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/sctp.h>
53 #include <linux/netdevice.h>
54 #ifdef CONFIG_NET_CLS_ACT
55 #include <net/pkt_sched.h>
57 #include <linux/string.h>
58 #include <linux/skbuff.h>
59 #include <linux/splice.h>
60 #include <linux/cache.h>
61 #include <linux/rtnetlink.h>
62 #include <linux/init.h>
63 #include <linux/scatterlist.h>
64 #include <linux/errqueue.h>
65 #include <linux/prefetch.h>
66 #include <linux/if_vlan.h>
68 #include <net/protocol.h>
71 #include <net/checksum.h>
72 #include <net/ip6_checksum.h>
75 #include <linux/uaccess.h>
76 #include <trace/events/skb.h>
77 #include <linux/highmem.h>
78 #include <linux/capability.h>
79 #include <linux/user_namespace.h>
81 struct kmem_cache *skbuff_head_cache __read_mostly;
82 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
83 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
84 EXPORT_SYMBOL(sysctl_max_skb_frags);
87 * skb_panic - private function for out-of-line support
91 * @msg: skb_over_panic or skb_under_panic
93 * Out-of-line support for skb_put() and skb_push().
94 * Called via the wrapper skb_over_panic() or skb_under_panic().
95 * Keep out of line to prevent kernel bloat.
96 * __builtin_return_address is not used because it is not always reliable.
98 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
101 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
102 msg, addr, skb->len, sz, skb->head, skb->data,
103 (unsigned long)skb->tail, (unsigned long)skb->end,
104 skb->dev ? skb->dev->name : "<NULL>");
108 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
110 skb_panic(skb, sz, addr, __func__);
113 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
115 skb_panic(skb, sz, addr, __func__);
119 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
120 * the caller if emergency pfmemalloc reserves are being used. If it is and
121 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
122 * may be used. Otherwise, the packet data may be discarded until enough
125 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
126 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
128 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
129 unsigned long ip, bool *pfmemalloc)
132 bool ret_pfmemalloc = false;
135 * Try a regular allocation, when that fails and we're not entitled
136 * to the reserves, fail.
138 obj = kmalloc_node_track_caller(size,
139 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
141 if (obj || !(gfp_pfmemalloc_allowed(flags)))
144 /* Try again but now we are using pfmemalloc reserves */
145 ret_pfmemalloc = true;
146 obj = kmalloc_node_track_caller(size, flags, node);
150 *pfmemalloc = ret_pfmemalloc;
155 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
156 * 'private' fields and also do memory statistics to find all the
161 struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
166 skb = kmem_cache_alloc_node(skbuff_head_cache,
167 gfp_mask & ~__GFP_DMA, node);
172 * Only clear those fields we need to clear, not those that we will
173 * actually initialise below. Hence, don't put any more fields after
174 * the tail pointer in struct sk_buff!
176 memset(skb, 0, offsetof(struct sk_buff, tail));
178 skb->truesize = sizeof(struct sk_buff);
179 atomic_set(&skb->users, 1);
181 skb->mac_header = (typeof(skb->mac_header))~0U;
187 * __alloc_skb - allocate a network buffer
188 * @size: size to allocate
189 * @gfp_mask: allocation mask
190 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
191 * instead of head cache and allocate a cloned (child) skb.
192 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
193 * allocations in case the data is required for writeback
194 * @node: numa node to allocate memory on
196 * Allocate a new &sk_buff. The returned buffer has no headroom and a
197 * tail room of at least size bytes. The object has a reference count
198 * of one. The return is the buffer. On a failure the return is %NULL.
200 * Buffers may only be allocated from interrupts using a @gfp_mask of
203 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
206 struct kmem_cache *cache;
207 struct skb_shared_info *shinfo;
212 cache = (flags & SKB_ALLOC_FCLONE)
213 ? skbuff_fclone_cache : skbuff_head_cache;
215 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
216 gfp_mask |= __GFP_MEMALLOC;
219 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
224 /* We do our best to align skb_shared_info on a separate cache
225 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
226 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
227 * Both skb->head and skb_shared_info are cache line aligned.
229 size = SKB_DATA_ALIGN(size);
230 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
231 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
234 /* kmalloc(size) might give us more room than requested.
235 * Put skb_shared_info exactly at the end of allocated zone,
236 * to allow max possible filling before reallocation.
238 size = SKB_WITH_OVERHEAD(ksize(data));
239 prefetchw(data + size);
242 * Only clear those fields we need to clear, not those that we will
243 * actually initialise below. Hence, don't put any more fields after
244 * the tail pointer in struct sk_buff!
246 memset(skb, 0, offsetof(struct sk_buff, tail));
247 /* Account for allocated memory : skb + skb->head */
248 skb->truesize = SKB_TRUESIZE(size);
249 skb->pfmemalloc = pfmemalloc;
250 atomic_set(&skb->users, 1);
253 skb_reset_tail_pointer(skb);
254 skb->end = skb->tail + size;
255 skb->mac_header = (typeof(skb->mac_header))~0U;
256 skb->transport_header = (typeof(skb->transport_header))~0U;
258 /* make sure we initialize shinfo sequentially */
259 shinfo = skb_shinfo(skb);
260 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
261 atomic_set(&shinfo->dataref, 1);
262 kmemcheck_annotate_variable(shinfo->destructor_arg);
264 if (flags & SKB_ALLOC_FCLONE) {
265 struct sk_buff_fclones *fclones;
267 fclones = container_of(skb, struct sk_buff_fclones, skb1);
269 kmemcheck_annotate_bitfield(&fclones->skb2, flags1);
270 skb->fclone = SKB_FCLONE_ORIG;
271 atomic_set(&fclones->fclone_ref, 1);
273 fclones->skb2.fclone = SKB_FCLONE_CLONE;
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;
356 unsigned int skb_count;
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 = page_frag_alloc(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 page_frag_alloc(&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 = page_frag_alloc(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 = page_frag_alloc(&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(skb));
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);
885 * You should not add any new code to this function. Add it to
886 * __copy_skb_header above instead.
888 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
890 #define C(x) n->x = skb->x
892 n->next = n->prev = NULL;
894 __copy_skb_header(n, skb);
899 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
902 n->destructor = NULL;
909 atomic_set(&n->users, 1);
911 atomic_inc(&(skb_shinfo(skb)->dataref));
919 * skb_morph - morph one skb into another
920 * @dst: the skb to receive the contents
921 * @src: the skb to supply the contents
923 * This is identical to skb_clone except that the target skb is
924 * supplied by the user.
926 * The target skb is returned upon exit.
928 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
930 skb_release_all(dst);
931 return __skb_clone(dst, src);
933 EXPORT_SYMBOL_GPL(skb_morph);
936 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
937 * @skb: the skb to modify
938 * @gfp_mask: allocation priority
940 * This must be called on SKBTX_DEV_ZEROCOPY skb.
941 * It will copy all frags into kernel and drop the reference
942 * to userspace pages.
944 * If this function is called from an interrupt gfp_mask() must be
947 * Returns 0 on success or a negative error code on failure
948 * to allocate kernel memory to copy to.
950 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
953 int num_frags = skb_shinfo(skb)->nr_frags;
954 struct page *page, *head = NULL;
955 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
957 for (i = 0; i < num_frags; i++) {
959 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
961 page = alloc_page(gfp_mask);
964 struct page *next = (struct page *)page_private(head);
970 vaddr = kmap_atomic(skb_frag_page(f));
971 memcpy(page_address(page),
972 vaddr + f->page_offset, skb_frag_size(f));
973 kunmap_atomic(vaddr);
974 set_page_private(page, (unsigned long)head);
978 /* skb frags release userspace buffers */
979 for (i = 0; i < num_frags; i++)
980 skb_frag_unref(skb, i);
982 uarg->callback(uarg, false);
984 /* skb frags point to kernel buffers */
985 for (i = num_frags - 1; i >= 0; i--) {
986 __skb_fill_page_desc(skb, i, head, 0,
987 skb_shinfo(skb)->frags[i].size);
988 head = (struct page *)page_private(head);
991 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
994 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
997 * skb_clone - duplicate an sk_buff
998 * @skb: buffer to clone
999 * @gfp_mask: allocation priority
1001 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1002 * copies share the same packet data but not structure. The new
1003 * buffer has a reference count of 1. If the allocation fails the
1004 * function returns %NULL otherwise the new buffer is returned.
1006 * If this function is called from an interrupt gfp_mask() must be
1010 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1012 struct sk_buff_fclones *fclones = container_of(skb,
1013 struct sk_buff_fclones,
1017 if (skb_orphan_frags(skb, gfp_mask))
1020 if (skb->fclone == SKB_FCLONE_ORIG &&
1021 atomic_read(&fclones->fclone_ref) == 1) {
1023 atomic_set(&fclones->fclone_ref, 2);
1025 if (skb_pfmemalloc(skb))
1026 gfp_mask |= __GFP_MEMALLOC;
1028 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1032 kmemcheck_annotate_bitfield(n, flags1);
1033 n->fclone = SKB_FCLONE_UNAVAILABLE;
1036 return __skb_clone(n, skb);
1038 EXPORT_SYMBOL(skb_clone);
1040 static void skb_headers_offset_update(struct sk_buff *skb, int off)
1042 /* Only adjust this if it actually is csum_start rather than csum */
1043 if (skb->ip_summed == CHECKSUM_PARTIAL)
1044 skb->csum_start += off;
1045 /* {transport,network,mac}_header and tail are relative to skb->head */
1046 skb->transport_header += off;
1047 skb->network_header += off;
1048 if (skb_mac_header_was_set(skb))
1049 skb->mac_header += off;
1050 skb->inner_transport_header += off;
1051 skb->inner_network_header += off;
1052 skb->inner_mac_header += off;
1055 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1057 __copy_skb_header(new, old);
1059 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1060 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1061 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1064 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1066 if (skb_pfmemalloc(skb))
1067 return SKB_ALLOC_RX;
1072 * skb_copy - create private copy of an sk_buff
1073 * @skb: buffer to copy
1074 * @gfp_mask: allocation priority
1076 * Make a copy of both an &sk_buff and its data. This is used when the
1077 * caller wishes to modify the data and needs a private copy of the
1078 * data to alter. Returns %NULL on failure or the pointer to the buffer
1079 * on success. The returned buffer has a reference count of 1.
1081 * As by-product this function converts non-linear &sk_buff to linear
1082 * one, so that &sk_buff becomes completely private and caller is allowed
1083 * to modify all the data of returned buffer. This means that this
1084 * function is not recommended for use in circumstances when only
1085 * header is going to be modified. Use pskb_copy() instead.
