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>
68 #include <net/ip6_checksum.h>
71 #include <asm/uaccess.h>
72 #include <trace/events/skb.h>
73 #include <linux/highmem.h>
75 struct kmem_cache *skbuff_head_cache __read_mostly;
76 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
78 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
79 struct pipe_buffer *buf)
84 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
85 struct pipe_buffer *buf)
90 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
91 struct pipe_buffer *buf)
97 /* Pipe buffer operations for a socket. */
98 static const struct pipe_buf_operations sock_pipe_buf_ops = {
100 .map = generic_pipe_buf_map,
101 .unmap = generic_pipe_buf_unmap,
102 .confirm = generic_pipe_buf_confirm,
103 .release = sock_pipe_buf_release,
104 .steal = sock_pipe_buf_steal,
105 .get = sock_pipe_buf_get,
109 * skb_panic - private function for out-of-line support
113 * @msg: skb_over_panic or skb_under_panic
115 * Out-of-line support for skb_put() and skb_push().
116 * Called via the wrapper skb_over_panic() or skb_under_panic().
117 * Keep out of line to prevent kernel bloat.
118 * __builtin_return_address is not used because it is not always reliable.
120 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
123 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
124 msg, addr, skb->len, sz, skb->head, skb->data,
125 (unsigned long)skb->tail, (unsigned long)skb->end,
126 skb->dev ? skb->dev->name : "<NULL>");
130 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
132 skb_panic(skb, sz, addr, __func__);
135 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
137 skb_panic(skb, sz, addr, __func__);
141 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
142 * the caller if emergency pfmemalloc reserves are being used. If it is and
143 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
144 * may be used. Otherwise, the packet data may be discarded until enough
147 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
148 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
150 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
151 unsigned long ip, bool *pfmemalloc)
154 bool ret_pfmemalloc = false;
157 * Try a regular allocation, when that fails and we're not entitled
158 * to the reserves, fail.
160 obj = kmalloc_node_track_caller(size,
161 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
163 if (obj || !(gfp_pfmemalloc_allowed(flags)))
166 /* Try again but now we are using pfmemalloc reserves */
167 ret_pfmemalloc = true;
168 obj = kmalloc_node_track_caller(size, flags, node);
172 *pfmemalloc = ret_pfmemalloc;
177 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
178 * 'private' fields and also do memory statistics to find all the
183 struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
188 skb = kmem_cache_alloc_node(skbuff_head_cache,
189 gfp_mask & ~__GFP_DMA, node);
194 * Only clear those fields we need to clear, not those that we will
195 * actually initialise below. Hence, don't put any more fields after
196 * the tail pointer in struct sk_buff!
198 memset(skb, 0, offsetof(struct sk_buff, tail));
200 skb->truesize = sizeof(struct sk_buff);
201 atomic_set(&skb->users, 1);
203 skb->mac_header = (typeof(skb->mac_header))~0U;
209 * __alloc_skb - allocate a network buffer
210 * @size: size to allocate
211 * @gfp_mask: allocation mask
212 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
213 * instead of head cache and allocate a cloned (child) skb.
214 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
215 * allocations in case the data is required for writeback
216 * @node: numa node to allocate memory on
218 * Allocate a new &sk_buff. The returned buffer has no headroom and a
219 * tail room of at least size bytes. The object has a reference count
220 * of one. The return is the buffer. On a failure the return is %NULL.
222 * Buffers may only be allocated from interrupts using a @gfp_mask of
225 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
228 struct kmem_cache *cache;
229 struct skb_shared_info *shinfo;
234 cache = (flags & SKB_ALLOC_FCLONE)
235 ? skbuff_fclone_cache : skbuff_head_cache;
237 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
238 gfp_mask |= __GFP_MEMALLOC;
241 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
246 /* We do our best to align skb_shared_info on a separate cache
247 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
248 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
249 * Both skb->head and skb_shared_info are cache line aligned.
251 size = SKB_DATA_ALIGN(size);
252 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
253 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
256 /* kmalloc(size) might give us more room than requested.
257 * Put skb_shared_info exactly at the end of allocated zone,
258 * to allow max possible filling before reallocation.
260 size = SKB_WITH_OVERHEAD(ksize(data));
261 prefetchw(data + size);
264 * Only clear those fields we need to clear, not those that we will
265 * actually initialise below. Hence, don't put any more fields after
266 * the tail pointer in struct sk_buff!
268 memset(skb, 0, offsetof(struct sk_buff, tail));
269 /* Account for allocated memory : skb + skb->head */
270 skb->truesize = SKB_TRUESIZE(size);
271 skb->pfmemalloc = pfmemalloc;
272 atomic_set(&skb->users, 1);
275 skb_reset_tail_pointer(skb);
276 skb->end = skb->tail + size;
277 skb->mac_header = (typeof(skb->mac_header))~0U;
278 skb->transport_header = (typeof(skb->transport_header))~0U;
280 /* make sure we initialize shinfo sequentially */
281 shinfo = skb_shinfo(skb);
282 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
283 atomic_set(&shinfo->dataref, 1);
284 kmemcheck_annotate_variable(shinfo->destructor_arg);
286 if (flags & SKB_ALLOC_FCLONE) {
287 struct sk_buff *child = skb + 1;
288 atomic_t *fclone_ref = (atomic_t *) (child + 1);
290 kmemcheck_annotate_bitfield(child, flags1);
291 kmemcheck_annotate_bitfield(child, flags2);
292 skb->fclone = SKB_FCLONE_ORIG;
293 atomic_set(fclone_ref, 1);
295 child->fclone = SKB_FCLONE_UNAVAILABLE;
296 child->pfmemalloc = pfmemalloc;
301 kmem_cache_free(cache, skb);
305 EXPORT_SYMBOL(__alloc_skb);
308 * build_skb - build a network buffer
309 * @data: data buffer provided by caller
310 * @frag_size: size of fragment, or 0 if head was kmalloced
312 * Allocate a new &sk_buff. Caller provides space holding head and
313 * skb_shared_info. @data must have been allocated by kmalloc() only if
314 * @frag_size is 0, otherwise data should come from the page allocator.
315 * The return is the new skb buffer.
316 * On a failure the return is %NULL, and @data is not freed.
318 * Before IO, driver allocates only data buffer where NIC put incoming frame
319 * Driver should add room at head (NET_SKB_PAD) and
320 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
321 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
322 * before giving packet to stack.
323 * RX rings only contains data buffers, not full skbs.
325 struct sk_buff *build_skb(void *data, unsigned int frag_size)
327 struct skb_shared_info *shinfo;
329 unsigned int size = frag_size ? : ksize(data);
331 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
335 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
337 memset(skb, 0, offsetof(struct sk_buff, tail));
338 skb->truesize = SKB_TRUESIZE(size);
339 skb->head_frag = frag_size != 0;
340 atomic_set(&skb->users, 1);
343 skb_reset_tail_pointer(skb);
344 skb->end = skb->tail + size;
345 skb->mac_header = (typeof(skb->mac_header))~0U;
346 skb->transport_header = (typeof(skb->transport_header))~0U;
348 /* make sure we initialize shinfo sequentially */
349 shinfo = skb_shinfo(skb);
350 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
351 atomic_set(&shinfo->dataref, 1);
352 kmemcheck_annotate_variable(shinfo->destructor_arg);
356 EXPORT_SYMBOL(build_skb);
358 struct netdev_alloc_cache {
359 struct page_frag frag;
360 /* we maintain a pagecount bias, so that we dont dirty cache line
361 * containing page->_count every time we allocate a fragment.
363 unsigned int pagecnt_bias;
365 static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
367 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
369 struct netdev_alloc_cache *nc;
374 local_irq_save(flags);
375 nc = &__get_cpu_var(netdev_alloc_cache);
376 if (unlikely(!nc->frag.page)) {
378 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
379 gfp_t gfp = gfp_mask;
382 gfp |= __GFP_COMP | __GFP_NOWARN;
383 nc->frag.page = alloc_pages(gfp, order);
384 if (likely(nc->frag.page))
389 nc->frag.size = PAGE_SIZE << order;
391 atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS);
392 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
396 if (nc->frag.offset + fragsz > nc->frag.size) {
397 /* avoid unnecessary locked operations if possible */
398 if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) ||
399 atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count))
404 data = page_address(nc->frag.page) + nc->frag.offset;
405 nc->frag.offset += fragsz;
408 local_irq_restore(flags);
413 * netdev_alloc_frag - allocate a page fragment
414 * @fragsz: fragment size
416 * Allocates a frag from a page for receive buffer.
417 * Uses GFP_ATOMIC allocations.
419 void *netdev_alloc_frag(unsigned int fragsz)
421 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
423 EXPORT_SYMBOL(netdev_alloc_frag);
426 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
427 * @dev: network device to receive on
428 * @length: length to allocate
429 * @gfp_mask: get_free_pages mask, passed to alloc_skb
431 * Allocate a new &sk_buff and assign it a usage count of one. The
432 * buffer has unspecified headroom built in. Users should allocate
433 * the headroom they think they need without accounting for the
434 * built in space. The built in space is used for optimisations.
436 * %NULL is returned if there is no free memory.
438 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
439 unsigned int length, gfp_t gfp_mask)
441 struct sk_buff *skb = NULL;
442 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
443 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
445 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
448 if (sk_memalloc_socks())
449 gfp_mask |= __GFP_MEMALLOC;
451 data = __netdev_alloc_frag(fragsz, gfp_mask);
454 skb = build_skb(data, fragsz);
456 put_page(virt_to_head_page(data));
459 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
460 SKB_ALLOC_RX, NUMA_NO_NODE);
463 skb_reserve(skb, NET_SKB_PAD);
468 EXPORT_SYMBOL(__netdev_alloc_skb);
470 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
471 int size, unsigned int truesize)
473 skb_fill_page_desc(skb, i, page, off, size);
475 skb->data_len += size;
476 skb->truesize += truesize;
478 EXPORT_SYMBOL(skb_add_rx_frag);
480 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
481 unsigned int truesize)
483 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
485 skb_frag_size_add(frag, size);
487 skb->data_len += size;
488 skb->truesize += truesize;
490 EXPORT_SYMBOL(skb_coalesce_rx_frag);
492 static void skb_drop_list(struct sk_buff **listp)
494 kfree_skb_list(*listp);
498 static inline void skb_drop_fraglist(struct sk_buff *skb)
500 skb_drop_list(&skb_shinfo(skb)->frag_list);
503 static void skb_clone_fraglist(struct sk_buff *skb)
505 struct sk_buff *list;
507 skb_walk_frags(skb, list)
511 static void skb_free_head(struct sk_buff *skb)
514 put_page(virt_to_head_page(skb->head));
519 static void skb_release_data(struct sk_buff *skb)
522 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
523 &skb_shinfo(skb)->dataref)) {
524 if (skb_shinfo(skb)->nr_frags) {
526 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
527 skb_frag_unref(skb, i);
531 * If skb buf is from userspace, we need to notify the caller
532 * the lower device DMA has done;
534 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
535 struct ubuf_info *uarg;
537 uarg = skb_shinfo(skb)->destructor_arg;
539 uarg->callback(uarg, true);
542 if (skb_has_frag_list(skb))
543 skb_drop_fraglist(skb);
550 * Free an skbuff by memory without cleaning the state.
