2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
8 * Alan Cox : Fixed the worst of the load
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
36 * The functions in this file will not compile correctly with gcc 2.4.x
39 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
44 #include <linux/kmemcheck.h>
46 #include <linux/interrupt.h>
48 #include <linux/inet.h>
49 #include <linux/slab.h>
50 #include <linux/netdevice.h>
51 #ifdef CONFIG_NET_CLS_ACT
52 #include <net/pkt_sched.h>
54 #include <linux/string.h>
55 #include <linux/skbuff.h>
56 #include <linux/splice.h>
57 #include <linux/cache.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/init.h>
60 #include <linux/scatterlist.h>
61 #include <linux/errqueue.h>
62 #include <linux/prefetch.h>
64 #include <net/protocol.h>
67 #include <net/checksum.h>
70 #include <asm/uaccess.h>
71 #include <trace/events/skb.h>
72 #include <linux/highmem.h>
74 struct kmem_cache *skbuff_head_cache __read_mostly;
75 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
77 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
78 struct pipe_buffer *buf)
83 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
84 struct pipe_buffer *buf)
89 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
90 struct pipe_buffer *buf)
96 /* Pipe buffer operations for a socket. */
97 static const struct pipe_buf_operations sock_pipe_buf_ops = {
99 .map = generic_pipe_buf_map,
100 .unmap = generic_pipe_buf_unmap,
101 .confirm = generic_pipe_buf_confirm,
102 .release = sock_pipe_buf_release,
103 .steal = sock_pipe_buf_steal,
104 .get = sock_pipe_buf_get,
108 * Keep out-of-line to prevent kernel bloat.
109 * __builtin_return_address is not used because it is not always
114 * skb_over_panic - private function
119 * Out of line support code for skb_put(). Not user callable.
121 static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
123 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
124 __func__, here, skb->len, sz, skb->head, skb->data,
125 (unsigned long)skb->tail, (unsigned long)skb->end,
126 skb->dev ? skb->dev->name : "<NULL>");
131 * skb_under_panic - private function
136 * Out of line support code for skb_push(). Not user callable.
139 static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
141 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
142 __func__, here, skb->len, sz, skb->head, skb->data,
143 (unsigned long)skb->tail, (unsigned long)skb->end,
144 skb->dev ? skb->dev->name : "<NULL>");
150 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
151 * the caller if emergency pfmemalloc reserves are being used. If it is and
152 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
153 * may be used. Otherwise, the packet data may be discarded until enough
156 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
157 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
158 void *__kmalloc_reserve(size_t size, gfp_t flags, int node, unsigned long ip,
162 bool ret_pfmemalloc = false;
165 * Try a regular allocation, when that fails and we're not entitled
166 * to the reserves, fail.
168 obj = kmalloc_node_track_caller(size,
169 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
171 if (obj || !(gfp_pfmemalloc_allowed(flags)))
174 /* Try again but now we are using pfmemalloc reserves */
175 ret_pfmemalloc = true;
176 obj = kmalloc_node_track_caller(size, flags, node);
180 *pfmemalloc = ret_pfmemalloc;
185 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
186 * 'private' fields and also do memory statistics to find all the
192 * __alloc_skb - allocate a network buffer
193 * @size: size to allocate
194 * @gfp_mask: allocation mask
195 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
196 * instead of head cache and allocate a cloned (child) skb.
197 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
198 * allocations in case the data is required for writeback
199 * @node: numa node to allocate memory on
201 * Allocate a new &sk_buff. The returned buffer has no headroom and a
202 * tail room of at least size bytes. The object has a reference count
203 * of one. The return is the buffer. On a failure the return is %NULL.
205 * Buffers may only be allocated from interrupts using a @gfp_mask of
208 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
211 struct kmem_cache *cache;
212 struct skb_shared_info *shinfo;
217 cache = (flags & SKB_ALLOC_FCLONE)
218 ? skbuff_fclone_cache : skbuff_head_cache;
220 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
221 gfp_mask |= __GFP_MEMALLOC;
224 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
229 /* We do our best to align skb_shared_info on a separate cache
230 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
231 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
232 * Both skb->head and skb_shared_info are cache line aligned.
234 size = SKB_DATA_ALIGN(size);
235 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
236 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
239 /* kmalloc(size) might give us more room than requested.
240 * Put skb_shared_info exactly at the end of allocated zone,
241 * to allow max possible filling before reallocation.
243 size = SKB_WITH_OVERHEAD(ksize(data));
244 prefetchw(data + size);
247 * Only clear those fields we need to clear, not those that we will
248 * actually initialise below. Hence, don't put any more fields after
249 * the tail pointer in struct sk_buff!
251 memset(skb, 0, offsetof(struct sk_buff, tail));
252 /* Account for allocated memory : skb + skb->head */
253 skb->truesize = SKB_TRUESIZE(size);
254 skb->pfmemalloc = pfmemalloc;
255 atomic_set(&skb->users, 1);
258 skb_reset_tail_pointer(skb);
259 skb->end = skb->tail + size;
260 #ifdef NET_SKBUFF_DATA_USES_OFFSET
261 skb->mac_header = ~0U;
264 /* make sure we initialize shinfo sequentially */
265 shinfo = skb_shinfo(skb);
266 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
267 atomic_set(&shinfo->dataref, 1);
268 kmemcheck_annotate_variable(shinfo->destructor_arg);
270 if (flags & SKB_ALLOC_FCLONE) {
271 struct sk_buff *child = skb + 1;
272 atomic_t *fclone_ref = (atomic_t *) (child + 1);
274 kmemcheck_annotate_bitfield(child, flags1);
275 kmemcheck_annotate_bitfield(child, flags2);
276 skb->fclone = SKB_FCLONE_ORIG;
277 atomic_set(fclone_ref, 1);
279 child->fclone = SKB_FCLONE_UNAVAILABLE;
280 child->pfmemalloc = pfmemalloc;
285 kmem_cache_free(cache, skb);
289 EXPORT_SYMBOL(__alloc_skb);
292 * build_skb - build a network buffer
293 * @data: data buffer provided by caller
294 * @frag_size: size of fragment, or 0 if head was kmalloced
296 * Allocate a new &sk_buff. Caller provides space holding head and
297 * skb_shared_info. @data must have been allocated by kmalloc()
298 * The return is the new skb buffer.
299 * On a failure the return is %NULL, and @data is not freed.
301 * Before IO, driver allocates only data buffer where NIC put incoming frame
302 * Driver should add room at head (NET_SKB_PAD) and
303 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
304 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
305 * before giving packet to stack.
306 * RX rings only contains data buffers, not full skbs.
308 struct sk_buff *build_skb(void *data, unsigned int frag_size)
310 struct skb_shared_info *shinfo;
312 unsigned int size = frag_size ? : ksize(data);
314 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
318 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
320 memset(skb, 0, offsetof(struct sk_buff, tail));
321 skb->truesize = SKB_TRUESIZE(size);
322 skb->head_frag = frag_size != 0;
323 atomic_set(&skb->users, 1);
326 skb_reset_tail_pointer(skb);
327 skb->end = skb->tail + size;
328 #ifdef NET_SKBUFF_DATA_USES_OFFSET
329 skb->mac_header = ~0U;
332 /* make sure we initialize shinfo sequentially */
333 shinfo = skb_shinfo(skb);
334 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
335 atomic_set(&shinfo->dataref, 1);
336 kmemcheck_annotate_variable(shinfo->destructor_arg);
340 EXPORT_SYMBOL(build_skb);
342 struct netdev_alloc_cache {
343 struct page_frag frag;
344 /* we maintain a pagecount bias, so that we dont dirty cache line
345 * containing page->_count every time we allocate a fragment.
347 unsigned int pagecnt_bias;
349 static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
351 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
352 #define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
353 #define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
355 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
357 struct netdev_alloc_cache *nc;
362 local_irq_save(flags);
363 nc = &__get_cpu_var(netdev_alloc_cache);
364 if (unlikely(!nc->frag.page)) {
366 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
367 gfp_t gfp = gfp_mask;
370 gfp |= __GFP_COMP | __GFP_NOWARN;
371 nc->frag.page = alloc_pages(gfp, order);
372 if (likely(nc->frag.page))
377 nc->frag.size = PAGE_SIZE << order;
379 atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS);
380 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
384 if (nc->frag.offset + fragsz > nc->frag.size) {
385 /* avoid unnecessary locked operations if possible */
386 if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) ||
387 atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count))
392 data = page_address(nc->frag.page) + nc->frag.offset;
393 nc->frag.offset += fragsz;
396 local_irq_restore(flags);
401 * netdev_alloc_frag - allocate a page fragment
402 * @fragsz: fragment size
404 * Allocates a frag from a page for receive buffer.
405 * Uses GFP_ATOMIC allocations.
407 void *netdev_alloc_frag(unsigned int fragsz)
409 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
411 EXPORT_SYMBOL(netdev_alloc_frag);
414 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
415 * @dev: network device to receive on
416 * @length: length to allocate
417 * @gfp_mask: get_free_pages mask, passed to alloc_skb
419 * Allocate a new &sk_buff and assign it a usage count of one. The
420 * buffer has unspecified headroom built in. Users should allocate
421 * the headroom they think they need without accounting for the
422 * built in space. The built in space is used for optimisations.
424 * %NULL is returned if there is no free memory.
