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
162 * __alloc_skb - allocate a network buffer
163 * @size: size to allocate
164 * @gfp_mask: allocation mask
165 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
166 * instead of head cache and allocate a cloned (child) skb.
167 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
168 * allocations in case the data is required for writeback
169 * @node: numa node to allocate memory on
171 * Allocate a new &sk_buff. The returned buffer has no headroom and a
172 * tail room of at least size bytes. The object has a reference count
173 * of one. The return is the buffer. On a failure the return is %NULL.
175 * Buffers may only be allocated from interrupts using a @gfp_mask of
178 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
181 struct kmem_cache *cache;
182 struct skb_shared_info *shinfo;
187 cache = (flags & SKB_ALLOC_FCLONE)
188 ? skbuff_fclone_cache : skbuff_head_cache;
190 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
191 gfp_mask |= __GFP_MEMALLOC;
194 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
199 /* We do our best to align skb_shared_info on a separate cache
200 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
201 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
202 * Both skb->head and skb_shared_info are cache line aligned.
204 size = SKB_DATA_ALIGN(size);
205 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
206 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
209 /* kmalloc(size) might give us more room than requested.
210 * Put skb_shared_info exactly at the end of allocated zone,
211 * to allow max possible filling before reallocation.
213 size = SKB_WITH_OVERHEAD(ksize(data));
214 prefetchw(data + size);
217 * Only clear those fields we need to clear, not those that we will
218 * actually initialise below. Hence, don't put any more fields after
219 * the tail pointer in struct sk_buff!
221 memset(skb, 0, offsetof(struct sk_buff, tail));
222 /* Account for allocated memory : skb + skb->head */
223 skb->truesize = SKB_TRUESIZE(size);
224 skb->pfmemalloc = pfmemalloc;
225 refcount_set(&skb->users, 1);
228 skb_reset_tail_pointer(skb);
229 skb->end = skb->tail + size;
230 skb->mac_header = (typeof(skb->mac_header))~0U;
231 skb->transport_header = (typeof(skb->transport_header))~0U;
233 /* make sure we initialize shinfo sequentially */
234 shinfo = skb_shinfo(skb);
235 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
236 atomic_set(&shinfo->dataref, 1);
237 kmemcheck_annotate_variable(shinfo->destructor_arg);
239 if (flags & SKB_ALLOC_FCLONE) {
240 struct sk_buff_fclones *fclones;
242 fclones = container_of(skb, struct sk_buff_fclones, skb1);
244 kmemcheck_annotate_bitfield(&fclones->skb2, flags1);
245 skb->fclone = SKB_FCLONE_ORIG;
246 refcount_set(&fclones->fclone_ref, 1);
248 fclones->skb2.fclone = SKB_FCLONE_CLONE;
253 kmem_cache_free(cache, skb);
257 EXPORT_SYMBOL(__alloc_skb);
260 * __build_skb - build a network buffer
261 * @data: data buffer provided by caller
262 * @frag_size: size of data, or 0 if head was kmalloced
264 * Allocate a new &sk_buff. Caller provides space holding head and
265 * skb_shared_info. @data must have been allocated by kmalloc() only if
266 * @frag_size is 0, otherwise data should come from the page allocator
268 * The return is the new skb buffer.
269 * On a failure the return is %NULL, and @data is not freed.
271 * Before IO, driver allocates only data buffer where NIC put incoming frame
272 * Driver should add room at head (NET_SKB_PAD) and
273 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
274 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
275 * before giving packet to stack.
276 * RX rings only contains data buffers, not full skbs.
278 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
280 struct skb_shared_info *shinfo;
282 unsigned int size = frag_size ? : ksize(data);
284 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
288 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
290 memset(skb, 0, offsetof(struct sk_buff, tail));
291 skb->truesize = SKB_TRUESIZE(size);
292 refcount_set(&skb->users, 1);
295 skb_reset_tail_pointer(skb);
296 skb->end = skb->tail + size;
297 skb->mac_header = (typeof(skb->mac_header))~0U;
298 skb->transport_header = (typeof(skb->transport_header))~0U;
300 /* make sure we initialize shinfo sequentially */
301 shinfo = skb_shinfo(skb);
302 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
303 atomic_set(&shinfo->dataref, 1);
304 kmemcheck_annotate_variable(shinfo->destructor_arg);
309 /* build_skb() is wrapper over __build_skb(), that specifically
310 * takes care of skb->head and skb->pfmemalloc
311 * This means that if @frag_size is not zero, then @data must be backed
312 * by a page fragment, not kmalloc() or vmalloc()
314 struct sk_buff *build_skb(void *data, unsigned int frag_size)
316 struct sk_buff *skb = __build_skb(data, frag_size);
318 if (skb && frag_size) {
320 if (page_is_pfmemalloc(virt_to_head_page(data)))
325 EXPORT_SYMBOL(build_skb);
327 #define NAPI_SKB_CACHE_SIZE 64
329 struct napi_alloc_cache {
330 struct page_frag_cache page;
331 unsigned int skb_count;
332 void *skb_cache[NAPI_SKB_CACHE_SIZE];
335 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
336 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
338 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
340 struct page_frag_cache *nc;
344 local_irq_save(flags);
345 nc = this_cpu_ptr(&netdev_alloc_cache);
346 data = page_frag_alloc(nc, fragsz, gfp_mask);
347 local_irq_restore(flags);
352 * netdev_alloc_frag - allocate a page fragment
353 * @fragsz: fragment size
355 * Allocates a frag from a page for receive buffer.
356 * Uses GFP_ATOMIC allocations.
358 void *netdev_alloc_frag(unsigned int fragsz)
360 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
362 EXPORT_SYMBOL(netdev_alloc_frag);
364 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
366 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
368 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
371 void *napi_alloc_frag(unsigned int fragsz)
373 return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
375 EXPORT_SYMBOL(napi_alloc_frag);
378 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
379 * @dev: network device to receive on
380 * @len: length to allocate
381 * @gfp_mask: get_free_pages mask, passed to alloc_skb
383 * Allocate a new &sk_buff and assign it a usage count of one. The
384 * buffer has NET_SKB_PAD headroom built in. Users should allocate
385 * the headroom they think they need without accounting for the
386 * built in space. The built in space is used for optimisations.
388 * %NULL is returned if there is no free memory.
390 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
393 struct page_frag_cache *nc;
401 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
402 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
403 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
409 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
410 len = SKB_DATA_ALIGN(len);
412 if (sk_memalloc_socks())
413 gfp_mask |= __GFP_MEMALLOC;
415 local_irq_save(flags);
417 nc = this_cpu_ptr(&netdev_alloc_cache);
418 data = page_frag_alloc(nc, len, gfp_mask);
419 pfmemalloc = nc->pfmemalloc;
421 local_irq_restore(flags);
426 skb = __build_skb(data, len);
427 if (unlikely(!skb)) {
432 /* use OR instead of assignment to avoid clearing of bits in mask */
438 skb_reserve(skb, NET_SKB_PAD);
444 EXPORT_SYMBOL(__netdev_alloc_skb);
447 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
448 * @napi: napi instance this buffer was allocated for
449 * @len: length to allocate
450 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
452 * Allocate a new sk_buff for use in NAPI receive. This buffer will
453 * attempt to allocate the head from a special reserved region used
454 * only for NAPI Rx allocation. By doing this we can save several
455 * CPU cycles by avoiding having to disable and re-enable IRQs.
457 * %NULL is returned if there is no free memory.
459 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
462 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
466 len += NET_SKB_PAD + NET_IP_ALIGN;
468 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
469 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
470 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
476 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
477 len = SKB_DATA_ALIGN(len);
479 if (sk_memalloc_socks())
480 gfp_mask |= __GFP_MEMALLOC;
482 data = page_frag_alloc(&nc->page, len, gfp_mask);
486 skb = __build_skb(data, len);
487 if (unlikely(!skb)) {
492 /* use OR instead of assignment to avoid clearing of bits in mask */
493 if (nc->page.pfmemalloc)
498 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
499 skb->dev = napi->dev;
504 EXPORT_SYMBOL(__napi_alloc_skb);
506 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
507 int size, unsigned int truesize)
509 skb_fill_page_desc(skb, i, page, off, size);
511 skb->data_len += size;
512 skb->truesize += truesize;
514 EXPORT_SYMBOL(skb_add_rx_frag);
516 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
517 unsigned int truesize)
519 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
521 skb_frag_size_add(frag, size);
523 skb->data_len += size;
524 skb->truesize += truesize;
526 EXPORT_SYMBOL(skb_coalesce_rx_frag);
528 static void skb_drop_list(struct sk_buff **listp)
530 kfree_skb_list(*listp);
534 static inline void skb_drop_fraglist(struct sk_buff *skb)
536 skb_drop_list(&skb_shinfo(skb)->frag_list);
539 static void skb_clone_fraglist(struct sk_buff *skb)
541 struct sk_buff *list;
543 skb_walk_frags(skb, list)
547 static void skb_free_head(struct sk_buff *skb)
549 unsigned char *head = skb->head;
557 static void skb_release_data(struct sk_buff *skb)
559 struct skb_shared_info *shinfo = skb_shinfo(skb);
563 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
567 for (i = 0; i < shinfo->nr_frags; i++)
568 __skb_frag_unref(&shinfo->frags[i]);
570 if (shinfo->frag_list)
571 kfree_skb_list(shinfo->frag_list);
573 skb_zcopy_clear(skb, true);
578 * Free an skbuff by memory without cleaning the state.
580 static void kfree_skbmem(struct sk_buff *skb)
582 struct sk_buff_fclones *fclones;
584 switch (skb->fclone) {
585 case SKB_FCLONE_UNAVAILABLE:
586 kmem_cache_free(skbuff_head_cache, skb);
589 case SKB_FCLONE_ORIG:
590 fclones = container_of(skb, struct sk_buff_fclones, skb1);
592 /* We usually free the clone (TX completion) before original skb
593 * This test would have no chance to be true for the clone,
594 * while here, branch prediction will be good.
596 if (refcount_read(&fclones->fclone_ref) == 1)
600 default: /* SKB_FCLONE_CLONE */
601 fclones = container_of(skb, struct sk_buff_fclones, skb2);
604 if (!refcount_dec_and_test(&fclones->fclone_ref))
607 kmem_cache_free(skbuff_fclone_cache, fclones);
610 void skb_release_head_state(struct sk_buff *skb)
614 if (skb->destructor) {
616 skb->destructor(skb);
618 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
619 nf_conntrack_put(skb_nfct(skb));
621 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
622 nf_bridge_put(skb->nf_bridge);
626 /* Free everything but the sk_buff shell. */
627 static void skb_release_all(struct sk_buff *skb)
629 skb_release_head_state(skb);
630 if (likely(skb->head))
631 skb_release_data(skb);
635 * __kfree_skb - private function
638 * Free an sk_buff. Release anything attached to the buffer.
639 * Clean the state. This is an internal helper function. Users should
640 * always call kfree_skb
643 void __kfree_skb(struct sk_buff *skb)
645 skb_release_all(skb);
648 EXPORT_SYMBOL(__kfree_skb);
651 * kfree_skb - free an sk_buff
652 * @skb: buffer to free
654 * Drop a reference to the buffer and free it if the usage count has
657 void kfree_skb(struct sk_buff *skb)
662 trace_kfree_skb(skb, __builtin_return_address(0));
665 EXPORT_SYMBOL(kfree_skb);
667 void kfree_skb_list(struct sk_buff *segs)
670 struct sk_buff *next = segs->next;
676 EXPORT_SYMBOL(kfree_skb_list);
679 * skb_tx_error - report an sk_buff xmit error
680 * @skb: buffer that triggered an error
682 * Report xmit error if a device callback is tracking this skb.
683 * skb must be freed afterwards.
685 void skb_tx_error(struct sk_buff *skb)
687 skb_zcopy_clear(skb, true);
689 EXPORT_SYMBOL(skb_tx_error);
692 * consume_skb - free an skbuff
693 * @skb: buffer to free
695 * Drop a ref to the buffer and free it if the usage count has hit zero
696 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
697 * is being dropped after a failure and notes that
699 void consume_skb(struct sk_buff *skb)
704 trace_consume_skb(skb);
707 EXPORT_SYMBOL(consume_skb);
710 * consume_stateless_skb - free an skbuff, assuming it is stateless
711 * @skb: buffer to free
713 * Alike consume_skb(), but this variant assumes that this is the last
714 * skb reference and all the head states have been already dropped
716 void __consume_stateless_skb(struct sk_buff *skb)
718 trace_consume_skb(skb);
719 skb_release_data(skb);
723 void __kfree_skb_flush(void)
725 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
727 /* flush skb_cache if containing objects */
729 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
735 static inline void _kfree_skb_defer(struct sk_buff *skb)
737 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
739 /* drop skb->head and call any destructors for packet */
740 skb_release_all(skb);
742 /* record skb to CPU local list */
743 nc->skb_cache[nc->skb_count++] = skb;
746 /* SLUB writes into objects when freeing */
750 /* flush skb_cache if it is filled */
751 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
752 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
757 void __kfree_skb_defer(struct sk_buff *skb)
759 _kfree_skb_defer(skb);
762 void napi_consume_skb(struct sk_buff *skb, int budget)
767 /* Zero budget indicate non-NAPI context called us, like netpoll */
768 if (unlikely(!budget)) {
769 dev_consume_skb_any(skb);
776 /* if reaching here SKB is ready to free */
777 trace_consume_skb(skb);
779 /* if SKB is a clone, don't handle this case */
780 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
785 _kfree_skb_defer(skb);
787 EXPORT_SYMBOL(napi_consume_skb);
789 /* Make sure a field is enclosed inside headers_start/headers_end section */
790 #define CHECK_SKB_FIELD(field) \
791 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
792 offsetof(struct sk_buff, headers_start)); \
793 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
794 offsetof(struct sk_buff, headers_end)); \
796 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
798 new->tstamp = old->tstamp;
799 /* We do not copy old->sk */
801 memcpy(new->cb, old->cb, sizeof(old->cb));
802 skb_dst_copy(new, old);
804 new->sp = secpath_get(old->sp);
806 __nf_copy(new, old, false);
808 /* Note : this field could be in headers_start/headers_end section
809 * It is not yet because we do not want to have a 16 bit hole
811 new->queue_mapping = old->queue_mapping;
813 memcpy(&new->headers_start, &old->headers_start,
814 offsetof(struct sk_buff, headers_end) -
815 offsetof(struct sk_buff, headers_start));
816 CHECK_SKB_FIELD(protocol);
817 CHECK_SKB_FIELD(csum);
818 CHECK_SKB_FIELD(hash);
819 CHECK_SKB_FIELD(priority);
820 CHECK_SKB_FIELD(skb_iif);
821 CHECK_SKB_FIELD(vlan_proto);
822 CHECK_SKB_FIELD(vlan_tci);
823 CHECK_SKB_FIELD(transport_header);
824 CHECK_SKB_FIELD(network_header);
825 CHECK_SKB_FIELD(mac_header);
826 CHECK_SKB_FIELD(inner_protocol);
827 CHECK_SKB_FIELD(inner_transport_header);
828 CHECK_SKB_FIELD(inner_network_header);
829 CHECK_SKB_FIELD(inner_mac_header);
830 CHECK_SKB_FIELD(mark);
831 #ifdef CONFIG_NETWORK_SECMARK
832 CHECK_SKB_FIELD(secmark);
834 #ifdef CONFIG_NET_RX_BUSY_POLL
835 CHECK_SKB_FIELD(napi_id);
838 CHECK_SKB_FIELD(sender_cpu);
840 #ifdef CONFIG_NET_SCHED
841 CHECK_SKB_FIELD(tc_index);
847 * You should not add any new code to this function. Add it to
848 * __copy_skb_header above instead.
