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 #ifdef NET_SKBUFF_DATA_USES_OFFSET
203 skb->mac_header = ~0U;
210 * __alloc_skb - allocate a network buffer
211 * @size: size to allocate
212 * @gfp_mask: allocation mask
213 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
214 * instead of head cache and allocate a cloned (child) skb.
215 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
216 * allocations in case the data is required for writeback
217 * @node: numa node to allocate memory on
219 * Allocate a new &sk_buff. The returned buffer has no headroom and a
220 * tail room of at least size bytes. The object has a reference count
221 * of one. The return is the buffer. On a failure the return is %NULL.
223 * Buffers may only be allocated from interrupts using a @gfp_mask of
226 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
229 struct kmem_cache *cache;
230 struct skb_shared_info *shinfo;
235 cache = (flags & SKB_ALLOC_FCLONE)
236 ? skbuff_fclone_cache : skbuff_head_cache;
238 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
239 gfp_mask |= __GFP_MEMALLOC;
242 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
247 /* We do our best to align skb_shared_info on a separate cache
248 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
249 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
250 * Both skb->head and skb_shared_info are cache line aligned.
252 size = SKB_DATA_ALIGN(size);
253 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
254 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
257 /* kmalloc(size) might give us more room than requested.
258 * Put skb_shared_info exactly at the end of allocated zone,
259 * to allow max possible filling before reallocation.
261 size = SKB_WITH_OVERHEAD(ksize(data));
262 prefetchw(data + size);
265 * Only clear those fields we need to clear, not those that we will
266 * actually initialise below. Hence, don't put any more fields after
267 * the tail pointer in struct sk_buff!
269 memset(skb, 0, offsetof(struct sk_buff, tail));
270 /* Account for allocated memory : skb + skb->head */
271 skb->truesize = SKB_TRUESIZE(size);
272 skb->pfmemalloc = pfmemalloc;
273 atomic_set(&skb->users, 1);
276 skb_reset_tail_pointer(skb);
277 skb->end = skb->tail + size;
278 #ifdef NET_SKBUFF_DATA_USES_OFFSET
279 skb->mac_header = ~0U;
280 skb->transport_header = ~0U;
283 /* make sure we initialize shinfo sequentially */
284 shinfo = skb_shinfo(skb);
285 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
286 atomic_set(&shinfo->dataref, 1);
287 kmemcheck_annotate_variable(shinfo->destructor_arg);
289 if (flags & SKB_ALLOC_FCLONE) {
290 struct sk_buff *child = skb + 1;
291 atomic_t *fclone_ref = (atomic_t *) (child + 1);
293 kmemcheck_annotate_bitfield(child, flags1);
294 kmemcheck_annotate_bitfield(child, flags2);
295 skb->fclone = SKB_FCLONE_ORIG;
296 atomic_set(fclone_ref, 1);
298 child->fclone = SKB_FCLONE_UNAVAILABLE;
299 child->pfmemalloc = pfmemalloc;
304 kmem_cache_free(cache, skb);
308 EXPORT_SYMBOL(__alloc_skb);
311 * build_skb - build a network buffer
312 * @data: data buffer provided by caller
313 * @frag_size: size of fragment, or 0 if head was kmalloced
315 * Allocate a new &sk_buff. Caller provides space holding head and
316 * skb_shared_info. @data must have been allocated by kmalloc()
317 * The return is the new skb buffer.
318 * On a failure the return is %NULL, and @data is not freed.
320 * Before IO, driver allocates only data buffer where NIC put incoming frame
321 * Driver should add room at head (NET_SKB_PAD) and
322 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
323 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
324 * before giving packet to stack.
325 * RX rings only contains data buffers, not full skbs.
327 struct sk_buff *build_skb(void *data, unsigned int frag_size)
329 struct skb_shared_info *shinfo;
331 unsigned int size = frag_size ? : ksize(data);
333 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
337 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
339 memset(skb, 0, offsetof(struct sk_buff, tail));
340 skb->truesize = SKB_TRUESIZE(size);
341 skb->head_frag = frag_size != 0;
342 atomic_set(&skb->users, 1);
345 skb_reset_tail_pointer(skb);
346 skb->end = skb->tail + size;
347 #ifdef NET_SKBUFF_DATA_USES_OFFSET
348 skb->mac_header = ~0U;
349 skb->transport_header = ~0U;
352 /* make sure we initialize shinfo sequentially */
353 shinfo = skb_shinfo(skb);
354 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
355 atomic_set(&shinfo->dataref, 1);
356 kmemcheck_annotate_variable(shinfo->destructor_arg);
360 EXPORT_SYMBOL(build_skb);
362 struct netdev_alloc_cache {
363 struct page_frag frag;
364 /* we maintain a pagecount bias, so that we dont dirty cache line
365 * containing page->_count every time we allocate a fragment.
367 unsigned int pagecnt_bias;
369 static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
371 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
373 struct netdev_alloc_cache *nc;
378 local_irq_save(flags);
379 nc = &__get_cpu_var(netdev_alloc_cache);
380 if (unlikely(!nc->frag.page)) {
382 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
383 gfp_t gfp = gfp_mask;
386 gfp |= __GFP_COMP | __GFP_NOWARN;
387 nc->frag.page = alloc_pages(gfp, order);
388 if (likely(nc->frag.page))
393 nc->frag.size = PAGE_SIZE << order;
395 atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS);
396 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
400 if (nc->frag.offset + fragsz > nc->frag.size) {
401 /* avoid unnecessary locked operations if possible */
402 if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) ||
403 atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count))
408 data = page_address(nc->frag.page) + nc->frag.offset;
409 nc->frag.offset += fragsz;
412 local_irq_restore(flags);
417 * netdev_alloc_frag - allocate a page fragment
418 * @fragsz: fragment size
420 * Allocates a frag from a page for receive buffer.
421 * Uses GFP_ATOMIC allocations.
423 void *netdev_alloc_frag(unsigned int fragsz)
425 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
427 EXPORT_SYMBOL(netdev_alloc_frag);
430 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
431 * @dev: network device to receive on
432 * @length: length to allocate
433 * @gfp_mask: get_free_pages mask, passed to alloc_skb
435 * Allocate a new &sk_buff and assign it a usage count of one. The
436 * buffer has unspecified headroom built in. Users should allocate
437 * the headroom they think they need without accounting for the
438 * built in space. The built in space is used for optimisations.
440 * %NULL is returned if there is no free memory.
442 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
443 unsigned int length, gfp_t gfp_mask)
445 struct sk_buff *skb = NULL;
446 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
447 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
449 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
452 if (sk_memalloc_socks())
453 gfp_mask |= __GFP_MEMALLOC;
455 data = __netdev_alloc_frag(fragsz, gfp_mask);
458 skb = build_skb(data, fragsz);
460 put_page(virt_to_head_page(data));
463 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
464 SKB_ALLOC_RX, NUMA_NO_NODE);
467 skb_reserve(skb, NET_SKB_PAD);
472 EXPORT_SYMBOL(__netdev_alloc_skb);
474 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
475 int size, unsigned int truesize)
477 skb_fill_page_desc(skb, i, page, off, size);
479 skb->data_len += size;
480 skb->truesize += truesize;
482 EXPORT_SYMBOL(skb_add_rx_frag);
484 static void skb_drop_list(struct sk_buff **listp)
486 kfree_skb_list(*listp);
490 static inline void skb_drop_fraglist(struct sk_buff *skb)
492 skb_drop_list(&skb_shinfo(skb)->frag_list);
495 static void skb_clone_fraglist(struct sk_buff *skb)
497 struct sk_buff *list;
499 skb_walk_frags(skb, list)
503 static void skb_free_head(struct sk_buff *skb)
506 put_page(virt_to_head_page(skb->head));
511 static void skb_release_data(struct sk_buff *skb)
514 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
515 &skb_shinfo(skb)->dataref)) {
516 if (skb_shinfo(skb)->nr_frags) {
518 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
519 skb_frag_unref(skb, i);
523 * If skb buf is from userspace, we need to notify the caller
524 * the lower device DMA has done;
526 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
527 struct ubuf_info *uarg;
529 uarg = skb_shinfo(skb)->destructor_arg;
531 uarg->callback(uarg, true);
534 if (skb_has_frag_list(skb))
535 skb_drop_fraglist(skb);
542 * Free an skbuff by memory without cleaning the state.
544 static void kfree_skbmem(struct sk_buff *skb)
546 struct sk_buff *other;
547 atomic_t *fclone_ref;
549 switch (skb->fclone) {
550 case SKB_FCLONE_UNAVAILABLE:
551 kmem_cache_free(skbuff_head_cache, skb);
554 case SKB_FCLONE_ORIG:
555 fclone_ref = (atomic_t *) (skb + 2);
556 if (atomic_dec_and_test(fclone_ref))
557 kmem_cache_free(skbuff_fclone_cache, skb);
560 case SKB_FCLONE_CLONE:
561 fclone_ref = (atomic_t *) (skb + 1);
564 /* The clone portion is available for
565 * fast-cloning again.