1088 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1090 int headerlen = skb_headroom(skb);
1091 unsigned int size = skb_end_offset(skb) + skb->data_len;
1092 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1093 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1098 /* Set the data pointer */
1099 skb_reserve(n, headerlen);
1100 /* Set the tail pointer and length */
1101 skb_put(n, skb->len);
1103 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1106 copy_skb_header(n, skb);
1109 EXPORT_SYMBOL(skb_copy);
1112 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1113 * @skb: buffer to copy
1114 * @headroom: headroom of new skb
1115 * @gfp_mask: allocation priority
1116 * @fclone: if true allocate the copy of the skb from the fclone
1117 * cache instead of the head cache; it is recommended to set this
1118 * to true for the cases where the copy will likely be cloned
1120 * Make a copy of both an &sk_buff and part of its data, located
1121 * in header. Fragmented data remain shared. This is used when
1122 * the caller wishes to modify only header of &sk_buff and needs
1123 * private copy of the header to alter. Returns %NULL on failure
1124 * or the pointer to the buffer on success.
1125 * The returned buffer has a reference count of 1.
1128 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1129 gfp_t gfp_mask, bool fclone)
1131 unsigned int size = skb_headlen(skb) + headroom;
1132 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1133 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1138 /* Set the data pointer */
1139 skb_reserve(n, headroom);
1140 /* Set the tail pointer and length */
1141 skb_put(n, skb_headlen(skb));
1142 /* Copy the bytes */
1143 skb_copy_from_linear_data(skb, n->data, n->len);
1145 n->truesize += skb->data_len;
1146 n->data_len = skb->data_len;
1149 if (skb_shinfo(skb)->nr_frags) {
1152 if (skb_orphan_frags(skb, gfp_mask)) {
1157 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1158 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1159 skb_frag_ref(skb, i);
1161 skb_shinfo(n)->nr_frags = i;
1164 if (skb_has_frag_list(skb)) {
1165 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1166 skb_clone_fraglist(n);
1169 copy_skb_header(n, skb);
1173 EXPORT_SYMBOL(__pskb_copy_fclone);
1176 * pskb_expand_head - reallocate header of &sk_buff
1177 * @skb: buffer to reallocate
1178 * @nhead: room to add at head
1179 * @ntail: room to add at tail
1180 * @gfp_mask: allocation priority
1182 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1183 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1184 * reference count of 1. Returns zero in the case of success or error,
1185 * if expansion failed. In the last case, &sk_buff is not changed.
1187 * All the pointers pointing into skb header may change and must be
1188 * reloaded after call to this function.
1191 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1194 int i, osize = skb_end_offset(skb);
1195 int size = osize + nhead + ntail;
1201 if (skb_shared(skb))
1204 size = SKB_DATA_ALIGN(size);
1206 if (skb_pfmemalloc(skb))
1207 gfp_mask |= __GFP_MEMALLOC;
1208 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1209 gfp_mask, NUMA_NO_NODE, NULL);
1212 size = SKB_WITH_OVERHEAD(ksize(data));
1214 /* Copy only real data... and, alas, header. This should be
1215 * optimized for the cases when header is void.
1217 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1219 memcpy((struct skb_shared_info *)(data + size),
1221 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1224 * if shinfo is shared we must drop the old head gracefully, but if it
1225 * is not we can just drop the old head and let the existing refcount
1226 * be since all we did is relocate the values
1228 if (skb_cloned(skb)) {
1229 /* copy this zero copy skb frags */
1230 if (skb_orphan_frags(skb, gfp_mask))
1232 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1233 skb_frag_ref(skb, i);
1235 if (skb_has_frag_list(skb))
1236 skb_clone_fraglist(skb);
1238 skb_release_data(skb);
1242 off = (data + nhead) - skb->head;
1247 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1251 skb->end = skb->head + size;
1254 skb_headers_offset_update(skb, nhead);
1258 atomic_set(&skb_shinfo(skb)->dataref, 1);
1260 /* It is not generally safe to change skb->truesize.
1261 * For the moment, we really care of rx path, or
1262 * when skb is orphaned (not attached to a socket).
1264 if (!skb->sk || skb->destructor == sock_edemux)
1265 skb->truesize += size - osize;
1274 EXPORT_SYMBOL(pskb_expand_head);
1276 /* Make private copy of skb with writable head and some headroom */
1278 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1280 struct sk_buff *skb2;
1281 int delta = headroom - skb_headroom(skb);
1284 skb2 = pskb_copy(skb, GFP_ATOMIC);
1286 skb2 = skb_clone(skb, GFP_ATOMIC);
1287 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1295 EXPORT_SYMBOL(skb_realloc_headroom);
1298 * skb_copy_expand - copy and expand sk_buff
1299 * @skb: buffer to copy
1300 * @newheadroom: new free bytes at head
1301 * @newtailroom: new free bytes at tail
1302 * @gfp_mask: allocation priority
1304 * Make a copy of both an &sk_buff and its data and while doing so
1305 * allocate additional space.
1307 * This is used when the caller wishes to modify the data and needs a
1308 * private copy of the data to alter as well as more space for new fields.
1309 * Returns %NULL on failure or the pointer to the buffer
1310 * on success. The returned buffer has a reference count of 1.
1312 * You must pass %GFP_ATOMIC as the allocation priority if this function
1313 * is called from an interrupt.
1315 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1316 int newheadroom, int newtailroom,
1320 * Allocate the copy buffer
1322 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1323 gfp_mask, skb_alloc_rx_flag(skb),
1325 int oldheadroom = skb_headroom(skb);
1326 int head_copy_len, head_copy_off;
1331 skb_reserve(n, newheadroom);
1333 /* Set the tail pointer and length */
1334 skb_put(n, skb->len);
1336 head_copy_len = oldheadroom;
1338 if (newheadroom <= head_copy_len)
1339 head_copy_len = newheadroom;
1341 head_copy_off = newheadroom - head_copy_len;
1343 /* Copy the linear header and data. */
1344 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1345 skb->len + head_copy_len))
1348 copy_skb_header(n, skb);
1350 skb_headers_offset_update(n, newheadroom - oldheadroom);
1354 EXPORT_SYMBOL(skb_copy_expand);
1357 * skb_pad - zero pad the tail of an skb
1358 * @skb: buffer to pad
1359 * @pad: space to pad
1361 * Ensure that a buffer is followed by a padding area that is zero
1362 * filled. Used by network drivers which may DMA or transfer data
1363 * beyond the buffer end onto the wire.
1365 * May return error in out of memory cases. The skb is freed on error.
1368 int skb_pad(struct sk_buff *skb, int pad)
1373 /* If the skbuff is non linear tailroom is always zero.. */
1374 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1375 memset(skb->data+skb->len, 0, pad);
1379 ntail = skb->data_len + pad - (skb->end - skb->tail);
1380 if (likely(skb_cloned(skb) || ntail > 0)) {
1381 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1386 /* FIXME: The use of this function with non-linear skb's really needs
1389 err = skb_linearize(skb);
1393 memset(skb->data + skb->len, 0, pad);
1400 EXPORT_SYMBOL(skb_pad);
1403 * pskb_put - add data to the tail of a potentially fragmented buffer
1404 * @skb: start of the buffer to use
1405 * @tail: tail fragment of the buffer to use
1406 * @len: amount of data to add
1408 * This function extends the used data area of the potentially
1409 * fragmented buffer. @tail must be the last fragment of @skb -- or
1410 * @skb itself. If this would exceed the total buffer size the kernel
1411 * will panic. A pointer to the first byte of the extra data is
1415 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1418 skb->data_len += len;
1421 return skb_put(tail, len);
1423 EXPORT_SYMBOL_GPL(pskb_put);
1426 * skb_put - add data to a buffer
1427 * @skb: buffer to use
1428 * @len: amount of data to add
1430 * This function extends the used data area of the buffer. If this would
1431 * exceed the total buffer size the kernel will panic. A pointer to the
1432 * first byte of the extra data is returned.
1434 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1436 unsigned char *tmp = skb_tail_pointer(skb);
1437 SKB_LINEAR_ASSERT(skb);
1440 if (unlikely(skb->tail > skb->end))
1441 skb_over_panic(skb, len, __builtin_return_address(0));
1444 EXPORT_SYMBOL(skb_put);
1447 * skb_push - add data to the start of a buffer
1448 * @skb: buffer to use
1449 * @len: amount of data to add
1451 * This function extends the used data area of the buffer at the buffer
1452 * start. If this would exceed the total buffer headroom the kernel will
1453 * panic. A pointer to the first byte of the extra data is returned.
1455 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1459 if (unlikely(skb->data<skb->head))
1460 skb_under_panic(skb, len, __builtin_return_address(0));
1463 EXPORT_SYMBOL(skb_push);
1466 * skb_pull - remove data from the start of a buffer
1467 * @skb: buffer to use
1468 * @len: amount of data to remove
1470 * This function removes data from the start of a buffer, returning
1471 * the memory to the headroom. A pointer to the next data in the buffer
1472 * is returned. Once the data has been pulled future pushes will overwrite
1475 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1477 return skb_pull_inline(skb, len);
1479 EXPORT_SYMBOL(skb_pull);
1482 * skb_trim - remove end from a buffer
1483 * @skb: buffer to alter
1486 * Cut the length of a buffer down by removing data from the tail. If
1487 * the buffer is already under the length specified it is not modified.
1488 * The skb must be linear.
1490 void skb_trim(struct sk_buff *skb, unsigned int len)
1493 __skb_trim(skb, len);
1495 EXPORT_SYMBOL(skb_trim);
1497 /* Trims skb to length len. It can change skb pointers.
1500 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1502 struct sk_buff **fragp;
1503 struct sk_buff *frag;
1504 int offset = skb_headlen(skb);
1505 int nfrags = skb_shinfo(skb)->nr_frags;
1509 if (skb_cloned(skb) &&
1510 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1517 for (; i < nfrags; i++) {
1518 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1525 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1528 skb_shinfo(skb)->nr_frags = i;
1530 for (; i < nfrags; i++)
1531 skb_frag_unref(skb, i);
1533 if (skb_has_frag_list(skb))
1534 skb_drop_fraglist(skb);
1538 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1539 fragp = &frag->next) {
1540 int end = offset + frag->len;
1542 if (skb_shared(frag)) {
1543 struct sk_buff *nfrag;
1545 nfrag = skb_clone(frag, GFP_ATOMIC);
1546 if (unlikely(!nfrag))
1549 nfrag->next = frag->next;
1561 unlikely((err = pskb_trim(frag, len - offset))))
1565 skb_drop_list(&frag->next);
1570 if (len > skb_headlen(skb)) {
1571 skb->data_len -= skb->len - len;
1576 skb_set_tail_pointer(skb, len);
1581 EXPORT_SYMBOL(___pskb_trim);
1584 * __pskb_pull_tail - advance tail of skb header
1585 * @skb: buffer to reallocate
1586 * @delta: number of bytes to advance tail
1588 * The function makes a sense only on a fragmented &sk_buff,
1589 * it expands header moving its tail forward and copying necessary
1590 * data from fragmented part.
1592 * &sk_buff MUST have reference count of 1.
