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, NUMA_NO_NODE);
571 nc = this_cpu_ptr(&napi_alloc_cache);
572 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
573 len = SKB_DATA_ALIGN(len);
575 if (sk_memalloc_socks())
576 gfp_mask |= __GFP_MEMALLOC;
578 data = page_frag_alloc(&nc->page, len, gfp_mask);
582 skb = __build_skb(data, len);
583 if (unlikely(!skb)) {
588 if (nc->page.pfmemalloc)
593 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
594 skb->dev = napi->dev;
599 EXPORT_SYMBOL(__napi_alloc_skb);
601 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
602 int size, unsigned int truesize)
604 skb_fill_page_desc(skb, i, page, off, size);
606 skb->data_len += size;
607 skb->truesize += truesize;
609 EXPORT_SYMBOL(skb_add_rx_frag);
611 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
612 unsigned int truesize)
614 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
616 skb_frag_size_add(frag, size);
618 skb->data_len += size;
619 skb->truesize += truesize;
621 EXPORT_SYMBOL(skb_coalesce_rx_frag);
623 static void skb_drop_list(struct sk_buff **listp)
625 kfree_skb_list(*listp);
629 static inline void skb_drop_fraglist(struct sk_buff *skb)
631 skb_drop_list(&skb_shinfo(skb)->frag_list);
634 static void skb_clone_fraglist(struct sk_buff *skb)
636 struct sk_buff *list;
638 skb_walk_frags(skb, list)
642 static void skb_free_head(struct sk_buff *skb)
644 unsigned char *head = skb->head;
652 static void skb_release_data(struct sk_buff *skb)
654 struct skb_shared_info *shinfo = skb_shinfo(skb);
658 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
662 skb_zcopy_clear(skb, true);
664 for (i = 0; i < shinfo->nr_frags; i++)
665 __skb_frag_unref(&shinfo->frags[i]);
667 if (shinfo->frag_list)
668 kfree_skb_list(shinfo->frag_list);
674 * Free an skbuff by memory without cleaning the state.
676 static void kfree_skbmem(struct sk_buff *skb)
678 struct sk_buff_fclones *fclones;
680 switch (skb->fclone) {
681 case SKB_FCLONE_UNAVAILABLE:
682 kmem_cache_free(skbuff_head_cache, skb);
685 case SKB_FCLONE_ORIG:
686 fclones = container_of(skb, struct sk_buff_fclones, skb1);
688 /* We usually free the clone (TX completion) before original skb
689 * This test would have no chance to be true for the clone,
690 * while here, branch prediction will be good.
692 if (refcount_read(&fclones->fclone_ref) == 1)
696 default: /* SKB_FCLONE_CLONE */
697 fclones = container_of(skb, struct sk_buff_fclones, skb2);
700 if (!refcount_dec_and_test(&fclones->fclone_ref))
703 kmem_cache_free(skbuff_fclone_cache, fclones);
706 void skb_release_head_state(struct sk_buff *skb)
709 if (skb->destructor) {
711 skb->destructor(skb);
713 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
714 nf_conntrack_put(skb_nfct(skb));
719 /* Free everything but the sk_buff shell. */
720 static void skb_release_all(struct sk_buff *skb)
722 skb_release_head_state(skb);
723 if (likely(skb->head))
724 skb_release_data(skb);
728 * __kfree_skb - private function
731 * Free an sk_buff. Release anything attached to the buffer.
732 * Clean the state. This is an internal helper function. Users should
733 * always call kfree_skb
736 void __kfree_skb(struct sk_buff *skb)
738 skb_release_all(skb);
741 EXPORT_SYMBOL(__kfree_skb);
744 * kfree_skb - free an sk_buff
745 * @skb: buffer to free
747 * Drop a reference to the buffer and free it if the usage count has
750 void kfree_skb(struct sk_buff *skb)
755 trace_kfree_skb(skb, __builtin_return_address(0));
758 EXPORT_SYMBOL(kfree_skb);
760 void kfree_skb_list(struct sk_buff *segs)
763 struct sk_buff *next = segs->next;
769 EXPORT_SYMBOL(kfree_skb_list);
771 /* Dump skb information and contents.
773 * Must only be called from net_ratelimit()-ed paths.
775 * Dumps whole packets if full_pkt, only headers otherwise.
