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 void skb_release_head_state(struct sk_buff *skb)
650 if (skb->destructor) {
652 skb->destructor(skb);
654 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
655 nf_conntrack_put(skb_nfct(skb));
657 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
658 nf_bridge_put(skb->nf_bridge);
662 /* Free everything but the sk_buff shell. */
663 static void skb_release_all(struct sk_buff *skb)
665 skb_release_head_state(skb);
666 if (likely(skb->head))
667 skb_release_data(skb);
671 * __kfree_skb - private function
674 * Free an sk_buff. Release anything attached to the buffer.
675 * Clean the state. This is an internal helper function. Users should
676 * always call kfree_skb
679 void __kfree_skb(struct sk_buff *skb)
681 skb_release_all(skb);
684 EXPORT_SYMBOL(__kfree_skb);
687 * kfree_skb - free an sk_buff
688 * @skb: buffer to free
690 * Drop a reference to the buffer and free it if the usage count has
693 void kfree_skb(struct sk_buff *skb)
698 trace_kfree_skb(skb, __builtin_return_address(0));
701 EXPORT_SYMBOL(kfree_skb);
703 void kfree_skb_list(struct sk_buff *segs)
706 struct sk_buff *next = segs->next;
712 EXPORT_SYMBOL(kfree_skb_list);
715 * skb_tx_error - report an sk_buff xmit error
716 * @skb: buffer that triggered an error
718 * Report xmit error if a device callback is tracking this skb.
719 * skb must be freed afterwards.
721 void skb_tx_error(struct sk_buff *skb)
723 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
724 struct ubuf_info *uarg;
726 uarg = skb_shinfo(skb)->destructor_arg;
728 uarg->callback(uarg, false);
729 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
732 EXPORT_SYMBOL(skb_tx_error);
735 * consume_skb - free an skbuff
736 * @skb: buffer to free
738 * Drop a ref to the buffer and free it if the usage count has hit zero
739 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
740 * is being dropped after a failure and notes that
742 void consume_skb(struct sk_buff *skb)
747 trace_consume_skb(skb);
750 EXPORT_SYMBOL(consume_skb);
753 * consume_stateless_skb - free an skbuff, assuming it is stateless
754 * @skb: buffer to free
756 * Works like consume_skb(), but this variant assumes that all the head
757 * states have been already dropped.
759 void consume_stateless_skb(struct sk_buff *skb)
764 trace_consume_skb(skb);
765 if (likely(skb->head))
766 skb_release_data(skb);
770 void __kfree_skb_flush(void)
772 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
774 /* flush skb_cache if containing objects */
776 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
782 static inline void _kfree_skb_defer(struct sk_buff *skb)
784 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
786 /* drop skb->head and call any destructors for packet */
787 skb_release_all(skb);
789 /* record skb to CPU local list */
790 nc->skb_cache[nc->skb_count++] = skb;
793 /* SLUB writes into objects when freeing */
797 /* flush skb_cache if it is filled */
798 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
799 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
804 void __kfree_skb_defer(struct sk_buff *skb)
806 _kfree_skb_defer(skb);
809 void napi_consume_skb(struct sk_buff *skb, int budget)
814 /* Zero budget indicate non-NAPI context called us, like netpoll */
815 if (unlikely(!budget)) {
816 dev_consume_skb_any(skb);
820 if (likely(atomic_read(&skb->users) == 1))
822 else if (likely(!atomic_dec_and_test(&skb->users)))
824 /* if reaching here SKB is ready to free */
825 trace_consume_skb(skb);
827 /* if SKB is a clone, don't handle this case */
828 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
833 _kfree_skb_defer(skb);
835 EXPORT_SYMBOL(napi_consume_skb);
837 /* Make sure a field is enclosed inside headers_start/headers_end section */
838 #define CHECK_SKB_FIELD(field) \
839 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
840 offsetof(struct sk_buff, headers_start)); \
841 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
842 offsetof(struct sk_buff, headers_end)); \
844 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
846 new->tstamp = old->tstamp;
847 /* We do not copy old->sk */
849 memcpy(new->cb, old->cb, sizeof(old->cb));
850 skb_dst_copy(new, old);
852 new->sp = secpath_get(old->sp);
854 __nf_copy(new, old, false);
856 /* Note : this field could be in headers_start/headers_end section
857 * It is not yet because we do not want to have a 16 bit hole
859 new->queue_mapping = old->queue_mapping;
861 memcpy(&new->headers_start, &old->headers_start,
862 offsetof(struct sk_buff, headers_end) -
863 offsetof(struct sk_buff, headers_start));
864 CHECK_SKB_FIELD(protocol);
865 CHECK_SKB_FIELD(csum);
866 CHECK_SKB_FIELD(hash);
867 CHECK_SKB_FIELD(priority);
868 CHECK_SKB_FIELD(skb_iif);
869 CHECK_SKB_FIELD(vlan_proto);
870 CHECK_SKB_FIELD(vlan_tci);
871 CHECK_SKB_FIELD(transport_header);
872 CHECK_SKB_FIELD(network_header);
873 CHECK_SKB_FIELD(mac_header);
874 CHECK_SKB_FIELD(inner_protocol);
875 CHECK_SKB_FIELD(inner_transport_header);
876 CHECK_SKB_FIELD(inner_network_header);
877 CHECK_SKB_FIELD(inner_mac_header);
878 CHECK_SKB_FIELD(mark);
879 #ifdef CONFIG_NETWORK_SECMARK
880 CHECK_SKB_FIELD(secmark);
882 #ifdef CONFIG_NET_RX_BUSY_POLL
883 CHECK_SKB_FIELD(napi_id);
886 CHECK_SKB_FIELD(sender_cpu);
888 #ifdef CONFIG_NET_SCHED
889 CHECK_SKB_FIELD(tc_index);
895 * You should not add any new code to this function. Add it to
896 * __copy_skb_header above instead.
898 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
900 #define C(x) n->x = skb->x
902 n->next = n->prev = NULL;
904 __copy_skb_header(n, skb);
909 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
912 n->destructor = NULL;
919 atomic_set(&n->users, 1);
921 atomic_inc(&(skb_shinfo(skb)->dataref));
929 * skb_morph - morph one skb into another
930 * @dst: the skb to receive the contents
931 * @src: the skb to supply the contents
933 * This is identical to skb_clone except that the target skb is
934 * supplied by the user.
936 * The target skb is returned upon exit.
938 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
940 skb_release_all(dst);
941 return __skb_clone(dst, src);
943 EXPORT_SYMBOL_GPL(skb_morph);
946 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
947 * @skb: the skb to modify
948 * @gfp_mask: allocation priority
950 * This must be called on SKBTX_DEV_ZEROCOPY skb.
951 * It will copy all frags into kernel and drop the reference
952 * to userspace pages.
954 * If this function is called from an interrupt gfp_mask() must be
957 * Returns 0 on success or a negative error code on failure
958 * to allocate kernel memory to copy to.
960 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
963 int num_frags = skb_shinfo(skb)->nr_frags;
964 struct page *page, *head = NULL;
965 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
967 for (i = 0; i < num_frags; i++) {
969 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
971 page = alloc_page(gfp_mask);
974 struct page *next = (struct page *)page_private(head);
980 vaddr = kmap_atomic(skb_frag_page(f));
981 memcpy(page_address(page),
982 vaddr + f->page_offset, skb_frag_size(f));
983 kunmap_atomic(vaddr);
984 set_page_private(page, (unsigned long)head);
988 /* skb frags release userspace buffers */
989 for (i = 0; i < num_frags; i++)
990 skb_frag_unref(skb, i);
992 uarg->callback(uarg, false);
994 /* skb frags point to kernel buffers */
995 for (i = num_frags - 1; i >= 0; i--) {
996 __skb_fill_page_desc(skb, i, head, 0,
997 skb_shinfo(skb)->frags[i].size);
998 head = (struct page *)page_private(head);
1001 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
1004 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1007 * skb_clone - duplicate an sk_buff
1008 * @skb: buffer to clone
1009 * @gfp_mask: allocation priority
1011 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1012 * copies share the same packet data but not structure. The new
1013 * buffer has a reference count of 1. If the allocation fails the
1014 * function returns %NULL otherwise the new buffer is returned.
1016 * If this function is called from an interrupt gfp_mask() must be
1020 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1022 struct sk_buff_fclones *fclones = container_of(skb,
1023 struct sk_buff_fclones,
1027 if (skb_orphan_frags(skb, gfp_mask))
1030 if (skb->fclone == SKB_FCLONE_ORIG &&
1031 atomic_read(&fclones->fclone_ref) == 1) {
1033 atomic_set(&fclones->fclone_ref, 2);
1035 if (skb_pfmemalloc(skb))
1036 gfp_mask |= __GFP_MEMALLOC;
1038 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1042 kmemcheck_annotate_bitfield(n, flags1);
1043 n->fclone = SKB_FCLONE_UNAVAILABLE;
1046 return __skb_clone(n, skb);
1048 EXPORT_SYMBOL(skb_clone);
1050 static void skb_headers_offset_update(struct sk_buff *skb, int off)
1052 /* Only adjust this if it actually is csum_start rather than csum */
1053 if (skb->ip_summed == CHECKSUM_PARTIAL)
1054 skb->csum_start += off;
1055 /* {transport,network,mac}_header and tail are relative to skb->head */
1056 skb->transport_header += off;
1057 skb->network_header += off;
1058 if (skb_mac_header_was_set(skb))
1059 skb->mac_header += off;
1060 skb->inner_transport_header += off;
1061 skb->inner_network_header += off;
1062 skb->inner_mac_header += off;
1065 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1067 __copy_skb_header(new, old);
1069 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1070 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1071 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1074 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1076 if (skb_pfmemalloc(skb))
1077 return SKB_ALLOC_RX;
1082 * skb_copy - create private copy of an sk_buff
1083 * @skb: buffer to copy
1084 * @gfp_mask: allocation priority
1086 * Make a copy of both an &sk_buff and its data. This is used when the
1087 * caller wishes to modify the data and needs a private copy of the
1088 * data to alter. Returns %NULL on failure or the pointer to the buffer
1089 * on success. The returned buffer has a reference count of 1.
1091 * As by-product this function converts non-linear &sk_buff to linear
1092 * one, so that &sk_buff becomes completely private and caller is allowed
1093 * to modify all the data of returned buffer. This means that this
1094 * function is not recommended for use in circumstances when only
1095 * header is going to be modified. Use pskb_copy() instead.
1098 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1100 int headerlen = skb_headroom(skb);
1101 unsigned int size = skb_end_offset(skb) + skb->data_len;
1102 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1103 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1108 /* Set the data pointer */
1109 skb_reserve(n, headerlen);
1110 /* Set the tail pointer and length */
1111 skb_put(n, skb->len);
1113 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1116 copy_skb_header(n, skb);
1119 EXPORT_SYMBOL(skb_copy);
1122 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1123 * @skb: buffer to copy
1124 * @headroom: headroom of new skb
1125 * @gfp_mask: allocation priority
1126 * @fclone: if true allocate the copy of the skb from the fclone
1127 * cache instead of the head cache; it is recommended to set this
1128 * to true for the cases where the copy will likely be cloned
1130 * Make a copy of both an &sk_buff and part of its data, located
1131 * in header. Fragmented data remain shared. This is used when
1132 * the caller wishes to modify only header of &sk_buff and needs
1133 * private copy of the header to alter. Returns %NULL on failure
1134 * or the pointer to the buffer on success.
1135 * The returned buffer has a reference count of 1.