1594 * Returns %NULL (and &sk_buff does not change) if pull failed
1595 * or value of new tail of skb in the case of success.
1597 * All the pointers pointing into skb header may change and must be
1598 * reloaded after call to this function.
1601 /* Moves tail of skb head forward, copying data from fragmented part,
1602 * when it is necessary.
1603 * 1. It may fail due to malloc failure.
1604 * 2. It may change skb pointers.
1606 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1608 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1610 /* If skb has not enough free space at tail, get new one
1611 * plus 128 bytes for future expansions. If we have enough
1612 * room at tail, reallocate without expansion only if skb is cloned.
1614 int i, k, eat = (skb->tail + delta) - skb->end;
1616 if (eat > 0 || skb_cloned(skb)) {
1617 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1622 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1625 /* Optimization: no fragments, no reasons to preestimate
1626 * size of pulled pages. Superb.
1628 if (!skb_has_frag_list(skb))
1631 /* Estimate size of pulled pages. */
1633 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1634 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1641 /* If we need update frag list, we are in troubles.
1642 * Certainly, it possible to add an offset to skb data,
1643 * but taking into account that pulling is expected to
1644 * be very rare operation, it is worth to fight against
1645 * further bloating skb head and crucify ourselves here instead.
1646 * Pure masohism, indeed. 8)8)
1649 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1650 struct sk_buff *clone = NULL;
1651 struct sk_buff *insp = NULL;
1656 if (list->len <= eat) {
1657 /* Eaten as whole. */
1662 /* Eaten partially. */
1664 if (skb_shared(list)) {
1665 /* Sucks! We need to fork list. :-( */
1666 clone = skb_clone(list, GFP_ATOMIC);
1672 /* This may be pulled without
1676 if (!pskb_pull(list, eat)) {
1684 /* Free pulled out fragments. */
1685 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1686 skb_shinfo(skb)->frag_list = list->next;
1689 /* And insert new clone at head. */
1692 skb_shinfo(skb)->frag_list = clone;
1695 /* Success! Now we may commit changes to skb data. */
1700 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1701 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1704 skb_frag_unref(skb, i);
1707 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1709 skb_shinfo(skb)->frags[k].page_offset += eat;
1710 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1716 skb_shinfo(skb)->nr_frags = k;
1719 skb->data_len -= delta;
1721 return skb_tail_pointer(skb);
1723 EXPORT_SYMBOL(__pskb_pull_tail);
1726 * skb_copy_bits - copy bits from skb to kernel buffer
1728 * @offset: offset in source
1729 * @to: destination buffer
1730 * @len: number of bytes to copy
1732 * Copy the specified number of bytes from the source skb to the
1733 * destination buffer.
1736 * If its prototype is ever changed,
1737 * check arch/{*}/net/{*}.S files,
1738 * since it is called from BPF assembly code.
1740 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1742 int start = skb_headlen(skb);
1743 struct sk_buff *frag_iter;
1746 if (offset > (int)skb->len - len)
1750 if ((copy = start - offset) > 0) {
1753 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1754 if ((len -= copy) == 0)
1760 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1762 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1764 WARN_ON(start > offset + len);
1766 end = start + skb_frag_size(f);
1767 if ((copy = end - offset) > 0) {
1773 vaddr = kmap_atomic(skb_frag_page(f));
1775 vaddr + f->page_offset + offset - start,
1777 kunmap_atomic(vaddr);
1779 if ((len -= copy) == 0)
1787 skb_walk_frags(skb, frag_iter) {
1790 WARN_ON(start > offset + len);
1792 end = start + frag_iter->len;
1793 if ((copy = end - offset) > 0) {
1796 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1798 if ((len -= copy) == 0)
1812 EXPORT_SYMBOL(skb_copy_bits);
1815 * Callback from splice_to_pipe(), if we need to release some pages
1816 * at the end of the spd in case we error'ed out in filling the pipe.
1818 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1820 put_page(spd->pages[i]);
1823 static struct page *linear_to_page(struct page *page, unsigned int *len,
1824 unsigned int *offset,
1827 struct page_frag *pfrag = sk_page_frag(sk);
1829 if (!sk_page_frag_refill(sk, pfrag))
1832 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1834 memcpy(page_address(pfrag->page) + pfrag->offset,
1835 page_address(page) + *offset, *len);
1836 *offset = pfrag->offset;
1837 pfrag->offset += *len;
1842 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1844 unsigned int offset)
1846 return spd->nr_pages &&
1847 spd->pages[spd->nr_pages - 1] == page &&
1848 (spd->partial[spd->nr_pages - 1].offset +
1849 spd->partial[spd->nr_pages - 1].len == offset);
1853 * Fill page/offset/length into spd, if it can hold more pages.
1855 static bool spd_fill_page(struct splice_pipe_desc *spd,
1856 struct pipe_inode_info *pipe, struct page *page,
1857 unsigned int *len, unsigned int offset,
1861 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1865 page = linear_to_page(page, len, &offset, sk);
1869 if (spd_can_coalesce(spd, page, offset)) {
1870 spd->partial[spd->nr_pages - 1].len += *len;
1874 spd->pages[spd->nr_pages] = page;
1875 spd->partial[spd->nr_pages].len = *len;
1876 spd->partial[spd->nr_pages].offset = offset;
1882 static bool __splice_segment(struct page *page, unsigned int poff,
1883 unsigned int plen, unsigned int *off,
1885 struct splice_pipe_desc *spd, bool linear,
1887 struct pipe_inode_info *pipe)
1892 /* skip this segment if already processed */
1898 /* ignore any bits we already processed */
1904 unsigned int flen = min(*len, plen);
1906 if (spd_fill_page(spd, pipe, page, &flen, poff,
1912 } while (*len && plen);
1918 * Map linear and fragment data from the skb to spd. It reports true if the
1919 * pipe is full or if we already spliced the requested length.
1921 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1922 unsigned int *offset, unsigned int *len,
1923 struct splice_pipe_desc *spd, struct sock *sk)
1926 struct sk_buff *iter;
1928 /* map the linear part :
1929 * If skb->head_frag is set, this 'linear' part is backed by a
1930 * fragment, and if the head is not shared with any clones then
1931 * we can avoid a copy since we own the head portion of this page.
1933 if (__splice_segment(virt_to_page(skb->data),
1934 (unsigned long) skb->data & (PAGE_SIZE - 1),
1937 skb_head_is_locked(skb),
1942 * then map the fragments
1944 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1945 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1947 if (__splice_segment(skb_frag_page(f),
1948 f->page_offset, skb_frag_size(f),
1949 offset, len, spd, false, sk, pipe))
1953 skb_walk_frags(skb, iter) {
1954 if (*offset >= iter->len) {
1955 *offset -= iter->len;
1958 /* __skb_splice_bits() only fails if the output has no room
1959 * left, so no point in going over the frag_list for the error
1962 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
1970 * Map data from the skb to a pipe. Should handle both the linear part,
1971 * the fragments, and the frag list.
1973 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
1974 struct pipe_inode_info *pipe, unsigned int tlen,
1977 struct partial_page partial[MAX_SKB_FRAGS];
1978 struct page *pages[MAX_SKB_FRAGS];
1979 struct splice_pipe_desc spd = {
1982 .nr_pages_max = MAX_SKB_FRAGS,
1984 .ops = &nosteal_pipe_buf_ops,
1985 .spd_release = sock_spd_release,
1989 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
1992 ret = splice_to_pipe(pipe, &spd);
1996 EXPORT_SYMBOL_GPL(skb_splice_bits);
1999 * skb_store_bits - store bits from kernel buffer to skb
2000 * @skb: destination buffer
2001 * @offset: offset in destination
2002 * @from: source buffer
2003 * @len: number of bytes to copy
2005 * Copy the specified number of bytes from the source buffer to the
2006 * destination skb. This function handles all the messy bits of
2007 * traversing fragment lists and such.
2010 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2012 int start = skb_headlen(skb);
2013 struct sk_buff *frag_iter;
2016 if (offset > (int)skb->len - len)
2019 if ((copy = start - offset) > 0) {
2022 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2023 if ((len -= copy) == 0)
2029 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2030 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2033 WARN_ON(start > offset + len);
2035 end = start + skb_frag_size(frag);
2036 if ((copy = end - offset) > 0) {
2042 vaddr = kmap_atomic(skb_frag_page(frag));
2043 memcpy(vaddr + frag->page_offset + offset - start,
2045 kunmap_atomic(vaddr);
2047 if ((len -= copy) == 0)
2055 skb_walk_frags(skb, frag_iter) {
2058 WARN_ON(start > offset + len);
2060 end = start + frag_iter->len;
2061 if ((copy = end - offset) > 0) {
2064 if (skb_store_bits(frag_iter, offset - start,
2067 if ((len -= copy) == 0)
2080 EXPORT_SYMBOL(skb_store_bits);
2082 /* Checksum skb data. */
2083 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2084 __wsum csum, const struct skb_checksum_ops *ops)
2086 int start = skb_headlen(skb);
2087 int i, copy = start - offset;
2088 struct sk_buff *frag_iter;
2091 /* Checksum header. */
2095 csum = ops->update(skb->data + offset, copy, csum);
2096 if ((len -= copy) == 0)
2102 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2104 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2106 WARN_ON(start > offset + len);
2108 end = start + skb_frag_size(frag);
2109 if ((copy = end - offset) > 0) {
2115 vaddr = kmap_atomic(skb_frag_page(frag));
2116 csum2 = ops->update(vaddr + frag->page_offset +
2117 offset - start, copy, 0);
2118 kunmap_atomic(vaddr);
2119 csum = ops->combine(csum, csum2, pos, copy);
2128 skb_walk_frags(skb, frag_iter) {
2131 WARN_ON(start > offset + len);
2133 end = start + frag_iter->len;
2134 if ((copy = end - offset) > 0) {
2138 csum2 = __skb_checksum(frag_iter, offset - start,
2140 csum = ops->combine(csum, csum2, pos, copy);
2141 if ((len -= copy) == 0)
2152 EXPORT_SYMBOL(__skb_checksum);
2154 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2155 int len, __wsum csum)
2157 const struct skb_checksum_ops ops = {
2158 .update = csum_partial_ext,
2159 .combine = csum_block_add_ext,
2162 return __skb_checksum(skb, offset, len, csum, &ops);
2164 EXPORT_SYMBOL(skb_checksum);
2166 /* Both of above in one bottle. */
2168 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2169 u8 *to, int len, __wsum csum)
2171 int start = skb_headlen(skb);
2172 int i, copy = start - offset;
2173 struct sk_buff *frag_iter;
2180 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2182 if ((len -= copy) == 0)
2189 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2192 WARN_ON(start > offset + len);
2194 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2195 if ((copy = end - offset) > 0) {
2198 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2202 vaddr = kmap_atomic(skb_frag_page(frag));
2203 csum2 = csum_partial_copy_nocheck(vaddr +
2207 kunmap_atomic(vaddr);
2208 csum = csum_block_add(csum, csum2, pos);
2218 skb_walk_frags(skb, frag_iter) {
2222 WARN_ON(start > offset + len);
2224 end = start + frag_iter->len;
2225 if ((copy = end - offset) > 0) {
2228 csum2 = skb_copy_and_csum_bits(frag_iter,
2231 csum = csum_block_add(csum, csum2, pos);
2232 if ((len -= copy) == 0)
2243 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2246 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2247 * @from: source buffer
2249 * Calculates the amount of linear headroom needed in the 'to' skb passed
2250 * into skb_zerocopy().