777 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
779 struct skb_shared_info *sh = skb_shinfo(skb);
780 struct net_device *dev = skb->dev;
781 struct sock *sk = skb->sk;
782 struct sk_buff *list_skb;
783 bool has_mac, has_trans;
784 int headroom, tailroom;
790 len = min_t(int, skb->len, MAX_HEADER + 128);
792 headroom = skb_headroom(skb);
793 tailroom = skb_tailroom(skb);
795 has_mac = skb_mac_header_was_set(skb);
796 has_trans = skb_transport_header_was_set(skb);
798 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
799 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
800 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
801 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
802 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
803 level, skb->len, headroom, skb_headlen(skb), tailroom,
804 has_mac ? skb->mac_header : -1,
805 has_mac ? skb_mac_header_len(skb) : -1,
807 has_trans ? skb_network_header_len(skb) : -1,
808 has_trans ? skb->transport_header : -1,
809 sh->tx_flags, sh->nr_frags,
810 sh->gso_size, sh->gso_type, sh->gso_segs,
811 skb->csum, skb->ip_summed, skb->csum_complete_sw,
812 skb->csum_valid, skb->csum_level,
813 skb->hash, skb->sw_hash, skb->l4_hash,
814 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
817 printk("%sdev name=%s feat=0x%pNF\n",
818 level, dev->name, &dev->features);
820 printk("%ssk family=%hu type=%u proto=%u\n",
821 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
823 if (full_pkt && headroom)
824 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
825 16, 1, skb->head, headroom, false);
827 seg_len = min_t(int, skb_headlen(skb), len);
829 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
830 16, 1, skb->data, seg_len, false);
833 if (full_pkt && tailroom)
834 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
835 16, 1, skb_tail_pointer(skb), tailroom, false);
837 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
838 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
839 u32 p_off, p_len, copied;
843 skb_frag_foreach_page(frag, skb_frag_off(frag),
844 skb_frag_size(frag), p, p_off, p_len,
846 seg_len = min_t(int, p_len, len);
847 vaddr = kmap_atomic(p);
848 print_hex_dump(level, "skb frag: ",
850 16, 1, vaddr + p_off, seg_len, false);
851 kunmap_atomic(vaddr);
858 if (full_pkt && skb_has_frag_list(skb)) {
859 printk("skb fraglist:\n");
860 skb_walk_frags(skb, list_skb)
861 skb_dump(level, list_skb, true);
864 EXPORT_SYMBOL(skb_dump);
867 * skb_tx_error - report an sk_buff xmit error
868 * @skb: buffer that triggered an error
870 * Report xmit error if a device callback is tracking this skb.
871 * skb must be freed afterwards.
873 void skb_tx_error(struct sk_buff *skb)
875 skb_zcopy_clear(skb, true);
877 EXPORT_SYMBOL(skb_tx_error);
879 #ifdef CONFIG_TRACEPOINTS
881 * consume_skb - free an skbuff
882 * @skb: buffer to free
884 * Drop a ref to the buffer and free it if the usage count has hit zero
885 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
886 * is being dropped after a failure and notes that
888 void consume_skb(struct sk_buff *skb)
893 trace_consume_skb(skb);
896 EXPORT_SYMBOL(consume_skb);
900 * __consume_stateless_skb - free an skbuff, assuming it is stateless
901 * @skb: buffer to free
903 * Alike consume_skb(), but this variant assumes that this is the last
904 * skb reference and all the head states have been already dropped
906 void __consume_stateless_skb(struct sk_buff *skb)
908 trace_consume_skb(skb);
909 skb_release_data(skb);
913 static void napi_skb_cache_put(struct sk_buff *skb)
915 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
918 /* drop skb->head and call any destructors for packet */
919 skb_release_all(skb);
921 kasan_poison_object_data(skbuff_head_cache, skb);