1138 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1139 gfp_t gfp_mask, bool fclone)
1141 unsigned int size = skb_headlen(skb) + headroom;
1142 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1143 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1148 /* Set the data pointer */
1149 skb_reserve(n, headroom);
1150 /* Set the tail pointer and length */
1151 skb_put(n, skb_headlen(skb));
1152 /* Copy the bytes */
1153 skb_copy_from_linear_data(skb, n->data, n->len);
1155 n->truesize += skb->data_len;
1156 n->data_len = skb->data_len;
1159 if (skb_shinfo(skb)->nr_frags) {
1162 if (skb_orphan_frags(skb, gfp_mask)) {
1167 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1168 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1169 skb_frag_ref(skb, i);
1171 skb_shinfo(n)->nr_frags = i;
1174 if (skb_has_frag_list(skb)) {
1175 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1176 skb_clone_fraglist(n);
1179 copy_skb_header(n, skb);
1183 EXPORT_SYMBOL(__pskb_copy_fclone);
1186 * pskb_expand_head - reallocate header of &sk_buff
1187 * @skb: buffer to reallocate
1188 * @nhead: room to add at head
1189 * @ntail: room to add at tail
1190 * @gfp_mask: allocation priority
1192 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1193 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1194 * reference count of 1. Returns zero in the case of success or error,
1195 * if expansion failed. In the last case, &sk_buff is not changed.
1197 * All the pointers pointing into skb header may change and must be
1198 * reloaded after call to this function.
1201 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1204 int i, osize = skb_end_offset(skb);
1205 int size = osize + nhead + ntail;
1211 if (skb_shared(skb))
1214 size = SKB_DATA_ALIGN(size);
1216 if (skb_pfmemalloc(skb))
1217 gfp_mask |= __GFP_MEMALLOC;
1218 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1219 gfp_mask, NUMA_NO_NODE, NULL);
1222 size = SKB_WITH_OVERHEAD(ksize(data));
1224 /* Copy only real data... and, alas, header. This should be
1225 * optimized for the cases when header is void.
1227 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1229 memcpy((struct skb_shared_info *)(data + size),
1231 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1234 * if shinfo is shared we must drop the old head gracefully, but if it
1235 * is not we can just drop the old head and let the existing refcount
1236 * be since all we did is relocate the values
1238 if (skb_cloned(skb)) {
1239 /* copy this zero copy skb frags */
1240 if (skb_orphan_frags(skb, gfp_mask))
1242 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1243 skb_frag_ref(skb, i);
1245 if (skb_has_frag_list(skb))
1246 skb_clone_fraglist(skb);
1248 skb_release_data(skb);
1252 off = (data + nhead) - skb->head;
1257 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1261 skb->end = skb->head + size;
1264 skb_headers_offset_update(skb, nhead);
1268 atomic_set(&skb_shinfo(skb)->dataref, 1);
1270 /* It is not generally safe to change skb->truesize.
1271 * For the moment, we really care of rx path, or
1272 * when skb is orphaned (not attached to a socket).
1274 if (!skb->sk || skb->destructor == sock_edemux)
1275 skb->truesize += size - osize;
1284 EXPORT_SYMBOL(pskb_expand_head);
1286 /* Make private copy of skb with writable head and some headroom */
1288 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1290 struct sk_buff *skb2;
1291 int delta = headroom - skb_headroom(skb);
1294 skb2 = pskb_copy(skb, GFP_ATOMIC);
1296 skb2 = skb_clone(skb, GFP_ATOMIC);
1297 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1305 EXPORT_SYMBOL(skb_realloc_headroom);
1308 * skb_copy_expand - copy and expand sk_buff
1309 * @skb: buffer to copy
1310 * @newheadroom: new free bytes at head
1311 * @newtailroom: new free bytes at tail
1312 * @gfp_mask: allocation priority
1314 * Make a copy of both an &sk_buff and its data and while doing so
1315 * allocate additional space.
1317 * This is used when the caller wishes to modify the data and needs a
1318 * private copy of the data to alter as well as more space for new fields.
1319 * Returns %NULL on failure or the pointer to the buffer
1320 * on success. The returned buffer has a reference count of 1.
1322 * You must pass %GFP_ATOMIC as the allocation priority if this function
1323 * is called from an interrupt.
1325 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1326 int newheadroom, int newtailroom,
1330 * Allocate the copy buffer
1332 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1333 gfp_mask, skb_alloc_rx_flag(skb),
1335 int oldheadroom = skb_headroom(skb);
1336 int head_copy_len, head_copy_off;
1341 skb_reserve(n, newheadroom);
1343 /* Set the tail pointer and length */
1344 skb_put(n, skb->len);
1346 head_copy_len = oldheadroom;
1348 if (newheadroom <= head_copy_len)
1349 head_copy_len = newheadroom;
1351 head_copy_off = newheadroom - head_copy_len;
1353 /* Copy the linear header and data. */
1354 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1355 skb->len + head_copy_len))
1358 copy_skb_header(n, skb);
1360 skb_headers_offset_update(n, newheadroom - oldheadroom);
1364 EXPORT_SYMBOL(skb_copy_expand);
1367 * skb_pad - zero pad the tail of an skb
1368 * @skb: buffer to pad
1369 * @pad: space to pad
1371 * Ensure that a buffer is followed by a padding area that is zero
1372 * filled. Used by network drivers which may DMA or transfer data
1373 * beyond the buffer end onto the wire.
1375 * May return error in out of memory cases. The skb is freed on error.
1378 int skb_pad(struct sk_buff *skb, int pad)
1383 /* If the skbuff is non linear tailroom is always zero.. */
1384 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1385 memset(skb->data+skb->len, 0, pad);
1389 ntail = skb->data_len + pad - (skb->end - skb->tail);
1390 if (likely(skb_cloned(skb) || ntail > 0)) {
1391 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1396 /* FIXME: The use of this function with non-linear skb's really needs
1399 err = skb_linearize(skb);
1403 memset(skb->data + skb->len, 0, pad);
1410 EXPORT_SYMBOL(skb_pad);
1413 * pskb_put - add data to the tail of a potentially fragmented buffer
1414 * @skb: start of the buffer to use
1415 * @tail: tail fragment of the buffer to use
1416 * @len: amount of data to add
1418 * This function extends the used data area of the potentially
1419 * fragmented buffer. @tail must be the last fragment of @skb -- or
1420 * @skb itself. If this would exceed the total buffer size the kernel
1421 * will panic. A pointer to the first byte of the extra data is
1425 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1428 skb->data_len += len;
1431 return skb_put(tail, len);
1433 EXPORT_SYMBOL_GPL(pskb_put);
1436 * skb_put - add data to a buffer
1437 * @skb: buffer to use
1438 * @len: amount of data to add
1440 * This function extends the used data area of the buffer. If this would
1441 * exceed the total buffer size the kernel will panic. A pointer to the
1442 * first byte of the extra data is returned.
1444 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1446 unsigned char *tmp = skb_tail_pointer(skb);
1447 SKB_LINEAR_ASSERT(skb);
1450 if (unlikely(skb->tail > skb->end))
1451 skb_over_panic(skb, len, __builtin_return_address(0));
1454 EXPORT_SYMBOL(skb_put);
1457 * skb_push - add data to the start of a buffer
1458 * @skb: buffer to use
1459 * @len: amount of data to add
1461 * This function extends the used data area of the buffer at the buffer
1462 * start. If this would exceed the total buffer headroom the kernel will
1463 * panic. A pointer to the first byte of the extra data is returned.
1465 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1469 if (unlikely(skb->data<skb->head))
1470 skb_under_panic(skb, len, __builtin_return_address(0));
1473 EXPORT_SYMBOL(skb_push);
1476 * skb_pull - remove data from the start of a buffer
1477 * @skb: buffer to use
1478 * @len: amount of data to remove
1480 * This function removes data from the start of a buffer, returning
1481 * the memory to the headroom. A pointer to the next data in the buffer
1482 * is returned. Once the data has been pulled future pushes will overwrite
1485 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1487 return skb_pull_inline(skb, len);
1489 EXPORT_SYMBOL(skb_pull);
1492 * skb_trim - remove end from a buffer
1493 * @skb: buffer to alter
1496 * Cut the length of a buffer down by removing data from the tail. If
1497 * the buffer is already under the length specified it is not modified.
1498 * The skb must be linear.
1500 void skb_trim(struct sk_buff *skb, unsigned int len)
1503 __skb_trim(skb, len);
1505 EXPORT_SYMBOL(skb_trim);
1507 /* Trims skb to length len. It can change skb pointers.
1510 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1512 struct sk_buff **fragp;
1513 struct sk_buff *frag;
1514 int offset = skb_headlen(skb);
1515 int nfrags = skb_shinfo(skb)->nr_frags;
1519 if (skb_cloned(skb) &&
1520 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1527 for (; i < nfrags; i++) {
1528 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1535 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1538 skb_shinfo(skb)->nr_frags = i;
1540 for (; i < nfrags; i++)
1541 skb_frag_unref(skb, i);
1543 if (skb_has_frag_list(skb))
1544 skb_drop_fraglist(skb);
1548 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1549 fragp = &frag->next) {
1550 int end = offset + frag->len;
1552 if (skb_shared(frag)) {
1553 struct sk_buff *nfrag;
1555 nfrag = skb_clone(frag, GFP_ATOMIC);
1556 if (unlikely(!nfrag))
1559 nfrag->next = frag->next;
1571 unlikely((err = pskb_trim(frag, len - offset))))
1575 skb_drop_list(&frag->next);
1580 if (len > skb_headlen(skb)) {
1581 skb->data_len -= skb->len - len;
1586 skb_set_tail_pointer(skb, len);
1589 if (!skb->sk || skb->destructor == sock_edemux)
1593 EXPORT_SYMBOL(___pskb_trim);
1596 * __pskb_pull_tail - advance tail of skb header
1597 * @skb: buffer to reallocate
1598 * @delta: number of bytes to advance tail
1600 * The function makes a sense only on a fragmented &sk_buff,
1601 * it expands header moving its tail forward and copying necessary
1602 * data from fragmented part.
1604 * &sk_buff MUST have reference count of 1.
1606 * Returns %NULL (and &sk_buff does not change) if pull failed
1607 * or value of new tail of skb in the case of success.
1609 * All the pointers pointing into skb header may change and must be
1610 * reloaded after call to this function.
1613 /* Moves tail of skb head forward, copying data from fragmented part,
1614 * when it is necessary.
1615 * 1. It may fail due to malloc failure.
1616 * 2. It may change skb pointers.
1618 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1620 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1622 /* If skb has not enough free space at tail, get new one
1623 * plus 128 bytes for future expansions. If we have enough
1624 * room at tail, reallocate without expansion only if skb is cloned.
1626 int i, k, eat = (skb->tail + delta) - skb->end;
1628 if (eat > 0 || skb_cloned(skb)) {
1629 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1634 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1637 /* Optimization: no fragments, no reasons to preestimate
1638 * size of pulled pages. Superb.
1640 if (!skb_has_frag_list(skb))
1643 /* Estimate size of pulled pages. */
1645 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1646 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1653 /* If we need update frag list, we are in troubles.
1654 * Certainly, it possible to add an offset to skb data,
1655 * but taking into account that pulling is expected to
1656 * be very rare operation, it is worth to fight against
1657 * further bloating skb head and crucify ourselves here instead.
