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/tcp.h>
51 #include <linux/udp.h>
52 #include <linux/netdevice.h>
53 #ifdef CONFIG_NET_CLS_ACT
54 #include <net/pkt_sched.h>
56 #include <linux/string.h>
57 #include <linux/skbuff.h>
58 #include <linux/splice.h>
59 #include <linux/cache.h>
60 #include <linux/rtnetlink.h>
61 #include <linux/init.h>
62 #include <linux/scatterlist.h>
63 #include <linux/errqueue.h>
64 #include <linux/prefetch.h>
65 #include <linux/if_vlan.h>
67 #include <net/protocol.h>
70 #include <net/checksum.h>
71 #include <net/ip6_checksum.h>
74 #include <asm/uaccess.h>
75 #include <trace/events/skb.h>
76 #include <linux/highmem.h>
77 #include <linux/capability.h>
78 #include <linux/user_namespace.h>
80 struct kmem_cache *skbuff_head_cache __read_mostly;
81 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
82 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
83 EXPORT_SYMBOL(sysctl_max_skb_frags);
86 * skb_panic - private function for out-of-line support
90 * @msg: skb_over_panic or skb_under_panic
92 * Out-of-line support for skb_put() and skb_push().
93 * Called via the wrapper skb_over_panic() or skb_under_panic().
94 * Keep out of line to prevent kernel bloat.
95 * __builtin_return_address is not used because it is not always reliable.
97 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
100 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
101 msg, addr, skb->len, sz, skb->head, skb->data,
102 (unsigned long)skb->tail, (unsigned long)skb->end,
103 skb->dev ? skb->dev->name : "<NULL>");
107 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
109 skb_panic(skb, sz, addr, __func__);
112 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
114 skb_panic(skb, sz, addr, __func__);
118 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
119 * the caller if emergency pfmemalloc reserves are being used. If it is and
120 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
121 * may be used. Otherwise, the packet data may be discarded until enough
124 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
125 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
127 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
128 unsigned long ip, bool *pfmemalloc)
131 bool ret_pfmemalloc = false;
134 * Try a regular allocation, when that fails and we're not entitled
135 * to the reserves, fail.
137 obj = kmalloc_node_track_caller(size,
138 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
140 if (obj || !(gfp_pfmemalloc_allowed(flags)))
143 /* Try again but now we are using pfmemalloc reserves */
144 ret_pfmemalloc = true;
145 obj = kmalloc_node_track_caller(size, flags, node);
149 *pfmemalloc = ret_pfmemalloc;
154 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
155 * 'private' fields and also do memory statistics to find all the
160 struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
165 skb = kmem_cache_alloc_node(skbuff_head_cache,
166 gfp_mask & ~__GFP_DMA, node);
171 * Only clear those fields we need to clear, not those that we will
172 * actually initialise below. Hence, don't put any more fields after
173 * the tail pointer in struct sk_buff!
175 memset(skb, 0, offsetof(struct sk_buff, tail));
177 skb->truesize = sizeof(struct sk_buff);
178 atomic_set(&skb->users, 1);
180 skb->mac_header = (typeof(skb->mac_header))~0U;
186 * __alloc_skb - allocate a network buffer
187 * @size: size to allocate
188 * @gfp_mask: allocation mask
189 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
190 * instead of head cache and allocate a cloned (child) skb.
191 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
192 * allocations in case the data is required for writeback
193 * @node: numa node to allocate memory on
195 * Allocate a new &sk_buff. The returned buffer has no headroom and a
196 * tail room of at least size bytes. The object has a reference count
197 * of one. The return is the buffer. On a failure the return is %NULL.
199 * Buffers may only be allocated from interrupts using a @gfp_mask of
202 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
205 struct kmem_cache *cache;
206 struct skb_shared_info *shinfo;
211 cache = (flags & SKB_ALLOC_FCLONE)
212 ? skbuff_fclone_cache : skbuff_head_cache;
214 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
215 gfp_mask |= __GFP_MEMALLOC;
218 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
223 /* We do our best to align skb_shared_info on a separate cache
224 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
225 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
226 * Both skb->head and skb_shared_info are cache line aligned.
228 size = SKB_DATA_ALIGN(size);
229 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
230 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
233 /* kmalloc(size) might give us more room than requested.
234 * Put skb_shared_info exactly at the end of allocated zone,
235 * to allow max possible filling before reallocation.
237 size = SKB_WITH_OVERHEAD(ksize(data));
238 prefetchw(data + size);
241 * Only clear those fields we need to clear, not those that we will
242 * actually initialise below. Hence, don't put any more fields after
243 * the tail pointer in struct sk_buff!
245 memset(skb, 0, offsetof(struct sk_buff, tail));
246 /* Account for allocated memory : skb + skb->head */
247 skb->truesize = SKB_TRUESIZE(size);
248 skb->pfmemalloc = pfmemalloc;
249 atomic_set(&skb->users, 1);
252 skb_reset_tail_pointer(skb);
253 skb->end = skb->tail + size;
254 skb->mac_header = (typeof(skb->mac_header))~0U;
255 skb->transport_header = (typeof(skb->transport_header))~0U;
257 /* make sure we initialize shinfo sequentially */
258 shinfo = skb_shinfo(skb);
259 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
260 atomic_set(&shinfo->dataref, 1);
261 kmemcheck_annotate_variable(shinfo->destructor_arg);
263 if (flags & SKB_ALLOC_FCLONE) {
264 struct sk_buff_fclones *fclones;
266 fclones = container_of(skb, struct sk_buff_fclones, skb1);
268 kmemcheck_annotate_bitfield(&fclones->skb2, flags1);
269 skb->fclone = SKB_FCLONE_ORIG;
270 atomic_set(&fclones->fclone_ref, 1);
272 fclones->skb2.fclone = SKB_FCLONE_CLONE;
273 fclones->skb2.pfmemalloc = pfmemalloc;
278 kmem_cache_free(cache, skb);
282 EXPORT_SYMBOL(__alloc_skb);
285 * __build_skb - build a network buffer
286 * @data: data buffer provided by caller
287 * @frag_size: size of data, or 0 if head was kmalloced
289 * Allocate a new &sk_buff. Caller provides space holding head and
290 * skb_shared_info. @data must have been allocated by kmalloc() only if
291 * @frag_size is 0, otherwise data should come from the page allocator
293 * The return is the new skb buffer.
294 * On a failure the return is %NULL, and @data is not freed.
296 * Before IO, driver allocates only data buffer where NIC put incoming frame
297 * Driver should add room at head (NET_SKB_PAD) and
298 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
299 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
300 * before giving packet to stack.
301 * RX rings only contains data buffers, not full skbs.
303 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
305 struct skb_shared_info *shinfo;
307 unsigned int size = frag_size ? : ksize(data);
309 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
313 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
315 memset(skb, 0, offsetof(struct sk_buff, tail));
316 skb->truesize = SKB_TRUESIZE(size);
317 atomic_set(&skb->users, 1);
320 skb_reset_tail_pointer(skb);
321 skb->end = skb->tail + size;
322 skb->mac_header = (typeof(skb->mac_header))~0U;
323 skb->transport_header = (typeof(skb->transport_header))~0U;
325 /* make sure we initialize shinfo sequentially */
326 shinfo = skb_shinfo(skb);
327 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
328 atomic_set(&shinfo->dataref, 1);
329 kmemcheck_annotate_variable(shinfo->destructor_arg);
334 /* build_skb() is wrapper over __build_skb(), that specifically
335 * takes care of skb->head and skb->pfmemalloc
336 * This means that if @frag_size is not zero, then @data must be backed
337 * by a page fragment, not kmalloc() or vmalloc()
339 struct sk_buff *build_skb(void *data, unsigned int frag_size)
341 struct sk_buff *skb = __build_skb(data, frag_size);
343 if (skb && frag_size) {
345 if (page_is_pfmemalloc(virt_to_head_page(data)))
350 EXPORT_SYMBOL(build_skb);
352 #define NAPI_SKB_CACHE_SIZE 64
354 struct napi_alloc_cache {
355 struct page_frag_cache page;
357 void *skb_cache[NAPI_SKB_CACHE_SIZE];
360 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
361 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
363 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
365 struct page_frag_cache *nc;
369 local_irq_save(flags);
370 nc = this_cpu_ptr(&netdev_alloc_cache);
371 data = __alloc_page_frag(nc, fragsz, gfp_mask);
372 local_irq_restore(flags);
377 * netdev_alloc_frag - allocate a page fragment
378 * @fragsz: fragment size
380 * Allocates a frag from a page for receive buffer.
381 * Uses GFP_ATOMIC allocations.
383 void *netdev_alloc_frag(unsigned int fragsz)
385 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
387 EXPORT_SYMBOL(netdev_alloc_frag);
389 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
391 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
393 return __alloc_page_frag(&nc->page, fragsz, gfp_mask);
396 void *napi_alloc_frag(unsigned int fragsz)
398 return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
400 EXPORT_SYMBOL(napi_alloc_frag);
403 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
404 * @dev: network device to receive on
405 * @len: length to allocate
406 * @gfp_mask: get_free_pages mask, passed to alloc_skb
408 * Allocate a new &sk_buff and assign it a usage count of one. The
409 * buffer has NET_SKB_PAD headroom built in. Users should allocate
410 * the headroom they think they need without accounting for the
411 * built in space. The built in space is used for optimisations.
413 * %NULL is returned if there is no free memory.
415 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
418 struct page_frag_cache *nc;
426 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
427 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
428 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
434 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
435 len = SKB_DATA_ALIGN(len);
437 if (sk_memalloc_socks())
438 gfp_mask |= __GFP_MEMALLOC;
440 local_irq_save(flags);
442 nc = this_cpu_ptr(&netdev_alloc_cache);
443 data = __alloc_page_frag(nc, len, gfp_mask);
444 pfmemalloc = nc->pfmemalloc;
446 local_irq_restore(flags);
451 skb = __build_skb(data, len);
452 if (unlikely(!skb)) {
457 /* use OR instead of assignment to avoid clearing of bits in mask */
463 skb_reserve(skb, NET_SKB_PAD);
469 EXPORT_SYMBOL(__netdev_alloc_skb);
472 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
473 * @napi: napi instance this buffer was allocated for
474 * @len: length to allocate
475 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
477 * Allocate a new sk_buff for use in NAPI receive. This buffer will
478 * attempt to allocate the head from a special reserved region used
479 * only for NAPI Rx allocation. By doing this we can save several
480 * CPU cycles by avoiding having to disable and re-enable IRQs.
482 * %NULL is returned if there is no free memory.
484 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
487 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
491 len += NET_SKB_PAD + NET_IP_ALIGN;
493 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
494 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
495 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
501 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
502 len = SKB_DATA_ALIGN(len);
504 if (sk_memalloc_socks())
505 gfp_mask |= __GFP_MEMALLOC;
507 data = __alloc_page_frag(&nc->page, len, gfp_mask);
511 skb = __build_skb(data, len);
512 if (unlikely(!skb)) {
517 /* use OR instead of assignment to avoid clearing of bits in mask */
518 if (nc->page.pfmemalloc)
523 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
524 skb->dev = napi->dev;
529 EXPORT_SYMBOL(__napi_alloc_skb);
531 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
532 int size, unsigned int truesize)
534 skb_fill_page_desc(skb, i, page, off, size);
536 skb->data_len += size;
537 skb->truesize += truesize;
539 EXPORT_SYMBOL(skb_add_rx_frag);
541 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
542 unsigned int truesize)
544 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
546 skb_frag_size_add(frag, size);
548 skb->data_len += size;
549 skb->truesize += truesize;
551 EXPORT_SYMBOL(skb_coalesce_rx_frag);
553 static void skb_drop_list(struct sk_buff **listp)
555 kfree_skb_list(*listp);
559 static inline void skb_drop_fraglist(struct sk_buff *skb)
561 skb_drop_list(&skb_shinfo(skb)->frag_list);
564 static void skb_clone_fraglist(struct sk_buff *skb)
566 struct sk_buff *list;
568 skb_walk_frags(skb, list)
572 static void skb_free_head(struct sk_buff *skb)
574 unsigned char *head = skb->head;
582 static void skb_release_data(struct sk_buff *skb)
584 struct skb_shared_info *shinfo = skb_shinfo(skb);
588 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
592 for (i = 0; i < shinfo->nr_frags; i++)
593 __skb_frag_unref(&shinfo->frags[i]);
596 * If skb buf is from userspace, we need to notify the caller
597 * the lower device DMA has done;
599 if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) {
600 struct ubuf_info *uarg;
602 uarg = shinfo->destructor_arg;
604 uarg->callback(uarg, true);
607 if (shinfo->frag_list)
608 kfree_skb_list(shinfo->frag_list);
614 * Free an skbuff by memory without cleaning the state.
