2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
8 * Alan Cox : Fixed the worst of the load
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
36 * The functions in this file will not compile correctly with gcc 2.4.x
39 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
44 #include <linux/kmemcheck.h>
46 #include <linux/interrupt.h>
48 #include <linux/inet.h>
49 #include <linux/slab.h>
50 #include <linux/tcp.h>
51 #include <linux/udp.h>
52 #include <linux/netdevice.h>
53 #ifdef CONFIG_NET_CLS_ACT
54 #include <net/pkt_sched.h>
56 #include <linux/string.h>
57 #include <linux/skbuff.h>
58 #include <linux/splice.h>
59 #include <linux/cache.h>
60 #include <linux/rtnetlink.h>
61 #include <linux/init.h>
62 #include <linux/scatterlist.h>
63 #include <linux/errqueue.h>
64 #include <linux/prefetch.h>
65 #include <linux/if_vlan.h>
67 #include <net/protocol.h>
70 #include <net/checksum.h>
71 #include <net/ip6_checksum.h>
74 #include <asm/uaccess.h>
75 #include <trace/events/skb.h>
76 #include <linux/highmem.h>
78 struct kmem_cache *skbuff_head_cache __read_mostly;
79 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
82 * skb_panic - private function for out-of-line support
86 * @msg: skb_over_panic or skb_under_panic
88 * Out-of-line support for skb_put() and skb_push().
89 * Called via the wrapper skb_over_panic() or skb_under_panic().
90 * Keep out of line to prevent kernel bloat.
91 * __builtin_return_address is not used because it is not always reliable.
93 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
96 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
97 msg, addr, skb->len, sz, skb->head, skb->data,
98 (unsigned long)skb->tail, (unsigned long)skb->end,
99 skb->dev ? skb->dev->name : "<NULL>");
103 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
105 skb_panic(skb, sz, addr, __func__);
108 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
110 skb_panic(skb, sz, addr, __func__);
114 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
115 * the caller if emergency pfmemalloc reserves are being used. If it is and
116 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
117 * may be used. Otherwise, the packet data may be discarded until enough
120 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
121 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
123 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
124 unsigned long ip, bool *pfmemalloc)
127 bool ret_pfmemalloc = false;
130 * Try a regular allocation, when that fails and we're not entitled
131 * to the reserves, fail.
133 obj = kmalloc_node_track_caller(size,
134 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
136 if (obj || !(gfp_pfmemalloc_allowed(flags)))
139 /* Try again but now we are using pfmemalloc reserves */
140 ret_pfmemalloc = true;
141 obj = kmalloc_node_track_caller(size, flags, node);
145 *pfmemalloc = ret_pfmemalloc;
150 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
151 * 'private' fields and also do memory statistics to find all the
156 struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
161 skb = kmem_cache_alloc_node(skbuff_head_cache,
162 gfp_mask & ~__GFP_DMA, node);
167 * Only clear those fields we need to clear, not those that we will
168 * actually initialise below. Hence, don't put any more fields after
169 * the tail pointer in struct sk_buff!
171 memset(skb, 0, offsetof(struct sk_buff, tail));
173 skb->truesize = sizeof(struct sk_buff);
174 atomic_set(&skb->users, 1);
176 skb->mac_header = (typeof(skb->mac_header))~0U;
182 * __alloc_skb - allocate a network buffer
183 * @size: size to allocate
184 * @gfp_mask: allocation mask
185 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
186 * instead of head cache and allocate a cloned (child) skb.
187 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
188 * allocations in case the data is required for writeback
189 * @node: numa node to allocate memory on
191 * Allocate a new &sk_buff. The returned buffer has no headroom and a
192 * tail room of at least size bytes. The object has a reference count
193 * of one. The return is the buffer. On a failure the return is %NULL.
195 * Buffers may only be allocated from interrupts using a @gfp_mask of
198 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
201 struct kmem_cache *cache;
202 struct skb_shared_info *shinfo;
207 cache = (flags & SKB_ALLOC_FCLONE)
208 ? skbuff_fclone_cache : skbuff_head_cache;
210 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
211 gfp_mask |= __GFP_MEMALLOC;
214 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
219 /* We do our best to align skb_shared_info on a separate cache
220 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
221 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
222 * Both skb->head and skb_shared_info are cache line aligned.
224 size = SKB_DATA_ALIGN(size);
225 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
226 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
229 /* kmalloc(size) might give us more room than requested.
230 * Put skb_shared_info exactly at the end of allocated zone,
231 * to allow max possible filling before reallocation.
233 size = SKB_WITH_OVERHEAD(ksize(data));
234 prefetchw(data + size);
237 * Only clear those fields we need to clear, not those that we will
238 * actually initialise below. Hence, don't put any more fields after
239 * the tail pointer in struct sk_buff!
241 memset(skb, 0, offsetof(struct sk_buff, tail));
242 /* Account for allocated memory : skb + skb->head */
243 skb->truesize = SKB_TRUESIZE(size);
244 skb->pfmemalloc = pfmemalloc;
245 atomic_set(&skb->users, 1);
248 skb_reset_tail_pointer(skb);
249 skb->end = skb->tail + size;
250 skb->mac_header = (typeof(skb->mac_header))~0U;
251 skb->transport_header = (typeof(skb->transport_header))~0U;
253 /* make sure we initialize shinfo sequentially */
254 shinfo = skb_shinfo(skb);
255 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
256 atomic_set(&shinfo->dataref, 1);
257 kmemcheck_annotate_variable(shinfo->destructor_arg);
259 if (flags & SKB_ALLOC_FCLONE) {
260 struct sk_buff_fclones *fclones;
262 fclones = container_of(skb, struct sk_buff_fclones, skb1);
264 kmemcheck_annotate_bitfield(&fclones->skb2, flags1);
265 skb->fclone = SKB_FCLONE_ORIG;
266 atomic_set(&fclones->fclone_ref, 1);
268 fclones->skb2.fclone = SKB_FCLONE_FREE;
269 fclones->skb2.pfmemalloc = pfmemalloc;
274 kmem_cache_free(cache, skb);
278 EXPORT_SYMBOL(__alloc_skb);
281 * build_skb - build a network buffer
282 * @data: data buffer provided by caller
283 * @frag_size: size of fragment, or 0 if head was kmalloced
285 * Allocate a new &sk_buff. Caller provides space holding head and
286 * skb_shared_info. @data must have been allocated by kmalloc() only if
287 * @frag_size is 0, otherwise data should come from the page allocator.
288 * The return is the new skb buffer.
289 * On a failure the return is %NULL, and @data is not freed.
291 * Before IO, driver allocates only data buffer where NIC put incoming frame
292 * Driver should add room at head (NET_SKB_PAD) and
293 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
294 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
295 * before giving packet to stack.
296 * RX rings only contains data buffers, not full skbs.
298 struct sk_buff *build_skb(void *data, unsigned int frag_size)
300 struct skb_shared_info *shinfo;
302 unsigned int size = frag_size ? : ksize(data);
304 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
308 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
310 memset(skb, 0, offsetof(struct sk_buff, tail));
311 skb->truesize = SKB_TRUESIZE(size);
312 skb->head_frag = frag_size != 0;
313 atomic_set(&skb->users, 1);
316 skb_reset_tail_pointer(skb);
317 skb->end = skb->tail + size;
318 skb->mac_header = (typeof(skb->mac_header))~0U;
319 skb->transport_header = (typeof(skb->transport_header))~0U;
321 /* make sure we initialize shinfo sequentially */
322 shinfo = skb_shinfo(skb);
323 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
324 atomic_set(&shinfo->dataref, 1);
325 kmemcheck_annotate_variable(shinfo->destructor_arg);
329 EXPORT_SYMBOL(build_skb);
331 struct netdev_alloc_cache {
332 struct page_frag frag;
333 /* we maintain a pagecount bias, so that we dont dirty cache line
334 * containing page->_count every time we allocate a fragment.
336 unsigned int pagecnt_bias;
338 static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
340 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
342 struct netdev_alloc_cache *nc;
347 local_irq_save(flags);
348 nc = this_cpu_ptr(&netdev_alloc_cache);
349 if (unlikely(!nc->frag.page)) {
351 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
352 gfp_t gfp = gfp_mask;
355 gfp |= __GFP_COMP | __GFP_NOWARN;
356 nc->frag.page = alloc_pages(gfp, order);
357 if (likely(nc->frag.page))
362 nc->frag.size = PAGE_SIZE << order;
363 /* Even if we own the page, we do not use atomic_set().
364 * This would break get_page_unless_zero() users.
366 atomic_add(NETDEV_PAGECNT_MAX_BIAS - 1,
367 &nc->frag.page->_count);
368 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
372 if (nc->frag.offset + fragsz > nc->frag.size) {
373 if (atomic_read(&nc->frag.page->_count) != nc->pagecnt_bias) {
374 if (!atomic_sub_and_test(nc->pagecnt_bias,
375 &nc->frag.page->_count))
377 /* OK, page count is 0, we can safely set it */
378 atomic_set(&nc->frag.page->_count,
379 NETDEV_PAGECNT_MAX_BIAS);
381 atomic_add(NETDEV_PAGECNT_MAX_BIAS - nc->pagecnt_bias,
382 &nc->frag.page->_count);
384 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
388 data = page_address(nc->frag.page) + nc->frag.offset;
389 nc->frag.offset += fragsz;
392 local_irq_restore(flags);
397 * netdev_alloc_frag - allocate a page fragment
398 * @fragsz: fragment size
400 * Allocates a frag from a page for receive buffer.
401 * Uses GFP_ATOMIC allocations.
403 void *netdev_alloc_frag(unsigned int fragsz)
405 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
407 EXPORT_SYMBOL(netdev_alloc_frag);
410 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
411 * @dev: network device to receive on
412 * @length: length to allocate
413 * @gfp_mask: get_free_pages mask, passed to alloc_skb
415 * Allocate a new &sk_buff and assign it a usage count of one. The
416 * buffer has unspecified headroom built in. Users should allocate
417 * the headroom they think they need without accounting for the
418 * built in space. The built in space is used for optimisations.
420 * %NULL is returned if there is no free memory.
422 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
423 unsigned int length, gfp_t gfp_mask)
425 struct sk_buff *skb = NULL;
426 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
427 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
429 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
432 if (sk_memalloc_socks())
433 gfp_mask |= __GFP_MEMALLOC;
435 data = __netdev_alloc_frag(fragsz, gfp_mask);
438 skb = build_skb(data, fragsz);
440 put_page(virt_to_head_page(data));
443 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
444 SKB_ALLOC_RX, NUMA_NO_NODE);
447 skb_reserve(skb, NET_SKB_PAD);
452 EXPORT_SYMBOL(__netdev_alloc_skb);
454 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
455 int size, unsigned int truesize)
457 skb_fill_page_desc(skb, i, page, off, size);
459 skb->data_len += size;
460 skb->truesize += truesize;
462 EXPORT_SYMBOL(skb_add_rx_frag);
464 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
465 unsigned int truesize)
467 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
469 skb_frag_size_add(frag, size);
471 skb->data_len += size;
472 skb->truesize += truesize;
474 EXPORT_SYMBOL(skb_coalesce_rx_frag);
476 static void skb_drop_list(struct sk_buff **listp)
478 kfree_skb_list(*listp);
482 static inline void skb_drop_fraglist(struct sk_buff *skb)
484 skb_drop_list(&skb_shinfo(skb)->frag_list);
487 static void skb_clone_fraglist(struct sk_buff *skb)
489 struct sk_buff *list;
491 skb_walk_frags(skb, list)
495 static void skb_free_head(struct sk_buff *skb)
498 put_page(virt_to_head_page(skb->head));
503 static void skb_release_data(struct sk_buff *skb)
505 struct skb_shared_info *shinfo = skb_shinfo(skb);
509 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
513 for (i = 0; i < shinfo->nr_frags; i++)
514 __skb_frag_unref(&shinfo->frags[i]);
517 * If skb buf is from userspace, we need to notify the caller
518 * the lower device DMA has done;
520 if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) {
521 struct ubuf_info *uarg;
523 uarg = shinfo->destructor_arg;
525 uarg->callback(uarg, true);
528 if (shinfo->frag_list)
529 kfree_skb_list(shinfo->frag_list);
535 * Free an skbuff by memory without cleaning the state.
