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>
63 #include <linux/kcov.h>
65 #include <net/protocol.h>
68 #include <net/checksum.h>
69 #include <net/ip6_checksum.h>
72 #include <net/mptcp.h>
74 #include <linux/uaccess.h>
75 #include <trace/events/skb.h>
76 #include <linux/highmem.h>
77 #include <linux/capability.h>
78 #include <linux/user_namespace.h>
79 #include <linux/indirect_call_wrapper.h>
83 struct kmem_cache *skbuff_head_cache __ro_after_init;
84 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
85 #ifdef CONFIG_SKB_EXTENSIONS
86 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
88 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
89 EXPORT_SYMBOL(sysctl_max_skb_frags);
92 * skb_panic - private function for out-of-line support
96 * @msg: skb_over_panic or skb_under_panic
98 * Out-of-line support for skb_put() and skb_push().
99 * Called via the wrapper skb_over_panic() or skb_under_panic().
100 * Keep out of line to prevent kernel bloat.
101 * __builtin_return_address is not used because it is not always reliable.
103 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
106 pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
107 msg, addr, skb->len, sz, skb->head, skb->data,
108 (unsigned long)skb->tail, (unsigned long)skb->end,
109 skb->dev ? skb->dev->name : "<NULL>");
113 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
115 skb_panic(skb, sz, addr, __func__);
118 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
120 skb_panic(skb, sz, addr, __func__);
123 #define NAPI_SKB_CACHE_SIZE 64
124 #define NAPI_SKB_CACHE_BULK 16
125 #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2)
127 struct napi_alloc_cache {
128 struct page_frag_cache page;
129 unsigned int skb_count;
130 void *skb_cache[NAPI_SKB_CACHE_SIZE];
133 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
134 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
136 static void *__alloc_frag_align(unsigned int fragsz, gfp_t gfp_mask,
137 unsigned int align_mask)
139 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
141 return page_frag_alloc_align(&nc->page, fragsz, gfp_mask, align_mask);
144 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
146 fragsz = SKB_DATA_ALIGN(fragsz);
148 return __alloc_frag_align(fragsz, GFP_ATOMIC, align_mask);
150 EXPORT_SYMBOL(__napi_alloc_frag_align);
152 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
154 struct page_frag_cache *nc;
157 fragsz = SKB_DATA_ALIGN(fragsz);
158 if (in_irq() || irqs_disabled()) {
159 nc = this_cpu_ptr(&netdev_alloc_cache);
160 data = page_frag_alloc_align(nc, fragsz, GFP_ATOMIC, align_mask);
163 data = __alloc_frag_align(fragsz, GFP_ATOMIC, align_mask);
168 EXPORT_SYMBOL(__netdev_alloc_frag_align);
170 static struct sk_buff *napi_skb_cache_get(void)
172 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
175 if (unlikely(!nc->skb_count))
176 nc->skb_count = kmem_cache_alloc_bulk(skbuff_head_cache,
180 if (unlikely(!nc->skb_count))
183 skb = nc->skb_cache[--nc->skb_count];
184 kasan_unpoison_object_data(skbuff_head_cache, skb);
189 /* Caller must provide SKB that is memset cleared */
190 static void __build_skb_around(struct sk_buff *skb, void *data,
191 unsigned int frag_size)
193 struct skb_shared_info *shinfo;
194 unsigned int size = frag_size ? : ksize(data);
196 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
198 /* Assumes caller memset cleared SKB */
199 skb->truesize = SKB_TRUESIZE(size);
200 refcount_set(&skb->users, 1);
203 skb_reset_tail_pointer(skb);
204 skb->end = skb->tail + size;
205 skb->mac_header = (typeof(skb->mac_header))~0U;
206 skb->transport_header = (typeof(skb->transport_header))~0U;
208 /* make sure we initialize shinfo sequentially */
209 shinfo = skb_shinfo(skb);
210 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
211 atomic_set(&shinfo->dataref, 1);
213 skb_set_kcov_handle(skb, kcov_common_handle());
217 * __build_skb - build a network buffer
218 * @data: data buffer provided by caller
219 * @frag_size: size of data, or 0 if head was kmalloced
221 * Allocate a new &sk_buff. Caller provides space holding head and
222 * skb_shared_info. @data must have been allocated by kmalloc() only if
223 * @frag_size is 0, otherwise data should come from the page allocator
225 * The return is the new skb buffer.
226 * On a failure the return is %NULL, and @data is not freed.
228 * Before IO, driver allocates only data buffer where NIC put incoming frame
229 * Driver should add room at head (NET_SKB_PAD) and
230 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
231 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
232 * before giving packet to stack.
233 * RX rings only contains data buffers, not full skbs.
235 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
239 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
243 memset(skb, 0, offsetof(struct sk_buff, tail));
244 __build_skb_around(skb, data, frag_size);
249 /* build_skb() is wrapper over __build_skb(), that specifically
250 * takes care of skb->head and skb->pfmemalloc
251 * This means that if @frag_size is not zero, then @data must be backed
252 * by a page fragment, not kmalloc() or vmalloc()
254 struct sk_buff *build_skb(void *data, unsigned int frag_size)
256 struct sk_buff *skb = __build_skb(data, frag_size);
258 if (skb && frag_size) {
260 if (page_is_pfmemalloc(virt_to_head_page(data)))
265 EXPORT_SYMBOL(build_skb);
268 * build_skb_around - build a network buffer around provided skb
269 * @skb: sk_buff provide by caller, must be memset cleared
270 * @data: data buffer provided by caller
271 * @frag_size: size of data, or 0 if head was kmalloced
273 struct sk_buff *build_skb_around(struct sk_buff *skb,
274 void *data, unsigned int frag_size)
279 __build_skb_around(skb, data, frag_size);
283 if (page_is_pfmemalloc(virt_to_head_page(data)))
288 EXPORT_SYMBOL(build_skb_around);
291 * __napi_build_skb - build a network buffer
292 * @data: data buffer provided by caller
293 * @frag_size: size of data, or 0 if head was kmalloced
295 * Version of __build_skb() that uses NAPI percpu caches to obtain
296 * skbuff_head instead of inplace allocation.
298 * Returns a new &sk_buff on success, %NULL on allocation failure.
300 static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size)
304 skb = napi_skb_cache_get();
308 memset(skb, 0, offsetof(struct sk_buff, tail));
309 __build_skb_around(skb, data, frag_size);
315 * napi_build_skb - build a network buffer
316 * @data: data buffer provided by caller
317 * @frag_size: size of data, or 0 if head was kmalloced
319 * Version of __napi_build_skb() that takes care of skb->head_frag
320 * and skb->pfmemalloc when the data is a page or page fragment.
322 * Returns a new &sk_buff on success, %NULL on allocation failure.
324 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size)
326 struct sk_buff *skb = __napi_build_skb(data, frag_size);
328 if (likely(skb) && frag_size) {
330 skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
335 EXPORT_SYMBOL(napi_build_skb);
338 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
339 * the caller if emergency pfmemalloc reserves are being used. If it is and
340 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
341 * may be used. Otherwise, the packet data may be discarded until enough
344 static void *kmalloc_reserve(size_t size, gfp_t flags, int node,
348 bool ret_pfmemalloc = false;
351 * Try a regular allocation, when that fails and we're not entitled
352 * to the reserves, fail.
354 obj = kmalloc_node_track_caller(size,
355 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
357 if (obj || !(gfp_pfmemalloc_allowed(flags)))
360 /* Try again but now we are using pfmemalloc reserves */
361 ret_pfmemalloc = true;
362 obj = kmalloc_node_track_caller(size, flags, node);
366 *pfmemalloc = ret_pfmemalloc;
371 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
372 * 'private' fields and also do memory statistics to find all the
378 * __alloc_skb - allocate a network buffer
379 * @size: size to allocate
380 * @gfp_mask: allocation mask
381 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
382 * instead of head cache and allocate a cloned (child) skb.
383 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
384 * allocations in case the data is required for writeback
385 * @node: numa node to allocate memory on
387 * Allocate a new &sk_buff. The returned buffer has no headroom and a
388 * tail room of at least size bytes. The object has a reference count
389 * of one. The return is the buffer. On a failure the return is %NULL.
391 * Buffers may only be allocated from interrupts using a @gfp_mask of
394 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
397 struct kmem_cache *cache;
402 cache = (flags & SKB_ALLOC_FCLONE)
403 ? skbuff_fclone_cache : skbuff_head_cache;
405 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
406 gfp_mask |= __GFP_MEMALLOC;
409 if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI &&
410 likely(node == NUMA_NO_NODE || node == numa_mem_id()))
411 skb = napi_skb_cache_get();
413 skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node);
418 /* We do our best to align skb_shared_info on a separate cache
419 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
420 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
421 * Both skb->head and skb_shared_info are cache line aligned.
423 size = SKB_DATA_ALIGN(size);
424 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
425 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
428 /* kmalloc(size) might give us more room than requested.
429 * Put skb_shared_info exactly at the end of allocated zone,
430 * to allow max possible filling before reallocation.
432 size = SKB_WITH_OVERHEAD(ksize(data));
433 prefetchw(data + size);
436 * Only clear those fields we need to clear, not those that we will
437 * actually initialise below. Hence, don't put any more fields after
438 * the tail pointer in struct sk_buff!
440 memset(skb, 0, offsetof(struct sk_buff, tail));
441 __build_skb_around(skb, data, 0);
442 skb->pfmemalloc = pfmemalloc;
444 if (flags & SKB_ALLOC_FCLONE) {
445 struct sk_buff_fclones *fclones;
447 fclones = container_of(skb, struct sk_buff_fclones, skb1);
449 skb->fclone = SKB_FCLONE_ORIG;
450 refcount_set(&fclones->fclone_ref, 1);
452 fclones->skb2.fclone = SKB_FCLONE_CLONE;
458 kmem_cache_free(cache, skb);
461 EXPORT_SYMBOL(__alloc_skb);
464 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
465 * @dev: network device to receive on
466 * @len: length to allocate
467 * @gfp_mask: get_free_pages mask, passed to alloc_skb
469 * Allocate a new &sk_buff and assign it a usage count of one. The
470 * buffer has NET_SKB_PAD headroom built in. Users should allocate
471 * the headroom they think they need without accounting for the
472 * built in space. The built in space is used for optimisations.
474 * %NULL is returned if there is no free memory.
476 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
479 struct page_frag_cache *nc;
486 /* If requested length is either too small or too big,
487 * we use kmalloc() for skb->head allocation.
489 if (len <= SKB_WITH_OVERHEAD(1024) ||
490 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
491 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
492 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
498 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
499 len = SKB_DATA_ALIGN(len);
501 if (sk_memalloc_socks())
502 gfp_mask |= __GFP_MEMALLOC;
504 if (in_irq() || irqs_disabled()) {
505 nc = this_cpu_ptr(&netdev_alloc_cache);
506 data = page_frag_alloc(nc, len, gfp_mask);
507 pfmemalloc = nc->pfmemalloc;
510 nc = this_cpu_ptr(&napi_alloc_cache.page);
511 data = page_frag_alloc(nc, len, gfp_mask);
512 pfmemalloc = nc->pfmemalloc;
519 skb = __build_skb(data, len);
520 if (unlikely(!skb)) {
530 skb_reserve(skb, NET_SKB_PAD);
536 EXPORT_SYMBOL(__netdev_alloc_skb);
539 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
540 * @napi: napi instance this buffer was allocated for
541 * @len: length to allocate
542 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
544 * Allocate a new sk_buff for use in NAPI receive. This buffer will
545 * attempt to allocate the head from a special reserved region used
546 * only for NAPI Rx allocation. By doing this we can save several
547 * CPU cycles by avoiding having to disable and re-enable IRQs.
549 * %NULL is returned if there is no free memory.
551 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
554 struct napi_alloc_cache *nc;
558 len += NET_SKB_PAD + NET_IP_ALIGN;
560 /* If requested length is either too small or too big,
561 * we use kmalloc() for skb->head allocation.
563 if (len <= SKB_WITH_OVERHEAD(1024) ||
564 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
565 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
566 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI,
573 nc = this_cpu_ptr(&napi_alloc_cache);
574 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
575 len = SKB_DATA_ALIGN(len);
577 if (sk_memalloc_socks())
578 gfp_mask |= __GFP_MEMALLOC;
580 data = page_frag_alloc(&nc->page, len, gfp_mask);
584 skb = __napi_build_skb(data, len);
585 if (unlikely(!skb)) {
590 if (nc->page.pfmemalloc)
595 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
596 skb->dev = napi->dev;
601 EXPORT_SYMBOL(__napi_alloc_skb);
603 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
604 int size, unsigned int truesize)
606 skb_fill_page_desc(skb, i, page, off, size);
608 skb->data_len += size;
609 skb->truesize += truesize;
611 EXPORT_SYMBOL(skb_add_rx_frag);
613 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
614 unsigned int truesize)
616 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
618 skb_frag_size_add(frag, size);
620 skb->data_len += size;
621 skb->truesize += truesize;
623 EXPORT_SYMBOL(skb_coalesce_rx_frag);
625 static void skb_drop_list(struct sk_buff **listp)
627 kfree_skb_list(*listp);
631 static inline void skb_drop_fraglist(struct sk_buff *skb)
633 skb_drop_list(&skb_shinfo(skb)->frag_list);
636 static void skb_clone_fraglist(struct sk_buff *skb)
638 struct sk_buff *list;
640 skb_walk_frags(skb, list)
644 static void skb_free_head(struct sk_buff *skb)
646 unsigned char *head = skb->head;
654 static void skb_release_data(struct sk_buff *skb)
656 struct skb_shared_info *shinfo = skb_shinfo(skb);
660 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
664 skb_zcopy_clear(skb, true);
666 for (i = 0; i < shinfo->nr_frags; i++)
667 __skb_frag_unref(&shinfo->frags[i]);
669 if (shinfo->frag_list)
670 kfree_skb_list(shinfo->frag_list);
676 * Free an skbuff by memory without cleaning the state.
678 static void kfree_skbmem(struct sk_buff *skb)
680 struct sk_buff_fclones *fclones;
682 switch (skb->fclone) {
683 case SKB_FCLONE_UNAVAILABLE:
684 kmem_cache_free(skbuff_head_cache, skb);
687 case SKB_FCLONE_ORIG:
688 fclones = container_of(skb, struct sk_buff_fclones, skb1);
690 /* We usually free the clone (TX completion) before original skb
691 * This test would have no chance to be true for the clone,
692 * while here, branch prediction will be good.
694 if (refcount_read(&fclones->fclone_ref) == 1)
698 default: /* SKB_FCLONE_CLONE */
699 fclones = container_of(skb, struct sk_buff_fclones, skb2);
702 if (!refcount_dec_and_test(&fclones->fclone_ref))
705 kmem_cache_free(skbuff_fclone_cache, fclones);
708 void skb_release_head_state(struct sk_buff *skb)
711 if (skb->destructor) {
713 skb->destructor(skb);
715 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
716 nf_conntrack_put(skb_nfct(skb));
721 /* Free everything but the sk_buff shell. */
722 static void skb_release_all(struct sk_buff *skb)
724 skb_release_head_state(skb);
725 if (likely(skb->head))
726 skb_release_data(skb);
730 * __kfree_skb - private function
733 * Free an sk_buff. Release anything attached to the buffer.
734 * Clean the state. This is an internal helper function. Users should
735 * always call kfree_skb
738 void __kfree_skb(struct sk_buff *skb)
740 skb_release_all(skb);
743 EXPORT_SYMBOL(__kfree_skb);
746 * kfree_skb - free an sk_buff
747 * @skb: buffer to free
749 * Drop a reference to the buffer and free it if the usage count has
752 void kfree_skb(struct sk_buff *skb)
757 trace_kfree_skb(skb, __builtin_return_address(0));
760 EXPORT_SYMBOL(kfree_skb);
762 void kfree_skb_list(struct sk_buff *segs)
765 struct sk_buff *next = segs->next;
771 EXPORT_SYMBOL(kfree_skb_list);
773 /* Dump skb information and contents.
775 * Must only be called from net_ratelimit()-ed paths.
777 * Dumps whole packets if full_pkt, only headers otherwise.
