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>
71 #include <net/mptcp.h>
73 #include <linux/uaccess.h>
74 #include <trace/events/skb.h>
75 #include <linux/highmem.h>
76 #include <linux/capability.h>
77 #include <linux/user_namespace.h>
78 #include <linux/indirect_call_wrapper.h>
82 struct kmem_cache *skbuff_head_cache __ro_after_init;
83 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
84 #ifdef CONFIG_SKB_EXTENSIONS
85 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
87 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
88 EXPORT_SYMBOL(sysctl_max_skb_frags);
91 * skb_panic - private function for out-of-line support
95 * @msg: skb_over_panic or skb_under_panic
97 * Out-of-line support for skb_put() and skb_push().
98 * Called via the wrapper skb_over_panic() or skb_under_panic().
99 * Keep out of line to prevent kernel bloat.
100 * __builtin_return_address is not used because it is not always reliable.
102 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
105 pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
106 msg, addr, skb->len, sz, skb->head, skb->data,
107 (unsigned long)skb->tail, (unsigned long)skb->end,
108 skb->dev ? skb->dev->name : "<NULL>");
112 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
114 skb_panic(skb, sz, addr, __func__);
117 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
119 skb_panic(skb, sz, addr, __func__);
123 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
124 * the caller if emergency pfmemalloc reserves are being used. If it is and
125 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
126 * may be used. Otherwise, the packet data may be discarded until enough
129 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
130 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
132 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
133 unsigned long ip, bool *pfmemalloc)
136 bool ret_pfmemalloc = false;
139 * Try a regular allocation, when that fails and we're not entitled
140 * to the reserves, fail.
142 obj = kmalloc_node_track_caller(size,
143 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
145 if (obj || !(gfp_pfmemalloc_allowed(flags)))
148 /* Try again but now we are using pfmemalloc reserves */
149 ret_pfmemalloc = true;
150 obj = kmalloc_node_track_caller(size, flags, node);
154 *pfmemalloc = ret_pfmemalloc;
159 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
160 * 'private' fields and also do memory statistics to find all the
166 * __alloc_skb - allocate a network buffer
167 * @size: size to allocate
168 * @gfp_mask: allocation mask
169 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
170 * instead of head cache and allocate a cloned (child) skb.
171 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
172 * allocations in case the data is required for writeback
173 * @node: numa node to allocate memory on
175 * Allocate a new &sk_buff. The returned buffer has no headroom and a
176 * tail room of at least size bytes. The object has a reference count
177 * of one. The return is the buffer. On a failure the return is %NULL.
179 * Buffers may only be allocated from interrupts using a @gfp_mask of
182 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
185 struct kmem_cache *cache;
186 struct skb_shared_info *shinfo;
191 cache = (flags & SKB_ALLOC_FCLONE)
192 ? skbuff_fclone_cache : skbuff_head_cache;
194 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
195 gfp_mask |= __GFP_MEMALLOC;
198 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
203 /* We do our best to align skb_shared_info on a separate cache
204 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
205 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
206 * Both skb->head and skb_shared_info are cache line aligned.
208 size = SKB_DATA_ALIGN(size);
209 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
210 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
213 /* kmalloc(size) might give us more room than requested.
214 * Put skb_shared_info exactly at the end of allocated zone,
215 * to allow max possible filling before reallocation.
217 size = SKB_WITH_OVERHEAD(ksize(data));
218 prefetchw(data + size);
221 * Only clear those fields we need to clear, not those that we will
222 * actually initialise below. Hence, don't put any more fields after
223 * the tail pointer in struct sk_buff!
225 memset(skb, 0, offsetof(struct sk_buff, tail));
226 /* Account for allocated memory : skb + skb->head */
227 skb->truesize = SKB_TRUESIZE(size);
228 skb->pfmemalloc = pfmemalloc;
229 refcount_set(&skb->users, 1);
232 skb_reset_tail_pointer(skb);
233 skb->end = skb->tail + size;
234 skb->mac_header = (typeof(skb->mac_header))~0U;
235 skb->transport_header = (typeof(skb->transport_header))~0U;
237 /* make sure we initialize shinfo sequentially */
238 shinfo = skb_shinfo(skb);
239 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
240 atomic_set(&shinfo->dataref, 1);
242 if (flags & SKB_ALLOC_FCLONE) {
243 struct sk_buff_fclones *fclones;
245 fclones = container_of(skb, struct sk_buff_fclones, skb1);
247 skb->fclone = SKB_FCLONE_ORIG;
248 refcount_set(&fclones->fclone_ref, 1);
250 fclones->skb2.fclone = SKB_FCLONE_CLONE;
255 kmem_cache_free(cache, skb);
259 EXPORT_SYMBOL(__alloc_skb);
261 /* Caller must provide SKB that is memset cleared */
262 static struct sk_buff *__build_skb_around(struct sk_buff *skb,
263 void *data, unsigned int frag_size)
265 struct skb_shared_info *shinfo;
266 unsigned int size = frag_size ? : ksize(data);
268 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
270 /* Assumes caller memset cleared SKB */
271 skb->truesize = SKB_TRUESIZE(size);
272 refcount_set(&skb->users, 1);
275 skb_reset_tail_pointer(skb);
276 skb->end = skb->tail + size;
277 skb->mac_header = (typeof(skb->mac_header))~0U;
278 skb->transport_header = (typeof(skb->transport_header))~0U;
280 /* make sure we initialize shinfo sequentially */
281 shinfo = skb_shinfo(skb);
282 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
283 atomic_set(&shinfo->dataref, 1);
289 * __build_skb - build a network buffer
290 * @data: data buffer provided by caller
291 * @frag_size: size of data, or 0 if head was kmalloced
293 * Allocate a new &sk_buff. Caller provides space holding head and
294 * skb_shared_info. @data must have been allocated by kmalloc() only if
295 * @frag_size is 0, otherwise data should come from the page allocator
297 * The return is the new skb buffer.
298 * On a failure the return is %NULL, and @data is not freed.
300 * Before IO, driver allocates only data buffer where NIC put incoming frame
301 * Driver should add room at head (NET_SKB_PAD) and
302 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
303 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
304 * before giving packet to stack.
305 * RX rings only contains data buffers, not full skbs.
307 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
311 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
315 memset(skb, 0, offsetof(struct sk_buff, tail));
317 return __build_skb_around(skb, data, frag_size);
320 /* build_skb() is wrapper over __build_skb(), that specifically
321 * takes care of skb->head and skb->pfmemalloc
322 * This means that if @frag_size is not zero, then @data must be backed
323 * by a page fragment, not kmalloc() or vmalloc()
325 struct sk_buff *build_skb(void *data, unsigned int frag_size)
327 struct sk_buff *skb = __build_skb(data, frag_size);
329 if (skb && frag_size) {
331 if (page_is_pfmemalloc(virt_to_head_page(data)))
336 EXPORT_SYMBOL(build_skb);
339 * build_skb_around - build a network buffer around provided skb
340 * @skb: sk_buff provide by caller, must be memset cleared
341 * @data: data buffer provided by caller
342 * @frag_size: size of data, or 0 if head was kmalloced
344 struct sk_buff *build_skb_around(struct sk_buff *skb,
345 void *data, unsigned int frag_size)
350 skb = __build_skb_around(skb, data, frag_size);
352 if (skb && frag_size) {
354 if (page_is_pfmemalloc(virt_to_head_page(data)))
359 EXPORT_SYMBOL(build_skb_around);
361 #define NAPI_SKB_CACHE_SIZE 64
363 struct napi_alloc_cache {
364 struct page_frag_cache page;
365 unsigned int skb_count;
366 void *skb_cache[NAPI_SKB_CACHE_SIZE];
369 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
370 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
372 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
374 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
376 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
379 void *napi_alloc_frag(unsigned int fragsz)
381 fragsz = SKB_DATA_ALIGN(fragsz);
383 return __napi_alloc_frag(fragsz, GFP_ATOMIC);
385 EXPORT_SYMBOL(napi_alloc_frag);
388 * netdev_alloc_frag - allocate a page fragment
389 * @fragsz: fragment size
391 * Allocates a frag from a page for receive buffer.
392 * Uses GFP_ATOMIC allocations.
394 void *netdev_alloc_frag(unsigned int fragsz)
396 struct page_frag_cache *nc;
399 fragsz = SKB_DATA_ALIGN(fragsz);
400 if (in_irq() || irqs_disabled()) {
401 nc = this_cpu_ptr(&netdev_alloc_cache);
402 data = page_frag_alloc(nc, fragsz, GFP_ATOMIC);
405 data = __napi_alloc_frag(fragsz, GFP_ATOMIC);
410 EXPORT_SYMBOL(netdev_alloc_frag);
413 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
414 * @dev: network device to receive on
415 * @len: length to allocate
416 * @gfp_mask: get_free_pages mask, passed to alloc_skb
418 * Allocate a new &sk_buff and assign it a usage count of one. The
419 * buffer has NET_SKB_PAD headroom built in. Users should allocate
420 * the headroom they think they need without accounting for the
421 * built in space. The built in space is used for optimisations.
423 * %NULL is returned if there is no free memory.
425 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
428 struct page_frag_cache *nc;
435 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
436 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
437 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
443 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
444 len = SKB_DATA_ALIGN(len);
446 if (sk_memalloc_socks())
447 gfp_mask |= __GFP_MEMALLOC;
449 if (in_irq() || irqs_disabled()) {
450 nc = this_cpu_ptr(&netdev_alloc_cache);
451 data = page_frag_alloc(nc, len, gfp_mask);
452 pfmemalloc = nc->pfmemalloc;
455 nc = this_cpu_ptr(&napi_alloc_cache.page);
456 data = page_frag_alloc(nc, len, gfp_mask);
457 pfmemalloc = nc->pfmemalloc;
464 skb = __build_skb(data, len);
465 if (unlikely(!skb)) {
475 skb_reserve(skb, NET_SKB_PAD);
481 EXPORT_SYMBOL(__netdev_alloc_skb);
484 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
485 * @napi: napi instance this buffer was allocated for
486 * @len: length to allocate
487 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
489 * Allocate a new sk_buff for use in NAPI receive. This buffer will
490 * attempt to allocate the head from a special reserved region used
491 * only for NAPI Rx allocation. By doing this we can save several
492 * CPU cycles by avoiding having to disable and re-enable IRQs.
494 * %NULL is returned if there is no free memory.
496 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
499 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
503 len += NET_SKB_PAD + NET_IP_ALIGN;
505 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
506 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
507 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
513 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
514 len = SKB_DATA_ALIGN(len);
516 if (sk_memalloc_socks())
517 gfp_mask |= __GFP_MEMALLOC;
519 data = page_frag_alloc(&nc->page, len, gfp_mask);
523 skb = __build_skb(data, len);
524 if (unlikely(!skb)) {
529 if (nc->page.pfmemalloc)
534 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
535 skb->dev = napi->dev;
540 EXPORT_SYMBOL(__napi_alloc_skb);
542 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
543 int size, unsigned int truesize)
545 skb_fill_page_desc(skb, i, page, off, size);
547 skb->data_len += size;
548 skb->truesize += truesize;
550 EXPORT_SYMBOL(skb_add_rx_frag);
552 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
553 unsigned int truesize)
555 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
557 skb_frag_size_add(frag, size);
559 skb->data_len += size;
560 skb->truesize += truesize;
562 EXPORT_SYMBOL(skb_coalesce_rx_frag);
564 static void skb_drop_list(struct sk_buff **listp)
566 kfree_skb_list(*listp);
570 static inline void skb_drop_fraglist(struct sk_buff *skb)
572 skb_drop_list(&skb_shinfo(skb)->frag_list);
575 static void skb_clone_fraglist(struct sk_buff *skb)
577 struct sk_buff *list;
579 skb_walk_frags(skb, list)
583 static void skb_free_head(struct sk_buff *skb)
585 unsigned char *head = skb->head;
593 static void skb_release_data(struct sk_buff *skb)
595 struct skb_shared_info *shinfo = skb_shinfo(skb);
599 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
603 for (i = 0; i < shinfo->nr_frags; i++)
604 __skb_frag_unref(&shinfo->frags[i]);
606 if (shinfo->frag_list)
607 kfree_skb_list(shinfo->frag_list);
609 skb_zcopy_clear(skb, true);
614 * Free an skbuff by memory without cleaning the state.
616 static void kfree_skbmem(struct sk_buff *skb)
618 struct sk_buff_fclones *fclones;
620 switch (skb->fclone) {
621 case SKB_FCLONE_UNAVAILABLE:
622 kmem_cache_free(skbuff_head_cache, skb);
625 case SKB_FCLONE_ORIG:
626 fclones = container_of(skb, struct sk_buff_fclones, skb1);
628 /* We usually free the clone (TX completion) before original skb
629 * This test would have no chance to be true for the clone,
630 * while here, branch prediction will be good.
632 if (refcount_read(&fclones->fclone_ref) == 1)
636 default: /* SKB_FCLONE_CLONE */
637 fclones = container_of(skb, struct sk_buff_fclones, skb2);
640 if (!refcount_dec_and_test(&fclones->fclone_ref))
643 kmem_cache_free(skbuff_fclone_cache, fclones);
646 void skb_release_head_state(struct sk_buff *skb)
649 if (skb->destructor) {
651 skb->destructor(skb);
653 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
654 nf_conntrack_put(skb_nfct(skb));
659 /* Free everything but the sk_buff shell. */
660 static void skb_release_all(struct sk_buff *skb)
662 skb_release_head_state(skb);
663 if (likely(skb->head))
664 skb_release_data(skb);
668 * __kfree_skb - private function
671 * Free an sk_buff. Release anything attached to the buffer.
672 * Clean the state. This is an internal helper function. Users should
673 * always call kfree_skb
676 void __kfree_skb(struct sk_buff *skb)
678 skb_release_all(skb);
681 EXPORT_SYMBOL(__kfree_skb);
684 * kfree_skb - free an sk_buff
685 * @skb: buffer to free
687 * Drop a reference to the buffer and free it if the usage count has
690 void kfree_skb(struct sk_buff *skb)
695 trace_kfree_skb(skb, __builtin_return_address(0));
698 EXPORT_SYMBOL(kfree_skb);
700 void kfree_skb_list(struct sk_buff *segs)
703 struct sk_buff *next = segs->next;
709 EXPORT_SYMBOL(kfree_skb_list);
711 /* Dump skb information and contents.
713 * Must only be called from net_ratelimit()-ed paths.
715 * Dumps up to can_dump_full whole packets if full_pkt, headers otherwise.
717 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
719 static atomic_t can_dump_full = ATOMIC_INIT(5);
720 struct skb_shared_info *sh = skb_shinfo(skb);
721 struct net_device *dev = skb->dev;
722 struct sock *sk = skb->sk;
723 struct sk_buff *list_skb;
724 bool has_mac, has_trans;
725 int headroom, tailroom;
729 full_pkt = atomic_dec_if_positive(&can_dump_full) >= 0;
734 len = min_t(int, skb->len, MAX_HEADER + 128);
736 headroom = skb_headroom(skb);
737 tailroom = skb_tailroom(skb);
739 has_mac = skb_mac_header_was_set(skb);
740 has_trans = skb_transport_header_was_set(skb);
742 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
743 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
744 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
745 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
746 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
747 level, skb->len, headroom, skb_headlen(skb), tailroom,
748 has_mac ? skb->mac_header : -1,
749 has_mac ? skb_mac_header_len(skb) : -1,
751 has_trans ? skb_network_header_len(skb) : -1,
752 has_trans ? skb->transport_header : -1,
753 sh->tx_flags, sh->nr_frags,
754 sh->gso_size, sh->gso_type, sh->gso_segs,
755 skb->csum, skb->ip_summed, skb->csum_complete_sw,
756 skb->csum_valid, skb->csum_level,
757 skb->hash, skb->sw_hash, skb->l4_hash,
758 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
761 printk("%sdev name=%s feat=0x%pNF\n",
762 level, dev->name, &dev->features);
764 printk("%ssk family=%hu type=%u proto=%u\n",
765 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
767 if (full_pkt && headroom)
768 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
769 16, 1, skb->head, headroom, false);
771 seg_len = min_t(int, skb_headlen(skb), len);
773 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
774 16, 1, skb->data, seg_len, false);
777 if (full_pkt && tailroom)
778 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
779 16, 1, skb_tail_pointer(skb), tailroom, false);
781 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
782 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
783 u32 p_off, p_len, copied;
787 skb_frag_foreach_page(frag, skb_frag_off(frag),
788 skb_frag_size(frag), p, p_off, p_len,
790 seg_len = min_t(int, p_len, len);
791 vaddr = kmap_atomic(p);
792 print_hex_dump(level, "skb frag: ",
794 16, 1, vaddr + p_off, seg_len, false);
795 kunmap_atomic(vaddr);
802 if (full_pkt && skb_has_frag_list(skb)) {
803 printk("skb fraglist:\n");
804 skb_walk_frags(skb, list_skb)
805 skb_dump(level, list_skb, true);
808 EXPORT_SYMBOL(skb_dump);
811 * skb_tx_error - report an sk_buff xmit error
812 * @skb: buffer that triggered an error
814 * Report xmit error if a device callback is tracking this skb.