850 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
852 #define C(x) n->x = skb->x
854 n->next = n->prev = NULL;
856 __copy_skb_header(n, skb);
861 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
864 n->destructor = NULL;
871 refcount_set(&n->users, 1);
873 atomic_inc(&(skb_shinfo(skb)->dataref));
881 * skb_morph - morph one skb into another
882 * @dst: the skb to receive the contents
883 * @src: the skb to supply the contents
885 * This is identical to skb_clone except that the target skb is
886 * supplied by the user.
888 * The target skb is returned upon exit.
890 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
892 skb_release_all(dst);
893 return __skb_clone(dst, src);
895 EXPORT_SYMBOL_GPL(skb_morph);
897 static int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
899 unsigned long max_pg, num_pg, new_pg, old_pg;
900 struct user_struct *user;
902 if (capable(CAP_IPC_LOCK) || !size)
905 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
906 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
907 user = mmp->user ? : current_user();
910 old_pg = atomic_long_read(&user->locked_vm);
911 new_pg = old_pg + num_pg;
914 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
918 mmp->user = get_uid(user);
919 mmp->num_pg = num_pg;
921 mmp->num_pg += num_pg;
927 static void mm_unaccount_pinned_pages(struct mmpin *mmp)
930 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
935 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
937 struct ubuf_info *uarg;
940 WARN_ON_ONCE(!in_task());
942 if (!sock_flag(sk, SOCK_ZEROCOPY))
945 skb = sock_omalloc(sk, 0, GFP_KERNEL);
949 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
950 uarg = (void *)skb->cb;
951 uarg->mmp.user = NULL;
953 if (mm_account_pinned_pages(&uarg->mmp, size)) {
958 uarg->callback = sock_zerocopy_callback;
959 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
961 uarg->bytelen = size;
963 refcount_set(&uarg->refcnt, 1);
968 EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
970 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
972 return container_of((void *)uarg, struct sk_buff, cb);
975 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
976 struct ubuf_info *uarg)
979 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
982 /* realloc only when socket is locked (TCP, UDP cork),
983 * so uarg->len and sk_zckey access is serialized
985 if (!sock_owned_by_user(sk)) {
990 bytelen = uarg->bytelen + size;
991 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
992 /* TCP can create new skb to attach new uarg */
993 if (sk->sk_type == SOCK_STREAM)
998 next = (u32)atomic_read(&sk->sk_zckey);
999 if ((u32)(uarg->id + uarg->len) == next) {
1000 if (mm_account_pinned_pages(&uarg->mmp, size))
1003 uarg->bytelen = bytelen;
1004 atomic_set(&sk->sk_zckey, ++next);
1005 sock_zerocopy_get(uarg);
1011 return sock_zerocopy_alloc(sk, size);
1013 EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1015 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1017 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1021 old_lo = serr->ee.ee_info;
1022 old_hi = serr->ee.ee_data;
1023 sum_len = old_hi - old_lo + 1ULL + len;
1025 if (sum_len >= (1ULL << 32))
1028 if (lo != old_hi + 1)
1031 serr->ee.ee_data += len;
1035 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1037 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1038 struct sock_exterr_skb *serr;
1039 struct sock *sk = skb->sk;
1040 struct sk_buff_head *q;
1041 unsigned long flags;
1045 mm_unaccount_pinned_pages(&uarg->mmp);
1047 /* if !len, there was only 1 call, and it was aborted
1048 * so do not queue a completion notification
1050 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1055 hi = uarg->id + len - 1;
1057 serr = SKB_EXT_ERR(skb);
1058 memset(serr, 0, sizeof(*serr));
1059 serr->ee.ee_errno = 0;
1060 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1061 serr->ee.ee_data = hi;
1062 serr->ee.ee_info = lo;
1064 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1066 q = &sk->sk_error_queue;
1067 spin_lock_irqsave(&q->lock, flags);
1068 tail = skb_peek_tail(q);
1069 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1070 !skb_zerocopy_notify_extend(tail, lo, len)) {
1071 __skb_queue_tail(q, skb);
1074 spin_unlock_irqrestore(&q->lock, flags);
1076 sk->sk_error_report(sk);
1082 EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1084 void sock_zerocopy_put(struct ubuf_info *uarg)
1086 if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1088 uarg->callback(uarg, uarg->zerocopy);
1090 consume_skb(skb_from_uarg(uarg));
1093 EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1095 void sock_zerocopy_put_abort(struct ubuf_info *uarg)
1098 struct sock *sk = skb_from_uarg(uarg)->sk;
1100 atomic_dec(&sk->sk_zckey);
1103 sock_zerocopy_put(uarg);
1106 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1108 extern int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb,
1109 struct iov_iter *from, size_t length);
1111 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1112 struct msghdr *msg, int len,
1113 struct ubuf_info *uarg)
1115 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1116 struct iov_iter orig_iter = msg->msg_iter;
1117 int err, orig_len = skb->len;
1119 /* An skb can only point to one uarg. This edge case happens when
1120 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1122 if (orig_uarg && uarg != orig_uarg)
1125 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1126 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1127 /* Streams do not free skb on error. Reset to prev state. */
1128 msg->msg_iter = orig_iter;
1129 ___pskb_trim(skb, orig_len);
1133 skb_zcopy_set(skb, uarg);
1134 return skb->len - orig_len;
1136 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1138 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1141 if (skb_zcopy(orig)) {
1142 if (skb_zcopy(nskb)) {
1143 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1148 if (skb_uarg(nskb) == skb_uarg(orig))
1150 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1153 skb_zcopy_set(nskb, skb_uarg(orig));
1159 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1160 * @skb: the skb to modify
1161 * @gfp_mask: allocation priority
1163 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1164 * It will copy all frags into kernel and drop the reference
1165 * to userspace pages.
1167 * If this function is called from an interrupt gfp_mask() must be
1170 * Returns 0 on success or a negative error code on failure
1171 * to allocate kernel memory to copy to.
1173 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1175 int num_frags = skb_shinfo(skb)->nr_frags;
1176 struct page *page, *head = NULL;
1183 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1186 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1187 for (i = 0; i < new_frags; i++) {
1188 page = alloc_page(gfp_mask);
1191 struct page *next = (struct page *)page_private(head);
1197 set_page_private(page, (unsigned long)head);
1203 for (i = 0; i < num_frags; i++) {
1204 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1205 u32 p_off, p_len, copied;
1209 skb_frag_foreach_page(f, f->page_offset, skb_frag_size(f),
1210 p, p_off, p_len, copied) {
1212 vaddr = kmap_atomic(p);
1214 while (done < p_len) {
1215 if (d_off == PAGE_SIZE) {
1217 page = (struct page *)page_private(page);
1219 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1220 memcpy(page_address(page) + d_off,
1221 vaddr + p_off + done, copy);
1225 kunmap_atomic(vaddr);
1229 /* skb frags release userspace buffers */
1230 for (i = 0; i < num_frags; i++)
1231 skb_frag_unref(skb, i);
1233 /* skb frags point to kernel buffers */
1234 for (i = 0; i < new_frags - 1; i++) {
1235 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1236 head = (struct page *)page_private(head);
1238 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1239 skb_shinfo(skb)->nr_frags = new_frags;
1241 skb_zcopy_clear(skb, false);
1244 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1247 * skb_clone - duplicate an sk_buff
1248 * @skb: buffer to clone
1249 * @gfp_mask: allocation priority
1251 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1252 * copies share the same packet data but not structure. The new
1253 * buffer has a reference count of 1. If the allocation fails the
1254 * function returns %NULL otherwise the new buffer is returned.
1256 * If this function is called from an interrupt gfp_mask() must be
1260 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1262 struct sk_buff_fclones *fclones = container_of(skb,
1263 struct sk_buff_fclones,
1267 if (skb_orphan_frags(skb, gfp_mask))
1270 if (skb->fclone == SKB_FCLONE_ORIG &&
1271 refcount_read(&fclones->fclone_ref) == 1) {
1273 refcount_set(&fclones->fclone_ref, 2);
1275 if (skb_pfmemalloc(skb))
1276 gfp_mask |= __GFP_MEMALLOC;
1278 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1282 kmemcheck_annotate_bitfield(n, flags1);
1283 n->fclone = SKB_FCLONE_UNAVAILABLE;
1286 return __skb_clone(n, skb);
1288 EXPORT_SYMBOL(skb_clone);
1290 static void skb_headers_offset_update(struct sk_buff *skb, int off)
1292 /* Only adjust this if it actually is csum_start rather than csum */
1293 if (skb->ip_summed == CHECKSUM_PARTIAL)
1294 skb->csum_start += off;
1295 /* {transport,network,mac}_header and tail are relative to skb->head */
1296 skb->transport_header += off;
1297 skb->network_header += off;
1298 if (skb_mac_header_was_set(skb))
1299 skb->mac_header += off;
1300 skb->inner_transport_header += off;
1301 skb->inner_network_header += off;
1302 skb->inner_mac_header += off;
1305 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1307 __copy_skb_header(new, old);
1309 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1310 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1311 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1314 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1316 if (skb_pfmemalloc(skb))
1317 return SKB_ALLOC_RX;
1322 * skb_copy - create private copy of an sk_buff
1323 * @skb: buffer to copy
1324 * @gfp_mask: allocation priority
1326 * Make a copy of both an &sk_buff and its data. This is used when the
1327 * caller wishes to modify the data and needs a private copy of the
1328 * data to alter. Returns %NULL on failure or the pointer to the buffer
1329 * on success. The returned buffer has a reference count of 1.
1331 * As by-product this function converts non-linear &sk_buff to linear
1332 * one, so that &sk_buff becomes completely private and caller is allowed
1333 * to modify all the data of returned buffer. This means that this
1334 * function is not recommended for use in circumstances when only
1335 * header is going to be modified. Use pskb_copy() instead.
1338 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1340 int headerlen = skb_headroom(skb);
1341 unsigned int size = skb_end_offset(skb) + skb->data_len;
1342 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1343 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1348 /* Set the data pointer */
1349 skb_reserve(n, headerlen);
1350 /* Set the tail pointer and length */
1351 skb_put(n, skb->len);
1353 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1356 copy_skb_header(n, skb);
1359 EXPORT_SYMBOL(skb_copy);
1362 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1363 * @skb: buffer to copy
1364 * @headroom: headroom of new skb
1365 * @gfp_mask: allocation priority
1366 * @fclone: if true allocate the copy of the skb from the fclone
1367 * cache instead of the head cache; it is recommended to set this
1368 * to true for the cases where the copy will likely be cloned
1370 * Make a copy of both an &sk_buff and part of its data, located
1371 * in header. Fragmented data remain shared. This is used when
1372 * the caller wishes to modify only header of &sk_buff and needs
1373 * private copy of the header to alter. Returns %NULL on failure
1374 * or the pointer to the buffer on success.
1375 * The returned buffer has a reference count of 1.
1378 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1379 gfp_t gfp_mask, bool fclone)
1381 unsigned int size = skb_headlen(skb) + headroom;
1382 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1383 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1388 /* Set the data pointer */
1389 skb_reserve(n, headroom);
1390 /* Set the tail pointer and length */
1391 skb_put(n, skb_headlen(skb));
1392 /* Copy the bytes */
1393 skb_copy_from_linear_data(skb, n->data, n->len);
1395 n->truesize += skb->data_len;
1396 n->data_len = skb->data_len;
1399 if (skb_shinfo(skb)->nr_frags) {
1402 if (skb_orphan_frags(skb, gfp_mask) ||
1403 skb_zerocopy_clone(n, skb, gfp_mask)) {
1408 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1409 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1410 skb_frag_ref(skb, i);
1412 skb_shinfo(n)->nr_frags = i;
1415 if (skb_has_frag_list(skb)) {
1416 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1417 skb_clone_fraglist(n);
1420 copy_skb_header(n, skb);
1424 EXPORT_SYMBOL(__pskb_copy_fclone);
1427 * pskb_expand_head - reallocate header of &sk_buff
1428 * @skb: buffer to reallocate
1429 * @nhead: room to add at head
1430 * @ntail: room to add at tail
1431 * @gfp_mask: allocation priority
1433 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1434 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1435 * reference count of 1. Returns zero in the case of success or error,
1436 * if expansion failed. In the last case, &sk_buff is not changed.