779 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
781 struct skb_shared_info *sh = skb_shinfo(skb);
782 struct net_device *dev = skb->dev;
783 struct sock *sk = skb->sk;
784 struct sk_buff *list_skb;
785 bool has_mac, has_trans;
786 int headroom, tailroom;
792 len = min_t(int, skb->len, MAX_HEADER + 128);
794 headroom = skb_headroom(skb);
795 tailroom = skb_tailroom(skb);
797 has_mac = skb_mac_header_was_set(skb);
798 has_trans = skb_transport_header_was_set(skb);
800 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
801 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
802 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
803 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
804 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
805 level, skb->len, headroom, skb_headlen(skb), tailroom,
806 has_mac ? skb->mac_header : -1,
807 has_mac ? skb_mac_header_len(skb) : -1,
809 has_trans ? skb_network_header_len(skb) : -1,
810 has_trans ? skb->transport_header : -1,
811 sh->tx_flags, sh->nr_frags,
812 sh->gso_size, sh->gso_type, sh->gso_segs,
813 skb->csum, skb->ip_summed, skb->csum_complete_sw,
814 skb->csum_valid, skb->csum_level,
815 skb->hash, skb->sw_hash, skb->l4_hash,
816 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
819 printk("%sdev name=%s feat=0x%pNF\n",
820 level, dev->name, &dev->features);
822 printk("%ssk family=%hu type=%u proto=%u\n",
823 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
825 if (full_pkt && headroom)
826 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
827 16, 1, skb->head, headroom, false);
829 seg_len = min_t(int, skb_headlen(skb), len);
831 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
832 16, 1, skb->data, seg_len, false);
835 if (full_pkt && tailroom)
836 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
837 16, 1, skb_tail_pointer(skb), tailroom, false);
839 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
840 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
841 u32 p_off, p_len, copied;
845 skb_frag_foreach_page(frag, skb_frag_off(frag),
846 skb_frag_size(frag), p, p_off, p_len,
848 seg_len = min_t(int, p_len, len);
849 vaddr = kmap_atomic(p);
850 print_hex_dump(level, "skb frag: ",
852 16, 1, vaddr + p_off, seg_len, false);
853 kunmap_atomic(vaddr);
860 if (full_pkt && skb_has_frag_list(skb)) {
861 printk("skb fraglist:\n");
862 skb_walk_frags(skb, list_skb)
863 skb_dump(level, list_skb, true);
866 EXPORT_SYMBOL(skb_dump);
869 * skb_tx_error - report an sk_buff xmit error
870 * @skb: buffer that triggered an error
872 * Report xmit error if a device callback is tracking this skb.
873 * skb must be freed afterwards.
875 void skb_tx_error(struct sk_buff *skb)
877 skb_zcopy_clear(skb, true);
879 EXPORT_SYMBOL(skb_tx_error);
881 #ifdef CONFIG_TRACEPOINTS
883 * consume_skb - free an skbuff
884 * @skb: buffer to free
886 * Drop a ref to the buffer and free it if the usage count has hit zero
887 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
888 * is being dropped after a failure and notes that
890 void consume_skb(struct sk_buff *skb)
895 trace_consume_skb(skb);
898 EXPORT_SYMBOL(consume_skb);
902 * __consume_stateless_skb - free an skbuff, assuming it is stateless
903 * @skb: buffer to free
905 * Alike consume_skb(), but this variant assumes that this is the last
906 * skb reference and all the head states have been already dropped
908 void __consume_stateless_skb(struct sk_buff *skb)
910 trace_consume_skb(skb);
911 skb_release_data(skb);
915 static void napi_skb_cache_put(struct sk_buff *skb)
917 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
920 kasan_poison_object_data(skbuff_head_cache, skb);
921 nc->skb_cache[nc->skb_count++] = skb;
923 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
924 for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++)
925 kasan_unpoison_object_data(skbuff_head_cache,
928 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF,
929 nc->skb_cache + NAPI_SKB_CACHE_HALF);
930 nc->skb_count = NAPI_SKB_CACHE_HALF;
934 void __kfree_skb_defer(struct sk_buff *skb)
936 skb_release_all(skb);
937 napi_skb_cache_put(skb);
940 void napi_skb_free_stolen_head(struct sk_buff *skb)
945 napi_skb_cache_put(skb);
948 void napi_consume_skb(struct sk_buff *skb, int budget)
950 /* Zero budget indicate non-NAPI context called us, like netpoll */
951 if (unlikely(!budget)) {
952 dev_consume_skb_any(skb);
956 lockdep_assert_in_softirq();
961 /* if reaching here SKB is ready to free */
962 trace_consume_skb(skb);
964 /* if SKB is a clone, don't handle this case */
965 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
970 skb_release_all(skb);
971 napi_skb_cache_put(skb);
973 EXPORT_SYMBOL(napi_consume_skb);
975 /* Make sure a field is enclosed inside headers_start/headers_end section */
976 #define CHECK_SKB_FIELD(field) \
977 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
978 offsetof(struct sk_buff, headers_start)); \
979 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
980 offsetof(struct sk_buff, headers_end)); \
982 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
984 new->tstamp = old->tstamp;
985 /* We do not copy old->sk */
987 memcpy(new->cb, old->cb, sizeof(old->cb));
988 skb_dst_copy(new, old);
989 __skb_ext_copy(new, old);
990 __nf_copy(new, old, false);
992 /* Note : this field could be in headers_start/headers_end section
993 * It is not yet because we do not want to have a 16 bit hole
995 new->queue_mapping = old->queue_mapping;
997 memcpy(&new->headers_start, &old->headers_start,
998 offsetof(struct sk_buff, headers_end) -
999 offsetof(struct sk_buff, headers_start));
1000 CHECK_SKB_FIELD(protocol);
1001 CHECK_SKB_FIELD(csum);
1002 CHECK_SKB_FIELD(hash);
1003 CHECK_SKB_FIELD(priority);
1004 CHECK_SKB_FIELD(skb_iif);
1005 CHECK_SKB_FIELD(vlan_proto);
1006 CHECK_SKB_FIELD(vlan_tci);
1007 CHECK_SKB_FIELD(transport_header);
1008 CHECK_SKB_FIELD(network_header);
1009 CHECK_SKB_FIELD(mac_header);
1010 CHECK_SKB_FIELD(inner_protocol);
1011 CHECK_SKB_FIELD(inner_transport_header);
1012 CHECK_SKB_FIELD(inner_network_header);
1013 CHECK_SKB_FIELD(inner_mac_header);
1014 CHECK_SKB_FIELD(mark);
1015 #ifdef CONFIG_NETWORK_SECMARK
1016 CHECK_SKB_FIELD(secmark);
1018 #ifdef CONFIG_NET_RX_BUSY_POLL
1019 CHECK_SKB_FIELD(napi_id);
1022 CHECK_SKB_FIELD(sender_cpu);
1024 #ifdef CONFIG_NET_SCHED
1025 CHECK_SKB_FIELD(tc_index);
1031 * You should not add any new code to this function. Add it to
1032 * __copy_skb_header above instead.
1034 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
1036 #define C(x) n->x = skb->x
1038 n->next = n->prev = NULL;
1040 __copy_skb_header(n, skb);
1045 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1050 n->destructor = NULL;
1057 refcount_set(&n->users, 1);
1059 atomic_inc(&(skb_shinfo(skb)->dataref));
1067 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1068 * @first: first sk_buff of the msg
1070 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1074 n = alloc_skb(0, GFP_ATOMIC);
1078 n->len = first->len;
1079 n->data_len = first->len;
1080 n->truesize = first->truesize;
1082 skb_shinfo(n)->frag_list = first;
1084 __copy_skb_header(n, first);
1085 n->destructor = NULL;
1089 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1092 * skb_morph - morph one skb into another
1093 * @dst: the skb to receive the contents
1094 * @src: the skb to supply the contents
1096 * This is identical to skb_clone except that the target skb is
1097 * supplied by the user.
1099 * The target skb is returned upon exit.
1101 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1103 skb_release_all(dst);
1104 return __skb_clone(dst, src);
1106 EXPORT_SYMBOL_GPL(skb_morph);
1108 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1110 unsigned long max_pg, num_pg, new_pg, old_pg;
1111 struct user_struct *user;
1113 if (capable(CAP_IPC_LOCK) || !size)
1116 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1117 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1118 user = mmp->user ? : current_user();
1121 old_pg = atomic_long_read(&user->locked_vm);
1122 new_pg = old_pg + num_pg;
1123 if (new_pg > max_pg)
1125 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1129 mmp->user = get_uid(user);
1130 mmp->num_pg = num_pg;
1132 mmp->num_pg += num_pg;
1137 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1139 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1142 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1143 free_uid(mmp->user);
1146 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1148 struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size)
1150 struct ubuf_info *uarg;
1151 struct sk_buff *skb;
1153 WARN_ON_ONCE(!in_task());
1155 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1159 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1160 uarg = (void *)skb->cb;
1161 uarg->mmp.user = NULL;
1163 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1168 uarg->callback = msg_zerocopy_callback;
1169 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1171 uarg->bytelen = size;
1173 uarg->flags = SKBFL_ZEROCOPY_FRAG;
1174 refcount_set(&uarg->refcnt, 1);
1179 EXPORT_SYMBOL_GPL(msg_zerocopy_alloc);
1181 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1183 return container_of((void *)uarg, struct sk_buff, cb);
1186 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1187 struct ubuf_info *uarg)
1190 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1193 /* realloc only when socket is locked (TCP, UDP cork),
1194 * so uarg->len and sk_zckey access is serialized
1196 if (!sock_owned_by_user(sk)) {
1201 bytelen = uarg->bytelen + size;
1202 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1203 /* TCP can create new skb to attach new uarg */
1204 if (sk->sk_type == SOCK_STREAM)
1209 next = (u32)atomic_read(&sk->sk_zckey);
1210 if ((u32)(uarg->id + uarg->len) == next) {
1211 if (mm_account_pinned_pages(&uarg->mmp, size))
1214 uarg->bytelen = bytelen;
1215 atomic_set(&sk->sk_zckey, ++next);
1217 /* no extra ref when appending to datagram (MSG_MORE) */
1218 if (sk->sk_type == SOCK_STREAM)
1219 net_zcopy_get(uarg);
1226 return msg_zerocopy_alloc(sk, size);
1228 EXPORT_SYMBOL_GPL(msg_zerocopy_realloc);
1230 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1232 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1236 old_lo = serr->ee.ee_info;
1237 old_hi = serr->ee.ee_data;
1238 sum_len = old_hi - old_lo + 1ULL + len;
1240 if (sum_len >= (1ULL << 32))
1243 if (lo != old_hi + 1)
1246 serr->ee.ee_data += len;
1250 static void __msg_zerocopy_callback(struct ubuf_info *uarg)
1252 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1253 struct sock_exterr_skb *serr;
1254 struct sock *sk = skb->sk;
1255 struct sk_buff_head *q;
1256 unsigned long flags;
1261 mm_unaccount_pinned_pages(&uarg->mmp);
1263 /* if !len, there was only 1 call, and it was aborted
1264 * so do not queue a completion notification
1266 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1271 hi = uarg->id + len - 1;
1272 is_zerocopy = uarg->zerocopy;
1274 serr = SKB_EXT_ERR(skb);
1275 memset(serr, 0, sizeof(*serr));
1276 serr->ee.ee_errno = 0;
1277 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1278 serr->ee.ee_data = hi;
1279 serr->ee.ee_info = lo;
1281 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1283 q = &sk->sk_error_queue;
1284 spin_lock_irqsave(&q->lock, flags);
1285 tail = skb_peek_tail(q);
1286 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1287 !skb_zerocopy_notify_extend(tail, lo, len)) {
1288 __skb_queue_tail(q, skb);
1291 spin_unlock_irqrestore(&q->lock, flags);
1293 sk->sk_error_report(sk);
1300 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1303 uarg->zerocopy = uarg->zerocopy & success;
1305 if (refcount_dec_and_test(&uarg->refcnt))
1306 __msg_zerocopy_callback(uarg);
1308 EXPORT_SYMBOL_GPL(msg_zerocopy_callback);
1310 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1312 struct sock *sk = skb_from_uarg(uarg)->sk;
1314 atomic_dec(&sk->sk_zckey);
1318 msg_zerocopy_callback(NULL, uarg, true);
1320 EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort);
1322 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1324 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1326 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1328 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1329 struct msghdr *msg, int len,
1330 struct ubuf_info *uarg)
1332 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1333 struct iov_iter orig_iter = msg->msg_iter;
1334 int err, orig_len = skb->len;
1336 /* An skb can only point to one uarg. This edge case happens when
1337 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1339 if (orig_uarg && uarg != orig_uarg)
1342 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1343 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1344 struct sock *save_sk = skb->sk;
1346 /* Streams do not free skb on error. Reset to prev state. */
1347 msg->msg_iter = orig_iter;
1349 ___pskb_trim(skb, orig_len);
1354 skb_zcopy_set(skb, uarg, NULL);
1355 return skb->len - orig_len;
1357 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1359 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1362 if (skb_zcopy(orig)) {
1363 if (skb_zcopy(nskb)) {
1364 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1369 if (skb_uarg(nskb) == skb_uarg(orig))
1371 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1374 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1380 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1381 * @skb: the skb to modify
1382 * @gfp_mask: allocation priority
1384 * This must be called on skb with SKBFL_ZEROCOPY_ENABLE.
1385 * It will copy all frags into kernel and drop the reference
1386 * to userspace pages.
1388 * If this function is called from an interrupt gfp_mask() must be
1391 * Returns 0 on success or a negative error code on failure
1392 * to allocate kernel memory to copy to.
1394 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1396 int num_frags = skb_shinfo(skb)->nr_frags;
1397 struct page *page, *head = NULL;
1401 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1407 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1408 for (i = 0; i < new_frags; i++) {
1409 page = alloc_page(gfp_mask);
1412 struct page *next = (struct page *)page_private(head);
1418 set_page_private(page, (unsigned long)head);
1424 for (i = 0; i < num_frags; i++) {
1425 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1426 u32 p_off, p_len, copied;
1430 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1431 p, p_off, p_len, copied) {
1433 vaddr = kmap_atomic(p);
1435 while (done < p_len) {
1436 if (d_off == PAGE_SIZE) {
1438 page = (struct page *)page_private(page);
1440 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1441 memcpy(page_address(page) + d_off,
1442 vaddr + p_off + done, copy);
1446 kunmap_atomic(vaddr);
1450 /* skb frags release userspace buffers */
1451 for (i = 0; i < num_frags; i++)
1452 skb_frag_unref(skb, i);
1454 /* skb frags point to kernel buffers */
1455 for (i = 0; i < new_frags - 1; i++) {
1456 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1457 head = (struct page *)page_private(head);
1459 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1460 skb_shinfo(skb)->nr_frags = new_frags;
1463 skb_zcopy_clear(skb, false);
1466 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1469 * skb_clone - duplicate an sk_buff
1470 * @skb: buffer to clone
1471 * @gfp_mask: allocation priority
1473 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1474 * copies share the same packet data but not structure. The new
1475 * buffer has a reference count of 1. If the allocation fails the
1476 * function returns %NULL otherwise the new buffer is returned.
1478 * If this function is called from an interrupt gfp_mask() must be
1482 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1484 struct sk_buff_fclones *fclones = container_of(skb,
1485 struct sk_buff_fclones,
1489 if (skb_orphan_frags(skb, gfp_mask))
1492 if (skb->fclone == SKB_FCLONE_ORIG &&
1493 refcount_read(&fclones->fclone_ref) == 1) {
1495 refcount_set(&fclones->fclone_ref, 2);
1497 if (skb_pfmemalloc(skb))
1498 gfp_mask |= __GFP_MEMALLOC;
1500 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1504 n->fclone = SKB_FCLONE_UNAVAILABLE;
1507 return __skb_clone(n, skb);
1509 EXPORT_SYMBOL(skb_clone);
1511 void skb_headers_offset_update(struct sk_buff *skb, int off)
1513 /* Only adjust this if it actually is csum_start rather than csum */
1514 if (skb->ip_summed == CHECKSUM_PARTIAL)
1515 skb->csum_start += off;
1516 /* {transport,network,mac}_header and tail are relative to skb->head */
1517 skb->transport_header += off;
1518 skb->network_header += off;
1519 if (skb_mac_header_was_set(skb))
1520 skb->mac_header += off;
1521 skb->inner_transport_header += off;
1522 skb->inner_network_header += off;
1523 skb->inner_mac_header += off;
1525 EXPORT_SYMBOL(skb_headers_offset_update);
1527 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1529 __copy_skb_header(new, old);
1531 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1532 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1533 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1535 EXPORT_SYMBOL(skb_copy_header);
1537 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1539 if (skb_pfmemalloc(skb))
1540 return SKB_ALLOC_RX;
1545 * skb_copy - create private copy of an sk_buff
1546 * @skb: buffer to copy
1547 * @gfp_mask: allocation priority
1549 * Make a copy of both an &sk_buff and its data. This is used when the
1550 * caller wishes to modify the data and needs a private copy of the
1551 * data to alter. Returns %NULL on failure or the pointer to the buffer
1552 * on success. The returned buffer has a reference count of 1.
1554 * As by-product this function converts non-linear &sk_buff to linear
1555 * one, so that &sk_buff becomes completely private and caller is allowed
1556 * to modify all the data of returned buffer. This means that this
1557 * function is not recommended for use in circumstances when only
1558 * header is going to be modified. Use pskb_copy() instead.