1658 * Pure masohism, indeed. 8)8)
1661 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1662 struct sk_buff *clone = NULL;
1663 struct sk_buff *insp = NULL;
1668 if (list->len <= eat) {
1669 /* Eaten as whole. */
1674 /* Eaten partially. */
1676 if (skb_shared(list)) {
1677 /* Sucks! We need to fork list. :-( */
1678 clone = skb_clone(list, GFP_ATOMIC);
1684 /* This may be pulled without
1688 if (!pskb_pull(list, eat)) {
1696 /* Free pulled out fragments. */
1697 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1698 skb_shinfo(skb)->frag_list = list->next;
1701 /* And insert new clone at head. */
1704 skb_shinfo(skb)->frag_list = clone;
1707 /* Success! Now we may commit changes to skb data. */
1712 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1713 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1716 skb_frag_unref(skb, i);
1719 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1721 skb_shinfo(skb)->frags[k].page_offset += eat;
1722 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1728 skb_shinfo(skb)->nr_frags = k;
1731 skb->data_len -= delta;
1733 return skb_tail_pointer(skb);
1735 EXPORT_SYMBOL(__pskb_pull_tail);
1738 * skb_copy_bits - copy bits from skb to kernel buffer
1740 * @offset: offset in source
1741 * @to: destination buffer
1742 * @len: number of bytes to copy
1744 * Copy the specified number of bytes from the source skb to the
1745 * destination buffer.
1748 * If its prototype is ever changed,
1749 * check arch/{*}/net/{*}.S files,
1750 * since it is called from BPF assembly code.
1752 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1754 int start = skb_headlen(skb);
1755 struct sk_buff *frag_iter;
1758 if (offset > (int)skb->len - len)
1762 if ((copy = start - offset) > 0) {
1765 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1766 if ((len -= copy) == 0)
1772 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1774 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1776 WARN_ON(start > offset + len);
1778 end = start + skb_frag_size(f);
1779 if ((copy = end - offset) > 0) {
1785 vaddr = kmap_atomic(skb_frag_page(f));
1787 vaddr + f->page_offset + offset - start,
1789 kunmap_atomic(vaddr);
1791 if ((len -= copy) == 0)
1799 skb_walk_frags(skb, frag_iter) {
1802 WARN_ON(start > offset + len);
1804 end = start + frag_iter->len;
1805 if ((copy = end - offset) > 0) {
1808 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1810 if ((len -= copy) == 0)
1824 EXPORT_SYMBOL(skb_copy_bits);
1827 * Callback from splice_to_pipe(), if we need to release some pages
1828 * at the end of the spd in case we error'ed out in filling the pipe.
1830 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1832 put_page(spd->pages[i]);
1835 static struct page *linear_to_page(struct page *page, unsigned int *len,
1836 unsigned int *offset,
1839 struct page_frag *pfrag = sk_page_frag(sk);
1841 if (!sk_page_frag_refill(sk, pfrag))
1844 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1846 memcpy(page_address(pfrag->page) + pfrag->offset,
1847 page_address(page) + *offset, *len);
1848 *offset = pfrag->offset;
1849 pfrag->offset += *len;
1854 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1856 unsigned int offset)
1858 return spd->nr_pages &&
1859 spd->pages[spd->nr_pages - 1] == page &&
1860 (spd->partial[spd->nr_pages - 1].offset +
1861 spd->partial[spd->nr_pages - 1].len == offset);
1865 * Fill page/offset/length into spd, if it can hold more pages.
1867 static bool spd_fill_page(struct splice_pipe_desc *spd,
1868 struct pipe_inode_info *pipe, struct page *page,
1869 unsigned int *len, unsigned int offset,
1873 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1877 page = linear_to_page(page, len, &offset, sk);
1881 if (spd_can_coalesce(spd, page, offset)) {
1882 spd->partial[spd->nr_pages - 1].len += *len;
1886 spd->pages[spd->nr_pages] = page;
1887 spd->partial[spd->nr_pages].len = *len;
1888 spd->partial[spd->nr_pages].offset = offset;
1894 static bool __splice_segment(struct page *page, unsigned int poff,
1895 unsigned int plen, unsigned int *off,
1897 struct splice_pipe_desc *spd, bool linear,
1899 struct pipe_inode_info *pipe)
1904 /* skip this segment if already processed */
1910 /* ignore any bits we already processed */
1916 unsigned int flen = min(*len, plen);
1918 if (spd_fill_page(spd, pipe, page, &flen, poff,
1924 } while (*len && plen);
1930 * Map linear and fragment data from the skb to spd. It reports true if the
1931 * pipe is full or if we already spliced the requested length.
1933 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1934 unsigned int *offset, unsigned int *len,
1935 struct splice_pipe_desc *spd, struct sock *sk)
1938 struct sk_buff *iter;
1940 /* map the linear part :
1941 * If skb->head_frag is set, this 'linear' part is backed by a
1942 * fragment, and if the head is not shared with any clones then
1943 * we can avoid a copy since we own the head portion of this page.
1945 if (__splice_segment(virt_to_page(skb->data),
1946 (unsigned long) skb->data & (PAGE_SIZE - 1),
1949 skb_head_is_locked(skb),
1954 * then map the fragments
1956 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1957 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1959 if (__splice_segment(skb_frag_page(f),
1960 f->page_offset, skb_frag_size(f),
1961 offset, len, spd, false, sk, pipe))
1965 skb_walk_frags(skb, iter) {
1966 if (*offset >= iter->len) {
1967 *offset -= iter->len;
1970 /* __skb_splice_bits() only fails if the output has no room
1971 * left, so no point in going over the frag_list for the error
1974 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
1982 * Map data from the skb to a pipe. Should handle both the linear part,
1983 * the fragments, and the frag list.
1985 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
1986 struct pipe_inode_info *pipe, unsigned int tlen,
1989 struct partial_page partial[MAX_SKB_FRAGS];
1990 struct page *pages[MAX_SKB_FRAGS];
1991 struct splice_pipe_desc spd = {
1994 .nr_pages_max = MAX_SKB_FRAGS,
1995 .ops = &nosteal_pipe_buf_ops,
1996 .spd_release = sock_spd_release,
2000 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2003 ret = splice_to_pipe(pipe, &spd);
2007 EXPORT_SYMBOL_GPL(skb_splice_bits);
2010 * skb_store_bits - store bits from kernel buffer to skb
2011 * @skb: destination buffer
2012 * @offset: offset in destination
2013 * @from: source buffer
2014 * @len: number of bytes to copy
2016 * Copy the specified number of bytes from the source buffer to the
2017 * destination skb. This function handles all the messy bits of
2018 * traversing fragment lists and such.
2021 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2023 int start = skb_headlen(skb);
2024 struct sk_buff *frag_iter;
2027 if (offset > (int)skb->len - len)
2030 if ((copy = start - offset) > 0) {
2033 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2034 if ((len -= copy) == 0)
2040 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2041 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2044 WARN_ON(start > offset + len);
2046 end = start + skb_frag_size(frag);
2047 if ((copy = end - offset) > 0) {
2053 vaddr = kmap_atomic(skb_frag_page(frag));
2054 memcpy(vaddr + frag->page_offset + offset - start,
2056 kunmap_atomic(vaddr);
2058 if ((len -= copy) == 0)
2066 skb_walk_frags(skb, frag_iter) {
2069 WARN_ON(start > offset + len);
2071 end = start + frag_iter->len;
2072 if ((copy = end - offset) > 0) {
2075 if (skb_store_bits(frag_iter, offset - start,
2078 if ((len -= copy) == 0)
2091 EXPORT_SYMBOL(skb_store_bits);
2093 /* Checksum skb data. */
2094 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2095 __wsum csum, const struct skb_checksum_ops *ops)
2097 int start = skb_headlen(skb);
2098 int i, copy = start - offset;
2099 struct sk_buff *frag_iter;
2102 /* Checksum header. */
2106 csum = ops->update(skb->data + offset, copy, csum);
2107 if ((len -= copy) == 0)
2113 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2115 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2117 WARN_ON(start > offset + len);
2119 end = start + skb_frag_size(frag);
2120 if ((copy = end - offset) > 0) {
2126 vaddr = kmap_atomic(skb_frag_page(frag));
2127 csum2 = ops->update(vaddr + frag->page_offset +
2128 offset - start, copy, 0);
2129 kunmap_atomic(vaddr);
2130 csum = ops->combine(csum, csum2, pos, copy);
2139 skb_walk_frags(skb, frag_iter) {
2142 WARN_ON(start > offset + len);
2144 end = start + frag_iter->len;
2145 if ((copy = end - offset) > 0) {
2149 csum2 = __skb_checksum(frag_iter, offset - start,
2151 csum = ops->combine(csum, csum2, pos, copy);
2152 if ((len -= copy) == 0)
2163 EXPORT_SYMBOL(__skb_checksum);
2165 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2166 int len, __wsum csum)
2168 const struct skb_checksum_ops ops = {
2169 .update = csum_partial_ext,
2170 .combine = csum_block_add_ext,
2173 return __skb_checksum(skb, offset, len, csum, &ops);
2175 EXPORT_SYMBOL(skb_checksum);
2177 /* Both of above in one bottle. */
2179 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2180 u8 *to, int len, __wsum csum)
2182 int start = skb_headlen(skb);
2183 int i, copy = start - offset;
2184 struct sk_buff *frag_iter;
2191 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2193 if ((len -= copy) == 0)
2200 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2203 WARN_ON(start > offset + len);
2205 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2206 if ((copy = end - offset) > 0) {
2209 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2213 vaddr = kmap_atomic(skb_frag_page(frag));
2214 csum2 = csum_partial_copy_nocheck(vaddr +
2218 kunmap_atomic(vaddr);
2219 csum = csum_block_add(csum, csum2, pos);
2229 skb_walk_frags(skb, frag_iter) {
2233 WARN_ON(start > offset + len);
2235 end = start + frag_iter->len;
2236 if ((copy = end - offset) > 0) {
2239 csum2 = skb_copy_and_csum_bits(frag_iter,
2242 csum = csum_block_add(csum, csum2, pos);
2243 if ((len -= copy) == 0)
2254 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2256 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2258 net_warn_ratelimited(
2259 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2264 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2265 int offset, int len)
2267 net_warn_ratelimited(
2268 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2273 static const struct skb_checksum_ops default_crc32c_ops = {
2274 .update = warn_crc32c_csum_update,
2275 .combine = warn_crc32c_csum_combine,
2278 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2279 &default_crc32c_ops;
2280 EXPORT_SYMBOL(crc32c_csum_stub);
2283 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2284 * @from: source buffer
2286 * Calculates the amount of linear headroom needed in the 'to' skb passed
2287 * into skb_zerocopy().
2290 skb_zerocopy_headlen(const struct sk_buff *from)
2292 unsigned int hlen = 0;
2294 if (!from->head_frag ||
2295 skb_headlen(from) < L1_CACHE_BYTES ||
2296 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2297 hlen = skb_headlen(from);
2299 if (skb_has_frag_list(from))
2304 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2307 * skb_zerocopy - Zero copy skb to skb
2308 * @to: destination buffer
2309 * @from: source buffer
2310 * @len: number of bytes to copy from source buffer
2311 * @hlen: size of linear headroom in destination buffer
2313 * Copies up to `len` bytes from `from` to `to` by creating references
2314 * to the frags in the source buffer.
2316 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2317 * headroom in the `to` buffer.
2320 * 0: everything is OK
2321 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2322 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2325 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2328 int plen = 0; /* length of skb->head fragment */
2331 unsigned int offset;
2333 BUG_ON(!from->head_frag && !hlen);
2335 /* dont bother with small payloads */
2336 if (len <= skb_tailroom(to))
2337 return skb_copy_bits(from, 0, skb_put(to, len), len);
2340 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2345 plen = min_t(int, skb_headlen(from), len);
2347 page = virt_to_head_page(from->head);
2348 offset = from->data - (unsigned char *)page_address(page);
2349 __skb_fill_page_desc(to, 0, page, offset, plen);
2356 to->truesize += len + plen;
2357 to->len += len + plen;
2358 to->data_len += len + plen;
2360 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2365 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2368 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2369 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2370 len -= skb_shinfo(to)->frags[j].size;
2371 skb_frag_ref(to, j);
2374 skb_shinfo(to)->nr_frags = j;
2378 EXPORT_SYMBOL_GPL(skb_zerocopy);
2380 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2385 if (skb->ip_summed == CHECKSUM_PARTIAL)
2386 csstart = skb_checksum_start_offset(skb);
2388 csstart = skb_headlen(skb);
2390 BUG_ON(csstart > skb_headlen(skb));
2392 skb_copy_from_linear_data(skb, to, csstart);
2395 if (csstart != skb->len)
2396 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2397 skb->len - csstart, 0);
2399 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2400 long csstuff = csstart + skb->csum_offset;
2402 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2405 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2408 * skb_dequeue - remove from the head of the queue
2409 * @list: list to dequeue from
2411 * Remove the head of the list. The list lock is taken so the function
2412 * may be used safely with other locking list functions. The head item is
2413 * returned or %NULL if the list is empty.