426 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
427 unsigned int length, gfp_t gfp_mask)
429 struct sk_buff *skb = NULL;
430 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
431 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
433 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
436 if (sk_memalloc_socks())
437 gfp_mask |= __GFP_MEMALLOC;
439 data = __netdev_alloc_frag(fragsz, gfp_mask);
442 skb = build_skb(data, fragsz);
444 put_page(virt_to_head_page(data));
447 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
448 SKB_ALLOC_RX, NUMA_NO_NODE);
451 skb_reserve(skb, NET_SKB_PAD);
456 EXPORT_SYMBOL(__netdev_alloc_skb);
458 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
459 int size, unsigned int truesize)
461 skb_fill_page_desc(skb, i, page, off, size);
463 skb->data_len += size;
464 skb->truesize += truesize;
466 EXPORT_SYMBOL(skb_add_rx_frag);
468 static void skb_drop_list(struct sk_buff **listp)
470 struct sk_buff *list = *listp;
475 struct sk_buff *this = list;
481 static inline void skb_drop_fraglist(struct sk_buff *skb)
483 skb_drop_list(&skb_shinfo(skb)->frag_list);
486 static void skb_clone_fraglist(struct sk_buff *skb)
488 struct sk_buff *list;
490 skb_walk_frags(skb, list)
494 static void skb_free_head(struct sk_buff *skb)
497 put_page(virt_to_head_page(skb->head));
502 static void skb_release_data(struct sk_buff *skb)
505 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
506 &skb_shinfo(skb)->dataref)) {
507 if (skb_shinfo(skb)->nr_frags) {
509 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
510 skb_frag_unref(skb, i);
514 * If skb buf is from userspace, we need to notify the caller
515 * the lower device DMA has done;
517 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
518 struct ubuf_info *uarg;
520 uarg = skb_shinfo(skb)->destructor_arg;
522 uarg->callback(uarg);
525 if (skb_has_frag_list(skb))
526 skb_drop_fraglist(skb);
533 * Free an skbuff by memory without cleaning the state.
535 static void kfree_skbmem(struct sk_buff *skb)
537 struct sk_buff *other;
538 atomic_t *fclone_ref;
540 switch (skb->fclone) {
541 case SKB_FCLONE_UNAVAILABLE:
542 kmem_cache_free(skbuff_head_cache, skb);
545 case SKB_FCLONE_ORIG:
546 fclone_ref = (atomic_t *) (skb + 2);
547 if (atomic_dec_and_test(fclone_ref))
548 kmem_cache_free(skbuff_fclone_cache, skb);
551 case SKB_FCLONE_CLONE:
552 fclone_ref = (atomic_t *) (skb + 1);
555 /* The clone portion is available for
556 * fast-cloning again.
558 skb->fclone = SKB_FCLONE_UNAVAILABLE;
560 if (atomic_dec_and_test(fclone_ref))
561 kmem_cache_free(skbuff_fclone_cache, other);
566 static void skb_release_head_state(struct sk_buff *skb)
570 secpath_put(skb->sp);
572 if (skb->destructor) {
574 skb->destructor(skb);
576 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
577 nf_conntrack_put(skb->nfct);
579 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
580 nf_conntrack_put_reasm(skb->nfct_reasm);
582 #ifdef CONFIG_BRIDGE_NETFILTER
583 nf_bridge_put(skb->nf_bridge);
585 /* XXX: IS this still necessary? - JHS */
586 #ifdef CONFIG_NET_SCHED
588 #ifdef CONFIG_NET_CLS_ACT
594 /* Free everything but the sk_buff shell. */
595 static void skb_release_all(struct sk_buff *skb)
597 skb_release_head_state(skb);
598 skb_release_data(skb);
602 * __kfree_skb - private function
605 * Free an sk_buff. Release anything attached to the buffer.
606 * Clean the state. This is an internal helper function. Users should
607 * always call kfree_skb
610 void __kfree_skb(struct sk_buff *skb)
612 skb_release_all(skb);
615 EXPORT_SYMBOL(__kfree_skb);
618 * kfree_skb - free an sk_buff
619 * @skb: buffer to free
621 * Drop a reference to the buffer and free it if the usage count has
624 void kfree_skb(struct sk_buff *skb)
628 if (likely(atomic_read(&skb->users) == 1))
630 else if (likely(!atomic_dec_and_test(&skb->users)))
632 trace_kfree_skb(skb, __builtin_return_address(0));
635 EXPORT_SYMBOL(kfree_skb);
638 * consume_skb - free an skbuff
639 * @skb: buffer to free
641 * Drop a ref to the buffer and free it if the usage count has hit zero
642 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
643 * is being dropped after a failure and notes that
645 void consume_skb(struct sk_buff *skb)
649 if (likely(atomic_read(&skb->users) == 1))
651 else if (likely(!atomic_dec_and_test(&skb->users)))
653 trace_consume_skb(skb);
656 EXPORT_SYMBOL(consume_skb);
659 * skb_recycle - clean up an skb for reuse
662 * Recycles the skb to be reused as a receive buffer. This
663 * function does any necessary reference count dropping, and
664 * cleans up the skbuff as if it just came from __alloc_skb().
666 void skb_recycle(struct sk_buff *skb)
668 struct skb_shared_info *shinfo;
670 skb_release_head_state(skb);
672 shinfo = skb_shinfo(skb);
673 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
674 atomic_set(&shinfo->dataref, 1);
676 memset(skb, 0, offsetof(struct sk_buff, tail));
677 skb->data = skb->head + NET_SKB_PAD;
678 skb_reset_tail_pointer(skb);
680 EXPORT_SYMBOL(skb_recycle);
683 * skb_recycle_check - check if skb can be reused for receive
685 * @skb_size: minimum receive buffer size
687 * Checks that the skb passed in is not shared or cloned, and
688 * that it is linear and its head portion at least as large as
689 * skb_size so that it can be recycled as a receive buffer.
690 * If these conditions are met, this function does any necessary
691 * reference count dropping and cleans up the skbuff as if it
692 * just came from __alloc_skb().
694 bool skb_recycle_check(struct sk_buff *skb, int skb_size)
696 if (!skb_is_recycleable(skb, skb_size))
703 EXPORT_SYMBOL(skb_recycle_check);
705 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
707 new->tstamp = old->tstamp;
709 new->transport_header = old->transport_header;
710 new->network_header = old->network_header;
711 new->mac_header = old->mac_header;
712 skb_dst_copy(new, old);
713 new->rxhash = old->rxhash;
714 new->ooo_okay = old->ooo_okay;
715 new->l4_rxhash = old->l4_rxhash;
716 new->no_fcs = old->no_fcs;
718 new->sp = secpath_get(old->sp);
720 memcpy(new->cb, old->cb, sizeof(old->cb));
721 new->csum = old->csum;
722 new->local_df = old->local_df;
723 new->pkt_type = old->pkt_type;
724 new->ip_summed = old->ip_summed;
725 skb_copy_queue_mapping(new, old);
726 new->priority = old->priority;
727 #if IS_ENABLED(CONFIG_IP_VS)
728 new->ipvs_property = old->ipvs_property;
730 new->pfmemalloc = old->pfmemalloc;
731 new->protocol = old->protocol;
732 new->mark = old->mark;
733 new->skb_iif = old->skb_iif;
735 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
736 new->nf_trace = old->nf_trace;
738 #ifdef CONFIG_NET_SCHED
739 new->tc_index = old->tc_index;
740 #ifdef CONFIG_NET_CLS_ACT
741 new->tc_verd = old->tc_verd;
744 new->vlan_tci = old->vlan_tci;
746 skb_copy_secmark(new, old);
750 * You should not add any new code to this function. Add it to
751 * __copy_skb_header above instead.
753 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
755 #define C(x) n->x = skb->x
757 n->next = n->prev = NULL;
759 __copy_skb_header(n, skb);
764 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
767 n->destructor = NULL;
774 atomic_set(&n->users, 1);
776 atomic_inc(&(skb_shinfo(skb)->dataref));
784 * skb_morph - morph one skb into another
785 * @dst: the skb to receive the contents
786 * @src: the skb to supply the contents
788 * This is identical to skb_clone except that the target skb is
789 * supplied by the user.
791 * The target skb is returned upon exit.
793 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
795 skb_release_all(dst);
796 return __skb_clone(dst, src);
798 EXPORT_SYMBOL_GPL(skb_morph);
801 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
802 * @skb: the skb to modify
803 * @gfp_mask: allocation priority
805 * This must be called on SKBTX_DEV_ZEROCOPY skb.
806 * It will copy all frags into kernel and drop the reference
807 * to userspace pages.
809 * If this function is called from an interrupt gfp_mask() must be
812 * Returns 0 on success or a negative error code on failure
813 * to allocate kernel memory to copy to.
815 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
818 int num_frags = skb_shinfo(skb)->nr_frags;
819 struct page *page, *head = NULL;
820 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
822 for (i = 0; i < num_frags; i++) {
824 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
826 page = alloc_page(gfp_mask);
829 struct page *next = (struct page *)head->private;
835 vaddr = kmap_atomic(skb_frag_page(f));
836 memcpy(page_address(page),
837 vaddr + f->page_offset, skb_frag_size(f));
838 kunmap_atomic(vaddr);
839 page->private = (unsigned long)head;
843 /* skb frags release userspace buffers */
844 for (i = 0; i < num_frags; i++)
845 skb_frag_unref(skb, i);
847 uarg->callback(uarg);
849 /* skb frags point to kernel buffers */
850 for (i = num_frags - 1; i >= 0; i--) {
851 __skb_fill_page_desc(skb, i, head, 0,
852 skb_shinfo(skb)->frags[i].size);
853 head = (struct page *)head->private;
856 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
859 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
862 * skb_clone - duplicate an sk_buff
863 * @skb: buffer to clone
864 * @gfp_mask: allocation priority
866 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
867 * copies share the same packet data but not structure. The new
868 * buffer has a reference count of 1. If the allocation fails the
869 * function returns %NULL otherwise the new buffer is returned.