1561 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1563 int headerlen = skb_headroom(skb);
1564 unsigned int size = skb_end_offset(skb) + skb->data_len;
1565 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1566 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1571 /* Set the data pointer */
1572 skb_reserve(n, headerlen);
1573 /* Set the tail pointer and length */
1574 skb_put(n, skb->len);
1576 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1578 skb_copy_header(n, skb);
1581 EXPORT_SYMBOL(skb_copy);
1584 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1585 * @skb: buffer to copy
1586 * @headroom: headroom of new skb
1587 * @gfp_mask: allocation priority
1588 * @fclone: if true allocate the copy of the skb from the fclone
1589 * cache instead of the head cache; it is recommended to set this
1590 * to true for the cases where the copy will likely be cloned
1592 * Make a copy of both an &sk_buff and part of its data, located
1593 * in header. Fragmented data remain shared. This is used when
1594 * the caller wishes to modify only header of &sk_buff and needs
1595 * private copy of the header to alter. Returns %NULL on failure
1596 * or the pointer to the buffer on success.
1597 * The returned buffer has a reference count of 1.
1600 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1601 gfp_t gfp_mask, bool fclone)
1603 unsigned int size = skb_headlen(skb) + headroom;
1604 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1605 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1610 /* Set the data pointer */
1611 skb_reserve(n, headroom);
1612 /* Set the tail pointer and length */
1613 skb_put(n, skb_headlen(skb));
1614 /* Copy the bytes */
1615 skb_copy_from_linear_data(skb, n->data, n->len);
1617 n->truesize += skb->data_len;
1618 n->data_len = skb->data_len;
1621 if (skb_shinfo(skb)->nr_frags) {
1624 if (skb_orphan_frags(skb, gfp_mask) ||
1625 skb_zerocopy_clone(n, skb, gfp_mask)) {
1630 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1631 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1632 skb_frag_ref(skb, i);
1634 skb_shinfo(n)->nr_frags = i;
1637 if (skb_has_frag_list(skb)) {
1638 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1639 skb_clone_fraglist(n);
1642 skb_copy_header(n, skb);
1646 EXPORT_SYMBOL(__pskb_copy_fclone);
1649 * pskb_expand_head - reallocate header of &sk_buff
1650 * @skb: buffer to reallocate
1651 * @nhead: room to add at head
1652 * @ntail: room to add at tail
1653 * @gfp_mask: allocation priority
1655 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1656 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1657 * reference count of 1. Returns zero in the case of success or error,
1658 * if expansion failed. In the last case, &sk_buff is not changed.
1660 * All the pointers pointing into skb header may change and must be
1661 * reloaded after call to this function.
1664 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1667 int i, osize = skb_end_offset(skb);
1668 int size = osize + nhead + ntail;
1674 BUG_ON(skb_shared(skb));
1676 size = SKB_DATA_ALIGN(size);
1678 if (skb_pfmemalloc(skb))
1679 gfp_mask |= __GFP_MEMALLOC;
1680 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1681 gfp_mask, NUMA_NO_NODE, NULL);
1684 size = SKB_WITH_OVERHEAD(ksize(data));
1686 /* Copy only real data... and, alas, header. This should be
1687 * optimized for the cases when header is void.
1689 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1691 memcpy((struct skb_shared_info *)(data + size),
1693 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1696 * if shinfo is shared we must drop the old head gracefully, but if it
1697 * is not we can just drop the old head and let the existing refcount
1698 * be since all we did is relocate the values
1700 if (skb_cloned(skb)) {
1701 if (skb_orphan_frags(skb, gfp_mask))
1704 refcount_inc(&skb_uarg(skb)->refcnt);
1705 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1706 skb_frag_ref(skb, i);
1708 if (skb_has_frag_list(skb))
1709 skb_clone_fraglist(skb);
1711 skb_release_data(skb);
1715 off = (data + nhead) - skb->head;
1720 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1724 skb->end = skb->head + size;
1727 skb_headers_offset_update(skb, nhead);
1731 atomic_set(&skb_shinfo(skb)->dataref, 1);
1733 skb_metadata_clear(skb);
1735 /* It is not generally safe to change skb->truesize.
1736 * For the moment, we really care of rx path, or
1737 * when skb is orphaned (not attached to a socket).
1739 if (!skb->sk || skb->destructor == sock_edemux)
1740 skb->truesize += size - osize;
1749 EXPORT_SYMBOL(pskb_expand_head);
1751 /* Make private copy of skb with writable head and some headroom */
1753 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1755 struct sk_buff *skb2;
1756 int delta = headroom - skb_headroom(skb);
1759 skb2 = pskb_copy(skb, GFP_ATOMIC);
1761 skb2 = skb_clone(skb, GFP_ATOMIC);
1762 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1770 EXPORT_SYMBOL(skb_realloc_headroom);
1773 * skb_copy_expand - copy and expand sk_buff
1774 * @skb: buffer to copy
1775 * @newheadroom: new free bytes at head
1776 * @newtailroom: new free bytes at tail
1777 * @gfp_mask: allocation priority
1779 * Make a copy of both an &sk_buff and its data and while doing so
1780 * allocate additional space.
1782 * This is used when the caller wishes to modify the data and needs a
1783 * private copy of the data to alter as well as more space for new fields.
1784 * Returns %NULL on failure or the pointer to the buffer
1785 * on success. The returned buffer has a reference count of 1.
1787 * You must pass %GFP_ATOMIC as the allocation priority if this function
1788 * is called from an interrupt.
1790 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1791 int newheadroom, int newtailroom,
1795 * Allocate the copy buffer
1797 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1798 gfp_mask, skb_alloc_rx_flag(skb),
1800 int oldheadroom = skb_headroom(skb);
1801 int head_copy_len, head_copy_off;
1806 skb_reserve(n, newheadroom);
1808 /* Set the tail pointer and length */
1809 skb_put(n, skb->len);
1811 head_copy_len = oldheadroom;
1813 if (newheadroom <= head_copy_len)
1814 head_copy_len = newheadroom;
1816 head_copy_off = newheadroom - head_copy_len;
1818 /* Copy the linear header and data. */
1819 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1820 skb->len + head_copy_len));
1822 skb_copy_header(n, skb);
1824 skb_headers_offset_update(n, newheadroom - oldheadroom);
1828 EXPORT_SYMBOL(skb_copy_expand);
1831 * __skb_pad - zero pad the tail of an skb
1832 * @skb: buffer to pad
1833 * @pad: space to pad
1834 * @free_on_error: free buffer on error
1836 * Ensure that a buffer is followed by a padding area that is zero
1837 * filled. Used by network drivers which may DMA or transfer data
1838 * beyond the buffer end onto the wire.
1840 * May return error in out of memory cases. The skb is freed on error
1841 * if @free_on_error is true.
1844 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1849 /* If the skbuff is non linear tailroom is always zero.. */
1850 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1851 memset(skb->data+skb->len, 0, pad);
1855 ntail = skb->data_len + pad - (skb->end - skb->tail);
1856 if (likely(skb_cloned(skb) || ntail > 0)) {
1857 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1862 /* FIXME: The use of this function with non-linear skb's really needs
1865 err = skb_linearize(skb);
1869 memset(skb->data + skb->len, 0, pad);
1877 EXPORT_SYMBOL(__skb_pad);
1880 * pskb_put - add data to the tail of a potentially fragmented buffer
1881 * @skb: start of the buffer to use
1882 * @tail: tail fragment of the buffer to use
1883 * @len: amount of data to add
1885 * This function extends the used data area of the potentially
1886 * fragmented buffer. @tail must be the last fragment of @skb -- or
1887 * @skb itself. If this would exceed the total buffer size the kernel
1888 * will panic. A pointer to the first byte of the extra data is
1892 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1895 skb->data_len += len;
1898 return skb_put(tail, len);
1900 EXPORT_SYMBOL_GPL(pskb_put);
1903 * skb_put - add data to a buffer
1904 * @skb: buffer to use
1905 * @len: amount of data to add
1907 * This function extends the used data area of the buffer. If this would
1908 * exceed the total buffer size the kernel will panic. A pointer to the
1909 * first byte of the extra data is returned.
1911 void *skb_put(struct sk_buff *skb, unsigned int len)
1913 void *tmp = skb_tail_pointer(skb);
1914 SKB_LINEAR_ASSERT(skb);
1917 if (unlikely(skb->tail > skb->end))
1918 skb_over_panic(skb, len, __builtin_return_address(0));
1921 EXPORT_SYMBOL(skb_put);
1924 * skb_push - add data to the start of a buffer
1925 * @skb: buffer to use
1926 * @len: amount of data to add
1928 * This function extends the used data area of the buffer at the buffer
1929 * start. If this would exceed the total buffer headroom the kernel will
1930 * panic. A pointer to the first byte of the extra data is returned.
1932 void *skb_push(struct sk_buff *skb, unsigned int len)
1936 if (unlikely(skb->data < skb->head))
1937 skb_under_panic(skb, len, __builtin_return_address(0));
1940 EXPORT_SYMBOL(skb_push);
1943 * skb_pull - remove data from the start of a buffer
1944 * @skb: buffer to use
1945 * @len: amount of data to remove
1947 * This function removes data from the start of a buffer, returning
1948 * the memory to the headroom. A pointer to the next data in the buffer
1949 * is returned. Once the data has been pulled future pushes will overwrite
1952 void *skb_pull(struct sk_buff *skb, unsigned int len)
1954 return skb_pull_inline(skb, len);
1956 EXPORT_SYMBOL(skb_pull);
1959 * skb_trim - remove end from a buffer
1960 * @skb: buffer to alter
1963 * Cut the length of a buffer down by removing data from the tail. If
1964 * the buffer is already under the length specified it is not modified.
1965 * The skb must be linear.
1967 void skb_trim(struct sk_buff *skb, unsigned int len)
1970 __skb_trim(skb, len);
1972 EXPORT_SYMBOL(skb_trim);
1974 /* Trims skb to length len. It can change skb pointers.
1977 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1979 struct sk_buff **fragp;
1980 struct sk_buff *frag;
1981 int offset = skb_headlen(skb);
1982 int nfrags = skb_shinfo(skb)->nr_frags;
1986 if (skb_cloned(skb) &&
1987 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1994 for (; i < nfrags; i++) {
1995 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2002 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
2005 skb_shinfo(skb)->nr_frags = i;
2007 for (; i < nfrags; i++)
2008 skb_frag_unref(skb, i);
2010 if (skb_has_frag_list(skb))
2011 skb_drop_fraglist(skb);
2015 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
2016 fragp = &frag->next) {
2017 int end = offset + frag->len;
2019 if (skb_shared(frag)) {
2020 struct sk_buff *nfrag;
2022 nfrag = skb_clone(frag, GFP_ATOMIC);
2023 if (unlikely(!nfrag))
2026 nfrag->next = frag->next;
2038 unlikely((err = pskb_trim(frag, len - offset))))
2042 skb_drop_list(&frag->next);
2047 if (len > skb_headlen(skb)) {
2048 skb->data_len -= skb->len - len;
2053 skb_set_tail_pointer(skb, len);
2056 if (!skb->sk || skb->destructor == sock_edemux)
2060 EXPORT_SYMBOL(___pskb_trim);
2062 /* Note : use pskb_trim_rcsum() instead of calling this directly
2064 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2066 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2067 int delta = skb->len - len;
2069 skb->csum = csum_block_sub(skb->csum,
2070 skb_checksum(skb, len, delta, 0),
2072 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2073 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2074 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2076 if (offset + sizeof(__sum16) > hdlen)
2079 return __pskb_trim(skb, len);
2081 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2084 * __pskb_pull_tail - advance tail of skb header
2085 * @skb: buffer to reallocate
2086 * @delta: number of bytes to advance tail
2088 * The function makes a sense only on a fragmented &sk_buff,
2089 * it expands header moving its tail forward and copying necessary
2090 * data from fragmented part.
2092 * &sk_buff MUST have reference count of 1.
2094 * Returns %NULL (and &sk_buff does not change) if pull failed
2095 * or value of new tail of skb in the case of success.
2097 * All the pointers pointing into skb header may change and must be
2098 * reloaded after call to this function.
2101 /* Moves tail of skb head forward, copying data from fragmented part,
2102 * when it is necessary.
2103 * 1. It may fail due to malloc failure.
2104 * 2. It may change skb pointers.
2106 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2108 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2110 /* If skb has not enough free space at tail, get new one
2111 * plus 128 bytes for future expansions. If we have enough
2112 * room at tail, reallocate without expansion only if skb is cloned.
2114 int i, k, eat = (skb->tail + delta) - skb->end;
2116 if (eat > 0 || skb_cloned(skb)) {
2117 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2122 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2123 skb_tail_pointer(skb), delta));
2125 /* Optimization: no fragments, no reasons to preestimate
2126 * size of pulled pages. Superb.
2128 if (!skb_has_frag_list(skb))
2131 /* Estimate size of pulled pages. */
2133 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2134 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2141 /* If we need update frag list, we are in troubles.
2142 * Certainly, it is possible to add an offset to skb data,
2143 * but taking into account that pulling is expected to
2144 * be very rare operation, it is worth to fight against
2145 * further bloating skb head and crucify ourselves here instead.
2146 * Pure masohism, indeed. 8)8)
2149 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2150 struct sk_buff *clone = NULL;
2151 struct sk_buff *insp = NULL;
2154 if (list->len <= eat) {
2155 /* Eaten as whole. */
2160 /* Eaten partially. */
2162 if (skb_shared(list)) {
2163 /* Sucks! We need to fork list. :-( */
2164 clone = skb_clone(list, GFP_ATOMIC);
2170 /* This may be pulled without
2174 if (!pskb_pull(list, eat)) {
2182 /* Free pulled out fragments. */
2183 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2184 skb_shinfo(skb)->frag_list = list->next;
2187 /* And insert new clone at head. */
2190 skb_shinfo(skb)->frag_list = clone;
2193 /* Success! Now we may commit changes to skb data. */
2198 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2199 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2202 skb_frag_unref(skb, i);
2205 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2207 *frag = skb_shinfo(skb)->frags[i];
2209 skb_frag_off_add(frag, eat);
2210 skb_frag_size_sub(frag, eat);
2218 skb_shinfo(skb)->nr_frags = k;
2222 skb->data_len -= delta;
2225 skb_zcopy_clear(skb, false);
2227 return skb_tail_pointer(skb);
2229 EXPORT_SYMBOL(__pskb_pull_tail);
2232 * skb_copy_bits - copy bits from skb to kernel buffer
2234 * @offset: offset in source
2235 * @to: destination buffer
2236 * @len: number of bytes to copy
2238 * Copy the specified number of bytes from the source skb to the
2239 * destination buffer.
2242 * If its prototype is ever changed,
2243 * check arch/{*}/net/{*}.S files,
2244 * since it is called from BPF assembly code.
2246 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2248 int start = skb_headlen(skb);
2249 struct sk_buff *frag_iter;
2252 if (offset > (int)skb->len - len)
2256 if ((copy = start - offset) > 0) {
2259 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2260 if ((len -= copy) == 0)
2266 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2268 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2270 WARN_ON(start > offset + len);
2272 end = start + skb_frag_size(f);
2273 if ((copy = end - offset) > 0) {
2274 u32 p_off, p_len, copied;
2281 skb_frag_foreach_page(f,
2282 skb_frag_off(f) + offset - start,
2283 copy, p, p_off, p_len, copied) {
2284 vaddr = kmap_atomic(p);
2285 memcpy(to + copied, vaddr + p_off, p_len);
2286 kunmap_atomic(vaddr);
2289 if ((len -= copy) == 0)
2297 skb_walk_frags(skb, frag_iter) {
2300 WARN_ON(start > offset + len);
2302 end = start + frag_iter->len;
2303 if ((copy = end - offset) > 0) {
2306 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2308 if ((len -= copy) == 0)
2322 EXPORT_SYMBOL(skb_copy_bits);
2325 * Callback from splice_to_pipe(), if we need to release some pages
2326 * at the end of the spd in case we error'ed out in filling the pipe.
2328 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2330 put_page(spd->pages[i]);
2333 static struct page *linear_to_page(struct page *page, unsigned int *len,
2334 unsigned int *offset,
2337 struct page_frag *pfrag = sk_page_frag(sk);
2339 if (!sk_page_frag_refill(sk, pfrag))
2342 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2344 memcpy(page_address(pfrag->page) + pfrag->offset,
2345 page_address(page) + *offset, *len);
2346 *offset = pfrag->offset;
2347 pfrag->offset += *len;
2352 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2354 unsigned int offset)
2356 return spd->nr_pages &&
2357 spd->pages[spd->nr_pages - 1] == page &&
2358 (spd->partial[spd->nr_pages - 1].offset +
2359 spd->partial[spd->nr_pages - 1].len == offset);
2363 * Fill page/offset/length into spd, if it can hold more pages.
