1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Routines having to do with the 'struct sk_buff' memory handlers.
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Florian La Roche <rzsfl@rz.uni-sb.de>
9 * Alan Cox : Fixed the worst of the load
11 * Dave Platt : Interrupt stacking fix.
12 * Richard Kooijman : Timestamp fixes.
13 * Alan Cox : Changed buffer format.
14 * Alan Cox : destructor hook for AF_UNIX etc.
15 * Linus Torvalds : Better skb_clone.
16 * Alan Cox : Added skb_copy.
17 * Alan Cox : Added all the changed routines Linus
18 * only put in the headers
19 * Ray VanTassle : Fixed --skb->lock in free
20 * Alan Cox : skb_copy copy arp field
21 * Andi Kleen : slabified it.
22 * Robert Olsson : Removed skb_head_pool
25 * The __skb_ routines should be called with interrupts
26 * disabled, or you better be *real* sure that the operation is atomic
27 * with respect to whatever list is being frobbed (e.g. via lock_sock()
28 * or via disabling bottom half handlers, etc).
32 * The functions in this file will not compile correctly with gcc 2.4.x
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/kernel.h>
41 #include <linux/interrupt.h>
43 #include <linux/inet.h>
44 #include <linux/slab.h>
45 #include <linux/tcp.h>
46 #include <linux/udp.h>
47 #include <linux/sctp.h>
48 #include <linux/netdevice.h>
49 #ifdef CONFIG_NET_CLS_ACT
50 #include <net/pkt_sched.h>
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/splice.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
59 #include <linux/errqueue.h>
60 #include <linux/prefetch.h>
61 #include <linux/if_vlan.h>
62 #include <linux/mpls.h>
64 #include <net/protocol.h>
67 #include <net/checksum.h>
68 #include <net/ip6_checksum.h>
71 #include <net/mptcp.h>
73 #include <linux/uaccess.h>
74 #include <trace/events/skb.h>
75 #include <linux/highmem.h>
76 #include <linux/capability.h>
77 #include <linux/user_namespace.h>
78 #include <linux/indirect_call_wrapper.h>
82 struct kmem_cache *skbuff_head_cache __ro_after_init;
83 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
84 #ifdef CONFIG_SKB_EXTENSIONS
85 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
87 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
88 EXPORT_SYMBOL(sysctl_max_skb_frags);
91 * skb_panic - private function for out-of-line support
95 * @msg: skb_over_panic or skb_under_panic
97 * Out-of-line support for skb_put() and skb_push().
98 * Called via the wrapper skb_over_panic() or skb_under_panic().
99 * Keep out of line to prevent kernel bloat.
100 * __builtin_return_address is not used because it is not always reliable.
102 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
105 pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
106 msg, addr, skb->len, sz, skb->head, skb->data,
107 (unsigned long)skb->tail, (unsigned long)skb->end,
108 skb->dev ? skb->dev->name : "<NULL>");
112 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
114 skb_panic(skb, sz, addr, __func__);
117 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
119 skb_panic(skb, sz, addr, __func__);
122 #define NAPI_SKB_CACHE_SIZE 64
123 #define NAPI_SKB_CACHE_BULK 16
124 #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2)
126 struct napi_alloc_cache {
127 struct page_frag_cache page;
128 unsigned int skb_count;
129 void *skb_cache[NAPI_SKB_CACHE_SIZE];
132 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
133 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
135 static void *__alloc_frag_align(unsigned int fragsz, gfp_t gfp_mask,
136 unsigned int align_mask)
138 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
140 return page_frag_alloc_align(&nc->page, fragsz, gfp_mask, align_mask);
143 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
145 fragsz = SKB_DATA_ALIGN(fragsz);
147 return __alloc_frag_align(fragsz, GFP_ATOMIC, align_mask);
149 EXPORT_SYMBOL(__napi_alloc_frag_align);
151 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
153 struct page_frag_cache *nc;
156 fragsz = SKB_DATA_ALIGN(fragsz);
157 if (in_irq() || irqs_disabled()) {
158 nc = this_cpu_ptr(&netdev_alloc_cache);
159 data = page_frag_alloc_align(nc, fragsz, GFP_ATOMIC, align_mask);
162 data = __alloc_frag_align(fragsz, GFP_ATOMIC, align_mask);
167 EXPORT_SYMBOL(__netdev_alloc_frag_align);
169 static struct sk_buff *napi_skb_cache_get(void)
171 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
174 if (unlikely(!nc->skb_count))
175 nc->skb_count = kmem_cache_alloc_bulk(skbuff_head_cache,
179 if (unlikely(!nc->skb_count))
182 skb = nc->skb_cache[--nc->skb_count];
183 kasan_unpoison_object_data(skbuff_head_cache, skb);
188 /* Caller must provide SKB that is memset cleared */
189 static void __build_skb_around(struct sk_buff *skb, void *data,
190 unsigned int frag_size)
192 struct skb_shared_info *shinfo;
193 unsigned int size = frag_size ? : ksize(data);
195 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
197 /* Assumes caller memset cleared SKB */
198 skb->truesize = SKB_TRUESIZE(size);
199 refcount_set(&skb->users, 1);
202 skb_reset_tail_pointer(skb);
203 skb->end = skb->tail + size;
204 skb->mac_header = (typeof(skb->mac_header))~0U;
205 skb->transport_header = (typeof(skb->transport_header))~0U;
207 /* make sure we initialize shinfo sequentially */
208 shinfo = skb_shinfo(skb);
209 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
210 atomic_set(&shinfo->dataref, 1);
212 skb_set_kcov_handle(skb, kcov_common_handle());
216 * __build_skb - build a network buffer
217 * @data: data buffer provided by caller
218 * @frag_size: size of data, or 0 if head was kmalloced
220 * Allocate a new &sk_buff. Caller provides space holding head and
221 * skb_shared_info. @data must have been allocated by kmalloc() only if
222 * @frag_size is 0, otherwise data should come from the page allocator
224 * The return is the new skb buffer.
225 * On a failure the return is %NULL, and @data is not freed.
227 * Before IO, driver allocates only data buffer where NIC put incoming frame
228 * Driver should add room at head (NET_SKB_PAD) and
229 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
230 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
231 * before giving packet to stack.
232 * RX rings only contains data buffers, not full skbs.
234 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
238 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
242 memset(skb, 0, offsetof(struct sk_buff, tail));
243 __build_skb_around(skb, data, frag_size);
248 /* build_skb() is wrapper over __build_skb(), that specifically
249 * takes care of skb->head and skb->pfmemalloc
250 * This means that if @frag_size is not zero, then @data must be backed
251 * by a page fragment, not kmalloc() or vmalloc()
253 struct sk_buff *build_skb(void *data, unsigned int frag_size)
255 struct sk_buff *skb = __build_skb(data, frag_size);
257 if (skb && frag_size) {
259 if (page_is_pfmemalloc(virt_to_head_page(data)))
264 EXPORT_SYMBOL(build_skb);
267 * build_skb_around - build a network buffer around provided skb
268 * @skb: sk_buff provide by caller, must be memset cleared
269 * @data: data buffer provided by caller
270 * @frag_size: size of data, or 0 if head was kmalloced
272 struct sk_buff *build_skb_around(struct sk_buff *skb,
273 void *data, unsigned int frag_size)
278 __build_skb_around(skb, data, frag_size);
282 if (page_is_pfmemalloc(virt_to_head_page(data)))
287 EXPORT_SYMBOL(build_skb_around);
290 * __napi_build_skb - build a network buffer
291 * @data: data buffer provided by caller
292 * @frag_size: size of data, or 0 if head was kmalloced
294 * Version of __build_skb() that uses NAPI percpu caches to obtain
295 * skbuff_head instead of inplace allocation.
297 * Returns a new &sk_buff on success, %NULL on allocation failure.
299 static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size)
303 skb = napi_skb_cache_get();
307 memset(skb, 0, offsetof(struct sk_buff, tail));
308 __build_skb_around(skb, data, frag_size);
314 * napi_build_skb - build a network buffer
315 * @data: data buffer provided by caller
316 * @frag_size: size of data, or 0 if head was kmalloced
318 * Version of __napi_build_skb() that takes care of skb->head_frag
319 * and skb->pfmemalloc when the data is a page or page fragment.
321 * Returns a new &sk_buff on success, %NULL on allocation failure.
323 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size)
325 struct sk_buff *skb = __napi_build_skb(data, frag_size);
327 if (likely(skb) && frag_size) {
329 skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
334 EXPORT_SYMBOL(napi_build_skb);
337 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
338 * the caller if emergency pfmemalloc reserves are being used. If it is and
339 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
340 * may be used. Otherwise, the packet data may be discarded until enough
343 static void *kmalloc_reserve(size_t size, gfp_t flags, int node,
347 bool ret_pfmemalloc = false;
350 * Try a regular allocation, when that fails and we're not entitled
351 * to the reserves, fail.
353 obj = kmalloc_node_track_caller(size,
354 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
356 if (obj || !(gfp_pfmemalloc_allowed(flags)))
359 /* Try again but now we are using pfmemalloc reserves */
360 ret_pfmemalloc = true;
361 obj = kmalloc_node_track_caller(size, flags, node);
365 *pfmemalloc = ret_pfmemalloc;
370 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
371 * 'private' fields and also do memory statistics to find all the
377 * __alloc_skb - allocate a network buffer
378 * @size: size to allocate
379 * @gfp_mask: allocation mask
380 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
381 * instead of head cache and allocate a cloned (child) skb.
382 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
383 * allocations in case the data is required for writeback
384 * @node: numa node to allocate memory on
386 * Allocate a new &sk_buff. The returned buffer has no headroom and a
387 * tail room of at least size bytes. The object has a reference count
388 * of one. The return is the buffer. On a failure the return is %NULL.
390 * Buffers may only be allocated from interrupts using a @gfp_mask of
393 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
396 struct kmem_cache *cache;
401 cache = (flags & SKB_ALLOC_FCLONE)
402 ? skbuff_fclone_cache : skbuff_head_cache;
404 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
405 gfp_mask |= __GFP_MEMALLOC;
408 if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI &&
409 likely(node == NUMA_NO_NODE || node == numa_mem_id()))
410 skb = napi_skb_cache_get();
412 skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node);
417 /* We do our best to align skb_shared_info on a separate cache
418 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
419 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
420 * Both skb->head and skb_shared_info are cache line aligned.
422 size = SKB_DATA_ALIGN(size);
423 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
424 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
427 /* kmalloc(size) might give us more room than requested.
428 * Put skb_shared_info exactly at the end of allocated zone,
429 * to allow max possible filling before reallocation.
431 size = SKB_WITH_OVERHEAD(ksize(data));
432 prefetchw(data + size);
435 * Only clear those fields we need to clear, not those that we will
436 * actually initialise below. Hence, don't put any more fields after
437 * the tail pointer in struct sk_buff!
439 memset(skb, 0, offsetof(struct sk_buff, tail));
440 __build_skb_around(skb, data, 0);
441 skb->pfmemalloc = pfmemalloc;
443 if (flags & SKB_ALLOC_FCLONE) {
444 struct sk_buff_fclones *fclones;
446 fclones = container_of(skb, struct sk_buff_fclones, skb1);
448 skb->fclone = SKB_FCLONE_ORIG;
449 refcount_set(&fclones->fclone_ref, 1);
451 fclones->skb2.fclone = SKB_FCLONE_CLONE;
457 kmem_cache_free(cache, skb);
460 EXPORT_SYMBOL(__alloc_skb);
463 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
464 * @dev: network device to receive on
465 * @len: length to allocate
466 * @gfp_mask: get_free_pages mask, passed to alloc_skb
468 * Allocate a new &sk_buff and assign it a usage count of one. The
469 * buffer has NET_SKB_PAD headroom built in. Users should allocate
470 * the headroom they think they need without accounting for the
471 * built in space. The built in space is used for optimisations.
473 * %NULL is returned if there is no free memory.
475 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
478 struct page_frag_cache *nc;
485 /* If requested length is either too small or too big,
486 * we use kmalloc() for skb->head allocation.
488 if (len <= SKB_WITH_OVERHEAD(1024) ||
489 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
490 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
491 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
497 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
498 len = SKB_DATA_ALIGN(len);
500 if (sk_memalloc_socks())
501 gfp_mask |= __GFP_MEMALLOC;
503 if (in_irq() || irqs_disabled()) {
504 nc = this_cpu_ptr(&netdev_alloc_cache);
505 data = page_frag_alloc(nc, len, gfp_mask);
506 pfmemalloc = nc->pfmemalloc;
509 nc = this_cpu_ptr(&napi_alloc_cache.page);
510 data = page_frag_alloc(nc, len, gfp_mask);
511 pfmemalloc = nc->pfmemalloc;
518 skb = __build_skb(data, len);
519 if (unlikely(!skb)) {
529 skb_reserve(skb, NET_SKB_PAD);
535 EXPORT_SYMBOL(__netdev_alloc_skb);
538 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
539 * @napi: napi instance this buffer was allocated for
540 * @len: length to allocate
541 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
543 * Allocate a new sk_buff for use in NAPI receive. This buffer will
544 * attempt to allocate the head from a special reserved region used
545 * only for NAPI Rx allocation. By doing this we can save several
546 * CPU cycles by avoiding having to disable and re-enable IRQs.
548 * %NULL is returned if there is no free memory.
550 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
553 struct napi_alloc_cache *nc;
557 len += NET_SKB_PAD + NET_IP_ALIGN;
559 /* If requested length is either too small or too big,
560 * we use kmalloc() for skb->head allocation.
562 if (len <= SKB_WITH_OVERHEAD(1024) ||
563 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
564 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
565 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI,
572 nc = this_cpu_ptr(&napi_alloc_cache);
573 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
574 len = SKB_DATA_ALIGN(len);
576 if (sk_memalloc_socks())
577 gfp_mask |= __GFP_MEMALLOC;
579 data = page_frag_alloc(&nc->page, len, gfp_mask);
583 skb = __napi_build_skb(data, len);
584 if (unlikely(!skb)) {
589 if (nc->page.pfmemalloc)
594 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
595 skb->dev = napi->dev;
600 EXPORT_SYMBOL(__napi_alloc_skb);
602 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
603 int size, unsigned int truesize)
605 skb_fill_page_desc(skb, i, page, off, size);
607 skb->data_len += size;
608 skb->truesize += truesize;
610 EXPORT_SYMBOL(skb_add_rx_frag);
612 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
613 unsigned int truesize)
615 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
617 skb_frag_size_add(frag, size);
619 skb->data_len += size;
620 skb->truesize += truesize;
622 EXPORT_SYMBOL(skb_coalesce_rx_frag);
624 static void skb_drop_list(struct sk_buff **listp)
626 kfree_skb_list(*listp);
630 static inline void skb_drop_fraglist(struct sk_buff *skb)
632 skb_drop_list(&skb_shinfo(skb)->frag_list);
635 static void skb_clone_fraglist(struct sk_buff *skb)
637 struct sk_buff *list;
639 skb_walk_frags(skb, list)
643 static void skb_free_head(struct sk_buff *skb)
645 unsigned char *head = skb->head;
653 static void skb_release_data(struct sk_buff *skb)
655 struct skb_shared_info *shinfo = skb_shinfo(skb);
659 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
663 skb_zcopy_clear(skb, true);
665 for (i = 0; i < shinfo->nr_frags; i++)
666 __skb_frag_unref(&shinfo->frags[i]);
668 if (shinfo->frag_list)
669 kfree_skb_list(shinfo->frag_list);
675 * Free an skbuff by memory without cleaning the state.
677 static void kfree_skbmem(struct sk_buff *skb)
679 struct sk_buff_fclones *fclones;
681 switch (skb->fclone) {
682 case SKB_FCLONE_UNAVAILABLE:
683 kmem_cache_free(skbuff_head_cache, skb);
686 case SKB_FCLONE_ORIG:
687 fclones = container_of(skb, struct sk_buff_fclones, skb1);
689 /* We usually free the clone (TX completion) before original skb
690 * This test would have no chance to be true for the clone,
691 * while here, branch prediction will be good.
693 if (refcount_read(&fclones->fclone_ref) == 1)
697 default: /* SKB_FCLONE_CLONE */
698 fclones = container_of(skb, struct sk_buff_fclones, skb2);
701 if (!refcount_dec_and_test(&fclones->fclone_ref))
704 kmem_cache_free(skbuff_fclone_cache, fclones);
707 void skb_release_head_state(struct sk_buff *skb)
710 if (skb->destructor) {
712 skb->destructor(skb);
714 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
715 nf_conntrack_put(skb_nfct(skb));
720 /* Free everything but the sk_buff shell. */
721 static void skb_release_all(struct sk_buff *skb)
723 skb_release_head_state(skb);
724 if (likely(skb->head))
725 skb_release_data(skb);
729 * __kfree_skb - private function
732 * Free an sk_buff. Release anything attached to the buffer.