815 * skb must be freed afterwards.
817 void skb_tx_error(struct sk_buff *skb)
819 skb_zcopy_clear(skb, true);
821 EXPORT_SYMBOL(skb_tx_error);
823 #ifdef CONFIG_TRACEPOINTS
825 * consume_skb - free an skbuff
826 * @skb: buffer to free
828 * Drop a ref to the buffer and free it if the usage count has hit zero
829 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
830 * is being dropped after a failure and notes that
832 void consume_skb(struct sk_buff *skb)
837 trace_consume_skb(skb);
840 EXPORT_SYMBOL(consume_skb);
844 * consume_stateless_skb - free an skbuff, assuming it is stateless
845 * @skb: buffer to free
847 * Alike consume_skb(), but this variant assumes that this is the last
848 * skb reference and all the head states have been already dropped
850 void __consume_stateless_skb(struct sk_buff *skb)
852 trace_consume_skb(skb);
853 skb_release_data(skb);
857 void __kfree_skb_flush(void)
859 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
861 /* flush skb_cache if containing objects */
863 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
869 static inline void _kfree_skb_defer(struct sk_buff *skb)
871 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
873 /* drop skb->head and call any destructors for packet */
874 skb_release_all(skb);
876 /* record skb to CPU local list */
877 nc->skb_cache[nc->skb_count++] = skb;
880 /* SLUB writes into objects when freeing */
884 /* flush skb_cache if it is filled */
885 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
886 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
891 void __kfree_skb_defer(struct sk_buff *skb)
893 _kfree_skb_defer(skb);
896 void napi_consume_skb(struct sk_buff *skb, int budget)
901 /* Zero budget indicate non-NAPI context called us, like netpoll */
902 if (unlikely(!budget)) {
903 dev_consume_skb_any(skb);
910 /* if reaching here SKB is ready to free */
911 trace_consume_skb(skb);
913 /* if SKB is a clone, don't handle this case */
914 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
919 _kfree_skb_defer(skb);
921 EXPORT_SYMBOL(napi_consume_skb);
923 /* Make sure a field is enclosed inside headers_start/headers_end section */
924 #define CHECK_SKB_FIELD(field) \
925 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
926 offsetof(struct sk_buff, headers_start)); \
927 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
928 offsetof(struct sk_buff, headers_end)); \
930 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
932 new->tstamp = old->tstamp;
933 /* We do not copy old->sk */
935 memcpy(new->cb, old->cb, sizeof(old->cb));
936 skb_dst_copy(new, old);
937 __skb_ext_copy(new, old);
938 __nf_copy(new, old, false);
940 /* Note : this field could be in headers_start/headers_end section
941 * It is not yet because we do not want to have a 16 bit hole
943 new->queue_mapping = old->queue_mapping;
945 memcpy(&new->headers_start, &old->headers_start,
946 offsetof(struct sk_buff, headers_end) -
947 offsetof(struct sk_buff, headers_start));
948 CHECK_SKB_FIELD(protocol);
949 CHECK_SKB_FIELD(csum);
950 CHECK_SKB_FIELD(hash);
951 CHECK_SKB_FIELD(priority);
952 CHECK_SKB_FIELD(skb_iif);
953 CHECK_SKB_FIELD(vlan_proto);
954 CHECK_SKB_FIELD(vlan_tci);
955 CHECK_SKB_FIELD(transport_header);
956 CHECK_SKB_FIELD(network_header);
957 CHECK_SKB_FIELD(mac_header);
958 CHECK_SKB_FIELD(inner_protocol);
959 CHECK_SKB_FIELD(inner_transport_header);
960 CHECK_SKB_FIELD(inner_network_header);
961 CHECK_SKB_FIELD(inner_mac_header);
962 CHECK_SKB_FIELD(mark);
963 #ifdef CONFIG_NETWORK_SECMARK
964 CHECK_SKB_FIELD(secmark);
966 #ifdef CONFIG_NET_RX_BUSY_POLL
967 CHECK_SKB_FIELD(napi_id);
970 CHECK_SKB_FIELD(sender_cpu);
972 #ifdef CONFIG_NET_SCHED
973 CHECK_SKB_FIELD(tc_index);
979 * You should not add any new code to this function. Add it to
980 * __copy_skb_header above instead.
982 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
984 #define C(x) n->x = skb->x
986 n->next = n->prev = NULL;
988 __copy_skb_header(n, skb);
993 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
998 n->destructor = NULL;
1005 refcount_set(&n->users, 1);
1007 atomic_inc(&(skb_shinfo(skb)->dataref));
1015 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1016 * @first: first sk_buff of the msg
1018 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1022 n = alloc_skb(0, GFP_ATOMIC);
1026 n->len = first->len;
1027 n->data_len = first->len;
1028 n->truesize = first->truesize;
1030 skb_shinfo(n)->frag_list = first;
1032 __copy_skb_header(n, first);
1033 n->destructor = NULL;
1037 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1040 * skb_morph - morph one skb into another
1041 * @dst: the skb to receive the contents
1042 * @src: the skb to supply the contents
1044 * This is identical to skb_clone except that the target skb is
1045 * supplied by the user.
1047 * The target skb is returned upon exit.
1049 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1051 skb_release_all(dst);
1052 return __skb_clone(dst, src);
1054 EXPORT_SYMBOL_GPL(skb_morph);
1056 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1058 unsigned long max_pg, num_pg, new_pg, old_pg;
1059 struct user_struct *user;
1061 if (capable(CAP_IPC_LOCK) || !size)
1064 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1065 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1066 user = mmp->user ? : current_user();
1069 old_pg = atomic_long_read(&user->locked_vm);
1070 new_pg = old_pg + num_pg;
1071 if (new_pg > max_pg)
1073 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1077 mmp->user = get_uid(user);
1078 mmp->num_pg = num_pg;
1080 mmp->num_pg += num_pg;
1085 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1087 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1090 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1091 free_uid(mmp->user);
1094 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1096 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
1098 struct ubuf_info *uarg;
1099 struct sk_buff *skb;
1101 WARN_ON_ONCE(!in_task());
1103 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1107 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1108 uarg = (void *)skb->cb;
1109 uarg->mmp.user = NULL;
1111 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1116 uarg->callback = sock_zerocopy_callback;
1117 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1119 uarg->bytelen = size;
1121 refcount_set(&uarg->refcnt, 1);
1126 EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
1128 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1130 return container_of((void *)uarg, struct sk_buff, cb);
1133 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
1134 struct ubuf_info *uarg)
1137 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1140 /* realloc only when socket is locked (TCP, UDP cork),
1141 * so uarg->len and sk_zckey access is serialized
1143 if (!sock_owned_by_user(sk)) {
1148 bytelen = uarg->bytelen + size;
1149 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1150 /* TCP can create new skb to attach new uarg */
1151 if (sk->sk_type == SOCK_STREAM)
1156 next = (u32)atomic_read(&sk->sk_zckey);
1157 if ((u32)(uarg->id + uarg->len) == next) {
1158 if (mm_account_pinned_pages(&uarg->mmp, size))
1161 uarg->bytelen = bytelen;
1162 atomic_set(&sk->sk_zckey, ++next);
1164 /* no extra ref when appending to datagram (MSG_MORE) */
1165 if (sk->sk_type == SOCK_STREAM)
1166 sock_zerocopy_get(uarg);
1173 return sock_zerocopy_alloc(sk, size);
1175 EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1177 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1179 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1183 old_lo = serr->ee.ee_info;
1184 old_hi = serr->ee.ee_data;
1185 sum_len = old_hi - old_lo + 1ULL + len;
1187 if (sum_len >= (1ULL << 32))
1190 if (lo != old_hi + 1)
1193 serr->ee.ee_data += len;
1197 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1199 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1200 struct sock_exterr_skb *serr;
1201 struct sock *sk = skb->sk;
1202 struct sk_buff_head *q;
1203 unsigned long flags;
1207 mm_unaccount_pinned_pages(&uarg->mmp);
1209 /* if !len, there was only 1 call, and it was aborted
1210 * so do not queue a completion notification
1212 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1217 hi = uarg->id + len - 1;
1219 serr = SKB_EXT_ERR(skb);
1220 memset(serr, 0, sizeof(*serr));
1221 serr->ee.ee_errno = 0;
1222 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1223 serr->ee.ee_data = hi;
1224 serr->ee.ee_info = lo;
1226 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1228 q = &sk->sk_error_queue;
1229 spin_lock_irqsave(&q->lock, flags);
1230 tail = skb_peek_tail(q);
1231 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1232 !skb_zerocopy_notify_extend(tail, lo, len)) {
1233 __skb_queue_tail(q, skb);
1236 spin_unlock_irqrestore(&q->lock, flags);
1238 sk->sk_error_report(sk);
1244 EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1246 void sock_zerocopy_put(struct ubuf_info *uarg)
1248 if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1250 uarg->callback(uarg, uarg->zerocopy);
1252 consume_skb(skb_from_uarg(uarg));
1255 EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1257 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1260 struct sock *sk = skb_from_uarg(uarg)->sk;
1262 atomic_dec(&sk->sk_zckey);
1266 sock_zerocopy_put(uarg);
1269 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1271 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1273 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1275 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1277 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1278 struct msghdr *msg, int len,
1279 struct ubuf_info *uarg)
1281 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1282 struct iov_iter orig_iter = msg->msg_iter;
1283 int err, orig_len = skb->len;
1285 /* An skb can only point to one uarg. This edge case happens when
1286 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1288 if (orig_uarg && uarg != orig_uarg)
1291 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1292 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1293 struct sock *save_sk = skb->sk;
1295 /* Streams do not free skb on error. Reset to prev state. */
1296 msg->msg_iter = orig_iter;
1298 ___pskb_trim(skb, orig_len);
1303 skb_zcopy_set(skb, uarg, NULL);
1304 return skb->len - orig_len;
1306 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1308 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1311 if (skb_zcopy(orig)) {
1312 if (skb_zcopy(nskb)) {
1313 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1318 if (skb_uarg(nskb) == skb_uarg(orig))
1320 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1323 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1329 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1330 * @skb: the skb to modify
1331 * @gfp_mask: allocation priority
1333 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1334 * It will copy all frags into kernel and drop the reference
1335 * to userspace pages.
1337 * If this function is called from an interrupt gfp_mask() must be
1340 * Returns 0 on success or a negative error code on failure
1341 * to allocate kernel memory to copy to.
1343 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1345 int num_frags = skb_shinfo(skb)->nr_frags;
1346 struct page *page, *head = NULL;
1350 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1356 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1357 for (i = 0; i < new_frags; i++) {
1358 page = alloc_page(gfp_mask);
1361 struct page *next = (struct page *)page_private(head);
1367 set_page_private(page, (unsigned long)head);
1373 for (i = 0; i < num_frags; i++) {
1374 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1375 u32 p_off, p_len, copied;
1379 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1380 p, p_off, p_len, copied) {
1382 vaddr = kmap_atomic(p);
1384 while (done < p_len) {
1385 if (d_off == PAGE_SIZE) {
1387 page = (struct page *)page_private(page);
1389 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1390 memcpy(page_address(page) + d_off,
1391 vaddr + p_off + done, copy);
1395 kunmap_atomic(vaddr);
1399 /* skb frags release userspace buffers */
1400 for (i = 0; i < num_frags; i++)
1401 skb_frag_unref(skb, i);
1403 /* skb frags point to kernel buffers */
1404 for (i = 0; i < new_frags - 1; i++) {
1405 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1406 head = (struct page *)page_private(head);
1408 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1409 skb_shinfo(skb)->nr_frags = new_frags;
1412 skb_zcopy_clear(skb, false);
1415 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1418 * skb_clone - duplicate an sk_buff
1419 * @skb: buffer to clone
1420 * @gfp_mask: allocation priority
1422 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1423 * copies share the same packet data but not structure. The new
1424 * buffer has a reference count of 1. If the allocation fails the
1425 * function returns %NULL otherwise the new buffer is returned.
1427 * If this function is called from an interrupt gfp_mask() must be
1431 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1433 struct sk_buff_fclones *fclones = container_of(skb,
1434 struct sk_buff_fclones,
1438 if (skb_orphan_frags(skb, gfp_mask))
1441 if (skb->fclone == SKB_FCLONE_ORIG &&
1442 refcount_read(&fclones->fclone_ref) == 1) {
1444 refcount_set(&fclones->fclone_ref, 2);
1446 if (skb_pfmemalloc(skb))
1447 gfp_mask |= __GFP_MEMALLOC;
1449 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1453 n->fclone = SKB_FCLONE_UNAVAILABLE;
1456 return __skb_clone(n, skb);
1458 EXPORT_SYMBOL(skb_clone);
1460 void skb_headers_offset_update(struct sk_buff *skb, int off)
1462 /* Only adjust this if it actually is csum_start rather than csum */
1463 if (skb->ip_summed == CHECKSUM_PARTIAL)
1464 skb->csum_start += off;
1465 /* {transport,network,mac}_header and tail are relative to skb->head */
1466 skb->transport_header += off;
1467 skb->network_header += off;
1468 if (skb_mac_header_was_set(skb))
1469 skb->mac_header += off;
1470 skb->inner_transport_header += off;
1471 skb->inner_network_header += off;
1472 skb->inner_mac_header += off;
1474 EXPORT_SYMBOL(skb_headers_offset_update);
1476 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1478 __copy_skb_header(new, old);
1480 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1481 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1482 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1484 EXPORT_SYMBOL(skb_copy_header);
1486 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1488 if (skb_pfmemalloc(skb))
1489 return SKB_ALLOC_RX;
1494 * skb_copy - create private copy of an sk_buff
1495 * @skb: buffer to copy
1496 * @gfp_mask: allocation priority
1498 * Make a copy of both an &sk_buff and its data. This is used when the
1499 * caller wishes to modify the data and needs a private copy of the
1500 * data to alter. Returns %NULL on failure or the pointer to the buffer
1501 * on success. The returned buffer has a reference count of 1.
1503 * As by-product this function converts non-linear &sk_buff to linear
1504 * one, so that &sk_buff becomes completely private and caller is allowed
1505 * to modify all the data of returned buffer. This means that this
1506 * function is not recommended for use in circumstances when only
1507 * header is going to be modified. Use pskb_copy() instead.
1510 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1512 int headerlen = skb_headroom(skb);
1513 unsigned int size = skb_end_offset(skb) + skb->data_len;
1514 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1515 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1520 /* Set the data pointer */
1521 skb_reserve(n, headerlen);
1522 /* Set the tail pointer and length */
1523 skb_put(n, skb->len);
1525 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1527 skb_copy_header(n, skb);
1530 EXPORT_SYMBOL(skb_copy);
1533 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1534 * @skb: buffer to copy
1535 * @headroom: headroom of new skb
1536 * @gfp_mask: allocation priority
1537 * @fclone: if true allocate the copy of the skb from the fclone
1538 * cache instead of the head cache; it is recommended to set this
1539 * to true for the cases where the copy will likely be cloned
1541 * Make a copy of both an &sk_buff and part of its data, located
1542 * in header. Fragmented data remain shared. This is used when
1543 * the caller wishes to modify only header of &sk_buff and needs
1544 * private copy of the header to alter. Returns %NULL on failure
1545 * or the pointer to the buffer on success.