2365 static bool spd_fill_page(struct splice_pipe_desc *spd,
2366 struct pipe_inode_info *pipe, struct page *page,
2367 unsigned int *len, unsigned int offset,
2371 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2375 page = linear_to_page(page, len, &offset, sk);
2379 if (spd_can_coalesce(spd, page, offset)) {
2380 spd->partial[spd->nr_pages - 1].len += *len;
2384 spd->pages[spd->nr_pages] = page;
2385 spd->partial[spd->nr_pages].len = *len;
2386 spd->partial[spd->nr_pages].offset = offset;
2392 static bool __splice_segment(struct page *page, unsigned int poff,
2393 unsigned int plen, unsigned int *off,
2395 struct splice_pipe_desc *spd, bool linear,
2397 struct pipe_inode_info *pipe)
2402 /* skip this segment if already processed */
2408 /* ignore any bits we already processed */
2414 unsigned int flen = min(*len, plen);
2416 if (spd_fill_page(spd, pipe, page, &flen, poff,
2422 } while (*len && plen);
2428 * Map linear and fragment data from the skb to spd. It reports true if the
2429 * pipe is full or if we already spliced the requested length.
2431 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2432 unsigned int *offset, unsigned int *len,
2433 struct splice_pipe_desc *spd, struct sock *sk)
2436 struct sk_buff *iter;
2438 /* map the linear part :
2439 * If skb->head_frag is set, this 'linear' part is backed by a
2440 * fragment, and if the head is not shared with any clones then
2441 * we can avoid a copy since we own the head portion of this page.
2443 if (__splice_segment(virt_to_page(skb->data),
2444 (unsigned long) skb->data & (PAGE_SIZE - 1),
2447 skb_head_is_locked(skb),
2452 * then map the fragments
2454 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2455 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2457 if (__splice_segment(skb_frag_page(f),
2458 skb_frag_off(f), skb_frag_size(f),
2459 offset, len, spd, false, sk, pipe))
2463 skb_walk_frags(skb, iter) {
2464 if (*offset >= iter->len) {
2465 *offset -= iter->len;
2468 /* __skb_splice_bits() only fails if the output has no room
2469 * left, so no point in going over the frag_list for the error
2472 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2480 * Map data from the skb to a pipe. Should handle both the linear part,
2481 * the fragments, and the frag list.
2483 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2484 struct pipe_inode_info *pipe, unsigned int tlen,
2487 struct partial_page partial[MAX_SKB_FRAGS];
2488 struct page *pages[MAX_SKB_FRAGS];
2489 struct splice_pipe_desc spd = {
2492 .nr_pages_max = MAX_SKB_FRAGS,
2493 .ops = &nosteal_pipe_buf_ops,
2494 .spd_release = sock_spd_release,
2498 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2501 ret = splice_to_pipe(pipe, &spd);
2505 EXPORT_SYMBOL_GPL(skb_splice_bits);
2507 static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg,
2508 struct kvec *vec, size_t num, size_t size)
2510 struct socket *sock = sk->sk_socket;
2514 return kernel_sendmsg(sock, msg, vec, num, size);
2517 static int sendpage_unlocked(struct sock *sk, struct page *page, int offset,
2518 size_t size, int flags)
2520 struct socket *sock = sk->sk_socket;
2524 return kernel_sendpage(sock, page, offset, size, flags);
2527 typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg,
2528 struct kvec *vec, size_t num, size_t size);
2529 typedef int (*sendpage_func)(struct sock *sk, struct page *page, int offset,
2530 size_t size, int flags);
2531 static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset,
2532 int len, sendmsg_func sendmsg, sendpage_func sendpage)
2534 unsigned int orig_len = len;
2535 struct sk_buff *head = skb;
2536 unsigned short fragidx;
2541 /* Deal with head data */
2542 while (offset < skb_headlen(skb) && len) {
2546 slen = min_t(int, len, skb_headlen(skb) - offset);
2547 kv.iov_base = skb->data + offset;
2549 memset(&msg, 0, sizeof(msg));
2550 msg.msg_flags = MSG_DONTWAIT;
2552 ret = INDIRECT_CALL_2(sendmsg, kernel_sendmsg_locked,
2553 sendmsg_unlocked, sk, &msg, &kv, 1, slen);
2561 /* All the data was skb head? */
2565 /* Make offset relative to start of frags */
2566 offset -= skb_headlen(skb);
2568 /* Find where we are in frag list */
2569 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2570 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2572 if (offset < skb_frag_size(frag))
2575 offset -= skb_frag_size(frag);
2578 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2579 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2581 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2584 ret = INDIRECT_CALL_2(sendpage, kernel_sendpage_locked,
2585 sendpage_unlocked, sk,
2586 skb_frag_page(frag),
2587 skb_frag_off(frag) + offset,
2588 slen, MSG_DONTWAIT);
2601 /* Process any frag lists */
2604 if (skb_has_frag_list(skb)) {
2605 skb = skb_shinfo(skb)->frag_list;
2608 } else if (skb->next) {
2615 return orig_len - len;
2618 return orig_len == len ? ret : orig_len - len;
2621 /* Send skb data on a socket. Socket must be locked. */
2622 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2625 return __skb_send_sock(sk, skb, offset, len, kernel_sendmsg_locked,
2626 kernel_sendpage_locked);
2628 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2630 /* Send skb data on a socket. Socket must be unlocked. */
2631 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
2633 return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked,
2638 * skb_store_bits - store bits from kernel buffer to skb
2639 * @skb: destination buffer
2640 * @offset: offset in destination
2641 * @from: source buffer
2642 * @len: number of bytes to copy
2644 * Copy the specified number of bytes from the source buffer to the
2645 * destination skb. This function handles all the messy bits of
2646 * traversing fragment lists and such.
2649 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2651 int start = skb_headlen(skb);
2652 struct sk_buff *frag_iter;
2655 if (offset > (int)skb->len - len)
2658 if ((copy = start - offset) > 0) {
2661 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2662 if ((len -= copy) == 0)
2668 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2669 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2672 WARN_ON(start > offset + len);
2674 end = start + skb_frag_size(frag);
2675 if ((copy = end - offset) > 0) {
2676 u32 p_off, p_len, copied;
2683 skb_frag_foreach_page(frag,
2684 skb_frag_off(frag) + offset - start,
2685 copy, p, p_off, p_len, copied) {
2686 vaddr = kmap_atomic(p);
2687 memcpy(vaddr + p_off, from + copied, p_len);
2688 kunmap_atomic(vaddr);
2691 if ((len -= copy) == 0)
2699 skb_walk_frags(skb, frag_iter) {
2702 WARN_ON(start > offset + len);
2704 end = start + frag_iter->len;
2705 if ((copy = end - offset) > 0) {
2708 if (skb_store_bits(frag_iter, offset - start,
2711 if ((len -= copy) == 0)
2724 EXPORT_SYMBOL(skb_store_bits);
2726 /* Checksum skb data. */
2727 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2728 __wsum csum, const struct skb_checksum_ops *ops)
2730 int start = skb_headlen(skb);
2731 int i, copy = start - offset;
2732 struct sk_buff *frag_iter;
2735 /* Checksum header. */
2739 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2740 skb->data + offset, copy, csum);
2741 if ((len -= copy) == 0)
2747 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2749 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2751 WARN_ON(start > offset + len);
2753 end = start + skb_frag_size(frag);
2754 if ((copy = end - offset) > 0) {
2755 u32 p_off, p_len, copied;
2763 skb_frag_foreach_page(frag,
2764 skb_frag_off(frag) + offset - start,
2765 copy, p, p_off, p_len, copied) {
2766 vaddr = kmap_atomic(p);
2767 csum2 = INDIRECT_CALL_1(ops->update,
2769 vaddr + p_off, p_len, 0);
2770 kunmap_atomic(vaddr);
2771 csum = INDIRECT_CALL_1(ops->combine,
2772 csum_block_add_ext, csum,
2784 skb_walk_frags(skb, frag_iter) {
2787 WARN_ON(start > offset + len);
2789 end = start + frag_iter->len;
2790 if ((copy = end - offset) > 0) {
2794 csum2 = __skb_checksum(frag_iter, offset - start,
2796 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2797 csum, csum2, pos, copy);
2798 if ((len -= copy) == 0)
2809 EXPORT_SYMBOL(__skb_checksum);
2811 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2812 int len, __wsum csum)
2814 const struct skb_checksum_ops ops = {
2815 .update = csum_partial_ext,
2816 .combine = csum_block_add_ext,
2819 return __skb_checksum(skb, offset, len, csum, &ops);
2821 EXPORT_SYMBOL(skb_checksum);
2823 /* Both of above in one bottle. */
2825 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2828 int start = skb_headlen(skb);
2829 int i, copy = start - offset;
2830 struct sk_buff *frag_iter;
2838 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2840 if ((len -= copy) == 0)
2847 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2850 WARN_ON(start > offset + len);
2852 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2853 if ((copy = end - offset) > 0) {
2854 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2855 u32 p_off, p_len, copied;
2863 skb_frag_foreach_page(frag,
2864 skb_frag_off(frag) + offset - start,
2865 copy, p, p_off, p_len, copied) {
2866 vaddr = kmap_atomic(p);
2867 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2870 kunmap_atomic(vaddr);
2871 csum = csum_block_add(csum, csum2, pos);
2883 skb_walk_frags(skb, frag_iter) {
2887 WARN_ON(start > offset + len);
2889 end = start + frag_iter->len;
2890 if ((copy = end - offset) > 0) {
2893 csum2 = skb_copy_and_csum_bits(frag_iter,
2896 csum = csum_block_add(csum, csum2, pos);
2897 if ((len -= copy) == 0)
2908 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2910 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2914 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2915 /* See comments in __skb_checksum_complete(). */
2917 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2918 !skb->csum_complete_sw)
2919 netdev_rx_csum_fault(skb->dev, skb);
2921 if (!skb_shared(skb))
2922 skb->csum_valid = !sum;
2925 EXPORT_SYMBOL(__skb_checksum_complete_head);
2927 /* This function assumes skb->csum already holds pseudo header's checksum,
2928 * which has been changed from the hardware checksum, for example, by
2929 * __skb_checksum_validate_complete(). And, the original skb->csum must
2930 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2932 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2933 * zero. The new checksum is stored back into skb->csum unless the skb is
2936 __sum16 __skb_checksum_complete(struct sk_buff *skb)
2941 csum = skb_checksum(skb, 0, skb->len, 0);
2943 sum = csum_fold(csum_add(skb->csum, csum));
2944 /* This check is inverted, because we already knew the hardware
2945 * checksum is invalid before calling this function. So, if the
2946 * re-computed checksum is valid instead, then we have a mismatch
2947 * between the original skb->csum and skb_checksum(). This means either
2948 * the original hardware checksum is incorrect or we screw up skb->csum
2949 * when moving skb->data around.
2952 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2953 !skb->csum_complete_sw)
2954 netdev_rx_csum_fault(skb->dev, skb);
2957 if (!skb_shared(skb)) {
2958 /* Save full packet checksum */
2960 skb->ip_summed = CHECKSUM_COMPLETE;
2961 skb->csum_complete_sw = 1;
2962 skb->csum_valid = !sum;
2967 EXPORT_SYMBOL(__skb_checksum_complete);
2969 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2971 net_warn_ratelimited(
2972 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2977 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2978 int offset, int len)
2980 net_warn_ratelimited(
2981 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2986 static const struct skb_checksum_ops default_crc32c_ops = {
2987 .update = warn_crc32c_csum_update,
2988 .combine = warn_crc32c_csum_combine,
2991 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2992 &default_crc32c_ops;
2993 EXPORT_SYMBOL(crc32c_csum_stub);
2996 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2997 * @from: source buffer
2999 * Calculates the amount of linear headroom needed in the 'to' skb passed
3000 * into skb_zerocopy().
3003 skb_zerocopy_headlen(const struct sk_buff *from)
3005 unsigned int hlen = 0;
3007 if (!from->head_frag ||
3008 skb_headlen(from) < L1_CACHE_BYTES ||
3009 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) {
3010 hlen = skb_headlen(from);
3015 if (skb_has_frag_list(from))
3020 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
3023 * skb_zerocopy - Zero copy skb to skb
3024 * @to: destination buffer
3025 * @from: source buffer
3026 * @len: number of bytes to copy from source buffer
3027 * @hlen: size of linear headroom in destination buffer
3029 * Copies up to `len` bytes from `from` to `to` by creating references
3030 * to the frags in the source buffer.
3032 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
3033 * headroom in the `to` buffer.
3036 * 0: everything is OK
3037 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
3038 * -EFAULT: skb_copy_bits() found some problem with skb geometry
3041 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
3044 int plen = 0; /* length of skb->head fragment */
3047 unsigned int offset;
3049 BUG_ON(!from->head_frag && !hlen);
3051 /* dont bother with small payloads */
3052 if (len <= skb_tailroom(to))
3053 return skb_copy_bits(from, 0, skb_put(to, len), len);
3056 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
3061 plen = min_t(int, skb_headlen(from), len);
3063 page = virt_to_head_page(from->head);
3064 offset = from->data - (unsigned char *)page_address(page);
3065 __skb_fill_page_desc(to, 0, page, offset, plen);
3072 to->truesize += len + plen;
3073 to->len += len + plen;
3074 to->data_len += len + plen;
3076 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
3080 skb_zerocopy_clone(to, from, GFP_ATOMIC);
3082 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3087 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3088 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3090 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3092 skb_frag_ref(to, j);
3095 skb_shinfo(to)->nr_frags = j;
3099 EXPORT_SYMBOL_GPL(skb_zerocopy);
3101 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3106 if (skb->ip_summed == CHECKSUM_PARTIAL)
3107 csstart = skb_checksum_start_offset(skb);
3109 csstart = skb_headlen(skb);
3111 BUG_ON(csstart > skb_headlen(skb));
3113 skb_copy_from_linear_data(skb, to, csstart);
3116 if (csstart != skb->len)
3117 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3118 skb->len - csstart);
3120 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3121 long csstuff = csstart + skb->csum_offset;
3123 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3126 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3129 * skb_dequeue - remove from the head of the queue
3130 * @list: list to dequeue from
3132 * Remove the head of the list. The list lock is taken so the function
3133 * may be used safely with other locking list functions. The head item is
3134 * returned or %NULL if the list is empty.
3137 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3139 unsigned long flags;
3140 struct sk_buff *result;
3142 spin_lock_irqsave(&list->lock, flags);
3143 result = __skb_dequeue(list);
3144 spin_unlock_irqrestore(&list->lock, flags);
3147 EXPORT_SYMBOL(skb_dequeue);
3150 * skb_dequeue_tail - remove from the tail of the queue
3151 * @list: list to dequeue from
3153 * Remove the tail of the list. The list lock is taken so the function
3154 * may be used safely with other locking list functions. The tail item is
3155 * returned or %NULL if the list is empty.
3157 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3159 unsigned long flags;
3160 struct sk_buff *result;
3162 spin_lock_irqsave(&list->lock, flags);
3163 result = __skb_dequeue_tail(list);
3164 spin_unlock_irqrestore(&list->lock, flags);
3167 EXPORT_SYMBOL(skb_dequeue_tail);
3170 * skb_queue_purge - empty a list
3171 * @list: list to empty
3173 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3174 * the list and one reference dropped. This function takes the list
3175 * lock and is atomic with respect to other list locking functions.
3177 void skb_queue_purge(struct sk_buff_head *list)
3179 struct sk_buff *skb;
3180 while ((skb = skb_dequeue(list)) != NULL)
3183 EXPORT_SYMBOL(skb_queue_purge);
3186 * skb_rbtree_purge - empty a skb rbtree
3187 * @root: root of the rbtree to empty
3188 * Return value: the sum of truesizes of all purged skbs.
3190 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3191 * the list and one reference dropped. This function does not take
3192 * any lock. Synchronization should be handled by the caller (e.g., TCP
3193 * out-of-order queue is protected by the socket lock).
3195 unsigned int skb_rbtree_purge(struct rb_root *root)
3197 struct rb_node *p = rb_first(root);
3198 unsigned int sum = 0;
3201 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3204 rb_erase(&skb->rbnode, root);
3205 sum += skb->truesize;
3212 * skb_queue_head - queue a buffer at the list head
3213 * @list: list to use
3214 * @newsk: buffer to queue
3216 * Queue a buffer at the start of the list. This function takes the
3217 * list lock and can be used safely with other locking &sk_buff functions
3220 * A buffer cannot be placed on two lists at the same time.
3222 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3224 unsigned long flags;
3226 spin_lock_irqsave(&list->lock, flags);
3227 __skb_queue_head(list, newsk);
3228 spin_unlock_irqrestore(&list->lock, flags);
3230 EXPORT_SYMBOL(skb_queue_head);
3233 * skb_queue_tail - queue a buffer at the list tail
3234 * @list: list to use
3235 * @newsk: buffer to queue
3237 * Queue a buffer at the tail of the list. This function takes the
3238 * list lock and can be used safely with other locking &sk_buff functions
3241 * A buffer cannot be placed on two lists at the same time.