552 static void kfree_skbmem(struct sk_buff *skb)
554 struct sk_buff *other;
555 atomic_t *fclone_ref;
557 switch (skb->fclone) {
558 case SKB_FCLONE_UNAVAILABLE:
559 kmem_cache_free(skbuff_head_cache, skb);
562 case SKB_FCLONE_ORIG:
563 fclone_ref = (atomic_t *) (skb + 2);
564 if (atomic_dec_and_test(fclone_ref))
565 kmem_cache_free(skbuff_fclone_cache, skb);
568 case SKB_FCLONE_CLONE:
569 fclone_ref = (atomic_t *) (skb + 1);
572 /* The clone portion is available for
573 * fast-cloning again.
575 skb->fclone = SKB_FCLONE_UNAVAILABLE;
577 if (atomic_dec_and_test(fclone_ref))
578 kmem_cache_free(skbuff_fclone_cache, other);
583 static void skb_release_head_state(struct sk_buff *skb)
587 secpath_put(skb->sp);
589 if (skb->destructor) {
591 skb->destructor(skb);
593 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
594 nf_conntrack_put(skb->nfct);
596 #ifdef CONFIG_BRIDGE_NETFILTER
597 nf_bridge_put(skb->nf_bridge);
599 /* XXX: IS this still necessary? - JHS */
600 #ifdef CONFIG_NET_SCHED
602 #ifdef CONFIG_NET_CLS_ACT
608 /* Free everything but the sk_buff shell. */
609 static void skb_release_all(struct sk_buff *skb)
611 skb_release_head_state(skb);
612 if (likely(skb->head))
613 skb_release_data(skb);
617 * __kfree_skb - private function
620 * Free an sk_buff. Release anything attached to the buffer.
621 * Clean the state. This is an internal helper function. Users should
622 * always call kfree_skb
625 void __kfree_skb(struct sk_buff *skb)
627 skb_release_all(skb);
630 EXPORT_SYMBOL(__kfree_skb);
633 * kfree_skb - free an sk_buff
634 * @skb: buffer to free
636 * Drop a reference to the buffer and free it if the usage count has
639 void kfree_skb(struct sk_buff *skb)
643 if (likely(atomic_read(&skb->users) == 1))
645 else if (likely(!atomic_dec_and_test(&skb->users)))
647 trace_kfree_skb(skb, __builtin_return_address(0));
650 EXPORT_SYMBOL(kfree_skb);
652 void kfree_skb_list(struct sk_buff *segs)
655 struct sk_buff *next = segs->next;
661 EXPORT_SYMBOL(kfree_skb_list);
664 * skb_tx_error - report an sk_buff xmit error
665 * @skb: buffer that triggered an error
667 * Report xmit error if a device callback is tracking this skb.
668 * skb must be freed afterwards.
670 void skb_tx_error(struct sk_buff *skb)
672 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
673 struct ubuf_info *uarg;
675 uarg = skb_shinfo(skb)->destructor_arg;
677 uarg->callback(uarg, false);
678 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
681 EXPORT_SYMBOL(skb_tx_error);
684 * consume_skb - free an skbuff
685 * @skb: buffer to free
687 * Drop a ref to the buffer and free it if the usage count has hit zero
688 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
689 * is being dropped after a failure and notes that
691 void consume_skb(struct sk_buff *skb)
695 if (likely(atomic_read(&skb->users) == 1))
697 else if (likely(!atomic_dec_and_test(&skb->users)))
699 trace_consume_skb(skb);
702 EXPORT_SYMBOL(consume_skb);
704 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
706 new->tstamp = old->tstamp;
708 new->transport_header = old->transport_header;
709 new->network_header = old->network_header;
710 new->mac_header = old->mac_header;
711 new->inner_protocol = old->inner_protocol;
712 new->inner_transport_header = old->inner_transport_header;
713 new->inner_network_header = old->inner_network_header;
714 new->inner_mac_header = old->inner_mac_header;
715 skb_dst_copy(new, old);
716 skb_copy_hash(new, old);
717 new->ooo_okay = old->ooo_okay;
718 new->no_fcs = old->no_fcs;
719 new->encapsulation = old->encapsulation;
721 new->sp = secpath_get(old->sp);
723 memcpy(new->cb, old->cb, sizeof(old->cb));
724 new->csum = old->csum;
725 new->local_df = old->local_df;
726 new->pkt_type = old->pkt_type;
727 new->ip_summed = old->ip_summed;
728 skb_copy_queue_mapping(new, old);
729 new->priority = old->priority;
730 #if IS_ENABLED(CONFIG_IP_VS)
731 new->ipvs_property = old->ipvs_property;
733 new->pfmemalloc = old->pfmemalloc;
734 new->protocol = old->protocol;
735 new->mark = old->mark;
736 new->skb_iif = old->skb_iif;
738 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
739 new->nf_trace = old->nf_trace;
741 #ifdef CONFIG_NET_SCHED
742 new->tc_index = old->tc_index;
743 #ifdef CONFIG_NET_CLS_ACT
744 new->tc_verd = old->tc_verd;
747 new->vlan_proto = old->vlan_proto;
748 new->vlan_tci = old->vlan_tci;
750 skb_copy_secmark(new, old);
752 #ifdef CONFIG_NET_RX_BUSY_POLL
753 new->napi_id = old->napi_id;
758 * You should not add any new code to this function. Add it to
759 * __copy_skb_header above instead.
761 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
763 #define C(x) n->x = skb->x
765 n->next = n->prev = NULL;
767 __copy_skb_header(n, skb);
772 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
775 n->destructor = NULL;
782 atomic_set(&n->users, 1);
784 atomic_inc(&(skb_shinfo(skb)->dataref));
792 * skb_morph - morph one skb into another
793 * @dst: the skb to receive the contents
794 * @src: the skb to supply the contents
796 * This is identical to skb_clone except that the target skb is
797 * supplied by the user.
799 * The target skb is returned upon exit.
801 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
803 skb_release_all(dst);
804 return __skb_clone(dst, src);
806 EXPORT_SYMBOL_GPL(skb_morph);
809 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
810 * @skb: the skb to modify
811 * @gfp_mask: allocation priority
813 * This must be called on SKBTX_DEV_ZEROCOPY skb.
814 * It will copy all frags into kernel and drop the reference
815 * to userspace pages.
817 * If this function is called from an interrupt gfp_mask() must be
820 * Returns 0 on success or a negative error code on failure
821 * to allocate kernel memory to copy to.
823 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
826 int num_frags = skb_shinfo(skb)->nr_frags;
827 struct page *page, *head = NULL;
828 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
830 for (i = 0; i < num_frags; i++) {
832 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
834 page = alloc_page(gfp_mask);
837 struct page *next = (struct page *)page_private(head);
843 vaddr = kmap_atomic(skb_frag_page(f));
844 memcpy(page_address(page),
845 vaddr + f->page_offset, skb_frag_size(f));
846 kunmap_atomic(vaddr);
847 set_page_private(page, (unsigned long)head);
851 /* skb frags release userspace buffers */
852 for (i = 0; i < num_frags; i++)
853 skb_frag_unref(skb, i);
855 uarg->callback(uarg, false);
857 /* skb frags point to kernel buffers */
858 for (i = num_frags - 1; i >= 0; i--) {
859 __skb_fill_page_desc(skb, i, head, 0,
860 skb_shinfo(skb)->frags[i].size);
861 head = (struct page *)page_private(head);
864 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
867 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
870 * skb_clone - duplicate an sk_buff
871 * @skb: buffer to clone
872 * @gfp_mask: allocation priority
874 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
875 * copies share the same packet data but not structure. The new
876 * buffer has a reference count of 1. If the allocation fails the
877 * function returns %NULL otherwise the new buffer is returned.
879 * If this function is called from an interrupt gfp_mask() must be
883 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
887 if (skb_orphan_frags(skb, gfp_mask))
891 if (skb->fclone == SKB_FCLONE_ORIG &&
892 n->fclone == SKB_FCLONE_UNAVAILABLE) {
893 atomic_t *fclone_ref = (atomic_t *) (n + 1);
894 n->fclone = SKB_FCLONE_CLONE;
895 atomic_inc(fclone_ref);
897 if (skb_pfmemalloc(skb))
898 gfp_mask |= __GFP_MEMALLOC;
900 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
904 kmemcheck_annotate_bitfield(n, flags1);
905 kmemcheck_annotate_bitfield(n, flags2);
906 n->fclone = SKB_FCLONE_UNAVAILABLE;
909 return __skb_clone(n, skb);
911 EXPORT_SYMBOL(skb_clone);
913 static void skb_headers_offset_update(struct sk_buff *skb, int off)
915 /* Only adjust this if it actually is csum_start rather than csum */
916 if (skb->ip_summed == CHECKSUM_PARTIAL)
917 skb->csum_start += off;
918 /* {transport,network,mac}_header and tail are relative to skb->head */
919 skb->transport_header += off;
920 skb->network_header += off;
921 if (skb_mac_header_was_set(skb))
922 skb->mac_header += off;
923 skb->inner_transport_header += off;
924 skb->inner_network_header += off;
925 skb->inner_mac_header += off;
928 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
930 __copy_skb_header(new, old);
932 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
933 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
934 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
937 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
939 if (skb_pfmemalloc(skb))
945 * skb_copy - create private copy of an sk_buff
946 * @skb: buffer to copy
947 * @gfp_mask: allocation priority
949 * Make a copy of both an &sk_buff and its data. This is used when the
950 * caller wishes to modify the data and needs a private copy of the
951 * data to alter. Returns %NULL on failure or the pointer to the buffer
952 * on success. The returned buffer has a reference count of 1.