567 skb->fclone = SKB_FCLONE_UNAVAILABLE;
569 if (atomic_dec_and_test(fclone_ref))
570 kmem_cache_free(skbuff_fclone_cache, other);
575 static void skb_release_head_state(struct sk_buff *skb)
579 secpath_put(skb->sp);
581 if (skb->destructor) {
583 skb->destructor(skb);
585 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
586 nf_conntrack_put(skb->nfct);
588 #ifdef CONFIG_BRIDGE_NETFILTER
589 nf_bridge_put(skb->nf_bridge);
591 /* XXX: IS this still necessary? - JHS */
592 #ifdef CONFIG_NET_SCHED
594 #ifdef CONFIG_NET_CLS_ACT
600 /* Free everything but the sk_buff shell. */
601 static void skb_release_all(struct sk_buff *skb)
603 skb_release_head_state(skb);
604 if (likely(skb->head))
605 skb_release_data(skb);
609 * __kfree_skb - private function
612 * Free an sk_buff. Release anything attached to the buffer.
613 * Clean the state. This is an internal helper function. Users should
614 * always call kfree_skb
617 void __kfree_skb(struct sk_buff *skb)
619 skb_release_all(skb);
622 EXPORT_SYMBOL(__kfree_skb);
625 * kfree_skb - free an sk_buff
626 * @skb: buffer to free
628 * Drop a reference to the buffer and free it if the usage count has
631 void kfree_skb(struct sk_buff *skb)
635 if (likely(atomic_read(&skb->users) == 1))
637 else if (likely(!atomic_dec_and_test(&skb->users)))
639 trace_kfree_skb(skb, __builtin_return_address(0));
642 EXPORT_SYMBOL(kfree_skb);
644 void kfree_skb_list(struct sk_buff *segs)
647 struct sk_buff *next = segs->next;
653 EXPORT_SYMBOL(kfree_skb_list);
656 * skb_tx_error - report an sk_buff xmit error
657 * @skb: buffer that triggered an error
659 * Report xmit error if a device callback is tracking this skb.
660 * skb must be freed afterwards.
662 void skb_tx_error(struct sk_buff *skb)
664 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
665 struct ubuf_info *uarg;
667 uarg = skb_shinfo(skb)->destructor_arg;
669 uarg->callback(uarg, false);
670 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
673 EXPORT_SYMBOL(skb_tx_error);
676 * consume_skb - free an skbuff
677 * @skb: buffer to free
679 * Drop a ref to the buffer and free it if the usage count has hit zero
680 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
681 * is being dropped after a failure and notes that
683 void consume_skb(struct sk_buff *skb)
687 if (likely(atomic_read(&skb->users) == 1))
689 else if (likely(!atomic_dec_and_test(&skb->users)))
691 trace_consume_skb(skb);
694 EXPORT_SYMBOL(consume_skb);
696 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
698 new->tstamp = old->tstamp;
700 new->transport_header = old->transport_header;
701 new->network_header = old->network_header;
702 new->mac_header = old->mac_header;
703 new->inner_transport_header = old->inner_transport_header;
704 new->inner_network_header = old->inner_network_header;
705 new->inner_mac_header = old->inner_mac_header;
706 skb_dst_copy(new, old);
707 new->rxhash = old->rxhash;
708 new->ooo_okay = old->ooo_okay;
709 new->l4_rxhash = old->l4_rxhash;
710 new->no_fcs = old->no_fcs;
711 new->encapsulation = old->encapsulation;
713 new->sp = secpath_get(old->sp);
715 memcpy(new->cb, old->cb, sizeof(old->cb));
716 new->csum = old->csum;
717 new->local_df = old->local_df;
718 new->pkt_type = old->pkt_type;
719 new->ip_summed = old->ip_summed;
720 skb_copy_queue_mapping(new, old);
721 new->priority = old->priority;
722 #if IS_ENABLED(CONFIG_IP_VS)
723 new->ipvs_property = old->ipvs_property;
725 new->pfmemalloc = old->pfmemalloc;
726 new->protocol = old->protocol;
727 new->mark = old->mark;
728 new->skb_iif = old->skb_iif;
730 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
731 new->nf_trace = old->nf_trace;
733 #ifdef CONFIG_NET_SCHED
734 new->tc_index = old->tc_index;
735 #ifdef CONFIG_NET_CLS_ACT
736 new->tc_verd = old->tc_verd;
739 new->vlan_proto = old->vlan_proto;
740 new->vlan_tci = old->vlan_tci;
742 skb_copy_secmark(new, old);
746 * You should not add any new code to this function. Add it to
747 * __copy_skb_header above instead.
749 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
751 #define C(x) n->x = skb->x
753 n->next = n->prev = NULL;
755 __copy_skb_header(n, skb);
760 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
763 n->destructor = NULL;
770 atomic_set(&n->users, 1);
772 atomic_inc(&(skb_shinfo(skb)->dataref));
780 * skb_morph - morph one skb into another
781 * @dst: the skb to receive the contents
782 * @src: the skb to supply the contents
784 * This is identical to skb_clone except that the target skb is
785 * supplied by the user.
787 * The target skb is returned upon exit.
789 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
791 skb_release_all(dst);
792 return __skb_clone(dst, src);
794 EXPORT_SYMBOL_GPL(skb_morph);
797 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
798 * @skb: the skb to modify
799 * @gfp_mask: allocation priority
801 * This must be called on SKBTX_DEV_ZEROCOPY skb.
802 * It will copy all frags into kernel and drop the reference
803 * to userspace pages.
805 * If this function is called from an interrupt gfp_mask() must be
808 * Returns 0 on success or a negative error code on failure
809 * to allocate kernel memory to copy to.
811 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
814 int num_frags = skb_shinfo(skb)->nr_frags;
815 struct page *page, *head = NULL;
816 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
818 for (i = 0; i < num_frags; i++) {
820 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
822 page = alloc_page(gfp_mask);
825 struct page *next = (struct page *)head->private;
831 vaddr = kmap_atomic(skb_frag_page(f));
832 memcpy(page_address(page),
833 vaddr + f->page_offset, skb_frag_size(f));
834 kunmap_atomic(vaddr);
835 page->private = (unsigned long)head;
839 /* skb frags release userspace buffers */
840 for (i = 0; i < num_frags; i++)
841 skb_frag_unref(skb, i);
843 uarg->callback(uarg, false);
845 /* skb frags point to kernel buffers */
846 for (i = num_frags - 1; i >= 0; i--) {
847 __skb_fill_page_desc(skb, i, head, 0,
848 skb_shinfo(skb)->frags[i].size);
849 head = (struct page *)head->private;
852 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
855 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
858 * skb_clone - duplicate an sk_buff
859 * @skb: buffer to clone
860 * @gfp_mask: allocation priority
862 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
863 * copies share the same packet data but not structure. The new
864 * buffer has a reference count of 1. If the allocation fails the
865 * function returns %NULL otherwise the new buffer is returned.
867 * If this function is called from an interrupt gfp_mask() must be
871 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
875 if (skb_orphan_frags(skb, gfp_mask))
879 if (skb->fclone == SKB_FCLONE_ORIG &&
880 n->fclone == SKB_FCLONE_UNAVAILABLE) {
881 atomic_t *fclone_ref = (atomic_t *) (n + 1);
882 n->fclone = SKB_FCLONE_CLONE;
883 atomic_inc(fclone_ref);
885 if (skb_pfmemalloc(skb))
886 gfp_mask |= __GFP_MEMALLOC;
888 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
892 kmemcheck_annotate_bitfield(n, flags1);
893 kmemcheck_annotate_bitfield(n, flags2);
894 n->fclone = SKB_FCLONE_UNAVAILABLE;
897 return __skb_clone(n, skb);
899 EXPORT_SYMBOL(skb_clone);
901 static void skb_headers_offset_update(struct sk_buff *skb, int off)
903 /* {transport,network,mac}_header and tail are relative to skb->head */
904 skb->transport_header += off;
905 skb->network_header += off;
906 if (skb_mac_header_was_set(skb))
907 skb->mac_header += off;
908 skb->inner_transport_header += off;
909 skb->inner_network_header += off;
910 skb->inner_mac_header += off;
913 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
915 #ifndef NET_SKBUFF_DATA_USES_OFFSET
917 * Shift between the two data areas in bytes
919 unsigned long offset = new->data - old->data;
922 __copy_skb_header(new, old);
924 #ifndef NET_SKBUFF_DATA_USES_OFFSET
925 skb_headers_offset_update(new, offset);
927 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
928 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
929 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
932 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
934 if (skb_pfmemalloc(skb))
940 * skb_copy - create private copy of an sk_buff
941 * @skb: buffer to copy
942 * @gfp_mask: allocation priority
944 * Make a copy of both an &sk_buff and its data. This is used when the
945 * caller wishes to modify the data and needs a private copy of the
946 * data to alter. Returns %NULL on failure or the pointer to the buffer
947 * on success. The returned buffer has a reference count of 1.