3243 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3245 unsigned long flags;
3247 spin_lock_irqsave(&list->lock, flags);
3248 __skb_queue_tail(list, newsk);
3249 spin_unlock_irqrestore(&list->lock, flags);
3251 EXPORT_SYMBOL(skb_queue_tail);
3254 * skb_unlink - remove a buffer from a list
3255 * @skb: buffer to remove
3256 * @list: list to use
3258 * Remove a packet from a list. The list locks are taken and this
3259 * function is atomic with respect to other list locked calls
3261 * You must know what list the SKB is on.
3263 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3265 unsigned long flags;
3267 spin_lock_irqsave(&list->lock, flags);
3268 __skb_unlink(skb, list);
3269 spin_unlock_irqrestore(&list->lock, flags);
3271 EXPORT_SYMBOL(skb_unlink);
3274 * skb_append - append a buffer
3275 * @old: buffer to insert after
3276 * @newsk: buffer to insert
3277 * @list: list to use
3279 * Place a packet after a given packet in a list. The list locks are taken
3280 * and this function is atomic with respect to other list locked calls.
3281 * A buffer cannot be placed on two lists at the same time.
3283 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3285 unsigned long flags;
3287 spin_lock_irqsave(&list->lock, flags);
3288 __skb_queue_after(list, old, newsk);
3289 spin_unlock_irqrestore(&list->lock, flags);
3291 EXPORT_SYMBOL(skb_append);
3293 static inline void skb_split_inside_header(struct sk_buff *skb,
3294 struct sk_buff* skb1,
3295 const u32 len, const int pos)
3299 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3301 /* And move data appendix as is. */
3302 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3303 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3305 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3306 skb_shinfo(skb)->nr_frags = 0;
3307 skb1->data_len = skb->data_len;
3308 skb1->len += skb1->data_len;
3311 skb_set_tail_pointer(skb, len);
3314 static inline void skb_split_no_header(struct sk_buff *skb,
3315 struct sk_buff* skb1,
3316 const u32 len, int pos)
3319 const int nfrags = skb_shinfo(skb)->nr_frags;
3321 skb_shinfo(skb)->nr_frags = 0;
3322 skb1->len = skb1->data_len = skb->len - len;
3324 skb->data_len = len - pos;
3326 for (i = 0; i < nfrags; i++) {
3327 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3329 if (pos + size > len) {
3330 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3334 * We have two variants in this case:
3335 * 1. Move all the frag to the second
3336 * part, if it is possible. F.e.
3337 * this approach is mandatory for TUX,
3338 * where splitting is expensive.
3339 * 2. Split is accurately. We make this.
3341 skb_frag_ref(skb, i);
3342 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3343 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3344 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3345 skb_shinfo(skb)->nr_frags++;
3349 skb_shinfo(skb)->nr_frags++;
3352 skb_shinfo(skb1)->nr_frags = k;
3356 * skb_split - Split fragmented skb to two parts at length len.
3357 * @skb: the buffer to split
3358 * @skb1: the buffer to receive the second part
3359 * @len: new length for skb
3361 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3363 int pos = skb_headlen(skb);
3365 skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3366 skb_zerocopy_clone(skb1, skb, 0);
3367 if (len < pos) /* Split line is inside header. */
3368 skb_split_inside_header(skb, skb1, len, pos);
3369 else /* Second chunk has no header, nothing to copy. */
3370 skb_split_no_header(skb, skb1, len, pos);
3372 EXPORT_SYMBOL(skb_split);
3374 /* Shifting from/to a cloned skb is a no-go.
3376 * Caller cannot keep skb_shinfo related pointers past calling here!
3378 static int skb_prepare_for_shift(struct sk_buff *skb)
3382 if (skb_cloned(skb)) {
3383 /* Save and restore truesize: pskb_expand_head() may reallocate
3384 * memory where ksize(kmalloc(S)) != ksize(kmalloc(S)), but we
3385 * cannot change truesize at this point.
3387 unsigned int save_truesize = skb->truesize;
3389 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3390 skb->truesize = save_truesize;
3396 * skb_shift - Shifts paged data partially from skb to another
3397 * @tgt: buffer into which tail data gets added
3398 * @skb: buffer from which the paged data comes from
3399 * @shiftlen: shift up to this many bytes
3401 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3402 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3403 * It's up to caller to free skb if everything was shifted.
3405 * If @tgt runs out of frags, the whole operation is aborted.
3407 * Skb cannot include anything else but paged data while tgt is allowed
3408 * to have non-paged data as well.
3410 * TODO: full sized shift could be optimized but that would need
3411 * specialized skb free'er to handle frags without up-to-date nr_frags.
3413 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3415 int from, to, merge, todo;
3416 skb_frag_t *fragfrom, *fragto;
3418 BUG_ON(shiftlen > skb->len);
3420 if (skb_headlen(skb))
3422 if (skb_zcopy(tgt) || skb_zcopy(skb))
3427 to = skb_shinfo(tgt)->nr_frags;
3428 fragfrom = &skb_shinfo(skb)->frags[from];
3430 /* Actual merge is delayed until the point when we know we can
3431 * commit all, so that we don't have to undo partial changes
3434 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3435 skb_frag_off(fragfrom))) {
3440 todo -= skb_frag_size(fragfrom);
3442 if (skb_prepare_for_shift(skb) ||
3443 skb_prepare_for_shift(tgt))
3446 /* All previous frag pointers might be stale! */
3447 fragfrom = &skb_shinfo(skb)->frags[from];
3448 fragto = &skb_shinfo(tgt)->frags[merge];
3450 skb_frag_size_add(fragto, shiftlen);
3451 skb_frag_size_sub(fragfrom, shiftlen);
3452 skb_frag_off_add(fragfrom, shiftlen);
3460 /* Skip full, not-fitting skb to avoid expensive operations */
3461 if ((shiftlen == skb->len) &&
3462 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3465 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3468 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3469 if (to == MAX_SKB_FRAGS)
3472 fragfrom = &skb_shinfo(skb)->frags[from];
3473 fragto = &skb_shinfo(tgt)->frags[to];
3475 if (todo >= skb_frag_size(fragfrom)) {
3476 *fragto = *fragfrom;
3477 todo -= skb_frag_size(fragfrom);
3482 __skb_frag_ref(fragfrom);
3483 skb_frag_page_copy(fragto, fragfrom);
3484 skb_frag_off_copy(fragto, fragfrom);
3485 skb_frag_size_set(fragto, todo);
3487 skb_frag_off_add(fragfrom, todo);
3488 skb_frag_size_sub(fragfrom, todo);
3496 /* Ready to "commit" this state change to tgt */
3497 skb_shinfo(tgt)->nr_frags = to;
3500 fragfrom = &skb_shinfo(skb)->frags[0];
3501 fragto = &skb_shinfo(tgt)->frags[merge];
3503 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3504 __skb_frag_unref(fragfrom);
3507 /* Reposition in the original skb */
3509 while (from < skb_shinfo(skb)->nr_frags)
3510 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3511 skb_shinfo(skb)->nr_frags = to;
3513 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3516 /* Most likely the tgt won't ever need its checksum anymore, skb on
3517 * the other hand might need it if it needs to be resent
3519 tgt->ip_summed = CHECKSUM_PARTIAL;
3520 skb->ip_summed = CHECKSUM_PARTIAL;
3522 /* Yak, is it really working this way? Some helper please? */
3523 skb->len -= shiftlen;
3524 skb->data_len -= shiftlen;
3525 skb->truesize -= shiftlen;
3526 tgt->len += shiftlen;
3527 tgt->data_len += shiftlen;
3528 tgt->truesize += shiftlen;
3534 * skb_prepare_seq_read - Prepare a sequential read of skb data
3535 * @skb: the buffer to read
3536 * @from: lower offset of data to be read
3537 * @to: upper offset of data to be read
3538 * @st: state variable
3540 * Initializes the specified state variable. Must be called before
3541 * invoking skb_seq_read() for the first time.
3543 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3544 unsigned int to, struct skb_seq_state *st)
3546 st->lower_offset = from;
3547 st->upper_offset = to;
3548 st->root_skb = st->cur_skb = skb;
3549 st->frag_idx = st->stepped_offset = 0;
3550 st->frag_data = NULL;
3553 EXPORT_SYMBOL(skb_prepare_seq_read);
3556 * skb_seq_read - Sequentially read skb data
3557 * @consumed: number of bytes consumed by the caller so far
3558 * @data: destination pointer for data to be returned
3559 * @st: state variable
3561 * Reads a block of skb data at @consumed relative to the
3562 * lower offset specified to skb_prepare_seq_read(). Assigns
3563 * the head of the data block to @data and returns the length
3564 * of the block or 0 if the end of the skb data or the upper
3565 * offset has been reached.
3567 * The caller is not required to consume all of the data
3568 * returned, i.e. @consumed is typically set to the number
3569 * of bytes already consumed and the next call to
3570 * skb_seq_read() will return the remaining part of the block.
3572 * Note 1: The size of each block of data returned can be arbitrary,
3573 * this limitation is the cost for zerocopy sequential
3574 * reads of potentially non linear data.
3576 * Note 2: Fragment lists within fragments are not implemented
3577 * at the moment, state->root_skb could be replaced with
3578 * a stack for this purpose.
3580 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3581 struct skb_seq_state *st)
3583 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3586 if (unlikely(abs_offset >= st->upper_offset)) {
3587 if (st->frag_data) {
3588 kunmap_atomic(st->frag_data);
3589 st->frag_data = NULL;
3595 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3597 if (abs_offset < block_limit && !st->frag_data) {
3598 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3599 return block_limit - abs_offset;
3602 if (st->frag_idx == 0 && !st->frag_data)
3603 st->stepped_offset += skb_headlen(st->cur_skb);
3605 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3606 unsigned int pg_idx, pg_off, pg_sz;
3608 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3611 pg_off = skb_frag_off(frag);
3612 pg_sz = skb_frag_size(frag);
3614 if (skb_frag_must_loop(skb_frag_page(frag))) {
3615 pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
3616 pg_off = offset_in_page(pg_off + st->frag_off);
3617 pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
3618 PAGE_SIZE - pg_off);
3621 block_limit = pg_sz + st->stepped_offset;
3622 if (abs_offset < block_limit) {
3624 st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
3626 *data = (u8 *)st->frag_data + pg_off +
3627 (abs_offset - st->stepped_offset);
3629 return block_limit - abs_offset;
3632 if (st->frag_data) {
3633 kunmap_atomic(st->frag_data);
3634 st->frag_data = NULL;
3637 st->stepped_offset += pg_sz;
3638 st->frag_off += pg_sz;
3639 if (st->frag_off == skb_frag_size(frag)) {
3645 if (st->frag_data) {
3646 kunmap_atomic(st->frag_data);
3647 st->frag_data = NULL;
3650 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3651 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3654 } else if (st->cur_skb->next) {
3655 st->cur_skb = st->cur_skb->next;
3662 EXPORT_SYMBOL(skb_seq_read);
3665 * skb_abort_seq_read - Abort a sequential read of skb data
3666 * @st: state variable
3668 * Must be called if skb_seq_read() was not called until it
3671 void skb_abort_seq_read(struct skb_seq_state *st)
3674 kunmap_atomic(st->frag_data);
3676 EXPORT_SYMBOL(skb_abort_seq_read);
3678 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3680 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3681 struct ts_config *conf,
3682 struct ts_state *state)
3684 return skb_seq_read(offset, text, TS_SKB_CB(state));
3687 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3689 skb_abort_seq_read(TS_SKB_CB(state));
3693 * skb_find_text - Find a text pattern in skb data
3694 * @skb: the buffer to look in
3695 * @from: search offset
3697 * @config: textsearch configuration
3699 * Finds a pattern in the skb data according to the specified
3700 * textsearch configuration. Use textsearch_next() to retrieve
3701 * subsequent occurrences of the pattern. Returns the offset
3702 * to the first occurrence or UINT_MAX if no match was found.
3704 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3705 unsigned int to, struct ts_config *config)
3707 struct ts_state state;
3710 BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb));
3712 config->get_next_block = skb_ts_get_next_block;
3713 config->finish = skb_ts_finish;
3715 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3717 ret = textsearch_find(config, &state);
3718 return (ret <= to - from ? ret : UINT_MAX);
3720 EXPORT_SYMBOL(skb_find_text);
3722 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3723 int offset, size_t size)
3725 int i = skb_shinfo(skb)->nr_frags;
3727 if (skb_can_coalesce(skb, i, page, offset)) {
3728 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3729 } else if (i < MAX_SKB_FRAGS) {
3731 skb_fill_page_desc(skb, i, page, offset, size);
3738 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3741 * skb_pull_rcsum - pull skb and update receive checksum
3742 * @skb: buffer to update
3743 * @len: length of data pulled
3745 * This function performs an skb_pull on the packet and updates
3746 * the CHECKSUM_COMPLETE checksum. It should be used on
3747 * receive path processing instead of skb_pull unless you know
3748 * that the checksum difference is zero (e.g., a valid IP header)
3749 * or you are setting ip_summed to CHECKSUM_NONE.
3751 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3753 unsigned char *data = skb->data;
3755 BUG_ON(len > skb->len);
3756 __skb_pull(skb, len);
3757 skb_postpull_rcsum(skb, data, len);
3760 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3762 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3764 skb_frag_t head_frag;
3767 page = virt_to_head_page(frag_skb->head);
3768 __skb_frag_set_page(&head_frag, page);
3769 skb_frag_off_set(&head_frag, frag_skb->data -
3770 (unsigned char *)page_address(page));
3771 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3775 struct sk_buff *skb_segment_list(struct sk_buff *skb,
3776 netdev_features_t features,
3777 unsigned int offset)
3779 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
3780 unsigned int tnl_hlen = skb_tnl_header_len(skb);
3781 unsigned int delta_truesize = 0;
3782 unsigned int delta_len = 0;
3783 struct sk_buff *tail = NULL;
3784 struct sk_buff *nskb, *tmp;
3787 skb_push(skb, -skb_network_offset(skb) + offset);
3789 skb_shinfo(skb)->frag_list = NULL;
3793 list_skb = list_skb->next;
3796 if (skb_shared(nskb)) {
3797 tmp = skb_clone(nskb, GFP_ATOMIC);
3801 err = skb_unclone(nskb, GFP_ATOMIC);
3812 if (unlikely(err)) {
3813 nskb->next = list_skb;
3819 delta_len += nskb->len;
3820 delta_truesize += nskb->truesize;
3822 skb_push(nskb, -skb_network_offset(nskb) + offset);
3824 skb_release_head_state(nskb);
3825 __copy_skb_header(nskb, skb);
3827 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
3828 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
3829 nskb->data - tnl_hlen,
3832 if (skb_needs_linearize(nskb, features) &&
3833 __skb_linearize(nskb))
3838 skb->truesize = skb->truesize - delta_truesize;
3839 skb->data_len = skb->data_len - delta_len;
3840 skb->len = skb->len - delta_len;
3846 if (skb_needs_linearize(skb, features) &&
3847 __skb_linearize(skb))
3855 kfree_skb_list(skb->next);
3857 return ERR_PTR(-ENOMEM);
3859 EXPORT_SYMBOL_GPL(skb_segment_list);
3861 int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
3863 if (unlikely(p->len + skb->len >= 65536))
3866 if (NAPI_GRO_CB(p)->last == p)
3867 skb_shinfo(p)->frag_list = skb;
3869 NAPI_GRO_CB(p)->last->next = skb;
3871 skb_pull(skb, skb_gro_offset(skb));
3873 NAPI_GRO_CB(p)->last = skb;
3874 NAPI_GRO_CB(p)->count++;
3875 p->data_len += skb->len;
3876 p->truesize += skb->truesize;
3879 NAPI_GRO_CB(skb)->same_flow = 1;
3885 * skb_segment - Perform protocol segmentation on skb.
3886 * @head_skb: buffer to segment
3887 * @features: features for the output path (see dev->features)
3889 * This function performs segmentation on the given skb. It returns
3890 * a pointer to the first in a list of new skbs for the segments.
3891 * In case of error it returns ERR_PTR(err).
3893 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3894 netdev_features_t features)
3896 struct sk_buff *segs = NULL;
3897 struct sk_buff *tail = NULL;
3898 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3899 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3900 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3901 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3902 struct sk_buff *frag_skb = head_skb;
3903 unsigned int offset = doffset;
3904 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3905 unsigned int partial_segs = 0;
3906 unsigned int headroom;
3907 unsigned int len = head_skb->len;
3910 int nfrags = skb_shinfo(head_skb)->nr_frags;
3915 if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3916 (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3917 /* gso_size is untrusted, and we have a frag_list with a linear
3918 * non head_frag head.
3920 * (we assume checking the first list_skb member suffices;
3921 * i.e if either of the list_skb members have non head_frag
3922 * head, then the first one has too).
3924 * If head_skb's headlen does not fit requested gso_size, it
3925 * means that the frag_list members do NOT terminate on exact
3926 * gso_size boundaries. Hence we cannot perform skb_frag_t page
3927 * sharing. Therefore we must fallback to copying the frag_list
3928 * skbs; we do so by disabling SG.
