1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Routines having to do with the 'struct sk_buff' memory handlers.
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Florian La Roche <rzsfl@rz.uni-sb.de>
9 * Alan Cox : Fixed the worst of the load
11 * Dave Platt : Interrupt stacking fix.
12 * Richard Kooijman : Timestamp fixes.
13 * Alan Cox : Changed buffer format.
14 * Alan Cox : destructor hook for AF_UNIX etc.
15 * Linus Torvalds : Better skb_clone.
16 * Alan Cox : Added skb_copy.
17 * Alan Cox : Added all the changed routines Linus
18 * only put in the headers
19 * Ray VanTassle : Fixed --skb->lock in free
20 * Alan Cox : skb_copy copy arp field
21 * Andi Kleen : slabified it.
22 * Robert Olsson : Removed skb_head_pool
25 * The __skb_ routines should be called with interrupts
26 * disabled, or you better be *real* sure that the operation is atomic
27 * with respect to whatever list is being frobbed (e.g. via lock_sock()
28 * or via disabling bottom half handlers, etc).
32 * The functions in this file will not compile correctly with gcc 2.4.x
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/kernel.h>
41 #include <linux/interrupt.h>
43 #include <linux/inet.h>
44 #include <linux/slab.h>
45 #include <linux/tcp.h>
46 #include <linux/udp.h>
47 #include <linux/sctp.h>
48 #include <linux/netdevice.h>
49 #ifdef CONFIG_NET_CLS_ACT
50 #include <net/pkt_sched.h>
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/splice.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
59 #include <linux/errqueue.h>
60 #include <linux/prefetch.h>
61 #include <linux/if_vlan.h>
62 #include <linux/mpls.h>
64 #include <net/protocol.h>
67 #include <net/checksum.h>
68 #include <net/ip6_checksum.h>
72 #include <linux/uaccess.h>
73 #include <trace/events/skb.h>
74 #include <linux/highmem.h>
75 #include <linux/capability.h>
76 #include <linux/user_namespace.h>
77 #include <linux/indirect_call_wrapper.h>
81 struct kmem_cache *skbuff_head_cache __ro_after_init;
82 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
83 #ifdef CONFIG_SKB_EXTENSIONS
84 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
86 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
87 EXPORT_SYMBOL(sysctl_max_skb_frags);
90 * skb_panic - private function for out-of-line support
94 * @msg: skb_over_panic or skb_under_panic
96 * Out-of-line support for skb_put() and skb_push().
97 * Called via the wrapper skb_over_panic() or skb_under_panic().
98 * Keep out of line to prevent kernel bloat.
99 * __builtin_return_address is not used because it is not always reliable.
101 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
104 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
105 msg, addr, skb->len, sz, skb->head, skb->data,
106 (unsigned long)skb->tail, (unsigned long)skb->end,
107 skb->dev ? skb->dev->name : "<NULL>");
111 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
113 skb_panic(skb, sz, addr, __func__);
116 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
118 skb_panic(skb, sz, addr, __func__);
122 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
123 * the caller if emergency pfmemalloc reserves are being used. If it is and
124 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
125 * may be used. Otherwise, the packet data may be discarded until enough
128 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
129 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
131 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
132 unsigned long ip, bool *pfmemalloc)
135 bool ret_pfmemalloc = false;
138 * Try a regular allocation, when that fails and we're not entitled
139 * to the reserves, fail.
141 obj = kmalloc_node_track_caller(size,
142 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
144 if (obj || !(gfp_pfmemalloc_allowed(flags)))
147 /* Try again but now we are using pfmemalloc reserves */
148 ret_pfmemalloc = true;
149 obj = kmalloc_node_track_caller(size, flags, node);
153 *pfmemalloc = ret_pfmemalloc;
158 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
159 * 'private' fields and also do memory statistics to find all the
165 * __alloc_skb - allocate a network buffer
166 * @size: size to allocate
167 * @gfp_mask: allocation mask
168 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
169 * instead of head cache and allocate a cloned (child) skb.
170 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
171 * allocations in case the data is required for writeback
172 * @node: numa node to allocate memory on
174 * Allocate a new &sk_buff. The returned buffer has no headroom and a
175 * tail room of at least size bytes. The object has a reference count
176 * of one. The return is the buffer. On a failure the return is %NULL.
178 * Buffers may only be allocated from interrupts using a @gfp_mask of
181 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
184 struct kmem_cache *cache;
185 struct skb_shared_info *shinfo;
190 cache = (flags & SKB_ALLOC_FCLONE)
191 ? skbuff_fclone_cache : skbuff_head_cache;
193 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
194 gfp_mask |= __GFP_MEMALLOC;
197 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
202 /* We do our best to align skb_shared_info on a separate cache
203 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
204 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
205 * Both skb->head and skb_shared_info are cache line aligned.
207 size = SKB_DATA_ALIGN(size);
208 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
209 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
212 /* kmalloc(size) might give us more room than requested.
213 * Put skb_shared_info exactly at the end of allocated zone,
214 * to allow max possible filling before reallocation.
216 size = SKB_WITH_OVERHEAD(ksize(data));
217 prefetchw(data + size);
220 * Only clear those fields we need to clear, not those that we will
221 * actually initialise below. Hence, don't put any more fields after
222 * the tail pointer in struct sk_buff!
224 memset(skb, 0, offsetof(struct sk_buff, tail));
225 /* Account for allocated memory : skb + skb->head */
226 skb->truesize = SKB_TRUESIZE(size);
227 skb->pfmemalloc = pfmemalloc;
228 refcount_set(&skb->users, 1);
231 skb_reset_tail_pointer(skb);
232 skb->end = skb->tail + size;
233 skb->mac_header = (typeof(skb->mac_header))~0U;
234 skb->transport_header = (typeof(skb->transport_header))~0U;
236 /* make sure we initialize shinfo sequentially */
237 shinfo = skb_shinfo(skb);
238 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
239 atomic_set(&shinfo->dataref, 1);
241 if (flags & SKB_ALLOC_FCLONE) {
242 struct sk_buff_fclones *fclones;
244 fclones = container_of(skb, struct sk_buff_fclones, skb1);
246 skb->fclone = SKB_FCLONE_ORIG;
247 refcount_set(&fclones->fclone_ref, 1);
249 fclones->skb2.fclone = SKB_FCLONE_CLONE;
254 kmem_cache_free(cache, skb);
258 EXPORT_SYMBOL(__alloc_skb);
260 /* Caller must provide SKB that is memset cleared */
261 static struct sk_buff *__build_skb_around(struct sk_buff *skb,
262 void *data, unsigned int frag_size)
264 struct skb_shared_info *shinfo;
265 unsigned int size = frag_size ? : ksize(data);
267 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
269 /* Assumes caller memset cleared SKB */
270 skb->truesize = SKB_TRUESIZE(size);
271 refcount_set(&skb->users, 1);
274 skb_reset_tail_pointer(skb);
275 skb->end = skb->tail + size;
276 skb->mac_header = (typeof(skb->mac_header))~0U;
277 skb->transport_header = (typeof(skb->transport_header))~0U;
279 /* make sure we initialize shinfo sequentially */
280 shinfo = skb_shinfo(skb);
281 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
282 atomic_set(&shinfo->dataref, 1);
288 * __build_skb - build a network buffer
289 * @data: data buffer provided by caller
290 * @frag_size: size of data, or 0 if head was kmalloced
292 * Allocate a new &sk_buff. Caller provides space holding head and
293 * skb_shared_info. @data must have been allocated by kmalloc() only if
294 * @frag_size is 0, otherwise data should come from the page allocator
296 * The return is the new skb buffer.
297 * On a failure the return is %NULL, and @data is not freed.
299 * Before IO, driver allocates only data buffer where NIC put incoming frame
300 * Driver should add room at head (NET_SKB_PAD) and
301 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
302 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
303 * before giving packet to stack.
304 * RX rings only contains data buffers, not full skbs.
306 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
310 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
314 memset(skb, 0, offsetof(struct sk_buff, tail));
316 return __build_skb_around(skb, data, frag_size);
319 /* build_skb() is wrapper over __build_skb(), that specifically
320 * takes care of skb->head and skb->pfmemalloc
321 * This means that if @frag_size is not zero, then @data must be backed
322 * by a page fragment, not kmalloc() or vmalloc()
324 struct sk_buff *build_skb(void *data, unsigned int frag_size)
326 struct sk_buff *skb = __build_skb(data, frag_size);
328 if (skb && frag_size) {
330 if (page_is_pfmemalloc(virt_to_head_page(data)))
335 EXPORT_SYMBOL(build_skb);
338 * build_skb_around - build a network buffer around provided skb
339 * @skb: sk_buff provide by caller, must be memset cleared
340 * @data: data buffer provided by caller
341 * @frag_size: size of data, or 0 if head was kmalloced
343 struct sk_buff *build_skb_around(struct sk_buff *skb,
344 void *data, unsigned int frag_size)
349 skb = __build_skb_around(skb, data, frag_size);
351 if (skb && frag_size) {
353 if (page_is_pfmemalloc(virt_to_head_page(data)))
358 EXPORT_SYMBOL(build_skb_around);
360 #define NAPI_SKB_CACHE_SIZE 64
362 struct napi_alloc_cache {
363 struct page_frag_cache page;
364 unsigned int skb_count;
365 void *skb_cache[NAPI_SKB_CACHE_SIZE];
368 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
369 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
371 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
373 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
375 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
378 void *napi_alloc_frag(unsigned int fragsz)
380 fragsz = SKB_DATA_ALIGN(fragsz);
382 return __napi_alloc_frag(fragsz, GFP_ATOMIC);
384 EXPORT_SYMBOL(napi_alloc_frag);
387 * netdev_alloc_frag - allocate a page fragment
388 * @fragsz: fragment size
390 * Allocates a frag from a page for receive buffer.
391 * Uses GFP_ATOMIC allocations.
393 void *netdev_alloc_frag(unsigned int fragsz)
395 struct page_frag_cache *nc;
398 fragsz = SKB_DATA_ALIGN(fragsz);
399 if (in_irq() || irqs_disabled()) {
400 nc = this_cpu_ptr(&netdev_alloc_cache);
401 data = page_frag_alloc(nc, fragsz, GFP_ATOMIC);
404 data = __napi_alloc_frag(fragsz, GFP_ATOMIC);
409 EXPORT_SYMBOL(netdev_alloc_frag);
412 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
413 * @dev: network device to receive on
414 * @len: length to allocate
415 * @gfp_mask: get_free_pages mask, passed to alloc_skb
417 * Allocate a new &sk_buff and assign it a usage count of one. The
418 * buffer has NET_SKB_PAD headroom built in. Users should allocate
419 * the headroom they think they need without accounting for the
420 * built in space. The built in space is used for optimisations.
422 * %NULL is returned if there is no free memory.
424 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
427 struct page_frag_cache *nc;
434 /* If requested length is either too small or too big,
435 * we use kmalloc() for skb->head allocation.
437 if (len <= SKB_WITH_OVERHEAD(1024) ||
438 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
439 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
440 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
446 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
447 len = SKB_DATA_ALIGN(len);
449 if (sk_memalloc_socks())
450 gfp_mask |= __GFP_MEMALLOC;
452 if (in_irq() || irqs_disabled()) {
453 nc = this_cpu_ptr(&netdev_alloc_cache);
454 data = page_frag_alloc(nc, len, gfp_mask);
455 pfmemalloc = nc->pfmemalloc;
458 nc = this_cpu_ptr(&napi_alloc_cache.page);
459 data = page_frag_alloc(nc, len, gfp_mask);
460 pfmemalloc = nc->pfmemalloc;
467 skb = __build_skb(data, len);
468 if (unlikely(!skb)) {
473 /* use OR instead of assignment to avoid clearing of bits in mask */
479 skb_reserve(skb, NET_SKB_PAD);
485 EXPORT_SYMBOL(__netdev_alloc_skb);
488 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
489 * @napi: napi instance this buffer was allocated for
490 * @len: length to allocate
491 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
493 * Allocate a new sk_buff for use in NAPI receive. This buffer will
494 * attempt to allocate the head from a special reserved region used
495 * only for NAPI Rx allocation. By doing this we can save several
496 * CPU cycles by avoiding having to disable and re-enable IRQs.
498 * %NULL is returned if there is no free memory.
500 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
503 struct napi_alloc_cache *nc;
507 len += NET_SKB_PAD + NET_IP_ALIGN;
509 /* If requested length is either too small or too big,
510 * we use kmalloc() for skb->head allocation.
512 if (len <= SKB_WITH_OVERHEAD(1024) ||
513 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
514 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
515 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
521 nc = this_cpu_ptr(&napi_alloc_cache);
522 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
523 len = SKB_DATA_ALIGN(len);
525 if (sk_memalloc_socks())
526 gfp_mask |= __GFP_MEMALLOC;
528 data = page_frag_alloc(&nc->page, len, gfp_mask);
532 skb = __build_skb(data, len);
533 if (unlikely(!skb)) {
538 /* use OR instead of assignment to avoid clearing of bits in mask */
539 if (nc->page.pfmemalloc)
544 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
545 skb->dev = napi->dev;
550 EXPORT_SYMBOL(__napi_alloc_skb);
552 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
553 int size, unsigned int truesize)
555 skb_fill_page_desc(skb, i, page, off, size);
557 skb->data_len += size;
558 skb->truesize += truesize;
560 EXPORT_SYMBOL(skb_add_rx_frag);
562 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
563 unsigned int truesize)
565 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
567 skb_frag_size_add(frag, size);
569 skb->data_len += size;
570 skb->truesize += truesize;
572 EXPORT_SYMBOL(skb_coalesce_rx_frag);
574 static void skb_drop_list(struct sk_buff **listp)
576 kfree_skb_list(*listp);
580 static inline void skb_drop_fraglist(struct sk_buff *skb)
582 skb_drop_list(&skb_shinfo(skb)->frag_list);
585 static void skb_clone_fraglist(struct sk_buff *skb)
587 struct sk_buff *list;
589 skb_walk_frags(skb, list)
593 static void skb_free_head(struct sk_buff *skb)
595 unsigned char *head = skb->head;
603 static void skb_release_data(struct sk_buff *skb)
605 struct skb_shared_info *shinfo = skb_shinfo(skb);
609 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
613 for (i = 0; i < shinfo->nr_frags; i++)
614 __skb_frag_unref(&shinfo->frags[i]);
616 if (shinfo->frag_list)
617 kfree_skb_list(shinfo->frag_list);
619 skb_zcopy_clear(skb, true);
624 * Free an skbuff by memory without cleaning the state.
626 static void kfree_skbmem(struct sk_buff *skb)
628 struct sk_buff_fclones *fclones;
630 switch (skb->fclone) {
631 case SKB_FCLONE_UNAVAILABLE:
632 kmem_cache_free(skbuff_head_cache, skb);
635 case SKB_FCLONE_ORIG:
636 fclones = container_of(skb, struct sk_buff_fclones, skb1);
638 /* We usually free the clone (TX completion) before original skb
639 * This test would have no chance to be true for the clone,
640 * while here, branch prediction will be good.
642 if (refcount_read(&fclones->fclone_ref) == 1)
646 default: /* SKB_FCLONE_CLONE */
647 fclones = container_of(skb, struct sk_buff_fclones, skb2);
650 if (!refcount_dec_and_test(&fclones->fclone_ref))
653 kmem_cache_free(skbuff_fclone_cache, fclones);
656 void skb_release_head_state(struct sk_buff *skb)
659 if (skb->destructor) {
661 skb->destructor(skb);
663 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
664 nf_conntrack_put(skb_nfct(skb));
669 /* Free everything but the sk_buff shell. */
670 static void skb_release_all(struct sk_buff *skb)
672 skb_release_head_state(skb);
673 if (likely(skb->head))
674 skb_release_data(skb);
678 * __kfree_skb - private function
681 * Free an sk_buff. Release anything attached to the buffer.