922 nc->skb_cache[nc->skb_count++] = skb;
924 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
925 for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++)
926 kasan_unpoison_object_data(skbuff_head_cache,
929 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF,
930 nc->skb_cache + NAPI_SKB_CACHE_HALF);
931 nc->skb_count = NAPI_SKB_CACHE_HALF;
935 void __kfree_skb_defer(struct sk_buff *skb)
937 napi_skb_cache_put(skb);
940 void napi_consume_skb(struct sk_buff *skb, int budget)
942 /* Zero budget indicate non-NAPI context called us, like netpoll */
943 if (unlikely(!budget)) {
944 dev_consume_skb_any(skb);
948 lockdep_assert_in_softirq();
953 /* if reaching here SKB is ready to free */
954 trace_consume_skb(skb);
956 /* if SKB is a clone, don't handle this case */
957 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
962 napi_skb_cache_put(skb);
964 EXPORT_SYMBOL(napi_consume_skb);
966 /* Make sure a field is enclosed inside headers_start/headers_end section */
967 #define CHECK_SKB_FIELD(field) \
968 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
969 offsetof(struct sk_buff, headers_start)); \
970 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
971 offsetof(struct sk_buff, headers_end)); \
973 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
975 new->tstamp = old->tstamp;
976 /* We do not copy old->sk */
978 memcpy(new->cb, old->cb, sizeof(old->cb));
979 skb_dst_copy(new, old);
980 __skb_ext_copy(new, old);
981 __nf_copy(new, old, false);
983 /* Note : this field could be in headers_start/headers_end section
984 * It is not yet because we do not want to have a 16 bit hole
986 new->queue_mapping = old->queue_mapping;
988 memcpy(&new->headers_start, &old->headers_start,
989 offsetof(struct sk_buff, headers_end) -
990 offsetof(struct sk_buff, headers_start));
991 CHECK_SKB_FIELD(protocol);
992 CHECK_SKB_FIELD(csum);
993 CHECK_SKB_FIELD(hash);
994 CHECK_SKB_FIELD(priority);
995 CHECK_SKB_FIELD(skb_iif);
996 CHECK_SKB_FIELD(vlan_proto);
997 CHECK_SKB_FIELD(vlan_tci);
998 CHECK_SKB_FIELD(transport_header);
999 CHECK_SKB_FIELD(network_header);
1000 CHECK_SKB_FIELD(mac_header);
1001 CHECK_SKB_FIELD(inner_protocol);
1002 CHECK_SKB_FIELD(inner_transport_header);
1003 CHECK_SKB_FIELD(inner_network_header);
1004 CHECK_SKB_FIELD(inner_mac_header);
1005 CHECK_SKB_FIELD(mark);
1006 #ifdef CONFIG_NETWORK_SECMARK
1007 CHECK_SKB_FIELD(secmark);
1009 #ifdef CONFIG_NET_RX_BUSY_POLL
1010 CHECK_SKB_FIELD(napi_id);
1013 CHECK_SKB_FIELD(sender_cpu);
1015 #ifdef CONFIG_NET_SCHED
1016 CHECK_SKB_FIELD(tc_index);
1022 * You should not add any new code to this function. Add it to
1023 * __copy_skb_header above instead.
1025 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
1027 #define C(x) n->x = skb->x
1029 n->next = n->prev = NULL;
1031 __copy_skb_header(n, skb);
1036 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1041 n->destructor = NULL;
1048 refcount_set(&n->users, 1);
1050 atomic_inc(&(skb_shinfo(skb)->dataref));
1058 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1059 * @first: first sk_buff of the msg
1061 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1065 n = alloc_skb(0, GFP_ATOMIC);
1069 n->len = first->len;
1070 n->data_len = first->len;
1071 n->truesize = first->truesize;
1073 skb_shinfo(n)->frag_list = first;
1075 __copy_skb_header(n, first);
1076 n->destructor = NULL;
1080 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1083 * skb_morph - morph one skb into another
1084 * @dst: the skb to receive the contents
1085 * @src: the skb to supply the contents
1087 * This is identical to skb_clone except that the target skb is
1088 * supplied by the user.
1090 * The target skb is returned upon exit.