954 * As by-product this function converts non-linear &sk_buff to linear
955 * one, so that &sk_buff becomes completely private and caller is allowed
956 * to modify all the data of returned buffer. This means that this
957 * function is not recommended for use in circumstances when only
958 * header is going to be modified. Use pskb_copy() instead.
961 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
963 int headerlen = skb_headroom(skb);
964 unsigned int size = skb_end_offset(skb) + skb->data_len;
965 struct sk_buff *n = __alloc_skb(size, gfp_mask,
966 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
971 /* Set the data pointer */
972 skb_reserve(n, headerlen);
973 /* Set the tail pointer and length */
974 skb_put(n, skb->len);
976 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
979 copy_skb_header(n, skb);
982 EXPORT_SYMBOL(skb_copy);
985 * __pskb_copy - create copy of an sk_buff with private head.
986 * @skb: buffer to copy
987 * @headroom: headroom of new skb
988 * @gfp_mask: allocation priority
990 * Make a copy of both an &sk_buff and part of its data, located
991 * in header. Fragmented data remain shared. This is used when
992 * the caller wishes to modify only header of &sk_buff and needs
993 * private copy of the header to alter. Returns %NULL on failure
994 * or the pointer to the buffer on success.
995 * The returned buffer has a reference count of 1.
998 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
1000 unsigned int size = skb_headlen(skb) + headroom;
1001 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1002 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1007 /* Set the data pointer */
1008 skb_reserve(n, headroom);
1009 /* Set the tail pointer and length */
1010 skb_put(n, skb_headlen(skb));
1011 /* Copy the bytes */
1012 skb_copy_from_linear_data(skb, n->data, n->len);
1014 n->truesize += skb->data_len;
1015 n->data_len = skb->data_len;
1018 if (skb_shinfo(skb)->nr_frags) {
1021 if (skb_orphan_frags(skb, gfp_mask)) {
1026 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1027 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1028 skb_frag_ref(skb, i);
1030 skb_shinfo(n)->nr_frags = i;
1033 if (skb_has_frag_list(skb)) {
1034 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1035 skb_clone_fraglist(n);
1038 copy_skb_header(n, skb);
1042 EXPORT_SYMBOL(__pskb_copy);
1045 * pskb_expand_head - reallocate header of &sk_buff
1046 * @skb: buffer to reallocate
1047 * @nhead: room to add at head
1048 * @ntail: room to add at tail
1049 * @gfp_mask: allocation priority
1051 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1052 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1053 * reference count of 1. Returns zero in the case of success or error,
1054 * if expansion failed. In the last case, &sk_buff is not changed.
1056 * All the pointers pointing into skb header may change and must be
1057 * reloaded after call to this function.
1060 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1065 int size = nhead + skb_end_offset(skb) + ntail;
1070 if (skb_shared(skb))
1073 size = SKB_DATA_ALIGN(size);
1075 if (skb_pfmemalloc(skb))
1076 gfp_mask |= __GFP_MEMALLOC;
1077 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1078 gfp_mask, NUMA_NO_NODE, NULL);
1081 size = SKB_WITH_OVERHEAD(ksize(data));
1083 /* Copy only real data... and, alas, header. This should be
1084 * optimized for the cases when header is void.
1086 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1088 memcpy((struct skb_shared_info *)(data + size),
1090 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1093 * if shinfo is shared we must drop the old head gracefully, but if it
1094 * is not we can just drop the old head and let the existing refcount
1095 * be since all we did is relocate the values
1097 if (skb_cloned(skb)) {
1098 /* copy this zero copy skb frags */
1099 if (skb_orphan_frags(skb, gfp_mask))
1101 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1102 skb_frag_ref(skb, i);
1104 if (skb_has_frag_list(skb))
1105 skb_clone_fraglist(skb);
1107 skb_release_data(skb);
1111 off = (data + nhead) - skb->head;
1116 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1120 skb->end = skb->head + size;
1123 skb_headers_offset_update(skb, nhead);
1127 atomic_set(&skb_shinfo(skb)->dataref, 1);
1135 EXPORT_SYMBOL(pskb_expand_head);
1137 /* Make private copy of skb with writable head and some headroom */
1139 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1141 struct sk_buff *skb2;
1142 int delta = headroom - skb_headroom(skb);
1145 skb2 = pskb_copy(skb, GFP_ATOMIC);
1147 skb2 = skb_clone(skb, GFP_ATOMIC);
1148 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1156 EXPORT_SYMBOL(skb_realloc_headroom);
1159 * skb_copy_expand - copy and expand sk_buff
1160 * @skb: buffer to copy
1161 * @newheadroom: new free bytes at head
1162 * @newtailroom: new free bytes at tail
1163 * @gfp_mask: allocation priority
1165 * Make a copy of both an &sk_buff and its data and while doing so
1166 * allocate additional space.
1168 * This is used when the caller wishes to modify the data and needs a
1169 * private copy of the data to alter as well as more space for new fields.
1170 * Returns %NULL on failure or the pointer to the buffer
1171 * on success. The returned buffer has a reference count of 1.
1173 * You must pass %GFP_ATOMIC as the allocation priority if this function
1174 * is called from an interrupt.
1176 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1177 int newheadroom, int newtailroom,
1181 * Allocate the copy buffer
1183 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1184 gfp_mask, skb_alloc_rx_flag(skb),
1186 int oldheadroom = skb_headroom(skb);
1187 int head_copy_len, head_copy_off;
1192 skb_reserve(n, newheadroom);
1194 /* Set the tail pointer and length */
1195 skb_put(n, skb->len);
1197 head_copy_len = oldheadroom;
1199 if (newheadroom <= head_copy_len)
1200 head_copy_len = newheadroom;
1202 head_copy_off = newheadroom - head_copy_len;
1204 /* Copy the linear header and data. */
1205 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1206 skb->len + head_copy_len))
1209 copy_skb_header(n, skb);
1211 skb_headers_offset_update(n, newheadroom - oldheadroom);
1215 EXPORT_SYMBOL(skb_copy_expand);
1218 * skb_pad - zero pad the tail of an skb
1219 * @skb: buffer to pad
1220 * @pad: space to pad
1222 * Ensure that a buffer is followed by a padding area that is zero
1223 * filled. Used by network drivers which may DMA or transfer data
1224 * beyond the buffer end onto the wire.
1226 * May return error in out of memory cases. The skb is freed on error.
1229 int skb_pad(struct sk_buff *skb, int pad)
1234 /* If the skbuff is non linear tailroom is always zero.. */
1235 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1236 memset(skb->data+skb->len, 0, pad);
1240 ntail = skb->data_len + pad - (skb->end - skb->tail);
1241 if (likely(skb_cloned(skb) || ntail > 0)) {
1242 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1247 /* FIXME: The use of this function with non-linear skb's really needs
1250 err = skb_linearize(skb);
1254 memset(skb->data + skb->len, 0, pad);
1261 EXPORT_SYMBOL(skb_pad);
1264 * pskb_put - add data to the tail of a potentially fragmented buffer
1265 * @skb: start of the buffer to use
1266 * @tail: tail fragment of the buffer to use
1267 * @len: amount of data to add
1269 * This function extends the used data area of the potentially
1270 * fragmented buffer. @tail must be the last fragment of @skb -- or
1271 * @skb itself. If this would exceed the total buffer size the kernel
1272 * will panic. A pointer to the first byte of the extra data is
1276 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1279 skb->data_len += len;
1282 return skb_put(tail, len);
1284 EXPORT_SYMBOL_GPL(pskb_put);
1287 * skb_put - add data to a buffer
1288 * @skb: buffer to use
1289 * @len: amount of data to add
1291 * This function extends the used data area of the buffer. If this would
1292 * exceed the total buffer size the kernel will panic. A pointer to the
1293 * first byte of the extra data is returned.
1295 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1297 unsigned char *tmp = skb_tail_pointer(skb);
1298 SKB_LINEAR_ASSERT(skb);
1301 if (unlikely(skb->tail > skb->end))
1302 skb_over_panic(skb, len, __builtin_return_address(0));
1305 EXPORT_SYMBOL(skb_put);
1308 * skb_push - add data to the start of a buffer
1309 * @skb: buffer to use
1310 * @len: amount of data to add
1312 * This function extends the used data area of the buffer at the buffer
1313 * start. If this would exceed the total buffer headroom the kernel will
1314 * panic. A pointer to the first byte of the extra data is returned.
1316 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1320 if (unlikely(skb->data<skb->head))
1321 skb_under_panic(skb, len, __builtin_return_address(0));
1324 EXPORT_SYMBOL(skb_push);
1327 * skb_pull - remove data from the start of a buffer
1328 * @skb: buffer to use
1329 * @len: amount of data to remove
1331 * This function removes data from the start of a buffer, returning
1332 * the memory to the headroom. A pointer to the next data in the buffer
1333 * is returned. Once the data has been pulled future pushes will overwrite
1336 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1338 return skb_pull_inline(skb, len);
1340 EXPORT_SYMBOL(skb_pull);
1343 * skb_trim - remove end from a buffer
1344 * @skb: buffer to alter
1347 * Cut the length of a buffer down by removing data from the tail. If
1348 * the buffer is already under the length specified it is not modified.
1349 * The skb must be linear.
1351 void skb_trim(struct sk_buff *skb, unsigned int len)
1354 __skb_trim(skb, len);
1356 EXPORT_SYMBOL(skb_trim);
1358 /* Trims skb to length len. It can change skb pointers.
1361 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1363 struct sk_buff **fragp;
1364 struct sk_buff *frag;
1365 int offset = skb_headlen(skb);
1366 int nfrags = skb_shinfo(skb)->nr_frags;
1370 if (skb_cloned(skb) &&
1371 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1378 for (; i < nfrags; i++) {
1379 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1386 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1389 skb_shinfo(skb)->nr_frags = i;
1391 for (; i < nfrags; i++)
1392 skb_frag_unref(skb, i);
1394 if (skb_has_frag_list(skb))
1395 skb_drop_fraglist(skb);
1399 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1400 fragp = &frag->next) {
1401 int end = offset + frag->len;
1403 if (skb_shared(frag)) {
1404 struct sk_buff *nfrag;
1406 nfrag = skb_clone(frag, GFP_ATOMIC);
1407 if (unlikely(!nfrag))
1410 nfrag->next = frag->next;
1422 unlikely((err = pskb_trim(frag, len - offset))))
1426 skb_drop_list(&frag->next);
1431 if (len > skb_headlen(skb)) {
1432 skb->data_len -= skb->len - len;
1437 skb_set_tail_pointer(skb, len);
1442 EXPORT_SYMBOL(___pskb_trim);
1445 * __pskb_pull_tail - advance tail of skb header
1446 * @skb: buffer to reallocate
1447 * @delta: number of bytes to advance tail
1449 * The function makes a sense only on a fragmented &sk_buff,
1450 * it expands header moving its tail forward and copying necessary
1451 * data from fragmented part.