871 * If this function is called from an interrupt gfp_mask() must be
875 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
879 if (skb_orphan_frags(skb, gfp_mask))
883 if (skb->fclone == SKB_FCLONE_ORIG &&
884 n->fclone == SKB_FCLONE_UNAVAILABLE) {
885 atomic_t *fclone_ref = (atomic_t *) (n + 1);
886 n->fclone = SKB_FCLONE_CLONE;
887 atomic_inc(fclone_ref);
889 if (skb_pfmemalloc(skb))
890 gfp_mask |= __GFP_MEMALLOC;
892 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
896 kmemcheck_annotate_bitfield(n, flags1);
897 kmemcheck_annotate_bitfield(n, flags2);
898 n->fclone = SKB_FCLONE_UNAVAILABLE;
901 return __skb_clone(n, skb);
903 EXPORT_SYMBOL(skb_clone);
905 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
907 #ifndef NET_SKBUFF_DATA_USES_OFFSET
909 * Shift between the two data areas in bytes
911 unsigned long offset = new->data - old->data;
914 __copy_skb_header(new, old);
916 #ifndef NET_SKBUFF_DATA_USES_OFFSET
917 /* {transport,network,mac}_header are relative to skb->head */
918 new->transport_header += offset;
919 new->network_header += offset;
920 if (skb_mac_header_was_set(new))
921 new->mac_header += offset;
923 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
924 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
925 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
928 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
930 if (skb_pfmemalloc(skb))
936 * skb_copy - create private copy of an sk_buff
937 * @skb: buffer to copy
938 * @gfp_mask: allocation priority
940 * Make a copy of both an &sk_buff and its data. This is used when the
941 * caller wishes to modify the data and needs a private copy of the
942 * data to alter. Returns %NULL on failure or the pointer to the buffer
943 * on success. The returned buffer has a reference count of 1.
945 * As by-product this function converts non-linear &sk_buff to linear
946 * one, so that &sk_buff becomes completely private and caller is allowed
947 * to modify all the data of returned buffer. This means that this
948 * function is not recommended for use in circumstances when only
949 * header is going to be modified. Use pskb_copy() instead.
952 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
954 int headerlen = skb_headroom(skb);
955 unsigned int size = skb_end_offset(skb) + skb->data_len;
956 struct sk_buff *n = __alloc_skb(size, gfp_mask,
957 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
962 /* Set the data pointer */
963 skb_reserve(n, headerlen);
964 /* Set the tail pointer and length */
965 skb_put(n, skb->len);
967 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
970 copy_skb_header(n, skb);
973 EXPORT_SYMBOL(skb_copy);
976 * __pskb_copy - create copy of an sk_buff with private head.
977 * @skb: buffer to copy
978 * @headroom: headroom of new skb
979 * @gfp_mask: allocation priority
981 * Make a copy of both an &sk_buff and part of its data, located
982 * in header. Fragmented data remain shared. This is used when
983 * the caller wishes to modify only header of &sk_buff and needs
984 * private copy of the header to alter. Returns %NULL on failure
985 * or the pointer to the buffer on success.
986 * The returned buffer has a reference count of 1.
989 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
991 unsigned int size = skb_headlen(skb) + headroom;
992 struct sk_buff *n = __alloc_skb(size, gfp_mask,
993 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
998 /* Set the data pointer */
999 skb_reserve(n, headroom);
1000 /* Set the tail pointer and length */
1001 skb_put(n, skb_headlen(skb));
1002 /* Copy the bytes */
1003 skb_copy_from_linear_data(skb, n->data, n->len);
1005 n->truesize += skb->data_len;
1006 n->data_len = skb->data_len;
1009 if (skb_shinfo(skb)->nr_frags) {
1012 if (skb_orphan_frags(skb, gfp_mask)) {
1017 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1018 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1019 skb_frag_ref(skb, i);
1021 skb_shinfo(n)->nr_frags = i;
1024 if (skb_has_frag_list(skb)) {
1025 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1026 skb_clone_fraglist(n);
1029 copy_skb_header(n, skb);
1033 EXPORT_SYMBOL(__pskb_copy);
1036 * pskb_expand_head - reallocate header of &sk_buff
1037 * @skb: buffer to reallocate
1038 * @nhead: room to add at head
1039 * @ntail: room to add at tail
1040 * @gfp_mask: allocation priority
1042 * Expands (or creates identical copy, if &nhead and &ntail are zero)
1043 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
1044 * reference count of 1. Returns zero in the case of success or error,
1045 * if expansion failed. In the last case, &sk_buff is not changed.
1047 * All the pointers pointing into skb header may change and must be
1048 * reloaded after call to this function.
1051 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1056 int size = nhead + skb_end_offset(skb) + ntail;
1061 if (skb_shared(skb))
1064 size = SKB_DATA_ALIGN(size);
1066 if (skb_pfmemalloc(skb))
1067 gfp_mask |= __GFP_MEMALLOC;
1068 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1069 gfp_mask, NUMA_NO_NODE, NULL);
1072 size = SKB_WITH_OVERHEAD(ksize(data));
1074 /* Copy only real data... and, alas, header. This should be
1075 * optimized for the cases when header is void.
1077 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1079 memcpy((struct skb_shared_info *)(data + size),
1081 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1084 * if shinfo is shared we must drop the old head gracefully, but if it
1085 * is not we can just drop the old head and let the existing refcount
1086 * be since all we did is relocate the values
1088 if (skb_cloned(skb)) {
1089 /* copy this zero copy skb frags */
1090 if (skb_orphan_frags(skb, gfp_mask))
1092 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1093 skb_frag_ref(skb, i);
1095 if (skb_has_frag_list(skb))
1096 skb_clone_fraglist(skb);
1098 skb_release_data(skb);
1102 off = (data + nhead) - skb->head;
1107 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1111 skb->end = skb->head + size;
1113 /* {transport,network,mac}_header and tail are relative to skb->head */
1115 skb->transport_header += off;
1116 skb->network_header += off;
1117 if (skb_mac_header_was_set(skb))
1118 skb->mac_header += off;
1119 /* Only adjust this if it actually is csum_start rather than csum */
1120 if (skb->ip_summed == CHECKSUM_PARTIAL)
1121 skb->csum_start += nhead;
1125 atomic_set(&skb_shinfo(skb)->dataref, 1);
1133 EXPORT_SYMBOL(pskb_expand_head);
1135 /* Make private copy of skb with writable head and some headroom */
1137 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1139 struct sk_buff *skb2;
1140 int delta = headroom - skb_headroom(skb);
1143 skb2 = pskb_copy(skb, GFP_ATOMIC);
1145 skb2 = skb_clone(skb, GFP_ATOMIC);
1146 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1154 EXPORT_SYMBOL(skb_realloc_headroom);
1157 * skb_copy_expand - copy and expand sk_buff
1158 * @skb: buffer to copy
1159 * @newheadroom: new free bytes at head
1160 * @newtailroom: new free bytes at tail
1161 * @gfp_mask: allocation priority
1163 * Make a copy of both an &sk_buff and its data and while doing so
1164 * allocate additional space.
1166 * This is used when the caller wishes to modify the data and needs a
1167 * private copy of the data to alter as well as more space for new fields.
1168 * Returns %NULL on failure or the pointer to the buffer
1169 * on success. The returned buffer has a reference count of 1.
1171 * You must pass %GFP_ATOMIC as the allocation priority if this function
1172 * is called from an interrupt.
1174 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1175 int newheadroom, int newtailroom,
1179 * Allocate the copy buffer
1181 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1182 gfp_mask, skb_alloc_rx_flag(skb),
1184 int oldheadroom = skb_headroom(skb);
1185 int head_copy_len, head_copy_off;
1191 skb_reserve(n, newheadroom);
1193 /* Set the tail pointer and length */
1194 skb_put(n, skb->len);
1196 head_copy_len = oldheadroom;
1198 if (newheadroom <= head_copy_len)
1199 head_copy_len = newheadroom;
1201 head_copy_off = newheadroom - head_copy_len;
1203 /* Copy the linear header and data. */
1204 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1205 skb->len + head_copy_len))
1208 copy_skb_header(n, skb);
1210 off = newheadroom - oldheadroom;
1211 if (n->ip_summed == CHECKSUM_PARTIAL)
1212 n->csum_start += off;
1213 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1214 n->transport_header += off;
1215 n->network_header += off;
1216 if (skb_mac_header_was_set(skb))
1217 n->mac_header += off;
1222 EXPORT_SYMBOL(skb_copy_expand);
1225 * skb_pad - zero pad the tail of an skb
1226 * @skb: buffer to pad
1227 * @pad: space to pad
1229 * Ensure that a buffer is followed by a padding area that is zero
1230 * filled. Used by network drivers which may DMA or transfer data
1231 * beyond the buffer end onto the wire.
1233 * May return error in out of memory cases. The skb is freed on error.
1236 int skb_pad(struct sk_buff *skb, int pad)
1241 /* If the skbuff is non linear tailroom is always zero.. */
1242 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1243 memset(skb->data+skb->len, 0, pad);
1247 ntail = skb->data_len + pad - (skb->end - skb->tail);
1248 if (likely(skb_cloned(skb) || ntail > 0)) {
1249 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1254 /* FIXME: The use of this function with non-linear skb's really needs
1257 err = skb_linearize(skb);
1261 memset(skb->data + skb->len, 0, pad);
1268 EXPORT_SYMBOL(skb_pad);
1271 * skb_put - add data to a buffer
1272 * @skb: buffer to use
1273 * @len: amount of data to add
1275 * This function extends the used data area of the buffer. If this would
1276 * exceed the total buffer size the kernel will panic. A pointer to the
1277 * first byte of the extra data is returned.