949 * As by-product this function converts non-linear &sk_buff to linear
950 * one, so that &sk_buff becomes completely private and caller is allowed
951 * to modify all the data of returned buffer. This means that this
952 * function is not recommended for use in circumstances when only
953 * header is going to be modified. Use pskb_copy() instead.
956 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
958 int headerlen = skb_headroom(skb);
959 unsigned int size = skb_end_offset(skb) + skb->data_len;
960 struct sk_buff *n = __alloc_skb(size, gfp_mask,
961 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
966 /* Set the data pointer */
967 skb_reserve(n, headerlen);
968 /* Set the tail pointer and length */
969 skb_put(n, skb->len);
971 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
974 copy_skb_header(n, skb);
977 EXPORT_SYMBOL(skb_copy);
980 * __pskb_copy - create copy of an sk_buff with private head.
981 * @skb: buffer to copy
982 * @headroom: headroom of new skb
983 * @gfp_mask: allocation priority
985 * Make a copy of both an &sk_buff and part of its data, located
986 * in header. Fragmented data remain shared. This is used when
987 * the caller wishes to modify only header of &sk_buff and needs
988 * private copy of the header to alter. Returns %NULL on failure
989 * or the pointer to the buffer on success.
990 * The returned buffer has a reference count of 1.
993 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
995 unsigned int size = skb_headlen(skb) + headroom;
996 struct sk_buff *n = __alloc_skb(size, gfp_mask,
997 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1002 /* Set the data pointer */
1003 skb_reserve(n, headroom);
1004 /* Set the tail pointer and length */
1005 skb_put(n, skb_headlen(skb));
1006 /* Copy the bytes */
1007 skb_copy_from_linear_data(skb, n->data, n->len);
1009 n->truesize += skb->data_len;
1010 n->data_len = skb->data_len;
1013 if (skb_shinfo(skb)->nr_frags) {
1016 if (skb_orphan_frags(skb, gfp_mask)) {
1021 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1022 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1023 skb_frag_ref(skb, i);
1025 skb_shinfo(n)->nr_frags = i;
1028 if (skb_has_frag_list(skb)) {
1029 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1030 skb_clone_fraglist(n);
1033 copy_skb_header(n, skb);
1037 EXPORT_SYMBOL(__pskb_copy);
1040 * pskb_expand_head - reallocate header of &sk_buff
1041 * @skb: buffer to reallocate
1042 * @nhead: room to add at head
1043 * @ntail: room to add at tail
1044 * @gfp_mask: allocation priority
1046 * Expands (or creates identical copy, if &nhead and &ntail are zero)
1047 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
1048 * reference count of 1. Returns zero in the case of success or error,
1049 * if expansion failed. In the last case, &sk_buff is not changed.
1051 * All the pointers pointing into skb header may change and must be
1052 * reloaded after call to this function.
1055 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1060 int size = nhead + skb_end_offset(skb) + ntail;
1065 if (skb_shared(skb))
1068 size = SKB_DATA_ALIGN(size);
1070 if (skb_pfmemalloc(skb))
1071 gfp_mask |= __GFP_MEMALLOC;
1072 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1073 gfp_mask, NUMA_NO_NODE, NULL);
1076 size = SKB_WITH_OVERHEAD(ksize(data));
1078 /* Copy only real data... and, alas, header. This should be
1079 * optimized for the cases when header is void.
1081 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1083 memcpy((struct skb_shared_info *)(data + size),
1085 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1088 * if shinfo is shared we must drop the old head gracefully, but if it
1089 * is not we can just drop the old head and let the existing refcount
1090 * be since all we did is relocate the values
1092 if (skb_cloned(skb)) {
1093 /* copy this zero copy skb frags */
1094 if (skb_orphan_frags(skb, gfp_mask))
1096 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1097 skb_frag_ref(skb, i);
1099 if (skb_has_frag_list(skb))
1100 skb_clone_fraglist(skb);
1102 skb_release_data(skb);
1106 off = (data + nhead) - skb->head;
1111 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1115 skb->end = skb->head + size;
1118 skb_headers_offset_update(skb, 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 skb_headers_offset_update(n, off);
1219 EXPORT_SYMBOL(skb_copy_expand);
1222 * skb_pad - zero pad the tail of an skb
1223 * @skb: buffer to pad
1224 * @pad: space to pad
1226 * Ensure that a buffer is followed by a padding area that is zero
1227 * filled. Used by network drivers which may DMA or transfer data
1228 * beyond the buffer end onto the wire.
1230 * May return error in out of memory cases. The skb is freed on error.
1233 int skb_pad(struct sk_buff *skb, int pad)
1238 /* If the skbuff is non linear tailroom is always zero.. */
1239 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1240 memset(skb->data+skb->len, 0, pad);
1244 ntail = skb->data_len + pad - (skb->end - skb->tail);
1245 if (likely(skb_cloned(skb) || ntail > 0)) {
1246 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1251 /* FIXME: The use of this function with non-linear skb's really needs
1254 err = skb_linearize(skb);
1258 memset(skb->data + skb->len, 0, pad);
1265 EXPORT_SYMBOL(skb_pad);
1268 * skb_put - add data to a buffer
1269 * @skb: buffer to use
1270 * @len: amount of data to add
1272 * This function extends the used data area of the buffer. If this would
1273 * exceed the total buffer size the kernel will panic. A pointer to the
1274 * first byte of the extra data is returned.
1276 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1278 unsigned char *tmp = skb_tail_pointer(skb);
1279 SKB_LINEAR_ASSERT(skb);
1282 if (unlikely(skb->tail > skb->end))
1283 skb_over_panic(skb, len, __builtin_return_address(0));
1286 EXPORT_SYMBOL(skb_put);
1289 * skb_push - add data to the start of a buffer
1290 * @skb: buffer to use
1291 * @len: amount of data to add
1293 * This function extends the used data area of the buffer at the buffer
1294 * start. If this would exceed the total buffer headroom the kernel will
1295 * panic. A pointer to the first byte of the extra data is returned.
1297 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1301 if (unlikely(skb->data<skb->head))
1302 skb_under_panic(skb, len, __builtin_return_address(0));
1305 EXPORT_SYMBOL(skb_push);
1308 * skb_pull - remove data from the start of a buffer
1309 * @skb: buffer to use
1310 * @len: amount of data to remove
1312 * This function removes data from the start of a buffer, returning
1313 * the memory to the headroom. A pointer to the next data in the buffer
1314 * is returned. Once the data has been pulled future pushes will overwrite
1317 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1319 return skb_pull_inline(skb, len);
1321 EXPORT_SYMBOL(skb_pull);
1324 * skb_trim - remove end from a buffer
1325 * @skb: buffer to alter
1328 * Cut the length of a buffer down by removing data from the tail. If
1329 * the buffer is already under the length specified it is not modified.
1330 * The skb must be linear.
1332 void skb_trim(struct sk_buff *skb, unsigned int len)
1335 __skb_trim(skb, len);
1337 EXPORT_SYMBOL(skb_trim);
1339 /* Trims skb to length len. It can change skb pointers.
1342 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1344 struct sk_buff **fragp;
1345 struct sk_buff *frag;
1346 int offset = skb_headlen(skb);
1347 int nfrags = skb_shinfo(skb)->nr_frags;
1351 if (skb_cloned(skb) &&
1352 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1359 for (; i < nfrags; i++) {
1360 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1367 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1370 skb_shinfo(skb)->nr_frags = i;
1372 for (; i < nfrags; i++)
1373 skb_frag_unref(skb, i);
1375 if (skb_has_frag_list(skb))
1376 skb_drop_fraglist(skb);
1380 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1381 fragp = &frag->next) {
1382 int end = offset + frag->len;
1384 if (skb_shared(frag)) {
1385 struct sk_buff *nfrag;
1387 nfrag = skb_clone(frag, GFP_ATOMIC);
1388 if (unlikely(!nfrag))
1391 nfrag->next = frag->next;
1403 unlikely((err = pskb_trim(frag, len - offset))))
1407 skb_drop_list(&frag->next);
1412 if (len > skb_headlen(skb)) {
1413 skb->data_len -= skb->len - len;
1418 skb_set_tail_pointer(skb, len);
1423 EXPORT_SYMBOL(___pskb_trim);
1426 * __pskb_pull_tail - advance tail of skb header
1427 * @skb: buffer to reallocate
1428 * @delta: number of bytes to advance tail
1430 * The function makes a sense only on a fragmented &sk_buff,
1431 * it expands header moving its tail forward and copying necessary
1432 * data from fragmented part.
1434 * &sk_buff MUST have reference count of 1.