1438 * All the pointers pointing into skb header may change and must be
1439 * reloaded after call to this function.
1442 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1445 int i, osize = skb_end_offset(skb);
1446 int size = osize + nhead + ntail;
1452 if (skb_shared(skb))
1455 size = SKB_DATA_ALIGN(size);
1457 if (skb_pfmemalloc(skb))
1458 gfp_mask |= __GFP_MEMALLOC;
1459 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1460 gfp_mask, NUMA_NO_NODE, NULL);
1463 size = SKB_WITH_OVERHEAD(ksize(data));
1465 /* Copy only real data... and, alas, header. This should be
1466 * optimized for the cases when header is void.
1468 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1470 memcpy((struct skb_shared_info *)(data + size),
1472 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1475 * if shinfo is shared we must drop the old head gracefully, but if it
1476 * is not we can just drop the old head and let the existing refcount
1477 * be since all we did is relocate the values
1479 if (skb_cloned(skb)) {
1480 if (skb_orphan_frags(skb, gfp_mask))
1483 refcount_inc(&skb_uarg(skb)->refcnt);
1484 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1485 skb_frag_ref(skb, i);
1487 if (skb_has_frag_list(skb))
1488 skb_clone_fraglist(skb);
1490 skb_release_data(skb);
1494 off = (data + nhead) - skb->head;
1499 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1503 skb->end = skb->head + size;
1506 skb_headers_offset_update(skb, nhead);
1510 atomic_set(&skb_shinfo(skb)->dataref, 1);
1512 /* It is not generally safe to change skb->truesize.
1513 * For the moment, we really care of rx path, or
1514 * when skb is orphaned (not attached to a socket).
1516 if (!skb->sk || skb->destructor == sock_edemux)
1517 skb->truesize += size - osize;
1526 EXPORT_SYMBOL(pskb_expand_head);
1528 /* Make private copy of skb with writable head and some headroom */
1530 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1532 struct sk_buff *skb2;
1533 int delta = headroom - skb_headroom(skb);
1536 skb2 = pskb_copy(skb, GFP_ATOMIC);
1538 skb2 = skb_clone(skb, GFP_ATOMIC);
1539 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1547 EXPORT_SYMBOL(skb_realloc_headroom);
1550 * skb_copy_expand - copy and expand sk_buff
1551 * @skb: buffer to copy
1552 * @newheadroom: new free bytes at head
1553 * @newtailroom: new free bytes at tail
1554 * @gfp_mask: allocation priority
1556 * Make a copy of both an &sk_buff and its data and while doing so
1557 * allocate additional space.
1559 * This is used when the caller wishes to modify the data and needs a
1560 * private copy of the data to alter as well as more space for new fields.
1561 * Returns %NULL on failure or the pointer to the buffer
1562 * on success. The returned buffer has a reference count of 1.
1564 * You must pass %GFP_ATOMIC as the allocation priority if this function
1565 * is called from an interrupt.
1567 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1568 int newheadroom, int newtailroom,
1572 * Allocate the copy buffer
1574 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1575 gfp_mask, skb_alloc_rx_flag(skb),
1577 int oldheadroom = skb_headroom(skb);
1578 int head_copy_len, head_copy_off;
1583 skb_reserve(n, newheadroom);
1585 /* Set the tail pointer and length */
1586 skb_put(n, skb->len);
1588 head_copy_len = oldheadroom;
1590 if (newheadroom <= head_copy_len)
1591 head_copy_len = newheadroom;
1593 head_copy_off = newheadroom - head_copy_len;
1595 /* Copy the linear header and data. */
1596 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1597 skb->len + head_copy_len))
1600 copy_skb_header(n, skb);
1602 skb_headers_offset_update(n, newheadroom - oldheadroom);
1606 EXPORT_SYMBOL(skb_copy_expand);
1609 * __skb_pad - zero pad the tail of an skb
1610 * @skb: buffer to pad
1611 * @pad: space to pad
1612 * @free_on_error: free buffer on error
1614 * Ensure that a buffer is followed by a padding area that is zero
1615 * filled. Used by network drivers which may DMA or transfer data
1616 * beyond the buffer end onto the wire.
1618 * May return error in out of memory cases. The skb is freed on error
1619 * if @free_on_error is true.
1622 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1627 /* If the skbuff is non linear tailroom is always zero.. */
1628 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1629 memset(skb->data+skb->len, 0, pad);
1633 ntail = skb->data_len + pad - (skb->end - skb->tail);
1634 if (likely(skb_cloned(skb) || ntail > 0)) {
1635 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1640 /* FIXME: The use of this function with non-linear skb's really needs
1643 err = skb_linearize(skb);
1647 memset(skb->data + skb->len, 0, pad);
1655 EXPORT_SYMBOL(__skb_pad);
1658 * pskb_put - add data to the tail of a potentially fragmented buffer
1659 * @skb: start of the buffer to use
1660 * @tail: tail fragment of the buffer to use
1661 * @len: amount of data to add
1663 * This function extends the used data area of the potentially
1664 * fragmented buffer. @tail must be the last fragment of @skb -- or
1665 * @skb itself. If this would exceed the total buffer size the kernel
1666 * will panic. A pointer to the first byte of the extra data is
1670 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1673 skb->data_len += len;
1676 return skb_put(tail, len);
1678 EXPORT_SYMBOL_GPL(pskb_put);
1681 * skb_put - add data to a buffer
1682 * @skb: buffer to use
1683 * @len: amount of data to add
1685 * This function extends the used data area of the buffer. If this would
1686 * exceed the total buffer size the kernel will panic. A pointer to the
1687 * first byte of the extra data is returned.
1689 void *skb_put(struct sk_buff *skb, unsigned int len)
1691 void *tmp = skb_tail_pointer(skb);
1692 SKB_LINEAR_ASSERT(skb);
1695 if (unlikely(skb->tail > skb->end))
1696 skb_over_panic(skb, len, __builtin_return_address(0));
1699 EXPORT_SYMBOL(skb_put);
1702 * skb_push - add data to the start of a buffer
1703 * @skb: buffer to use
1704 * @len: amount of data to add
1706 * This function extends the used data area of the buffer at the buffer
1707 * start. If this would exceed the total buffer headroom the kernel will
1708 * panic. A pointer to the first byte of the extra data is returned.
1710 void *skb_push(struct sk_buff *skb, unsigned int len)
1714 if (unlikely(skb->data<skb->head))
1715 skb_under_panic(skb, len, __builtin_return_address(0));
1718 EXPORT_SYMBOL(skb_push);
1721 * skb_pull - remove data from the start of a buffer
1722 * @skb: buffer to use
1723 * @len: amount of data to remove
1725 * This function removes data from the start of a buffer, returning
1726 * the memory to the headroom. A pointer to the next data in the buffer
1727 * is returned. Once the data has been pulled future pushes will overwrite
1730 void *skb_pull(struct sk_buff *skb, unsigned int len)
1732 return skb_pull_inline(skb, len);
1734 EXPORT_SYMBOL(skb_pull);
1737 * skb_trim - remove end from a buffer
1738 * @skb: buffer to alter
1741 * Cut the length of a buffer down by removing data from the tail. If
1742 * the buffer is already under the length specified it is not modified.
1743 * The skb must be linear.
1745 void skb_trim(struct sk_buff *skb, unsigned int len)
1748 __skb_trim(skb, len);
1750 EXPORT_SYMBOL(skb_trim);
1752 /* Trims skb to length len. It can change skb pointers.
1755 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1757 struct sk_buff **fragp;
1758 struct sk_buff *frag;
1759 int offset = skb_headlen(skb);
1760 int nfrags = skb_shinfo(skb)->nr_frags;
1764 if (skb_cloned(skb) &&
1765 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1772 for (; i < nfrags; i++) {
1773 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1780 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1783 skb_shinfo(skb)->nr_frags = i;
1785 for (; i < nfrags; i++)
1786 skb_frag_unref(skb, i);
1788 if (skb_has_frag_list(skb))
1789 skb_drop_fraglist(skb);
1793 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1794 fragp = &frag->next) {
1795 int end = offset + frag->len;
1797 if (skb_shared(frag)) {
1798 struct sk_buff *nfrag;
1800 nfrag = skb_clone(frag, GFP_ATOMIC);
1801 if (unlikely(!nfrag))
1804 nfrag->next = frag->next;
1816 unlikely((err = pskb_trim(frag, len - offset))))
1820 skb_drop_list(&frag->next);
1825 if (len > skb_headlen(skb)) {
1826 skb->data_len -= skb->len - len;
1831 skb_set_tail_pointer(skb, len);
1834 if (!skb->sk || skb->destructor == sock_edemux)
1838 EXPORT_SYMBOL(___pskb_trim);
1841 * __pskb_pull_tail - advance tail of skb header
1842 * @skb: buffer to reallocate
1843 * @delta: number of bytes to advance tail
1845 * The function makes a sense only on a fragmented &sk_buff,
1846 * it expands header moving its tail forward and copying necessary
1847 * data from fragmented part.
1849 * &sk_buff MUST have reference count of 1.
1851 * Returns %NULL (and &sk_buff does not change) if pull failed
1852 * or value of new tail of skb in the case of success.
1854 * All the pointers pointing into skb header may change and must be
1855 * reloaded after call to this function.
1858 /* Moves tail of skb head forward, copying data from fragmented part,
1859 * when it is necessary.
1860 * 1. It may fail due to malloc failure.
1861 * 2. It may change skb pointers.
1863 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1865 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
1867 /* If skb has not enough free space at tail, get new one
1868 * plus 128 bytes for future expansions. If we have enough
1869 * room at tail, reallocate without expansion only if skb is cloned.
1871 int i, k, eat = (skb->tail + delta) - skb->end;
1873 if (eat > 0 || skb_cloned(skb)) {
1874 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1879 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1882 /* Optimization: no fragments, no reasons to preestimate
1883 * size of pulled pages. Superb.
1885 if (!skb_has_frag_list(skb))
1888 /* Estimate size of pulled pages. */
1890 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1891 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1898 /* If we need update frag list, we are in troubles.
1899 * Certainly, it possible to add an offset to skb data,
1900 * but taking into account that pulling is expected to
1901 * be very rare operation, it is worth to fight against
1902 * further bloating skb head and crucify ourselves here instead.
1903 * Pure masohism, indeed. 8)8)
1906 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1907 struct sk_buff *clone = NULL;
1908 struct sk_buff *insp = NULL;
1913 if (list->len <= eat) {
1914 /* Eaten as whole. */
1919 /* Eaten partially. */
1921 if (skb_shared(list)) {
1922 /* Sucks! We need to fork list. :-( */
1923 clone = skb_clone(list, GFP_ATOMIC);
1929 /* This may be pulled without
1933 if (!pskb_pull(list, eat)) {
1941 /* Free pulled out fragments. */
1942 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1943 skb_shinfo(skb)->frag_list = list->next;
1946 /* And insert new clone at head. */
1949 skb_shinfo(skb)->frag_list = clone;
1952 /* Success! Now we may commit changes to skb data. */
1957 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1958 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1961 skb_frag_unref(skb, i);
1964 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1966 skb_shinfo(skb)->frags[k].page_offset += eat;
1967 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1975 skb_shinfo(skb)->nr_frags = k;
1979 skb->data_len -= delta;
1982 skb_zcopy_clear(skb, false);
1984 return skb_tail_pointer(skb);
1986 EXPORT_SYMBOL(__pskb_pull_tail);
1989 * skb_copy_bits - copy bits from skb to kernel buffer
1991 * @offset: offset in source
1992 * @to: destination buffer
1993 * @len: number of bytes to copy
1995 * Copy the specified number of bytes from the source skb to the
1996 * destination buffer.
1999 * If its prototype is ever changed,
2000 * check arch/{*}/net/{*}.S files,
2001 * since it is called from BPF assembly code.
2003 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2005 int start = skb_headlen(skb);
2006 struct sk_buff *frag_iter;
2009 if (offset > (int)skb->len - len)
2013 if ((copy = start - offset) > 0) {
2016 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2017 if ((len -= copy) == 0)
2023 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2025 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2027 WARN_ON(start > offset + len);
2029 end = start + skb_frag_size(f);
2030 if ((copy = end - offset) > 0) {
2031 u32 p_off, p_len, copied;
2038 skb_frag_foreach_page(f,
2039 f->page_offset + offset - start,
2040 copy, p, p_off, p_len, copied) {
2041 vaddr = kmap_atomic(p);
2042 memcpy(to + copied, vaddr + p_off, p_len);
2043 kunmap_atomic(vaddr);
2046 if ((len -= copy) == 0)
2054 skb_walk_frags(skb, frag_iter) {
2057 WARN_ON(start > offset + len);
2059 end = start + frag_iter->len;
2060 if ((copy = end - offset) > 0) {
2063 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2065 if ((len -= copy) == 0)
2079 EXPORT_SYMBOL(skb_copy_bits);
2082 * Callback from splice_to_pipe(), if we need to release some pages
2083 * at the end of the spd in case we error'ed out in filling the pipe.
2085 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2087 put_page(spd->pages[i]);
2090 static struct page *linear_to_page(struct page *page, unsigned int *len,
2091 unsigned int *offset,
2094 struct page_frag *pfrag = sk_page_frag(sk);
2096 if (!sk_page_frag_refill(sk, pfrag))
2099 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2101 memcpy(page_address(pfrag->page) + pfrag->offset,
2102 page_address(page) + *offset, *len);
2103 *offset = pfrag->offset;
2104 pfrag->offset += *len;
2109 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2111 unsigned int offset)
2113 return spd->nr_pages &&
2114 spd->pages[spd->nr_pages - 1] == page &&
2115 (spd->partial[spd->nr_pages - 1].offset +
2116 spd->partial[spd->nr_pages - 1].len == offset);
2120 * Fill page/offset/length into spd, if it can hold more pages.