1453 * &sk_buff MUST have reference count of 1.
1455 * Returns %NULL (and &sk_buff does not change) if pull failed
1456 * or value of new tail of skb in the case of success.
1458 * All the pointers pointing into skb header may change and must be
1459 * reloaded after call to this function.
1462 /* Moves tail of skb head forward, copying data from fragmented part,
1463 * when it is necessary.
1464 * 1. It may fail due to malloc failure.
1465 * 2. It may change skb pointers.
1467 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1469 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1471 /* If skb has not enough free space at tail, get new one
1472 * plus 128 bytes for future expansions. If we have enough
1473 * room at tail, reallocate without expansion only if skb is cloned.
1475 int i, k, eat = (skb->tail + delta) - skb->end;
1477 if (eat > 0 || skb_cloned(skb)) {
1478 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1483 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1486 /* Optimization: no fragments, no reasons to preestimate
1487 * size of pulled pages. Superb.
1489 if (!skb_has_frag_list(skb))
1492 /* Estimate size of pulled pages. */
1494 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1495 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1502 /* If we need update frag list, we are in troubles.
1503 * Certainly, it possible to add an offset to skb data,
1504 * but taking into account that pulling is expected to
1505 * be very rare operation, it is worth to fight against
1506 * further bloating skb head and crucify ourselves here instead.
1507 * Pure masohism, indeed. 8)8)
1510 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1511 struct sk_buff *clone = NULL;
1512 struct sk_buff *insp = NULL;
1517 if (list->len <= eat) {
1518 /* Eaten as whole. */
1523 /* Eaten partially. */
1525 if (skb_shared(list)) {
1526 /* Sucks! We need to fork list. :-( */
1527 clone = skb_clone(list, GFP_ATOMIC);
1533 /* This may be pulled without
1537 if (!pskb_pull(list, eat)) {
1545 /* Free pulled out fragments. */
1546 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1547 skb_shinfo(skb)->frag_list = list->next;
1550 /* And insert new clone at head. */
1553 skb_shinfo(skb)->frag_list = clone;
1556 /* Success! Now we may commit changes to skb data. */
1561 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1562 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1565 skb_frag_unref(skb, i);
1568 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1570 skb_shinfo(skb)->frags[k].page_offset += eat;
1571 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1577 skb_shinfo(skb)->nr_frags = k;
1580 skb->data_len -= delta;
1582 return skb_tail_pointer(skb);
1584 EXPORT_SYMBOL(__pskb_pull_tail);
1587 * skb_copy_bits - copy bits from skb to kernel buffer
1589 * @offset: offset in source
1590 * @to: destination buffer
1591 * @len: number of bytes to copy
1593 * Copy the specified number of bytes from the source skb to the
1594 * destination buffer.
1597 * If its prototype is ever changed,
1598 * check arch/{*}/net/{*}.S files,
1599 * since it is called from BPF assembly code.
1601 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1603 int start = skb_headlen(skb);
1604 struct sk_buff *frag_iter;
1607 if (offset > (int)skb->len - len)
1611 if ((copy = start - offset) > 0) {
1614 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1615 if ((len -= copy) == 0)
1621 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1623 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1625 WARN_ON(start > offset + len);
1627 end = start + skb_frag_size(f);
1628 if ((copy = end - offset) > 0) {
1634 vaddr = kmap_atomic(skb_frag_page(f));
1636 vaddr + f->page_offset + offset - start,
1638 kunmap_atomic(vaddr);
1640 if ((len -= copy) == 0)
1648 skb_walk_frags(skb, frag_iter) {
1651 WARN_ON(start > offset + len);
1653 end = start + frag_iter->len;
1654 if ((copy = end - offset) > 0) {
1657 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1659 if ((len -= copy) == 0)
1673 EXPORT_SYMBOL(skb_copy_bits);
1676 * Callback from splice_to_pipe(), if we need to release some pages
1677 * at the end of the spd in case we error'ed out in filling the pipe.
1679 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1681 put_page(spd->pages[i]);
1684 static struct page *linear_to_page(struct page *page, unsigned int *len,
1685 unsigned int *offset,
1688 struct page_frag *pfrag = sk_page_frag(sk);
1690 if (!sk_page_frag_refill(sk, pfrag))
1693 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1695 memcpy(page_address(pfrag->page) + pfrag->offset,
1696 page_address(page) + *offset, *len);
1697 *offset = pfrag->offset;
1698 pfrag->offset += *len;
1703 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1705 unsigned int offset)
1707 return spd->nr_pages &&
1708 spd->pages[spd->nr_pages - 1] == page &&
1709 (spd->partial[spd->nr_pages - 1].offset +
1710 spd->partial[spd->nr_pages - 1].len == offset);
1714 * Fill page/offset/length into spd, if it can hold more pages.
1716 static bool spd_fill_page(struct splice_pipe_desc *spd,
1717 struct pipe_inode_info *pipe, struct page *page,
1718 unsigned int *len, unsigned int offset,
1722 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1726 page = linear_to_page(page, len, &offset, sk);
1730 if (spd_can_coalesce(spd, page, offset)) {
1731 spd->partial[spd->nr_pages - 1].len += *len;
1735 spd->pages[spd->nr_pages] = page;
1736 spd->partial[spd->nr_pages].len = *len;
1737 spd->partial[spd->nr_pages].offset = offset;
1743 static bool __splice_segment(struct page *page, unsigned int poff,
1744 unsigned int plen, unsigned int *off,
1746 struct splice_pipe_desc *spd, bool linear,
1748 struct pipe_inode_info *pipe)
1753 /* skip this segment if already processed */
1759 /* ignore any bits we already processed */
1765 unsigned int flen = min(*len, plen);
1767 if (spd_fill_page(spd, pipe, page, &flen, poff,
1773 } while (*len && plen);
1779 * Map linear and fragment data from the skb to spd. It reports true if the
1780 * pipe is full or if we already spliced the requested length.
1782 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1783 unsigned int *offset, unsigned int *len,
1784 struct splice_pipe_desc *spd, struct sock *sk)
1788 /* map the linear part :
1789 * If skb->head_frag is set, this 'linear' part is backed by a
1790 * fragment, and if the head is not shared with any clones then
1791 * we can avoid a copy since we own the head portion of this page.
1793 if (__splice_segment(virt_to_page(skb->data),
1794 (unsigned long) skb->data & (PAGE_SIZE - 1),
1797 skb_head_is_locked(skb),
1802 * then map the fragments
1804 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1805 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1807 if (__splice_segment(skb_frag_page(f),
1808 f->page_offset, skb_frag_size(f),
1809 offset, len, spd, false, sk, pipe))
1817 * Map data from the skb to a pipe. Should handle both the linear part,
1818 * the fragments, and the frag list. It does NOT handle frag lists within
1819 * the frag list, if such a thing exists. We'd probably need to recurse to
1820 * handle that cleanly.
1822 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1823 struct pipe_inode_info *pipe, unsigned int tlen,
1826 struct partial_page partial[MAX_SKB_FRAGS];
1827 struct page *pages[MAX_SKB_FRAGS];
1828 struct splice_pipe_desc spd = {
1831 .nr_pages_max = MAX_SKB_FRAGS,
1833 .ops = &sock_pipe_buf_ops,
1834 .spd_release = sock_spd_release,
1836 struct sk_buff *frag_iter;
1837 struct sock *sk = skb->sk;
1841 * __skb_splice_bits() only fails if the output has no room left,
1842 * so no point in going over the frag_list for the error case.
1844 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1850 * now see if we have a frag_list to map
1852 skb_walk_frags(skb, frag_iter) {
1855 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1862 * Drop the socket lock, otherwise we have reverse
1863 * locking dependencies between sk_lock and i_mutex
1864 * here as compared to sendfile(). We enter here
1865 * with the socket lock held, and splice_to_pipe() will
1866 * grab the pipe inode lock. For sendfile() emulation,
1867 * we call into ->sendpage() with the i_mutex lock held
1868 * and networking will grab the socket lock.
1871 ret = splice_to_pipe(pipe, &spd);
1879 * skb_store_bits - store bits from kernel buffer to skb
1880 * @skb: destination buffer
1881 * @offset: offset in destination
1882 * @from: source buffer
1883 * @len: number of bytes to copy
1885 * Copy the specified number of bytes from the source buffer to the
1886 * destination skb. This function handles all the messy bits of
1887 * traversing fragment lists and such.
