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 * Works like consume_skb(), but this variant assumes that all the head
714 * states have been already dropped.
716 void consume_stateless_skb(struct sk_buff *skb)
721 trace_consume_skb(skb);
722 skb_release_data(skb);
726 void __kfree_skb_flush(void)
728 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
730 /* flush skb_cache if containing objects */
732 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
738 static inline void _kfree_skb_defer(struct sk_buff *skb)
740 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
742 /* drop skb->head and call any destructors for packet */
743 skb_release_all(skb);
745 /* record skb to CPU local list */
746 nc->skb_cache[nc->skb_count++] = skb;
749 /* SLUB writes into objects when freeing */
753 /* flush skb_cache if it is filled */
754 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
755 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
760 void __kfree_skb_defer(struct sk_buff *skb)
762 _kfree_skb_defer(skb);
765 void napi_consume_skb(struct sk_buff *skb, int budget)
770 /* Zero budget indicate non-NAPI context called us, like netpoll */
771 if (unlikely(!budget)) {
772 dev_consume_skb_any(skb);
779 /* if reaching here SKB is ready to free */
780 trace_consume_skb(skb);
782 /* if SKB is a clone, don't handle this case */
783 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
788 _kfree_skb_defer(skb);
790 EXPORT_SYMBOL(napi_consume_skb);
792 /* Make sure a field is enclosed inside headers_start/headers_end section */
793 #define CHECK_SKB_FIELD(field) \
794 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
795 offsetof(struct sk_buff, headers_start)); \
796 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
797 offsetof(struct sk_buff, headers_end)); \
799 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
801 new->tstamp = old->tstamp;
802 /* We do not copy old->sk */
804 memcpy(new->cb, old->cb, sizeof(old->cb));
805 skb_dst_copy(new, old);
807 new->sp = secpath_get(old->sp);
809 __nf_copy(new, old, false);
811 /* Note : this field could be in headers_start/headers_end section
812 * It is not yet because we do not want to have a 16 bit hole
814 new->queue_mapping = old->queue_mapping;
816 memcpy(&new->headers_start, &old->headers_start,
817 offsetof(struct sk_buff, headers_end) -
818 offsetof(struct sk_buff, headers_start));
819 CHECK_SKB_FIELD(protocol);
820 CHECK_SKB_FIELD(csum);
821 CHECK_SKB_FIELD(hash);
822 CHECK_SKB_FIELD(priority);
823 CHECK_SKB_FIELD(skb_iif);
824 CHECK_SKB_FIELD(vlan_proto);
825 CHECK_SKB_FIELD(vlan_tci);
826 CHECK_SKB_FIELD(transport_header);
827 CHECK_SKB_FIELD(network_header);
828 CHECK_SKB_FIELD(mac_header);
829 CHECK_SKB_FIELD(inner_protocol);
830 CHECK_SKB_FIELD(inner_transport_header);
831 CHECK_SKB_FIELD(inner_network_header);
832 CHECK_SKB_FIELD(inner_mac_header);
833 CHECK_SKB_FIELD(mark);
834 #ifdef CONFIG_NETWORK_SECMARK
835 CHECK_SKB_FIELD(secmark);
837 #ifdef CONFIG_NET_RX_BUSY_POLL
838 CHECK_SKB_FIELD(napi_id);
841 CHECK_SKB_FIELD(sender_cpu);
843 #ifdef CONFIG_NET_SCHED
844 CHECK_SKB_FIELD(tc_index);
850 * You should not add any new code to this function. Add it to
851 * __copy_skb_header above instead.
853 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
855 #define C(x) n->x = skb->x
857 n->next = n->prev = NULL;
859 __copy_skb_header(n, skb);
864 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
867 n->destructor = NULL;
874 refcount_set(&n->users, 1);
876 atomic_inc(&(skb_shinfo(skb)->dataref));
884 * skb_morph - morph one skb into another
885 * @dst: the skb to receive the contents
886 * @src: the skb to supply the contents
888 * This is identical to skb_clone except that the target skb is
889 * supplied by the user.
891 * The target skb is returned upon exit.
893 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
895 skb_release_all(dst);
896 return __skb_clone(dst, src);
898 EXPORT_SYMBOL_GPL(skb_morph);
900 static int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
902 unsigned long max_pg, num_pg, new_pg, old_pg;
903 struct user_struct *user;
905 if (capable(CAP_IPC_LOCK) || !size)
908 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
909 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
910 user = mmp->user ? : current_user();
913 old_pg = atomic_long_read(&user->locked_vm);
914 new_pg = old_pg + num_pg;
917 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
921 mmp->user = get_uid(user);
922 mmp->num_pg = num_pg;
924 mmp->num_pg += num_pg;
930 static void mm_unaccount_pinned_pages(struct mmpin *mmp)
933 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
938 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
940 struct ubuf_info *uarg;
943 WARN_ON_ONCE(!in_task());
945 if (!sock_flag(sk, SOCK_ZEROCOPY))
948 skb = sock_omalloc(sk, 0, GFP_KERNEL);
952 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
953 uarg = (void *)skb->cb;
954 uarg->mmp.user = NULL;
956 if (mm_account_pinned_pages(&uarg->mmp, size)) {
961 uarg->callback = sock_zerocopy_callback;
962 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
964 uarg->bytelen = size;
966 atomic_set(&uarg->refcnt, 0);
971 EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
973 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
975 return container_of((void *)uarg, struct sk_buff, cb);
978 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
979 struct ubuf_info *uarg)
982 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
985 /* realloc only when socket is locked (TCP, UDP cork),
986 * so uarg->len and sk_zckey access is serialized
988 if (!sock_owned_by_user(sk)) {
993 bytelen = uarg->bytelen + size;
994 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
995 /* TCP can create new skb to attach new uarg */
996 if (sk->sk_type == SOCK_STREAM)
1001 next = (u32)atomic_read(&sk->sk_zckey);
1002 if ((u32)(uarg->id + uarg->len) == next) {
1003 if (mm_account_pinned_pages(&uarg->mmp, size))
1006 uarg->bytelen = bytelen;
1007 atomic_set(&sk->sk_zckey, ++next);
1013 return sock_zerocopy_alloc(sk, size);
1015 EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1017 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1019 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1023 old_lo = serr->ee.ee_info;
1024 old_hi = serr->ee.ee_data;
1025 sum_len = old_hi - old_lo + 1ULL + len;
1027 if (sum_len >= (1ULL << 32))
1030 if (lo != old_hi + 1)
1033 serr->ee.ee_data += len;
1037 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1039 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1040 struct sock_exterr_skb *serr;
1041 struct sock *sk = skb->sk;
1042 struct sk_buff_head *q;
1043 unsigned long flags;
1047 mm_unaccount_pinned_pages(&uarg->mmp);
1049 /* if !len, there was only 1 call, and it was aborted
1050 * so do not queue a completion notification
1052 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1057 hi = uarg->id + len - 1;
1059 serr = SKB_EXT_ERR(skb);
1060 memset(serr, 0, sizeof(*serr));
1061 serr->ee.ee_errno = 0;
1062 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1063 serr->ee.ee_data = hi;
1064 serr->ee.ee_info = lo;
1066 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1068 q = &sk->sk_error_queue;
1069 spin_lock_irqsave(&q->lock, flags);
1070 tail = skb_peek_tail(q);
1071 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1072 !skb_zerocopy_notify_extend(tail, lo, len)) {
1073 __skb_queue_tail(q, skb);
1076 spin_unlock_irqrestore(&q->lock, flags);
1078 sk->sk_error_report(sk);
1084 EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1086 void sock_zerocopy_put(struct ubuf_info *uarg)
1088 if (uarg && atomic_dec_and_test(&uarg->refcnt)) {
1090 uarg->callback(uarg, uarg->zerocopy);
1092 consume_skb(skb_from_uarg(uarg));
1095 EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1097 void sock_zerocopy_put_abort(struct ubuf_info *uarg)
1100 struct sock *sk = skb_from_uarg(uarg)->sk;
1102 atomic_dec(&sk->sk_zckey);
1105 /* sock_zerocopy_put expects a ref. Most sockets take one per
1106 * skb, which is zero on abort. tcp_sendmsg holds one extra, to
1107 * avoid an skb send inside the main loop triggering uarg free.
1109 if (sk->sk_type != SOCK_STREAM)
1110 atomic_inc(&uarg->refcnt);
1112 sock_zerocopy_put(uarg);
1115 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1117 extern int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb,
1118 struct iov_iter *from, size_t length);
1120 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1121 struct msghdr *msg, int len,
1122 struct ubuf_info *uarg)
1124 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1125 struct iov_iter orig_iter = msg->msg_iter;
1126 int err, orig_len = skb->len;
1128 /* An skb can only point to one uarg. This edge case happens when
1129 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1131 if (orig_uarg && uarg != orig_uarg)
1134 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1135 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1136 /* Streams do not free skb on error. Reset to prev state. */
1137 msg->msg_iter = orig_iter;
1138 ___pskb_trim(skb, orig_len);
1142 skb_zcopy_set(skb, uarg);
1143 return skb->len - orig_len;
1145 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1147 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1150 if (skb_zcopy(orig)) {
1151 if (skb_zcopy(nskb)) {
1152 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1157 if (skb_uarg(nskb) == skb_uarg(orig))
1159 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1162 skb_zcopy_set(nskb, skb_uarg(orig));
1168 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1169 * @skb: the skb to modify
1170 * @gfp_mask: allocation priority
1172 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1173 * It will copy all frags into kernel and drop the reference
1174 * to userspace pages.
1176 * If this function is called from an interrupt gfp_mask() must be
1179 * Returns 0 on success or a negative error code on failure
1180 * to allocate kernel memory to copy to.
1182 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1184 int num_frags = skb_shinfo(skb)->nr_frags;
1185 struct page *page, *head = NULL;
1192 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1195 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1196 for (i = 0; i < new_frags; i++) {
1197 page = alloc_page(gfp_mask);
1200 struct page *next = (struct page *)page_private(head);
1206 set_page_private(page, (unsigned long)head);
1212 for (i = 0; i < num_frags; i++) {
1213 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1214 u32 p_off, p_len, copied;
1218 skb_frag_foreach_page(f, f->page_offset, skb_frag_size(f),
1219 p, p_off, p_len, copied) {
1221 vaddr = kmap_atomic(p);
1223 while (done < p_len) {
1224 if (d_off == PAGE_SIZE) {
1226 page = (struct page *)page_private(page);
1228 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1229 memcpy(page_address(page) + d_off,
1230 vaddr + p_off + done, copy);
1234 kunmap_atomic(vaddr);
1238 /* skb frags release userspace buffers */
1239 for (i = 0; i < num_frags; i++)
1240 skb_frag_unref(skb, i);
1242 /* skb frags point to kernel buffers */
1243 for (i = 0; i < new_frags - 1; i++) {
1244 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1245 head = (struct page *)page_private(head);
1247 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1248 skb_shinfo(skb)->nr_frags = new_frags;
1250 skb_zcopy_clear(skb, false);
1253 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1256 * skb_clone - duplicate an sk_buff
1257 * @skb: buffer to clone
1258 * @gfp_mask: allocation priority
1260 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1261 * copies share the same packet data but not structure. The new
1262 * buffer has a reference count of 1. If the allocation fails the
1263 * function returns %NULL otherwise the new buffer is returned.
1265 * If this function is called from an interrupt gfp_mask() must be
1269 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1271 struct sk_buff_fclones *fclones = container_of(skb,
1272 struct sk_buff_fclones,
1276 if (skb_orphan_frags(skb, gfp_mask))
1279 if (skb->fclone == SKB_FCLONE_ORIG &&
1280 refcount_read(&fclones->fclone_ref) == 1) {
1282 refcount_set(&fclones->fclone_ref, 2);
1284 if (skb_pfmemalloc(skb))
1285 gfp_mask |= __GFP_MEMALLOC;
1287 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1291 kmemcheck_annotate_bitfield(n, flags1);
1292 n->fclone = SKB_FCLONE_UNAVAILABLE;
1295 return __skb_clone(n, skb);
1297 EXPORT_SYMBOL(skb_clone);
1299 static void skb_headers_offset_update(struct sk_buff *skb, int off)
1301 /* Only adjust this if it actually is csum_start rather than csum */
1302 if (skb->ip_summed == CHECKSUM_PARTIAL)
1303 skb->csum_start += off;
1304 /* {transport,network,mac}_header and tail are relative to skb->head */
1305 skb->transport_header += off;
1306 skb->network_header += off;
1307 if (skb_mac_header_was_set(skb))
1308 skb->mac_header += off;
1309 skb->inner_transport_header += off;
1310 skb->inner_network_header += off;
1311 skb->inner_mac_header += off;
1314 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1316 __copy_skb_header(new, old);
1318 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1319 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1320 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1323 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1325 if (skb_pfmemalloc(skb))
1326 return SKB_ALLOC_RX;
1331 * skb_copy - create private copy of an sk_buff
1332 * @skb: buffer to copy
1333 * @gfp_mask: allocation priority
1335 * Make a copy of both an &sk_buff and its data. This is used when the
1336 * caller wishes to modify the data and needs a private copy of the
1337 * data to alter. Returns %NULL on failure or the pointer to the buffer
1338 * on success. The returned buffer has a reference count of 1.