1279 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1281 unsigned char *tmp = skb_tail_pointer(skb);
1282 SKB_LINEAR_ASSERT(skb);
1285 if (unlikely(skb->tail > skb->end))
1286 skb_over_panic(skb, len, __builtin_return_address(0));
1289 EXPORT_SYMBOL(skb_put);
1292 * skb_push - add data to the start of a buffer
1293 * @skb: buffer to use
1294 * @len: amount of data to add
1296 * This function extends the used data area of the buffer at the buffer
1297 * start. If this would exceed the total buffer headroom the kernel will
1298 * panic. A pointer to the first byte of the extra data is returned.
1300 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1304 if (unlikely(skb->data<skb->head))
1305 skb_under_panic(skb, len, __builtin_return_address(0));
1308 EXPORT_SYMBOL(skb_push);
1311 * skb_pull - remove data from the start of a buffer
1312 * @skb: buffer to use
1313 * @len: amount of data to remove
1315 * This function removes data from the start of a buffer, returning
1316 * the memory to the headroom. A pointer to the next data in the buffer
1317 * is returned. Once the data has been pulled future pushes will overwrite
1320 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1322 return skb_pull_inline(skb, len);
1324 EXPORT_SYMBOL(skb_pull);
1327 * skb_trim - remove end from a buffer
1328 * @skb: buffer to alter
1331 * Cut the length of a buffer down by removing data from the tail. If
1332 * the buffer is already under the length specified it is not modified.
1333 * The skb must be linear.
1335 void skb_trim(struct sk_buff *skb, unsigned int len)
1338 __skb_trim(skb, len);
1340 EXPORT_SYMBOL(skb_trim);
1342 /* Trims skb to length len. It can change skb pointers.
1345 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1347 struct sk_buff **fragp;
1348 struct sk_buff *frag;
1349 int offset = skb_headlen(skb);
1350 int nfrags = skb_shinfo(skb)->nr_frags;
1354 if (skb_cloned(skb) &&
1355 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1362 for (; i < nfrags; i++) {
1363 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1370 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1373 skb_shinfo(skb)->nr_frags = i;
1375 for (; i < nfrags; i++)
1376 skb_frag_unref(skb, i);
1378 if (skb_has_frag_list(skb))
1379 skb_drop_fraglist(skb);
1383 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1384 fragp = &frag->next) {
1385 int end = offset + frag->len;
1387 if (skb_shared(frag)) {
1388 struct sk_buff *nfrag;
1390 nfrag = skb_clone(frag, GFP_ATOMIC);
1391 if (unlikely(!nfrag))
1394 nfrag->next = frag->next;
1406 unlikely((err = pskb_trim(frag, len - offset))))
1410 skb_drop_list(&frag->next);
1415 if (len > skb_headlen(skb)) {
1416 skb->data_len -= skb->len - len;
1421 skb_set_tail_pointer(skb, len);
1426 EXPORT_SYMBOL(___pskb_trim);
1429 * __pskb_pull_tail - advance tail of skb header
1430 * @skb: buffer to reallocate
1431 * @delta: number of bytes to advance tail
1433 * The function makes a sense only on a fragmented &sk_buff,
1434 * it expands header moving its tail forward and copying necessary
1435 * data from fragmented part.
1437 * &sk_buff MUST have reference count of 1.
1439 * Returns %NULL (and &sk_buff does not change) if pull failed
1440 * or value of new tail of skb in the case of success.
1442 * All the pointers pointing into skb header may change and must be
1443 * reloaded after call to this function.
1446 /* Moves tail of skb head forward, copying data from fragmented part,
1447 * when it is necessary.
1448 * 1. It may fail due to malloc failure.
1449 * 2. It may change skb pointers.
1451 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1453 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1455 /* If skb has not enough free space at tail, get new one
1456 * plus 128 bytes for future expansions. If we have enough
1457 * room at tail, reallocate without expansion only if skb is cloned.
1459 int i, k, eat = (skb->tail + delta) - skb->end;
1461 if (eat > 0 || skb_cloned(skb)) {
1462 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1467 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1470 /* Optimization: no fragments, no reasons to preestimate
1471 * size of pulled pages. Superb.
1473 if (!skb_has_frag_list(skb))
1476 /* Estimate size of pulled pages. */
1478 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1479 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1486 /* If we need update frag list, we are in troubles.
1487 * Certainly, it possible to add an offset to skb data,
1488 * but taking into account that pulling is expected to
1489 * be very rare operation, it is worth to fight against
1490 * further bloating skb head and crucify ourselves here instead.
1491 * Pure masohism, indeed. 8)8)
1494 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1495 struct sk_buff *clone = NULL;
1496 struct sk_buff *insp = NULL;
1501 if (list->len <= eat) {
1502 /* Eaten as whole. */
1507 /* Eaten partially. */
1509 if (skb_shared(list)) {
1510 /* Sucks! We need to fork list. :-( */
1511 clone = skb_clone(list, GFP_ATOMIC);
1517 /* This may be pulled without
1521 if (!pskb_pull(list, eat)) {
1529 /* Free pulled out fragments. */
1530 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1531 skb_shinfo(skb)->frag_list = list->next;
1534 /* And insert new clone at head. */
1537 skb_shinfo(skb)->frag_list = clone;
1540 /* Success! Now we may commit changes to skb data. */
1545 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1546 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1549 skb_frag_unref(skb, i);
1552 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1554 skb_shinfo(skb)->frags[k].page_offset += eat;
1555 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1561 skb_shinfo(skb)->nr_frags = k;
1564 skb->data_len -= delta;
1566 return skb_tail_pointer(skb);
1568 EXPORT_SYMBOL(__pskb_pull_tail);
1571 * skb_copy_bits - copy bits from skb to kernel buffer
1573 * @offset: offset in source
1574 * @to: destination buffer
1575 * @len: number of bytes to copy
1577 * Copy the specified number of bytes from the source skb to the
1578 * destination buffer.
1581 * If its prototype is ever changed,
1582 * check arch/{*}/net/{*}.S files,
1583 * since it is called from BPF assembly code.
1585 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1587 int start = skb_headlen(skb);
1588 struct sk_buff *frag_iter;
1591 if (offset > (int)skb->len - len)
1595 if ((copy = start - offset) > 0) {
1598 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1599 if ((len -= copy) == 0)
1605 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1607 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1609 WARN_ON(start > offset + len);
1611 end = start + skb_frag_size(f);
1612 if ((copy = end - offset) > 0) {
1618 vaddr = kmap_atomic(skb_frag_page(f));
1620 vaddr + f->page_offset + offset - start,
1622 kunmap_atomic(vaddr);
1624 if ((len -= copy) == 0)
1632 skb_walk_frags(skb, frag_iter) {
1635 WARN_ON(start > offset + len);
1637 end = start + frag_iter->len;
1638 if ((copy = end - offset) > 0) {
1641 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1643 if ((len -= copy) == 0)
1657 EXPORT_SYMBOL(skb_copy_bits);
1660 * Callback from splice_to_pipe(), if we need to release some pages
1661 * at the end of the spd in case we error'ed out in filling the pipe.
1663 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1665 put_page(spd->pages[i]);
1668 static struct page *linear_to_page(struct page *page, unsigned int *len,
1669 unsigned int *offset,
1670 struct sk_buff *skb, struct sock *sk)
1672 struct page_frag *pfrag = sk_page_frag(sk);
1674 if (!sk_page_frag_refill(sk, pfrag))
1677 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1679 memcpy(page_address(pfrag->page) + pfrag->offset,
1680 page_address(page) + *offset, *len);
1681 *offset = pfrag->offset;
1682 pfrag->offset += *len;
1687 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1689 unsigned int offset)
1691 return spd->nr_pages &&
1692 spd->pages[spd->nr_pages - 1] == page &&
1693 (spd->partial[spd->nr_pages - 1].offset +
1694 spd->partial[spd->nr_pages - 1].len == offset);
1698 * Fill page/offset/length into spd, if it can hold more pages.
1700 static bool spd_fill_page(struct splice_pipe_desc *spd,
1701 struct pipe_inode_info *pipe, struct page *page,
1702 unsigned int *len, unsigned int offset,
1703 struct sk_buff *skb, bool linear,
1706 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1710 page = linear_to_page(page, len, &offset, skb, sk);
1714 if (spd_can_coalesce(spd, page, offset)) {
1715 spd->partial[spd->nr_pages - 1].len += *len;
1719 spd->pages[spd->nr_pages] = page;
1720 spd->partial[spd->nr_pages].len = *len;
1721 spd->partial[spd->nr_pages].offset = offset;
1727 static inline void __segment_seek(struct page **page, unsigned int *poff,
1728 unsigned int *plen, unsigned int off)
1733 n = *poff / PAGE_SIZE;
1735 *page = nth_page(*page, n);
1737 *poff = *poff % PAGE_SIZE;
1741 static bool __splice_segment(struct page *page, unsigned int poff,
1742 unsigned int plen, unsigned int *off,
1743 unsigned int *len, struct sk_buff *skb,
1744 struct splice_pipe_desc *spd, bool linear,
1746 struct pipe_inode_info *pipe)
1751 /* skip this segment if already processed */
1757 /* ignore any bits we already processed */
1759 __segment_seek(&page, &poff, &plen, *off);
1764 unsigned int flen = min(*len, plen);
1766 /* the linear region may spread across several pages */
1767 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1769 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1772 __segment_seek(&page, &poff, &plen, flen);
1775 } while (*len && plen);
1781 * Map linear and fragment data from the skb to spd. It reports true if the
1782 * pipe is full or if we already spliced the requested length.