1890 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1892 int start = skb_headlen(skb);
1893 struct sk_buff *frag_iter;
1896 if (offset > (int)skb->len - len)
1899 if ((copy = start - offset) > 0) {
1902 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1903 if ((len -= copy) == 0)
1909 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1910 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1913 WARN_ON(start > offset + len);
1915 end = start + skb_frag_size(frag);
1916 if ((copy = end - offset) > 0) {
1922 vaddr = kmap_atomic(skb_frag_page(frag));
1923 memcpy(vaddr + frag->page_offset + offset - start,
1925 kunmap_atomic(vaddr);
1927 if ((len -= copy) == 0)
1935 skb_walk_frags(skb, frag_iter) {
1938 WARN_ON(start > offset + len);
1940 end = start + frag_iter->len;
1941 if ((copy = end - offset) > 0) {
1944 if (skb_store_bits(frag_iter, offset - start,
1947 if ((len -= copy) == 0)
1960 EXPORT_SYMBOL(skb_store_bits);
1962 /* Checksum skb data. */
1963 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
1964 __wsum csum, const struct skb_checksum_ops *ops)
1966 int start = skb_headlen(skb);
1967 int i, copy = start - offset;
1968 struct sk_buff *frag_iter;
1971 /* Checksum header. */
1975 csum = ops->update(skb->data + offset, copy, csum);
1976 if ((len -= copy) == 0)
1982 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1984 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1986 WARN_ON(start > offset + len);
1988 end = start + skb_frag_size(frag);
1989 if ((copy = end - offset) > 0) {
1995 vaddr = kmap_atomic(skb_frag_page(frag));
1996 csum2 = ops->update(vaddr + frag->page_offset +
1997 offset - start, copy, 0);
1998 kunmap_atomic(vaddr);
1999 csum = ops->combine(csum, csum2, pos, copy);
2008 skb_walk_frags(skb, frag_iter) {
2011 WARN_ON(start > offset + len);
2013 end = start + frag_iter->len;
2014 if ((copy = end - offset) > 0) {
2018 csum2 = __skb_checksum(frag_iter, offset - start,
2020 csum = ops->combine(csum, csum2, pos, copy);
2021 if ((len -= copy) == 0)
2032 EXPORT_SYMBOL(__skb_checksum);
2034 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2035 int len, __wsum csum)
2037 const struct skb_checksum_ops ops = {
2038 .update = csum_partial_ext,
2039 .combine = csum_block_add_ext,
2042 return __skb_checksum(skb, offset, len, csum, &ops);
2044 EXPORT_SYMBOL(skb_checksum);
2046 /* Both of above in one bottle. */
2048 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2049 u8 *to, int len, __wsum csum)
2051 int start = skb_headlen(skb);
2052 int i, copy = start - offset;
2053 struct sk_buff *frag_iter;
2060 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2062 if ((len -= copy) == 0)
2069 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2072 WARN_ON(start > offset + len);
2074 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2075 if ((copy = end - offset) > 0) {
2078 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2082 vaddr = kmap_atomic(skb_frag_page(frag));
2083 csum2 = csum_partial_copy_nocheck(vaddr +
2087 kunmap_atomic(vaddr);
2088 csum = csum_block_add(csum, csum2, pos);
2098 skb_walk_frags(skb, frag_iter) {
2102 WARN_ON(start > offset + len);
2104 end = start + frag_iter->len;
2105 if ((copy = end - offset) > 0) {
2108 csum2 = skb_copy_and_csum_bits(frag_iter,
2111 csum = csum_block_add(csum, csum2, pos);
2112 if ((len -= copy) == 0)
2123 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2126 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2127 * @from: source buffer
2129 * Calculates the amount of linear headroom needed in the 'to' skb passed
2130 * into skb_zerocopy().
2133 skb_zerocopy_headlen(const struct sk_buff *from)
2135 unsigned int hlen = 0;
2137 if (!from->head_frag ||
2138 skb_headlen(from) < L1_CACHE_BYTES ||
2139 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2140 hlen = skb_headlen(from);
2142 if (skb_has_frag_list(from))
2147 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2150 * skb_zerocopy - Zero copy skb to skb
2151 * @to: destination buffer
2152 * @source: source buffer
2153 * @len: number of bytes to copy from source buffer
2154 * @hlen: size of linear headroom in destination buffer
2156 * Copies up to `len` bytes from `from` to `to` by creating references
2157 * to the frags in the source buffer.
2159 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2160 * headroom in the `to` buffer.
2163 skb_zerocopy(struct sk_buff *to, const struct sk_buff *from, int len, int hlen)
2166 int plen = 0; /* length of skb->head fragment */
2168 unsigned int offset;
2170 BUG_ON(!from->head_frag && !hlen);
2172 /* dont bother with small payloads */
2173 if (len <= skb_tailroom(to)) {
2174 skb_copy_bits(from, 0, skb_put(to, len), len);
2179 skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2182 plen = min_t(int, skb_headlen(from), len);
2184 page = virt_to_head_page(from->head);
2185 offset = from->data - (unsigned char *)page_address(page);
2186 __skb_fill_page_desc(to, 0, page, offset, plen);
2193 to->truesize += len + plen;
2194 to->len += len + plen;
2195 to->data_len += len + plen;
2197 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2200 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2201 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2202 len -= skb_shinfo(to)->frags[j].size;
2203 skb_frag_ref(to, j);
2206 skb_shinfo(to)->nr_frags = j;
2208 EXPORT_SYMBOL_GPL(skb_zerocopy);
2210 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2215 if (skb->ip_summed == CHECKSUM_PARTIAL)
2216 csstart = skb_checksum_start_offset(skb);
2218 csstart = skb_headlen(skb);
2220 BUG_ON(csstart > skb_headlen(skb));
2222 skb_copy_from_linear_data(skb, to, csstart);
2225 if (csstart != skb->len)
2226 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2227 skb->len - csstart, 0);
2229 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2230 long csstuff = csstart + skb->csum_offset;
2232 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2235 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2238 * skb_dequeue - remove from the head of the queue
2239 * @list: list to dequeue from
2241 * Remove the head of the list. The list lock is taken so the function
2242 * may be used safely with other locking list functions. The head item is
2243 * returned or %NULL if the list is empty.
2246 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2248 unsigned long flags;
2249 struct sk_buff *result;
2251 spin_lock_irqsave(&list->lock, flags);
2252 result = __skb_dequeue(list);
2253 spin_unlock_irqrestore(&list->lock, flags);
2256 EXPORT_SYMBOL(skb_dequeue);
2259 * skb_dequeue_tail - remove from the tail of the queue
2260 * @list: list to dequeue from
2262 * Remove the tail of the list. The list lock is taken so the function
2263 * may be used safely with other locking list functions. The tail item is
2264 * returned or %NULL if the list is empty.
2266 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2268 unsigned long flags;
2269 struct sk_buff *result;
2271 spin_lock_irqsave(&list->lock, flags);
2272 result = __skb_dequeue_tail(list);
2273 spin_unlock_irqrestore(&list->lock, flags);
2276 EXPORT_SYMBOL(skb_dequeue_tail);
2279 * skb_queue_purge - empty a list
2280 * @list: list to empty
2282 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2283 * the list and one reference dropped. This function takes the list
2284 * lock and is atomic with respect to other list locking functions.
2286 void skb_queue_purge(struct sk_buff_head *list)
2288 struct sk_buff *skb;
2289 while ((skb = skb_dequeue(list)) != NULL)
2292 EXPORT_SYMBOL(skb_queue_purge);
2295 * skb_queue_head - queue a buffer at the list head
2296 * @list: list to use
2297 * @newsk: buffer to queue
2299 * Queue a buffer at the start of the list. This function takes the
2300 * list lock and can be used safely with other locking &sk_buff functions
2303 * A buffer cannot be placed on two lists at the same time.
2305 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2307 unsigned long flags;
2309 spin_lock_irqsave(&list->lock, flags);
2310 __skb_queue_head(list, newsk);
2311 spin_unlock_irqrestore(&list->lock, flags);
2313 EXPORT_SYMBOL(skb_queue_head);
2316 * skb_queue_tail - queue a buffer at the list tail
2317 * @list: list to use
2318 * @newsk: buffer to queue
2320 * Queue a buffer at the tail of the list. This function takes the
2321 * list lock and can be used safely with other locking &sk_buff functions
2324 * A buffer cannot be placed on two lists at the same time.
2326 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2328 unsigned long flags;
2330 spin_lock_irqsave(&list->lock, flags);
2331 __skb_queue_tail(list, newsk);
2332 spin_unlock_irqrestore(&list->lock, flags);
2334 EXPORT_SYMBOL(skb_queue_tail);
2337 * skb_unlink - remove a buffer from a list
2338 * @skb: buffer to remove
2339 * @list: list to use
2341 * Remove a packet from a list. The list locks are taken and this
2342 * function is atomic with respect to other list locked calls
2344 * You must know what list the SKB is on.
2346 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2348 unsigned long flags;
2350 spin_lock_irqsave(&list->lock, flags);
2351 __skb_unlink(skb, list);
2352 spin_unlock_irqrestore(&list->lock, flags);
2354 EXPORT_SYMBOL(skb_unlink);
2357 * skb_append - append a buffer
2358 * @old: buffer to insert after
2359 * @newsk: buffer to insert
2360 * @list: list to use
2362 * Place a packet after a given packet in a list. The list locks are taken
2363 * and this function is atomic with respect to other list locked calls.
2364 * A buffer cannot be placed on two lists at the same time.
2366 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2368 unsigned long flags;
2370 spin_lock_irqsave(&list->lock, flags);
2371 __skb_queue_after(list, old, newsk);
2372 spin_unlock_irqrestore(&list->lock, flags);
2374 EXPORT_SYMBOL(skb_append);
2377 * skb_insert - insert a buffer
2378 * @old: buffer to insert before
2379 * @newsk: buffer to insert
2380 * @list: list to use
2382 * Place a packet before a given packet in a list. The list locks are
2383 * taken and this function is atomic with respect to other list locked
2386 * A buffer cannot be placed on two lists at the same time.
2388 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2390 unsigned long flags;
2392 spin_lock_irqsave(&list->lock, flags);
2393 __skb_insert(newsk, old->prev, old, list);
2394 spin_unlock_irqrestore(&list->lock, flags);
2396 EXPORT_SYMBOL(skb_insert);
2398 static inline void skb_split_inside_header(struct sk_buff *skb,
2399 struct sk_buff* skb1,
2400 const u32 len, const int pos)
2404 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2406 /* And move data appendix as is. */
2407 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2408 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2410 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2411 skb_shinfo(skb)->nr_frags = 0;
2412 skb1->data_len = skb->data_len;
2413 skb1->len += skb1->data_len;
2416 skb_set_tail_pointer(skb, len);
2419 static inline void skb_split_no_header(struct sk_buff *skb,
2420 struct sk_buff* skb1,
2421 const u32 len, int pos)
2424 const int nfrags = skb_shinfo(skb)->nr_frags;
2426 skb_shinfo(skb)->nr_frags = 0;
2427 skb1->len = skb1->data_len = skb->len - len;
2429 skb->data_len = len - pos;
2431 for (i = 0; i < nfrags; i++) {
2432 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2434 if (pos + size > len) {
2435 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2439 * We have two variants in this case:
2440 * 1. Move all the frag to the second
2441 * part, if it is possible. F.e.
2442 * this approach is mandatory for TUX,
2443 * where splitting is expensive.
2444 * 2. Split is accurately. We make this.
2446 skb_frag_ref(skb, i);
2447 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2448 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2449 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2450 skb_shinfo(skb)->nr_frags++;
2454 skb_shinfo(skb)->nr_frags++;
2457 skb_shinfo(skb1)->nr_frags = k;
2461 * skb_split - Split fragmented skb to two parts at length len.