1340 * As by-product this function converts non-linear &sk_buff to linear
1341 * one, so that &sk_buff becomes completely private and caller is allowed
1342 * to modify all the data of returned buffer. This means that this
1343 * function is not recommended for use in circumstances when only
1344 * header is going to be modified. Use pskb_copy() instead.
1347 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1349 int headerlen = skb_headroom(skb);
1350 unsigned int size = skb_end_offset(skb) + skb->data_len;
1351 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1352 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1357 /* Set the data pointer */
1358 skb_reserve(n, headerlen);
1359 /* Set the tail pointer and length */
1360 skb_put(n, skb->len);
1362 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1365 copy_skb_header(n, skb);
1368 EXPORT_SYMBOL(skb_copy);
1371 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1372 * @skb: buffer to copy
1373 * @headroom: headroom of new skb
1374 * @gfp_mask: allocation priority
1375 * @fclone: if true allocate the copy of the skb from the fclone
1376 * cache instead of the head cache; it is recommended to set this
1377 * to true for the cases where the copy will likely be cloned
1379 * Make a copy of both an &sk_buff and part of its data, located
1380 * in header. Fragmented data remain shared. This is used when
1381 * the caller wishes to modify only header of &sk_buff and needs
1382 * private copy of the header to alter. Returns %NULL on failure
1383 * or the pointer to the buffer on success.
1384 * The returned buffer has a reference count of 1.
1387 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1388 gfp_t gfp_mask, bool fclone)
1390 unsigned int size = skb_headlen(skb) + headroom;
1391 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1392 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1397 /* Set the data pointer */
1398 skb_reserve(n, headroom);
1399 /* Set the tail pointer and length */
1400 skb_put(n, skb_headlen(skb));
1401 /* Copy the bytes */
1402 skb_copy_from_linear_data(skb, n->data, n->len);
1404 n->truesize += skb->data_len;
1405 n->data_len = skb->data_len;
1408 if (skb_shinfo(skb)->nr_frags) {
1411 if (skb_orphan_frags(skb, gfp_mask) ||
1412 skb_zerocopy_clone(n, skb, gfp_mask)) {
1417 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1418 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1419 skb_frag_ref(skb, i);
1421 skb_shinfo(n)->nr_frags = i;
1424 if (skb_has_frag_list(skb)) {
1425 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1426 skb_clone_fraglist(n);
1429 copy_skb_header(n, skb);
1433 EXPORT_SYMBOL(__pskb_copy_fclone);
1436 * pskb_expand_head - reallocate header of &sk_buff
1437 * @skb: buffer to reallocate
1438 * @nhead: room to add at head
1439 * @ntail: room to add at tail
1440 * @gfp_mask: allocation priority
1442 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1443 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1444 * reference count of 1. Returns zero in the case of success or error,
1445 * if expansion failed. In the last case, &sk_buff is not changed.
1447 * All the pointers pointing into skb header may change and must be
1448 * reloaded after call to this function.
1451 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1454 int i, osize = skb_end_offset(skb);
1455 int size = osize + nhead + ntail;
1461 if (skb_shared(skb))
1464 size = SKB_DATA_ALIGN(size);
1466 if (skb_pfmemalloc(skb))
1467 gfp_mask |= __GFP_MEMALLOC;
1468 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1469 gfp_mask, NUMA_NO_NODE, NULL);
1472 size = SKB_WITH_OVERHEAD(ksize(data));
1474 /* Copy only real data... and, alas, header. This should be
1475 * optimized for the cases when header is void.
1477 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1479 memcpy((struct skb_shared_info *)(data + size),
1481 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1484 * if shinfo is shared we must drop the old head gracefully, but if it
1485 * is not we can just drop the old head and let the existing refcount
1486 * be since all we did is relocate the values
1488 if (skb_cloned(skb)) {
1489 if (skb_orphan_frags(skb, gfp_mask))
1492 atomic_inc(&skb_uarg(skb)->refcnt);
1493 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1494 skb_frag_ref(skb, i);
1496 if (skb_has_frag_list(skb))
1497 skb_clone_fraglist(skb);
1499 skb_release_data(skb);
1503 off = (data + nhead) - skb->head;
1508 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1512 skb->end = skb->head + size;
1515 skb_headers_offset_update(skb, nhead);
1519 atomic_set(&skb_shinfo(skb)->dataref, 1);
1521 /* It is not generally safe to change skb->truesize.
1522 * For the moment, we really care of rx path, or
1523 * when skb is orphaned (not attached to a socket).
1525 if (!skb->sk || skb->destructor == sock_edemux)
1526 skb->truesize += size - osize;
1535 EXPORT_SYMBOL(pskb_expand_head);
1537 /* Make private copy of skb with writable head and some headroom */
1539 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1541 struct sk_buff *skb2;
1542 int delta = headroom - skb_headroom(skb);
1545 skb2 = pskb_copy(skb, GFP_ATOMIC);
1547 skb2 = skb_clone(skb, GFP_ATOMIC);
1548 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1556 EXPORT_SYMBOL(skb_realloc_headroom);
1559 * skb_copy_expand - copy and expand sk_buff
1560 * @skb: buffer to copy
1561 * @newheadroom: new free bytes at head
1562 * @newtailroom: new free bytes at tail
1563 * @gfp_mask: allocation priority
1565 * Make a copy of both an &sk_buff and its data and while doing so
1566 * allocate additional space.
1568 * This is used when the caller wishes to modify the data and needs a
1569 * private copy of the data to alter as well as more space for new fields.
1570 * Returns %NULL on failure or the pointer to the buffer
1571 * on success. The returned buffer has a reference count of 1.
1573 * You must pass %GFP_ATOMIC as the allocation priority if this function
1574 * is called from an interrupt.
1576 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1577 int newheadroom, int newtailroom,
1581 * Allocate the copy buffer
1583 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1584 gfp_mask, skb_alloc_rx_flag(skb),
1586 int oldheadroom = skb_headroom(skb);
1587 int head_copy_len, head_copy_off;
1592 skb_reserve(n, newheadroom);
1594 /* Set the tail pointer and length */
1595 skb_put(n, skb->len);
1597 head_copy_len = oldheadroom;
1599 if (newheadroom <= head_copy_len)
1600 head_copy_len = newheadroom;
1602 head_copy_off = newheadroom - head_copy_len;
1604 /* Copy the linear header and data. */
1605 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1606 skb->len + head_copy_len))
1609 copy_skb_header(n, skb);
1611 skb_headers_offset_update(n, newheadroom - oldheadroom);
1615 EXPORT_SYMBOL(skb_copy_expand);
1618 * skb_pad - zero pad the tail of an skb
1619 * @skb: buffer to pad
1620 * @pad: space to pad
1622 * Ensure that a buffer is followed by a padding area that is zero
1623 * filled. Used by network drivers which may DMA or transfer data
1624 * beyond the buffer end onto the wire.
1626 * May return error in out of memory cases. The skb is freed on error.
1629 int skb_pad(struct sk_buff *skb, int pad)
1634 /* If the skbuff is non linear tailroom is always zero.. */
1635 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1636 memset(skb->data+skb->len, 0, pad);
1640 ntail = skb->data_len + pad - (skb->end - skb->tail);
1641 if (likely(skb_cloned(skb) || ntail > 0)) {
1642 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1647 /* FIXME: The use of this function with non-linear skb's really needs
1650 err = skb_linearize(skb);
1654 memset(skb->data + skb->len, 0, pad);
1661 EXPORT_SYMBOL(skb_pad);
1664 * pskb_put - add data to the tail of a potentially fragmented buffer
1665 * @skb: start of the buffer to use
1666 * @tail: tail fragment of the buffer to use
1667 * @len: amount of data to add
1669 * This function extends the used data area of the potentially
1670 * fragmented buffer. @tail must be the last fragment of @skb -- or
1671 * @skb itself. If this would exceed the total buffer size the kernel
1672 * will panic. A pointer to the first byte of the extra data is
1676 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1679 skb->data_len += len;
1682 return skb_put(tail, len);
1684 EXPORT_SYMBOL_GPL(pskb_put);
1687 * skb_put - add data to a buffer
1688 * @skb: buffer to use
1689 * @len: amount of data to add
1691 * This function extends the used data area of the buffer. If this would
1692 * exceed the total buffer size the kernel will panic. A pointer to the
1693 * first byte of the extra data is returned.
1695 void *skb_put(struct sk_buff *skb, unsigned int len)
1697 void *tmp = skb_tail_pointer(skb);
1698 SKB_LINEAR_ASSERT(skb);
1701 if (unlikely(skb->tail > skb->end))
1702 skb_over_panic(skb, len, __builtin_return_address(0));
1705 EXPORT_SYMBOL(skb_put);
1708 * skb_push - add data to the start of a buffer
1709 * @skb: buffer to use
1710 * @len: amount of data to add
1712 * This function extends the used data area of the buffer at the buffer
1713 * start. If this would exceed the total buffer headroom the kernel will
1714 * panic. A pointer to the first byte of the extra data is returned.
1716 void *skb_push(struct sk_buff *skb, unsigned int len)
1720 if (unlikely(skb->data<skb->head))
1721 skb_under_panic(skb, len, __builtin_return_address(0));
1724 EXPORT_SYMBOL(skb_push);
1727 * skb_pull - remove data from the start of a buffer
1728 * @skb: buffer to use
1729 * @len: amount of data to remove
1731 * This function removes data from the start of a buffer, returning
1732 * the memory to the headroom. A pointer to the next data in the buffer
1733 * is returned. Once the data has been pulled future pushes will overwrite
1736 void *skb_pull(struct sk_buff *skb, unsigned int len)
1738 return skb_pull_inline(skb, len);
1740 EXPORT_SYMBOL(skb_pull);
1743 * skb_trim - remove end from a buffer
1744 * @skb: buffer to alter
1747 * Cut the length of a buffer down by removing data from the tail. If
1748 * the buffer is already under the length specified it is not modified.
1749 * The skb must be linear.
1751 void skb_trim(struct sk_buff *skb, unsigned int len)
1754 __skb_trim(skb, len);
1756 EXPORT_SYMBOL(skb_trim);
1758 /* Trims skb to length len. It can change skb pointers.
1761 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1763 struct sk_buff **fragp;
1764 struct sk_buff *frag;
1765 int offset = skb_headlen(skb);
1766 int nfrags = skb_shinfo(skb)->nr_frags;
1770 if (skb_cloned(skb) &&
1771 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1778 for (; i < nfrags; i++) {
1779 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1786 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1789 skb_shinfo(skb)->nr_frags = i;
1791 for (; i < nfrags; i++)
1792 skb_frag_unref(skb, i);
1794 if (skb_has_frag_list(skb))
1795 skb_drop_fraglist(skb);
1799 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1800 fragp = &frag->next) {
1801 int end = offset + frag->len;
1803 if (skb_shared(frag)) {
1804 struct sk_buff *nfrag;
1806 nfrag = skb_clone(frag, GFP_ATOMIC);
1807 if (unlikely(!nfrag))
1810 nfrag->next = frag->next;
1822 unlikely((err = pskb_trim(frag, len - offset))))
1826 skb_drop_list(&frag->next);
1831 if (len > skb_headlen(skb)) {
1832 skb->data_len -= skb->len - len;
1837 skb_set_tail_pointer(skb, len);
1840 if (!skb->sk || skb->destructor == sock_edemux)
1844 EXPORT_SYMBOL(___pskb_trim);
1847 * __pskb_pull_tail - advance tail of skb header
1848 * @skb: buffer to reallocate
1849 * @delta: number of bytes to advance tail
1851 * The function makes a sense only on a fragmented &sk_buff,
1852 * it expands header moving its tail forward and copying necessary
1853 * data from fragmented part.