682 * Clean the state. This is an internal helper function. Users should
683 * always call kfree_skb
686 void __kfree_skb(struct sk_buff *skb)
688 skb_release_all(skb);
691 EXPORT_SYMBOL(__kfree_skb);
694 * kfree_skb - free an sk_buff
695 * @skb: buffer to free
697 * Drop a reference to the buffer and free it if the usage count has
700 void kfree_skb(struct sk_buff *skb)
705 trace_kfree_skb(skb, __builtin_return_address(0));
708 EXPORT_SYMBOL(kfree_skb);
710 void kfree_skb_list(struct sk_buff *segs)
713 struct sk_buff *next = segs->next;
719 EXPORT_SYMBOL(kfree_skb_list);
721 /* Dump skb information and contents.
723 * Must only be called from net_ratelimit()-ed paths.
725 * Dumps up to can_dump_full whole packets if full_pkt, headers otherwise.
727 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
729 static atomic_t can_dump_full = ATOMIC_INIT(5);
730 struct skb_shared_info *sh = skb_shinfo(skb);
731 struct net_device *dev = skb->dev;
732 struct sock *sk = skb->sk;
733 struct sk_buff *list_skb;
734 bool has_mac, has_trans;
735 int headroom, tailroom;
739 full_pkt = atomic_dec_if_positive(&can_dump_full) >= 0;
744 len = min_t(int, skb->len, MAX_HEADER + 128);
746 headroom = skb_headroom(skb);
747 tailroom = skb_tailroom(skb);
749 has_mac = skb_mac_header_was_set(skb);
750 has_trans = skb_transport_header_was_set(skb);
752 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
753 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
754 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
755 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
756 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
757 level, skb->len, headroom, skb_headlen(skb), tailroom,
758 has_mac ? skb->mac_header : -1,
759 has_mac ? skb_mac_header_len(skb) : -1,
761 has_trans ? skb_network_header_len(skb) : -1,
762 has_trans ? skb->transport_header : -1,
763 sh->tx_flags, sh->nr_frags,
764 sh->gso_size, sh->gso_type, sh->gso_segs,
765 skb->csum, skb->ip_summed, skb->csum_complete_sw,
766 skb->csum_valid, skb->csum_level,
767 skb->hash, skb->sw_hash, skb->l4_hash,
768 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
771 printk("%sdev name=%s feat=%pNF\n",
772 level, dev->name, &dev->features);
774 printk("%ssk family=%hu type=%u proto=%u\n",
775 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
777 if (full_pkt && headroom)
778 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
779 16, 1, skb->head, headroom, false);
781 seg_len = min_t(int, skb_headlen(skb), len);
783 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
784 16, 1, skb->data, seg_len, false);
787 if (full_pkt && tailroom)
788 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
789 16, 1, skb_tail_pointer(skb), tailroom, false);
791 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
792 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
793 u32 p_off, p_len, copied;
797 skb_frag_foreach_page(frag, skb_frag_off(frag),
798 skb_frag_size(frag), p, p_off, p_len,
800 seg_len = min_t(int, p_len, len);
801 vaddr = kmap_atomic(p);
802 print_hex_dump(level, "skb frag: ",
804 16, 1, vaddr + p_off, seg_len, false);
805 kunmap_atomic(vaddr);
812 if (full_pkt && skb_has_frag_list(skb)) {
813 printk("skb fraglist:\n");
814 skb_walk_frags(skb, list_skb)
815 skb_dump(level, list_skb, true);
818 EXPORT_SYMBOL(skb_dump);
821 * skb_tx_error - report an sk_buff xmit error
822 * @skb: buffer that triggered an error
824 * Report xmit error if a device callback is tracking this skb.
825 * skb must be freed afterwards.
827 void skb_tx_error(struct sk_buff *skb)
829 skb_zcopy_clear(skb, true);
831 EXPORT_SYMBOL(skb_tx_error);
834 * consume_skb - free an skbuff
835 * @skb: buffer to free
837 * Drop a ref to the buffer and free it if the usage count has hit zero
838 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
839 * is being dropped after a failure and notes that
841 void consume_skb(struct sk_buff *skb)
846 trace_consume_skb(skb);
849 EXPORT_SYMBOL(consume_skb);
852 * consume_stateless_skb - free an skbuff, assuming it is stateless
853 * @skb: buffer to free
855 * Alike consume_skb(), but this variant assumes that this is the last
856 * skb reference and all the head states have been already dropped
858 void __consume_stateless_skb(struct sk_buff *skb)
860 trace_consume_skb(skb);
861 skb_release_data(skb);
865 void __kfree_skb_flush(void)
867 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
869 /* flush skb_cache if containing objects */
871 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
877 static inline void _kfree_skb_defer(struct sk_buff *skb)
879 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
881 /* drop skb->head and call any destructors for packet */
882 skb_release_all(skb);
884 /* record skb to CPU local list */
885 nc->skb_cache[nc->skb_count++] = skb;
888 /* SLUB writes into objects when freeing */
892 /* flush skb_cache if it is filled */
893 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
894 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
899 void __kfree_skb_defer(struct sk_buff *skb)
901 _kfree_skb_defer(skb);
904 void napi_consume_skb(struct sk_buff *skb, int budget)
909 /* Zero budget indicate non-NAPI context called us, like netpoll */
910 if (unlikely(!budget)) {
911 dev_consume_skb_any(skb);
918 /* if reaching here SKB is ready to free */
919 trace_consume_skb(skb);
921 /* if SKB is a clone, don't handle this case */
922 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
927 _kfree_skb_defer(skb);
929 EXPORT_SYMBOL(napi_consume_skb);
931 /* Make sure a field is enclosed inside headers_start/headers_end section */
932 #define CHECK_SKB_FIELD(field) \
933 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
934 offsetof(struct sk_buff, headers_start)); \
935 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
936 offsetof(struct sk_buff, headers_end)); \
938 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
940 new->tstamp = old->tstamp;
941 /* We do not copy old->sk */
943 memcpy(new->cb, old->cb, sizeof(old->cb));
944 skb_dst_copy(new, old);
945 __skb_ext_copy(new, old);
946 __nf_copy(new, old, false);
948 /* Note : this field could be in headers_start/headers_end section
949 * It is not yet because we do not want to have a 16 bit hole
951 new->queue_mapping = old->queue_mapping;
953 memcpy(&new->headers_start, &old->headers_start,
954 offsetof(struct sk_buff, headers_end) -
955 offsetof(struct sk_buff, headers_start));
956 CHECK_SKB_FIELD(protocol);
957 CHECK_SKB_FIELD(csum);
958 CHECK_SKB_FIELD(hash);
959 CHECK_SKB_FIELD(priority);
960 CHECK_SKB_FIELD(skb_iif);
961 CHECK_SKB_FIELD(vlan_proto);
962 CHECK_SKB_FIELD(vlan_tci);
963 CHECK_SKB_FIELD(transport_header);
964 CHECK_SKB_FIELD(network_header);
965 CHECK_SKB_FIELD(mac_header);
966 CHECK_SKB_FIELD(inner_protocol);
967 CHECK_SKB_FIELD(inner_transport_header);
968 CHECK_SKB_FIELD(inner_network_header);
969 CHECK_SKB_FIELD(inner_mac_header);
970 CHECK_SKB_FIELD(mark);
971 #ifdef CONFIG_NETWORK_SECMARK
972 CHECK_SKB_FIELD(secmark);
974 #ifdef CONFIG_NET_RX_BUSY_POLL
975 CHECK_SKB_FIELD(napi_id);
978 CHECK_SKB_FIELD(sender_cpu);
980 #ifdef CONFIG_NET_SCHED
981 CHECK_SKB_FIELD(tc_index);
987 * You should not add any new code to this function. Add it to
988 * __copy_skb_header above instead.
990 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
992 #define C(x) n->x = skb->x
994 n->next = n->prev = NULL;
996 __copy_skb_header(n, skb);
1001 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1006 n->destructor = NULL;
1013 refcount_set(&n->users, 1);
1015 atomic_inc(&(skb_shinfo(skb)->dataref));
1023 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1024 * @first: first sk_buff of the msg
1026 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1030 n = alloc_skb(0, GFP_ATOMIC);
1034 n->len = first->len;
1035 n->data_len = first->len;
1036 n->truesize = first->truesize;
1038 skb_shinfo(n)->frag_list = first;
1040 __copy_skb_header(n, first);
1041 n->destructor = NULL;
1045 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1048 * skb_morph - morph one skb into another
1049 * @dst: the skb to receive the contents
1050 * @src: the skb to supply the contents
1052 * This is identical to skb_clone except that the target skb is
1053 * supplied by the user.
1055 * The target skb is returned upon exit.
1057 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1059 skb_release_all(dst);
1060 return __skb_clone(dst, src);
1062 EXPORT_SYMBOL_GPL(skb_morph);
1064 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1066 unsigned long max_pg, num_pg, new_pg, old_pg;
1067 struct user_struct *user;
1069 if (capable(CAP_IPC_LOCK) || !size)
1072 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1073 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1074 user = mmp->user ? : current_user();
1077 old_pg = atomic_long_read(&user->locked_vm);
1078 new_pg = old_pg + num_pg;
1079 if (new_pg > max_pg)
1081 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1085 mmp->user = get_uid(user);
1086 mmp->num_pg = num_pg;
1088 mmp->num_pg += num_pg;
1093 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1095 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1098 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1099 free_uid(mmp->user);
1102 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1104 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
1106 struct ubuf_info *uarg;
1107 struct sk_buff *skb;
1109 WARN_ON_ONCE(!in_task());
1111 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1115 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1116 uarg = (void *)skb->cb;
1117 uarg->mmp.user = NULL;
1119 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1124 uarg->callback = sock_zerocopy_callback;
1125 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1127 uarg->bytelen = size;
1129 refcount_set(&uarg->refcnt, 1);
1134 EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
1136 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1138 return container_of((void *)uarg, struct sk_buff, cb);
1141 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
1142 struct ubuf_info *uarg)
1145 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1148 /* realloc only when socket is locked (TCP, UDP cork),
1149 * so uarg->len and sk_zckey access is serialized
1151 if (!sock_owned_by_user(sk)) {
1156 bytelen = uarg->bytelen + size;
1157 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1158 /* TCP can create new skb to attach new uarg */
1159 if (sk->sk_type == SOCK_STREAM)
1164 next = (u32)atomic_read(&sk->sk_zckey);
1165 if ((u32)(uarg->id + uarg->len) == next) {
1166 if (mm_account_pinned_pages(&uarg->mmp, size))
1169 uarg->bytelen = bytelen;
1170 atomic_set(&sk->sk_zckey, ++next);
1172 /* no extra ref when appending to datagram (MSG_MORE) */
1173 if (sk->sk_type == SOCK_STREAM)
1174 sock_zerocopy_get(uarg);
1181 return sock_zerocopy_alloc(sk, size);
1183 EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1185 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1187 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1191 old_lo = serr->ee.ee_info;
1192 old_hi = serr->ee.ee_data;
1193 sum_len = old_hi - old_lo + 1ULL + len;
1195 if (sum_len >= (1ULL << 32))
1198 if (lo != old_hi + 1)
1201 serr->ee.ee_data += len;
1205 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1207 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1208 struct sock_exterr_skb *serr;
1209 struct sock *sk = skb->sk;
1210 struct sk_buff_head *q;
1211 unsigned long flags;
1215 mm_unaccount_pinned_pages(&uarg->mmp);
1217 /* if !len, there was only 1 call, and it was aborted
1218 * so do not queue a completion notification
1220 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1225 hi = uarg->id + len - 1;
1227 serr = SKB_EXT_ERR(skb);
1228 memset(serr, 0, sizeof(*serr));
1229 serr->ee.ee_errno = 0;
1230 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1231 serr->ee.ee_data = hi;
1232 serr->ee.ee_info = lo;
1234 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1236 q = &sk->sk_error_queue;
1237 spin_lock_irqsave(&q->lock, flags);
1238 tail = skb_peek_tail(q);
1239 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1240 !skb_zerocopy_notify_extend(tail, lo, len)) {
1241 __skb_queue_tail(q, skb);
1244 spin_unlock_irqrestore(&q->lock, flags);
1246 sk->sk_error_report(sk);
1252 EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1254 void sock_zerocopy_put(struct ubuf_info *uarg)
1256 if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1258 uarg->callback(uarg, uarg->zerocopy);
1260 consume_skb(skb_from_uarg(uarg));
1263 EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1265 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1268 struct sock *sk = skb_from_uarg(uarg)->sk;
1270 atomic_dec(&sk->sk_zckey);
1274 sock_zerocopy_put(uarg);
1277 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1279 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1281 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1283 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1285 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1286 struct msghdr *msg, int len,
1287 struct ubuf_info *uarg)
1289 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1290 struct iov_iter orig_iter = msg->msg_iter;
1291 int err, orig_len = skb->len;
1293 /* An skb can only point to one uarg. This edge case happens when
1294 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1296 if (orig_uarg && uarg != orig_uarg)
1299 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1300 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1301 struct sock *save_sk = skb->sk;
1303 /* Streams do not free skb on error. Reset to prev state. */
1304 msg->msg_iter = orig_iter;
1306 ___pskb_trim(skb, orig_len);
1311 skb_zcopy_set(skb, uarg, NULL);
1312 return skb->len - orig_len;
1314 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1316 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1319 if (skb_zcopy(orig)) {
1320 if (skb_zcopy(nskb)) {
1321 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1326 if (skb_uarg(nskb) == skb_uarg(orig))
1328 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1331 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1337 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1338 * @skb: the skb to modify
1339 * @gfp_mask: allocation priority
1341 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1342 * It will copy all frags into kernel and drop the reference
1343 * to userspace pages.
1345 * If this function is called from an interrupt gfp_mask() must be
1348 * Returns 0 on success or a negative error code on failure
1349 * to allocate kernel memory to copy to.
1351 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1353 int num_frags = skb_shinfo(skb)->nr_frags;
1354 struct page *page, *head = NULL;
1358 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1364 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1365 for (i = 0; i < new_frags; i++) {
1366 page = alloc_page(gfp_mask);
1369 struct page *next = (struct page *)page_private(head);
1375 set_page_private(page, (unsigned long)head);
1381 for (i = 0; i < num_frags; i++) {
1382 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1383 u32 p_off, p_len, copied;
1387 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1388 p, p_off, p_len, copied) {
1390 vaddr = kmap_atomic(p);
1392 while (done < p_len) {
1393 if (d_off == PAGE_SIZE) {
1395 page = (struct page *)page_private(page);
1397 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1398 memcpy(page_address(page) + d_off,
1399 vaddr + p_off + done, copy);
1403 kunmap_atomic(vaddr);
1407 /* skb frags release userspace buffers */
1408 for (i = 0; i < num_frags; i++)
1409 skb_frag_unref(skb, i);
1411 /* skb frags point to kernel buffers */
1412 for (i = 0; i < new_frags - 1; i++) {
1413 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1414 head = (struct page *)page_private(head);
1416 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1417 skb_shinfo(skb)->nr_frags = new_frags;
1420 skb_zcopy_clear(skb, false);
1423 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1426 * skb_clone - duplicate an sk_buff
1427 * @skb: buffer to clone
1428 * @gfp_mask: allocation priority
1430 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1431 * copies share the same packet data but not structure. The new
1432 * buffer has a reference count of 1. If the allocation fails the
1433 * function returns %NULL otherwise the new buffer is returned.
1435 * If this function is called from an interrupt gfp_mask() must be
1439 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1441 struct sk_buff_fclones *fclones = container_of(skb,
1442 struct sk_buff_fclones,
1446 if (skb_orphan_frags(skb, gfp_mask))
1449 if (skb->fclone == SKB_FCLONE_ORIG &&
1450 refcount_read(&fclones->fclone_ref) == 1) {
1452 refcount_set(&fclones->fclone_ref, 2);
1454 if (skb_pfmemalloc(skb))
1455 gfp_mask |= __GFP_MEMALLOC;
1457 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1461 n->fclone = SKB_FCLONE_UNAVAILABLE;
1464 return __skb_clone(n, skb);
1466 EXPORT_SYMBOL(skb_clone);
1468 void skb_headers_offset_update(struct sk_buff *skb, int off)
1470 /* Only adjust this if it actually is csum_start rather than csum */
1471 if (skb->ip_summed == CHECKSUM_PARTIAL)
1472 skb->csum_start += off;
1473 /* {transport,network,mac}_header and tail are relative to skb->head */
1474 skb->transport_header += off;
1475 skb->network_header += off;
1476 if (skb_mac_header_was_set(skb))
1477 skb->mac_header += off;
1478 skb->inner_transport_header += off;
1479 skb->inner_network_header += off;
1480 skb->inner_mac_header += off;
1482 EXPORT_SYMBOL(skb_headers_offset_update);
1484 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1486 __copy_skb_header(new, old);
1488 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1489 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1490 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1492 EXPORT_SYMBOL(skb_copy_header);
1494 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1496 if (skb_pfmemalloc(skb))
1497 return SKB_ALLOC_RX;
1502 * skb_copy - create private copy of an sk_buff
1503 * @skb: buffer to copy
1504 * @gfp_mask: allocation priority
1506 * Make a copy of both an &sk_buff and its data. This is used when the
1507 * caller wishes to modify the data and needs a private copy of the
1508 * data to alter. Returns %NULL on failure or the pointer to the buffer
1509 * on success. The returned buffer has a reference count of 1.