1436 * Returns %NULL (and &sk_buff does not change) if pull failed
1437 * or value of new tail of skb in the case of success.
1439 * All the pointers pointing into skb header may change and must be
1440 * reloaded after call to this function.
1443 /* Moves tail of skb head forward, copying data from fragmented part,
1444 * when it is necessary.
1445 * 1. It may fail due to malloc failure.
1446 * 2. It may change skb pointers.
1448 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1450 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1452 /* If skb has not enough free space at tail, get new one
1453 * plus 128 bytes for future expansions. If we have enough
1454 * room at tail, reallocate without expansion only if skb is cloned.
1456 int i, k, eat = (skb->tail + delta) - skb->end;
1458 if (eat > 0 || skb_cloned(skb)) {
1459 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1464 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1467 /* Optimization: no fragments, no reasons to preestimate
1468 * size of pulled pages. Superb.
1470 if (!skb_has_frag_list(skb))
1473 /* Estimate size of pulled pages. */
1475 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1476 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1483 /* If we need update frag list, we are in troubles.
1484 * Certainly, it possible to add an offset to skb data,
1485 * but taking into account that pulling is expected to
1486 * be very rare operation, it is worth to fight against
1487 * further bloating skb head and crucify ourselves here instead.
1488 * Pure masohism, indeed. 8)8)
1491 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1492 struct sk_buff *clone = NULL;
1493 struct sk_buff *insp = NULL;
1498 if (list->len <= eat) {
1499 /* Eaten as whole. */
1504 /* Eaten partially. */
1506 if (skb_shared(list)) {
1507 /* Sucks! We need to fork list. :-( */
1508 clone = skb_clone(list, GFP_ATOMIC);
1514 /* This may be pulled without
1518 if (!pskb_pull(list, eat)) {
1526 /* Free pulled out fragments. */
1527 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1528 skb_shinfo(skb)->frag_list = list->next;
1531 /* And insert new clone at head. */
1534 skb_shinfo(skb)->frag_list = clone;
1537 /* Success! Now we may commit changes to skb data. */
1542 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1543 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1546 skb_frag_unref(skb, i);
1549 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1551 skb_shinfo(skb)->frags[k].page_offset += eat;
1552 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1558 skb_shinfo(skb)->nr_frags = k;
1561 skb->data_len -= delta;
1563 return skb_tail_pointer(skb);
1565 EXPORT_SYMBOL(__pskb_pull_tail);
1568 * skb_copy_bits - copy bits from skb to kernel buffer
1570 * @offset: offset in source
1571 * @to: destination buffer
1572 * @len: number of bytes to copy
1574 * Copy the specified number of bytes from the source skb to the
1575 * destination buffer.
1578 * If its prototype is ever changed,
1579 * check arch/{*}/net/{*}.S files,
1580 * since it is called from BPF assembly code.
1582 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1584 int start = skb_headlen(skb);
1585 struct sk_buff *frag_iter;
1588 if (offset > (int)skb->len - len)
1592 if ((copy = start - offset) > 0) {
1595 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1596 if ((len -= copy) == 0)
1602 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1604 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1606 WARN_ON(start > offset + len);
1608 end = start + skb_frag_size(f);
1609 if ((copy = end - offset) > 0) {
1615 vaddr = kmap_atomic(skb_frag_page(f));
1617 vaddr + f->page_offset + offset - start,
1619 kunmap_atomic(vaddr);
1621 if ((len -= copy) == 0)
1629 skb_walk_frags(skb, frag_iter) {
1632 WARN_ON(start > offset + len);
1634 end = start + frag_iter->len;
1635 if ((copy = end - offset) > 0) {
1638 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1640 if ((len -= copy) == 0)
1654 EXPORT_SYMBOL(skb_copy_bits);
1657 * Callback from splice_to_pipe(), if we need to release some pages
1658 * at the end of the spd in case we error'ed out in filling the pipe.
1660 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1662 put_page(spd->pages[i]);
1665 static struct page *linear_to_page(struct page *page, unsigned int *len,
1666 unsigned int *offset,
1669 struct page_frag *pfrag = sk_page_frag(sk);
1671 if (!sk_page_frag_refill(sk, pfrag))
1674 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1676 memcpy(page_address(pfrag->page) + pfrag->offset,
1677 page_address(page) + *offset, *len);
1678 *offset = pfrag->offset;
1679 pfrag->offset += *len;
1684 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1686 unsigned int offset)
1688 return spd->nr_pages &&
1689 spd->pages[spd->nr_pages - 1] == page &&
1690 (spd->partial[spd->nr_pages - 1].offset +
1691 spd->partial[spd->nr_pages - 1].len == offset);
1695 * Fill page/offset/length into spd, if it can hold more pages.
1697 static bool spd_fill_page(struct splice_pipe_desc *spd,
1698 struct pipe_inode_info *pipe, struct page *page,
1699 unsigned int *len, unsigned int offset,
1703 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1707 page = linear_to_page(page, len, &offset, sk);
1711 if (spd_can_coalesce(spd, page, offset)) {
1712 spd->partial[spd->nr_pages - 1].len += *len;
1716 spd->pages[spd->nr_pages] = page;
1717 spd->partial[spd->nr_pages].len = *len;
1718 spd->partial[spd->nr_pages].offset = offset;
1724 static bool __splice_segment(struct page *page, unsigned int poff,
1725 unsigned int plen, unsigned int *off,
1727 struct splice_pipe_desc *spd, bool linear,
1729 struct pipe_inode_info *pipe)
1734 /* skip this segment if already processed */
1740 /* ignore any bits we already processed */
1746 unsigned int flen = min(*len, plen);
1748 if (spd_fill_page(spd, pipe, page, &flen, poff,
1754 } while (*len && plen);
1760 * Map linear and fragment data from the skb to spd. It reports true if the
1761 * pipe is full or if we already spliced the requested length.
1763 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1764 unsigned int *offset, unsigned int *len,
1765 struct splice_pipe_desc *spd, struct sock *sk)
1769 /* map the linear part :
1770 * If skb->head_frag is set, this 'linear' part is backed by a
1771 * fragment, and if the head is not shared with any clones then
1772 * we can avoid a copy since we own the head portion of this page.
1774 if (__splice_segment(virt_to_page(skb->data),
1775 (unsigned long) skb->data & (PAGE_SIZE - 1),
1778 skb_head_is_locked(skb),
1783 * then map the fragments
1785 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1786 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1788 if (__splice_segment(skb_frag_page(f),
1789 f->page_offset, skb_frag_size(f),
1790 offset, len, spd, false, sk, pipe))
1798 * Map data from the skb to a pipe. Should handle both the linear part,
1799 * the fragments, and the frag list. It does NOT handle frag lists within
1800 * the frag list, if such a thing exists. We'd probably need to recurse to
1801 * handle that cleanly.
1803 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1804 struct pipe_inode_info *pipe, unsigned int tlen,
1807 struct partial_page partial[MAX_SKB_FRAGS];
1808 struct page *pages[MAX_SKB_FRAGS];
1809 struct splice_pipe_desc spd = {
1812 .nr_pages_max = MAX_SKB_FRAGS,
1814 .ops = &sock_pipe_buf_ops,
1815 .spd_release = sock_spd_release,
1817 struct sk_buff *frag_iter;
1818 struct sock *sk = skb->sk;
1822 * __skb_splice_bits() only fails if the output has no room left,
1823 * so no point in going over the frag_list for the error case.
1825 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1831 * now see if we have a frag_list to map
1833 skb_walk_frags(skb, frag_iter) {
1836 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1843 * Drop the socket lock, otherwise we have reverse
1844 * locking dependencies between sk_lock and i_mutex
1845 * here as compared to sendfile(). We enter here
1846 * with the socket lock held, and splice_to_pipe() will
1847 * grab the pipe inode lock. For sendfile() emulation,
1848 * we call into ->sendpage() with the i_mutex lock held
1849 * and networking will grab the socket lock.
1852 ret = splice_to_pipe(pipe, &spd);
1860 * skb_store_bits - store bits from kernel buffer to skb
1861 * @skb: destination buffer
1862 * @offset: offset in destination
1863 * @from: source buffer
1864 * @len: number of bytes to copy
1866 * Copy the specified number of bytes from the source buffer to the
1867 * destination skb. This function handles all the messy bits of
1868 * traversing fragment lists and such.