1784 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1785 unsigned int *offset, unsigned int *len,
1786 struct splice_pipe_desc *spd, struct sock *sk)
1790 /* map the linear part :
1791 * If skb->head_frag is set, this 'linear' part is backed by a
1792 * fragment, and if the head is not shared with any clones then
1793 * we can avoid a copy since we own the head portion of this page.
1795 if (__splice_segment(virt_to_page(skb->data),
1796 (unsigned long) skb->data & (PAGE_SIZE - 1),
1798 offset, len, skb, spd,
1799 skb_head_is_locked(skb),
1804 * then map the fragments
1806 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1807 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1809 if (__splice_segment(skb_frag_page(f),
1810 f->page_offset, skb_frag_size(f),
1811 offset, len, skb, spd, false, sk, pipe))
1819 * Map data from the skb to a pipe. Should handle both the linear part,
1820 * the fragments, and the frag list. It does NOT handle frag lists within
1821 * the frag list, if such a thing exists. We'd probably need to recurse to
1822 * handle that cleanly.
1824 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1825 struct pipe_inode_info *pipe, unsigned int tlen,
1828 struct partial_page partial[MAX_SKB_FRAGS];
1829 struct page *pages[MAX_SKB_FRAGS];
1830 struct splice_pipe_desc spd = {
1833 .nr_pages_max = MAX_SKB_FRAGS,
1835 .ops = &sock_pipe_buf_ops,
1836 .spd_release = sock_spd_release,
1838 struct sk_buff *frag_iter;
1839 struct sock *sk = skb->sk;
1843 * __skb_splice_bits() only fails if the output has no room left,
1844 * so no point in going over the frag_list for the error case.
1846 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1852 * now see if we have a frag_list to map
1854 skb_walk_frags(skb, frag_iter) {
1857 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1864 * Drop the socket lock, otherwise we have reverse
1865 * locking dependencies between sk_lock and i_mutex
1866 * here as compared to sendfile(). We enter here
1867 * with the socket lock held, and splice_to_pipe() will
1868 * grab the pipe inode lock. For sendfile() emulation,
1869 * we call into ->sendpage() with the i_mutex lock held
1870 * and networking will grab the socket lock.
1873 ret = splice_to_pipe(pipe, &spd);
1881 * skb_store_bits - store bits from kernel buffer to skb
1882 * @skb: destination buffer
1883 * @offset: offset in destination
1884 * @from: source buffer
1885 * @len: number of bytes to copy
1887 * Copy the specified number of bytes from the source buffer to the
1888 * destination skb. This function handles all the messy bits of
1889 * traversing fragment lists and such.
1892 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1894 int start = skb_headlen(skb);
1895 struct sk_buff *frag_iter;
1898 if (offset > (int)skb->len - len)
1901 if ((copy = start - offset) > 0) {
1904 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1905 if ((len -= copy) == 0)
1911 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1912 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1915 WARN_ON(start > offset + len);
1917 end = start + skb_frag_size(frag);
1918 if ((copy = end - offset) > 0) {
1924 vaddr = kmap_atomic(skb_frag_page(frag));
1925 memcpy(vaddr + frag->page_offset + offset - start,
1927 kunmap_atomic(vaddr);
1929 if ((len -= copy) == 0)
1937 skb_walk_frags(skb, frag_iter) {
1940 WARN_ON(start > offset + len);
1942 end = start + frag_iter->len;
1943 if ((copy = end - offset) > 0) {
1946 if (skb_store_bits(frag_iter, offset - start,
1949 if ((len -= copy) == 0)
1962 EXPORT_SYMBOL(skb_store_bits);
1964 /* Checksum skb data. */
1966 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1967 int len, __wsum csum)
1969 int start = skb_headlen(skb);
1970 int i, copy = start - offset;
1971 struct sk_buff *frag_iter;
1974 /* Checksum header. */
1978 csum = csum_partial(skb->data + offset, copy, csum);
1979 if ((len -= copy) == 0)
1985 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1987 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1989 WARN_ON(start > offset + len);
1991 end = start + skb_frag_size(frag);
1992 if ((copy = end - offset) > 0) {
1998 vaddr = kmap_atomic(skb_frag_page(frag));
1999 csum2 = csum_partial(vaddr + frag->page_offset +
2000 offset - start, copy, 0);
2001 kunmap_atomic(vaddr);
2002 csum = csum_block_add(csum, csum2, pos);
2011 skb_walk_frags(skb, frag_iter) {
2014 WARN_ON(start > offset + len);
2016 end = start + frag_iter->len;
2017 if ((copy = end - offset) > 0) {
2021 csum2 = skb_checksum(frag_iter, offset - start,
2023 csum = csum_block_add(csum, csum2, pos);
2024 if ((len -= copy) == 0)
2035 EXPORT_SYMBOL(skb_checksum);
2037 /* Both of above in one bottle. */
2039 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2040 u8 *to, int len, __wsum csum)
2042 int start = skb_headlen(skb);
2043 int i, copy = start - offset;
2044 struct sk_buff *frag_iter;
2051 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2053 if ((len -= copy) == 0)
2060 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2063 WARN_ON(start > offset + len);
2065 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2066 if ((copy = end - offset) > 0) {
2069 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2073 vaddr = kmap_atomic(skb_frag_page(frag));
2074 csum2 = csum_partial_copy_nocheck(vaddr +
2078 kunmap_atomic(vaddr);
2079 csum = csum_block_add(csum, csum2, pos);
2089 skb_walk_frags(skb, frag_iter) {
2093 WARN_ON(start > offset + len);
2095 end = start + frag_iter->len;
2096 if ((copy = end - offset) > 0) {
2099 csum2 = skb_copy_and_csum_bits(frag_iter,
2102 csum = csum_block_add(csum, csum2, pos);
2103 if ((len -= copy) == 0)
2114 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2116 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2121 if (skb->ip_summed == CHECKSUM_PARTIAL)
2122 csstart = skb_checksum_start_offset(skb);
2124 csstart = skb_headlen(skb);
2126 BUG_ON(csstart > skb_headlen(skb));
2128 skb_copy_from_linear_data(skb, to, csstart);
2131 if (csstart != skb->len)
2132 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2133 skb->len - csstart, 0);
2135 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2136 long csstuff = csstart + skb->csum_offset;
2138 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2141 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2144 * skb_dequeue - remove from the head of the queue
2145 * @list: list to dequeue from
2147 * Remove the head of the list. The list lock is taken so the function
2148 * may be used safely with other locking list functions. The head item is
2149 * returned or %NULL if the list is empty.
2152 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2154 unsigned long flags;
2155 struct sk_buff *result;
2157 spin_lock_irqsave(&list->lock, flags);
2158 result = __skb_dequeue(list);
2159 spin_unlock_irqrestore(&list->lock, flags);
2162 EXPORT_SYMBOL(skb_dequeue);
2165 * skb_dequeue_tail - remove from the tail of the queue
2166 * @list: list to dequeue from
2168 * Remove the tail of the list. The list lock is taken so the function
2169 * may be used safely with other locking list functions. The tail item is
2170 * returned or %NULL if the list is empty.
2172 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2174 unsigned long flags;
2175 struct sk_buff *result;
2177 spin_lock_irqsave(&list->lock, flags);
2178 result = __skb_dequeue_tail(list);
2179 spin_unlock_irqrestore(&list->lock, flags);
2182 EXPORT_SYMBOL(skb_dequeue_tail);
2185 * skb_queue_purge - empty a list
2186 * @list: list to empty
2188 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2189 * the list and one reference dropped. This function takes the list
2190 * lock and is atomic with respect to other list locking functions.
2192 void skb_queue_purge(struct sk_buff_head *list)
2194 struct sk_buff *skb;
2195 while ((skb = skb_dequeue(list)) != NULL)
2198 EXPORT_SYMBOL(skb_queue_purge);
2201 * skb_queue_head - queue a buffer at the list head
2202 * @list: list to use
2203 * @newsk: buffer to queue
2205 * Queue a buffer at the start of the list. This function takes the
2206 * list lock and can be used safely with other locking &sk_buff functions
2209 * A buffer cannot be placed on two lists at the same time.
2211 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2213 unsigned long flags;
2215 spin_lock_irqsave(&list->lock, flags);
2216 __skb_queue_head(list, newsk);
2217 spin_unlock_irqrestore(&list->lock, flags);
2219 EXPORT_SYMBOL(skb_queue_head);
2222 * skb_queue_tail - queue a buffer at the list tail
2223 * @list: list to use
2224 * @newsk: buffer to queue
2226 * Queue a buffer at the tail of the list. This function takes the
2227 * list lock and can be used safely with other locking &sk_buff functions
2230 * A buffer cannot be placed on two lists at the same time.
2232 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2234 unsigned long flags;
2236 spin_lock_irqsave(&list->lock, flags);
2237 __skb_queue_tail(list, newsk);
2238 spin_unlock_irqrestore(&list->lock, flags);
2240 EXPORT_SYMBOL(skb_queue_tail);
2243 * skb_unlink - remove a buffer from a list
2244 * @skb: buffer to remove
2245 * @list: list to use
2247 * Remove a packet from a list. The list locks are taken and this
2248 * function is atomic with respect to other list locked calls
2250 * You must know what list the SKB is on.