733 * Clean the state. This is an internal helper function. Users should
734 * always call kfree_skb
737 void __kfree_skb(struct sk_buff *skb)
739 skb_release_all(skb);
742 EXPORT_SYMBOL(__kfree_skb);
745 * kfree_skb - free an sk_buff
746 * @skb: buffer to free
748 * Drop a reference to the buffer and free it if the usage count has
751 void kfree_skb(struct sk_buff *skb)
756 trace_kfree_skb(skb, __builtin_return_address(0));
759 EXPORT_SYMBOL(kfree_skb);
761 void kfree_skb_list(struct sk_buff *segs)
764 struct sk_buff *next = segs->next;
770 EXPORT_SYMBOL(kfree_skb_list);
772 /* Dump skb information and contents.
774 * Must only be called from net_ratelimit()-ed paths.
776 * Dumps whole packets if full_pkt, only headers otherwise.
778 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
780 struct skb_shared_info *sh = skb_shinfo(skb);
781 struct net_device *dev = skb->dev;
782 struct sock *sk = skb->sk;
783 struct sk_buff *list_skb;
784 bool has_mac, has_trans;
785 int headroom, tailroom;
791 len = min_t(int, skb->len, MAX_HEADER + 128);
793 headroom = skb_headroom(skb);
794 tailroom = skb_tailroom(skb);
796 has_mac = skb_mac_header_was_set(skb);
797 has_trans = skb_transport_header_was_set(skb);
799 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
800 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
801 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
802 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
803 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
804 level, skb->len, headroom, skb_headlen(skb), tailroom,
805 has_mac ? skb->mac_header : -1,
806 has_mac ? skb_mac_header_len(skb) : -1,
808 has_trans ? skb_network_header_len(skb) : -1,
809 has_trans ? skb->transport_header : -1,
810 sh->tx_flags, sh->nr_frags,
811 sh->gso_size, sh->gso_type, sh->gso_segs,
812 skb->csum, skb->ip_summed, skb->csum_complete_sw,
813 skb->csum_valid, skb->csum_level,
814 skb->hash, skb->sw_hash, skb->l4_hash,
815 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
818 printk("%sdev name=%s feat=0x%pNF\n",
819 level, dev->name, &dev->features);
821 printk("%ssk family=%hu type=%u proto=%u\n",
822 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
824 if (full_pkt && headroom)
825 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
826 16, 1, skb->head, headroom, false);
828 seg_len = min_t(int, skb_headlen(skb), len);
830 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
831 16, 1, skb->data, seg_len, false);
834 if (full_pkt && tailroom)
835 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
836 16, 1, skb_tail_pointer(skb), tailroom, false);
838 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
839 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
840 u32 p_off, p_len, copied;
844 skb_frag_foreach_page(frag, skb_frag_off(frag),
845 skb_frag_size(frag), p, p_off, p_len,
847 seg_len = min_t(int, p_len, len);
848 vaddr = kmap_atomic(p);
849 print_hex_dump(level, "skb frag: ",
851 16, 1, vaddr + p_off, seg_len, false);
852 kunmap_atomic(vaddr);
859 if (full_pkt && skb_has_frag_list(skb)) {
860 printk("skb fraglist:\n");
861 skb_walk_frags(skb, list_skb)
862 skb_dump(level, list_skb, true);
865 EXPORT_SYMBOL(skb_dump);
868 * skb_tx_error - report an sk_buff xmit error
869 * @skb: buffer that triggered an error
871 * Report xmit error if a device callback is tracking this skb.
872 * skb must be freed afterwards.
874 void skb_tx_error(struct sk_buff *skb)
876 skb_zcopy_clear(skb, true);
878 EXPORT_SYMBOL(skb_tx_error);
880 #ifdef CONFIG_TRACEPOINTS
882 * consume_skb - free an skbuff
883 * @skb: buffer to free
885 * Drop a ref to the buffer and free it if the usage count has hit zero
886 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
887 * is being dropped after a failure and notes that
889 void consume_skb(struct sk_buff *skb)
894 trace_consume_skb(skb);
897 EXPORT_SYMBOL(consume_skb);
901 * __consume_stateless_skb - free an skbuff, assuming it is stateless
902 * @skb: buffer to free
904 * Alike consume_skb(), but this variant assumes that this is the last
905 * skb reference and all the head states have been already dropped
907 void __consume_stateless_skb(struct sk_buff *skb)
909 trace_consume_skb(skb);
910 skb_release_data(skb);
914 static void napi_skb_cache_put(struct sk_buff *skb)
916 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
919 kasan_poison_object_data(skbuff_head_cache, skb);
920 nc->skb_cache[nc->skb_count++] = skb;
922 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
923 for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++)
924 kasan_unpoison_object_data(skbuff_head_cache,
927 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF,
928 nc->skb_cache + NAPI_SKB_CACHE_HALF);
929 nc->skb_count = NAPI_SKB_CACHE_HALF;
933 void __kfree_skb_defer(struct sk_buff *skb)
935 skb_release_all(skb);
936 napi_skb_cache_put(skb);
939 void napi_skb_free_stolen_head(struct sk_buff *skb)
943 napi_skb_cache_put(skb);
946 void napi_consume_skb(struct sk_buff *skb, int budget)
948 /* Zero budget indicate non-NAPI context called us, like netpoll */
949 if (unlikely(!budget)) {
950 dev_consume_skb_any(skb);
954 lockdep_assert_in_softirq();
959 /* if reaching here SKB is ready to free */
960 trace_consume_skb(skb);
962 /* if SKB is a clone, don't handle this case */
963 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
968 skb_release_all(skb);
969 napi_skb_cache_put(skb);
971 EXPORT_SYMBOL(napi_consume_skb);
973 /* Make sure a field is enclosed inside headers_start/headers_end section */
974 #define CHECK_SKB_FIELD(field) \
975 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
976 offsetof(struct sk_buff, headers_start)); \
977 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
978 offsetof(struct sk_buff, headers_end)); \
980 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
982 new->tstamp = old->tstamp;
983 /* We do not copy old->sk */
985 memcpy(new->cb, old->cb, sizeof(old->cb));
986 skb_dst_copy(new, old);
987 __skb_ext_copy(new, old);
988 __nf_copy(new, old, false);
990 /* Note : this field could be in headers_start/headers_end section
991 * It is not yet because we do not want to have a 16 bit hole
993 new->queue_mapping = old->queue_mapping;
995 memcpy(&new->headers_start, &old->headers_start,
996 offsetof(struct sk_buff, headers_end) -
997 offsetof(struct sk_buff, headers_start));
998 CHECK_SKB_FIELD(protocol);
999 CHECK_SKB_FIELD(csum);
1000 CHECK_SKB_FIELD(hash);
1001 CHECK_SKB_FIELD(priority);
1002 CHECK_SKB_FIELD(skb_iif);
1003 CHECK_SKB_FIELD(vlan_proto);
1004 CHECK_SKB_FIELD(vlan_tci);
1005 CHECK_SKB_FIELD(transport_header);
1006 CHECK_SKB_FIELD(network_header);
1007 CHECK_SKB_FIELD(mac_header);
1008 CHECK_SKB_FIELD(inner_protocol);
1009 CHECK_SKB_FIELD(inner_transport_header);
1010 CHECK_SKB_FIELD(inner_network_header);
1011 CHECK_SKB_FIELD(inner_mac_header);
1012 CHECK_SKB_FIELD(mark);
1013 #ifdef CONFIG_NETWORK_SECMARK
1014 CHECK_SKB_FIELD(secmark);
1016 #ifdef CONFIG_NET_RX_BUSY_POLL
1017 CHECK_SKB_FIELD(napi_id);
1020 CHECK_SKB_FIELD(sender_cpu);
1022 #ifdef CONFIG_NET_SCHED
1023 CHECK_SKB_FIELD(tc_index);
1029 * You should not add any new code to this function. Add it to
1030 * __copy_skb_header above instead.
1032 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
1034 #define C(x) n->x = skb->x
1036 n->next = n->prev = NULL;
1038 __copy_skb_header(n, skb);
1043 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1048 n->destructor = NULL;
1055 refcount_set(&n->users, 1);
1057 atomic_inc(&(skb_shinfo(skb)->dataref));
1065 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1066 * @first: first sk_buff of the msg
1068 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1072 n = alloc_skb(0, GFP_ATOMIC);
1076 n->len = first->len;
1077 n->data_len = first->len;
1078 n->truesize = first->truesize;
1080 skb_shinfo(n)->frag_list = first;
1082 __copy_skb_header(n, first);
1083 n->destructor = NULL;
1087 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1090 * skb_morph - morph one skb into another
1091 * @dst: the skb to receive the contents
1092 * @src: the skb to supply the contents
1094 * This is identical to skb_clone except that the target skb is
1095 * supplied by the user.
1097 * The target skb is returned upon exit.
1099 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1101 skb_release_all(dst);
1102 return __skb_clone(dst, src);
1104 EXPORT_SYMBOL_GPL(skb_morph);
1106 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1108 unsigned long max_pg, num_pg, new_pg, old_pg;
1109 struct user_struct *user;
1111 if (capable(CAP_IPC_LOCK) || !size)
1114 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1115 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1116 user = mmp->user ? : current_user();
1119 old_pg = atomic_long_read(&user->locked_vm);
1120 new_pg = old_pg + num_pg;
1121 if (new_pg > max_pg)
1123 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1127 mmp->user = get_uid(user);
1128 mmp->num_pg = num_pg;
1130 mmp->num_pg += num_pg;
1135 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1137 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1140 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1141 free_uid(mmp->user);
1144 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1146 struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size)
1148 struct ubuf_info *uarg;
1149 struct sk_buff *skb;
1151 WARN_ON_ONCE(!in_task());
1153 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1157 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1158 uarg = (void *)skb->cb;
1159 uarg->mmp.user = NULL;
1161 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1166 uarg->callback = msg_zerocopy_callback;
1167 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1169 uarg->bytelen = size;
1171 uarg->flags = SKBFL_ZEROCOPY_FRAG;
1172 refcount_set(&uarg->refcnt, 1);
1177 EXPORT_SYMBOL_GPL(msg_zerocopy_alloc);
1179 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1181 return container_of((void *)uarg, struct sk_buff, cb);
1184 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1185 struct ubuf_info *uarg)
1188 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1191 /* realloc only when socket is locked (TCP, UDP cork),
1192 * so uarg->len and sk_zckey access is serialized
1194 if (!sock_owned_by_user(sk)) {
1199 bytelen = uarg->bytelen + size;
1200 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1201 /* TCP can create new skb to attach new uarg */
1202 if (sk->sk_type == SOCK_STREAM)
1207 next = (u32)atomic_read(&sk->sk_zckey);
1208 if ((u32)(uarg->id + uarg->len) == next) {
1209 if (mm_account_pinned_pages(&uarg->mmp, size))
1212 uarg->bytelen = bytelen;
1213 atomic_set(&sk->sk_zckey, ++next);
1215 /* no extra ref when appending to datagram (MSG_MORE) */
1216 if (sk->sk_type == SOCK_STREAM)
1217 net_zcopy_get(uarg);
1224 return msg_zerocopy_alloc(sk, size);
1226 EXPORT_SYMBOL_GPL(msg_zerocopy_realloc);
1228 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1230 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1234 old_lo = serr->ee.ee_info;
1235 old_hi = serr->ee.ee_data;
1236 sum_len = old_hi - old_lo + 1ULL + len;
1238 if (sum_len >= (1ULL << 32))
1241 if (lo != old_hi + 1)
1244 serr->ee.ee_data += len;
1248 static void __msg_zerocopy_callback(struct ubuf_info *uarg)
1250 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1251 struct sock_exterr_skb *serr;
1252 struct sock *sk = skb->sk;
1253 struct sk_buff_head *q;
1254 unsigned long flags;
1258 mm_unaccount_pinned_pages(&uarg->mmp);
1260 /* if !len, there was only 1 call, and it was aborted
1261 * so do not queue a completion notification
1263 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1268 hi = uarg->id + len - 1;
1270 serr = SKB_EXT_ERR(skb);
1271 memset(serr, 0, sizeof(*serr));
1272 serr->ee.ee_errno = 0;
1273 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1274 serr->ee.ee_data = hi;
1275 serr->ee.ee_info = lo;
1276 if (!uarg->zerocopy)
1277 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1279 q = &sk->sk_error_queue;
1280 spin_lock_irqsave(&q->lock, flags);
1281 tail = skb_peek_tail(q);
1282 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1283 !skb_zerocopy_notify_extend(tail, lo, len)) {
1284 __skb_queue_tail(q, skb);
1287 spin_unlock_irqrestore(&q->lock, flags);
1289 sk->sk_error_report(sk);
1296 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1299 uarg->zerocopy = uarg->zerocopy & success;
1301 if (refcount_dec_and_test(&uarg->refcnt))
1302 __msg_zerocopy_callback(uarg);
1304 EXPORT_SYMBOL_GPL(msg_zerocopy_callback);
1306 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1308 struct sock *sk = skb_from_uarg(uarg)->sk;
1310 atomic_dec(&sk->sk_zckey);
1314 msg_zerocopy_callback(NULL, uarg, true);
1316 EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort);
1318 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1320 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1322 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1324 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1325 struct msghdr *msg, int len,
1326 struct ubuf_info *uarg)
1328 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1329 struct iov_iter orig_iter = msg->msg_iter;
1330 int err, orig_len = skb->len;
1332 /* An skb can only point to one uarg. This edge case happens when
1333 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1335 if (orig_uarg && uarg != orig_uarg)
1338 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1339 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1340 struct sock *save_sk = skb->sk;
1342 /* Streams do not free skb on error. Reset to prev state. */
1343 msg->msg_iter = orig_iter;
1345 ___pskb_trim(skb, orig_len);
1350 skb_zcopy_set(skb, uarg, NULL);
1351 return skb->len - orig_len;
1353 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1355 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1358 if (skb_zcopy(orig)) {
1359 if (skb_zcopy(nskb)) {
1360 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1365 if (skb_uarg(nskb) == skb_uarg(orig))
1367 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1370 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1376 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1377 * @skb: the skb to modify
1378 * @gfp_mask: allocation priority
1380 * This must be called on skb with SKBFL_ZEROCOPY_ENABLE.
1381 * It will copy all frags into kernel and drop the reference
1382 * to userspace pages.
1384 * If this function is called from an interrupt gfp_mask() must be
1387 * Returns 0 on success or a negative error code on failure
1388 * to allocate kernel memory to copy to.
1390 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1392 int num_frags = skb_shinfo(skb)->nr_frags;
1393 struct page *page, *head = NULL;
1397 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1403 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1404 for (i = 0; i < new_frags; i++) {
1405 page = alloc_page(gfp_mask);
1408 struct page *next = (struct page *)page_private(head);
1414 set_page_private(page, (unsigned long)head);
1420 for (i = 0; i < num_frags; i++) {
1421 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1422 u32 p_off, p_len, copied;
1426 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1427 p, p_off, p_len, copied) {
1429 vaddr = kmap_atomic(p);
1431 while (done < p_len) {
1432 if (d_off == PAGE_SIZE) {
1434 page = (struct page *)page_private(page);
1436 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1437 memcpy(page_address(page) + d_off,
1438 vaddr + p_off + done, copy);
1442 kunmap_atomic(vaddr);
1446 /* skb frags release userspace buffers */
1447 for (i = 0; i < num_frags; i++)
1448 skb_frag_unref(skb, i);
1450 /* skb frags point to kernel buffers */
1451 for (i = 0; i < new_frags - 1; i++) {
1452 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1453 head = (struct page *)page_private(head);
1455 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1456 skb_shinfo(skb)->nr_frags = new_frags;
1459 skb_zcopy_clear(skb, false);
1462 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1465 * skb_clone - duplicate an sk_buff
1466 * @skb: buffer to clone
1467 * @gfp_mask: allocation priority
1469 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1470 * copies share the same packet data but not structure. The new
1471 * buffer has a reference count of 1. If the allocation fails the
1472 * function returns %NULL otherwise the new buffer is returned.
1474 * If this function is called from an interrupt gfp_mask() must be
1478 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1480 struct sk_buff_fclones *fclones = container_of(skb,
1481 struct sk_buff_fclones,
1485 if (skb_orphan_frags(skb, gfp_mask))
1488 if (skb->fclone == SKB_FCLONE_ORIG &&
1489 refcount_read(&fclones->fclone_ref) == 1) {
1491 refcount_set(&fclones->fclone_ref, 2);
1493 if (skb_pfmemalloc(skb))
1494 gfp_mask |= __GFP_MEMALLOC;
1496 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1500 n->fclone = SKB_FCLONE_UNAVAILABLE;
1503 return __skb_clone(n, skb);
1505 EXPORT_SYMBOL(skb_clone);
1507 void skb_headers_offset_update(struct sk_buff *skb, int off)
1509 /* Only adjust this if it actually is csum_start rather than csum */
1510 if (skb->ip_summed == CHECKSUM_PARTIAL)
1511 skb->csum_start += off;
1512 /* {transport,network,mac}_header and tail are relative to skb->head */
1513 skb->transport_header += off;
1514 skb->network_header += off;
1515 if (skb_mac_header_was_set(skb))
1516 skb->mac_header += off;
1517 skb->inner_transport_header += off;
1518 skb->inner_network_header += off;
1519 skb->inner_mac_header += off;
1521 EXPORT_SYMBOL(skb_headers_offset_update);
1523 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1525 __copy_skb_header(new, old);
1527 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1528 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1529 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1531 EXPORT_SYMBOL(skb_copy_header);
1533 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1535 if (skb_pfmemalloc(skb))
1536 return SKB_ALLOC_RX;
1541 * skb_copy - create private copy of an sk_buff
1542 * @skb: buffer to copy
1543 * @gfp_mask: allocation priority
1545 * Make a copy of both an &sk_buff and its data. This is used when the
1546 * caller wishes to modify the data and needs a private copy of the
1547 * data to alter. Returns %NULL on failure or the pointer to the buffer
1548 * on success. The returned buffer has a reference count of 1.