1871 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1873 int start = skb_headlen(skb);
1874 struct sk_buff *frag_iter;
1877 if (offset > (int)skb->len - len)
1880 if ((copy = start - offset) > 0) {
1883 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1884 if ((len -= copy) == 0)
1890 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1891 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1894 WARN_ON(start > offset + len);
1896 end = start + skb_frag_size(frag);
1897 if ((copy = end - offset) > 0) {
1903 vaddr = kmap_atomic(skb_frag_page(frag));
1904 memcpy(vaddr + frag->page_offset + offset - start,
1906 kunmap_atomic(vaddr);
1908 if ((len -= copy) == 0)
1916 skb_walk_frags(skb, frag_iter) {
1919 WARN_ON(start > offset + len);
1921 end = start + frag_iter->len;
1922 if ((copy = end - offset) > 0) {
1925 if (skb_store_bits(frag_iter, offset - start,
1928 if ((len -= copy) == 0)
1941 EXPORT_SYMBOL(skb_store_bits);
1943 /* Checksum skb data. */
1945 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1946 int len, __wsum csum)
1948 int start = skb_headlen(skb);
1949 int i, copy = start - offset;
1950 struct sk_buff *frag_iter;
1953 /* Checksum header. */
1957 csum = csum_partial(skb->data + offset, copy, csum);
1958 if ((len -= copy) == 0)
1964 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1966 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1968 WARN_ON(start > offset + len);
1970 end = start + skb_frag_size(frag);
1971 if ((copy = end - offset) > 0) {
1977 vaddr = kmap_atomic(skb_frag_page(frag));
1978 csum2 = csum_partial(vaddr + frag->page_offset +
1979 offset - start, copy, 0);
1980 kunmap_atomic(vaddr);
1981 csum = csum_block_add(csum, csum2, pos);
1990 skb_walk_frags(skb, frag_iter) {
1993 WARN_ON(start > offset + len);
1995 end = start + frag_iter->len;
1996 if ((copy = end - offset) > 0) {
2000 csum2 = skb_checksum(frag_iter, offset - start,
2002 csum = csum_block_add(csum, csum2, pos);
2003 if ((len -= copy) == 0)
2014 EXPORT_SYMBOL(skb_checksum);
2016 /* Both of above in one bottle. */
2018 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2019 u8 *to, int len, __wsum csum)
2021 int start = skb_headlen(skb);
2022 int i, copy = start - offset;
2023 struct sk_buff *frag_iter;
2030 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2032 if ((len -= copy) == 0)
2039 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2042 WARN_ON(start > offset + len);
2044 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2045 if ((copy = end - offset) > 0) {
2048 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2052 vaddr = kmap_atomic(skb_frag_page(frag));
2053 csum2 = csum_partial_copy_nocheck(vaddr +
2057 kunmap_atomic(vaddr);
2058 csum = csum_block_add(csum, csum2, pos);
2068 skb_walk_frags(skb, frag_iter) {
2072 WARN_ON(start > offset + len);
2074 end = start + frag_iter->len;
2075 if ((copy = end - offset) > 0) {
2078 csum2 = skb_copy_and_csum_bits(frag_iter,
2081 csum = csum_block_add(csum, csum2, pos);
2082 if ((len -= copy) == 0)
2093 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2095 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2100 if (skb->ip_summed == CHECKSUM_PARTIAL)
2101 csstart = skb_checksum_start_offset(skb);
2103 csstart = skb_headlen(skb);
2105 BUG_ON(csstart > skb_headlen(skb));
2107 skb_copy_from_linear_data(skb, to, csstart);
2110 if (csstart != skb->len)
2111 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2112 skb->len - csstart, 0);
2114 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2115 long csstuff = csstart + skb->csum_offset;
2117 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2120 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2123 * skb_dequeue - remove from the head of the queue
2124 * @list: list to dequeue from
2126 * Remove the head of the list. The list lock is taken so the function
2127 * may be used safely with other locking list functions. The head item is
2128 * returned or %NULL if the list is empty.
2131 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2133 unsigned long flags;
2134 struct sk_buff *result;
2136 spin_lock_irqsave(&list->lock, flags);
2137 result = __skb_dequeue(list);
2138 spin_unlock_irqrestore(&list->lock, flags);
2141 EXPORT_SYMBOL(skb_dequeue);
2144 * skb_dequeue_tail - remove from the tail of the queue
2145 * @list: list to dequeue from
2147 * Remove the tail of the list. The list lock is taken so the function
2148 * may be used safely with other locking list functions. The tail item is
2149 * returned or %NULL if the list is empty.
2151 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2153 unsigned long flags;
2154 struct sk_buff *result;
2156 spin_lock_irqsave(&list->lock, flags);
2157 result = __skb_dequeue_tail(list);
2158 spin_unlock_irqrestore(&list->lock, flags);
2161 EXPORT_SYMBOL(skb_dequeue_tail);
2164 * skb_queue_purge - empty a list
2165 * @list: list to empty
2167 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2168 * the list and one reference dropped. This function takes the list
2169 * lock and is atomic with respect to other list locking functions.
2171 void skb_queue_purge(struct sk_buff_head *list)
2173 struct sk_buff *skb;
2174 while ((skb = skb_dequeue(list)) != NULL)
2177 EXPORT_SYMBOL(skb_queue_purge);
2180 * skb_queue_head - queue a buffer at the list head
2181 * @list: list to use
2182 * @newsk: buffer to queue
2184 * Queue a buffer at the start of the list. This function takes the
2185 * list lock and can be used safely with other locking &sk_buff functions
2188 * A buffer cannot be placed on two lists at the same time.
2190 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2192 unsigned long flags;
2194 spin_lock_irqsave(&list->lock, flags);
2195 __skb_queue_head(list, newsk);
2196 spin_unlock_irqrestore(&list->lock, flags);
2198 EXPORT_SYMBOL(skb_queue_head);
2201 * skb_queue_tail - queue a buffer at the list tail
2202 * @list: list to use
2203 * @newsk: buffer to queue
2205 * Queue a buffer at the tail 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_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2213 unsigned long flags;
2215 spin_lock_irqsave(&list->lock, flags);
2216 __skb_queue_tail(list, newsk);
2217 spin_unlock_irqrestore(&list->lock, flags);
2219 EXPORT_SYMBOL(skb_queue_tail);
2222 * skb_unlink - remove a buffer from a list
2223 * @skb: buffer to remove
2224 * @list: list to use
2226 * Remove a packet from a list. The list locks are taken and this
2227 * function is atomic with respect to other list locked calls
2229 * You must know what list the SKB is on.
2231 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2233 unsigned long flags;
2235 spin_lock_irqsave(&list->lock, flags);
2236 __skb_unlink(skb, list);
2237 spin_unlock_irqrestore(&list->lock, flags);
2239 EXPORT_SYMBOL(skb_unlink);
2242 * skb_append - append a buffer
2243 * @old: buffer to insert after
2244 * @newsk: buffer to insert
2245 * @list: list to use
2247 * Place a packet after a given packet in a list. The list locks are taken
2248 * and this function is atomic with respect to other list locked calls.
2249 * A buffer cannot be placed on two lists at the same time.
2251 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2253 unsigned long flags;
2255 spin_lock_irqsave(&list->lock, flags);
2256 __skb_queue_after(list, old, newsk);
2257 spin_unlock_irqrestore(&list->lock, flags);
2259 EXPORT_SYMBOL(skb_append);
2262 * skb_insert - insert a buffer
2263 * @old: buffer to insert before
2264 * @newsk: buffer to insert
2265 * @list: list to use
2267 * Place a packet before a given packet in a list. The list locks are
2268 * taken and this function is atomic with respect to other list locked
2271 * A buffer cannot be placed on two lists at the same time.
2273 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2275 unsigned long flags;
2277 spin_lock_irqsave(&list->lock, flags);
2278 __skb_insert(newsk, old->prev, old, list);
2279 spin_unlock_irqrestore(&list->lock, flags);
2281 EXPORT_SYMBOL(skb_insert);
2283 static inline void skb_split_inside_header(struct sk_buff *skb,
2284 struct sk_buff* skb1,
2285 const u32 len, const int pos)
2289 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2291 /* And move data appendix as is. */
2292 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2293 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2295 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2296 skb_shinfo(skb)->nr_frags = 0;
2297 skb1->data_len = skb->data_len;
2298 skb1->len += skb1->data_len;
2301 skb_set_tail_pointer(skb, len);
2304 static inline void skb_split_no_header(struct sk_buff *skb,
2305 struct sk_buff* skb1,
2306 const u32 len, int pos)
2309 const int nfrags = skb_shinfo(skb)->nr_frags;
2311 skb_shinfo(skb)->nr_frags = 0;
2312 skb1->len = skb1->data_len = skb->len - len;
2314 skb->data_len = len - pos;
2316 for (i = 0; i < nfrags; i++) {
2317 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2319 if (pos + size > len) {
2320 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2324 * We have two variants in this case:
2325 * 1. Move all the frag to the second
2326 * part, if it is possible. F.e.
2327 * this approach is mandatory for TUX,
2328 * where splitting is expensive.
2329 * 2. Split is accurately. We make this.
2331 skb_frag_ref(skb, i);
2332 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2333 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2334 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2335 skb_shinfo(skb)->nr_frags++;
2339 skb_shinfo(skb)->nr_frags++;
2342 skb_shinfo(skb1)->nr_frags = k;
2346 * skb_split - Split fragmented skb to two parts at length len.