616 static void kfree_skbmem(struct sk_buff *skb)
618 struct sk_buff_fclones *fclones;
620 switch (skb->fclone) {
621 case SKB_FCLONE_UNAVAILABLE:
622 kmem_cache_free(skbuff_head_cache, skb);
625 case SKB_FCLONE_ORIG:
626 fclones = container_of(skb, struct sk_buff_fclones, skb1);
628 /* We usually free the clone (TX completion) before original skb
629 * This test would have no chance to be true for the clone,
630 * while here, branch prediction will be good.
632 if (atomic_read(&fclones->fclone_ref) == 1)
636 default: /* SKB_FCLONE_CLONE */
637 fclones = container_of(skb, struct sk_buff_fclones, skb2);
640 if (!atomic_dec_and_test(&fclones->fclone_ref))
643 kmem_cache_free(skbuff_fclone_cache, fclones);
646 static void skb_release_head_state(struct sk_buff *skb)
650 secpath_put(skb->sp);
652 if (skb->destructor) {
654 skb->destructor(skb);
656 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
657 nf_conntrack_put(skb->nfct);
659 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
660 nf_bridge_put(skb->nf_bridge);
664 /* Free everything but the sk_buff shell. */
665 static void skb_release_all(struct sk_buff *skb)
667 skb_release_head_state(skb);
668 if (likely(skb->head))
669 skb_release_data(skb);
673 * __kfree_skb - private function
676 * Free an sk_buff. Release anything attached to the buffer.
677 * Clean the state. This is an internal helper function. Users should
678 * always call kfree_skb
681 void __kfree_skb(struct sk_buff *skb)
683 skb_release_all(skb);
686 EXPORT_SYMBOL(__kfree_skb);
689 * kfree_skb - free an sk_buff
690 * @skb: buffer to free
692 * Drop a reference to the buffer and free it if the usage count has
695 void kfree_skb(struct sk_buff *skb)
699 if (likely(atomic_read(&skb->users) == 1))
701 else if (likely(!atomic_dec_and_test(&skb->users)))
703 trace_kfree_skb(skb, __builtin_return_address(0));
706 EXPORT_SYMBOL(kfree_skb);
708 void kfree_skb_list(struct sk_buff *segs)
711 struct sk_buff *next = segs->next;
717 EXPORT_SYMBOL(kfree_skb_list);
720 * skb_tx_error - report an sk_buff xmit error
721 * @skb: buffer that triggered an error
723 * Report xmit error if a device callback is tracking this skb.
724 * skb must be freed afterwards.
726 void skb_tx_error(struct sk_buff *skb)
728 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
729 struct ubuf_info *uarg;
731 uarg = skb_shinfo(skb)->destructor_arg;
733 uarg->callback(uarg, false);
734 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
737 EXPORT_SYMBOL(skb_tx_error);
740 * consume_skb - free an skbuff
741 * @skb: buffer to free
743 * Drop a ref to the buffer and free it if the usage count has hit zero
744 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
745 * is being dropped after a failure and notes that
747 void consume_skb(struct sk_buff *skb)
751 if (likely(atomic_read(&skb->users) == 1))
753 else if (likely(!atomic_dec_and_test(&skb->users)))
755 trace_consume_skb(skb);
758 EXPORT_SYMBOL(consume_skb);
760 void __kfree_skb_flush(void)
762 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
764 /* flush skb_cache if containing objects */
766 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
772 static inline void _kfree_skb_defer(struct sk_buff *skb)
774 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
776 /* drop skb->head and call any destructors for packet */
777 skb_release_all(skb);
779 /* record skb to CPU local list */
780 nc->skb_cache[nc->skb_count++] = skb;
783 /* SLUB writes into objects when freeing */
787 /* flush skb_cache if it is filled */
788 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
789 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
794 void __kfree_skb_defer(struct sk_buff *skb)
796 _kfree_skb_defer(skb);
799 void napi_consume_skb(struct sk_buff *skb, int budget)
804 /* Zero budget indicate non-NAPI context called us, like netpoll */
805 if (unlikely(!budget)) {
806 dev_consume_skb_any(skb);
810 if (likely(atomic_read(&skb->users) == 1))
812 else if (likely(!atomic_dec_and_test(&skb->users)))
814 /* if reaching here SKB is ready to free */
815 trace_consume_skb(skb);
817 /* if SKB is a clone, don't handle this case */
818 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
823 _kfree_skb_defer(skb);
825 EXPORT_SYMBOL(napi_consume_skb);
827 /* Make sure a field is enclosed inside headers_start/headers_end section */
828 #define CHECK_SKB_FIELD(field) \
829 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
830 offsetof(struct sk_buff, headers_start)); \
831 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
832 offsetof(struct sk_buff, headers_end)); \
834 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
836 new->tstamp = old->tstamp;
837 /* We do not copy old->sk */
839 memcpy(new->cb, old->cb, sizeof(old->cb));
840 skb_dst_copy(new, old);
842 new->sp = secpath_get(old->sp);
844 __nf_copy(new, old, false);
846 /* Note : this field could be in headers_start/headers_end section
847 * It is not yet because we do not want to have a 16 bit hole
849 new->queue_mapping = old->queue_mapping;
851 memcpy(&new->headers_start, &old->headers_start,
852 offsetof(struct sk_buff, headers_end) -
853 offsetof(struct sk_buff, headers_start));
854 CHECK_SKB_FIELD(protocol);
855 CHECK_SKB_FIELD(csum);
856 CHECK_SKB_FIELD(hash);
857 CHECK_SKB_FIELD(priority);
858 CHECK_SKB_FIELD(skb_iif);
859 CHECK_SKB_FIELD(vlan_proto);
860 CHECK_SKB_FIELD(vlan_tci);
861 CHECK_SKB_FIELD(transport_header);
862 CHECK_SKB_FIELD(network_header);
863 CHECK_SKB_FIELD(mac_header);
864 CHECK_SKB_FIELD(inner_protocol);
865 CHECK_SKB_FIELD(inner_transport_header);
866 CHECK_SKB_FIELD(inner_network_header);
867 CHECK_SKB_FIELD(inner_mac_header);
868 CHECK_SKB_FIELD(mark);
869 #ifdef CONFIG_NETWORK_SECMARK
870 CHECK_SKB_FIELD(secmark);
872 #ifdef CONFIG_NET_RX_BUSY_POLL
873 CHECK_SKB_FIELD(napi_id);
876 CHECK_SKB_FIELD(sender_cpu);
878 #ifdef CONFIG_NET_SCHED
879 CHECK_SKB_FIELD(tc_index);
880 #ifdef CONFIG_NET_CLS_ACT
881 CHECK_SKB_FIELD(tc_verd);
888 * You should not add any new code to this function. Add it to
889 * __copy_skb_header above instead.
891 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
893 #define C(x) n->x = skb->x
895 n->next = n->prev = NULL;
897 __copy_skb_header(n, skb);
902 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
905 n->destructor = NULL;
912 atomic_set(&n->users, 1);
914 atomic_inc(&(skb_shinfo(skb)->dataref));
922 * skb_morph - morph one skb into another
923 * @dst: the skb to receive the contents
924 * @src: the skb to supply the contents
926 * This is identical to skb_clone except that the target skb is
927 * supplied by the user.
929 * The target skb is returned upon exit.
931 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
933 skb_release_all(dst);
934 return __skb_clone(dst, src);
936 EXPORT_SYMBOL_GPL(skb_morph);
939 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
940 * @skb: the skb to modify
941 * @gfp_mask: allocation priority
943 * This must be called on SKBTX_DEV_ZEROCOPY skb.
944 * It will copy all frags into kernel and drop the reference
945 * to userspace pages.
947 * If this function is called from an interrupt gfp_mask() must be
950 * Returns 0 on success or a negative error code on failure
951 * to allocate kernel memory to copy to.
953 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
956 int num_frags = skb_shinfo(skb)->nr_frags;
957 struct page *page, *head = NULL;
958 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
960 for (i = 0; i < num_frags; i++) {
962 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
964 page = alloc_page(gfp_mask);
967 struct page *next = (struct page *)page_private(head);
973 vaddr = kmap_atomic(skb_frag_page(f));
974 memcpy(page_address(page),
975 vaddr + f->page_offset, skb_frag_size(f));
976 kunmap_atomic(vaddr);
977 set_page_private(page, (unsigned long)head);
981 /* skb frags release userspace buffers */
982 for (i = 0; i < num_frags; i++)
983 skb_frag_unref(skb, i);
985 uarg->callback(uarg, false);
987 /* skb frags point to kernel buffers */
988 for (i = num_frags - 1; i >= 0; i--) {
989 __skb_fill_page_desc(skb, i, head, 0,
990 skb_shinfo(skb)->frags[i].size);
991 head = (struct page *)page_private(head);
994 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
997 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1000 * skb_clone - duplicate an sk_buff
1001 * @skb: buffer to clone
1002 * @gfp_mask: allocation priority
1004 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1005 * copies share the same packet data but not structure. The new
1006 * buffer has a reference count of 1. If the allocation fails the
1007 * function returns %NULL otherwise the new buffer is returned.
1009 * If this function is called from an interrupt gfp_mask() must be
1013 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1015 struct sk_buff_fclones *fclones = container_of(skb,
1016 struct sk_buff_fclones,
1020 if (skb_orphan_frags(skb, gfp_mask))
1023 if (skb->fclone == SKB_FCLONE_ORIG &&
1024 atomic_read(&fclones->fclone_ref) == 1) {
1026 atomic_set(&fclones->fclone_ref, 2);
1028 if (skb_pfmemalloc(skb))
1029 gfp_mask |= __GFP_MEMALLOC;
1031 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1035 kmemcheck_annotate_bitfield(n, flags1);
1036 n->fclone = SKB_FCLONE_UNAVAILABLE;
1039 return __skb_clone(n, skb);
1041 EXPORT_SYMBOL(skb_clone);
1043 static void skb_headers_offset_update(struct sk_buff *skb, int off)
1045 /* Only adjust this if it actually is csum_start rather than csum */
1046 if (skb->ip_summed == CHECKSUM_PARTIAL)
1047 skb->csum_start += off;
1048 /* {transport,network,mac}_header and tail are relative to skb->head */
1049 skb->transport_header += off;
1050 skb->network_header += off;
1051 if (skb_mac_header_was_set(skb))
1052 skb->mac_header += off;
1053 skb->inner_transport_header += off;
1054 skb->inner_network_header += off;
1055 skb->inner_mac_header += off;
1058 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1060 __copy_skb_header(new, old);
1062 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1063 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1064 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1067 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1069 if (skb_pfmemalloc(skb))
1070 return SKB_ALLOC_RX;
1075 * skb_copy - create private copy of an sk_buff
1076 * @skb: buffer to copy
1077 * @gfp_mask: allocation priority
1079 * Make a copy of both an &sk_buff and its data. This is used when the
1080 * caller wishes to modify the data and needs a private copy of the
1081 * data to alter. Returns %NULL on failure or the pointer to the buffer
1082 * on success. The returned buffer has a reference count of 1.
1084 * As by-product this function converts non-linear &sk_buff to linear
1085 * one, so that &sk_buff becomes completely private and caller is allowed
1086 * to modify all the data of returned buffer. This means that this
1087 * function is not recommended for use in circumstances when only
1088 * header is going to be modified. Use pskb_copy() instead.
1091 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1093 int headerlen = skb_headroom(skb);
1094 unsigned int size = skb_end_offset(skb) + skb->data_len;
1095 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1096 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1101 /* Set the data pointer */
1102 skb_reserve(n, headerlen);
1103 /* Set the tail pointer and length */
1104 skb_put(n, skb->len);
1106 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1109 copy_skb_header(n, skb);
1112 EXPORT_SYMBOL(skb_copy);
1115 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1116 * @skb: buffer to copy
1117 * @headroom: headroom of new skb
1118 * @gfp_mask: allocation priority
1119 * @fclone: if true allocate the copy of the skb from the fclone
1120 * cache instead of the head cache; it is recommended to set this
1121 * to true for the cases where the copy will likely be cloned
1123 * Make a copy of both an &sk_buff and part of its data, located
1124 * in header. Fragmented data remain shared. This is used when
1125 * the caller wishes to modify only header of &sk_buff and needs
1126 * private copy of the header to alter. Returns %NULL on failure
1127 * or the pointer to the buffer on success.
1128 * The returned buffer has a reference count of 1.