1546 * The returned buffer has a reference count of 1.
1549 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1550 gfp_t gfp_mask, bool fclone)
1552 unsigned int size = skb_headlen(skb) + headroom;
1553 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1554 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1559 /* Set the data pointer */
1560 skb_reserve(n, headroom);
1561 /* Set the tail pointer and length */
1562 skb_put(n, skb_headlen(skb));
1563 /* Copy the bytes */
1564 skb_copy_from_linear_data(skb, n->data, n->len);
1566 n->truesize += skb->data_len;
1567 n->data_len = skb->data_len;
1570 if (skb_shinfo(skb)->nr_frags) {
1573 if (skb_orphan_frags(skb, gfp_mask) ||
1574 skb_zerocopy_clone(n, skb, gfp_mask)) {
1579 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1580 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1581 skb_frag_ref(skb, i);
1583 skb_shinfo(n)->nr_frags = i;
1586 if (skb_has_frag_list(skb)) {
1587 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1588 skb_clone_fraglist(n);
1591 skb_copy_header(n, skb);
1595 EXPORT_SYMBOL(__pskb_copy_fclone);
1598 * pskb_expand_head - reallocate header of &sk_buff
1599 * @skb: buffer to reallocate
1600 * @nhead: room to add at head
1601 * @ntail: room to add at tail
1602 * @gfp_mask: allocation priority
1604 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1605 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1606 * reference count of 1. Returns zero in the case of success or error,
1607 * if expansion failed. In the last case, &sk_buff is not changed.
1609 * All the pointers pointing into skb header may change and must be
1610 * reloaded after call to this function.
1613 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1616 int i, osize = skb_end_offset(skb);
1617 int size = osize + nhead + ntail;
1623 BUG_ON(skb_shared(skb));
1625 size = SKB_DATA_ALIGN(size);
1627 if (skb_pfmemalloc(skb))
1628 gfp_mask |= __GFP_MEMALLOC;
1629 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1630 gfp_mask, NUMA_NO_NODE, NULL);
1633 size = SKB_WITH_OVERHEAD(ksize(data));
1635 /* Copy only real data... and, alas, header. This should be
1636 * optimized for the cases when header is void.
1638 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1640 memcpy((struct skb_shared_info *)(data + size),
1642 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1645 * if shinfo is shared we must drop the old head gracefully, but if it
1646 * is not we can just drop the old head and let the existing refcount
1647 * be since all we did is relocate the values
1649 if (skb_cloned(skb)) {
1650 if (skb_orphan_frags(skb, gfp_mask))
1653 refcount_inc(&skb_uarg(skb)->refcnt);
1654 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1655 skb_frag_ref(skb, i);
1657 if (skb_has_frag_list(skb))
1658 skb_clone_fraglist(skb);
1660 skb_release_data(skb);
1664 off = (data + nhead) - skb->head;
1669 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1673 skb->end = skb->head + size;
1676 skb_headers_offset_update(skb, nhead);
1680 atomic_set(&skb_shinfo(skb)->dataref, 1);
1682 skb_metadata_clear(skb);
1684 /* It is not generally safe to change skb->truesize.
1685 * For the moment, we really care of rx path, or
1686 * when skb is orphaned (not attached to a socket).
1688 if (!skb->sk || skb->destructor == sock_edemux)
1689 skb->truesize += size - osize;
1698 EXPORT_SYMBOL(pskb_expand_head);
1700 /* Make private copy of skb with writable head and some headroom */
1702 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1704 struct sk_buff *skb2;
1705 int delta = headroom - skb_headroom(skb);
1708 skb2 = pskb_copy(skb, GFP_ATOMIC);
1710 skb2 = skb_clone(skb, GFP_ATOMIC);
1711 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1719 EXPORT_SYMBOL(skb_realloc_headroom);
1722 * skb_copy_expand - copy and expand sk_buff
1723 * @skb: buffer to copy
1724 * @newheadroom: new free bytes at head
1725 * @newtailroom: new free bytes at tail
1726 * @gfp_mask: allocation priority
1728 * Make a copy of both an &sk_buff and its data and while doing so
1729 * allocate additional space.
1731 * This is used when the caller wishes to modify the data and needs a
1732 * private copy of the data to alter as well as more space for new fields.
1733 * Returns %NULL on failure or the pointer to the buffer
1734 * on success. The returned buffer has a reference count of 1.
1736 * You must pass %GFP_ATOMIC as the allocation priority if this function
1737 * is called from an interrupt.
1739 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1740 int newheadroom, int newtailroom,
1744 * Allocate the copy buffer
1746 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1747 gfp_mask, skb_alloc_rx_flag(skb),
1749 int oldheadroom = skb_headroom(skb);
1750 int head_copy_len, head_copy_off;
1755 skb_reserve(n, newheadroom);
1757 /* Set the tail pointer and length */
1758 skb_put(n, skb->len);
1760 head_copy_len = oldheadroom;
1762 if (newheadroom <= head_copy_len)
1763 head_copy_len = newheadroom;
1765 head_copy_off = newheadroom - head_copy_len;
1767 /* Copy the linear header and data. */
1768 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1769 skb->len + head_copy_len));
1771 skb_copy_header(n, skb);
1773 skb_headers_offset_update(n, newheadroom - oldheadroom);
1777 EXPORT_SYMBOL(skb_copy_expand);
1780 * __skb_pad - zero pad the tail of an skb
1781 * @skb: buffer to pad
1782 * @pad: space to pad
1783 * @free_on_error: free buffer on error
1785 * Ensure that a buffer is followed by a padding area that is zero
1786 * filled. Used by network drivers which may DMA or transfer data
1787 * beyond the buffer end onto the wire.
1789 * May return error in out of memory cases. The skb is freed on error
1790 * if @free_on_error is true.
1793 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1798 /* If the skbuff is non linear tailroom is always zero.. */
1799 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1800 memset(skb->data+skb->len, 0, pad);
1804 ntail = skb->data_len + pad - (skb->end - skb->tail);
1805 if (likely(skb_cloned(skb) || ntail > 0)) {
1806 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1811 /* FIXME: The use of this function with non-linear skb's really needs
1814 err = skb_linearize(skb);
1818 memset(skb->data + skb->len, 0, pad);
1826 EXPORT_SYMBOL(__skb_pad);
1829 * pskb_put - add data to the tail of a potentially fragmented buffer
1830 * @skb: start of the buffer to use
1831 * @tail: tail fragment of the buffer to use
1832 * @len: amount of data to add
1834 * This function extends the used data area of the potentially
1835 * fragmented buffer. @tail must be the last fragment of @skb -- or
1836 * @skb itself. If this would exceed the total buffer size the kernel
1837 * will panic. A pointer to the first byte of the extra data is
1841 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1844 skb->data_len += len;
1847 return skb_put(tail, len);
1849 EXPORT_SYMBOL_GPL(pskb_put);
1852 * skb_put - add data to a buffer
1853 * @skb: buffer to use
1854 * @len: amount of data to add
1856 * This function extends the used data area of the buffer. If this would
1857 * exceed the total buffer size the kernel will panic. A pointer to the
1858 * first byte of the extra data is returned.
1860 void *skb_put(struct sk_buff *skb, unsigned int len)
1862 void *tmp = skb_tail_pointer(skb);
1863 SKB_LINEAR_ASSERT(skb);
1866 if (unlikely(skb->tail > skb->end))
1867 skb_over_panic(skb, len, __builtin_return_address(0));
1870 EXPORT_SYMBOL(skb_put);
1873 * skb_push - add data to the start of a buffer
1874 * @skb: buffer to use
1875 * @len: amount of data to add
1877 * This function extends the used data area of the buffer at the buffer
1878 * start. If this would exceed the total buffer headroom the kernel will
1879 * panic. A pointer to the first byte of the extra data is returned.
1881 void *skb_push(struct sk_buff *skb, unsigned int len)
1885 if (unlikely(skb->data < skb->head))
1886 skb_under_panic(skb, len, __builtin_return_address(0));
1889 EXPORT_SYMBOL(skb_push);
1892 * skb_pull - remove data from the start of a buffer
1893 * @skb: buffer to use
1894 * @len: amount of data to remove
1896 * This function removes data from the start of a buffer, returning
1897 * the memory to the headroom. A pointer to the next data in the buffer
1898 * is returned. Once the data has been pulled future pushes will overwrite
1901 void *skb_pull(struct sk_buff *skb, unsigned int len)
1903 return skb_pull_inline(skb, len);
1905 EXPORT_SYMBOL(skb_pull);
1908 * skb_trim - remove end from a buffer
1909 * @skb: buffer to alter
1912 * Cut the length of a buffer down by removing data from the tail. If
1913 * the buffer is already under the length specified it is not modified.
1914 * The skb must be linear.
1916 void skb_trim(struct sk_buff *skb, unsigned int len)
1919 __skb_trim(skb, len);
1921 EXPORT_SYMBOL(skb_trim);
1923 /* Trims skb to length len. It can change skb pointers.
1926 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1928 struct sk_buff **fragp;
1929 struct sk_buff *frag;
1930 int offset = skb_headlen(skb);
1931 int nfrags = skb_shinfo(skb)->nr_frags;
1935 if (skb_cloned(skb) &&
1936 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1943 for (; i < nfrags; i++) {
1944 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1951 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1954 skb_shinfo(skb)->nr_frags = i;
1956 for (; i < nfrags; i++)
1957 skb_frag_unref(skb, i);
1959 if (skb_has_frag_list(skb))
1960 skb_drop_fraglist(skb);
1964 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1965 fragp = &frag->next) {
1966 int end = offset + frag->len;
1968 if (skb_shared(frag)) {
1969 struct sk_buff *nfrag;
1971 nfrag = skb_clone(frag, GFP_ATOMIC);
1972 if (unlikely(!nfrag))
1975 nfrag->next = frag->next;
1987 unlikely((err = pskb_trim(frag, len - offset))))
1991 skb_drop_list(&frag->next);
1996 if (len > skb_headlen(skb)) {
1997 skb->data_len -= skb->len - len;
2002 skb_set_tail_pointer(skb, len);
2005 if (!skb->sk || skb->destructor == sock_edemux)
2009 EXPORT_SYMBOL(___pskb_trim);
2011 /* Note : use pskb_trim_rcsum() instead of calling this directly
2013 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2015 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2016 int delta = skb->len - len;
2018 skb->csum = csum_block_sub(skb->csum,
2019 skb_checksum(skb, len, delta, 0),
2022 return __pskb_trim(skb, len);
2024 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2027 * __pskb_pull_tail - advance tail of skb header
2028 * @skb: buffer to reallocate
2029 * @delta: number of bytes to advance tail
2031 * The function makes a sense only on a fragmented &sk_buff,
2032 * it expands header moving its tail forward and copying necessary
2033 * data from fragmented part.
2035 * &sk_buff MUST have reference count of 1.
2037 * Returns %NULL (and &sk_buff does not change) if pull failed
2038 * or value of new tail of skb in the case of success.
2040 * All the pointers pointing into skb header may change and must be
2041 * reloaded after call to this function.
2044 /* Moves tail of skb head forward, copying data from fragmented part,
2045 * when it is necessary.
2046 * 1. It may fail due to malloc failure.
2047 * 2. It may change skb pointers.
2049 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2051 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2053 /* If skb has not enough free space at tail, get new one
2054 * plus 128 bytes for future expansions. If we have enough
2055 * room at tail, reallocate without expansion only if skb is cloned.
2057 int i, k, eat = (skb->tail + delta) - skb->end;
2059 if (eat > 0 || skb_cloned(skb)) {
2060 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2065 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2066 skb_tail_pointer(skb), delta));
2068 /* Optimization: no fragments, no reasons to preestimate
2069 * size of pulled pages. Superb.
2071 if (!skb_has_frag_list(skb))
2074 /* Estimate size of pulled pages. */
2076 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2077 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2084 /* If we need update frag list, we are in troubles.
2085 * Certainly, it is possible to add an offset to skb data,
2086 * but taking into account that pulling is expected to
2087 * be very rare operation, it is worth to fight against
2088 * further bloating skb head and crucify ourselves here instead.
2089 * Pure masohism, indeed. 8)8)
2092 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2093 struct sk_buff *clone = NULL;
2094 struct sk_buff *insp = NULL;
2097 if (list->len <= eat) {
2098 /* Eaten as whole. */
2103 /* Eaten partially. */
2105 if (skb_shared(list)) {
2106 /* Sucks! We need to fork list. :-( */
2107 clone = skb_clone(list, GFP_ATOMIC);
2113 /* This may be pulled without
2117 if (!pskb_pull(list, eat)) {
2125 /* Free pulled out fragments. */
2126 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2127 skb_shinfo(skb)->frag_list = list->next;
2130 /* And insert new clone at head. */
2133 skb_shinfo(skb)->frag_list = clone;
2136 /* Success! Now we may commit changes to skb data. */
2141 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2142 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2145 skb_frag_unref(skb, i);
2148 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2150 *frag = skb_shinfo(skb)->frags[i];
2152 skb_frag_off_add(frag, eat);
2153 skb_frag_size_sub(frag, eat);
2161 skb_shinfo(skb)->nr_frags = k;
2165 skb->data_len -= delta;
2168 skb_zcopy_clear(skb, false);
2170 return skb_tail_pointer(skb);
2172 EXPORT_SYMBOL(__pskb_pull_tail);
2175 * skb_copy_bits - copy bits from skb to kernel buffer
2177 * @offset: offset in source
2178 * @to: destination buffer
2179 * @len: number of bytes to copy
2181 * Copy the specified number of bytes from the source skb to the
2182 * destination buffer.
2185 * If its prototype is ever changed,
2186 * check arch/{*}/net/{*}.S files,
2187 * since it is called from BPF assembly code.
2189 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2191 int start = skb_headlen(skb);
2192 struct sk_buff *frag_iter;
2195 if (offset > (int)skb->len - len)
2199 if ((copy = start - offset) > 0) {
2202 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2203 if ((len -= copy) == 0)
2209 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2211 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2213 WARN_ON(start > offset + len);
2215 end = start + skb_frag_size(f);
2216 if ((copy = end - offset) > 0) {
2217 u32 p_off, p_len, copied;
2224 skb_frag_foreach_page(f,
2225 skb_frag_off(f) + offset - start,
2226 copy, p, p_off, p_len, copied) {
2227 vaddr = kmap_atomic(p);
2228 memcpy(to + copied, vaddr + p_off, p_len);
2229 kunmap_atomic(vaddr);
2232 if ((len -= copy) == 0)
2240 skb_walk_frags(skb, frag_iter) {
2243 WARN_ON(start > offset + len);
2245 end = start + frag_iter->len;
2246 if ((copy = end - offset) > 0) {
2249 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2251 if ((len -= copy) == 0)
2265 EXPORT_SYMBOL(skb_copy_bits);
2268 * Callback from splice_to_pipe(), if we need to release some pages
2269 * at the end of the spd in case we error'ed out in filling the pipe.
2271 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2273 put_page(spd->pages[i]);
2276 static struct page *linear_to_page(struct page *page, unsigned int *len,
2277 unsigned int *offset,
2280 struct page_frag *pfrag = sk_page_frag(sk);
2282 if (!sk_page_frag_refill(sk, pfrag))
2285 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2287 memcpy(page_address(pfrag->page) + pfrag->offset,
2288 page_address(page) + *offset, *len);
2289 *offset = pfrag->offset;
2290 pfrag->offset += *len;
2295 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2297 unsigned int offset)
2299 return spd->nr_pages &&
2300 spd->pages[spd->nr_pages - 1] == page &&
2301 (spd->partial[spd->nr_pages - 1].offset +
2302 spd->partial[spd->nr_pages - 1].len == offset);
2306 * Fill page/offset/length into spd, if it can hold more pages.