3930 if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3931 features &= ~NETIF_F_SG;
3934 __skb_push(head_skb, doffset);
3935 proto = skb_network_protocol(head_skb, NULL);
3936 if (unlikely(!proto))
3937 return ERR_PTR(-EINVAL);
3939 sg = !!(features & NETIF_F_SG);
3940 csum = !!can_checksum_protocol(features, proto);
3942 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3943 if (!(features & NETIF_F_GSO_PARTIAL)) {
3944 struct sk_buff *iter;
3945 unsigned int frag_len;
3948 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3951 /* If we get here then all the required
3952 * GSO features except frag_list are supported.
3953 * Try to split the SKB to multiple GSO SKBs
3954 * with no frag_list.
3955 * Currently we can do that only when the buffers don't
3956 * have a linear part and all the buffers except
3957 * the last are of the same length.
3959 frag_len = list_skb->len;
3960 skb_walk_frags(head_skb, iter) {
3961 if (frag_len != iter->len && iter->next)
3963 if (skb_headlen(iter) && !iter->head_frag)
3969 if (len != frag_len)
3973 /* GSO partial only requires that we trim off any excess that
3974 * doesn't fit into an MSS sized block, so take care of that
3977 partial_segs = len / mss;
3978 if (partial_segs > 1)
3979 mss *= partial_segs;
3985 headroom = skb_headroom(head_skb);
3986 pos = skb_headlen(head_skb);
3989 struct sk_buff *nskb;
3990 skb_frag_t *nskb_frag;
3994 if (unlikely(mss == GSO_BY_FRAGS)) {
3995 len = list_skb->len;
3997 len = head_skb->len - offset;
4002 hsize = skb_headlen(head_skb) - offset;
4004 if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) &&
4005 (skb_headlen(list_skb) == len || sg)) {
4006 BUG_ON(skb_headlen(list_skb) > len);
4009 nfrags = skb_shinfo(list_skb)->nr_frags;
4010 frag = skb_shinfo(list_skb)->frags;
4011 frag_skb = list_skb;
4012 pos += skb_headlen(list_skb);
4014 while (pos < offset + len) {
4015 BUG_ON(i >= nfrags);
4017 size = skb_frag_size(frag);
4018 if (pos + size > offset + len)
4026 nskb = skb_clone(list_skb, GFP_ATOMIC);
4027 list_skb = list_skb->next;
4029 if (unlikely(!nskb))
4032 if (unlikely(pskb_trim(nskb, len))) {
4037 hsize = skb_end_offset(nskb);
4038 if (skb_cow_head(nskb, doffset + headroom)) {
4043 nskb->truesize += skb_end_offset(nskb) - hsize;
4044 skb_release_head_state(nskb);
4045 __skb_push(nskb, doffset);
4049 if (hsize > len || !sg)
4052 nskb = __alloc_skb(hsize + doffset + headroom,
4053 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
4056 if (unlikely(!nskb))
4059 skb_reserve(nskb, headroom);
4060 __skb_put(nskb, doffset);
4069 __copy_skb_header(nskb, head_skb);
4071 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
4072 skb_reset_mac_len(nskb);
4074 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
4075 nskb->data - tnl_hlen,
4076 doffset + tnl_hlen);
4078 if (nskb->len == len + doffset)
4079 goto perform_csum_check;
4083 if (!nskb->remcsum_offload)
4084 nskb->ip_summed = CHECKSUM_NONE;
4085 SKB_GSO_CB(nskb)->csum =
4086 skb_copy_and_csum_bits(head_skb, offset,
4090 SKB_GSO_CB(nskb)->csum_start =
4091 skb_headroom(nskb) + doffset;
4093 skb_copy_bits(head_skb, offset,
4100 nskb_frag = skb_shinfo(nskb)->frags;
4102 skb_copy_from_linear_data_offset(head_skb, offset,
4103 skb_put(nskb, hsize), hsize);
4105 skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags &
4108 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4109 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4112 while (pos < offset + len) {
4115 nfrags = skb_shinfo(list_skb)->nr_frags;
4116 frag = skb_shinfo(list_skb)->frags;
4117 frag_skb = list_skb;
4118 if (!skb_headlen(list_skb)) {
4121 BUG_ON(!list_skb->head_frag);
4123 /* to make room for head_frag. */
4127 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4128 skb_zerocopy_clone(nskb, frag_skb,
4132 list_skb = list_skb->next;
4135 if (unlikely(skb_shinfo(nskb)->nr_frags >=
4137 net_warn_ratelimited(
4138 "skb_segment: too many frags: %u %u\n",
4144 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4145 __skb_frag_ref(nskb_frag);
4146 size = skb_frag_size(nskb_frag);
4149 skb_frag_off_add(nskb_frag, offset - pos);
4150 skb_frag_size_sub(nskb_frag, offset - pos);
4153 skb_shinfo(nskb)->nr_frags++;
4155 if (pos + size <= offset + len) {
4160 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4168 nskb->data_len = len - hsize;
4169 nskb->len += nskb->data_len;
4170 nskb->truesize += nskb->data_len;
4174 if (skb_has_shared_frag(nskb) &&
4175 __skb_linearize(nskb))
4178 if (!nskb->remcsum_offload)
4179 nskb->ip_summed = CHECKSUM_NONE;
4180 SKB_GSO_CB(nskb)->csum =
4181 skb_checksum(nskb, doffset,
4182 nskb->len - doffset, 0);
4183 SKB_GSO_CB(nskb)->csum_start =
4184 skb_headroom(nskb) + doffset;
4186 } while ((offset += len) < head_skb->len);
4188 /* Some callers want to get the end of the list.
4189 * Put it in segs->prev to avoid walking the list.
4190 * (see validate_xmit_skb_list() for example)
4195 struct sk_buff *iter;
4196 int type = skb_shinfo(head_skb)->gso_type;
4197 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4199 /* Update type to add partial and then remove dodgy if set */
4200 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4201 type &= ~SKB_GSO_DODGY;
4203 /* Update GSO info and prepare to start updating headers on
4204 * our way back down the stack of protocols.
4206 for (iter = segs; iter; iter = iter->next) {
4207 skb_shinfo(iter)->gso_size = gso_size;
4208 skb_shinfo(iter)->gso_segs = partial_segs;
4209 skb_shinfo(iter)->gso_type = type;
4210 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4213 if (tail->len - doffset <= gso_size)
4214 skb_shinfo(tail)->gso_size = 0;
4215 else if (tail != segs)
4216 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4219 /* Following permits correct backpressure, for protocols
4220 * using skb_set_owner_w().
4221 * Idea is to tranfert ownership from head_skb to last segment.
4223 if (head_skb->destructor == sock_wfree) {
4224 swap(tail->truesize, head_skb->truesize);
4225 swap(tail->destructor, head_skb->destructor);
4226 swap(tail->sk, head_skb->sk);
4231 kfree_skb_list(segs);
4232 return ERR_PTR(err);
4234 EXPORT_SYMBOL_GPL(skb_segment);
4236 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4238 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4239 unsigned int offset = skb_gro_offset(skb);
4240 unsigned int headlen = skb_headlen(skb);
4241 unsigned int len = skb_gro_len(skb);
4242 unsigned int delta_truesize;
4245 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4248 lp = NAPI_GRO_CB(p)->last;
4249 pinfo = skb_shinfo(lp);
4251 if (headlen <= offset) {
4254 int i = skbinfo->nr_frags;
4255 int nr_frags = pinfo->nr_frags + i;
4257 if (nr_frags > MAX_SKB_FRAGS)
4261 pinfo->nr_frags = nr_frags;
4262 skbinfo->nr_frags = 0;
4264 frag = pinfo->frags + nr_frags;
4265 frag2 = skbinfo->frags + i;
4270 skb_frag_off_add(frag, offset);
4271 skb_frag_size_sub(frag, offset);
4273 /* all fragments truesize : remove (head size + sk_buff) */
4274 delta_truesize = skb->truesize -
4275 SKB_TRUESIZE(skb_end_offset(skb));
4277 skb->truesize -= skb->data_len;
4278 skb->len -= skb->data_len;
4281 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4283 } else if (skb->head_frag) {
4284 int nr_frags = pinfo->nr_frags;
4285 skb_frag_t *frag = pinfo->frags + nr_frags;
4286 struct page *page = virt_to_head_page(skb->head);
4287 unsigned int first_size = headlen - offset;
4288 unsigned int first_offset;
4290 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4293 first_offset = skb->data -
4294 (unsigned char *)page_address(page) +
4297 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4299 __skb_frag_set_page(frag, page);
4300 skb_frag_off_set(frag, first_offset);
4301 skb_frag_size_set(frag, first_size);
4303 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4304 /* We dont need to clear skbinfo->nr_frags here */
4306 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4307 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4312 delta_truesize = skb->truesize;
4313 if (offset > headlen) {
4314 unsigned int eat = offset - headlen;
4316 skb_frag_off_add(&skbinfo->frags[0], eat);
4317 skb_frag_size_sub(&skbinfo->frags[0], eat);
4318 skb->data_len -= eat;
4323 __skb_pull(skb, offset);
4325 if (NAPI_GRO_CB(p)->last == p)
4326 skb_shinfo(p)->frag_list = skb;
4328 NAPI_GRO_CB(p)->last->next = skb;
4329 NAPI_GRO_CB(p)->last = skb;
4330 __skb_header_release(skb);
4334 NAPI_GRO_CB(p)->count++;
4336 p->truesize += delta_truesize;
4339 lp->data_len += len;
4340 lp->truesize += delta_truesize;
4343 NAPI_GRO_CB(skb)->same_flow = 1;
4347 #ifdef CONFIG_SKB_EXTENSIONS
4348 #define SKB_EXT_ALIGN_VALUE 8
4349 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4351 static const u8 skb_ext_type_len[] = {
4352 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4353 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4356 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4358 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4359 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4361 #if IS_ENABLED(CONFIG_MPTCP)
4362 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4366 static __always_inline unsigned int skb_ext_total_length(void)
4368 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4369 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4370 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4373 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4375 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4376 skb_ext_type_len[TC_SKB_EXT] +
4378 #if IS_ENABLED(CONFIG_MPTCP)
4379 skb_ext_type_len[SKB_EXT_MPTCP] +
4384 static void skb_extensions_init(void)
4386 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4387 BUILD_BUG_ON(skb_ext_total_length() > 255);
4389 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4390 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4392 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4396 static void skb_extensions_init(void) {}
4399 void __init skb_init(void)
4401 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4402 sizeof(struct sk_buff),
4404 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4405 offsetof(struct sk_buff, cb),
4406 sizeof_field(struct sk_buff, cb),
4408 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4409 sizeof(struct sk_buff_fclones),
4411 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4413 skb_extensions_init();
4417 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4418 unsigned int recursion_level)
4420 int start = skb_headlen(skb);
4421 int i, copy = start - offset;
4422 struct sk_buff *frag_iter;
4425 if (unlikely(recursion_level >= 24))
4431 sg_set_buf(sg, skb->data + offset, copy);
4433 if ((len -= copy) == 0)
4438 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4441 WARN_ON(start > offset + len);
4443 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4444 if ((copy = end - offset) > 0) {
4445 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4446 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4451 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4452 skb_frag_off(frag) + offset - start);
4461 skb_walk_frags(skb, frag_iter) {
4464 WARN_ON(start > offset + len);
4466 end = start + frag_iter->len;
4467 if ((copy = end - offset) > 0) {
4468 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4473 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4474 copy, recursion_level + 1);
4475 if (unlikely(ret < 0))
4478 if ((len -= copy) == 0)
4489 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4490 * @skb: Socket buffer containing the buffers to be mapped
4491 * @sg: The scatter-gather list to map into
4492 * @offset: The offset into the buffer's contents to start mapping
4493 * @len: Length of buffer space to be mapped
4495 * Fill the specified scatter-gather list with mappings/pointers into a
4496 * region of the buffer space attached to a socket buffer. Returns either
4497 * the number of scatterlist items used, or -EMSGSIZE if the contents
4500 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4502 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4507 sg_mark_end(&sg[nsg - 1]);
4511 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4513 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4514 * sglist without mark the sg which contain last skb data as the end.
4515 * So the caller can mannipulate sg list as will when padding new data after
4516 * the first call without calling sg_unmark_end to expend sg list.
4518 * Scenario to use skb_to_sgvec_nomark:
4520 * 2. skb_to_sgvec_nomark(payload1)
4521 * 3. skb_to_sgvec_nomark(payload2)
4523 * This is equivalent to:
4525 * 2. skb_to_sgvec(payload1)
4527 * 4. skb_to_sgvec(payload2)
4529 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4530 * is more preferable.
4532 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4533 int offset, int len)
4535 return __skb_to_sgvec(skb, sg, offset, len, 0);
4537 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4542 * skb_cow_data - Check that a socket buffer's data buffers are writable
4543 * @skb: The socket buffer to check.
4544 * @tailbits: Amount of trailing space to be added
4545 * @trailer: Returned pointer to the skb where the @tailbits space begins
4547 * Make sure that the data buffers attached to a socket buffer are
4548 * writable. If they are not, private copies are made of the data buffers
4549 * and the socket buffer is set to use these instead.
4551 * If @tailbits is given, make sure that there is space to write @tailbits
4552 * bytes of data beyond current end of socket buffer. @trailer will be
4553 * set to point to the skb in which this space begins.
4555 * The number of scatterlist elements required to completely map the
4556 * COW'd and extended socket buffer will be returned.
4558 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4562 struct sk_buff *skb1, **skb_p;
4564 /* If skb is cloned or its head is paged, reallocate
4565 * head pulling out all the pages (pages are considered not writable
4566 * at the moment even if they are anonymous).
4568 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4569 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4572 /* Easy case. Most of packets will go this way. */
4573 if (!skb_has_frag_list(skb)) {
4574 /* A little of trouble, not enough of space for trailer.
4575 * This should not happen, when stack is tuned to generate
4576 * good frames. OK, on miss we reallocate and reserve even more
4577 * space, 128 bytes is fair. */
4579 if (skb_tailroom(skb) < tailbits &&
4580 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4588 /* Misery. We are in troubles, going to mincer fragments... */
4591 skb_p = &skb_shinfo(skb)->frag_list;
4594 while ((skb1 = *skb_p) != NULL) {
4597 /* The fragment is partially pulled by someone,
4598 * this can happen on input. Copy it and everything
4601 if (skb_shared(skb1))
4604 /* If the skb is the last, worry about trailer. */
4606 if (skb1->next == NULL && tailbits) {
4607 if (skb_shinfo(skb1)->nr_frags ||
4608 skb_has_frag_list(skb1) ||
4609 skb_tailroom(skb1) < tailbits)
4610 ntail = tailbits + 128;
4616 skb_shinfo(skb1)->nr_frags ||
4617 skb_has_frag_list(skb1)) {
4618 struct sk_buff *skb2;
4620 /* Fuck, we are miserable poor guys... */
4622 skb2 = skb_copy(skb1, GFP_ATOMIC);
4624 skb2 = skb_copy_expand(skb1,
4628 if (unlikely(skb2 == NULL))
4632 skb_set_owner_w(skb2, skb1->sk);
4634 /* Looking around. Are we still alive?
4635 * OK, link new skb, drop old one */
4637 skb2->next = skb1->next;
4644 skb_p = &skb1->next;
4649 EXPORT_SYMBOL_GPL(skb_cow_data);
4651 static void sock_rmem_free(struct sk_buff *skb)
4653 struct sock *sk = skb->sk;
4655 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4658 static void skb_set_err_queue(struct sk_buff *skb)
4660 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4661 * So, it is safe to (mis)use it to mark skbs on the error queue.
4663 skb->pkt_type = PACKET_OUTGOING;
4664 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4668 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4670 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4672 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4673 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4678 skb->destructor = sock_rmem_free;
4679 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4680 skb_set_err_queue(skb);
4682 /* before exiting rcu section, make sure dst is refcounted */
4685 skb_queue_tail(&sk->sk_error_queue, skb);
4686 if (!sock_flag(sk, SOCK_DEAD))
4687 sk->sk_error_report(sk);
4690 EXPORT_SYMBOL(sock_queue_err_skb);
4692 static bool is_icmp_err_skb(const struct sk_buff *skb)
4694 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4695 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4698 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4700 struct sk_buff_head *q = &sk->sk_error_queue;
4701 struct sk_buff *skb, *skb_next = NULL;
4702 bool icmp_next = false;
4703 unsigned long flags;
4705 spin_lock_irqsave(&q->lock, flags);
4706 skb = __skb_dequeue(q);
4707 if (skb && (skb_next = skb_peek(q))) {
4708 icmp_next = is_icmp_err_skb(skb_next);
4710 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
4712 spin_unlock_irqrestore(&q->lock, flags);
4714 if (is_icmp_err_skb(skb) && !icmp_next)
4718 sk->sk_error_report(sk);
4722 EXPORT_SYMBOL(sock_dequeue_err_skb);
4725 * skb_clone_sk - create clone of skb, and take reference to socket
4726 * @skb: the skb to clone
4728 * This function creates a clone of a buffer that holds a reference on
4729 * sk_refcnt. Buffers created via this function are meant to be
4730 * returned using sock_queue_err_skb, or free via kfree_skb.