1131 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1132 gfp_t gfp_mask, bool fclone)
1134 unsigned int size = skb_headlen(skb) + headroom;
1135 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1136 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1141 /* Set the data pointer */
1142 skb_reserve(n, headroom);
1143 /* Set the tail pointer and length */
1144 skb_put(n, skb_headlen(skb));
1145 /* Copy the bytes */
1146 skb_copy_from_linear_data(skb, n->data, n->len);
1148 n->truesize += skb->data_len;
1149 n->data_len = skb->data_len;
1152 if (skb_shinfo(skb)->nr_frags) {
1155 if (skb_orphan_frags(skb, gfp_mask)) {
1160 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1161 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1162 skb_frag_ref(skb, i);
1164 skb_shinfo(n)->nr_frags = i;
1167 if (skb_has_frag_list(skb)) {
1168 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1169 skb_clone_fraglist(n);
1172 copy_skb_header(n, skb);
1176 EXPORT_SYMBOL(__pskb_copy_fclone);
1179 * pskb_expand_head - reallocate header of &sk_buff
1180 * @skb: buffer to reallocate
1181 * @nhead: room to add at head
1182 * @ntail: room to add at tail
1183 * @gfp_mask: allocation priority
1185 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1186 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1187 * reference count of 1. Returns zero in the case of success or error,
1188 * if expansion failed. In the last case, &sk_buff is not changed.
1190 * All the pointers pointing into skb header may change and must be
1191 * reloaded after call to this function.
1194 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1199 int size = nhead + skb_end_offset(skb) + ntail;
1204 if (skb_shared(skb))
1207 size = SKB_DATA_ALIGN(size);
1209 if (skb_pfmemalloc(skb))
1210 gfp_mask |= __GFP_MEMALLOC;
1211 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1212 gfp_mask, NUMA_NO_NODE, NULL);
1215 size = SKB_WITH_OVERHEAD(ksize(data));
1217 /* Copy only real data... and, alas, header. This should be
1218 * optimized for the cases when header is void.
1220 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1222 memcpy((struct skb_shared_info *)(data + size),
1224 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1227 * if shinfo is shared we must drop the old head gracefully, but if it
1228 * is not we can just drop the old head and let the existing refcount
1229 * be since all we did is relocate the values
1231 if (skb_cloned(skb)) {
1232 /* copy this zero copy skb frags */
1233 if (skb_orphan_frags(skb, gfp_mask))
1235 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1236 skb_frag_ref(skb, i);
1238 if (skb_has_frag_list(skb))
1239 skb_clone_fraglist(skb);
1241 skb_release_data(skb);
1245 off = (data + nhead) - skb->head;
1250 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1254 skb->end = skb->head + size;
1257 skb_headers_offset_update(skb, nhead);
1261 atomic_set(&skb_shinfo(skb)->dataref, 1);
1269 EXPORT_SYMBOL(pskb_expand_head);
1271 /* Make private copy of skb with writable head and some headroom */
1273 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1275 struct sk_buff *skb2;
1276 int delta = headroom - skb_headroom(skb);
1279 skb2 = pskb_copy(skb, GFP_ATOMIC);
1281 skb2 = skb_clone(skb, GFP_ATOMIC);
1282 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1290 EXPORT_SYMBOL(skb_realloc_headroom);
1293 * skb_copy_expand - copy and expand sk_buff
1294 * @skb: buffer to copy
1295 * @newheadroom: new free bytes at head
1296 * @newtailroom: new free bytes at tail
1297 * @gfp_mask: allocation priority
1299 * Make a copy of both an &sk_buff and its data and while doing so
1300 * allocate additional space.
1302 * This is used when the caller wishes to modify the data and needs a
1303 * private copy of the data to alter as well as more space for new fields.
1304 * Returns %NULL on failure or the pointer to the buffer
1305 * on success. The returned buffer has a reference count of 1.
1307 * You must pass %GFP_ATOMIC as the allocation priority if this function
1308 * is called from an interrupt.
1310 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1311 int newheadroom, int newtailroom,
1315 * Allocate the copy buffer
1317 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1318 gfp_mask, skb_alloc_rx_flag(skb),
1320 int oldheadroom = skb_headroom(skb);
1321 int head_copy_len, head_copy_off;
1326 skb_reserve(n, newheadroom);
1328 /* Set the tail pointer and length */
1329 skb_put(n, skb->len);
1331 head_copy_len = oldheadroom;
1333 if (newheadroom <= head_copy_len)
1334 head_copy_len = newheadroom;
1336 head_copy_off = newheadroom - head_copy_len;
1338 /* Copy the linear header and data. */
1339 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1340 skb->len + head_copy_len))
1343 copy_skb_header(n, skb);
1345 skb_headers_offset_update(n, newheadroom - oldheadroom);
1349 EXPORT_SYMBOL(skb_copy_expand);
1352 * skb_pad - zero pad the tail of an skb
1353 * @skb: buffer to pad
1354 * @pad: space to pad
1356 * Ensure that a buffer is followed by a padding area that is zero
1357 * filled. Used by network drivers which may DMA or transfer data
1358 * beyond the buffer end onto the wire.
1360 * May return error in out of memory cases. The skb is freed on error.
1363 int skb_pad(struct sk_buff *skb, int pad)
1368 /* If the skbuff is non linear tailroom is always zero.. */
1369 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1370 memset(skb->data+skb->len, 0, pad);
1374 ntail = skb->data_len + pad - (skb->end - skb->tail);
1375 if (likely(skb_cloned(skb) || ntail > 0)) {
1376 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1381 /* FIXME: The use of this function with non-linear skb's really needs
1384 err = skb_linearize(skb);
1388 memset(skb->data + skb->len, 0, pad);
1395 EXPORT_SYMBOL(skb_pad);
1398 * pskb_put - add data to the tail of a potentially fragmented buffer
1399 * @skb: start of the buffer to use
1400 * @tail: tail fragment of the buffer to use
1401 * @len: amount of data to add
1403 * This function extends the used data area of the potentially
1404 * fragmented buffer. @tail must be the last fragment of @skb -- or
1405 * @skb itself. If this would exceed the total buffer size the kernel
1406 * will panic. A pointer to the first byte of the extra data is
1410 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1413 skb->data_len += len;
1416 return skb_put(tail, len);
1418 EXPORT_SYMBOL_GPL(pskb_put);
1421 * skb_put - add data to a buffer
1422 * @skb: buffer to use
1423 * @len: amount of data to add
1425 * This function extends the used data area of the buffer. If this would
1426 * exceed the total buffer size the kernel will panic. A pointer to the
1427 * first byte of the extra data is returned.
1429 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1431 unsigned char *tmp = skb_tail_pointer(skb);
1432 SKB_LINEAR_ASSERT(skb);
1435 if (unlikely(skb->tail > skb->end))
1436 skb_over_panic(skb, len, __builtin_return_address(0));
1439 EXPORT_SYMBOL(skb_put);
1442 * skb_push - add data to the start of a buffer
1443 * @skb: buffer to use
1444 * @len: amount of data to add
1446 * This function extends the used data area of the buffer at the buffer
1447 * start. If this would exceed the total buffer headroom the kernel will
1448 * panic. A pointer to the first byte of the extra data is returned.
1450 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1454 if (unlikely(skb->data<skb->head))
1455 skb_under_panic(skb, len, __builtin_return_address(0));
1458 EXPORT_SYMBOL(skb_push);
1461 * skb_pull - remove data from the start of a buffer
1462 * @skb: buffer to use
1463 * @len: amount of data to remove
1465 * This function removes data from the start of a buffer, returning
1466 * the memory to the headroom. A pointer to the next data in the buffer
1467 * is returned. Once the data has been pulled future pushes will overwrite
1470 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1472 return skb_pull_inline(skb, len);
1474 EXPORT_SYMBOL(skb_pull);
1477 * skb_trim - remove end from a buffer
1478 * @skb: buffer to alter
1481 * Cut the length of a buffer down by removing data from the tail. If
1482 * the buffer is already under the length specified it is not modified.
1483 * The skb must be linear.
1485 void skb_trim(struct sk_buff *skb, unsigned int len)
1488 __skb_trim(skb, len);
1490 EXPORT_SYMBOL(skb_trim);
1492 /* Trims skb to length len. It can change skb pointers.
1495 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1497 struct sk_buff **fragp;
1498 struct sk_buff *frag;
1499 int offset = skb_headlen(skb);
1500 int nfrags = skb_shinfo(skb)->nr_frags;
1504 if (skb_cloned(skb) &&
1505 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1512 for (; i < nfrags; i++) {
1513 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1520 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1523 skb_shinfo(skb)->nr_frags = i;
1525 for (; i < nfrags; i++)
1526 skb_frag_unref(skb, i);
1528 if (skb_has_frag_list(skb))
1529 skb_drop_fraglist(skb);
1533 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1534 fragp = &frag->next) {
1535 int end = offset + frag->len;
1537 if (skb_shared(frag)) {
1538 struct sk_buff *nfrag;
1540 nfrag = skb_clone(frag, GFP_ATOMIC);
1541 if (unlikely(!nfrag))
1544 nfrag->next = frag->next;
1556 unlikely((err = pskb_trim(frag, len - offset))))
1560 skb_drop_list(&frag->next);
1565 if (len > skb_headlen(skb)) {
1566 skb->data_len -= skb->len - len;
1571 skb_set_tail_pointer(skb, len);
1576 EXPORT_SYMBOL(___pskb_trim);
1579 * __pskb_pull_tail - advance tail of skb header
1580 * @skb: buffer to reallocate
1581 * @delta: number of bytes to advance tail
1583 * The function makes a sense only on a fragmented &sk_buff,
1584 * it expands header moving its tail forward and copying necessary
1585 * data from fragmented part.
1587 * &sk_buff MUST have reference count of 1.
1589 * Returns %NULL (and &sk_buff does not change) if pull failed
1590 * or value of new tail of skb in the case of success.
1592 * All the pointers pointing into skb header may change and must be
1593 * reloaded after call to this function.
1596 /* Moves tail of skb head forward, copying data from fragmented part,
1597 * when it is necessary.
1598 * 1. It may fail due to malloc failure.
1599 * 2. It may change skb pointers.
1601 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1603 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1605 /* If skb has not enough free space at tail, get new one
1606 * plus 128 bytes for future expansions. If we have enough
1607 * room at tail, reallocate without expansion only if skb is cloned.
1609 int i, k, eat = (skb->tail + delta) - skb->end;
1611 if (eat > 0 || skb_cloned(skb)) {
1612 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1617 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1620 /* Optimization: no fragments, no reasons to preestimate
1621 * size of pulled pages. Superb.
1623 if (!skb_has_frag_list(skb))
1626 /* Estimate size of pulled pages. */
1628 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1629 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1636 /* If we need update frag list, we are in troubles.
1637 * Certainly, it possible to add an offset to skb data,
1638 * but taking into account that pulling is expected to
1639 * be very rare operation, it is worth to fight against
1640 * further bloating skb head and crucify ourselves here instead.
1641 * Pure masohism, indeed. 8)8)
1644 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1645 struct sk_buff *clone = NULL;
1646 struct sk_buff *insp = NULL;
1651 if (list->len <= eat) {
1652 /* Eaten as whole. */
1657 /* Eaten partially. */
1659 if (skb_shared(list)) {
1660 /* Sucks! We need to fork list. :-( */
1661 clone = skb_clone(list, GFP_ATOMIC);
1667 /* This may be pulled without
1671 if (!pskb_pull(list, eat)) {
1679 /* Free pulled out fragments. */
1680 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1681 skb_shinfo(skb)->frag_list = list->next;
1684 /* And insert new clone at head. */
1687 skb_shinfo(skb)->frag_list = clone;
1690 /* Success! Now we may commit changes to skb data. */
1695 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1696 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1699 skb_frag_unref(skb, i);
1702 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1704 skb_shinfo(skb)->frags[k].page_offset += eat;
1705 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1711 skb_shinfo(skb)->nr_frags = k;
1714 skb->data_len -= delta;
1716 return skb_tail_pointer(skb);
1718 EXPORT_SYMBOL(__pskb_pull_tail);
1721 * skb_copy_bits - copy bits from skb to kernel buffer
1723 * @offset: offset in source
1724 * @to: destination buffer
1725 * @len: number of bytes to copy
1727 * Copy the specified number of bytes from the source skb to the
1728 * destination buffer.
1731 * If its prototype is ever changed,
1732 * check arch/{*}/net/{*}.S files,
1733 * since it is called from BPF assembly code.
1735 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1737 int start = skb_headlen(skb);
1738 struct sk_buff *frag_iter;
1741 if (offset > (int)skb->len - len)
1745 if ((copy = start - offset) > 0) {
1748 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1749 if ((len -= copy) == 0)
1755 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1757 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1759 WARN_ON(start > offset + len);
1761 end = start + skb_frag_size(f);
1762 if ((copy = end - offset) > 0) {
1768 vaddr = kmap_atomic(skb_frag_page(f));
1770 vaddr + f->page_offset + offset - start,
1772 kunmap_atomic(vaddr);
1774 if ((len -= copy) == 0)
1782 skb_walk_frags(skb, frag_iter) {
1785 WARN_ON(start > offset + len);
1787 end = start + frag_iter->len;
1788 if ((copy = end - offset) > 0) {
1791 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1793 if ((len -= copy) == 0)
1807 EXPORT_SYMBOL(skb_copy_bits);
1810 * Callback from splice_to_pipe(), if we need to release some pages
1811 * at the end of the spd in case we error'ed out in filling the pipe.
