_abort_seq_read(TS_SKB_CB(state));
2613 * skb_find_text - Find a text pattern in skb data
2614 * @skb: the buffer to look in
2615 * @from: search offset
2617 * @config: textsearch configuration
2618 * @state: uninitialized textsearch state variable
2620 * Finds a pattern in the skb data according to the specified
2621 * textsearch configuration. Use textsearch_next() to retrieve
2622 * subsequent occurrences of the pattern. Returns the offset
2623 * to the first occurrence or UINT_MAX if no match was found.
2625 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2626 unsigned int to, struct ts_config *config,
2627 struct ts_state *state)
2631 config->get_next_block = skb_ts_get_next_block;
2632 config->finish = skb_ts_finish;
2634 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2636 ret = textsearch_find(config, state);
2637 return (ret <= to - from ? ret : UINT_MAX);
2639 EXPORT_SYMBOL(skb_find_text);
2642 * skb_append_datato_frags - append the user data to a skb
2643 * @sk: sock structure
2644 * @skb: skb structure to be appened with user data.
2645 * @getfrag: call back function to be used for getting the user data
2646 * @from: pointer to user message iov
2647 * @length: length of the iov message
2649 * Description: This procedure append the user data in the fragment part
2650 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2652 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2653 int (*getfrag)(void *from, char *to, int offset,
2654 int len, int odd, struct sk_buff *skb),
2655 void *from, int length)
2657 int frg_cnt = skb_shinfo(skb)->nr_frags;
2661 struct page_frag *pfrag = ¤t->task_frag;
2664 /* Return error if we don't have space for new frag */
2665 if (frg_cnt >= MAX_SKB_FRAGS)
2668 if (!sk_page_frag_refill(sk, pfrag))
2671 /* copy the user data to page */
2672 copy = min_t(int, length, pfrag->size - pfrag->offset);
2674 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2675 offset, copy, 0, skb);
2679 /* copy was successful so update the size parameters */
2680 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2683 pfrag->offset += copy;
2684 get_page(pfrag->page);
2686 skb->truesize += copy;
2687 atomic_add(copy, &sk->sk_wmem_alloc);
2689 skb->data_len += copy;
2693 } while (length > 0);
2697 EXPORT_SYMBOL(skb_append_datato_frags);
2700 * skb_pull_rcsum - pull skb and update receive checksum
2701 * @skb: buffer to update
2702 * @len: length of data pulled
2704 * This function performs an skb_pull on the packet and updates
2705 * the CHECKSUM_COMPLETE checksum. It should be used on
2706 * receive path processing instead of skb_pull unless you know
2707 * that the checksum difference is zero (e.g., a valid IP header)
2708 * or you are setting ip_summed to CHECKSUM_NONE.
2710 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2712 BUG_ON(len > skb->len);
2714 BUG_ON(skb->len < skb->data_len);
2715 skb_postpull_rcsum(skb, skb->data, len);
2716 return skb->data += len;
2718 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2721 * skb_segment - Perform protocol segmentation on skb.
2722 * @skb: buffer to segment
2723 * @features: features for the output path (see dev->features)
2725 * This function performs segmentation on the given skb. It returns
2726 * a pointer to the first in a list of new skbs for the segments.
2727 * In case of error it returns ERR_PTR(err).
2729 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2731 struct sk_buff *segs = NULL;
2732 struct sk_buff *tail = NULL;
2733 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2734 unsigned int mss = skb_shinfo(skb)->gso_size;
2735 unsigned int doffset = skb->data - skb_mac_header(skb);
2736 unsigned int offset = doffset;
2737 unsigned int tnl_hlen = skb_tnl_header_len(skb);
2738 unsigned int headroom;
2742 int sg = !!(features & NETIF_F_SG);
2743 int nfrags = skb_shinfo(skb)->nr_frags;
2748 proto = skb_network_protocol(skb);
2749 if (unlikely(!proto))
2750 return ERR_PTR(-EINVAL);
2752 csum = !!can_checksum_protocol(features, proto);
2753 __skb_push(skb, doffset);
2754 headroom = skb_headroom(skb);
2755 pos = skb_headlen(skb);
2758 struct sk_buff *nskb;
2763 len = skb->len - offset;
2767 hsize = skb_headlen(skb) - offset;
2770 if (hsize > len || !sg)
2773 if (!hsize && i >= nfrags) {
2774 BUG_ON(fskb->len != len);
2777 nskb = skb_clone(fskb, GFP_ATOMIC);
2780 if (unlikely(!nskb))
2783 hsize = skb_end_offset(nskb);
2784 if (skb_cow_head(nskb, doffset + headroom)) {
2789 nskb->truesize += skb_end_offset(nskb) - hsize;
2790 skb_release_head_state(nskb);
2791 __skb_push(nskb, doffset);
2793 nskb = __alloc_skb(hsize + doffset + headroom,
2794 GFP_ATOMIC, skb_alloc_rx_flag(skb),
2797 if (unlikely(!nskb))
2800 skb_reserve(nskb, headroom);