2308 static bool spd_fill_page(struct splice_pipe_desc *spd,
2309 struct pipe_inode_info *pipe, struct page *page,
2310 unsigned int *len, unsigned int offset,
2314 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2318 page = linear_to_page(page, len, &offset, sk);
2322 if (spd_can_coalesce(spd, page, offset)) {
2323 spd->partial[spd->nr_pages - 1].len += *len;
2327 spd->pages[spd->nr_pages] = page;
2328 spd->partial[spd->nr_pages].len = *len;
2329 spd->partial[spd->nr_pages].offset = offset;
2335 static bool __splice_segment(struct page *page, unsigned int poff,
2336 unsigned int plen, unsigned int *off,
2338 struct splice_pipe_desc *spd, bool linear,
2340 struct pipe_inode_info *pipe)
2345 /* skip this segment if already processed */
2351 /* ignore any bits we already processed */
2357 unsigned int flen = min(*len, plen);
2359 if (spd_fill_page(spd, pipe, page, &flen, poff,
2365 } while (*len && plen);
2371 * Map linear and fragment data from the skb to spd. It reports true if the
2372 * pipe is full or if we already spliced the requested length.
2374 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2375 unsigned int *offset, unsigned int *len,
2376 struct splice_pipe_desc *spd, struct sock *sk)
2379 struct sk_buff *iter;
2381 /* map the linear part :
2382 * If skb->head_frag is set, this 'linear' part is backed by a
2383 * fragment, and if the head is not shared with any clones then
2384 * we can avoid a copy since we own the head portion of this page.
2386 if (__splice_segment(virt_to_page(skb->data),
2387 (unsigned long) skb->data & (PAGE_SIZE - 1),
2390 skb_head_is_locked(skb),
2395 * then map the fragments
2397 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2398 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2400 if (__splice_segment(skb_frag_page(f),
2401 skb_frag_off(f), skb_frag_size(f),
2402 offset, len, spd, false, sk, pipe))
2406 skb_walk_frags(skb, iter) {
2407 if (*offset >= iter->len) {
2408 *offset -= iter->len;
2411 /* __skb_splice_bits() only fails if the output has no room
2412 * left, so no point in going over the frag_list for the error
2415 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2423 * Map data from the skb to a pipe. Should handle both the linear part,
2424 * the fragments, and the frag list.
2426 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2427 struct pipe_inode_info *pipe, unsigned int tlen,
2430 struct partial_page partial[MAX_SKB_FRAGS];
2431 struct page *pages[MAX_SKB_FRAGS];
2432 struct splice_pipe_desc spd = {
2435 .nr_pages_max = MAX_SKB_FRAGS,
2436 .ops = &nosteal_pipe_buf_ops,
2437 .spd_release = sock_spd_release,
2441 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2444 ret = splice_to_pipe(pipe, &spd);
2448 EXPORT_SYMBOL_GPL(skb_splice_bits);
2450 /* Send skb data on a socket. Socket must be locked. */
2451 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2454 unsigned int orig_len = len;
2455 struct sk_buff *head = skb;
2456 unsigned short fragidx;
2461 /* Deal with head data */
2462 while (offset < skb_headlen(skb) && len) {
2466 slen = min_t(int, len, skb_headlen(skb) - offset);
2467 kv.iov_base = skb->data + offset;
2469 memset(&msg, 0, sizeof(msg));
2470 msg.msg_flags = MSG_DONTWAIT;
2472 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2480 /* All the data was skb head? */
2484 /* Make offset relative to start of frags */
2485 offset -= skb_headlen(skb);
2487 /* Find where we are in frag list */
2488 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2489 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2491 if (offset < skb_frag_size(frag))
2494 offset -= skb_frag_size(frag);
2497 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2498 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2500 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2503 ret = kernel_sendpage_locked(sk, skb_frag_page(frag),
2504 skb_frag_off(frag) + offset,
2505 slen, MSG_DONTWAIT);
2518 /* Process any frag lists */
2521 if (skb_has_frag_list(skb)) {
2522 skb = skb_shinfo(skb)->frag_list;
2525 } else if (skb->next) {
2532 return orig_len - len;
2535 return orig_len == len ? ret : orig_len - len;
2537 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2540 * skb_store_bits - store bits from kernel buffer to skb
2541 * @skb: destination buffer
2542 * @offset: offset in destination
2543 * @from: source buffer
2544 * @len: number of bytes to copy
2546 * Copy the specified number of bytes from the source buffer to the
2547 * destination skb. This function handles all the messy bits of
2548 * traversing fragment lists and such.
2551 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2553 int start = skb_headlen(skb);
2554 struct sk_buff *frag_iter;
2557 if (offset > (int)skb->len - len)
2560 if ((copy = start - offset) > 0) {
2563 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2564 if ((len -= copy) == 0)
2570 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2571 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2574 WARN_ON(start > offset + len);
2576 end = start + skb_frag_size(frag);
2577 if ((copy = end - offset) > 0) {
2578 u32 p_off, p_len, copied;
2585 skb_frag_foreach_page(frag,
2586 skb_frag_off(frag) + offset - start,
2587 copy, p, p_off, p_len, copied) {
2588 vaddr = kmap_atomic(p);
2589 memcpy(vaddr + p_off, from + copied, p_len);
2590 kunmap_atomic(vaddr);
2593 if ((len -= copy) == 0)
2601 skb_walk_frags(skb, frag_iter) {
2604 WARN_ON(start > offset + len);
2606 end = start + frag_iter->len;
2607 if ((copy = end - offset) > 0) {
2610 if (skb_store_bits(frag_iter, offset - start,
2613 if ((len -= copy) == 0)
2626 EXPORT_SYMBOL(skb_store_bits);
2628 /* Checksum skb data. */
2629 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2630 __wsum csum, const struct skb_checksum_ops *ops)
2632 int start = skb_headlen(skb);
2633 int i, copy = start - offset;
2634 struct sk_buff *frag_iter;
2637 /* Checksum header. */
2641 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2642 skb->data + offset, copy, csum);
2643 if ((len -= copy) == 0)
2649 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2651 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2653 WARN_ON(start > offset + len);
2655 end = start + skb_frag_size(frag);
2656 if ((copy = end - offset) > 0) {
2657 u32 p_off, p_len, copied;
2665 skb_frag_foreach_page(frag,
2666 skb_frag_off(frag) + offset - start,
2667 copy, p, p_off, p_len, copied) {
2668 vaddr = kmap_atomic(p);
2669 csum2 = INDIRECT_CALL_1(ops->update,
2671 vaddr + p_off, p_len, 0);
2672 kunmap_atomic(vaddr);
2673 csum = INDIRECT_CALL_1(ops->combine,
2674 csum_block_add_ext, csum,
2686 skb_walk_frags(skb, frag_iter) {
2689 WARN_ON(start > offset + len);
2691 end = start + frag_iter->len;
2692 if ((copy = end - offset) > 0) {
2696 csum2 = __skb_checksum(frag_iter, offset - start,
2698 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2699 csum, csum2, pos, copy);
2700 if ((len -= copy) == 0)
2711 EXPORT_SYMBOL(__skb_checksum);
2713 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2714 int len, __wsum csum)
2716 const struct skb_checksum_ops ops = {
2717 .update = csum_partial_ext,
2718 .combine = csum_block_add_ext,
2721 return __skb_checksum(skb, offset, len, csum, &ops);
2723 EXPORT_SYMBOL(skb_checksum);
2725 /* Both of above in one bottle. */
2727 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2728 u8 *to, int len, __wsum csum)
2730 int start = skb_headlen(skb);
2731 int i, copy = start - offset;
2732 struct sk_buff *frag_iter;
2739 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2741 if ((len -= copy) == 0)
2748 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2751 WARN_ON(start > offset + len);
2753 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2754 if ((copy = end - offset) > 0) {
2755 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2756 u32 p_off, p_len, copied;
2764 skb_frag_foreach_page(frag,
2765 skb_frag_off(frag) + offset - start,
2766 copy, p, p_off, p_len, copied) {
2767 vaddr = kmap_atomic(p);
2768 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2771 kunmap_atomic(vaddr);
2772 csum = csum_block_add(csum, csum2, pos);
2784 skb_walk_frags(skb, frag_iter) {
2788 WARN_ON(start > offset + len);
2790 end = start + frag_iter->len;
2791 if ((copy = end - offset) > 0) {
2794 csum2 = skb_copy_and_csum_bits(frag_iter,
2797 csum = csum_block_add(csum, csum2, pos);
2798 if ((len -= copy) == 0)
2809 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2811 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2815 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2816 /* See comments in __skb_checksum_complete(). */
2818 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2819 !skb->csum_complete_sw)
2820 netdev_rx_csum_fault(skb->dev, skb);
2822 if (!skb_shared(skb))
2823 skb->csum_valid = !sum;
2826 EXPORT_SYMBOL(__skb_checksum_complete_head);
2828 /* This function assumes skb->csum already holds pseudo header's checksum,
2829 * which has been changed from the hardware checksum, for example, by
2830 * __skb_checksum_validate_complete(). And, the original skb->csum must
2831 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2833 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2834 * zero. The new checksum is stored back into skb->csum unless the skb is
2837 __sum16 __skb_checksum_complete(struct sk_buff *skb)
2842 csum = skb_checksum(skb, 0, skb->len, 0);
2844 sum = csum_fold(csum_add(skb->csum, csum));
2845 /* This check is inverted, because we already knew the hardware
2846 * checksum is invalid before calling this function. So, if the
2847 * re-computed checksum is valid instead, then we have a mismatch
2848 * between the original skb->csum and skb_checksum(). This means either
2849 * the original hardware checksum is incorrect or we screw up skb->csum
2850 * when moving skb->data around.
2853 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2854 !skb->csum_complete_sw)
2855 netdev_rx_csum_fault(skb->dev, skb);
2858 if (!skb_shared(skb)) {
2859 /* Save full packet checksum */
2861 skb->ip_summed = CHECKSUM_COMPLETE;
2862 skb->csum_complete_sw = 1;
2863 skb->csum_valid = !sum;
2868 EXPORT_SYMBOL(__skb_checksum_complete);
2870 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2872 net_warn_ratelimited(
2873 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2878 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2879 int offset, int len)
2881 net_warn_ratelimited(
2882 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2887 static const struct skb_checksum_ops default_crc32c_ops = {
2888 .update = warn_crc32c_csum_update,
2889 .combine = warn_crc32c_csum_combine,
2892 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2893 &default_crc32c_ops;
2894 EXPORT_SYMBOL(crc32c_csum_stub);
2897 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2898 * @from: source buffer
2900 * Calculates the amount of linear headroom needed in the 'to' skb passed
2901 * into skb_zerocopy().
2904 skb_zerocopy_headlen(const struct sk_buff *from)
2906 unsigned int hlen = 0;
2908 if (!from->head_frag ||
2909 skb_headlen(from) < L1_CACHE_BYTES ||
2910 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2911 hlen = skb_headlen(from);
2913 if (skb_has_frag_list(from))
2918 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2921 * skb_zerocopy - Zero copy skb to skb
2922 * @to: destination buffer
2923 * @from: source buffer
2924 * @len: number of bytes to copy from source buffer
2925 * @hlen: size of linear headroom in destination buffer
2927 * Copies up to `len` bytes from `from` to `to` by creating references
2928 * to the frags in the source buffer.
2930 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2931 * headroom in the `to` buffer.
2934 * 0: everything is OK
2935 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2936 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2939 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2942 int plen = 0; /* length of skb->head fragment */
2945 unsigned int offset;
2947 BUG_ON(!from->head_frag && !hlen);
2949 /* dont bother with small payloads */
2950 if (len <= skb_tailroom(to))
2951 return skb_copy_bits(from, 0, skb_put(to, len), len);
2954 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2959 plen = min_t(int, skb_headlen(from), len);
2961 page = virt_to_head_page(from->head);
2962 offset = from->data - (unsigned char *)page_address(page);
2963 __skb_fill_page_desc(to, 0, page, offset, plen);
2970 to->truesize += len + plen;
2971 to->len += len + plen;
2972 to->data_len += len + plen;
2974 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2978 skb_zerocopy_clone(to, from, GFP_ATOMIC);
2980 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2985 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2986 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
2988 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
2990 skb_frag_ref(to, j);
2993 skb_shinfo(to)->nr_frags = j;
2997 EXPORT_SYMBOL_GPL(skb_zerocopy);
2999 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3004 if (skb->ip_summed == CHECKSUM_PARTIAL)
3005 csstart = skb_checksum_start_offset(skb);
3007 csstart = skb_headlen(skb);
3009 BUG_ON(csstart > skb_headlen(skb));
3011 skb_copy_from_linear_data(skb, to, csstart);
3014 if (csstart != skb->len)
3015 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3016 skb->len - csstart, 0);
3018 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3019 long csstuff = csstart + skb->csum_offset;
3021 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3024 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3027 * skb_dequeue - remove from the head of the queue
3028 * @list: list to dequeue from
3030 * Remove the head of the list. The list lock is taken so the function
3031 * may be used safely with other locking list functions. The head item is
3032 * returned or %NULL if the list is empty.
3035 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3037 unsigned long flags;
3038 struct sk_buff *result;
3040 spin_lock_irqsave(&list->lock, flags);
3041 result = __skb_dequeue(list);
3042 spin_unlock_irqrestore(&list->lock, flags);
3045 EXPORT_SYMBOL(skb_dequeue);
3048 * skb_dequeue_tail - remove from the tail of the queue
3049 * @list: list to dequeue from
3051 * Remove the tail of the list. The list lock is taken so the function
3052 * may be used safely with other locking list functions. The tail item is
3053 * returned or %NULL if the list is empty.
3055 struct sk_buff *skb_dequeue_tail(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_tail(list);
3062 spin_unlock_irqrestore(&list->lock, flags);
3065 EXPORT_SYMBOL(skb_dequeue_tail);
3068 * skb_queue_purge - empty a list
3069 * @list: list to empty
3071 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3072 * the list and one reference dropped. This function takes the list
3073 * lock and is atomic with respect to other list locking functions.
3075 void skb_queue_purge(struct sk_buff_head *list)
3077 struct sk_buff *skb;
3078 while ((skb = skb_dequeue(list)) != NULL)
3081 EXPORT_SYMBOL(skb_queue_purge);
3084 * skb_rbtree_purge - empty a skb rbtree
3085 * @root: root of the rbtree to empty
3086 * Return value: the sum of truesizes of all purged skbs.
3088 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3089 * the list and one reference dropped. This function does not take
3090 * any lock. Synchronization should be handled by the caller (e.g., TCP
3091 * out-of-order queue is protected by the socket lock).
3093 unsigned int skb_rbtree_purge(struct rb_root *root)
3095 struct rb_node *p = rb_first(root);
3096 unsigned int sum = 0;
3099 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3102 rb_erase(&skb->rbnode, root);
3103 sum += skb->truesize;
3110 * skb_queue_head - queue a buffer at the list head
3111 * @list: list to use
3112 * @newsk: buffer to queue
3114 * Queue a buffer at the start of the list. This function takes the
3115 * list lock and can be used safely with other locking &sk_buff functions
3118 * A buffer cannot be placed on two lists at the same time.
3120 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3122 unsigned long flags;
3124 spin_lock_irqsave(&list->lock, flags);
3125 __skb_queue_head(list, newsk);
3126 spin_unlock_irqrestore(&list->lock, flags);
3128 EXPORT_SYMBOL(skb_queue_head);
3131 * skb_queue_tail - queue a buffer at the list tail
3132 * @list: list to use
3133 * @newsk: buffer to queue
3135 * Queue a buffer at the tail of the list. This function takes the
3136 * list lock and can be used safely with other locking &sk_buff functions
3139 * A buffer cannot be placed on two lists at the same time.
3141 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3143 unsigned long flags;
3145 spin_lock_irqsave(&list->lock, flags);
3146 __skb_queue_tail(list, newsk);
3147 spin_unlock_irqrestore(&list->lock, flags);
3149 EXPORT_SYMBOL(skb_queue_tail);
3152 * skb_unlink - remove a buffer from a list
3153 * @skb: buffer to remove
3154 * @list: list to use
3156 * Remove a packet from a list. The list locks are taken and this
3157 * function is atomic with respect to other list locked calls
3159 * You must know what list the SKB is on.