1511 * As by-product this function converts non-linear &sk_buff to linear
1512 * one, so that &sk_buff becomes completely private and caller is allowed
1513 * to modify all the data of returned buffer. This means that this
1514 * function is not recommended for use in circumstances when only
1515 * header is going to be modified. Use pskb_copy() instead.
1518 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1520 int headerlen = skb_headroom(skb);
1521 unsigned int size = skb_end_offset(skb) + skb->data_len;
1522 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1523 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1528 /* Set the data pointer */
1529 skb_reserve(n, headerlen);
1530 /* Set the tail pointer and length */
1531 skb_put(n, skb->len);
1533 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1535 skb_copy_header(n, skb);
1538 EXPORT_SYMBOL(skb_copy);
1541 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1542 * @skb: buffer to copy
1543 * @headroom: headroom of new skb
1544 * @gfp_mask: allocation priority
1545 * @fclone: if true allocate the copy of the skb from the fclone
1546 * cache instead of the head cache; it is recommended to set this
1547 * to true for the cases where the copy will likely be cloned
1549 * Make a copy of both an &sk_buff and part of its data, located
1550 * in header. Fragmented data remain shared. This is used when
1551 * the caller wishes to modify only header of &sk_buff and needs
1552 * private copy of the header to alter. Returns %NULL on failure
1553 * or the pointer to the buffer on success.
1554 * The returned buffer has a reference count of 1.
1557 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1558 gfp_t gfp_mask, bool fclone)
1560 unsigned int size = skb_headlen(skb) + headroom;
1561 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1562 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1567 /* Set the data pointer */
1568 skb_reserve(n, headroom);
1569 /* Set the tail pointer and length */
1570 skb_put(n, skb_headlen(skb));
1571 /* Copy the bytes */
1572 skb_copy_from_linear_data(skb, n->data, n->len);
1574 n->truesize += skb->data_len;
1575 n->data_len = skb->data_len;
1578 if (skb_shinfo(skb)->nr_frags) {
1581 if (skb_orphan_frags(skb, gfp_mask) ||
1582 skb_zerocopy_clone(n, skb, gfp_mask)) {
1587 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1588 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1589 skb_frag_ref(skb, i);
1591 skb_shinfo(n)->nr_frags = i;
1594 if (skb_has_frag_list(skb)) {
1595 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1596 skb_clone_fraglist(n);
1599 skb_copy_header(n, skb);
1603 EXPORT_SYMBOL(__pskb_copy_fclone);
1606 * pskb_expand_head - reallocate header of &sk_buff
1607 * @skb: buffer to reallocate
1608 * @nhead: room to add at head
1609 * @ntail: room to add at tail
1610 * @gfp_mask: allocation priority
1612 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1613 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1614 * reference count of 1. Returns zero in the case of success or error,
1615 * if expansion failed. In the last case, &sk_buff is not changed.
1617 * All the pointers pointing into skb header may change and must be
1618 * reloaded after call to this function.
1621 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1624 int i, osize = skb_end_offset(skb);
1625 int size = osize + nhead + ntail;
1631 BUG_ON(skb_shared(skb));
1633 size = SKB_DATA_ALIGN(size);
1635 if (skb_pfmemalloc(skb))
1636 gfp_mask |= __GFP_MEMALLOC;
1637 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1638 gfp_mask, NUMA_NO_NODE, NULL);
1641 size = SKB_WITH_OVERHEAD(ksize(data));
1643 /* Copy only real data... and, alas, header. This should be
1644 * optimized for the cases when header is void.
1646 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1648 memcpy((struct skb_shared_info *)(data + size),
1650 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1653 * if shinfo is shared we must drop the old head gracefully, but if it
1654 * is not we can just drop the old head and let the existing refcount
1655 * be since all we did is relocate the values
1657 if (skb_cloned(skb)) {
1658 if (skb_orphan_frags(skb, gfp_mask))
1661 refcount_inc(&skb_uarg(skb)->refcnt);
1662 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1663 skb_frag_ref(skb, i);
1665 if (skb_has_frag_list(skb))
1666 skb_clone_fraglist(skb);
1668 skb_release_data(skb);
1672 off = (data + nhead) - skb->head;
1677 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1681 skb->end = skb->head + size;
1684 skb_headers_offset_update(skb, nhead);
1688 atomic_set(&skb_shinfo(skb)->dataref, 1);
1690 skb_metadata_clear(skb);
1692 /* It is not generally safe to change skb->truesize.
1693 * For the moment, we really care of rx path, or
1694 * when skb is orphaned (not attached to a socket).
1696 if (!skb->sk || skb->destructor == sock_edemux)
1697 skb->truesize += size - osize;
1706 EXPORT_SYMBOL(pskb_expand_head);
1708 /* Make private copy of skb with writable head and some headroom */
1710 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1712 struct sk_buff *skb2;
1713 int delta = headroom - skb_headroom(skb);
1716 skb2 = pskb_copy(skb, GFP_ATOMIC);
1718 skb2 = skb_clone(skb, GFP_ATOMIC);
1719 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1727 EXPORT_SYMBOL(skb_realloc_headroom);
1730 * skb_copy_expand - copy and expand sk_buff
1731 * @skb: buffer to copy
1732 * @newheadroom: new free bytes at head
1733 * @newtailroom: new free bytes at tail
1734 * @gfp_mask: allocation priority
1736 * Make a copy of both an &sk_buff and its data and while doing so
1737 * allocate additional space.
1739 * This is used when the caller wishes to modify the data and needs a
1740 * private copy of the data to alter as well as more space for new fields.
1741 * Returns %NULL on failure or the pointer to the buffer
1742 * on success. The returned buffer has a reference count of 1.
1744 * You must pass %GFP_ATOMIC as the allocation priority if this function
1745 * is called from an interrupt.
1747 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1748 int newheadroom, int newtailroom,
1752 * Allocate the copy buffer
1754 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1755 gfp_mask, skb_alloc_rx_flag(skb),
1757 int oldheadroom = skb_headroom(skb);
1758 int head_copy_len, head_copy_off;
1763 skb_reserve(n, newheadroom);
1765 /* Set the tail pointer and length */
1766 skb_put(n, skb->len);
1768 head_copy_len = oldheadroom;
1770 if (newheadroom <= head_copy_len)
1771 head_copy_len = newheadroom;
1773 head_copy_off = newheadroom - head_copy_len;
1775 /* Copy the linear header and data. */
1776 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1777 skb->len + head_copy_len));
1779 skb_copy_header(n, skb);
1781 skb_headers_offset_update(n, newheadroom - oldheadroom);
1785 EXPORT_SYMBOL(skb_copy_expand);
1788 * __skb_pad - zero pad the tail of an skb
1789 * @skb: buffer to pad
1790 * @pad: space to pad
1791 * @free_on_error: free buffer on error
1793 * Ensure that a buffer is followed by a padding area that is zero
1794 * filled. Used by network drivers which may DMA or transfer data
1795 * beyond the buffer end onto the wire.
1797 * May return error in out of memory cases. The skb is freed on error
1798 * if @free_on_error is true.
1801 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1806 /* If the skbuff is non linear tailroom is always zero.. */
1807 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1808 memset(skb->data+skb->len, 0, pad);
1812 ntail = skb->data_len + pad - (skb->end - skb->tail);
1813 if (likely(skb_cloned(skb) || ntail > 0)) {
1814 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1819 /* FIXME: The use of this function with non-linear skb's really needs
1822 err = skb_linearize(skb);
1826 memset(skb->data + skb->len, 0, pad);
1834 EXPORT_SYMBOL(__skb_pad);
1837 * pskb_put - add data to the tail of a potentially fragmented buffer
1838 * @skb: start of the buffer to use
1839 * @tail: tail fragment of the buffer to use
1840 * @len: amount of data to add
1842 * This function extends the used data area of the potentially
1843 * fragmented buffer. @tail must be the last fragment of @skb -- or
1844 * @skb itself. If this would exceed the total buffer size the kernel
1845 * will panic. A pointer to the first byte of the extra data is
1849 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1852 skb->data_len += len;
1855 return skb_put(tail, len);
1857 EXPORT_SYMBOL_GPL(pskb_put);
1860 * skb_put - add data to a buffer
1861 * @skb: buffer to use
1862 * @len: amount of data to add
1864 * This function extends the used data area of the buffer. If this would
1865 * exceed the total buffer size the kernel will panic. A pointer to the
1866 * first byte of the extra data is returned.
1868 void *skb_put(struct sk_buff *skb, unsigned int len)
1870 void *tmp = skb_tail_pointer(skb);
1871 SKB_LINEAR_ASSERT(skb);
1874 if (unlikely(skb->tail > skb->end))
1875 skb_over_panic(skb, len, __builtin_return_address(0));
1878 EXPORT_SYMBOL(skb_put);
1881 * skb_push - add data to the start of a buffer
1882 * @skb: buffer to use
1883 * @len: amount of data to add
1885 * This function extends the used data area of the buffer at the buffer
1886 * start. If this would exceed the total buffer headroom the kernel will
1887 * panic. A pointer to the first byte of the extra data is returned.
1889 void *skb_push(struct sk_buff *skb, unsigned int len)
1893 if (unlikely(skb->data < skb->head))
1894 skb_under_panic(skb, len, __builtin_return_address(0));
1897 EXPORT_SYMBOL(skb_push);
1900 * skb_pull - remove data from the start of a buffer
1901 * @skb: buffer to use
1902 * @len: amount of data to remove
1904 * This function removes data from the start of a buffer, returning
1905 * the memory to the headroom. A pointer to the next data in the buffer
1906 * is returned. Once the data has been pulled future pushes will overwrite
1909 void *skb_pull(struct sk_buff *skb, unsigned int len)
1911 return skb_pull_inline(skb, len);
1913 EXPORT_SYMBOL(skb_pull);
1916 * skb_trim - remove end from a buffer
1917 * @skb: buffer to alter
1920 * Cut the length of a buffer down by removing data from the tail. If
1921 * the buffer is already under the length specified it is not modified.
1922 * The skb must be linear.
1924 void skb_trim(struct sk_buff *skb, unsigned int len)
1927 __skb_trim(skb, len);
1929 EXPORT_SYMBOL(skb_trim);
1931 /* Trims skb to length len. It can change skb pointers.
1934 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1936 struct sk_buff **fragp;
1937 struct sk_buff *frag;
1938 int offset = skb_headlen(skb);
1939 int nfrags = skb_shinfo(skb)->nr_frags;
1943 if (skb_cloned(skb) &&
1944 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1951 for (; i < nfrags; i++) {
1952 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1959 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1962 skb_shinfo(skb)->nr_frags = i;
1964 for (; i < nfrags; i++)
1965 skb_frag_unref(skb, i);
1967 if (skb_has_frag_list(skb))
1968 skb_drop_fraglist(skb);
1972 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1973 fragp = &frag->next) {
1974 int end = offset + frag->len;
1976 if (skb_shared(frag)) {
1977 struct sk_buff *nfrag;
1979 nfrag = skb_clone(frag, GFP_ATOMIC);
1980 if (unlikely(!nfrag))
1983 nfrag->next = frag->next;
1995 unlikely((err = pskb_trim(frag, len - offset))))
1999 skb_drop_list(&frag->next);
2004 if (len > skb_headlen(skb)) {
2005 skb->data_len -= skb->len - len;
2010 skb_set_tail_pointer(skb, len);
2013 if (!skb->sk || skb->destructor == sock_edemux)
2017 EXPORT_SYMBOL(___pskb_trim);
2019 /* Note : use pskb_trim_rcsum() instead of calling this directly
2021 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2023 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2024 int delta = skb->len - len;
2026 skb->csum = csum_block_sub(skb->csum,
2027 skb_checksum(skb, len, delta, 0),
2029 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2030 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2031 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2033 if (offset + sizeof(__sum16) > hdlen)
2036 return __pskb_trim(skb, len);
2038 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2041 * __pskb_pull_tail - advance tail of skb header
2042 * @skb: buffer to reallocate
2043 * @delta: number of bytes to advance tail
2045 * The function makes a sense only on a fragmented &sk_buff,
2046 * it expands header moving its tail forward and copying necessary
2047 * data from fragmented part.
2049 * &sk_buff MUST have reference count of 1.
2051 * Returns %NULL (and &sk_buff does not change) if pull failed
2052 * or value of new tail of skb in the case of success.
2054 * All the pointers pointing into skb header may change and must be
2055 * reloaded after call to this function.
2058 /* Moves tail of skb head forward, copying data from fragmented part,
2059 * when it is necessary.
2060 * 1. It may fail due to malloc failure.
2061 * 2. It may change skb pointers.
2063 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2065 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2067 /* If skb has not enough free space at tail, get new one
2068 * plus 128 bytes for future expansions. If we have enough
2069 * room at tail, reallocate without expansion only if skb is cloned.
2071 int i, k, eat = (skb->tail + delta) - skb->end;
2073 if (eat > 0 || skb_cloned(skb)) {
2074 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2079 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2080 skb_tail_pointer(skb), delta));
2082 /* Optimization: no fragments, no reasons to preestimate
2083 * size of pulled pages. Superb.
2085 if (!skb_has_frag_list(skb))
2088 /* Estimate size of pulled pages. */
2090 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2091 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2098 /* If we need update frag list, we are in troubles.
2099 * Certainly, it is possible to add an offset to skb data,
2100 * but taking into account that pulling is expected to
2101 * be very rare operation, it is worth to fight against
2102 * further bloating skb head and crucify ourselves here instead.
2103 * Pure masohism, indeed. 8)8)
2106 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2107 struct sk_buff *clone = NULL;
2108 struct sk_buff *insp = NULL;
2111 if (list->len <= eat) {
2112 /* Eaten as whole. */
2117 /* Eaten partially. */
2118 if (skb_is_gso(skb) && !list->head_frag &&
2120 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2122 if (skb_shared(list)) {
2123 /* Sucks! We need to fork list. :-( */
2124 clone = skb_clone(list, GFP_ATOMIC);
2130 /* This may be pulled without
2134 if (!pskb_pull(list, eat)) {
2142 /* Free pulled out fragments. */
2143 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2144 skb_shinfo(skb)->frag_list = list->next;
2147 /* And insert new clone at head. */
2150 skb_shinfo(skb)->frag_list = clone;
2153 /* Success! Now we may commit changes to skb data. */
2158 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2159 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2162 skb_frag_unref(skb, i);
2165 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2167 *frag = skb_shinfo(skb)->frags[i];
2169 skb_frag_off_add(frag, eat);
2170 skb_frag_size_sub(frag, eat);
2178 skb_shinfo(skb)->nr_frags = k;
2182 skb->data_len -= delta;
2185 skb_zcopy_clear(skb, false);
2187 return skb_tail_pointer(skb);
2189 EXPORT_SYMBOL(__pskb_pull_tail);
2192 * skb_copy_bits - copy bits from skb to kernel buffer
2194 * @offset: offset in source
2195 * @to: destination buffer
2196 * @len: number of bytes to copy
2198 * Copy the specified number of bytes from the source skb to the
2199 * destination buffer.
2202 * If its prototype is ever changed,
2203 * check arch/{*}/net/{*}.S files,
2204 * since it is called from BPF assembly code.