1855 * &sk_buff MUST have reference count of 1.
1857 * Returns %NULL (and &sk_buff does not change) if pull failed
1858 * or value of new tail of skb in the case of success.
1860 * All the pointers pointing into skb header may change and must be
1861 * reloaded after call to this function.
1864 /* Moves tail of skb head forward, copying data from fragmented part,
1865 * when it is necessary.
1866 * 1. It may fail due to malloc failure.
1867 * 2. It may change skb pointers.
1869 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1871 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
1873 /* If skb has not enough free space at tail, get new one
1874 * plus 128 bytes for future expansions. If we have enough
1875 * room at tail, reallocate without expansion only if skb is cloned.
1877 int i, k, eat = (skb->tail + delta) - skb->end;
1879 if (eat > 0 || skb_cloned(skb)) {
1880 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1885 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1888 /* Optimization: no fragments, no reasons to preestimate
1889 * size of pulled pages. Superb.
1891 if (!skb_has_frag_list(skb))
1894 /* Estimate size of pulled pages. */
1896 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1897 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1904 /* If we need update frag list, we are in troubles.
1905 * Certainly, it possible to add an offset to skb data,
1906 * but taking into account that pulling is expected to
1907 * be very rare operation, it is worth to fight against
1908 * further bloating skb head and crucify ourselves here instead.
1909 * Pure masohism, indeed. 8)8)
1912 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1913 struct sk_buff *clone = NULL;
1914 struct sk_buff *insp = NULL;
1919 if (list->len <= eat) {
1920 /* Eaten as whole. */
1925 /* Eaten partially. */
1927 if (skb_shared(list)) {
1928 /* Sucks! We need to fork list. :-( */
1929 clone = skb_clone(list, GFP_ATOMIC);
1935 /* This may be pulled without
1939 if (!pskb_pull(list, eat)) {
1947 /* Free pulled out fragments. */
1948 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1949 skb_shinfo(skb)->frag_list = list->next;
1952 /* And insert new clone at head. */
1955 skb_shinfo(skb)->frag_list = clone;
1958 /* Success! Now we may commit changes to skb data. */
1963 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1964 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1967 skb_frag_unref(skb, i);
1970 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1972 skb_shinfo(skb)->frags[k].page_offset += eat;
1973 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1981 skb_shinfo(skb)->nr_frags = k;
1985 skb->data_len -= delta;
1988 skb_zcopy_clear(skb, false);
1990 return skb_tail_pointer(skb);
1992 EXPORT_SYMBOL(__pskb_pull_tail);
1995 * skb_copy_bits - copy bits from skb to kernel buffer
1997 * @offset: offset in source
1998 * @to: destination buffer
1999 * @len: number of bytes to copy
2001 * Copy the specified number of bytes from the source skb to the
2002 * destination buffer.
2005 * If its prototype is ever changed,
2006 * check arch/{*}/net/{*}.S files,
2007 * since it is called from BPF assembly code.
2009 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2011 int start = skb_headlen(skb);
2012 struct sk_buff *frag_iter;
2015 if (offset > (int)skb->len - len)
2019 if ((copy = start - offset) > 0) {
2022 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2023 if ((len -= copy) == 0)
2029 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2031 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2033 WARN_ON(start > offset + len);
2035 end = start + skb_frag_size(f);
2036 if ((copy = end - offset) > 0) {
2037 u32 p_off, p_len, copied;
2044 skb_frag_foreach_page(f,
2045 f->page_offset + offset - start,
2046 copy, p, p_off, p_len, copied) {
2047 vaddr = kmap_atomic(p);
2048 memcpy(to + copied, vaddr + p_off, p_len);
2049 kunmap_atomic(vaddr);
2052 if ((len -= copy) == 0)
2060 skb_walk_frags(skb, frag_iter) {
2063 WARN_ON(start > offset + len);
2065 end = start + frag_iter->len;
2066 if ((copy = end - offset) > 0) {
2069 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2071 if ((len -= copy) == 0)
2085 EXPORT_SYMBOL(skb_copy_bits);
2088 * Callback from splice_to_pipe(), if we need to release some pages
2089 * at the end of the spd in case we error'ed out in filling the pipe.
2091 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2093 put_page(spd->pages[i]);
2096 static struct page *linear_to_page(struct page *page, unsigned int *len,
2097 unsigned int *offset,
2100 struct page_frag *pfrag = sk_page_frag(sk);
2102 if (!sk_page_frag_refill(sk, pfrag))
2105 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2107 memcpy(page_address(pfrag->page) + pfrag->offset,
2108 page_address(page) + *offset, *len);
2109 *offset = pfrag->offset;
2110 pfrag->offset += *len;
2115 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2117 unsigned int offset)
2119 return spd->nr_pages &&
2120 spd->pages[spd->nr_pages - 1] == page &&
2121 (spd->partial[spd->nr_pages - 1].offset +
2122 spd->partial[spd->nr_pages - 1].len == offset);
2126 * Fill page/offset/length into spd, if it can hold more pages.
2128 static bool spd_fill_page(struct splice_pipe_desc *spd,
2129 struct pipe_inode_info *pipe, struct page *page,
2130 unsigned int *len, unsigned int offset,
2134 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2138 page = linear_to_page(page, len, &offset, sk);
2142 if (spd_can_coalesce(spd, page, offset)) {
2143 spd->partial[spd->nr_pages - 1].len += *len;
2147 spd->pages[spd->nr_pages] = page;
2148 spd->partial[spd->nr_pages].len = *len;
2149 spd->partial[spd->nr_pages].offset = offset;
2155 static bool __splice_segment(struct page *page, unsigned int poff,
2156 unsigned int plen, unsigned int *off,
2158 struct splice_pipe_desc *spd, bool linear,
2160 struct pipe_inode_info *pipe)
2165 /* skip this segment if already processed */
2171 /* ignore any bits we already processed */
2177 unsigned int flen = min(*len, plen);
2179 if (spd_fill_page(spd, pipe, page, &flen, poff,
2185 } while (*len && plen);
2191 * Map linear and fragment data from the skb to spd. It reports true if the
2192 * pipe is full or if we already spliced the requested length.
2194 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2195 unsigned int *offset, unsigned int *len,
2196 struct splice_pipe_desc *spd, struct sock *sk)
2199 struct sk_buff *iter;
2201 /* map the linear part :
2202 * If skb->head_frag is set, this 'linear' part is backed by a
2203 * fragment, and if the head is not shared with any clones then
2204 * we can avoid a copy since we own the head portion of this page.
2206 if (__splice_segment(virt_to_page(skb->data),
2207 (unsigned long) skb->data & (PAGE_SIZE - 1),
2210 skb_head_is_locked(skb),
2215 * then map the fragments
2217 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2218 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2220 if (__splice_segment(skb_frag_page(f),
2221 f->page_offset, skb_frag_size(f),
2222 offset, len, spd, false, sk, pipe))
2226 skb_walk_frags(skb, iter) {
2227 if (*offset >= iter->len) {
2228 *offset -= iter->len;
2231 /* __skb_splice_bits() only fails if the output has no room
2232 * left, so no point in going over the frag_list for the error
2235 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2243 * Map data from the skb to a pipe. Should handle both the linear part,
2244 * the fragments, and the frag list.
2246 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2247 struct pipe_inode_info *pipe, unsigned int tlen,
2250 struct partial_page partial[MAX_SKB_FRAGS];
2251 struct page *pages[MAX_SKB_FRAGS];
2252 struct splice_pipe_desc spd = {
2255 .nr_pages_max = MAX_SKB_FRAGS,
2256 .ops = &nosteal_pipe_buf_ops,
2257 .spd_release = sock_spd_release,
2261 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2264 ret = splice_to_pipe(pipe, &spd);
2268 EXPORT_SYMBOL_GPL(skb_splice_bits);
2270 /* Send skb data on a socket. Socket must be locked. */
2271 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2274 unsigned int orig_len = len;
2275 struct sk_buff *head = skb;
2276 unsigned short fragidx;
2281 /* Deal with head data */
2282 while (offset < skb_headlen(skb) && len) {
2286 slen = min_t(int, len, skb_headlen(skb) - offset);
2287 kv.iov_base = skb->data + offset;
2289 memset(&msg, 0, sizeof(msg));
2291 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2299 /* All the data was skb head? */
2303 /* Make offset relative to start of frags */
2304 offset -= skb_headlen(skb);
2306 /* Find where we are in frag list */
2307 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2308 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2310 if (offset < frag->size)
2313 offset -= frag->size;
2316 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2317 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2319 slen = min_t(size_t, len, frag->size - offset);
2322 ret = kernel_sendpage_locked(sk, frag->page.p,
2323 frag->page_offset + offset,
2324 slen, MSG_DONTWAIT);
2337 /* Process any frag lists */
2340 if (skb_has_frag_list(skb)) {
2341 skb = skb_shinfo(skb)->frag_list;
2344 } else if (skb->next) {
2351 return orig_len - len;
2354 return orig_len == len ? ret : orig_len - len;
2356 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2358 /* Send skb data on a socket. */
2359 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
2364 ret = skb_send_sock_locked(sk, skb, offset, len);
2369 EXPORT_SYMBOL_GPL(skb_send_sock);
2372 * skb_store_bits - store bits from kernel buffer to skb
2373 * @skb: destination buffer
2374 * @offset: offset in destination
2375 * @from: source buffer
2376 * @len: number of bytes to copy
2378 * Copy the specified number of bytes from the source buffer to the
2379 * destination skb. This function handles all the messy bits of
2380 * traversing fragment lists and such.
2383 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2385 int start = skb_headlen(skb);
2386 struct sk_buff *frag_iter;
2389 if (offset > (int)skb->len - len)
2392 if ((copy = start - offset) > 0) {
2395 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2396 if ((len -= copy) == 0)
2402 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2403 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2406 WARN_ON(start > offset + len);
2408 end = start + skb_frag_size(frag);
2409 if ((copy = end - offset) > 0) {
2410 u32 p_off, p_len, copied;
2417 skb_frag_foreach_page(frag,
2418 frag->page_offset + offset - start,
2419 copy, p, p_off, p_len, copied) {
2420 vaddr = kmap_atomic(p);
2421 memcpy(vaddr + p_off, from + copied, p_len);
2422 kunmap_atomic(vaddr);
2425 if ((len -= copy) == 0)
2433 skb_walk_frags(skb, frag_iter) {
2436 WARN_ON(start > offset + len);
2438 end = start + frag_iter->len;
2439 if ((copy = end - offset) > 0) {
2442 if (skb_store_bits(frag_iter, offset - start,
2445 if ((len -= copy) == 0)
2458 EXPORT_SYMBOL(skb_store_bits);
2460 /* Checksum skb data. */
2461 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2462 __wsum csum, const struct skb_checksum_ops *ops)
2464 int start = skb_headlen(skb);
2465 int i, copy = start - offset;
2466 struct sk_buff *frag_iter;
2469 /* Checksum header. */
2473 csum = ops->update(skb->data + offset, copy, csum);
2474 if ((len -= copy) == 0)
2480 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2482 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2484 WARN_ON(start > offset + len);
2486 end = start + skb_frag_size(frag);
2487 if ((copy = end - offset) > 0) {
2488 u32 p_off, p_len, copied;
2496 skb_frag_foreach_page(frag,
2497 frag->page_offset + offset - start,
2498 copy, p, p_off, p_len, copied) {
2499 vaddr = kmap_atomic(p);
2500 csum2 = ops->update(vaddr + p_off, p_len, 0);
2501 kunmap_atomic(vaddr);
2502 csum = ops->combine(csum, csum2, pos, p_len);
2513 skb_walk_frags(skb, frag_iter) {
2516 WARN_ON(start > offset + len);
2518 end = start + frag_iter->len;
2519 if ((copy = end - offset) > 0) {
2523 csum2 = __skb_checksum(frag_iter, offset - start,
2525 csum = ops->combine(csum, csum2, pos, copy);
2526 if ((len -= copy) == 0)
2537 EXPORT_SYMBOL(__skb_checksum);
2539 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2540 int len, __wsum csum)
2542 const struct skb_checksum_ops ops = {
2543 .update = csum_partial_ext,
2544 .combine = csum_block_add_ext,
2547 return __skb_checksum(skb, offset, len, csum, &ops);
2549 EXPORT_SYMBOL(skb_checksum);
2551 /* Both of above in one bottle. */
2553 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2554 u8 *to, int len, __wsum csum)
2556 int start = skb_headlen(skb);
2557 int i, copy = start - offset;
2558 struct sk_buff *frag_iter;
2565 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2567 if ((len -= copy) == 0)
2574 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2577 WARN_ON(start > offset + len);
2579 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2580 if ((copy = end - offset) > 0) {
2581 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2582 u32 p_off, p_len, copied;
2590 skb_frag_foreach_page(frag,
2591 frag->page_offset + offset - start,
2592 copy, p, p_off, p_len, copied) {
2593 vaddr = kmap_atomic(p);
2594 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2597 kunmap_atomic(vaddr);
2598 csum = csum_block_add(csum, csum2, pos);
2610 skb_walk_frags(skb, frag_iter) {
2614 WARN_ON(start > offset + len);
2616 end = start + frag_iter->len;
2617 if ((copy = end - offset) > 0) {
2620 csum2 = skb_copy_and_csum_bits(frag_iter,
2623 csum = csum_block_add(csum, csum2, pos);
2624 if ((len -= copy) == 0)
2635 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2637 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2639 net_warn_ratelimited(
2640 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2645 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2646 int offset, int len)
2648 net_warn_ratelimited(
2649 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2654 static const struct skb_checksum_ops default_crc32c_ops = {
2655 .update = warn_crc32c_csum_update,
2656 .combine = warn_crc32c_csum_combine,
2659 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2660 &default_crc32c_ops;
2661 EXPORT_SYMBOL(crc32c_csum_stub);
2664 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2665 * @from: source buffer
2667 * Calculates the amount of linear headroom needed in the 'to' skb passed
2668 * into skb_zerocopy().