3161 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3163 unsigned long flags;
3165 spin_lock_irqsave(&list->lock, flags);
3166 __skb_unlink(skb, list);
3167 spin_unlock_irqrestore(&list->lock, flags);
3169 EXPORT_SYMBOL(skb_unlink);
3172 * skb_append - append a buffer
3173 * @old: buffer to insert after
3174 * @newsk: buffer to insert
3175 * @list: list to use
3177 * Place a packet after a given packet in a list. The list locks are taken
3178 * and this function is atomic with respect to other list locked calls.
3179 * A buffer cannot be placed on two lists at the same time.
3181 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3183 unsigned long flags;
3185 spin_lock_irqsave(&list->lock, flags);
3186 __skb_queue_after(list, old, newsk);
3187 spin_unlock_irqrestore(&list->lock, flags);
3189 EXPORT_SYMBOL(skb_append);
3191 static inline void skb_split_inside_header(struct sk_buff *skb,
3192 struct sk_buff* skb1,
3193 const u32 len, const int pos)
3197 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3199 /* And move data appendix as is. */
3200 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3201 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3203 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3204 skb_shinfo(skb)->nr_frags = 0;
3205 skb1->data_len = skb->data_len;
3206 skb1->len += skb1->data_len;
3209 skb_set_tail_pointer(skb, len);
3212 static inline void skb_split_no_header(struct sk_buff *skb,
3213 struct sk_buff* skb1,
3214 const u32 len, int pos)
3217 const int nfrags = skb_shinfo(skb)->nr_frags;
3219 skb_shinfo(skb)->nr_frags = 0;
3220 skb1->len = skb1->data_len = skb->len - len;
3222 skb->data_len = len - pos;
3224 for (i = 0; i < nfrags; i++) {
3225 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3227 if (pos + size > len) {
3228 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3232 * We have two variants in this case:
3233 * 1. Move all the frag to the second
3234 * part, if it is possible. F.e.
3235 * this approach is mandatory for TUX,
3236 * where splitting is expensive.
3237 * 2. Split is accurately. We make this.
3239 skb_frag_ref(skb, i);
3240 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3241 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3242 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3243 skb_shinfo(skb)->nr_frags++;
3247 skb_shinfo(skb)->nr_frags++;
3250 skb_shinfo(skb1)->nr_frags = k;
3254 * skb_split - Split fragmented skb to two parts at length len.
3255 * @skb: the buffer to split
3256 * @skb1: the buffer to receive the second part
3257 * @len: new length for skb
3259 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3261 int pos = skb_headlen(skb);
3263 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3265 skb_zerocopy_clone(skb1, skb, 0);
3266 if (len < pos) /* Split line is inside header. */
3267 skb_split_inside_header(skb, skb1, len, pos);
3268 else /* Second chunk has no header, nothing to copy. */
3269 skb_split_no_header(skb, skb1, len, pos);
3271 EXPORT_SYMBOL(skb_split);
3273 /* Shifting from/to a cloned skb is a no-go.
3275 * Caller cannot keep skb_shinfo related pointers past calling here!
3277 static int skb_prepare_for_shift(struct sk_buff *skb)
3279 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3283 * skb_shift - Shifts paged data partially from skb to another
3284 * @tgt: buffer into which tail data gets added
3285 * @skb: buffer from which the paged data comes from
3286 * @shiftlen: shift up to this many bytes
3288 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3289 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3290 * It's up to caller to free skb if everything was shifted.
3292 * If @tgt runs out of frags, the whole operation is aborted.
3294 * Skb cannot include anything else but paged data while tgt is allowed
3295 * to have non-paged data as well.
3297 * TODO: full sized shift could be optimized but that would need
3298 * specialized skb free'er to handle frags without up-to-date nr_frags.
3300 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3302 int from, to, merge, todo;
3303 skb_frag_t *fragfrom, *fragto;
3305 BUG_ON(shiftlen > skb->len);
3307 if (skb_headlen(skb))
3309 if (skb_zcopy(tgt) || skb_zcopy(skb))
3314 to = skb_shinfo(tgt)->nr_frags;
3315 fragfrom = &skb_shinfo(skb)->frags[from];
3317 /* Actual merge is delayed until the point when we know we can
3318 * commit all, so that we don't have to undo partial changes
3321 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3322 skb_frag_off(fragfrom))) {
3327 todo -= skb_frag_size(fragfrom);
3329 if (skb_prepare_for_shift(skb) ||
3330 skb_prepare_for_shift(tgt))
3333 /* All previous frag pointers might be stale! */
3334 fragfrom = &skb_shinfo(skb)->frags[from];
3335 fragto = &skb_shinfo(tgt)->frags[merge];
3337 skb_frag_size_add(fragto, shiftlen);
3338 skb_frag_size_sub(fragfrom, shiftlen);
3339 skb_frag_off_add(fragfrom, shiftlen);
3347 /* Skip full, not-fitting skb to avoid expensive operations */
3348 if ((shiftlen == skb->len) &&
3349 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3352 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3355 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3356 if (to == MAX_SKB_FRAGS)
3359 fragfrom = &skb_shinfo(skb)->frags[from];
3360 fragto = &skb_shinfo(tgt)->frags[to];
3362 if (todo >= skb_frag_size(fragfrom)) {
3363 *fragto = *fragfrom;
3364 todo -= skb_frag_size(fragfrom);
3369 __skb_frag_ref(fragfrom);
3370 skb_frag_page_copy(fragto, fragfrom);
3371 skb_frag_off_copy(fragto, fragfrom);
3372 skb_frag_size_set(fragto, todo);
3374 skb_frag_off_add(fragfrom, todo);
3375 skb_frag_size_sub(fragfrom, todo);
3383 /* Ready to "commit" this state change to tgt */
3384 skb_shinfo(tgt)->nr_frags = to;
3387 fragfrom = &skb_shinfo(skb)->frags[0];
3388 fragto = &skb_shinfo(tgt)->frags[merge];
3390 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3391 __skb_frag_unref(fragfrom);
3394 /* Reposition in the original skb */
3396 while (from < skb_shinfo(skb)->nr_frags)
3397 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3398 skb_shinfo(skb)->nr_frags = to;
3400 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3403 /* Most likely the tgt won't ever need its checksum anymore, skb on
3404 * the other hand might need it if it needs to be resent
3406 tgt->ip_summed = CHECKSUM_PARTIAL;
3407 skb->ip_summed = CHECKSUM_PARTIAL;
3409 /* Yak, is it really working this way? Some helper please? */
3410 skb->len -= shiftlen;
3411 skb->data_len -= shiftlen;
3412 skb->truesize -= shiftlen;
3413 tgt->len += shiftlen;
3414 tgt->data_len += shiftlen;
3415 tgt->truesize += shiftlen;
3421 * skb_prepare_seq_read - Prepare a sequential read of skb data
3422 * @skb: the buffer to read
3423 * @from: lower offset of data to be read
3424 * @to: upper offset of data to be read
3425 * @st: state variable
3427 * Initializes the specified state variable. Must be called before
3428 * invoking skb_seq_read() for the first time.
3430 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3431 unsigned int to, struct skb_seq_state *st)
3433 st->lower_offset = from;
3434 st->upper_offset = to;
3435 st->root_skb = st->cur_skb = skb;
3436 st->frag_idx = st->stepped_offset = 0;
3437 st->frag_data = NULL;
3439 EXPORT_SYMBOL(skb_prepare_seq_read);
3442 * skb_seq_read - Sequentially read skb data
3443 * @consumed: number of bytes consumed by the caller so far
3444 * @data: destination pointer for data to be returned
3445 * @st: state variable
3447 * Reads a block of skb data at @consumed relative to the
3448 * lower offset specified to skb_prepare_seq_read(). Assigns
3449 * the head of the data block to @data and returns the length
3450 * of the block or 0 if the end of the skb data or the upper
3451 * offset has been reached.
3453 * The caller is not required to consume all of the data
3454 * returned, i.e. @consumed is typically set to the number
3455 * of bytes already consumed and the next call to
3456 * skb_seq_read() will return the remaining part of the block.
3458 * Note 1: The size of each block of data returned can be arbitrary,
3459 * this limitation is the cost for zerocopy sequential
3460 * reads of potentially non linear data.
3462 * Note 2: Fragment lists within fragments are not implemented
3463 * at the moment, state->root_skb could be replaced with
3464 * a stack for this purpose.
3466 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3467 struct skb_seq_state *st)
3469 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3472 if (unlikely(abs_offset >= st->upper_offset)) {
3473 if (st->frag_data) {
3474 kunmap_atomic(st->frag_data);
3475 st->frag_data = NULL;
3481 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3483 if (abs_offset < block_limit && !st->frag_data) {
3484 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3485 return block_limit - abs_offset;
3488 if (st->frag_idx == 0 && !st->frag_data)
3489 st->stepped_offset += skb_headlen(st->cur_skb);
3491 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3492 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3493 block_limit = skb_frag_size(frag) + st->stepped_offset;
3495 if (abs_offset < block_limit) {
3497 st->frag_data = kmap_atomic(skb_frag_page(frag));
3499 *data = (u8 *) st->frag_data + skb_frag_off(frag) +
3500 (abs_offset - st->stepped_offset);
3502 return block_limit - abs_offset;
3505 if (st->frag_data) {
3506 kunmap_atomic(st->frag_data);
3507 st->frag_data = NULL;
3511 st->stepped_offset += skb_frag_size(frag);
3514 if (st->frag_data) {
3515 kunmap_atomic(st->frag_data);
3516 st->frag_data = NULL;
3519 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3520 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3523 } else if (st->cur_skb->next) {
3524 st->cur_skb = st->cur_skb->next;
3531 EXPORT_SYMBOL(skb_seq_read);
3534 * skb_abort_seq_read - Abort a sequential read of skb data
3535 * @st: state variable
3537 * Must be called if skb_seq_read() was not called until it
3540 void skb_abort_seq_read(struct skb_seq_state *st)
3543 kunmap_atomic(st->frag_data);
3545 EXPORT_SYMBOL(skb_abort_seq_read);
3547 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3549 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3550 struct ts_config *conf,
3551 struct ts_state *state)
3553 return skb_seq_read(offset, text, TS_SKB_CB(state));
3556 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3558 skb_abort_seq_read(TS_SKB_CB(state));
3562 * skb_find_text - Find a text pattern in skb data
3563 * @skb: the buffer to look in
3564 * @from: search offset
3566 * @config: textsearch configuration
3568 * Finds a pattern in the skb data according to the specified
3569 * textsearch configuration. Use textsearch_next() to retrieve
3570 * subsequent occurrences of the pattern. Returns the offset
3571 * to the first occurrence or UINT_MAX if no match was found.
3573 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3574 unsigned int to, struct ts_config *config)
3576 struct ts_state state;
3579 config->get_next_block = skb_ts_get_next_block;
3580 config->finish = skb_ts_finish;
3582 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3584 ret = textsearch_find(config, &state);
3585 return (ret <= to - from ? ret : UINT_MAX);
3587 EXPORT_SYMBOL(skb_find_text);
3589 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3590 int offset, size_t size)
3592 int i = skb_shinfo(skb)->nr_frags;
3594 if (skb_can_coalesce(skb, i, page, offset)) {
3595 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3596 } else if (i < MAX_SKB_FRAGS) {
3598 skb_fill_page_desc(skb, i, page, offset, size);
3605 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3608 * skb_pull_rcsum - pull skb and update receive checksum
3609 * @skb: buffer to update
3610 * @len: length of data pulled
3612 * This function performs an skb_pull on the packet and updates
3613 * the CHECKSUM_COMPLETE checksum. It should be used on
3614 * receive path processing instead of skb_pull unless you know
3615 * that the checksum difference is zero (e.g., a valid IP header)
3616 * or you are setting ip_summed to CHECKSUM_NONE.
3618 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3620 unsigned char *data = skb->data;
3622 BUG_ON(len > skb->len);
3623 __skb_pull(skb, len);
3624 skb_postpull_rcsum(skb, data, len);
3627 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3629 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3631 skb_frag_t head_frag;
3634 page = virt_to_head_page(frag_skb->head);
3635 __skb_frag_set_page(&head_frag, page);
3636 skb_frag_off_set(&head_frag, frag_skb->data -
3637 (unsigned char *)page_address(page));
3638 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3642 struct sk_buff *skb_segment_list(struct sk_buff *skb,
3643 netdev_features_t features,
3644 unsigned int offset)
3646 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
3647 unsigned int tnl_hlen = skb_tnl_header_len(skb);
3648 unsigned int delta_truesize = 0;
3649 unsigned int delta_len = 0;
3650 struct sk_buff *tail = NULL;
3651 struct sk_buff *nskb;
3653 skb_push(skb, -skb_network_offset(skb) + offset);
3655 skb_shinfo(skb)->frag_list = NULL;
3659 list_skb = list_skb->next;
3668 delta_len += nskb->len;
3669 delta_truesize += nskb->truesize;
3671 skb_push(nskb, -skb_network_offset(nskb) + offset);
3673 skb_release_head_state(nskb);
3674 __copy_skb_header(nskb, skb);
3676 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
3677 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
3678 nskb->data - tnl_hlen,
3681 if (skb_needs_linearize(nskb, features) &&
3682 __skb_linearize(nskb))
3687 skb->truesize = skb->truesize - delta_truesize;
3688 skb->data_len = skb->data_len - delta_len;
3689 skb->len = skb->len - delta_len;
3695 if (skb_needs_linearize(skb, features) &&
3696 __skb_linearize(skb))
3704 kfree_skb_list(skb->next);
3706 return ERR_PTR(-ENOMEM);
3708 EXPORT_SYMBOL_GPL(skb_segment_list);
3710 int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
3712 if (unlikely(p->len + skb->len >= 65536))
3715 if (NAPI_GRO_CB(p)->last == p)
3716 skb_shinfo(p)->frag_list = skb;
3718 NAPI_GRO_CB(p)->last->next = skb;
3720 skb_pull(skb, skb_gro_offset(skb));
3722 NAPI_GRO_CB(p)->last = skb;
3723 NAPI_GRO_CB(p)->count++;
3724 p->data_len += skb->len;
3725 p->truesize += skb->truesize;
3728 NAPI_GRO_CB(skb)->same_flow = 1;
3734 * skb_segment - Perform protocol segmentation on skb.
3735 * @head_skb: buffer to segment
3736 * @features: features for the output path (see dev->features)
3738 * This function performs segmentation on the given skb. It returns
3739 * a pointer to the first in a list of new skbs for the segments.
3740 * In case of error it returns ERR_PTR(err).
3742 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3743 netdev_features_t features)
3745 struct sk_buff *segs = NULL;
3746 struct sk_buff *tail = NULL;
3747 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3748 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3749 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3750 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3751 struct sk_buff *frag_skb = head_skb;
3752 unsigned int offset = doffset;
3753 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3754 unsigned int partial_segs = 0;
3755 unsigned int headroom;
3756 unsigned int len = head_skb->len;
3759 int nfrags = skb_shinfo(head_skb)->nr_frags;
3764 if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3765 (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3766 /* gso_size is untrusted, and we have a frag_list with a linear
3767 * non head_frag head.
3769 * (we assume checking the first list_skb member suffices;
3770 * i.e if either of the list_skb members have non head_frag
3771 * head, then the first one has too).
3773 * If head_skb's headlen does not fit requested gso_size, it
3774 * means that the frag_list members do NOT terminate on exact
3775 * gso_size boundaries. Hence we cannot perform skb_frag_t page
3776 * sharing. Therefore we must fallback to copying the frag_list
3777 * skbs; we do so by disabling SG.
3779 if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3780 features &= ~NETIF_F_SG;
3783 __skb_push(head_skb, doffset);
3784 proto = skb_network_protocol(head_skb, NULL);
3785 if (unlikely(!proto))
3786 return ERR_PTR(-EINVAL);
3788 sg = !!(features & NETIF_F_SG);
3789 csum = !!can_checksum_protocol(features, proto);
3791 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3792 if (!(features & NETIF_F_GSO_PARTIAL)) {
3793 struct sk_buff *iter;
3794 unsigned int frag_len;
3797 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3800 /* If we get here then all the required
3801 * GSO features except frag_list are supported.
3802 * Try to split the SKB to multiple GSO SKBs
3803 * with no frag_list.