2206 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2208 int start = skb_headlen(skb);
2209 struct sk_buff *frag_iter;
2212 if (offset > (int)skb->len - len)
2216 if ((copy = start - offset) > 0) {
2219 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2220 if ((len -= copy) == 0)
2226 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2228 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2230 WARN_ON(start > offset + len);
2232 end = start + skb_frag_size(f);
2233 if ((copy = end - offset) > 0) {
2234 u32 p_off, p_len, copied;
2241 skb_frag_foreach_page(f,
2242 skb_frag_off(f) + offset - start,
2243 copy, p, p_off, p_len, copied) {
2244 vaddr = kmap_atomic(p);
2245 memcpy(to + copied, vaddr + p_off, p_len);
2246 kunmap_atomic(vaddr);
2249 if ((len -= copy) == 0)
2257 skb_walk_frags(skb, frag_iter) {
2260 WARN_ON(start > offset + len);
2262 end = start + frag_iter->len;
2263 if ((copy = end - offset) > 0) {
2266 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2268 if ((len -= copy) == 0)
2282 EXPORT_SYMBOL(skb_copy_bits);
2285 * Callback from splice_to_pipe(), if we need to release some pages
2286 * at the end of the spd in case we error'ed out in filling the pipe.
2288 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2290 put_page(spd->pages[i]);
2293 static struct page *linear_to_page(struct page *page, unsigned int *len,
2294 unsigned int *offset,
2297 struct page_frag *pfrag = sk_page_frag(sk);
2299 if (!sk_page_frag_refill(sk, pfrag))
2302 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2304 memcpy(page_address(pfrag->page) + pfrag->offset,
2305 page_address(page) + *offset, *len);
2306 *offset = pfrag->offset;
2307 pfrag->offset += *len;
2312 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2314 unsigned int offset)
2316 return spd->nr_pages &&
2317 spd->pages[spd->nr_pages - 1] == page &&
2318 (spd->partial[spd->nr_pages - 1].offset +
2319 spd->partial[spd->nr_pages - 1].len == offset);
2323 * Fill page/offset/length into spd, if it can hold more pages.
2325 static bool spd_fill_page(struct splice_pipe_desc *spd,
2326 struct pipe_inode_info *pipe, struct page *page,
2327 unsigned int *len, unsigned int offset,
2331 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2335 page = linear_to_page(page, len, &offset, sk);
2339 if (spd_can_coalesce(spd, page, offset)) {
2340 spd->partial[spd->nr_pages - 1].len += *len;
2344 spd->pages[spd->nr_pages] = page;
2345 spd->partial[spd->nr_pages].len = *len;
2346 spd->partial[spd->nr_pages].offset = offset;
2352 static bool __splice_segment(struct page *page, unsigned int poff,
2353 unsigned int plen, unsigned int *off,
2355 struct splice_pipe_desc *spd, bool linear,
2357 struct pipe_inode_info *pipe)
2362 /* skip this segment if already processed */
2368 /* ignore any bits we already processed */
2374 unsigned int flen = min(*len, plen);
2376 if (spd_fill_page(spd, pipe, page, &flen, poff,
2382 } while (*len && plen);
2388 * Map linear and fragment data from the skb to spd. It reports true if the
2389 * pipe is full or if we already spliced the requested length.
2391 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2392 unsigned int *offset, unsigned int *len,
2393 struct splice_pipe_desc *spd, struct sock *sk)
2396 struct sk_buff *iter;
2398 /* map the linear part :
2399 * If skb->head_frag is set, this 'linear' part is backed by a
2400 * fragment, and if the head is not shared with any clones then
2401 * we can avoid a copy since we own the head portion of this page.
2403 if (__splice_segment(virt_to_page(skb->data),
2404 (unsigned long) skb->data & (PAGE_SIZE - 1),
2407 skb_head_is_locked(skb),
2412 * then map the fragments
2414 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2415 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2417 if (__splice_segment(skb_frag_page(f),
2418 skb_frag_off(f), skb_frag_size(f),
2419 offset, len, spd, false, sk, pipe))
2423 skb_walk_frags(skb, iter) {
2424 if (*offset >= iter->len) {
2425 *offset -= iter->len;
2428 /* __skb_splice_bits() only fails if the output has no room
2429 * left, so no point in going over the frag_list for the error
2432 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2440 * Map data from the skb to a pipe. Should handle both the linear part,
2441 * the fragments, and the frag list.
2443 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2444 struct pipe_inode_info *pipe, unsigned int tlen,
2447 struct partial_page partial[MAX_SKB_FRAGS];
2448 struct page *pages[MAX_SKB_FRAGS];
2449 struct splice_pipe_desc spd = {
2452 .nr_pages_max = MAX_SKB_FRAGS,
2453 .ops = &nosteal_pipe_buf_ops,
2454 .spd_release = sock_spd_release,
2458 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2461 ret = splice_to_pipe(pipe, &spd);
2465 EXPORT_SYMBOL_GPL(skb_splice_bits);
2467 /* Send skb data on a socket. Socket must be locked. */
2468 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2471 unsigned int orig_len = len;
2472 struct sk_buff *head = skb;
2473 unsigned short fragidx;
2478 /* Deal with head data */
2479 while (offset < skb_headlen(skb) && len) {
2483 slen = min_t(int, len, skb_headlen(skb) - offset);
2484 kv.iov_base = skb->data + offset;
2486 memset(&msg, 0, sizeof(msg));
2487 msg.msg_flags = MSG_DONTWAIT;
2489 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2497 /* All the data was skb head? */
2501 /* Make offset relative to start of frags */
2502 offset -= skb_headlen(skb);
2504 /* Find where we are in frag list */
2505 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2506 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2508 if (offset < skb_frag_size(frag))
2511 offset -= skb_frag_size(frag);
2514 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2515 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2517 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2520 ret = kernel_sendpage_locked(sk, skb_frag_page(frag),
2521 skb_frag_off(frag) + offset,
2522 slen, MSG_DONTWAIT);
2535 /* Process any frag lists */
2538 if (skb_has_frag_list(skb)) {
2539 skb = skb_shinfo(skb)->frag_list;
2542 } else if (skb->next) {
2549 return orig_len - len;
2552 return orig_len == len ? ret : orig_len - len;
2554 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2557 * skb_store_bits - store bits from kernel buffer to skb
2558 * @skb: destination buffer
2559 * @offset: offset in destination
2560 * @from: source buffer
2561 * @len: number of bytes to copy
2563 * Copy the specified number of bytes from the source buffer to the
2564 * destination skb. This function handles all the messy bits of
2565 * traversing fragment lists and such.
2568 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2570 int start = skb_headlen(skb);
2571 struct sk_buff *frag_iter;
2574 if (offset > (int)skb->len - len)
2577 if ((copy = start - offset) > 0) {
2580 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2581 if ((len -= copy) == 0)
2587 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2588 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2591 WARN_ON(start > offset + len);
2593 end = start + skb_frag_size(frag);
2594 if ((copy = end - offset) > 0) {
2595 u32 p_off, p_len, copied;
2602 skb_frag_foreach_page(frag,
2603 skb_frag_off(frag) + offset - start,
2604 copy, p, p_off, p_len, copied) {
2605 vaddr = kmap_atomic(p);
2606 memcpy(vaddr + p_off, from + copied, p_len);
2607 kunmap_atomic(vaddr);
2610 if ((len -= copy) == 0)
2618 skb_walk_frags(skb, frag_iter) {
2621 WARN_ON(start > offset + len);
2623 end = start + frag_iter->len;
2624 if ((copy = end - offset) > 0) {
2627 if (skb_store_bits(frag_iter, offset - start,
2630 if ((len -= copy) == 0)
2643 EXPORT_SYMBOL(skb_store_bits);
2645 /* Checksum skb data. */
2646 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2647 __wsum csum, const struct skb_checksum_ops *ops)
2649 int start = skb_headlen(skb);
2650 int i, copy = start - offset;
2651 struct sk_buff *frag_iter;
2654 /* Checksum header. */
2658 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2659 skb->data + offset, copy, csum);
2660 if ((len -= copy) == 0)
2666 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2668 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2670 WARN_ON(start > offset + len);
2672 end = start + skb_frag_size(frag);
2673 if ((copy = end - offset) > 0) {
2674 u32 p_off, p_len, copied;
2682 skb_frag_foreach_page(frag,
2683 skb_frag_off(frag) + offset - start,
2684 copy, p, p_off, p_len, copied) {
2685 vaddr = kmap_atomic(p);
2686 csum2 = INDIRECT_CALL_1(ops->update,
2688 vaddr + p_off, p_len, 0);
2689 kunmap_atomic(vaddr);
2690 csum = INDIRECT_CALL_1(ops->combine,
2691 csum_block_add_ext, csum,
2703 skb_walk_frags(skb, frag_iter) {
2706 WARN_ON(start > offset + len);
2708 end = start + frag_iter->len;
2709 if ((copy = end - offset) > 0) {
2713 csum2 = __skb_checksum(frag_iter, offset - start,
2715 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2716 csum, csum2, pos, copy);
2717 if ((len -= copy) == 0)
2728 EXPORT_SYMBOL(__skb_checksum);
2730 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2731 int len, __wsum csum)
2733 const struct skb_checksum_ops ops = {
2734 .update = csum_partial_ext,
2735 .combine = csum_block_add_ext,
2738 return __skb_checksum(skb, offset, len, csum, &ops);
2740 EXPORT_SYMBOL(skb_checksum);
2742 /* Both of above in one bottle. */
2744 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2745 u8 *to, int len, __wsum csum)
2747 int start = skb_headlen(skb);
2748 int i, copy = start - offset;
2749 struct sk_buff *frag_iter;
2756 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2758 if ((len -= copy) == 0)
2765 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2768 WARN_ON(start > offset + len);
2770 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2771 if ((copy = end - offset) > 0) {
2772 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2773 u32 p_off, p_len, copied;
2781 skb_frag_foreach_page(frag,
2782 skb_frag_off(frag) + offset - start,
2783 copy, p, p_off, p_len, copied) {
2784 vaddr = kmap_atomic(p);
2785 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2788 kunmap_atomic(vaddr);
2789 csum = csum_block_add(csum, csum2, pos);
2801 skb_walk_frags(skb, frag_iter) {
2805 WARN_ON(start > offset + len);
2807 end = start + frag_iter->len;
2808 if ((copy = end - offset) > 0) {
2811 csum2 = skb_copy_and_csum_bits(frag_iter,
2814 csum = csum_block_add(csum, csum2, pos);
2815 if ((len -= copy) == 0)
2826 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2828 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2832 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2833 /* See comments in __skb_checksum_complete(). */
2835 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2836 !skb->csum_complete_sw)
2837 netdev_rx_csum_fault(skb->dev, skb);
2839 if (!skb_shared(skb))
2840 skb->csum_valid = !sum;
2843 EXPORT_SYMBOL(__skb_checksum_complete_head);
2845 /* This function assumes skb->csum already holds pseudo header's checksum,
2846 * which has been changed from the hardware checksum, for example, by
2847 * __skb_checksum_validate_complete(). And, the original skb->csum must
2848 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2850 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2851 * zero. The new checksum is stored back into skb->csum unless the skb is
2854 __sum16 __skb_checksum_complete(struct sk_buff *skb)
2859 csum = skb_checksum(skb, 0, skb->len, 0);
2861 sum = csum_fold(csum_add(skb->csum, csum));
2862 /* This check is inverted, because we already knew the hardware
2863 * checksum is invalid before calling this function. So, if the
2864 * re-computed checksum is valid instead, then we have a mismatch
2865 * between the original skb->csum and skb_checksum(). This means either
2866 * the original hardware checksum is incorrect or we screw up skb->csum
2867 * when moving skb->data around.
2870 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2871 !skb->csum_complete_sw)
2872 netdev_rx_csum_fault(skb->dev, skb);
2875 if (!skb_shared(skb)) {
2876 /* Save full packet checksum */
2878 skb->ip_summed = CHECKSUM_COMPLETE;
2879 skb->csum_complete_sw = 1;
2880 skb->csum_valid = !sum;
2885 EXPORT_SYMBOL(__skb_checksum_complete);
2887 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2889 net_warn_ratelimited(
2890 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2895 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2896 int offset, int len)
2898 net_warn_ratelimited(
2899 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2904 static const struct skb_checksum_ops default_crc32c_ops = {
2905 .update = warn_crc32c_csum_update,
2906 .combine = warn_crc32c_csum_combine,
2909 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2910 &default_crc32c_ops;
2911 EXPORT_SYMBOL(crc32c_csum_stub);
2914 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2915 * @from: source buffer
2917 * Calculates the amount of linear headroom needed in the 'to' skb passed
2918 * into skb_zerocopy().
2921 skb_zerocopy_headlen(const struct sk_buff *from)
2923 unsigned int hlen = 0;
2925 if (!from->head_frag ||
2926 skb_headlen(from) < L1_CACHE_BYTES ||
2927 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) {
2928 hlen = skb_headlen(from);
2933 if (skb_has_frag_list(from))
2938 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2941 * skb_zerocopy - Zero copy skb to skb
2942 * @to: destination buffer
2943 * @from: source buffer
2944 * @len: number of bytes to copy from source buffer
2945 * @hlen: size of linear headroom in destination buffer
2947 * Copies up to `len` bytes from `from` to `to` by creating references
2948 * to the frags in the source buffer.
2950 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2951 * headroom in the `to` buffer.
2954 * 0: everything is OK
2955 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2956 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2959 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2962 int plen = 0; /* length of skb->head fragment */
2965 unsigned int offset;
2967 BUG_ON(!from->head_frag && !hlen);
2969 /* dont bother with small payloads */
2970 if (len <= skb_tailroom(to))
2971 return skb_copy_bits(from, 0, skb_put(to, len), len);
2974 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2979 plen = min_t(int, skb_headlen(from), len);
2981 page = virt_to_head_page(from->head);
2982 offset = from->data - (unsigned char *)page_address(page);
2983 __skb_fill_page_desc(to, 0, page, offset, plen);
2990 to->truesize += len + plen;
2991 to->len += len + plen;
2992 to->data_len += len + plen;
2994 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2998 skb_zerocopy_clone(to, from, GFP_ATOMIC);
3000 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3005 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3006 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3008 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3010 skb_frag_ref(to, j);
3013 skb_shinfo(to)->nr_frags = j;
3017 EXPORT_SYMBOL_GPL(skb_zerocopy);
3019 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3024 if (skb->ip_summed == CHECKSUM_PARTIAL)
3025 csstart = skb_checksum_start_offset(skb);
3027 csstart = skb_headlen(skb);
3029 BUG_ON(csstart > skb_headlen(skb));
3031 skb_copy_from_linear_data(skb, to, csstart);
3034 if (csstart != skb->len)
3035 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3036 skb->len - csstart, 0);
3038 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3039 long csstuff = csstart + skb->csum_offset;
3041 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3044 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3047 * skb_dequeue - remove from the head of the queue
3048 * @list: list to dequeue from
3050 * Remove the head of the list. The list lock is taken so the function
3051 * may be used safely with other locking list functions. The head item is
3052 * returned or %NULL if the list is empty.
3055 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3057 unsigned long flags;
3058 struct sk_buff *result;
3060 spin_lock_irqsave(&list->lock, flags);
3061 result = __skb_dequeue(list);
3062 spin_unlock_irqrestore(&list->lock, flags);
3065 EXPORT_SYMBOL(skb_dequeue);
3068 * skb_dequeue_tail - remove from the tail of the queue
3069 * @list: list to dequeue from
3071 * Remove the tail of the list. The list lock is taken so the function
3072 * may be used safely with other locking list functions. The tail item is
3073 * returned or %NULL if the list is empty.
3075 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3077 unsigned long flags;
3078 struct sk_buff *result;
3080 spin_lock_irqsave(&list->lock, flags);
3081 result = __skb_dequeue_tail(list);
3082 spin_unlock_irqrestore(&list->lock, flags);
3085 EXPORT_SYMBOL(skb_dequeue_tail);
3088 * skb_queue_purge - empty a list
3089 * @list: list to empty
3091 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3092 * the list and one reference dropped. This function takes the list
3093 * lock and is atomic with respect to other list locking functions.
