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
182 struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
187 skb = kmem_cache_alloc_node(skbuff_head_cache,
188 gfp_mask & ~__GFP_DMA, node);
193 * Only clear those fields we need to clear, not those that we will
194 * actually initialise below. Hence, don't put any more fields after
195 * the tail pointer in struct sk_buff!
197 memset(skb, 0, offsetof(struct sk_buff, tail));
199 skb->truesize = sizeof(struct sk_buff);
200 atomic_set(&skb->users, 1);
202 skb->mac_header = (typeof(skb->mac_header))~0U;
208 * __alloc_skb - allocate a network buffer
209 * @size: size to allocate
210 * @gfp_mask: allocation mask
211 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
212 * instead of head cache and allocate a cloned (child) skb.
213 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
214 * allocations in case the data is required for writeback
215 * @node: numa node to allocate memory on
217 * Allocate a new &sk_buff. The returned buffer has no headroom and a
218 * tail room of at least size bytes. The object has a reference count
219 * of one. The return is the buffer. On a failure the return is %NULL.
221 * Buffers may only be allocated from interrupts using a @gfp_mask of
224 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
227 struct kmem_cache *cache;
228 struct skb_shared_info *shinfo;
233 cache = (flags & SKB_ALLOC_FCLONE)
234 ? skbuff_fclone_cache : skbuff_head_cache;
236 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
237 gfp_mask |= __GFP_MEMALLOC;
240 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
245 /* We do our best to align skb_shared_info on a separate cache
246 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
247 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
248 * Both skb->head and skb_shared_info are cache line aligned.
250 size = SKB_DATA_ALIGN(size);
251 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
252 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
255 /* kmalloc(size) might give us more room than requested.
256 * Put skb_shared_info exactly at the end of allocated zone,
257 * to allow max possible filling before reallocation.
259 size = SKB_WITH_OVERHEAD(ksize(data));
260 prefetchw(data + size);
263 * Only clear those fields we need to clear, not those that we will
264 * actually initialise below. Hence, don't put any more fields after
265 * the tail pointer in struct sk_buff!
267 memset(skb, 0, offsetof(struct sk_buff, tail));
268 /* Account for allocated memory : skb + skb->head */
269 skb->truesize = SKB_TRUESIZE(size);
270 skb->pfmemalloc = pfmemalloc;
271 atomic_set(&skb->users, 1);
274 skb_reset_tail_pointer(skb);
275 skb->end = skb->tail + size;
276 skb->mac_header = (typeof(skb->mac_header))~0U;
277 skb->transport_header = (typeof(skb->transport_header))~0U;
279 /* make sure we initialize shinfo sequentially */
280 shinfo = skb_shinfo(skb);
281 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
282 atomic_set(&shinfo->dataref, 1);
283 kmemcheck_annotate_variable(shinfo->destructor_arg);
285 if (flags & SKB_ALLOC_FCLONE) {
286 struct sk_buff *child = skb + 1;
287 atomic_t *fclone_ref = (atomic_t *) (child + 1);
289 kmemcheck_annotate_bitfield(child, flags1);
290 kmemcheck_annotate_bitfield(child, flags2);
291 skb->fclone = SKB_FCLONE_ORIG;
292 atomic_set(fclone_ref, 1);
294 child->fclone = SKB_FCLONE_UNAVAILABLE;
295 child->pfmemalloc = pfmemalloc;
300 kmem_cache_free(cache, skb);
304 EXPORT_SYMBOL(__alloc_skb);
307 * build_skb - build a network buffer
308 * @data: data buffer provided by caller
309 * @frag_size: size of fragment, or 0 if head was kmalloced
311 * Allocate a new &sk_buff. Caller provides space holding head and
312 * skb_shared_info. @data must have been allocated by kmalloc()
313 * The return is the new skb buffer.
314 * On a failure the return is %NULL, and @data is not freed.
316 * Before IO, driver allocates only data buffer where NIC put incoming frame
317 * Driver should add room at head (NET_SKB_PAD) and
318 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
319 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
320 * before giving packet to stack.
321 * RX rings only contains data buffers, not full skbs.
323 struct sk_buff *build_skb(void *data, unsigned int frag_size)
325 struct skb_shared_info *shinfo;
327 unsigned int size = frag_size ? : ksize(data);
329 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
333 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
335 memset(skb, 0, offsetof(struct sk_buff, tail));
336 skb->truesize = SKB_TRUESIZE(size);
337 skb->head_frag = frag_size != 0;
338 atomic_set(&skb->users, 1);
341 skb_reset_tail_pointer(skb);
342 skb->end = skb->tail + size;
343 skb->mac_header = (typeof(skb->mac_header))~0U;
344 skb->transport_header = (typeof(skb->transport_header))~0U;
346 /* make sure we initialize shinfo sequentially */
347 shinfo = skb_shinfo(skb);
348 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
349 atomic_set(&shinfo->dataref, 1);
350 kmemcheck_annotate_variable(shinfo->destructor_arg);
354 EXPORT_SYMBOL(build_skb);
356 struct netdev_alloc_cache {
357 struct page_frag frag;
358 /* we maintain a pagecount bias, so that we dont dirty cache line
359 * containing page->_count every time we allocate a fragment.
361 unsigned int pagecnt_bias;
363 static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
365 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
367 struct netdev_alloc_cache *nc;
372 local_irq_save(flags);
373 nc = &__get_cpu_var(netdev_alloc_cache);
374 if (unlikely(!nc->frag.page)) {
376 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
377 gfp_t gfp = gfp_mask;
380 gfp |= __GFP_COMP | __GFP_NOWARN;
381 nc->frag.page = alloc_pages(gfp, order);
382 if (likely(nc->frag.page))
387 nc->frag.size = PAGE_SIZE << order;
389 atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS);
390 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
394 if (nc->frag.offset + fragsz > nc->frag.size) {
395 /* avoid unnecessary locked operations if possible */
396 if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) ||
397 atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count))
402 data = page_address(nc->frag.page) + nc->frag.offset;
403 nc->frag.offset += fragsz;
406 local_irq_restore(flags);
411 * netdev_alloc_frag - allocate a page fragment
412 * @fragsz: fragment size
414 * Allocates a frag from a page for receive buffer.
415 * Uses GFP_ATOMIC allocations.
417 void *netdev_alloc_frag(unsigned int fragsz)
419 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
421 EXPORT_SYMBOL(netdev_alloc_frag);
424 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
425 * @dev: network device to receive on
426 * @length: length to allocate
427 * @gfp_mask: get_free_pages mask, passed to alloc_skb
429 * Allocate a new &sk_buff and assign it a usage count of one. The
430 * buffer has unspecified headroom built in. Users should allocate
431 * the headroom they think they need without accounting for the
432 * built in space. The built in space is used for optimisations.
434 * %NULL is returned if there is no free memory.
436 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
437 unsigned int length, gfp_t gfp_mask)
439 struct sk_buff *skb = NULL;
440 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
441 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
443 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
446 if (sk_memalloc_socks())
447 gfp_mask |= __GFP_MEMALLOC;
449 data = __netdev_alloc_frag(fragsz, gfp_mask);
452 skb = build_skb(data, fragsz);
454 put_page(virt_to_head_page(data));
457 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
458 SKB_ALLOC_RX, NUMA_NO_NODE);
461 skb_reserve(skb, NET_SKB_PAD);
466 EXPORT_SYMBOL(__netdev_alloc_skb);
468 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
469 int size, unsigned int truesize)
471 skb_fill_page_desc(skb, i, page, off, size);
473 skb->data_len += size;
474 skb->truesize += truesize;
476 EXPORT_SYMBOL(skb_add_rx_frag);
478 static void skb_drop_list(struct sk_buff **listp)
480 struct sk_buff *list = *listp;
485 struct sk_buff *this = list;
491 static inline void skb_drop_fraglist(struct sk_buff *skb)
493 skb_drop_list(&skb_shinfo(skb)->frag_list);
496 static void skb_clone_fraglist(struct sk_buff *skb)
498 struct sk_buff *list;
500 skb_walk_frags(skb, list)
504 static void skb_free_head(struct sk_buff *skb)
507 put_page(virt_to_head_page(skb->head));
512 static void skb_release_data(struct sk_buff *skb)
515 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
516 &skb_shinfo(skb)->dataref)) {
517 if (skb_shinfo(skb)->nr_frags) {
519 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
520 skb_frag_unref(skb, i);
524 * If skb buf is from userspace, we need to notify the caller
525 * the lower device DMA has done;
527 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
528 struct ubuf_info *uarg;
530 uarg = skb_shinfo(skb)->destructor_arg;
532 uarg->callback(uarg, true);
535 if (skb_has_frag_list(skb))
536 skb_drop_fraglist(skb);
543 * Free an skbuff by memory without cleaning the state.
545 static void kfree_skbmem(struct sk_buff *skb)
547 struct sk_buff *other;
548 atomic_t *fclone_ref;
550 switch (skb->fclone) {
551 case SKB_FCLONE_UNAVAILABLE:
552 kmem_cache_free(skbuff_head_cache, skb);
555 case SKB_FCLONE_ORIG:
556 fclone_ref = (atomic_t *) (skb + 2);
557 if (atomic_dec_and_test(fclone_ref))
558 kmem_cache_free(skbuff_fclone_cache, skb);
561 case SKB_FCLONE_CLONE:
562 fclone_ref = (atomic_t *) (skb + 1);
565 /* The clone portion is available for
566 * fast-cloning again.