2122 static bool spd_fill_page(struct splice_pipe_desc *spd,
2123 struct pipe_inode_info *pipe, struct page *page,
2124 unsigned int *len, unsigned int offset,
2128 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2132 page = linear_to_page(page, len, &offset, sk);
2136 if (spd_can_coalesce(spd, page, offset)) {
2137 spd->partial[spd->nr_pages - 1].len += *len;
2141 spd->pages[spd->nr_pages] = page;
2142 spd->partial[spd->nr_pages].len = *len;
2143 spd->partial[spd->nr_pages].offset = offset;
2149 static bool __splice_segment(struct page *page, unsigned int poff,
2150 unsigned int plen, unsigned int *off,
2152 struct splice_pipe_desc *spd, bool linear,
2154 struct pipe_inode_info *pipe)
2159 /* skip this segment if already processed */
2165 /* ignore any bits we already processed */
2171 unsigned int flen = min(*len, plen);
2173 if (spd_fill_page(spd, pipe, page, &flen, poff,
2179 } while (*len && plen);
2185 * Map linear and fragment data from the skb to spd. It reports true if the
2186 * pipe is full or if we already spliced the requested length.
2188 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2189 unsigned int *offset, unsigned int *len,
2190 struct splice_pipe_desc *spd, struct sock *sk)
2193 struct sk_buff *iter;
2195 /* map the linear part :
2196 * If skb->head_frag is set, this 'linear' part is backed by a
2197 * fragment, and if the head is not shared with any clones then
2198 * we can avoid a copy since we own the head portion of this page.
2200 if (__splice_segment(virt_to_page(skb->data),
2201 (unsigned long) skb->data & (PAGE_SIZE - 1),
2204 skb_head_is_locked(skb),
2209 * then map the fragments
2211 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2212 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2214 if (__splice_segment(skb_frag_page(f),
2215 f->page_offset, skb_frag_size(f),
2216 offset, len, spd, false, sk, pipe))
2220 skb_walk_frags(skb, iter) {
2221 if (*offset >= iter->len) {
2222 *offset -= iter->len;
2225 /* __skb_splice_bits() only fails if the output has no room
2226 * left, so no point in going over the frag_list for the error
2229 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2237 * Map data from the skb to a pipe. Should handle both the linear part,
2238 * the fragments, and the frag list.
2240 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2241 struct pipe_inode_info *pipe, unsigned int tlen,
2244 struct partial_page partial[MAX_SKB_FRAGS];
2245 struct page *pages[MAX_SKB_FRAGS];
2246 struct splice_pipe_desc spd = {
2249 .nr_pages_max = MAX_SKB_FRAGS,
2250 .ops = &nosteal_pipe_buf_ops,
2251 .spd_release = sock_spd_release,
2255 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2258 ret = splice_to_pipe(pipe, &spd);
2262 EXPORT_SYMBOL_GPL(skb_splice_bits);
2264 /* Send skb data on a socket. Socket must be locked. */
2265 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2268 unsigned int orig_len = len;
2269 struct sk_buff *head = skb;
2270 unsigned short fragidx;
2275 /* Deal with head data */
2276 while (offset < skb_headlen(skb) && len) {
2280 slen = min_t(int, len, skb_headlen(skb) - offset);
2281 kv.iov_base = skb->data + offset;
2283 memset(&msg, 0, sizeof(msg));
2285 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2293 /* All the data was skb head? */
2297 /* Make offset relative to start of frags */
2298 offset -= skb_headlen(skb);
2300 /* Find where we are in frag list */
2301 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2302 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2304 if (offset < frag->size)
2307 offset -= frag->size;
2310 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2311 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2313 slen = min_t(size_t, len, frag->size - offset);
2316 ret = kernel_sendpage_locked(sk, frag->page.p,
2317 frag->page_offset + offset,
2318 slen, MSG_DONTWAIT);
2331 /* Process any frag lists */
2334 if (skb_has_frag_list(skb)) {
2335 skb = skb_shinfo(skb)->frag_list;
2338 } else if (skb->next) {
2345 return orig_len - len;
2348 return orig_len == len ? ret : orig_len - len;
2350 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2352 /* Send skb data on a socket. */
2353 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
2358 ret = skb_send_sock_locked(sk, skb, offset, len);
2363 EXPORT_SYMBOL_GPL(skb_send_sock);
2366 * skb_store_bits - store bits from kernel buffer to skb
2367 * @skb: destination buffer
2368 * @offset: offset in destination
2369 * @from: source buffer
2370 * @len: number of bytes to copy
2372 * Copy the specified number of bytes from the source buffer to the
2373 * destination skb. This function handles all the messy bits of
2374 * traversing fragment lists and such.
2377 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2379 int start = skb_headlen(skb);
2380 struct sk_buff *frag_iter;
2383 if (offset > (int)skb->len - len)
2386 if ((copy = start - offset) > 0) {
2389 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2390 if ((len -= copy) == 0)
2396 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2397 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2400 WARN_ON(start > offset + len);
2402 end = start + skb_frag_size(frag);
2403 if ((copy = end - offset) > 0) {
2404 u32 p_off, p_len, copied;
2411 skb_frag_foreach_page(frag,
2412 frag->page_offset + offset - start,
2413 copy, p, p_off, p_len, copied) {
2414 vaddr = kmap_atomic(p);
2415 memcpy(vaddr + p_off, from + copied, p_len);
2416 kunmap_atomic(vaddr);
2419 if ((len -= copy) == 0)
2427 skb_walk_frags(skb, frag_iter) {
2430 WARN_ON(start > offset + len);
2432 end = start + frag_iter->len;
2433 if ((copy = end - offset) > 0) {
2436 if (skb_store_bits(frag_iter, offset - start,
2439 if ((len -= copy) == 0)
2452 EXPORT_SYMBOL(skb_store_bits);
2454 /* Checksum skb data. */
2455 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2456 __wsum csum, const struct skb_checksum_ops *ops)
2458 int start = skb_headlen(skb);
2459 int i, copy = start - offset;
2460 struct sk_buff *frag_iter;
2463 /* Checksum header. */
2467 csum = ops->update(skb->data + offset, copy, csum);
2468 if ((len -= copy) == 0)
2474 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2476 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2478 WARN_ON(start > offset + len);
2480 end = start + skb_frag_size(frag);
2481 if ((copy = end - offset) > 0) {
2482 u32 p_off, p_len, copied;
2490 skb_frag_foreach_page(frag,
2491 frag->page_offset + offset - start,
2492 copy, p, p_off, p_len, copied) {
2493 vaddr = kmap_atomic(p);
2494 csum2 = ops->update(vaddr + p_off, p_len, 0);
2495 kunmap_atomic(vaddr);
2496 csum = ops->combine(csum, csum2, pos, p_len);
2507 skb_walk_frags(skb, frag_iter) {
2510 WARN_ON(start > offset + len);
2512 end = start + frag_iter->len;
2513 if ((copy = end - offset) > 0) {
2517 csum2 = __skb_checksum(frag_iter, offset - start,
2519 csum = ops->combine(csum, csum2, pos, copy);
2520 if ((len -= copy) == 0)
2531 EXPORT_SYMBOL(__skb_checksum);
2533 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2534 int len, __wsum csum)
2536 const struct skb_checksum_ops ops = {
2537 .update = csum_partial_ext,
2538 .combine = csum_block_add_ext,
2541 return __skb_checksum(skb, offset, len, csum, &ops);
2543 EXPORT_SYMBOL(skb_checksum);
2545 /* Both of above in one bottle. */
2547 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2548 u8 *to, int len, __wsum csum)
2550 int start = skb_headlen(skb);
2551 int i, copy = start - offset;
2552 struct sk_buff *frag_iter;
2559 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2561 if ((len -= copy) == 0)
2568 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2571 WARN_ON(start > offset + len);
2573 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2574 if ((copy = end - offset) > 0) {
2575 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2576 u32 p_off, p_len, copied;
2584 skb_frag_foreach_page(frag,
2585 frag->page_offset + offset - start,
2586 copy, p, p_off, p_len, copied) {
2587 vaddr = kmap_atomic(p);
2588 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2591 kunmap_atomic(vaddr);
2592 csum = csum_block_add(csum, csum2, pos);
2604 skb_walk_frags(skb, frag_iter) {
2608 WARN_ON(start > offset + len);
2610 end = start + frag_iter->len;
2611 if ((copy = end - offset) > 0) {
2614 csum2 = skb_copy_and_csum_bits(frag_iter,
2617 csum = csum_block_add(csum, csum2, pos);
2618 if ((len -= copy) == 0)
2629 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2631 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2633 net_warn_ratelimited(
2634 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2639 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2640 int offset, int len)
2642 net_warn_ratelimited(
2643 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2648 static const struct skb_checksum_ops default_crc32c_ops = {
2649 .update = warn_crc32c_csum_update,
2650 .combine = warn_crc32c_csum_combine,
2653 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2654 &default_crc32c_ops;
2655 EXPORT_SYMBOL(crc32c_csum_stub);
2658 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2659 * @from: source buffer
2661 * Calculates the amount of linear headroom needed in the 'to' skb passed
2662 * into skb_zerocopy().
2665 skb_zerocopy_headlen(const struct sk_buff *from)
2667 unsigned int hlen = 0;
2669 if (!from->head_frag ||
2670 skb_headlen(from) < L1_CACHE_BYTES ||
2671 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2672 hlen = skb_headlen(from);
2674 if (skb_has_frag_list(from))
2679 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2682 * skb_zerocopy - Zero copy skb to skb
2683 * @to: destination buffer
2684 * @from: source buffer
2685 * @len: number of bytes to copy from source buffer
2686 * @hlen: size of linear headroom in destination buffer
2688 * Copies up to `len` bytes from `from` to `to` by creating references
2689 * to the frags in the source buffer.
2691 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2692 * headroom in the `to` buffer.
2695 * 0: everything is OK
2696 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2697 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2700 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2703 int plen = 0; /* length of skb->head fragment */
2706 unsigned int offset;
2708 BUG_ON(!from->head_frag && !hlen);
2710 /* dont bother with small payloads */
2711 if (len <= skb_tailroom(to))
2712 return skb_copy_bits(from, 0, skb_put(to, len), len);
2715 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2720 plen = min_t(int, skb_headlen(from), len);
2722 page = virt_to_head_page(from->head);
2723 offset = from->data - (unsigned char *)page_address(page);
2724 __skb_fill_page_desc(to, 0, page, offset, plen);
2731 to->truesize += len + plen;
2732 to->len += len + plen;
2733 to->data_len += len + plen;
2735 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2739 skb_zerocopy_clone(to, from, GFP_ATOMIC);
2741 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2744 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2745 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2746 len -= skb_shinfo(to)->frags[j].size;
2747 skb_frag_ref(to, j);
2750 skb_shinfo(to)->nr_frags = j;
2754 EXPORT_SYMBOL_GPL(skb_zerocopy);
2756 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2761 if (skb->ip_summed == CHECKSUM_PARTIAL)
2762 csstart = skb_checksum_start_offset(skb);
2764 csstart = skb_headlen(skb);
2766 BUG_ON(csstart > skb_headlen(skb));
2768 skb_copy_from_linear_data(skb, to, csstart);
2771 if (csstart != skb->len)
2772 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2773 skb->len - csstart, 0);
2775 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2776 long csstuff = csstart + skb->csum_offset;
2778 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2781 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2784 * skb_dequeue - remove from the head of the queue
2785 * @list: list to dequeue from
2787 * Remove the head of the list. The list lock is taken so the function
2788 * may be used safely with other locking list functions. The head item is
2789 * returned or %NULL if the list is empty.
2792 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2794 unsigned long flags;
2795 struct sk_buff *result;
2797 spin_lock_irqsave(&list->lock, flags);
2798 result = __skb_dequeue(list);
2799 spin_unlock_irqrestore(&list->lock, flags);
2802 EXPORT_SYMBOL(skb_dequeue);
2805 * skb_dequeue_tail - remove from the tail of the queue
2806 * @list: list to dequeue from
2808 * Remove the tail of the list. The list lock is taken so the function
2809 * may be used safely with other locking list functions. The tail item is
2810 * returned or %NULL if the list is empty.
2812 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2814 unsigned long flags;
2815 struct sk_buff *result;
2817 spin_lock_irqsave(&list->lock, flags);
2818 result = __skb_dequeue_tail(list);
2819 spin_unlock_irqrestore(&list->lock, flags);
2822 EXPORT_SYMBOL(skb_dequeue_tail);
2825 * skb_queue_purge - empty a list
2826 * @list: list to empty
2828 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2829 * the list and one reference dropped. This function takes the list
2830 * lock and is atomic with respect to other list locking functions.
2832 void skb_queue_purge(struct sk_buff_head *list)
2834 struct sk_buff *skb;
2835 while ((skb = skb_dequeue(list)) != NULL)
2838 EXPORT_SYMBOL(skb_queue_purge);
2841 * skb_rbtree_purge - empty a skb rbtree
2842 * @root: root of the rbtree to empty
2844 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
2845 * the list and one reference dropped. This function does not take
2846 * any lock. Synchronization should be handled by the caller (e.g., TCP
2847 * out-of-order queue is protected by the socket lock).
2849 void skb_rbtree_purge(struct rb_root *root)
2851 struct sk_buff *skb, *next;
2853 rbtree_postorder_for_each_entry_safe(skb, next, root, rbnode)
2860 * skb_queue_head - queue a buffer at the list head
2861 * @list: list to use
2862 * @newsk: buffer to queue
2864 * Queue a buffer at the start of the list. This function takes the
2865 * list lock and can be used safely with other locking &sk_buff functions
2868 * A buffer cannot be placed on two lists at the same time.