3095 void skb_queue_purge(struct sk_buff_head *list)
3097 struct sk_buff *skb;
3098 while ((skb = skb_dequeue(list)) != NULL)
3101 EXPORT_SYMBOL(skb_queue_purge);
3104 * skb_rbtree_purge - empty a skb rbtree
3105 * @root: root of the rbtree to empty
3106 * Return value: the sum of truesizes of all purged skbs.
3108 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3109 * the list and one reference dropped. This function does not take
3110 * any lock. Synchronization should be handled by the caller (e.g., TCP
3111 * out-of-order queue is protected by the socket lock).
3113 unsigned int skb_rbtree_purge(struct rb_root *root)
3115 struct rb_node *p = rb_first(root);
3116 unsigned int sum = 0;
3119 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3122 rb_erase(&skb->rbnode, root);
3123 sum += skb->truesize;
3130 * skb_queue_head - queue a buffer at the list head
3131 * @list: list to use
3132 * @newsk: buffer to queue
3134 * Queue a buffer at the start of the list. This function takes the
3135 * list lock and can be used safely with other locking &sk_buff functions
3138 * A buffer cannot be placed on two lists at the same time.
3140 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3142 unsigned long flags;
3144 spin_lock_irqsave(&list->lock, flags);
3145 __skb_queue_head(list, newsk);
3146 spin_unlock_irqrestore(&list->lock, flags);
3148 EXPORT_SYMBOL(skb_queue_head);
3151 * skb_queue_tail - queue a buffer at the list tail
3152 * @list: list to use
3153 * @newsk: buffer to queue
3155 * Queue a buffer at the tail of the list. This function takes the
3156 * list lock and can be used safely with other locking &sk_buff functions
3159 * A buffer cannot be placed on two lists at the same time.
3161 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3163 unsigned long flags;
3165 spin_lock_irqsave(&list->lock, flags);
3166 __skb_queue_tail(list, newsk);
3167 spin_unlock_irqrestore(&list->lock, flags);
3169 EXPORT_SYMBOL(skb_queue_tail);
3172 * skb_unlink - remove a buffer from a list
3173 * @skb: buffer to remove
3174 * @list: list to use
3176 * Remove a packet from a list. The list locks are taken and this
3177 * function is atomic with respect to other list locked calls
3179 * You must know what list the SKB is on.
3181 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3183 unsigned long flags;
3185 spin_lock_irqsave(&list->lock, flags);
3186 __skb_unlink(skb, list);
3187 spin_unlock_irqrestore(&list->lock, flags);
3189 EXPORT_SYMBOL(skb_unlink);
3192 * skb_append - append a buffer
3193 * @old: buffer to insert after
3194 * @newsk: buffer to insert
3195 * @list: list to use
3197 * Place a packet after a given packet in a list. The list locks are taken
3198 * and this function is atomic with respect to other list locked calls.
3199 * A buffer cannot be placed on two lists at the same time.
3201 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3203 unsigned long flags;
3205 spin_lock_irqsave(&list->lock, flags);
3206 __skb_queue_after(list, old, newsk);
3207 spin_unlock_irqrestore(&list->lock, flags);
3209 EXPORT_SYMBOL(skb_append);
3211 static inline void skb_split_inside_header(struct sk_buff *skb,
3212 struct sk_buff* skb1,
3213 const u32 len, const int pos)
3217 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3219 /* And move data appendix as is. */
3220 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3221 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3223 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3224 skb_shinfo(skb)->nr_frags = 0;
3225 skb1->data_len = skb->data_len;
3226 skb1->len += skb1->data_len;
3229 skb_set_tail_pointer(skb, len);
3232 static inline void skb_split_no_header(struct sk_buff *skb,
3233 struct sk_buff* skb1,
3234 const u32 len, int pos)
3237 const int nfrags = skb_shinfo(skb)->nr_frags;
3239 skb_shinfo(skb)->nr_frags = 0;
3240 skb1->len = skb1->data_len = skb->len - len;
3242 skb->data_len = len - pos;
3244 for (i = 0; i < nfrags; i++) {
3245 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3247 if (pos + size > len) {
3248 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3252 * We have two variants in this case:
3253 * 1. Move all the frag to the second
3254 * part, if it is possible. F.e.
3255 * this approach is mandatory for TUX,
3256 * where splitting is expensive.
3257 * 2. Split is accurately. We make this.
3259 skb_frag_ref(skb, i);
3260 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3261 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3262 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3263 skb_shinfo(skb)->nr_frags++;
3267 skb_shinfo(skb)->nr_frags++;
3270 skb_shinfo(skb1)->nr_frags = k;
3274 * skb_split - Split fragmented skb to two parts at length len.
3275 * @skb: the buffer to split
3276 * @skb1: the buffer to receive the second part
3277 * @len: new length for skb
3279 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3281 int pos = skb_headlen(skb);
3283 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3285 skb_zerocopy_clone(skb1, skb, 0);
3286 if (len < pos) /* Split line is inside header. */
3287 skb_split_inside_header(skb, skb1, len, pos);
3288 else /* Second chunk has no header, nothing to copy. */
3289 skb_split_no_header(skb, skb1, len, pos);
3291 EXPORT_SYMBOL(skb_split);
3293 /* Shifting from/to a cloned skb is a no-go.
3295 * Caller cannot keep skb_shinfo related pointers past calling here!
3297 static int skb_prepare_for_shift(struct sk_buff *skb)
3301 if (skb_cloned(skb)) {
3302 /* Save and restore truesize: pskb_expand_head() may reallocate
3303 * memory where ksize(kmalloc(S)) != ksize(kmalloc(S)), but we
3304 * cannot change truesize at this point.
3306 unsigned int save_truesize = skb->truesize;
3308 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3309 skb->truesize = save_truesize;
3315 * skb_shift - Shifts paged data partially from skb to another
3316 * @tgt: buffer into which tail data gets added
3317 * @skb: buffer from which the paged data comes from
3318 * @shiftlen: shift up to this many bytes
3320 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3321 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3322 * It's up to caller to free skb if everything was shifted.
3324 * If @tgt runs out of frags, the whole operation is aborted.
3326 * Skb cannot include anything else but paged data while tgt is allowed
3327 * to have non-paged data as well.
3329 * TODO: full sized shift could be optimized but that would need
3330 * specialized skb free'er to handle frags without up-to-date nr_frags.
3332 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3334 int from, to, merge, todo;
3335 skb_frag_t *fragfrom, *fragto;
3337 BUG_ON(shiftlen > skb->len);
3339 if (skb_headlen(skb))
3341 if (skb_zcopy(tgt) || skb_zcopy(skb))
3346 to = skb_shinfo(tgt)->nr_frags;
3347 fragfrom = &skb_shinfo(skb)->frags[from];
3349 /* Actual merge is delayed until the point when we know we can
3350 * commit all, so that we don't have to undo partial changes
3353 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3354 skb_frag_off(fragfrom))) {
3359 todo -= skb_frag_size(fragfrom);
3361 if (skb_prepare_for_shift(skb) ||
3362 skb_prepare_for_shift(tgt))
3365 /* All previous frag pointers might be stale! */
3366 fragfrom = &skb_shinfo(skb)->frags[from];
3367 fragto = &skb_shinfo(tgt)->frags[merge];
3369 skb_frag_size_add(fragto, shiftlen);
3370 skb_frag_size_sub(fragfrom, shiftlen);
3371 skb_frag_off_add(fragfrom, shiftlen);
3379 /* Skip full, not-fitting skb to avoid expensive operations */
3380 if ((shiftlen == skb->len) &&
3381 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3384 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3387 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3388 if (to == MAX_SKB_FRAGS)
3391 fragfrom = &skb_shinfo(skb)->frags[from];
3392 fragto = &skb_shinfo(tgt)->frags[to];
3394 if (todo >= skb_frag_size(fragfrom)) {
3395 *fragto = *fragfrom;
3396 todo -= skb_frag_size(fragfrom);
3401 __skb_frag_ref(fragfrom);
3402 skb_frag_page_copy(fragto, fragfrom);
3403 skb_frag_off_copy(fragto, fragfrom);
3404 skb_frag_size_set(fragto, todo);
3406 skb_frag_off_add(fragfrom, todo);
3407 skb_frag_size_sub(fragfrom, todo);
3415 /* Ready to "commit" this state change to tgt */
3416 skb_shinfo(tgt)->nr_frags = to;
3419 fragfrom = &skb_shinfo(skb)->frags[0];
3420 fragto = &skb_shinfo(tgt)->frags[merge];
3422 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3423 __skb_frag_unref(fragfrom);
3426 /* Reposition in the original skb */
3428 while (from < skb_shinfo(skb)->nr_frags)
3429 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3430 skb_shinfo(skb)->nr_frags = to;
3432 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3435 /* Most likely the tgt won't ever need its checksum anymore, skb on
3436 * the other hand might need it if it needs to be resent
3438 tgt->ip_summed = CHECKSUM_PARTIAL;
3439 skb->ip_summed = CHECKSUM_PARTIAL;
3441 /* Yak, is it really working this way? Some helper please? */
3442 skb->len -= shiftlen;
3443 skb->data_len -= shiftlen;
3444 skb->truesize -= shiftlen;
3445 tgt->len += shiftlen;
3446 tgt->data_len += shiftlen;
3447 tgt->truesize += shiftlen;
3453 * skb_prepare_seq_read - Prepare a sequential read of skb data
3454 * @skb: the buffer to read
3455 * @from: lower offset of data to be read
3456 * @to: upper offset of data to be read
3457 * @st: state variable
3459 * Initializes the specified state variable. Must be called before
3460 * invoking skb_seq_read() for the first time.
3462 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3463 unsigned int to, struct skb_seq_state *st)
3465 st->lower_offset = from;
3466 st->upper_offset = to;
3467 st->root_skb = st->cur_skb = skb;
3468 st->frag_idx = st->stepped_offset = 0;
3469 st->frag_data = NULL;
3471 EXPORT_SYMBOL(skb_prepare_seq_read);
3474 * skb_seq_read - Sequentially read skb data
3475 * @consumed: number of bytes consumed by the caller so far
3476 * @data: destination pointer for data to be returned
3477 * @st: state variable
3479 * Reads a block of skb data at @consumed relative to the
3480 * lower offset specified to skb_prepare_seq_read(). Assigns
3481 * the head of the data block to @data and returns the length
3482 * of the block or 0 if the end of the skb data or the upper
3483 * offset has been reached.
3485 * The caller is not required to consume all of the data
3486 * returned, i.e. @consumed is typically set to the number
3487 * of bytes already consumed and the next call to
3488 * skb_seq_read() will return the remaining part of the block.
3490 * Note 1: The size of each block of data returned can be arbitrary,
3491 * this limitation is the cost for zerocopy sequential
3492 * reads of potentially non linear data.
3494 * Note 2: Fragment lists within fragments are not implemented
3495 * at the moment, state->root_skb could be replaced with
3496 * a stack for this purpose.
3498 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3499 struct skb_seq_state *st)
3501 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3504 if (unlikely(abs_offset >= st->upper_offset)) {
3505 if (st->frag_data) {
3506 kunmap_atomic(st->frag_data);
3507 st->frag_data = NULL;
3513 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3515 if (abs_offset < block_limit && !st->frag_data) {
3516 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3517 return block_limit - abs_offset;
3520 if (st->frag_idx == 0 && !st->frag_data)
3521 st->stepped_offset += skb_headlen(st->cur_skb);
3523 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3524 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3525 block_limit = skb_frag_size(frag) + st->stepped_offset;
3527 if (abs_offset < block_limit) {
3529 st->frag_data = kmap_atomic(skb_frag_page(frag));
3531 *data = (u8 *) st->frag_data + skb_frag_off(frag) +
3532 (abs_offset - st->stepped_offset);
3534 return block_limit - abs_offset;
3537 if (st->frag_data) {
3538 kunmap_atomic(st->frag_data);
3539 st->frag_data = NULL;
3543 st->stepped_offset += skb_frag_size(frag);
3546 if (st->frag_data) {
3547 kunmap_atomic(st->frag_data);
3548 st->frag_data = NULL;
3551 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3552 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3555 } else if (st->cur_skb->next) {
3556 st->cur_skb = st->cur_skb->next;
3563 EXPORT_SYMBOL(skb_seq_read);
3566 * skb_abort_seq_read - Abort a sequential read of skb data
3567 * @st: state variable
3569 * Must be called if skb_seq_read() was not called until it
3572 void skb_abort_seq_read(struct skb_seq_state *st)
3575 kunmap_atomic(st->frag_data);
3577 EXPORT_SYMBOL(skb_abort_seq_read);
3579 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3581 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3582 struct ts_config *conf,
3583 struct ts_state *state)
3585 return skb_seq_read(offset, text, TS_SKB_CB(state));
3588 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3590 skb_abort_seq_read(TS_SKB_CB(state));
3594 * skb_find_text - Find a text pattern in skb data
3595 * @skb: the buffer to look in
3596 * @from: search offset
3598 * @config: textsearch configuration
3600 * Finds a pattern in the skb data according to the specified
3601 * textsearch configuration. Use textsearch_next() to retrieve
3602 * subsequent occurrences of the pattern. Returns the offset
3603 * to the first occurrence or UINT_MAX if no match was found.
3605 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3606 unsigned int to, struct ts_config *config)
3608 struct ts_state state;
3611 config->get_next_block = skb_ts_get_next_block;
3612 config->finish = skb_ts_finish;
3614 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3616 ret = textsearch_find(config, &state);
3617 return (ret <= to - from ? ret : UINT_MAX);
3619 EXPORT_SYMBOL(skb_find_text);
3621 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3622 int offset, size_t size)
3624 int i = skb_shinfo(skb)->nr_frags;
3626 if (skb_can_coalesce(skb, i, page, offset)) {
3627 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3628 } else if (i < MAX_SKB_FRAGS) {
3630 skb_fill_page_desc(skb, i, page, offset, size);
3637 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3640 * skb_pull_rcsum - pull skb and update receive checksum
3641 * @skb: buffer to update
3642 * @len: length of data pulled
3644 * This function performs an skb_pull on the packet and updates
3645 * the CHECKSUM_COMPLETE checksum. It should be used on
3646 * receive path processing instead of skb_pull unless you know
3647 * that the checksum difference is zero (e.g., a valid IP header)
3648 * or you are setting ip_summed to CHECKSUM_NONE.
3650 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3652 unsigned char *data = skb->data;
3654 BUG_ON(len > skb->len);
3655 __skb_pull(skb, len);
3656 skb_postpull_rcsum(skb, data, len);
3659 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3661 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3663 skb_frag_t head_frag;
3666 page = virt_to_head_page(frag_skb->head);
3667 __skb_frag_set_page(&head_frag, page);
3668 skb_frag_off_set(&head_frag, frag_skb->data -
3669 (unsigned char *)page_address(page));
3670 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3675 * skb_segment - Perform protocol segmentation on skb.
3676 * @head_skb: buffer to segment
3677 * @features: features for the output path (see dev->features)
3679 * This function performs segmentation on the given skb. It returns
3680 * a pointer to the first in a list of new skbs for the segments.
3681 * In case of error it returns ERR_PTR(err).
3683 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3684 netdev_features_t features)
3686 struct sk_buff *segs = NULL;
3687 struct sk_buff *tail = NULL;
3688 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3689 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3690 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3691 unsigned int offset = doffset;
3692 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3693 unsigned int partial_segs = 0;
3694 unsigned int headroom;
3695 unsigned int len = head_skb->len;
3696 struct sk_buff *frag_skb;
3705 if ((skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY) &&
3706 mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) {
3707 struct sk_buff *check_skb;
3709 for (check_skb = list_skb; check_skb; check_skb = check_skb->next) {
3710 if (skb_headlen(check_skb) && !check_skb->head_frag) {
3711 /* gso_size is untrusted, and we have a frag_list with
3712 * a linear non head_frag item.
3714 * If head_skb's headlen does not fit requested gso_size,
3715 * it means that the frag_list members do NOT terminate
3716 * on exact gso_size boundaries. Hence we cannot perform
3717 * skb_frag_t page sharing. Therefore we must fallback to
3718 * copying the frag_list skbs; we do so by disabling SG.