4732 * When passing buffers allocated with this function to sock_queue_err_skb
4733 * it is necessary to wrap the call with sock_hold/sock_put in order to
4734 * prevent the socket from being released prior to being enqueued on
4735 * the sk_error_queue.
4737 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4739 struct sock *sk = skb->sk;
4740 struct sk_buff *clone;
4742 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4745 clone = skb_clone(skb, GFP_ATOMIC);
4752 clone->destructor = sock_efree;
4756 EXPORT_SYMBOL(skb_clone_sk);
4758 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4763 struct sock_exterr_skb *serr;
4766 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4768 serr = SKB_EXT_ERR(skb);
4769 memset(serr, 0, sizeof(*serr));
4770 serr->ee.ee_errno = ENOMSG;
4771 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4772 serr->ee.ee_info = tstype;
4773 serr->opt_stats = opt_stats;
4774 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4775 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4776 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4777 if (sk->sk_protocol == IPPROTO_TCP &&
4778 sk->sk_type == SOCK_STREAM)
4779 serr->ee.ee_data -= sk->sk_tskey;
4782 err = sock_queue_err_skb(sk, skb);
4788 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4792 if (likely(sysctl_tstamp_allow_data || tsonly))
4795 read_lock_bh(&sk->sk_callback_lock);
4796 ret = sk->sk_socket && sk->sk_socket->file &&
4797 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4798 read_unlock_bh(&sk->sk_callback_lock);
4802 void skb_complete_tx_timestamp(struct sk_buff *skb,
4803 struct skb_shared_hwtstamps *hwtstamps)
4805 struct sock *sk = skb->sk;
4807 if (!skb_may_tx_timestamp(sk, false))
4810 /* Take a reference to prevent skb_orphan() from freeing the socket,
4811 * but only if the socket refcount is not zero.
4813 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4814 *skb_hwtstamps(skb) = *hwtstamps;
4815 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4823 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4825 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4826 const struct sk_buff *ack_skb,
4827 struct skb_shared_hwtstamps *hwtstamps,
4828 struct sock *sk, int tstype)
4830 struct sk_buff *skb;
4831 bool tsonly, opt_stats = false;
4836 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4837 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4840 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4841 if (!skb_may_tx_timestamp(sk, tsonly))
4846 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4847 sk->sk_protocol == IPPROTO_TCP &&
4848 sk->sk_type == SOCK_STREAM) {
4849 skb = tcp_get_timestamping_opt_stats(sk, orig_skb,
4854 skb = alloc_skb(0, GFP_ATOMIC);
4856 skb = skb_clone(orig_skb, GFP_ATOMIC);
4862 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4864 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4868 *skb_hwtstamps(skb) = *hwtstamps;
4870 skb->tstamp = ktime_get_real();
4872 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4874 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4876 void skb_tstamp_tx(struct sk_buff *orig_skb,
4877 struct skb_shared_hwtstamps *hwtstamps)
4879 return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk,
4882 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4884 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4886 struct sock *sk = skb->sk;
4887 struct sock_exterr_skb *serr;
4890 skb->wifi_acked_valid = 1;
4891 skb->wifi_acked = acked;
4893 serr = SKB_EXT_ERR(skb);
4894 memset(serr, 0, sizeof(*serr));
4895 serr->ee.ee_errno = ENOMSG;
4896 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4898 /* Take a reference to prevent skb_orphan() from freeing the socket,
4899 * but only if the socket refcount is not zero.
4901 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4902 err = sock_queue_err_skb(sk, skb);
4908 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4911 * skb_partial_csum_set - set up and verify partial csum values for packet
4912 * @skb: the skb to set
4913 * @start: the number of bytes after skb->data to start checksumming.
4914 * @off: the offset from start to place the checksum.
4916 * For untrusted partially-checksummed packets, we need to make sure the values
4917 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4919 * This function checks and sets those values and skb->ip_summed: if this
4920 * returns false you should drop the packet.
4922 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4924 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4925 u32 csum_start = skb_headroom(skb) + (u32)start;
4927 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4928 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4929 start, off, skb_headroom(skb), skb_headlen(skb));
4932 skb->ip_summed = CHECKSUM_PARTIAL;
4933 skb->csum_start = csum_start;
4934 skb->csum_offset = off;
4935 skb_set_transport_header(skb, start);
4938 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4940 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4943 if (skb_headlen(skb) >= len)
4946 /* If we need to pullup then pullup to the max, so we
4947 * won't need to do it again.
4952 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4955 if (skb_headlen(skb) < len)
4961 #define MAX_TCP_HDR_LEN (15 * 4)
4963 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4964 typeof(IPPROTO_IP) proto,
4971 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4972 off + MAX_TCP_HDR_LEN);
4973 if (!err && !skb_partial_csum_set(skb, off,
4974 offsetof(struct tcphdr,
4977 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4980 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4981 off + sizeof(struct udphdr));
4982 if (!err && !skb_partial_csum_set(skb, off,
4983 offsetof(struct udphdr,
4986 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4989 return ERR_PTR(-EPROTO);
4992 /* This value should be large enough to cover a tagged ethernet header plus
4993 * maximally sized IP and TCP or UDP headers.
4995 #define MAX_IP_HDR_LEN 128
4997 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
5006 err = skb_maybe_pull_tail(skb,
5007 sizeof(struct iphdr),
5012 if (ip_is_fragment(ip_hdr(skb)))
5015 off = ip_hdrlen(skb);
5022 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
5024 return PTR_ERR(csum);
5027 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
5030 ip_hdr(skb)->protocol, 0);
5037 /* This value should be large enough to cover a tagged ethernet header plus
5038 * an IPv6 header, all options, and a maximal TCP or UDP header.
5040 #define MAX_IPV6_HDR_LEN 256
5042 #define OPT_HDR(type, skb, off) \
5043 (type *)(skb_network_header(skb) + (off))
5045 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
5058 off = sizeof(struct ipv6hdr);
5060 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
5064 nexthdr = ipv6_hdr(skb)->nexthdr;
5066 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
5067 while (off <= len && !done) {
5069 case IPPROTO_DSTOPTS:
5070 case IPPROTO_HOPOPTS:
5071 case IPPROTO_ROUTING: {
5072 struct ipv6_opt_hdr *hp;
5074 err = skb_maybe_pull_tail(skb,
5076 sizeof(struct ipv6_opt_hdr),
5081 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
5082 nexthdr = hp->nexthdr;
5083 off += ipv6_optlen(hp);
5087 struct ip_auth_hdr *hp;
5089 err = skb_maybe_pull_tail(skb,
5091 sizeof(struct ip_auth_hdr),
5096 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
5097 nexthdr = hp->nexthdr;
5098 off += ipv6_authlen(hp);
5101 case IPPROTO_FRAGMENT: {
5102 struct frag_hdr *hp;
5104 err = skb_maybe_pull_tail(skb,
5106 sizeof(struct frag_hdr),
5111 hp = OPT_HDR(struct frag_hdr, skb, off);
5113 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5116 nexthdr = hp->nexthdr;
5117 off += sizeof(struct frag_hdr);
5128 if (!done || fragment)
5131 csum = skb_checksum_setup_ip(skb, nexthdr, off);
5133 return PTR_ERR(csum);
5136 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5137 &ipv6_hdr(skb)->daddr,
5138 skb->len - off, nexthdr, 0);
5146 * skb_checksum_setup - set up partial checksum offset
5147 * @skb: the skb to set up
5148 * @recalculate: if true the pseudo-header checksum will be recalculated
5150 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5154 switch (skb->protocol) {
5155 case htons(ETH_P_IP):
5156 err = skb_checksum_setup_ipv4(skb, recalculate);
5159 case htons(ETH_P_IPV6):
5160 err = skb_checksum_setup_ipv6(skb, recalculate);
5170 EXPORT_SYMBOL(skb_checksum_setup);
5173 * skb_checksum_maybe_trim - maybe trims the given skb
5174 * @skb: the skb to check
5175 * @transport_len: the data length beyond the network header
5177 * Checks whether the given skb has data beyond the given transport length.
5178 * If so, returns a cloned skb trimmed to this transport length.
5179 * Otherwise returns the provided skb. Returns NULL in error cases
5180 * (e.g. transport_len exceeds skb length or out-of-memory).
5182 * Caller needs to set the skb transport header and free any returned skb if it
5183 * differs from the provided skb.
5185 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5186 unsigned int transport_len)
5188 struct sk_buff *skb_chk;
5189 unsigned int len = skb_transport_offset(skb) + transport_len;
5194 else if (skb->len == len)
5197 skb_chk = skb_clone(skb, GFP_ATOMIC);
5201 ret = pskb_trim_rcsum(skb_chk, len);
5211 * skb_checksum_trimmed - validate checksum of an skb
5212 * @skb: the skb to check
5213 * @transport_len: the data length beyond the network header
5214 * @skb_chkf: checksum function to use
5216 * Applies the given checksum function skb_chkf to the provided skb.
5217 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5219 * If the skb has data beyond the given transport length, then a
5220 * trimmed & cloned skb is checked and returned.
5222 * Caller needs to set the skb transport header and free any returned skb if it
5223 * differs from the provided skb.
5225 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5226 unsigned int transport_len,
5227 __sum16(*skb_chkf)(struct sk_buff *skb))
5229 struct sk_buff *skb_chk;
5230 unsigned int offset = skb_transport_offset(skb);
5233 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5237 if (!pskb_may_pull(skb_chk, offset))
5240 skb_pull_rcsum(skb_chk, offset);
5241 ret = skb_chkf(skb_chk);
5242 skb_push_rcsum(skb_chk, offset);
5250 if (skb_chk && skb_chk != skb)
5256 EXPORT_SYMBOL(skb_checksum_trimmed);
5258 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5260 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5263 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5265 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5268 skb_release_head_state(skb);
5269 kmem_cache_free(skbuff_head_cache, skb);
5274 EXPORT_SYMBOL(kfree_skb_partial);
5277 * skb_try_coalesce - try to merge skb to prior one
5279 * @from: buffer to add
5280 * @fragstolen: pointer to boolean
5281 * @delta_truesize: how much more was allocated than was requested
5283 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5284 bool *fragstolen, int *delta_truesize)
5286 struct skb_shared_info *to_shinfo, *from_shinfo;
5287 int i, delta, len = from->len;
5289 *fragstolen = false;
5294 if (len <= skb_tailroom(to)) {
5296 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5297 *delta_truesize = 0;
5301 to_shinfo = skb_shinfo(to);
5302 from_shinfo = skb_shinfo(from);
5303 if (to_shinfo->frag_list || from_shinfo->frag_list)
5305 if (skb_zcopy(to) || skb_zcopy(from))
5308 if (skb_headlen(from) != 0) {
5310 unsigned int offset;
5312 if (to_shinfo->nr_frags +
5313 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5316 if (skb_head_is_locked(from))
5319 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5321 page = virt_to_head_page(from->head);
5322 offset = from->data - (unsigned char *)page_address(page);
5324 skb_fill_page_desc(to, to_shinfo->nr_frags,
5325 page, offset, skb_headlen(from));
5328 if (to_shinfo->nr_frags +
5329 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5332 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5335 WARN_ON_ONCE(delta < len);
5337 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5339 from_shinfo->nr_frags * sizeof(skb_frag_t));
5340 to_shinfo->nr_frags += from_shinfo->nr_frags;
5342 if (!skb_cloned(from))
5343 from_shinfo->nr_frags = 0;
5345 /* if the skb is not cloned this does nothing
5346 * since we set nr_frags to 0.
5348 for (i = 0; i < from_shinfo->nr_frags; i++)
5349 __skb_frag_ref(&from_shinfo->frags[i]);
5351 to->truesize += delta;
5353 to->data_len += len;
5355 *delta_truesize = delta;
5358 EXPORT_SYMBOL(skb_try_coalesce);
5361 * skb_scrub_packet - scrub an skb
5363 * @skb: buffer to clean
5364 * @xnet: packet is crossing netns
5366 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5367 * into/from a tunnel. Some information have to be cleared during these
5369 * skb_scrub_packet can also be used to clean a skb before injecting it in
5370 * another namespace (@xnet == true). We have to clear all information in the
5371 * skb that could impact namespace isolation.
5373 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5375 skb->pkt_type = PACKET_HOST;
5381 nf_reset_trace(skb);
5383 #ifdef CONFIG_NET_SWITCHDEV
5384 skb->offload_fwd_mark = 0;
5385 skb->offload_l3_fwd_mark = 0;
5395 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5398 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5402 * skb_gso_transport_seglen is used to determine the real size of the
5403 * individual segments, including Layer4 headers (TCP/UDP).
5405 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5407 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5409 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5410 unsigned int thlen = 0;
5412 if (skb->encapsulation) {
5413 thlen = skb_inner_transport_header(skb) -
5414 skb_transport_header(skb);
5416 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5417 thlen += inner_tcp_hdrlen(skb);
5418 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5419 thlen = tcp_hdrlen(skb);
5420 } else if (unlikely(skb_is_gso_sctp(skb))) {
5421 thlen = sizeof(struct sctphdr);
5422 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5423 thlen = sizeof(struct udphdr);
5425 /* UFO sets gso_size to the size of the fragmentation
5426 * payload, i.e. the size of the L4 (UDP) header is already
5429 return thlen + shinfo->gso_size;
5433 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5437 * skb_gso_network_seglen is used to determine the real size of the
5438 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5440 * The MAC/L2 header is not accounted for.
5442 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5444 unsigned int hdr_len = skb_transport_header(skb) -
5445 skb_network_header(skb);
5447 return hdr_len + skb_gso_transport_seglen(skb);
5451 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5455 * skb_gso_mac_seglen is used to determine the real size of the
5456 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5457 * headers (TCP/UDP).
5459 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5461 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5463 return hdr_len + skb_gso_transport_seglen(skb);
5467 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5469 * There are a couple of instances where we have a GSO skb, and we
5470 * want to determine what size it would be after it is segmented.
5472 * We might want to check:
5473 * - L3+L4+payload size (e.g. IP forwarding)
5474 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5476 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5480 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5481 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5483 * @max_len: The maximum permissible length.
5485 * Returns true if the segmented length <= max length.
5487 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5488 unsigned int seg_len,
5489 unsigned int max_len) {
5490 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5491 const struct sk_buff *iter;
5493 if (shinfo->gso_size != GSO_BY_FRAGS)
5494 return seg_len <= max_len;
5496 /* Undo this so we can re-use header sizes */
5497 seg_len -= GSO_BY_FRAGS;
5499 skb_walk_frags(skb, iter) {
5500 if (seg_len + skb_headlen(iter) > max_len)
5508 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5511 * @mtu: MTU to validate against
5513 * skb_gso_validate_network_len validates if a given skb will fit a
5514 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5517 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5519 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5521 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5524 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5527 * @len: length to validate against
5529 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5530 * length once split, including L2, L3 and L4 headers and the payload.
5532 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5534 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5536 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5538 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5540 int mac_len, meta_len;
5543 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5548 mac_len = skb->data - skb_mac_header(skb);
5549 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5550 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5551 mac_len - VLAN_HLEN - ETH_TLEN);
5554 meta_len = skb_metadata_len(skb);
5556 meta = skb_metadata_end(skb) - meta_len;
5557 memmove(meta + VLAN_HLEN, meta, meta_len);
5560 skb->mac_header += VLAN_HLEN;
5564 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5566 struct vlan_hdr *vhdr;
5569 if (unlikely(skb_vlan_tag_present(skb))) {
5570 /* vlan_tci is already set-up so leave this for another time */
5574 skb = skb_share_check(skb, GFP_ATOMIC);
5577 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5578 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5581 vhdr = (struct vlan_hdr *)skb->data;
5582 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5583 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5585 skb_pull_rcsum(skb, VLAN_HLEN);
5586 vlan_set_encap_proto(skb, vhdr);
5588 skb = skb_reorder_vlan_header(skb);
5592 skb_reset_network_header(skb);
5593 if (!skb_transport_header_was_set(skb))
5594 skb_reset_transport_header(skb);
5595 skb_reset_mac_len(skb);
5603 EXPORT_SYMBOL(skb_vlan_untag);
5605 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5607 if (!pskb_may_pull(skb, write_len))
5610 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5613 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5615 EXPORT_SYMBOL(skb_ensure_writable);
5617 /* remove VLAN header from packet and update csum accordingly.
5618 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5620 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5622 struct vlan_hdr *vhdr;
5623 int offset = skb->data - skb_mac_header(skb);
5626 if (WARN_ONCE(offset,
5627 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5632 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5636 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5638 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5639 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5641 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5642 __skb_pull(skb, VLAN_HLEN);
5644 vlan_set_encap_proto(skb, vhdr);
5645 skb->mac_header += VLAN_HLEN;
5647 if (skb_network_offset(skb) < ETH_HLEN)
5648 skb_set_network_header(skb, ETH_HLEN);
5650 skb_reset_mac_len(skb);
5654 EXPORT_SYMBOL(__skb_vlan_pop);
5656 /* Pop a vlan tag either from hwaccel or from payload.