568 skb->fclone = SKB_FCLONE_UNAVAILABLE;
570 if (atomic_dec_and_test(fclone_ref))
571 kmem_cache_free(skbuff_fclone_cache, other);
576 static void skb_release_head_state(struct sk_buff *skb)
580 secpath_put(skb->sp);
582 if (skb->destructor) {
584 skb->destructor(skb);
586 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
587 nf_conntrack_put(skb->nfct);
589 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
590 nf_conntrack_put_reasm(skb->nfct_reasm);
592 #ifdef CONFIG_BRIDGE_NETFILTER
593 nf_bridge_put(skb->nf_bridge);
595 /* XXX: IS this still necessary? - JHS */
596 #ifdef CONFIG_NET_SCHED
598 #ifdef CONFIG_NET_CLS_ACT
604 /* Free everything but the sk_buff shell. */
605 static void skb_release_all(struct sk_buff *skb)
607 skb_release_head_state(skb);
608 if (likely(skb->head))
609 skb_release_data(skb);
613 * __kfree_skb - private function
616 * Free an sk_buff. Release anything attached to the buffer.
617 * Clean the state. This is an internal helper function. Users should
618 * always call kfree_skb
621 void __kfree_skb(struct sk_buff *skb)
623 skb_release_all(skb);
626 EXPORT_SYMBOL(__kfree_skb);
629 * kfree_skb - free an sk_buff
630 * @skb: buffer to free
632 * Drop a reference to the buffer and free it if the usage count has
635 void kfree_skb(struct sk_buff *skb)
639 if (likely(atomic_read(&skb->users) == 1))
641 else if (likely(!atomic_dec_and_test(&skb->users)))
643 trace_kfree_skb(skb, __builtin_return_address(0));
646 EXPORT_SYMBOL(kfree_skb);
649 * skb_tx_error - report an sk_buff xmit error
650 * @skb: buffer that triggered an error
652 * Report xmit error if a device callback is tracking this skb.
653 * skb must be freed afterwards.
655 void skb_tx_error(struct sk_buff *skb)
657 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
658 struct ubuf_info *uarg;
660 uarg = skb_shinfo(skb)->destructor_arg;
662 uarg->callback(uarg, false);
663 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
666 EXPORT_SYMBOL(skb_tx_error);
669 * consume_skb - free an skbuff
670 * @skb: buffer to free
672 * Drop a ref to the buffer and free it if the usage count has hit zero
673 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
674 * is being dropped after a failure and notes that
676 void consume_skb(struct sk_buff *skb)
680 if (likely(atomic_read(&skb->users) == 1))
682 else if (likely(!atomic_dec_and_test(&skb->users)))
684 trace_consume_skb(skb);
687 EXPORT_SYMBOL(consume_skb);
689 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
691 new->tstamp = old->tstamp;
693 new->transport_header = old->transport_header;
694 new->network_header = old->network_header;
695 new->mac_header = old->mac_header;
696 new->inner_transport_header = old->inner_transport_header;
697 new->inner_network_header = old->inner_network_header;
698 new->inner_mac_header = old->inner_mac_header;
699 skb_dst_copy(new, old);
700 new->rxhash = old->rxhash;
701 new->ooo_okay = old->ooo_okay;
702 new->l4_rxhash = old->l4_rxhash;
703 new->no_fcs = old->no_fcs;
704 new->encapsulation = old->encapsulation;
706 new->sp = secpath_get(old->sp);
708 memcpy(new->cb, old->cb, sizeof(old->cb));
709 new->csum = old->csum;
710 new->local_df = old->local_df;
711 new->pkt_type = old->pkt_type;
712 new->ip_summed = old->ip_summed;
713 skb_copy_queue_mapping(new, old);
714 new->priority = old->priority;
715 #if IS_ENABLED(CONFIG_IP_VS)
716 new->ipvs_property = old->ipvs_property;
718 new->pfmemalloc = old->pfmemalloc;
719 new->protocol = old->protocol;
720 new->mark = old->mark;
721 new->skb_iif = old->skb_iif;
723 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
724 new->nf_trace = old->nf_trace;
726 #ifdef CONFIG_NET_SCHED
727 new->tc_index = old->tc_index;
728 #ifdef CONFIG_NET_CLS_ACT
729 new->tc_verd = old->tc_verd;
732 new->vlan_proto = old->vlan_proto;
733 new->vlan_tci = old->vlan_tci;
735 skb_copy_secmark(new, old);
737 #ifdef CONFIG_NET_LL_RX_POLL
738 new->napi_id = old->napi_id;
743 * You should not add any new code to this function. Add it to
744 * __copy_skb_header above instead.
746 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
748 #define C(x) n->x = skb->x
750 n->next = n->prev = NULL;
752 __copy_skb_header(n, skb);
757 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
760 n->destructor = NULL;
767 atomic_set(&n->users, 1);
769 atomic_inc(&(skb_shinfo(skb)->dataref));
777 * skb_morph - morph one skb into another
778 * @dst: the skb to receive the contents
779 * @src: the skb to supply the contents
781 * This is identical to skb_clone except that the target skb is
782 * supplied by the user.
784 * The target skb is returned upon exit.
786 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
788 skb_release_all(dst);
789 return __skb_clone(dst, src);
791 EXPORT_SYMBOL_GPL(skb_morph);
794 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
795 * @skb: the skb to modify
796 * @gfp_mask: allocation priority
798 * This must be called on SKBTX_DEV_ZEROCOPY skb.
799 * It will copy all frags into kernel and drop the reference
800 * to userspace pages.
802 * If this function is called from an interrupt gfp_mask() must be
805 * Returns 0 on success or a negative error code on failure
806 * to allocate kernel memory to copy to.
808 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
811 int num_frags = skb_shinfo(skb)->nr_frags;
812 struct page *page, *head = NULL;
813 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
815 for (i = 0; i < num_frags; i++) {
817 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
819 page = alloc_page(gfp_mask);
822 struct page *next = (struct page *)head->private;
828 vaddr = kmap_atomic(skb_frag_page(f));
829 memcpy(page_address(page),
830 vaddr + f->page_offset, skb_frag_size(f));
831 kunmap_atomic(vaddr);
832 page->private = (unsigned long)head;
836 /* skb frags release userspace buffers */
837 for (i = 0; i < num_frags; i++)
838 skb_frag_unref(skb, i);
840 uarg->callback(uarg, false);
842 /* skb frags point to kernel buffers */
843 for (i = num_frags - 1; i >= 0; i--) {
844 __skb_fill_page_desc(skb, i, head, 0,
845 skb_shinfo(skb)->frags[i].size);
846 head = (struct page *)head->private;
849 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
852 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
855 * skb_clone - duplicate an sk_buff
856 * @skb: buffer to clone
857 * @gfp_mask: allocation priority
859 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
860 * copies share the same packet data but not structure. The new
861 * buffer has a reference count of 1. If the allocation fails the
862 * function returns %NULL otherwise the new buffer is returned.
864 * If this function is called from an interrupt gfp_mask() must be
868 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
872 if (skb_orphan_frags(skb, gfp_mask))
876 if (skb->fclone == SKB_FCLONE_ORIG &&
877 n->fclone == SKB_FCLONE_UNAVAILABLE) {
878 atomic_t *fclone_ref = (atomic_t *) (n + 1);
879 n->fclone = SKB_FCLONE_CLONE;
880 atomic_inc(fclone_ref);
882 if (skb_pfmemalloc(skb))
883 gfp_mask |= __GFP_MEMALLOC;
885 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
889 kmemcheck_annotate_bitfield(n, flags1);
890 kmemcheck_annotate_bitfield(n, flags2);
891 n->fclone = SKB_FCLONE_UNAVAILABLE;
894 return __skb_clone(n, skb);
896 EXPORT_SYMBOL(skb_clone);
898 static void skb_headers_offset_update(struct sk_buff *skb, int off)
900 /* {transport,network,mac}_header and tail are relative to skb->head */
901 skb->transport_header += off;
902 skb->network_header += off;
903 if (skb_mac_header_was_set(skb))
904 skb->mac_header += off;
905 skb->inner_transport_header += off;
906 skb->inner_network_header += off;
907 skb->inner_mac_header += off;
910 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
912 __copy_skb_header(new, old);
914 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
915 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
916 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
919 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
921 if (skb_pfmemalloc(skb))
927 * skb_copy - create private copy of an sk_buff
928 * @skb: buffer to copy
929 * @gfp_mask: allocation priority
931 * Make a copy of both an &sk_buff and its data. This is used when the
932 * caller wishes to modify the data and needs a private copy of the
933 * data to alter. Returns %NULL on failure or the pointer to the buffer
934 * on success. The returned buffer has a reference count of 1.
936 * As by-product this function converts non-linear &sk_buff to linear
937 * one, so that &sk_buff becomes completely private and caller is allowed
938 * to modify all the data of returned buffer. This means that this
939 * function is not recommended for use in circumstances when only
940 * header is going to be modified. Use pskb_copy() instead.