3720 features &= ~NETIF_F_SG;
3726 __skb_push(head_skb, doffset);
3727 proto = skb_network_protocol(head_skb, &dummy);
3728 if (unlikely(!proto))
3729 return ERR_PTR(-EINVAL);
3731 sg = !!(features & NETIF_F_SG);
3732 csum = !!can_checksum_protocol(features, proto);
3734 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3735 if (!(features & NETIF_F_GSO_PARTIAL)) {
3736 struct sk_buff *iter;
3737 unsigned int frag_len;
3740 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3743 /* If we get here then all the required
3744 * GSO features except frag_list are supported.
3745 * Try to split the SKB to multiple GSO SKBs
3746 * with no frag_list.
3747 * Currently we can do that only when the buffers don't
3748 * have a linear part and all the buffers except
3749 * the last are of the same length.
3751 frag_len = list_skb->len;
3752 skb_walk_frags(head_skb, iter) {
3753 if (frag_len != iter->len && iter->next)
3755 if (skb_headlen(iter) && !iter->head_frag)
3761 if (len != frag_len)
3765 /* GSO partial only requires that we trim off any excess that
3766 * doesn't fit into an MSS sized block, so take care of that
3769 partial_segs = len / mss;
3770 if (partial_segs > 1)
3771 mss *= partial_segs;
3777 headroom = skb_headroom(head_skb);
3778 pos = skb_headlen(head_skb);
3780 if (skb_orphan_frags(head_skb, GFP_ATOMIC))
3781 return ERR_PTR(-ENOMEM);
3783 nfrags = skb_shinfo(head_skb)->nr_frags;
3784 frag = skb_shinfo(head_skb)->frags;
3785 frag_skb = head_skb;
3788 struct sk_buff *nskb;
3789 skb_frag_t *nskb_frag;
3793 if (unlikely(mss == GSO_BY_FRAGS)) {
3794 len = list_skb->len;
3796 len = head_skb->len - offset;
3801 hsize = skb_headlen(head_skb) - offset;
3804 if (hsize > len || !sg)
3807 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3808 (skb_headlen(list_skb) == len || sg)) {
3809 BUG_ON(skb_headlen(list_skb) > len);
3811 nskb = skb_clone(list_skb, GFP_ATOMIC);
3812 if (unlikely(!nskb))
3816 nfrags = skb_shinfo(list_skb)->nr_frags;
3817 frag = skb_shinfo(list_skb)->frags;
3818 frag_skb = list_skb;
3819 pos += skb_headlen(list_skb);
3821 while (pos < offset + len) {
3822 BUG_ON(i >= nfrags);
3824 size = skb_frag_size(frag);
3825 if (pos + size > offset + len)
3833 list_skb = list_skb->next;
3835 if (unlikely(pskb_trim(nskb, len))) {
3840 hsize = skb_end_offset(nskb);
3841 if (skb_cow_head(nskb, doffset + headroom)) {
3846 nskb->truesize += skb_end_offset(nskb) - hsize;
3847 skb_release_head_state(nskb);
3848 __skb_push(nskb, doffset);
3850 nskb = __alloc_skb(hsize + doffset + headroom,
3851 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3854 if (unlikely(!nskb))
3857 skb_reserve(nskb, headroom);
3858 __skb_put(nskb, doffset);
3867 __copy_skb_header(nskb, head_skb);
3869 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3870 skb_reset_mac_len(nskb);
3872 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3873 nskb->data - tnl_hlen,
3874 doffset + tnl_hlen);
3876 if (nskb->len == len + doffset)
3877 goto perform_csum_check;
3880 if (!nskb->remcsum_offload)
3881 nskb->ip_summed = CHECKSUM_NONE;
3882 SKB_GSO_CB(nskb)->csum =
3883 skb_copy_and_csum_bits(head_skb, offset,
3886 SKB_GSO_CB(nskb)->csum_start =
3887 skb_headroom(nskb) + doffset;
3891 nskb_frag = skb_shinfo(nskb)->frags;
3893 skb_copy_from_linear_data_offset(head_skb, offset,
3894 skb_put(nskb, hsize), hsize);
3896 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3899 if (skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
3902 while (pos < offset + len) {
3904 if (skb_orphan_frags(list_skb, GFP_ATOMIC) ||
3905 skb_zerocopy_clone(nskb, list_skb,
3910 nfrags = skb_shinfo(list_skb)->nr_frags;
3911 frag = skb_shinfo(list_skb)->frags;
3912 frag_skb = list_skb;
3913 if (!skb_headlen(list_skb)) {
3916 BUG_ON(!list_skb->head_frag);
3918 /* to make room for head_frag. */
3923 list_skb = list_skb->next;
3926 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3928 net_warn_ratelimited(
3929 "skb_segment: too many frags: %u %u\n",
3935 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
3936 __skb_frag_ref(nskb_frag);
3937 size = skb_frag_size(nskb_frag);
3940 skb_frag_off_add(nskb_frag, offset - pos);
3941 skb_frag_size_sub(nskb_frag, offset - pos);
3944 skb_shinfo(nskb)->nr_frags++;
3946 if (pos + size <= offset + len) {
3951 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3959 nskb->data_len = len - hsize;
3960 nskb->len += nskb->data_len;
3961 nskb->truesize += nskb->data_len;
3965 if (skb_has_shared_frag(nskb) &&
3966 __skb_linearize(nskb))
3969 if (!nskb->remcsum_offload)
3970 nskb->ip_summed = CHECKSUM_NONE;
3971 SKB_GSO_CB(nskb)->csum =
3972 skb_checksum(nskb, doffset,
3973 nskb->len - doffset, 0);
3974 SKB_GSO_CB(nskb)->csum_start =
3975 skb_headroom(nskb) + doffset;
3977 } while ((offset += len) < head_skb->len);
3979 /* Some callers want to get the end of the list.
3980 * Put it in segs->prev to avoid walking the list.
3981 * (see validate_xmit_skb_list() for example)
3986 struct sk_buff *iter;
3987 int type = skb_shinfo(head_skb)->gso_type;
3988 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
3990 /* Update type to add partial and then remove dodgy if set */
3991 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
3992 type &= ~SKB_GSO_DODGY;
3994 /* Update GSO info and prepare to start updating headers on
3995 * our way back down the stack of protocols.
3997 for (iter = segs; iter; iter = iter->next) {
3998 skb_shinfo(iter)->gso_size = gso_size;
3999 skb_shinfo(iter)->gso_segs = partial_segs;
4000 skb_shinfo(iter)->gso_type = type;
4001 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4004 if (tail->len - doffset <= gso_size)
4005 skb_shinfo(tail)->gso_size = 0;
4006 else if (tail != segs)
4007 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4010 /* Following permits correct backpressure, for protocols
4011 * using skb_set_owner_w().
4012 * Idea is to tranfert ownership from head_skb to last segment.
4014 if (head_skb->destructor == sock_wfree) {
4015 swap(tail->truesize, head_skb->truesize);
4016 swap(tail->destructor, head_skb->destructor);
4017 swap(tail->sk, head_skb->sk);
4022 kfree_skb_list(segs);
4023 return ERR_PTR(err);
4025 EXPORT_SYMBOL_GPL(skb_segment);
4027 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4029 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4030 unsigned int offset = skb_gro_offset(skb);
4031 unsigned int headlen = skb_headlen(skb);
4032 unsigned int len = skb_gro_len(skb);
4033 unsigned int delta_truesize;
4036 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4039 lp = NAPI_GRO_CB(p)->last;
4040 pinfo = skb_shinfo(lp);
4042 if (headlen <= offset) {
4045 int i = skbinfo->nr_frags;
4046 int nr_frags = pinfo->nr_frags + i;
4048 if (nr_frags > MAX_SKB_FRAGS)
4052 pinfo->nr_frags = nr_frags;
4053 skbinfo->nr_frags = 0;
4055 frag = pinfo->frags + nr_frags;
4056 frag2 = skbinfo->frags + i;
4061 skb_frag_off_add(frag, offset);
4062 skb_frag_size_sub(frag, offset);
4064 /* all fragments truesize : remove (head size + sk_buff) */
4065 delta_truesize = skb->truesize -
4066 SKB_TRUESIZE(skb_end_offset(skb));
4068 skb->truesize -= skb->data_len;
4069 skb->len -= skb->data_len;
4072 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4074 } else if (skb->head_frag) {
4075 int nr_frags = pinfo->nr_frags;
4076 skb_frag_t *frag = pinfo->frags + nr_frags;
4077 struct page *page = virt_to_head_page(skb->head);
4078 unsigned int first_size = headlen - offset;
4079 unsigned int first_offset;
4081 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4084 first_offset = skb->data -
4085 (unsigned char *)page_address(page) +
4088 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4090 __skb_frag_set_page(frag, page);
4091 skb_frag_off_set(frag, first_offset);
4092 skb_frag_size_set(frag, first_size);
4094 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4095 /* We dont need to clear skbinfo->nr_frags here */
4097 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4098 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4103 delta_truesize = skb->truesize;
4104 if (offset > headlen) {
4105 unsigned int eat = offset - headlen;
4107 skb_frag_off_add(&skbinfo->frags[0], eat);
4108 skb_frag_size_sub(&skbinfo->frags[0], eat);
4109 skb->data_len -= eat;
4114 __skb_pull(skb, offset);
4116 if (NAPI_GRO_CB(p)->last == p)
4117 skb_shinfo(p)->frag_list = skb;
4119 NAPI_GRO_CB(p)->last->next = skb;
4120 NAPI_GRO_CB(p)->last = skb;
4121 __skb_header_release(skb);
4125 NAPI_GRO_CB(p)->count++;
4127 p->truesize += delta_truesize;
4130 lp->data_len += len;
4131 lp->truesize += delta_truesize;
4134 NAPI_GRO_CB(skb)->same_flow = 1;
4137 EXPORT_SYMBOL_GPL(skb_gro_receive);
4139 #ifdef CONFIG_SKB_EXTENSIONS
4140 #define SKB_EXT_ALIGN_VALUE 8
4141 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4143 static const u8 skb_ext_type_len[] = {
4144 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4145 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4148 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4150 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4151 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4155 static __always_inline unsigned int skb_ext_total_length(void)
4157 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4158 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4159 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4162 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4164 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4165 skb_ext_type_len[TC_SKB_EXT] +
4170 static void skb_extensions_init(void)
4172 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4173 BUILD_BUG_ON(skb_ext_total_length() > 255);
4175 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4176 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4178 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4182 static void skb_extensions_init(void) {}
4185 void __init skb_init(void)
4187 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4188 sizeof(struct sk_buff),
4190 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4191 offsetof(struct sk_buff, cb),
4192 sizeof_field(struct sk_buff, cb),
4194 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4195 sizeof(struct sk_buff_fclones),
4197 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4199 skb_extensions_init();
4203 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4204 unsigned int recursion_level)
4206 int start = skb_headlen(skb);
4207 int i, copy = start - offset;
4208 struct sk_buff *frag_iter;
4211 if (unlikely(recursion_level >= 24))
4217 sg_set_buf(sg, skb->data + offset, copy);
4219 if ((len -= copy) == 0)
4224 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4227 WARN_ON(start > offset + len);
4229 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4230 if ((copy = end - offset) > 0) {
4231 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4232 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4237 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4238 skb_frag_off(frag) + offset - start);
4247 skb_walk_frags(skb, frag_iter) {
4250 WARN_ON(start > offset + len);
4252 end = start + frag_iter->len;
4253 if ((copy = end - offset) > 0) {
4254 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4259 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4260 copy, recursion_level + 1);
4261 if (unlikely(ret < 0))
4264 if ((len -= copy) == 0)
4275 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4276 * @skb: Socket buffer containing the buffers to be mapped
4277 * @sg: The scatter-gather list to map into
4278 * @offset: The offset into the buffer's contents to start mapping
4279 * @len: Length of buffer space to be mapped
4281 * Fill the specified scatter-gather list with mappings/pointers into a
4282 * region of the buffer space attached to a socket buffer. Returns either
4283 * the number of scatterlist items used, or -EMSGSIZE if the contents
4286 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4288 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4293 sg_mark_end(&sg[nsg - 1]);
4297 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4299 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4300 * sglist without mark the sg which contain last skb data as the end.
4301 * So the caller can mannipulate sg list as will when padding new data after
4302 * the first call without calling sg_unmark_end to expend sg list.
4304 * Scenario to use skb_to_sgvec_nomark:
4306 * 2. skb_to_sgvec_nomark(payload1)
4307 * 3. skb_to_sgvec_nomark(payload2)
4309 * This is equivalent to:
4311 * 2. skb_to_sgvec(payload1)
4313 * 4. skb_to_sgvec(payload2)
4315 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4316 * is more preferable.
4318 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4319 int offset, int len)
4321 return __skb_to_sgvec(skb, sg, offset, len, 0);
4323 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4328 * skb_cow_data - Check that a socket buffer's data buffers are writable
4329 * @skb: The socket buffer to check.
4330 * @tailbits: Amount of trailing space to be added
4331 * @trailer: Returned pointer to the skb where the @tailbits space begins
4333 * Make sure that the data buffers attached to a socket buffer are
4334 * writable. If they are not, private copies are made of the data buffers
4335 * and the socket buffer is set to use these instead.
4337 * If @tailbits is given, make sure that there is space to write @tailbits
4338 * bytes of data beyond current end of socket buffer. @trailer will be
4339 * set to point to the skb in which this space begins.
4341 * The number of scatterlist elements required to completely map the
4342 * COW'd and extended socket buffer will be returned.
4344 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4348 struct sk_buff *skb1, **skb_p;
4350 /* If skb is cloned or its head is paged, reallocate
4351 * head pulling out all the pages (pages are considered not writable
4352 * at the moment even if they are anonymous).
4354 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4355 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
4358 /* Easy case. Most of packets will go this way. */
4359 if (!skb_has_frag_list(skb)) {
4360 /* A little of trouble, not enough of space for trailer.
4361 * This should not happen, when stack is tuned to generate
4362 * good frames. OK, on miss we reallocate and reserve even more
4363 * space, 128 bytes is fair. */
4365 if (skb_tailroom(skb) < tailbits &&
4366 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4374 /* Misery. We are in troubles, going to mincer fragments... */
4377 skb_p = &skb_shinfo(skb)->frag_list;
4380 while ((skb1 = *skb_p) != NULL) {
4383 /* The fragment is partially pulled by someone,
4384 * this can happen on input. Copy it and everything
4387 if (skb_shared(skb1))
4390 /* If the skb is the last, worry about trailer. */
4392 if (skb1->next == NULL && tailbits) {
4393 if (skb_shinfo(skb1)->nr_frags ||
4394 skb_has_frag_list(skb1) ||
4395 skb_tailroom(skb1) < tailbits)
4396 ntail = tailbits + 128;
4402 skb_shinfo(skb1)->nr_frags ||
4403 skb_has_frag_list(skb1)) {
4404 struct sk_buff *skb2;
4406 /* Fuck, we are miserable poor guys... */
4408 skb2 = skb_copy(skb1, GFP_ATOMIC);
4410 skb2 = skb_copy_expand(skb1,
4414 if (unlikely(skb2 == NULL))
4418 skb_set_owner_w(skb2, skb1->sk);
4420 /* Looking around. Are we still alive?
4421 * OK, link new skb, drop old one */
4423 skb2->next = skb1->next;
4430 skb_p = &skb1->next;
4435 EXPORT_SYMBOL_GPL(skb_cow_data);
4437 static void sock_rmem_free(struct sk_buff *skb)
4439 struct sock *sk = skb->sk;
4441 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4444 static void skb_set_err_queue(struct sk_buff *skb)
4446 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4447 * So, it is safe to (mis)use it to mark skbs on the error queue.