1092 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1094 skb_release_all(dst);
1095 return __skb_clone(dst, src);
1097 EXPORT_SYMBOL_GPL(skb_morph);
1099 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1101 unsigned long max_pg, num_pg, new_pg, old_pg;
1102 struct user_struct *user;
1104 if (capable(CAP_IPC_LOCK) || !size)
1107 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1108 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1109 user = mmp->user ? : current_user();
1112 old_pg = atomic_long_read(&user->locked_vm);
1113 new_pg = old_pg + num_pg;
1114 if (new_pg > max_pg)
1116 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1120 mmp->user = get_uid(user);
1121 mmp->num_pg = num_pg;
1123 mmp->num_pg += num_pg;
1128 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1130 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1133 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1134 free_uid(mmp->user);
1137 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1139 struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size)
1141 struct ubuf_info *uarg;
1142 struct sk_buff *skb;
1144 WARN_ON_ONCE(!in_task());
1146 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1150 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1151 uarg = (void *)skb->cb;
1152 uarg->mmp.user = NULL;
1154 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1159 uarg->callback = msg_zerocopy_callback;
1160 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1162 uarg->bytelen = size;
1164 uarg->flags = SKBFL_ZEROCOPY_FRAG;
1165 refcount_set(&uarg->refcnt, 1);
1170 EXPORT_SYMBOL_GPL(msg_zerocopy_alloc);
1172 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1174 return container_of((void *)uarg, struct sk_buff, cb);
1177 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1178 struct ubuf_info *uarg)
1181 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1184 /* realloc only when socket is locked (TCP, UDP cork),
1185 * so uarg->len and sk_zckey access is serialized
1187 if (!sock_owned_by_user(sk)) {
1192 bytelen = uarg->bytelen + size;
1193 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1194 /* TCP can create new skb to attach new uarg */
1195 if (sk->sk_type == SOCK_STREAM)
1200 next = (u32)atomic_read(&sk->sk_zckey);
1201 if ((u32)(uarg->id + uarg->len) == next) {
1202 if (mm_account_pinned_pages(&uarg->mmp, size))
1205 uarg->bytelen = bytelen;
1206 atomic_set(&sk->sk_zckey, ++next);
1208 /* no extra ref when appending to datagram (MSG_MORE) */
1209 if (sk->sk_type == SOCK_STREAM)
1210 net_zcopy_get(uarg);
1217 return msg_zerocopy_alloc(sk, size);
1219 EXPORT_SYMBOL_GPL(msg_zerocopy_realloc);
1221 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1223 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1227 old_lo = serr->ee.ee_info;
1228 old_hi = serr->ee.ee_data;
1229 sum_len = old_hi - old_lo + 1ULL + len;
1231 if (sum_len >= (1ULL << 32))
1234 if (lo != old_hi + 1)
1237 serr->ee.ee_data += len;
1241 static void __msg_zerocopy_callback(struct ubuf_info *uarg)
1243 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1244 struct sock_exterr_skb *serr;
1245 struct sock *sk = skb->sk;
1246 struct sk_buff_head *q;
1247 unsigned long flags;
1251 mm_unaccount_pinned_pages(&uarg->mmp);
1253 /* if !len, there was only 1 call, and it was aborted
1254 * so do not queue a completion notification
1256 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1261 hi = uarg->id + len - 1;
1263 serr = SKB_EXT_ERR(skb);
1264 memset(serr, 0, sizeof(*serr));
1265 serr->ee.ee_errno = 0;
1266 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1267 serr->ee.ee_data = hi;
1268 serr->ee.ee_info = lo;
1269 if (!uarg->zerocopy)
1270 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1272 q = &sk->sk_error_queue;
1273 spin_lock_irqsave(&q->lock, flags);
1274 tail = skb_peek_tail(q);
1275 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1276 !skb_zerocopy_notify_extend(tail, lo, len)) {
1277 __skb_queue_tail(q, skb);
1280 spin_unlock_irqrestore(&q->lock, flags);
1282 sk->sk_error_report(sk);
1289 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1292 uarg->zerocopy = uarg->zerocopy & success;
1294 if (refcount_dec_and_test(&uarg->refcnt))
1295 __msg_zerocopy_callback(uarg);
1297 EXPORT_SYMBOL_GPL(msg_zerocopy_callback);
1299 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1301 struct sock *sk = skb_from_uarg(uarg)->sk;
1303 atomic_dec(&sk->sk_zckey);
1307 msg_zerocopy_callback(NULL, uarg, true);
1309 EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort);
1311 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1313 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1315 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1317 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1318 struct msghdr *msg, int len,
1319 struct ubuf_info *uarg)
1321 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1322 struct iov_iter orig_iter = msg->msg_iter;
1323 int err, orig_len = skb->len;
1325 /* An skb can only point to one uarg. This edge case happens when
1326 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1328 if (orig_uarg && uarg != orig_uarg)
1331 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1332 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1333 struct sock *save_sk = skb->sk;
1335 /* Streams do not free skb on error. Reset to prev state. */
1336 msg->msg_iter = orig_iter;
1338 ___pskb_trim(skb, orig_len);
1343 skb_zcopy_set(skb, uarg, NULL);
1344 return skb->len - orig_len;
1346 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1348 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1351 if (skb_zcopy(orig)) {
1352 if (skb_zcopy(nskb)) {
1353 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1358 if (skb_uarg(nskb) == skb_uarg(orig))
1360 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1363 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1369 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1370 * @skb: the skb to modify
1371 * @gfp_mask: allocation priority
1373 * This must be called on skb with SKBFL_ZEROCOPY_ENABLE.
1374 * It will copy all frags into kernel and drop the reference
1375 * to userspace pages.
1377 * If this function is called from an interrupt gfp_mask() must be
1380 * Returns 0 on success or a negative error code on failure
1381 * to allocate kernel memory to copy to.