2870 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2872 unsigned long flags;
2874 spin_lock_irqsave(&list->lock, flags);
2875 __skb_queue_head(list, newsk);
2876 spin_unlock_irqrestore(&list->lock, flags);
2878 EXPORT_SYMBOL(skb_queue_head);
2881 * skb_queue_tail - queue a buffer at the list tail
2882 * @list: list to use
2883 * @newsk: buffer to queue
2885 * Queue a buffer at the tail of the list. This function takes the
2886 * list lock and can be used safely with other locking &sk_buff functions
2889 * A buffer cannot be placed on two lists at the same time.
2891 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2893 unsigned long flags;
2895 spin_lock_irqsave(&list->lock, flags);
2896 __skb_queue_tail(list, newsk);
2897 spin_unlock_irqrestore(&list->lock, flags);
2899 EXPORT_SYMBOL(skb_queue_tail);
2902 * skb_unlink - remove a buffer from a list
2903 * @skb: buffer to remove
2904 * @list: list to use
2906 * Remove a packet from a list. The list locks are taken and this
2907 * function is atomic with respect to other list locked calls
2909 * You must know what list the SKB is on.
2911 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2913 unsigned long flags;
2915 spin_lock_irqsave(&list->lock, flags);
2916 __skb_unlink(skb, list);
2917 spin_unlock_irqrestore(&list->lock, flags);
2919 EXPORT_SYMBOL(skb_unlink);
2922 * skb_append - append a buffer
2923 * @old: buffer to insert after
2924 * @newsk: buffer to insert
2925 * @list: list to use
2927 * Place a packet after a given packet in a list. The list locks are taken
2928 * and this function is atomic with respect to other list locked calls.
2929 * A buffer cannot be placed on two lists at the same time.
2931 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2933 unsigned long flags;
2935 spin_lock_irqsave(&list->lock, flags);
2936 __skb_queue_after(list, old, newsk);
2937 spin_unlock_irqrestore(&list->lock, flags);
2939 EXPORT_SYMBOL(skb_append);
2942 * skb_insert - insert a buffer
2943 * @old: buffer to insert before
2944 * @newsk: buffer to insert
2945 * @list: list to use
2947 * Place a packet before a given packet in a list. The list locks are
2948 * taken and this function is atomic with respect to other list locked
2951 * A buffer cannot be placed on two lists at the same time.
2953 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2955 unsigned long flags;
2957 spin_lock_irqsave(&list->lock, flags);
2958 __skb_insert(newsk, old->prev, old, list);
2959 spin_unlock_irqrestore(&list->lock, flags);
2961 EXPORT_SYMBOL(skb_insert);
2963 static inline void skb_split_inside_header(struct sk_buff *skb,
2964 struct sk_buff* skb1,
2965 const u32 len, const int pos)
2969 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2971 /* And move data appendix as is. */
2972 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2973 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2975 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2976 skb_shinfo(skb)->nr_frags = 0;
2977 skb1->data_len = skb->data_len;
2978 skb1->len += skb1->data_len;
2981 skb_set_tail_pointer(skb, len);
2984 static inline void skb_split_no_header(struct sk_buff *skb,
2985 struct sk_buff* skb1,
2986 const u32 len, int pos)
2989 const int nfrags = skb_shinfo(skb)->nr_frags;
2991 skb_shinfo(skb)->nr_frags = 0;
2992 skb1->len = skb1->data_len = skb->len - len;
2994 skb->data_len = len - pos;
2996 for (i = 0; i < nfrags; i++) {
2997 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2999 if (pos + size > len) {
3000 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3004 * We have two variants in this case:
3005 * 1. Move all the frag to the second
3006 * part, if it is possible. F.e.
3007 * this approach is mandatory for TUX,
3008 * where splitting is expensive.
3009 * 2. Split is accurately. We make this.
3011 skb_frag_ref(skb, i);
3012 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
3013 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3014 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3015 skb_shinfo(skb)->nr_frags++;
3019 skb_shinfo(skb)->nr_frags++;
3022 skb_shinfo(skb1)->nr_frags = k;
3026 * skb_split - Split fragmented skb to two parts at length len.
3027 * @skb: the buffer to split
3028 * @skb1: the buffer to receive the second part
3029 * @len: new length for skb
3031 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3033 int pos = skb_headlen(skb);
3035 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3037 skb_zerocopy_clone(skb1, skb, 0);
3038 if (len < pos) /* Split line is inside header. */
3039 skb_split_inside_header(skb, skb1, len, pos);
3040 else /* Second chunk has no header, nothing to copy. */
3041 skb_split_no_header(skb, skb1, len, pos);
3043 EXPORT_SYMBOL(skb_split);
3045 /* Shifting from/to a cloned skb is a no-go.
3047 * Caller cannot keep skb_shinfo related pointers past calling here!
3049 static int skb_prepare_for_shift(struct sk_buff *skb)
3051 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3055 * skb_shift - Shifts paged data partially from skb to another
3056 * @tgt: buffer into which tail data gets added
3057 * @skb: buffer from which the paged data comes from
3058 * @shiftlen: shift up to this many bytes
3060 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3061 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3062 * It's up to caller to free skb if everything was shifted.
3064 * If @tgt runs out of frags, the whole operation is aborted.
3066 * Skb cannot include anything else but paged data while tgt is allowed
3067 * to have non-paged data as well.
3069 * TODO: full sized shift could be optimized but that would need
3070 * specialized skb free'er to handle frags without up-to-date nr_frags.
3072 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3074 int from, to, merge, todo;
3075 struct skb_frag_struct *fragfrom, *fragto;
3077 BUG_ON(shiftlen > skb->len);
3079 if (skb_headlen(skb))
3081 if (skb_zcopy(tgt) || skb_zcopy(skb))
3086 to = skb_shinfo(tgt)->nr_frags;
3087 fragfrom = &skb_shinfo(skb)->frags[from];
3089 /* Actual merge is delayed until the point when we know we can
3090 * commit all, so that we don't have to undo partial changes
3093 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3094 fragfrom->page_offset)) {
3099 todo -= skb_frag_size(fragfrom);
3101 if (skb_prepare_for_shift(skb) ||
3102 skb_prepare_for_shift(tgt))
3105 /* All previous frag pointers might be stale! */
3106 fragfrom = &skb_shinfo(skb)->frags[from];
3107 fragto = &skb_shinfo(tgt)->frags[merge];
3109 skb_frag_size_add(fragto, shiftlen);
3110 skb_frag_size_sub(fragfrom, shiftlen);
3111 fragfrom->page_offset += shiftlen;
3119 /* Skip full, not-fitting skb to avoid expensive operations */
3120 if ((shiftlen == skb->len) &&
3121 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3124 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3127 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3128 if (to == MAX_SKB_FRAGS)
3131 fragfrom = &skb_shinfo(skb)->frags[from];
3132 fragto = &skb_shinfo(tgt)->frags[to];
3134 if (todo >= skb_frag_size(fragfrom)) {
3135 *fragto = *fragfrom;
3136 todo -= skb_frag_size(fragfrom);
3141 __skb_frag_ref(fragfrom);
3142 fragto->page = fragfrom->page;
3143 fragto->page_offset = fragfrom->page_offset;
3144 skb_frag_size_set(fragto, todo);
3146 fragfrom->page_offset += todo;
3147 skb_frag_size_sub(fragfrom, todo);
3155 /* Ready to "commit" this state change to tgt */
3156 skb_shinfo(tgt)->nr_frags = to;
3159 fragfrom = &skb_shinfo(skb)->frags[0];
3160 fragto = &skb_shinfo(tgt)->frags[merge];
3162 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3163 __skb_frag_unref(fragfrom);
3166 /* Reposition in the original skb */
3168 while (from < skb_shinfo(skb)->nr_frags)
3169 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3170 skb_shinfo(skb)->nr_frags = to;
3172 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3175 /* Most likely the tgt won't ever need its checksum anymore, skb on
3176 * the other hand might need it if it needs to be resent
3178 tgt->ip_summed = CHECKSUM_PARTIAL;
3179 skb->ip_summed = CHECKSUM_PARTIAL;
3181 /* Yak, is it really working this way? Some helper please? */
3182 skb->len -= shiftlen;
3183 skb->data_len -= shiftlen;
3184 skb->truesize -= shiftlen;
3185 tgt->len += shiftlen;
3186 tgt->data_len += shiftlen;
3187 tgt->truesize += shiftlen;
3193 * skb_prepare_seq_read - Prepare a sequential read of skb data
3194 * @skb: the buffer to read
3195 * @from: lower offset of data to be read
3196 * @to: upper offset of data to be read
3197 * @st: state variable
3199 * Initializes the specified state variable. Must be called before
3200 * invoking skb_seq_read() for the first time.
3202 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3203 unsigned int to, struct skb_seq_state *st)
3205 st->lower_offset = from;
3206 st->upper_offset = to;
3207 st->root_skb = st->cur_skb = skb;
3208 st->frag_idx = st->stepped_offset = 0;
3209 st->frag_data = NULL;
3211 EXPORT_SYMBOL(skb_prepare_seq_read);
3214 * skb_seq_read - Sequentially read skb data
3215 * @consumed: number of bytes consumed by the caller so far
3216 * @data: destination pointer for data to be returned
3217 * @st: state variable
3219 * Reads a block of skb data at @consumed relative to the
3220 * lower offset specified to skb_prepare_seq_read(). Assigns
3221 * the head of the data block to @data and returns the length
3222 * of the block or 0 if the end of the skb data or the upper
3223 * offset has been reached.
3225 * The caller is not required to consume all of the data
3226 * returned, i.e. @consumed is typically set to the number
3227 * of bytes already consumed and the next call to
3228 * skb_seq_read() will return the remaining part of the block.
3230 * Note 1: The size of each block of data returned can be arbitrary,
3231 * this limitation is the cost for zerocopy sequential
3232 * reads of potentially non linear data.
3234 * Note 2: Fragment lists within fragments are not implemented
3235 * at the moment, state->root_skb could be replaced with
3236 * a stack for this purpose.
3238 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3239 struct skb_seq_state *st)
3241 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3244 if (unlikely(abs_offset >= st->upper_offset)) {
3245 if (st->frag_data) {
3246 kunmap_atomic(st->frag_data);
3247 st->frag_data = NULL;
3253 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3255 if (abs_offset < block_limit && !st->frag_data) {
3256 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3257 return block_limit - abs_offset;
3260 if (st->frag_idx == 0 && !st->frag_data)
3261 st->stepped_offset += skb_headlen(st->cur_skb);
3263 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3264 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3265 block_limit = skb_frag_size(frag) + st->stepped_offset;
3267 if (abs_offset < block_limit) {
3269 st->frag_data = kmap_atomic(skb_frag_page(frag));
3271 *data = (u8 *) st->frag_data + frag->page_offset +
3272 (abs_offset - st->stepped_offset);
3274 return block_limit - abs_offset;
3277 if (st->frag_data) {
3278 kunmap_atomic(st->frag_data);
3279 st->frag_data = NULL;
3283 st->stepped_offset += skb_frag_size(frag);
3286 if (st->frag_data) {
3287 kunmap_atomic(st->frag_data);
3288 st->frag_data = NULL;
3291 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3292 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3295 } else if (st->cur_skb->next) {
3296 st->cur_skb = st->cur_skb->next;
3303 EXPORT_SYMBOL(skb_seq_read);
3306 * skb_abort_seq_read - Abort a sequential read of skb data
3307 * @st: state variable
3309 * Must be called if skb_seq_read() was not called until it
3312 void skb_abort_seq_read(struct skb_seq_state *st)
3315 kunmap_atomic(st->frag_data);
3317 EXPORT_SYMBOL(skb_abort_seq_read);
3319 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3321 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3322 struct ts_config *conf,
3323 struct ts_state *state)
3325 return skb_seq_read(offset, text, TS_SKB_CB(state));
3328 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3330 skb_abort_seq_read(TS_SKB_CB(state));
3334 * skb_find_text - Find a text pattern in skb data
3335 * @skb: the buffer to look in
3336 * @from: search offset
3338 * @config: textsearch configuration
3340 * Finds a pattern in the skb data according to the specified
3341 * textsearch configuration. Use textsearch_next() to retrieve
3342 * subsequent occurrences of the pattern. Returns the offset
3343 * to the first occurrence or UINT_MAX if no match was found.
3345 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3346 unsigned int to, struct ts_config *config)
3348 struct ts_state state;
3351 config->get_next_block = skb_ts_get_next_block;
3352 config->finish = skb_ts_finish;
3354 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3356 ret = textsearch_find(config, &state);
3357 return (ret <= to - from ? ret : UINT_MAX);
3359 EXPORT_SYMBOL(skb_find_text);
3362 * skb_append_datato_frags - append the user data to a skb
3363 * @sk: sock structure
3364 * @skb: skb structure to be appended with user data.
3365 * @getfrag: call back function to be used for getting the user data
3366 * @from: pointer to user message iov
3367 * @length: length of the iov message
3369 * Description: This procedure append the user data in the fragment part
3370 * of the skb if any page alloc fails user this procedure returns -ENOMEM
3372 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
3373 int (*getfrag)(void *from, char *to, int offset,
3374 int len, int odd, struct sk_buff *skb),
3375 void *from, int length)
3377 int frg_cnt = skb_shinfo(skb)->nr_frags;
3381 struct page_frag *pfrag = ¤t->task_frag;
3384 /* Return error if we don't have space for new frag */
3385 if (frg_cnt >= MAX_SKB_FRAGS)
3388 if (!sk_page_frag_refill(sk, pfrag))
3391 /* copy the user data to page */
3392 copy = min_t(int, length, pfrag->size - pfrag->offset);
3394 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
3395 offset, copy, 0, skb);
3399 /* copy was successful so update the size parameters */
3400 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
3403 pfrag->offset += copy;
3404 get_page(pfrag->page);
3406 skb->truesize += copy;
3407 refcount_add(copy, &sk->sk_wmem_alloc);
3409 skb->data_len += copy;
3413 } while (length > 0);
3417 EXPORT_SYMBOL(skb_append_datato_frags);
3419 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3420 int offset, size_t size)
3422 int i = skb_shinfo(skb)->nr_frags;
3424 if (skb_can_coalesce(skb, i, page, offset)) {
3425 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3426 } else if (i < MAX_SKB_FRAGS) {
3428 skb_fill_page_desc(skb, i, page, offset, size);
3435 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3438 * skb_pull_rcsum - pull skb and update receive checksum
3439 * @skb: buffer to update
3440 * @len: length of data pulled
3442 * This function performs an skb_pull on the packet and updates
3443 * the CHECKSUM_COMPLETE checksum. It should be used on
3444 * receive path processing instead of skb_pull unless you know
3445 * that the checksum difference is zero (e.g., a valid IP header)
3446 * or you are setting ip_summed to CHECKSUM_NONE.