2253 skb_zerocopy_headlen(const struct sk_buff *from)
2255 unsigned int hlen = 0;
2257 if (!from->head_frag ||
2258 skb_headlen(from) < L1_CACHE_BYTES ||
2259 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2260 hlen = skb_headlen(from);
2262 if (skb_has_frag_list(from))
2267 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2270 * skb_zerocopy - Zero copy skb to skb
2271 * @to: destination buffer
2272 * @from: source buffer
2273 * @len: number of bytes to copy from source buffer
2274 * @hlen: size of linear headroom in destination buffer
2276 * Copies up to `len` bytes from `from` to `to` by creating references
2277 * to the frags in the source buffer.
2279 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2280 * headroom in the `to` buffer.
2283 * 0: everything is OK
2284 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2285 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2288 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2291 int plen = 0; /* length of skb->head fragment */
2294 unsigned int offset;
2296 BUG_ON(!from->head_frag && !hlen);
2298 /* dont bother with small payloads */
2299 if (len <= skb_tailroom(to))
2300 return skb_copy_bits(from, 0, skb_put(to, len), len);
2303 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2308 plen = min_t(int, skb_headlen(from), len);
2310 page = virt_to_head_page(from->head);
2311 offset = from->data - (unsigned char *)page_address(page);
2312 __skb_fill_page_desc(to, 0, page, offset, plen);
2319 to->truesize += len + plen;
2320 to->len += len + plen;
2321 to->data_len += len + plen;
2323 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2328 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2331 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2332 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2333 len -= skb_shinfo(to)->frags[j].size;
2334 skb_frag_ref(to, j);
2337 skb_shinfo(to)->nr_frags = j;
2341 EXPORT_SYMBOL_GPL(skb_zerocopy);
2343 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2348 if (skb->ip_summed == CHECKSUM_PARTIAL)
2349 csstart = skb_checksum_start_offset(skb);
2351 csstart = skb_headlen(skb);
2353 BUG_ON(csstart > skb_headlen(skb));
2355 skb_copy_from_linear_data(skb, to, csstart);
2358 if (csstart != skb->len)
2359 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2360 skb->len - csstart, 0);
2362 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2363 long csstuff = csstart + skb->csum_offset;
2365 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2368 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2371 * skb_dequeue - remove from the head of the queue
2372 * @list: list to dequeue from
2374 * Remove the head of the list. The list lock is taken so the function
2375 * may be used safely with other locking list functions. The head item is
2376 * returned or %NULL if the list is empty.
2379 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2381 unsigned long flags;
2382 struct sk_buff *result;
2384 spin_lock_irqsave(&list->lock, flags);
2385 result = __skb_dequeue(list);
2386 spin_unlock_irqrestore(&list->lock, flags);
2389 EXPORT_SYMBOL(skb_dequeue);
2392 * skb_dequeue_tail - remove from the tail of the queue
2393 * @list: list to dequeue from
2395 * Remove the tail of the list. The list lock is taken so the function
2396 * may be used safely with other locking list functions. The tail item is
2397 * returned or %NULL if the list is empty.
2399 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2401 unsigned long flags;
2402 struct sk_buff *result;
2404 spin_lock_irqsave(&list->lock, flags);
2405 result = __skb_dequeue_tail(list);
2406 spin_unlock_irqrestore(&list->lock, flags);
2409 EXPORT_SYMBOL(skb_dequeue_tail);
2412 * skb_queue_purge - empty a list
2413 * @list: list to empty
2415 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2416 * the list and one reference dropped. This function takes the list
2417 * lock and is atomic with respect to other list locking functions.
2419 void skb_queue_purge(struct sk_buff_head *list)
2421 struct sk_buff *skb;
2422 while ((skb = skb_dequeue(list)) != NULL)
2425 EXPORT_SYMBOL(skb_queue_purge);
2428 * skb_rbtree_purge - empty a skb rbtree
2429 * @root: root of the rbtree to empty
2431 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
2432 * the list and one reference dropped. This function does not take
2433 * any lock. Synchronization should be handled by the caller (e.g., TCP
2434 * out-of-order queue is protected by the socket lock).
2436 void skb_rbtree_purge(struct rb_root *root)
2438 struct sk_buff *skb, *next;
2440 rbtree_postorder_for_each_entry_safe(skb, next, root, rbnode)
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);
2664 if (skb_headlen(skb))
2669 to = skb_shinfo(tgt)->nr_frags;
2670 fragfrom = &skb_shinfo(skb)->frags[from];
2672 /* Actual merge is delayed until the point when we know we can
2673 * commit all, so that we don't have to undo partial changes
2676 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2677 fragfrom->page_offset)) {
2682 todo -= skb_frag_size(fragfrom);
2684 if (skb_prepare_for_shift(skb) ||
2685 skb_prepare_for_shift(tgt))
2688 /* All previous frag pointers might be stale! */
2689 fragfrom = &skb_shinfo(skb)->frags[from];
2690 fragto = &skb_shinfo(tgt)->frags[merge];
2692 skb_frag_size_add(fragto, shiftlen);
2693 skb_frag_size_sub(fragfrom, shiftlen);
2694 fragfrom->page_offset += shiftlen;
2702 /* Skip full, not-fitting skb to avoid expensive operations */
2703 if ((shiftlen == skb->len) &&
2704 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2707 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2710 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2711 if (to == MAX_SKB_FRAGS)
2714 fragfrom = &skb_shinfo(skb)->frags[from];
2715 fragto = &skb_shinfo(tgt)->frags[to];
2717 if (todo >= skb_frag_size(fragfrom)) {
2718 *fragto = *fragfrom;
2719 todo -= skb_frag_size(fragfrom);
2724 __skb_frag_ref(fragfrom);
2725 fragto->page = fragfrom->page;
2726 fragto->page_offset = fragfrom->page_offset;
2727 skb_frag_size_set(fragto, todo);
2729 fragfrom->page_offset += todo;
2730 skb_frag_size_sub(fragfrom, todo);
2738 /* Ready to "commit" this state change to tgt */
2739 skb_shinfo(tgt)->nr_frags = to;
2742 fragfrom = &skb_shinfo(skb)->frags[0];
2743 fragto = &skb_shinfo(tgt)->frags[merge];
2745 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2746 __skb_frag_unref(fragfrom);
2749 /* Reposition in the original skb */
2751 while (from < skb_shinfo(skb)->nr_frags)
2752 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2753 skb_shinfo(skb)->nr_frags = to;
2755 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2758 /* Most likely the tgt won't ever need its checksum anymore, skb on
2759 * the other hand might need it if it needs to be resent
2761 tgt->ip_summed = CHECKSUM_PARTIAL;
2762 skb->ip_summed = CHECKSUM_PARTIAL;
2764 /* Yak, is it really working this way? Some helper please? */
2765 skb->len -= shiftlen;
2766 skb->data_len -= shiftlen;
2767 skb->truesize -= shiftlen;
2768 tgt->len += shiftlen;
2769 tgt->data_len += shiftlen;
2770 tgt->truesize += shiftlen;
2776 * skb_prepare_seq_read - Prepare a sequential read of skb data
2777 * @skb: the buffer to read
2778 * @from: lower offset of data to be read
2779 * @to: upper offset of data to be read
2780 * @st: state variable
2782 * Initializes the specified state variable. Must be called before
2783 * invoking skb_seq_read() for the first time.
2785 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2786 unsigned int to, struct skb_seq_state *st)
2788 st->lower_offset = from;
2789 st->upper_offset = to;
2790 st->root_skb = st->cur_skb = skb;
2791 st->frag_idx = st->stepped_offset = 0;
2792 st->frag_data = NULL;
2794 EXPORT_SYMBOL(skb_prepare_seq_read);
2797 * skb_seq_read - Sequentially read skb data
2798 * @consumed: number of bytes consumed by the caller so far
2799 * @data: destination pointer for data to be returned
2800 * @st: state variable
2802 * Reads a block of skb data at @consumed relative to the
2803 * lower offset specified to skb_prepare_seq_read(). Assigns
2804 * the head of the data block to @data and returns the length
2805 * of the block or 0 if the end of the skb data or the upper
2806 * offset has been reached.
2808 * The caller is not required to consume all of the data
2809 * returned, i.e. @consumed is typically set to the number
2810 * of bytes already consumed and the next call to
2811 * skb_seq_read() will return the remaining part of the block.
2813 * Note 1: The size of each block of data returned can be arbitrary,
2814 * this limitation is the cost for zerocopy sequential
2815 * reads of potentially non linear data.
2817 * Note 2: Fragment lists within fragments are not implemented
2818 * at the moment, state->root_skb could be replaced with
2819 * a stack for this purpose.
2821 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2822 struct skb_seq_state *st)
2824 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2827 if (unlikely(abs_offset >= st->upper_offset)) {
2828 if (st->frag_data) {
2829 kunmap_atomic(st->frag_data);
2830 st->frag_data = NULL;
2836 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2838 if (abs_offset < block_limit && !st->frag_data) {
2839 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2840 return block_limit - abs_offset;
2843 if (st->frag_idx == 0 && !st->frag_data)
2844 st->stepped_offset += skb_headlen(st->cur_skb);
2846 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2847 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2848 block_limit = skb_frag_size(frag) + st->stepped_offset;
2850 if (abs_offset < block_limit) {
2852 st->frag_data = kmap_atomic(skb_frag_page(frag));
2854 *data = (u8 *) st->frag_data + frag->page_offset +
2855 (abs_offset - st->stepped_offset);
2857 return block_limit - abs_offset;
2860 if (st->frag_data) {
2861 kunmap_atomic(st->frag_data);
2862 st->frag_data = NULL;
2866 st->stepped_offset += skb_frag_size(frag);
2869 if (st->frag_data) {
2870 kunmap_atomic(st->frag_data);
2871 st->frag_data = NULL;
2874 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2875 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2878 } else if (st->cur_skb->next) {
2879 st->cur_skb = st->cur_skb->next;
2886 EXPORT_SYMBOL(skb_seq_read);
2889 * skb_abort_seq_read - Abort a sequential read of skb data
2890 * @st: state variable
2892 * Must be called if skb_seq_read() was not called until it
2895 void skb_abort_seq_read(struct skb_seq_state *st)
2898 kunmap_atomic(st->frag_data);
2900 EXPORT_SYMBOL(skb_abort_seq_read);
2902 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2904 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2905 struct ts_config *conf,
2906 struct ts_state *state)
2908 return skb_seq_read(offset, text, TS_SKB_CB(state));
2911 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2913 skb_abort_seq_read(TS_SKB_CB(state));
2917 * skb_find_text - Find a text pattern in skb data
2918 * @skb: the buffer to look in
2919 * @from: search offset
2921 * @config: textsearch configuration
2923 * Finds a pattern in the skb data according to the specified
2924 * textsearch configuration. Use textsearch_next() to retrieve
2925 * subsequent occurrences of the pattern. Returns the offset
2926 * to the first occurrence or UINT_MAX if no match was found.