1383 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1385 int num_frags = skb_shinfo(skb)->nr_frags;
1386 struct page *page, *head = NULL;
1390 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1396 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1397 for (i = 0; i < new_frags; i++) {
1398 page = alloc_page(gfp_mask);
1401 struct page *next = (struct page *)page_private(head);
1407 set_page_private(page, (unsigned long)head);
1413 for (i = 0; i < num_frags; i++) {
1414 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1415 u32 p_off, p_len, copied;
1419 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1420 p, p_off, p_len, copied) {
1422 vaddr = kmap_atomic(p);
1424 while (done < p_len) {
1425 if (d_off == PAGE_SIZE) {
1427 page = (struct page *)page_private(page);
1429 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1430 memcpy(page_address(page) + d_off,
1431 vaddr + p_off + done, copy);
1435 kunmap_atomic(vaddr);
1439 /* skb frags release userspace buffers */
1440 for (i = 0; i < num_frags; i++)
1441 skb_frag_unref(skb, i);
1443 /* skb frags point to kernel buffers */
1444 for (i = 0; i < new_frags - 1; i++) {
1445 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1446 head = (struct page *)page_private(head);
1448 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1449 skb_shinfo(skb)->nr_frags = new_frags;
1452 skb_zcopy_clear(skb, false);
1455 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1458 * skb_clone - duplicate an sk_buff
1459 * @skb: buffer to clone
1460 * @gfp_mask: allocation priority
1462 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1463 * copies share the same packet data but not structure. The new
1464 * buffer has a reference count of 1. If the allocation fails the
1465 * function returns %NULL otherwise the new buffer is returned.
1467 * If this function is called from an interrupt gfp_mask() must be
1471 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1473 struct sk_buff_fclones *fclones = container_of(skb,
1474 struct sk_buff_fclones,
1478 if (skb_orphan_frags(skb, gfp_mask))
1481 if (skb->fclone == SKB_FCLONE_ORIG &&
1482 refcount_read(&fclones->fclone_ref) == 1) {
1484 refcount_set(&fclones->fclone_ref, 2);
1486 if (skb_pfmemalloc(skb))
1487 gfp_mask |= __GFP_MEMALLOC;
1489 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1493 n->fclone = SKB_FCLONE_UNAVAILABLE;
1496 return __skb_clone(n, skb);
1498 EXPORT_SYMBOL(skb_clone);
1500 void skb_headers_offset_update(struct sk_buff *skb, int off)
1502 /* Only adjust this if it actually is csum_start rather than csum */
1503 if (skb->ip_summed == CHECKSUM_PARTIAL)
1504 skb->csum_start += off;
1505 /* {transport,network,mac}_header and tail are relative to skb->head */
1506 skb->transport_header += off;
1507 skb->network_header += off;
1508 if (skb_mac_header_was_set(skb))
1509 skb->mac_header += off;
1510 skb->inner_transport_header += off;
1511 skb->inner_network_header += off;
1512 skb->inner_mac_header += off;
1514 EXPORT_SYMBOL(skb_headers_offset_update);
1516 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1518 __copy_skb_header(new, old);
1520 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1521 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1522 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1524 EXPORT_SYMBOL(skb_copy_header);
1526 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1528 if (skb_pfmemalloc(skb))
1529 return SKB_ALLOC_RX;
1534 * skb_copy - create private copy of an sk_buff
1535 * @skb: buffer to copy
1536 * @gfp_mask: allocation priority
1538 * Make a copy of both an &sk_buff and its data. This is used when the
1539 * caller wishes to modify the data and needs a private copy of the
1540 * data to alter. Returns %NULL on failure or the pointer to the buffer
1541 * on success. The returned buffer has a reference count of 1.
1543 * As by-product this function converts non-linear &sk_buff to linear
1544 * one, so that &sk_buff becomes completely private and caller is allowed
1545 * to modify all the data of returned buffer. This means that this
1546 * function is not recommended for use in circumstances when only
1547 * header is going to be modified. Use pskb_copy() instead.