2671 skb_zerocopy_headlen(const struct sk_buff *from)
2673 unsigned int hlen = 0;
2675 if (!from->head_frag ||
2676 skb_headlen(from) < L1_CACHE_BYTES ||
2677 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2678 hlen = skb_headlen(from);
2680 if (skb_has_frag_list(from))
2685 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2688 * skb_zerocopy - Zero copy skb to skb
2689 * @to: destination buffer
2690 * @from: source buffer
2691 * @len: number of bytes to copy from source buffer
2692 * @hlen: size of linear headroom in destination buffer
2694 * Copies up to `len` bytes from `from` to `to` by creating references
2695 * to the frags in the source buffer.
2697 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2698 * headroom in the `to` buffer.
2701 * 0: everything is OK
2702 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2703 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2706 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2709 int plen = 0; /* length of skb->head fragment */
2712 unsigned int offset;
2714 BUG_ON(!from->head_frag && !hlen);
2716 /* dont bother with small payloads */
2717 if (len <= skb_tailroom(to))
2718 return skb_copy_bits(from, 0, skb_put(to, len), len);
2721 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2726 plen = min_t(int, skb_headlen(from), len);
2728 page = virt_to_head_page(from->head);
2729 offset = from->data - (unsigned char *)page_address(page);
2730 __skb_fill_page_desc(to, 0, page, offset, plen);
2737 to->truesize += len + plen;
2738 to->len += len + plen;
2739 to->data_len += len + plen;
2741 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2745 skb_zerocopy_clone(to, from, GFP_ATOMIC);
2747 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2750 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2751 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2752 len -= skb_shinfo(to)->frags[j].size;
2753 skb_frag_ref(to, j);
2756 skb_shinfo(to)->nr_frags = j;
2760 EXPORT_SYMBOL_GPL(skb_zerocopy);
2762 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2767 if (skb->ip_summed == CHECKSUM_PARTIAL)
2768 csstart = skb_checksum_start_offset(skb);
2770 csstart = skb_headlen(skb);
2772 BUG_ON(csstart > skb_headlen(skb));
2774 skb_copy_from_linear_data(skb, to, csstart);
2777 if (csstart != skb->len)
2778 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2779 skb->len - csstart, 0);
2781 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2782 long csstuff = csstart + skb->csum_offset;
2784 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2787 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2790 * skb_dequeue - remove from the head of the queue
2791 * @list: list to dequeue from
2793 * Remove the head of the list. The list lock is taken so the function
2794 * may be used safely with other locking list functions. The head item is
2795 * returned or %NULL if the list is empty.
2798 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2800 unsigned long flags;
2801 struct sk_buff *result;
2803 spin_lock_irqsave(&list->lock, flags);
2804 result = __skb_dequeue(list);
2805 spin_unlock_irqrestore(&list->lock, flags);
2808 EXPORT_SYMBOL(skb_dequeue);
2811 * skb_dequeue_tail - remove from the tail of the queue
2812 * @list: list to dequeue from
2814 * Remove the tail of the list. The list lock is taken so the function
2815 * may be used safely with other locking list functions. The tail item is
2816 * returned or %NULL if the list is empty.
2818 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2820 unsigned long flags;
2821 struct sk_buff *result;
2823 spin_lock_irqsave(&list->lock, flags);
2824 result = __skb_dequeue_tail(list);
2825 spin_unlock_irqrestore(&list->lock, flags);
2828 EXPORT_SYMBOL(skb_dequeue_tail);
2831 * skb_queue_purge - empty a list
2832 * @list: list to empty
2834 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2835 * the list and one reference dropped. This function takes the list
2836 * lock and is atomic with respect to other list locking functions.
2838 void skb_queue_purge(struct sk_buff_head *list)
2840 struct sk_buff *skb;
2841 while ((skb = skb_dequeue(list)) != NULL)
2844 EXPORT_SYMBOL(skb_queue_purge);
2847 * skb_rbtree_purge - empty a skb rbtree
2848 * @root: root of the rbtree to empty
2850 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
2851 * the list and one reference dropped. This function does not take
2852 * any lock. Synchronization should be handled by the caller (e.g., TCP
2853 * out-of-order queue is protected by the socket lock).
2855 void skb_rbtree_purge(struct rb_root *root)
2857 struct sk_buff *skb, *next;
2859 rbtree_postorder_for_each_entry_safe(skb, next, root, rbnode)
2866 * skb_queue_head - queue a buffer at the list head
2867 * @list: list to use
2868 * @newsk: buffer to queue
2870 * Queue a buffer at the start of the list. This function takes the
2871 * list lock and can be used safely with other locking &sk_buff functions
2874 * A buffer cannot be placed on two lists at the same time.
2876 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2878 unsigned long flags;
2880 spin_lock_irqsave(&list->lock, flags);
2881 __skb_queue_head(list, newsk);
2882 spin_unlock_irqrestore(&list->lock, flags);
2884 EXPORT_SYMBOL(skb_queue_head);
2887 * skb_queue_tail - queue a buffer at the list tail
2888 * @list: list to use
2889 * @newsk: buffer to queue
2891 * Queue a buffer at the tail of the list. This function takes the
2892 * list lock and can be used safely with other locking &sk_buff functions
2895 * A buffer cannot be placed on two lists at the same time.
2897 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2899 unsigned long flags;
2901 spin_lock_irqsave(&list->lock, flags);
2902 __skb_queue_tail(list, newsk);
2903 spin_unlock_irqrestore(&list->lock, flags);
2905 EXPORT_SYMBOL(skb_queue_tail);
2908 * skb_unlink - remove a buffer from a list
2909 * @skb: buffer to remove
2910 * @list: list to use
2912 * Remove a packet from a list. The list locks are taken and this
2913 * function is atomic with respect to other list locked calls
2915 * You must know what list the SKB is on.
2917 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2919 unsigned long flags;
2921 spin_lock_irqsave(&list->lock, flags);
2922 __skb_unlink(skb, list);
2923 spin_unlock_irqrestore(&list->lock, flags);
2925 EXPORT_SYMBOL(skb_unlink);
2928 * skb_append - append a buffer
2929 * @old: buffer to insert after
2930 * @newsk: buffer to insert
2931 * @list: list to use
2933 * Place a packet after a given packet in a list. The list locks are taken
2934 * and this function is atomic with respect to other list locked calls.
2935 * A buffer cannot be placed on two lists at the same time.
2937 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2939 unsigned long flags;
2941 spin_lock_irqsave(&list->lock, flags);
2942 __skb_queue_after(list, old, newsk);
2943 spin_unlock_irqrestore(&list->lock, flags);
2945 EXPORT_SYMBOL(skb_append);
2948 * skb_insert - insert a buffer
2949 * @old: buffer to insert before
2950 * @newsk: buffer to insert
2951 * @list: list to use
2953 * Place a packet before a given packet in a list. The list locks are
2954 * taken and this function is atomic with respect to other list locked
2957 * A buffer cannot be placed on two lists at the same time.
2959 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2961 unsigned long flags;
2963 spin_lock_irqsave(&list->lock, flags);
2964 __skb_insert(newsk, old->prev, old, list);
2965 spin_unlock_irqrestore(&list->lock, flags);
2967 EXPORT_SYMBOL(skb_insert);
2969 static inline void skb_split_inside_header(struct sk_buff *skb,
2970 struct sk_buff* skb1,
2971 const u32 len, const int pos)
2975 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2977 /* And move data appendix as is. */
2978 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2979 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2981 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2982 skb_shinfo(skb)->nr_frags = 0;
2983 skb1->data_len = skb->data_len;
2984 skb1->len += skb1->data_len;
2987 skb_set_tail_pointer(skb, len);
2990 static inline void skb_split_no_header(struct sk_buff *skb,
2991 struct sk_buff* skb1,
2992 const u32 len, int pos)
2995 const int nfrags = skb_shinfo(skb)->nr_frags;
2997 skb_shinfo(skb)->nr_frags = 0;
2998 skb1->len = skb1->data_len = skb->len - len;
3000 skb->data_len = len - pos;
3002 for (i = 0; i < nfrags; i++) {
3003 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3005 if (pos + size > len) {
3006 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3010 * We have two variants in this case:
3011 * 1. Move all the frag to the second
3012 * part, if it is possible. F.e.
3013 * this approach is mandatory for TUX,
3014 * where splitting is expensive.
3015 * 2. Split is accurately. We make this.
3017 skb_frag_ref(skb, i);
3018 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
3019 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3020 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3021 skb_shinfo(skb)->nr_frags++;
3025 skb_shinfo(skb)->nr_frags++;
3028 skb_shinfo(skb1)->nr_frags = k;
3032 * skb_split - Split fragmented skb to two parts at length len.
3033 * @skb: the buffer to split
3034 * @skb1: the buffer to receive the second part
3035 * @len: new length for skb
3037 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3039 int pos = skb_headlen(skb);
3041 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3043 skb_zerocopy_clone(skb1, skb, 0);
3044 if (len < pos) /* Split line is inside header. */
3045 skb_split_inside_header(skb, skb1, len, pos);
3046 else /* Second chunk has no header, nothing to copy. */
3047 skb_split_no_header(skb, skb1, len, pos);
3049 EXPORT_SYMBOL(skb_split);
3051 /* Shifting from/to a cloned skb is a no-go.
3053 * Caller cannot keep skb_shinfo related pointers past calling here!
3055 static int skb_prepare_for_shift(struct sk_buff *skb)
3057 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3061 * skb_shift - Shifts paged data partially from skb to another
3062 * @tgt: buffer into which tail data gets added
3063 * @skb: buffer from which the paged data comes from
3064 * @shiftlen: shift up to this many bytes
3066 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3067 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3068 * It's up to caller to free skb if everything was shifted.
3070 * If @tgt runs out of frags, the whole operation is aborted.
3072 * Skb cannot include anything else but paged data while tgt is allowed
3073 * to have non-paged data as well.
3075 * TODO: full sized shift could be optimized but that would need
3076 * specialized skb free'er to handle frags without up-to-date nr_frags.