2252 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2254 unsigned long flags;
2256 spin_lock_irqsave(&list->lock, flags);
2257 __skb_unlink(skb, list);
2258 spin_unlock_irqrestore(&list->lock, flags);
2260 EXPORT_SYMBOL(skb_unlink);
2263 * skb_append - append a buffer
2264 * @old: buffer to insert after
2265 * @newsk: buffer to insert
2266 * @list: list to use
2268 * Place a packet after a given packet in a list. The list locks are taken
2269 * and this function is atomic with respect to other list locked calls.
2270 * A buffer cannot be placed on two lists at the same time.
2272 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2274 unsigned long flags;
2276 spin_lock_irqsave(&list->lock, flags);
2277 __skb_queue_after(list, old, newsk);
2278 spin_unlock_irqrestore(&list->lock, flags);
2280 EXPORT_SYMBOL(skb_append);
2283 * skb_insert - insert a buffer
2284 * @old: buffer to insert before
2285 * @newsk: buffer to insert
2286 * @list: list to use
2288 * Place a packet before a given packet in a list. The list locks are
2289 * taken and this function is atomic with respect to other list locked
2292 * A buffer cannot be placed on two lists at the same time.
2294 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2296 unsigned long flags;
2298 spin_lock_irqsave(&list->lock, flags);
2299 __skb_insert(newsk, old->prev, old, list);
2300 spin_unlock_irqrestore(&list->lock, flags);
2302 EXPORT_SYMBOL(skb_insert);
2304 static inline void skb_split_inside_header(struct sk_buff *skb,
2305 struct sk_buff* skb1,
2306 const u32 len, const int pos)
2310 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2312 /* And move data appendix as is. */
2313 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2314 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2316 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2317 skb_shinfo(skb)->nr_frags = 0;
2318 skb1->data_len = skb->data_len;
2319 skb1->len += skb1->data_len;
2322 skb_set_tail_pointer(skb, len);
2325 static inline void skb_split_no_header(struct sk_buff *skb,
2326 struct sk_buff* skb1,
2327 const u32 len, int pos)
2330 const int nfrags = skb_shinfo(skb)->nr_frags;
2332 skb_shinfo(skb)->nr_frags = 0;
2333 skb1->len = skb1->data_len = skb->len - len;
2335 skb->data_len = len - pos;
2337 for (i = 0; i < nfrags; i++) {
2338 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2340 if (pos + size > len) {
2341 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2345 * We have two variants in this case:
2346 * 1. Move all the frag to the second
2347 * part, if it is possible. F.e.
2348 * this approach is mandatory for TUX,
2349 * where splitting is expensive.
2350 * 2. Split is accurately. We make this.
2352 skb_frag_ref(skb, i);
2353 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2354 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2355 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2356 skb_shinfo(skb)->nr_frags++;
2360 skb_shinfo(skb)->nr_frags++;
2363 skb_shinfo(skb1)->nr_frags = k;
2367 * skb_split - Split fragmented skb to two parts at length len.
2368 * @skb: the buffer to split
2369 * @skb1: the buffer to receive the second part
2370 * @len: new length for skb
2372 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2374 int pos = skb_headlen(skb);
2376 if (len < pos) /* Split line is inside header. */
2377 skb_split_inside_header(skb, skb1, len, pos);
2378 else /* Second chunk has no header, nothing to copy. */
2379 skb_split_no_header(skb, skb1, len, pos);
2381 EXPORT_SYMBOL(skb_split);
2383 /* Shifting from/to a cloned skb is a no-go.
2385 * Caller cannot keep skb_shinfo related pointers past calling here!
2387 static int skb_prepare_for_shift(struct sk_buff *skb)
2389 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2393 * skb_shift - Shifts paged data partially from skb to another
2394 * @tgt: buffer into which tail data gets added
2395 * @skb: buffer from which the paged data comes from
2396 * @shiftlen: shift up to this many bytes
2398 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2399 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2400 * It's up to caller to free skb if everything was shifted.
2402 * If @tgt runs out of frags, the whole operation is aborted.
2404 * Skb cannot include anything else but paged data while tgt is allowed
2405 * to have non-paged data as well.
2407 * TODO: full sized shift could be optimized but that would need
2408 * specialized skb free'er to handle frags without up-to-date nr_frags.
2410 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2412 int from, to, merge, todo;
2413 struct skb_frag_struct *fragfrom, *fragto;
2415 BUG_ON(shiftlen > skb->len);
2416 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2420 to = skb_shinfo(tgt)->nr_frags;
2421 fragfrom = &skb_shinfo(skb)->frags[from];
2423 /* Actual merge is delayed until the point when we know we can
2424 * commit all, so that we don't have to undo partial changes
2427 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2428 fragfrom->page_offset)) {
2433 todo -= skb_frag_size(fragfrom);
2435 if (skb_prepare_for_shift(skb) ||
2436 skb_prepare_for_shift(tgt))
2439 /* All previous frag pointers might be stale! */
2440 fragfrom = &skb_shinfo(skb)->frags[from];
2441 fragto = &skb_shinfo(tgt)->frags[merge];
2443 skb_frag_size_add(fragto, shiftlen);
2444 skb_frag_size_sub(fragfrom, shiftlen);
2445 fragfrom->page_offset += shiftlen;
2453 /* Skip full, not-fitting skb to avoid expensive operations */
2454 if ((shiftlen == skb->len) &&
2455 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2458 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2461 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2462 if (to == MAX_SKB_FRAGS)
2465 fragfrom = &skb_shinfo(skb)->frags[from];
2466 fragto = &skb_shinfo(tgt)->frags[to];
2468 if (todo >= skb_frag_size(fragfrom)) {
2469 *fragto = *fragfrom;
2470 todo -= skb_frag_size(fragfrom);
2475 __skb_frag_ref(fragfrom);
2476 fragto->page = fragfrom->page;
2477 fragto->page_offset = fragfrom->page_offset;
2478 skb_frag_size_set(fragto, todo);
2480 fragfrom->page_offset += todo;
2481 skb_frag_size_sub(fragfrom, todo);
2489 /* Ready to "commit" this state change to tgt */
2490 skb_shinfo(tgt)->nr_frags = to;
2493 fragfrom = &skb_shinfo(skb)->frags[0];
2494 fragto = &skb_shinfo(tgt)->frags[merge];
2496 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2497 __skb_frag_unref(fragfrom);
2500 /* Reposition in the original skb */
2502 while (from < skb_shinfo(skb)->nr_frags)
2503 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2504 skb_shinfo(skb)->nr_frags = to;
2506 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2509 /* Most likely the tgt won't ever need its checksum anymore, skb on
2510 * the other hand might need it if it needs to be resent
2512 tgt->ip_summed = CHECKSUM_PARTIAL;
2513 skb->ip_summed = CHECKSUM_PARTIAL;
2515 /* Yak, is it really working this way? Some helper please? */
2516 skb->len -= shiftlen;
2517 skb->data_len -= shiftlen;
2518 skb->truesize -= shiftlen;
2519 tgt->len += shiftlen;
2520 tgt->data_len += shiftlen;
2521 tgt->truesize += shiftlen;
2527 * skb_prepare_seq_read - Prepare a sequential read of skb data
2528 * @skb: the buffer to read
2529 * @from: lower offset of data to be read
2530 * @to: upper offset of data to be read
2531 * @st: state variable
2533 * Initializes the specified state variable. Must be called before
2534 * invoking skb_seq_read() for the first time.
2536 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2537 unsigned int to, struct skb_seq_state *st)
2539 st->lower_offset = from;
2540 st->upper_offset = to;
2541 st->root_skb = st->cur_skb = skb;
2542 st->frag_idx = st->stepped_offset = 0;
2543 st->frag_data = NULL;
2545 EXPORT_SYMBOL(skb_prepare_seq_read);
2548 * skb_seq_read - Sequentially read skb data
2549 * @consumed: number of bytes consumed by the caller so far
2550 * @data: destination pointer for data to be returned
2551 * @st: state variable
2553 * Reads a block of skb data at &consumed relative to the
2554 * lower offset specified to skb_prepare_seq_read(). Assigns
2555 * the head of the data block to &data and returns the length
2556 * of the block or 0 if the end of the skb data or the upper
2557 * offset has been reached.
2559 * The caller is not required to consume all of the data
2560 * returned, i.e. &consumed is typically set to the number
2561 * of bytes already consumed and the next call to
2562 * skb_seq_read() will return the remaining part of the block.
2564 * Note 1: The size of each block of data returned can be arbitrary,
2565 * this limitation is the cost for zerocopy seqeuental
2566 * reads of potentially non linear data.
2568 * Note 2: Fragment lists within fragments are not implemented
2569 * at the moment, state->root_skb could be replaced with
2570 * a stack for this purpose.