2801 __skb_put(nskb, doffset);
2810 __copy_skb_header(nskb, skb);
2811 nskb->mac_len = skb->mac_len;
2813 /* nskb and skb might have different headroom */
2814 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2815 nskb->csum_start += skb_headroom(nskb) - headroom;
2817 skb_reset_mac_header(nskb);
2818 skb_set_network_header(nskb, skb->mac_len);
2819 nskb->transport_header = (nskb->network_header +
2820 skb_network_header_len(skb));
2822 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
2823 nskb->data - tnl_hlen,
2824 doffset + tnl_hlen);
2826 if (fskb != skb_shinfo(skb)->frag_list)
2830 nskb->ip_summed = CHECKSUM_NONE;
2831 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2837 frag = skb_shinfo(nskb)->frags;
2839 skb_copy_from_linear_data_offset(skb, offset,
2840 skb_put(nskb, hsize), hsize);
2842 skb_shinfo(nskb)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2844 while (pos < offset + len && i < nfrags) {
2845 *frag = skb_shinfo(skb)->frags[i];
2846 __skb_frag_ref(frag);
2847 size = skb_frag_size(frag);
2850 frag->page_offset += offset - pos;
2851 skb_frag_size_sub(frag, offset - pos);
2854 skb_shinfo(nskb)->nr_frags++;
2856 if (pos + size <= offset + len) {
2860 skb_frag_size_sub(frag, pos + size - (offset + len));
2867 if (pos < offset + len) {
2868 struct sk_buff *fskb2 = fskb;
2870 BUG_ON(pos + fskb->len != offset + len);
2876 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2882 SKB_FRAG_ASSERT(nskb);
2883 skb_shinfo(nskb)->frag_list = fskb2;
2887 nskb->data_len = len - hsize;
2888 nskb->len += nskb->data_len;
2889 nskb->truesize += nskb->data_len;
2892 nskb->csum = skb_checksum(nskb, doffset,
2893 nskb->len - doffset, 0);
2894 nskb->ip_summed = CHECKSUM_NONE;
2896 } while ((offset += len) < skb->len);
2901 while ((skb = segs)) {
2905 return ERR_PTR(err);
2907 EXPORT_SYMBOL_GPL(skb_segment);
2909 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2911 struct sk_buff *p = *head;
2912 struct sk_buff *nskb;
2913 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2914 struct skb_shared_info *pinfo = skb_shinfo(p);
2915 unsigned int headroom;
2916 unsigned int len = skb_gro_len(skb);
2917 unsigned int offset = skb_gro_offset(skb);
2918 unsigned int headlen = skb_headlen(skb);
2919 unsigned int delta_truesize;
2921 if (p->len + len >= 65536)
2924 if (pinfo->frag_list)
2926 else if (headlen <= offset) {
2929 int i = skbinfo->nr_frags;
2930 int nr_frags = pinfo->nr_frags + i;
2934 if (nr_frags > MAX_SKB_FRAGS)
2937 pinfo->nr_frags = nr_frags;
2938 skbinfo->nr_frags = 0;
2940 frag = pinfo->frags + nr_frags;
2941 frag2 = skbinfo->frags + i;
2946 frag->page_offset += offset;
2947 skb_frag_size_sub(frag, offset);
2949 /* all fragments truesize : remove (head size + sk_buff) */
2950 delta_truesize = skb->truesize -
2951 SKB_TRUESIZE(skb_end_offset(skb));
2953 skb->truesize -= skb->data_len;
2954 skb->len -= skb->data_len;
2957 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
2959 } else if (skb->head_frag) {
2960 int nr_frags = pinfo->nr_frags;
2961 skb_frag_t *frag = pinfo->frags + nr_frags;
2962 struct page *page = virt_to_head_page(skb->head);
2963 unsigned int first_size = headlen - offset;
2964 unsigned int first_offset;
2966 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
2969 first_offset = skb->data -
2970 (unsigned char *)page_address(page) +
2973 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
2975 frag->page.p = page;
2976 frag->page_offset = first_offset;
2977 skb_frag_size_set(frag, first_size);
2979 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
2980 /* We dont need to clear skbinfo->nr_frags here */
2982 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
2983 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
2985 } else if (skb_gro_len(p) != pinfo->gso_size)
2988 headroom = skb_headroom(p);
2989 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2990 if (unlikely(!nskb))
2993 __copy_skb_header(nskb, p);
2994 nskb->mac_len = p->mac_len;
2996 skb_reserve(nskb, headroom);
2997 __skb_put(nskb, skb_gro_offset(p));
2999 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3000 skb_set_network_header(nskb, skb_network_offset(p));
3001 skb_set_transport_header(nskb, skb_transport_offset(p));
3003 __skb_pull(p, skb_gro_offset(p));
3004 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3005 p->data - skb_mac_header(p));
3007 skb_shinfo(nskb)->frag_list = p;
3008 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3009 pinfo->gso_size = 0;
3010 skb_header_release(p);
3011 NAPI_GRO_CB(nskb)->last = p;
3013 nskb->data_len += p->len;
3014 nskb->truesize += p->truesize;
3015 nskb->len += p->len;
3018 nskb->next = p->next;
3024 delta_truesize = skb->truesize;