3078 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3080 int from, to, merge, todo;
3081 struct skb_frag_struct *fragfrom, *fragto;
3083 BUG_ON(shiftlen > skb->len);
3085 if (skb_headlen(skb))
3087 if (skb_zcopy(tgt) || skb_zcopy(skb))
3092 to = skb_shinfo(tgt)->nr_frags;
3093 fragfrom = &skb_shinfo(skb)->frags[from];
3095 /* Actual merge is delayed until the point when we know we can
3096 * commit all, so that we don't have to undo partial changes
3099 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3100 fragfrom->page_offset)) {
3105 todo -= skb_frag_size(fragfrom);
3107 if (skb_prepare_for_shift(skb) ||
3108 skb_prepare_for_shift(tgt))
3111 /* All previous frag pointers might be stale! */
3112 fragfrom = &skb_shinfo(skb)->frags[from];
3113 fragto = &skb_shinfo(tgt)->frags[merge];
3115 skb_frag_size_add(fragto, shiftlen);
3116 skb_frag_size_sub(fragfrom, shiftlen);
3117 fragfrom->page_offset += shiftlen;
3125 /* Skip full, not-fitting skb to avoid expensive operations */
3126 if ((shiftlen == skb->len) &&
3127 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3130 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3133 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3134 if (to == MAX_SKB_FRAGS)
3137 fragfrom = &skb_shinfo(skb)->frags[from];
3138 fragto = &skb_shinfo(tgt)->frags[to];
3140 if (todo >= skb_frag_size(fragfrom)) {
3141 *fragto = *fragfrom;
3142 todo -= skb_frag_size(fragfrom);
3147 __skb_frag_ref(fragfrom);
3148 fragto->page = fragfrom->page;
3149 fragto->page_offset = fragfrom->page_offset;
3150 skb_frag_size_set(fragto, todo);
3152 fragfrom->page_offset += todo;
3153 skb_frag_size_sub(fragfrom, todo);
3161 /* Ready to "commit" this state change to tgt */
3162 skb_shinfo(tgt)->nr_frags = to;
3165 fragfrom = &skb_shinfo(skb)->frags[0];
3166 fragto = &skb_shinfo(tgt)->frags[merge];
3168 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3169 __skb_frag_unref(fragfrom);
3172 /* Reposition in the original skb */
3174 while (from < skb_shinfo(skb)->nr_frags)
3175 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3176 skb_shinfo(skb)->nr_frags = to;
3178 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3181 /* Most likely the tgt won't ever need its checksum anymore, skb on
3182 * the other hand might need it if it needs to be resent
3184 tgt->ip_summed = CHECKSUM_PARTIAL;
3185 skb->ip_summed = CHECKSUM_PARTIAL;
3187 /* Yak, is it really working this way? Some helper please? */
3188 skb->len -= shiftlen;
3189 skb->data_len -= shiftlen;
3190 skb->truesize -= shiftlen;
3191 tgt->len += shiftlen;
3192 tgt->data_len += shiftlen;
3193 tgt->truesize += shiftlen;
3199 * skb_prepare_seq_read - Prepare a sequential read of skb data
3200 * @skb: the buffer to read
3201 * @from: lower offset of data to be read
3202 * @to: upper offset of data to be read
3203 * @st: state variable
3205 * Initializes the specified state variable. Must be called before
3206 * invoking skb_seq_read() for the first time.
3208 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3209 unsigned int to, struct skb_seq_state *st)
3211 st->lower_offset = from;
3212 st->upper_offset = to;
3213 st->root_skb = st->cur_skb = skb;
3214 st->frag_idx = st->stepped_offset = 0;
3215 st->frag_data = NULL;
3217 EXPORT_SYMBOL(skb_prepare_seq_read);
3220 * skb_seq_read - Sequentially read skb data
3221 * @consumed: number of bytes consumed by the caller so far
3222 * @data: destination pointer for data to be returned
3223 * @st: state variable
3225 * Reads a block of skb data at @consumed relative to the
3226 * lower offset specified to skb_prepare_seq_read(). Assigns
3227 * the head of the data block to @data and returns the length
3228 * of the block or 0 if the end of the skb data or the upper
3229 * offset has been reached.
3231 * The caller is not required to consume all of the data
3232 * returned, i.e. @consumed is typically set to the number
3233 * of bytes already consumed and the next call to
3234 * skb_seq_read() will return the remaining part of the block.
3236 * Note 1: The size of each block of data returned can be arbitrary,
3237 * this limitation is the cost for zerocopy sequential
3238 * reads of potentially non linear data.
3240 * Note 2: Fragment lists within fragments are not implemented
3241 * at the moment, state->root_skb could be replaced with
3242 * a stack for this purpose.
3244 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3245 struct skb_seq_state *st)
3247 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3250 if (unlikely(abs_offset >= st->upper_offset)) {
3251 if (st->frag_data) {
3252 kunmap_atomic(st->frag_data);
3253 st->frag_data = NULL;
3259 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3261 if (abs_offset < block_limit && !st->frag_data) {
3262 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3263 return block_limit - abs_offset;
3266 if (st->frag_idx == 0 && !st->frag_data)
3267 st->stepped_offset += skb_headlen(st->cur_skb);
3269 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3270 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3271 block_limit = skb_frag_size(frag) + st->stepped_offset;
3273 if (abs_offset < block_limit) {
3275 st->frag_data = kmap_atomic(skb_frag_page(frag));
3277 *data = (u8 *) st->frag_data + frag->page_offset +
3278 (abs_offset - st->stepped_offset);
3280 return block_limit - abs_offset;
3283 if (st->frag_data) {
3284 kunmap_atomic(st->frag_data);
3285 st->frag_data = NULL;
3289 st->stepped_offset += skb_frag_size(frag);
3292 if (st->frag_data) {
3293 kunmap_atomic(st->frag_data);
3294 st->frag_data = NULL;
3297 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3298 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3301 } else if (st->cur_skb->next) {
3302 st->cur_skb = st->cur_skb->next;
3309 EXPORT_SYMBOL(skb_seq_read);
3312 * skb_abort_seq_read - Abort a sequential read of skb data
3313 * @st: state variable
3315 * Must be called if skb_seq_read() was not called until it
3318 void skb_abort_seq_read(struct skb_seq_state *st)
3321 kunmap_atomic(st->frag_data);
3323 EXPORT_SYMBOL(skb_abort_seq_read);
3325 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3327 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3328 struct ts_config *conf,
3329 struct ts_state *state)
3331 return skb_seq_read(offset, text, TS_SKB_CB(state));
3334 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3336 skb_abort_seq_read(TS_SKB_CB(state));
3340 * skb_find_text - Find a text pattern in skb data
3341 * @skb: the buffer to look in
3342 * @from: search offset
3344 * @config: textsearch configuration
3346 * Finds a pattern in the skb data according to the specified
3347 * textsearch configuration. Use textsearch_next() to retrieve
3348 * subsequent occurrences of the pattern. Returns the offset
3349 * to the first occurrence or UINT_MAX if no match was found.
3351 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3352 unsigned int to, struct ts_config *config)
3354 struct ts_state state;
3357 config->get_next_block = skb_ts_get_next_block;
3358 config->finish = skb_ts_finish;
3360 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3362 ret = textsearch_find(config, &state);
3363 return (ret <= to - from ? ret : UINT_MAX);
3365 EXPORT_SYMBOL(skb_find_text);
3368 * skb_append_datato_frags - append the user data to a skb
3369 * @sk: sock structure
3370 * @skb: skb structure to be appended with user data.
3371 * @getfrag: call back function to be used for getting the user data
3372 * @from: pointer to user message iov
3373 * @length: length of the iov message
3375 * Description: This procedure append the user data in the fragment part
3376 * of the skb if any page alloc fails user this procedure returns -ENOMEM
3378 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
3379 int (*getfrag)(void *from, char *to, int offset,
3380 int len, int odd, struct sk_buff *skb),
3381 void *from, int length)
3383 int frg_cnt = skb_shinfo(skb)->nr_frags;
3387 struct page_frag *pfrag = ¤t->task_frag;
3390 /* Return error if we don't have space for new frag */
3391 if (frg_cnt >= MAX_SKB_FRAGS)
3394 if (!sk_page_frag_refill(sk, pfrag))
3397 /* copy the user data to page */
3398 copy = min_t(int, length, pfrag->size - pfrag->offset);
3400 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
3401 offset, copy, 0, skb);
3405 /* copy was successful so update the size parameters */
3406 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
3409 pfrag->offset += copy;
3410 get_page(pfrag->page);
3412 skb->truesize += copy;
3413 refcount_add(copy, &sk->sk_wmem_alloc);
3415 skb->data_len += copy;
3419 } while (length > 0);
3423 EXPORT_SYMBOL(skb_append_datato_frags);
3425 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3426 int offset, size_t size)
3428 int i = skb_shinfo(skb)->nr_frags;
3430 if (skb_can_coalesce(skb, i, page, offset)) {
3431 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3432 } else if (i < MAX_SKB_FRAGS) {
3434 skb_fill_page_desc(skb, i, page, offset, size);
3441 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3444 * skb_pull_rcsum - pull skb and update receive checksum
3445 * @skb: buffer to update
3446 * @len: length of data pulled
3448 * This function performs an skb_pull on the packet and updates
3449 * the CHECKSUM_COMPLETE checksum. It should be used on
3450 * receive path processing instead of skb_pull unless you know
3451 * that the checksum difference is zero (e.g., a valid IP header)
3452 * or you are setting ip_summed to CHECKSUM_NONE.
3454 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3456 unsigned char *data = skb->data;
3458 BUG_ON(len > skb->len);
3459 __skb_pull(skb, len);
3460 skb_postpull_rcsum(skb, data, len);
3463 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3466 * skb_segment - Perform protocol segmentation on skb.
3467 * @head_skb: buffer to segment
3468 * @features: features for the output path (see dev->features)
3470 * This function performs segmentation on the given skb. It returns
3471 * a pointer to the first in a list of new skbs for the segments.
3472 * In case of error it returns ERR_PTR(err).
3474 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3475 netdev_features_t features)
3477 struct sk_buff *segs = NULL;
3478 struct sk_buff *tail = NULL;
3479 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3480 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3481 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3482 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3483 struct sk_buff *frag_skb = head_skb;
3484 unsigned int offset = doffset;
3485 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3486 unsigned int partial_segs = 0;
3487 unsigned int headroom;
3488 unsigned int len = head_skb->len;
3491 int nfrags = skb_shinfo(head_skb)->nr_frags;
3497 __skb_push(head_skb, doffset);
3498 proto = skb_network_protocol(head_skb, &dummy);
3499 if (unlikely(!proto))
3500 return ERR_PTR(-EINVAL);
3502 sg = !!(features & NETIF_F_SG);
3503 csum = !!can_checksum_protocol(features, proto);
3505 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3506 if (!(features & NETIF_F_GSO_PARTIAL)) {
3507 struct sk_buff *iter;
3508 unsigned int frag_len;
3511 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3514 /* If we get here then all the required
3515 * GSO features except frag_list are supported.
3516 * Try to split the SKB to multiple GSO SKBs
3517 * with no frag_list.
3518 * Currently we can do that only when the buffers don't
3519 * have a linear part and all the buffers except
3520 * the last are of the same length.