5657 * Expects skb->data at mac header.
5659 int skb_vlan_pop(struct sk_buff *skb)
5665 if (likely(skb_vlan_tag_present(skb))) {
5666 __vlan_hwaccel_clear_tag(skb);
5668 if (unlikely(!eth_type_vlan(skb->protocol)))
5671 err = __skb_vlan_pop(skb, &vlan_tci);
5675 /* move next vlan tag to hw accel tag */
5676 if (likely(!eth_type_vlan(skb->protocol)))
5679 vlan_proto = skb->protocol;
5680 err = __skb_vlan_pop(skb, &vlan_tci);
5684 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5687 EXPORT_SYMBOL(skb_vlan_pop);
5689 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5690 * Expects skb->data at mac header.
5692 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5694 if (skb_vlan_tag_present(skb)) {
5695 int offset = skb->data - skb_mac_header(skb);
5698 if (WARN_ONCE(offset,
5699 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5704 err = __vlan_insert_tag(skb, skb->vlan_proto,
5705 skb_vlan_tag_get(skb));
5709 skb->protocol = skb->vlan_proto;
5710 skb->mac_len += VLAN_HLEN;
5712 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5714 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5717 EXPORT_SYMBOL(skb_vlan_push);
5720 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
5722 * @skb: Socket buffer to modify
5724 * Drop the Ethernet header of @skb.
5726 * Expects that skb->data points to the mac header and that no VLAN tags are
5729 * Returns 0 on success, -errno otherwise.
5731 int skb_eth_pop(struct sk_buff *skb)
5733 if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
5734 skb_network_offset(skb) < ETH_HLEN)
5737 skb_pull_rcsum(skb, ETH_HLEN);
5738 skb_reset_mac_header(skb);
5739 skb_reset_mac_len(skb);
5743 EXPORT_SYMBOL(skb_eth_pop);
5746 * skb_eth_push() - Add a new Ethernet header at the head of a packet
5748 * @skb: Socket buffer to modify
5749 * @dst: Destination MAC address of the new header
5750 * @src: Source MAC address of the new header
5752 * Prepend @skb with a new Ethernet header.
5754 * Expects that skb->data points to the mac header, which must be empty.
5756 * Returns 0 on success, -errno otherwise.
5758 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
5759 const unsigned char *src)
5764 if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
5767 err = skb_cow_head(skb, sizeof(*eth));
5771 skb_push(skb, sizeof(*eth));
5772 skb_reset_mac_header(skb);
5773 skb_reset_mac_len(skb);
5776 ether_addr_copy(eth->h_dest, dst);
5777 ether_addr_copy(eth->h_source, src);
5778 eth->h_proto = skb->protocol;
5780 skb_postpush_rcsum(skb, eth, sizeof(*eth));
5784 EXPORT_SYMBOL(skb_eth_push);
5786 /* Update the ethertype of hdr and the skb csum value if required. */
5787 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5790 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5791 __be16 diff[] = { ~hdr->h_proto, ethertype };
5793 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5796 hdr->h_proto = ethertype;
5800 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
5804 * @mpls_lse: MPLS label stack entry to push
5805 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5806 * @mac_len: length of the MAC header
5807 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
5810 * Expects skb->data at mac header.
5812 * Returns 0 on success, -errno otherwise.
5814 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5815 int mac_len, bool ethernet)
5817 struct mpls_shim_hdr *lse;
5820 if (unlikely(!eth_p_mpls(mpls_proto)))
5823 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5824 if (skb->encapsulation)
5827 err = skb_cow_head(skb, MPLS_HLEN);
5831 if (!skb->inner_protocol) {
5832 skb_set_inner_network_header(skb, skb_network_offset(skb));
5833 skb_set_inner_protocol(skb, skb->protocol);
5836 skb_push(skb, MPLS_HLEN);
5837 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5839 skb_reset_mac_header(skb);
5840 skb_set_network_header(skb, mac_len);
5841 skb_reset_mac_len(skb);
5843 lse = mpls_hdr(skb);
5844 lse->label_stack_entry = mpls_lse;
5845 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5847 if (ethernet && mac_len >= ETH_HLEN)
5848 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5849 skb->protocol = mpls_proto;
5853 EXPORT_SYMBOL_GPL(skb_mpls_push);
5856 * skb_mpls_pop() - pop the outermost MPLS header
5859 * @next_proto: ethertype of header after popped MPLS header
5860 * @mac_len: length of the MAC header
5861 * @ethernet: flag to indicate if the packet is ethernet
5863 * Expects skb->data at mac header.
5865 * Returns 0 on success, -errno otherwise.
5867 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5872 if (unlikely(!eth_p_mpls(skb->protocol)))
5875 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5879 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5880 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5883 __skb_pull(skb, MPLS_HLEN);
5884 skb_reset_mac_header(skb);
5885 skb_set_network_header(skb, mac_len);
5887 if (ethernet && mac_len >= ETH_HLEN) {
5890 /* use mpls_hdr() to get ethertype to account for VLANs. */
5891 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5892 skb_mod_eth_type(skb, hdr, next_proto);
5894 skb->protocol = next_proto;
5898 EXPORT_SYMBOL_GPL(skb_mpls_pop);
5901 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5904 * @mpls_lse: new MPLS label stack entry to update to
5906 * Expects skb->data at mac header.
5908 * Returns 0 on success, -errno otherwise.
5910 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5914 if (unlikely(!eth_p_mpls(skb->protocol)))
5917 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5921 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5922 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5924 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5927 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5931 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5934 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5938 * Expects skb->data at mac header.
5940 * Returns 0 on success, -errno otherwise.
5942 int skb_mpls_dec_ttl(struct sk_buff *skb)
5947 if (unlikely(!eth_p_mpls(skb->protocol)))
5950 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
5953 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5954 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5958 lse &= ~MPLS_LS_TTL_MASK;
5959 lse |= ttl << MPLS_LS_TTL_SHIFT;
5961 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5963 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5966 * alloc_skb_with_frags - allocate skb with page frags
5968 * @header_len: size of linear part
5969 * @data_len: needed length in frags
5970 * @max_page_order: max page order desired.
5971 * @errcode: pointer to error code if any
5972 * @gfp_mask: allocation mask
5974 * This can be used to allocate a paged skb, given a maximal order for frags.
5976 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5977 unsigned long data_len,
5982 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5983 unsigned long chunk;
5984 struct sk_buff *skb;
5988 *errcode = -EMSGSIZE;
5989 /* Note this test could be relaxed, if we succeed to allocate
5990 * high order pages...
5992 if (npages > MAX_SKB_FRAGS)
5995 *errcode = -ENOBUFS;
5996 skb = alloc_skb(header_len, gfp_mask);
6000 skb->truesize += npages << PAGE_SHIFT;
6002 for (i = 0; npages > 0; i++) {
6003 int order = max_page_order;
6006 if (npages >= 1 << order) {
6007 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
6013 /* Do not retry other high order allocations */
6019 page = alloc_page(gfp_mask);
6023 chunk = min_t(unsigned long, data_len,
6024 PAGE_SIZE << order);
6025 skb_fill_page_desc(skb, i, page, 0, chunk);
6027 npages -= 1 << order;
6035 EXPORT_SYMBOL(alloc_skb_with_frags);
6037 /* carve out the first off bytes from skb when off < headlen */
6038 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
6039 const int headlen, gfp_t gfp_mask)
6042 int size = skb_end_offset(skb);
6043 int new_hlen = headlen - off;
6046 size = SKB_DATA_ALIGN(size);
6048 if (skb_pfmemalloc(skb))
6049 gfp_mask |= __GFP_MEMALLOC;
6050 data = kmalloc_reserve(size +
6051 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6052 gfp_mask, NUMA_NO_NODE, NULL);
6056 size = SKB_WITH_OVERHEAD(ksize(data));
6058 /* Copy real data, and all frags */
6059 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
6062 memcpy((struct skb_shared_info *)(data + size),
6064 offsetof(struct skb_shared_info,
6065 frags[skb_shinfo(skb)->nr_frags]));
6066 if (skb_cloned(skb)) {
6067 /* drop the old head gracefully */
6068 if (skb_orphan_frags(skb, gfp_mask)) {
6072 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
6073 skb_frag_ref(skb, i);
6074 if (skb_has_frag_list(skb))
6075 skb_clone_fraglist(skb);
6076 skb_release_data(skb);
6078 /* we can reuse existing recount- all we did was
6087 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6090 skb->end = skb->head + size;
6092 skb_set_tail_pointer(skb, skb_headlen(skb));
6093 skb_headers_offset_update(skb, 0);
6097 atomic_set(&skb_shinfo(skb)->dataref, 1);
6102 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6104 /* carve out the first eat bytes from skb's frag_list. May recurse into
6107 static int pskb_carve_frag_list(struct sk_buff *skb,
6108 struct skb_shared_info *shinfo, int eat,
6111 struct sk_buff *list = shinfo->frag_list;
6112 struct sk_buff *clone = NULL;
6113 struct sk_buff *insp = NULL;
6117 pr_err("Not enough bytes to eat. Want %d\n", eat);
6120 if (list->len <= eat) {
6121 /* Eaten as whole. */
6126 /* Eaten partially. */
6127 if (skb_shared(list)) {
6128 clone = skb_clone(list, gfp_mask);
6134 /* This may be pulled without problems. */
6137 if (pskb_carve(list, eat, gfp_mask) < 0) {
6145 /* Free pulled out fragments. */
6146 while ((list = shinfo->frag_list) != insp) {
6147 shinfo->frag_list = list->next;
6150 /* And insert new clone at head. */
6153 shinfo->frag_list = clone;
6158 /* carve off first len bytes from skb. Split line (off) is in the
6159 * non-linear part of skb
6161 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6162 int pos, gfp_t gfp_mask)
6165 int size = skb_end_offset(skb);
6167 const int nfrags = skb_shinfo(skb)->nr_frags;
6168 struct skb_shared_info *shinfo;
6170 size = SKB_DATA_ALIGN(size);
6172 if (skb_pfmemalloc(skb))
6173 gfp_mask |= __GFP_MEMALLOC;
6174 data = kmalloc_reserve(size +
6175 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6176 gfp_mask, NUMA_NO_NODE, NULL);
6180 size = SKB_WITH_OVERHEAD(ksize(data));
6182 memcpy((struct skb_shared_info *)(data + size),
6183 skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6184 if (skb_orphan_frags(skb, gfp_mask)) {
6188 shinfo = (struct skb_shared_info *)(data + size);
6189 for (i = 0; i < nfrags; i++) {
6190 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6192 if (pos + fsize > off) {
6193 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6197 * We have two variants in this case:
6198 * 1. Move all the frag to the second
6199 * part, if it is possible. F.e.
6200 * this approach is mandatory for TUX,
6201 * where splitting is expensive.
6202 * 2. Split is accurately. We make this.
6204 skb_frag_off_add(&shinfo->frags[0], off - pos);
6205 skb_frag_size_sub(&shinfo->frags[0], off - pos);
6207 skb_frag_ref(skb, i);
6212 shinfo->nr_frags = k;
6213 if (skb_has_frag_list(skb))
6214 skb_clone_fraglist(skb);
6216 /* split line is in frag list */
6217 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6218 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6219 if (skb_has_frag_list(skb))
6220 kfree_skb_list(skb_shinfo(skb)->frag_list);
6224 skb_release_data(skb);
6229 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6232 skb->end = skb->head + size;
6234 skb_reset_tail_pointer(skb);
6235 skb_headers_offset_update(skb, 0);
6240 skb->data_len = skb->len;
6241 atomic_set(&skb_shinfo(skb)->dataref, 1);
6245 /* remove len bytes from the beginning of the skb */
6246 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6248 int headlen = skb_headlen(skb);
6251 return pskb_carve_inside_header(skb, len, headlen, gfp);
6253 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6256 /* Extract to_copy bytes starting at off from skb, and return this in
6259 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6260 int to_copy, gfp_t gfp)
6262 struct sk_buff *clone = skb_clone(skb, gfp);
6267 if (pskb_carve(clone, off, gfp) < 0 ||
6268 pskb_trim(clone, to_copy)) {
6274 EXPORT_SYMBOL(pskb_extract);
6277 * skb_condense - try to get rid of fragments/frag_list if possible
6280 * Can be used to save memory before skb is added to a busy queue.
6281 * If packet has bytes in frags and enough tail room in skb->head,
6282 * pull all of them, so that we can free the frags right now and adjust
6285 * We do not reallocate skb->head thus can not fail.
6286 * Caller must re-evaluate skb->truesize if needed.
6288 void skb_condense(struct sk_buff *skb)
6290 if (skb->data_len) {
6291 if (skb->data_len > skb->end - skb->tail ||
6295 /* Nice, we can free page frag(s) right now */
6296 __pskb_pull_tail(skb, skb->data_len);
6298 /* At this point, skb->truesize might be over estimated,
6299 * because skb had a fragment, and fragments do not tell
6301 * When we pulled its content into skb->head, fragment
6302 * was freed, but __pskb_pull_tail() could not possibly
6303 * adjust skb->truesize, not knowing the frag truesize.
6305 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6308 #ifdef CONFIG_SKB_EXTENSIONS
6309 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6311 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6315 * __skb_ext_alloc - allocate a new skb extensions storage
6317 * @flags: See kmalloc().
6319 * Returns the newly allocated pointer. The pointer can later attached to a
6320 * skb via __skb_ext_set().
6321 * Note: caller must handle the skb_ext as an opaque data.
6323 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6325 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6328 memset(new->offset, 0, sizeof(new->offset));
6329 refcount_set(&new->refcnt, 1);
6335 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6336 unsigned int old_active)
6338 struct skb_ext *new;
6340 if (refcount_read(&old->refcnt) == 1)
6343 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6347 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6348 refcount_set(&new->refcnt, 1);
6351 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6352 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6355 for (i = 0; i < sp->len; i++)
6356 xfrm_state_hold(sp->xvec[i]);
6364 * __skb_ext_set - attach the specified extension storage to this skb
6367 * @ext: extension storage previously allocated via __skb_ext_alloc()
6369 * Existing extensions, if any, are cleared.
6371 * Returns the pointer to the extension.
6373 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6374 struct skb_ext *ext)
6376 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6379 newlen = newoff + skb_ext_type_len[id];
6380 ext->chunks = newlen;
6381 ext->offset[id] = newoff;
6382 skb->extensions = ext;
6383 skb->active_extensions = 1 << id;
6384 return skb_ext_get_ptr(ext, id);
6388 * skb_ext_add - allocate space for given extension, COW if needed
6390 * @id: extension to allocate space for
6392 * Allocates enough space for the given extension.
6393 * If the extension is already present, a pointer to that extension
6396 * If the skb was cloned, COW applies and the returned memory can be
6397 * modified without changing the extension space of clones buffers.
6399 * Returns pointer to the extension or NULL on allocation failure.
6401 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6403 struct skb_ext *new, *old = NULL;
6404 unsigned int newlen, newoff;
6406 if (skb->active_extensions) {
6407 old = skb->extensions;
6409 new = skb_ext_maybe_cow(old, skb->active_extensions);
6413 if (__skb_ext_exist(new, id))
6416 newoff = new->chunks;
6418 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6420 new = __skb_ext_alloc(GFP_ATOMIC);
6425 newlen = newoff + skb_ext_type_len[id];
6426 new->chunks = newlen;
6427 new->offset[id] = newoff;
6429 skb->extensions = new;
6430 skb->active_extensions |= 1 << id;
6431 return skb_ext_get_ptr(new, id);
6433 EXPORT_SYMBOL(skb_ext_add);
6436 static void skb_ext_put_sp(struct sec_path *sp)
6440 for (i = 0; i < sp->len; i++)
6441 xfrm_state_put(sp->xvec[i]);
6445 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6447 struct skb_ext *ext = skb->extensions;
6449 skb->active_extensions &= ~(1 << id);
6450 if (skb->active_extensions == 0) {
6451 skb->extensions = NULL;
6454 } else if (id == SKB_EXT_SEC_PATH &&
6455 refcount_read(&ext->refcnt) == 1) {
6456 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6463 EXPORT_SYMBOL(__skb_ext_del);
6465 void __skb_ext_put(struct skb_ext *ext)
6467 /* If this is last clone, nothing can increment
6468 * it after check passes. Avoids one atomic op.
6470 if (refcount_read(&ext->refcnt) == 1)
6473 if (!refcount_dec_and_test(&ext->refcnt))
6477 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6478 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6481 kmem_cache_free(skbuff_ext_cache, ext);
6483 EXPORT_SYMBOL(__skb_ext_put);
6484 #endif /* CONFIG_SKB_EXTENSIONS */