2572 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2573 struct skb_seq_state *st)
2575 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2578 if (unlikely(abs_offset >= st->upper_offset))
2582 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2584 if (abs_offset < block_limit && !st->frag_data) {
2585 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2586 return block_limit - abs_offset;
2589 if (st->frag_idx == 0 && !st->frag_data)
2590 st->stepped_offset += skb_headlen(st->cur_skb);
2592 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2593 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2594 block_limit = skb_frag_size(frag) + st->stepped_offset;
2596 if (abs_offset < block_limit) {
2598 st->frag_data = kmap_atomic(skb_frag_page(frag));
2600 *data = (u8 *) st->frag_data + frag->page_offset +
2601 (abs_offset - st->stepped_offset);
2603 return block_limit - abs_offset;
2606 if (st->frag_data) {
2607 kunmap_atomic(st->frag_data);
2608 st->frag_data = NULL;
2612 st->stepped_offset += skb_frag_size(frag);
2615 if (st->frag_data) {
2616 kunmap_atomic(st->frag_data);
2617 st->frag_data = NULL;
2620 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2621 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2624 } else if (st->cur_skb->next) {
2625 st->cur_skb = st->cur_skb->next;
2632 EXPORT_SYMBOL(skb_seq_read);
2635 * skb_abort_seq_read - Abort a sequential read of skb data
2636 * @st: state variable
2638 * Must be called if skb_seq_read() was not called until it
2641 void skb_abort_seq_read(struct skb_seq_state *st)
2644 kunmap_atomic(st->frag_data);
2646 EXPORT_SYMBOL(skb_abort_seq_read);
2648 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2650 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2651 struct ts_config *conf,
2652 struct ts_state *state)
2654 return skb_seq_read(offset, text, TS_SKB_CB(state));
2657 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2659 skb_abort_seq_read(TS_SKB_CB(state));
2663 * skb_find_text - Find a text pattern in skb data
2664 * @skb: the buffer to look in
2665 * @from: search offset
2667 * @config: textsearch configuration
2668 * @state: uninitialized textsearch state variable
2670 * Finds a pattern in the skb data according to the specified
2671 * textsearch configuration. Use textsearch_next() to retrieve
2672 * subsequent occurrences of the pattern. Returns the offset
2673 * to the first occurrence or UINT_MAX if no match was found.
2675 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2676 unsigned int to, struct ts_config *config,
2677 struct ts_state *state)
2681 config->get_next_block = skb_ts_get_next_block;
2682 config->finish = skb_ts_finish;
2684 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2686 ret = textsearch_find(config, state);
2687 return (ret <= to - from ? ret : UINT_MAX);
2689 EXPORT_SYMBOL(skb_find_text);
2692 * skb_append_datato_frags - append the user data to a skb
2693 * @sk: sock structure
2694 * @skb: skb structure to be appened with user data.
2695 * @getfrag: call back function to be used for getting the user data
2696 * @from: pointer to user message iov
2697 * @length: length of the iov message
2699 * Description: This procedure append the user data in the fragment part
2700 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2702 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2703 int (*getfrag)(void *from, char *to, int offset,
2704 int len, int odd, struct sk_buff *skb),
2705 void *from, int length)
2708 skb_frag_t *frag = NULL;
2709 struct page *page = NULL;
2715 /* Return error if we don't have space for new frag */
2716 frg_cnt = skb_shinfo(skb)->nr_frags;
2717 if (frg_cnt >= MAX_SKB_FRAGS)
2720 /* allocate a new page for next frag */
2721 page = alloc_pages(sk->sk_allocation, 0);
2723 /* If alloc_page fails just return failure and caller will
2724 * free previous allocated pages by doing kfree_skb()
2729 /* initialize the next frag */
2730 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2731 skb->truesize += PAGE_SIZE;
2732 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2734 /* get the new initialized frag */
2735 frg_cnt = skb_shinfo(skb)->nr_frags;
2736 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2738 /* copy the user data to page */
2739 left = PAGE_SIZE - frag->page_offset;
2740 copy = (length > left)? left : length;
2742 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
2743 offset, copy, 0, skb);
2747 /* copy was successful so update the size parameters */
2748 skb_frag_size_add(frag, copy);
2750 skb->data_len += copy;
2754 } while (length > 0);
2758 EXPORT_SYMBOL(skb_append_datato_frags);
2761 * skb_pull_rcsum - pull skb and update receive checksum
2762 * @skb: buffer to update
2763 * @len: length of data pulled
2765 * This function performs an skb_pull on the packet and updates
2766 * the CHECKSUM_COMPLETE checksum. It should be used on
2767 * receive path processing instead of skb_pull unless you know
2768 * that the checksum difference is zero (e.g., a valid IP header)
2769 * or you are setting ip_summed to CHECKSUM_NONE.
2771 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2773 BUG_ON(len > skb->len);
2775 BUG_ON(skb->len < skb->data_len);
2776 skb_postpull_rcsum(skb, skb->data, len);
2777 return skb->data += len;
2779 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2782 * skb_segment - Perform protocol segmentation on skb.
2783 * @skb: buffer to segment
2784 * @features: features for the output path (see dev->features)
2786 * This function performs segmentation on the given skb. It returns
2787 * a pointer to the first in a list of new skbs for the segments.
2788 * In case of error it returns ERR_PTR(err).
2790 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2792 struct sk_buff *segs = NULL;
2793 struct sk_buff *tail = NULL;
2794 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2795 unsigned int mss = skb_shinfo(skb)->gso_size;
2796 unsigned int doffset = skb->data - skb_mac_header(skb);
2797 unsigned int offset = doffset;
2798 unsigned int headroom;
2800 int sg = !!(features & NETIF_F_SG);
2801 int nfrags = skb_shinfo(skb)->nr_frags;
2806 __skb_push(skb, doffset);
2807 headroom = skb_headroom(skb);
2808 pos = skb_headlen(skb);
2811 struct sk_buff *nskb;
2816 len = skb->len - offset;
2820 hsize = skb_headlen(skb) - offset;
2823 if (hsize > len || !sg)
2826 if (!hsize && i >= nfrags) {
2827 BUG_ON(fskb->len != len);
2830 nskb = skb_clone(fskb, GFP_ATOMIC);
2833 if (unlikely(!nskb))
2836 hsize = skb_end_offset(nskb);
2837 if (skb_cow_head(nskb, doffset + headroom)) {
2842 nskb->truesize += skb_end_offset(nskb) - hsize;
2843 skb_release_head_state(nskb);
2844 __skb_push(nskb, doffset);
2846 nskb = __alloc_skb(hsize + doffset + headroom,
2847 GFP_ATOMIC, skb_alloc_rx_flag(skb),
2850 if (unlikely(!nskb))
2853 skb_reserve(nskb, headroom);
2854 __skb_put(nskb, doffset);
2863 __copy_skb_header(nskb, skb);
2864 nskb->mac_len = skb->mac_len;
2866 /* nskb and skb might have different headroom */
2867 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2868 nskb->csum_start += skb_headroom(nskb) - headroom;
2870 skb_reset_mac_header(nskb);
2871 skb_set_network_header(nskb, skb->mac_len);
2872 nskb->transport_header = (nskb->network_header +
2873 skb_network_header_len(skb));
2874 skb_copy_from_linear_data(skb, nskb->data, doffset);
2876 if (fskb != skb_shinfo(skb)->frag_list)
2880 nskb->ip_summed = CHECKSUM_NONE;
2881 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2887 frag = skb_shinfo(nskb)->frags;
2889 skb_copy_from_linear_data_offset(skb, offset,
2890 skb_put(nskb, hsize), hsize);
2892 while (pos < offset + len && i < nfrags) {
2893 *frag = skb_shinfo(skb)->frags[i];
2894 __skb_frag_ref(frag);
2895 size = skb_frag_size(frag);
2898 frag->page_offset += offset - pos;
2899 skb_frag_size_sub(frag, offset - pos);
2902 skb_shinfo(nskb)->nr_frags++;
2904 if (pos + size <= offset + len) {
2908 skb_frag_size_sub(frag, pos + size - (offset + len));
2915 if (pos < offset + len) {
2916 struct sk_buff *fskb2 = fskb;
2918 BUG_ON(pos + fskb->len != offset + len);
2924 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2930 SKB_FRAG_ASSERT(nskb);
2931 skb_shinfo(nskb)->frag_list = fskb2;
2935 nskb->data_len = len - hsize;
2936 nskb->len += nskb->data_len;
2937 nskb->truesize += nskb->data_len;
2938 } while ((offset += len) < skb->len);
2943 while ((skb = segs)) {
2947 return ERR_PTR(err);
2949 EXPORT_SYMBOL_GPL(skb_segment);
2951 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2953 struct sk_buff *p = *head;
2954 struct sk_buff *nskb;
2955 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2956 struct skb_shared_info *pinfo = skb_shinfo(p);
2957 unsigned int headroom;
2958 unsigned int len = skb_gro_len(skb);
2959 unsigned int offset = skb_gro_offset(skb);
2960 unsigned int headlen = skb_headlen(skb);
2961 unsigned int delta_truesize;
2963 if (p->len + len >= 65536)
2966 if (pinfo->frag_list)
2968 else if (headlen <= offset) {
2971 int i = skbinfo->nr_frags;
2972 int nr_frags = pinfo->nr_frags + i;
2976 if (nr_frags > MAX_SKB_FRAGS)
2979 pinfo->nr_frags = nr_frags;
2980 skbinfo->nr_frags = 0;
2982 frag = pinfo->frags + nr_frags;
2983 frag2 = skbinfo->frags + i;
2988 frag->page_offset += offset;
2989 skb_frag_size_sub(frag, offset);
2991 /* all fragments truesize : remove (head size + sk_buff) */
2992 delta_truesize = skb->truesize -
2993 SKB_TRUESIZE(skb_end_offset(skb));
2995 skb->truesize -= skb->data_len;
2996 skb->len -= skb->data_len;
2999 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3001 } else if (skb->head_frag) {
3002 int nr_frags = pinfo->nr_frags;
3003 skb_frag_t *frag = pinfo->frags + nr_frags;
3004 struct page *page = virt_to_head_page(skb->head);
3005 unsigned int first_size = headlen - offset;
3006 unsigned int first_offset;
3008 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3011 first_offset = skb->data -
3012 (unsigned char *)page_address(page) +
3015 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3017 frag->page.p = page;
3018 frag->page_offset = first_offset;
3019 skb_frag_size_set(frag, first_size);
3021 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3022 /* We dont need to clear skbinfo->nr_frags here */
3024 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3025 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3027 } else if (skb_gro_len(p) != pinfo->gso_size)
3030 headroom = skb_headroom(p);
3031 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
3032 if (unlikely(!nskb))
3035 __copy_skb_header(nskb, p);
3036 nskb->mac_len = p->mac_len;
3038 skb_reserve(nskb, headroom);
3039 __skb_put(nskb, skb_gro_offset(p));
3041 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3042 skb_set_network_header(nskb, skb_network_offset(p));
3043 skb_set_transport_header(nskb, skb_transport_offset(p));
3045 __skb_pull(p, skb_gro_offset(p));
3046 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3047 p->data - skb_mac_header(p));
3049 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
3050 skb_shinfo(nskb)->frag_list = p;
3051 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3052 pinfo->gso_size = 0;
3053 skb_header_release(p);
3056 nskb->data_len += p->len;
3057 nskb->truesize += p->truesize;
3058 nskb->len += p->len;
3061 nskb->next = p->next;
3067 delta_truesize = skb->truesize;
3068 if (offset > headlen) {
3069 unsigned int eat = offset - headlen;
3071 skbinfo->frags[0].page_offset += eat;
3072 skb_frag_size_sub(&skbinfo->frags[0], eat);
3073 skb->data_len -= eat;
3078 __skb_pull(skb, offset);
3080 p->prev->next = skb;
3082 skb_header_release(skb);
3085 NAPI_GRO_CB(p)->count++;
3087 p->truesize += delta_truesize;
3090 NAPI_GRO_CB(skb)->same_flow = 1;
3093 EXPORT_SYMBOL_GPL(skb_gro_receive);
3095 void __init skb_init(void)
3097 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3098 sizeof(struct sk_buff),
3100 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3102 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3103 (2*sizeof(struct sk_buff)) +
3106 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3111 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3112 * @skb: Socket buffer containing the buffers to be mapped
3113 * @sg: The scatter-gather list to map into
3114 * @offset: The offset into the buffer's contents to start mapping
3115 * @len: Length of buffer space to be mapped
3117 * Fill the specified scatter-gather list with mappings/pointers into a
3118 * region of the buffer space attached to a socket buffer.