2928 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2929 unsigned int to, struct ts_config *config)
2931 struct ts_state state;
2934 config->get_next_block = skb_ts_get_next_block;
2935 config->finish = skb_ts_finish;
2937 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
2939 ret = textsearch_find(config, &state);
2940 return (ret <= to - from ? ret : UINT_MAX);
2942 EXPORT_SYMBOL(skb_find_text);
2945 * skb_append_datato_frags - append the user data to a skb
2946 * @sk: sock structure
2947 * @skb: skb structure to be appended with user data.
2948 * @getfrag: call back function to be used for getting the user data
2949 * @from: pointer to user message iov
2950 * @length: length of the iov message
2952 * Description: This procedure append the user data in the fragment part
2953 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2955 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2956 int (*getfrag)(void *from, char *to, int offset,
2957 int len, int odd, struct sk_buff *skb),
2958 void *from, int length)
2960 int frg_cnt = skb_shinfo(skb)->nr_frags;
2964 struct page_frag *pfrag = ¤t->task_frag;
2967 /* Return error if we don't have space for new frag */
2968 if (frg_cnt >= MAX_SKB_FRAGS)
2971 if (!sk_page_frag_refill(sk, pfrag))
2974 /* copy the user data to page */
2975 copy = min_t(int, length, pfrag->size - pfrag->offset);
2977 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2978 offset, copy, 0, skb);
2982 /* copy was successful so update the size parameters */
2983 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2986 pfrag->offset += copy;
2987 get_page(pfrag->page);
2989 skb->truesize += copy;
2990 atomic_add(copy, &sk->sk_wmem_alloc);
2992 skb->data_len += copy;
2996 } while (length > 0);
3000 EXPORT_SYMBOL(skb_append_datato_frags);
3002 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3003 int offset, size_t size)
3005 int i = skb_shinfo(skb)->nr_frags;
3007 if (skb_can_coalesce(skb, i, page, offset)) {
3008 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3009 } else if (i < MAX_SKB_FRAGS) {
3011 skb_fill_page_desc(skb, i, page, offset, size);
3018 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3021 * skb_pull_rcsum - pull skb and update receive checksum
3022 * @skb: buffer to update
3023 * @len: length of data pulled
3025 * This function performs an skb_pull on the packet and updates
3026 * the CHECKSUM_COMPLETE checksum. It should be used on
3027 * receive path processing instead of skb_pull unless you know
3028 * that the checksum difference is zero (e.g., a valid IP header)
3029 * or you are setting ip_summed to CHECKSUM_NONE.
3031 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3033 unsigned char *data = skb->data;
3035 BUG_ON(len > skb->len);
3036 __skb_pull(skb, len);
3037 skb_postpull_rcsum(skb, data, len);
3040 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3043 * skb_segment - Perform protocol segmentation on skb.
3044 * @head_skb: buffer to segment
3045 * @features: features for the output path (see dev->features)
3047 * This function performs segmentation on the given skb. It returns
3048 * a pointer to the first in a list of new skbs for the segments.
3049 * In case of error it returns ERR_PTR(err).
3051 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3052 netdev_features_t features)
3054 struct sk_buff *segs = NULL;
3055 struct sk_buff *tail = NULL;
3056 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3057 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3058 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3059 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3060 struct sk_buff *frag_skb = head_skb;
3061 unsigned int offset = doffset;
3062 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3063 unsigned int partial_segs = 0;
3064 unsigned int headroom;
3065 unsigned int len = head_skb->len;
3068 int nfrags = skb_shinfo(head_skb)->nr_frags;
3074 __skb_push(head_skb, doffset);
3075 proto = skb_network_protocol(head_skb, &dummy);
3076 if (unlikely(!proto))
3077 return ERR_PTR(-EINVAL);
3079 sg = !!(features & NETIF_F_SG);
3080 csum = !!can_checksum_protocol(features, proto);
3082 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3083 if (!(features & NETIF_F_GSO_PARTIAL)) {
3084 struct sk_buff *iter;
3087 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3090 /* Split the buffer at the frag_list pointer.
3091 * This is based on the assumption that all
3092 * buffers in the chain excluding the last
3093 * containing the same amount of data.
3095 skb_walk_frags(head_skb, iter) {
3096 if (skb_headlen(iter))
3103 /* GSO partial only requires that we trim off any excess that
3104 * doesn't fit into an MSS sized block, so take care of that
3107 partial_segs = len / mss;
3108 if (partial_segs > 1)
3109 mss *= partial_segs;
3115 headroom = skb_headroom(head_skb);
3116 pos = skb_headlen(head_skb);
3119 struct sk_buff *nskb;
3120 skb_frag_t *nskb_frag;
3124 if (unlikely(mss == GSO_BY_FRAGS)) {
3125 len = list_skb->len;
3127 len = head_skb->len - offset;
3132 hsize = skb_headlen(head_skb) - offset;
3135 if (hsize > len || !sg)
3138 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3139 (skb_headlen(list_skb) == len || sg)) {
3140 BUG_ON(skb_headlen(list_skb) > len);
3143 nfrags = skb_shinfo(list_skb)->nr_frags;
3144 frag = skb_shinfo(list_skb)->frags;
3145 frag_skb = list_skb;
3146 pos += skb_headlen(list_skb);
3148 while (pos < offset + len) {
3149 BUG_ON(i >= nfrags);
3151 size = skb_frag_size(frag);
3152 if (pos + size > offset + len)
3160 nskb = skb_clone(list_skb, GFP_ATOMIC);
3161 list_skb = list_skb->next;
3163 if (unlikely(!nskb))
3166 if (unlikely(pskb_trim(nskb, len))) {
3171 hsize = skb_end_offset(nskb);
3172 if (skb_cow_head(nskb, doffset + headroom)) {
3177 nskb->truesize += skb_end_offset(nskb) - hsize;
3178 skb_release_head_state(nskb);
3179 __skb_push(nskb, doffset);
3181 nskb = __alloc_skb(hsize + doffset + headroom,
3182 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3185 if (unlikely(!nskb))
3188 skb_reserve(nskb, headroom);
3189 __skb_put(nskb, doffset);
3198 __copy_skb_header(nskb, head_skb);
3200 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3201 skb_reset_mac_len(nskb);
3203 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3204 nskb->data - tnl_hlen,
3205 doffset + tnl_hlen);
3207 if (nskb->len == len + doffset)
3208 goto perform_csum_check;
3211 if (!nskb->remcsum_offload)
3212 nskb->ip_summed = CHECKSUM_NONE;
3213 SKB_GSO_CB(nskb)->csum =
3214 skb_copy_and_csum_bits(head_skb, offset,
3217 SKB_GSO_CB(nskb)->csum_start =
3218 skb_headroom(nskb) + doffset;
3222 nskb_frag = skb_shinfo(nskb)->frags;
3224 skb_copy_from_linear_data_offset(head_skb, offset,
3225 skb_put(nskb, hsize), hsize);
3227 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3230 while (pos < offset + len) {
3232 BUG_ON(skb_headlen(list_skb));
3235 nfrags = skb_shinfo(list_skb)->nr_frags;
3236 frag = skb_shinfo(list_skb)->frags;
3237 frag_skb = list_skb;
3241 list_skb = list_skb->next;
3244 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3246 net_warn_ratelimited(
3247 "skb_segment: too many frags: %u %u\n",
3252 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3256 __skb_frag_ref(nskb_frag);
3257 size = skb_frag_size(nskb_frag);
3260 nskb_frag->page_offset += offset - pos;
3261 skb_frag_size_sub(nskb_frag, offset - pos);
3264 skb_shinfo(nskb)->nr_frags++;
3266 if (pos + size <= offset + len) {
3271 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3279 nskb->data_len = len - hsize;
3280 nskb->len += nskb->data_len;
3281 nskb->truesize += nskb->data_len;
3285 if (skb_has_shared_frag(nskb)) {
3286 err = __skb_linearize(nskb);
3290 if (!nskb->remcsum_offload)
3291 nskb->ip_summed = CHECKSUM_NONE;
3292 SKB_GSO_CB(nskb)->csum =
3293 skb_checksum(nskb, doffset,
3294 nskb->len - doffset, 0);
3295 SKB_GSO_CB(nskb)->csum_start =
3296 skb_headroom(nskb) + doffset;
3298 } while ((offset += len) < head_skb->len);
3300 /* Some callers want to get the end of the list.
3301 * Put it in segs->prev to avoid walking the list.
3302 * (see validate_xmit_skb_list() for example)
3307 struct sk_buff *iter;
3308 int type = skb_shinfo(head_skb)->gso_type;
3309 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
3311 /* Update type to add partial and then remove dodgy if set */
3312 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
3313 type &= ~SKB_GSO_DODGY;
3315 /* Update GSO info and prepare to start updating headers on
3316 * our way back down the stack of protocols.
3318 for (iter = segs; iter; iter = iter->next) {
3319 skb_shinfo(iter)->gso_size = gso_size;
3320 skb_shinfo(iter)->gso_segs = partial_segs;
3321 skb_shinfo(iter)->gso_type = type;
3322 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
3325 if (tail->len - doffset <= gso_size)
3326 skb_shinfo(tail)->gso_size = 0;
3327 else if (tail != segs)
3328 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
3331 /* Following permits correct backpressure, for protocols
3332 * using skb_set_owner_w().
3333 * Idea is to tranfert ownership from head_skb to last segment.