537 static void kfree_skbmem(struct sk_buff *skb)
539 struct sk_buff_fclones *fclones;
541 switch (skb->fclone) {
542 case SKB_FCLONE_UNAVAILABLE:
543 kmem_cache_free(skbuff_head_cache, skb);
546 case SKB_FCLONE_ORIG:
547 fclones = container_of(skb, struct sk_buff_fclones, skb1);
548 if (atomic_dec_and_test(&fclones->fclone_ref))
549 kmem_cache_free(skbuff_fclone_cache, fclones);
552 case SKB_FCLONE_CLONE:
553 fclones = container_of(skb, struct sk_buff_fclones, skb2);
555 /* The clone portion is available for
556 * fast-cloning again.
558 skb->fclone = SKB_FCLONE_FREE;
560 if (atomic_dec_and_test(&fclones->fclone_ref))
561 kmem_cache_free(skbuff_fclone_cache, fclones);
566 static void skb_release_head_state(struct sk_buff *skb)
570 secpath_put(skb->sp);
572 if (skb->destructor) {
574 skb->destructor(skb);
576 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
577 nf_conntrack_put(skb->nfct);
579 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
580 nf_bridge_put(skb->nf_bridge);
582 /* XXX: IS this still necessary? - JHS */
583 #ifdef CONFIG_NET_SCHED
585 #ifdef CONFIG_NET_CLS_ACT
591 /* Free everything but the sk_buff shell. */
592 static void skb_release_all(struct sk_buff *skb)
594 skb_release_head_state(skb);
595 if (likely(skb->head))
596 skb_release_data(skb);
600 * __kfree_skb - private function
603 * Free an sk_buff. Release anything attached to the buffer.
604 * Clean the state. This is an internal helper function. Users should
605 * always call kfree_skb
608 void __kfree_skb(struct sk_buff *skb)
610 skb_release_all(skb);
613 EXPORT_SYMBOL(__kfree_skb);
616 * kfree_skb - free an sk_buff
617 * @skb: buffer to free
619 * Drop a reference to the buffer and free it if the usage count has
622 void kfree_skb(struct sk_buff *skb)
626 if (likely(atomic_read(&skb->users) == 1))
628 else if (likely(!atomic_dec_and_test(&skb->users)))
630 trace_kfree_skb(skb, __builtin_return_address(0));
633 EXPORT_SYMBOL(kfree_skb);
635 void kfree_skb_list(struct sk_buff *segs)
638 struct sk_buff *next = segs->next;
644 EXPORT_SYMBOL(kfree_skb_list);
647 * skb_tx_error - report an sk_buff xmit error
648 * @skb: buffer that triggered an error
650 * Report xmit error if a device callback is tracking this skb.
651 * skb must be freed afterwards.
653 void skb_tx_error(struct sk_buff *skb)
655 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
656 struct ubuf_info *uarg;
658 uarg = skb_shinfo(skb)->destructor_arg;
660 uarg->callback(uarg, false);
661 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
664 EXPORT_SYMBOL(skb_tx_error);
667 * consume_skb - free an skbuff
668 * @skb: buffer to free
670 * Drop a ref to the buffer and free it if the usage count has hit zero
671 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
672 * is being dropped after a failure and notes that
674 void consume_skb(struct sk_buff *skb)
678 if (likely(atomic_read(&skb->users) == 1))
680 else if (likely(!atomic_dec_and_test(&skb->users)))
682 trace_consume_skb(skb);
685 EXPORT_SYMBOL(consume_skb);
687 /* Make sure a field is enclosed inside headers_start/headers_end section */
688 #define CHECK_SKB_FIELD(field) \
689 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
690 offsetof(struct sk_buff, headers_start)); \
691 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
692 offsetof(struct sk_buff, headers_end)); \
694 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
696 new->tstamp = old->tstamp;
697 /* We do not copy old->sk */
699 memcpy(new->cb, old->cb, sizeof(old->cb));
700 skb_dst_copy(new, old);
702 new->sp = secpath_get(old->sp);
704 __nf_copy(new, old, false);
706 /* Note : this field could be in headers_start/headers_end section
707 * It is not yet because we do not want to have a 16 bit hole
709 new->queue_mapping = old->queue_mapping;
711 memcpy(&new->headers_start, &old->headers_start,
712 offsetof(struct sk_buff, headers_end) -
713 offsetof(struct sk_buff, headers_start));
714 CHECK_SKB_FIELD(protocol);
715 CHECK_SKB_FIELD(csum);
716 CHECK_SKB_FIELD(hash);
717 CHECK_SKB_FIELD(priority);
718 CHECK_SKB_FIELD(skb_iif);
719 CHECK_SKB_FIELD(vlan_proto);
720 CHECK_SKB_FIELD(vlan_tci);
721 CHECK_SKB_FIELD(transport_header);
722 CHECK_SKB_FIELD(network_header);
723 CHECK_SKB_FIELD(mac_header);
724 CHECK_SKB_FIELD(inner_protocol);
725 CHECK_SKB_FIELD(inner_transport_header);
726 CHECK_SKB_FIELD(inner_network_header);
727 CHECK_SKB_FIELD(inner_mac_header);
728 CHECK_SKB_FIELD(mark);
729 #ifdef CONFIG_NETWORK_SECMARK
730 CHECK_SKB_FIELD(secmark);
732 #ifdef CONFIG_NET_RX_BUSY_POLL
733 CHECK_SKB_FIELD(napi_id);
735 #ifdef CONFIG_NET_SCHED
736 CHECK_SKB_FIELD(tc_index);
737 #ifdef CONFIG_NET_CLS_ACT
738 CHECK_SKB_FIELD(tc_verd);
745 * You should not add any new code to this function. Add it to
746 * __copy_skb_header above instead.
748 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
750 #define C(x) n->x = skb->x
752 n->next = n->prev = NULL;
754 __copy_skb_header(n, skb);
759 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
762 n->destructor = NULL;
769 atomic_set(&n->users, 1);
771 atomic_inc(&(skb_shinfo(skb)->dataref));
779 * skb_morph - morph one skb into another
780 * @dst: the skb to receive the contents
781 * @src: the skb to supply the contents
783 * This is identical to skb_clone except that the target skb is
784 * supplied by the user.
786 * The target skb is returned upon exit.
788 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
790 skb_release_all(dst);
791 return __skb_clone(dst, src);
793 EXPORT_SYMBOL_GPL(skb_morph);
796 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
797 * @skb: the skb to modify
798 * @gfp_mask: allocation priority
800 * This must be called on SKBTX_DEV_ZEROCOPY skb.
801 * It will copy all frags into kernel and drop the reference
802 * to userspace pages.
804 * If this function is called from an interrupt gfp_mask() must be
807 * Returns 0 on success or a negative error code on failure
808 * to allocate kernel memory to copy to.
810 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
813 int num_frags = skb_shinfo(skb)->nr_frags;
814 struct page *page, *head = NULL;
815 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
817 for (i = 0; i < num_frags; i++) {
819 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
821 page = alloc_page(gfp_mask);
824 struct page *next = (struct page *)page_private(head);
830 vaddr = kmap_atomic(skb_frag_page(f));
831 memcpy(page_address(page),
832 vaddr + f->page_offset, skb_frag_size(f));
833 kunmap_atomic(vaddr);
834 set_page_private(page, (unsigned long)head);
838 /* skb frags release userspace buffers */
839 for (i = 0; i < num_frags; i++)
840 skb_frag_unref(skb, i);
842 uarg->callback(uarg, false);
844 /* skb frags point to kernel buffers */
845 for (i = num_frags - 1; i >= 0; i--) {
846 __skb_fill_page_desc(skb, i, head, 0,
847 skb_shinfo(skb)->frags[i].size);
848 head = (struct page *)page_private(head);
851 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
854 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
857 * skb_clone - duplicate an sk_buff
858 * @skb: buffer to clone
859 * @gfp_mask: allocation priority
861 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
862 * copies share the same packet data but not structure. The new
863 * buffer has a reference count of 1. If the allocation fails the
864 * function returns %NULL otherwise the new buffer is returned.
866 * If this function is called from an interrupt gfp_mask() must be
870 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
872 struct sk_buff_fclones *fclones = container_of(skb,
873 struct sk_buff_fclones,
875 struct sk_buff *n = &fclones->skb2;
877 if (skb_orphan_frags(skb, gfp_mask))
880 if (skb->fclone == SKB_FCLONE_ORIG &&
881 n->fclone == SKB_FCLONE_FREE) {
882 n->fclone = SKB_FCLONE_CLONE;
883 atomic_inc(&fclones->fclone_ref);
885 if (skb_pfmemalloc(skb))
886 gfp_mask |= __GFP_MEMALLOC;
888 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
892 kmemcheck_annotate_bitfield(n, flags1);
893 n->fclone = SKB_FCLONE_UNAVAILABLE;
896 return __skb_clone(n, skb);
898 EXPORT_SYMBOL(skb_clone);
900 static void skb_headers_offset_update(struct sk_buff *skb, int off)
902 /* Only adjust this if it actually is csum_start rather than csum */
903 if (skb->ip_summed == CHECKSUM_PARTIAL)
904 skb->csum_start += off;
905 /* {transport,network,mac}_header and tail are relative to skb->head */
906 skb->transport_header += off;
907 skb->network_header += off;
908 if (skb_mac_header_was_set(skb))
909 skb->mac_header += off;
910 skb->inner_transport_header += off;
911 skb->inner_network_header += off;
912 skb->inner_mac_header += off;
915 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
917 __copy_skb_header(new, old);
919 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
920 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
921 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
924 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
926 if (skb_pfmemalloc(skb))
932 * skb_copy - create private copy of an sk_buff
933 * @skb: buffer to copy
934 * @gfp_mask: allocation priority
936 * Make a copy of both an &sk_buff and its data. This is used when the
937 * caller wishes to modify the data and needs a private copy of the
938 * data to alter. Returns %NULL on failure or the pointer to the buffer
939 * on success. The returned buffer has a reference count of 1.
941 * As by-product this function converts non-linear &sk_buff to linear
942 * one, so that &sk_buff becomes completely private and caller is allowed
943 * to modify all the data of returned buffer. This means that this
944 * function is not recommended for use in circumstances when only
945 * header is going to be modified. Use pskb_copy() instead.