2347 * @skb: the buffer to split
2348 * @skb1: the buffer to receive the second part
2349 * @len: new length for skb
2351 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2353 int pos = skb_headlen(skb);
2355 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2356 if (len < pos) /* Split line is inside header. */
2357 skb_split_inside_header(skb, skb1, len, pos);
2358 else /* Second chunk has no header, nothing to copy. */
2359 skb_split_no_header(skb, skb1, len, pos);
2361 EXPORT_SYMBOL(skb_split);
2363 /* Shifting from/to a cloned skb is a no-go.
2365 * Caller cannot keep skb_shinfo related pointers past calling here!
2367 static int skb_prepare_for_shift(struct sk_buff *skb)
2369 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2373 * skb_shift - Shifts paged data partially from skb to another
2374 * @tgt: buffer into which tail data gets added
2375 * @skb: buffer from which the paged data comes from
2376 * @shiftlen: shift up to this many bytes
2378 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2379 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2380 * It's up to caller to free skb if everything was shifted.
2382 * If @tgt runs out of frags, the whole operation is aborted.
2384 * Skb cannot include anything else but paged data while tgt is allowed
2385 * to have non-paged data as well.
2387 * TODO: full sized shift could be optimized but that would need
2388 * specialized skb free'er to handle frags without up-to-date nr_frags.
2390 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2392 int from, to, merge, todo;
2393 struct skb_frag_struct *fragfrom, *fragto;
2395 BUG_ON(shiftlen > skb->len);
2396 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2400 to = skb_shinfo(tgt)->nr_frags;
2401 fragfrom = &skb_shinfo(skb)->frags[from];
2403 /* Actual merge is delayed until the point when we know we can
2404 * commit all, so that we don't have to undo partial changes
2407 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2408 fragfrom->page_offset)) {
2413 todo -= skb_frag_size(fragfrom);
2415 if (skb_prepare_for_shift(skb) ||
2416 skb_prepare_for_shift(tgt))
2419 /* All previous frag pointers might be stale! */
2420 fragfrom = &skb_shinfo(skb)->frags[from];
2421 fragto = &skb_shinfo(tgt)->frags[merge];
2423 skb_frag_size_add(fragto, shiftlen);
2424 skb_frag_size_sub(fragfrom, shiftlen);
2425 fragfrom->page_offset += shiftlen;
2433 /* Skip full, not-fitting skb to avoid expensive operations */
2434 if ((shiftlen == skb->len) &&
2435 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2438 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2441 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2442 if (to == MAX_SKB_FRAGS)
2445 fragfrom = &skb_shinfo(skb)->frags[from];
2446 fragto = &skb_shinfo(tgt)->frags[to];
2448 if (todo >= skb_frag_size(fragfrom)) {
2449 *fragto = *fragfrom;
2450 todo -= skb_frag_size(fragfrom);
2455 __skb_frag_ref(fragfrom);
2456 fragto->page = fragfrom->page;
2457 fragto->page_offset = fragfrom->page_offset;
2458 skb_frag_size_set(fragto, todo);
2460 fragfrom->page_offset += todo;
2461 skb_frag_size_sub(fragfrom, todo);
2469 /* Ready to "commit" this state change to tgt */
2470 skb_shinfo(tgt)->nr_frags = to;
2473 fragfrom = &skb_shinfo(skb)->frags[0];
2474 fragto = &skb_shinfo(tgt)->frags[merge];
2476 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2477 __skb_frag_unref(fragfrom);
2480 /* Reposition in the original skb */
2482 while (from < skb_shinfo(skb)->nr_frags)
2483 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2484 skb_shinfo(skb)->nr_frags = to;
2486 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2489 /* Most likely the tgt won't ever need its checksum anymore, skb on
2490 * the other hand might need it if it needs to be resent
2492 tgt->ip_summed = CHECKSUM_PARTIAL;
2493 skb->ip_summed = CHECKSUM_PARTIAL;
2495 /* Yak, is it really working this way? Some helper please? */
2496 skb->len -= shiftlen;
2497 skb->data_len -= shiftlen;
2498 skb->truesize -= shiftlen;
2499 tgt->len += shiftlen;
2500 tgt->data_len += shiftlen;
2501 tgt->truesize += shiftlen;
2507 * skb_prepare_seq_read - Prepare a sequential read of skb data
2508 * @skb: the buffer to read
2509 * @from: lower offset of data to be read
2510 * @to: upper offset of data to be read
2511 * @st: state variable
2513 * Initializes the specified state variable. Must be called before
2514 * invoking skb_seq_read() for the first time.
2516 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2517 unsigned int to, struct skb_seq_state *st)
2519 st->lower_offset = from;
2520 st->upper_offset = to;
2521 st->root_skb = st->cur_skb = skb;
2522 st->frag_idx = st->stepped_offset = 0;
2523 st->frag_data = NULL;
2525 EXPORT_SYMBOL(skb_prepare_seq_read);
2528 * skb_seq_read - Sequentially read skb data
2529 * @consumed: number of bytes consumed by the caller so far
2530 * @data: destination pointer for data to be returned
2531 * @st: state variable
2533 * Reads a block of skb data at &consumed relative to the
2534 * lower offset specified to skb_prepare_seq_read(). Assigns
2535 * the head of the data block to &data and returns the length
2536 * of the block or 0 if the end of the skb data or the upper
2537 * offset has been reached.
2539 * The caller is not required to consume all of the data
2540 * returned, i.e. &consumed is typically set to the number
2541 * of bytes already consumed and the next call to
2542 * skb_seq_read() will return the remaining part of the block.
2544 * Note 1: The size of each block of data returned can be arbitrary,
2545 * this limitation is the cost for zerocopy seqeuental
2546 * reads of potentially non linear data.
2548 * Note 2: Fragment lists within fragments are not implemented
2549 * at the moment, state->root_skb could be replaced with
2550 * a stack for this purpose.
2552 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2553 struct skb_seq_state *st)
2555 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2558 if (unlikely(abs_offset >= st->upper_offset))
2562 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2564 if (abs_offset < block_limit && !st->frag_data) {
2565 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2566 return block_limit - abs_offset;
2569 if (st->frag_idx == 0 && !st->frag_data)
2570 st->stepped_offset += skb_headlen(st->cur_skb);
2572 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2573 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2574 block_limit = skb_frag_size(frag) + st->stepped_offset;
2576 if (abs_offset < block_limit) {
2578 st->frag_data = kmap_atomic(skb_frag_page(frag));
2580 *data = (u8 *) st->frag_data + frag->page_offset +
2581 (abs_offset - st->stepped_offset);
2583 return block_limit - abs_offset;
2586 if (st->frag_data) {
2587 kunmap_atomic(st->frag_data);
2588 st->frag_data = NULL;
2592 st->stepped_offset += skb_frag_size(frag);
2595 if (st->frag_data) {
2596 kunmap_atomic(st->frag_data);
2597 st->frag_data = NULL;
2600 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2601 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2604 } else if (st->cur_skb->next) {
2605 st->cur_skb = st->cur_skb->next;
2612 EXPORT_SYMBOL(skb_seq_read);
2615 * skb_abort_seq_read - Abort a sequential read of skb data
2616 * @st: state variable
2618 * Must be called if skb_seq_read() was not called until it
2621 void skb_abort_seq_read(struct skb_seq_state *st)
2624 kunmap_atomic(st->frag_data);
2626 EXPORT_SYMBOL(skb_abort_seq_read);
2628 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2630 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2631 struct ts_config *conf,
2632 struct ts_state *state)
2634 return skb_seq_read(offset, text, TS_SKB_CB(state));
2637 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2639 skb_abort_seq_read(TS_SKB_CB(state));
2643 * skb_find_text - Find a text pattern in skb data
2644 * @skb: the buffer to look in
2645 * @from: search offset
2647 * @config: textsearch configuration
2648 * @state: uninitialized textsearch state variable
2650 * Finds a pattern in the skb data according to the specified
2651 * textsearch configuration. Use textsearch_next() to retrieve
2652 * subsequent occurrences of the pattern. Returns the offset
2653 * to the first occurrence or UINT_MAX if no match was found.
2655 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2656 unsigned int to, struct ts_config *config,
2657 struct ts_state *state)
2661 config->get_next_block = skb_ts_get_next_block;
2662 config->finish = skb_ts_finish;
2664 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2666 ret = textsearch_find(config, state);
2667 return (ret <= to - from ? ret : UINT_MAX);
2669 EXPORT_SYMBOL(skb_find_text);
2672 * skb_append_datato_frags - append the user data to a skb
2673 * @sk: sock structure
2674 * @skb: skb structure to be appened with user data.