3448 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3450 unsigned char *data = skb->data;
3452 BUG_ON(len > skb->len);
3453 __skb_pull(skb, len);
3454 skb_postpull_rcsum(skb, data, len);
3457 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3460 * skb_segment - Perform protocol segmentation on skb.
3461 * @head_skb: buffer to segment
3462 * @features: features for the output path (see dev->features)
3464 * This function performs segmentation on the given skb. It returns
3465 * a pointer to the first in a list of new skbs for the segments.
3466 * In case of error it returns ERR_PTR(err).
3468 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3469 netdev_features_t features)
3471 struct sk_buff *segs = NULL;
3472 struct sk_buff *tail = NULL;
3473 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3474 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3475 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3476 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3477 struct sk_buff *frag_skb = head_skb;
3478 unsigned int offset = doffset;
3479 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3480 unsigned int partial_segs = 0;
3481 unsigned int headroom;
3482 unsigned int len = head_skb->len;
3485 int nfrags = skb_shinfo(head_skb)->nr_frags;
3491 __skb_push(head_skb, doffset);
3492 proto = skb_network_protocol(head_skb, &dummy);
3493 if (unlikely(!proto))
3494 return ERR_PTR(-EINVAL);
3496 sg = !!(features & NETIF_F_SG);
3497 csum = !!can_checksum_protocol(features, proto);
3499 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3500 if (!(features & NETIF_F_GSO_PARTIAL)) {
3501 struct sk_buff *iter;
3502 unsigned int frag_len;
3505 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3508 /* If we get here then all the required
3509 * GSO features except frag_list are supported.
3510 * Try to split the SKB to multiple GSO SKBs
3511 * with no frag_list.
3512 * Currently we can do that only when the buffers don't
3513 * have a linear part and all the buffers except
3514 * the last are of the same length.
3516 frag_len = list_skb->len;
3517 skb_walk_frags(head_skb, iter) {
3518 if (frag_len != iter->len && iter->next)
3520 if (skb_headlen(iter) && !iter->head_frag)
3526 if (len != frag_len)
3530 /* GSO partial only requires that we trim off any excess that
3531 * doesn't fit into an MSS sized block, so take care of that
3534 partial_segs = len / mss;
3535 if (partial_segs > 1)
3536 mss *= partial_segs;
3542 headroom = skb_headroom(head_skb);
3543 pos = skb_headlen(head_skb);
3546 struct sk_buff *nskb;
3547 skb_frag_t *nskb_frag;
3551 if (unlikely(mss == GSO_BY_FRAGS)) {
3552 len = list_skb->len;
3554 len = head_skb->len - offset;
3559 hsize = skb_headlen(head_skb) - offset;
3562 if (hsize > len || !sg)
3565 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3566 (skb_headlen(list_skb) == len || sg)) {
3567 BUG_ON(skb_headlen(list_skb) > len);
3570 nfrags = skb_shinfo(list_skb)->nr_frags;
3571 frag = skb_shinfo(list_skb)->frags;
3572 frag_skb = list_skb;
3573 pos += skb_headlen(list_skb);
3575 while (pos < offset + len) {
3576 BUG_ON(i >= nfrags);
3578 size = skb_frag_size(frag);
3579 if (pos + size > offset + len)
3587 nskb = skb_clone(list_skb, GFP_ATOMIC);
3588 list_skb = list_skb->next;
3590 if (unlikely(!nskb))
3593 if (unlikely(pskb_trim(nskb, len))) {
3598 hsize = skb_end_offset(nskb);
3599 if (skb_cow_head(nskb, doffset + headroom)) {
3604 nskb->truesize += skb_end_offset(nskb) - hsize;
3605 skb_release_head_state(nskb);
3606 __skb_push(nskb, doffset);
3608 nskb = __alloc_skb(hsize + doffset + headroom,
3609 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3612 if (unlikely(!nskb))
3615 skb_reserve(nskb, headroom);
3616 __skb_put(nskb, doffset);
3625 __copy_skb_header(nskb, head_skb);
3627 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3628 skb_reset_mac_len(nskb);
3630 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3631 nskb->data - tnl_hlen,
3632 doffset + tnl_hlen);
3634 if (nskb->len == len + doffset)
3635 goto perform_csum_check;
3638 if (!nskb->remcsum_offload)
3639 nskb->ip_summed = CHECKSUM_NONE;
3640 SKB_GSO_CB(nskb)->csum =
3641 skb_copy_and_csum_bits(head_skb, offset,
3644 SKB_GSO_CB(nskb)->csum_start =
3645 skb_headroom(nskb) + doffset;
3649 nskb_frag = skb_shinfo(nskb)->frags;
3651 skb_copy_from_linear_data_offset(head_skb, offset,
3652 skb_put(nskb, hsize), hsize);
3654 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3656 if (skb_zerocopy_clone(nskb, head_skb, GFP_ATOMIC))
3659 while (pos < offset + len) {
3661 BUG_ON(skb_headlen(list_skb));
3664 nfrags = skb_shinfo(list_skb)->nr_frags;
3665 frag = skb_shinfo(list_skb)->frags;
3666 frag_skb = list_skb;
3670 list_skb = list_skb->next;
3673 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3675 net_warn_ratelimited(
3676 "skb_segment: too many frags: %u %u\n",
3681 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3685 __skb_frag_ref(nskb_frag);
3686 size = skb_frag_size(nskb_frag);
3689 nskb_frag->page_offset += offset - pos;
3690 skb_frag_size_sub(nskb_frag, offset - pos);
3693 skb_shinfo(nskb)->nr_frags++;
3695 if (pos + size <= offset + len) {
3700 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3708 nskb->data_len = len - hsize;
3709 nskb->len += nskb->data_len;
3710 nskb->truesize += nskb->data_len;
3714 if (skb_has_shared_frag(nskb)) {
3715 err = __skb_linearize(nskb);
3719 if (!nskb->remcsum_offload)
3720 nskb->ip_summed = CHECKSUM_NONE;
3721 SKB_GSO_CB(nskb)->csum =
3722 skb_checksum(nskb, doffset,
3723 nskb->len - doffset, 0);
3724 SKB_GSO_CB(nskb)->csum_start =
3725 skb_headroom(nskb) + doffset;
3727 } while ((offset += len) < head_skb->len);
3729 /* Some callers want to get the end of the list.
3730 * Put it in segs->prev to avoid walking the list.
3731 * (see validate_xmit_skb_list() for example)
3736 struct sk_buff *iter;
3737 int type = skb_shinfo(head_skb)->gso_type;
3738 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
3740 /* Update type to add partial and then remove dodgy if set */
3741 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
3742 type &= ~SKB_GSO_DODGY;
3744 /* Update GSO info and prepare to start updating headers on
3745 * our way back down the stack of protocols.
3747 for (iter = segs; iter; iter = iter->next) {
3748 skb_shinfo(iter)->gso_size = gso_size;
3749 skb_shinfo(iter)->gso_segs = partial_segs;
3750 skb_shinfo(iter)->gso_type = type;
3751 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
3754 if (tail->len - doffset <= gso_size)
3755 skb_shinfo(tail)->gso_size = 0;
3756 else if (tail != segs)
3757 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
3760 /* Following permits correct backpressure, for protocols
3761 * using skb_set_owner_w().
3762 * Idea is to tranfert ownership from head_skb to last segment.
3764 if (head_skb->destructor == sock_wfree) {
3765 swap(tail->truesize, head_skb->truesize);
3766 swap(tail->destructor, head_skb->destructor);
3767 swap(tail->sk, head_skb->sk);
3772 kfree_skb_list(segs);
3773 return ERR_PTR(err);
3775 EXPORT_SYMBOL_GPL(skb_segment);
3777 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3779 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3780 unsigned int offset = skb_gro_offset(skb);
3781 unsigned int headlen = skb_headlen(skb);
3782 unsigned int len = skb_gro_len(skb);
3783 struct sk_buff *lp, *p = *head;
3784 unsigned int delta_truesize;
3786 if (unlikely(p->len + len >= 65536))
3789 lp = NAPI_GRO_CB(p)->last;
3790 pinfo = skb_shinfo(lp);
3792 if (headlen <= offset) {
3795 int i = skbinfo->nr_frags;
3796 int nr_frags = pinfo->nr_frags + i;
3798 if (nr_frags > MAX_SKB_FRAGS)
3802 pinfo->nr_frags = nr_frags;
3803 skbinfo->nr_frags = 0;
3805 frag = pinfo->frags + nr_frags;
3806 frag2 = skbinfo->frags + i;
3811 frag->page_offset += offset;
3812 skb_frag_size_sub(frag, offset);
3814 /* all fragments truesize : remove (head size + sk_buff) */
3815 delta_truesize = skb->truesize -
3816 SKB_TRUESIZE(skb_end_offset(skb));
3818 skb->truesize -= skb->data_len;
3819 skb->len -= skb->data_len;
3822 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3824 } else if (skb->head_frag) {
3825 int nr_frags = pinfo->nr_frags;
3826 skb_frag_t *frag = pinfo->frags + nr_frags;
3827 struct page *page = virt_to_head_page(skb->head);
3828 unsigned int first_size = headlen - offset;
3829 unsigned int first_offset;
3831 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3834 first_offset = skb->data -
3835 (unsigned char *)page_address(page) +
3838 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3840 frag->page.p = page;
3841 frag->page_offset = first_offset;
3842 skb_frag_size_set(frag, first_size);
3844 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3845 /* We dont need to clear skbinfo->nr_frags here */
3847 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3848 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3853 delta_truesize = skb->truesize;
3854 if (offset > headlen) {
3855 unsigned int eat = offset - headlen;
3857 skbinfo->frags[0].page_offset += eat;
3858 skb_frag_size_sub(&skbinfo->frags[0], eat);
3859 skb->data_len -= eat;
3864 __skb_pull(skb, offset);
3866 if (NAPI_GRO_CB(p)->last == p)
3867 skb_shinfo(p)->frag_list = skb;
3869 NAPI_GRO_CB(p)->last->next = skb;
3870 NAPI_GRO_CB(p)->last = skb;
3871 __skb_header_release(skb);
3875 NAPI_GRO_CB(p)->count++;
3877 p->truesize += delta_truesize;
3880 lp->data_len += len;
3881 lp->truesize += delta_truesize;
3884 NAPI_GRO_CB(skb)->same_flow = 1;
3887 EXPORT_SYMBOL_GPL(skb_gro_receive);
3889 void __init skb_init(void)
3891 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3892 sizeof(struct sk_buff),
3894 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3896 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3897 sizeof(struct sk_buff_fclones),
3899 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3904 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
3905 unsigned int recursion_level)
3907 int start = skb_headlen(skb);
3908 int i, copy = start - offset;
3909 struct sk_buff *frag_iter;
3912 if (unlikely(recursion_level >= 24))
3918 sg_set_buf(sg, skb->data + offset, copy);
3920 if ((len -= copy) == 0)
3925 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3928 WARN_ON(start > offset + len);
3930 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3931 if ((copy = end - offset) > 0) {
3932 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3933 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
3938 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3939 frag->page_offset+offset-start);
3948 skb_walk_frags(skb, frag_iter) {
3951 WARN_ON(start > offset + len);
3953 end = start + frag_iter->len;
3954 if ((copy = end - offset) > 0) {
3955 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
3960 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3961 copy, recursion_level + 1);
3962 if (unlikely(ret < 0))
3965 if ((len -= copy) == 0)
3976 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3977 * @skb: Socket buffer containing the buffers to be mapped
3978 * @sg: The scatter-gather list to map into
3979 * @offset: The offset into the buffer's contents to start mapping
3980 * @len: Length of buffer space to be mapped
3982 * Fill the specified scatter-gather list with mappings/pointers into a
3983 * region of the buffer space attached to a socket buffer. Returns either
3984 * the number of scatterlist items used, or -EMSGSIZE if the contents
3987 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3989 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
3994 sg_mark_end(&sg[nsg - 1]);
3998 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4000 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4001 * sglist without mark the sg which contain last skb data as the end.
4002 * So the caller can mannipulate sg list as will when padding new data after
4003 * the first call without calling sg_unmark_end to expend sg list.
4005 * Scenario to use skb_to_sgvec_nomark:
4007 * 2. skb_to_sgvec_nomark(payload1)
4008 * 3. skb_to_sgvec_nomark(payload2)
4010 * This is equivalent to:
4012 * 2. skb_to_sgvec(payload1)
4014 * 4. skb_to_sgvec(payload2)
4016 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4017 * is more preferable.
4019 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4020 int offset, int len)
4022 return __skb_to_sgvec(skb, sg, offset, len, 0);
4024 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4029 * skb_cow_data - Check that a socket buffer's data buffers are writable
4030 * @skb: The socket buffer to check.
4031 * @tailbits: Amount of trailing space to be added
4032 * @trailer: Returned pointer to the skb where the @tailbits space begins
4034 * Make sure that the data buffers attached to a socket buffer are
4035 * writable. If they are not, private copies are made of the data buffers
4036 * and the socket buffer is set to use these instead.
4038 * If @tailbits is given, make sure that there is space to write @tailbits
4039 * bytes of data beyond current end of socket buffer. @trailer will be
4040 * set to point to the skb in which this space begins.