3804 * Currently we can do that only when the buffers don't
3805 * have a linear part and all the buffers except
3806 * the last are of the same length.
3808 frag_len = list_skb->len;
3809 skb_walk_frags(head_skb, iter) {
3810 if (frag_len != iter->len && iter->next)
3812 if (skb_headlen(iter) && !iter->head_frag)
3818 if (len != frag_len)
3822 /* GSO partial only requires that we trim off any excess that
3823 * doesn't fit into an MSS sized block, so take care of that
3826 partial_segs = len / mss;
3827 if (partial_segs > 1)
3828 mss *= partial_segs;
3834 headroom = skb_headroom(head_skb);
3835 pos = skb_headlen(head_skb);
3838 struct sk_buff *nskb;
3839 skb_frag_t *nskb_frag;
3843 if (unlikely(mss == GSO_BY_FRAGS)) {
3844 len = list_skb->len;
3846 len = head_skb->len - offset;
3851 hsize = skb_headlen(head_skb) - offset;
3854 if (hsize > len || !sg)
3857 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3858 (skb_headlen(list_skb) == len || sg)) {
3859 BUG_ON(skb_headlen(list_skb) > len);
3862 nfrags = skb_shinfo(list_skb)->nr_frags;
3863 frag = skb_shinfo(list_skb)->frags;
3864 frag_skb = list_skb;
3865 pos += skb_headlen(list_skb);
3867 while (pos < offset + len) {
3868 BUG_ON(i >= nfrags);
3870 size = skb_frag_size(frag);
3871 if (pos + size > offset + len)
3879 nskb = skb_clone(list_skb, GFP_ATOMIC);
3880 list_skb = list_skb->next;
3882 if (unlikely(!nskb))
3885 if (unlikely(pskb_trim(nskb, len))) {
3890 hsize = skb_end_offset(nskb);
3891 if (skb_cow_head(nskb, doffset + headroom)) {
3896 nskb->truesize += skb_end_offset(nskb) - hsize;
3897 skb_release_head_state(nskb);
3898 __skb_push(nskb, doffset);
3900 nskb = __alloc_skb(hsize + doffset + headroom,
3901 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3904 if (unlikely(!nskb))
3907 skb_reserve(nskb, headroom);
3908 __skb_put(nskb, doffset);
3917 __copy_skb_header(nskb, head_skb);
3919 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3920 skb_reset_mac_len(nskb);
3922 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3923 nskb->data - tnl_hlen,
3924 doffset + tnl_hlen);
3926 if (nskb->len == len + doffset)
3927 goto perform_csum_check;
3931 if (!nskb->remcsum_offload)
3932 nskb->ip_summed = CHECKSUM_NONE;
3933 SKB_GSO_CB(nskb)->csum =
3934 skb_copy_and_csum_bits(head_skb, offset,
3938 SKB_GSO_CB(nskb)->csum_start =
3939 skb_headroom(nskb) + doffset;
3941 skb_copy_bits(head_skb, offset,
3948 nskb_frag = skb_shinfo(nskb)->frags;
3950 skb_copy_from_linear_data_offset(head_skb, offset,
3951 skb_put(nskb, hsize), hsize);
3953 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3956 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3957 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
3960 while (pos < offset + len) {
3963 nfrags = skb_shinfo(list_skb)->nr_frags;
3964 frag = skb_shinfo(list_skb)->frags;
3965 frag_skb = list_skb;
3966 if (!skb_headlen(list_skb)) {
3969 BUG_ON(!list_skb->head_frag);
3971 /* to make room for head_frag. */
3975 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3976 skb_zerocopy_clone(nskb, frag_skb,
3980 list_skb = list_skb->next;
3983 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3985 net_warn_ratelimited(
3986 "skb_segment: too many frags: %u %u\n",
3992 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
3993 __skb_frag_ref(nskb_frag);
3994 size = skb_frag_size(nskb_frag);
3997 skb_frag_off_add(nskb_frag, offset - pos);
3998 skb_frag_size_sub(nskb_frag, offset - pos);
4001 skb_shinfo(nskb)->nr_frags++;
4003 if (pos + size <= offset + len) {
4008 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4016 nskb->data_len = len - hsize;
4017 nskb->len += nskb->data_len;
4018 nskb->truesize += nskb->data_len;
4022 if (skb_has_shared_frag(nskb) &&
4023 __skb_linearize(nskb))
4026 if (!nskb->remcsum_offload)
4027 nskb->ip_summed = CHECKSUM_NONE;
4028 SKB_GSO_CB(nskb)->csum =
4029 skb_checksum(nskb, doffset,
4030 nskb->len - doffset, 0);
4031 SKB_GSO_CB(nskb)->csum_start =
4032 skb_headroom(nskb) + doffset;
4034 } while ((offset += len) < head_skb->len);
4036 /* Some callers want to get the end of the list.
4037 * Put it in segs->prev to avoid walking the list.
4038 * (see validate_xmit_skb_list() for example)
4043 struct sk_buff *iter;
4044 int type = skb_shinfo(head_skb)->gso_type;
4045 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4047 /* Update type to add partial and then remove dodgy if set */
4048 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4049 type &= ~SKB_GSO_DODGY;
4051 /* Update GSO info and prepare to start updating headers on
4052 * our way back down the stack of protocols.
4054 for (iter = segs; iter; iter = iter->next) {
4055 skb_shinfo(iter)->gso_size = gso_size;
4056 skb_shinfo(iter)->gso_segs = partial_segs;
4057 skb_shinfo(iter)->gso_type = type;
4058 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4061 if (tail->len - doffset <= gso_size)
4062 skb_shinfo(tail)->gso_size = 0;
4063 else if (tail != segs)
4064 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4067 /* Following permits correct backpressure, for protocols
4068 * using skb_set_owner_w().
4069 * Idea is to tranfert ownership from head_skb to last segment.
4071 if (head_skb->destructor == sock_wfree) {
4072 swap(tail->truesize, head_skb->truesize);
4073 swap(tail->destructor, head_skb->destructor);
4074 swap(tail->sk, head_skb->sk);
4079 kfree_skb_list(segs);
4080 return ERR_PTR(err);
4082 EXPORT_SYMBOL_GPL(skb_segment);
4084 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4086 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4087 unsigned int offset = skb_gro_offset(skb);
4088 unsigned int headlen = skb_headlen(skb);
4089 unsigned int len = skb_gro_len(skb);
4090 unsigned int delta_truesize;
4093 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4096 lp = NAPI_GRO_CB(p)->last;
4097 pinfo = skb_shinfo(lp);
4099 if (headlen <= offset) {
4102 int i = skbinfo->nr_frags;
4103 int nr_frags = pinfo->nr_frags + i;
4105 if (nr_frags > MAX_SKB_FRAGS)
4109 pinfo->nr_frags = nr_frags;
4110 skbinfo->nr_frags = 0;
4112 frag = pinfo->frags + nr_frags;
4113 frag2 = skbinfo->frags + i;
4118 skb_frag_off_add(frag, offset);
4119 skb_frag_size_sub(frag, offset);
4121 /* all fragments truesize : remove (head size + sk_buff) */
4122 delta_truesize = skb->truesize -
4123 SKB_TRUESIZE(skb_end_offset(skb));
4125 skb->truesize -= skb->data_len;
4126 skb->len -= skb->data_len;
4129 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4131 } else if (skb->head_frag) {
4132 int nr_frags = pinfo->nr_frags;
4133 skb_frag_t *frag = pinfo->frags + nr_frags;
4134 struct page *page = virt_to_head_page(skb->head);
4135 unsigned int first_size = headlen - offset;
4136 unsigned int first_offset;
4138 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4141 first_offset = skb->data -
4142 (unsigned char *)page_address(page) +
4145 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4147 __skb_frag_set_page(frag, page);
4148 skb_frag_off_set(frag, first_offset);
4149 skb_frag_size_set(frag, first_size);
4151 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4152 /* We dont need to clear skbinfo->nr_frags here */
4154 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4155 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4160 delta_truesize = skb->truesize;
4161 if (offset > headlen) {
4162 unsigned int eat = offset - headlen;
4164 skb_frag_off_add(&skbinfo->frags[0], eat);
4165 skb_frag_size_sub(&skbinfo->frags[0], eat);
4166 skb->data_len -= eat;
4171 __skb_pull(skb, offset);
4173 if (NAPI_GRO_CB(p)->last == p)
4174 skb_shinfo(p)->frag_list = skb;
4176 NAPI_GRO_CB(p)->last->next = skb;
4177 NAPI_GRO_CB(p)->last = skb;
4178 __skb_header_release(skb);
4182 NAPI_GRO_CB(p)->count++;
4184 p->truesize += delta_truesize;
4187 lp->data_len += len;
4188 lp->truesize += delta_truesize;
4191 NAPI_GRO_CB(skb)->same_flow = 1;
4195 #ifdef CONFIG_SKB_EXTENSIONS
4196 #define SKB_EXT_ALIGN_VALUE 8
4197 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4199 static const u8 skb_ext_type_len[] = {
4200 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4201 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4204 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4206 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4207 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4209 #if IS_ENABLED(CONFIG_MPTCP)
4210 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4214 static __always_inline unsigned int skb_ext_total_length(void)
4216 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4217 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4218 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4221 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4223 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4224 skb_ext_type_len[TC_SKB_EXT] +
4226 #if IS_ENABLED(CONFIG_MPTCP)
4227 skb_ext_type_len[SKB_EXT_MPTCP] +
4232 static void skb_extensions_init(void)
4234 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4235 BUILD_BUG_ON(skb_ext_total_length() > 255);
4237 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4238 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4240 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4244 static void skb_extensions_init(void) {}
4247 void __init skb_init(void)
4249 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4250 sizeof(struct sk_buff),
4252 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4253 offsetof(struct sk_buff, cb),
4254 sizeof_field(struct sk_buff, cb),
4256 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4257 sizeof(struct sk_buff_fclones),
4259 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4261 skb_extensions_init();
4265 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4266 unsigned int recursion_level)
4268 int start = skb_headlen(skb);
4269 int i, copy = start - offset;
4270 struct sk_buff *frag_iter;
4273 if (unlikely(recursion_level >= 24))
4279 sg_set_buf(sg, skb->data + offset, copy);
4281 if ((len -= copy) == 0)
4286 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4289 WARN_ON(start > offset + len);
4291 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4292 if ((copy = end - offset) > 0) {
4293 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4294 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4299 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4300 skb_frag_off(frag) + offset - start);
4309 skb_walk_frags(skb, frag_iter) {
4312 WARN_ON(start > offset + len);
4314 end = start + frag_iter->len;
4315 if ((copy = end - offset) > 0) {
4316 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4321 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4322 copy, recursion_level + 1);
4323 if (unlikely(ret < 0))
4326 if ((len -= copy) == 0)
4337 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4338 * @skb: Socket buffer containing the buffers to be mapped
4339 * @sg: The scatter-gather list to map into
4340 * @offset: The offset into the buffer's contents to start mapping
4341 * @len: Length of buffer space to be mapped
4343 * Fill the specified scatter-gather list with mappings/pointers into a
4344 * region of the buffer space attached to a socket buffer. Returns either
4345 * the number of scatterlist items used, or -EMSGSIZE if the contents
4348 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4350 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4355 sg_mark_end(&sg[nsg - 1]);
4359 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4361 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4362 * sglist without mark the sg which contain last skb data as the end.
4363 * So the caller can mannipulate sg list as will when padding new data after
4364 * the first call without calling sg_unmark_end to expend sg list.
4366 * Scenario to use skb_to_sgvec_nomark:
4368 * 2. skb_to_sgvec_nomark(payload1)
4369 * 3. skb_to_sgvec_nomark(payload2)
4371 * This is equivalent to:
4373 * 2. skb_to_sgvec(payload1)
4375 * 4. skb_to_sgvec(payload2)
4377 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4378 * is more preferable.
4380 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4381 int offset, int len)
4383 return __skb_to_sgvec(skb, sg, offset, len, 0);
4385 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4390 * skb_cow_data - Check that a socket buffer's data buffers are writable
4391 * @skb: The socket buffer to check.
4392 * @tailbits: Amount of trailing space to be added
4393 * @trailer: Returned pointer to the skb where the @tailbits space begins
4395 * Make sure that the data buffers attached to a socket buffer are
4396 * writable. If they are not, private copies are made of the data buffers
4397 * and the socket buffer is set to use these instead.
4399 * If @tailbits is given, make sure that there is space to write @tailbits
4400 * bytes of data beyond current end of socket buffer. @trailer will be
4401 * set to point to the skb in which this space begins.
4403 * The number of scatterlist elements required to completely map the
4404 * COW'd and extended socket buffer will be returned.
4406 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4410 struct sk_buff *skb1, **skb_p;
4412 /* If skb is cloned or its head is paged, reallocate
4413 * head pulling out all the pages (pages are considered not writable
4414 * at the moment even if they are anonymous).
4416 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4417 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4420 /* Easy case. Most of packets will go this way. */
4421 if (!skb_has_frag_list(skb)) {
4422 /* A little of trouble, not enough of space for trailer.
4423 * This should not happen, when stack is tuned to generate
4424 * good frames. OK, on miss we reallocate and reserve even more
4425 * space, 128 bytes is fair. */
4427 if (skb_tailroom(skb) < tailbits &&
4428 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4436 /* Misery. We are in troubles, going to mincer fragments... */
4439 skb_p = &skb_shinfo(skb)->frag_list;
4442 while ((skb1 = *skb_p) != NULL) {
4445 /* The fragment is partially pulled by someone,
4446 * this can happen on input. Copy it and everything
4449 if (skb_shared(skb1))
4452 /* If the skb is the last, worry about trailer. */
4454 if (skb1->next == NULL && tailbits) {
4455 if (skb_shinfo(skb1)->nr_frags ||
4456 skb_has_frag_list(skb1) ||
4457 skb_tailroom(skb1) < tailbits)
4458 ntail = tailbits + 128;
4464 skb_shinfo(skb1)->nr_frags ||
4465 skb_has_frag_list(skb1)) {
4466 struct sk_buff *skb2;
4468 /* Fuck, we are miserable poor guys... */
4470 skb2 = skb_copy(skb1, GFP_ATOMIC);
4472 skb2 = skb_copy_expand(skb1,
4476 if (unlikely(skb2 == NULL))
4480 skb_set_owner_w(skb2, skb1->sk);
4482 /* Looking around. Are we still alive?
4483 * OK, link new skb, drop old one */
4485 skb2->next = skb1->next;
4492 skb_p = &skb1->next;
4497 EXPORT_SYMBOL_GPL(skb_cow_data);
4499 static void sock_rmem_free(struct sk_buff *skb)
4501 struct sock *sk = skb->sk;
4503 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4506 static void skb_set_err_queue(struct sk_buff *skb)
4508 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4509 * So, it is safe to (mis)use it to mark skbs on the error queue.
4511 skb->pkt_type = PACKET_OUTGOING;
4512 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4516 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4518 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4520 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4521 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4526 skb->destructor = sock_rmem_free;
4527 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4528 skb_set_err_queue(skb);
4530 /* before exiting rcu section, make sure dst is refcounted */
4533 skb_queue_tail(&sk->sk_error_queue, skb);
4534 if (!sock_flag(sk, SOCK_DEAD))
4535 sk->sk_error_report(sk);
4538 EXPORT_SYMBOL(sock_queue_err_skb);
4540 static bool is_icmp_err_skb(const struct sk_buff *skb)
4542 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4543 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4546 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4548 struct sk_buff_head *q = &sk->sk_error_queue;
4549 struct sk_buff *skb, *skb_next = NULL;
4550 bool icmp_next = false;
4551 unsigned long flags;
4553 spin_lock_irqsave(&q->lock, flags);
4554 skb = __skb_dequeue(q);
4555 if (skb && (skb_next = skb_peek(q))) {
4556 icmp_next = is_icmp_err_skb(skb_next);
4558 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin;
4560 spin_unlock_irqrestore(&q->lock, flags);
4562 if (is_icmp_err_skb(skb) && !icmp_next)
4566 sk->sk_error_report(sk);
4570 EXPORT_SYMBOL(sock_dequeue_err_skb);
4573 * skb_clone_sk - create clone of skb, and take reference to socket
4574 * @skb: the skb to clone
4576 * This function creates a clone of a buffer that holds a reference on
4577 * sk_refcnt. Buffers created via this function are meant to be
4578 * returned using sock_queue_err_skb, or free via kfree_skb.