3335 if (head_skb->destructor == sock_wfree) {
3336 swap(tail->truesize, head_skb->truesize);
3337 swap(tail->destructor, head_skb->destructor);
3338 swap(tail->sk, head_skb->sk);
3343 kfree_skb_list(segs);
3344 return ERR_PTR(err);
3346 EXPORT_SYMBOL_GPL(skb_segment);
3348 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3350 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3351 unsigned int offset = skb_gro_offset(skb);
3352 unsigned int headlen = skb_headlen(skb);
3353 unsigned int len = skb_gro_len(skb);
3354 struct sk_buff *lp, *p = *head;
3355 unsigned int delta_truesize;
3357 if (unlikely(p->len + len >= 65536))
3360 lp = NAPI_GRO_CB(p)->last;
3361 pinfo = skb_shinfo(lp);
3363 if (headlen <= offset) {
3366 int i = skbinfo->nr_frags;
3367 int nr_frags = pinfo->nr_frags + i;
3369 if (nr_frags > MAX_SKB_FRAGS)
3373 pinfo->nr_frags = nr_frags;
3374 skbinfo->nr_frags = 0;
3376 frag = pinfo->frags + nr_frags;
3377 frag2 = skbinfo->frags + i;
3382 frag->page_offset += offset;
3383 skb_frag_size_sub(frag, offset);
3385 /* all fragments truesize : remove (head size + sk_buff) */
3386 delta_truesize = skb->truesize -
3387 SKB_TRUESIZE(skb_end_offset(skb));
3389 skb->truesize -= skb->data_len;
3390 skb->len -= skb->data_len;
3393 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3395 } else if (skb->head_frag) {
3396 int nr_frags = pinfo->nr_frags;
3397 skb_frag_t *frag = pinfo->frags + nr_frags;
3398 struct page *page = virt_to_head_page(skb->head);
3399 unsigned int first_size = headlen - offset;
3400 unsigned int first_offset;
3402 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3405 first_offset = skb->data -
3406 (unsigned char *)page_address(page) +
3409 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3411 frag->page.p = page;
3412 frag->page_offset = first_offset;
3413 skb_frag_size_set(frag, first_size);
3415 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3416 /* We dont need to clear skbinfo->nr_frags here */
3418 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3419 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3424 delta_truesize = skb->truesize;
3425 if (offset > headlen) {
3426 unsigned int eat = offset - headlen;
3428 skbinfo->frags[0].page_offset += eat;
3429 skb_frag_size_sub(&skbinfo->frags[0], eat);
3430 skb->data_len -= eat;
3435 __skb_pull(skb, offset);
3437 if (NAPI_GRO_CB(p)->last == p)
3438 skb_shinfo(p)->frag_list = skb;
3440 NAPI_GRO_CB(p)->last->next = skb;
3441 NAPI_GRO_CB(p)->last = skb;
3442 __skb_header_release(skb);
3446 NAPI_GRO_CB(p)->count++;
3448 p->truesize += delta_truesize;
3451 lp->data_len += len;
3452 lp->truesize += delta_truesize;
3455 NAPI_GRO_CB(skb)->same_flow = 1;
3458 EXPORT_SYMBOL_GPL(skb_gro_receive);
3460 void __init skb_init(void)
3462 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3463 sizeof(struct sk_buff),
3465 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3467 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3468 sizeof(struct sk_buff_fclones),
3470 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3475 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3476 * @skb: Socket buffer containing the buffers to be mapped
3477 * @sg: The scatter-gather list to map into
3478 * @offset: The offset into the buffer's contents to start mapping
3479 * @len: Length of buffer space to be mapped
3481 * Fill the specified scatter-gather list with mappings/pointers into a
3482 * region of the buffer space attached to a socket buffer.
3485 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3487 int start = skb_headlen(skb);
3488 int i, copy = start - offset;
3489 struct sk_buff *frag_iter;
3495 sg_set_buf(sg, skb->data + offset, copy);
3497 if ((len -= copy) == 0)
3502 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3505 WARN_ON(start > offset + len);
3507 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3508 if ((copy = end - offset) > 0) {
3509 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3513 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3514 frag->page_offset+offset-start);
3523 skb_walk_frags(skb, frag_iter) {
3526 WARN_ON(start > offset + len);
3528 end = start + frag_iter->len;
3529 if ((copy = end - offset) > 0) {
3532 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3534 if ((len -= copy) == 0)
3544 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3545 * sglist without mark the sg which contain last skb data as the end.
3546 * So the caller can mannipulate sg list as will when padding new data after
3547 * the first call without calling sg_unmark_end to expend sg list.
3549 * Scenario to use skb_to_sgvec_nomark:
3551 * 2. skb_to_sgvec_nomark(payload1)
3552 * 3. skb_to_sgvec_nomark(payload2)
3554 * This is equivalent to:
3556 * 2. skb_to_sgvec(payload1)
3558 * 4. skb_to_sgvec(payload2)
3560 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3561 * is more preferable.
3563 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3564 int offset, int len)
3566 return __skb_to_sgvec(skb, sg, offset, len);
3568 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3570 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3572 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3574 sg_mark_end(&sg[nsg - 1]);
3578 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3581 * skb_cow_data - Check that a socket buffer's data buffers are writable
3582 * @skb: The socket buffer to check.
3583 * @tailbits: Amount of trailing space to be added
3584 * @trailer: Returned pointer to the skb where the @tailbits space begins
3586 * Make sure that the data buffers attached to a socket buffer are
3587 * writable. If they are not, private copies are made of the data buffers
3588 * and the socket buffer is set to use these instead.
3590 * If @tailbits is given, make sure that there is space to write @tailbits
3591 * bytes of data beyond current end of socket buffer. @trailer will be
3592 * set to point to the skb in which this space begins.
3594 * The number of scatterlist elements required to completely map the
3595 * COW'd and extended socket buffer will be returned.
3597 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3601 struct sk_buff *skb1, **skb_p;
3603 /* If skb is cloned or its head is paged, reallocate
3604 * head pulling out all the pages (pages are considered not writable
3605 * at the moment even if they are anonymous).
3607 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3608 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3611 /* Easy case. Most of packets will go this way. */
3612 if (!skb_has_frag_list(skb)) {
3613 /* A little of trouble, not enough of space for trailer.
3614 * This should not happen, when stack is tuned to generate
3615 * good frames. OK, on miss we reallocate and reserve even more
3616 * space, 128 bytes is fair. */
3618 if (skb_tailroom(skb) < tailbits &&
3619 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3627 /* Misery. We are in troubles, going to mincer fragments... */
3630 skb_p = &skb_shinfo(skb)->frag_list;
3633 while ((skb1 = *skb_p) != NULL) {
3636 /* The fragment is partially pulled by someone,
3637 * this can happen on input. Copy it and everything
3640 if (skb_shared(skb1))
3643 /* If the skb is the last, worry about trailer. */
3645 if (skb1->next == NULL && tailbits) {
3646 if (skb_shinfo(skb1)->nr_frags ||
3647 skb_has_frag_list(skb1) ||
3648 skb_tailroom(skb1) < tailbits)
3649 ntail = tailbits + 128;
3655 skb_shinfo(skb1)->nr_frags ||
3656 skb_has_frag_list(skb1)) {
3657 struct sk_buff *skb2;
3659 /* Fuck, we are miserable poor guys... */
3661 skb2 = skb_copy(skb1, GFP_ATOMIC);
3663 skb2 = skb_copy_expand(skb1,
3667 if (unlikely(skb2 == NULL))
3671 skb_set_owner_w(skb2, skb1->sk);
3673 /* Looking around. Are we still alive?
3674 * OK, link new skb, drop old one */
3676 skb2->next = skb1->next;
3683 skb_p = &skb1->next;
3688 EXPORT_SYMBOL_GPL(skb_cow_data);
3690 static void sock_rmem_free(struct sk_buff *skb)
3692 struct sock *sk = skb->sk;
3694 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3698 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3700 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3702 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3703 (unsigned int)sk->sk_rcvbuf)
3708 skb->destructor = sock_rmem_free;
3709 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3711 /* before exiting rcu section, make sure dst is refcounted */
3714 skb_queue_tail(&sk->sk_error_queue, skb);
3715 if (!sock_flag(sk, SOCK_DEAD))
3716 sk->sk_data_ready(sk);
3719 EXPORT_SYMBOL(sock_queue_err_skb);
3721 static bool is_icmp_err_skb(const struct sk_buff *skb)
3723 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
3724 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
3727 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3729 struct sk_buff_head *q = &sk->sk_error_queue;
3730 struct sk_buff *skb, *skb_next = NULL;
3731 bool icmp_next = false;
3732 unsigned long flags;
3734 spin_lock_irqsave(&q->lock, flags);
3735 skb = __skb_dequeue(q);
3736 if (skb && (skb_next = skb_peek(q)))
3737 icmp_next = is_icmp_err_skb(skb_next);
3738 spin_unlock_irqrestore(&q->lock, flags);
3740 if (is_icmp_err_skb(skb) && !icmp_next)
3744 sk->sk_error_report(sk);
3748 EXPORT_SYMBOL(sock_dequeue_err_skb);
3751 * skb_clone_sk - create clone of skb, and take reference to socket
3752 * @skb: the skb to clone
3754 * This function creates a clone of a buffer that holds a reference on
3755 * sk_refcnt. Buffers created via this function are meant to be
3756 * returned using sock_queue_err_skb, or free via kfree_skb.
3758 * When passing buffers allocated with this function to sock_queue_err_skb
3759 * it is necessary to wrap the call with sock_hold/sock_put in order to
3760 * prevent the socket from being released prior to being enqueued on
3761 * the sk_error_queue.
3763 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3765 struct sock *sk = skb->sk;
3766 struct sk_buff *clone;
3768 if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3771 clone = skb_clone(skb, GFP_ATOMIC);
3778 clone->destructor = sock_efree;
3782 EXPORT_SYMBOL(skb_clone_sk);
3784 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3788 struct sock_exterr_skb *serr;
3791 serr = SKB_EXT_ERR(skb);
3792 memset(serr, 0, sizeof(*serr));
3793 serr->ee.ee_errno = ENOMSG;
3794 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3795 serr->ee.ee_info = tstype;
3796 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3797 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3798 if (sk->sk_protocol == IPPROTO_TCP &&
3799 sk->sk_type == SOCK_STREAM)
3800 serr->ee.ee_data -= sk->sk_tskey;
3803 err = sock_queue_err_skb(sk, skb);
3809 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
3813 if (likely(sysctl_tstamp_allow_data || tsonly))
3816 read_lock_bh(&sk->sk_callback_lock);
3817 ret = sk->sk_socket && sk->sk_socket->file &&
3818 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
3819 read_unlock_bh(&sk->sk_callback_lock);
3823 void skb_complete_tx_timestamp(struct sk_buff *skb,
3824 struct skb_shared_hwtstamps *hwtstamps)
3826 struct sock *sk = skb->sk;
3828 if (!skb_may_tx_timestamp(sk, false))
3831 /* Take a reference to prevent skb_orphan() from freeing the socket,
3832 * but only if the socket refcount is not zero.