1550 * As by-product this function converts non-linear &sk_buff to linear
1551 * one, so that &sk_buff becomes completely private and caller is allowed
1552 * to modify all the data of returned buffer. This means that this
1553 * function is not recommended for use in circumstances when only
1554 * header is going to be modified. Use pskb_copy() instead.
1557 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1559 int headerlen = skb_headroom(skb);
1560 unsigned int size = skb_end_offset(skb) + skb->data_len;
1561 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1562 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1567 /* Set the data pointer */
1568 skb_reserve(n, headerlen);
1569 /* Set the tail pointer and length */
1570 skb_put(n, skb->len);
1572 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1574 skb_copy_header(n, skb);
1577 EXPORT_SYMBOL(skb_copy);
1580 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1581 * @skb: buffer to copy
1582 * @headroom: headroom of new skb
1583 * @gfp_mask: allocation priority
1584 * @fclone: if true allocate the copy of the skb from the fclone
1585 * cache instead of the head cache; it is recommended to set this
1586 * to true for the cases where the copy will likely be cloned
1588 * Make a copy of both an &sk_buff and part of its data, located
1589 * in header. Fragmented data remain shared. This is used when
1590 * the caller wishes to modify only header of &sk_buff and needs
1591 * private copy of the header to alter. Returns %NULL on failure
1592 * or the pointer to the buffer on success.
1593 * The returned buffer has a reference count of 1.
1596 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1597 gfp_t gfp_mask, bool fclone)
1599 unsigned int size = skb_headlen(skb) + headroom;
1600 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1601 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1606 /* Set the data pointer */
1607 skb_reserve(n, headroom);
1608 /* Set the tail pointer and length */
1609 skb_put(n, skb_headlen(skb));
1610 /* Copy the bytes */
1611 skb_copy_from_linear_data(skb, n->data, n->len);
1613 n->truesize += skb->data_len;
1614 n->data_len = skb->data_len;
1617 if (skb_shinfo(skb)->nr_frags) {
1620 if (skb_orphan_frags(skb, gfp_mask) ||
1621 skb_zerocopy_clone(n, skb, gfp_mask)) {
1626 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1627 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1628 skb_frag_ref(skb, i);
1630 skb_shinfo(n)->nr_frags = i;
1633 if (skb_has_frag_list(skb)) {
1634 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1635 skb_clone_fraglist(n);
1638 skb_copy_header(n, skb);
1642 EXPORT_SYMBOL(__pskb_copy_fclone);
1645 * pskb_expand_head - reallocate header of &sk_buff
1646 * @skb: buffer to reallocate
1647 * @nhead: room to add at head
1648 * @ntail: room to add at tail
1649 * @gfp_mask: allocation priority
1651 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1652 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1653 * reference count of 1. Returns zero in the case of success or error,
1654 * if expansion failed. In the last case, &sk_buff is not changed.
1656 * All the pointers pointing into skb header may change and must be
1657 * reloaded after call to this function.
1660 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1663 int i, osize = skb_end_offset(skb);
1664 int size = osize + nhead + ntail;
1670 BUG_ON(skb_shared(skb));
1672 size = SKB_DATA_ALIGN(size);
1674 if (skb_pfmemalloc(skb))
1675 gfp_mask |= __GFP_MEMALLOC;
1676 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1677 gfp_mask, NUMA_NO_NODE, NULL);
1680 size = SKB_WITH_OVERHEAD(ksize(data));
1682 /* Copy only real data... and, alas, header. This should be
1683 * optimized for the cases when header is void.
1685 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1687 memcpy((struct skb_shared_info *)(data + size),
1689 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1692 * if shinfo is shared we must drop the old head gracefully, but if it
1693 * is not we can just drop the old head and let the existing refcount
1694 * be since all we did is relocate the values
1696 if (skb_cloned(skb)) {
1697 if (skb_orphan_frags(skb, gfp_mask))
1700 refcount_inc(&skb_uarg(skb)->refcnt);
1701 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1702 skb_frag_ref(skb, i);
1704 if (skb_has_frag_list(skb))
1705 skb_clone_fraglist(skb);
1707 skb_release_data(skb);
1711 off = (data + nhead) - skb->head;
1716 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1720 skb->end = skb->head + size;
1723 skb_headers_offset_update(skb, nhead);
1727 atomic_set(&skb_shinfo(skb)->dataref, 1);
1729 skb_metadata_clear(skb);
1731 /* It is not generally safe to change skb->truesize.
1732 * For the moment, we really care of rx path, or
1733 * when skb is orphaned (not attached to a socket).
1735 if (!skb->sk || skb->destructor == sock_edemux)
1736 skb->truesize += size - osize;
1745 EXPORT_SYMBOL(pskb_expand_head);
1747 /* Make private copy of skb with writable head and some headroom */
1749 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1751 struct sk_buff *skb2;
1752 int delta = headroom - skb_headroom(skb);
1755 skb2 = pskb_copy(skb, GFP_ATOMIC);
1757 skb2 = skb_clone(skb, GFP_ATOMIC);
1758 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1766 EXPORT_SYMBOL(skb_realloc_headroom);
1769 * skb_copy_expand - copy and expand sk_buff
1770 * @skb: buffer to copy
1771 * @newheadroom: new free bytes at head
1772 * @newtailroom: new free bytes at tail
1773 * @gfp_mask: allocation priority
1775 * Make a copy of both an &sk_buff and its data and while doing so
1776 * allocate additional space.
1778 * This is used when the caller wishes to modify the data and needs a
1779 * private copy of the data to alter as well as more space for new fields.
1780 * Returns %NULL on failure or the pointer to the buffer
1781 * on success. The returned buffer has a reference count of 1.
1783 * You must pass %GFP_ATOMIC as the allocation priority if this function
1784 * is called from an interrupt.
1786 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1787 int newheadroom, int newtailroom,
1791 * Allocate the copy buffer
1793 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1794 gfp_mask, skb_alloc_rx_flag(skb),
1796 int oldheadroom = skb_headroom(skb);
1797 int head_copy_len, head_copy_off;
1802 skb_reserve(n, newheadroom);
1804 /* Set the tail pointer and length */
1805 skb_put(n, skb->len);
1807 head_copy_len = oldheadroom;
1809 if (newheadroom <= head_copy_len)
1810 head_copy_len = newheadroom;
1812 head_copy_off = newheadroom - head_copy_len;
1814 /* Copy the linear header and data. */
1815 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1816 skb->len + head_copy_len));
1818 skb_copy_header(n, skb);
1820 skb_headers_offset_update(n, newheadroom - oldheadroom);
1824 EXPORT_SYMBOL(skb_copy_expand);
1827 * __skb_pad - zero pad the tail of an skb
1828 * @skb: buffer to pad
1829 * @pad: space to pad
1830 * @free_on_error: free buffer on error
1832 * Ensure that a buffer is followed by a padding area that is zero
1833 * filled. Used by network drivers which may DMA or transfer data
1834 * beyond the buffer end onto the wire.
1836 * May return error in out of memory cases. The skb is freed on error
1837 * if @free_on_error is true.
1840 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1845 /* If the skbuff is non linear tailroom is always zero.. */
1846 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1847 memset(skb->data+skb->len, 0, pad);
1851 ntail = skb->data_len + pad - (skb->end - skb->tail);
1852 if (likely(skb_cloned(skb) || ntail > 0)) {
1853 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1858 /* FIXME: The use of this function with non-linear skb's really needs
1861 err = skb_linearize(skb);
1865 memset(skb->data + skb->len, 0, pad);
1873 EXPORT_SYMBOL(__skb_pad);
1876 * pskb_put - add data to the tail of a potentially fragmented buffer
1877 * @skb: start of the buffer to use
1878 * @tail: tail fragment of the buffer to use
1879 * @len: amount of data to add
1881 * This function extends the used data area of the potentially
1882 * fragmented buffer. @tail must be the last fragment of @skb -- or
1883 * @skb itself. If this would exceed the total buffer size the kernel
1884 * will panic. A pointer to the first byte of the extra data is
1888 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1891 skb->data_len += len;
1894 return skb_put(tail, len);
1896 EXPORT_SYMBOL_GPL(pskb_put);
1899 * skb_put - add data to a buffer
1900 * @skb: buffer to use
1901 * @len: amount of data to add
1903 * This function extends the used data area of the buffer. If this would
1904 * exceed the total buffer size the kernel will panic. A pointer to the
1905 * first byte of the extra data is returned.
1907 void *skb_put(struct sk_buff *skb, unsigned int len)
1909 void *tmp = skb_tail_pointer(skb);
1910 SKB_LINEAR_ASSERT(skb);
1913 if (unlikely(skb->tail > skb->end))
1914 skb_over_panic(skb, len, __builtin_return_address(0));
1917 EXPORT_SYMBOL(skb_put);
1920 * skb_push - add data to the start of a buffer
1921 * @skb: buffer to use
1922 * @len: amount of data to add
1924 * This function extends the used data area of the buffer at the buffer
1925 * start. If this would exceed the total buffer headroom the kernel will
1926 * panic. A pointer to the first byte of the extra data is returned.
1928 void *skb_push(struct sk_buff *skb, unsigned int len)
1932 if (unlikely(skb->data < skb->head))
1933 skb_under_panic(skb, len, __builtin_return_address(0));
1936 EXPORT_SYMBOL(skb_push);
1939 * skb_pull - remove data from the start of a buffer
1940 * @skb: buffer to use
1941 * @len: amount of data to remove
1943 * This function removes data from the start of a buffer, returning
1944 * the memory to the headroom. A pointer to the next data in the buffer
1945 * is returned. Once the data has been pulled future pushes will overwrite
1948 void *skb_pull(struct sk_buff *skb, unsigned int len)
1950 return skb_pull_inline(skb, len);
1952 EXPORT_SYMBOL(skb_pull);
1955 * skb_trim - remove end from a buffer
1956 * @skb: buffer to alter
1959 * Cut the length of a buffer down by removing data from the tail. If
1960 * the buffer is already under the length specified it is not modified.
1961 * The skb must be linear.
1963 void skb_trim(struct sk_buff *skb, unsigned int len)
1966 __skb_trim(skb, len);
1968 EXPORT_SYMBOL(skb_trim);
1970 /* Trims skb to length len. It can change skb pointers.
1973 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1975 struct sk_buff **fragp;
1976 struct sk_buff *frag;
1977 int offset = skb_headlen(skb);
1978 int nfrags = skb_shinfo(skb)->nr_frags;
1982 if (skb_cloned(skb) &&
1983 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1990 for (; i < nfrags; i++) {
1991 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1998 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
2001 skb_shinfo(skb)->nr_frags = i;
2003 for (; i < nfrags; i++)
2004 skb_frag_unref(skb, i);
2006 if (skb_has_frag_list(skb))
2007 skb_drop_fraglist(skb);
2011 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
2012 fragp = &frag->next) {
2013 int end = offset + frag->len;
2015 if (skb_shared(frag)) {
2016 struct sk_buff *nfrag;
2018 nfrag = skb_clone(frag, GFP_ATOMIC);
2019 if (unlikely(!nfrag))
2022 nfrag->next = frag->next;
2034 unlikely((err = pskb_trim(frag, len - offset))))
2038 skb_drop_list(&frag->next);
2043 if (len > skb_headlen(skb)) {
2044 skb->data_len -= skb->len - len;
2049 skb_set_tail_pointer(skb, len);
2052 if (!skb->sk || skb->destructor == sock_edemux)
2056 EXPORT_SYMBOL(___pskb_trim);
2058 /* Note : use pskb_trim_rcsum() instead of calling this directly
2060 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2062 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2063 int delta = skb->len - len;
2065 skb->csum = csum_block_sub(skb->csum,
2066 skb_checksum(skb, len, delta, 0),
2068 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2069 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2070 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2072 if (offset + sizeof(__sum16) > hdlen)
2075 return __pskb_trim(skb, len);
2077 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2080 * __pskb_pull_tail - advance tail of skb header
2081 * @skb: buffer to reallocate
2082 * @delta: number of bytes to advance tail
2084 * The function makes a sense only on a fragmented &sk_buff,
2085 * it expands header moving its tail forward and copying necessary
2086 * data from fragmented part.
2088 * &sk_buff MUST have reference count of 1.
2090 * Returns %NULL (and &sk_buff does not change) if pull failed
2091 * or value of new tail of skb in the case of success.
2093 * All the pointers pointing into skb header may change and must be
2094 * reloaded after call to this function.
2097 /* Moves tail of skb head forward, copying data from fragmented part,
2098 * when it is necessary.
2099 * 1. It may fail due to malloc failure.
2100 * 2. It may change skb pointers.
2102 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2104 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2106 /* If skb has not enough free space at tail, get new one
2107 * plus 128 bytes for future expansions. If we have enough
2108 * room at tail, reallocate without expansion only if skb is cloned.
2110 int i, k, eat = (skb->tail + delta) - skb->end;
2112 if (eat > 0 || skb_cloned(skb)) {
2113 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2118 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2119 skb_tail_pointer(skb), delta));
2121 /* Optimization: no fragments, no reasons to preestimate
2122 * size of pulled pages. Superb.
2124 if (!skb_has_frag_list(skb))
2127 /* Estimate size of pulled pages. */
2129 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2130 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2137 /* If we need update frag list, we are in troubles.
2138 * Certainly, it is possible to add an offset to skb data,
2139 * but taking into account that pulling is expected to
2140 * be very rare operation, it is worth to fight against
2141 * further bloating skb head and crucify ourselves here instead.
2142 * Pure masohism, indeed. 8)8)
2145 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2146 struct sk_buff *clone = NULL;
2147 struct sk_buff *insp = NULL;
2150 if (list->len <= eat) {
2151 /* Eaten as whole. */
2156 /* Eaten partially. */
2158 if (skb_shared(list)) {
2159 /* Sucks! We need to fork list. :-( */
2160 clone = skb_clone(list, GFP_ATOMIC);
2166 /* This may be pulled without
2170 if (!pskb_pull(list, eat)) {
2178 /* Free pulled out fragments. */
2179 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2180 skb_shinfo(skb)->frag_list = list->next;
2183 /* And insert new clone at head. */
2186 skb_shinfo(skb)->frag_list = clone;
2189 /* Success! Now we may commit changes to skb data. */
2194 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2195 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2198 skb_frag_unref(skb, i);
2201 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2203 *frag = skb_shinfo(skb)->frags[i];
2205 skb_frag_off_add(frag, eat);
2206 skb_frag_size_sub(frag, eat);
2214 skb_shinfo(skb)->nr_frags = k;
2218 skb->data_len -= delta;
2221 skb_zcopy_clear(skb, false);
2223 return skb_tail_pointer(skb);
2225 EXPORT_SYMBOL(__pskb_pull_tail);
2228 * skb_copy_bits - copy bits from skb to kernel buffer
2230 * @offset: offset in source
2231 * @to: destination buffer
2232 * @len: number of bytes to copy
2234 * Copy the specified number of bytes from the source skb to the
2235 * destination buffer.
2238 * If its prototype is ever changed,
2239 * check arch/{*}/net/{*}.S files,
2240 * since it is called from BPF assembly code.
2242 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2244 int start = skb_headlen(skb);
2245 struct sk_buff *frag_iter;
2248 if (offset > (int)skb->len - len)
2252 if ((copy = start - offset) > 0) {
2255 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2256 if ((len -= copy) == 0)
2262 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2264 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2266 WARN_ON(start > offset + len);
2268 end = start + skb_frag_size(f);
2269 if ((copy = end - offset) > 0) {
2270 u32 p_off, p_len, copied;
2277 skb_frag_foreach_page(f,
2278 skb_frag_off(f) + offset - start,
2279 copy, p, p_off, p_len, copied) {
2280 vaddr = kmap_atomic(p);
2281 memcpy(to + copied, vaddr + p_off, p_len);
2282 kunmap_atomic(vaddr);
2285 if ((len -= copy) == 0)
2293 skb_walk_frags(skb, frag_iter) {
2296 WARN_ON(start > offset + len);
2298 end = start + frag_iter->len;
2299 if ((copy = end - offset) > 0) {
2302 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2304 if ((len -= copy) == 0)
2318 EXPORT_SYMBOL(skb_copy_bits);
2321 * Callback from splice_to_pipe(), if we need to release some pages
2322 * at the end of the spd in case we error'ed out in filling the pipe.