3025 if (offset > headlen) {
3026 unsigned int eat = offset - headlen;
3028 skbinfo->frags[0].page_offset += eat;
3029 skb_frag_size_sub(&skbinfo->frags[0], eat);
3030 skb->data_len -= eat;
3035 __skb_pull(skb, offset);
3037 NAPI_GRO_CB(p)->last->next = skb;
3038 NAPI_GRO_CB(p)->last = skb;
3039 skb_header_release(skb);
3042 NAPI_GRO_CB(p)->count++;
3044 p->truesize += delta_truesize;
3047 NAPI_GRO_CB(skb)->same_flow = 1;
3050 EXPORT_SYMBOL_GPL(skb_gro_receive);
3052 void __init skb_init(void)
3054 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3055 sizeof(struct sk_buff),
3057 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3059 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3060 (2*sizeof(struct sk_buff)) +
3063 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3068 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3069 * @skb: Socket buffer containing the buffers to be mapped
3070 * @sg: The scatter-gather list to map into
3071 * @offset: The offset into the buffer's contents to start mapping
3072 * @len: Length of buffer space to be mapped
3074 * Fill the specified scatter-gather list with mappings/pointers into a
3075 * region of the buffer space attached to a socket buffer.
3078 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3080 int start = skb_headlen(skb);
3081 int i, copy = start - offset;
3082 struct sk_buff *frag_iter;
3088 sg_set_buf(sg, skb->data + offset, copy);
3090 if ((len -= copy) == 0)
3095 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3098 WARN_ON(start > offset + len);
3100 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3101 if ((copy = end - offset) > 0) {
3102 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3106 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3107 frag->page_offset+offset-start);
3116 skb_walk_frags(skb, frag_iter) {
3119 WARN_ON(start > offset + len);
3121 end = start + frag_iter->len;
3122 if ((copy = end - offset) > 0) {
3125 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3127 if ((len -= copy) == 0)
3137 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3139 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3141 sg_mark_end(&sg[nsg - 1]);
3145 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3148 * skb_cow_data - Check that a socket buffer's data buffers are writable
3149 * @skb: The socket buffer to check.
3150 * @tailbits: Amount of trailing space to be added
3151 * @trailer: Returned pointer to the skb where the @tailbits space begins
3153 * Make sure that the data buffers attached to a socket buffer are
3154 * writable. If they are not, private copies are made of the data buffers
3155 * and the socket buffer is set to use these instead.
3157 * If @tailbits is given, make sure that there is space to write @tailbits
3158 * bytes of data beyond current end of socket buffer. @trailer will be
3159 * set to point to the skb in which this space begins.
3161 * The number of scatterlist elements required to completely map the
3162 * COW'd and extended socket buffer will be returned.
3164 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3168 struct sk_buff *skb1, **skb_p;
3170 /* If skb is cloned or its head is paged, reallocate
3171 * head pulling out all the pages (pages are considered not writable
3172 * at the moment even if they are anonymous).
3174 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3175 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3178 /* Easy case. Most of packets will go this way. */
3179 if (!skb_has_frag_list(skb)) {
3180 /* A little of trouble, not enough of space for trailer.
3181 * This should not happen, when stack is tuned to generate
3182 * good frames. OK, on miss we reallocate and reserve even more
3183 * space, 128 bytes is fair. */
3185 if (skb_tailroom(skb) < tailbits &&
3186 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3194 /* Misery. We are in troubles, going to mincer fragments... */
3197 skb_p = &skb_shinfo(skb)->frag_list;
3200 while ((skb1 = *skb_p) != NULL) {
3203 /* The fragment is partially pulled by someone,
3204 * this can happen on input. Copy it and everything
3207 if (skb_shared(skb1))
3210 /* If the skb is the last, worry about trailer. */
3212 if (skb1->next == NULL && tailbits) {
3213 if (skb_shinfo(skb1)->nr_frags ||
3214 skb_has_frag_list(skb1) ||
3215 skb_tailroom(skb1) < tailbits)
3216 ntail = tailbits + 128;
3222 skb_shinfo(skb1)->nr_frags ||
3223 skb_has_frag_list(skb1)) {
3224 struct sk_buff *skb2;
3226 /* Fuck, we are miserable poor guys... */
3228 skb2 = skb_copy(skb1, GFP_ATOMIC);
3230 skb2 = skb_copy_expand(skb1,
3234 if (unlikely(skb2 == NULL))
3238 skb_set_owner_w(skb2, skb1->sk);
3240 /* Looking around. Are we still alive?