2462 * @skb: the buffer to split
2463 * @skb1: the buffer to receive the second part
2464 * @len: new length for skb
2466 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2468 int pos = skb_headlen(skb);
2470 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2471 if (len < pos) /* Split line is inside header. */
2472 skb_split_inside_header(skb, skb1, len, pos);
2473 else /* Second chunk has no header, nothing to copy. */
2474 skb_split_no_header(skb, skb1, len, pos);
2476 EXPORT_SYMBOL(skb_split);
2478 /* Shifting from/to a cloned skb is a no-go.
2480 * Caller cannot keep skb_shinfo related pointers past calling here!
2482 static int skb_prepare_for_shift(struct sk_buff *skb)
2484 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2488 * skb_shift - Shifts paged data partially from skb to another
2489 * @tgt: buffer into which tail data gets added
2490 * @skb: buffer from which the paged data comes from
2491 * @shiftlen: shift up to this many bytes
2493 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2494 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2495 * It's up to caller to free skb if everything was shifted.
2497 * If @tgt runs out of frags, the whole operation is aborted.
2499 * Skb cannot include anything else but paged data while tgt is allowed
2500 * to have non-paged data as well.
2502 * TODO: full sized shift could be optimized but that would need
2503 * specialized skb free'er to handle frags without up-to-date nr_frags.
2505 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2507 int from, to, merge, todo;
2508 struct skb_frag_struct *fragfrom, *fragto;
2510 BUG_ON(shiftlen > skb->len);
2511 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2515 to = skb_shinfo(tgt)->nr_frags;
2516 fragfrom = &skb_shinfo(skb)->frags[from];
2518 /* Actual merge is delayed until the point when we know we can
2519 * commit all, so that we don't have to undo partial changes
2522 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2523 fragfrom->page_offset)) {
2528 todo -= skb_frag_size(fragfrom);
2530 if (skb_prepare_for_shift(skb) ||
2531 skb_prepare_for_shift(tgt))
2534 /* All previous frag pointers might be stale! */
2535 fragfrom = &skb_shinfo(skb)->frags[from];
2536 fragto = &skb_shinfo(tgt)->frags[merge];
2538 skb_frag_size_add(fragto, shiftlen);
2539 skb_frag_size_sub(fragfrom, shiftlen);
2540 fragfrom->page_offset += shiftlen;
2548 /* Skip full, not-fitting skb to avoid expensive operations */
2549 if ((shiftlen == skb->len) &&
2550 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2553 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2556 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2557 if (to == MAX_SKB_FRAGS)
2560 fragfrom = &skb_shinfo(skb)->frags[from];
2561 fragto = &skb_shinfo(tgt)->frags[to];
2563 if (todo >= skb_frag_size(fragfrom)) {
2564 *fragto = *fragfrom;
2565 todo -= skb_frag_size(fragfrom);
2570 __skb_frag_ref(fragfrom);
2571 fragto->page = fragfrom->page;
2572 fragto->page_offset = fragfrom->page_offset;
2573 skb_frag_size_set(fragto, todo);
2575 fragfrom->page_offset += todo;
2576 skb_frag_size_sub(fragfrom, todo);
2584 /* Ready to "commit" this state change to tgt */
2585 skb_shinfo(tgt)->nr_frags = to;
2588 fragfrom = &skb_shinfo(skb)->frags[0];
2589 fragto = &skb_shinfo(tgt)->frags[merge];
2591 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2592 __skb_frag_unref(fragfrom);
2595 /* Reposition in the original skb */
2597 while (from < skb_shinfo(skb)->nr_frags)
2598 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2599 skb_shinfo(skb)->nr_frags = to;
2601 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2604 /* Most likely the tgt won't ever need its checksum anymore, skb on
2605 * the other hand might need it if it needs to be resent
2607 tgt->ip_summed = CHECKSUM_PARTIAL;
2608 skb->ip_summed = CHECKSUM_PARTIAL;
2610 /* Yak, is it really working this way? Some helper please? */
2611 skb->len -= shiftlen;
2612 skb->data_len -= shiftlen;
2613 skb->truesize -= shiftlen;
2614 tgt->len += shiftlen;
2615 tgt->data_len += shiftlen;
2616 tgt->truesize += shiftlen;
2622 * skb_prepare_seq_read - Prepare a sequential read of skb data
2623 * @skb: the buffer to read
2624 * @from: lower offset of data to be read
2625 * @to: upper offset of data to be read
2626 * @st: state variable
2628 * Initializes the specified state variable. Must be called before
2629 * invoking skb_seq_read() for the first time.
2631 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2632 unsigned int to, struct skb_seq_state *st)
2634 st->lower_offset = from;
2635 st->upper_offset = to;
2636 st->root_skb = st->cur_skb = skb;
2637 st->frag_idx = st->stepped_offset = 0;
2638 st->frag_data = NULL;
2640 EXPORT_SYMBOL(skb_prepare_seq_read);
2643 * skb_seq_read - Sequentially read skb data
2644 * @consumed: number of bytes consumed by the caller so far
2645 * @data: destination pointer for data to be returned
2646 * @st: state variable
2648 * Reads a block of skb data at @consumed relative to the
2649 * lower offset specified to skb_prepare_seq_read(). Assigns
2650 * the head of the data block to @data and returns the length
2651 * of the block or 0 if the end of the skb data or the upper
2652 * offset has been reached.
2654 * The caller is not required to consume all of the data
2655 * returned, i.e. @consumed is typically set to the number
2656 * of bytes already consumed and the next call to
2657 * skb_seq_read() will return the remaining part of the block.
2659 * Note 1: The size of each block of data returned can be arbitrary,
2660 * this limitation is the cost for zerocopy seqeuental
2661 * reads of potentially non linear data.
2663 * Note 2: Fragment lists within fragments are not implemented
2664 * at the moment, state->root_skb could be replaced with
2665 * a stack for this purpose.
2667 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2668 struct skb_seq_state *st)
2670 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2673 if (unlikely(abs_offset >= st->upper_offset)) {
2674 if (st->frag_data) {
2675 kunmap_atomic(st->frag_data);
2676 st->frag_data = NULL;
2682 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2684 if (abs_offset < block_limit && !st->frag_data) {
2685 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2686 return block_limit - abs_offset;
2689 if (st->frag_idx == 0 && !st->frag_data)
2690 st->stepped_offset += skb_headlen(st->cur_skb);
2692 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2693 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2694 block_limit = skb_frag_size(frag) + st->stepped_offset;
2696 if (abs_offset < block_limit) {
2698 st->frag_data = kmap_atomic(skb_frag_page(frag));
2700 *data = (u8 *) st->frag_data + frag->page_offset +
2701 (abs_offset - st->stepped_offset);
2703 return block_limit - abs_offset;
2706 if (st->frag_data) {
2707 kunmap_atomic(st->frag_data);
2708 st->frag_data = NULL;
2712 st->stepped_offset += skb_frag_size(frag);
2715 if (st->frag_data) {
2716 kunmap_atomic(st->frag_data);
2717 st->frag_data = NULL;
2720 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2721 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2724 } else if (st->cur_skb->next) {
2725 st->cur_skb = st->cur_skb->next;
2732 EXPORT_SYMBOL(skb_seq_read);
2735 * skb_abort_seq_read - Abort a sequential read of skb data
2736 * @st: state variable
2738 * Must be called if skb_seq_read() was not called until it
2741 void skb_abort_seq_read(struct skb_seq_state *st)
2744 kunmap_atomic(st->frag_data);
2746 EXPORT_SYMBOL(skb_abort_seq_read);
2748 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2750 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2751 struct ts_config *conf,
2752 struct ts_state *state)
2754 return skb_seq_read(offset, text, TS_SKB_CB(state));
2757 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2759 skb_abort_seq_read(TS_SKB_CB(state));
2763 * skb_find_text - Find a text pattern in skb data
2764 * @skb: the buffer to look in
2765 * @from: search offset
2767 * @config: textsearch configuration
2768 * @state: uninitialized textsearch state variable
2770 * Finds a pattern in the skb data according to the specified
2771 * textsearch configuration. Use textsearch_next() to retrieve
2772 * subsequent occurrences of the pattern. Returns the offset
2773 * to the first occurrence or UINT_MAX if no match was found.
2775 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2776 unsigned int to, struct ts_config *config,
2777 struct ts_state *state)
2781 config->get_next_block = skb_ts_get_next_block;
2782 config->finish = skb_ts_finish;
2784 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2786 ret = textsearch_find(config, state);
2787 return (ret <= to - from ? ret : UINT_MAX);
2789 EXPORT_SYMBOL(skb_find_text);
2792 * skb_append_datato_frags - append the user data to a skb
2793 * @sk: sock structure
2794 * @skb: skb structure to be appened with user data.
2795 * @getfrag: call back function to be used for getting the user data
2796 * @from: pointer to user message iov
2797 * @length: length of the iov message
2799 * Description: This procedure append the user data in the fragment part
2800 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2802 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2803 int (*getfrag)(void *from, char *to, int offset,
2804 int len, int odd, struct sk_buff *skb),
2805 void *from, int length)
2807 int frg_cnt = skb_shinfo(skb)->nr_frags;
2811 struct page_frag *pfrag = ¤t->task_frag;
2814 /* Return error if we don't have space for new frag */
2815 if (frg_cnt >= MAX_SKB_FRAGS)
2818 if (!sk_page_frag_refill(sk, pfrag))
2821 /* copy the user data to page */
2822 copy = min_t(int, length, pfrag->size - pfrag->offset);
2824 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2825 offset, copy, 0, skb);
2829 /* copy was successful so update the size parameters */
2830 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2833 pfrag->offset += copy;
2834 get_page(pfrag->page);
2836 skb->truesize += copy;
2837 atomic_add(copy, &sk->sk_wmem_alloc);
2839 skb->data_len += copy;
2843 } while (length > 0);
2847 EXPORT_SYMBOL(skb_append_datato_frags);
2850 * skb_pull_rcsum - pull skb and update receive checksum
2851 * @skb: buffer to update
2852 * @len: length of data pulled
2854 * This function performs an skb_pull on the packet and updates
2855 * the CHECKSUM_COMPLETE checksum. It should be used on
2856 * receive path processing instead of skb_pull unless you know
2857 * that the checksum difference is zero (e.g., a valid IP header)
2858 * or you are setting ip_summed to CHECKSUM_NONE.
2860 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2862 BUG_ON(len > skb->len);
2864 BUG_ON(skb->len < skb->data_len);
2865 skb_postpull_rcsum(skb, skb->data, len);
2866 return skb->data += len;
2868 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2871 * skb_segment - Perform protocol segmentation on skb.