2324 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2326 put_page(spd->pages[i]);
2329 static struct page *linear_to_page(struct page *page, unsigned int *len,
2330 unsigned int *offset,
2333 struct page_frag *pfrag = sk_page_frag(sk);
2335 if (!sk_page_frag_refill(sk, pfrag))
2338 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2340 memcpy(page_address(pfrag->page) + pfrag->offset,
2341 page_address(page) + *offset, *len);
2342 *offset = pfrag->offset;
2343 pfrag->offset += *len;
2348 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2350 unsigned int offset)
2352 return spd->nr_pages &&
2353 spd->pages[spd->nr_pages - 1] == page &&
2354 (spd->partial[spd->nr_pages - 1].offset +
2355 spd->partial[spd->nr_pages - 1].len == offset);
2359 * Fill page/offset/length into spd, if it can hold more pages.
2361 static bool spd_fill_page(struct splice_pipe_desc *spd,
2362 struct pipe_inode_info *pipe, struct page *page,
2363 unsigned int *len, unsigned int offset,
2367 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2371 page = linear_to_page(page, len, &offset, sk);
2375 if (spd_can_coalesce(spd, page, offset)) {
2376 spd->partial[spd->nr_pages - 1].len += *len;
2380 spd->pages[spd->nr_pages] = page;
2381 spd->partial[spd->nr_pages].len = *len;
2382 spd->partial[spd->nr_pages].offset = offset;
2388 static bool __splice_segment(struct page *page, unsigned int poff,
2389 unsigned int plen, unsigned int *off,
2391 struct splice_pipe_desc *spd, bool linear,
2393 struct pipe_inode_info *pipe)
2398 /* skip this segment if already processed */
2404 /* ignore any bits we already processed */
2410 unsigned int flen = min(*len, plen);
2412 if (spd_fill_page(spd, pipe, page, &flen, poff,
2418 } while (*len && plen);
2424 * Map linear and fragment data from the skb to spd. It reports true if the
2425 * pipe is full or if we already spliced the requested length.
2427 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2428 unsigned int *offset, unsigned int *len,
2429 struct splice_pipe_desc *spd, struct sock *sk)
2432 struct sk_buff *iter;
2434 /* map the linear part :
2435 * If skb->head_frag is set, this 'linear' part is backed by a
2436 * fragment, and if the head is not shared with any clones then
2437 * we can avoid a copy since we own the head portion of this page.
2439 if (__splice_segment(virt_to_page(skb->data),
2440 (unsigned long) skb->data & (PAGE_SIZE - 1),
2443 skb_head_is_locked(skb),
2448 * then map the fragments
2450 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2451 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2453 if (__splice_segment(skb_frag_page(f),
2454 skb_frag_off(f), skb_frag_size(f),
2455 offset, len, spd, false, sk, pipe))
2459 skb_walk_frags(skb, iter) {
2460 if (*offset >= iter->len) {
2461 *offset -= iter->len;
2464 /* __skb_splice_bits() only fails if the output has no room
2465 * left, so no point in going over the frag_list for the error
2468 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2476 * Map data from the skb to a pipe. Should handle both the linear part,
2477 * the fragments, and the frag list.
2479 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2480 struct pipe_inode_info *pipe, unsigned int tlen,
2483 struct partial_page partial[MAX_SKB_FRAGS];
2484 struct page *pages[MAX_SKB_FRAGS];
2485 struct splice_pipe_desc spd = {
2488 .nr_pages_max = MAX_SKB_FRAGS,
2489 .ops = &nosteal_pipe_buf_ops,
2490 .spd_release = sock_spd_release,
2494 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2497 ret = splice_to_pipe(pipe, &spd);
2501 EXPORT_SYMBOL_GPL(skb_splice_bits);
2503 static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg,
2504 struct kvec *vec, size_t num, size_t size)
2506 struct socket *sock = sk->sk_socket;
2510 return kernel_sendmsg(sock, msg, vec, num, size);
2513 static int sendpage_unlocked(struct sock *sk, struct page *page, int offset,
2514 size_t size, int flags)
2516 struct socket *sock = sk->sk_socket;
2520 return kernel_sendpage(sock, page, offset, size, flags);
2523 typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg,
2524 struct kvec *vec, size_t num, size_t size);
2525 typedef int (*sendpage_func)(struct sock *sk, struct page *page, int offset,
2526 size_t size, int flags);
2527 static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset,
2528 int len, sendmsg_func sendmsg, sendpage_func sendpage)
2530 unsigned int orig_len = len;
2531 struct sk_buff *head = skb;
2532 unsigned short fragidx;
2537 /* Deal with head data */
2538 while (offset < skb_headlen(skb) && len) {
2542 slen = min_t(int, len, skb_headlen(skb) - offset);
2543 kv.iov_base = skb->data + offset;
2545 memset(&msg, 0, sizeof(msg));
2546 msg.msg_flags = MSG_DONTWAIT;
2548 ret = INDIRECT_CALL_2(sendmsg, kernel_sendmsg_locked,
2549 sendmsg_unlocked, sk, &msg, &kv, 1, slen);
2557 /* All the data was skb head? */
2561 /* Make offset relative to start of frags */
2562 offset -= skb_headlen(skb);
2564 /* Find where we are in frag list */
2565 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2566 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2568 if (offset < skb_frag_size(frag))
2571 offset -= skb_frag_size(frag);
2574 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2575 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2577 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2580 ret = INDIRECT_CALL_2(sendpage, kernel_sendpage_locked,
2581 sendpage_unlocked, sk,
2582 skb_frag_page(frag),
2583 skb_frag_off(frag) + offset,
2584 slen, MSG_DONTWAIT);
2597 /* Process any frag lists */
2600 if (skb_has_frag_list(skb)) {
2601 skb = skb_shinfo(skb)->frag_list;
2604 } else if (skb->next) {
2611 return orig_len - len;
2614 return orig_len == len ? ret : orig_len - len;
2617 /* Send skb data on a socket. Socket must be locked. */
2618 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2621 return __skb_send_sock(sk, skb, offset, len, kernel_sendmsg_locked,
2622 kernel_sendpage_locked);
2624 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2626 /* Send skb data on a socket. Socket must be unlocked. */
2627 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
2629 return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked,
2634 * skb_store_bits - store bits from kernel buffer to skb
2635 * @skb: destination buffer
2636 * @offset: offset in destination
2637 * @from: source buffer
2638 * @len: number of bytes to copy
2640 * Copy the specified number of bytes from the source buffer to the
2641 * destination skb. This function handles all the messy bits of
2642 * traversing fragment lists and such.
2645 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2647 int start = skb_headlen(skb);
2648 struct sk_buff *frag_iter;
2651 if (offset > (int)skb->len - len)
2654 if ((copy = start - offset) > 0) {
2657 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2658 if ((len -= copy) == 0)
2664 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2665 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2668 WARN_ON(start > offset + len);
2670 end = start + skb_frag_size(frag);
2671 if ((copy = end - offset) > 0) {
2672 u32 p_off, p_len, copied;
2679 skb_frag_foreach_page(frag,
2680 skb_frag_off(frag) + offset - start,
2681 copy, p, p_off, p_len, copied) {
2682 vaddr = kmap_atomic(p);
2683 memcpy(vaddr + p_off, from + copied, p_len);
2684 kunmap_atomic(vaddr);
2687 if ((len -= copy) == 0)
2695 skb_walk_frags(skb, frag_iter) {
2698 WARN_ON(start > offset + len);
2700 end = start + frag_iter->len;
2701 if ((copy = end - offset) > 0) {
2704 if (skb_store_bits(frag_iter, offset - start,
2707 if ((len -= copy) == 0)
2720 EXPORT_SYMBOL(skb_store_bits);
2722 /* Checksum skb data. */
2723 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2724 __wsum csum, const struct skb_checksum_ops *ops)
2726 int start = skb_headlen(skb);
2727 int i, copy = start - offset;
2728 struct sk_buff *frag_iter;
2731 /* Checksum header. */
2735 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2736 skb->data + offset, copy, csum);
2737 if ((len -= copy) == 0)
2743 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2745 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2747 WARN_ON(start > offset + len);
2749 end = start + skb_frag_size(frag);
2750 if ((copy = end - offset) > 0) {
2751 u32 p_off, p_len, copied;
2759 skb_frag_foreach_page(frag,
2760 skb_frag_off(frag) + offset - start,
2761 copy, p, p_off, p_len, copied) {
2762 vaddr = kmap_atomic(p);
2763 csum2 = INDIRECT_CALL_1(ops->update,
2765 vaddr + p_off, p_len, 0);
2766 kunmap_atomic(vaddr);
2767 csum = INDIRECT_CALL_1(ops->combine,
2768 csum_block_add_ext, csum,
2780 skb_walk_frags(skb, frag_iter) {
2783 WARN_ON(start > offset + len);
2785 end = start + frag_iter->len;
2786 if ((copy = end - offset) > 0) {
2790 csum2 = __skb_checksum(frag_iter, offset - start,
2792 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2793 csum, csum2, pos, copy);
2794 if ((len -= copy) == 0)
2805 EXPORT_SYMBOL(__skb_checksum);
2807 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2808 int len, __wsum csum)
2810 const struct skb_checksum_ops ops = {
2811 .update = csum_partial_ext,
2812 .combine = csum_block_add_ext,
2815 return __skb_checksum(skb, offset, len, csum, &ops);
2817 EXPORT_SYMBOL(skb_checksum);
2819 /* Both of above in one bottle. */
2821 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2824 int start = skb_headlen(skb);
2825 int i, copy = start - offset;
2826 struct sk_buff *frag_iter;
2834 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2836 if ((len -= copy) == 0)
2843 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2846 WARN_ON(start > offset + len);
2848 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2849 if ((copy = end - offset) > 0) {
2850 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2851 u32 p_off, p_len, copied;
2859 skb_frag_foreach_page(frag,
2860 skb_frag_off(frag) + offset - start,
2861 copy, p, p_off, p_len, copied) {
2862 vaddr = kmap_atomic(p);
2863 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2866 kunmap_atomic(vaddr);
2867 csum = csum_block_add(csum, csum2, pos);
2879 skb_walk_frags(skb, frag_iter) {
2883 WARN_ON(start > offset + len);
2885 end = start + frag_iter->len;
2886 if ((copy = end - offset) > 0) {
2889 csum2 = skb_copy_and_csum_bits(frag_iter,
2892 csum = csum_block_add(csum, csum2, pos);
2893 if ((len -= copy) == 0)
2904 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2906 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2910 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2911 /* See comments in __skb_checksum_complete(). */
2913 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2914 !skb->csum_complete_sw)
2915 netdev_rx_csum_fault(skb->dev, skb);
2917 if (!skb_shared(skb))
2918 skb->csum_valid = !sum;
2921 EXPORT_SYMBOL(__skb_checksum_complete_head);
2923 /* This function assumes skb->csum already holds pseudo header's checksum,
2924 * which has been changed from the hardware checksum, for example, by
2925 * __skb_checksum_validate_complete(). And, the original skb->csum must
2926 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2928 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2929 * zero. The new checksum is stored back into skb->csum unless the skb is
2932 __sum16 __skb_checksum_complete(struct sk_buff *skb)
2937 csum = skb_checksum(skb, 0, skb->len, 0);
2939 sum = csum_fold(csum_add(skb->csum, csum));
2940 /* This check is inverted, because we already knew the hardware
2941 * checksum is invalid before calling this function. So, if the
2942 * re-computed checksum is valid instead, then we have a mismatch
2943 * between the original skb->csum and skb_checksum(). This means either
2944 * the original hardware checksum is incorrect or we screw up skb->csum
2945 * when moving skb->data around.
2948 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2949 !skb->csum_complete_sw)
2950 netdev_rx_csum_fault(skb->dev, skb);
2953 if (!skb_shared(skb)) {
2954 /* Save full packet checksum */
2956 skb->ip_summed = CHECKSUM_COMPLETE;
2957 skb->csum_complete_sw = 1;
2958 skb->csum_valid = !sum;
2963 EXPORT_SYMBOL(__skb_checksum_complete);
2965 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2967 net_warn_ratelimited(
2968 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2973 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2974 int offset, int len)
2976 net_warn_ratelimited(
2977 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2982 static const struct skb_checksum_ops default_crc32c_ops = {
2983 .update = warn_crc32c_csum_update,
2984 .combine = warn_crc32c_csum_combine,
2987 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2988 &default_crc32c_ops;
2989 EXPORT_SYMBOL(crc32c_csum_stub);
2992 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2993 * @from: source buffer
2995 * Calculates the amount of linear headroom needed in the 'to' skb passed
2996 * into skb_zerocopy().
2999 skb_zerocopy_headlen(const struct sk_buff *from)
3001 unsigned int hlen = 0;
3003 if (!from->head_frag ||
3004 skb_headlen(from) < L1_CACHE_BYTES ||
3005 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3006 hlen = skb_headlen(from);
3008 if (skb_has_frag_list(from))
3013 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
3016 * skb_zerocopy - Zero copy skb to skb
3017 * @to: destination buffer
3018 * @from: source buffer
3019 * @len: number of bytes to copy from source buffer
3020 * @hlen: size of linear headroom in destination buffer
3022 * Copies up to `len` bytes from `from` to `to` by creating references
3023 * to the frags in the source buffer.
3025 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
3026 * headroom in the `to` buffer.
3029 * 0: everything is OK
3030 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
3031 * -EFAULT: skb_copy_bits() found some problem with skb geometry
3034 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
3037 int plen = 0; /* length of skb->head fragment */
3040 unsigned int offset;
3042 BUG_ON(!from->head_frag && !hlen);
3044 /* dont bother with small payloads */
3045 if (len <= skb_tailroom(to))
3046 return skb_copy_bits(from, 0, skb_put(to, len), len);
3049 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
3054 plen = min_t(int, skb_headlen(from), len);
3056 page = virt_to_head_page(from->head);
3057 offset = from->data - (unsigned char *)page_address(page);
3058 __skb_fill_page_desc(to, 0, page, offset, plen);
3065 to->truesize += len + plen;
3066 to->len += len + plen;
3067 to->data_len += len + plen;
3069 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
3073 skb_zerocopy_clone(to, from, GFP_ATOMIC);
3075 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3080 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3081 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3083 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3085 skb_frag_ref(to, j);
3088 skb_shinfo(to)->nr_frags = j;
3092 EXPORT_SYMBOL_GPL(skb_zerocopy);
3094 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3099 if (skb->ip_summed == CHECKSUM_PARTIAL)
3100 csstart = skb_checksum_start_offset(skb);
3102 csstart = skb_headlen(skb);
3104 BUG_ON(csstart > skb_headlen(skb));
3106 skb_copy_from_linear_data(skb, to, csstart);
3109 if (csstart != skb->len)
3110 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3111 skb->len - csstart);
3113 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3114 long csstuff = csstart + skb->csum_offset;
3116 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3119 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3122 * skb_dequeue - remove from the head of the queue
3123 * @list: list to dequeue from
3125 * Remove the head of the list. The list lock is taken so the function
3126 * may be used safely with other locking list functions. The head item is
3127 * returned or %NULL if the list is empty.
3130 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3132 unsigned long flags;
3133 struct sk_buff *result;
3135 spin_lock_irqsave(&list->lock, flags);
3136 result = __skb_dequeue(list);
3137 spin_unlock_irqrestore(&list->lock, flags);
3140 EXPORT_SYMBOL(skb_dequeue);
3143 * skb_dequeue_tail - remove from the tail of the queue
3144 * @list: list to dequeue from
3146 * Remove the tail of the list. The list lock is taken so the function
3147 * may be used safely with other locking list functions. The tail item is
3148 * returned or %NULL if the list is empty.
3150 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3152 unsigned long flags;
3153 struct sk_buff *result;
3155 spin_lock_irqsave(&list->lock, flags);
3156 result = __skb_dequeue_tail(list);
3157 spin_unlock_irqrestore(&list->lock, flags);
3160 EXPORT_SYMBOL(skb_dequeue_tail);
3163 * skb_queue_purge - empty a list
3164 * @list: list to empty
3166 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3167 * the list and one reference dropped. This function takes the list
3168 * lock and is atomic with respect to other list locking functions.
3170 void skb_queue_purge(struct sk_buff_head *list)
3172 struct sk_buff *skb;
3173 while ((skb = skb_dequeue(list)) != NULL)
3176 EXPORT_SYMBOL(skb_queue_purge);
3179 * skb_rbtree_purge - empty a skb rbtree
3180 * @root: root of the rbtree to empty
3181 * Return value: the sum of truesizes of all purged skbs.
3183 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3184 * the list and one reference dropped. This function does not take
3185 * any lock. Synchronization should be handled by the caller (e.g., TCP
3186 * out-of-order queue is protected by the socket lock).
3188 unsigned int skb_rbtree_purge(struct rb_root *root)
3190 struct rb_node *p = rb_first(root);
3191 unsigned int sum = 0;
3194 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3197 rb_erase(&skb->rbnode, root);
3198 sum += skb->truesize;
3205 * skb_queue_head - queue a buffer at the list head
3206 * @list: list to use
3207 * @newsk: buffer to queue
3209 * Queue a buffer at the start of the list. This function takes the
3210 * list lock and can be used safely with other locking &sk_buff functions
3213 * A buffer cannot be placed on two lists at the same time.