1550 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1552 int headerlen = skb_headroom(skb);
1553 unsigned int size = skb_end_offset(skb) + skb->data_len;
1554 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1555 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1560 /* Set the data pointer */
1561 skb_reserve(n, headerlen);
1562 /* Set the tail pointer and length */
1563 skb_put(n, skb->len);
1565 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1567 skb_copy_header(n, skb);
1570 EXPORT_SYMBOL(skb_copy);
1573 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1574 * @skb: buffer to copy
1575 * @headroom: headroom of new skb
1576 * @gfp_mask: allocation priority
1577 * @fclone: if true allocate the copy of the skb from the fclone
1578 * cache instead of the head cache; it is recommended to set this
1579 * to true for the cases where the copy will likely be cloned
1581 * Make a copy of both an &sk_buff and part of its data, located
1582 * in header. Fragmented data remain shared. This is used when
1583 * the caller wishes to modify only header of &sk_buff and needs
1584 * private copy of the header to alter. Returns %NULL on failure
1585 * or the pointer to the buffer on success.
1586 * The returned buffer has a reference count of 1.
1589 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1590 gfp_t gfp_mask, bool fclone)
1592 unsigned int size = skb_headlen(skb) + headroom;
1593 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1594 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1599 /* Set the data pointer */
1600 skb_reserve(n, headroom);
1601 /* Set the tail pointer and length */
1602 skb_put(n, skb_headlen(skb));
1603 /* Copy the bytes */
1604 skb_copy_from_linear_data(skb, n->data, n->len);
1606 n->truesize += skb->data_len;
1607 n->data_len = skb->data_len;
1610 if (skb_shinfo(skb)->nr_frags) {
1613 if (skb_orphan_frags(skb, gfp_mask) ||
1614 skb_zerocopy_clone(n, skb, gfp_mask)) {
1619 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1620 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1621 skb_frag_ref(skb, i);
1623 skb_shinfo(n)->nr_frags = i;
1626 if (skb_has_frag_list(skb)) {
1627 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1628 skb_clone_fraglist(n);
1631 skb_copy_header(n, skb);
1635 EXPORT_SYMBOL(__pskb_copy_fclone);
1638 * pskb_expand_head - reallocate header of &sk_buff
1639 * @skb: buffer to reallocate
1640 * @nhead: room to add at head
1641 * @ntail: room to add at tail
1642 * @gfp_mask: allocation priority
1644 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1645 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1646 * reference count of 1. Returns zero in the case of success or error,
1647 * if expansion failed. In the last case, &sk_buff is not changed.
1649 * All the pointers pointing into skb header may change and must be
1650 * reloaded after call to this function.
1653 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1656 int i, osize = skb_end_offset(skb);
1657 int size = osize + nhead + ntail;
1663 BUG_ON(skb_shared(skb));
1665 size = SKB_DATA_ALIGN(size);
1667 if (skb_pfmemalloc(skb))
1668 gfp_mask |= __GFP_MEMALLOC;
1669 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1670 gfp_mask, NUMA_NO_NODE, NULL);
1673 size = SKB_WITH_OVERHEAD(ksize(data));
1675 /* Copy only real data... and, alas, header. This should be
1676 * optimized for the cases when header is void.
1678 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1680 memcpy((struct skb_shared_info *)(data + size),
1682 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1685 * if shinfo is shared we must drop the old head gracefully, but if it
1686 * is not we can just drop the old head and let the existing refcount
1687 * be since all we did is relocate the values
1689 if (skb_cloned(skb)) {
1690 if (skb_orphan_frags(skb, gfp_mask))
1693 refcount_inc(&skb_uarg(skb)->refcnt);
1694 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1695 skb_frag_ref(skb, i);
1697 if (skb_has_frag_list(skb))
1698 skb_clone_fraglist(skb);
1700 skb_release_data(skb);
1704 off = (data + nhead) - skb->head;
1709 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1713 skb->end = skb->head + size;
1716 skb_headers_offset_update(skb, nhead);
1720 atomic_set(&skb_shinfo(skb)->dataref, 1);
1722 skb_metadata_clear(skb);
1724 /* It is not generally safe to change skb->truesize.
1725 * For the moment, we really care of rx path, or
1726 * when skb is orphaned (not attached to a socket).
1728 if (!skb->sk || skb->destructor == sock_edemux)
1729 skb->truesize += size - osize;
1738 EXPORT_SYMBOL(pskb_expand_head);
1740 /* Make private copy of skb with writable head and some headroom */
1742 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1744 struct sk_buff *skb2;
1745 int delta = headroom - skb_headroom(skb);
1748 skb2 = pskb_copy(skb, GFP_ATOMIC);
1750 skb2 = skb_clone(skb, GFP_ATOMIC);
1751 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1759 EXPORT_SYMBOL(skb_realloc_headroom);
1762 * skb_copy_expand - copy and expand sk_buff
1763 * @skb: buffer to copy
1764 * @newheadroom: new free bytes at head
1765 * @newtailroom: new free bytes at tail
1766 * @gfp_mask: allocation priority
1768 * Make a copy of both an &sk_buff and its data and while doing so
1769 * allocate additional space.