2416 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2418 unsigned long flags;
2419 struct sk_buff *result;
2421 spin_lock_irqsave(&list->lock, flags);
2422 result = __skb_dequeue(list);
2423 spin_unlock_irqrestore(&list->lock, flags);
2426 EXPORT_SYMBOL(skb_dequeue);
2429 * skb_dequeue_tail - remove from the tail of the queue
2430 * @list: list to dequeue from
2432 * Remove the tail of the list. The list lock is taken so the function
2433 * may be used safely with other locking list functions. The tail item is
2434 * returned or %NULL if the list is empty.
2436 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2438 unsigned long flags;
2439 struct sk_buff *result;
2441 spin_lock_irqsave(&list->lock, flags);
2442 result = __skb_dequeue_tail(list);
2443 spin_unlock_irqrestore(&list->lock, flags);
2446 EXPORT_SYMBOL(skb_dequeue_tail);
2449 * skb_queue_purge - empty a list
2450 * @list: list to empty
2452 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2453 * the list and one reference dropped. This function takes the list
2454 * lock and is atomic with respect to other list locking functions.
2456 void skb_queue_purge(struct sk_buff_head *list)
2458 struct sk_buff *skb;
2459 while ((skb = skb_dequeue(list)) != NULL)
2462 EXPORT_SYMBOL(skb_queue_purge);
2465 * skb_rbtree_purge - empty a skb rbtree
2466 * @root: root of the rbtree to empty
2468 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
2469 * the list and one reference dropped. This function does not take
2470 * any lock. Synchronization should be handled by the caller (e.g., TCP
2471 * out-of-order queue is protected by the socket lock).
2473 void skb_rbtree_purge(struct rb_root *root)
2475 struct sk_buff *skb, *next;
2477 rbtree_postorder_for_each_entry_safe(skb, next, root, rbnode)
2484 * skb_queue_head - queue a buffer at the list head
2485 * @list: list to use
2486 * @newsk: buffer to queue
2488 * Queue a buffer at the start of the list. This function takes the
2489 * list lock and can be used safely with other locking &sk_buff functions
2492 * A buffer cannot be placed on two lists at the same time.
2494 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2496 unsigned long flags;
2498 spin_lock_irqsave(&list->lock, flags);
2499 __skb_queue_head(list, newsk);
2500 spin_unlock_irqrestore(&list->lock, flags);
2502 EXPORT_SYMBOL(skb_queue_head);
2505 * skb_queue_tail - queue a buffer at the list tail
2506 * @list: list to use
2507 * @newsk: buffer to queue
2509 * Queue a buffer at the tail of the list. This function takes the
2510 * list lock and can be used safely with other locking &sk_buff functions
2513 * A buffer cannot be placed on two lists at the same time.
2515 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2517 unsigned long flags;
2519 spin_lock_irqsave(&list->lock, flags);
2520 __skb_queue_tail(list, newsk);
2521 spin_unlock_irqrestore(&list->lock, flags);
2523 EXPORT_SYMBOL(skb_queue_tail);
2526 * skb_unlink - remove a buffer from a list
2527 * @skb: buffer to remove
2528 * @list: list to use
2530 * Remove a packet from a list. The list locks are taken and this
2531 * function is atomic with respect to other list locked calls
2533 * You must know what list the SKB is on.
2535 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2537 unsigned long flags;
2539 spin_lock_irqsave(&list->lock, flags);
2540 __skb_unlink(skb, list);
2541 spin_unlock_irqrestore(&list->lock, flags);
2543 EXPORT_SYMBOL(skb_unlink);
2546 * skb_append - append a buffer
2547 * @old: buffer to insert after
2548 * @newsk: buffer to insert
2549 * @list: list to use
2551 * Place a packet after a given packet in a list. The list locks are taken
2552 * and this function is atomic with respect to other list locked calls.
2553 * A buffer cannot be placed on two lists at the same time.
2555 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2557 unsigned long flags;
2559 spin_lock_irqsave(&list->lock, flags);
2560 __skb_queue_after(list, old, newsk);
2561 spin_unlock_irqrestore(&list->lock, flags);
2563 EXPORT_SYMBOL(skb_append);
2566 * skb_insert - insert a buffer
2567 * @old: buffer to insert before
2568 * @newsk: buffer to insert
2569 * @list: list to use
2571 * Place a packet before a given packet in a list. The list locks are
2572 * taken and this function is atomic with respect to other list locked
2575 * A buffer cannot be placed on two lists at the same time.
2577 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2579 unsigned long flags;
2581 spin_lock_irqsave(&list->lock, flags);
2582 __skb_insert(newsk, old->prev, old, list);
2583 spin_unlock_irqrestore(&list->lock, flags);
2585 EXPORT_SYMBOL(skb_insert);
2587 static inline void skb_split_inside_header(struct sk_buff *skb,
2588 struct sk_buff* skb1,
2589 const u32 len, const int pos)
2593 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2595 /* And move data appendix as is. */
2596 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2597 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2599 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2600 skb_shinfo(skb)->nr_frags = 0;
2601 skb1->data_len = skb->data_len;
2602 skb1->len += skb1->data_len;
2605 skb_set_tail_pointer(skb, len);
2608 static inline void skb_split_no_header(struct sk_buff *skb,
2609 struct sk_buff* skb1,
2610 const u32 len, int pos)
2613 const int nfrags = skb_shinfo(skb)->nr_frags;
2615 skb_shinfo(skb)->nr_frags = 0;
2616 skb1->len = skb1->data_len = skb->len - len;
2618 skb->data_len = len - pos;
2620 for (i = 0; i < nfrags; i++) {
2621 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2623 if (pos + size > len) {
2624 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2628 * We have two variants in this case:
2629 * 1. Move all the frag to the second
2630 * part, if it is possible. F.e.
2631 * this approach is mandatory for TUX,
2632 * where splitting is expensive.
2633 * 2. Split is accurately. We make this.
2635 skb_frag_ref(skb, i);
2636 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2637 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2638 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2639 skb_shinfo(skb)->nr_frags++;
2643 skb_shinfo(skb)->nr_frags++;
2646 skb_shinfo(skb1)->nr_frags = k;
2650 * skb_split - Split fragmented skb to two parts at length len.
2651 * @skb: the buffer to split
2652 * @skb1: the buffer to receive the second part
2653 * @len: new length for skb
2655 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2657 int pos = skb_headlen(skb);
2659 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
2661 if (len < pos) /* Split line is inside header. */
2662 skb_split_inside_header(skb, skb1, len, pos);
2663 else /* Second chunk has no header, nothing to copy. */
2664 skb_split_no_header(skb, skb1, len, pos);
2666 EXPORT_SYMBOL(skb_split);
2668 /* Shifting from/to a cloned skb is a no-go.
2670 * Caller cannot keep skb_shinfo related pointers past calling here!
2672 static int skb_prepare_for_shift(struct sk_buff *skb)
2674 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2678 * skb_shift - Shifts paged data partially from skb to another
2679 * @tgt: buffer into which tail data gets added
2680 * @skb: buffer from which the paged data comes from
2681 * @shiftlen: shift up to this many bytes
2683 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2684 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2685 * It's up to caller to free skb if everything was shifted.
2687 * If @tgt runs out of frags, the whole operation is aborted.
2689 * Skb cannot include anything else but paged data while tgt is allowed
2690 * to have non-paged data as well.
2692 * TODO: full sized shift could be optimized but that would need
2693 * specialized skb free'er to handle frags without up-to-date nr_frags.
2695 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2697 int from, to, merge, todo;
2698 struct skb_frag_struct *fragfrom, *fragto;
2700 BUG_ON(shiftlen > skb->len);
2702 if (skb_headlen(skb))
2707 to = skb_shinfo(tgt)->nr_frags;
2708 fragfrom = &skb_shinfo(skb)->frags[from];
2710 /* Actual merge is delayed until the point when we know we can
2711 * commit all, so that we don't have to undo partial changes
2714 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2715 fragfrom->page_offset)) {
2720 todo -= skb_frag_size(fragfrom);
2722 if (skb_prepare_for_shift(skb) ||
2723 skb_prepare_for_shift(tgt))
2726 /* All previous frag pointers might be stale! */
2727 fragfrom = &skb_shinfo(skb)->frags[from];
2728 fragto = &skb_shinfo(tgt)->frags[merge];
2730 skb_frag_size_add(fragto, shiftlen);
2731 skb_frag_size_sub(fragfrom, shiftlen);
2732 fragfrom->page_offset += shiftlen;
2740 /* Skip full, not-fitting skb to avoid expensive operations */
2741 if ((shiftlen == skb->len) &&
2742 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2745 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2748 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2749 if (to == MAX_SKB_FRAGS)
2752 fragfrom = &skb_shinfo(skb)->frags[from];
2753 fragto = &skb_shinfo(tgt)->frags[to];
2755 if (todo >= skb_frag_size(fragfrom)) {
2756 *fragto = *fragfrom;
2757 todo -= skb_frag_size(fragfrom);
2762 __skb_frag_ref(fragfrom);
2763 fragto->page = fragfrom->page;
2764 fragto->page_offset = fragfrom->page_offset;
2765 skb_frag_size_set(fragto, todo);
2767 fragfrom->page_offset += todo;
2768 skb_frag_size_sub(fragfrom, todo);
2776 /* Ready to "commit" this state change to tgt */
2777 skb_shinfo(tgt)->nr_frags = to;
2780 fragfrom = &skb_shinfo(skb)->frags[0];
2781 fragto = &skb_shinfo(tgt)->frags[merge];
2783 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2784 __skb_frag_unref(fragfrom);
2787 /* Reposition in the original skb */
2789 while (from < skb_shinfo(skb)->nr_frags)
2790 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2791 skb_shinfo(skb)->nr_frags = to;
2793 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2796 /* Most likely the tgt won't ever need its checksum anymore, skb on
2797 * the other hand might need it if it needs to be resent
2799 tgt->ip_summed = CHECKSUM_PARTIAL;
2800 skb->ip_summed = CHECKSUM_PARTIAL;
2802 /* Yak, is it really working this way? Some helper please? */
2803 skb->len -= shiftlen;
2804 skb->data_len -= shiftlen;
2805 skb->truesize -= shiftlen;
2806 tgt->len += shiftlen;
2807 tgt->data_len += shiftlen;
2808 tgt->truesize += shiftlen;
2814 * skb_prepare_seq_read - Prepare a sequential read of skb data
2815 * @skb: the buffer to read
2816 * @from: lower offset of data to be read
2817 * @to: upper offset of data to be read
2818 * @st: state variable
2820 * Initializes the specified state variable. Must be called before
2821 * invoking skb_seq_read() for the first time.
2823 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2824 unsigned int to, struct skb_seq_state *st)
2826 st->lower_offset = from;
2827 st->upper_offset = to;
2828 st->root_skb = st->cur_skb = skb;
2829 st->frag_idx = st->stepped_offset = 0;
2830 st->frag_data = NULL;
2832 EXPORT_SYMBOL(skb_prepare_seq_read);
2835 * skb_seq_read - Sequentially read skb data
2836 * @consumed: number of bytes consumed by the caller so far
2837 * @data: destination pointer for data to be returned
2838 * @st: state variable
2840 * Reads a block of skb data at @consumed relative to the
2841 * lower offset specified to skb_prepare_seq_read(). Assigns
2842 * the head of the data block to @data and returns the length
2843 * of the block or 0 if the end of the skb data or the upper
2844 * offset has been reached.