943 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
945 int headerlen = skb_headroom(skb);
946 unsigned int size = skb_end_offset(skb) + skb->data_len;
947 struct sk_buff *n = __alloc_skb(size, gfp_mask,
948 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
953 /* Set the data pointer */
954 skb_reserve(n, headerlen);
955 /* Set the tail pointer and length */
956 skb_put(n, skb->len);
958 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
961 copy_skb_header(n, skb);
964 EXPORT_SYMBOL(skb_copy);
967 * __pskb_copy - create copy of an sk_buff with private head.
968 * @skb: buffer to copy
969 * @headroom: headroom of new skb
970 * @gfp_mask: allocation priority
972 * Make a copy of both an &sk_buff and part of its data, located
973 * in header. Fragmented data remain shared. This is used when
974 * the caller wishes to modify only header of &sk_buff and needs
975 * private copy of the header to alter. Returns %NULL on failure
976 * or the pointer to the buffer on success.
977 * The returned buffer has a reference count of 1.
980 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
982 unsigned int size = skb_headlen(skb) + headroom;
983 struct sk_buff *n = __alloc_skb(size, gfp_mask,
984 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
989 /* Set the data pointer */
990 skb_reserve(n, headroom);
991 /* Set the tail pointer and length */
992 skb_put(n, skb_headlen(skb));
994 skb_copy_from_linear_data(skb, n->data, n->len);
996 n->truesize += skb->data_len;
997 n->data_len = skb->data_len;
1000 if (skb_shinfo(skb)->nr_frags) {
1003 if (skb_orphan_frags(skb, gfp_mask)) {
1008 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1009 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1010 skb_frag_ref(skb, i);
1012 skb_shinfo(n)->nr_frags = i;
1015 if (skb_has_frag_list(skb)) {
1016 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1017 skb_clone_fraglist(n);
1020 copy_skb_header(n, skb);
1024 EXPORT_SYMBOL(__pskb_copy);
1027 * pskb_expand_head - reallocate header of &sk_buff
1028 * @skb: buffer to reallocate
1029 * @nhead: room to add at head
1030 * @ntail: room to add at tail
1031 * @gfp_mask: allocation priority
1033 * Expands (or creates identical copy, if &nhead and &ntail are zero)
1034 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
1035 * reference count of 1. Returns zero in the case of success or error,
1036 * if expansion failed. In the last case, &sk_buff is not changed.
1038 * All the pointers pointing into skb header may change and must be
1039 * reloaded after call to this function.
1042 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1047 int size = nhead + skb_end_offset(skb) + ntail;
1052 if (skb_shared(skb))
1055 size = SKB_DATA_ALIGN(size);
1057 if (skb_pfmemalloc(skb))
1058 gfp_mask |= __GFP_MEMALLOC;
1059 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1060 gfp_mask, NUMA_NO_NODE, NULL);
1063 size = SKB_WITH_OVERHEAD(ksize(data));
1065 /* Copy only real data... and, alas, header. This should be
1066 * optimized for the cases when header is void.
1068 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1070 memcpy((struct skb_shared_info *)(data + size),
1072 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1075 * if shinfo is shared we must drop the old head gracefully, but if it
1076 * is not we can just drop the old head and let the existing refcount
1077 * be since all we did is relocate the values
1079 if (skb_cloned(skb)) {
1080 /* copy this zero copy skb frags */
1081 if (skb_orphan_frags(skb, gfp_mask))
1083 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1084 skb_frag_ref(skb, i);
1086 if (skb_has_frag_list(skb))
1087 skb_clone_fraglist(skb);
1089 skb_release_data(skb);
1093 off = (data + nhead) - skb->head;
1098 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1102 skb->end = skb->head + size;
1105 skb_headers_offset_update(skb, nhead);
1106 /* Only adjust this if it actually is csum_start rather than csum */
1107 if (skb->ip_summed == CHECKSUM_PARTIAL)
1108 skb->csum_start += nhead;
1112 atomic_set(&skb_shinfo(skb)->dataref, 1);
1120 EXPORT_SYMBOL(pskb_expand_head);
1122 /* Make private copy of skb with writable head and some headroom */
1124 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1126 struct sk_buff *skb2;
1127 int delta = headroom - skb_headroom(skb);
1130 skb2 = pskb_copy(skb, GFP_ATOMIC);
1132 skb2 = skb_clone(skb, GFP_ATOMIC);
1133 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1141 EXPORT_SYMBOL(skb_realloc_headroom);
1144 * skb_copy_expand - copy and expand sk_buff
1145 * @skb: buffer to copy
1146 * @newheadroom: new free bytes at head
1147 * @newtailroom: new free bytes at tail
1148 * @gfp_mask: allocation priority
1150 * Make a copy of both an &sk_buff and its data and while doing so
1151 * allocate additional space.
1153 * This is used when the caller wishes to modify the data and needs a
1154 * private copy of the data to alter as well as more space for new fields.
1155 * Returns %NULL on failure or the pointer to the buffer
1156 * on success. The returned buffer has a reference count of 1.
1158 * You must pass %GFP_ATOMIC as the allocation priority if this function
1159 * is called from an interrupt.
1161 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1162 int newheadroom, int newtailroom,
1166 * Allocate the copy buffer
1168 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1169 gfp_mask, skb_alloc_rx_flag(skb),
1171 int oldheadroom = skb_headroom(skb);
1172 int head_copy_len, head_copy_off;
1178 skb_reserve(n, newheadroom);
1180 /* Set the tail pointer and length */
1181 skb_put(n, skb->len);
1183 head_copy_len = oldheadroom;
1185 if (newheadroom <= head_copy_len)
1186 head_copy_len = newheadroom;
1188 head_copy_off = newheadroom - head_copy_len;
1190 /* Copy the linear header and data. */
1191 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1192 skb->len + head_copy_len))
1195 copy_skb_header(n, skb);
1197 off = newheadroom - oldheadroom;
1198 if (n->ip_summed == CHECKSUM_PARTIAL)
1199 n->csum_start += off;
1201 skb_headers_offset_update(n, off);
1205 EXPORT_SYMBOL(skb_copy_expand);
1208 * skb_pad - zero pad the tail of an skb
1209 * @skb: buffer to pad
1210 * @pad: space to pad
1212 * Ensure that a buffer is followed by a padding area that is zero
1213 * filled. Used by network drivers which may DMA or transfer data
1214 * beyond the buffer end onto the wire.
1216 * May return error in out of memory cases. The skb is freed on error.
1219 int skb_pad(struct sk_buff *skb, int pad)
1224 /* If the skbuff is non linear tailroom is always zero.. */
1225 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1226 memset(skb->data+skb->len, 0, pad);
1230 ntail = skb->data_len + pad - (skb->end - skb->tail);
1231 if (likely(skb_cloned(skb) || ntail > 0)) {
1232 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1237 /* FIXME: The use of this function with non-linear skb's really needs
1240 err = skb_linearize(skb);
1244 memset(skb->data + skb->len, 0, pad);
1251 EXPORT_SYMBOL(skb_pad);
1254 * skb_put - add data to a buffer
1255 * @skb: buffer to use
1256 * @len: amount of data to add
1258 * This function extends the used data area of the buffer. If this would
1259 * exceed the total buffer size the kernel will panic. A pointer to the
1260 * first byte of the extra data is returned.
1262 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1264 unsigned char *tmp = skb_tail_pointer(skb);
1265 SKB_LINEAR_ASSERT(skb);
1268 if (unlikely(skb->tail > skb->end))
1269 skb_over_panic(skb, len, __builtin_return_address(0));
1272 EXPORT_SYMBOL(skb_put);
1275 * skb_push - add data to the start of a buffer
1276 * @skb: buffer to use
1277 * @len: amount of data to add
1279 * This function extends the used data area of the buffer at the buffer
1280 * start. If this would exceed the total buffer headroom the kernel will
1281 * panic. A pointer to the first byte of the extra data is returned.
1283 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1287 if (unlikely(skb->data<skb->head))
1288 skb_under_panic(skb, len, __builtin_return_address(0));
1291 EXPORT_SYMBOL(skb_push);
1294 * skb_pull - remove data from the start of a buffer
1295 * @skb: buffer to use
1296 * @len: amount of data to remove
1298 * This function removes data from the start of a buffer, returning
1299 * the memory to the headroom. A pointer to the next data in the buffer
1300 * is returned. Once the data has been pulled future pushes will overwrite
1303 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1305 return skb_pull_inline(skb, len);
1307 EXPORT_SYMBOL(skb_pull);
1310 * skb_trim - remove end from a buffer
1311 * @skb: buffer to alter
1314 * Cut the length of a buffer down by removing data from the tail. If
1315 * the buffer is already under the length specified it is not modified.
1316 * The skb must be linear.
1318 void skb_trim(struct sk_buff *skb, unsigned int len)
1321 __skb_trim(skb, len);
1323 EXPORT_SYMBOL(skb_trim);
1325 /* Trims skb to length len. It can change skb pointers.