1771 * This is used when the caller wishes to modify the data and needs a
1772 * private copy of the data to alter as well as more space for new fields.
1773 * Returns %NULL on failure or the pointer to the buffer
1774 * on success. The returned buffer has a reference count of 1.
1776 * You must pass %GFP_ATOMIC as the allocation priority if this function
1777 * is called from an interrupt.
1779 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1780 int newheadroom, int newtailroom,
1784 * Allocate the copy buffer
1786 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1787 gfp_mask, skb_alloc_rx_flag(skb),
1789 int oldheadroom = skb_headroom(skb);
1790 int head_copy_len, head_copy_off;
1795 skb_reserve(n, newheadroom);
1797 /* Set the tail pointer and length */
1798 skb_put(n, skb->len);
1800 head_copy_len = oldheadroom;
1802 if (newheadroom <= head_copy_len)
1803 head_copy_len = newheadroom;
1805 head_copy_off = newheadroom - head_copy_len;
1807 /* Copy the linear header and data. */
1808 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1809 skb->len + head_copy_len));
1811 skb_copy_header(n, skb);
1813 skb_headers_offset_update(n, newheadroom - oldheadroom);
1817 EXPORT_SYMBOL(skb_copy_expand);
1820 * __skb_pad - zero pad the tail of an skb
1821 * @skb: buffer to pad
1822 * @pad: space to pad
1823 * @free_on_error: free buffer on error
1825 * Ensure that a buffer is followed by a padding area that is zero
1826 * filled. Used by network drivers which may DMA or transfer data
1827 * beyond the buffer end onto the wire.
1829 * May return error in out of memory cases. The skb is freed on error
1830 * if @free_on_error is true.
1833 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1838 /* If the skbuff is non linear tailroom is always zero.. */
1839 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1840 memset(skb->data+skb->len, 0, pad);
1844 ntail = skb->data_len + pad - (skb->end - skb->tail);
1845 if (likely(skb_cloned(skb) || ntail > 0)) {
1846 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1851 /* FIXME: The use of this function with non-linear skb's really needs
1854 err = skb_linearize(skb);
1858 memset(skb->data + skb->len, 0, pad);
1866 EXPORT_SYMBOL(__skb_pad);
1869 * pskb_put - add data to the tail of a potentially fragmented buffer
1870 * @skb: start of the buffer to use
1871 * @tail: tail fragment of the buffer to use
1872 * @len: amount of data to add
1874 * This function extends the used data area of the potentially
1875 * fragmented buffer. @tail must be the last fragment of @skb -- or
1876 * @skb itself. If this would exceed the total buffer size the kernel
1877 * will panic. A pointer to the first byte of the extra data is
1881 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1884 skb->data_len += len;
1887 return skb_put(tail, len);
1889 EXPORT_SYMBOL_GPL(pskb_put);
1892 * skb_put - add data to a buffer
1893 * @skb: buffer to use
1894 * @len: amount of data to add
1896 * This function extends the used data area of the buffer. If this would
1897 * exceed the total buffer size the kernel will panic. A pointer to the
1898 * first byte of the extra data is returned.
1900 void *skb_put(struct sk_buff *skb, unsigned int len)
1902 void *tmp = skb_tail_pointer(skb);
1903 SKB_LINEAR_ASSERT(skb);
1906 if (unlikely(skb->tail > skb->end))
1907 skb_over_panic(skb, len, __builtin_return_address(0));
1910 EXPORT_SYMBOL(skb_put);
1913 * skb_push - add data to the start of a buffer
1914 * @skb: buffer to use
1915 * @len: amount of data to add
1917 * This function extends the used data area of the buffer at the buffer
1918 * start. If this would exceed the total buffer headroom the kernel will
1919 * panic. A pointer to the first byte of the extra data is returned.
1921 void *skb_push(struct sk_buff *skb, unsigned int len)
1925 if (unlikely(skb->data < skb->head))
1926 skb_under_panic(skb, len, __builtin_return_address(0));
1929 EXPORT_SYMBOL(skb_push);
1932 * skb_pull - remove data from the start of a buffer
1933 * @skb: buffer to use
1934 * @len: amount of data to remove
1936 * This function removes data from the start of a buffer, returning
1937 * the memory to the headroom. A pointer to the next data in the buffer
1938 * is returned. Once the data has been pulled future pushes will overwrite
1941 void *skb_pull(struct sk_buff *skb, unsigned int len)
1943 return skb_pull_inline(skb, len);
1945 EXPORT_SYMBOL(skb_pull);
1948 * skb_trim - remove end from a buffer
1949 * @skb: buffer to alter
1952 * Cut the length of a buffer down by removing data from the tail. If
1953 * the buffer is already under the length specified it is not modified.