3215 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3217 unsigned long flags;
3219 spin_lock_irqsave(&list->lock, flags);
3220 __skb_queue_head(list, newsk);
3221 spin_unlock_irqrestore(&list->lock, flags);
3223 EXPORT_SYMBOL(skb_queue_head);
3226 * skb_queue_tail - queue a buffer at the list tail
3227 * @list: list to use
3228 * @newsk: buffer to queue
3230 * Queue a buffer at the tail of the list. This function takes the
3231 * list lock and can be used safely with other locking &sk_buff functions
3234 * A buffer cannot be placed on two lists at the same time.
3236 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3238 unsigned long flags;
3240 spin_lock_irqsave(&list->lock, flags);
3241 __skb_queue_tail(list, newsk);
3242 spin_unlock_irqrestore(&list->lock, flags);
3244 EXPORT_SYMBOL(skb_queue_tail);
3247 * skb_unlink - remove a buffer from a list
3248 * @skb: buffer to remove
3249 * @list: list to use
3251 * Remove a packet from a list. The list locks are taken and this
3252 * function is atomic with respect to other list locked calls
3254 * You must know what list the SKB is on.
3256 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3258 unsigned long flags;
3260 spin_lock_irqsave(&list->lock, flags);
3261 __skb_unlink(skb, list);
3262 spin_unlock_irqrestore(&list->lock, flags);
3264 EXPORT_SYMBOL(skb_unlink);
3267 * skb_append - append a buffer
3268 * @old: buffer to insert after
3269 * @newsk: buffer to insert
3270 * @list: list to use
3272 * Place a packet after a given packet in a list. The list locks are taken
3273 * and this function is atomic with respect to other list locked calls.
3274 * A buffer cannot be placed on two lists at the same time.
3276 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3278 unsigned long flags;
3280 spin_lock_irqsave(&list->lock, flags);
3281 __skb_queue_after(list, old, newsk);
3282 spin_unlock_irqrestore(&list->lock, flags);
3284 EXPORT_SYMBOL(skb_append);
3286 static inline void skb_split_inside_header(struct sk_buff *skb,
3287 struct sk_buff* skb1,
3288 const u32 len, const int pos)
3292 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3294 /* And move data appendix as is. */
3295 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3296 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3298 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3299 skb_shinfo(skb)->nr_frags = 0;
3300 skb1->data_len = skb->data_len;
3301 skb1->len += skb1->data_len;
3304 skb_set_tail_pointer(skb, len);
3307 static inline void skb_split_no_header(struct sk_buff *skb,
3308 struct sk_buff* skb1,
3309 const u32 len, int pos)
3312 const int nfrags = skb_shinfo(skb)->nr_frags;
3314 skb_shinfo(skb)->nr_frags = 0;
3315 skb1->len = skb1->data_len = skb->len - len;
3317 skb->data_len = len - pos;
3319 for (i = 0; i < nfrags; i++) {
3320 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3322 if (pos + size > len) {
3323 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3327 * We have two variants in this case:
3328 * 1. Move all the frag to the second
3329 * part, if it is possible. F.e.
3330 * this approach is mandatory for TUX,
3331 * where splitting is expensive.
3332 * 2. Split is accurately. We make this.
3334 skb_frag_ref(skb, i);
3335 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3336 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3337 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3338 skb_shinfo(skb)->nr_frags++;
3342 skb_shinfo(skb)->nr_frags++;
3345 skb_shinfo(skb1)->nr_frags = k;
3349 * skb_split - Split fragmented skb to two parts at length len.
3350 * @skb: the buffer to split
3351 * @skb1: the buffer to receive the second part
3352 * @len: new length for skb
3354 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3356 int pos = skb_headlen(skb);
3358 skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3359 skb_zerocopy_clone(skb1, skb, 0);
3360 if (len < pos) /* Split line is inside header. */
3361 skb_split_inside_header(skb, skb1, len, pos);
3362 else /* Second chunk has no header, nothing to copy. */
3363 skb_split_no_header(skb, skb1, len, pos);
3365 EXPORT_SYMBOL(skb_split);
3367 /* Shifting from/to a cloned skb is a no-go.
3369 * Caller cannot keep skb_shinfo related pointers past calling here!
3371 static int skb_prepare_for_shift(struct sk_buff *skb)
3375 if (skb_cloned(skb)) {
3376 /* Save and restore truesize: pskb_expand_head() may reallocate
3377 * memory where ksize(kmalloc(S)) != ksize(kmalloc(S)), but we
3378 * cannot change truesize at this point.
3380 unsigned int save_truesize = skb->truesize;
3382 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3383 skb->truesize = save_truesize;
3389 * skb_shift - Shifts paged data partially from skb to another
3390 * @tgt: buffer into which tail data gets added
3391 * @skb: buffer from which the paged data comes from
3392 * @shiftlen: shift up to this many bytes
3394 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3395 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3396 * It's up to caller to free skb if everything was shifted.
3398 * If @tgt runs out of frags, the whole operation is aborted.
3400 * Skb cannot include anything else but paged data while tgt is allowed
3401 * to have non-paged data as well.
3403 * TODO: full sized shift could be optimized but that would need
3404 * specialized skb free'er to handle frags without up-to-date nr_frags.
3406 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3408 int from, to, merge, todo;
3409 skb_frag_t *fragfrom, *fragto;
3411 BUG_ON(shiftlen > skb->len);
3413 if (skb_headlen(skb))
3415 if (skb_zcopy(tgt) || skb_zcopy(skb))
3420 to = skb_shinfo(tgt)->nr_frags;
3421 fragfrom = &skb_shinfo(skb)->frags[from];
3423 /* Actual merge is delayed until the point when we know we can
3424 * commit all, so that we don't have to undo partial changes
3427 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3428 skb_frag_off(fragfrom))) {
3433 todo -= skb_frag_size(fragfrom);
3435 if (skb_prepare_for_shift(skb) ||
3436 skb_prepare_for_shift(tgt))
3439 /* All previous frag pointers might be stale! */
3440 fragfrom = &skb_shinfo(skb)->frags[from];
3441 fragto = &skb_shinfo(tgt)->frags[merge];
3443 skb_frag_size_add(fragto, shiftlen);
3444 skb_frag_size_sub(fragfrom, shiftlen);
3445 skb_frag_off_add(fragfrom, shiftlen);
3453 /* Skip full, not-fitting skb to avoid expensive operations */
3454 if ((shiftlen == skb->len) &&
3455 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3458 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3461 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3462 if (to == MAX_SKB_FRAGS)
3465 fragfrom = &skb_shinfo(skb)->frags[from];
3466 fragto = &skb_shinfo(tgt)->frags[to];
3468 if (todo >= skb_frag_size(fragfrom)) {
3469 *fragto = *fragfrom;
3470 todo -= skb_frag_size(fragfrom);
3475 __skb_frag_ref(fragfrom);
3476 skb_frag_page_copy(fragto, fragfrom);
3477 skb_frag_off_copy(fragto, fragfrom);
3478 skb_frag_size_set(fragto, todo);
3480 skb_frag_off_add(fragfrom, todo);
3481 skb_frag_size_sub(fragfrom, todo);
3489 /* Ready to "commit" this state change to tgt */
3490 skb_shinfo(tgt)->nr_frags = to;
3493 fragfrom = &skb_shinfo(skb)->frags[0];
3494 fragto = &skb_shinfo(tgt)->frags[merge];
3496 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3497 __skb_frag_unref(fragfrom);
3500 /* Reposition in the original skb */
3502 while (from < skb_shinfo(skb)->nr_frags)
3503 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3504 skb_shinfo(skb)->nr_frags = to;
3506 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3509 /* Most likely the tgt won't ever need its checksum anymore, skb on
3510 * the other hand might need it if it needs to be resent
3512 tgt->ip_summed = CHECKSUM_PARTIAL;
3513 skb->ip_summed = CHECKSUM_PARTIAL;
3515 /* Yak, is it really working this way? Some helper please? */
3516 skb->len -= shiftlen;
3517 skb->data_len -= shiftlen;
3518 skb->truesize -= shiftlen;
3519 tgt->len += shiftlen;
3520 tgt->data_len += shiftlen;
3521 tgt->truesize += shiftlen;
3527 * skb_prepare_seq_read - Prepare a sequential read of skb data
3528 * @skb: the buffer to read
3529 * @from: lower offset of data to be read
3530 * @to: upper offset of data to be read
3531 * @st: state variable
3533 * Initializes the specified state variable. Must be called before
3534 * invoking skb_seq_read() for the first time.
3536 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3537 unsigned int to, struct skb_seq_state *st)
3539 st->lower_offset = from;
3540 st->upper_offset = to;
3541 st->root_skb = st->cur_skb = skb;
3542 st->frag_idx = st->stepped_offset = 0;
3543 st->frag_data = NULL;
3546 EXPORT_SYMBOL(skb_prepare_seq_read);
3549 * skb_seq_read - Sequentially read skb data
3550 * @consumed: number of bytes consumed by the caller so far
3551 * @data: destination pointer for data to be returned
3552 * @st: state variable
3554 * Reads a block of skb data at @consumed relative to the
3555 * lower offset specified to skb_prepare_seq_read(). Assigns
3556 * the head of the data block to @data and returns the length
3557 * of the block or 0 if the end of the skb data or the upper
3558 * offset has been reached.
3560 * The caller is not required to consume all of the data
3561 * returned, i.e. @consumed is typically set to the number
3562 * of bytes already consumed and the next call to
3563 * skb_seq_read() will return the remaining part of the block.
3565 * Note 1: The size of each block of data returned can be arbitrary,
3566 * this limitation is the cost for zerocopy sequential
3567 * reads of potentially non linear data.
3569 * Note 2: Fragment lists within fragments are not implemented
3570 * at the moment, state->root_skb could be replaced with
3571 * a stack for this purpose.
3573 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3574 struct skb_seq_state *st)
3576 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3579 if (unlikely(abs_offset >= st->upper_offset)) {
3580 if (st->frag_data) {
3581 kunmap_atomic(st->frag_data);
3582 st->frag_data = NULL;
3588 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3590 if (abs_offset < block_limit && !st->frag_data) {
3591 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3592 return block_limit - abs_offset;
3595 if (st->frag_idx == 0 && !st->frag_data)
3596 st->stepped_offset += skb_headlen(st->cur_skb);
3598 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3599 unsigned int pg_idx, pg_off, pg_sz;
3601 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3604 pg_off = skb_frag_off(frag);
3605 pg_sz = skb_frag_size(frag);
3607 if (skb_frag_must_loop(skb_frag_page(frag))) {
3608 pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
3609 pg_off = offset_in_page(pg_off + st->frag_off);
3610 pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
3611 PAGE_SIZE - pg_off);
3614 block_limit = pg_sz + st->stepped_offset;
3615 if (abs_offset < block_limit) {
3617 st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
3619 *data = (u8 *)st->frag_data + pg_off +
3620 (abs_offset - st->stepped_offset);
3622 return block_limit - abs_offset;
3625 if (st->frag_data) {
3626 kunmap_atomic(st->frag_data);
3627 st->frag_data = NULL;
3630 st->stepped_offset += pg_sz;
3631 st->frag_off += pg_sz;
3632 if (st->frag_off == skb_frag_size(frag)) {
3638 if (st->frag_data) {
3639 kunmap_atomic(st->frag_data);
3640 st->frag_data = NULL;
3643 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3644 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3647 } else if (st->cur_skb->next) {
3648 st->cur_skb = st->cur_skb->next;
3655 EXPORT_SYMBOL(skb_seq_read);
3658 * skb_abort_seq_read - Abort a sequential read of skb data
3659 * @st: state variable
3661 * Must be called if skb_seq_read() was not called until it
3664 void skb_abort_seq_read(struct skb_seq_state *st)
3667 kunmap_atomic(st->frag_data);
3669 EXPORT_SYMBOL(skb_abort_seq_read);
3671 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3673 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3674 struct ts_config *conf,
3675 struct ts_state *state)
3677 return skb_seq_read(offset, text, TS_SKB_CB(state));
3680 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3682 skb_abort_seq_read(TS_SKB_CB(state));
3686 * skb_find_text - Find a text pattern in skb data
3687 * @skb: the buffer to look in
3688 * @from: search offset
3690 * @config: textsearch configuration
3692 * Finds a pattern in the skb data according to the specified
3693 * textsearch configuration. Use textsearch_next() to retrieve
3694 * subsequent occurrences of the pattern. Returns the offset
3695 * to the first occurrence or UINT_MAX if no match was found.
3697 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3698 unsigned int to, struct ts_config *config)
3700 struct ts_state state;
3703 BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb));
3705 config->get_next_block = skb_ts_get_next_block;
3706 config->finish = skb_ts_finish;
3708 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3710 ret = textsearch_find(config, &state);
3711 return (ret <= to - from ? ret : UINT_MAX);
3713 EXPORT_SYMBOL(skb_find_text);
3715 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3716 int offset, size_t size)
3718 int i = skb_shinfo(skb)->nr_frags;
3720 if (skb_can_coalesce(skb, i, page, offset)) {
3721 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3722 } else if (i < MAX_SKB_FRAGS) {
3724 skb_fill_page_desc(skb, i, page, offset, size);
3731 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3734 * skb_pull_rcsum - pull skb and update receive checksum
3735 * @skb: buffer to update
3736 * @len: length of data pulled
3738 * This function performs an skb_pull on the packet and updates
3739 * the CHECKSUM_COMPLETE checksum. It should be used on
3740 * receive path processing instead of skb_pull unless you know
3741 * that the checksum difference is zero (e.g., a valid IP header)
3742 * or you are setting ip_summed to CHECKSUM_NONE.
3744 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3746 unsigned char *data = skb->data;
3748 BUG_ON(len > skb->len);
3749 __skb_pull(skb, len);
3750 skb_postpull_rcsum(skb, data, len);
3753 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3755 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3757 skb_frag_t head_frag;
3760 page = virt_to_head_page(frag_skb->head);
3761 __skb_frag_set_page(&head_frag, page);
3762 skb_frag_off_set(&head_frag, frag_skb->data -
3763 (unsigned char *)page_address(page));
3764 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3768 struct sk_buff *skb_segment_list(struct sk_buff *skb,
3769 netdev_features_t features,
3770 unsigned int offset)
3772 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
3773 unsigned int tnl_hlen = skb_tnl_header_len(skb);
3774 unsigned int delta_truesize = 0;
3775 unsigned int delta_len = 0;
3776 struct sk_buff *nskb, *tmp;
3779 skb_push(skb, -skb_network_offset(skb) + offset);
3781 skb_shinfo(skb)->frag_list = NULL;
3782 skb->next = list_skb;
3786 list_skb = list_skb->next;
3789 if (skb_shared(nskb)) {
3790 tmp = skb_clone(nskb, GFP_ATOMIC);
3794 err = skb_unclone(nskb, GFP_ATOMIC);
3803 delta_len += nskb->len;
3804 delta_truesize += nskb->truesize;
3806 skb_push(nskb, -skb_network_offset(nskb) + offset);
3808 skb_release_head_state(nskb);
3809 __copy_skb_header(nskb, skb);
3811 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
3812 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
3813 nskb->data - tnl_hlen,
3816 if (skb_needs_linearize(nskb, features) &&
3817 __skb_linearize(nskb))
3822 skb->truesize = skb->truesize - delta_truesize;
3823 skb->data_len = skb->data_len - delta_len;
3824 skb->len = skb->len - delta_len;
3830 if (skb_needs_linearize(skb, features) &&
3831 __skb_linearize(skb))
3839 kfree_skb_list(skb->next);
3841 return ERR_PTR(-ENOMEM);
3843 EXPORT_SYMBOL_GPL(skb_segment_list);
3845 int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
3847 if (unlikely(p->len + skb->len >= 65536))
3850 if (NAPI_GRO_CB(p)->last == p)
3851 skb_shinfo(p)->frag_list = skb;
3853 NAPI_GRO_CB(p)->last->next = skb;
3855 skb_pull(skb, skb_gro_offset(skb));
3857 NAPI_GRO_CB(p)->last = skb;
3858 NAPI_GRO_CB(p)->count++;
3859 p->data_len += skb->len;
3860 p->truesize += skb->truesize;
3863 NAPI_GRO_CB(skb)->same_flow = 1;
3869 * skb_segment - Perform protocol segmentation on skb.
3870 * @head_skb: buffer to segment
3871 * @features: features for the output path (see dev->features)
3873 * This function performs segmentation on the given skb. It returns
3874 * a pointer to the first in a list of new skbs for the segments.
3875 * In case of error it returns ERR_PTR(err).
3877 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3878 netdev_features_t features)
3880 struct sk_buff *segs = NULL;
3881 struct sk_buff *tail = NULL;
3882 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3883 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3884 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3885 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3886 struct sk_buff *frag_skb = head_skb;
3887 unsigned int offset = doffset;
3888 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3889 unsigned int partial_segs = 0;
3890 unsigned int headroom;
3891 unsigned int len = head_skb->len;
3894 int nfrags = skb_shinfo(head_skb)->nr_frags;
3899 if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3900 (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3901 /* gso_size is untrusted, and we have a frag_list with a linear
3902 * non head_frag head.