2675 * @getfrag: call back function to be used for getting the user data
2676 * @from: pointer to user message iov
2677 * @length: length of the iov message
2679 * Description: This procedure append the user data in the fragment part
2680 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2682 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2683 int (*getfrag)(void *from, char *to, int offset,
2684 int len, int odd, struct sk_buff *skb),
2685 void *from, int length)
2687 int frg_cnt = skb_shinfo(skb)->nr_frags;
2691 struct page_frag *pfrag = ¤t->task_frag;
2694 /* Return error if we don't have space for new frag */
2695 if (frg_cnt >= MAX_SKB_FRAGS)
2698 if (!sk_page_frag_refill(sk, pfrag))
2701 /* copy the user data to page */
2702 copy = min_t(int, length, pfrag->size - pfrag->offset);
2704 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2705 offset, copy, 0, skb);
2709 /* copy was successful so update the size parameters */
2710 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2713 pfrag->offset += copy;
2714 get_page(pfrag->page);
2716 skb->truesize += copy;
2717 atomic_add(copy, &sk->sk_wmem_alloc);
2719 skb->data_len += copy;
2723 } while (length > 0);
2727 EXPORT_SYMBOL(skb_append_datato_frags);
2730 * skb_pull_rcsum - pull skb and update receive checksum
2731 * @skb: buffer to update
2732 * @len: length of data pulled
2734 * This function performs an skb_pull on the packet and updates
2735 * the CHECKSUM_COMPLETE checksum. It should be used on
2736 * receive path processing instead of skb_pull unless you know
2737 * that the checksum difference is zero (e.g., a valid IP header)
2738 * or you are setting ip_summed to CHECKSUM_NONE.
2740 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2742 BUG_ON(len > skb->len);
2744 BUG_ON(skb->len < skb->data_len);
2745 skb_postpull_rcsum(skb, skb->data, len);
2746 return skb->data += len;
2748 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2751 * skb_segment - Perform protocol segmentation on skb.
2752 * @skb: buffer to segment
2753 * @features: features for the output path (see dev->features)
2755 * This function performs segmentation on the given skb. It returns
2756 * a pointer to the first in a list of new skbs for the segments.
2757 * In case of error it returns ERR_PTR(err).
2759 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2761 struct sk_buff *segs = NULL;
2762 struct sk_buff *tail = NULL;
2763 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2764 unsigned int mss = skb_shinfo(skb)->gso_size;
2765 unsigned int doffset = skb->data - skb_mac_header(skb);
2766 unsigned int offset = doffset;
2767 unsigned int tnl_hlen = skb_tnl_header_len(skb);
2768 unsigned int headroom;
2772 int sg = !!(features & NETIF_F_SG);
2773 int nfrags = skb_shinfo(skb)->nr_frags;
2778 proto = skb_network_protocol(skb);
2779 if (unlikely(!proto))
2780 return ERR_PTR(-EINVAL);
2782 csum = !!can_checksum_protocol(features, proto);
2783 __skb_push(skb, doffset);
2784 headroom = skb_headroom(skb);
2785 pos = skb_headlen(skb);
2788 struct sk_buff *nskb;
2793 len = skb->len - offset;
2797 hsize = skb_headlen(skb) - offset;
2800 if (hsize > len || !sg)
2803 if (!hsize && i >= nfrags) {
2804 BUG_ON(fskb->len != len);
2807 nskb = skb_clone(fskb, GFP_ATOMIC);
2810 if (unlikely(!nskb))
2813 hsize = skb_end_offset(nskb);
2814 if (skb_cow_head(nskb, doffset + headroom)) {
2819 nskb->truesize += skb_end_offset(nskb) - hsize;
2820 skb_release_head_state(nskb);
2821 __skb_push(nskb, doffset);
2823 nskb = __alloc_skb(hsize + doffset + headroom,
2824 GFP_ATOMIC, skb_alloc_rx_flag(skb),
2827 if (unlikely(!nskb))
2830 skb_reserve(nskb, headroom);
2831 __skb_put(nskb, doffset);
2840 __copy_skb_header(nskb, skb);
2841 nskb->mac_len = skb->mac_len;
2843 /* nskb and skb might have different headroom */
2844 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2845 nskb->csum_start += skb_headroom(nskb) - headroom;
2847 skb_reset_mac_header(nskb);
2848 skb_set_network_header(nskb, skb->mac_len);
2849 nskb->transport_header = (nskb->network_header +
2850 skb_network_header_len(skb));
2852 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
2853 nskb->data - tnl_hlen,
2854 doffset + tnl_hlen);
2856 if (fskb != skb_shinfo(skb)->frag_list)
2857 goto perform_csum_check;
2860 nskb->ip_summed = CHECKSUM_NONE;
2861 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2867 frag = skb_shinfo(nskb)->frags;
2869 skb_copy_from_linear_data_offset(skb, offset,
2870 skb_put(nskb, hsize), hsize);
2872 skb_shinfo(nskb)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2874 while (pos < offset + len && i < nfrags) {
2875 *frag = skb_shinfo(skb)->frags[i];
2876 __skb_frag_ref(frag);
2877 size = skb_frag_size(frag);
2880 frag->page_offset += offset - pos;
2881 skb_frag_size_sub(frag, offset - pos);
2884 skb_shinfo(nskb)->nr_frags++;
2886 if (pos + size <= offset + len) {
2890 skb_frag_size_sub(frag, pos + size - (offset + len));
2897 if (pos < offset + len) {
2898 struct sk_buff *fskb2 = fskb;
2900 BUG_ON(pos + fskb->len != offset + len);
2906 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2912 SKB_FRAG_ASSERT(nskb);
2913 skb_shinfo(nskb)->frag_list = fskb2;
2917 nskb->data_len = len - hsize;
2918 nskb->len += nskb->data_len;
2919 nskb->truesize += nskb->data_len;
2923 nskb->csum = skb_checksum(nskb, doffset,
2924 nskb->len - doffset, 0);
2925 nskb->ip_summed = CHECKSUM_NONE;
2927 } while ((offset += len) < skb->len);
2932 while ((skb = segs)) {
2936 return ERR_PTR(err);
2938 EXPORT_SYMBOL_GPL(skb_segment);
2940 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2942 struct sk_buff *p = *head;
2943 struct sk_buff *nskb;
2944 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2945 struct skb_shared_info *pinfo = skb_shinfo(p);
2946 unsigned int headroom;
2947 unsigned int len = skb_gro_len(skb);
2948 unsigned int offset = skb_gro_offset(skb);
2949 unsigned int headlen = skb_headlen(skb);
2950 unsigned int delta_truesize;
2952 if (p->len + len >= 65536)
2955 if (pinfo->frag_list)
2957 else if (headlen <= offset) {
2960 int i = skbinfo->nr_frags;
2961 int nr_frags = pinfo->nr_frags + i;
2965 if (nr_frags > MAX_SKB_FRAGS)
2968 pinfo->nr_frags = nr_frags;
2969 skbinfo->nr_frags = 0;
2971 frag = pinfo->frags + nr_frags;
2972 frag2 = skbinfo->frags + i;
2977 frag->page_offset += offset;
2978 skb_frag_size_sub(frag, offset);
2980 /* all fragments truesize : remove (head size + sk_buff) */
2981 delta_truesize = skb->truesize -
2982 SKB_TRUESIZE(skb_end_offset(skb));
2984 skb->truesize -= skb->data_len;
2985 skb->len -= skb->data_len;
2988 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
2990 } else if (skb->head_frag) {
2991 int nr_frags = pinfo->nr_frags;
2992 skb_frag_t *frag = pinfo->frags + nr_frags;
2993 struct page *page = virt_to_head_page(skb->head);
2994 unsigned int first_size = headlen - offset;
2995 unsigned int first_offset;
2997 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3000 first_offset = skb->data -
3001 (unsigned char *)page_address(page) +
3004 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3006 frag->page.p = page;
3007 frag->page_offset = first_offset;
3008 skb_frag_size_set(frag, first_size);
3010 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3011 /* We dont need to clear skbinfo->nr_frags here */
3013 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3014 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3016 } else if (skb_gro_len(p) != pinfo->gso_size)
3019 headroom = skb_headroom(p);
3020 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
3021 if (unlikely(!nskb))
3024 __copy_skb_header(nskb, p);
3025 nskb->mac_len = p->mac_len;
3027 skb_reserve(nskb, headroom);
3028 __skb_put(nskb, skb_gro_offset(p));
3030 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3031 skb_set_network_header(nskb, skb_network_offset(p));
3032 skb_set_transport_header(nskb, skb_transport_offset(p));
3034 __skb_pull(p, skb_gro_offset(p));
3035 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3036 p->data - skb_mac_header(p));
3038 skb_shinfo(nskb)->frag_list = p;
3039 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3040 pinfo->gso_size = 0;
3041 skb_header_release(p);
3042 NAPI_GRO_CB(nskb)->last = p;
3044 nskb->data_len += p->len;
3045 nskb->truesize += p->truesize;
3046 nskb->len += p->len;
3049 nskb->next = p->next;
3055 delta_truesize = skb->truesize;
3056 if (offset > headlen) {
3057 unsigned int eat = offset - headlen;
3059 skbinfo->frags[0].page_offset += eat;
3060 skb_frag_size_sub(&skbinfo->frags[0], eat);
3061 skb->data_len -= eat;
3066 __skb_pull(skb, offset);
3068 NAPI_GRO_CB(p)->last->next = skb;
3069 NAPI_GRO_CB(p)->last = skb;
3070 skb_header_release(skb);
3073 NAPI_GRO_CB(p)->count++;
3075 p->truesize += delta_truesize;
3078 NAPI_GRO_CB(skb)->same_flow = 1;
3081 EXPORT_SYMBOL_GPL(skb_gro_receive);
3083 void __init skb_init(void)
3085 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3086 sizeof(struct sk_buff),
3088 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3090 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3091 (2*sizeof(struct sk_buff)) +
3094 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3099 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3100 * @skb: Socket buffer containing the buffers to be mapped
3101 * @sg: The scatter-gather list to map into
3102 * @offset: The offset into the buffer's contents to start mapping
3103 * @len: Length of buffer space to be mapped
3105 * Fill the specified scatter-gather list with mappings/pointers into a
3106 * region of the buffer space attached to a socket buffer.