2846 * The caller is not required to consume all of the data
2847 * returned, i.e. @consumed is typically set to the number
2848 * of bytes already consumed and the next call to
2849 * skb_seq_read() will return the remaining part of the block.
2851 * Note 1: The size of each block of data returned can be arbitrary,
2852 * this limitation is the cost for zerocopy sequential
2853 * reads of potentially non linear data.
2855 * Note 2: Fragment lists within fragments are not implemented
2856 * at the moment, state->root_skb could be replaced with
2857 * a stack for this purpose.
2859 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2860 struct skb_seq_state *st)
2862 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2865 if (unlikely(abs_offset >= st->upper_offset)) {
2866 if (st->frag_data) {
2867 kunmap_atomic(st->frag_data);
2868 st->frag_data = NULL;
2874 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2876 if (abs_offset < block_limit && !st->frag_data) {
2877 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2878 return block_limit - abs_offset;
2881 if (st->frag_idx == 0 && !st->frag_data)
2882 st->stepped_offset += skb_headlen(st->cur_skb);
2884 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2885 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2886 block_limit = skb_frag_size(frag) + st->stepped_offset;
2888 if (abs_offset < block_limit) {
2890 st->frag_data = kmap_atomic(skb_frag_page(frag));
2892 *data = (u8 *) st->frag_data + frag->page_offset +
2893 (abs_offset - st->stepped_offset);
2895 return block_limit - abs_offset;
2898 if (st->frag_data) {
2899 kunmap_atomic(st->frag_data);
2900 st->frag_data = NULL;
2904 st->stepped_offset += skb_frag_size(frag);
2907 if (st->frag_data) {
2908 kunmap_atomic(st->frag_data);
2909 st->frag_data = NULL;
2912 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2913 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2916 } else if (st->cur_skb->next) {
2917 st->cur_skb = st->cur_skb->next;
2924 EXPORT_SYMBOL(skb_seq_read);
2927 * skb_abort_seq_read - Abort a sequential read of skb data
2928 * @st: state variable
2930 * Must be called if skb_seq_read() was not called until it
2933 void skb_abort_seq_read(struct skb_seq_state *st)
2936 kunmap_atomic(st->frag_data);
2938 EXPORT_SYMBOL(skb_abort_seq_read);
2940 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2942 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2943 struct ts_config *conf,
2944 struct ts_state *state)
2946 return skb_seq_read(offset, text, TS_SKB_CB(state));
2949 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2951 skb_abort_seq_read(TS_SKB_CB(state));
2955 * skb_find_text - Find a text pattern in skb data
2956 * @skb: the buffer to look in
2957 * @from: search offset
2959 * @config: textsearch configuration
2961 * Finds a pattern in the skb data according to the specified
2962 * textsearch configuration. Use textsearch_next() to retrieve
2963 * subsequent occurrences of the pattern. Returns the offset
2964 * to the first occurrence or UINT_MAX if no match was found.
2966 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2967 unsigned int to, struct ts_config *config)
2969 struct ts_state state;
2972 config->get_next_block = skb_ts_get_next_block;
2973 config->finish = skb_ts_finish;
2975 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
2977 ret = textsearch_find(config, &state);
2978 return (ret <= to - from ? ret : UINT_MAX);
2980 EXPORT_SYMBOL(skb_find_text);
2983 * skb_append_datato_frags - append the user data to a skb
2984 * @sk: sock structure
2985 * @skb: skb structure to be appended with user data.
2986 * @getfrag: call back function to be used for getting the user data
2987 * @from: pointer to user message iov
2988 * @length: length of the iov message
2990 * Description: This procedure append the user data in the fragment part
2991 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2993 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2994 int (*getfrag)(void *from, char *to, int offset,
2995 int len, int odd, struct sk_buff *skb),
2996 void *from, int length)
2998 int frg_cnt = skb_shinfo(skb)->nr_frags;
3002 struct page_frag *pfrag = ¤t->task_frag;
3005 /* Return error if we don't have space for new frag */
3006 if (frg_cnt >= MAX_SKB_FRAGS)
3009 if (!sk_page_frag_refill(sk, pfrag))
3012 /* copy the user data to page */
3013 copy = min_t(int, length, pfrag->size - pfrag->offset);
3015 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
3016 offset, copy, 0, skb);
3020 /* copy was successful so update the size parameters */
3021 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
3024 pfrag->offset += copy;
3025 get_page(pfrag->page);
3027 skb->truesize += copy;
3028 atomic_add(copy, &sk->sk_wmem_alloc);
3030 skb->data_len += copy;
3034 } while (length > 0);
3038 EXPORT_SYMBOL(skb_append_datato_frags);
3040 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3041 int offset, size_t size)
3043 int i = skb_shinfo(skb)->nr_frags;
3045 if (skb_can_coalesce(skb, i, page, offset)) {
3046 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3047 } else if (i < MAX_SKB_FRAGS) {
3049 skb_fill_page_desc(skb, i, page, offset, size);
3056 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3059 * skb_pull_rcsum - pull skb and update receive checksum
3060 * @skb: buffer to update
3061 * @len: length of data pulled
3063 * This function performs an skb_pull on the packet and updates
3064 * the CHECKSUM_COMPLETE checksum. It should be used on
3065 * receive path processing instead of skb_pull unless you know
3066 * that the checksum difference is zero (e.g., a valid IP header)
3067 * or you are setting ip_summed to CHECKSUM_NONE.
3069 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3071 unsigned char *data = skb->data;
3073 BUG_ON(len > skb->len);
3074 __skb_pull(skb, len);
3075 skb_postpull_rcsum(skb, data, len);
3078 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3081 * skb_segment - Perform protocol segmentation on skb.
3082 * @head_skb: buffer to segment
3083 * @features: features for the output path (see dev->features)
3085 * This function performs segmentation on the given skb. It returns
3086 * a pointer to the first in a list of new skbs for the segments.
3087 * In case of error it returns ERR_PTR(err).
3089 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3090 netdev_features_t features)
3092 struct sk_buff *segs = NULL;
3093 struct sk_buff *tail = NULL;
3094 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3095 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3096 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3097 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3098 struct sk_buff *frag_skb = head_skb;
3099 unsigned int offset = doffset;
3100 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3101 unsigned int partial_segs = 0;
3102 unsigned int headroom;
3103 unsigned int len = head_skb->len;
3106 int nfrags = skb_shinfo(head_skb)->nr_frags;
3112 __skb_push(head_skb, doffset);
3113 proto = skb_network_protocol(head_skb, &dummy);
3114 if (unlikely(!proto))
3115 return ERR_PTR(-EINVAL);
3117 sg = !!(features & NETIF_F_SG);
3118 csum = !!can_checksum_protocol(features, proto);
3120 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3121 if (!(features & NETIF_F_GSO_PARTIAL)) {
3122 struct sk_buff *iter;
3123 unsigned int frag_len;
3126 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3129 /* If we get here then all the required
3130 * GSO features except frag_list are supported.
3131 * Try to split the SKB to multiple GSO SKBs
3132 * with no frag_list.
3133 * Currently we can do that only when the buffers don't
3134 * have a linear part and all the buffers except
3135 * the last are of the same length.
3137 frag_len = list_skb->len;
3138 skb_walk_frags(head_skb, iter) {
3139 if (frag_len != iter->len && iter->next)
3141 if (skb_headlen(iter) && !iter->head_frag)
3147 if (len != frag_len)
3151 /* GSO partial only requires that we trim off any excess that
3152 * doesn't fit into an MSS sized block, so take care of that
3155 partial_segs = len / mss;
3156 if (partial_segs > 1)
3157 mss *= partial_segs;
3163 headroom = skb_headroom(head_skb);
3164 pos = skb_headlen(head_skb);
3167 struct sk_buff *nskb;
3168 skb_frag_t *nskb_frag;
3172 if (unlikely(mss == GSO_BY_FRAGS)) {
3173 len = list_skb->len;
3175 len = head_skb->len - offset;
3180 hsize = skb_headlen(head_skb) - offset;
3183 if (hsize > len || !sg)
3186 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3187 (skb_headlen(list_skb) == len || sg)) {
3188 BUG_ON(skb_headlen(list_skb) > len);
3191 nfrags = skb_shinfo(list_skb)->nr_frags;
3192 frag = skb_shinfo(list_skb)->frags;
3193 frag_skb = list_skb;
3194 pos += skb_headlen(list_skb);
3196 while (pos < offset + len) {
3197 BUG_ON(i >= nfrags);
3199 size = skb_frag_size(frag);
3200 if (pos + size > offset + len)
3208 nskb = skb_clone(list_skb, GFP_ATOMIC);
3209 list_skb = list_skb->next;
3211 if (unlikely(!nskb))
3214 if (unlikely(pskb_trim(nskb, len))) {
3219 hsize = skb_end_offset(nskb);
3220 if (skb_cow_head(nskb, doffset + headroom)) {
3225 nskb->truesize += skb_end_offset(nskb) - hsize;
3226 skb_release_head_state(nskb);
3227 __skb_push(nskb, doffset);
3229 nskb = __alloc_skb(hsize + doffset + headroom,
3230 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3233 if (unlikely(!nskb))
3236 skb_reserve(nskb, headroom);
3237 __skb_put(nskb, doffset);
3246 __copy_skb_header(nskb, head_skb);
3248 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3249 skb_reset_mac_len(nskb);
3251 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3252 nskb->data - tnl_hlen,
3253 doffset + tnl_hlen);
3255 if (nskb->len == len + doffset)
3256 goto perform_csum_check;
3259 if (!nskb->remcsum_offload)
3260 nskb->ip_summed = CHECKSUM_NONE;
3261 SKB_GSO_CB(nskb)->csum =
3262 skb_copy_and_csum_bits(head_skb, offset,
3265 SKB_GSO_CB(nskb)->csum_start =
3266 skb_headroom(nskb) + doffset;
3270 nskb_frag = skb_shinfo(nskb)->frags;
3272 skb_copy_from_linear_data_offset(head_skb, offset,
3273 skb_put(nskb, hsize), hsize);
3275 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3278 while (pos < offset + len) {
3280 BUG_ON(skb_headlen(list_skb));
3283 nfrags = skb_shinfo(list_skb)->nr_frags;
3284 frag = skb_shinfo(list_skb)->frags;
3285 frag_skb = list_skb;
3289 list_skb = list_skb->next;
3292 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3294 net_warn_ratelimited(
3295 "skb_segment: too many frags: %u %u\n",
3300 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3304 __skb_frag_ref(nskb_frag);
3305 size = skb_frag_size(nskb_frag);
3308 nskb_frag->page_offset += offset - pos;
3309 skb_frag_size_sub(nskb_frag, offset - pos);
3312 skb_shinfo(nskb)->nr_frags++;
3314 if (pos + size <= offset + len) {
3319 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3327 nskb->data_len = len - hsize;
3328 nskb->len += nskb->data_len;
3329 nskb->truesize += nskb->data_len;
3333 if (skb_has_shared_frag(nskb)) {
3334 err = __skb_linearize(nskb);
3338 if (!nskb->remcsum_offload)
3339 nskb->ip_summed = CHECKSUM_NONE;
3340 SKB_GSO_CB(nskb)->csum =
3341 skb_checksum(nskb, doffset,
3342 nskb->len - doffset, 0);
3343 SKB_GSO_CB(nskb)->csum_start =
3344 skb_headroom(nskb) + doffset;
3346 } while ((offset += len) < head_skb->len);
3348 /* Some callers want to get the end of the list.