3904 * (we assume checking the first list_skb member suffices;
3905 * i.e if either of the list_skb members have non head_frag
3906 * head, then the first one has too).
3908 * If head_skb's headlen does not fit requested gso_size, it
3909 * means that the frag_list members do NOT terminate on exact
3910 * gso_size boundaries. Hence we cannot perform skb_frag_t page
3911 * sharing. Therefore we must fallback to copying the frag_list
3912 * skbs; we do so by disabling SG.
3914 if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3915 features &= ~NETIF_F_SG;
3918 __skb_push(head_skb, doffset);
3919 proto = skb_network_protocol(head_skb, NULL);
3920 if (unlikely(!proto))
3921 return ERR_PTR(-EINVAL);
3923 sg = !!(features & NETIF_F_SG);
3924 csum = !!can_checksum_protocol(features, proto);
3926 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3927 if (!(features & NETIF_F_GSO_PARTIAL)) {
3928 struct sk_buff *iter;
3929 unsigned int frag_len;
3932 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3935 /* If we get here then all the required
3936 * GSO features except frag_list are supported.
3937 * Try to split the SKB to multiple GSO SKBs
3938 * with no frag_list.
3939 * Currently we can do that only when the buffers don't
3940 * have a linear part and all the buffers except
3941 * the last are of the same length.
3943 frag_len = list_skb->len;
3944 skb_walk_frags(head_skb, iter) {
3945 if (frag_len != iter->len && iter->next)
3947 if (skb_headlen(iter) && !iter->head_frag)
3953 if (len != frag_len)
3957 /* GSO partial only requires that we trim off any excess that
3958 * doesn't fit into an MSS sized block, so take care of that
3961 partial_segs = len / mss;
3962 if (partial_segs > 1)
3963 mss *= partial_segs;
3969 headroom = skb_headroom(head_skb);
3970 pos = skb_headlen(head_skb);
3973 struct sk_buff *nskb;
3974 skb_frag_t *nskb_frag;
3978 if (unlikely(mss == GSO_BY_FRAGS)) {
3979 len = list_skb->len;
3981 len = head_skb->len - offset;
3986 hsize = skb_headlen(head_skb) - offset;
3988 if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) &&
3989 (skb_headlen(list_skb) == len || sg)) {
3990 BUG_ON(skb_headlen(list_skb) > len);
3993 nfrags = skb_shinfo(list_skb)->nr_frags;
3994 frag = skb_shinfo(list_skb)->frags;
3995 frag_skb = list_skb;
3996 pos += skb_headlen(list_skb);
3998 while (pos < offset + len) {
3999 BUG_ON(i >= nfrags);
4001 size = skb_frag_size(frag);
4002 if (pos + size > offset + len)
4010 nskb = skb_clone(list_skb, GFP_ATOMIC);
4011 list_skb = list_skb->next;
4013 if (unlikely(!nskb))
4016 if (unlikely(pskb_trim(nskb, len))) {
4021 hsize = skb_end_offset(nskb);
4022 if (skb_cow_head(nskb, doffset + headroom)) {
4027 nskb->truesize += skb_end_offset(nskb) - hsize;
4028 skb_release_head_state(nskb);
4029 __skb_push(nskb, doffset);
4033 if (hsize > len || !sg)
4036 nskb = __alloc_skb(hsize + doffset + headroom,
4037 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
4040 if (unlikely(!nskb))
4043 skb_reserve(nskb, headroom);
4044 __skb_put(nskb, doffset);
4053 __copy_skb_header(nskb, head_skb);
4055 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
4056 skb_reset_mac_len(nskb);
4058 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
4059 nskb->data - tnl_hlen,
4060 doffset + tnl_hlen);
4062 if (nskb->len == len + doffset)
4063 goto perform_csum_check;
4067 if (!nskb->remcsum_offload)
4068 nskb->ip_summed = CHECKSUM_NONE;
4069 SKB_GSO_CB(nskb)->csum =
4070 skb_copy_and_csum_bits(head_skb, offset,
4074 SKB_GSO_CB(nskb)->csum_start =
4075 skb_headroom(nskb) + doffset;
4077 skb_copy_bits(head_skb, offset,
4084 nskb_frag = skb_shinfo(nskb)->frags;
4086 skb_copy_from_linear_data_offset(head_skb, offset,
4087 skb_put(nskb, hsize), hsize);
4089 skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags &
4092 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4093 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4096 while (pos < offset + len) {
4099 nfrags = skb_shinfo(list_skb)->nr_frags;
4100 frag = skb_shinfo(list_skb)->frags;
4101 frag_skb = list_skb;
4102 if (!skb_headlen(list_skb)) {
4105 BUG_ON(!list_skb->head_frag);
4107 /* to make room for head_frag. */
4111 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4112 skb_zerocopy_clone(nskb, frag_skb,
4116 list_skb = list_skb->next;
4119 if (unlikely(skb_shinfo(nskb)->nr_frags >=
4121 net_warn_ratelimited(
4122 "skb_segment: too many frags: %u %u\n",
4128 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4129 __skb_frag_ref(nskb_frag);
4130 size = skb_frag_size(nskb_frag);
4133 skb_frag_off_add(nskb_frag, offset - pos);
4134 skb_frag_size_sub(nskb_frag, offset - pos);
4137 skb_shinfo(nskb)->nr_frags++;
4139 if (pos + size <= offset + len) {
4144 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4152 nskb->data_len = len - hsize;
4153 nskb->len += nskb->data_len;
4154 nskb->truesize += nskb->data_len;
4158 if (skb_has_shared_frag(nskb) &&
4159 __skb_linearize(nskb))
4162 if (!nskb->remcsum_offload)
4163 nskb->ip_summed = CHECKSUM_NONE;
4164 SKB_GSO_CB(nskb)->csum =
4165 skb_checksum(nskb, doffset,
4166 nskb->len - doffset, 0);
4167 SKB_GSO_CB(nskb)->csum_start =
4168 skb_headroom(nskb) + doffset;
4170 } while ((offset += len) < head_skb->len);
4172 /* Some callers want to get the end of the list.
4173 * Put it in segs->prev to avoid walking the list.
4174 * (see validate_xmit_skb_list() for example)
4179 struct sk_buff *iter;
4180 int type = skb_shinfo(head_skb)->gso_type;
4181 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4183 /* Update type to add partial and then remove dodgy if set */
4184 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4185 type &= ~SKB_GSO_DODGY;
4187 /* Update GSO info and prepare to start updating headers on
4188 * our way back down the stack of protocols.
4190 for (iter = segs; iter; iter = iter->next) {
4191 skb_shinfo(iter)->gso_size = gso_size;
4192 skb_shinfo(iter)->gso_segs = partial_segs;
4193 skb_shinfo(iter)->gso_type = type;
4194 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4197 if (tail->len - doffset <= gso_size)
4198 skb_shinfo(tail)->gso_size = 0;
4199 else if (tail != segs)
4200 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4203 /* Following permits correct backpressure, for protocols
4204 * using skb_set_owner_w().
4205 * Idea is to tranfert ownership from head_skb to last segment.
4207 if (head_skb->destructor == sock_wfree) {
4208 swap(tail->truesize, head_skb->truesize);
4209 swap(tail->destructor, head_skb->destructor);
4210 swap(tail->sk, head_skb->sk);
4215 kfree_skb_list(segs);
4216 return ERR_PTR(err);
4218 EXPORT_SYMBOL_GPL(skb_segment);
4220 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4222 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4223 unsigned int offset = skb_gro_offset(skb);
4224 unsigned int headlen = skb_headlen(skb);
4225 unsigned int len = skb_gro_len(skb);
4226 unsigned int delta_truesize;
4229 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4232 lp = NAPI_GRO_CB(p)->last;
4233 pinfo = skb_shinfo(lp);
4235 if (headlen <= offset) {
4238 int i = skbinfo->nr_frags;
4239 int nr_frags = pinfo->nr_frags + i;
4241 if (nr_frags > MAX_SKB_FRAGS)
4245 pinfo->nr_frags = nr_frags;
4246 skbinfo->nr_frags = 0;
4248 frag = pinfo->frags + nr_frags;
4249 frag2 = skbinfo->frags + i;
4254 skb_frag_off_add(frag, offset);
4255 skb_frag_size_sub(frag, offset);
4257 /* all fragments truesize : remove (head size + sk_buff) */
4258 delta_truesize = skb->truesize -
4259 SKB_TRUESIZE(skb_end_offset(skb));
4261 skb->truesize -= skb->data_len;
4262 skb->len -= skb->data_len;
4265 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4267 } else if (skb->head_frag) {
4268 int nr_frags = pinfo->nr_frags;
4269 skb_frag_t *frag = pinfo->frags + nr_frags;
4270 struct page *page = virt_to_head_page(skb->head);
4271 unsigned int first_size = headlen - offset;
4272 unsigned int first_offset;
4274 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4277 first_offset = skb->data -
4278 (unsigned char *)page_address(page) +
4281 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4283 __skb_frag_set_page(frag, page);
4284 skb_frag_off_set(frag, first_offset);
4285 skb_frag_size_set(frag, first_size);
4287 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4288 /* We dont need to clear skbinfo->nr_frags here */
4290 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4291 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4296 delta_truesize = skb->truesize;
4297 if (offset > headlen) {
4298 unsigned int eat = offset - headlen;
4300 skb_frag_off_add(&skbinfo->frags[0], eat);
4301 skb_frag_size_sub(&skbinfo->frags[0], eat);
4302 skb->data_len -= eat;
4307 __skb_pull(skb, offset);
4309 if (NAPI_GRO_CB(p)->last == p)
4310 skb_shinfo(p)->frag_list = skb;
4312 NAPI_GRO_CB(p)->last->next = skb;
4313 NAPI_GRO_CB(p)->last = skb;
4314 __skb_header_release(skb);
4318 NAPI_GRO_CB(p)->count++;
4320 p->truesize += delta_truesize;
4323 lp->data_len += len;
4324 lp->truesize += delta_truesize;
4327 NAPI_GRO_CB(skb)->same_flow = 1;
4331 #ifdef CONFIG_SKB_EXTENSIONS
4332 #define SKB_EXT_ALIGN_VALUE 8
4333 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4335 static const u8 skb_ext_type_len[] = {
4336 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4337 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4340 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4342 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4343 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4345 #if IS_ENABLED(CONFIG_MPTCP)
4346 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4350 static __always_inline unsigned int skb_ext_total_length(void)
4352 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4353 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4354 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4357 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4359 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4360 skb_ext_type_len[TC_SKB_EXT] +
4362 #if IS_ENABLED(CONFIG_MPTCP)
4363 skb_ext_type_len[SKB_EXT_MPTCP] +
4368 static void skb_extensions_init(void)
4370 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4371 BUILD_BUG_ON(skb_ext_total_length() > 255);
4373 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4374 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4376 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4380 static void skb_extensions_init(void) {}
4383 void __init skb_init(void)
4385 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4386 sizeof(struct sk_buff),
4388 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4389 offsetof(struct sk_buff, cb),
4390 sizeof_field(struct sk_buff, cb),
4392 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4393 sizeof(struct sk_buff_fclones),
4395 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4397 skb_extensions_init();
4401 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4402 unsigned int recursion_level)
4404 int start = skb_headlen(skb);
4405 int i, copy = start - offset;
4406 struct sk_buff *frag_iter;
4409 if (unlikely(recursion_level >= 24))
4415 sg_set_buf(sg, skb->data + offset, copy);
4417 if ((len -= copy) == 0)
4422 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4425 WARN_ON(start > offset + len);
4427 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4428 if ((copy = end - offset) > 0) {
4429 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4430 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4435 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4436 skb_frag_off(frag) + offset - start);
4445 skb_walk_frags(skb, frag_iter) {
4448 WARN_ON(start > offset + len);
4450 end = start + frag_iter->len;
4451 if ((copy = end - offset) > 0) {
4452 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4457 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4458 copy, recursion_level + 1);
4459 if (unlikely(ret < 0))
4462 if ((len -= copy) == 0)
4473 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4474 * @skb: Socket buffer containing the buffers to be mapped
4475 * @sg: The scatter-gather list to map into
4476 * @offset: The offset into the buffer's contents to start mapping
4477 * @len: Length of buffer space to be mapped
4479 * Fill the specified scatter-gather list with mappings/pointers into a
4480 * region of the buffer space attached to a socket buffer. Returns either
4481 * the number of scatterlist items used, or -EMSGSIZE if the contents
4484 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4486 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4491 sg_mark_end(&sg[nsg - 1]);
4495 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4497 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4498 * sglist without mark the sg which contain last skb data as the end.
4499 * So the caller can mannipulate sg list as will when padding new data after
4500 * the first call without calling sg_unmark_end to expend sg list.
4502 * Scenario to use skb_to_sgvec_nomark:
4504 * 2. skb_to_sgvec_nomark(payload1)
4505 * 3. skb_to_sgvec_nomark(payload2)
4507 * This is equivalent to:
4509 * 2. skb_to_sgvec(payload1)
4511 * 4. skb_to_sgvec(payload2)
4513 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4514 * is more preferable.
4516 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4517 int offset, int len)
4519 return __skb_to_sgvec(skb, sg, offset, len, 0);
4521 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4526 * skb_cow_data - Check that a socket buffer's data buffers are writable
4527 * @skb: The socket buffer to check.
4528 * @tailbits: Amount of trailing space to be added
4529 * @trailer: Returned pointer to the skb where the @tailbits space begins
4531 * Make sure that the data buffers attached to a socket buffer are
4532 * writable. If they are not, private copies are made of the data buffers
4533 * and the socket buffer is set to use these instead.
4535 * If @tailbits is given, make sure that there is space to write @tailbits
4536 * bytes of data beyond current end of socket buffer. @trailer will be
4537 * set to point to the skb in which this space begins.
4539 * The number of scatterlist elements required to completely map the
4540 * COW'd and extended socket buffer will be returned.
4542 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4546 struct sk_buff *skb1, **skb_p;
4548 /* If skb is cloned or its head is paged, reallocate
4549 * head pulling out all the pages (pages are considered not writable
4550 * at the moment even if they are anonymous).
4552 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4553 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4556 /* Easy case. Most of packets will go this way. */
4557 if (!skb_has_frag_list(skb)) {
4558 /* A little of trouble, not enough of space for trailer.
4559 * This should not happen, when stack is tuned to generate
4560 * good frames. OK, on miss we reallocate and reserve even more
4561 * space, 128 bytes is fair. */
4563 if (skb_tailroom(skb) < tailbits &&
4564 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4572 /* Misery. We are in troubles, going to mincer fragments... */
4575 skb_p = &skb_shinfo(skb)->frag_list;
4578 while ((skb1 = *skb_p) != NULL) {
4581 /* The fragment is partially pulled by someone,
4582 * this can happen on input. Copy it and everything
4585 if (skb_shared(skb1))
4588 /* If the skb is the last, worry about trailer. */
4590 if (skb1->next == NULL && tailbits) {
4591 if (skb_shinfo(skb1)->nr_frags ||
4592 skb_has_frag_list(skb1) ||
4593 skb_tailroom(skb1) < tailbits)
4594 ntail = tailbits + 128;
4600 skb_shinfo(skb1)->nr_frags ||
4601 skb_has_frag_list(skb1)) {
4602 struct sk_buff *skb2;
4604 /* Fuck, we are miserable poor guys... */
4606 skb2 = skb_copy(skb1, GFP_ATOMIC);
4608 skb2 = skb_copy_expand(skb1,
4612 if (unlikely(skb2 == NULL))
4616 skb_set_owner_w(skb2, skb1->sk);
4618 /* Looking around. Are we still alive?
4619 * OK, link new skb, drop old one */
4621 skb2->next = skb1->next;
4628 skb_p = &skb1->next;
4633 EXPORT_SYMBOL_GPL(skb_cow_data);
4635 static void sock_rmem_free(struct sk_buff *skb)
4637 struct sock *sk = skb->sk;
4639 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4642 static void skb_set_err_queue(struct sk_buff *skb)
4644 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4645 * So, it is safe to (mis)use it to mark skbs on the error queue.
4647 skb->pkt_type = PACKET_OUTGOING;
4648 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4652 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4654 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4656 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4657 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4662 skb->destructor = sock_rmem_free;
4663 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4664 skb_set_err_queue(skb);
4666 /* before exiting rcu section, make sure dst is refcounted */
4669 skb_queue_tail(&sk->sk_error_queue, skb);
4670 if (!sock_flag(sk, SOCK_DEAD))
4671 sk->sk_error_report(sk);
4674 EXPORT_SYMBOL(sock_queue_err_skb);
4676 static bool is_icmp_err_skb(const struct sk_buff *skb)
4678 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4679 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4682 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4684 struct sk_buff_head *q = &sk->sk_error_queue;
4685 struct sk_buff *skb, *skb_next = NULL;
4686 bool icmp_next = false;
4687 unsigned long flags;
4689 spin_lock_irqsave(&q->lock, flags);
4690 skb = __skb_dequeue(q);
4691 if (skb && (skb_next = skb_peek(q))) {
4692 icmp_next = is_icmp_err_skb(skb_next);
4694 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
4696 spin_unlock_irqrestore(&q->lock, flags);
4698 if (is_icmp_err_skb(skb) && !icmp_next)
4702 sk->sk_error_report(sk);
4706 EXPORT_SYMBOL(sock_dequeue_err_skb);
4709 * skb_clone_sk - create clone of skb, and take reference to socket
4710 * @skb: the skb to clone
4712 * This function creates a clone of a buffer that holds a reference on
4713 * sk_refcnt. Buffers created via this function are meant to be
4714 * returned using sock_queue_err_skb, or free via kfree_skb.