3834 if (likely(atomic_inc_not_zero(&sk->sk_refcnt))) {
3835 *skb_hwtstamps(skb) = *hwtstamps;
3836 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
3840 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3842 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3843 struct skb_shared_hwtstamps *hwtstamps,
3844 struct sock *sk, int tstype)
3846 struct sk_buff *skb;
3852 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
3853 if (!skb_may_tx_timestamp(sk, tsonly))
3858 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
3859 sk->sk_protocol == IPPROTO_TCP &&
3860 sk->sk_type == SOCK_STREAM)
3861 skb = tcp_get_timestamping_opt_stats(sk);
3864 skb = alloc_skb(0, GFP_ATOMIC);
3866 skb = skb_clone(orig_skb, GFP_ATOMIC);
3872 skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags;
3873 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
3877 *skb_hwtstamps(skb) = *hwtstamps;
3879 skb->tstamp = ktime_get_real();
3881 __skb_complete_tx_timestamp(skb, sk, tstype);
3883 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3885 void skb_tstamp_tx(struct sk_buff *orig_skb,
3886 struct skb_shared_hwtstamps *hwtstamps)
3888 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3891 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3893 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3895 struct sock *sk = skb->sk;
3896 struct sock_exterr_skb *serr;
3899 skb->wifi_acked_valid = 1;
3900 skb->wifi_acked = acked;
3902 serr = SKB_EXT_ERR(skb);
3903 memset(serr, 0, sizeof(*serr));
3904 serr->ee.ee_errno = ENOMSG;
3905 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3907 /* Take a reference to prevent skb_orphan() from freeing the socket,
3908 * but only if the socket refcount is not zero.
3910 if (likely(atomic_inc_not_zero(&sk->sk_refcnt))) {
3911 err = sock_queue_err_skb(sk, skb);
3917 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3920 * skb_partial_csum_set - set up and verify partial csum values for packet
3921 * @skb: the skb to set
3922 * @start: the number of bytes after skb->data to start checksumming.
3923 * @off: the offset from start to place the checksum.
3925 * For untrusted partially-checksummed packets, we need to make sure the values
3926 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3928 * This function checks and sets those values and skb->ip_summed: if this
3929 * returns false you should drop the packet.
3931 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3933 if (unlikely(start > skb_headlen(skb)) ||
3934 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3935 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3936 start, off, skb_headlen(skb));
3939 skb->ip_summed = CHECKSUM_PARTIAL;
3940 skb->csum_start = skb_headroom(skb) + start;
3941 skb->csum_offset = off;
3942 skb_set_transport_header(skb, start);
3945 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3947 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3950 if (skb_headlen(skb) >= len)
3953 /* If we need to pullup then pullup to the max, so we
3954 * won't need to do it again.
3959 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3962 if (skb_headlen(skb) < len)
3968 #define MAX_TCP_HDR_LEN (15 * 4)
3970 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3971 typeof(IPPROTO_IP) proto,
3978 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3979 off + MAX_TCP_HDR_LEN);
3980 if (!err && !skb_partial_csum_set(skb, off,
3981 offsetof(struct tcphdr,
3984 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3987 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3988 off + sizeof(struct udphdr));
3989 if (!err && !skb_partial_csum_set(skb, off,
3990 offsetof(struct udphdr,
3993 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3996 return ERR_PTR(-EPROTO);
3999 /* This value should be large enough to cover a tagged ethernet header plus
4000 * maximally sized IP and TCP or UDP headers.
4002 #define MAX_IP_HDR_LEN 128
4004 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4013 err = skb_maybe_pull_tail(skb,
4014 sizeof(struct iphdr),
4019 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4022 off = ip_hdrlen(skb);
4029 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4031 return PTR_ERR(csum);
4034 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4037 ip_hdr(skb)->protocol, 0);
4044 /* This value should be large enough to cover a tagged ethernet header plus
4045 * an IPv6 header, all options, and a maximal TCP or UDP header.
4047 #define MAX_IPV6_HDR_LEN 256
4049 #define OPT_HDR(type, skb, off) \
4050 (type *)(skb_network_header(skb) + (off))
4052 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4065 off = sizeof(struct ipv6hdr);
4067 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4071 nexthdr = ipv6_hdr(skb)->nexthdr;
4073 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4074 while (off <= len && !done) {
4076 case IPPROTO_DSTOPTS:
4077 case IPPROTO_HOPOPTS:
4078 case IPPROTO_ROUTING: {
4079 struct ipv6_opt_hdr *hp;
4081 err = skb_maybe_pull_tail(skb,
4083 sizeof(struct ipv6_opt_hdr),
4088 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4089 nexthdr = hp->nexthdr;
4090 off += ipv6_optlen(hp);
4094 struct ip_auth_hdr *hp;
4096 err = skb_maybe_pull_tail(skb,
4098 sizeof(struct ip_auth_hdr),
4103 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4104 nexthdr = hp->nexthdr;
4105 off += ipv6_authlen(hp);
4108 case IPPROTO_FRAGMENT: {
4109 struct frag_hdr *hp;
4111 err = skb_maybe_pull_tail(skb,
4113 sizeof(struct frag_hdr),
4118 hp = OPT_HDR(struct frag_hdr, skb, off);
4120 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4123 nexthdr = hp->nexthdr;
4124 off += sizeof(struct frag_hdr);
4135 if (!done || fragment)
4138 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4140 return PTR_ERR(csum);
4143 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4144 &ipv6_hdr(skb)->daddr,
4145 skb->len - off, nexthdr, 0);
4153 * skb_checksum_setup - set up partial checksum offset
4154 * @skb: the skb to set up
4155 * @recalculate: if true the pseudo-header checksum will be recalculated
4157 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4161 switch (skb->protocol) {
4162 case htons(ETH_P_IP):
4163 err = skb_checksum_setup_ipv4(skb, recalculate);
4166 case htons(ETH_P_IPV6):
4167 err = skb_checksum_setup_ipv6(skb, recalculate);
4177 EXPORT_SYMBOL(skb_checksum_setup);
4180 * skb_checksum_maybe_trim - maybe trims the given skb
4181 * @skb: the skb to check
4182 * @transport_len: the data length beyond the network header
4184 * Checks whether the given skb has data beyond the given transport length.
4185 * If so, returns a cloned skb trimmed to this transport length.
4186 * Otherwise returns the provided skb. Returns NULL in error cases
4187 * (e.g. transport_len exceeds skb length or out-of-memory).
4189 * Caller needs to set the skb transport header and free any returned skb if it
4190 * differs from the provided skb.
4192 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4193 unsigned int transport_len)
4195 struct sk_buff *skb_chk;
4196 unsigned int len = skb_transport_offset(skb) + transport_len;
4201 else if (skb->len == len)
4204 skb_chk = skb_clone(skb, GFP_ATOMIC);
4208 ret = pskb_trim_rcsum(skb_chk, len);
4218 * skb_checksum_trimmed - validate checksum of an skb
4219 * @skb: the skb to check
4220 * @transport_len: the data length beyond the network header
4221 * @skb_chkf: checksum function to use
4223 * Applies the given checksum function skb_chkf to the provided skb.
4224 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4226 * If the skb has data beyond the given transport length, then a
4227 * trimmed & cloned skb is checked and returned.
4229 * Caller needs to set the skb transport header and free any returned skb if it
4230 * differs from the provided skb.
4232 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4233 unsigned int transport_len,
4234 __sum16(*skb_chkf)(struct sk_buff *skb))
4236 struct sk_buff *skb_chk;
4237 unsigned int offset = skb_transport_offset(skb);
4240 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4244 if (!pskb_may_pull(skb_chk, offset))
4247 skb_pull_rcsum(skb_chk, offset);
4248 ret = skb_chkf(skb_chk);
4249 skb_push_rcsum(skb_chk, offset);
4257 if (skb_chk && skb_chk != skb)
4263 EXPORT_SYMBOL(skb_checksum_trimmed);
4265 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4267 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4270 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4272 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4275 skb_release_head_state(skb);
4276 kmem_cache_free(skbuff_head_cache, skb);
4281 EXPORT_SYMBOL(kfree_skb_partial);
4284 * skb_try_coalesce - try to merge skb to prior one
4286 * @from: buffer to add
4287 * @fragstolen: pointer to boolean
4288 * @delta_truesize: how much more was allocated than was requested
4290 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4291 bool *fragstolen, int *delta_truesize)
4293 int i, delta, len = from->len;
4295 *fragstolen = false;
4300 if (len <= skb_tailroom(to)) {
4302 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4303 *delta_truesize = 0;
4307 if (skb_has_frag_list(to) || skb_has_frag_list(from))
4310 if (skb_headlen(from) != 0) {
4312 unsigned int offset;
4314 if (skb_shinfo(to)->nr_frags +
4315 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4318 if (skb_head_is_locked(from))
4321 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4323 page = virt_to_head_page(from->head);
4324 offset = from->data - (unsigned char *)page_address(page);
4326 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4327 page, offset, skb_headlen(from));
4330 if (skb_shinfo(to)->nr_frags +
4331 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4334 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4337 WARN_ON_ONCE(delta < len);
4339 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4340 skb_shinfo(from)->frags,
4341 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4342 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4344 if (!skb_cloned(from))
4345 skb_shinfo(from)->nr_frags = 0;
4347 /* if the skb is not cloned this does nothing
4348 * since we set nr_frags to 0.
4350 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4351 skb_frag_ref(from, i);
4353 to->truesize += delta;
4355 to->data_len += len;
4357 *delta_truesize = delta;
4360 EXPORT_SYMBOL(skb_try_coalesce);
4363 * skb_scrub_packet - scrub an skb
4365 * @skb: buffer to clean
4366 * @xnet: packet is crossing netns
4368 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4369 * into/from a tunnel. Some information have to be cleared during these
4371 * skb_scrub_packet can also be used to clean a skb before injecting it in
4372 * another namespace (@xnet == true). We have to clear all information in the
4373 * skb that could impact namespace isolation.
4375 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4378 skb->pkt_type = PACKET_HOST;
4384 nf_reset_trace(skb);
4392 EXPORT_SYMBOL_GPL(skb_scrub_packet);
4395 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4399 * skb_gso_transport_seglen is used to determine the real size of the
4400 * individual segments, including Layer4 headers (TCP/UDP).