1813 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1815 put_page(spd->pages[i]);
1818 static struct page *linear_to_page(struct page *page, unsigned int *len,
1819 unsigned int *offset,
1822 struct page_frag *pfrag = sk_page_frag(sk);
1824 if (!sk_page_frag_refill(sk, pfrag))
1827 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1829 memcpy(page_address(pfrag->page) + pfrag->offset,
1830 page_address(page) + *offset, *len);
1831 *offset = pfrag->offset;
1832 pfrag->offset += *len;
1837 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1839 unsigned int offset)
1841 return spd->nr_pages &&
1842 spd->pages[spd->nr_pages - 1] == page &&
1843 (spd->partial[spd->nr_pages - 1].offset +
1844 spd->partial[spd->nr_pages - 1].len == offset);
1848 * Fill page/offset/length into spd, if it can hold more pages.
1850 static bool spd_fill_page(struct splice_pipe_desc *spd,
1851 struct pipe_inode_info *pipe, struct page *page,
1852 unsigned int *len, unsigned int offset,
1856 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1860 page = linear_to_page(page, len, &offset, sk);
1864 if (spd_can_coalesce(spd, page, offset)) {
1865 spd->partial[spd->nr_pages - 1].len += *len;
1869 spd->pages[spd->nr_pages] = page;
1870 spd->partial[spd->nr_pages].len = *len;
1871 spd->partial[spd->nr_pages].offset = offset;
1877 static bool __splice_segment(struct page *page, unsigned int poff,
1878 unsigned int plen, unsigned int *off,
1880 struct splice_pipe_desc *spd, bool linear,
1882 struct pipe_inode_info *pipe)
1887 /* skip this segment if already processed */
1893 /* ignore any bits we already processed */
1899 unsigned int flen = min(*len, plen);
1901 if (spd_fill_page(spd, pipe, page, &flen, poff,
1907 } while (*len && plen);
1913 * Map linear and fragment data from the skb to spd. It reports true if the
1914 * pipe is full or if we already spliced the requested length.
1916 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1917 unsigned int *offset, unsigned int *len,
1918 struct splice_pipe_desc *spd, struct sock *sk)
1921 struct sk_buff *iter;
1923 /* map the linear part :
1924 * If skb->head_frag is set, this 'linear' part is backed by a
1925 * fragment, and if the head is not shared with any clones then
1926 * we can avoid a copy since we own the head portion of this page.
1928 if (__splice_segment(virt_to_page(skb->data),
1929 (unsigned long) skb->data & (PAGE_SIZE - 1),
1932 skb_head_is_locked(skb),
1937 * then map the fragments
1939 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1940 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1942 if (__splice_segment(skb_frag_page(f),
1943 f->page_offset, skb_frag_size(f),
1944 offset, len, spd, false, sk, pipe))
1948 skb_walk_frags(skb, iter) {
1949 if (*offset >= iter->len) {
1950 *offset -= iter->len;
1953 /* __skb_splice_bits() only fails if the output has no room
1954 * left, so no point in going over the frag_list for the error
1957 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
1964 ssize_t skb_socket_splice(struct sock *sk,
1965 struct pipe_inode_info *pipe,
1966 struct splice_pipe_desc *spd)
1970 /* Drop the socket lock, otherwise we have reverse
1971 * locking dependencies between sk_lock and i_mutex
1972 * here as compared to sendfile(). We enter here
1973 * with the socket lock held, and splice_to_pipe() will
1974 * grab the pipe inode lock. For sendfile() emulation,
1975 * we call into ->sendpage() with the i_mutex lock held
1976 * and networking will grab the socket lock.
1979 ret = splice_to_pipe(pipe, spd);
1986 * Map data from the skb to a pipe. Should handle both the linear part,
1987 * the fragments, and the frag list.
1989 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
1990 struct pipe_inode_info *pipe, unsigned int tlen,
1992 ssize_t (*splice_cb)(struct sock *,
1993 struct pipe_inode_info *,
1994 struct splice_pipe_desc *))
1996 struct partial_page partial[MAX_SKB_FRAGS];
1997 struct page *pages[MAX_SKB_FRAGS];
1998 struct splice_pipe_desc spd = {
2001 .nr_pages_max = MAX_SKB_FRAGS,
2003 .ops = &nosteal_pipe_buf_ops,
2004 .spd_release = sock_spd_release,
2008 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2011 ret = splice_cb(sk, pipe, &spd);
2015 EXPORT_SYMBOL_GPL(skb_splice_bits);
2018 * skb_store_bits - store bits from kernel buffer to skb
2019 * @skb: destination buffer
2020 * @offset: offset in destination
2021 * @from: source buffer
2022 * @len: number of bytes to copy
2024 * Copy the specified number of bytes from the source buffer to the
2025 * destination skb. This function handles all the messy bits of
2026 * traversing fragment lists and such.
2029 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2031 int start = skb_headlen(skb);
2032 struct sk_buff *frag_iter;
2035 if (offset > (int)skb->len - len)
2038 if ((copy = start - offset) > 0) {
2041 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2042 if ((len -= copy) == 0)
2048 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2049 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2052 WARN_ON(start > offset + len);
2054 end = start + skb_frag_size(frag);
2055 if ((copy = end - offset) > 0) {
2061 vaddr = kmap_atomic(skb_frag_page(frag));
2062 memcpy(vaddr + frag->page_offset + offset - start,
2064 kunmap_atomic(vaddr);
2066 if ((len -= copy) == 0)
2074 skb_walk_frags(skb, frag_iter) {
2077 WARN_ON(start > offset + len);
2079 end = start + frag_iter->len;
2080 if ((copy = end - offset) > 0) {
2083 if (skb_store_bits(frag_iter, offset - start,
2086 if ((len -= copy) == 0)
2099 EXPORT_SYMBOL(skb_store_bits);
2101 /* Checksum skb data. */
2102 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2103 __wsum csum, const struct skb_checksum_ops *ops)
2105 int start = skb_headlen(skb);
2106 int i, copy = start - offset;
2107 struct sk_buff *frag_iter;
2110 /* Checksum header. */
2114 csum = ops->update(skb->data + offset, copy, csum);
2115 if ((len -= copy) == 0)
2121 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2123 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2125 WARN_ON(start > offset + len);
2127 end = start + skb_frag_size(frag);
2128 if ((copy = end - offset) > 0) {
2134 vaddr = kmap_atomic(skb_frag_page(frag));
2135 csum2 = ops->update(vaddr + frag->page_offset +
2136 offset - start, copy, 0);
2137 kunmap_atomic(vaddr);
2138 csum = ops->combine(csum, csum2, pos, copy);
2147 skb_walk_frags(skb, frag_iter) {
2150 WARN_ON(start > offset + len);
2152 end = start + frag_iter->len;
2153 if ((copy = end - offset) > 0) {
2157 csum2 = __skb_checksum(frag_iter, offset - start,
2159 csum = ops->combine(csum, csum2, pos, copy);
2160 if ((len -= copy) == 0)
2171 EXPORT_SYMBOL(__skb_checksum);
2173 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2174 int len, __wsum csum)
2176 const struct skb_checksum_ops ops = {
2177 .update = csum_partial_ext,
2178 .combine = csum_block_add_ext,
2181 return __skb_checksum(skb, offset, len, csum, &ops);
2183 EXPORT_SYMBOL(skb_checksum);
2185 /* Both of above in one bottle. */
2187 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2188 u8 *to, int len, __wsum csum)
2190 int start = skb_headlen(skb);
2191 int i, copy = start - offset;
2192 struct sk_buff *frag_iter;
2199 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2201 if ((len -= copy) == 0)
2208 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2211 WARN_ON(start > offset + len);
2213 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2214 if ((copy = end - offset) > 0) {
2217 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2221 vaddr = kmap_atomic(skb_frag_page(frag));
2222 csum2 = csum_partial_copy_nocheck(vaddr +
2226 kunmap_atomic(vaddr);
2227 csum = csum_block_add(csum, csum2, pos);
2237 skb_walk_frags(skb, frag_iter) {
2241 WARN_ON(start > offset + len);
2243 end = start + frag_iter->len;
2244 if ((copy = end - offset) > 0) {
2247 csum2 = skb_copy_and_csum_bits(frag_iter,
2250 csum = csum_block_add(csum, csum2, pos);
2251 if ((len -= copy) == 0)
2262 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2265 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2266 * @from: source buffer
2268 * Calculates the amount of linear headroom needed in the 'to' skb passed
2269 * into skb_zerocopy().
2272 skb_zerocopy_headlen(const struct sk_buff *from)
2274 unsigned int hlen = 0;
2276 if (!from->head_frag ||
2277 skb_headlen(from) < L1_CACHE_BYTES ||
2278 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2279 hlen = skb_headlen(from);
2281 if (skb_has_frag_list(from))
2286 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2289 * skb_zerocopy - Zero copy skb to skb
2290 * @to: destination buffer
2291 * @from: source buffer
2292 * @len: number of bytes to copy from source buffer
2293 * @hlen: size of linear headroom in destination buffer
2295 * Copies up to `len` bytes from `from` to `to` by creating references
2296 * to the frags in the source buffer.
2298 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2299 * headroom in the `to` buffer.
2302 * 0: everything is OK
2303 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2304 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2307 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2310 int plen = 0; /* length of skb->head fragment */
2313 unsigned int offset;
2315 BUG_ON(!from->head_frag && !hlen);
2317 /* dont bother with small payloads */
2318 if (len <= skb_tailroom(to))
2319 return skb_copy_bits(from, 0, skb_put(to, len), len);
2322 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2327 plen = min_t(int, skb_headlen(from), len);
2329 page = virt_to_head_page(from->head);
2330 offset = from->data - (unsigned char *)page_address(page);
2331 __skb_fill_page_desc(to, 0, page, offset, plen);
2338 to->truesize += len + plen;
2339 to->len += len + plen;
2340 to->data_len += len + plen;
2342 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2347 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2350 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2351 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2352 len -= skb_shinfo(to)->frags[j].size;
2353 skb_frag_ref(to, j);
2356 skb_shinfo(to)->nr_frags = j;
2360 EXPORT_SYMBOL_GPL(skb_zerocopy);
2362 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2367 if (skb->ip_summed == CHECKSUM_PARTIAL)
2368 csstart = skb_checksum_start_offset(skb);
2370 csstart = skb_headlen(skb);
2372 BUG_ON(csstart > skb_headlen(skb));
2374 skb_copy_from_linear_data(skb, to, csstart);
2377 if (csstart != skb->len)
2378 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2379 skb->len - csstart, 0);
2381 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2382 long csstuff = csstart + skb->csum_offset;
2384 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2387 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2390 * skb_dequeue - remove from the head of the queue
2391 * @list: list to dequeue from
2393 * Remove the head of the list. The list lock is taken so the function
2394 * may be used safely with other locking list functions. The head item is
2395 * returned or %NULL if the list is empty.
2398 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2400 unsigned long flags;
2401 struct sk_buff *result;
2403 spin_lock_irqsave(&list->lock, flags);
2404 result = __skb_dequeue(list);
2405 spin_unlock_irqrestore(&list->lock, flags);
2408 EXPORT_SYMBOL(skb_dequeue);
2411 * skb_dequeue_tail - remove from the tail of the queue
2412 * @list: list to dequeue from
2414 * Remove the tail of the list. The list lock is taken so the function
2415 * may be used safely with other locking list functions. The tail item is
2416 * returned or %NULL if the list is empty.
2418 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2420 unsigned long flags;
2421 struct sk_buff *result;
2423 spin_lock_irqsave(&list->lock, flags);
2424 result = __skb_dequeue_tail(list);
2425 spin_unlock_irqrestore(&list->lock, flags);
2428 EXPORT_SYMBOL(skb_dequeue_tail);
2431 * skb_queue_purge - empty a list
2432 * @list: list to empty
2434 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2435 * the list and one reference dropped. This function takes the list
2436 * lock and is atomic with respect to other list locking functions.
2438 void skb_queue_purge(struct sk_buff_head *list)
2440 struct sk_buff *skb;
2441 while ((skb = skb_dequeue(list)) != NULL)
2444 EXPORT_SYMBOL(skb_queue_purge);
2447 * skb_queue_head - queue a buffer at the list head
2448 * @list: list to use
2449 * @newsk: buffer to queue
2451 * Queue a buffer at the start of the list. This function takes the
2452 * list lock and can be used safely with other locking &sk_buff functions
2455 * A buffer cannot be placed on two lists at the same time.
2457 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2459 unsigned long flags;
2461 spin_lock_irqsave(&list->lock, flags);
2462 __skb_queue_head(list, newsk);
2463 spin_unlock_irqrestore(&list->lock, flags);
2465 EXPORT_SYMBOL(skb_queue_head);
2468 * skb_queue_tail - queue a buffer at the list tail
2469 * @list: list to use
2470 * @newsk: buffer to queue
2472 * Queue a buffer at the tail of the list. This function takes the
2473 * list lock and can be used safely with other locking &sk_buff functions
2476 * A buffer cannot be placed on two lists at the same time.