4449 skb->pkt_type = PACKET_OUTGOING;
4450 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4454 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4456 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4458 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4459 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4464 skb->destructor = sock_rmem_free;
4465 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4466 skb_set_err_queue(skb);
4468 /* before exiting rcu section, make sure dst is refcounted */
4471 skb_queue_tail(&sk->sk_error_queue, skb);
4472 if (!sock_flag(sk, SOCK_DEAD))
4473 sk->sk_error_report(sk);
4476 EXPORT_SYMBOL(sock_queue_err_skb);
4478 static bool is_icmp_err_skb(const struct sk_buff *skb)
4480 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4481 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4484 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4486 struct sk_buff_head *q = &sk->sk_error_queue;
4487 struct sk_buff *skb, *skb_next = NULL;
4488 bool icmp_next = false;
4489 unsigned long flags;
4491 spin_lock_irqsave(&q->lock, flags);
4492 skb = __skb_dequeue(q);
4493 if (skb && (skb_next = skb_peek(q))) {
4494 icmp_next = is_icmp_err_skb(skb_next);
4496 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
4498 spin_unlock_irqrestore(&q->lock, flags);
4500 if (is_icmp_err_skb(skb) && !icmp_next)
4504 sk->sk_error_report(sk);
4508 EXPORT_SYMBOL(sock_dequeue_err_skb);
4511 * skb_clone_sk - create clone of skb, and take reference to socket
4512 * @skb: the skb to clone
4514 * This function creates a clone of a buffer that holds a reference on
4515 * sk_refcnt. Buffers created via this function are meant to be
4516 * returned using sock_queue_err_skb, or free via kfree_skb.
4518 * When passing buffers allocated with this function to sock_queue_err_skb
4519 * it is necessary to wrap the call with sock_hold/sock_put in order to
4520 * prevent the socket from being released prior to being enqueued on
4521 * the sk_error_queue.
4523 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4525 struct sock *sk = skb->sk;
4526 struct sk_buff *clone;
4528 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4531 clone = skb_clone(skb, GFP_ATOMIC);
4538 clone->destructor = sock_efree;
4542 EXPORT_SYMBOL(skb_clone_sk);
4544 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4549 struct sock_exterr_skb *serr;
4552 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4554 serr = SKB_EXT_ERR(skb);
4555 memset(serr, 0, sizeof(*serr));
4556 serr->ee.ee_errno = ENOMSG;
4557 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4558 serr->ee.ee_info = tstype;
4559 serr->opt_stats = opt_stats;
4560 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4561 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4562 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4563 if (sk->sk_protocol == IPPROTO_TCP &&
4564 sk->sk_type == SOCK_STREAM)
4565 serr->ee.ee_data -= sk->sk_tskey;
4568 err = sock_queue_err_skb(sk, skb);
4574 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4578 if (likely(READ_ONCE(sysctl_tstamp_allow_data) || tsonly))
4581 read_lock_bh(&sk->sk_callback_lock);
4582 ret = sk->sk_socket && sk->sk_socket->file &&
4583 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4584 read_unlock_bh(&sk->sk_callback_lock);
4588 void skb_complete_tx_timestamp(struct sk_buff *skb,
4589 struct skb_shared_hwtstamps *hwtstamps)
4591 struct sock *sk = skb->sk;
4593 if (!skb_may_tx_timestamp(sk, false))
4596 /* Take a reference to prevent skb_orphan() from freeing the socket,
4597 * but only if the socket refcount is not zero.
4599 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4600 *skb_hwtstamps(skb) = *hwtstamps;
4601 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4609 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4611 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4612 struct skb_shared_hwtstamps *hwtstamps,
4613 struct sock *sk, int tstype)
4615 struct sk_buff *skb;
4616 bool tsonly, opt_stats = false;
4621 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4622 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4625 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4626 if (!skb_may_tx_timestamp(sk, tsonly))
4631 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4632 sk->sk_protocol == IPPROTO_TCP &&
4633 sk->sk_type == SOCK_STREAM) {
4634 skb = tcp_get_timestamping_opt_stats(sk);
4638 skb = alloc_skb(0, GFP_ATOMIC);
4640 skb = skb_clone(orig_skb, GFP_ATOMIC);
4642 if (skb_orphan_frags_rx(skb, GFP_ATOMIC)) {
4651 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4653 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4657 *skb_hwtstamps(skb) = *hwtstamps;
4659 skb->tstamp = ktime_get_real();
4661 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4663 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4665 void skb_tstamp_tx(struct sk_buff *orig_skb,
4666 struct skb_shared_hwtstamps *hwtstamps)
4668 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4671 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4673 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4675 struct sock *sk = skb->sk;
4676 struct sock_exterr_skb *serr;
4679 skb->wifi_acked_valid = 1;
4680 skb->wifi_acked = acked;
4682 serr = SKB_EXT_ERR(skb);
4683 memset(serr, 0, sizeof(*serr));
4684 serr->ee.ee_errno = ENOMSG;
4685 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4687 /* Take a reference to prevent skb_orphan() from freeing the socket,
4688 * but only if the socket refcount is not zero.
4690 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4691 err = sock_queue_err_skb(sk, skb);
4697 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4700 * skb_partial_csum_set - set up and verify partial csum values for packet
4701 * @skb: the skb to set
4702 * @start: the number of bytes after skb->data to start checksumming.
4703 * @off: the offset from start to place the checksum.
4705 * For untrusted partially-checksummed packets, we need to make sure the values
4706 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4708 * This function checks and sets those values and skb->ip_summed: if this
4709 * returns false you should drop the packet.
4711 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4713 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4714 u32 csum_start = skb_headroom(skb) + (u32)start;
4716 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4717 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4718 start, off, skb_headroom(skb), skb_headlen(skb));
4721 skb->ip_summed = CHECKSUM_PARTIAL;
4722 skb->csum_start = csum_start;
4723 skb->csum_offset = off;
4724 skb_set_transport_header(skb, start);
4727 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4729 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4732 if (skb_headlen(skb) >= len)
4735 /* If we need to pullup then pullup to the max, so we
4736 * won't need to do it again.
4741 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4744 if (skb_headlen(skb) < len)
4750 #define MAX_TCP_HDR_LEN (15 * 4)
4752 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4753 typeof(IPPROTO_IP) proto,
4760 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4761 off + MAX_TCP_HDR_LEN);
4762 if (!err && !skb_partial_csum_set(skb, off,
4763 offsetof(struct tcphdr,
4766 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4769 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4770 off + sizeof(struct udphdr));
4771 if (!err && !skb_partial_csum_set(skb, off,
4772 offsetof(struct udphdr,
4775 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4778 return ERR_PTR(-EPROTO);
4781 /* This value should be large enough to cover a tagged ethernet header plus
4782 * maximally sized IP and TCP or UDP headers.
4784 #define MAX_IP_HDR_LEN 128
4786 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4795 err = skb_maybe_pull_tail(skb,
4796 sizeof(struct iphdr),
4801 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4804 off = ip_hdrlen(skb);
4811 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4813 return PTR_ERR(csum);
4816 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4819 ip_hdr(skb)->protocol, 0);
4826 /* This value should be large enough to cover a tagged ethernet header plus
4827 * an IPv6 header, all options, and a maximal TCP or UDP header.
4829 #define MAX_IPV6_HDR_LEN 256
4831 #define OPT_HDR(type, skb, off) \
4832 (type *)(skb_network_header(skb) + (off))
4834 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4847 off = sizeof(struct ipv6hdr);
4849 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4853 nexthdr = ipv6_hdr(skb)->nexthdr;
4855 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4856 while (off <= len && !done) {
4858 case IPPROTO_DSTOPTS:
4859 case IPPROTO_HOPOPTS:
4860 case IPPROTO_ROUTING: {
4861 struct ipv6_opt_hdr *hp;
4863 err = skb_maybe_pull_tail(skb,
4865 sizeof(struct ipv6_opt_hdr),
4870 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4871 nexthdr = hp->nexthdr;
4872 off += ipv6_optlen(hp);
4876 struct ip_auth_hdr *hp;
4878 err = skb_maybe_pull_tail(skb,
4880 sizeof(struct ip_auth_hdr),
4885 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4886 nexthdr = hp->nexthdr;
4887 off += ipv6_authlen(hp);
4890 case IPPROTO_FRAGMENT: {
4891 struct frag_hdr *hp;
4893 err = skb_maybe_pull_tail(skb,
4895 sizeof(struct frag_hdr),
4900 hp = OPT_HDR(struct frag_hdr, skb, off);
4902 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4905 nexthdr = hp->nexthdr;
4906 off += sizeof(struct frag_hdr);
4917 if (!done || fragment)
4920 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4922 return PTR_ERR(csum);
4925 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4926 &ipv6_hdr(skb)->daddr,
4927 skb->len - off, nexthdr, 0);
4935 * skb_checksum_setup - set up partial checksum offset
4936 * @skb: the skb to set up
4937 * @recalculate: if true the pseudo-header checksum will be recalculated
4939 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4943 switch (skb->protocol) {
4944 case htons(ETH_P_IP):
4945 err = skb_checksum_setup_ipv4(skb, recalculate);
4948 case htons(ETH_P_IPV6):
4949 err = skb_checksum_setup_ipv6(skb, recalculate);
4959 EXPORT_SYMBOL(skb_checksum_setup);
4962 * skb_checksum_maybe_trim - maybe trims the given skb
4963 * @skb: the skb to check
4964 * @transport_len: the data length beyond the network header
4966 * Checks whether the given skb has data beyond the given transport length.
4967 * If so, returns a cloned skb trimmed to this transport length.
4968 * Otherwise returns the provided skb. Returns NULL in error cases
4969 * (e.g. transport_len exceeds skb length or out-of-memory).
4971 * Caller needs to set the skb transport header and free any returned skb if it
4972 * differs from the provided skb.
4974 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4975 unsigned int transport_len)
4977 struct sk_buff *skb_chk;
4978 unsigned int len = skb_transport_offset(skb) + transport_len;
4983 else if (skb->len == len)
4986 skb_chk = skb_clone(skb, GFP_ATOMIC);
4990 ret = pskb_trim_rcsum(skb_chk, len);
5000 * skb_checksum_trimmed - validate checksum of an skb
5001 * @skb: the skb to check
5002 * @transport_len: the data length beyond the network header
5003 * @skb_chkf: checksum function to use
5005 * Applies the given checksum function skb_chkf to the provided skb.
5006 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5008 * If the skb has data beyond the given transport length, then a
5009 * trimmed & cloned skb is checked and returned.
5011 * Caller needs to set the skb transport header and free any returned skb if it
5012 * differs from the provided skb.
5014 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5015 unsigned int transport_len,
5016 __sum16(*skb_chkf)(struct sk_buff *skb))
5018 struct sk_buff *skb_chk;
5019 unsigned int offset = skb_transport_offset(skb);
5022 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5026 if (!pskb_may_pull(skb_chk, offset))
5029 skb_pull_rcsum(skb_chk, offset);
5030 ret = skb_chkf(skb_chk);
5031 skb_push_rcsum(skb_chk, offset);
5039 if (skb_chk && skb_chk != skb)
5045 EXPORT_SYMBOL(skb_checksum_trimmed);
5047 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5049 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5052 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5054 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5057 skb_release_head_state(skb);
5058 kmem_cache_free(skbuff_head_cache, skb);
5063 EXPORT_SYMBOL(kfree_skb_partial);
5066 * skb_try_coalesce - try to merge skb to prior one
5068 * @from: buffer to add
5069 * @fragstolen: pointer to boolean
5070 * @delta_truesize: how much more was allocated than was requested
5072 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5073 bool *fragstolen, int *delta_truesize)
5075 struct skb_shared_info *to_shinfo, *from_shinfo;
5076 int i, delta, len = from->len;
5078 *fragstolen = false;
5083 if (len <= skb_tailroom(to)) {
5085 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5086 *delta_truesize = 0;
5090 to_shinfo = skb_shinfo(to);
5091 from_shinfo = skb_shinfo(from);
5092 if (to_shinfo->frag_list || from_shinfo->frag_list)
5094 if (skb_zcopy(to) || skb_zcopy(from))
5097 if (skb_headlen(from) != 0) {
5099 unsigned int offset;
5101 if (to_shinfo->nr_frags +
5102 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5105 if (skb_head_is_locked(from))
5108 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5110 page = virt_to_head_page(from->head);
5111 offset = from->data - (unsigned char *)page_address(page);
5113 skb_fill_page_desc(to, to_shinfo->nr_frags,
5114 page, offset, skb_headlen(from));
5117 if (to_shinfo->nr_frags +
5118 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5121 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5124 WARN_ON_ONCE(delta < len);
5126 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5128 from_shinfo->nr_frags * sizeof(skb_frag_t));
5129 to_shinfo->nr_frags += from_shinfo->nr_frags;
5131 if (!skb_cloned(from))
5132 from_shinfo->nr_frags = 0;
5134 /* if the skb is not cloned this does nothing
5135 * since we set nr_frags to 0.
5137 for (i = 0; i < from_shinfo->nr_frags; i++)
5138 __skb_frag_ref(&from_shinfo->frags[i]);
5140 to->truesize += delta;
5142 to->data_len += len;
5144 *delta_truesize = delta;
5147 EXPORT_SYMBOL(skb_try_coalesce);
5150 * skb_scrub_packet - scrub an skb
5152 * @skb: buffer to clean
5153 * @xnet: packet is crossing netns
5155 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5156 * into/from a tunnel. Some information have to be cleared during these
5158 * skb_scrub_packet can also be used to clean a skb before injecting it in
5159 * another namespace (@xnet == true). We have to clear all information in the
5160 * skb that could impact namespace isolation.
5162 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5164 skb->pkt_type = PACKET_HOST;
5170 nf_reset_trace(skb);
5172 #ifdef CONFIG_NET_SWITCHDEV
5173 skb->offload_fwd_mark = 0;
5174 skb->offload_l3_fwd_mark = 0;
5184 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5187 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5191 * skb_gso_transport_seglen is used to determine the real size of the
5192 * individual segments, including Layer4 headers (TCP/UDP).
5194 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5196 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5198 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5199 unsigned int thlen = 0;
5201 if (skb->encapsulation) {
5202 thlen = skb_inner_transport_header(skb) -
5203 skb_transport_header(skb);
5205 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5206 thlen += inner_tcp_hdrlen(skb);
5207 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5208 thlen = tcp_hdrlen(skb);
5209 } else if (unlikely(skb_is_gso_sctp(skb))) {
5210 thlen = sizeof(struct sctphdr);
5211 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5212 thlen = sizeof(struct udphdr);
5214 /* UFO sets gso_size to the size of the fragmentation
5215 * payload, i.e. the size of the L4 (UDP) header is already
5218 return thlen + shinfo->gso_size;
5222 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5226 * skb_gso_network_seglen is used to determine the real size of the
5227 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5229 * The MAC/L2 header is not accounted for.
5231 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5233 unsigned int hdr_len = skb_transport_header(skb) -
5234 skb_network_header(skb);
5236 return hdr_len + skb_gso_transport_seglen(skb);
5240 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5244 * skb_gso_mac_seglen is used to determine the real size of the
5245 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5246 * headers (TCP/UDP).
5248 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5250 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5252 return hdr_len + skb_gso_transport_seglen(skb);
5256 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5258 * There are a couple of instances where we have a GSO skb, and we
5259 * want to determine what size it would be after it is segmented.
5261 * We might want to check:
5262 * - L3+L4+payload size (e.g. IP forwarding)
5263 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5265 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5269 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5270 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5272 * @max_len: The maximum permissible length.
5274 * Returns true if the segmented length <= max length.
5276 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5277 unsigned int seg_len,
5278 unsigned int max_len) {
5279 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5280 const struct sk_buff *iter;
5282 if (shinfo->gso_size != GSO_BY_FRAGS)
5283 return seg_len <= max_len;
5285 /* Undo this so we can re-use header sizes */
5286 seg_len -= GSO_BY_FRAGS;
5288 skb_walk_frags(skb, iter) {
5289 if (seg_len + skb_headlen(iter) > max_len)
5297 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5300 * @mtu: MTU to validate against
5302 * skb_gso_validate_network_len validates if a given skb will fit a
5303 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5306 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5308 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5310 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5313 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5316 * @len: length to validate against
5318 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5319 * length once split, including L2, L3 and L4 headers and the payload.