3522 frag_len = list_skb->len;
3523 skb_walk_frags(head_skb, iter) {
3524 if (frag_len != iter->len && iter->next)
3526 if (skb_headlen(iter) && !iter->head_frag)
3532 if (len != frag_len)
3536 /* GSO partial only requires that we trim off any excess that
3537 * doesn't fit into an MSS sized block, so take care of that
3540 partial_segs = len / mss;
3541 if (partial_segs > 1)
3542 mss *= partial_segs;
3548 headroom = skb_headroom(head_skb);
3549 pos = skb_headlen(head_skb);
3552 struct sk_buff *nskb;
3553 skb_frag_t *nskb_frag;
3557 if (unlikely(mss == GSO_BY_FRAGS)) {
3558 len = list_skb->len;
3560 len = head_skb->len - offset;
3565 hsize = skb_headlen(head_skb) - offset;
3568 if (hsize > len || !sg)
3571 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3572 (skb_headlen(list_skb) == len || sg)) {
3573 BUG_ON(skb_headlen(list_skb) > len);
3576 nfrags = skb_shinfo(list_skb)->nr_frags;
3577 frag = skb_shinfo(list_skb)->frags;
3578 frag_skb = list_skb;
3579 pos += skb_headlen(list_skb);
3581 while (pos < offset + len) {
3582 BUG_ON(i >= nfrags);
3584 size = skb_frag_size(frag);
3585 if (pos + size > offset + len)
3593 nskb = skb_clone(list_skb, GFP_ATOMIC);
3594 list_skb = list_skb->next;
3596 if (unlikely(!nskb))
3599 if (unlikely(pskb_trim(nskb, len))) {
3604 hsize = skb_end_offset(nskb);
3605 if (skb_cow_head(nskb, doffset + headroom)) {
3610 nskb->truesize += skb_end_offset(nskb) - hsize;
3611 skb_release_head_state(nskb);
3612 __skb_push(nskb, doffset);
3614 nskb = __alloc_skb(hsize + doffset + headroom,
3615 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3618 if (unlikely(!nskb))
3621 skb_reserve(nskb, headroom);
3622 __skb_put(nskb, doffset);
3631 __copy_skb_header(nskb, head_skb);
3633 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3634 skb_reset_mac_len(nskb);
3636 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3637 nskb->data - tnl_hlen,
3638 doffset + tnl_hlen);
3640 if (nskb->len == len + doffset)
3641 goto perform_csum_check;
3644 if (!nskb->remcsum_offload)
3645 nskb->ip_summed = CHECKSUM_NONE;
3646 SKB_GSO_CB(nskb)->csum =
3647 skb_copy_and_csum_bits(head_skb, offset,
3650 SKB_GSO_CB(nskb)->csum_start =
3651 skb_headroom(nskb) + doffset;
3655 nskb_frag = skb_shinfo(nskb)->frags;
3657 skb_copy_from_linear_data_offset(head_skb, offset,
3658 skb_put(nskb, hsize), hsize);
3660 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3662 if (skb_zerocopy_clone(nskb, head_skb, GFP_ATOMIC))
3665 while (pos < offset + len) {
3667 BUG_ON(skb_headlen(list_skb));
3670 nfrags = skb_shinfo(list_skb)->nr_frags;
3671 frag = skb_shinfo(list_skb)->frags;
3672 frag_skb = list_skb;
3676 list_skb = list_skb->next;
3679 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3681 net_warn_ratelimited(
3682 "skb_segment: too many frags: %u %u\n",
3687 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3691 __skb_frag_ref(nskb_frag);
3692 size = skb_frag_size(nskb_frag);
3695 nskb_frag->page_offset += offset - pos;
3696 skb_frag_size_sub(nskb_frag, offset - pos);
3699 skb_shinfo(nskb)->nr_frags++;
3701 if (pos + size <= offset + len) {
3706 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3714 nskb->data_len = len - hsize;
3715 nskb->len += nskb->data_len;
3716 nskb->truesize += nskb->data_len;
3720 if (skb_has_shared_frag(nskb)) {
3721 err = __skb_linearize(nskb);
3725 if (!nskb->remcsum_offload)
3726 nskb->ip_summed = CHECKSUM_NONE;
3727 SKB_GSO_CB(nskb)->csum =
3728 skb_checksum(nskb, doffset,
3729 nskb->len - doffset, 0);
3730 SKB_GSO_CB(nskb)->csum_start =
3731 skb_headroom(nskb) + doffset;
3733 } while ((offset += len) < head_skb->len);
3735 /* Some callers want to get the end of the list.
3736 * Put it in segs->prev to avoid walking the list.
3737 * (see validate_xmit_skb_list() for example)
3742 struct sk_buff *iter;
3743 int type = skb_shinfo(head_skb)->gso_type;
3744 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
3746 /* Update type to add partial and then remove dodgy if set */
3747 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
3748 type &= ~SKB_GSO_DODGY;
3750 /* Update GSO info and prepare to start updating headers on
3751 * our way back down the stack of protocols.
3753 for (iter = segs; iter; iter = iter->next) {
3754 skb_shinfo(iter)->gso_size = gso_size;
3755 skb_shinfo(iter)->gso_segs = partial_segs;
3756 skb_shinfo(iter)->gso_type = type;
3757 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
3760 if (tail->len - doffset <= gso_size)
3761 skb_shinfo(tail)->gso_size = 0;
3762 else if (tail != segs)
3763 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
3766 /* Following permits correct backpressure, for protocols
3767 * using skb_set_owner_w().
3768 * Idea is to tranfert ownership from head_skb to last segment.
3770 if (head_skb->destructor == sock_wfree) {
3771 swap(tail->truesize, head_skb->truesize);
3772 swap(tail->destructor, head_skb->destructor);
3773 swap(tail->sk, head_skb->sk);
3778 kfree_skb_list(segs);
3779 return ERR_PTR(err);
3781 EXPORT_SYMBOL_GPL(skb_segment);
3783 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3785 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3786 unsigned int offset = skb_gro_offset(skb);
3787 unsigned int headlen = skb_headlen(skb);
3788 unsigned int len = skb_gro_len(skb);
3789 struct sk_buff *lp, *p = *head;
3790 unsigned int delta_truesize;
3792 if (unlikely(p->len + len >= 65536))
3795 lp = NAPI_GRO_CB(p)->last;
3796 pinfo = skb_shinfo(lp);
3798 if (headlen <= offset) {
3801 int i = skbinfo->nr_frags;
3802 int nr_frags = pinfo->nr_frags + i;
3804 if (nr_frags > MAX_SKB_FRAGS)
3808 pinfo->nr_frags = nr_frags;
3809 skbinfo->nr_frags = 0;
3811 frag = pinfo->frags + nr_frags;
3812 frag2 = skbinfo->frags + i;
3817 frag->page_offset += offset;
3818 skb_frag_size_sub(frag, offset);
3820 /* all fragments truesize : remove (head size + sk_buff) */
3821 delta_truesize = skb->truesize -
3822 SKB_TRUESIZE(skb_end_offset(skb));
3824 skb->truesize -= skb->data_len;
3825 skb->len -= skb->data_len;
3828 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3830 } else if (skb->head_frag) {
3831 int nr_frags = pinfo->nr_frags;
3832 skb_frag_t *frag = pinfo->frags + nr_frags;
3833 struct page *page = virt_to_head_page(skb->head);
3834 unsigned int first_size = headlen - offset;
3835 unsigned int first_offset;
3837 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3840 first_offset = skb->data -
3841 (unsigned char *)page_address(page) +
3844 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3846 frag->page.p = page;
3847 frag->page_offset = first_offset;
3848 skb_frag_size_set(frag, first_size);
3850 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3851 /* We dont need to clear skbinfo->nr_frags here */
3853 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3854 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3859 delta_truesize = skb->truesize;
3860 if (offset > headlen) {
3861 unsigned int eat = offset - headlen;
3863 skbinfo->frags[0].page_offset += eat;
3864 skb_frag_size_sub(&skbinfo->frags[0], eat);
3865 skb->data_len -= eat;
3870 __skb_pull(skb, offset);
3872 if (NAPI_GRO_CB(p)->last == p)
3873 skb_shinfo(p)->frag_list = skb;
3875 NAPI_GRO_CB(p)->last->next = skb;
3876 NAPI_GRO_CB(p)->last = skb;
3877 __skb_header_release(skb);
3881 NAPI_GRO_CB(p)->count++;
3883 p->truesize += delta_truesize;
3886 lp->data_len += len;
3887 lp->truesize += delta_truesize;
3890 NAPI_GRO_CB(skb)->same_flow = 1;
3893 EXPORT_SYMBOL_GPL(skb_gro_receive);
3895 void __init skb_init(void)
3897 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3898 sizeof(struct sk_buff),
3900 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3902 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3903 sizeof(struct sk_buff_fclones),
3905 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3910 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
3911 unsigned int recursion_level)
3913 int start = skb_headlen(skb);
3914 int i, copy = start - offset;
3915 struct sk_buff *frag_iter;
3918 if (unlikely(recursion_level >= 24))
3924 sg_set_buf(sg, skb->data + offset, copy);
3926 if ((len -= copy) == 0)
3931 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3934 WARN_ON(start > offset + len);
3936 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3937 if ((copy = end - offset) > 0) {
3938 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3939 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
3944 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3945 frag->page_offset+offset-start);
3954 skb_walk_frags(skb, frag_iter) {
3957 WARN_ON(start > offset + len);
3959 end = start + frag_iter->len;
3960 if ((copy = end - offset) > 0) {
3961 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
3966 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3967 copy, recursion_level + 1);
3968 if (unlikely(ret < 0))
3971 if ((len -= copy) == 0)
3982 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3983 * @skb: Socket buffer containing the buffers to be mapped
3984 * @sg: The scatter-gather list to map into
3985 * @offset: The offset into the buffer's contents to start mapping
3986 * @len: Length of buffer space to be mapped
3988 * Fill the specified scatter-gather list with mappings/pointers into a
3989 * region of the buffer space attached to a socket buffer. Returns either
3990 * the number of scatterlist items used, or -EMSGSIZE if the contents
3993 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3995 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4000 sg_mark_end(&sg[nsg - 1]);
4004 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4006 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4007 * sglist without mark the sg which contain last skb data as the end.
4008 * So the caller can mannipulate sg list as will when padding new data after
4009 * the first call without calling sg_unmark_end to expend sg list.
4011 * Scenario to use skb_to_sgvec_nomark:
4013 * 2. skb_to_sgvec_nomark(payload1)
4014 * 3. skb_to_sgvec_nomark(payload2)
4016 * This is equivalent to:
4018 * 2. skb_to_sgvec(payload1)
4020 * 4. skb_to_sgvec(payload2)
4022 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4023 * is more preferable.
4025 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4026 int offset, int len)
4028 return __skb_to_sgvec(skb, sg, offset, len, 0);
4030 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4035 * skb_cow_data - Check that a socket buffer's data buffers are writable
4036 * @skb: The socket buffer to check.
4037 * @tailbits: Amount of trailing space to be added
4038 * @trailer: Returned pointer to the skb where the @tailbits space begins
4040 * Make sure that the data buffers attached to a socket buffer are
4041 * writable. If they are not, private copies are made of the data buffers
4042 * and the socket buffer is set to use these instead.
4044 * If @tailbits is given, make sure that there is space to write @tailbits
4045 * bytes of data beyond current end of socket buffer. @trailer will be
4046 * set to point to the skb in which this space begins.
4048 * The number of scatterlist elements required to completely map the
4049 * COW'd and extended socket buffer will be returned.
4051 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4055 struct sk_buff *skb1, **skb_p;
4057 /* If skb is cloned or its head is paged, reallocate
4058 * head pulling out all the pages (pages are considered not writable
4059 * at the moment even if they are anonymous).
4061 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4062 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
4065 /* Easy case. Most of packets will go this way. */
4066 if (!skb_has_frag_list(skb)) {
4067 /* A little of trouble, not enough of space for trailer.
4068 * This should not happen, when stack is tuned to generate
4069 * good frames. OK, on miss we reallocate and reserve even more
4070 * space, 128 bytes is fair. */
4072 if (skb_tailroom(skb) < tailbits &&
4073 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4081 /* Misery. We are in troubles, going to mincer fragments... */
4084 skb_p = &skb_shinfo(skb)->frag_list;
4087 while ((skb1 = *skb_p) != NULL) {
4090 /* The fragment is partially pulled by someone,
4091 * this can happen on input. Copy it and everything
4094 if (skb_shared(skb1))
4097 /* If the skb is the last, worry about trailer. */
4099 if (skb1->next == NULL && tailbits) {
4100 if (skb_shinfo(skb1)->nr_frags ||
4101 skb_has_frag_list(skb1) ||
4102 skb_tailroom(skb1) < tailbits)
4103 ntail = tailbits + 128;
4109 skb_shinfo(skb1)->nr_frags ||
4110 skb_has_frag_list(skb1)) {
4111 struct sk_buff *skb2;
4113 /* Fuck, we are miserable poor guys... */
4115 skb2 = skb_copy(skb1, GFP_ATOMIC);
4117 skb2 = skb_copy_expand(skb1,
4121 if (unlikely(skb2 == NULL))
4125 skb_set_owner_w(skb2, skb1->sk);
4127 /* Looking around. Are we still alive?
4128 * OK, link new skb, drop old one */
4130 skb2->next = skb1->next;
4137 skb_p = &skb1->next;
4142 EXPORT_SYMBOL_GPL(skb_cow_data);
4144 static void sock_rmem_free(struct sk_buff *skb)
4146 struct sock *sk = skb->sk;
4148 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4151 static void skb_set_err_queue(struct sk_buff *skb)
4153 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4154 * So, it is safe to (mis)use it to mark skbs on the error queue.