1328 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1330 struct sk_buff **fragp;
1331 struct sk_buff *frag;
1332 int offset = skb_headlen(skb);
1333 int nfrags = skb_shinfo(skb)->nr_frags;
1337 if (skb_cloned(skb) &&
1338 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1345 for (; i < nfrags; i++) {
1346 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1353 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1356 skb_shinfo(skb)->nr_frags = i;
1358 for (; i < nfrags; i++)
1359 skb_frag_unref(skb, i);
1361 if (skb_has_frag_list(skb))
1362 skb_drop_fraglist(skb);
1366 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1367 fragp = &frag->next) {
1368 int end = offset + frag->len;
1370 if (skb_shared(frag)) {
1371 struct sk_buff *nfrag;
1373 nfrag = skb_clone(frag, GFP_ATOMIC);
1374 if (unlikely(!nfrag))
1377 nfrag->next = frag->next;
1389 unlikely((err = pskb_trim(frag, len - offset))))
1393 skb_drop_list(&frag->next);
1398 if (len > skb_headlen(skb)) {
1399 skb->data_len -= skb->len - len;
1404 skb_set_tail_pointer(skb, len);
1409 EXPORT_SYMBOL(___pskb_trim);
1412 * __pskb_pull_tail - advance tail of skb header
1413 * @skb: buffer to reallocate
1414 * @delta: number of bytes to advance tail
1416 * The function makes a sense only on a fragmented &sk_buff,
1417 * it expands header moving its tail forward and copying necessary
1418 * data from fragmented part.
1420 * &sk_buff MUST have reference count of 1.
1422 * Returns %NULL (and &sk_buff does not change) if pull failed
1423 * or value of new tail of skb in the case of success.
1425 * All the pointers pointing into skb header may change and must be
1426 * reloaded after call to this function.
1429 /* Moves tail of skb head forward, copying data from fragmented part,
1430 * when it is necessary.
1431 * 1. It may fail due to malloc failure.
1432 * 2. It may change skb pointers.
1434 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1436 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1438 /* If skb has not enough free space at tail, get new one
1439 * plus 128 bytes for future expansions. If we have enough
1440 * room at tail, reallocate without expansion only if skb is cloned.
1442 int i, k, eat = (skb->tail + delta) - skb->end;
1444 if (eat > 0 || skb_cloned(skb)) {
1445 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1450 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1453 /* Optimization: no fragments, no reasons to preestimate
1454 * size of pulled pages. Superb.
1456 if (!skb_has_frag_list(skb))
1459 /* Estimate size of pulled pages. */
1461 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1462 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1469 /* If we need update frag list, we are in troubles.
1470 * Certainly, it possible to add an offset to skb data,
1471 * but taking into account that pulling is expected to
1472 * be very rare operation, it is worth to fight against
1473 * further bloating skb head and crucify ourselves here instead.
1474 * Pure masohism, indeed. 8)8)
1477 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1478 struct sk_buff *clone = NULL;
1479 struct sk_buff *insp = NULL;
1484 if (list->len <= eat) {
1485 /* Eaten as whole. */
1490 /* Eaten partially. */
1492 if (skb_shared(list)) {
1493 /* Sucks! We need to fork list. :-( */
1494 clone = skb_clone(list, GFP_ATOMIC);
1500 /* This may be pulled without
1504 if (!pskb_pull(list, eat)) {
1512 /* Free pulled out fragments. */
1513 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1514 skb_shinfo(skb)->frag_list = list->next;
1517 /* And insert new clone at head. */
1520 skb_shinfo(skb)->frag_list = clone;
1523 /* Success! Now we may commit changes to skb data. */
1528 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1529 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1532 skb_frag_unref(skb, i);
1535 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1537 skb_shinfo(skb)->frags[k].page_offset += eat;
1538 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1544 skb_shinfo(skb)->nr_frags = k;
1547 skb->data_len -= delta;
1549 return skb_tail_pointer(skb);
1551 EXPORT_SYMBOL(__pskb_pull_tail);
1554 * skb_copy_bits - copy bits from skb to kernel buffer
1556 * @offset: offset in source
1557 * @to: destination buffer
1558 * @len: number of bytes to copy
1560 * Copy the specified number of bytes from the source skb to the
1561 * destination buffer.
1564 * If its prototype is ever changed,
1565 * check arch/{*}/net/{*}.S files,
1566 * since it is called from BPF assembly code.
1568 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1570 int start = skb_headlen(skb);
1571 struct sk_buff *frag_iter;
1574 if (offset > (int)skb->len - len)
1578 if ((copy = start - offset) > 0) {
1581 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1582 if ((len -= copy) == 0)
1588 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1590 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1592 WARN_ON(start > offset + len);
1594 end = start + skb_frag_size(f);
1595 if ((copy = end - offset) > 0) {
1601 vaddr = kmap_atomic(skb_frag_page(f));
1603 vaddr + f->page_offset + offset - start,
1605 kunmap_atomic(vaddr);
1607 if ((len -= copy) == 0)
1615 skb_walk_frags(skb, frag_iter) {
1618 WARN_ON(start > offset + len);
1620 end = start + frag_iter->len;
1621 if ((copy = end - offset) > 0) {
1624 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1626 if ((len -= copy) == 0)
1640 EXPORT_SYMBOL(skb_copy_bits);
1643 * Callback from splice_to_pipe(), if we need to release some pages
1644 * at the end of the spd in case we error'ed out in filling the pipe.
1646 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1648 put_page(spd->pages[i]);
1651 static struct page *linear_to_page(struct page *page, unsigned int *len,
1652 unsigned int *offset,
1655 struct page_frag *pfrag = sk_page_frag(sk);
1657 if (!sk_page_frag_refill(sk, pfrag))
1660 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1662 memcpy(page_address(pfrag->page) + pfrag->offset,
1663 page_address(page) + *offset, *len);
1664 *offset = pfrag->offset;
1665 pfrag->offset += *len;
1670 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1672 unsigned int offset)
1674 return spd->nr_pages &&
1675 spd->pages[spd->nr_pages - 1] == page &&
1676 (spd->partial[spd->nr_pages - 1].offset +
1677 spd->partial[spd->nr_pages - 1].len == offset);
1681 * Fill page/offset/length into spd, if it can hold more pages.
1683 static bool spd_fill_page(struct splice_pipe_desc *spd,
1684 struct pipe_inode_info *pipe, struct page *page,
1685 unsigned int *len, unsigned int offset,
1689 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1693 page = linear_to_page(page, len, &offset, sk);
1697 if (spd_can_coalesce(spd, page, offset)) {
1698 spd->partial[spd->nr_pages - 1].len += *len;
1702 spd->pages[spd->nr_pages] = page;
1703 spd->partial[spd->nr_pages].len = *len;
1704 spd->partial[spd->nr_pages].offset = offset;
1710 static bool __splice_segment(struct page *page, unsigned int poff,
1711 unsigned int plen, unsigned int *off,
1713 struct splice_pipe_desc *spd, bool linear,
1715 struct pipe_inode_info *pipe)
1720 /* skip this segment if already processed */
1726 /* ignore any bits we already processed */
1732 unsigned int flen = min(*len, plen);
1734 if (spd_fill_page(spd, pipe, page, &flen, poff,
1740 } while (*len && plen);
1746 * Map linear and fragment data from the skb to spd. It reports true if the
1747 * pipe is full or if we already spliced the requested length.
1749 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1750 unsigned int *offset, unsigned int *len,
1751 struct splice_pipe_desc *spd, struct sock *sk)
1755 /* map the linear part :
1756 * If skb->head_frag is set, this 'linear' part is backed by a
1757 * fragment, and if the head is not shared with any clones then
1758 * we can avoid a copy since we own the head portion of this page.
1760 if (__splice_segment(virt_to_page(skb->data),
1761 (unsigned long) skb->data & (PAGE_SIZE - 1),
1764 skb_head_is_locked(skb),
1769 * then map the fragments
1771 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1772 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1774 if (__splice_segment(skb_frag_page(f),
1775 f->page_offset, skb_frag_size(f),
1776 offset, len, spd, false, sk, pipe))
1784 * Map data from the skb to a pipe. Should handle both the linear part,
1785 * the fragments, and the frag list. It does NOT handle frag lists within
1786 * the frag list, if such a thing exists. We'd probably need to recurse to
1787 * handle that cleanly.
1789 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1790 struct pipe_inode_info *pipe, unsigned int tlen,
1793 struct partial_page partial[MAX_SKB_FRAGS];
1794 struct page *pages[MAX_SKB_FRAGS];
1795 struct splice_pipe_desc spd = {
1798 .nr_pages_max = MAX_SKB_FRAGS,
1800 .ops = &sock_pipe_buf_ops,
1801 .spd_release = sock_spd_release,
1803 struct sk_buff *frag_iter;
1804 struct sock *sk = skb->sk;
1808 * __skb_splice_bits() only fails if the output has no room left,
1809 * so no point in going over the frag_list for the error case.
1811 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1817 * now see if we have a frag_list to map
1819 skb_walk_frags(skb, frag_iter) {
1822 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1829 * Drop the socket lock, otherwise we have reverse
1830 * locking dependencies between sk_lock and i_mutex
1831 * here as compared to sendfile(). We enter here
1832 * with the socket lock held, and splice_to_pipe() will
1833 * grab the pipe inode lock. For sendfile() emulation,
1834 * we call into ->sendpage() with the i_mutex lock held
1835 * and networking will grab the socket lock.