2478 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2480 unsigned long flags;
2482 spin_lock_irqsave(&list->lock, flags);
2483 __skb_queue_tail(list, newsk);
2484 spin_unlock_irqrestore(&list->lock, flags);
2486 EXPORT_SYMBOL(skb_queue_tail);
2489 * skb_unlink - remove a buffer from a list
2490 * @skb: buffer to remove
2491 * @list: list to use
2493 * Remove a packet from a list. The list locks are taken and this
2494 * function is atomic with respect to other list locked calls
2496 * You must know what list the SKB is on.
2498 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2500 unsigned long flags;
2502 spin_lock_irqsave(&list->lock, flags);
2503 __skb_unlink(skb, list);
2504 spin_unlock_irqrestore(&list->lock, flags);
2506 EXPORT_SYMBOL(skb_unlink);
2509 * skb_append - append a buffer
2510 * @old: buffer to insert after
2511 * @newsk: buffer to insert
2512 * @list: list to use
2514 * Place a packet after a given packet in a list. The list locks are taken
2515 * and this function is atomic with respect to other list locked calls.
2516 * A buffer cannot be placed on two lists at the same time.
2518 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2520 unsigned long flags;
2522 spin_lock_irqsave(&list->lock, flags);
2523 __skb_queue_after(list, old, newsk);
2524 spin_unlock_irqrestore(&list->lock, flags);
2526 EXPORT_SYMBOL(skb_append);
2529 * skb_insert - insert a buffer
2530 * @old: buffer to insert before
2531 * @newsk: buffer to insert
2532 * @list: list to use
2534 * Place a packet before a given packet in a list. The list locks are
2535 * taken and this function is atomic with respect to other list locked
2538 * A buffer cannot be placed on two lists at the same time.
2540 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2542 unsigned long flags;
2544 spin_lock_irqsave(&list->lock, flags);
2545 __skb_insert(newsk, old->prev, old, list);
2546 spin_unlock_irqrestore(&list->lock, flags);
2548 EXPORT_SYMBOL(skb_insert);
2550 static inline void skb_split_inside_header(struct sk_buff *skb,
2551 struct sk_buff* skb1,
2552 const u32 len, const int pos)
2556 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2558 /* And move data appendix as is. */
2559 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2560 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2562 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2563 skb_shinfo(skb)->nr_frags = 0;
2564 skb1->data_len = skb->data_len;
2565 skb1->len += skb1->data_len;
2568 skb_set_tail_pointer(skb, len);
2571 static inline void skb_split_no_header(struct sk_buff *skb,
2572 struct sk_buff* skb1,
2573 const u32 len, int pos)
2576 const int nfrags = skb_shinfo(skb)->nr_frags;
2578 skb_shinfo(skb)->nr_frags = 0;
2579 skb1->len = skb1->data_len = skb->len - len;
2581 skb->data_len = len - pos;
2583 for (i = 0; i < nfrags; i++) {
2584 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2586 if (pos + size > len) {
2587 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2591 * We have two variants in this case:
2592 * 1. Move all the frag to the second
2593 * part, if it is possible. F.e.
2594 * this approach is mandatory for TUX,
2595 * where splitting is expensive.
2596 * 2. Split is accurately. We make this.
2598 skb_frag_ref(skb, i);
2599 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2600 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2601 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2602 skb_shinfo(skb)->nr_frags++;
2606 skb_shinfo(skb)->nr_frags++;
2609 skb_shinfo(skb1)->nr_frags = k;
2613 * skb_split - Split fragmented skb to two parts at length len.
2614 * @skb: the buffer to split
2615 * @skb1: the buffer to receive the second part
2616 * @len: new length for skb
2618 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2620 int pos = skb_headlen(skb);
2622 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2623 if (len < pos) /* Split line is inside header. */
2624 skb_split_inside_header(skb, skb1, len, pos);
2625 else /* Second chunk has no header, nothing to copy. */
2626 skb_split_no_header(skb, skb1, len, pos);
2628 EXPORT_SYMBOL(skb_split);
2630 /* Shifting from/to a cloned skb is a no-go.
2632 * Caller cannot keep skb_shinfo related pointers past calling here!
2634 static int skb_prepare_for_shift(struct sk_buff *skb)
2636 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2640 * skb_shift - Shifts paged data partially from skb to another
2641 * @tgt: buffer into which tail data gets added
2642 * @skb: buffer from which the paged data comes from
2643 * @shiftlen: shift up to this many bytes
2645 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2646 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2647 * It's up to caller to free skb if everything was shifted.
2649 * If @tgt runs out of frags, the whole operation is aborted.
2651 * Skb cannot include anything else but paged data while tgt is allowed
2652 * to have non-paged data as well.
2654 * TODO: full sized shift could be optimized but that would need
2655 * specialized skb free'er to handle frags without up-to-date nr_frags.
2657 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2659 int from, to, merge, todo;
2660 struct skb_frag_struct *fragfrom, *fragto;
2662 BUG_ON(shiftlen > skb->len);
2663 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2667 to = skb_shinfo(tgt)->nr_frags;
2668 fragfrom = &skb_shinfo(skb)->frags[from];
2670 /* Actual merge is delayed until the point when we know we can
2671 * commit all, so that we don't have to undo partial changes
2674 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2675 fragfrom->page_offset)) {
2680 todo -= skb_frag_size(fragfrom);
2682 if (skb_prepare_for_shift(skb) ||
2683 skb_prepare_for_shift(tgt))
2686 /* All previous frag pointers might be stale! */
2687 fragfrom = &skb_shinfo(skb)->frags[from];
2688 fragto = &skb_shinfo(tgt)->frags[merge];
2690 skb_frag_size_add(fragto, shiftlen);
2691 skb_frag_size_sub(fragfrom, shiftlen);
2692 fragfrom->page_offset += shiftlen;
2700 /* Skip full, not-fitting skb to avoid expensive operations */
2701 if ((shiftlen == skb->len) &&
2702 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2705 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2708 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2709 if (to == MAX_SKB_FRAGS)
2712 fragfrom = &skb_shinfo(skb)->frags[from];
2713 fragto = &skb_shinfo(tgt)->frags[to];
2715 if (todo >= skb_frag_size(fragfrom)) {
2716 *fragto = *fragfrom;
2717 todo -= skb_frag_size(fragfrom);
2722 __skb_frag_ref(fragfrom);
2723 fragto->page = fragfrom->page;
2724 fragto->page_offset = fragfrom->page_offset;
2725 skb_frag_size_set(fragto, todo);
2727 fragfrom->page_offset += todo;
2728 skb_frag_size_sub(fragfrom, todo);
2736 /* Ready to "commit" this state change to tgt */
2737 skb_shinfo(tgt)->nr_frags = to;
2740 fragfrom = &skb_shinfo(skb)->frags[0];
2741 fragto = &skb_shinfo(tgt)->frags[merge];
2743 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2744 __skb_frag_unref(fragfrom);
2747 /* Reposition in the original skb */
2749 while (from < skb_shinfo(skb)->nr_frags)
2750 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2751 skb_shinfo(skb)->nr_frags = to;
2753 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2756 /* Most likely the tgt won't ever need its checksum anymore, skb on
2757 * the other hand might need it if it needs to be resent
2759 tgt->ip_summed = CHECKSUM_PARTIAL;
2760 skb->ip_summed = CHECKSUM_PARTIAL;
2762 /* Yak, is it really working this way? Some helper please? */
2763 skb->len -= shiftlen;
2764 skb->data_len -= shiftlen;
2765 skb->truesize -= shiftlen;
2766 tgt->len += shiftlen;
2767 tgt->data_len += shiftlen;
2768 tgt->truesize += shiftlen;
2774 * skb_prepare_seq_read - Prepare a sequential read of skb data
2775 * @skb: the buffer to read
2776 * @from: lower offset of data to be read
2777 * @to: upper offset of data to be read
2778 * @st: state variable
2780 * Initializes the specified state variable. Must be called before
2781 * invoking skb_seq_read() for the first time.
2783 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2784 unsigned int to, struct skb_seq_state *st)
2786 st->lower_offset = from;
2787 st->upper_offset = to;
2788 st->root_skb = st->cur_skb = skb;
2789 st->frag_idx = st->stepped_offset = 0;
2790 st->frag_data = NULL;
2792 EXPORT_SYMBOL(skb_prepare_seq_read);
2795 * skb_seq_read - Sequentially read skb data
2796 * @consumed: number of bytes consumed by the caller so far
2797 * @data: destination pointer for data to be returned
2798 * @st: state variable
2800 * Reads a block of skb data at @consumed relative to the
2801 * lower offset specified to skb_prepare_seq_read(). Assigns
2802 * the head of the data block to @data and returns the length
2803 * of the block or 0 if the end of the skb data or the upper
2804 * offset has been reached.
2806 * The caller is not required to consume all of the data
2807 * returned, i.e. @consumed is typically set to the number
2808 * of bytes already consumed and the next call to
2809 * skb_seq_read() will return the remaining part of the block.
2811 * Note 1: The size of each block of data returned can be arbitrary,
2812 * this limitation is the cost for zerocopy sequential
2813 * reads of potentially non linear data.
2815 * Note 2: Fragment lists within fragments are not implemented
2816 * at the moment, state->root_skb could be replaced with
2817 * a stack for this purpose.
2819 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2820 struct skb_seq_state *st)
2822 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2825 if (unlikely(abs_offset >= st->upper_offset)) {
2826 if (st->frag_data) {
2827 kunmap_atomic(st->frag_data);
2828 st->frag_data = NULL;
2834 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2836 if (abs_offset < block_limit && !st->frag_data) {
2837 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2838 return block_limit - abs_offset;
2841 if (st->frag_idx == 0 && !st->frag_data)
2842 st->stepped_offset += skb_headlen(st->cur_skb);
2844 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2845 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2846 block_limit = skb_frag_size(frag) + st->stepped_offset;
2848 if (abs_offset < block_limit) {
2850 st->frag_data = kmap_atomic(skb_frag_page(frag));
2852 *data = (u8 *) st->frag_data + frag->page_offset +
2853 (abs_offset - st->stepped_offset);
2855 return block_limit - abs_offset;
2858 if (st->frag_data) {
2859 kunmap_atomic(st->frag_data);
2860 st->frag_data = NULL;
2864 st->stepped_offset += skb_frag_size(frag);
2867 if (st->frag_data) {
2868 kunmap_atomic(st->frag_data);
2869 st->frag_data = NULL;
2872 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2873 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2876 } else if (st->cur_skb->next) {
2877 st->cur_skb = st->cur_skb->next;
2884 EXPORT_SYMBOL(skb_seq_read);
2887 * skb_abort_seq_read - Abort a sequential read of skb data
2888 * @st: state variable
2890 * Must be called if skb_seq_read() was not called until it
2893 void skb_abort_seq_read(struct skb_seq_state *st)
2896 kunmap_atomic(st->frag_data);
2898 EXPORT_SYMBOL(skb_abort_seq_read);
2900 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2902 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2903 struct ts_config *conf,
2904 struct ts_state *state)
2906 return skb_seq_read(offset, text, TS_SKB_CB(state));
2909 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2911 skb_abort_seq_read(TS_SKB_CB(state));
2915 * skb_find_text - Find a text pattern in skb data
2916 * @skb: the buffer to look in
2917 * @from: search offset
2919 * @config: textsearch configuration
2921 * Finds a pattern in the skb data according to the specified
2922 * textsearch configuration. Use textsearch_next() to retrieve
2923 * subsequent occurrences of the pattern. Returns the offset
2924 * to the first occurrence or UINT_MAX if no match was found.
2926 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2927 unsigned int to, struct ts_config *config)
2929 struct ts_state state;
2932 config->get_next_block = skb_ts_get_next_block;
2933 config->finish = skb_ts_finish;
2935 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
2937 ret = textsearch_find(config, &state);
2938 return (ret <= to - from ? ret : UINT_MAX);
2940 EXPORT_SYMBOL(skb_find_text);
2943 * skb_append_datato_frags - append the user data to a skb
2944 * @sk: sock structure
2945 * @skb: skb structure to be appended with user data.