948 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
950 int headerlen = skb_headroom(skb);
951 unsigned int size = skb_end_offset(skb) + skb->data_len;
952 struct sk_buff *n = __alloc_skb(size, gfp_mask,
953 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
958 /* Set the data pointer */
959 skb_reserve(n, headerlen);
960 /* Set the tail pointer and length */
961 skb_put(n, skb->len);
963 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
966 copy_skb_header(n, skb);
969 EXPORT_SYMBOL(skb_copy);
972 * __pskb_copy_fclone - create copy of an sk_buff with private head.
973 * @skb: buffer to copy
974 * @headroom: headroom of new skb
975 * @gfp_mask: allocation priority
976 * @fclone: if true allocate the copy of the skb from the fclone
977 * cache instead of the head cache; it is recommended to set this
978 * to true for the cases where the copy will likely be cloned
980 * Make a copy of both an &sk_buff and part of its data, located
981 * in header. Fragmented data remain shared. This is used when
982 * the caller wishes to modify only header of &sk_buff and needs
983 * private copy of the header to alter. Returns %NULL on failure
984 * or the pointer to the buffer on success.
985 * The returned buffer has a reference count of 1.
988 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
989 gfp_t gfp_mask, bool fclone)
991 unsigned int size = skb_headlen(skb) + headroom;
992 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
993 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
998 /* Set the data pointer */
999 skb_reserve(n, headroom);
1000 /* Set the tail pointer and length */
1001 skb_put(n, skb_headlen(skb));
1002 /* Copy the bytes */
1003 skb_copy_from_linear_data(skb, n->data, n->len);
1005 n->truesize += skb->data_len;
1006 n->data_len = skb->data_len;
1009 if (skb_shinfo(skb)->nr_frags) {
1012 if (skb_orphan_frags(skb, gfp_mask)) {
1017 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1018 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1019 skb_frag_ref(skb, i);
1021 skb_shinfo(n)->nr_frags = i;
1024 if (skb_has_frag_list(skb)) {
1025 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1026 skb_clone_fraglist(n);
1029 copy_skb_header(n, skb);
1033 EXPORT_SYMBOL(__pskb_copy_fclone);
1036 * pskb_expand_head - reallocate header of &sk_buff
1037 * @skb: buffer to reallocate
1038 * @nhead: room to add at head
1039 * @ntail: room to add at tail
1040 * @gfp_mask: allocation priority
1042 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1043 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1044 * reference count of 1. Returns zero in the case of success or error,
1045 * if expansion failed. In the last case, &sk_buff is not changed.
1047 * All the pointers pointing into skb header may change and must be
1048 * reloaded after call to this function.
1051 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1056 int size = nhead + skb_end_offset(skb) + ntail;
1061 if (skb_shared(skb))
1064 size = SKB_DATA_ALIGN(size);
1066 if (skb_pfmemalloc(skb))
1067 gfp_mask |= __GFP_MEMALLOC;
1068 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1069 gfp_mask, NUMA_NO_NODE, NULL);
1072 size = SKB_WITH_OVERHEAD(ksize(data));
1074 /* Copy only real data... and, alas, header. This should be
1075 * optimized for the cases when header is void.
1077 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1079 memcpy((struct skb_shared_info *)(data + size),
1081 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1084 * if shinfo is shared we must drop the old head gracefully, but if it
1085 * is not we can just drop the old head and let the existing refcount
1086 * be since all we did is relocate the values
1088 if (skb_cloned(skb)) {
1089 /* copy this zero copy skb frags */
1090 if (skb_orphan_frags(skb, gfp_mask))
1092 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1093 skb_frag_ref(skb, i);
1095 if (skb_has_frag_list(skb))
1096 skb_clone_fraglist(skb);
1098 skb_release_data(skb);
1102 off = (data + nhead) - skb->head;
1107 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1111 skb->end = skb->head + size;
1114 skb_headers_offset_update(skb, nhead);
1118 atomic_set(&skb_shinfo(skb)->dataref, 1);
1126 EXPORT_SYMBOL(pskb_expand_head);
1128 /* Make private copy of skb with writable head and some headroom */
1130 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1132 struct sk_buff *skb2;
1133 int delta = headroom - skb_headroom(skb);
1136 skb2 = pskb_copy(skb, GFP_ATOMIC);
1138 skb2 = skb_clone(skb, GFP_ATOMIC);
1139 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1147 EXPORT_SYMBOL(skb_realloc_headroom);
1150 * skb_copy_expand - copy and expand sk_buff
1151 * @skb: buffer to copy
1152 * @newheadroom: new free bytes at head
1153 * @newtailroom: new free bytes at tail
1154 * @gfp_mask: allocation priority
1156 * Make a copy of both an &sk_buff and its data and while doing so
1157 * allocate additional space.
1159 * This is used when the caller wishes to modify the data and needs a
1160 * private copy of the data to alter as well as more space for new fields.
1161 * Returns %NULL on failure or the pointer to the buffer
1162 * on success. The returned buffer has a reference count of 1.
1164 * You must pass %GFP_ATOMIC as the allocation priority if this function
1165 * is called from an interrupt.
1167 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1168 int newheadroom, int newtailroom,
1172 * Allocate the copy buffer
1174 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1175 gfp_mask, skb_alloc_rx_flag(skb),
1177 int oldheadroom = skb_headroom(skb);
1178 int head_copy_len, head_copy_off;
1183 skb_reserve(n, newheadroom);
1185 /* Set the tail pointer and length */
1186 skb_put(n, skb->len);
1188 head_copy_len = oldheadroom;
1190 if (newheadroom <= head_copy_len)
1191 head_copy_len = newheadroom;
1193 head_copy_off = newheadroom - head_copy_len;
1195 /* Copy the linear header and data. */
1196 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1197 skb->len + head_copy_len))
1200 copy_skb_header(n, skb);
1202 skb_headers_offset_update(n, newheadroom - oldheadroom);
1206 EXPORT_SYMBOL(skb_copy_expand);
1209 * skb_pad - zero pad the tail of an skb
1210 * @skb: buffer to pad
1211 * @pad: space to pad
1213 * Ensure that a buffer is followed by a padding area that is zero
1214 * filled. Used by network drivers which may DMA or transfer data
1215 * beyond the buffer end onto the wire.
1217 * May return error in out of memory cases. The skb is freed on error.
1220 int skb_pad(struct sk_buff *skb, int pad)
1225 /* If the skbuff is non linear tailroom is always zero.. */
1226 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1227 memset(skb->data+skb->len, 0, pad);
1231 ntail = skb->data_len + pad - (skb->end - skb->tail);
1232 if (likely(skb_cloned(skb) || ntail > 0)) {
1233 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1238 /* FIXME: The use of this function with non-linear skb's really needs
1241 err = skb_linearize(skb);
1245 memset(skb->data + skb->len, 0, pad);
1252 EXPORT_SYMBOL(skb_pad);
1255 * pskb_put - add data to the tail of a potentially fragmented buffer
1256 * @skb: start of the buffer to use
1257 * @tail: tail fragment of the buffer to use
1258 * @len: amount of data to add
1260 * This function extends the used data area of the potentially
1261 * fragmented buffer. @tail must be the last fragment of @skb -- or
1262 * @skb itself. If this would exceed the total buffer size the kernel
1263 * will panic. A pointer to the first byte of the extra data is
1267 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1270 skb->data_len += len;
1273 return skb_put(tail, len);
1275 EXPORT_SYMBOL_GPL(pskb_put);
1278 * skb_put - add data to a buffer
1279 * @skb: buffer to use
1280 * @len: amount of data to add
1282 * This function extends the used data area of the buffer. If this would
1283 * exceed the total buffer size the kernel will panic. A pointer to the
1284 * first byte of the extra data is returned.
1286 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1288 unsigned char *tmp = skb_tail_pointer(skb);
1289 SKB_LINEAR_ASSERT(skb);
1292 if (unlikely(skb->tail > skb->end))
1293 skb_over_panic(skb, len, __builtin_return_address(0));
1296 EXPORT_SYMBOL(skb_put);
1299 * skb_push - add data to the start of a buffer
1300 * @skb: buffer to use
1301 * @len: amount of data to add
1303 * This function extends the used data area of the buffer at the buffer
1304 * start. If this would exceed the total buffer headroom the kernel will
1305 * panic. A pointer to the first byte of the extra data is returned.
1307 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1311 if (unlikely(skb->data<skb->head))
1312 skb_under_panic(skb, len, __builtin_return_address(0));
1315 EXPORT_SYMBOL(skb_push);
1318 * skb_pull - remove data from the start of a buffer
1319 * @skb: buffer to use
1320 * @len: amount of data to remove
1322 * This function removes data from the start of a buffer, returning
1323 * the memory to the headroom. A pointer to the next data in the buffer
1324 * is returned. Once the data has been pulled future pushes will overwrite
1327 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1329 return skb_pull_inline(skb, len);
1331 EXPORT_SYMBOL(skb_pull);
1334 * skb_trim - remove end from a buffer
1335 * @skb: buffer to alter
1338 * Cut the length of a buffer down by removing data from the tail. If
1339 * the buffer is already under the length specified it is not modified.
1340 * The skb must be linear.
1342 void skb_trim(struct sk_buff *skb, unsigned int len)
1345 __skb_trim(skb, len);
1347 EXPORT_SYMBOL(skb_trim);
1349 /* Trims skb to length len. It can change skb pointers.
1352 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1354 struct sk_buff **fragp;
1355 struct sk_buff *frag;
1356 int offset = skb_headlen(skb);
1357 int nfrags = skb_shinfo(skb)->nr_frags;
1361 if (skb_cloned(skb) &&
1362 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1369 for (; i < nfrags; i++) {
1370 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1377 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1380 skb_shinfo(skb)->nr_frags = i;
1382 for (; i < nfrags; i++)
1383 skb_frag_unref(skb, i);
1385 if (skb_has_frag_list(skb))
1386 skb_drop_fraglist(skb);
1390 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1391 fragp = &frag->next) {
1392 int end = offset + frag->len;
1394 if (skb_shared(frag)) {
1395 struct sk_buff *nfrag;
1397 nfrag = skb_clone(frag, GFP_ATOMIC);
1398 if (unlikely(!nfrag))
1401 nfrag->next = frag->next;
1413 unlikely((err = pskb_trim(frag, len - offset))))
1417 skb_drop_list(&frag->next);
1422 if (len > skb_headlen(skb)) {
1423 skb->data_len -= skb->len - len;
1428 skb_set_tail_pointer(skb, len);
1433 EXPORT_SYMBOL(___pskb_trim);
1436 * __pskb_pull_tail - advance tail of skb header
1437 * @skb: buffer to reallocate
1438 * @delta: number of bytes to advance tail
1440 * The function makes a sense only on a fragmented &sk_buff,
1441 * it expands header moving its tail forward and copying necessary
1442 * data from fragmented part.
1444 * &sk_buff MUST have reference count of 1.
1446 * Returns %NULL (and &sk_buff does not change) if pull failed
1447 * or value of new tail of skb in the case of success.
1449 * All the pointers pointing into skb header may change and must be
1450 * reloaded after call to this function.
1453 /* Moves tail of skb head forward, copying data from fragmented part,
1454 * when it is necessary.
1455 * 1. It may fail due to malloc failure.
1456 * 2. It may change skb pointers.
1458 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1460 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1462 /* If skb has not enough free space at tail, get new one
1463 * plus 128 bytes for future expansions. If we have enough
1464 * room at tail, reallocate without expansion only if skb is cloned.