1838 ret = splice_to_pipe(pipe, &spd);
1846 * skb_store_bits - store bits from kernel buffer to skb
1847 * @skb: destination buffer
1848 * @offset: offset in destination
1849 * @from: source buffer
1850 * @len: number of bytes to copy
1852 * Copy the specified number of bytes from the source buffer to the
1853 * destination skb. This function handles all the messy bits of
1854 * traversing fragment lists and such.
1857 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1859 int start = skb_headlen(skb);
1860 struct sk_buff *frag_iter;
1863 if (offset > (int)skb->len - len)
1866 if ((copy = start - offset) > 0) {
1869 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1870 if ((len -= copy) == 0)
1876 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1877 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1880 WARN_ON(start > offset + len);
1882 end = start + skb_frag_size(frag);
1883 if ((copy = end - offset) > 0) {
1889 vaddr = kmap_atomic(skb_frag_page(frag));
1890 memcpy(vaddr + frag->page_offset + offset - start,
1892 kunmap_atomic(vaddr);
1894 if ((len -= copy) == 0)
1902 skb_walk_frags(skb, frag_iter) {
1905 WARN_ON(start > offset + len);
1907 end = start + frag_iter->len;
1908 if ((copy = end - offset) > 0) {
1911 if (skb_store_bits(frag_iter, offset - start,
1914 if ((len -= copy) == 0)
1927 EXPORT_SYMBOL(skb_store_bits);
1929 /* Checksum skb data. */
1931 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1932 int len, __wsum csum)
1934 int start = skb_headlen(skb);
1935 int i, copy = start - offset;
1936 struct sk_buff *frag_iter;
1939 /* Checksum header. */
1943 csum = csum_partial(skb->data + offset, copy, csum);
1944 if ((len -= copy) == 0)
1950 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1952 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1954 WARN_ON(start > offset + len);
1956 end = start + skb_frag_size(frag);
1957 if ((copy = end - offset) > 0) {
1963 vaddr = kmap_atomic(skb_frag_page(frag));
1964 csum2 = csum_partial(vaddr + frag->page_offset +
1965 offset - start, copy, 0);
1966 kunmap_atomic(vaddr);
1967 csum = csum_block_add(csum, csum2, pos);
1976 skb_walk_frags(skb, frag_iter) {
1979 WARN_ON(start > offset + len);
1981 end = start + frag_iter->len;
1982 if ((copy = end - offset) > 0) {
1986 csum2 = skb_checksum(frag_iter, offset - start,
1988 csum = csum_block_add(csum, csum2, pos);
1989 if ((len -= copy) == 0)
2000 EXPORT_SYMBOL(skb_checksum);
2002 /* Both of above in one bottle. */
2004 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2005 u8 *to, int len, __wsum csum)
2007 int start = skb_headlen(skb);
2008 int i, copy = start - offset;
2009 struct sk_buff *frag_iter;
2016 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2018 if ((len -= copy) == 0)
2025 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2028 WARN_ON(start > offset + len);
2030 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2031 if ((copy = end - offset) > 0) {
2034 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2038 vaddr = kmap_atomic(skb_frag_page(frag));
2039 csum2 = csum_partial_copy_nocheck(vaddr +
2043 kunmap_atomic(vaddr);
2044 csum = csum_block_add(csum, csum2, pos);
2054 skb_walk_frags(skb, frag_iter) {
2058 WARN_ON(start > offset + len);
2060 end = start + frag_iter->len;
2061 if ((copy = end - offset) > 0) {
2064 csum2 = skb_copy_and_csum_bits(frag_iter,
2067 csum = csum_block_add(csum, csum2, pos);
2068 if ((len -= copy) == 0)
2079 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2081 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2086 if (skb->ip_summed == CHECKSUM_PARTIAL)
2087 csstart = skb_checksum_start_offset(skb);
2089 csstart = skb_headlen(skb);
2091 BUG_ON(csstart > skb_headlen(skb));
2093 skb_copy_from_linear_data(skb, to, csstart);
2096 if (csstart != skb->len)
2097 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2098 skb->len - csstart, 0);
2100 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2101 long csstuff = csstart + skb->csum_offset;
2103 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2106 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2109 * skb_dequeue - remove from the head of the queue
2110 * @list: list to dequeue from
2112 * Remove the head of the list. The list lock is taken so the function
2113 * may be used safely with other locking list functions. The head item is
2114 * returned or %NULL if the list is empty.
2117 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2119 unsigned long flags;
2120 struct sk_buff *result;
2122 spin_lock_irqsave(&list->lock, flags);
2123 result = __skb_dequeue(list);
2124 spin_unlock_irqrestore(&list->lock, flags);
2127 EXPORT_SYMBOL(skb_dequeue);
2130 * skb_dequeue_tail - remove from the tail of the queue
2131 * @list: list to dequeue from
2133 * Remove the tail of the list. The list lock is taken so the function
2134 * may be used safely with other locking list functions. The tail item is
2135 * returned or %NULL if the list is empty.
2137 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2139 unsigned long flags;
2140 struct sk_buff *result;
2142 spin_lock_irqsave(&list->lock, flags);
2143 result = __skb_dequeue_tail(list);
2144 spin_unlock_irqrestore(&list->lock, flags);
2147 EXPORT_SYMBOL(skb_dequeue_tail);
2150 * skb_queue_purge - empty a list
2151 * @list: list to empty
2153 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2154 * the list and one reference dropped. This function takes the list
2155 * lock and is atomic with respect to other list locking functions.
2157 void skb_queue_purge(struct sk_buff_head *list)
2159 struct sk_buff *skb;
2160 while ((skb = skb_dequeue(list)) != NULL)
2163 EXPORT_SYMBOL(skb_queue_purge);
2166 * skb_queue_head - queue a buffer at the list head
2167 * @list: list to use
2168 * @newsk: buffer to queue
2170 * Queue a buffer at the start of the list. This function takes the
2171 * list lock and can be used safely with other locking &sk_buff functions
2174 * A buffer cannot be placed on two lists at the same time.
2176 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2178 unsigned long flags;
2180 spin_lock_irqsave(&list->lock, flags);
2181 __skb_queue_head(list, newsk);
2182 spin_unlock_irqrestore(&list->lock, flags);
2184 EXPORT_SYMBOL(skb_queue_head);
2187 * skb_queue_tail - queue a buffer at the list tail
2188 * @list: list to use
2189 * @newsk: buffer to queue
2191 * Queue a buffer at the tail of the list. This function takes the
2192 * list lock and can be used safely with other locking &sk_buff functions
2195 * A buffer cannot be placed on two lists at the same time.
2197 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2199 unsigned long flags;
2201 spin_lock_irqsave(&list->lock, flags);
2202 __skb_queue_tail(list, newsk);
2203 spin_unlock_irqrestore(&list->lock, flags);
2205 EXPORT_SYMBOL(skb_queue_tail);
2208 * skb_unlink - remove a buffer from a list
2209 * @skb: buffer to remove
2210 * @list: list to use
2212 * Remove a packet from a list. The list locks are taken and this
2213 * function is atomic with respect to other list locked calls
2215 * You must know what list the SKB is on.
2217 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2219 unsigned long flags;
2221 spin_lock_irqsave(&list->lock, flags);
2222 __skb_unlink(skb, list);
2223 spin_unlock_irqrestore(&list->lock, flags);
2225 EXPORT_SYMBOL(skb_unlink);
2228 * skb_append - append a buffer
2229 * @old: buffer to insert after
2230 * @newsk: buffer to insert
2231 * @list: list to use
2233 * Place a packet after a given packet in a list. The list locks are taken
2234 * and this function is atomic with respect to other list locked calls.
2235 * A buffer cannot be placed on two lists at the same time.
2237 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2239 unsigned long flags;
2241 spin_lock_irqsave(&list->lock, flags);
2242 __skb_queue_after(list, old, newsk);
2243 spin_unlock_irqrestore(&list->lock, flags);
2245 EXPORT_SYMBOL(skb_append);
2248 * skb_insert - insert a buffer
2249 * @old: buffer to insert before
2250 * @newsk: buffer to insert
2251 * @list: list to use
2253 * Place a packet before a given packet in a list. The list locks are
2254 * taken and this function is atomic with respect to other list locked
2257 * A buffer cannot be placed on two lists at the same time.
2259 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2261 unsigned long flags;
2263 spin_lock_irqsave(&list->lock, flags);
2264 __skb_insert(newsk, old->prev, old, list);
2265 spin_unlock_irqrestore(&list->lock, flags);
2267 EXPORT_SYMBOL(skb_insert);
2269 static inline void skb_split_inside_header(struct sk_buff *skb,
2270 struct sk_buff* skb1,
2271 const u32 len, const int pos)
2275 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2277 /* And move data appendix as is. */
2278 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2279 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2281 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2282 skb_shinfo(skb)->nr_frags = 0;
2283 skb1->data_len = skb->data_len;
2284 skb1->len += skb1->data_len;
2287 skb_set_tail_pointer(skb, len);
2290 static inline void skb_split_no_header(struct sk_buff *skb,
2291 struct sk_buff* skb1,
2292 const u32 len, int pos)
2295 const int nfrags = skb_shinfo(skb)->nr_frags;
2297 skb_shinfo(skb)->nr_frags = 0;
2298 skb1->len = skb1->data_len = skb->len - len;
2300 skb->data_len = len - pos;
2302 for (i = 0; i < nfrags; i++) {
2303 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2305 if (pos + size > len) {
2306 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2310 * We have two variants in this case:
2311 * 1. Move all the frag to the second
2312 * part, if it is possible. F.e.