2946 * @getfrag: call back function to be used for getting the user data
2947 * @from: pointer to user message iov
2948 * @length: length of the iov message
2950 * Description: This procedure append the user data in the fragment part
2951 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2953 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2954 int (*getfrag)(void *from, char *to, int offset,
2955 int len, int odd, struct sk_buff *skb),
2956 void *from, int length)
2958 int frg_cnt = skb_shinfo(skb)->nr_frags;
2962 struct page_frag *pfrag = ¤t->task_frag;
2965 /* Return error if we don't have space for new frag */
2966 if (frg_cnt >= MAX_SKB_FRAGS)
2969 if (!sk_page_frag_refill(sk, pfrag))
2972 /* copy the user data to page */
2973 copy = min_t(int, length, pfrag->size - pfrag->offset);
2975 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2976 offset, copy, 0, skb);
2980 /* copy was successful so update the size parameters */
2981 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2984 pfrag->offset += copy;
2985 get_page(pfrag->page);
2987 skb->truesize += copy;
2988 atomic_add(copy, &sk->sk_wmem_alloc);
2990 skb->data_len += copy;
2994 } while (length > 0);
2998 EXPORT_SYMBOL(skb_append_datato_frags);
3000 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3001 int offset, size_t size)
3003 int i = skb_shinfo(skb)->nr_frags;
3005 if (skb_can_coalesce(skb, i, page, offset)) {
3006 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3007 } else if (i < MAX_SKB_FRAGS) {
3009 skb_fill_page_desc(skb, i, page, offset, size);
3016 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3019 * skb_pull_rcsum - pull skb and update receive checksum
3020 * @skb: buffer to update
3021 * @len: length of data pulled
3023 * This function performs an skb_pull on the packet and updates
3024 * the CHECKSUM_COMPLETE checksum. It should be used on
3025 * receive path processing instead of skb_pull unless you know
3026 * that the checksum difference is zero (e.g., a valid IP header)
3027 * or you are setting ip_summed to CHECKSUM_NONE.
3029 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3031 unsigned char *data = skb->data;
3033 BUG_ON(len > skb->len);
3034 __skb_pull(skb, len);
3035 skb_postpull_rcsum(skb, data, len);
3038 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3041 * skb_segment - Perform protocol segmentation on skb.
3042 * @head_skb: buffer to segment
3043 * @features: features for the output path (see dev->features)
3045 * This function performs segmentation on the given skb. It returns
3046 * a pointer to the first in a list of new skbs for the segments.
3047 * In case of error it returns ERR_PTR(err).
3049 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3050 netdev_features_t features)
3052 struct sk_buff *segs = NULL;
3053 struct sk_buff *tail = NULL;
3054 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3055 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3056 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3057 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3058 struct sk_buff *frag_skb = head_skb;
3059 unsigned int offset = doffset;
3060 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3061 unsigned int partial_segs = 0;
3062 unsigned int headroom;
3063 unsigned int len = head_skb->len;
3066 int nfrags = skb_shinfo(head_skb)->nr_frags;
3072 __skb_push(head_skb, doffset);
3073 proto = skb_network_protocol(head_skb, &dummy);
3074 if (unlikely(!proto))
3075 return ERR_PTR(-EINVAL);
3077 sg = !!(features & NETIF_F_SG);
3078 csum = !!can_checksum_protocol(features, proto);
3080 /* GSO partial only requires that we trim off any excess that
3081 * doesn't fit into an MSS sized block, so take care of that
3084 if (sg && csum && (features & NETIF_F_GSO_PARTIAL)) {
3085 partial_segs = len / mss;
3086 if (partial_segs > 1)
3087 mss *= partial_segs;
3092 headroom = skb_headroom(head_skb);
3093 pos = skb_headlen(head_skb);
3096 struct sk_buff *nskb;
3097 skb_frag_t *nskb_frag;
3101 len = head_skb->len - offset;
3105 hsize = skb_headlen(head_skb) - offset;
3108 if (hsize > len || !sg)
3111 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3112 (skb_headlen(list_skb) == len || sg)) {
3113 BUG_ON(skb_headlen(list_skb) > len);
3116 nfrags = skb_shinfo(list_skb)->nr_frags;
3117 frag = skb_shinfo(list_skb)->frags;
3118 frag_skb = list_skb;
3119 pos += skb_headlen(list_skb);
3121 while (pos < offset + len) {
3122 BUG_ON(i >= nfrags);
3124 size = skb_frag_size(frag);
3125 if (pos + size > offset + len)
3133 nskb = skb_clone(list_skb, GFP_ATOMIC);
3134 list_skb = list_skb->next;
3136 if (unlikely(!nskb))
3139 if (unlikely(pskb_trim(nskb, len))) {
3144 hsize = skb_end_offset(nskb);
3145 if (skb_cow_head(nskb, doffset + headroom)) {
3150 nskb->truesize += skb_end_offset(nskb) - hsize;
3151 skb_release_head_state(nskb);
3152 __skb_push(nskb, doffset);
3154 nskb = __alloc_skb(hsize + doffset + headroom,
3155 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3158 if (unlikely(!nskb))
3161 skb_reserve(nskb, headroom);
3162 __skb_put(nskb, doffset);
3171 __copy_skb_header(nskb, head_skb);
3173 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3174 skb_reset_mac_len(nskb);
3176 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3177 nskb->data - tnl_hlen,
3178 doffset + tnl_hlen);
3180 if (nskb->len == len + doffset)
3181 goto perform_csum_check;
3184 if (!nskb->remcsum_offload)
3185 nskb->ip_summed = CHECKSUM_NONE;
3186 SKB_GSO_CB(nskb)->csum =
3187 skb_copy_and_csum_bits(head_skb, offset,
3190 SKB_GSO_CB(nskb)->csum_start =
3191 skb_headroom(nskb) + doffset;
3195 nskb_frag = skb_shinfo(nskb)->frags;
3197 skb_copy_from_linear_data_offset(head_skb, offset,
3198 skb_put(nskb, hsize), hsize);
3200 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3203 while (pos < offset + len) {
3205 BUG_ON(skb_headlen(list_skb));
3208 nfrags = skb_shinfo(list_skb)->nr_frags;
3209 frag = skb_shinfo(list_skb)->frags;
3210 frag_skb = list_skb;
3214 list_skb = list_skb->next;
3217 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3219 net_warn_ratelimited(
3220 "skb_segment: too many frags: %u %u\n",
3225 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3229 __skb_frag_ref(nskb_frag);
3230 size = skb_frag_size(nskb_frag);
3233 nskb_frag->page_offset += offset - pos;
3234 skb_frag_size_sub(nskb_frag, offset - pos);
3237 skb_shinfo(nskb)->nr_frags++;
3239 if (pos + size <= offset + len) {
3244 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3252 nskb->data_len = len - hsize;
3253 nskb->len += nskb->data_len;
3254 nskb->truesize += nskb->data_len;
3258 if (skb_has_shared_frag(nskb)) {
3259 err = __skb_linearize(nskb);
3263 if (!nskb->remcsum_offload)
3264 nskb->ip_summed = CHECKSUM_NONE;
3265 SKB_GSO_CB(nskb)->csum =
3266 skb_checksum(nskb, doffset,
3267 nskb->len - doffset, 0);
3268 SKB_GSO_CB(nskb)->csum_start =
3269 skb_headroom(nskb) + doffset;
3271 } while ((offset += len) < head_skb->len);
3273 /* Some callers want to get the end of the list.
3274 * Put it in segs->prev to avoid walking the list.
3275 * (see validate_xmit_skb_list() for example)
3279 /* Update GSO info on first skb in partial sequence. */
3281 int type = skb_shinfo(head_skb)->gso_type;
3283 /* Update type to add partial and then remove dodgy if set */
3284 type |= SKB_GSO_PARTIAL;
3285 type &= ~SKB_GSO_DODGY;
3287 /* Update GSO info and prepare to start updating headers on
3288 * our way back down the stack of protocols.
3290 skb_shinfo(segs)->gso_size = skb_shinfo(head_skb)->gso_size;
3291 skb_shinfo(segs)->gso_segs = partial_segs;
3292 skb_shinfo(segs)->gso_type = type;
3293 SKB_GSO_CB(segs)->data_offset = skb_headroom(segs) + doffset;
3296 /* Following permits correct backpressure, for protocols
3297 * using skb_set_owner_w().
3298 * Idea is to tranfert ownership from head_skb to last segment.
3300 if (head_skb->destructor == sock_wfree) {
3301 swap(tail->truesize, head_skb->truesize);
3302 swap(tail->destructor, head_skb->destructor);
3303 swap(tail->sk, head_skb->sk);
3308 kfree_skb_list(segs);
3309 return ERR_PTR(err);
3311 EXPORT_SYMBOL_GPL(skb_segment);
3313 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3315 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3316 unsigned int offset = skb_gro_offset(skb);
3317 unsigned int headlen = skb_headlen(skb);
3318 unsigned int len = skb_gro_len(skb);
3319 struct sk_buff *lp, *p = *head;
3320 unsigned int delta_truesize;
3322 if (unlikely(p->len + len >= 65536))
3325 lp = NAPI_GRO_CB(p)->last;
3326 pinfo = skb_shinfo(lp);
3328 if (headlen <= offset) {
3331 int i = skbinfo->nr_frags;
3332 int nr_frags = pinfo->nr_frags + i;
3334 if (nr_frags > MAX_SKB_FRAGS)
3338 pinfo->nr_frags = nr_frags;
3339 skbinfo->nr_frags = 0;
3341 frag = pinfo->frags + nr_frags;
3342 frag2 = skbinfo->frags + i;
3347 frag->page_offset += offset;
3348 skb_frag_size_sub(frag, offset);
3350 /* all fragments truesize : remove (head size + sk_buff) */
3351 delta_truesize = skb->truesize -
3352 SKB_TRUESIZE(skb_end_offset(skb));
3354 skb->truesize -= skb->data_len;
3355 skb->len -= skb->data_len;
3358 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3360 } else if (skb->head_frag) {
3361 int nr_frags = pinfo->nr_frags;
3362 skb_frag_t *frag = pinfo->frags + nr_frags;
3363 struct page *page = virt_to_head_page(skb->head);
3364 unsigned int first_size = headlen - offset;
3365 unsigned int first_offset;
3367 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3370 first_offset = skb->data -
3371 (unsigned char *)page_address(page) +
3374 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3376 frag->page.p = page;
3377 frag->page_offset = first_offset;
3378 skb_frag_size_set(frag, first_size);
3380 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3381 /* We dont need to clear skbinfo->nr_frags here */
3383 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3384 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3389 delta_truesize = skb->truesize;
3390 if (offset > headlen) {
3391 unsigned int eat = offset - headlen;
3393 skbinfo->frags[0].page_offset += eat;
3394 skb_frag_size_sub(&skbinfo->frags[0], eat);
3395 skb->data_len -= eat;
3400 __skb_pull(skb, offset);
3402 if (NAPI_GRO_CB(p)->last == p)
3403 skb_shinfo(p)->frag_list = skb;
3405 NAPI_GRO_CB(p)->last->next = skb;
3406 NAPI_GRO_CB(p)->last = skb;
3407 __skb_header_release(skb);
3411 NAPI_GRO_CB(p)->count++;
3413 p->truesize += delta_truesize;
3416 lp->data_len += len;
3417 lp->truesize += delta_truesize;
3420 NAPI_GRO_CB(skb)->same_flow = 1;
3424 void __init skb_init(void)
3426 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3427 sizeof(struct sk_buff),
3429 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3431 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3432 sizeof(struct sk_buff_fclones),
3434 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3439 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3440 * @skb: Socket buffer containing the buffers to be mapped
3441 * @sg: The scatter-gather list to map into
3442 * @offset: The offset into the buffer's contents to start mapping
3443 * @len: Length of buffer space to be mapped
3445 * Fill the specified scatter-gather list with mappings/pointers into a
3446 * region of the buffer space attached to a socket buffer.
3449 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3451 int start = skb_headlen(skb);
3452 int i, copy = start - offset;
3453 struct sk_buff *frag_iter;
3459 sg_set_buf(sg, skb->data + offset, copy);
3461 if ((len -= copy) == 0)
3466 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3469 WARN_ON(start > offset + len);
3471 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3472 if ((copy = end - offset) > 0) {
3473 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3477 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3478 frag->page_offset+offset-start);
3487 skb_walk_frags(skb, frag_iter) {
3490 WARN_ON(start > offset + len);
3492 end = start + frag_iter->len;
3493 if ((copy = end - offset) > 0) {
3496 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3498 if ((len -= copy) == 0)
3508 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3509 * sglist without mark the sg which contain last skb data as the end.
3510 * So the caller can mannipulate sg list as will when padding new data after
3511 * the first call without calling sg_unmark_end to expend sg list.
3513 * Scenario to use skb_to_sgvec_nomark:
3515 * 2. skb_to_sgvec_nomark(payload1)
3516 * 3. skb_to_sgvec_nomark(payload2)
3518 * This is equivalent to:
3520 * 2. skb_to_sgvec(payload1)
3522 * 4. skb_to_sgvec(payload2)
3524 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3525 * is more preferable.
3527 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3528 int offset, int len)
3530 return __skb_to_sgvec(skb, sg, offset, len);
3532 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3534 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3536 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3538 sg_mark_end(&sg[nsg - 1]);
3542 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3545 * skb_cow_data - Check that a socket buffer's data buffers are writable
3546 * @skb: The socket buffer to check.