1466 int i, k, eat = (skb->tail + delta) - skb->end;
1468 if (eat > 0 || skb_cloned(skb)) {
1469 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1474 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1477 /* Optimization: no fragments, no reasons to preestimate
1478 * size of pulled pages. Superb.
1480 if (!skb_has_frag_list(skb))
1483 /* Estimate size of pulled pages. */
1485 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1486 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1493 /* If we need update frag list, we are in troubles.
1494 * Certainly, it possible to add an offset to skb data,
1495 * but taking into account that pulling is expected to
1496 * be very rare operation, it is worth to fight against
1497 * further bloating skb head and crucify ourselves here instead.
1498 * Pure masohism, indeed. 8)8)
1501 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1502 struct sk_buff *clone = NULL;
1503 struct sk_buff *insp = NULL;
1508 if (list->len <= eat) {
1509 /* Eaten as whole. */
1514 /* Eaten partially. */
1516 if (skb_shared(list)) {
1517 /* Sucks! We need to fork list. :-( */
1518 clone = skb_clone(list, GFP_ATOMIC);
1524 /* This may be pulled without
1528 if (!pskb_pull(list, eat)) {
1536 /* Free pulled out fragments. */
1537 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1538 skb_shinfo(skb)->frag_list = list->next;
1541 /* And insert new clone at head. */
1544 skb_shinfo(skb)->frag_list = clone;
1547 /* Success! Now we may commit changes to skb data. */
1552 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1553 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1556 skb_frag_unref(skb, i);
1559 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1561 skb_shinfo(skb)->frags[k].page_offset += eat;
1562 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1568 skb_shinfo(skb)->nr_frags = k;
1571 skb->data_len -= delta;
1573 return skb_tail_pointer(skb);
1575 EXPORT_SYMBOL(__pskb_pull_tail);
1578 * skb_copy_bits - copy bits from skb to kernel buffer
1580 * @offset: offset in source
1581 * @to: destination buffer
1582 * @len: number of bytes to copy
1584 * Copy the specified number of bytes from the source skb to the
1585 * destination buffer.
1588 * If its prototype is ever changed,
1589 * check arch/{*}/net/{*}.S files,
1590 * since it is called from BPF assembly code.
1592 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1594 int start = skb_headlen(skb);
1595 struct sk_buff *frag_iter;
1598 if (offset > (int)skb->len - len)
1602 if ((copy = start - offset) > 0) {
1605 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1606 if ((len -= copy) == 0)
1612 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1614 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1616 WARN_ON(start > offset + len);
1618 end = start + skb_frag_size(f);
1619 if ((copy = end - offset) > 0) {
1625 vaddr = kmap_atomic(skb_frag_page(f));
1627 vaddr + f->page_offset + offset - start,
1629 kunmap_atomic(vaddr);
1631 if ((len -= copy) == 0)
1639 skb_walk_frags(skb, frag_iter) {
1642 WARN_ON(start > offset + len);
1644 end = start + frag_iter->len;
1645 if ((copy = end - offset) > 0) {
1648 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1650 if ((len -= copy) == 0)
1664 EXPORT_SYMBOL(skb_copy_bits);
1667 * Callback from splice_to_pipe(), if we need to release some pages
1668 * at the end of the spd in case we error'ed out in filling the pipe.
1670 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1672 put_page(spd->pages[i]);
1675 static struct page *linear_to_page(struct page *page, unsigned int *len,
1676 unsigned int *offset,
1679 struct page_frag *pfrag = sk_page_frag(sk);
1681 if (!sk_page_frag_refill(sk, pfrag))
1684 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1686 memcpy(page_address(pfrag->page) + pfrag->offset,
1687 page_address(page) + *offset, *len);
1688 *offset = pfrag->offset;
1689 pfrag->offset += *len;
1694 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1696 unsigned int offset)
1698 return spd->nr_pages &&
1699 spd->pages[spd->nr_pages - 1] == page &&
1700 (spd->partial[spd->nr_pages - 1].offset +
1701 spd->partial[spd->nr_pages - 1].len == offset);
1705 * Fill page/offset/length into spd, if it can hold more pages.
1707 static bool spd_fill_page(struct splice_pipe_desc *spd,
1708 struct pipe_inode_info *pipe, struct page *page,
1709 unsigned int *len, unsigned int offset,
1713 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1717 page = linear_to_page(page, len, &offset, sk);
1721 if (spd_can_coalesce(spd, page, offset)) {
1722 spd->partial[spd->nr_pages - 1].len += *len;
1726 spd->pages[spd->nr_pages] = page;
1727 spd->partial[spd->nr_pages].len = *len;
1728 spd->partial[spd->nr_pages].offset = offset;
1734 static bool __splice_segment(struct page *page, unsigned int poff,
1735 unsigned int plen, unsigned int *off,
1737 struct splice_pipe_desc *spd, bool linear,
1739 struct pipe_inode_info *pipe)
1744 /* skip this segment if already processed */
1750 /* ignore any bits we already processed */
1756 unsigned int flen = min(*len, plen);
1758 if (spd_fill_page(spd, pipe, page, &flen, poff,
1764 } while (*len && plen);
1770 * Map linear and fragment data from the skb to spd. It reports true if the
1771 * pipe is full or if we already spliced the requested length.
1773 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1774 unsigned int *offset, unsigned int *len,
1775 struct splice_pipe_desc *spd, struct sock *sk)
1779 /* map the linear part :
1780 * If skb->head_frag is set, this 'linear' part is backed by a
1781 * fragment, and if the head is not shared with any clones then
1782 * we can avoid a copy since we own the head portion of this page.
1784 if (__splice_segment(virt_to_page(skb->data),
1785 (unsigned long) skb->data & (PAGE_SIZE - 1),
1788 skb_head_is_locked(skb),
1793 * then map the fragments
1795 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1796 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1798 if (__splice_segment(skb_frag_page(f),
1799 f->page_offset, skb_frag_size(f),
1800 offset, len, spd, false, sk, pipe))
1808 * Map data from the skb to a pipe. Should handle both the linear part,
1809 * the fragments, and the frag list. It does NOT handle frag lists within
1810 * the frag list, if such a thing exists. We'd probably need to recurse to
1811 * handle that cleanly.
1813 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1814 struct pipe_inode_info *pipe, unsigned int tlen,
1817 struct partial_page partial[MAX_SKB_FRAGS];
1818 struct page *pages[MAX_SKB_FRAGS];
1819 struct splice_pipe_desc spd = {
1822 .nr_pages_max = MAX_SKB_FRAGS,
1824 .ops = &nosteal_pipe_buf_ops,
1825 .spd_release = sock_spd_release,
1827 struct sk_buff *frag_iter;
1828 struct sock *sk = skb->sk;
1832 * __skb_splice_bits() only fails if the output has no room left,
1833 * so no point in going over the frag_list for the error case.
1835 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1841 * now see if we have a frag_list to map
1843 skb_walk_frags(skb, frag_iter) {
1846 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1853 * Drop the socket lock, otherwise we have reverse
1854 * locking dependencies between sk_lock and i_mutex
1855 * here as compared to sendfile(). We enter here
1856 * with the socket lock held, and splice_to_pipe() will
1857 * grab the pipe inode lock. For sendfile() emulation,
1858 * we call into ->sendpage() with the i_mutex lock held
1859 * and networking will grab the socket lock.
1862 ret = splice_to_pipe(pipe, &spd);
1870 * skb_store_bits - store bits from kernel buffer to skb
1871 * @skb: destination buffer
1872 * @offset: offset in destination
1873 * @from: source buffer
1874 * @len: number of bytes to copy
1876 * Copy the specified number of bytes from the source buffer to the
1877 * destination skb. This function handles all the messy bits of
1878 * traversing fragment lists and such.
1881 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1883 int start = skb_headlen(skb);
1884 struct sk_buff *frag_iter;
1887 if (offset > (int)skb->len - len)
1890 if ((copy = start - offset) > 0) {
1893 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1894 if ((len -= copy) == 0)
1900 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1901 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1904 WARN_ON(start > offset + len);
1906 end = start + skb_frag_size(frag);
1907 if ((copy = end - offset) > 0) {
1913 vaddr = kmap_atomic(skb_frag_page(frag));
1914 memcpy(vaddr + frag->page_offset + offset - start,
1916 kunmap_atomic(vaddr);
1918 if ((len -= copy) == 0)
1926 skb_walk_frags(skb, frag_iter) {
1929 WARN_ON(start > offset + len);
1931 end = start + frag_iter->len;
1932 if ((copy = end - offset) > 0) {
1935 if (skb_store_bits(frag_iter, offset - start,
1938 if ((len -= copy) == 0)
1951 EXPORT_SYMBOL(skb_store_bits);
1953 /* Checksum skb data. */
1954 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
1955 __wsum csum, const struct skb_checksum_ops *ops)
1957 int start = skb_headlen(skb);
1958 int i, copy = start - offset;
1959 struct sk_buff *frag_iter;
1962 /* Checksum header. */
1966 csum = ops->update(skb->data + offset, copy, csum);
1967 if ((len -= copy) == 0)
1973 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1975 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1977 WARN_ON(start > offset + len);
1979 end = start + skb_frag_size(frag);
1980 if ((copy = end - offset) > 0) {
1986 vaddr = kmap_atomic(skb_frag_page(frag));
1987 csum2 = ops->update(vaddr + frag->page_offset +
1988 offset - start, copy, 0);
1989 kunmap_atomic(vaddr);
1990 csum = ops->combine(csum, csum2, pos, copy);
1999 skb_walk_frags(skb, frag_iter) {
2002 WARN_ON(start > offset + len);
2004 end = start + frag_iter->len;
2005 if ((copy = end - offset) > 0) {
2009 csum2 = __skb_checksum(frag_iter, offset - start,
2011 csum = ops->combine(csum, csum2, pos, copy);
2012 if ((len -= copy) == 0)
2023 EXPORT_SYMBOL(__skb_checksum);
2025 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2026 int len, __wsum csum)
2028 const struct skb_checksum_ops ops = {
2029 .update = csum_partial_ext,
2030 .combine = csum_block_add_ext,
2033 return __skb_checksum(skb, offset, len, csum, &ops);
2035 EXPORT_SYMBOL(skb_checksum);
2037 /* Both of above in one bottle. */
2039 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2040 u8 *to, int len, __wsum csum)
2042 int start = skb_headlen(skb);
2043 int i, copy = start - offset;
2044 struct sk_buff *frag_iter;
2051 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2053 if ((len -= copy) == 0)
2060 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2063 WARN_ON(start > offset + len);
2065 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2066 if ((copy = end - offset) > 0) {
2069 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2073 vaddr = kmap_atomic(skb_frag_page(frag));
2074 csum2 = csum_partial_copy_nocheck(vaddr +
2078 kunmap_atomic(vaddr);
2079 csum = csum_block_add(csum, csum2, pos);
2089 skb_walk_frags(skb, frag_iter) {
2093 WARN_ON(start > offset + len);
2095 end = start + frag_iter->len;
2096 if ((copy = end - offset) > 0) {
2099 csum2 = skb_copy_and_csum_bits(frag_iter,
2102 csum = csum_block_add(csum, csum2, pos);
2103 if ((len -= copy) == 0)
2114 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2117 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2118 * @from: source buffer
2120 * Calculates the amount of linear headroom needed in the 'to' skb passed
2121 * into skb_zerocopy().