2872 * @skb: buffer to segment
2873 * @features: features for the output path (see dev->features)
2875 * This function performs segmentation on the given skb. It returns
2876 * a pointer to the first in a list of new skbs for the segments.
2877 * In case of error it returns ERR_PTR(err).
2879 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2881 struct sk_buff *segs = NULL;
2882 struct sk_buff *tail = NULL;
2883 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2884 skb_frag_t *skb_frag = skb_shinfo(skb)->frags;
2885 unsigned int mss = skb_shinfo(skb)->gso_size;
2886 unsigned int doffset = skb->data - skb_mac_header(skb);
2887 unsigned int offset = doffset;
2888 unsigned int tnl_hlen = skb_tnl_header_len(skb);
2889 unsigned int headroom;
2893 int sg = !!(features & NETIF_F_SG);
2894 int nfrags = skb_shinfo(skb)->nr_frags;
2899 proto = skb_network_protocol(skb);
2900 if (unlikely(!proto))
2901 return ERR_PTR(-EINVAL);
2903 csum = !!can_checksum_protocol(features, proto);
2904 __skb_push(skb, doffset);
2905 headroom = skb_headroom(skb);
2906 pos = skb_headlen(skb);
2909 struct sk_buff *nskb;
2914 len = skb->len - offset;
2918 hsize = skb_headlen(skb) - offset;
2921 if (hsize > len || !sg)
2924 if (!hsize && i >= nfrags && skb_headlen(fskb) &&
2925 (skb_headlen(fskb) == len || sg)) {
2926 BUG_ON(skb_headlen(fskb) > len);
2929 nfrags = skb_shinfo(fskb)->nr_frags;
2930 skb_frag = skb_shinfo(fskb)->frags;
2931 pos += skb_headlen(fskb);
2933 while (pos < offset + len) {
2934 BUG_ON(i >= nfrags);
2936 size = skb_frag_size(skb_frag);
2937 if (pos + size > offset + len)
2945 nskb = skb_clone(fskb, GFP_ATOMIC);
2948 if (unlikely(!nskb))
2951 if (unlikely(pskb_trim(nskb, len))) {
2956 hsize = skb_end_offset(nskb);
2957 if (skb_cow_head(nskb, doffset + headroom)) {
2962 nskb->truesize += skb_end_offset(nskb) - hsize;
2963 skb_release_head_state(nskb);
2964 __skb_push(nskb, doffset);
2966 nskb = __alloc_skb(hsize + doffset + headroom,
2967 GFP_ATOMIC, skb_alloc_rx_flag(skb),
2970 if (unlikely(!nskb))
2973 skb_reserve(nskb, headroom);
2974 __skb_put(nskb, doffset);
2983 __copy_skb_header(nskb, skb);
2984 nskb->mac_len = skb->mac_len;
2986 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
2988 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
2989 nskb->data - tnl_hlen,
2990 doffset + tnl_hlen);
2992 if (nskb->len == len + doffset)
2993 goto perform_csum_check;
2996 nskb->ip_summed = CHECKSUM_NONE;
2997 nskb->csum = skb_copy_and_csum_bits(skb, offset,
3003 frag = skb_shinfo(nskb)->frags;
3005 skb_copy_from_linear_data_offset(skb, offset,
3006 skb_put(nskb, hsize), hsize);
3008 skb_shinfo(nskb)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3010 while (pos < offset + len) {
3012 BUG_ON(skb_headlen(fskb));
3015 nfrags = skb_shinfo(fskb)->nr_frags;
3016 skb_frag = skb_shinfo(fskb)->frags;
3023 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3025 net_warn_ratelimited(
3026 "skb_segment: too many frags: %u %u\n",
3032 __skb_frag_ref(frag);
3033 size = skb_frag_size(frag);
3036 frag->page_offset += offset - pos;
3037 skb_frag_size_sub(frag, offset - pos);
3040 skb_shinfo(nskb)->nr_frags++;
3042 if (pos + size <= offset + len) {
3047 skb_frag_size_sub(frag, pos + size - (offset + len));
3055 nskb->data_len = len - hsize;
3056 nskb->len += nskb->data_len;
3057 nskb->truesize += nskb->data_len;
3061 nskb->csum = skb_checksum(nskb, doffset,
3062 nskb->len - doffset, 0);
3063 nskb->ip_summed = CHECKSUM_NONE;
3065 } while ((offset += len) < skb->len);
3070 kfree_skb_list(segs);
3071 return ERR_PTR(err);
3073 EXPORT_SYMBOL_GPL(skb_segment);
3075 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3077 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3078 unsigned int offset = skb_gro_offset(skb);
3079 unsigned int headlen = skb_headlen(skb);
3080 struct sk_buff *nskb, *lp, *p = *head;
3081 unsigned int len = skb_gro_len(skb);
3082 unsigned int delta_truesize;
3083 unsigned int headroom;
3085 if (unlikely(p->len + len >= 65536))
3088 lp = NAPI_GRO_CB(p)->last ?: p;
3089 pinfo = skb_shinfo(lp);
3091 if (headlen <= offset) {
3094 int i = skbinfo->nr_frags;
3095 int nr_frags = pinfo->nr_frags + i;
3097 if (nr_frags > MAX_SKB_FRAGS)
3101 pinfo->nr_frags = nr_frags;
3102 skbinfo->nr_frags = 0;
3104 frag = pinfo->frags + nr_frags;
3105 frag2 = skbinfo->frags + i;
3110 frag->page_offset += offset;
3111 skb_frag_size_sub(frag, offset);
3113 /* all fragments truesize : remove (head size + sk_buff) */
3114 delta_truesize = skb->truesize -
3115 SKB_TRUESIZE(skb_end_offset(skb));
3117 skb->truesize -= skb->data_len;
3118 skb->len -= skb->data_len;
3121 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3123 } else if (skb->head_frag) {
3124 int nr_frags = pinfo->nr_frags;
3125 skb_frag_t *frag = pinfo->frags + nr_frags;
3126 struct page *page = virt_to_head_page(skb->head);
3127 unsigned int first_size = headlen - offset;
3128 unsigned int first_offset;
3130 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3133 first_offset = skb->data -
3134 (unsigned char *)page_address(page) +
3137 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3139 frag->page.p = page;
3140 frag->page_offset = first_offset;
3141 skb_frag_size_set(frag, first_size);
3143 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3144 /* We dont need to clear skbinfo->nr_frags here */
3146 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3147 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3150 if (pinfo->frag_list)
3152 if (skb_gro_len(p) != pinfo->gso_size)
3155 headroom = skb_headroom(p);
3156 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
3157 if (unlikely(!nskb))
3160 __copy_skb_header(nskb, p);
3161 nskb->mac_len = p->mac_len;
3163 skb_reserve(nskb, headroom);
3164 __skb_put(nskb, skb_gro_offset(p));
3166 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3167 skb_set_network_header(nskb, skb_network_offset(p));
3168 skb_set_transport_header(nskb, skb_transport_offset(p));
3170 __skb_pull(p, skb_gro_offset(p));
3171 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3172 p->data - skb_mac_header(p));
3174 skb_shinfo(nskb)->frag_list = p;
3175 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3176 pinfo->gso_size = 0;
3177 skb_header_release(p);
3178 NAPI_GRO_CB(nskb)->last = p;
3180 nskb->data_len += p->len;
3181 nskb->truesize += p->truesize;
3182 nskb->len += p->len;
3185 nskb->next = p->next;
3191 delta_truesize = skb->truesize;
3192 if (offset > headlen) {
3193 unsigned int eat = offset - headlen;
3195 skbinfo->frags[0].page_offset += eat;
3196 skb_frag_size_sub(&skbinfo->frags[0], eat);
3197 skb->data_len -= eat;
3202 __skb_pull(skb, offset);
3204 if (!NAPI_GRO_CB(p)->last)
3205 skb_shinfo(p)->frag_list = skb;
3207 NAPI_GRO_CB(p)->last->next = skb;
3208 NAPI_GRO_CB(p)->last = skb;
3209 skb_header_release(skb);
3213 NAPI_GRO_CB(p)->count++;
3215 p->truesize += delta_truesize;
3218 lp->data_len += len;
3219 lp->truesize += delta_truesize;
3222 NAPI_GRO_CB(skb)->same_flow = 1;
3225 EXPORT_SYMBOL_GPL(skb_gro_receive);
3227 void __init skb_init(void)
3229 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3230 sizeof(struct sk_buff),
3232 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3234 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3235 (2*sizeof(struct sk_buff)) +
3238 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3243 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3244 * @skb: Socket buffer containing the buffers to be mapped
3245 * @sg: The scatter-gather list to map into
3246 * @offset: The offset into the buffer's contents to start mapping
3247 * @len: Length of buffer space to be mapped
3249 * Fill the specified scatter-gather list with mappings/pointers into a
3250 * region of the buffer space attached to a socket buffer.
3253 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3255 int start = skb_headlen(skb);
3256 int i, copy = start - offset;
3257 struct sk_buff *frag_iter;
3263 sg_set_buf(sg, skb->data + offset, copy);
3265 if ((len -= copy) == 0)
3270 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3273 WARN_ON(start > offset + len);
3275 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3276 if ((copy = end - offset) > 0) {
3277 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3281 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3282 frag->page_offset+offset-start);
3291 skb_walk_frags(skb, frag_iter) {
3294 WARN_ON(start > offset + len);
3296 end = start + frag_iter->len;
3297 if ((copy = end - offset) > 0) {
3300 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3302 if ((len -= copy) == 0)
3312 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3314 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3316 sg_mark_end(&sg[nsg - 1]);
3320 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3323 * skb_cow_data - Check that a socket buffer's data buffers are writable
3324 * @skb: The socket buffer to check.
3325 * @tailbits: Amount of trailing space to be added
3326 * @trailer: Returned pointer to the skb where the @tailbits space begins
3328 * Make sure that the data buffers attached to a socket buffer are
3329 * writable. If they are not, private copies are made of the data buffers
3330 * and the socket buffer is set to use these instead.
3332 * If @tailbits is given, make sure that there is space to write @tailbits
3333 * bytes of data beyond current end of socket buffer. @trailer will be
3334 * set to point to the skb in which this space begins.
3336 * The number of scatterlist elements required to completely map the
3337 * COW'd and extended socket buffer will be returned.
3339 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3343 struct sk_buff *skb1, **skb_p;
3345 /* If skb is cloned or its head is paged, reallocate
3346 * head pulling out all the pages (pages are considered not writable
3347 * at the moment even if they are anonymous).