4716 * When passing buffers allocated with this function to sock_queue_err_skb
4717 * it is necessary to wrap the call with sock_hold/sock_put in order to
4718 * prevent the socket from being released prior to being enqueued on
4719 * the sk_error_queue.
4721 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4723 struct sock *sk = skb->sk;
4724 struct sk_buff *clone;
4726 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4729 clone = skb_clone(skb, GFP_ATOMIC);
4736 clone->destructor = sock_efree;
4740 EXPORT_SYMBOL(skb_clone_sk);
4742 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4747 struct sock_exterr_skb *serr;
4750 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4752 serr = SKB_EXT_ERR(skb);
4753 memset(serr, 0, sizeof(*serr));
4754 serr->ee.ee_errno = ENOMSG;
4755 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4756 serr->ee.ee_info = tstype;
4757 serr->opt_stats = opt_stats;
4758 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4759 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4760 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4761 if (sk->sk_protocol == IPPROTO_TCP &&
4762 sk->sk_type == SOCK_STREAM)
4763 serr->ee.ee_data -= sk->sk_tskey;
4766 err = sock_queue_err_skb(sk, skb);
4772 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4776 if (likely(sysctl_tstamp_allow_data || tsonly))
4779 read_lock_bh(&sk->sk_callback_lock);
4780 ret = sk->sk_socket && sk->sk_socket->file &&
4781 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4782 read_unlock_bh(&sk->sk_callback_lock);
4786 void skb_complete_tx_timestamp(struct sk_buff *skb,
4787 struct skb_shared_hwtstamps *hwtstamps)
4789 struct sock *sk = skb->sk;
4791 if (!skb_may_tx_timestamp(sk, false))
4794 /* Take a reference to prevent skb_orphan() from freeing the socket,
4795 * but only if the socket refcount is not zero.
4797 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4798 *skb_hwtstamps(skb) = *hwtstamps;
4799 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4807 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4809 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4810 const struct sk_buff *ack_skb,
4811 struct skb_shared_hwtstamps *hwtstamps,
4812 struct sock *sk, int tstype)
4814 struct sk_buff *skb;
4815 bool tsonly, opt_stats = false;
4820 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4821 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4824 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4825 if (!skb_may_tx_timestamp(sk, tsonly))
4830 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4831 sk->sk_protocol == IPPROTO_TCP &&
4832 sk->sk_type == SOCK_STREAM) {
4833 skb = tcp_get_timestamping_opt_stats(sk, orig_skb,
4838 skb = alloc_skb(0, GFP_ATOMIC);
4840 skb = skb_clone(orig_skb, GFP_ATOMIC);
4846 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4848 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4852 *skb_hwtstamps(skb) = *hwtstamps;
4854 skb->tstamp = ktime_get_real();
4856 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4858 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4860 void skb_tstamp_tx(struct sk_buff *orig_skb,
4861 struct skb_shared_hwtstamps *hwtstamps)
4863 return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk,
4866 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4868 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4870 struct sock *sk = skb->sk;
4871 struct sock_exterr_skb *serr;
4874 skb->wifi_acked_valid = 1;
4875 skb->wifi_acked = acked;
4877 serr = SKB_EXT_ERR(skb);
4878 memset(serr, 0, sizeof(*serr));
4879 serr->ee.ee_errno = ENOMSG;
4880 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4882 /* Take a reference to prevent skb_orphan() from freeing the socket,
4883 * but only if the socket refcount is not zero.
4885 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4886 err = sock_queue_err_skb(sk, skb);
4892 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4895 * skb_partial_csum_set - set up and verify partial csum values for packet
4896 * @skb: the skb to set
4897 * @start: the number of bytes after skb->data to start checksumming.
4898 * @off: the offset from start to place the checksum.
4900 * For untrusted partially-checksummed packets, we need to make sure the values
4901 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4903 * This function checks and sets those values and skb->ip_summed: if this
4904 * returns false you should drop the packet.
4906 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4908 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4909 u32 csum_start = skb_headroom(skb) + (u32)start;
4911 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4912 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4913 start, off, skb_headroom(skb), skb_headlen(skb));
4916 skb->ip_summed = CHECKSUM_PARTIAL;
4917 skb->csum_start = csum_start;
4918 skb->csum_offset = off;
4919 skb_set_transport_header(skb, start);
4922 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4924 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4927 if (skb_headlen(skb) >= len)
4930 /* If we need to pullup then pullup to the max, so we
4931 * won't need to do it again.
4936 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4939 if (skb_headlen(skb) < len)
4945 #define MAX_TCP_HDR_LEN (15 * 4)
4947 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4948 typeof(IPPROTO_IP) proto,
4955 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4956 off + MAX_TCP_HDR_LEN);
4957 if (!err && !skb_partial_csum_set(skb, off,
4958 offsetof(struct tcphdr,
4961 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4964 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4965 off + sizeof(struct udphdr));
4966 if (!err && !skb_partial_csum_set(skb, off,
4967 offsetof(struct udphdr,
4970 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4973 return ERR_PTR(-EPROTO);
4976 /* This value should be large enough to cover a tagged ethernet header plus
4977 * maximally sized IP and TCP or UDP headers.
4979 #define MAX_IP_HDR_LEN 128
4981 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4990 err = skb_maybe_pull_tail(skb,
4991 sizeof(struct iphdr),
4996 if (ip_is_fragment(ip_hdr(skb)))
4999 off = ip_hdrlen(skb);
5006 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
5008 return PTR_ERR(csum);
5011 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
5014 ip_hdr(skb)->protocol, 0);
5021 /* This value should be large enough to cover a tagged ethernet header plus
5022 * an IPv6 header, all options, and a maximal TCP or UDP header.
5024 #define MAX_IPV6_HDR_LEN 256
5026 #define OPT_HDR(type, skb, off) \
5027 (type *)(skb_network_header(skb) + (off))
5029 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
5042 off = sizeof(struct ipv6hdr);
5044 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
5048 nexthdr = ipv6_hdr(skb)->nexthdr;
5050 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
5051 while (off <= len && !done) {
5053 case IPPROTO_DSTOPTS:
5054 case IPPROTO_HOPOPTS:
5055 case IPPROTO_ROUTING: {
5056 struct ipv6_opt_hdr *hp;
5058 err = skb_maybe_pull_tail(skb,
5060 sizeof(struct ipv6_opt_hdr),
5065 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
5066 nexthdr = hp->nexthdr;
5067 off += ipv6_optlen(hp);
5071 struct ip_auth_hdr *hp;
5073 err = skb_maybe_pull_tail(skb,
5075 sizeof(struct ip_auth_hdr),
5080 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
5081 nexthdr = hp->nexthdr;
5082 off += ipv6_authlen(hp);
5085 case IPPROTO_FRAGMENT: {
5086 struct frag_hdr *hp;
5088 err = skb_maybe_pull_tail(skb,
5090 sizeof(struct frag_hdr),
5095 hp = OPT_HDR(struct frag_hdr, skb, off);
5097 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5100 nexthdr = hp->nexthdr;
5101 off += sizeof(struct frag_hdr);
5112 if (!done || fragment)
5115 csum = skb_checksum_setup_ip(skb, nexthdr, off);
5117 return PTR_ERR(csum);
5120 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5121 &ipv6_hdr(skb)->daddr,
5122 skb->len - off, nexthdr, 0);
5130 * skb_checksum_setup - set up partial checksum offset
5131 * @skb: the skb to set up
5132 * @recalculate: if true the pseudo-header checksum will be recalculated
5134 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5138 switch (skb->protocol) {
5139 case htons(ETH_P_IP):
5140 err = skb_checksum_setup_ipv4(skb, recalculate);
5143 case htons(ETH_P_IPV6):
5144 err = skb_checksum_setup_ipv6(skb, recalculate);
5154 EXPORT_SYMBOL(skb_checksum_setup);
5157 * skb_checksum_maybe_trim - maybe trims the given skb
5158 * @skb: the skb to check
5159 * @transport_len: the data length beyond the network header
5161 * Checks whether the given skb has data beyond the given transport length.
5162 * If so, returns a cloned skb trimmed to this transport length.
5163 * Otherwise returns the provided skb. Returns NULL in error cases
5164 * (e.g. transport_len exceeds skb length or out-of-memory).
5166 * Caller needs to set the skb transport header and free any returned skb if it
5167 * differs from the provided skb.
5169 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5170 unsigned int transport_len)
5172 struct sk_buff *skb_chk;
5173 unsigned int len = skb_transport_offset(skb) + transport_len;
5178 else if (skb->len == len)
5181 skb_chk = skb_clone(skb, GFP_ATOMIC);
5185 ret = pskb_trim_rcsum(skb_chk, len);
5195 * skb_checksum_trimmed - validate checksum of an skb
5196 * @skb: the skb to check
5197 * @transport_len: the data length beyond the network header
5198 * @skb_chkf: checksum function to use
5200 * Applies the given checksum function skb_chkf to the provided skb.
5201 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5203 * If the skb has data beyond the given transport length, then a
5204 * trimmed & cloned skb is checked and returned.
5206 * Caller needs to set the skb transport header and free any returned skb if it
5207 * differs from the provided skb.
5209 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5210 unsigned int transport_len,
5211 __sum16(*skb_chkf)(struct sk_buff *skb))
5213 struct sk_buff *skb_chk;
5214 unsigned int offset = skb_transport_offset(skb);
5217 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5221 if (!pskb_may_pull(skb_chk, offset))
5224 skb_pull_rcsum(skb_chk, offset);
5225 ret = skb_chkf(skb_chk);
5226 skb_push_rcsum(skb_chk, offset);
5234 if (skb_chk && skb_chk != skb)
5240 EXPORT_SYMBOL(skb_checksum_trimmed);
5242 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5244 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5247 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5249 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5252 skb_release_head_state(skb);
5253 kmem_cache_free(skbuff_head_cache, skb);
5258 EXPORT_SYMBOL(kfree_skb_partial);
5261 * skb_try_coalesce - try to merge skb to prior one
5263 * @from: buffer to add
5264 * @fragstolen: pointer to boolean
5265 * @delta_truesize: how much more was allocated than was requested
5267 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5268 bool *fragstolen, int *delta_truesize)
5270 struct skb_shared_info *to_shinfo, *from_shinfo;
5271 int i, delta, len = from->len;
5273 *fragstolen = false;
5278 if (len <= skb_tailroom(to)) {
5280 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5281 *delta_truesize = 0;
5285 to_shinfo = skb_shinfo(to);
5286 from_shinfo = skb_shinfo(from);
5287 if (to_shinfo->frag_list || from_shinfo->frag_list)
5289 if (skb_zcopy(to) || skb_zcopy(from))
5292 if (skb_headlen(from) != 0) {
5294 unsigned int offset;
5296 if (to_shinfo->nr_frags +
5297 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5300 if (skb_head_is_locked(from))
5303 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5305 page = virt_to_head_page(from->head);
5306 offset = from->data - (unsigned char *)page_address(page);
5308 skb_fill_page_desc(to, to_shinfo->nr_frags,
5309 page, offset, skb_headlen(from));
5312 if (to_shinfo->nr_frags +
5313 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5316 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5319 WARN_ON_ONCE(delta < len);
5321 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5323 from_shinfo->nr_frags * sizeof(skb_frag_t));
5324 to_shinfo->nr_frags += from_shinfo->nr_frags;
5326 if (!skb_cloned(from))
5327 from_shinfo->nr_frags = 0;
5329 /* if the skb is not cloned this does nothing
5330 * since we set nr_frags to 0.
5332 for (i = 0; i < from_shinfo->nr_frags; i++)
5333 __skb_frag_ref(&from_shinfo->frags[i]);
5335 to->truesize += delta;
5337 to->data_len += len;
5339 *delta_truesize = delta;
5342 EXPORT_SYMBOL(skb_try_coalesce);
5345 * skb_scrub_packet - scrub an skb
5347 * @skb: buffer to clean
5348 * @xnet: packet is crossing netns
5350 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5351 * into/from a tunnel. Some information have to be cleared during these
5353 * skb_scrub_packet can also be used to clean a skb before injecting it in
5354 * another namespace (@xnet == true). We have to clear all information in the
5355 * skb that could impact namespace isolation.
5357 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5359 skb->pkt_type = PACKET_HOST;
5365 nf_reset_trace(skb);
5367 #ifdef CONFIG_NET_SWITCHDEV
5368 skb->offload_fwd_mark = 0;
5369 skb->offload_l3_fwd_mark = 0;
5379 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5382 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5386 * skb_gso_transport_seglen is used to determine the real size of the
5387 * individual segments, including Layer4 headers (TCP/UDP).
5389 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5391 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5393 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5394 unsigned int thlen = 0;
5396 if (skb->encapsulation) {
5397 thlen = skb_inner_transport_header(skb) -
5398 skb_transport_header(skb);
5400 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5401 thlen += inner_tcp_hdrlen(skb);
5402 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5403 thlen = tcp_hdrlen(skb);
5404 } else if (unlikely(skb_is_gso_sctp(skb))) {
5405 thlen = sizeof(struct sctphdr);
5406 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5407 thlen = sizeof(struct udphdr);
5409 /* UFO sets gso_size to the size of the fragmentation
5410 * payload, i.e. the size of the L4 (UDP) header is already
5413 return thlen + shinfo->gso_size;
5417 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5421 * skb_gso_network_seglen is used to determine the real size of the
5422 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5424 * The MAC/L2 header is not accounted for.
5426 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5428 unsigned int hdr_len = skb_transport_header(skb) -
5429 skb_network_header(skb);
5431 return hdr_len + skb_gso_transport_seglen(skb);
5435 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5439 * skb_gso_mac_seglen is used to determine the real size of the
5440 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5441 * headers (TCP/UDP).
5443 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5445 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5447 return hdr_len + skb_gso_transport_seglen(skb);
5451 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5453 * There are a couple of instances where we have a GSO skb, and we
5454 * want to determine what size it would be after it is segmented.
5456 * We might want to check:
5457 * - L3+L4+payload size (e.g. IP forwarding)
5458 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5460 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5464 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5465 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5467 * @max_len: The maximum permissible length.
5469 * Returns true if the segmented length <= max length.
5471 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5472 unsigned int seg_len,
5473 unsigned int max_len) {
5474 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5475 const struct sk_buff *iter;
5477 if (shinfo->gso_size != GSO_BY_FRAGS)
5478 return seg_len <= max_len;
5480 /* Undo this so we can re-use header sizes */
5481 seg_len -= GSO_BY_FRAGS;
5483 skb_walk_frags(skb, iter) {
5484 if (seg_len + skb_headlen(iter) > max_len)
5492 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5495 * @mtu: MTU to validate against
5497 * skb_gso_validate_network_len validates if a given skb will fit a
5498 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5501 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5503 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5505 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5508 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5511 * @len: length to validate against
5513 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5514 * length once split, including L2, L3 and L4 headers and the payload.
5516 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5518 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5520 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5522 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5524 int mac_len, meta_len;
5527 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5532 mac_len = skb->data - skb_mac_header(skb);
5533 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5534 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5535 mac_len - VLAN_HLEN - ETH_TLEN);
5538 meta_len = skb_metadata_len(skb);
5540 meta = skb_metadata_end(skb) - meta_len;
5541 memmove(meta + VLAN_HLEN, meta, meta_len);
5544 skb->mac_header += VLAN_HLEN;
5548 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5550 struct vlan_hdr *vhdr;
5553 if (unlikely(skb_vlan_tag_present(skb))) {
5554 /* vlan_tci is already set-up so leave this for another time */
5558 skb = skb_share_check(skb, GFP_ATOMIC);
5561 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5562 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5565 vhdr = (struct vlan_hdr *)skb->data;
5566 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5567 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5569 skb_pull_rcsum(skb, VLAN_HLEN);
5570 vlan_set_encap_proto(skb, vhdr);
5572 skb = skb_reorder_vlan_header(skb);
5576 skb_reset_network_header(skb);
5577 if (!skb_transport_header_was_set(skb))
5578 skb_reset_transport_header(skb);
5579 skb_reset_mac_len(skb);
5587 EXPORT_SYMBOL(skb_vlan_untag);
5589 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5591 if (!pskb_may_pull(skb, write_len))
5594 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5597 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5599 EXPORT_SYMBOL(skb_ensure_writable);
5601 /* remove VLAN header from packet and update csum accordingly.