2124 skb_zerocopy_headlen(const struct sk_buff *from)
2126 unsigned int hlen = 0;
2128 if (!from->head_frag ||
2129 skb_headlen(from) < L1_CACHE_BYTES ||
2130 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2131 hlen = skb_headlen(from);
2133 if (skb_has_frag_list(from))
2138 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2141 * skb_zerocopy - Zero copy skb to skb
2142 * @to: destination buffer
2143 * @from: source buffer
2144 * @len: number of bytes to copy from source buffer
2145 * @hlen: size of linear headroom in destination buffer
2147 * Copies up to `len` bytes from `from` to `to` by creating references
2148 * to the frags in the source buffer.
2150 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2151 * headroom in the `to` buffer.
2154 * 0: everything is OK
2155 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2156 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2159 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2162 int plen = 0; /* length of skb->head fragment */
2165 unsigned int offset;
2167 BUG_ON(!from->head_frag && !hlen);
2169 /* dont bother with small payloads */
2170 if (len <= skb_tailroom(to))
2171 return skb_copy_bits(from, 0, skb_put(to, len), len);
2174 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2179 plen = min_t(int, skb_headlen(from), len);
2181 page = virt_to_head_page(from->head);
2182 offset = from->data - (unsigned char *)page_address(page);
2183 __skb_fill_page_desc(to, 0, page, offset, plen);
2190 to->truesize += len + plen;
2191 to->len += len + plen;
2192 to->data_len += len + plen;
2194 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2199 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2202 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2203 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2204 len -= skb_shinfo(to)->frags[j].size;
2205 skb_frag_ref(to, j);
2208 skb_shinfo(to)->nr_frags = j;
2212 EXPORT_SYMBOL_GPL(skb_zerocopy);
2214 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2219 if (skb->ip_summed == CHECKSUM_PARTIAL)
2220 csstart = skb_checksum_start_offset(skb);
2222 csstart = skb_headlen(skb);
2224 BUG_ON(csstart > skb_headlen(skb));
2226 skb_copy_from_linear_data(skb, to, csstart);
2229 if (csstart != skb->len)
2230 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2231 skb->len - csstart, 0);
2233 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2234 long csstuff = csstart + skb->csum_offset;
2236 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2239 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2242 * skb_dequeue - remove from the head of the queue
2243 * @list: list to dequeue from
2245 * Remove the head of the list. The list lock is taken so the function
2246 * may be used safely with other locking list functions. The head item is
2247 * returned or %NULL if the list is empty.
2250 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2252 unsigned long flags;
2253 struct sk_buff *result;
2255 spin_lock_irqsave(&list->lock, flags);
2256 result = __skb_dequeue(list);
2257 spin_unlock_irqrestore(&list->lock, flags);
2260 EXPORT_SYMBOL(skb_dequeue);
2263 * skb_dequeue_tail - remove from the tail of the queue
2264 * @list: list to dequeue from
2266 * Remove the tail of the list. The list lock is taken so the function
2267 * may be used safely with other locking list functions. The tail item is
2268 * returned or %NULL if the list is empty.
2270 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2272 unsigned long flags;
2273 struct sk_buff *result;
2275 spin_lock_irqsave(&list->lock, flags);
2276 result = __skb_dequeue_tail(list);
2277 spin_unlock_irqrestore(&list->lock, flags);
2280 EXPORT_SYMBOL(skb_dequeue_tail);
2283 * skb_queue_purge - empty a list
2284 * @list: list to empty
2286 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2287 * the list and one reference dropped. This function takes the list
2288 * lock and is atomic with respect to other list locking functions.
2290 void skb_queue_purge(struct sk_buff_head *list)
2292 struct sk_buff *skb;
2293 while ((skb = skb_dequeue(list)) != NULL)
2296 EXPORT_SYMBOL(skb_queue_purge);
2299 * skb_queue_head - queue a buffer at the list head
2300 * @list: list to use
2301 * @newsk: buffer to queue
2303 * Queue a buffer at the start of the list. This function takes the
2304 * list lock and can be used safely with other locking &sk_buff functions
2307 * A buffer cannot be placed on two lists at the same time.
2309 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2311 unsigned long flags;
2313 spin_lock_irqsave(&list->lock, flags);
2314 __skb_queue_head(list, newsk);
2315 spin_unlock_irqrestore(&list->lock, flags);
2317 EXPORT_SYMBOL(skb_queue_head);
2320 * skb_queue_tail - queue a buffer at the list tail
2321 * @list: list to use
2322 * @newsk: buffer to queue
2324 * Queue a buffer at the tail of the list. This function takes the
2325 * list lock and can be used safely with other locking &sk_buff functions
2328 * A buffer cannot be placed on two lists at the same time.
2330 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2332 unsigned long flags;
2334 spin_lock_irqsave(&list->lock, flags);
2335 __skb_queue_tail(list, newsk);
2336 spin_unlock_irqrestore(&list->lock, flags);
2338 EXPORT_SYMBOL(skb_queue_tail);
2341 * skb_unlink - remove a buffer from a list
2342 * @skb: buffer to remove
2343 * @list: list to use
2345 * Remove a packet from a list. The list locks are taken and this
2346 * function is atomic with respect to other list locked calls
2348 * You must know what list the SKB is on.
2350 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2352 unsigned long flags;
2354 spin_lock_irqsave(&list->lock, flags);
2355 __skb_unlink(skb, list);
2356 spin_unlock_irqrestore(&list->lock, flags);
2358 EXPORT_SYMBOL(skb_unlink);
2361 * skb_append - append a buffer
2362 * @old: buffer to insert after
2363 * @newsk: buffer to insert
2364 * @list: list to use
2366 * Place a packet after a given packet in a list. The list locks are taken
2367 * and this function is atomic with respect to other list locked calls.
2368 * A buffer cannot be placed on two lists at the same time.
2370 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2372 unsigned long flags;
2374 spin_lock_irqsave(&list->lock, flags);
2375 __skb_queue_after(list, old, newsk);
2376 spin_unlock_irqrestore(&list->lock, flags);
2378 EXPORT_SYMBOL(skb_append);
2381 * skb_insert - insert a buffer
2382 * @old: buffer to insert before
2383 * @newsk: buffer to insert
2384 * @list: list to use
2386 * Place a packet before a given packet in a list. The list locks are
2387 * taken and this function is atomic with respect to other list locked
2390 * A buffer cannot be placed on two lists at the same time.
2392 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2394 unsigned long flags;
2396 spin_lock_irqsave(&list->lock, flags);
2397 __skb_insert(newsk, old->prev, old, list);
2398 spin_unlock_irqrestore(&list->lock, flags);
2400 EXPORT_SYMBOL(skb_insert);
2402 static inline void skb_split_inside_header(struct sk_buff *skb,
2403 struct sk_buff* skb1,
2404 const u32 len, const int pos)
2408 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2410 /* And move data appendix as is. */
2411 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2412 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2414 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2415 skb_shinfo(skb)->nr_frags = 0;
2416 skb1->data_len = skb->data_len;
2417 skb1->len += skb1->data_len;
2420 skb_set_tail_pointer(skb, len);
2423 static inline void skb_split_no_header(struct sk_buff *skb,
2424 struct sk_buff* skb1,
2425 const u32 len, int pos)
2428 const int nfrags = skb_shinfo(skb)->nr_frags;
2430 skb_shinfo(skb)->nr_frags = 0;
2431 skb1->len = skb1->data_len = skb->len - len;
2433 skb->data_len = len - pos;
2435 for (i = 0; i < nfrags; i++) {
2436 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2438 if (pos + size > len) {
2439 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2443 * We have two variants in this case:
2444 * 1. Move all the frag to the second
2445 * part, if it is possible. F.e.
2446 * this approach is mandatory for TUX,
2447 * where splitting is expensive.
2448 * 2. Split is accurately. We make this.
2450 skb_frag_ref(skb, i);
2451 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2452 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2453 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2454 skb_shinfo(skb)->nr_frags++;
2458 skb_shinfo(skb)->nr_frags++;
2461 skb_shinfo(skb1)->nr_frags = k;
2465 * skb_split - Split fragmented skb to two parts at length len.
2466 * @skb: the buffer to split
2467 * @skb1: the buffer to receive the second part
2468 * @len: new length for skb
2470 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2472 int pos = skb_headlen(skb);
2474 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2475 if (len < pos) /* Split line is inside header. */
2476 skb_split_inside_header(skb, skb1, len, pos);
2477 else /* Second chunk has no header, nothing to copy. */
2478 skb_split_no_header(skb, skb1, len, pos);
2480 EXPORT_SYMBOL(skb_split);
2482 /* Shifting from/to a cloned skb is a no-go.
2484 * Caller cannot keep skb_shinfo related pointers past calling here!
2486 static int skb_prepare_for_shift(struct sk_buff *skb)
2488 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2492 * skb_shift - Shifts paged data partially from skb to another
2493 * @tgt: buffer into which tail data gets added
2494 * @skb: buffer from which the paged data comes from
2495 * @shiftlen: shift up to this many bytes
2497 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2498 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2499 * It's up to caller to free skb if everything was shifted.
2501 * If @tgt runs out of frags, the whole operation is aborted.
2503 * Skb cannot include anything else but paged data while tgt is allowed
2504 * to have non-paged data as well.
2506 * TODO: full sized shift could be optimized but that would need
2507 * specialized skb free'er to handle frags without up-to-date nr_frags.
2509 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2511 int from, to, merge, todo;
2512 struct skb_frag_struct *fragfrom, *fragto;
2514 BUG_ON(shiftlen > skb->len);
2515 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2519 to = skb_shinfo(tgt)->nr_frags;
2520 fragfrom = &skb_shinfo(skb)->frags[from];
2522 /* Actual merge is delayed until the point when we know we can
2523 * commit all, so that we don't have to undo partial changes
2526 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2527 fragfrom->page_offset)) {
2532 todo -= skb_frag_size(fragfrom);
2534 if (skb_prepare_for_shift(skb) ||
2535 skb_prepare_for_shift(tgt))
2538 /* All previous frag pointers might be stale! */
2539 fragfrom = &skb_shinfo(skb)->frags[from];
2540 fragto = &skb_shinfo(tgt)->frags[merge];
2542 skb_frag_size_add(fragto, shiftlen);
2543 skb_frag_size_sub(fragfrom, shiftlen);
2544 fragfrom->page_offset += shiftlen;
2552 /* Skip full, not-fitting skb to avoid expensive operations */
2553 if ((shiftlen == skb->len) &&
2554 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2557 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2560 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2561 if (to == MAX_SKB_FRAGS)
2564 fragfrom = &skb_shinfo(skb)->frags[from];
2565 fragto = &skb_shinfo(tgt)->frags[to];
2567 if (todo >= skb_frag_size(fragfrom)) {
2568 *fragto = *fragfrom;
2569 todo -= skb_frag_size(fragfrom);
2574 __skb_frag_ref(fragfrom);
2575 fragto->page = fragfrom->page;
2576 fragto->page_offset = fragfrom->page_offset;
2577 skb_frag_size_set(fragto, todo);
2579 fragfrom->page_offset += todo;
2580 skb_frag_size_sub(fragfrom, todo);
2588 /* Ready to "commit" this state change to tgt */
2589 skb_shinfo(tgt)->nr_frags = to;
2592 fragfrom = &skb_shinfo(skb)->frags[0];
2593 fragto = &skb_shinfo(tgt)->frags[merge];
2595 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2596 __skb_frag_unref(fragfrom);
2599 /* Reposition in the original skb */
2601 while (from < skb_shinfo(skb)->nr_frags)
2602 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2603 skb_shinfo(skb)->nr_frags = to;
2605 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2608 /* Most likely the tgt won't ever need its checksum anymore, skb on
2609 * the other hand might need it if it needs to be resent
2611 tgt->ip_summed = CHECKSUM_PARTIAL;
2612 skb->ip_summed = CHECKSUM_PARTIAL;
2614 /* Yak, is it really working this way? Some helper please? */
2615 skb->len -= shiftlen;
2616 skb->data_len -= shiftlen;
2617 skb->truesize -= shiftlen;
2618 tgt->len += shiftlen;
2619 tgt->data_len += shiftlen;
2620 tgt->truesize += shiftlen;
2626 * skb_prepare_seq_read - Prepare a sequential read of skb data
2627 * @skb: the buffer to read
2628 * @from: lower offset of data to be read
2629 * @to: upper offset of data to be read
2630 * @st: state variable
2632 * Initializes the specified state variable. Must be called before
2633 * invoking skb_seq_read() for the first time.