3349 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3350 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3353 /* Easy case. Most of packets will go this way. */
3354 if (!skb_has_frag_list(skb)) {
3355 /* A little of trouble, not enough of space for trailer.
3356 * This should not happen, when stack is tuned to generate
3357 * good frames. OK, on miss we reallocate and reserve even more
3358 * space, 128 bytes is fair. */
3360 if (skb_tailroom(skb) < tailbits &&
3361 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3369 /* Misery. We are in troubles, going to mincer fragments... */
3372 skb_p = &skb_shinfo(skb)->frag_list;
3375 while ((skb1 = *skb_p) != NULL) {
3378 /* The fragment is partially pulled by someone,
3379 * this can happen on input. Copy it and everything
3382 if (skb_shared(skb1))
3385 /* If the skb is the last, worry about trailer. */
3387 if (skb1->next == NULL && tailbits) {
3388 if (skb_shinfo(skb1)->nr_frags ||
3389 skb_has_frag_list(skb1) ||
3390 skb_tailroom(skb1) < tailbits)
3391 ntail = tailbits + 128;
3397 skb_shinfo(skb1)->nr_frags ||
3398 skb_has_frag_list(skb1)) {
3399 struct sk_buff *skb2;
3401 /* Fuck, we are miserable poor guys... */
3403 skb2 = skb_copy(skb1, GFP_ATOMIC);
3405 skb2 = skb_copy_expand(skb1,
3409 if (unlikely(skb2 == NULL))
3413 skb_set_owner_w(skb2, skb1->sk);
3415 /* Looking around. Are we still alive?
3416 * OK, link new skb, drop old one */
3418 skb2->next = skb1->next;
3425 skb_p = &skb1->next;
3430 EXPORT_SYMBOL_GPL(skb_cow_data);
3432 static void sock_rmem_free(struct sk_buff *skb)
3434 struct sock *sk = skb->sk;
3436 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3440 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3442 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3446 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3447 (unsigned int)sk->sk_rcvbuf)
3452 skb->destructor = sock_rmem_free;
3453 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3455 /* before exiting rcu section, make sure dst is refcounted */
3458 skb_queue_tail(&sk->sk_error_queue, skb);
3459 if (!sock_flag(sk, SOCK_DEAD))
3460 sk->sk_data_ready(sk, len);
3463 EXPORT_SYMBOL(sock_queue_err_skb);
3465 void skb_tstamp_tx(struct sk_buff *orig_skb,
3466 struct skb_shared_hwtstamps *hwtstamps)
3468 struct sock *sk = orig_skb->sk;
3469 struct sock_exterr_skb *serr;
3470 struct sk_buff *skb;
3477 *skb_hwtstamps(orig_skb) =
3481 * no hardware time stamps available,
3482 * so keep the shared tx_flags and only
3483 * store software time stamp
3485 orig_skb->tstamp = ktime_get_real();
3488 skb = skb_clone(orig_skb, GFP_ATOMIC);
3492 serr = SKB_EXT_ERR(skb);
3493 memset(serr, 0, sizeof(*serr));
3494 serr->ee.ee_errno = ENOMSG;
3495 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3497 err = sock_queue_err_skb(sk, skb);
3502 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3504 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3506 struct sock *sk = skb->sk;
3507 struct sock_exterr_skb *serr;
3510 skb->wifi_acked_valid = 1;
3511 skb->wifi_acked = acked;
3513 serr = SKB_EXT_ERR(skb);
3514 memset(serr, 0, sizeof(*serr));
3515 serr->ee.ee_errno = ENOMSG;
3516 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3518 err = sock_queue_err_skb(sk, skb);
3522 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3526 * skb_partial_csum_set - set up and verify partial csum values for packet
3527 * @skb: the skb to set
3528 * @start: the number of bytes after skb->data to start checksumming.
3529 * @off: the offset from start to place the checksum.
3531 * For untrusted partially-checksummed packets, we need to make sure the values
3532 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3534 * This function checks and sets those values and skb->ip_summed: if this
3535 * returns false you should drop the packet.
3537 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3539 if (unlikely(start > skb_headlen(skb)) ||
3540 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3541 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3542 start, off, skb_headlen(skb));
3545 skb->ip_summed = CHECKSUM_PARTIAL;
3546 skb->csum_start = skb_headroom(skb) + start;
3547 skb->csum_offset = off;
3548 skb_set_transport_header(skb, start);
3551 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3553 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3556 if (skb_headlen(skb) >= len)
3559 /* If we need to pullup then pullup to the max, so we
3560 * won't need to do it again.
3565 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3568 if (skb_headlen(skb) < len)
3574 /* This value should be large enough to cover a tagged ethernet header plus
3575 * maximally sized IP and TCP or UDP headers.
3577 #define MAX_IP_HDR_LEN 128
3579 static int skb_checksum_setup_ip(struct sk_buff *skb, bool recalculate)
3587 err = skb_maybe_pull_tail(skb,
3588 sizeof(struct iphdr),
3593 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3596 off = ip_hdrlen(skb);
3603 switch (ip_hdr(skb)->protocol) {
3605 err = skb_maybe_pull_tail(skb,
3606 off + sizeof(struct tcphdr),
3611 if (!skb_partial_csum_set(skb, off,
3612 offsetof(struct tcphdr, check))) {
3618 tcp_hdr(skb)->check =
3619 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3625 err = skb_maybe_pull_tail(skb,
3626 off + sizeof(struct udphdr),
3631 if (!skb_partial_csum_set(skb, off,
3632 offsetof(struct udphdr, check))) {
3638 udp_hdr(skb)->check =
3639 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3654 /* This value should be large enough to cover a tagged ethernet header plus
3655 * an IPv6 header, all options, and a maximal TCP or UDP header.
3657 #define MAX_IPV6_HDR_LEN 256
3659 #define OPT_HDR(type, skb, off) \
3660 (type *)(skb_network_header(skb) + (off))
3662 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
3674 off = sizeof(struct ipv6hdr);
3676 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
3680 nexthdr = ipv6_hdr(skb)->nexthdr;
3682 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
3683 while (off <= len && !done) {
3685 case IPPROTO_DSTOPTS:
3686 case IPPROTO_HOPOPTS:
3687 case IPPROTO_ROUTING: {
3688 struct ipv6_opt_hdr *hp;
3690 err = skb_maybe_pull_tail(skb,
3692 sizeof(struct ipv6_opt_hdr),
3697 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
3698 nexthdr = hp->nexthdr;
3699 off += ipv6_optlen(hp);
3703 struct ip_auth_hdr *hp;
3705 err = skb_maybe_pull_tail(skb,
3707 sizeof(struct ip_auth_hdr),
3712 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
3713 nexthdr = hp->nexthdr;
3714 off += ipv6_authlen(hp);
3717 case IPPROTO_FRAGMENT: {
3718 struct frag_hdr *hp;
3720 err = skb_maybe_pull_tail(skb,
3722 sizeof(struct frag_hdr),
3727 hp = OPT_HDR(struct frag_hdr, skb, off);
3729 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
3732 nexthdr = hp->nexthdr;
3733 off += sizeof(struct frag_hdr);
3744 if (!done || fragment)
3749 err = skb_maybe_pull_tail(skb,
3750 off + sizeof(struct tcphdr),
3755 if (!skb_partial_csum_set(skb, off,
3756 offsetof(struct tcphdr, check))) {
3762 tcp_hdr(skb)->check =
3763 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3764 &ipv6_hdr(skb)->daddr,
3769 err = skb_maybe_pull_tail(skb,
3770 off + sizeof(struct udphdr),
3775 if (!skb_partial_csum_set(skb, off,
3776 offsetof(struct udphdr, check))) {
3782 udp_hdr(skb)->check =
3783 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3784 &ipv6_hdr(skb)->daddr,
3799 * skb_checksum_setup - set up partial checksum offset
3800 * @skb: the skb to set up
3801 * @recalculate: if true the pseudo-header checksum will be recalculated
3803 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
3807 switch (skb->protocol) {
3808 case htons(ETH_P_IP):
3809 err = skb_checksum_setup_ip(skb, recalculate);
3812 case htons(ETH_P_IPV6):
3813 err = skb_checksum_setup_ipv6(skb, recalculate);
3823 EXPORT_SYMBOL(skb_checksum_setup);
3825 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3827 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3830 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3832 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3835 skb_release_head_state(skb);
3836 kmem_cache_free(skbuff_head_cache, skb);
3841 EXPORT_SYMBOL(kfree_skb_partial);
3844 * skb_try_coalesce - try to merge skb to prior one
3846 * @from: buffer to add
3847 * @fragstolen: pointer to boolean
3848 * @delta_truesize: how much more was allocated than was requested
3850 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3851 bool *fragstolen, int *delta_truesize)
3853 int i, delta, len = from->len;
3855 *fragstolen = false;
3860 if (len <= skb_tailroom(to)) {
3861 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3862 *delta_truesize = 0;
3866 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3869 if (skb_headlen(from) != 0) {
3871 unsigned int offset;
3873 if (skb_shinfo(to)->nr_frags +
3874 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3877 if (skb_head_is_locked(from))
3880 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3882 page = virt_to_head_page(from->head);
3883 offset = from->data - (unsigned char *)page_address(page);
3885 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3886 page, offset, skb_headlen(from));
3889 if (skb_shinfo(to)->nr_frags +
3890 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3893 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
3896 WARN_ON_ONCE(delta < len);
3898 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3899 skb_shinfo(from)->frags,
3900 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3901 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3903 if (!skb_cloned(from))
3904 skb_shinfo(from)->nr_frags = 0;
3906 /* if the skb is not cloned this does nothing
3907 * since we set nr_frags to 0.
3909 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3910 skb_frag_ref(from, i);
3912 to->truesize += delta;
3914 to->data_len += len;
3916 *delta_truesize = delta;
3919 EXPORT_SYMBOL(skb_try_coalesce);
3922 * skb_scrub_packet - scrub an skb
3924 * @skb: buffer to clean
3925 * @xnet: packet is crossing netns
3927 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
3928 * into/from a tunnel. Some information have to be cleared during these
3930 * skb_scrub_packet can also be used to clean a skb before injecting it in
3931 * another namespace (@xnet == true). We have to clear all information in the
3932 * skb that could impact namespace isolation.
3934 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
3938 skb->tstamp.tv64 = 0;
3939 skb->pkt_type = PACKET_HOST;
3946 nf_reset_trace(skb);
3948 EXPORT_SYMBOL_GPL(skb_scrub_packet);