3349 * Put it in segs->prev to avoid walking the list.
3350 * (see validate_xmit_skb_list() for example)
3355 struct sk_buff *iter;
3356 int type = skb_shinfo(head_skb)->gso_type;
3357 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
3359 /* Update type to add partial and then remove dodgy if set */
3360 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
3361 type &= ~SKB_GSO_DODGY;
3363 /* Update GSO info and prepare to start updating headers on
3364 * our way back down the stack of protocols.
3366 for (iter = segs; iter; iter = iter->next) {
3367 skb_shinfo(iter)->gso_size = gso_size;
3368 skb_shinfo(iter)->gso_segs = partial_segs;
3369 skb_shinfo(iter)->gso_type = type;
3370 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
3373 if (tail->len - doffset <= gso_size)
3374 skb_shinfo(tail)->gso_size = 0;
3375 else if (tail != segs)
3376 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
3379 /* Following permits correct backpressure, for protocols
3380 * using skb_set_owner_w().
3381 * Idea is to tranfert ownership from head_skb to last segment.
3383 if (head_skb->destructor == sock_wfree) {
3384 swap(tail->truesize, head_skb->truesize);
3385 swap(tail->destructor, head_skb->destructor);
3386 swap(tail->sk, head_skb->sk);
3391 kfree_skb_list(segs);
3392 return ERR_PTR(err);
3394 EXPORT_SYMBOL_GPL(skb_segment);
3396 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3398 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3399 unsigned int offset = skb_gro_offset(skb);
3400 unsigned int headlen = skb_headlen(skb);
3401 unsigned int len = skb_gro_len(skb);
3402 struct sk_buff *lp, *p = *head;
3403 unsigned int delta_truesize;
3405 if (unlikely(p->len + len >= 65536))
3408 lp = NAPI_GRO_CB(p)->last;
3409 pinfo = skb_shinfo(lp);
3411 if (headlen <= offset) {
3414 int i = skbinfo->nr_frags;
3415 int nr_frags = pinfo->nr_frags + i;
3417 if (nr_frags > MAX_SKB_FRAGS)
3421 pinfo->nr_frags = nr_frags;
3422 skbinfo->nr_frags = 0;
3424 frag = pinfo->frags + nr_frags;
3425 frag2 = skbinfo->frags + i;
3430 frag->page_offset += offset;
3431 skb_frag_size_sub(frag, offset);
3433 /* all fragments truesize : remove (head size + sk_buff) */
3434 delta_truesize = skb->truesize -
3435 SKB_TRUESIZE(skb_end_offset(skb));
3437 skb->truesize -= skb->data_len;
3438 skb->len -= skb->data_len;
3441 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3443 } else if (skb->head_frag) {
3444 int nr_frags = pinfo->nr_frags;
3445 skb_frag_t *frag = pinfo->frags + nr_frags;
3446 struct page *page = virt_to_head_page(skb->head);
3447 unsigned int first_size = headlen - offset;
3448 unsigned int first_offset;
3450 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3453 first_offset = skb->data -
3454 (unsigned char *)page_address(page) +
3457 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3459 frag->page.p = page;
3460 frag->page_offset = first_offset;
3461 skb_frag_size_set(frag, first_size);
3463 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3464 /* We dont need to clear skbinfo->nr_frags here */
3466 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3467 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3472 delta_truesize = skb->truesize;
3473 if (offset > headlen) {
3474 unsigned int eat = offset - headlen;
3476 skbinfo->frags[0].page_offset += eat;
3477 skb_frag_size_sub(&skbinfo->frags[0], eat);
3478 skb->data_len -= eat;
3483 __skb_pull(skb, offset);
3485 if (NAPI_GRO_CB(p)->last == p)
3486 skb_shinfo(p)->frag_list = skb;
3488 NAPI_GRO_CB(p)->last->next = skb;
3489 NAPI_GRO_CB(p)->last = skb;
3490 __skb_header_release(skb);
3494 NAPI_GRO_CB(p)->count++;
3496 p->truesize += delta_truesize;
3499 lp->data_len += len;
3500 lp->truesize += delta_truesize;
3503 NAPI_GRO_CB(skb)->same_flow = 1;
3506 EXPORT_SYMBOL_GPL(skb_gro_receive);
3508 void __init skb_init(void)
3510 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3511 sizeof(struct sk_buff),
3513 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3515 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3516 sizeof(struct sk_buff_fclones),
3518 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3523 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
3524 unsigned int recursion_level)
3526 int start = skb_headlen(skb);
3527 int i, copy = start - offset;
3528 struct sk_buff *frag_iter;
3531 if (unlikely(recursion_level >= 24))
3537 sg_set_buf(sg, skb->data + offset, copy);
3539 if ((len -= copy) == 0)
3544 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3547 WARN_ON(start > offset + len);
3549 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3550 if ((copy = end - offset) > 0) {
3551 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3552 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
3557 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3558 frag->page_offset+offset-start);
3567 skb_walk_frags(skb, frag_iter) {
3570 WARN_ON(start > offset + len);
3572 end = start + frag_iter->len;
3573 if ((copy = end - offset) > 0) {
3574 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
3579 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3580 copy, recursion_level + 1);
3581 if (unlikely(ret < 0))
3584 if ((len -= copy) == 0)
3595 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3596 * @skb: Socket buffer containing the buffers to be mapped
3597 * @sg: The scatter-gather list to map into
3598 * @offset: The offset into the buffer's contents to start mapping
3599 * @len: Length of buffer space to be mapped
3601 * Fill the specified scatter-gather list with mappings/pointers into a
3602 * region of the buffer space attached to a socket buffer. Returns either
3603 * the number of scatterlist items used, or -EMSGSIZE if the contents
3606 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3608 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
3613 sg_mark_end(&sg[nsg - 1]);
3617 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3619 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3620 * sglist without mark the sg which contain last skb data as the end.
3621 * So the caller can mannipulate sg list as will when padding new data after
3622 * the first call without calling sg_unmark_end to expend sg list.
3624 * Scenario to use skb_to_sgvec_nomark:
3626 * 2. skb_to_sgvec_nomark(payload1)
3627 * 3. skb_to_sgvec_nomark(payload2)
3629 * This is equivalent to:
3631 * 2. skb_to_sgvec(payload1)
3633 * 4. skb_to_sgvec(payload2)
3635 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3636 * is more preferable.
3638 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3639 int offset, int len)
3641 return __skb_to_sgvec(skb, sg, offset, len, 0);
3643 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3648 * skb_cow_data - Check that a socket buffer's data buffers are writable
3649 * @skb: The socket buffer to check.
3650 * @tailbits: Amount of trailing space to be added
3651 * @trailer: Returned pointer to the skb where the @tailbits space begins
3653 * Make sure that the data buffers attached to a socket buffer are
3654 * writable. If they are not, private copies are made of the data buffers
3655 * and the socket buffer is set to use these instead.
3657 * If @tailbits is given, make sure that there is space to write @tailbits
3658 * bytes of data beyond current end of socket buffer. @trailer will be
3659 * set to point to the skb in which this space begins.
3661 * The number of scatterlist elements required to completely map the
3662 * COW'd and extended socket buffer will be returned.
3664 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3668 struct sk_buff *skb1, **skb_p;
3670 /* If skb is cloned or its head is paged, reallocate
3671 * head pulling out all the pages (pages are considered not writable
3672 * at the moment even if they are anonymous).
3674 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3675 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3678 /* Easy case. Most of packets will go this way. */
3679 if (!skb_has_frag_list(skb)) {
3680 /* A little of trouble, not enough of space for trailer.
3681 * This should not happen, when stack is tuned to generate
3682 * good frames. OK, on miss we reallocate and reserve even more
3683 * space, 128 bytes is fair. */
3685 if (skb_tailroom(skb) < tailbits &&
3686 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3694 /* Misery. We are in troubles, going to mincer fragments... */
3697 skb_p = &skb_shinfo(skb)->frag_list;
3700 while ((skb1 = *skb_p) != NULL) {
3703 /* The fragment is partially pulled by someone,
3704 * this can happen on input. Copy it and everything
3707 if (skb_shared(skb1))
3710 /* If the skb is the last, worry about trailer. */
3712 if (skb1->next == NULL && tailbits) {
3713 if (skb_shinfo(skb1)->nr_frags ||
3714 skb_has_frag_list(skb1) ||
3715 skb_tailroom(skb1) < tailbits)
3716 ntail = tailbits + 128;
3722 skb_shinfo(skb1)->nr_frags ||
3723 skb_has_frag_list(skb1)) {
3724 struct sk_buff *skb2;
3726 /* Fuck, we are miserable poor guys... */
3728 skb2 = skb_copy(skb1, GFP_ATOMIC);
3730 skb2 = skb_copy_expand(skb1,
3734 if (unlikely(skb2 == NULL))
3738 skb_set_owner_w(skb2, skb1->sk);
3740 /* Looking around. Are we still alive?
3741 * OK, link new skb, drop old one */
3743 skb2->next = skb1->next;
3750 skb_p = &skb1->next;
3755 EXPORT_SYMBOL_GPL(skb_cow_data);
3757 static void sock_rmem_free(struct sk_buff *skb)
3759 struct sock *sk = skb->sk;
3761 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3764 static void skb_set_err_queue(struct sk_buff *skb)
3766 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
3767 * So, it is safe to (mis)use it to mark skbs on the error queue.
3769 skb->pkt_type = PACKET_OUTGOING;
3770 BUILD_BUG_ON(PACKET_OUTGOING == 0);
3774 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3776 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3778 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3779 (unsigned int)sk->sk_rcvbuf)
3784 skb->destructor = sock_rmem_free;
3785 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3786 skb_set_err_queue(skb);
3788 /* before exiting rcu section, make sure dst is refcounted */
3791 skb_queue_tail(&sk->sk_error_queue, skb);
3792 if (!sock_flag(sk, SOCK_DEAD))
3793 sk->sk_data_ready(sk);
3796 EXPORT_SYMBOL(sock_queue_err_skb);
3798 static bool is_icmp_err_skb(const struct sk_buff *skb)
3800 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
3801 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
3804 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3806 struct sk_buff_head *q = &sk->sk_error_queue;
3807 struct sk_buff *skb, *skb_next = NULL;
3808 bool icmp_next = false;
3809 unsigned long flags;
3811 spin_lock_irqsave(&q->lock, flags);
3812 skb = __skb_dequeue(q);
3813 if (skb && (skb_next = skb_peek(q))) {
3814 icmp_next = is_icmp_err_skb(skb_next);
3816 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin;
3818 spin_unlock_irqrestore(&q->lock, flags);
3820 if (is_icmp_err_skb(skb) && !icmp_next)
3824 sk->sk_error_report(sk);
3828 EXPORT_SYMBOL(sock_dequeue_err_skb);
3831 * skb_clone_sk - create clone of skb, and take reference to socket
3832 * @skb: the skb to clone
3834 * This function creates a clone of a buffer that holds a reference on
3835 * sk_refcnt. Buffers created via this function are meant to be
3836 * returned using sock_queue_err_skb, or free via kfree_skb.
3838 * When passing buffers allocated with this function to sock_queue_err_skb
3839 * it is necessary to wrap the call with sock_hold/sock_put in order to
3840 * prevent the socket from being released prior to being enqueued on
3841 * the sk_error_queue.