4402 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4404 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4406 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4407 unsigned int thlen = 0;
4409 if (skb->encapsulation) {
4410 thlen = skb_inner_transport_header(skb) -
4411 skb_transport_header(skb);
4413 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4414 thlen += inner_tcp_hdrlen(skb);
4415 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4416 thlen = tcp_hdrlen(skb);
4417 } else if (unlikely(shinfo->gso_type & SKB_GSO_SCTP)) {
4418 thlen = sizeof(struct sctphdr);
4420 /* UFO sets gso_size to the size of the fragmentation
4421 * payload, i.e. the size of the L4 (UDP) header is already
4424 return thlen + shinfo->gso_size;
4426 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4429 * skb_gso_validate_mtu - Return in case such skb fits a given MTU
4432 * @mtu: MTU to validate against
4434 * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU
4437 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu)
4439 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4440 const struct sk_buff *iter;
4443 hlen = skb_gso_network_seglen(skb);
4445 if (shinfo->gso_size != GSO_BY_FRAGS)
4448 /* Undo this so we can re-use header sizes */
4449 hlen -= GSO_BY_FRAGS;
4451 skb_walk_frags(skb, iter) {
4452 if (hlen + skb_headlen(iter) > mtu)
4458 EXPORT_SYMBOL_GPL(skb_gso_validate_mtu);
4460 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4462 if (skb_cow(skb, skb_headroom(skb)) < 0) {
4467 memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN,
4469 skb->mac_header += VLAN_HLEN;
4473 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4475 struct vlan_hdr *vhdr;
4478 if (unlikely(skb_vlan_tag_present(skb))) {
4479 /* vlan_tci is already set-up so leave this for another time */
4483 skb = skb_share_check(skb, GFP_ATOMIC);
4487 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4490 vhdr = (struct vlan_hdr *)skb->data;
4491 vlan_tci = ntohs(vhdr->h_vlan_TCI);
4492 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4494 skb_pull_rcsum(skb, VLAN_HLEN);
4495 vlan_set_encap_proto(skb, vhdr);
4497 skb = skb_reorder_vlan_header(skb);
4501 skb_reset_network_header(skb);
4502 skb_reset_transport_header(skb);
4503 skb_reset_mac_len(skb);
4511 EXPORT_SYMBOL(skb_vlan_untag);
4513 int skb_ensure_writable(struct sk_buff *skb, int write_len)
4515 if (!pskb_may_pull(skb, write_len))
4518 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4521 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4523 EXPORT_SYMBOL(skb_ensure_writable);
4525 /* remove VLAN header from packet and update csum accordingly.
4526 * expects a non skb_vlan_tag_present skb with a vlan tag payload
4528 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
4530 struct vlan_hdr *vhdr;
4531 int offset = skb->data - skb_mac_header(skb);
4534 if (WARN_ONCE(offset,
4535 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
4540 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
4544 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4546 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
4547 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
4549 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
4550 __skb_pull(skb, VLAN_HLEN);
4552 vlan_set_encap_proto(skb, vhdr);
4553 skb->mac_header += VLAN_HLEN;
4555 if (skb_network_offset(skb) < ETH_HLEN)
4556 skb_set_network_header(skb, ETH_HLEN);
4558 skb_reset_mac_len(skb);
4562 EXPORT_SYMBOL(__skb_vlan_pop);
4564 /* Pop a vlan tag either from hwaccel or from payload.
4565 * Expects skb->data at mac header.
4567 int skb_vlan_pop(struct sk_buff *skb)
4573 if (likely(skb_vlan_tag_present(skb))) {
4576 if (unlikely(!eth_type_vlan(skb->protocol)))
4579 err = __skb_vlan_pop(skb, &vlan_tci);
4583 /* move next vlan tag to hw accel tag */
4584 if (likely(!eth_type_vlan(skb->protocol)))
4587 vlan_proto = skb->protocol;
4588 err = __skb_vlan_pop(skb, &vlan_tci);
4592 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4595 EXPORT_SYMBOL(skb_vlan_pop);
4597 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
4598 * Expects skb->data at mac header.
4600 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
4602 if (skb_vlan_tag_present(skb)) {
4603 int offset = skb->data - skb_mac_header(skb);
4606 if (WARN_ONCE(offset,
4607 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
4612 err = __vlan_insert_tag(skb, skb->vlan_proto,
4613 skb_vlan_tag_get(skb));
4617 skb->protocol = skb->vlan_proto;
4618 skb->mac_len += VLAN_HLEN;
4620 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4622 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4625 EXPORT_SYMBOL(skb_vlan_push);
4628 * alloc_skb_with_frags - allocate skb with page frags
4630 * @header_len: size of linear part
4631 * @data_len: needed length in frags
4632 * @max_page_order: max page order desired.
4633 * @errcode: pointer to error code if any
4634 * @gfp_mask: allocation mask
4636 * This can be used to allocate a paged skb, given a maximal order for frags.
4638 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4639 unsigned long data_len,
4644 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4645 unsigned long chunk;
4646 struct sk_buff *skb;
4651 *errcode = -EMSGSIZE;
4652 /* Note this test could be relaxed, if we succeed to allocate
4653 * high order pages...
4655 if (npages > MAX_SKB_FRAGS)
4658 gfp_head = gfp_mask;
4659 if (gfp_head & __GFP_DIRECT_RECLAIM)
4660 gfp_head |= __GFP_REPEAT;
4662 *errcode = -ENOBUFS;
4663 skb = alloc_skb(header_len, gfp_head);
4667 skb->truesize += npages << PAGE_SHIFT;
4669 for (i = 0; npages > 0; i++) {
4670 int order = max_page_order;
4673 if (npages >= 1 << order) {
4674 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
4681 /* Do not retry other high order allocations */
4687 page = alloc_page(gfp_mask);
4691 chunk = min_t(unsigned long, data_len,
4692 PAGE_SIZE << order);
4693 skb_fill_page_desc(skb, i, page, 0, chunk);
4695 npages -= 1 << order;
4703 EXPORT_SYMBOL(alloc_skb_with_frags);
4705 /* carve out the first off bytes from skb when off < headlen */
4706 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
4707 const int headlen, gfp_t gfp_mask)
4710 int size = skb_end_offset(skb);
4711 int new_hlen = headlen - off;
4714 size = SKB_DATA_ALIGN(size);
4716 if (skb_pfmemalloc(skb))
4717 gfp_mask |= __GFP_MEMALLOC;
4718 data = kmalloc_reserve(size +
4719 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4720 gfp_mask, NUMA_NO_NODE, NULL);
4724 size = SKB_WITH_OVERHEAD(ksize(data));
4726 /* Copy real data, and all frags */
4727 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
4730 memcpy((struct skb_shared_info *)(data + size),
4732 offsetof(struct skb_shared_info,
4733 frags[skb_shinfo(skb)->nr_frags]));
4734 if (skb_cloned(skb)) {
4735 /* drop the old head gracefully */
4736 if (skb_orphan_frags(skb, gfp_mask)) {
4740 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
4741 skb_frag_ref(skb, i);
4742 if (skb_has_frag_list(skb))
4743 skb_clone_fraglist(skb);
4744 skb_release_data(skb);
4746 /* we can reuse existing recount- all we did was
4755 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4758 skb->end = skb->head + size;
4760 skb_set_tail_pointer(skb, skb_headlen(skb));
4761 skb_headers_offset_update(skb, 0);
4765 atomic_set(&skb_shinfo(skb)->dataref, 1);
4770 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
4772 /* carve out the first eat bytes from skb's frag_list. May recurse into
4775 static int pskb_carve_frag_list(struct sk_buff *skb,
4776 struct skb_shared_info *shinfo, int eat,
4779 struct sk_buff *list = shinfo->frag_list;
4780 struct sk_buff *clone = NULL;
4781 struct sk_buff *insp = NULL;
4785 pr_err("Not enough bytes to eat. Want %d\n", eat);
4788 if (list->len <= eat) {
4789 /* Eaten as whole. */
4794 /* Eaten partially. */
4795 if (skb_shared(list)) {
4796 clone = skb_clone(list, gfp_mask);
4802 /* This may be pulled without problems. */
4805 if (pskb_carve(list, eat, gfp_mask) < 0) {
4813 /* Free pulled out fragments. */
4814 while ((list = shinfo->frag_list) != insp) {
4815 shinfo->frag_list = list->next;
4818 /* And insert new clone at head. */
4821 shinfo->frag_list = clone;
4826 /* carve off first len bytes from skb. Split line (off) is in the
4827 * non-linear part of skb
4829 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
4830 int pos, gfp_t gfp_mask)
4833 int size = skb_end_offset(skb);
4835 const int nfrags = skb_shinfo(skb)->nr_frags;
4836 struct skb_shared_info *shinfo;
4838 size = SKB_DATA_ALIGN(size);
4840 if (skb_pfmemalloc(skb))
4841 gfp_mask |= __GFP_MEMALLOC;
4842 data = kmalloc_reserve(size +
4843 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4844 gfp_mask, NUMA_NO_NODE, NULL);
4848 size = SKB_WITH_OVERHEAD(ksize(data));
4850 memcpy((struct skb_shared_info *)(data + size),
4851 skb_shinfo(skb), offsetof(struct skb_shared_info,
4852 frags[skb_shinfo(skb)->nr_frags]));
4853 if (skb_orphan_frags(skb, gfp_mask)) {
4857 shinfo = (struct skb_shared_info *)(data + size);
4858 for (i = 0; i < nfrags; i++) {
4859 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
4861 if (pos + fsize > off) {
4862 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
4866 * We have two variants in this case:
4867 * 1. Move all the frag to the second
4868 * part, if it is possible. F.e.
4869 * this approach is mandatory for TUX,
4870 * where splitting is expensive.
4871 * 2. Split is accurately. We make this.
4873 shinfo->frags[0].page_offset += off - pos;
4874 skb_frag_size_sub(&shinfo->frags[0], off - pos);
4876 skb_frag_ref(skb, i);
4881 shinfo->nr_frags = k;
4882 if (skb_has_frag_list(skb))
4883 skb_clone_fraglist(skb);
4886 /* split line is in frag list */
4887 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
4889 skb_release_data(skb);
4894 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4897 skb->end = skb->head + size;
4899 skb_reset_tail_pointer(skb);
4900 skb_headers_offset_update(skb, 0);
4905 skb->data_len = skb->len;
4906 atomic_set(&skb_shinfo(skb)->dataref, 1);
4910 /* remove len bytes from the beginning of the skb */
4911 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
4913 int headlen = skb_headlen(skb);
4916 return pskb_carve_inside_header(skb, len, headlen, gfp);
4918 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
4921 /* Extract to_copy bytes starting at off from skb, and return this in
4924 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
4925 int to_copy, gfp_t gfp)
4927 struct sk_buff *clone = skb_clone(skb, gfp);
4932 if (pskb_carve(clone, off, gfp) < 0 ||
4933 pskb_trim(clone, to_copy)) {
4939 EXPORT_SYMBOL(pskb_extract);
4942 * skb_condense - try to get rid of fragments/frag_list if possible
4945 * Can be used to save memory before skb is added to a busy queue.
4946 * If packet has bytes in frags and enough tail room in skb->head,
4947 * pull all of them, so that we can free the frags right now and adjust
4950 * We do not reallocate skb->head thus can not fail.
4951 * Caller must re-evaluate skb->truesize if needed.
4953 void skb_condense(struct sk_buff *skb)
4955 if (skb->data_len) {
4956 if (skb->data_len > skb->end - skb->tail ||
4960 /* Nice, we can free page frag(s) right now */
4961 __pskb_pull_tail(skb, skb->data_len);
4963 /* At this point, skb->truesize might be over estimated,
4964 * because skb had a fragment, and fragments do not tell
4966 * When we pulled its content into skb->head, fragment
4967 * was freed, but __pskb_pull_tail() could not possibly
4968 * adjust skb->truesize, not knowing the frag truesize.
4970 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));