3121 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3123 int start = skb_headlen(skb);
3124 int i, copy = start - offset;
3125 struct sk_buff *frag_iter;
3131 sg_set_buf(sg, skb->data + offset, copy);
3133 if ((len -= copy) == 0)
3138 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3141 WARN_ON(start > offset + len);
3143 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3144 if ((copy = end - offset) > 0) {
3145 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3149 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3150 frag->page_offset+offset-start);
3159 skb_walk_frags(skb, frag_iter) {
3162 WARN_ON(start > offset + len);
3164 end = start + frag_iter->len;
3165 if ((copy = end - offset) > 0) {
3168 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3170 if ((len -= copy) == 0)
3180 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3182 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3184 sg_mark_end(&sg[nsg - 1]);
3188 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3191 * skb_cow_data - Check that a socket buffer's data buffers are writable
3192 * @skb: The socket buffer to check.
3193 * @tailbits: Amount of trailing space to be added
3194 * @trailer: Returned pointer to the skb where the @tailbits space begins
3196 * Make sure that the data buffers attached to a socket buffer are
3197 * writable. If they are not, private copies are made of the data buffers
3198 * and the socket buffer is set to use these instead.
3200 * If @tailbits is given, make sure that there is space to write @tailbits
3201 * bytes of data beyond current end of socket buffer. @trailer will be
3202 * set to point to the skb in which this space begins.
3204 * The number of scatterlist elements required to completely map the
3205 * COW'd and extended socket buffer will be returned.
3207 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3211 struct sk_buff *skb1, **skb_p;
3213 /* If skb is cloned or its head is paged, reallocate
3214 * head pulling out all the pages (pages are considered not writable
3215 * at the moment even if they are anonymous).
3217 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3218 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3221 /* Easy case. Most of packets will go this way. */
3222 if (!skb_has_frag_list(skb)) {
3223 /* A little of trouble, not enough of space for trailer.
3224 * This should not happen, when stack is tuned to generate
3225 * good frames. OK, on miss we reallocate and reserve even more
3226 * space, 128 bytes is fair. */
3228 if (skb_tailroom(skb) < tailbits &&
3229 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3237 /* Misery. We are in troubles, going to mincer fragments... */
3240 skb_p = &skb_shinfo(skb)->frag_list;
3243 while ((skb1 = *skb_p) != NULL) {
3246 /* The fragment is partially pulled by someone,
3247 * this can happen on input. Copy it and everything
3250 if (skb_shared(skb1))
3253 /* If the skb is the last, worry about trailer. */
3255 if (skb1->next == NULL && tailbits) {
3256 if (skb_shinfo(skb1)->nr_frags ||
3257 skb_has_frag_list(skb1) ||
3258 skb_tailroom(skb1) < tailbits)
3259 ntail = tailbits + 128;
3265 skb_shinfo(skb1)->nr_frags ||
3266 skb_has_frag_list(skb1)) {
3267 struct sk_buff *skb2;
3269 /* Fuck, we are miserable poor guys... */
3271 skb2 = skb_copy(skb1, GFP_ATOMIC);
3273 skb2 = skb_copy_expand(skb1,
3277 if (unlikely(skb2 == NULL))
3281 skb_set_owner_w(skb2, skb1->sk);
3283 /* Looking around. Are we still alive?
3284 * OK, link new skb, drop old one */
3286 skb2->next = skb1->next;
3293 skb_p = &skb1->next;
3298 EXPORT_SYMBOL_GPL(skb_cow_data);
3300 static void sock_rmem_free(struct sk_buff *skb)
3302 struct sock *sk = skb->sk;
3304 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3308 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3310 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3314 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3315 (unsigned int)sk->sk_rcvbuf)
3320 skb->destructor = sock_rmem_free;
3321 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3323 /* before exiting rcu section, make sure dst is refcounted */
3326 skb_queue_tail(&sk->sk_error_queue, skb);
3327 if (!sock_flag(sk, SOCK_DEAD))
3328 sk->sk_data_ready(sk, len);
3331 EXPORT_SYMBOL(sock_queue_err_skb);
3333 void skb_tstamp_tx(struct sk_buff *orig_skb,
3334 struct skb_shared_hwtstamps *hwtstamps)
3336 struct sock *sk = orig_skb->sk;
3337 struct sock_exterr_skb *serr;
3338 struct sk_buff *skb;
3344 skb = skb_clone(orig_skb, GFP_ATOMIC);
3349 *skb_hwtstamps(skb) =
3353 * no hardware time stamps available,
3354 * so keep the shared tx_flags and only
3355 * store software time stamp
3357 skb->tstamp = ktime_get_real();
3360 serr = SKB_EXT_ERR(skb);
3361 memset(serr, 0, sizeof(*serr));
3362 serr->ee.ee_errno = ENOMSG;
3363 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3365 err = sock_queue_err_skb(sk, skb);
3370 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3372 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3374 struct sock *sk = skb->sk;
3375 struct sock_exterr_skb *serr;
3378 skb->wifi_acked_valid = 1;
3379 skb->wifi_acked = acked;
3381 serr = SKB_EXT_ERR(skb);
3382 memset(serr, 0, sizeof(*serr));
3383 serr->ee.ee_errno = ENOMSG;
3384 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3386 err = sock_queue_err_skb(sk, skb);
3390 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3394 * skb_partial_csum_set - set up and verify partial csum values for packet
3395 * @skb: the skb to set
3396 * @start: the number of bytes after skb->data to start checksumming.
3397 * @off: the offset from start to place the checksum.
3399 * For untrusted partially-checksummed packets, we need to make sure the values
3400 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3402 * This function checks and sets those values and skb->ip_summed: if this
3403 * returns false you should drop the packet.
3405 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3407 if (unlikely(start > skb_headlen(skb)) ||
3408 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3409 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3410 start, off, skb_headlen(skb));
3413 skb->ip_summed = CHECKSUM_PARTIAL;
3414 skb->csum_start = skb_headroom(skb) + start;
3415 skb->csum_offset = off;
3418 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3420 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3422 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3425 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3427 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3430 kmem_cache_free(skbuff_head_cache, skb);
3434 EXPORT_SYMBOL(kfree_skb_partial);
3437 * skb_try_coalesce - try to merge skb to prior one
3439 * @from: buffer to add
3440 * @fragstolen: pointer to boolean
3441 * @delta_truesize: how much more was allocated than was requested
3443 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3444 bool *fragstolen, int *delta_truesize)
3446 int i, delta, len = from->len;
3448 *fragstolen = false;
3453 if (len <= skb_tailroom(to)) {
3454 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3455 *delta_truesize = 0;
3459 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3462 if (skb_headlen(from) != 0) {
3464 unsigned int offset;
3466 if (skb_shinfo(to)->nr_frags +
3467 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3470 if (skb_head_is_locked(from))
3473 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3475 page = virt_to_head_page(from->head);
3476 offset = from->data - (unsigned char *)page_address(page);
3478 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3479 page, offset, skb_headlen(from));
3482 if (skb_shinfo(to)->nr_frags +
3483 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3486 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
3489 WARN_ON_ONCE(delta < len);
3491 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3492 skb_shinfo(from)->frags,
3493 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3494 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3496 if (!skb_cloned(from))
3497 skb_shinfo(from)->nr_frags = 0;
3499 /* if the skb is not cloned this does nothing
3500 * since we set nr_frags to 0.
3502 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3503 skb_frag_ref(from, i);
3505 to->truesize += delta;
3507 to->data_len += len;
3509 *delta_truesize = delta;
3512 EXPORT_SYMBOL(skb_try_coalesce);