4580 * When passing buffers allocated with this function to sock_queue_err_skb
4581 * it is necessary to wrap the call with sock_hold/sock_put in order to
4582 * prevent the socket from being released prior to being enqueued on
4583 * the sk_error_queue.
4585 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4587 struct sock *sk = skb->sk;
4588 struct sk_buff *clone;
4590 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4593 clone = skb_clone(skb, GFP_ATOMIC);
4600 clone->destructor = sock_efree;
4604 EXPORT_SYMBOL(skb_clone_sk);
4606 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4611 struct sock_exterr_skb *serr;
4614 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4616 serr = SKB_EXT_ERR(skb);
4617 memset(serr, 0, sizeof(*serr));
4618 serr->ee.ee_errno = ENOMSG;
4619 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4620 serr->ee.ee_info = tstype;
4621 serr->opt_stats = opt_stats;
4622 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4623 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4624 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4625 if (sk->sk_protocol == IPPROTO_TCP &&
4626 sk->sk_type == SOCK_STREAM)
4627 serr->ee.ee_data -= sk->sk_tskey;
4630 err = sock_queue_err_skb(sk, skb);
4636 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4640 if (likely(sysctl_tstamp_allow_data || tsonly))
4643 read_lock_bh(&sk->sk_callback_lock);
4644 ret = sk->sk_socket && sk->sk_socket->file &&
4645 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4646 read_unlock_bh(&sk->sk_callback_lock);
4650 void skb_complete_tx_timestamp(struct sk_buff *skb,
4651 struct skb_shared_hwtstamps *hwtstamps)
4653 struct sock *sk = skb->sk;
4655 if (!skb_may_tx_timestamp(sk, false))
4658 /* Take a reference to prevent skb_orphan() from freeing the socket,
4659 * but only if the socket refcount is not zero.
4661 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4662 *skb_hwtstamps(skb) = *hwtstamps;
4663 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4671 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4673 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4674 struct skb_shared_hwtstamps *hwtstamps,
4675 struct sock *sk, int tstype)
4677 struct sk_buff *skb;
4678 bool tsonly, opt_stats = false;
4683 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4684 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4687 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4688 if (!skb_may_tx_timestamp(sk, tsonly))
4693 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4694 sk->sk_protocol == IPPROTO_TCP &&
4695 sk->sk_type == SOCK_STREAM) {
4696 skb = tcp_get_timestamping_opt_stats(sk, orig_skb);
4700 skb = alloc_skb(0, GFP_ATOMIC);
4702 skb = skb_clone(orig_skb, GFP_ATOMIC);
4708 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4710 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4714 *skb_hwtstamps(skb) = *hwtstamps;
4716 skb->tstamp = ktime_get_real();
4718 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4720 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4722 void skb_tstamp_tx(struct sk_buff *orig_skb,
4723 struct skb_shared_hwtstamps *hwtstamps)
4725 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4728 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4730 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4732 struct sock *sk = skb->sk;
4733 struct sock_exterr_skb *serr;
4736 skb->wifi_acked_valid = 1;
4737 skb->wifi_acked = acked;
4739 serr = SKB_EXT_ERR(skb);
4740 memset(serr, 0, sizeof(*serr));
4741 serr->ee.ee_errno = ENOMSG;
4742 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4744 /* Take a reference to prevent skb_orphan() from freeing the socket,
4745 * but only if the socket refcount is not zero.
4747 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4748 err = sock_queue_err_skb(sk, skb);
4754 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4757 * skb_partial_csum_set - set up and verify partial csum values for packet
4758 * @skb: the skb to set
4759 * @start: the number of bytes after skb->data to start checksumming.
4760 * @off: the offset from start to place the checksum.
4762 * For untrusted partially-checksummed packets, we need to make sure the values
4763 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4765 * This function checks and sets those values and skb->ip_summed: if this
4766 * returns false you should drop the packet.
4768 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4770 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4771 u32 csum_start = skb_headroom(skb) + (u32)start;
4773 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4774 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4775 start, off, skb_headroom(skb), skb_headlen(skb));
4778 skb->ip_summed = CHECKSUM_PARTIAL;
4779 skb->csum_start = csum_start;
4780 skb->csum_offset = off;
4781 skb_set_transport_header(skb, start);
4784 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4786 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4789 if (skb_headlen(skb) >= len)
4792 /* If we need to pullup then pullup to the max, so we
4793 * won't need to do it again.
4798 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4801 if (skb_headlen(skb) < len)
4807 #define MAX_TCP_HDR_LEN (15 * 4)
4809 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4810 typeof(IPPROTO_IP) proto,
4817 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4818 off + MAX_TCP_HDR_LEN);
4819 if (!err && !skb_partial_csum_set(skb, off,
4820 offsetof(struct tcphdr,
4823 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4826 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4827 off + sizeof(struct udphdr));
4828 if (!err && !skb_partial_csum_set(skb, off,
4829 offsetof(struct udphdr,
4832 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4835 return ERR_PTR(-EPROTO);
4838 /* This value should be large enough to cover a tagged ethernet header plus
4839 * maximally sized IP and TCP or UDP headers.
4841 #define MAX_IP_HDR_LEN 128
4843 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4852 err = skb_maybe_pull_tail(skb,
4853 sizeof(struct iphdr),
4858 if (ip_is_fragment(ip_hdr(skb)))
4861 off = ip_hdrlen(skb);
4868 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4870 return PTR_ERR(csum);
4873 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4876 ip_hdr(skb)->protocol, 0);
4883 /* This value should be large enough to cover a tagged ethernet header plus
4884 * an IPv6 header, all options, and a maximal TCP or UDP header.
4886 #define MAX_IPV6_HDR_LEN 256
4888 #define OPT_HDR(type, skb, off) \
4889 (type *)(skb_network_header(skb) + (off))
4891 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4904 off = sizeof(struct ipv6hdr);
4906 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4910 nexthdr = ipv6_hdr(skb)->nexthdr;
4912 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4913 while (off <= len && !done) {
4915 case IPPROTO_DSTOPTS:
4916 case IPPROTO_HOPOPTS:
4917 case IPPROTO_ROUTING: {
4918 struct ipv6_opt_hdr *hp;
4920 err = skb_maybe_pull_tail(skb,
4922 sizeof(struct ipv6_opt_hdr),
4927 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4928 nexthdr = hp->nexthdr;
4929 off += ipv6_optlen(hp);
4933 struct ip_auth_hdr *hp;
4935 err = skb_maybe_pull_tail(skb,
4937 sizeof(struct ip_auth_hdr),
4942 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4943 nexthdr = hp->nexthdr;
4944 off += ipv6_authlen(hp);
4947 case IPPROTO_FRAGMENT: {
4948 struct frag_hdr *hp;
4950 err = skb_maybe_pull_tail(skb,
4952 sizeof(struct frag_hdr),
4957 hp = OPT_HDR(struct frag_hdr, skb, off);
4959 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4962 nexthdr = hp->nexthdr;
4963 off += sizeof(struct frag_hdr);
4974 if (!done || fragment)
4977 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4979 return PTR_ERR(csum);
4982 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4983 &ipv6_hdr(skb)->daddr,
4984 skb->len - off, nexthdr, 0);
4992 * skb_checksum_setup - set up partial checksum offset
4993 * @skb: the skb to set up
4994 * @recalculate: if true the pseudo-header checksum will be recalculated
4996 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5000 switch (skb->protocol) {
5001 case htons(ETH_P_IP):
5002 err = skb_checksum_setup_ipv4(skb, recalculate);
5005 case htons(ETH_P_IPV6):
5006 err = skb_checksum_setup_ipv6(skb, recalculate);
5016 EXPORT_SYMBOL(skb_checksum_setup);
5019 * skb_checksum_maybe_trim - maybe trims the given skb
5020 * @skb: the skb to check
5021 * @transport_len: the data length beyond the network header
5023 * Checks whether the given skb has data beyond the given transport length.
5024 * If so, returns a cloned skb trimmed to this transport length.
5025 * Otherwise returns the provided skb. Returns NULL in error cases
5026 * (e.g. transport_len exceeds skb length or out-of-memory).
5028 * Caller needs to set the skb transport header and free any returned skb if it
5029 * differs from the provided skb.
5031 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5032 unsigned int transport_len)
5034 struct sk_buff *skb_chk;
5035 unsigned int len = skb_transport_offset(skb) + transport_len;
5040 else if (skb->len == len)
5043 skb_chk = skb_clone(skb, GFP_ATOMIC);
5047 ret = pskb_trim_rcsum(skb_chk, len);
5057 * skb_checksum_trimmed - validate checksum of an skb
5058 * @skb: the skb to check
5059 * @transport_len: the data length beyond the network header
5060 * @skb_chkf: checksum function to use
5062 * Applies the given checksum function skb_chkf to the provided skb.
5063 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5065 * If the skb has data beyond the given transport length, then a
5066 * trimmed & cloned skb is checked and returned.
5068 * Caller needs to set the skb transport header and free any returned skb if it
5069 * differs from the provided skb.
5071 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5072 unsigned int transport_len,
5073 __sum16(*skb_chkf)(struct sk_buff *skb))
5075 struct sk_buff *skb_chk;
5076 unsigned int offset = skb_transport_offset(skb);
5079 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5083 if (!pskb_may_pull(skb_chk, offset))
5086 skb_pull_rcsum(skb_chk, offset);
5087 ret = skb_chkf(skb_chk);
5088 skb_push_rcsum(skb_chk, offset);
5096 if (skb_chk && skb_chk != skb)
5102 EXPORT_SYMBOL(skb_checksum_trimmed);
5104 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5106 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5109 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5111 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5114 skb_release_head_state(skb);
5115 kmem_cache_free(skbuff_head_cache, skb);
5120 EXPORT_SYMBOL(kfree_skb_partial);
5123 * skb_try_coalesce - try to merge skb to prior one
5125 * @from: buffer to add
5126 * @fragstolen: pointer to boolean
5127 * @delta_truesize: how much more was allocated than was requested
5129 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5130 bool *fragstolen, int *delta_truesize)
5132 struct skb_shared_info *to_shinfo, *from_shinfo;
5133 int i, delta, len = from->len;
5135 *fragstolen = false;
5140 if (len <= skb_tailroom(to)) {
5142 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5143 *delta_truesize = 0;
5147 to_shinfo = skb_shinfo(to);
5148 from_shinfo = skb_shinfo(from);
5149 if (to_shinfo->frag_list || from_shinfo->frag_list)
5151 if (skb_zcopy(to) || skb_zcopy(from))
5154 if (skb_headlen(from) != 0) {
5156 unsigned int offset;
5158 if (to_shinfo->nr_frags +
5159 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5162 if (skb_head_is_locked(from))
5165 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5167 page = virt_to_head_page(from->head);
5168 offset = from->data - (unsigned char *)page_address(page);
5170 skb_fill_page_desc(to, to_shinfo->nr_frags,
5171 page, offset, skb_headlen(from));
5174 if (to_shinfo->nr_frags +
5175 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5178 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5181 WARN_ON_ONCE(delta < len);
5183 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5185 from_shinfo->nr_frags * sizeof(skb_frag_t));
5186 to_shinfo->nr_frags += from_shinfo->nr_frags;
5188 if (!skb_cloned(from))
5189 from_shinfo->nr_frags = 0;
5191 /* if the skb is not cloned this does nothing
5192 * since we set nr_frags to 0.
5194 for (i = 0; i < from_shinfo->nr_frags; i++)
5195 __skb_frag_ref(&from_shinfo->frags[i]);
5197 to->truesize += delta;
5199 to->data_len += len;
5201 *delta_truesize = delta;
5204 EXPORT_SYMBOL(skb_try_coalesce);
5207 * skb_scrub_packet - scrub an skb
5209 * @skb: buffer to clean
5210 * @xnet: packet is crossing netns
5212 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5213 * into/from a tunnel. Some information have to be cleared during these
5215 * skb_scrub_packet can also be used to clean a skb before injecting it in
5216 * another namespace (@xnet == true). We have to clear all information in the
5217 * skb that could impact namespace isolation.
5219 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5221 skb->pkt_type = PACKET_HOST;
5227 nf_reset_trace(skb);
5229 #ifdef CONFIG_NET_SWITCHDEV
5230 skb->offload_fwd_mark = 0;
5231 skb->offload_l3_fwd_mark = 0;
5241 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5244 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5248 * skb_gso_transport_seglen is used to determine the real size of the
5249 * individual segments, including Layer4 headers (TCP/UDP).
5251 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5253 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5255 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5256 unsigned int thlen = 0;
5258 if (skb->encapsulation) {
5259 thlen = skb_inner_transport_header(skb) -
5260 skb_transport_header(skb);
5262 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5263 thlen += inner_tcp_hdrlen(skb);
5264 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5265 thlen = tcp_hdrlen(skb);
5266 } else if (unlikely(skb_is_gso_sctp(skb))) {
5267 thlen = sizeof(struct sctphdr);
5268 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5269 thlen = sizeof(struct udphdr);
5271 /* UFO sets gso_size to the size of the fragmentation
5272 * payload, i.e. the size of the L4 (UDP) header is already
5275 return thlen + shinfo->gso_size;
5279 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5283 * skb_gso_network_seglen is used to determine the real size of the
5284 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5286 * The MAC/L2 header is not accounted for.
5288 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5290 unsigned int hdr_len = skb_transport_header(skb) -
5291 skb_network_header(skb);
5293 return hdr_len + skb_gso_transport_seglen(skb);
5297 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5301 * skb_gso_mac_seglen is used to determine the real size of the
5302 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5303 * headers (TCP/UDP).
5305 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5307 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5309 return hdr_len + skb_gso_transport_seglen(skb);
5313 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5315 * There are a couple of instances where we have a GSO skb, and we
5316 * want to determine what size it would be after it is segmented.
5318 * We might want to check:
5319 * - L3+L4+payload size (e.g. IP forwarding)
5320 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5322 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5326 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5327 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5329 * @max_len: The maximum permissible length.
5331 * Returns true if the segmented length <= max length.
5333 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5334 unsigned int seg_len,
5335 unsigned int max_len) {
5336 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5337 const struct sk_buff *iter;
5339 if (shinfo->gso_size != GSO_BY_FRAGS)
5340 return seg_len <= max_len;
5342 /* Undo this so we can re-use header sizes */
5343 seg_len -= GSO_BY_FRAGS;
5345 skb_walk_frags(skb, iter) {
5346 if (seg_len + skb_headlen(iter) > max_len)
5354 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5357 * @mtu: MTU to validate against
5359 * skb_gso_validate_network_len validates if a given skb will fit a
5360 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5363 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5365 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5367 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5370 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5373 * @len: length to validate against
5375 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5376 * length once split, including L2, L3 and L4 headers and the payload.
5378 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5380 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5382 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5384 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5386 int mac_len, meta_len;
5389 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5394 mac_len = skb->data - skb_mac_header(skb);
5395 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5396 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5397 mac_len - VLAN_HLEN - ETH_TLEN);
5400 meta_len = skb_metadata_len(skb);
5402 meta = skb_metadata_end(skb) - meta_len;
5403 memmove(meta + VLAN_HLEN, meta, meta_len);
5406 skb->mac_header += VLAN_HLEN;
5410 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5412 struct vlan_hdr *vhdr;
5415 if (unlikely(skb_vlan_tag_present(skb))) {
5416 /* vlan_tci is already set-up so leave this for another time */
5420 skb = skb_share_check(skb, GFP_ATOMIC);
5423 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5424 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5427 vhdr = (struct vlan_hdr *)skb->data;
5428 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5429 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5431 skb_pull_rcsum(skb, VLAN_HLEN);
5432 vlan_set_encap_proto(skb, vhdr);
5434 skb = skb_reorder_vlan_header(skb);
5438 skb_reset_network_header(skb);
5439 skb_reset_transport_header(skb);
5440 skb_reset_mac_len(skb);
5448 EXPORT_SYMBOL(skb_vlan_untag);
5450 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5452 if (!pskb_may_pull(skb, write_len))
5455 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5458 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5460 EXPORT_SYMBOL(skb_ensure_writable);
5462 /* remove VLAN header from packet and update csum accordingly.