4042 * The number of scatterlist elements required to completely map the
4043 * COW'd and extended socket buffer will be returned.
4045 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4049 struct sk_buff *skb1, **skb_p;
4051 /* If skb is cloned or its head is paged, reallocate
4052 * head pulling out all the pages (pages are considered not writable
4053 * at the moment even if they are anonymous).
4055 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4056 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
4059 /* Easy case. Most of packets will go this way. */
4060 if (!skb_has_frag_list(skb)) {
4061 /* A little of trouble, not enough of space for trailer.
4062 * This should not happen, when stack is tuned to generate
4063 * good frames. OK, on miss we reallocate and reserve even more
4064 * space, 128 bytes is fair. */
4066 if (skb_tailroom(skb) < tailbits &&
4067 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4075 /* Misery. We are in troubles, going to mincer fragments... */
4078 skb_p = &skb_shinfo(skb)->frag_list;
4081 while ((skb1 = *skb_p) != NULL) {
4084 /* The fragment is partially pulled by someone,
4085 * this can happen on input. Copy it and everything
4088 if (skb_shared(skb1))
4091 /* If the skb is the last, worry about trailer. */
4093 if (skb1->next == NULL && tailbits) {
4094 if (skb_shinfo(skb1)->nr_frags ||
4095 skb_has_frag_list(skb1) ||
4096 skb_tailroom(skb1) < tailbits)
4097 ntail = tailbits + 128;
4103 skb_shinfo(skb1)->nr_frags ||
4104 skb_has_frag_list(skb1)) {
4105 struct sk_buff *skb2;
4107 /* Fuck, we are miserable poor guys... */
4109 skb2 = skb_copy(skb1, GFP_ATOMIC);
4111 skb2 = skb_copy_expand(skb1,
4115 if (unlikely(skb2 == NULL))
4119 skb_set_owner_w(skb2, skb1->sk);
4121 /* Looking around. Are we still alive?
4122 * OK, link new skb, drop old one */
4124 skb2->next = skb1->next;
4131 skb_p = &skb1->next;
4136 EXPORT_SYMBOL_GPL(skb_cow_data);
4138 static void sock_rmem_free(struct sk_buff *skb)
4140 struct sock *sk = skb->sk;
4142 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4145 static void skb_set_err_queue(struct sk_buff *skb)
4147 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4148 * So, it is safe to (mis)use it to mark skbs on the error queue.
4150 skb->pkt_type = PACKET_OUTGOING;
4151 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4155 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4157 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4159 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4160 (unsigned int)sk->sk_rcvbuf)
4165 skb->destructor = sock_rmem_free;
4166 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4167 skb_set_err_queue(skb);
4169 /* before exiting rcu section, make sure dst is refcounted */
4172 skb_queue_tail(&sk->sk_error_queue, skb);
4173 if (!sock_flag(sk, SOCK_DEAD))
4174 sk->sk_data_ready(sk);
4177 EXPORT_SYMBOL(sock_queue_err_skb);
4179 static bool is_icmp_err_skb(const struct sk_buff *skb)
4181 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4182 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4185 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4187 struct sk_buff_head *q = &sk->sk_error_queue;
4188 struct sk_buff *skb, *skb_next = NULL;
4189 bool icmp_next = false;
4190 unsigned long flags;
4192 spin_lock_irqsave(&q->lock, flags);
4193 skb = __skb_dequeue(q);
4194 if (skb && (skb_next = skb_peek(q))) {
4195 icmp_next = is_icmp_err_skb(skb_next);
4197 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin;
4199 spin_unlock_irqrestore(&q->lock, flags);
4201 if (is_icmp_err_skb(skb) && !icmp_next)
4205 sk->sk_error_report(sk);
4209 EXPORT_SYMBOL(sock_dequeue_err_skb);
4212 * skb_clone_sk - create clone of skb, and take reference to socket
4213 * @skb: the skb to clone
4215 * This function creates a clone of a buffer that holds a reference on
4216 * sk_refcnt. Buffers created via this function are meant to be
4217 * returned using sock_queue_err_skb, or free via kfree_skb.
4219 * When passing buffers allocated with this function to sock_queue_err_skb
4220 * it is necessary to wrap the call with sock_hold/sock_put in order to
4221 * prevent the socket from being released prior to being enqueued on
4222 * the sk_error_queue.
4224 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4226 struct sock *sk = skb->sk;
4227 struct sk_buff *clone;
4229 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4232 clone = skb_clone(skb, GFP_ATOMIC);
4239 clone->destructor = sock_efree;
4243 EXPORT_SYMBOL(skb_clone_sk);
4245 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4250 struct sock_exterr_skb *serr;
4253 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4255 serr = SKB_EXT_ERR(skb);
4256 memset(serr, 0, sizeof(*serr));
4257 serr->ee.ee_errno = ENOMSG;
4258 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4259 serr->ee.ee_info = tstype;
4260 serr->opt_stats = opt_stats;
4261 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4262 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4263 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4264 if (sk->sk_protocol == IPPROTO_TCP &&
4265 sk->sk_type == SOCK_STREAM)
4266 serr->ee.ee_data -= sk->sk_tskey;
4269 err = sock_queue_err_skb(sk, skb);
4275 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4279 if (likely(sysctl_tstamp_allow_data || tsonly))
4282 read_lock_bh(&sk->sk_callback_lock);
4283 ret = sk->sk_socket && sk->sk_socket->file &&
4284 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4285 read_unlock_bh(&sk->sk_callback_lock);
4289 void skb_complete_tx_timestamp(struct sk_buff *skb,
4290 struct skb_shared_hwtstamps *hwtstamps)
4292 struct sock *sk = skb->sk;
4294 if (!skb_may_tx_timestamp(sk, false))
4297 /* Take a reference to prevent skb_orphan() from freeing the socket,
4298 * but only if the socket refcount is not zero.
4300 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4301 *skb_hwtstamps(skb) = *hwtstamps;
4302 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4306 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4308 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4309 struct skb_shared_hwtstamps *hwtstamps,
4310 struct sock *sk, int tstype)
4312 struct sk_buff *skb;
4313 bool tsonly, opt_stats = false;
4318 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4319 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4322 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4323 if (!skb_may_tx_timestamp(sk, tsonly))
4328 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4329 sk->sk_protocol == IPPROTO_TCP &&
4330 sk->sk_type == SOCK_STREAM) {
4331 skb = tcp_get_timestamping_opt_stats(sk);
4335 skb = alloc_skb(0, GFP_ATOMIC);
4337 skb = skb_clone(orig_skb, GFP_ATOMIC);
4343 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4345 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4349 *skb_hwtstamps(skb) = *hwtstamps;
4351 skb->tstamp = ktime_get_real();
4353 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4355 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4357 void skb_tstamp_tx(struct sk_buff *orig_skb,
4358 struct skb_shared_hwtstamps *hwtstamps)
4360 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4363 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4365 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4367 struct sock *sk = skb->sk;
4368 struct sock_exterr_skb *serr;
4371 skb->wifi_acked_valid = 1;
4372 skb->wifi_acked = acked;
4374 serr = SKB_EXT_ERR(skb);
4375 memset(serr, 0, sizeof(*serr));
4376 serr->ee.ee_errno = ENOMSG;
4377 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4379 /* Take a reference to prevent skb_orphan() from freeing the socket,
4380 * but only if the socket refcount is not zero.
4382 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4383 err = sock_queue_err_skb(sk, skb);
4389 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4392 * skb_partial_csum_set - set up and verify partial csum values for packet
4393 * @skb: the skb to set
4394 * @start: the number of bytes after skb->data to start checksumming.
4395 * @off: the offset from start to place the checksum.
4397 * For untrusted partially-checksummed packets, we need to make sure the values
4398 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4400 * This function checks and sets those values and skb->ip_summed: if this
4401 * returns false you should drop the packet.
4403 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4405 if (unlikely(start > skb_headlen(skb)) ||
4406 unlikely((int)start + off > skb_headlen(skb) - 2)) {
4407 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
4408 start, off, skb_headlen(skb));
4411 skb->ip_summed = CHECKSUM_PARTIAL;
4412 skb->csum_start = skb_headroom(skb) + start;
4413 skb->csum_offset = off;
4414 skb_set_transport_header(skb, start);
4417 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4419 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4422 if (skb_headlen(skb) >= len)
4425 /* If we need to pullup then pullup to the max, so we
4426 * won't need to do it again.
4431 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4434 if (skb_headlen(skb) < len)
4440 #define MAX_TCP_HDR_LEN (15 * 4)
4442 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4443 typeof(IPPROTO_IP) proto,
4450 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4451 off + MAX_TCP_HDR_LEN);
4452 if (!err && !skb_partial_csum_set(skb, off,
4453 offsetof(struct tcphdr,
4456 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4459 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4460 off + sizeof(struct udphdr));
4461 if (!err && !skb_partial_csum_set(skb, off,
4462 offsetof(struct udphdr,
4465 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4468 return ERR_PTR(-EPROTO);
4471 /* This value should be large enough to cover a tagged ethernet header plus
4472 * maximally sized IP and TCP or UDP headers.
4474 #define MAX_IP_HDR_LEN 128
4476 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4485 err = skb_maybe_pull_tail(skb,
4486 sizeof(struct iphdr),
4491 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4494 off = ip_hdrlen(skb);
4501 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4503 return PTR_ERR(csum);
4506 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4509 ip_hdr(skb)->protocol, 0);
4516 /* This value should be large enough to cover a tagged ethernet header plus
4517 * an IPv6 header, all options, and a maximal TCP or UDP header.
4519 #define MAX_IPV6_HDR_LEN 256
4521 #define OPT_HDR(type, skb, off) \
4522 (type *)(skb_network_header(skb) + (off))
4524 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4537 off = sizeof(struct ipv6hdr);
4539 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4543 nexthdr = ipv6_hdr(skb)->nexthdr;
4545 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4546 while (off <= len && !done) {
4548 case IPPROTO_DSTOPTS:
4549 case IPPROTO_HOPOPTS:
4550 case IPPROTO_ROUTING: {
4551 struct ipv6_opt_hdr *hp;
4553 err = skb_maybe_pull_tail(skb,
4555 sizeof(struct ipv6_opt_hdr),
4560 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4561 nexthdr = hp->nexthdr;
4562 off += ipv6_optlen(hp);
4566 struct ip_auth_hdr *hp;
4568 err = skb_maybe_pull_tail(skb,
4570 sizeof(struct ip_auth_hdr),
4575 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4576 nexthdr = hp->nexthdr;
4577 off += ipv6_authlen(hp);
4580 case IPPROTO_FRAGMENT: {
4581 struct frag_hdr *hp;
4583 err = skb_maybe_pull_tail(skb,
4585 sizeof(struct frag_hdr),
4590 hp = OPT_HDR(struct frag_hdr, skb, off);
4592 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4595 nexthdr = hp->nexthdr;
4596 off += sizeof(struct frag_hdr);
4607 if (!done || fragment)
4610 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4612 return PTR_ERR(csum);
4615 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4616 &ipv6_hdr(skb)->daddr,
4617 skb->len - off, nexthdr, 0);
4625 * skb_checksum_setup - set up partial checksum offset
4626 * @skb: the skb to set up
4627 * @recalculate: if true the pseudo-header checksum will be recalculated
4629 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4633 switch (skb->protocol) {
4634 case htons(ETH_P_IP):
4635 err = skb_checksum_setup_ipv4(skb, recalculate);
4638 case htons(ETH_P_IPV6):
4639 err = skb_checksum_setup_ipv6(skb, recalculate);
4649 EXPORT_SYMBOL(skb_checksum_setup);
4652 * skb_checksum_maybe_trim - maybe trims the given skb
4653 * @skb: the skb to check
4654 * @transport_len: the data length beyond the network header
4656 * Checks whether the given skb has data beyond the given transport length.
4657 * If so, returns a cloned skb trimmed to this transport length.
4658 * Otherwise returns the provided skb. Returns NULL in error cases
4659 * (e.g. transport_len exceeds skb length or out-of-memory).
4661 * Caller needs to set the skb transport header and free any returned skb if it
4662 * differs from the provided skb.
4664 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4665 unsigned int transport_len)
4667 struct sk_buff *skb_chk;
4668 unsigned int len = skb_transport_offset(skb) + transport_len;
4673 else if (skb->len == len)
4676 skb_chk = skb_clone(skb, GFP_ATOMIC);
4680 ret = pskb_trim_rcsum(skb_chk, len);
4690 * skb_checksum_trimmed - validate checksum of an skb
4691 * @skb: the skb to check
4692 * @transport_len: the data length beyond the network header
4693 * @skb_chkf: checksum function to use
4695 * Applies the given checksum function skb_chkf to the provided skb.
4696 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4698 * If the skb has data beyond the given transport length, then a
4699 * trimmed & cloned skb is checked and returned.
4701 * Caller needs to set the skb transport header and free any returned skb if it
4702 * differs from the provided skb.