3843 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3845 struct sock *sk = skb->sk;
3846 struct sk_buff *clone;
3848 if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3851 clone = skb_clone(skb, GFP_ATOMIC);
3858 clone->destructor = sock_efree;
3862 EXPORT_SYMBOL(skb_clone_sk);
3864 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3869 struct sock_exterr_skb *serr;
3872 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
3874 serr = SKB_EXT_ERR(skb);
3875 memset(serr, 0, sizeof(*serr));
3876 serr->ee.ee_errno = ENOMSG;
3877 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3878 serr->ee.ee_info = tstype;
3879 serr->opt_stats = opt_stats;
3880 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
3881 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3882 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3883 if (sk->sk_protocol == IPPROTO_TCP &&
3884 sk->sk_type == SOCK_STREAM)
3885 serr->ee.ee_data -= sk->sk_tskey;
3888 err = sock_queue_err_skb(sk, skb);
3894 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
3898 if (likely(sysctl_tstamp_allow_data || tsonly))
3901 read_lock_bh(&sk->sk_callback_lock);
3902 ret = sk->sk_socket && sk->sk_socket->file &&
3903 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
3904 read_unlock_bh(&sk->sk_callback_lock);
3908 void skb_complete_tx_timestamp(struct sk_buff *skb,
3909 struct skb_shared_hwtstamps *hwtstamps)
3911 struct sock *sk = skb->sk;
3913 if (!skb_may_tx_timestamp(sk, false))
3916 /* Take a reference to prevent skb_orphan() from freeing the socket,
3917 * but only if the socket refcount is not zero.
3919 if (likely(atomic_inc_not_zero(&sk->sk_refcnt))) {
3920 *skb_hwtstamps(skb) = *hwtstamps;
3921 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
3925 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3927 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3928 struct skb_shared_hwtstamps *hwtstamps,
3929 struct sock *sk, int tstype)
3931 struct sk_buff *skb;
3932 bool tsonly, opt_stats = false;
3937 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
3938 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
3941 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
3942 if (!skb_may_tx_timestamp(sk, tsonly))
3947 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
3948 sk->sk_protocol == IPPROTO_TCP &&
3949 sk->sk_type == SOCK_STREAM) {
3950 skb = tcp_get_timestamping_opt_stats(sk);
3954 skb = alloc_skb(0, GFP_ATOMIC);
3956 skb = skb_clone(orig_skb, GFP_ATOMIC);
3962 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
3964 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
3968 *skb_hwtstamps(skb) = *hwtstamps;
3970 skb->tstamp = ktime_get_real();
3972 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
3974 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3976 void skb_tstamp_tx(struct sk_buff *orig_skb,
3977 struct skb_shared_hwtstamps *hwtstamps)
3979 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3982 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3984 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3986 struct sock *sk = skb->sk;
3987 struct sock_exterr_skb *serr;
3990 skb->wifi_acked_valid = 1;
3991 skb->wifi_acked = acked;
3993 serr = SKB_EXT_ERR(skb);
3994 memset(serr, 0, sizeof(*serr));
3995 serr->ee.ee_errno = ENOMSG;
3996 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3998 /* Take a reference to prevent skb_orphan() from freeing the socket,
3999 * but only if the socket refcount is not zero.
4001 if (likely(atomic_inc_not_zero(&sk->sk_refcnt))) {
4002 err = sock_queue_err_skb(sk, skb);
4008 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4011 * skb_partial_csum_set - set up and verify partial csum values for packet
4012 * @skb: the skb to set
4013 * @start: the number of bytes after skb->data to start checksumming.
4014 * @off: the offset from start to place the checksum.
4016 * For untrusted partially-checksummed packets, we need to make sure the values
4017 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4019 * This function checks and sets those values and skb->ip_summed: if this
4020 * returns false you should drop the packet.
4022 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4024 if (unlikely(start > skb_headlen(skb)) ||
4025 unlikely((int)start + off > skb_headlen(skb) - 2)) {
4026 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
4027 start, off, skb_headlen(skb));
4030 skb->ip_summed = CHECKSUM_PARTIAL;
4031 skb->csum_start = skb_headroom(skb) + start;
4032 skb->csum_offset = off;
4033 skb_set_transport_header(skb, start);
4036 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4038 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4041 if (skb_headlen(skb) >= len)
4044 /* If we need to pullup then pullup to the max, so we
4045 * won't need to do it again.
4050 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4053 if (skb_headlen(skb) < len)
4059 #define MAX_TCP_HDR_LEN (15 * 4)
4061 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4062 typeof(IPPROTO_IP) proto,
4069 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4070 off + MAX_TCP_HDR_LEN);
4071 if (!err && !skb_partial_csum_set(skb, off,
4072 offsetof(struct tcphdr,
4075 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4078 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4079 off + sizeof(struct udphdr));
4080 if (!err && !skb_partial_csum_set(skb, off,
4081 offsetof(struct udphdr,
4084 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4087 return ERR_PTR(-EPROTO);
4090 /* This value should be large enough to cover a tagged ethernet header plus
4091 * maximally sized IP and TCP or UDP headers.
4093 #define MAX_IP_HDR_LEN 128
4095 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4104 err = skb_maybe_pull_tail(skb,
4105 sizeof(struct iphdr),
4110 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4113 off = ip_hdrlen(skb);
4120 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4122 return PTR_ERR(csum);
4125 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4128 ip_hdr(skb)->protocol, 0);
4135 /* This value should be large enough to cover a tagged ethernet header plus
4136 * an IPv6 header, all options, and a maximal TCP or UDP header.
4138 #define MAX_IPV6_HDR_LEN 256
4140 #define OPT_HDR(type, skb, off) \
4141 (type *)(skb_network_header(skb) + (off))
4143 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4156 off = sizeof(struct ipv6hdr);
4158 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4162 nexthdr = ipv6_hdr(skb)->nexthdr;
4164 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4165 while (off <= len && !done) {
4167 case IPPROTO_DSTOPTS:
4168 case IPPROTO_HOPOPTS:
4169 case IPPROTO_ROUTING: {
4170 struct ipv6_opt_hdr *hp;
4172 err = skb_maybe_pull_tail(skb,
4174 sizeof(struct ipv6_opt_hdr),
4179 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4180 nexthdr = hp->nexthdr;
4181 off += ipv6_optlen(hp);
4185 struct ip_auth_hdr *hp;
4187 err = skb_maybe_pull_tail(skb,
4189 sizeof(struct ip_auth_hdr),
4194 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4195 nexthdr = hp->nexthdr;
4196 off += ipv6_authlen(hp);
4199 case IPPROTO_FRAGMENT: {
4200 struct frag_hdr *hp;
4202 err = skb_maybe_pull_tail(skb,
4204 sizeof(struct frag_hdr),
4209 hp = OPT_HDR(struct frag_hdr, skb, off);
4211 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4214 nexthdr = hp->nexthdr;
4215 off += sizeof(struct frag_hdr);
4226 if (!done || fragment)
4229 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4231 return PTR_ERR(csum);
4234 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4235 &ipv6_hdr(skb)->daddr,
4236 skb->len - off, nexthdr, 0);
4244 * skb_checksum_setup - set up partial checksum offset
4245 * @skb: the skb to set up
4246 * @recalculate: if true the pseudo-header checksum will be recalculated
4248 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4252 switch (skb->protocol) {
4253 case htons(ETH_P_IP):
4254 err = skb_checksum_setup_ipv4(skb, recalculate);
4257 case htons(ETH_P_IPV6):
4258 err = skb_checksum_setup_ipv6(skb, recalculate);
4268 EXPORT_SYMBOL(skb_checksum_setup);
4271 * skb_checksum_maybe_trim - maybe trims the given skb
4272 * @skb: the skb to check
4273 * @transport_len: the data length beyond the network header
4275 * Checks whether the given skb has data beyond the given transport length.
4276 * If so, returns a cloned skb trimmed to this transport length.
4277 * Otherwise returns the provided skb. Returns NULL in error cases
4278 * (e.g. transport_len exceeds skb length or out-of-memory).
4280 * Caller needs to set the skb transport header and free any returned skb if it
4281 * differs from the provided skb.
4283 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4284 unsigned int transport_len)
4286 struct sk_buff *skb_chk;
4287 unsigned int len = skb_transport_offset(skb) + transport_len;
4292 else if (skb->len == len)
4295 skb_chk = skb_clone(skb, GFP_ATOMIC);
4299 ret = pskb_trim_rcsum(skb_chk, len);
4309 * skb_checksum_trimmed - validate checksum of an skb
4310 * @skb: the skb to check
4311 * @transport_len: the data length beyond the network header
4312 * @skb_chkf: checksum function to use
4314 * Applies the given checksum function skb_chkf to the provided skb.
4315 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4317 * If the skb has data beyond the given transport length, then a
4318 * trimmed & cloned skb is checked and returned.
4320 * Caller needs to set the skb transport header and free any returned skb if it
4321 * differs from the provided skb.
4323 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4324 unsigned int transport_len,
4325 __sum16(*skb_chkf)(struct sk_buff *skb))
4327 struct sk_buff *skb_chk;
4328 unsigned int offset = skb_transport_offset(skb);
4331 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4335 if (!pskb_may_pull(skb_chk, offset))
4338 skb_pull_rcsum(skb_chk, offset);
4339 ret = skb_chkf(skb_chk);
4340 skb_push_rcsum(skb_chk, offset);
4348 if (skb_chk && skb_chk != skb)
4354 EXPORT_SYMBOL(skb_checksum_trimmed);
4356 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4358 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4361 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4363 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4366 skb_release_head_state(skb);
4367 kmem_cache_free(skbuff_head_cache, skb);
4372 EXPORT_SYMBOL(kfree_skb_partial);
4375 * skb_try_coalesce - try to merge skb to prior one
4377 * @from: buffer to add
4378 * @fragstolen: pointer to boolean
4379 * @delta_truesize: how much more was allocated than was requested
4381 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4382 bool *fragstolen, int *delta_truesize)
4384 int i, delta, len = from->len;
4386 *fragstolen = false;
4391 if (len <= skb_tailroom(to)) {
4393 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4394 *delta_truesize = 0;
4398 if (skb_has_frag_list(to) || skb_has_frag_list(from))
4401 if (skb_headlen(from) != 0) {
4403 unsigned int offset;
4405 if (skb_shinfo(to)->nr_frags +
4406 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4409 if (skb_head_is_locked(from))
4412 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4414 page = virt_to_head_page(from->head);
4415 offset = from->data - (unsigned char *)page_address(page);
4417 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4418 page, offset, skb_headlen(from));
4421 if (skb_shinfo(to)->nr_frags +
4422 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4425 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4428 WARN_ON_ONCE(delta < len);
4430 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4431 skb_shinfo(from)->frags,
4432 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4433 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4435 if (!skb_cloned(from))
4436 skb_shinfo(from)->nr_frags = 0;
4438 /* if the skb is not cloned this does nothing
4439 * since we set nr_frags to 0.
4441 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4442 skb_frag_ref(from, i);
4444 to->truesize += delta;
4446 to->data_len += len;
4448 *delta_truesize = delta;
4451 EXPORT_SYMBOL(skb_try_coalesce);
4454 * skb_scrub_packet - scrub an skb
4456 * @skb: buffer to clean
4457 * @xnet: packet is crossing netns
4459 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4460 * into/from a tunnel. Some information have to be cleared during these
4462 * skb_scrub_packet can also be used to clean a skb before injecting it in
4463 * another namespace (@xnet == true). We have to clear all information in the
4464 * skb that could impact namespace isolation.
4466 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4469 skb->pkt_type = PACKET_HOST;
4475 nf_reset_trace(skb);
4483 EXPORT_SYMBOL_GPL(skb_scrub_packet);
4486 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4490 * skb_gso_transport_seglen is used to determine the real size of the
4491 * individual segments, including Layer4 headers (TCP/UDP).