3241 * OK, link new skb, drop old one */
3243 skb2->next = skb1->next;
3250 skb_p = &skb1->next;
3255 EXPORT_SYMBOL_GPL(skb_cow_data);
3257 static void sock_rmem_free(struct sk_buff *skb)
3259 struct sock *sk = skb->sk;
3261 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3265 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3267 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3271 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3272 (unsigned int)sk->sk_rcvbuf)
3277 skb->destructor = sock_rmem_free;
3278 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3280 /* before exiting rcu section, make sure dst is refcounted */
3283 skb_queue_tail(&sk->sk_error_queue, skb);
3284 if (!sock_flag(sk, SOCK_DEAD))
3285 sk->sk_data_ready(sk, len);
3288 EXPORT_SYMBOL(sock_queue_err_skb);
3290 void skb_tstamp_tx(struct sk_buff *orig_skb,
3291 struct skb_shared_hwtstamps *hwtstamps)
3293 struct sock *sk = orig_skb->sk;
3294 struct sock_exterr_skb *serr;
3295 struct sk_buff *skb;
3301 skb = skb_clone(orig_skb, GFP_ATOMIC);
3306 *skb_hwtstamps(skb) =
3310 * no hardware time stamps available,
3311 * so keep the shared tx_flags and only
3312 * store software time stamp
3314 skb->tstamp = ktime_get_real();
3317 serr = SKB_EXT_ERR(skb);
3318 memset(serr, 0, sizeof(*serr));
3319 serr->ee.ee_errno = ENOMSG;
3320 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3322 err = sock_queue_err_skb(sk, skb);
3327 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3329 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3331 struct sock *sk = skb->sk;
3332 struct sock_exterr_skb *serr;
3335 skb->wifi_acked_valid = 1;
3336 skb->wifi_acked = acked;
3338 serr = SKB_EXT_ERR(skb);
3339 memset(serr, 0, sizeof(*serr));
3340 serr->ee.ee_errno = ENOMSG;
3341 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3343 err = sock_queue_err_skb(sk, skb);
3347 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3351 * skb_partial_csum_set - set up and verify partial csum values for packet
3352 * @skb: the skb to set
3353 * @start: the number of bytes after skb->data to start checksumming.
3354 * @off: the offset from start to place the checksum.
3356 * For untrusted partially-checksummed packets, we need to make sure the values
3357 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3359 * This function checks and sets those values and skb->ip_summed: if this
3360 * returns false you should drop the packet.
3362 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3364 if (unlikely(start > skb_headlen(skb)) ||
3365 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3366 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3367 start, off, skb_headlen(skb));
3370 skb->ip_summed = CHECKSUM_PARTIAL;
3371 skb->csum_start = skb_headroom(skb) + start;
3372 skb->csum_offset = off;
3373 skb_set_transport_header(skb, start);
3376 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3378 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3380 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3383 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3385 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3388 skb_release_head_state(skb);
3389 kmem_cache_free(skbuff_head_cache, skb);
3394 EXPORT_SYMBOL(kfree_skb_partial);
3397 * skb_try_coalesce - try to merge skb to prior one
3399 * @from: buffer to add
3400 * @fragstolen: pointer to boolean
3401 * @delta_truesize: how much more was allocated than was requested
3403 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3404 bool *fragstolen, int *delta_truesize)
3406 int i, delta, len = from->len;
3408 *fragstolen = false;
3413 if (len <= skb_tailroom(to)) {
3414 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3415 *delta_truesize = 0;
3419 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3422 if (skb_headlen(from) != 0) {
3424 unsigned int offset;
3426 if (skb_shinfo(to)->nr_frags +
3427 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3430 if (skb_head_is_locked(from))
3433 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3435 page = virt_to_head_page(from->head);
3436 offset = from->data - (unsigned char *)page_address(page);
3438 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3439 page, offset, skb_headlen(from));
3442 if (skb_shinfo(to)->nr_frags +
3443 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3446 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
3449 WARN_ON_ONCE(delta < len);
3451 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3452 skb_shinfo(from)->frags,
3453 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3454 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3456 if (!skb_cloned(from))
3457 skb_shinfo(from)->nr_frags = 0;
3459 /* if the skb is not cloned this does nothing
3460 * since we set nr_frags to 0.
3462 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3463 skb_frag_ref(from, i);
3465 to->truesize += delta;
3467 to->data_len += len;
3469 *delta_truesize = delta;
3472 EXPORT_SYMBOL(skb_try_coalesce);