2635 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2636 unsigned int to, struct skb_seq_state *st)
2638 st->lower_offset = from;
2639 st->upper_offset = to;
2640 st->root_skb = st->cur_skb = skb;
2641 st->frag_idx = st->stepped_offset = 0;
2642 st->frag_data = NULL;
2644 EXPORT_SYMBOL(skb_prepare_seq_read);
2647 * skb_seq_read - Sequentially read skb data
2648 * @consumed: number of bytes consumed by the caller so far
2649 * @data: destination pointer for data to be returned
2650 * @st: state variable
2652 * Reads a block of skb data at @consumed relative to the
2653 * lower offset specified to skb_prepare_seq_read(). Assigns
2654 * the head of the data block to @data and returns the length
2655 * of the block or 0 if the end of the skb data or the upper
2656 * offset has been reached.
2658 * The caller is not required to consume all of the data
2659 * returned, i.e. @consumed is typically set to the number
2660 * of bytes already consumed and the next call to
2661 * skb_seq_read() will return the remaining part of the block.
2663 * Note 1: The size of each block of data returned can be arbitrary,
2664 * this limitation is the cost for zerocopy sequential
2665 * reads of potentially non linear data.
2667 * Note 2: Fragment lists within fragments are not implemented
2668 * at the moment, state->root_skb could be replaced with
2669 * a stack for this purpose.
2671 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2672 struct skb_seq_state *st)
2674 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2677 if (unlikely(abs_offset >= st->upper_offset)) {
2678 if (st->frag_data) {
2679 kunmap_atomic(st->frag_data);
2680 st->frag_data = NULL;
2686 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2688 if (abs_offset < block_limit && !st->frag_data) {
2689 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2690 return block_limit - abs_offset;
2693 if (st->frag_idx == 0 && !st->frag_data)
2694 st->stepped_offset += skb_headlen(st->cur_skb);
2696 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2697 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2698 block_limit = skb_frag_size(frag) + st->stepped_offset;
2700 if (abs_offset < block_limit) {
2702 st->frag_data = kmap_atomic(skb_frag_page(frag));
2704 *data = (u8 *) st->frag_data + frag->page_offset +
2705 (abs_offset - st->stepped_offset);
2707 return block_limit - abs_offset;
2710 if (st->frag_data) {
2711 kunmap_atomic(st->frag_data);
2712 st->frag_data = NULL;
2716 st->stepped_offset += skb_frag_size(frag);
2719 if (st->frag_data) {
2720 kunmap_atomic(st->frag_data);
2721 st->frag_data = NULL;
2724 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2725 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2728 } else if (st->cur_skb->next) {
2729 st->cur_skb = st->cur_skb->next;
2736 EXPORT_SYMBOL(skb_seq_read);
2739 * skb_abort_seq_read - Abort a sequential read of skb data
2740 * @st: state variable
2742 * Must be called if skb_seq_read() was not called until it
2745 void skb_abort_seq_read(struct skb_seq_state *st)
2748 kunmap_atomic(st->frag_data);
2750 EXPORT_SYMBOL(skb_abort_seq_read);
2752 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2754 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2755 struct ts_config *conf,
2756 struct ts_state *state)
2758 return skb_seq_read(offset, text, TS_SKB_CB(state));
2761 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2763 skb_abort_seq_read(TS_SKB_CB(state));
2767 * skb_find_text - Find a text pattern in skb data
2768 * @skb: the buffer to look in
2769 * @from: search offset
2771 * @config: textsearch configuration
2772 * @state: uninitialized textsearch state variable
2774 * Finds a pattern in the skb data according to the specified
2775 * textsearch configuration. Use textsearch_next() to retrieve
2776 * subsequent occurrences of the pattern. Returns the offset
2777 * to the first occurrence or UINT_MAX if no match was found.
2779 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2780 unsigned int to, struct ts_config *config,
2781 struct ts_state *state)
2785 config->get_next_block = skb_ts_get_next_block;
2786 config->finish = skb_ts_finish;
2788 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2790 ret = textsearch_find(config, state);
2791 return (ret <= to - from ? ret : UINT_MAX);
2793 EXPORT_SYMBOL(skb_find_text);
2796 * skb_append_datato_frags - append the user data to a skb
2797 * @sk: sock structure
2798 * @skb: skb structure to be appended with user data.
2799 * @getfrag: call back function to be used for getting the user data
2800 * @from: pointer to user message iov
2801 * @length: length of the iov message
2803 * Description: This procedure append the user data in the fragment part
2804 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2806 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2807 int (*getfrag)(void *from, char *to, int offset,
2808 int len, int odd, struct sk_buff *skb),
2809 void *from, int length)
2811 int frg_cnt = skb_shinfo(skb)->nr_frags;
2815 struct page_frag *pfrag = ¤t->task_frag;
2818 /* Return error if we don't have space for new frag */
2819 if (frg_cnt >= MAX_SKB_FRAGS)
2822 if (!sk_page_frag_refill(sk, pfrag))
2825 /* copy the user data to page */
2826 copy = min_t(int, length, pfrag->size - pfrag->offset);
2828 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2829 offset, copy, 0, skb);
2833 /* copy was successful so update the size parameters */
2834 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2837 pfrag->offset += copy;
2838 get_page(pfrag->page);
2840 skb->truesize += copy;
2841 atomic_add(copy, &sk->sk_wmem_alloc);
2843 skb->data_len += copy;
2847 } while (length > 0);
2851 EXPORT_SYMBOL(skb_append_datato_frags);
2854 * skb_pull_rcsum - pull skb and update receive checksum
2855 * @skb: buffer to update
2856 * @len: length of data pulled
2858 * This function performs an skb_pull on the packet and updates
2859 * the CHECKSUM_COMPLETE checksum. It should be used on
2860 * receive path processing instead of skb_pull unless you know
2861 * that the checksum difference is zero (e.g., a valid IP header)
2862 * or you are setting ip_summed to CHECKSUM_NONE.
2864 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2866 BUG_ON(len > skb->len);
2868 BUG_ON(skb->len < skb->data_len);
2869 skb_postpull_rcsum(skb, skb->data, len);
2870 return skb->data += len;
2872 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2875 * skb_segment - Perform protocol segmentation on skb.
2876 * @head_skb: buffer to segment
2877 * @features: features for the output path (see dev->features)
2879 * This function performs segmentation on the given skb. It returns
2880 * a pointer to the first in a list of new skbs for the segments.
2881 * In case of error it returns ERR_PTR(err).
2883 struct sk_buff *skb_segment(struct sk_buff *head_skb,
2884 netdev_features_t features)
2886 struct sk_buff *segs = NULL;
2887 struct sk_buff *tail = NULL;
2888 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
2889 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
2890 unsigned int mss = skb_shinfo(head_skb)->gso_size;
2891 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
2892 struct sk_buff *frag_skb = head_skb;
2893 unsigned int offset = doffset;
2894 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
2895 unsigned int headroom;
2899 int sg = !!(features & NETIF_F_SG);
2900 int nfrags = skb_shinfo(head_skb)->nr_frags;
2906 __skb_push(head_skb, doffset);
2907 proto = skb_network_protocol(head_skb, &dummy);
2908 if (unlikely(!proto))
2909 return ERR_PTR(-EINVAL);
2911 csum = !head_skb->encap_hdr_csum &&
2912 !!can_checksum_protocol(features, proto);
2914 headroom = skb_headroom(head_skb);
2915 pos = skb_headlen(head_skb);
2918 struct sk_buff *nskb;
2919 skb_frag_t *nskb_frag;
2923 len = head_skb->len - offset;
2927 hsize = skb_headlen(head_skb) - offset;
2930 if (hsize > len || !sg)
2933 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
2934 (skb_headlen(list_skb) == len || sg)) {
2935 BUG_ON(skb_headlen(list_skb) > len);
2938 nfrags = skb_shinfo(list_skb)->nr_frags;
2939 frag = skb_shinfo(list_skb)->frags;
2940 frag_skb = list_skb;
2941 pos += skb_headlen(list_skb);
2943 while (pos < offset + len) {
2944 BUG_ON(i >= nfrags);
2946 size = skb_frag_size(frag);
2947 if (pos + size > offset + len)
2955 nskb = skb_clone(list_skb, GFP_ATOMIC);
2956 list_skb = list_skb->next;
2958 if (unlikely(!nskb))
2961 if (unlikely(pskb_trim(nskb, len))) {
2966 hsize = skb_end_offset(nskb);
2967 if (skb_cow_head(nskb, doffset + headroom)) {
2972 nskb->truesize += skb_end_offset(nskb) - hsize;
2973 skb_release_head_state(nskb);
2974 __skb_push(nskb, doffset);
2976 nskb = __alloc_skb(hsize + doffset + headroom,
2977 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
2980 if (unlikely(!nskb))
2983 skb_reserve(nskb, headroom);
2984 __skb_put(nskb, doffset);
2993 __copy_skb_header(nskb, head_skb);
2995 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
2996 skb_reset_mac_len(nskb);
2998 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
2999 nskb->data - tnl_hlen,
3000 doffset + tnl_hlen);
3002 if (nskb->len == len + doffset)
3003 goto perform_csum_check;
3006 nskb->ip_summed = CHECKSUM_NONE;
3007 nskb->csum = skb_copy_and_csum_bits(head_skb, offset,
3010 SKB_GSO_CB(nskb)->csum_start =
3011 skb_headroom(nskb) + doffset;
3015 nskb_frag = skb_shinfo(nskb)->frags;
3017 skb_copy_from_linear_data_offset(head_skb, offset,
3018 skb_put(nskb, hsize), hsize);
3020 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3023 while (pos < offset + len) {
3025 BUG_ON(skb_headlen(list_skb));
3028 nfrags = skb_shinfo(list_skb)->nr_frags;
3029 frag = skb_shinfo(list_skb)->frags;
3030 frag_skb = list_skb;
3034 list_skb = list_skb->next;
3037 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3039 net_warn_ratelimited(
3040 "skb_segment: too many frags: %u %u\n",
3045 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3049 __skb_frag_ref(nskb_frag);
3050 size = skb_frag_size(nskb_frag);
3053 nskb_frag->page_offset += offset - pos;
3054 skb_frag_size_sub(nskb_frag, offset - pos);
3057 skb_shinfo(nskb)->nr_frags++;
3059 if (pos + size <= offset + len) {
3064 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3072 nskb->data_len = len - hsize;
3073 nskb->len += nskb->data_len;
3074 nskb->truesize += nskb->data_len;
3078 nskb->csum = skb_checksum(nskb, doffset,
3079 nskb->len - doffset, 0);
3080 nskb->ip_summed = CHECKSUM_NONE;
3081 SKB_GSO_CB(nskb)->csum_start =
3082 skb_headroom(nskb) + doffset;
3084 } while ((offset += len) < head_skb->len);
3086 /* Some callers want to get the end of the list.