3547 * @tailbits: Amount of trailing space to be added
3548 * @trailer: Returned pointer to the skb where the @tailbits space begins
3550 * Make sure that the data buffers attached to a socket buffer are
3551 * writable. If they are not, private copies are made of the data buffers
3552 * and the socket buffer is set to use these instead.
3554 * If @tailbits is given, make sure that there is space to write @tailbits
3555 * bytes of data beyond current end of socket buffer. @trailer will be
3556 * set to point to the skb in which this space begins.
3558 * The number of scatterlist elements required to completely map the
3559 * COW'd and extended socket buffer will be returned.
3561 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3565 struct sk_buff *skb1, **skb_p;
3567 /* If skb is cloned or its head is paged, reallocate
3568 * head pulling out all the pages (pages are considered not writable
3569 * at the moment even if they are anonymous).
3571 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3572 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3575 /* Easy case. Most of packets will go this way. */
3576 if (!skb_has_frag_list(skb)) {
3577 /* A little of trouble, not enough of space for trailer.
3578 * This should not happen, when stack is tuned to generate
3579 * good frames. OK, on miss we reallocate and reserve even more
3580 * space, 128 bytes is fair. */
3582 if (skb_tailroom(skb) < tailbits &&
3583 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3591 /* Misery. We are in troubles, going to mincer fragments... */
3594 skb_p = &skb_shinfo(skb)->frag_list;
3597 while ((skb1 = *skb_p) != NULL) {
3600 /* The fragment is partially pulled by someone,
3601 * this can happen on input. Copy it and everything
3604 if (skb_shared(skb1))
3607 /* If the skb is the last, worry about trailer. */
3609 if (skb1->next == NULL && tailbits) {
3610 if (skb_shinfo(skb1)->nr_frags ||
3611 skb_has_frag_list(skb1) ||
3612 skb_tailroom(skb1) < tailbits)
3613 ntail = tailbits + 128;
3619 skb_shinfo(skb1)->nr_frags ||
3620 skb_has_frag_list(skb1)) {
3621 struct sk_buff *skb2;
3623 /* Fuck, we are miserable poor guys... */
3625 skb2 = skb_copy(skb1, GFP_ATOMIC);
3627 skb2 = skb_copy_expand(skb1,
3631 if (unlikely(skb2 == NULL))
3635 skb_set_owner_w(skb2, skb1->sk);
3637 /* Looking around. Are we still alive?
3638 * OK, link new skb, drop old one */
3640 skb2->next = skb1->next;
3647 skb_p = &skb1->next;
3652 EXPORT_SYMBOL_GPL(skb_cow_data);
3654 static void sock_rmem_free(struct sk_buff *skb)
3656 struct sock *sk = skb->sk;
3658 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3662 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3664 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3666 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3667 (unsigned int)sk->sk_rcvbuf)
3672 skb->destructor = sock_rmem_free;
3673 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3675 /* before exiting rcu section, make sure dst is refcounted */
3678 skb_queue_tail(&sk->sk_error_queue, skb);
3679 if (!sock_flag(sk, SOCK_DEAD))
3680 sk->sk_data_ready(sk);
3683 EXPORT_SYMBOL(sock_queue_err_skb);
3685 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3687 struct sk_buff_head *q = &sk->sk_error_queue;
3688 struct sk_buff *skb, *skb_next;
3689 unsigned long flags;
3692 spin_lock_irqsave(&q->lock, flags);
3693 skb = __skb_dequeue(q);
3694 if (skb && (skb_next = skb_peek(q)))
3695 err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
3696 spin_unlock_irqrestore(&q->lock, flags);
3700 sk->sk_error_report(sk);
3704 EXPORT_SYMBOL(sock_dequeue_err_skb);
3707 * skb_clone_sk - create clone of skb, and take reference to socket
3708 * @skb: the skb to clone
3710 * This function creates a clone of a buffer that holds a reference on
3711 * sk_refcnt. Buffers created via this function are meant to be
3712 * returned using sock_queue_err_skb, or free via kfree_skb.
3714 * When passing buffers allocated with this function to sock_queue_err_skb
3715 * it is necessary to wrap the call with sock_hold/sock_put in order to
3716 * prevent the socket from being released prior to being enqueued on
3717 * the sk_error_queue.
3719 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3721 struct sock *sk = skb->sk;
3722 struct sk_buff *clone;
3724 if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3727 clone = skb_clone(skb, GFP_ATOMIC);
3734 clone->destructor = sock_efree;
3738 EXPORT_SYMBOL(skb_clone_sk);
3740 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3744 struct sock_exterr_skb *serr;
3747 serr = SKB_EXT_ERR(skb);
3748 memset(serr, 0, sizeof(*serr));
3749 serr->ee.ee_errno = ENOMSG;
3750 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3751 serr->ee.ee_info = tstype;
3752 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3753 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3754 if (sk->sk_protocol == IPPROTO_TCP &&
3755 sk->sk_type == SOCK_STREAM)
3756 serr->ee.ee_data -= sk->sk_tskey;
3759 err = sock_queue_err_skb(sk, skb);
3765 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
3769 if (likely(sysctl_tstamp_allow_data || tsonly))
3772 read_lock_bh(&sk->sk_callback_lock);
3773 ret = sk->sk_socket && sk->sk_socket->file &&
3774 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
3775 read_unlock_bh(&sk->sk_callback_lock);
3779 void skb_complete_tx_timestamp(struct sk_buff *skb,
3780 struct skb_shared_hwtstamps *hwtstamps)
3782 struct sock *sk = skb->sk;
3784 if (!skb_may_tx_timestamp(sk, false))
3787 /* take a reference to prevent skb_orphan() from freeing the socket */
3790 *skb_hwtstamps(skb) = *hwtstamps;
3791 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
3795 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3797 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3798 struct skb_shared_hwtstamps *hwtstamps,
3799 struct sock *sk, int tstype)
3801 struct sk_buff *skb;
3807 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
3808 if (!skb_may_tx_timestamp(sk, tsonly))
3812 skb = alloc_skb(0, GFP_ATOMIC);
3814 skb = skb_clone(orig_skb, GFP_ATOMIC);
3819 skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags;
3820 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
3824 *skb_hwtstamps(skb) = *hwtstamps;
3826 skb->tstamp = ktime_get_real();
3828 __skb_complete_tx_timestamp(skb, sk, tstype);
3830 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3832 void skb_tstamp_tx(struct sk_buff *orig_skb,
3833 struct skb_shared_hwtstamps *hwtstamps)
3835 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3838 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3840 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3842 struct sock *sk = skb->sk;
3843 struct sock_exterr_skb *serr;
3846 skb->wifi_acked_valid = 1;
3847 skb->wifi_acked = acked;
3849 serr = SKB_EXT_ERR(skb);
3850 memset(serr, 0, sizeof(*serr));
3851 serr->ee.ee_errno = ENOMSG;
3852 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3854 /* take a reference to prevent skb_orphan() from freeing the socket */
3857 err = sock_queue_err_skb(sk, skb);
3863 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3866 * skb_partial_csum_set - set up and verify partial csum values for packet
3867 * @skb: the skb to set
3868 * @start: the number of bytes after skb->data to start checksumming.
3869 * @off: the offset from start to place the checksum.
3871 * For untrusted partially-checksummed packets, we need to make sure the values
3872 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3874 * This function checks and sets those values and skb->ip_summed: if this
3875 * returns false you should drop the packet.
3877 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3879 if (unlikely(start > skb_headlen(skb)) ||
3880 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3881 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3882 start, off, skb_headlen(skb));
3885 skb->ip_summed = CHECKSUM_PARTIAL;
3886 skb->csum_start = skb_headroom(skb) + start;
3887 skb->csum_offset = off;
3888 skb_set_transport_header(skb, start);
3891 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3893 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3896 if (skb_headlen(skb) >= len)
3899 /* If we need to pullup then pullup to the max, so we
3900 * won't need to do it again.
3905 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3908 if (skb_headlen(skb) < len)
3914 #define MAX_TCP_HDR_LEN (15 * 4)
3916 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3917 typeof(IPPROTO_IP) proto,
3924 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3925 off + MAX_TCP_HDR_LEN);
3926 if (!err && !skb_partial_csum_set(skb, off,
3927 offsetof(struct tcphdr,
3930 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3933 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3934 off + sizeof(struct udphdr));
3935 if (!err && !skb_partial_csum_set(skb, off,
3936 offsetof(struct udphdr,
3939 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3942 return ERR_PTR(-EPROTO);
3945 /* This value should be large enough to cover a tagged ethernet header plus
3946 * maximally sized IP and TCP or UDP headers.
3948 #define MAX_IP_HDR_LEN 128
3950 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
3959 err = skb_maybe_pull_tail(skb,
3960 sizeof(struct iphdr),
3965 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3968 off = ip_hdrlen(skb);
3975 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
3977 return PTR_ERR(csum);
3980 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3983 ip_hdr(skb)->protocol, 0);
3990 /* This value should be large enough to cover a tagged ethernet header plus
3991 * an IPv6 header, all options, and a maximal TCP or UDP header.
3993 #define MAX_IPV6_HDR_LEN 256
3995 #define OPT_HDR(type, skb, off) \
3996 (type *)(skb_network_header(skb) + (off))
3998 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4011 off = sizeof(struct ipv6hdr);
4013 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4017 nexthdr = ipv6_hdr(skb)->nexthdr;
4019 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4020 while (off <= len && !done) {
4022 case IPPROTO_DSTOPTS:
4023 case IPPROTO_HOPOPTS:
4024 case IPPROTO_ROUTING: {
4025 struct ipv6_opt_hdr *hp;
4027 err = skb_maybe_pull_tail(skb,
4029 sizeof(struct ipv6_opt_hdr),
4034 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4035 nexthdr = hp->nexthdr;
4036 off += ipv6_optlen(hp);
4040 struct ip_auth_hdr *hp;
4042 err = skb_maybe_pull_tail(skb,
4044 sizeof(struct ip_auth_hdr),
4049 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4050 nexthdr = hp->nexthdr;
4051 off += ipv6_authlen(hp);
4054 case IPPROTO_FRAGMENT: {
4055 struct frag_hdr *hp;
4057 err = skb_maybe_pull_tail(skb,
4059 sizeof(struct frag_hdr),
4064 hp = OPT_HDR(struct frag_hdr, skb, off);
4066 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4069 nexthdr = hp->nexthdr;
4070 off += sizeof(struct frag_hdr);
4081 if (!done || fragment)
4084 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4086 return PTR_ERR(csum);
4089 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4090 &ipv6_hdr(skb)->daddr,
4091 skb->len - off, nexthdr, 0);
4099 * skb_checksum_setup - set up partial checksum offset
4100 * @skb: the skb to set up
4101 * @recalculate: if true the pseudo-header checksum will be recalculated
4103 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4107 switch (skb->protocol) {
4108 case htons(ETH_P_IP):
4109 err = skb_checksum_setup_ipv4(skb, recalculate);
4112 case htons(ETH_P_IPV6):
4113 err = skb_checksum_setup_ipv6(skb, recalculate);
4123 EXPORT_SYMBOL(skb_checksum_setup);
4126 * skb_checksum_maybe_trim - maybe trims the given skb
4127 * @skb: the skb to check
4128 * @transport_len: the data length beyond the network header
4130 * Checks whether the given skb has data beyond the given transport length.
4131 * If so, returns a cloned skb trimmed to this transport length.
4132 * Otherwise returns the provided skb. Returns NULL in error cases
4133 * (e.g. transport_len exceeds skb length or out-of-memory).
4135 * Caller needs to set the skb transport header and free any returned skb if it
4136 * differs from the provided skb.
4138 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4139 unsigned int transport_len)
4141 struct sk_buff *skb_chk;
4142 unsigned int len = skb_transport_offset(skb) + transport_len;
4147 else if (skb->len == len)
4150 skb_chk = skb_clone(skb, GFP_ATOMIC);
4154 ret = pskb_trim_rcsum(skb_chk, len);
4164 * skb_checksum_trimmed - validate checksum of an skb
4165 * @skb: the skb to check
4166 * @transport_len: the data length beyond the network header
4167 * @skb_chkf: checksum function to use
4169 * Applies the given checksum function skb_chkf to the provided skb.
4170 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4172 * If the skb has data beyond the given transport length, then a
4173 * trimmed & cloned skb is checked and returned.
4175 * Caller needs to set the skb transport header and free any returned skb if it
4176 * differs from the provided skb.