4156 skb->pkt_type = PACKET_OUTGOING;
4157 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4161 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4163 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4165 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4166 (unsigned int)sk->sk_rcvbuf)
4171 skb->destructor = sock_rmem_free;
4172 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4173 skb_set_err_queue(skb);
4175 /* before exiting rcu section, make sure dst is refcounted */
4178 skb_queue_tail(&sk->sk_error_queue, skb);
4179 if (!sock_flag(sk, SOCK_DEAD))
4180 sk->sk_data_ready(sk);
4183 EXPORT_SYMBOL(sock_queue_err_skb);
4185 static bool is_icmp_err_skb(const struct sk_buff *skb)
4187 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4188 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4191 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4193 struct sk_buff_head *q = &sk->sk_error_queue;
4194 struct sk_buff *skb, *skb_next = NULL;
4195 bool icmp_next = false;
4196 unsigned long flags;
4198 spin_lock_irqsave(&q->lock, flags);
4199 skb = __skb_dequeue(q);
4200 if (skb && (skb_next = skb_peek(q))) {
4201 icmp_next = is_icmp_err_skb(skb_next);
4203 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin;
4205 spin_unlock_irqrestore(&q->lock, flags);
4207 if (is_icmp_err_skb(skb) && !icmp_next)
4211 sk->sk_error_report(sk);
4215 EXPORT_SYMBOL(sock_dequeue_err_skb);
4218 * skb_clone_sk - create clone of skb, and take reference to socket
4219 * @skb: the skb to clone
4221 * This function creates a clone of a buffer that holds a reference on
4222 * sk_refcnt. Buffers created via this function are meant to be
4223 * returned using sock_queue_err_skb, or free via kfree_skb.
4225 * When passing buffers allocated with this function to sock_queue_err_skb
4226 * it is necessary to wrap the call with sock_hold/sock_put in order to
4227 * prevent the socket from being released prior to being enqueued on
4228 * the sk_error_queue.
4230 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4232 struct sock *sk = skb->sk;
4233 struct sk_buff *clone;
4235 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4238 clone = skb_clone(skb, GFP_ATOMIC);
4245 clone->destructor = sock_efree;
4249 EXPORT_SYMBOL(skb_clone_sk);
4251 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4256 struct sock_exterr_skb *serr;
4259 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4261 serr = SKB_EXT_ERR(skb);
4262 memset(serr, 0, sizeof(*serr));
4263 serr->ee.ee_errno = ENOMSG;
4264 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4265 serr->ee.ee_info = tstype;
4266 serr->opt_stats = opt_stats;
4267 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4268 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4269 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4270 if (sk->sk_protocol == IPPROTO_TCP &&
4271 sk->sk_type == SOCK_STREAM)
4272 serr->ee.ee_data -= sk->sk_tskey;
4275 err = sock_queue_err_skb(sk, skb);
4281 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4285 if (likely(sysctl_tstamp_allow_data || tsonly))
4288 read_lock_bh(&sk->sk_callback_lock);
4289 ret = sk->sk_socket && sk->sk_socket->file &&
4290 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4291 read_unlock_bh(&sk->sk_callback_lock);
4295 void skb_complete_tx_timestamp(struct sk_buff *skb,
4296 struct skb_shared_hwtstamps *hwtstamps)
4298 struct sock *sk = skb->sk;
4300 if (!skb_may_tx_timestamp(sk, false))
4303 /* Take a reference to prevent skb_orphan() from freeing the socket,
4304 * but only if the socket refcount is not zero.
4306 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4307 *skb_hwtstamps(skb) = *hwtstamps;
4308 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4312 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4314 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4315 struct skb_shared_hwtstamps *hwtstamps,
4316 struct sock *sk, int tstype)
4318 struct sk_buff *skb;
4319 bool tsonly, opt_stats = false;
4324 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4325 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4328 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4329 if (!skb_may_tx_timestamp(sk, tsonly))
4334 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4335 sk->sk_protocol == IPPROTO_TCP &&
4336 sk->sk_type == SOCK_STREAM) {
4337 skb = tcp_get_timestamping_opt_stats(sk);
4341 skb = alloc_skb(0, GFP_ATOMIC);
4343 skb = skb_clone(orig_skb, GFP_ATOMIC);
4349 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4351 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4355 *skb_hwtstamps(skb) = *hwtstamps;
4357 skb->tstamp = ktime_get_real();
4359 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4361 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4363 void skb_tstamp_tx(struct sk_buff *orig_skb,
4364 struct skb_shared_hwtstamps *hwtstamps)
4366 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4369 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4371 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4373 struct sock *sk = skb->sk;
4374 struct sock_exterr_skb *serr;
4377 skb->wifi_acked_valid = 1;
4378 skb->wifi_acked = acked;
4380 serr = SKB_EXT_ERR(skb);
4381 memset(serr, 0, sizeof(*serr));
4382 serr->ee.ee_errno = ENOMSG;
4383 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4385 /* Take a reference to prevent skb_orphan() from freeing the socket,
4386 * but only if the socket refcount is not zero.
4388 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4389 err = sock_queue_err_skb(sk, skb);
4395 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4398 * skb_partial_csum_set - set up and verify partial csum values for packet
4399 * @skb: the skb to set
4400 * @start: the number of bytes after skb->data to start checksumming.
4401 * @off: the offset from start to place the checksum.
4403 * For untrusted partially-checksummed packets, we need to make sure the values
4404 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4406 * This function checks and sets those values and skb->ip_summed: if this
4407 * returns false you should drop the packet.
4409 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4411 if (unlikely(start > skb_headlen(skb)) ||
4412 unlikely((int)start + off > skb_headlen(skb) - 2)) {
4413 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
4414 start, off, skb_headlen(skb));
4417 skb->ip_summed = CHECKSUM_PARTIAL;
4418 skb->csum_start = skb_headroom(skb) + start;
4419 skb->csum_offset = off;
4420 skb_set_transport_header(skb, start);
4423 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4425 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4428 if (skb_headlen(skb) >= len)
4431 /* If we need to pullup then pullup to the max, so we
4432 * won't need to do it again.
4437 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4440 if (skb_headlen(skb) < len)
4446 #define MAX_TCP_HDR_LEN (15 * 4)
4448 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4449 typeof(IPPROTO_IP) proto,
4456 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4457 off + MAX_TCP_HDR_LEN);
4458 if (!err && !skb_partial_csum_set(skb, off,
4459 offsetof(struct tcphdr,
4462 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4465 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4466 off + sizeof(struct udphdr));
4467 if (!err && !skb_partial_csum_set(skb, off,
4468 offsetof(struct udphdr,
4471 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4474 return ERR_PTR(-EPROTO);
4477 /* This value should be large enough to cover a tagged ethernet header plus
4478 * maximally sized IP and TCP or UDP headers.
4480 #define MAX_IP_HDR_LEN 128
4482 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4491 err = skb_maybe_pull_tail(skb,
4492 sizeof(struct iphdr),
4497 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4500 off = ip_hdrlen(skb);
4507 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4509 return PTR_ERR(csum);
4512 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4515 ip_hdr(skb)->protocol, 0);
4522 /* This value should be large enough to cover a tagged ethernet header plus
4523 * an IPv6 header, all options, and a maximal TCP or UDP header.
4525 #define MAX_IPV6_HDR_LEN 256
4527 #define OPT_HDR(type, skb, off) \
4528 (type *)(skb_network_header(skb) + (off))
4530 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4543 off = sizeof(struct ipv6hdr);
4545 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4549 nexthdr = ipv6_hdr(skb)->nexthdr;
4551 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4552 while (off <= len && !done) {
4554 case IPPROTO_DSTOPTS:
4555 case IPPROTO_HOPOPTS:
4556 case IPPROTO_ROUTING: {
4557 struct ipv6_opt_hdr *hp;
4559 err = skb_maybe_pull_tail(skb,
4561 sizeof(struct ipv6_opt_hdr),
4566 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4567 nexthdr = hp->nexthdr;
4568 off += ipv6_optlen(hp);
4572 struct ip_auth_hdr *hp;
4574 err = skb_maybe_pull_tail(skb,
4576 sizeof(struct ip_auth_hdr),
4581 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4582 nexthdr = hp->nexthdr;
4583 off += ipv6_authlen(hp);
4586 case IPPROTO_FRAGMENT: {
4587 struct frag_hdr *hp;
4589 err = skb_maybe_pull_tail(skb,
4591 sizeof(struct frag_hdr),
4596 hp = OPT_HDR(struct frag_hdr, skb, off);
4598 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4601 nexthdr = hp->nexthdr;
4602 off += sizeof(struct frag_hdr);
4613 if (!done || fragment)
4616 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4618 return PTR_ERR(csum);
4621 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4622 &ipv6_hdr(skb)->daddr,
4623 skb->len - off, nexthdr, 0);
4631 * skb_checksum_setup - set up partial checksum offset
4632 * @skb: the skb to set up
4633 * @recalculate: if true the pseudo-header checksum will be recalculated
4635 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4639 switch (skb->protocol) {
4640 case htons(ETH_P_IP):
4641 err = skb_checksum_setup_ipv4(skb, recalculate);
4644 case htons(ETH_P_IPV6):
4645 err = skb_checksum_setup_ipv6(skb, recalculate);
4655 EXPORT_SYMBOL(skb_checksum_setup);
4658 * skb_checksum_maybe_trim - maybe trims the given skb
4659 * @skb: the skb to check
4660 * @transport_len: the data length beyond the network header
4662 * Checks whether the given skb has data beyond the given transport length.
4663 * If so, returns a cloned skb trimmed to this transport length.
4664 * Otherwise returns the provided skb. Returns NULL in error cases
4665 * (e.g. transport_len exceeds skb length or out-of-memory).
4667 * Caller needs to set the skb transport header and free any returned skb if it
4668 * differs from the provided skb.
4670 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4671 unsigned int transport_len)
4673 struct sk_buff *skb_chk;
4674 unsigned int len = skb_transport_offset(skb) + transport_len;
4679 else if (skb->len == len)
4682 skb_chk = skb_clone(skb, GFP_ATOMIC);
4686 ret = pskb_trim_rcsum(skb_chk, len);
4696 * skb_checksum_trimmed - validate checksum of an skb
4697 * @skb: the skb to check
4698 * @transport_len: the data length beyond the network header
4699 * @skb_chkf: checksum function to use
4701 * Applies the given checksum function skb_chkf to the provided skb.
4702 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4704 * If the skb has data beyond the given transport length, then a
4705 * trimmed & cloned skb is checked and returned.
4707 * Caller needs to set the skb transport header and free any returned skb if it
4708 * differs from the provided skb.
4710 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4711 unsigned int transport_len,
4712 __sum16(*skb_chkf)(struct sk_buff *skb))
4714 struct sk_buff *skb_chk;
4715 unsigned int offset = skb_transport_offset(skb);
4718 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4722 if (!pskb_may_pull(skb_chk, offset))
4725 skb_pull_rcsum(skb_chk, offset);
4726 ret = skb_chkf(skb_chk);
4727 skb_push_rcsum(skb_chk, offset);
4735 if (skb_chk && skb_chk != skb)
4741 EXPORT_SYMBOL(skb_checksum_trimmed);
4743 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4745 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4748 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4750 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4753 skb_release_head_state(skb);
4754 kmem_cache_free(skbuff_head_cache, skb);
4759 EXPORT_SYMBOL(kfree_skb_partial);
4762 * skb_try_coalesce - try to merge skb to prior one
4764 * @from: buffer to add
4765 * @fragstolen: pointer to boolean
4766 * @delta_truesize: how much more was allocated than was requested
4768 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4769 bool *fragstolen, int *delta_truesize)
4771 int i, delta, len = from->len;
4773 *fragstolen = false;
4778 if (len <= skb_tailroom(to)) {
4780 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4781 *delta_truesize = 0;
4785 if (skb_has_frag_list(to) || skb_has_frag_list(from))
4787 if (skb_zcopy(to) || skb_zcopy(from))
4790 if (skb_headlen(from) != 0) {
4792 unsigned int offset;
4794 if (skb_shinfo(to)->nr_frags +
4795 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4798 if (skb_head_is_locked(from))
4801 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4803 page = virt_to_head_page(from->head);
4804 offset = from->data - (unsigned char *)page_address(page);
4806 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4807 page, offset, skb_headlen(from));
4810 if (skb_shinfo(to)->nr_frags +
4811 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4814 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4817 WARN_ON_ONCE(delta < len);
4819 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4820 skb_shinfo(from)->frags,
4821 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4822 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4824 if (!skb_cloned(from))
4825 skb_shinfo(from)->nr_frags = 0;
4827 /* if the skb is not cloned this does nothing
4828 * since we set nr_frags to 0.
4830 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4831 skb_frag_ref(from, i);
4833 to->truesize += delta;
4835 to->data_len += len;
4837 *delta_truesize = delta;
4840 EXPORT_SYMBOL(skb_try_coalesce);
4843 * skb_scrub_packet - scrub an skb
4845 * @skb: buffer to clean
4846 * @xnet: packet is crossing netns
4848 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4849 * into/from a tunnel. Some information have to be cleared during these
4851 * skb_scrub_packet can also be used to clean a skb before injecting it in
4852 * another namespace (@xnet == true). We have to clear all information in the
4853 * skb that could impact namespace isolation.