3109 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3111 int start = skb_headlen(skb);
3112 int i, copy = start - offset;
3113 struct sk_buff *frag_iter;
3119 sg_set_buf(sg, skb->data + offset, copy);
3121 if ((len -= copy) == 0)
3126 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3129 WARN_ON(start > offset + len);
3131 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3132 if ((copy = end - offset) > 0) {
3133 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3137 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3138 frag->page_offset+offset-start);
3147 skb_walk_frags(skb, frag_iter) {
3150 WARN_ON(start > offset + len);
3152 end = start + frag_iter->len;
3153 if ((copy = end - offset) > 0) {
3156 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3158 if ((len -= copy) == 0)
3168 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3170 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3172 sg_mark_end(&sg[nsg - 1]);
3176 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3179 * skb_cow_data - Check that a socket buffer's data buffers are writable
3180 * @skb: The socket buffer to check.
3181 * @tailbits: Amount of trailing space to be added
3182 * @trailer: Returned pointer to the skb where the @tailbits space begins
3184 * Make sure that the data buffers attached to a socket buffer are
3185 * writable. If they are not, private copies are made of the data buffers
3186 * and the socket buffer is set to use these instead.
3188 * If @tailbits is given, make sure that there is space to write @tailbits
3189 * bytes of data beyond current end of socket buffer. @trailer will be
3190 * set to point to the skb in which this space begins.
3192 * The number of scatterlist elements required to completely map the
3193 * COW'd and extended socket buffer will be returned.
3195 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3199 struct sk_buff *skb1, **skb_p;
3201 /* If skb is cloned or its head is paged, reallocate
3202 * head pulling out all the pages (pages are considered not writable
3203 * at the moment even if they are anonymous).
3205 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3206 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3209 /* Easy case. Most of packets will go this way. */
3210 if (!skb_has_frag_list(skb)) {
3211 /* A little of trouble, not enough of space for trailer.
3212 * This should not happen, when stack is tuned to generate
3213 * good frames. OK, on miss we reallocate and reserve even more
3214 * space, 128 bytes is fair. */
3216 if (skb_tailroom(skb) < tailbits &&
3217 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3225 /* Misery. We are in troubles, going to mincer fragments... */
3228 skb_p = &skb_shinfo(skb)->frag_list;
3231 while ((skb1 = *skb_p) != NULL) {
3234 /* The fragment is partially pulled by someone,
3235 * this can happen on input. Copy it and everything
3238 if (skb_shared(skb1))
3241 /* If the skb is the last, worry about trailer. */
3243 if (skb1->next == NULL && tailbits) {
3244 if (skb_shinfo(skb1)->nr_frags ||
3245 skb_has_frag_list(skb1) ||
3246 skb_tailroom(skb1) < tailbits)
3247 ntail = tailbits + 128;
3253 skb_shinfo(skb1)->nr_frags ||
3254 skb_has_frag_list(skb1)) {
3255 struct sk_buff *skb2;
3257 /* Fuck, we are miserable poor guys... */
3259 skb2 = skb_copy(skb1, GFP_ATOMIC);
3261 skb2 = skb_copy_expand(skb1,
3265 if (unlikely(skb2 == NULL))
3269 skb_set_owner_w(skb2, skb1->sk);
3271 /* Looking around. Are we still alive?
3272 * OK, link new skb, drop old one */
3274 skb2->next = skb1->next;
3281 skb_p = &skb1->next;
3286 EXPORT_SYMBOL_GPL(skb_cow_data);
3288 static void sock_rmem_free(struct sk_buff *skb)
3290 struct sock *sk = skb->sk;
3292 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3296 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3298 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3302 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3303 (unsigned int)sk->sk_rcvbuf)
3308 skb->destructor = sock_rmem_free;
3309 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3311 /* before exiting rcu section, make sure dst is refcounted */
3314 skb_queue_tail(&sk->sk_error_queue, skb);
3315 if (!sock_flag(sk, SOCK_DEAD))
3316 sk->sk_data_ready(sk, len);
3319 EXPORT_SYMBOL(sock_queue_err_skb);
3321 void skb_tstamp_tx(struct sk_buff *orig_skb,
3322 struct skb_shared_hwtstamps *hwtstamps)
3324 struct sock *sk = orig_skb->sk;
3325 struct sock_exterr_skb *serr;
3326 struct sk_buff *skb;
3333 *skb_hwtstamps(orig_skb) =
3337 * no hardware time stamps available,
3338 * so keep the shared tx_flags and only
3339 * store software time stamp
3341 orig_skb->tstamp = ktime_get_real();
3344 skb = skb_clone(orig_skb, GFP_ATOMIC);
3348 serr = SKB_EXT_ERR(skb);
3349 memset(serr, 0, sizeof(*serr));
3350 serr->ee.ee_errno = ENOMSG;
3351 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3353 err = sock_queue_err_skb(sk, skb);
3358 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3360 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3362 struct sock *sk = skb->sk;
3363 struct sock_exterr_skb *serr;
3366 skb->wifi_acked_valid = 1;
3367 skb->wifi_acked = acked;
3369 serr = SKB_EXT_ERR(skb);
3370 memset(serr, 0, sizeof(*serr));
3371 serr->ee.ee_errno = ENOMSG;
3372 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3374 err = sock_queue_err_skb(sk, skb);
3378 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3382 * skb_partial_csum_set - set up and verify partial csum values for packet
3383 * @skb: the skb to set
3384 * @start: the number of bytes after skb->data to start checksumming.
3385 * @off: the offset from start to place the checksum.
3387 * For untrusted partially-checksummed packets, we need to make sure the values
3388 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3390 * This function checks and sets those values and skb->ip_summed: if this
3391 * returns false you should drop the packet.
3393 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3395 if (unlikely(start > skb_headlen(skb)) ||
3396 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3397 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3398 start, off, skb_headlen(skb));
3401 skb->ip_summed = CHECKSUM_PARTIAL;
3402 skb->csum_start = skb_headroom(skb) + start;
3403 skb->csum_offset = off;
3404 skb_set_transport_header(skb, start);
3407 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3409 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3411 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3414 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3416 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3419 skb_release_head_state(skb);
3420 kmem_cache_free(skbuff_head_cache, skb);
3425 EXPORT_SYMBOL(kfree_skb_partial);
3428 * skb_try_coalesce - try to merge skb to prior one
3430 * @from: buffer to add
3431 * @fragstolen: pointer to boolean
3432 * @delta_truesize: how much more was allocated than was requested
3434 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3435 bool *fragstolen, int *delta_truesize)
3437 int i, delta, len = from->len;
3439 *fragstolen = false;
3444 if (len <= skb_tailroom(to)) {
3445 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3446 *delta_truesize = 0;
3450 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3453 if (skb_headlen(from) != 0) {
3455 unsigned int offset;
3457 if (skb_shinfo(to)->nr_frags +
3458 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3461 if (skb_head_is_locked(from))
3464 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3466 page = virt_to_head_page(from->head);
3467 offset = from->data - (unsigned char *)page_address(page);
3469 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3470 page, offset, skb_headlen(from));
3473 if (skb_shinfo(to)->nr_frags +
3474 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3477 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
3480 WARN_ON_ONCE(delta < len);
3482 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3483 skb_shinfo(from)->frags,
3484 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3485 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3487 if (!skb_cloned(from))
3488 skb_shinfo(from)->nr_frags = 0;
3490 /* if the skb is not cloned this does nothing
3491 * since we set nr_frags to 0.
3493 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3494 skb_frag_ref(from, i);
3496 to->truesize += delta;
3498 to->data_len += len;
3500 *delta_truesize = delta;
3503 EXPORT_SYMBOL(skb_try_coalesce);