5463 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5465 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5467 struct vlan_hdr *vhdr;
5468 int offset = skb->data - skb_mac_header(skb);
5471 if (WARN_ONCE(offset,
5472 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5477 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5481 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5483 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5484 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5486 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5487 __skb_pull(skb, VLAN_HLEN);
5489 vlan_set_encap_proto(skb, vhdr);
5490 skb->mac_header += VLAN_HLEN;
5492 if (skb_network_offset(skb) < ETH_HLEN)
5493 skb_set_network_header(skb, ETH_HLEN);
5495 skb_reset_mac_len(skb);
5499 EXPORT_SYMBOL(__skb_vlan_pop);
5501 /* Pop a vlan tag either from hwaccel or from payload.
5502 * Expects skb->data at mac header.
5504 int skb_vlan_pop(struct sk_buff *skb)
5510 if (likely(skb_vlan_tag_present(skb))) {
5511 __vlan_hwaccel_clear_tag(skb);
5513 if (unlikely(!eth_type_vlan(skb->protocol)))
5516 err = __skb_vlan_pop(skb, &vlan_tci);
5520 /* move next vlan tag to hw accel tag */
5521 if (likely(!eth_type_vlan(skb->protocol)))
5524 vlan_proto = skb->protocol;
5525 err = __skb_vlan_pop(skb, &vlan_tci);
5529 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5532 EXPORT_SYMBOL(skb_vlan_pop);
5534 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5535 * Expects skb->data at mac header.
5537 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5539 if (skb_vlan_tag_present(skb)) {
5540 int offset = skb->data - skb_mac_header(skb);
5543 if (WARN_ONCE(offset,
5544 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5549 err = __vlan_insert_tag(skb, skb->vlan_proto,
5550 skb_vlan_tag_get(skb));
5554 skb->protocol = skb->vlan_proto;
5555 skb->mac_len += VLAN_HLEN;
5557 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5559 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5562 EXPORT_SYMBOL(skb_vlan_push);
5564 /* Update the ethertype of hdr and the skb csum value if required. */
5565 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5568 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5569 __be16 diff[] = { ~hdr->h_proto, ethertype };
5571 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5574 hdr->h_proto = ethertype;
5578 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
5582 * @mpls_lse: MPLS label stack entry to push
5583 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5584 * @mac_len: length of the MAC header
5585 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
5588 * Expects skb->data at mac header.
5590 * Returns 0 on success, -errno otherwise.
5592 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5593 int mac_len, bool ethernet)
5595 struct mpls_shim_hdr *lse;
5598 if (unlikely(!eth_p_mpls(mpls_proto)))
5601 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5602 if (skb->encapsulation)
5605 err = skb_cow_head(skb, MPLS_HLEN);
5609 if (!skb->inner_protocol) {
5610 skb_set_inner_network_header(skb, skb_network_offset(skb));
5611 skb_set_inner_protocol(skb, skb->protocol);
5614 skb_push(skb, MPLS_HLEN);
5615 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5617 skb_reset_mac_header(skb);
5618 skb_set_network_header(skb, mac_len);
5619 skb_reset_mac_len(skb);
5621 lse = mpls_hdr(skb);
5622 lse->label_stack_entry = mpls_lse;
5623 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5626 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5627 skb->protocol = mpls_proto;
5631 EXPORT_SYMBOL_GPL(skb_mpls_push);
5634 * skb_mpls_pop() - pop the outermost MPLS header
5637 * @next_proto: ethertype of header after popped MPLS header
5638 * @mac_len: length of the MAC header
5639 * @ethernet: flag to indicate if the packet is ethernet
5641 * Expects skb->data at mac header.
5643 * Returns 0 on success, -errno otherwise.
5645 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5650 if (unlikely(!eth_p_mpls(skb->protocol)))
5653 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5657 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5658 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5661 __skb_pull(skb, MPLS_HLEN);
5662 skb_reset_mac_header(skb);
5663 skb_set_network_header(skb, mac_len);
5668 /* use mpls_hdr() to get ethertype to account for VLANs. */
5669 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5670 skb_mod_eth_type(skb, hdr, next_proto);
5672 skb->protocol = next_proto;
5676 EXPORT_SYMBOL_GPL(skb_mpls_pop);
5679 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5682 * @mpls_lse: new MPLS label stack entry to update to
5684 * Expects skb->data at mac header.
5686 * Returns 0 on success, -errno otherwise.
5688 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5692 if (unlikely(!eth_p_mpls(skb->protocol)))
5695 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5699 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5700 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5702 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5705 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5709 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5712 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5716 * Expects skb->data at mac header.
5718 * Returns 0 on success, -errno otherwise.
5720 int skb_mpls_dec_ttl(struct sk_buff *skb)
5725 if (unlikely(!eth_p_mpls(skb->protocol)))
5728 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5729 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5733 lse &= ~MPLS_LS_TTL_MASK;
5734 lse |= ttl << MPLS_LS_TTL_SHIFT;
5736 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5738 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5741 * alloc_skb_with_frags - allocate skb with page frags
5743 * @header_len: size of linear part
5744 * @data_len: needed length in frags
5745 * @max_page_order: max page order desired.
5746 * @errcode: pointer to error code if any
5747 * @gfp_mask: allocation mask
5749 * This can be used to allocate a paged skb, given a maximal order for frags.
5751 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5752 unsigned long data_len,
5757 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5758 unsigned long chunk;
5759 struct sk_buff *skb;
5763 *errcode = -EMSGSIZE;
5764 /* Note this test could be relaxed, if we succeed to allocate
5765 * high order pages...
5767 if (npages > MAX_SKB_FRAGS)
5770 *errcode = -ENOBUFS;
5771 skb = alloc_skb(header_len, gfp_mask);
5775 skb->truesize += npages << PAGE_SHIFT;
5777 for (i = 0; npages > 0; i++) {
5778 int order = max_page_order;
5781 if (npages >= 1 << order) {
5782 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5788 /* Do not retry other high order allocations */
5794 page = alloc_page(gfp_mask);
5798 chunk = min_t(unsigned long, data_len,
5799 PAGE_SIZE << order);
5800 skb_fill_page_desc(skb, i, page, 0, chunk);
5802 npages -= 1 << order;
5810 EXPORT_SYMBOL(alloc_skb_with_frags);
5812 /* carve out the first off bytes from skb when off < headlen */
5813 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5814 const int headlen, gfp_t gfp_mask)
5817 int size = skb_end_offset(skb);
5818 int new_hlen = headlen - off;
5821 size = SKB_DATA_ALIGN(size);
5823 if (skb_pfmemalloc(skb))
5824 gfp_mask |= __GFP_MEMALLOC;
5825 data = kmalloc_reserve(size +
5826 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5827 gfp_mask, NUMA_NO_NODE, NULL);
5831 size = SKB_WITH_OVERHEAD(ksize(data));
5833 /* Copy real data, and all frags */
5834 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5837 memcpy((struct skb_shared_info *)(data + size),
5839 offsetof(struct skb_shared_info,
5840 frags[skb_shinfo(skb)->nr_frags]));
5841 if (skb_cloned(skb)) {
5842 /* drop the old head gracefully */
5843 if (skb_orphan_frags(skb, gfp_mask)) {
5847 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5848 skb_frag_ref(skb, i);
5849 if (skb_has_frag_list(skb))
5850 skb_clone_fraglist(skb);
5851 skb_release_data(skb);
5853 /* we can reuse existing recount- all we did was
5862 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5865 skb->end = skb->head + size;
5867 skb_set_tail_pointer(skb, skb_headlen(skb));
5868 skb_headers_offset_update(skb, 0);
5872 atomic_set(&skb_shinfo(skb)->dataref, 1);
5877 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
5879 /* carve out the first eat bytes from skb's frag_list. May recurse into
5882 static int pskb_carve_frag_list(struct sk_buff *skb,
5883 struct skb_shared_info *shinfo, int eat,
5886 struct sk_buff *list = shinfo->frag_list;
5887 struct sk_buff *clone = NULL;
5888 struct sk_buff *insp = NULL;
5892 pr_err("Not enough bytes to eat. Want %d\n", eat);
5895 if (list->len <= eat) {
5896 /* Eaten as whole. */
5901 /* Eaten partially. */
5902 if (skb_shared(list)) {
5903 clone = skb_clone(list, gfp_mask);
5909 /* This may be pulled without problems. */
5912 if (pskb_carve(list, eat, gfp_mask) < 0) {
5920 /* Free pulled out fragments. */
5921 while ((list = shinfo->frag_list) != insp) {
5922 shinfo->frag_list = list->next;
5925 /* And insert new clone at head. */
5928 shinfo->frag_list = clone;
5933 /* carve off first len bytes from skb. Split line (off) is in the
5934 * non-linear part of skb
5936 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
5937 int pos, gfp_t gfp_mask)
5940 int size = skb_end_offset(skb);
5942 const int nfrags = skb_shinfo(skb)->nr_frags;
5943 struct skb_shared_info *shinfo;
5945 size = SKB_DATA_ALIGN(size);
5947 if (skb_pfmemalloc(skb))
5948 gfp_mask |= __GFP_MEMALLOC;
5949 data = kmalloc_reserve(size +
5950 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5951 gfp_mask, NUMA_NO_NODE, NULL);
5955 size = SKB_WITH_OVERHEAD(ksize(data));
5957 memcpy((struct skb_shared_info *)(data + size),
5958 skb_shinfo(skb), offsetof(struct skb_shared_info,
5959 frags[skb_shinfo(skb)->nr_frags]));
5960 if (skb_orphan_frags(skb, gfp_mask)) {
5964 shinfo = (struct skb_shared_info *)(data + size);
5965 for (i = 0; i < nfrags; i++) {
5966 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
5968 if (pos + fsize > off) {
5969 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
5973 * We have two variants in this case:
5974 * 1. Move all the frag to the second
5975 * part, if it is possible. F.e.
5976 * this approach is mandatory for TUX,
5977 * where splitting is expensive.
5978 * 2. Split is accurately. We make this.
5980 skb_frag_off_add(&shinfo->frags[0], off - pos);
5981 skb_frag_size_sub(&shinfo->frags[0], off - pos);
5983 skb_frag_ref(skb, i);
5988 shinfo->nr_frags = k;
5989 if (skb_has_frag_list(skb))
5990 skb_clone_fraglist(skb);
5992 /* split line is in frag list */
5993 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
5994 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
5995 if (skb_has_frag_list(skb))
5996 kfree_skb_list(skb_shinfo(skb)->frag_list);
6000 skb_release_data(skb);
6005 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6008 skb->end = skb->head + size;
6010 skb_reset_tail_pointer(skb);
6011 skb_headers_offset_update(skb, 0);
6016 skb->data_len = skb->len;
6017 atomic_set(&skb_shinfo(skb)->dataref, 1);
6021 /* remove len bytes from the beginning of the skb */
6022 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6024 int headlen = skb_headlen(skb);
6027 return pskb_carve_inside_header(skb, len, headlen, gfp);
6029 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6032 /* Extract to_copy bytes starting at off from skb, and return this in
6035 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6036 int to_copy, gfp_t gfp)
6038 struct sk_buff *clone = skb_clone(skb, gfp);
6043 if (pskb_carve(clone, off, gfp) < 0 ||
6044 pskb_trim(clone, to_copy)) {
6050 EXPORT_SYMBOL(pskb_extract);
6053 * skb_condense - try to get rid of fragments/frag_list if possible
6056 * Can be used to save memory before skb is added to a busy queue.
6057 * If packet has bytes in frags and enough tail room in skb->head,
6058 * pull all of them, so that we can free the frags right now and adjust
6061 * We do not reallocate skb->head thus can not fail.
6062 * Caller must re-evaluate skb->truesize if needed.
6064 void skb_condense(struct sk_buff *skb)
6066 if (skb->data_len) {
6067 if (skb->data_len > skb->end - skb->tail ||
6071 /* Nice, we can free page frag(s) right now */
6072 __pskb_pull_tail(skb, skb->data_len);
6074 /* At this point, skb->truesize might be over estimated,
6075 * because skb had a fragment, and fragments do not tell
6077 * When we pulled its content into skb->head, fragment
6078 * was freed, but __pskb_pull_tail() could not possibly
6079 * adjust skb->truesize, not knowing the frag truesize.
6081 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6084 #ifdef CONFIG_SKB_EXTENSIONS
6085 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6087 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6091 * __skb_ext_alloc - allocate a new skb extensions storage
6093 * @flags: See kmalloc().
6095 * Returns the newly allocated pointer. The pointer can later attached to a
6096 * skb via __skb_ext_set().
6097 * Note: caller must handle the skb_ext as an opaque data.
6099 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6101 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6104 memset(new->offset, 0, sizeof(new->offset));
6105 refcount_set(&new->refcnt, 1);
6111 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6112 unsigned int old_active)
6114 struct skb_ext *new;
6116 if (refcount_read(&old->refcnt) == 1)
6119 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6123 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6124 refcount_set(&new->refcnt, 1);
6127 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6128 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6131 for (i = 0; i < sp->len; i++)
6132 xfrm_state_hold(sp->xvec[i]);
6140 * __skb_ext_set - attach the specified extension storage to this skb
6143 * @ext: extension storage previously allocated via __skb_ext_alloc()
6145 * Existing extensions, if any, are cleared.
6147 * Returns the pointer to the extension.
6149 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6150 struct skb_ext *ext)
6152 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6155 newlen = newoff + skb_ext_type_len[id];
6156 ext->chunks = newlen;
6157 ext->offset[id] = newoff;
6158 skb->extensions = ext;
6159 skb->active_extensions = 1 << id;
6160 return skb_ext_get_ptr(ext, id);
6164 * skb_ext_add - allocate space for given extension, COW if needed
6166 * @id: extension to allocate space for
6168 * Allocates enough space for the given extension.
6169 * If the extension is already present, a pointer to that extension
6172 * If the skb was cloned, COW applies and the returned memory can be
6173 * modified without changing the extension space of clones buffers.
6175 * Returns pointer to the extension or NULL on allocation failure.
6177 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6179 struct skb_ext *new, *old = NULL;
6180 unsigned int newlen, newoff;
6182 if (skb->active_extensions) {
6183 old = skb->extensions;
6185 new = skb_ext_maybe_cow(old, skb->active_extensions);
6189 if (__skb_ext_exist(new, id))
6192 newoff = new->chunks;
6194 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6196 new = __skb_ext_alloc(GFP_ATOMIC);
6201 newlen = newoff + skb_ext_type_len[id];
6202 new->chunks = newlen;
6203 new->offset[id] = newoff;
6205 skb->extensions = new;
6206 skb->active_extensions |= 1 << id;
6207 return skb_ext_get_ptr(new, id);
6209 EXPORT_SYMBOL(skb_ext_add);
6212 static void skb_ext_put_sp(struct sec_path *sp)
6216 for (i = 0; i < sp->len; i++)
6217 xfrm_state_put(sp->xvec[i]);
6221 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6223 struct skb_ext *ext = skb->extensions;
6225 skb->active_extensions &= ~(1 << id);
6226 if (skb->active_extensions == 0) {
6227 skb->extensions = NULL;
6230 } else if (id == SKB_EXT_SEC_PATH &&
6231 refcount_read(&ext->refcnt) == 1) {
6232 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6239 EXPORT_SYMBOL(__skb_ext_del);
6241 void __skb_ext_put(struct skb_ext *ext)
6243 /* If this is last clone, nothing can increment
6244 * it after check passes. Avoids one atomic op.
6246 if (refcount_read(&ext->refcnt) == 1)
6249 if (!refcount_dec_and_test(&ext->refcnt))
6253 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6254 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6257 kmem_cache_free(skbuff_ext_cache, ext);
6259 EXPORT_SYMBOL(__skb_ext_put);
6260 #endif /* CONFIG_SKB_EXTENSIONS */