5321 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5323 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5325 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5327 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5329 int mac_len, meta_len;
5332 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5337 mac_len = skb->data - skb_mac_header(skb);
5338 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5339 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5340 mac_len - VLAN_HLEN - ETH_TLEN);
5343 meta_len = skb_metadata_len(skb);
5345 meta = skb_metadata_end(skb) - meta_len;
5346 memmove(meta + VLAN_HLEN, meta, meta_len);
5349 skb->mac_header += VLAN_HLEN;
5353 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5355 struct vlan_hdr *vhdr;
5358 if (unlikely(skb_vlan_tag_present(skb))) {
5359 /* vlan_tci is already set-up so leave this for another time */
5363 skb = skb_share_check(skb, GFP_ATOMIC);
5366 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5367 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5370 vhdr = (struct vlan_hdr *)skb->data;
5371 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5372 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5374 skb_pull_rcsum(skb, VLAN_HLEN);
5375 vlan_set_encap_proto(skb, vhdr);
5377 skb = skb_reorder_vlan_header(skb);
5381 skb_reset_network_header(skb);
5382 skb_reset_transport_header(skb);
5383 skb_reset_mac_len(skb);
5391 EXPORT_SYMBOL(skb_vlan_untag);
5393 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5395 if (!pskb_may_pull(skb, write_len))
5398 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5401 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5403 EXPORT_SYMBOL(skb_ensure_writable);
5405 /* remove VLAN header from packet and update csum accordingly.
5406 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5408 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5410 struct vlan_hdr *vhdr;
5411 int offset = skb->data - skb_mac_header(skb);
5414 if (WARN_ONCE(offset,
5415 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5420 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5424 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5426 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5427 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5429 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5430 __skb_pull(skb, VLAN_HLEN);
5432 vlan_set_encap_proto(skb, vhdr);
5433 skb->mac_header += VLAN_HLEN;
5435 if (skb_network_offset(skb) < ETH_HLEN)
5436 skb_set_network_header(skb, ETH_HLEN);
5438 skb_reset_mac_len(skb);
5442 EXPORT_SYMBOL(__skb_vlan_pop);
5444 /* Pop a vlan tag either from hwaccel or from payload.
5445 * Expects skb->data at mac header.
5447 int skb_vlan_pop(struct sk_buff *skb)
5453 if (likely(skb_vlan_tag_present(skb))) {
5454 __vlan_hwaccel_clear_tag(skb);
5456 if (unlikely(!eth_type_vlan(skb->protocol)))
5459 err = __skb_vlan_pop(skb, &vlan_tci);
5463 /* move next vlan tag to hw accel tag */
5464 if (likely(!eth_type_vlan(skb->protocol)))
5467 vlan_proto = skb->protocol;
5468 err = __skb_vlan_pop(skb, &vlan_tci);
5472 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5475 EXPORT_SYMBOL(skb_vlan_pop);
5477 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5478 * Expects skb->data at mac header.
5480 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5482 if (skb_vlan_tag_present(skb)) {
5483 int offset = skb->data - skb_mac_header(skb);
5486 if (WARN_ONCE(offset,
5487 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5492 err = __vlan_insert_tag(skb, skb->vlan_proto,
5493 skb_vlan_tag_get(skb));
5497 skb->protocol = skb->vlan_proto;
5498 skb->mac_len += VLAN_HLEN;
5500 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5502 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5505 EXPORT_SYMBOL(skb_vlan_push);
5507 /* Update the ethertype of hdr and the skb csum value if required. */
5508 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5511 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5512 __be16 diff[] = { ~hdr->h_proto, ethertype };
5514 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5517 hdr->h_proto = ethertype;
5521 * skb_mpls_push() - push a new MPLS header after the mac header
5524 * @mpls_lse: MPLS label stack entry to push
5525 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5526 * @mac_len: length of the MAC header
5528 * Expects skb->data at mac header.
5530 * Returns 0 on success, -errno otherwise.
5532 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5533 int mac_len, bool ethernet)
5535 struct mpls_shim_hdr *lse;
5538 if (unlikely(!eth_p_mpls(mpls_proto)))
5541 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5542 if (skb->encapsulation)
5545 err = skb_cow_head(skb, MPLS_HLEN);
5549 if (!skb->inner_protocol) {
5550 skb_set_inner_network_header(skb, mac_len);
5551 skb_set_inner_protocol(skb, skb->protocol);
5554 skb_push(skb, MPLS_HLEN);
5555 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5557 skb_reset_mac_header(skb);
5558 skb_set_network_header(skb, mac_len);
5560 lse = mpls_hdr(skb);
5561 lse->label_stack_entry = mpls_lse;
5562 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5564 if (ethernet && mac_len >= ETH_HLEN)
5565 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5566 skb->protocol = mpls_proto;
5570 EXPORT_SYMBOL_GPL(skb_mpls_push);
5573 * skb_mpls_pop() - pop the outermost MPLS header
5576 * @next_proto: ethertype of header after popped MPLS header
5577 * @mac_len: length of the MAC header
5578 * @ethernet: flag to indicate if ethernet header is present in packet
5580 * Expects skb->data at mac header.
5582 * Returns 0 on success, -errno otherwise.
5584 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5589 if (unlikely(!eth_p_mpls(skb->protocol)))
5592 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5596 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5597 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5600 __skb_pull(skb, MPLS_HLEN);
5601 skb_reset_mac_header(skb);
5602 skb_set_network_header(skb, mac_len);
5604 if (ethernet && mac_len >= ETH_HLEN) {
5607 /* use mpls_hdr() to get ethertype to account for VLANs. */
5608 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5609 skb_mod_eth_type(skb, hdr, next_proto);
5611 skb->protocol = next_proto;
5615 EXPORT_SYMBOL_GPL(skb_mpls_pop);
5618 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5621 * @mpls_lse: new MPLS label stack entry to update to
5623 * Expects skb->data at mac header.
5625 * Returns 0 on success, -errno otherwise.
5627 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5631 if (unlikely(!eth_p_mpls(skb->protocol)))
5634 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5638 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5639 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5641 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5644 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5648 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5651 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5655 * Expects skb->data at mac header.
5657 * Returns 0 on success, -errno otherwise.
5659 int skb_mpls_dec_ttl(struct sk_buff *skb)
5664 if (unlikely(!eth_p_mpls(skb->protocol)))
5667 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
5670 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5671 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5675 lse &= ~MPLS_LS_TTL_MASK;
5676 lse |= ttl << MPLS_LS_TTL_SHIFT;
5678 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5680 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5683 * alloc_skb_with_frags - allocate skb with page frags
5685 * @header_len: size of linear part
5686 * @data_len: needed length in frags
5687 * @max_page_order: max page order desired.
5688 * @errcode: pointer to error code if any
5689 * @gfp_mask: allocation mask
5691 * This can be used to allocate a paged skb, given a maximal order for frags.
5693 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5694 unsigned long data_len,
5699 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5700 unsigned long chunk;
5701 struct sk_buff *skb;
5705 *errcode = -EMSGSIZE;
5706 /* Note this test could be relaxed, if we succeed to allocate
5707 * high order pages...
5709 if (npages > MAX_SKB_FRAGS)
5712 *errcode = -ENOBUFS;
5713 skb = alloc_skb(header_len, gfp_mask);
5717 skb->truesize += npages << PAGE_SHIFT;
5719 for (i = 0; npages > 0; i++) {
5720 int order = max_page_order;
5723 if (npages >= 1 << order) {
5724 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5730 /* Do not retry other high order allocations */
5736 page = alloc_page(gfp_mask);
5740 chunk = min_t(unsigned long, data_len,
5741 PAGE_SIZE << order);
5742 skb_fill_page_desc(skb, i, page, 0, chunk);
5744 npages -= 1 << order;
5752 EXPORT_SYMBOL(alloc_skb_with_frags);
5754 /* carve out the first off bytes from skb when off < headlen */
5755 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5756 const int headlen, gfp_t gfp_mask)
5759 int size = skb_end_offset(skb);
5760 int new_hlen = headlen - off;
5763 size = SKB_DATA_ALIGN(size);
5765 if (skb_pfmemalloc(skb))
5766 gfp_mask |= __GFP_MEMALLOC;
5767 data = kmalloc_reserve(size +
5768 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5769 gfp_mask, NUMA_NO_NODE, NULL);
5773 size = SKB_WITH_OVERHEAD(ksize(data));
5775 /* Copy real data, and all frags */
5776 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5779 memcpy((struct skb_shared_info *)(data + size),
5781 offsetof(struct skb_shared_info,
5782 frags[skb_shinfo(skb)->nr_frags]));
5783 if (skb_cloned(skb)) {
5784 /* drop the old head gracefully */
5785 if (skb_orphan_frags(skb, gfp_mask)) {
5789 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5790 skb_frag_ref(skb, i);
5791 if (skb_has_frag_list(skb))
5792 skb_clone_fraglist(skb);
5793 skb_release_data(skb);
5795 /* we can reuse existing recount- all we did was
5804 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5807 skb->end = skb->head + size;
5809 skb_set_tail_pointer(skb, skb_headlen(skb));
5810 skb_headers_offset_update(skb, 0);
5814 atomic_set(&skb_shinfo(skb)->dataref, 1);
5819 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
5821 /* carve out the first eat bytes from skb's frag_list. May recurse into
5824 static int pskb_carve_frag_list(struct sk_buff *skb,
5825 struct skb_shared_info *shinfo, int eat,
5828 struct sk_buff *list = shinfo->frag_list;
5829 struct sk_buff *clone = NULL;
5830 struct sk_buff *insp = NULL;
5834 pr_err("Not enough bytes to eat. Want %d\n", eat);
5837 if (list->len <= eat) {
5838 /* Eaten as whole. */
5843 /* Eaten partially. */
5844 if (skb_shared(list)) {
5845 clone = skb_clone(list, gfp_mask);
5851 /* This may be pulled without problems. */
5854 if (pskb_carve(list, eat, gfp_mask) < 0) {
5862 /* Free pulled out fragments. */
5863 while ((list = shinfo->frag_list) != insp) {
5864 shinfo->frag_list = list->next;
5867 /* And insert new clone at head. */
5870 shinfo->frag_list = clone;
5875 /* carve off first len bytes from skb. Split line (off) is in the
5876 * non-linear part of skb
5878 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
5879 int pos, gfp_t gfp_mask)
5882 int size = skb_end_offset(skb);
5884 const int nfrags = skb_shinfo(skb)->nr_frags;
5885 struct skb_shared_info *shinfo;
5887 size = SKB_DATA_ALIGN(size);
5889 if (skb_pfmemalloc(skb))
5890 gfp_mask |= __GFP_MEMALLOC;
5891 data = kmalloc_reserve(size +
5892 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5893 gfp_mask, NUMA_NO_NODE, NULL);
5897 size = SKB_WITH_OVERHEAD(ksize(data));
5899 memcpy((struct skb_shared_info *)(data + size),
5900 skb_shinfo(skb), offsetof(struct skb_shared_info,
5901 frags[skb_shinfo(skb)->nr_frags]));
5902 if (skb_orphan_frags(skb, gfp_mask)) {
5906 shinfo = (struct skb_shared_info *)(data + size);
5907 for (i = 0; i < nfrags; i++) {
5908 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
5910 if (pos + fsize > off) {
5911 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
5915 * We have two variants in this case:
5916 * 1. Move all the frag to the second
5917 * part, if it is possible. F.e.
5918 * this approach is mandatory for TUX,
5919 * where splitting is expensive.
5920 * 2. Split is accurately. We make this.
5922 skb_frag_off_add(&shinfo->frags[0], off - pos);
5923 skb_frag_size_sub(&shinfo->frags[0], off - pos);
5925 skb_frag_ref(skb, i);
5930 shinfo->nr_frags = k;
5931 if (skb_has_frag_list(skb))
5932 skb_clone_fraglist(skb);
5934 /* split line is in frag list */
5935 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
5936 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
5937 if (skb_has_frag_list(skb))
5938 kfree_skb_list(skb_shinfo(skb)->frag_list);
5942 skb_release_data(skb);
5947 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5950 skb->end = skb->head + size;
5952 skb_reset_tail_pointer(skb);
5953 skb_headers_offset_update(skb, 0);
5958 skb->data_len = skb->len;
5959 atomic_set(&skb_shinfo(skb)->dataref, 1);
5963 /* remove len bytes from the beginning of the skb */
5964 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
5966 int headlen = skb_headlen(skb);
5969 return pskb_carve_inside_header(skb, len, headlen, gfp);
5971 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
5974 /* Extract to_copy bytes starting at off from skb, and return this in
5977 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
5978 int to_copy, gfp_t gfp)
5980 struct sk_buff *clone = skb_clone(skb, gfp);
5985 if (pskb_carve(clone, off, gfp) < 0 ||
5986 pskb_trim(clone, to_copy)) {
5992 EXPORT_SYMBOL(pskb_extract);
5995 * skb_condense - try to get rid of fragments/frag_list if possible
5998 * Can be used to save memory before skb is added to a busy queue.
5999 * If packet has bytes in frags and enough tail room in skb->head,
6000 * pull all of them, so that we can free the frags right now and adjust
6003 * We do not reallocate skb->head thus can not fail.
6004 * Caller must re-evaluate skb->truesize if needed.
6006 void skb_condense(struct sk_buff *skb)
6008 if (skb->data_len) {
6009 if (skb->data_len > skb->end - skb->tail ||
6013 /* Nice, we can free page frag(s) right now */
6014 __pskb_pull_tail(skb, skb->data_len);
6016 /* At this point, skb->truesize might be over estimated,
6017 * because skb had a fragment, and fragments do not tell
6019 * When we pulled its content into skb->head, fragment
6020 * was freed, but __pskb_pull_tail() could not possibly
6021 * adjust skb->truesize, not knowing the frag truesize.
6023 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6026 #ifdef CONFIG_SKB_EXTENSIONS
6027 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6029 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6032 static struct skb_ext *skb_ext_alloc(void)
6034 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6037 memset(new->offset, 0, sizeof(new->offset));
6038 refcount_set(&new->refcnt, 1);
6044 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6045 unsigned int old_active)
6047 struct skb_ext *new;
6049 if (refcount_read(&old->refcnt) == 1)
6052 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6056 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6057 refcount_set(&new->refcnt, 1);
6060 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6061 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6064 for (i = 0; i < sp->len; i++)
6065 xfrm_state_hold(sp->xvec[i]);
6073 * skb_ext_add - allocate space for given extension, COW if needed
6075 * @id: extension to allocate space for
6077 * Allocates enough space for the given extension.
6078 * If the extension is already present, a pointer to that extension
6081 * If the skb was cloned, COW applies and the returned memory can be
6082 * modified without changing the extension space of clones buffers.
6084 * Returns pointer to the extension or NULL on allocation failure.
6086 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6088 struct skb_ext *new, *old = NULL;
6089 unsigned int newlen, newoff;
6091 if (skb->active_extensions) {
6092 old = skb->extensions;
6094 new = skb_ext_maybe_cow(old, skb->active_extensions);
6098 if (__skb_ext_exist(new, id))
6101 newoff = new->chunks;
6103 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6105 new = skb_ext_alloc();
6110 newlen = newoff + skb_ext_type_len[id];
6111 new->chunks = newlen;
6112 new->offset[id] = newoff;
6114 skb->extensions = new;
6115 skb->active_extensions |= 1 << id;
6116 return skb_ext_get_ptr(new, id);
6118 EXPORT_SYMBOL(skb_ext_add);
6121 static void skb_ext_put_sp(struct sec_path *sp)
6125 for (i = 0; i < sp->len; i++)
6126 xfrm_state_put(sp->xvec[i]);
6130 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6132 struct skb_ext *ext = skb->extensions;
6134 skb->active_extensions &= ~(1 << id);
6135 if (skb->active_extensions == 0) {
6136 skb->extensions = NULL;
6139 } else if (id == SKB_EXT_SEC_PATH &&
6140 refcount_read(&ext->refcnt) == 1) {
6141 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6148 EXPORT_SYMBOL(__skb_ext_del);
6150 void __skb_ext_put(struct skb_ext *ext)
6152 /* If this is last clone, nothing can increment
6153 * it after check passes. Avoids one atomic op.
6155 if (refcount_read(&ext->refcnt) == 1)
6158 if (!refcount_dec_and_test(&ext->refcnt))
6162 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6163 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6166 kmem_cache_free(skbuff_ext_cache, ext);
6168 EXPORT_SYMBOL(__skb_ext_put);
6169 #endif /* CONFIG_SKB_EXTENSIONS */