3087 * Put it in segs->prev to avoid walking the list.
3088 * (see validate_xmit_skb_list() for example)
3094 kfree_skb_list(segs);
3095 return ERR_PTR(err);
3097 EXPORT_SYMBOL_GPL(skb_segment);
3099 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3101 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3102 unsigned int offset = skb_gro_offset(skb);
3103 unsigned int headlen = skb_headlen(skb);
3104 struct sk_buff *nskb, *lp, *p = *head;
3105 unsigned int len = skb_gro_len(skb);
3106 unsigned int delta_truesize;
3107 unsigned int headroom;
3109 if (unlikely(p->len + len >= 65536))
3112 lp = NAPI_GRO_CB(p)->last;
3113 pinfo = skb_shinfo(lp);
3115 if (headlen <= offset) {
3118 int i = skbinfo->nr_frags;
3119 int nr_frags = pinfo->nr_frags + i;
3121 if (nr_frags > MAX_SKB_FRAGS)
3125 pinfo->nr_frags = nr_frags;
3126 skbinfo->nr_frags = 0;
3128 frag = pinfo->frags + nr_frags;
3129 frag2 = skbinfo->frags + i;
3134 frag->page_offset += offset;
3135 skb_frag_size_sub(frag, offset);
3137 /* all fragments truesize : remove (head size + sk_buff) */
3138 delta_truesize = skb->truesize -
3139 SKB_TRUESIZE(skb_end_offset(skb));
3141 skb->truesize -= skb->data_len;
3142 skb->len -= skb->data_len;
3145 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3147 } else if (skb->head_frag) {
3148 int nr_frags = pinfo->nr_frags;
3149 skb_frag_t *frag = pinfo->frags + nr_frags;
3150 struct page *page = virt_to_head_page(skb->head);
3151 unsigned int first_size = headlen - offset;
3152 unsigned int first_offset;
3154 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3157 first_offset = skb->data -
3158 (unsigned char *)page_address(page) +
3161 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3163 frag->page.p = page;
3164 frag->page_offset = first_offset;
3165 skb_frag_size_set(frag, first_size);
3167 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3168 /* We dont need to clear skbinfo->nr_frags here */
3170 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3171 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3174 /* switch back to head shinfo */
3175 pinfo = skb_shinfo(p);
3177 if (pinfo->frag_list)
3179 if (skb_gro_len(p) != pinfo->gso_size)
3182 headroom = skb_headroom(p);
3183 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
3184 if (unlikely(!nskb))
3187 __copy_skb_header(nskb, p);
3188 nskb->mac_len = p->mac_len;
3190 skb_reserve(nskb, headroom);
3191 __skb_put(nskb, skb_gro_offset(p));
3193 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3194 skb_set_network_header(nskb, skb_network_offset(p));
3195 skb_set_transport_header(nskb, skb_transport_offset(p));
3197 __skb_pull(p, skb_gro_offset(p));
3198 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3199 p->data - skb_mac_header(p));
3201 skb_shinfo(nskb)->frag_list = p;
3202 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3203 pinfo->gso_size = 0;
3204 __skb_header_release(p);
3205 NAPI_GRO_CB(nskb)->last = p;
3207 nskb->data_len += p->len;
3208 nskb->truesize += p->truesize;
3209 nskb->len += p->len;
3212 nskb->next = p->next;
3218 delta_truesize = skb->truesize;
3219 if (offset > headlen) {
3220 unsigned int eat = offset - headlen;
3222 skbinfo->frags[0].page_offset += eat;
3223 skb_frag_size_sub(&skbinfo->frags[0], eat);
3224 skb->data_len -= eat;
3229 __skb_pull(skb, offset);
3231 if (NAPI_GRO_CB(p)->last == p)
3232 skb_shinfo(p)->frag_list = skb;
3234 NAPI_GRO_CB(p)->last->next = skb;
3235 NAPI_GRO_CB(p)->last = skb;
3236 __skb_header_release(skb);
3240 NAPI_GRO_CB(p)->count++;
3242 p->truesize += delta_truesize;
3245 lp->data_len += len;
3246 lp->truesize += delta_truesize;
3249 NAPI_GRO_CB(skb)->same_flow = 1;
3253 void __init skb_init(void)
3255 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3256 sizeof(struct sk_buff),
3258 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3260 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3261 sizeof(struct sk_buff_fclones),
3263 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3268 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3269 * @skb: Socket buffer containing the buffers to be mapped
3270 * @sg: The scatter-gather list to map into
3271 * @offset: The offset into the buffer's contents to start mapping
3272 * @len: Length of buffer space to be mapped
3274 * Fill the specified scatter-gather list with mappings/pointers into a
3275 * region of the buffer space attached to a socket buffer.
3278 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3280 int start = skb_headlen(skb);
3281 int i, copy = start - offset;
3282 struct sk_buff *frag_iter;
3288 sg_set_buf(sg, skb->data + offset, copy);
3290 if ((len -= copy) == 0)
3295 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3298 WARN_ON(start > offset + len);
3300 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3301 if ((copy = end - offset) > 0) {
3302 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3306 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3307 frag->page_offset+offset-start);
3316 skb_walk_frags(skb, frag_iter) {
3319 WARN_ON(start > offset + len);
3321 end = start + frag_iter->len;
3322 if ((copy = end - offset) > 0) {
3325 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3327 if ((len -= copy) == 0)
3337 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3338 * sglist without mark the sg which contain last skb data as the end.
3339 * So the caller can mannipulate sg list as will when padding new data after
3340 * the first call without calling sg_unmark_end to expend sg list.
3342 * Scenario to use skb_to_sgvec_nomark:
3344 * 2. skb_to_sgvec_nomark(payload1)
3345 * 3. skb_to_sgvec_nomark(payload2)
3347 * This is equivalent to:
3349 * 2. skb_to_sgvec(payload1)
3351 * 4. skb_to_sgvec(payload2)
3353 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3354 * is more preferable.
3356 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3357 int offset, int len)
3359 return __skb_to_sgvec(skb, sg, offset, len);
3361 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3363 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3365 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3367 sg_mark_end(&sg[nsg - 1]);
3371 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3374 * skb_cow_data - Check that a socket buffer's data buffers are writable
3375 * @skb: The socket buffer to check.
3376 * @tailbits: Amount of trailing space to be added
3377 * @trailer: Returned pointer to the skb where the @tailbits space begins
3379 * Make sure that the data buffers attached to a socket buffer are
3380 * writable. If they are not, private copies are made of the data buffers
3381 * and the socket buffer is set to use these instead.
3383 * If @tailbits is given, make sure that there is space to write @tailbits
3384 * bytes of data beyond current end of socket buffer. @trailer will be
3385 * set to point to the skb in which this space begins.
3387 * The number of scatterlist elements required to completely map the
3388 * COW'd and extended socket buffer will be returned.
3390 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3394 struct sk_buff *skb1, **skb_p;
3396 /* If skb is cloned or its head is paged, reallocate
3397 * head pulling out all the pages (pages are considered not writable
3398 * at the moment even if they are anonymous).
3400 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3401 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3404 /* Easy case. Most of packets will go this way. */
3405 if (!skb_has_frag_list(skb)) {
3406 /* A little of trouble, not enough of space for trailer.
3407 * This should not happen, when stack is tuned to generate
3408 * good frames. OK, on miss we reallocate and reserve even more
3409 * space, 128 bytes is fair. */
3411 if (skb_tailroom(skb) < tailbits &&
3412 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3420 /* Misery. We are in troubles, going to mincer fragments... */
3423 skb_p = &skb_shinfo(skb)->frag_list;
3426 while ((skb1 = *skb_p) != NULL) {
3429 /* The fragment is partially pulled by someone,
3430 * this can happen on input. Copy it and everything
3433 if (skb_shared(skb1))
3436 /* If the skb is the last, worry about trailer. */
3438 if (skb1->next == NULL && tailbits) {
3439 if (skb_shinfo(skb1)->nr_frags ||
3440 skb_has_frag_list(skb1) ||
3441 skb_tailroom(skb1) < tailbits)
3442 ntail = tailbits + 128;
3448 skb_shinfo(skb1)->nr_frags ||
3449 skb_has_frag_list(skb1)) {
3450 struct sk_buff *skb2;
3452 /* Fuck, we are miserable poor guys... */
3454 skb2 = skb_copy(skb1, GFP_ATOMIC);
3456 skb2 = skb_copy_expand(skb1,
3460 if (unlikely(skb2 == NULL))
3464 skb_set_owner_w(skb2, skb1->sk);
3466 /* Looking around. Are we still alive?
3467 * OK, link new skb, drop old one */
3469 skb2->next = skb1->next;
3476 skb_p = &skb1->next;
3481 EXPORT_SYMBOL_GPL(skb_cow_data);
3483 static void sock_rmem_free(struct sk_buff *skb)
3485 struct sock *sk = skb->sk;
3487 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3491 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3493 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3495 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3496 (unsigned int)sk->sk_rcvbuf)
3501 skb->destructor = sock_rmem_free;
3502 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3504 /* before exiting rcu section, make sure dst is refcounted */
3507 skb_queue_tail(&sk->sk_error_queue, skb);
3508 if (!sock_flag(sk, SOCK_DEAD))
3509 sk->sk_data_ready(sk);
3512 EXPORT_SYMBOL(sock_queue_err_skb);
3514 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3516 struct sk_buff_head *q = &sk->sk_error_queue;
3517 struct sk_buff *skb, *skb_next;
3520 spin_lock_bh(&q->lock);
3521 skb = __skb_dequeue(q);
3522 if (skb && (skb_next = skb_peek(q)))
3523 err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
3524 spin_unlock_bh(&q->lock);
3528 sk->sk_error_report(sk);
3532 EXPORT_SYMBOL(sock_dequeue_err_skb);
3535 * skb_clone_sk - create clone of skb, and take reference to socket
3536 * @skb: the skb to clone
3538 * This function creates a clone of a buffer that holds a reference on
3539 * sk_refcnt. Buffers created via this function are meant to be
3540 * returned using sock_queue_err_skb, or free via kfree_skb.
3542 * When passing buffers allocated with this function to sock_queue_err_skb
3543 * it is necessary to wrap the call with sock_hold/sock_put in order to
3544 * prevent the socket from being released prior to being enqueued on
3545 * the sk_error_queue.