4493 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4495 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4497 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4498 unsigned int thlen = 0;
4500 if (skb->encapsulation) {
4501 thlen = skb_inner_transport_header(skb) -
4502 skb_transport_header(skb);
4504 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4505 thlen += inner_tcp_hdrlen(skb);
4506 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4507 thlen = tcp_hdrlen(skb);
4508 } else if (unlikely(shinfo->gso_type & SKB_GSO_SCTP)) {
4509 thlen = sizeof(struct sctphdr);
4511 /* UFO sets gso_size to the size of the fragmentation
4512 * payload, i.e. the size of the L4 (UDP) header is already
4515 return thlen + shinfo->gso_size;
4517 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4520 * skb_gso_validate_mtu - Return in case such skb fits a given MTU
4523 * @mtu: MTU to validate against
4525 * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU
4528 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu)
4530 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4531 const struct sk_buff *iter;
4534 hlen = skb_gso_network_seglen(skb);
4536 if (shinfo->gso_size != GSO_BY_FRAGS)
4539 /* Undo this so we can re-use header sizes */
4540 hlen -= GSO_BY_FRAGS;
4542 skb_walk_frags(skb, iter) {
4543 if (hlen + skb_headlen(iter) > mtu)
4549 EXPORT_SYMBOL_GPL(skb_gso_validate_mtu);
4551 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4553 if (skb_cow(skb, skb_headroom(skb)) < 0) {
4558 memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN,
4560 skb->mac_header += VLAN_HLEN;
4564 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4566 struct vlan_hdr *vhdr;
4569 if (unlikely(skb_vlan_tag_present(skb))) {
4570 /* vlan_tci is already set-up so leave this for another time */
4574 skb = skb_share_check(skb, GFP_ATOMIC);
4578 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4581 vhdr = (struct vlan_hdr *)skb->data;
4582 vlan_tci = ntohs(vhdr->h_vlan_TCI);
4583 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4585 skb_pull_rcsum(skb, VLAN_HLEN);
4586 vlan_set_encap_proto(skb, vhdr);
4588 skb = skb_reorder_vlan_header(skb);
4592 skb_reset_network_header(skb);
4593 skb_reset_transport_header(skb);
4594 skb_reset_mac_len(skb);
4602 EXPORT_SYMBOL(skb_vlan_untag);
4604 int skb_ensure_writable(struct sk_buff *skb, int write_len)
4606 if (!pskb_may_pull(skb, write_len))
4609 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4612 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4614 EXPORT_SYMBOL(skb_ensure_writable);
4616 /* remove VLAN header from packet and update csum accordingly.
4617 * expects a non skb_vlan_tag_present skb with a vlan tag payload
4619 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
4621 struct vlan_hdr *vhdr;
4622 int offset = skb->data - skb_mac_header(skb);
4625 if (WARN_ONCE(offset,
4626 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
4631 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
4635 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4637 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
4638 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
4640 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
4641 __skb_pull(skb, VLAN_HLEN);
4643 vlan_set_encap_proto(skb, vhdr);
4644 skb->mac_header += VLAN_HLEN;
4646 if (skb_network_offset(skb) < ETH_HLEN)
4647 skb_set_network_header(skb, ETH_HLEN);
4649 skb_reset_mac_len(skb);
4653 EXPORT_SYMBOL(__skb_vlan_pop);
4655 /* Pop a vlan tag either from hwaccel or from payload.
4656 * Expects skb->data at mac header.
4658 int skb_vlan_pop(struct sk_buff *skb)
4664 if (likely(skb_vlan_tag_present(skb))) {
4667 if (unlikely(!eth_type_vlan(skb->protocol)))
4670 err = __skb_vlan_pop(skb, &vlan_tci);
4674 /* move next vlan tag to hw accel tag */
4675 if (likely(!eth_type_vlan(skb->protocol)))
4678 vlan_proto = skb->protocol;
4679 err = __skb_vlan_pop(skb, &vlan_tci);
4683 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4686 EXPORT_SYMBOL(skb_vlan_pop);
4688 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
4689 * Expects skb->data at mac header.
4691 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
4693 if (skb_vlan_tag_present(skb)) {
4694 int offset = skb->data - skb_mac_header(skb);
4697 if (WARN_ONCE(offset,
4698 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
4703 err = __vlan_insert_tag(skb, skb->vlan_proto,
4704 skb_vlan_tag_get(skb));
4708 skb->protocol = skb->vlan_proto;
4709 skb->mac_len += VLAN_HLEN;
4711 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4713 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4716 EXPORT_SYMBOL(skb_vlan_push);
4719 * alloc_skb_with_frags - allocate skb with page frags
4721 * @header_len: size of linear part
4722 * @data_len: needed length in frags
4723 * @max_page_order: max page order desired.
4724 * @errcode: pointer to error code if any
4725 * @gfp_mask: allocation mask
4727 * This can be used to allocate a paged skb, given a maximal order for frags.
4729 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4730 unsigned long data_len,
4735 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4736 unsigned long chunk;
4737 struct sk_buff *skb;
4742 *errcode = -EMSGSIZE;
4743 /* Note this test could be relaxed, if we succeed to allocate
4744 * high order pages...
4746 if (npages > MAX_SKB_FRAGS)
4749 gfp_head = gfp_mask;
4750 if (gfp_head & __GFP_DIRECT_RECLAIM)
4751 gfp_head |= __GFP_REPEAT;
4753 *errcode = -ENOBUFS;
4754 skb = alloc_skb(header_len, gfp_head);
4758 skb->truesize += npages << PAGE_SHIFT;
4760 for (i = 0; npages > 0; i++) {
4761 int order = max_page_order;
4764 if (npages >= 1 << order) {
4765 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
4772 /* Do not retry other high order allocations */
4778 page = alloc_page(gfp_mask);
4782 chunk = min_t(unsigned long, data_len,
4783 PAGE_SIZE << order);
4784 skb_fill_page_desc(skb, i, page, 0, chunk);
4786 npages -= 1 << order;
4794 EXPORT_SYMBOL(alloc_skb_with_frags);
4796 /* carve out the first off bytes from skb when off < headlen */
4797 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
4798 const int headlen, gfp_t gfp_mask)
4801 int size = skb_end_offset(skb);
4802 int new_hlen = headlen - off;
4805 size = SKB_DATA_ALIGN(size);
4807 if (skb_pfmemalloc(skb))
4808 gfp_mask |= __GFP_MEMALLOC;
4809 data = kmalloc_reserve(size +
4810 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4811 gfp_mask, NUMA_NO_NODE, NULL);
4815 size = SKB_WITH_OVERHEAD(ksize(data));
4817 /* Copy real data, and all frags */
4818 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
4821 memcpy((struct skb_shared_info *)(data + size),
4823 offsetof(struct skb_shared_info,
4824 frags[skb_shinfo(skb)->nr_frags]));
4825 if (skb_cloned(skb)) {
4826 /* drop the old head gracefully */
4827 if (skb_orphan_frags(skb, gfp_mask)) {
4831 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
4832 skb_frag_ref(skb, i);
4833 if (skb_has_frag_list(skb))
4834 skb_clone_fraglist(skb);
4835 skb_release_data(skb);
4837 /* we can reuse existing recount- all we did was
4846 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4849 skb->end = skb->head + size;
4851 skb_set_tail_pointer(skb, skb_headlen(skb));
4852 skb_headers_offset_update(skb, 0);
4856 atomic_set(&skb_shinfo(skb)->dataref, 1);
4861 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
4863 /* carve out the first eat bytes from skb's frag_list. May recurse into
4866 static int pskb_carve_frag_list(struct sk_buff *skb,
4867 struct skb_shared_info *shinfo, int eat,
4870 struct sk_buff *list = shinfo->frag_list;
4871 struct sk_buff *clone = NULL;
4872 struct sk_buff *insp = NULL;
4876 pr_err("Not enough bytes to eat. Want %d\n", eat);
4879 if (list->len <= eat) {
4880 /* Eaten as whole. */
4885 /* Eaten partially. */
4886 if (skb_shared(list)) {
4887 clone = skb_clone(list, gfp_mask);
4893 /* This may be pulled without problems. */
4896 if (pskb_carve(list, eat, gfp_mask) < 0) {
4904 /* Free pulled out fragments. */
4905 while ((list = shinfo->frag_list) != insp) {
4906 shinfo->frag_list = list->next;
4909 /* And insert new clone at head. */
4912 shinfo->frag_list = clone;
4917 /* carve off first len bytes from skb. Split line (off) is in the
4918 * non-linear part of skb
4920 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
4921 int pos, gfp_t gfp_mask)
4924 int size = skb_end_offset(skb);
4926 const int nfrags = skb_shinfo(skb)->nr_frags;
4927 struct skb_shared_info *shinfo;
4929 size = SKB_DATA_ALIGN(size);
4931 if (skb_pfmemalloc(skb))
4932 gfp_mask |= __GFP_MEMALLOC;
4933 data = kmalloc_reserve(size +
4934 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4935 gfp_mask, NUMA_NO_NODE, NULL);
4939 size = SKB_WITH_OVERHEAD(ksize(data));
4941 memcpy((struct skb_shared_info *)(data + size),
4942 skb_shinfo(skb), offsetof(struct skb_shared_info,
4943 frags[skb_shinfo(skb)->nr_frags]));
4944 if (skb_orphan_frags(skb, gfp_mask)) {
4948 shinfo = (struct skb_shared_info *)(data + size);
4949 for (i = 0; i < nfrags; i++) {
4950 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
4952 if (pos + fsize > off) {
4953 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
4957 * We have two variants in this case:
4958 * 1. Move all the frag to the second
4959 * part, if it is possible. F.e.
4960 * this approach is mandatory for TUX,
4961 * where splitting is expensive.
4962 * 2. Split is accurately. We make this.
4964 shinfo->frags[0].page_offset += off - pos;
4965 skb_frag_size_sub(&shinfo->frags[0], off - pos);
4967 skb_frag_ref(skb, i);
4972 shinfo->nr_frags = k;
4973 if (skb_has_frag_list(skb))
4974 skb_clone_fraglist(skb);
4977 /* split line is in frag list */
4978 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
4980 skb_release_data(skb);
4985 #ifdef NET_SKBUFF_DATA_USES_OFFSET
4988 skb->end = skb->head + size;
4990 skb_reset_tail_pointer(skb);
4991 skb_headers_offset_update(skb, 0);
4996 skb->data_len = skb->len;
4997 atomic_set(&skb_shinfo(skb)->dataref, 1);
5001 /* remove len bytes from the beginning of the skb */
5002 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
5004 int headlen = skb_headlen(skb);
5007 return pskb_carve_inside_header(skb, len, headlen, gfp);
5009 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
5012 /* Extract to_copy bytes starting at off from skb, and return this in
5015 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
5016 int to_copy, gfp_t gfp)
5018 struct sk_buff *clone = skb_clone(skb, gfp);
5023 if (pskb_carve(clone, off, gfp) < 0 ||
5024 pskb_trim(clone, to_copy)) {
5030 EXPORT_SYMBOL(pskb_extract);
5033 * skb_condense - try to get rid of fragments/frag_list if possible
5036 * Can be used to save memory before skb is added to a busy queue.
5037 * If packet has bytes in frags and enough tail room in skb->head,
5038 * pull all of them, so that we can free the frags right now and adjust
5041 * We do not reallocate skb->head thus can not fail.
5042 * Caller must re-evaluate skb->truesize if needed.
5044 void skb_condense(struct sk_buff *skb)
5046 if (skb->data_len) {
5047 if (skb->data_len > skb->end - skb->tail ||
5051 /* Nice, we can free page frag(s) right now */
5052 __pskb_pull_tail(skb, skb->data_len);
5054 /* At this point, skb->truesize might be over estimated,
5055 * because skb had a fragment, and fragments do not tell
5057 * When we pulled its content into skb->head, fragment
5058 * was freed, but __pskb_pull_tail() could not possibly
5059 * adjust skb->truesize, not knowing the frag truesize.
5061 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));