2313 * this approach is mandatory for TUX,
2314 * where splitting is expensive.
2315 * 2. Split is accurately. We make this.
2317 skb_frag_ref(skb, i);
2318 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2319 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2320 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2321 skb_shinfo(skb)->nr_frags++;
2325 skb_shinfo(skb)->nr_frags++;
2328 skb_shinfo(skb1)->nr_frags = k;
2332 * skb_split - Split fragmented skb to two parts at length len.
2333 * @skb: the buffer to split
2334 * @skb1: the buffer to receive the second part
2335 * @len: new length for skb
2337 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2339 int pos = skb_headlen(skb);
2341 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2342 if (len < pos) /* Split line is inside header. */
2343 skb_split_inside_header(skb, skb1, len, pos);
2344 else /* Second chunk has no header, nothing to copy. */
2345 skb_split_no_header(skb, skb1, len, pos);
2347 EXPORT_SYMBOL(skb_split);
2349 /* Shifting from/to a cloned skb is a no-go.
2351 * Caller cannot keep skb_shinfo related pointers past calling here!
2353 static int skb_prepare_for_shift(struct sk_buff *skb)
2355 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2359 * skb_shift - Shifts paged data partially from skb to another
2360 * @tgt: buffer into which tail data gets added
2361 * @skb: buffer from which the paged data comes from
2362 * @shiftlen: shift up to this many bytes
2364 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2365 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2366 * It's up to caller to free skb if everything was shifted.
2368 * If @tgt runs out of frags, the whole operation is aborted.
2370 * Skb cannot include anything else but paged data while tgt is allowed
2371 * to have non-paged data as well.
2373 * TODO: full sized shift could be optimized but that would need
2374 * specialized skb free'er to handle frags without up-to-date nr_frags.
2376 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2378 int from, to, merge, todo;
2379 struct skb_frag_struct *fragfrom, *fragto;
2381 BUG_ON(shiftlen > skb->len);
2382 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2386 to = skb_shinfo(tgt)->nr_frags;
2387 fragfrom = &skb_shinfo(skb)->frags[from];
2389 /* Actual merge is delayed until the point when we know we can
2390 * commit all, so that we don't have to undo partial changes
2393 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2394 fragfrom->page_offset)) {
2399 todo -= skb_frag_size(fragfrom);
2401 if (skb_prepare_for_shift(skb) ||
2402 skb_prepare_for_shift(tgt))
2405 /* All previous frag pointers might be stale! */
2406 fragfrom = &skb_shinfo(skb)->frags[from];
2407 fragto = &skb_shinfo(tgt)->frags[merge];
2409 skb_frag_size_add(fragto, shiftlen);
2410 skb_frag_size_sub(fragfrom, shiftlen);
2411 fragfrom->page_offset += shiftlen;
2419 /* Skip full, not-fitting skb to avoid expensive operations */
2420 if ((shiftlen == skb->len) &&
2421 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2424 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2427 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2428 if (to == MAX_SKB_FRAGS)
2431 fragfrom = &skb_shinfo(skb)->frags[from];
2432 fragto = &skb_shinfo(tgt)->frags[to];
2434 if (todo >= skb_frag_size(fragfrom)) {
2435 *fragto = *fragfrom;
2436 todo -= skb_frag_size(fragfrom);
2441 __skb_frag_ref(fragfrom);
2442 fragto->page = fragfrom->page;
2443 fragto->page_offset = fragfrom->page_offset;
2444 skb_frag_size_set(fragto, todo);
2446 fragfrom->page_offset += todo;
2447 skb_frag_size_sub(fragfrom, todo);
2455 /* Ready to "commit" this state change to tgt */
2456 skb_shinfo(tgt)->nr_frags = to;
2459 fragfrom = &skb_shinfo(skb)->frags[0];
2460 fragto = &skb_shinfo(tgt)->frags[merge];
2462 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2463 __skb_frag_unref(fragfrom);
2466 /* Reposition in the original skb */
2468 while (from < skb_shinfo(skb)->nr_frags)
2469 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2470 skb_shinfo(skb)->nr_frags = to;
2472 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2475 /* Most likely the tgt won't ever need its checksum anymore, skb on
2476 * the other hand might need it if it needs to be resent
2478 tgt->ip_summed = CHECKSUM_PARTIAL;
2479 skb->ip_summed = CHECKSUM_PARTIAL;
2481 /* Yak, is it really working this way? Some helper please? */
2482 skb->len -= shiftlen;
2483 skb->data_len -= shiftlen;
2484 skb->truesize -= shiftlen;
2485 tgt->len += shiftlen;
2486 tgt->data_len += shiftlen;
2487 tgt->truesize += shiftlen;
2493 * skb_prepare_seq_read - Prepare a sequential read of skb data
2494 * @skb: the buffer to read
2495 * @from: lower offset of data to be read
2496 * @to: upper offset of data to be read
2497 * @st: state variable
2499 * Initializes the specified state variable. Must be called before
2500 * invoking skb_seq_read() for the first time.
2502 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2503 unsigned int to, struct skb_seq_state *st)
2505 st->lower_offset = from;
2506 st->upper_offset = to;
2507 st->root_skb = st->cur_skb = skb;
2508 st->frag_idx = st->stepped_offset = 0;
2509 st->frag_data = NULL;
2511 EXPORT_SYMBOL(skb_prepare_seq_read);
2514 * skb_seq_read - Sequentially read skb data
2515 * @consumed: number of bytes consumed by the caller so far
2516 * @data: destination pointer for data to be returned
2517 * @st: state variable
2519 * Reads a block of skb data at &consumed relative to the
2520 * lower offset specified to skb_prepare_seq_read(). Assigns
2521 * the head of the data block to &data and returns the length
2522 * of the block or 0 if the end of the skb data or the upper
2523 * offset has been reached.
2525 * The caller is not required to consume all of the data
2526 * returned, i.e. &consumed is typically set to the number
2527 * of bytes already consumed and the next call to
2528 * skb_seq_read() will return the remaining part of the block.
2530 * Note 1: The size of each block of data returned can be arbitrary,
2531 * this limitation is the cost for zerocopy seqeuental
2532 * reads of potentially non linear data.
2534 * Note 2: Fragment lists within fragments are not implemented
2535 * at the moment, state->root_skb could be replaced with
2536 * a stack for this purpose.
2538 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2539 struct skb_seq_state *st)
2541 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2544 if (unlikely(abs_offset >= st->upper_offset))
2548 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2550 if (abs_offset < block_limit && !st->frag_data) {
2551 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2552 return block_limit - abs_offset;
2555 if (st->frag_idx == 0 && !st->frag_data)
2556 st->stepped_offset += skb_headlen(st->cur_skb);
2558 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2559 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2560 block_limit = skb_frag_size(frag) + st->stepped_offset;
2562 if (abs_offset < block_limit) {
2564 st->frag_data = kmap_atomic(skb_frag_page(frag));
2566 *data = (u8 *) st->frag_data + frag->page_offset +
2567 (abs_offset - st->stepped_offset);
2569 return block_limit - abs_offset;
2572 if (st->frag_data) {
2573 kunmap_atomic(st->frag_data);
2574 st->frag_data = NULL;
2578 st->stepped_offset += skb_frag_size(frag);
2581 if (st->frag_data) {
2582 kunmap_atomic(st->frag_data);
2583 st->frag_data = NULL;
2586 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2587 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2590 } else if (st->cur_skb->next) {
2591 st->cur_skb = st->cur_skb->next;
2598 EXPORT_SYMBOL(skb_seq_read);
2601 * skb_abort_seq_read - Abort a sequential read of skb data
2602 * @st: state variable
2604 * Must be called if skb_seq_read() was not called until it
2607 void skb_abort_seq_read(struct skb_seq_state *st)
2610 kunmap_atomic(st->frag_data);
2612 EXPORT_SYMBOL(skb_abort_seq_read);
2614 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2616 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2617 struct ts_config *conf,
2618 struct ts_state *state)
2620 return skb_seq_read(offset, text, TS_SKB_CB(state));
2623 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2625 skb_abort_seq_read(TS_SKB_CB(state));
2629 * skb_find_text - Find a text pattern in skb data
2630 * @skb: the buffer to look in
2631 * @from: search offset
2633 * @config: textsearch configuration
2634 * @state: uninitialized textsearch state variable
2636 * Finds a pattern in the skb data according to the specified
2637 * textsearch configuration. Use textsearch_next() to retrieve
2638 * subsequent occurrences of the pattern. Returns the offset
2639 * to the first occurrence or UINT_MAX if no match was found.