4178 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4179 unsigned int transport_len,
4180 __sum16(*skb_chkf)(struct sk_buff *skb))
4182 struct sk_buff *skb_chk;
4183 unsigned int offset = skb_transport_offset(skb);
4186 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4190 if (!pskb_may_pull(skb_chk, offset))
4193 skb_pull_rcsum(skb_chk, offset);
4194 ret = skb_chkf(skb_chk);
4195 skb_push_rcsum(skb_chk, offset);
4203 if (skb_chk && skb_chk != skb)
4209 EXPORT_SYMBOL(skb_checksum_trimmed);
4211 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4213 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4216 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4218 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4221 skb_release_head_state(skb);
4222 kmem_cache_free(skbuff_head_cache, skb);
4227 EXPORT_SYMBOL(kfree_skb_partial);
4230 * skb_try_coalesce - try to merge skb to prior one
4232 * @from: buffer to add
4233 * @fragstolen: pointer to boolean
4234 * @delta_truesize: how much more was allocated than was requested
4236 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4237 bool *fragstolen, int *delta_truesize)
4239 int i, delta, len = from->len;
4241 *fragstolen = false;
4246 if (len <= skb_tailroom(to)) {
4248 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4249 *delta_truesize = 0;
4253 if (skb_has_frag_list(to) || skb_has_frag_list(from))
4256 if (skb_headlen(from) != 0) {
4258 unsigned int offset;
4260 if (skb_shinfo(to)->nr_frags +
4261 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4264 if (skb_head_is_locked(from))
4267 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4269 page = virt_to_head_page(from->head);
4270 offset = from->data - (unsigned char *)page_address(page);
4272 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4273 page, offset, skb_headlen(from));
4276 if (skb_shinfo(to)->nr_frags +
4277 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4280 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4283 WARN_ON_ONCE(delta < len);
4285 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4286 skb_shinfo(from)->frags,
4287 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4288 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4290 if (!skb_cloned(from))
4291 skb_shinfo(from)->nr_frags = 0;
4293 /* if the skb is not cloned this does nothing
4294 * since we set nr_frags to 0.
4296 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4297 skb_frag_ref(from, i);
4299 to->truesize += delta;
4301 to->data_len += len;
4303 *delta_truesize = delta;
4306 EXPORT_SYMBOL(skb_try_coalesce);
4309 * skb_scrub_packet - scrub an skb
4311 * @skb: buffer to clean
4312 * @xnet: packet is crossing netns
4314 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4315 * into/from a tunnel. Some information have to be cleared during these
4317 * skb_scrub_packet can also be used to clean a skb before injecting it in
4318 * another namespace (@xnet == true). We have to clear all information in the
4319 * skb that could impact namespace isolation.
4321 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4323 skb->tstamp.tv64 = 0;
4324 skb->pkt_type = PACKET_HOST;
4330 nf_reset_trace(skb);
4338 EXPORT_SYMBOL_GPL(skb_scrub_packet);
4341 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4345 * skb_gso_transport_seglen is used to determine the real size of the
4346 * individual segments, including Layer4 headers (TCP/UDP).
4348 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4350 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4352 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4353 unsigned int thlen = 0;
4355 if (skb->encapsulation) {
4356 thlen = skb_inner_transport_header(skb) -
4357 skb_transport_header(skb);
4359 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4360 thlen += inner_tcp_hdrlen(skb);
4361 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4362 thlen = tcp_hdrlen(skb);
4364 /* UFO sets gso_size to the size of the fragmentation
4365 * payload, i.e. the size of the L4 (UDP) header is already
4368 return thlen + shinfo->gso_size;
4370 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4372 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4374 if (skb_cow(skb, skb_headroom(skb)) < 0) {
4379 memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN,
4381 skb->mac_header += VLAN_HLEN;
4385 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4387 struct vlan_hdr *vhdr;
4390 if (unlikely(skb_vlan_tag_present(skb))) {
4391 /* vlan_tci is already set-up so leave this for another time */
4395 skb = skb_share_check(skb, GFP_ATOMIC);
4399 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4402 vhdr = (struct vlan_hdr *)skb->data;
4403 vlan_tci = ntohs(vhdr->h_vlan_TCI);
4404 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4406 skb_pull_rcsum(skb, VLAN_HLEN);
4407 vlan_set_encap_proto(skb, vhdr);
4409 skb = skb_reorder_vlan_header(skb);
4413 skb_reset_network_header(skb);
4414 skb_reset_transport_header(skb);
4415 skb_reset_mac_len(skb);
4423 EXPORT_SYMBOL(skb_vlan_untag);
4425 int skb_ensure_writable(struct sk_buff *skb, int write_len)
4427 if (!pskb_may_pull(skb, write_len))
4430 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4433 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4435 EXPORT_SYMBOL(skb_ensure_writable);
4437 /* remove VLAN header from packet and update csum accordingly. */
4438 static int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
4440 struct vlan_hdr *vhdr;
4441 unsigned int offset = skb->data - skb_mac_header(skb);
4444 __skb_push(skb, offset);
4445 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
4449 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4451 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
4452 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
4454 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
4455 __skb_pull(skb, VLAN_HLEN);
4457 vlan_set_encap_proto(skb, vhdr);
4458 skb->mac_header += VLAN_HLEN;
4460 if (skb_network_offset(skb) < ETH_HLEN)
4461 skb_set_network_header(skb, ETH_HLEN);
4463 skb_reset_mac_len(skb);
4465 __skb_pull(skb, offset);
4470 int skb_vlan_pop(struct sk_buff *skb)
4476 if (likely(skb_vlan_tag_present(skb))) {
4479 if (unlikely((skb->protocol != htons(ETH_P_8021Q) &&
4480 skb->protocol != htons(ETH_P_8021AD)) ||
4481 skb->len < VLAN_ETH_HLEN))
4484 err = __skb_vlan_pop(skb, &vlan_tci);
4488 /* move next vlan tag to hw accel tag */
4489 if (likely((skb->protocol != htons(ETH_P_8021Q) &&
4490 skb->protocol != htons(ETH_P_8021AD)) ||
4491 skb->len < VLAN_ETH_HLEN))
4494 vlan_proto = skb->protocol;
4495 err = __skb_vlan_pop(skb, &vlan_tci);
4499 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4502 EXPORT_SYMBOL(skb_vlan_pop);
4504 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
4506 if (skb_vlan_tag_present(skb)) {
4507 unsigned int offset = skb->data - skb_mac_header(skb);
4510 /* __vlan_insert_tag expect skb->data pointing to mac header.
4511 * So change skb->data before calling it and change back to
4512 * original position later
4514 __skb_push(skb, offset);
4515 err = __vlan_insert_tag(skb, skb->vlan_proto,
4516 skb_vlan_tag_get(skb));
4518 __skb_pull(skb, offset);
4522 skb->protocol = skb->vlan_proto;
4523 skb->mac_len += VLAN_HLEN;
4525 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4526 __skb_pull(skb, offset);
4528 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4531 EXPORT_SYMBOL(skb_vlan_push);
4534 * alloc_skb_with_frags - allocate skb with page frags
4536 * @header_len: size of linear part
4537 * @data_len: needed length in frags
4538 * @max_page_order: max page order desired.
4539 * @errcode: pointer to error code if any
4540 * @gfp_mask: allocation mask
4542 * This can be used to allocate a paged skb, given a maximal order for frags.
4544 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4545 unsigned long data_len,
4550 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4551 unsigned long chunk;
4552 struct sk_buff *skb;
4557 *errcode = -EMSGSIZE;
4558 /* Note this test could be relaxed, if we succeed to allocate
4559 * high order pages...
4561 if (npages > MAX_SKB_FRAGS)
4564 gfp_head = gfp_mask;
4565 if (gfp_head & __GFP_DIRECT_RECLAIM)
4566 gfp_head |= __GFP_REPEAT;
4568 *errcode = -ENOBUFS;
4569 skb = alloc_skb(header_len, gfp_head);
4573 skb->truesize += npages << PAGE_SHIFT;
4575 for (i = 0; npages > 0; i++) {
4576 int order = max_page_order;
4579 if (npages >= 1 << order) {
4580 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
4587 /* Do not retry other high order allocations */
4593 page = alloc_page(gfp_mask);
4597 chunk = min_t(unsigned long, data_len,
4598 PAGE_SIZE << order);
4599 skb_fill_page_desc(skb, i, page, 0, chunk);
4601 npages -= 1 << order;
4609 EXPORT_SYMBOL(alloc_skb_with_frags);
4611 /* carve out the first off bytes from skb when off < headlen */
4612 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
4613 const int headlen, gfp_t gfp_mask)
4616 int size = skb_end_offset(skb);
4617 int new_hlen = headlen - off;
4620 size = SKB_DATA_ALIGN(size);
4622 if (skb_pfmemalloc(skb))
4623 gfp_mask |= __GFP_MEMALLOC;
4624 data = kmalloc_reserve(size +
4625 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4626 gfp_mask, NUMA_NO_NODE, NULL);
4630 size = SKB_WITH_OVERHEAD(ksize(data));
4632 /* Copy real data, and all frags */
4633 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
4636 memcpy((struct skb_shared_info *)(data + size),
4638 offsetof(struct skb_shared_info,
4639 frags[skb_shinfo(skb)->nr_frags]));
4640 if (skb_cloned(skb)) {
4641 /* drop the old head gracefully */
4642 if (skb_orphan_frags(skb, gfp_mask)) {
4646 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
4647 skb_frag_ref(skb, i);
4648 if (skb_has_frag_list(skb))
4649 skb_clone_fraglist(skb);
4650 skb_release_data(skb);
4652 /* we can reuse existing recount- all we did was
4661 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4664 skb->end = skb->head + size;
4666 skb_set_tail_pointer(skb, skb_headlen(skb));
4667 skb_headers_offset_update(skb, 0);
4671 atomic_set(&skb_shinfo(skb)->dataref, 1);
4676 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
4678 /* carve out the first eat bytes from skb's frag_list. May recurse into
4681 static int pskb_carve_frag_list(struct sk_buff *skb,
4682 struct skb_shared_info *shinfo, int eat,
4685 struct sk_buff *list = shinfo->frag_list;
4686 struct sk_buff *clone = NULL;
4687 struct sk_buff *insp = NULL;
4691 pr_err("Not enough bytes to eat. Want %d\n", eat);
4694 if (list->len <= eat) {
4695 /* Eaten as whole. */
4700 /* Eaten partially. */
4701 if (skb_shared(list)) {
4702 clone = skb_clone(list, gfp_mask);
4708 /* This may be pulled without problems. */
4711 if (pskb_carve(list, eat, gfp_mask) < 0) {
4719 /* Free pulled out fragments. */
4720 while ((list = shinfo->frag_list) != insp) {
4721 shinfo->frag_list = list->next;
4724 /* And insert new clone at head. */
4727 shinfo->frag_list = clone;
4732 /* carve off first len bytes from skb. Split line (off) is in the
4733 * non-linear part of skb
4735 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
4736 int pos, gfp_t gfp_mask)
4739 int size = skb_end_offset(skb);
4741 const int nfrags = skb_shinfo(skb)->nr_frags;
4742 struct skb_shared_info *shinfo;
4744 size = SKB_DATA_ALIGN(size);
4746 if (skb_pfmemalloc(skb))
4747 gfp_mask |= __GFP_MEMALLOC;
4748 data = kmalloc_reserve(size +
4749 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4750 gfp_mask, NUMA_NO_NODE, NULL);
4754 size = SKB_WITH_OVERHEAD(ksize(data));
4756 memcpy((struct skb_shared_info *)(data + size),
4757 skb_shinfo(skb), offsetof(struct skb_shared_info,
4758 frags[skb_shinfo(skb)->nr_frags]));
4759 if (skb_orphan_frags(skb, gfp_mask)) {
4763 shinfo = (struct skb_shared_info *)(data + size);
4764 for (i = 0; i < nfrags; i++) {
4765 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
4767 if (pos + fsize > off) {
4768 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
4772 * We have two variants in this case:
4773 * 1. Move all the frag to the second
4774 * part, if it is possible. F.e.
4775 * this approach is mandatory for TUX,
4776 * where splitting is expensive.
4777 * 2. Split is accurately. We make this.
4779 shinfo->frags[0].page_offset += off - pos;
4780 skb_frag_size_sub(&shinfo->frags[0], off - pos);
4782 skb_frag_ref(skb, i);
4787 shinfo->nr_frags = k;
4788 if (skb_has_frag_list(skb))
4789 skb_clone_fraglist(skb);
4792 /* split line is in frag list */
4793 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
4795 skb_release_data(skb);
4800 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4803 skb->end = skb->head + size;
4805 skb_reset_tail_pointer(skb);
4806 skb_headers_offset_update(skb, 0);
4811 skb->data_len = skb->len;
4812 atomic_set(&skb_shinfo(skb)->dataref, 1);
4816 /* remove len bytes from the beginning of the skb */
4817 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
4819 int headlen = skb_headlen(skb);
4822 return pskb_carve_inside_header(skb, len, headlen, gfp);
4824 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
4827 /* Extract to_copy bytes starting at off from skb, and return this in
4830 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
4831 int to_copy, gfp_t gfp)
4833 struct sk_buff *clone = skb_clone(skb, gfp);
4838 if (pskb_carve(clone, off, gfp) < 0 ||
4839 pskb_trim(clone, to_copy)) {
4845 EXPORT_SYMBOL(pskb_extract);