4704 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4705 unsigned int transport_len,
4706 __sum16(*skb_chkf)(struct sk_buff *skb))
4708 struct sk_buff *skb_chk;
4709 unsigned int offset = skb_transport_offset(skb);
4712 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4716 if (!pskb_may_pull(skb_chk, offset))
4719 skb_pull_rcsum(skb_chk, offset);
4720 ret = skb_chkf(skb_chk);
4721 skb_push_rcsum(skb_chk, offset);
4729 if (skb_chk && skb_chk != skb)
4735 EXPORT_SYMBOL(skb_checksum_trimmed);
4737 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4739 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4742 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4744 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4747 skb_release_head_state(skb);
4748 kmem_cache_free(skbuff_head_cache, skb);
4753 EXPORT_SYMBOL(kfree_skb_partial);
4756 * skb_try_coalesce - try to merge skb to prior one
4758 * @from: buffer to add
4759 * @fragstolen: pointer to boolean
4760 * @delta_truesize: how much more was allocated than was requested
4762 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4763 bool *fragstolen, int *delta_truesize)
4765 int i, delta, len = from->len;
4767 *fragstolen = false;
4772 if (len <= skb_tailroom(to)) {
4774 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4775 *delta_truesize = 0;
4779 if (skb_has_frag_list(to) || skb_has_frag_list(from))
4781 if (skb_zcopy(to) || skb_zcopy(from))
4784 if (skb_headlen(from) != 0) {
4786 unsigned int offset;
4788 if (skb_shinfo(to)->nr_frags +
4789 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4792 if (skb_head_is_locked(from))
4795 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4797 page = virt_to_head_page(from->head);
4798 offset = from->data - (unsigned char *)page_address(page);
4800 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4801 page, offset, skb_headlen(from));
4804 if (skb_shinfo(to)->nr_frags +
4805 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4808 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4811 WARN_ON_ONCE(delta < len);
4813 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4814 skb_shinfo(from)->frags,
4815 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4816 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4818 if (!skb_cloned(from))
4819 skb_shinfo(from)->nr_frags = 0;
4821 /* if the skb is not cloned this does nothing
4822 * since we set nr_frags to 0.
4824 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4825 skb_frag_ref(from, i);
4827 to->truesize += delta;
4829 to->data_len += len;
4831 *delta_truesize = delta;
4834 EXPORT_SYMBOL(skb_try_coalesce);
4837 * skb_scrub_packet - scrub an skb
4839 * @skb: buffer to clean
4840 * @xnet: packet is crossing netns
4842 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4843 * into/from a tunnel. Some information have to be cleared during these
4845 * skb_scrub_packet can also be used to clean a skb before injecting it in
4846 * another namespace (@xnet == true). We have to clear all information in the
4847 * skb that could impact namespace isolation.
4849 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4852 skb->pkt_type = PACKET_HOST;
4858 nf_reset_trace(skb);
4866 EXPORT_SYMBOL_GPL(skb_scrub_packet);
4869 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4873 * skb_gso_transport_seglen is used to determine the real size of the
4874 * individual segments, including Layer4 headers (TCP/UDP).
4876 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4878 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4880 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4881 unsigned int thlen = 0;
4883 if (skb->encapsulation) {
4884 thlen = skb_inner_transport_header(skb) -
4885 skb_transport_header(skb);
4887 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4888 thlen += inner_tcp_hdrlen(skb);
4889 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4890 thlen = tcp_hdrlen(skb);
4891 } else if (unlikely(shinfo->gso_type & SKB_GSO_SCTP)) {
4892 thlen = sizeof(struct sctphdr);
4894 /* UFO sets gso_size to the size of the fragmentation
4895 * payload, i.e. the size of the L4 (UDP) header is already
4898 return thlen + shinfo->gso_size;
4900 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4903 * skb_gso_validate_mtu - Return in case such skb fits a given MTU
4906 * @mtu: MTU to validate against
4908 * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU
4911 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu)
4913 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4914 const struct sk_buff *iter;
4917 hlen = skb_gso_network_seglen(skb);
4919 if (shinfo->gso_size != GSO_BY_FRAGS)
4922 /* Undo this so we can re-use header sizes */
4923 hlen -= GSO_BY_FRAGS;
4925 skb_walk_frags(skb, iter) {
4926 if (hlen + skb_headlen(iter) > mtu)
4932 EXPORT_SYMBOL_GPL(skb_gso_validate_mtu);
4934 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4936 if (skb_cow(skb, skb_headroom(skb)) < 0) {
4941 memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN,
4943 skb->mac_header += VLAN_HLEN;
4947 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4949 struct vlan_hdr *vhdr;
4952 if (unlikely(skb_vlan_tag_present(skb))) {
4953 /* vlan_tci is already set-up so leave this for another time */
4957 skb = skb_share_check(skb, GFP_ATOMIC);
4961 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4964 vhdr = (struct vlan_hdr *)skb->data;
4965 vlan_tci = ntohs(vhdr->h_vlan_TCI);
4966 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4968 skb_pull_rcsum(skb, VLAN_HLEN);
4969 vlan_set_encap_proto(skb, vhdr);
4971 skb = skb_reorder_vlan_header(skb);
4975 skb_reset_network_header(skb);
4976 skb_reset_transport_header(skb);
4977 skb_reset_mac_len(skb);
4985 EXPORT_SYMBOL(skb_vlan_untag);
4987 int skb_ensure_writable(struct sk_buff *skb, int write_len)
4989 if (!pskb_may_pull(skb, write_len))
4992 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4995 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4997 EXPORT_SYMBOL(skb_ensure_writable);
4999 /* remove VLAN header from packet and update csum accordingly.
5000 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5002 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5004 struct vlan_hdr *vhdr;
5005 int offset = skb->data - skb_mac_header(skb);
5008 if (WARN_ONCE(offset,
5009 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5014 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5018 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5020 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5021 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5023 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5024 __skb_pull(skb, VLAN_HLEN);
5026 vlan_set_encap_proto(skb, vhdr);
5027 skb->mac_header += VLAN_HLEN;
5029 if (skb_network_offset(skb) < ETH_HLEN)
5030 skb_set_network_header(skb, ETH_HLEN);
5032 skb_reset_mac_len(skb);
5036 EXPORT_SYMBOL(__skb_vlan_pop);
5038 /* Pop a vlan tag either from hwaccel or from payload.
5039 * Expects skb->data at mac header.
5041 int skb_vlan_pop(struct sk_buff *skb)
5047 if (likely(skb_vlan_tag_present(skb))) {
5050 if (unlikely(!eth_type_vlan(skb->protocol)))
5053 err = __skb_vlan_pop(skb, &vlan_tci);
5057 /* move next vlan tag to hw accel tag */
5058 if (likely(!eth_type_vlan(skb->protocol)))
5061 vlan_proto = skb->protocol;
5062 err = __skb_vlan_pop(skb, &vlan_tci);
5066 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5069 EXPORT_SYMBOL(skb_vlan_pop);
5071 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5072 * Expects skb->data at mac header.
5074 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5076 if (skb_vlan_tag_present(skb)) {
5077 int offset = skb->data - skb_mac_header(skb);
5080 if (WARN_ONCE(offset,
5081 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5086 err = __vlan_insert_tag(skb, skb->vlan_proto,
5087 skb_vlan_tag_get(skb));
5091 skb->protocol = skb->vlan_proto;
5092 skb->mac_len += VLAN_HLEN;
5094 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5096 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5099 EXPORT_SYMBOL(skb_vlan_push);
5102 * alloc_skb_with_frags - allocate skb with page frags
5104 * @header_len: size of linear part
5105 * @data_len: needed length in frags
5106 * @max_page_order: max page order desired.
5107 * @errcode: pointer to error code if any
5108 * @gfp_mask: allocation mask
5110 * This can be used to allocate a paged skb, given a maximal order for frags.
5112 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5113 unsigned long data_len,
5118 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5119 unsigned long chunk;
5120 struct sk_buff *skb;
5125 *errcode = -EMSGSIZE;
5126 /* Note this test could be relaxed, if we succeed to allocate
5127 * high order pages...
5129 if (npages > MAX_SKB_FRAGS)
5132 gfp_head = gfp_mask;
5133 if (gfp_head & __GFP_DIRECT_RECLAIM)
5134 gfp_head |= __GFP_RETRY_MAYFAIL;
5136 *errcode = -ENOBUFS;
5137 skb = alloc_skb(header_len, gfp_head);
5141 skb->truesize += npages << PAGE_SHIFT;
5143 for (i = 0; npages > 0; i++) {
5144 int order = max_page_order;
5147 if (npages >= 1 << order) {
5148 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5155 /* Do not retry other high order allocations */
5161 page = alloc_page(gfp_mask);
5165 chunk = min_t(unsigned long, data_len,
5166 PAGE_SIZE << order);
5167 skb_fill_page_desc(skb, i, page, 0, chunk);
5169 npages -= 1 << order;
5177 EXPORT_SYMBOL(alloc_skb_with_frags);
5179 /* carve out the first off bytes from skb when off < headlen */
5180 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5181 const int headlen, gfp_t gfp_mask)
5184 int size = skb_end_offset(skb);
5185 int new_hlen = headlen - off;
5188 size = SKB_DATA_ALIGN(size);
5190 if (skb_pfmemalloc(skb))
5191 gfp_mask |= __GFP_MEMALLOC;
5192 data = kmalloc_reserve(size +
5193 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5194 gfp_mask, NUMA_NO_NODE, NULL);
5198 size = SKB_WITH_OVERHEAD(ksize(data));
5200 /* Copy real data, and all frags */
5201 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5204 memcpy((struct skb_shared_info *)(data + size),
5206 offsetof(struct skb_shared_info,
5207 frags[skb_shinfo(skb)->nr_frags]));
5208 if (skb_cloned(skb)) {
5209 /* drop the old head gracefully */
5210 if (skb_orphan_frags(skb, gfp_mask)) {
5214 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5215 skb_frag_ref(skb, i);
5216 if (skb_has_frag_list(skb))
5217 skb_clone_fraglist(skb);
5218 skb_release_data(skb);
5220 /* we can reuse existing recount- all we did was
5229 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5232 skb->end = skb->head + size;
5234 skb_set_tail_pointer(skb, skb_headlen(skb));
5235 skb_headers_offset_update(skb, 0);
5239 atomic_set(&skb_shinfo(skb)->dataref, 1);
5244 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
5246 /* carve out the first eat bytes from skb's frag_list. May recurse into
5249 static int pskb_carve_frag_list(struct sk_buff *skb,
5250 struct skb_shared_info *shinfo, int eat,
5253 struct sk_buff *list = shinfo->frag_list;
5254 struct sk_buff *clone = NULL;
5255 struct sk_buff *insp = NULL;
5259 pr_err("Not enough bytes to eat. Want %d\n", eat);
5262 if (list->len <= eat) {
5263 /* Eaten as whole. */
5268 /* Eaten partially. */
5269 if (skb_shared(list)) {
5270 clone = skb_clone(list, gfp_mask);
5276 /* This may be pulled without problems. */
5279 if (pskb_carve(list, eat, gfp_mask) < 0) {
5287 /* Free pulled out fragments. */
5288 while ((list = shinfo->frag_list) != insp) {
5289 shinfo->frag_list = list->next;
5292 /* And insert new clone at head. */
5295 shinfo->frag_list = clone;
5300 /* carve off first len bytes from skb. Split line (off) is in the
5301 * non-linear part of skb
5303 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
5304 int pos, gfp_t gfp_mask)
5307 int size = skb_end_offset(skb);
5309 const int nfrags = skb_shinfo(skb)->nr_frags;
5310 struct skb_shared_info *shinfo;
5312 size = SKB_DATA_ALIGN(size);
5314 if (skb_pfmemalloc(skb))
5315 gfp_mask |= __GFP_MEMALLOC;
5316 data = kmalloc_reserve(size +
5317 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5318 gfp_mask, NUMA_NO_NODE, NULL);
5322 size = SKB_WITH_OVERHEAD(ksize(data));
5324 memcpy((struct skb_shared_info *)(data + size),
5325 skb_shinfo(skb), offsetof(struct skb_shared_info,
5326 frags[skb_shinfo(skb)->nr_frags]));
5327 if (skb_orphan_frags(skb, gfp_mask)) {
5331 shinfo = (struct skb_shared_info *)(data + size);
5332 for (i = 0; i < nfrags; i++) {
5333 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
5335 if (pos + fsize > off) {
5336 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
5340 * We have two variants in this case:
5341 * 1. Move all the frag to the second
5342 * part, if it is possible. F.e.
5343 * this approach is mandatory for TUX,
5344 * where splitting is expensive.
5345 * 2. Split is accurately. We make this.
5347 shinfo->frags[0].page_offset += off - pos;
5348 skb_frag_size_sub(&shinfo->frags[0], off - pos);
5350 skb_frag_ref(skb, i);
5355 shinfo->nr_frags = k;
5356 if (skb_has_frag_list(skb))
5357 skb_clone_fraglist(skb);
5360 /* split line is in frag list */
5361 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
5363 skb_release_data(skb);
5368 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5371 skb->end = skb->head + size;
5373 skb_reset_tail_pointer(skb);
5374 skb_headers_offset_update(skb, 0);
5379 skb->data_len = skb->len;
5380 atomic_set(&skb_shinfo(skb)->dataref, 1);
5384 /* remove len bytes from the beginning of the skb */
5385 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
5387 int headlen = skb_headlen(skb);
5390 return pskb_carve_inside_header(skb, len, headlen, gfp);
5392 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
5395 /* Extract to_copy bytes starting at off from skb, and return this in
5398 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
5399 int to_copy, gfp_t gfp)
5401 struct sk_buff *clone = skb_clone(skb, gfp);
5406 if (pskb_carve(clone, off, gfp) < 0 ||
5407 pskb_trim(clone, to_copy)) {
5413 EXPORT_SYMBOL(pskb_extract);
5416 * skb_condense - try to get rid of fragments/frag_list if possible
5419 * Can be used to save memory before skb is added to a busy queue.
5420 * If packet has bytes in frags and enough tail room in skb->head,
5421 * pull all of them, so that we can free the frags right now and adjust
5424 * We do not reallocate skb->head thus can not fail.
5425 * Caller must re-evaluate skb->truesize if needed.
5427 void skb_condense(struct sk_buff *skb)
5429 if (skb->data_len) {
5430 if (skb->data_len > skb->end - skb->tail ||
5434 /* Nice, we can free page frag(s) right now */
5435 __pskb_pull_tail(skb, skb->data_len);
5437 /* At this point, skb->truesize might be over estimated,
5438 * because skb had a fragment, and fragments do not tell
5440 * When we pulled its content into skb->head, fragment
5441 * was freed, but __pskb_pull_tail() could not possibly
5442 * adjust skb->truesize, not knowing the frag truesize.
5444 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));