2641 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2642 unsigned int to, struct ts_config *config,
2643 struct ts_state *state)
2647 config->get_next_block = skb_ts_get_next_block;
2648 config->finish = skb_ts_finish;
2650 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2652 ret = textsearch_find(config, state);
2653 return (ret <= to - from ? ret : UINT_MAX);
2655 EXPORT_SYMBOL(skb_find_text);
2658 * skb_append_datato_frags - append the user data to a skb
2659 * @sk: sock structure
2660 * @skb: skb structure to be appened with user data.
2661 * @getfrag: call back function to be used for getting the user data
2662 * @from: pointer to user message iov
2663 * @length: length of the iov message
2665 * Description: This procedure append the user data in the fragment part
2666 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2668 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2669 int (*getfrag)(void *from, char *to, int offset,
2670 int len, int odd, struct sk_buff *skb),
2671 void *from, int length)
2673 int frg_cnt = skb_shinfo(skb)->nr_frags;
2677 struct page_frag *pfrag = ¤t->task_frag;
2680 /* Return error if we don't have space for new frag */
2681 if (frg_cnt >= MAX_SKB_FRAGS)
2684 if (!sk_page_frag_refill(sk, pfrag))
2687 /* copy the user data to page */
2688 copy = min_t(int, length, pfrag->size - pfrag->offset);
2690 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2691 offset, copy, 0, skb);
2695 /* copy was successful so update the size parameters */
2696 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2699 pfrag->offset += copy;
2700 get_page(pfrag->page);
2702 skb->truesize += copy;
2703 atomic_add(copy, &sk->sk_wmem_alloc);
2705 skb->data_len += copy;
2709 } while (length > 0);
2713 EXPORT_SYMBOL(skb_append_datato_frags);
2716 * skb_pull_rcsum - pull skb and update receive checksum
2717 * @skb: buffer to update
2718 * @len: length of data pulled
2720 * This function performs an skb_pull on the packet and updates
2721 * the CHECKSUM_COMPLETE checksum. It should be used on
2722 * receive path processing instead of skb_pull unless you know
2723 * that the checksum difference is zero (e.g., a valid IP header)
2724 * or you are setting ip_summed to CHECKSUM_NONE.
2726 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2728 BUG_ON(len > skb->len);
2730 BUG_ON(skb->len < skb->data_len);
2731 skb_postpull_rcsum(skb, skb->data, len);
2732 return skb->data += len;
2734 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2737 * skb_segment - Perform protocol segmentation on skb.
2738 * @skb: buffer to segment
2739 * @features: features for the output path (see dev->features)
2741 * This function performs segmentation on the given skb. It returns
2742 * a pointer to the first in a list of new skbs for the segments.
2743 * In case of error it returns ERR_PTR(err).
2745 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2747 struct sk_buff *segs = NULL;
2748 struct sk_buff *tail = NULL;
2749 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2750 unsigned int mss = skb_shinfo(skb)->gso_size;
2751 unsigned int doffset = skb->data - skb_mac_header(skb);
2752 unsigned int offset = doffset;
2753 unsigned int tnl_hlen = skb_tnl_header_len(skb);
2754 unsigned int headroom;
2758 int sg = !!(features & NETIF_F_SG);
2759 int nfrags = skb_shinfo(skb)->nr_frags;
2764 proto = skb_network_protocol(skb);
2765 if (unlikely(!proto))
2766 return ERR_PTR(-EINVAL);
2768 csum = !!can_checksum_protocol(features, proto);
2769 __skb_push(skb, doffset);
2770 headroom = skb_headroom(skb);
2771 pos = skb_headlen(skb);
2774 struct sk_buff *nskb;
2779 len = skb->len - offset;
2783 hsize = skb_headlen(skb) - offset;
2786 if (hsize > len || !sg)
2789 if (!hsize && i >= nfrags) {
2790 BUG_ON(fskb->len != len);
2793 nskb = skb_clone(fskb, GFP_ATOMIC);
2796 if (unlikely(!nskb))
2799 hsize = skb_end_offset(nskb);
2800 if (skb_cow_head(nskb, doffset + headroom)) {
2805 nskb->truesize += skb_end_offset(nskb) - hsize;
2806 skb_release_head_state(nskb);
2807 __skb_push(nskb, doffset);
2809 nskb = __alloc_skb(hsize + doffset + headroom,
2810 GFP_ATOMIC, skb_alloc_rx_flag(skb),
2813 if (unlikely(!nskb))
2816 skb_reserve(nskb, headroom);
2817 __skb_put(nskb, doffset);
2826 __copy_skb_header(nskb, skb);
2827 nskb->mac_len = skb->mac_len;
2829 /* nskb and skb might have different headroom */
2830 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2831 nskb->csum_start += skb_headroom(nskb) - headroom;
2833 skb_reset_mac_header(nskb);
2834 skb_set_network_header(nskb, skb->mac_len);
2835 nskb->transport_header = (nskb->network_header +
2836 skb_network_header_len(skb));
2838 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
2839 nskb->data - tnl_hlen,
2840 doffset + tnl_hlen);
2842 if (fskb != skb_shinfo(skb)->frag_list)
2843 goto perform_csum_check;
2846 nskb->ip_summed = CHECKSUM_NONE;
2847 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2853 frag = skb_shinfo(nskb)->frags;
2855 skb_copy_from_linear_data_offset(skb, offset,
2856 skb_put(nskb, hsize), hsize);
2858 skb_shinfo(nskb)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2860 while (pos < offset + len && i < nfrags) {
2861 *frag = skb_shinfo(skb)->frags[i];
2862 __skb_frag_ref(frag);
2863 size = skb_frag_size(frag);
2866 frag->page_offset += offset - pos;
2867 skb_frag_size_sub(frag, offset - pos);
2870 skb_shinfo(nskb)->nr_frags++;
2872 if (pos + size <= offset + len) {
2876 skb_frag_size_sub(frag, pos + size - (offset + len));
2883 if (pos < offset + len) {
2884 struct sk_buff *fskb2 = fskb;
2886 BUG_ON(pos + fskb->len != offset + len);
2892 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2898 SKB_FRAG_ASSERT(nskb);
2899 skb_shinfo(nskb)->frag_list = fskb2;
2903 nskb->data_len = len - hsize;
2904 nskb->len += nskb->data_len;
2905 nskb->truesize += nskb->data_len;
2909 nskb->csum = skb_checksum(nskb, doffset,
2910 nskb->len - doffset, 0);
2911 nskb->ip_summed = CHECKSUM_NONE;
2913 } while ((offset += len) < skb->len);
2918 while ((skb = segs)) {
2922 return ERR_PTR(err);
2924 EXPORT_SYMBOL_GPL(skb_segment);
2926 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2928 struct sk_buff *p = *head;
2929 struct sk_buff *nskb;
2930 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2931 struct skb_shared_info *pinfo = skb_shinfo(p);
2932 unsigned int headroom;
2933 unsigned int len = skb_gro_len(skb);
2934 unsigned int offset = skb_gro_offset(skb);
2935 unsigned int headlen = skb_headlen(skb);
2936 unsigned int delta_truesize;
2938 if (p->len + len >= 65536)
2941 if (pinfo->frag_list)
2943 else if (headlen <= offset) {
2946 int i = skbinfo->nr_frags;
2947 int nr_frags = pinfo->nr_frags + i;
2951 if (nr_frags > MAX_SKB_FRAGS)
2954 pinfo->nr_frags = nr_frags;
2955 skbinfo->nr_frags = 0;
2957 frag = pinfo->frags + nr_frags;
2958 frag2 = skbinfo->frags + i;
2963 frag->page_offset += offset;
2964 skb_frag_size_sub(frag, offset);
2966 /* all fragments truesize : remove (head size + sk_buff) */
2967 delta_truesize = skb->truesize -
2968 SKB_TRUESIZE(skb_end_offset(skb));
2970 skb->truesize -= skb->data_len;
2971 skb->len -= skb->data_len;
2974 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
2976 } else if (skb->head_frag) {
2977 int nr_frags = pinfo->nr_frags;
2978 skb_frag_t *frag = pinfo->frags + nr_frags;
2979 struct page *page = virt_to_head_page(skb->head);
2980 unsigned int first_size = headlen - offset;
2981 unsigned int first_offset;
2983 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
2986 first_offset = skb->data -
2987 (unsigned char *)page_address(page) +
2990 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
2992 frag->page.p = page;
2993 frag->page_offset = first_offset;
2994 skb_frag_size_set(frag, first_size);
2996 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
2997 /* We dont need to clear skbinfo->nr_frags here */
2999 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3000 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3002 } else if (skb_gro_len(p) != pinfo->gso_size)
3005 headroom = skb_headroom(p);
3006 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
3007 if (unlikely(!nskb))
3010 __copy_skb_header(nskb, p);
3011 nskb->mac_len = p->mac_len;
3013 skb_reserve(nskb, headroom);
3014 __skb_put(nskb, skb_gro_offset(p));
3016 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3017 skb_set_network_header(nskb, skb_network_offset(p));
3018 skb_set_transport_header(nskb, skb_transport_offset(p));
3020 __skb_pull(p, skb_gro_offset(p));
3021 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3022 p->data - skb_mac_header(p));
3024 skb_shinfo(nskb)->frag_list = p;
3025 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3026 pinfo->gso_size = 0;
3027 skb_header_release(p);
3028 NAPI_GRO_CB(nskb)->last = p;
3030 nskb->data_len += p->len;
3031 nskb->truesize += p->truesize;
3032 nskb->len += p->len;
3035 nskb->next = p->next;
3041 delta_truesize = skb->truesize;
3042 if (offset > headlen) {
3043 unsigned int eat = offset - headlen;
3045 skbinfo->frags[0].page_offset += eat;
3046 skb_frag_size_sub(&skbinfo->frags[0], eat);
3047 skb->data_len -= eat;
3052 __skb_pull(skb, offset);
3054 NAPI_GRO_CB(p)->last->next = skb;
3055 NAPI_GRO_CB(p)->last = skb;
3056 skb_header_release(skb);
3059 NAPI_GRO_CB(p)->count++;
3061 p->truesize += delta_truesize;
3064 NAPI_GRO_CB(skb)->same_flow = 1;
3067 EXPORT_SYMBOL_GPL(skb_gro_receive);
3069 void __init skb_init(void)
3071 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3072 sizeof(struct sk_buff),
3074 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3076 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3077 (2*sizeof(struct sk_buff)) +
3080 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3085 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3086 * @skb: Socket buffer containing the buffers to be mapped
3087 * @sg: The scatter-gather list to map into
3088 * @offset: The offset into the buffer's contents to start mapping
3089 * @len: Length of buffer space to be mapped
3091 * Fill the specified scatter-gather list with mappings/pointers into a
3092 * region of the buffer space attached to a socket buffer.