4855 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4858 skb->pkt_type = PACKET_HOST;
4864 nf_reset_trace(skb);
4872 EXPORT_SYMBOL_GPL(skb_scrub_packet);
4875 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4879 * skb_gso_transport_seglen is used to determine the real size of the
4880 * individual segments, including Layer4 headers (TCP/UDP).
4882 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4884 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4886 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4887 unsigned int thlen = 0;
4889 if (skb->encapsulation) {
4890 thlen = skb_inner_transport_header(skb) -
4891 skb_transport_header(skb);
4893 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4894 thlen += inner_tcp_hdrlen(skb);
4895 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4896 thlen = tcp_hdrlen(skb);
4897 } else if (unlikely(shinfo->gso_type & SKB_GSO_SCTP)) {
4898 thlen = sizeof(struct sctphdr);
4900 /* UFO sets gso_size to the size of the fragmentation
4901 * payload, i.e. the size of the L4 (UDP) header is already
4904 return thlen + shinfo->gso_size;
4906 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4909 * skb_gso_validate_mtu - Return in case such skb fits a given MTU
4912 * @mtu: MTU to validate against
4914 * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU
4917 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu)
4919 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4920 const struct sk_buff *iter;
4923 hlen = skb_gso_network_seglen(skb);
4925 if (shinfo->gso_size != GSO_BY_FRAGS)
4928 /* Undo this so we can re-use header sizes */
4929 hlen -= GSO_BY_FRAGS;
4931 skb_walk_frags(skb, iter) {
4932 if (hlen + skb_headlen(iter) > mtu)
4938 EXPORT_SYMBOL_GPL(skb_gso_validate_mtu);
4940 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4942 if (skb_cow(skb, skb_headroom(skb)) < 0) {
4947 memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN,
4949 skb->mac_header += VLAN_HLEN;
4953 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4955 struct vlan_hdr *vhdr;
4958 if (unlikely(skb_vlan_tag_present(skb))) {
4959 /* vlan_tci is already set-up so leave this for another time */
4963 skb = skb_share_check(skb, GFP_ATOMIC);
4967 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4970 vhdr = (struct vlan_hdr *)skb->data;
4971 vlan_tci = ntohs(vhdr->h_vlan_TCI);
4972 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4974 skb_pull_rcsum(skb, VLAN_HLEN);
4975 vlan_set_encap_proto(skb, vhdr);
4977 skb = skb_reorder_vlan_header(skb);
4981 skb_reset_network_header(skb);
4982 skb_reset_transport_header(skb);
4983 skb_reset_mac_len(skb);
4991 EXPORT_SYMBOL(skb_vlan_untag);
4993 int skb_ensure_writable(struct sk_buff *skb, int write_len)
4995 if (!pskb_may_pull(skb, write_len))
4998 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5001 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5003 EXPORT_SYMBOL(skb_ensure_writable);
5005 /* remove VLAN header from packet and update csum accordingly.
5006 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5008 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5010 struct vlan_hdr *vhdr;
5011 int offset = skb->data - skb_mac_header(skb);
5014 if (WARN_ONCE(offset,
5015 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5020 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5024 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5026 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5027 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5029 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5030 __skb_pull(skb, VLAN_HLEN);
5032 vlan_set_encap_proto(skb, vhdr);
5033 skb->mac_header += VLAN_HLEN;
5035 if (skb_network_offset(skb) < ETH_HLEN)
5036 skb_set_network_header(skb, ETH_HLEN);
5038 skb_reset_mac_len(skb);
5042 EXPORT_SYMBOL(__skb_vlan_pop);
5044 /* Pop a vlan tag either from hwaccel or from payload.
5045 * Expects skb->data at mac header.
5047 int skb_vlan_pop(struct sk_buff *skb)
5053 if (likely(skb_vlan_tag_present(skb))) {
5056 if (unlikely(!eth_type_vlan(skb->protocol)))
5059 err = __skb_vlan_pop(skb, &vlan_tci);
5063 /* move next vlan tag to hw accel tag */
5064 if (likely(!eth_type_vlan(skb->protocol)))
5067 vlan_proto = skb->protocol;
5068 err = __skb_vlan_pop(skb, &vlan_tci);
5072 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5075 EXPORT_SYMBOL(skb_vlan_pop);
5077 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5078 * Expects skb->data at mac header.
5080 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5082 if (skb_vlan_tag_present(skb)) {
5083 int offset = skb->data - skb_mac_header(skb);
5086 if (WARN_ONCE(offset,
5087 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5092 err = __vlan_insert_tag(skb, skb->vlan_proto,
5093 skb_vlan_tag_get(skb));
5097 skb->protocol = skb->vlan_proto;
5098 skb->mac_len += VLAN_HLEN;
5100 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5102 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5105 EXPORT_SYMBOL(skb_vlan_push);
5108 * alloc_skb_with_frags - allocate skb with page frags
5110 * @header_len: size of linear part
5111 * @data_len: needed length in frags
5112 * @max_page_order: max page order desired.
5113 * @errcode: pointer to error code if any
5114 * @gfp_mask: allocation mask
5116 * This can be used to allocate a paged skb, given a maximal order for frags.
5118 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5119 unsigned long data_len,
5124 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5125 unsigned long chunk;
5126 struct sk_buff *skb;
5131 *errcode = -EMSGSIZE;
5132 /* Note this test could be relaxed, if we succeed to allocate
5133 * high order pages...
5135 if (npages > MAX_SKB_FRAGS)
5138 gfp_head = gfp_mask;
5139 if (gfp_head & __GFP_DIRECT_RECLAIM)
5140 gfp_head |= __GFP_RETRY_MAYFAIL;
5142 *errcode = -ENOBUFS;
5143 skb = alloc_skb(header_len, gfp_head);
5147 skb->truesize += npages << PAGE_SHIFT;
5149 for (i = 0; npages > 0; i++) {
5150 int order = max_page_order;
5153 if (npages >= 1 << order) {
5154 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5161 /* Do not retry other high order allocations */
5167 page = alloc_page(gfp_mask);
5171 chunk = min_t(unsigned long, data_len,
5172 PAGE_SIZE << order);
5173 skb_fill_page_desc(skb, i, page, 0, chunk);
5175 npages -= 1 << order;
5183 EXPORT_SYMBOL(alloc_skb_with_frags);
5185 /* carve out the first off bytes from skb when off < headlen */
5186 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5187 const int headlen, gfp_t gfp_mask)
5190 int size = skb_end_offset(skb);
5191 int new_hlen = headlen - off;
5194 size = SKB_DATA_ALIGN(size);
5196 if (skb_pfmemalloc(skb))
5197 gfp_mask |= __GFP_MEMALLOC;
5198 data = kmalloc_reserve(size +
5199 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5200 gfp_mask, NUMA_NO_NODE, NULL);
5204 size = SKB_WITH_OVERHEAD(ksize(data));
5206 /* Copy real data, and all frags */
5207 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5210 memcpy((struct skb_shared_info *)(data + size),
5212 offsetof(struct skb_shared_info,
5213 frags[skb_shinfo(skb)->nr_frags]));
5214 if (skb_cloned(skb)) {
5215 /* drop the old head gracefully */
5216 if (skb_orphan_frags(skb, gfp_mask)) {
5220 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5221 skb_frag_ref(skb, i);
5222 if (skb_has_frag_list(skb))
5223 skb_clone_fraglist(skb);
5224 skb_release_data(skb);
5226 /* we can reuse existing recount- all we did was
5235 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5238 skb->end = skb->head + size;
5240 skb_set_tail_pointer(skb, skb_headlen(skb));
5241 skb_headers_offset_update(skb, 0);
5245 atomic_set(&skb_shinfo(skb)->dataref, 1);
5250 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
5252 /* carve out the first eat bytes from skb's frag_list. May recurse into
5255 static int pskb_carve_frag_list(struct sk_buff *skb,
5256 struct skb_shared_info *shinfo, int eat,
5259 struct sk_buff *list = shinfo->frag_list;
5260 struct sk_buff *clone = NULL;
5261 struct sk_buff *insp = NULL;
5265 pr_err("Not enough bytes to eat. Want %d\n", eat);
5268 if (list->len <= eat) {
5269 /* Eaten as whole. */
5274 /* Eaten partially. */
5275 if (skb_shared(list)) {
5276 clone = skb_clone(list, gfp_mask);
5282 /* This may be pulled without problems. */
5285 if (pskb_carve(list, eat, gfp_mask) < 0) {
5293 /* Free pulled out fragments. */
5294 while ((list = shinfo->frag_list) != insp) {
5295 shinfo->frag_list = list->next;
5298 /* And insert new clone at head. */
5301 shinfo->frag_list = clone;
5306 /* carve off first len bytes from skb. Split line (off) is in the
5307 * non-linear part of skb
5309 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
5310 int pos, gfp_t gfp_mask)
5313 int size = skb_end_offset(skb);
5315 const int nfrags = skb_shinfo(skb)->nr_frags;
5316 struct skb_shared_info *shinfo;
5318 size = SKB_DATA_ALIGN(size);
5320 if (skb_pfmemalloc(skb))
5321 gfp_mask |= __GFP_MEMALLOC;
5322 data = kmalloc_reserve(size +
5323 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5324 gfp_mask, NUMA_NO_NODE, NULL);
5328 size = SKB_WITH_OVERHEAD(ksize(data));
5330 memcpy((struct skb_shared_info *)(data + size),
5331 skb_shinfo(skb), offsetof(struct skb_shared_info,
5332 frags[skb_shinfo(skb)->nr_frags]));
5333 if (skb_orphan_frags(skb, gfp_mask)) {
5337 shinfo = (struct skb_shared_info *)(data + size);
5338 for (i = 0; i < nfrags; i++) {
5339 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
5341 if (pos + fsize > off) {
5342 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
5346 * We have two variants in this case:
5347 * 1. Move all the frag to the second
5348 * part, if it is possible. F.e.
5349 * this approach is mandatory for TUX,
5350 * where splitting is expensive.
5351 * 2. Split is accurately. We make this.
5353 shinfo->frags[0].page_offset += off - pos;
5354 skb_frag_size_sub(&shinfo->frags[0], off - pos);
5356 skb_frag_ref(skb, i);
5361 shinfo->nr_frags = k;
5362 if (skb_has_frag_list(skb))
5363 skb_clone_fraglist(skb);
5366 /* split line is in frag list */
5367 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
5369 skb_release_data(skb);
5374 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5377 skb->end = skb->head + size;
5379 skb_reset_tail_pointer(skb);
5380 skb_headers_offset_update(skb, 0);
5385 skb->data_len = skb->len;
5386 atomic_set(&skb_shinfo(skb)->dataref, 1);
5390 /* remove len bytes from the beginning of the skb */
5391 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
5393 int headlen = skb_headlen(skb);
5396 return pskb_carve_inside_header(skb, len, headlen, gfp);
5398 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
5401 /* Extract to_copy bytes starting at off from skb, and return this in
5404 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
5405 int to_copy, gfp_t gfp)
5407 struct sk_buff *clone = skb_clone(skb, gfp);
5412 if (pskb_carve(clone, off, gfp) < 0 ||
5413 pskb_trim(clone, to_copy)) {
5419 EXPORT_SYMBOL(pskb_extract);
5422 * skb_condense - try to get rid of fragments/frag_list if possible
5425 * Can be used to save memory before skb is added to a busy queue.
5426 * If packet has bytes in frags and enough tail room in skb->head,
5427 * pull all of them, so that we can free the frags right now and adjust
5430 * We do not reallocate skb->head thus can not fail.
5431 * Caller must re-evaluate skb->truesize if needed.
5433 void skb_condense(struct sk_buff *skb)
5435 if (skb->data_len) {
5436 if (skb->data_len > skb->end - skb->tail ||
5440 /* Nice, we can free page frag(s) right now */
5441 __pskb_pull_tail(skb, skb->data_len);
5443 /* At this point, skb->truesize might be over estimated,
5444 * because skb had a fragment, and fragments do not tell
5446 * When we pulled its content into skb->head, fragment
5447 * was freed, but __pskb_pull_tail() could not possibly
5448 * adjust skb->truesize, not knowing the frag truesize.
5450 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));