5602 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5604 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5606 struct vlan_hdr *vhdr;
5607 int offset = skb->data - skb_mac_header(skb);
5610 if (WARN_ONCE(offset,
5611 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5616 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5620 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5622 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5623 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5625 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5626 __skb_pull(skb, VLAN_HLEN);
5628 vlan_set_encap_proto(skb, vhdr);
5629 skb->mac_header += VLAN_HLEN;
5631 if (skb_network_offset(skb) < ETH_HLEN)
5632 skb_set_network_header(skb, ETH_HLEN);
5634 skb_reset_mac_len(skb);
5638 EXPORT_SYMBOL(__skb_vlan_pop);
5640 /* Pop a vlan tag either from hwaccel or from payload.
5641 * Expects skb->data at mac header.
5643 int skb_vlan_pop(struct sk_buff *skb)
5649 if (likely(skb_vlan_tag_present(skb))) {
5650 __vlan_hwaccel_clear_tag(skb);
5652 if (unlikely(!eth_type_vlan(skb->protocol)))
5655 err = __skb_vlan_pop(skb, &vlan_tci);
5659 /* move next vlan tag to hw accel tag */
5660 if (likely(!eth_type_vlan(skb->protocol)))
5663 vlan_proto = skb->protocol;
5664 err = __skb_vlan_pop(skb, &vlan_tci);
5668 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5671 EXPORT_SYMBOL(skb_vlan_pop);
5673 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5674 * Expects skb->data at mac header.
5676 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5678 if (skb_vlan_tag_present(skb)) {
5679 int offset = skb->data - skb_mac_header(skb);
5682 if (WARN_ONCE(offset,
5683 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5688 err = __vlan_insert_tag(skb, skb->vlan_proto,
5689 skb_vlan_tag_get(skb));
5693 skb->protocol = skb->vlan_proto;
5694 skb->mac_len += VLAN_HLEN;
5696 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5698 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5701 EXPORT_SYMBOL(skb_vlan_push);
5704 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
5706 * @skb: Socket buffer to modify
5708 * Drop the Ethernet header of @skb.
5710 * Expects that skb->data points to the mac header and that no VLAN tags are
5713 * Returns 0 on success, -errno otherwise.
5715 int skb_eth_pop(struct sk_buff *skb)
5717 if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
5718 skb_network_offset(skb) < ETH_HLEN)
5721 skb_pull_rcsum(skb, ETH_HLEN);
5722 skb_reset_mac_header(skb);
5723 skb_reset_mac_len(skb);
5727 EXPORT_SYMBOL(skb_eth_pop);
5730 * skb_eth_push() - Add a new Ethernet header at the head of a packet
5732 * @skb: Socket buffer to modify
5733 * @dst: Destination MAC address of the new header
5734 * @src: Source MAC address of the new header
5736 * Prepend @skb with a new Ethernet header.
5738 * Expects that skb->data points to the mac header, which must be empty.
5740 * Returns 0 on success, -errno otherwise.
5742 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
5743 const unsigned char *src)
5748 if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
5751 err = skb_cow_head(skb, sizeof(*eth));
5755 skb_push(skb, sizeof(*eth));
5756 skb_reset_mac_header(skb);
5757 skb_reset_mac_len(skb);
5760 ether_addr_copy(eth->h_dest, dst);
5761 ether_addr_copy(eth->h_source, src);
5762 eth->h_proto = skb->protocol;
5764 skb_postpush_rcsum(skb, eth, sizeof(*eth));
5768 EXPORT_SYMBOL(skb_eth_push);
5770 /* Update the ethertype of hdr and the skb csum value if required. */
5771 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5774 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5775 __be16 diff[] = { ~hdr->h_proto, ethertype };
5777 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5780 hdr->h_proto = ethertype;
5784 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
5788 * @mpls_lse: MPLS label stack entry to push
5789 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5790 * @mac_len: length of the MAC header
5791 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
5794 * Expects skb->data at mac header.
5796 * Returns 0 on success, -errno otherwise.
5798 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5799 int mac_len, bool ethernet)
5801 struct mpls_shim_hdr *lse;
5804 if (unlikely(!eth_p_mpls(mpls_proto)))
5807 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5808 if (skb->encapsulation)
5811 err = skb_cow_head(skb, MPLS_HLEN);
5815 if (!skb->inner_protocol) {
5816 skb_set_inner_network_header(skb, skb_network_offset(skb));
5817 skb_set_inner_protocol(skb, skb->protocol);
5820 skb_push(skb, MPLS_HLEN);
5821 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5823 skb_reset_mac_header(skb);
5824 skb_set_network_header(skb, mac_len);
5825 skb_reset_mac_len(skb);
5827 lse = mpls_hdr(skb);
5828 lse->label_stack_entry = mpls_lse;
5829 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5831 if (ethernet && mac_len >= ETH_HLEN)
5832 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5833 skb->protocol = mpls_proto;
5837 EXPORT_SYMBOL_GPL(skb_mpls_push);
5840 * skb_mpls_pop() - pop the outermost MPLS header
5843 * @next_proto: ethertype of header after popped MPLS header
5844 * @mac_len: length of the MAC header
5845 * @ethernet: flag to indicate if the packet is ethernet
5847 * Expects skb->data at mac header.
5849 * Returns 0 on success, -errno otherwise.
5851 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5856 if (unlikely(!eth_p_mpls(skb->protocol)))
5859 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5863 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5864 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5867 __skb_pull(skb, MPLS_HLEN);
5868 skb_reset_mac_header(skb);
5869 skb_set_network_header(skb, mac_len);
5871 if (ethernet && mac_len >= ETH_HLEN) {
5874 /* use mpls_hdr() to get ethertype to account for VLANs. */
5875 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5876 skb_mod_eth_type(skb, hdr, next_proto);
5878 skb->protocol = next_proto;
5882 EXPORT_SYMBOL_GPL(skb_mpls_pop);
5885 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5888 * @mpls_lse: new MPLS label stack entry to update to
5890 * Expects skb->data at mac header.
5892 * Returns 0 on success, -errno otherwise.
5894 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5898 if (unlikely(!eth_p_mpls(skb->protocol)))
5901 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5905 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5906 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5908 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5911 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5915 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5918 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5922 * Expects skb->data at mac header.
5924 * Returns 0 on success, -errno otherwise.
5926 int skb_mpls_dec_ttl(struct sk_buff *skb)
5931 if (unlikely(!eth_p_mpls(skb->protocol)))
5934 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
5937 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5938 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5942 lse &= ~MPLS_LS_TTL_MASK;
5943 lse |= ttl << MPLS_LS_TTL_SHIFT;
5945 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5947 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5950 * alloc_skb_with_frags - allocate skb with page frags
5952 * @header_len: size of linear part
5953 * @data_len: needed length in frags
5954 * @max_page_order: max page order desired.
5955 * @errcode: pointer to error code if any
5956 * @gfp_mask: allocation mask
5958 * This can be used to allocate a paged skb, given a maximal order for frags.
5960 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5961 unsigned long data_len,
5966 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5967 unsigned long chunk;
5968 struct sk_buff *skb;
5972 *errcode = -EMSGSIZE;
5973 /* Note this test could be relaxed, if we succeed to allocate
5974 * high order pages...
5976 if (npages > MAX_SKB_FRAGS)
5979 *errcode = -ENOBUFS;
5980 skb = alloc_skb(header_len, gfp_mask);
5984 skb->truesize += npages << PAGE_SHIFT;
5986 for (i = 0; npages > 0; i++) {
5987 int order = max_page_order;
5990 if (npages >= 1 << order) {
5991 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5997 /* Do not retry other high order allocations */
6003 page = alloc_page(gfp_mask);
6007 chunk = min_t(unsigned long, data_len,
6008 PAGE_SIZE << order);
6009 skb_fill_page_desc(skb, i, page, 0, chunk);
6011 npages -= 1 << order;
6019 EXPORT_SYMBOL(alloc_skb_with_frags);
6021 /* carve out the first off bytes from skb when off < headlen */
6022 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
6023 const int headlen, gfp_t gfp_mask)
6026 int size = skb_end_offset(skb);
6027 int new_hlen = headlen - off;
6030 size = SKB_DATA_ALIGN(size);
6032 if (skb_pfmemalloc(skb))
6033 gfp_mask |= __GFP_MEMALLOC;
6034 data = kmalloc_reserve(size +
6035 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6036 gfp_mask, NUMA_NO_NODE, NULL);
6040 size = SKB_WITH_OVERHEAD(ksize(data));
6042 /* Copy real data, and all frags */
6043 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
6046 memcpy((struct skb_shared_info *)(data + size),
6048 offsetof(struct skb_shared_info,
6049 frags[skb_shinfo(skb)->nr_frags]));
6050 if (skb_cloned(skb)) {
6051 /* drop the old head gracefully */
6052 if (skb_orphan_frags(skb, gfp_mask)) {
6056 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
6057 skb_frag_ref(skb, i);
6058 if (skb_has_frag_list(skb))
6059 skb_clone_fraglist(skb);
6060 skb_release_data(skb);
6062 /* we can reuse existing recount- all we did was
6071 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6074 skb->end = skb->head + size;
6076 skb_set_tail_pointer(skb, skb_headlen(skb));
6077 skb_headers_offset_update(skb, 0);
6081 atomic_set(&skb_shinfo(skb)->dataref, 1);
6086 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6088 /* carve out the first eat bytes from skb's frag_list. May recurse into
6091 static int pskb_carve_frag_list(struct sk_buff *skb,
6092 struct skb_shared_info *shinfo, int eat,
6095 struct sk_buff *list = shinfo->frag_list;
6096 struct sk_buff *clone = NULL;
6097 struct sk_buff *insp = NULL;
6101 pr_err("Not enough bytes to eat. Want %d\n", eat);
6104 if (list->len <= eat) {
6105 /* Eaten as whole. */
6110 /* Eaten partially. */
6111 if (skb_shared(list)) {
6112 clone = skb_clone(list, gfp_mask);
6118 /* This may be pulled without problems. */
6121 if (pskb_carve(list, eat, gfp_mask) < 0) {
6129 /* Free pulled out fragments. */
6130 while ((list = shinfo->frag_list) != insp) {
6131 shinfo->frag_list = list->next;
6134 /* And insert new clone at head. */
6137 shinfo->frag_list = clone;
6142 /* carve off first len bytes from skb. Split line (off) is in the
6143 * non-linear part of skb
6145 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6146 int pos, gfp_t gfp_mask)
6149 int size = skb_end_offset(skb);
6151 const int nfrags = skb_shinfo(skb)->nr_frags;
6152 struct skb_shared_info *shinfo;
6154 size = SKB_DATA_ALIGN(size);
6156 if (skb_pfmemalloc(skb))
6157 gfp_mask |= __GFP_MEMALLOC;
6158 data = kmalloc_reserve(size +
6159 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6160 gfp_mask, NUMA_NO_NODE, NULL);
6164 size = SKB_WITH_OVERHEAD(ksize(data));
6166 memcpy((struct skb_shared_info *)(data + size),
6167 skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6168 if (skb_orphan_frags(skb, gfp_mask)) {
6172 shinfo = (struct skb_shared_info *)(data + size);
6173 for (i = 0; i < nfrags; i++) {
6174 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6176 if (pos + fsize > off) {
6177 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6181 * We have two variants in this case:
6182 * 1. Move all the frag to the second
6183 * part, if it is possible. F.e.
6184 * this approach is mandatory for TUX,
6185 * where splitting is expensive.
6186 * 2. Split is accurately. We make this.
6188 skb_frag_off_add(&shinfo->frags[0], off - pos);
6189 skb_frag_size_sub(&shinfo->frags[0], off - pos);
6191 skb_frag_ref(skb, i);
6196 shinfo->nr_frags = k;
6197 if (skb_has_frag_list(skb))
6198 skb_clone_fraglist(skb);
6200 /* split line is in frag list */
6201 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6202 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6203 if (skb_has_frag_list(skb))
6204 kfree_skb_list(skb_shinfo(skb)->frag_list);
6208 skb_release_data(skb);
6213 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6216 skb->end = skb->head + size;
6218 skb_reset_tail_pointer(skb);
6219 skb_headers_offset_update(skb, 0);
6224 skb->data_len = skb->len;
6225 atomic_set(&skb_shinfo(skb)->dataref, 1);
6229 /* remove len bytes from the beginning of the skb */
6230 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6232 int headlen = skb_headlen(skb);
6235 return pskb_carve_inside_header(skb, len, headlen, gfp);
6237 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6240 /* Extract to_copy bytes starting at off from skb, and return this in
6243 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6244 int to_copy, gfp_t gfp)
6246 struct sk_buff *clone = skb_clone(skb, gfp);
6251 if (pskb_carve(clone, off, gfp) < 0 ||
6252 pskb_trim(clone, to_copy)) {
6258 EXPORT_SYMBOL(pskb_extract);
6261 * skb_condense - try to get rid of fragments/frag_list if possible
6264 * Can be used to save memory before skb is added to a busy queue.
6265 * If packet has bytes in frags and enough tail room in skb->head,
6266 * pull all of them, so that we can free the frags right now and adjust
6269 * We do not reallocate skb->head thus can not fail.
6270 * Caller must re-evaluate skb->truesize if needed.
6272 void skb_condense(struct sk_buff *skb)
6274 if (skb->data_len) {
6275 if (skb->data_len > skb->end - skb->tail ||
6279 /* Nice, we can free page frag(s) right now */
6280 __pskb_pull_tail(skb, skb->data_len);
6282 /* At this point, skb->truesize might be over estimated,
6283 * because skb had a fragment, and fragments do not tell
6285 * When we pulled its content into skb->head, fragment
6286 * was freed, but __pskb_pull_tail() could not possibly
6287 * adjust skb->truesize, not knowing the frag truesize.
6289 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6292 #ifdef CONFIG_SKB_EXTENSIONS
6293 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6295 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6299 * __skb_ext_alloc - allocate a new skb extensions storage
6301 * @flags: See kmalloc().
6303 * Returns the newly allocated pointer. The pointer can later attached to a
6304 * skb via __skb_ext_set().
6305 * Note: caller must handle the skb_ext as an opaque data.
6307 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6309 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6312 memset(new->offset, 0, sizeof(new->offset));
6313 refcount_set(&new->refcnt, 1);
6319 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6320 unsigned int old_active)
6322 struct skb_ext *new;
6324 if (refcount_read(&old->refcnt) == 1)
6327 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6331 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6332 refcount_set(&new->refcnt, 1);
6335 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6336 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6339 for (i = 0; i < sp->len; i++)
6340 xfrm_state_hold(sp->xvec[i]);
6348 * __skb_ext_set - attach the specified extension storage to this skb
6351 * @ext: extension storage previously allocated via __skb_ext_alloc()
6353 * Existing extensions, if any, are cleared.
6355 * Returns the pointer to the extension.
6357 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6358 struct skb_ext *ext)
6360 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6363 newlen = newoff + skb_ext_type_len[id];
6364 ext->chunks = newlen;
6365 ext->offset[id] = newoff;
6366 skb->extensions = ext;
6367 skb->active_extensions = 1 << id;
6368 return skb_ext_get_ptr(ext, id);
6372 * skb_ext_add - allocate space for given extension, COW if needed
6374 * @id: extension to allocate space for
6376 * Allocates enough space for the given extension.
6377 * If the extension is already present, a pointer to that extension
6380 * If the skb was cloned, COW applies and the returned memory can be
6381 * modified without changing the extension space of clones buffers.
6383 * Returns pointer to the extension or NULL on allocation failure.
6385 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6387 struct skb_ext *new, *old = NULL;
6388 unsigned int newlen, newoff;
6390 if (skb->active_extensions) {
6391 old = skb->extensions;
6393 new = skb_ext_maybe_cow(old, skb->active_extensions);
6397 if (__skb_ext_exist(new, id))
6400 newoff = new->chunks;
6402 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6404 new = __skb_ext_alloc(GFP_ATOMIC);
6409 newlen = newoff + skb_ext_type_len[id];
6410 new->chunks = newlen;
6411 new->offset[id] = newoff;
6413 skb->extensions = new;
6414 skb->active_extensions |= 1 << id;
6415 return skb_ext_get_ptr(new, id);
6417 EXPORT_SYMBOL(skb_ext_add);
6420 static void skb_ext_put_sp(struct sec_path *sp)
6424 for (i = 0; i < sp->len; i++)
6425 xfrm_state_put(sp->xvec[i]);
6429 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6431 struct skb_ext *ext = skb->extensions;
6433 skb->active_extensions &= ~(1 << id);
6434 if (skb->active_extensions == 0) {
6435 skb->extensions = NULL;
6438 } else if (id == SKB_EXT_SEC_PATH &&
6439 refcount_read(&ext->refcnt) == 1) {
6440 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6447 EXPORT_SYMBOL(__skb_ext_del);
6449 void __skb_ext_put(struct skb_ext *ext)
6451 /* If this is last clone, nothing can increment
6452 * it after check passes. Avoids one atomic op.
6454 if (refcount_read(&ext->refcnt) == 1)
6457 if (!refcount_dec_and_test(&ext->refcnt))
6461 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6462 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6465 kmem_cache_free(skbuff_ext_cache, ext);
6467 EXPORT_SYMBOL(__skb_ext_put);
6468 #endif /* CONFIG_SKB_EXTENSIONS */