3095 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3097 int start = skb_headlen(skb);
3098 int i, copy = start - offset;
3099 struct sk_buff *frag_iter;
3105 sg_set_buf(sg, skb->data + offset, copy);
3107 if ((len -= copy) == 0)
3112 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3115 WARN_ON(start > offset + len);
3117 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3118 if ((copy = end - offset) > 0) {
3119 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3123 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3124 frag->page_offset+offset-start);
3133 skb_walk_frags(skb, frag_iter) {
3136 WARN_ON(start > offset + len);
3138 end = start + frag_iter->len;
3139 if ((copy = end - offset) > 0) {
3142 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3144 if ((len -= copy) == 0)
3154 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3156 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3158 sg_mark_end(&sg[nsg - 1]);
3162 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3165 * skb_cow_data - Check that a socket buffer's data buffers are writable
3166 * @skb: The socket buffer to check.
3167 * @tailbits: Amount of trailing space to be added
3168 * @trailer: Returned pointer to the skb where the @tailbits space begins
3170 * Make sure that the data buffers attached to a socket buffer are
3171 * writable. If they are not, private copies are made of the data buffers
3172 * and the socket buffer is set to use these instead.
3174 * If @tailbits is given, make sure that there is space to write @tailbits
3175 * bytes of data beyond current end of socket buffer. @trailer will be
3176 * set to point to the skb in which this space begins.
3178 * The number of scatterlist elements required to completely map the
3179 * COW'd and extended socket buffer will be returned.
3181 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3185 struct sk_buff *skb1, **skb_p;
3187 /* If skb is cloned or its head is paged, reallocate
3188 * head pulling out all the pages (pages are considered not writable
3189 * at the moment even if they are anonymous).
3191 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3192 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3195 /* Easy case. Most of packets will go this way. */
3196 if (!skb_has_frag_list(skb)) {
3197 /* A little of trouble, not enough of space for trailer.
3198 * This should not happen, when stack is tuned to generate
3199 * good frames. OK, on miss we reallocate and reserve even more
3200 * space, 128 bytes is fair. */
3202 if (skb_tailroom(skb) < tailbits &&
3203 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3211 /* Misery. We are in troubles, going to mincer fragments... */
3214 skb_p = &skb_shinfo(skb)->frag_list;
3217 while ((skb1 = *skb_p) != NULL) {
3220 /* The fragment is partially pulled by someone,
3221 * this can happen on input. Copy it and everything
3224 if (skb_shared(skb1))
3227 /* If the skb is the last, worry about trailer. */
3229 if (skb1->next == NULL && tailbits) {
3230 if (skb_shinfo(skb1)->nr_frags ||
3231 skb_has_frag_list(skb1) ||
3232 skb_tailroom(skb1) < tailbits)
3233 ntail = tailbits + 128;
3239 skb_shinfo(skb1)->nr_frags ||
3240 skb_has_frag_list(skb1)) {
3241 struct sk_buff *skb2;
3243 /* Fuck, we are miserable poor guys... */
3245 skb2 = skb_copy(skb1, GFP_ATOMIC);
3247 skb2 = skb_copy_expand(skb1,
3251 if (unlikely(skb2 == NULL))
3255 skb_set_owner_w(skb2, skb1->sk);
3257 /* Looking around. Are we still alive?
3258 * OK, link new skb, drop old one */
3260 skb2->next = skb1->next;
3267 skb_p = &skb1->next;
3272 EXPORT_SYMBOL_GPL(skb_cow_data);
3274 static void sock_rmem_free(struct sk_buff *skb)
3276 struct sock *sk = skb->sk;
3278 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3282 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3284 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3288 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3289 (unsigned int)sk->sk_rcvbuf)
3294 skb->destructor = sock_rmem_free;
3295 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3297 /* before exiting rcu section, make sure dst is refcounted */
3300 skb_queue_tail(&sk->sk_error_queue, skb);
3301 if (!sock_flag(sk, SOCK_DEAD))
3302 sk->sk_data_ready(sk, len);
3305 EXPORT_SYMBOL(sock_queue_err_skb);
3307 void skb_tstamp_tx(struct sk_buff *orig_skb,
3308 struct skb_shared_hwtstamps *hwtstamps)
3310 struct sock *sk = orig_skb->sk;
3311 struct sock_exterr_skb *serr;
3312 struct sk_buff *skb;
3319 *skb_hwtstamps(orig_skb) =
3323 * no hardware time stamps available,
3324 * so keep the shared tx_flags and only
3325 * store software time stamp
3327 orig_skb->tstamp = ktime_get_real();
3330 skb = skb_clone(orig_skb, GFP_ATOMIC);
3334 serr = SKB_EXT_ERR(skb);
3335 memset(serr, 0, sizeof(*serr));
3336 serr->ee.ee_errno = ENOMSG;
3337 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3339 err = sock_queue_err_skb(sk, skb);
3344 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3346 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3348 struct sock *sk = skb->sk;
3349 struct sock_exterr_skb *serr;
3352 skb->wifi_acked_valid = 1;
3353 skb->wifi_acked = acked;
3355 serr = SKB_EXT_ERR(skb);
3356 memset(serr, 0, sizeof(*serr));
3357 serr->ee.ee_errno = ENOMSG;
3358 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3360 err = sock_queue_err_skb(sk, skb);
3364 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3368 * skb_partial_csum_set - set up and verify partial csum values for packet
3369 * @skb: the skb to set
3370 * @start: the number of bytes after skb->data to start checksumming.
3371 * @off: the offset from start to place the checksum.
3373 * For untrusted partially-checksummed packets, we need to make sure the values
3374 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3376 * This function checks and sets those values and skb->ip_summed: if this
3377 * returns false you should drop the packet.
3379 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3381 if (unlikely(start > skb_headlen(skb)) ||
3382 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3383 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3384 start, off, skb_headlen(skb));
3387 skb->ip_summed = CHECKSUM_PARTIAL;
3388 skb->csum_start = skb_headroom(skb) + start;
3389 skb->csum_offset = off;
3390 skb_set_transport_header(skb, start);
3393 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3395 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3397 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3400 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3402 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3405 skb_release_head_state(skb);
3406 kmem_cache_free(skbuff_head_cache, skb);
3411 EXPORT_SYMBOL(kfree_skb_partial);
3414 * skb_try_coalesce - try to merge skb to prior one
3416 * @from: buffer to add
3417 * @fragstolen: pointer to boolean
3418 * @delta_truesize: how much more was allocated than was requested
3420 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3421 bool *fragstolen, int *delta_truesize)
3423 int i, delta, len = from->len;
3425 *fragstolen = false;
3430 if (len <= skb_tailroom(to)) {
3431 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3432 *delta_truesize = 0;
3436 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3439 if (skb_headlen(from) != 0) {
3441 unsigned int offset;
3443 if (skb_shinfo(to)->nr_frags +
3444 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3447 if (skb_head_is_locked(from))
3450 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3452 page = virt_to_head_page(from->head);
3453 offset = from->data - (unsigned char *)page_address(page);
3455 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3456 page, offset, skb_headlen(from));
3459 if (skb_shinfo(to)->nr_frags +
3460 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3463 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
3466 WARN_ON_ONCE(delta < len);
3468 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3469 skb_shinfo(from)->frags,
3470 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3471 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3473 if (!skb_cloned(from))
3474 skb_shinfo(from)->nr_frags = 0;
3476 /* if the skb is not cloned this does nothing
3477 * since we set nr_frags to 0.
3479 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3480 skb_frag_ref(from, i);
3482 to->truesize += delta;
3484 to->data_len += len;
3486 *delta_truesize = delta;
3489 EXPORT_SYMBOL(skb_try_coalesce);