3547 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3549 struct sock *sk = skb->sk;
3550 struct sk_buff *clone;
3552 if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3555 clone = skb_clone(skb, GFP_ATOMIC);
3562 clone->destructor = sock_efree;
3566 EXPORT_SYMBOL(skb_clone_sk);
3568 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3572 struct sock_exterr_skb *serr;
3575 serr = SKB_EXT_ERR(skb);
3576 memset(serr, 0, sizeof(*serr));
3577 serr->ee.ee_errno = ENOMSG;
3578 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3579 serr->ee.ee_info = tstype;
3580 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3581 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3582 if (sk->sk_protocol == IPPROTO_TCP)
3583 serr->ee.ee_data -= sk->sk_tskey;
3586 err = sock_queue_err_skb(sk, skb);
3592 void skb_complete_tx_timestamp(struct sk_buff *skb,
3593 struct skb_shared_hwtstamps *hwtstamps)
3595 struct sock *sk = skb->sk;
3597 /* take a reference to prevent skb_orphan() from freeing the socket */
3600 *skb_hwtstamps(skb) = *hwtstamps;
3601 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
3605 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3607 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3608 struct skb_shared_hwtstamps *hwtstamps,
3609 struct sock *sk, int tstype)
3611 struct sk_buff *skb;
3617 *skb_hwtstamps(orig_skb) = *hwtstamps;
3619 orig_skb->tstamp = ktime_get_real();
3621 skb = skb_clone(orig_skb, GFP_ATOMIC);
3625 __skb_complete_tx_timestamp(skb, sk, tstype);
3627 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3629 void skb_tstamp_tx(struct sk_buff *orig_skb,
3630 struct skb_shared_hwtstamps *hwtstamps)
3632 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3635 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3637 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3639 struct sock *sk = skb->sk;
3640 struct sock_exterr_skb *serr;
3643 skb->wifi_acked_valid = 1;
3644 skb->wifi_acked = acked;
3646 serr = SKB_EXT_ERR(skb);
3647 memset(serr, 0, sizeof(*serr));
3648 serr->ee.ee_errno = ENOMSG;
3649 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3651 /* take a reference to prevent skb_orphan() from freeing the socket */
3654 err = sock_queue_err_skb(sk, skb);
3660 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3664 * skb_partial_csum_set - set up and verify partial csum values for packet
3665 * @skb: the skb to set
3666 * @start: the number of bytes after skb->data to start checksumming.
3667 * @off: the offset from start to place the checksum.
3669 * For untrusted partially-checksummed packets, we need to make sure the values
3670 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3672 * This function checks and sets those values and skb->ip_summed: if this
3673 * returns false you should drop the packet.
3675 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3677 if (unlikely(start > skb_headlen(skb)) ||
3678 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3679 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3680 start, off, skb_headlen(skb));
3683 skb->ip_summed = CHECKSUM_PARTIAL;
3684 skb->csum_start = skb_headroom(skb) + start;
3685 skb->csum_offset = off;
3686 skb_set_transport_header(skb, start);
3689 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3691 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3694 if (skb_headlen(skb) >= len)
3697 /* If we need to pullup then pullup to the max, so we
3698 * won't need to do it again.
3703 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3706 if (skb_headlen(skb) < len)
3712 #define MAX_TCP_HDR_LEN (15 * 4)
3714 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3715 typeof(IPPROTO_IP) proto,
3722 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3723 off + MAX_TCP_HDR_LEN);
3724 if (!err && !skb_partial_csum_set(skb, off,
3725 offsetof(struct tcphdr,
3728 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3731 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3732 off + sizeof(struct udphdr));
3733 if (!err && !skb_partial_csum_set(skb, off,
3734 offsetof(struct udphdr,
3737 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3740 return ERR_PTR(-EPROTO);
3743 /* This value should be large enough to cover a tagged ethernet header plus
3744 * maximally sized IP and TCP or UDP headers.
3746 #define MAX_IP_HDR_LEN 128
3748 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
3757 err = skb_maybe_pull_tail(skb,
3758 sizeof(struct iphdr),
3763 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3766 off = ip_hdrlen(skb);
3773 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
3775 return PTR_ERR(csum);
3778 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3781 ip_hdr(skb)->protocol, 0);
3788 /* This value should be large enough to cover a tagged ethernet header plus
3789 * an IPv6 header, all options, and a maximal TCP or UDP header.
3791 #define MAX_IPV6_HDR_LEN 256
3793 #define OPT_HDR(type, skb, off) \
3794 (type *)(skb_network_header(skb) + (off))
3796 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
3809 off = sizeof(struct ipv6hdr);
3811 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
3815 nexthdr = ipv6_hdr(skb)->nexthdr;
3817 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
3818 while (off <= len && !done) {
3820 case IPPROTO_DSTOPTS:
3821 case IPPROTO_HOPOPTS:
3822 case IPPROTO_ROUTING: {
3823 struct ipv6_opt_hdr *hp;
3825 err = skb_maybe_pull_tail(skb,
3827 sizeof(struct ipv6_opt_hdr),
3832 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
3833 nexthdr = hp->nexthdr;
3834 off += ipv6_optlen(hp);
3838 struct ip_auth_hdr *hp;
3840 err = skb_maybe_pull_tail(skb,
3842 sizeof(struct ip_auth_hdr),
3847 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
3848 nexthdr = hp->nexthdr;
3849 off += ipv6_authlen(hp);
3852 case IPPROTO_FRAGMENT: {
3853 struct frag_hdr *hp;
3855 err = skb_maybe_pull_tail(skb,
3857 sizeof(struct frag_hdr),
3862 hp = OPT_HDR(struct frag_hdr, skb, off);
3864 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
3867 nexthdr = hp->nexthdr;
3868 off += sizeof(struct frag_hdr);
3879 if (!done || fragment)
3882 csum = skb_checksum_setup_ip(skb, nexthdr, off);
3884 return PTR_ERR(csum);
3887 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3888 &ipv6_hdr(skb)->daddr,
3889 skb->len - off, nexthdr, 0);
3897 * skb_checksum_setup - set up partial checksum offset
3898 * @skb: the skb to set up
3899 * @recalculate: if true the pseudo-header checksum will be recalculated
3901 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
3905 switch (skb->protocol) {
3906 case htons(ETH_P_IP):
3907 err = skb_checksum_setup_ipv4(skb, recalculate);
3910 case htons(ETH_P_IPV6):
3911 err = skb_checksum_setup_ipv6(skb, recalculate);
3921 EXPORT_SYMBOL(skb_checksum_setup);
3923 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3925 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3928 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3930 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3933 skb_release_head_state(skb);
3934 kmem_cache_free(skbuff_head_cache, skb);
3939 EXPORT_SYMBOL(kfree_skb_partial);
3942 * skb_try_coalesce - try to merge skb to prior one
3944 * @from: buffer to add
3945 * @fragstolen: pointer to boolean
3946 * @delta_truesize: how much more was allocated than was requested
3948 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3949 bool *fragstolen, int *delta_truesize)
3951 int i, delta, len = from->len;
3953 *fragstolen = false;
3958 if (len <= skb_tailroom(to)) {
3960 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3961 *delta_truesize = 0;
3965 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3968 if (skb_headlen(from) != 0) {
3970 unsigned int offset;
3972 if (skb_shinfo(to)->nr_frags +
3973 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3976 if (skb_head_is_locked(from))
3979 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3981 page = virt_to_head_page(from->head);
3982 offset = from->data - (unsigned char *)page_address(page);
3984 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3985 page, offset, skb_headlen(from));
3988 if (skb_shinfo(to)->nr_frags +
3989 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3992 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
3995 WARN_ON_ONCE(delta < len);
3997 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3998 skb_shinfo(from)->frags,
3999 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4000 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4002 if (!skb_cloned(from))
4003 skb_shinfo(from)->nr_frags = 0;
4005 /* if the skb is not cloned this does nothing
4006 * since we set nr_frags to 0.
4008 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4009 skb_frag_ref(from, i);
4011 to->truesize += delta;
4013 to->data_len += len;
4015 *delta_truesize = delta;
4018 EXPORT_SYMBOL(skb_try_coalesce);
4021 * skb_scrub_packet - scrub an skb
4023 * @skb: buffer to clean
4024 * @xnet: packet is crossing netns
4026 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4027 * into/from a tunnel. Some information have to be cleared during these
4029 * skb_scrub_packet can also be used to clean a skb before injecting it in
4030 * another namespace (@xnet == true). We have to clear all information in the
4031 * skb that could impact namespace isolation.
4033 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4037 skb->tstamp.tv64 = 0;
4038 skb->pkt_type = PACKET_HOST;
4045 nf_reset_trace(skb);
4047 EXPORT_SYMBOL_GPL(skb_scrub_packet);
4050 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4054 * skb_gso_transport_seglen is used to determine the real size of the
4055 * individual segments, including Layer4 headers (TCP/UDP).
4057 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4059 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4061 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4062 unsigned int thlen = 0;
4064 if (skb->encapsulation) {
4065 thlen = skb_inner_transport_header(skb) -
4066 skb_transport_header(skb);
4068 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4069 thlen += inner_tcp_hdrlen(skb);
4070 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4071 thlen = tcp_hdrlen(skb);
4073 /* UFO sets gso_size to the size of the fragmentation
4074 * payload, i.e. the size of the L4 (UDP) header is already
4077 return thlen + shinfo->gso_size;
4079 EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4081 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4083 if (skb_cow(skb, skb_headroom(skb)) < 0) {
4088 memmove(skb->data - ETH_HLEN, skb->data - VLAN_ETH_HLEN, 2 * ETH_ALEN);
4089 skb->mac_header += VLAN_HLEN;
4093 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4095 struct vlan_hdr *vhdr;
4098 if (unlikely(vlan_tx_tag_present(skb))) {
4099 /* vlan_tci is already set-up so leave this for another time */
4103 skb = skb_share_check(skb, GFP_ATOMIC);
4107 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4110 vhdr = (struct vlan_hdr *)skb->data;
4111 vlan_tci = ntohs(vhdr->h_vlan_TCI);
4112 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4114 skb_pull_rcsum(skb, VLAN_HLEN);
4115 vlan_set_encap_proto(skb, vhdr);
4117 skb = skb_reorder_vlan_header(skb);
4121 skb_reset_network_header(skb);
4122 skb_reset_transport_header(skb);
4123 skb_reset_mac_len(skb);
4131 EXPORT_SYMBOL(skb_vlan_untag);
4134 * alloc_skb_with_frags - allocate skb with page frags
4136 * @header_len: size of linear part
4137 * @data_len: needed length in frags
4138 * @max_page_order: max page order desired.
4139 * @errcode: pointer to error code if any
4140 * @gfp_mask: allocation mask
4142 * This can be used to allocate a paged skb, given a maximal order for frags.
4144 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4145 unsigned long data_len,
4150 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4151 unsigned long chunk;
4152 struct sk_buff *skb;
4157 *errcode = -EMSGSIZE;
4158 /* Note this test could be relaxed, if we succeed to allocate
4159 * high order pages...
4161 if (npages > MAX_SKB_FRAGS)
4164 gfp_head = gfp_mask;
4165 if (gfp_head & __GFP_WAIT)
4166 gfp_head |= __GFP_REPEAT;
4168 *errcode = -ENOBUFS;
4169 skb = alloc_skb(header_len, gfp_head);
4173 skb->truesize += npages << PAGE_SHIFT;
4175 for (i = 0; npages > 0; i++) {
4176 int order = max_page_order;
4179 if (npages >= 1 << order) {
4180 page = alloc_pages(gfp_mask |
4187 /* Do not retry other high order allocations */
4193 page = alloc_page(gfp_mask);
4197 chunk = min_t(unsigned long, data_len,
4198 PAGE_SIZE << order);
4199 skb_fill_page_desc(skb, i, page, 0, chunk);
4201 npages -= 1 << order;
4209 EXPORT_SYMBOL(alloc_skb_with_frags);