1954 * The skb must be linear.
1956 void skb_trim(struct sk_buff *skb, unsigned int len)
1959 __skb_trim(skb, len);
1961 EXPORT_SYMBOL(skb_trim);
1963 /* Trims skb to length len. It can change skb pointers.
1966 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1968 struct sk_buff **fragp;
1969 struct sk_buff *frag;
1970 int offset = skb_headlen(skb);
1971 int nfrags = skb_shinfo(skb)->nr_frags;
1975 if (skb_cloned(skb) &&
1976 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1983 for (; i < nfrags; i++) {
1984 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1991 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1994 skb_shinfo(skb)->nr_frags = i;
1996 for (; i < nfrags; i++)
1997 skb_frag_unref(skb, i);
1999 if (skb_has_frag_list(skb))
2000 skb_drop_fraglist(skb);
2004 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
2005 fragp = &frag->next) {
2006 int end = offset + frag->len;
2008 if (skb_shared(frag)) {
2009 struct sk_buff *nfrag;
2011 nfrag = skb_clone(frag, GFP_ATOMIC);
2012 if (unlikely(!nfrag))
2015 nfrag->next = frag->next;
2027 unlikely((err = pskb_trim(frag, len - offset))))
2031 skb_drop_list(&frag->next);
2036 if (len > skb_headlen(skb)) {
2037 skb->data_len -= skb->len - len;
2042 skb_set_tail_pointer(skb, len);
2045 if (!skb->sk || skb->destructor == sock_edemux)
2049 EXPORT_SYMBOL(___pskb_trim);
2051 /* Note : use pskb_trim_rcsum() instead of calling this directly
2053 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2055 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2056 int delta = skb->len - len;
2058 skb->csum = csum_block_sub(skb->csum,
2059 skb_checksum(skb, len, delta, 0),
2061 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2062 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2063 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2065 if (offset + sizeof(__sum16) > hdlen)
2068 return __pskb_trim(skb, len);
2070 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2073 * __pskb_pull_tail - advance tail of skb header
2074 * @skb: buffer to reallocate
2075 * @delta: number of bytes to advance tail
2077 * The function makes a sense only on a fragmented &sk_buff,
2078 * it expands header moving its tail forward and copying necessary
2079 * data from fragmented part.
2081 * &sk_buff MUST have reference count of 1.
2083 * Returns %NULL (and &sk_buff does not change) if pull failed
2084 * or value of new tail of skb in the case of success.
2086 * All the pointers pointing into skb header may change and must be
2087 * reloaded after call to this function.
2090 /* Moves tail of skb head forward, copying data from fragmented part,
2091 * when it is necessary.
2092 * 1. It may fail due to malloc failure.
2093 * 2. It may change skb pointers.
2095 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2097 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2099 /* If skb has not enough free space at tail, get new one
2100 * plus 128 bytes for future expansions. If we have enough
2101 * room at tail, reallocate without expansion only if skb is cloned.
2103 int i, k, eat = (skb->tail + delta) - skb->end;
2105 if (eat > 0 || skb_cloned(skb)) {
2106 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2111 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2112 skb_tail_pointer(skb), delta));
2114 /* Optimization: no fragments, no reasons to preestimate
2115 * size of pulled pages. Superb.
2117 if (!skb_has_frag_list(skb))
2120 /* Estimate size of pulled pages. */
2122 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2123 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2130 /* If we need update frag list, we are in troubles.
2131 * Certainly, it is possible to add an offset to skb data,
2132 * but taking into account that pulling is expected to
2133 * be very rare operation, it is worth to fight against
2134 * further bloating skb head and crucify ourselves here instead.
2135 * Pure masohism, indeed. 8)8)
2138 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2139 struct sk_buff *clone = NULL;
2140 struct sk_buff *insp = NULL;
2143 if (list->len <= eat) {
2144 /* Eaten as whole. */
2149 /* Eaten partially. */
2151 if (skb_shared(list)) {
2152 /* Sucks! We need to fork list. :-( */
2153 clone = skb_clone(list, GFP_ATOMIC);
2159 /* This may be pulled without
2163 if (!pskb_pull(list, eat)) {
2171 /* Free pulled out fragments. */
2172 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2173 skb_shinfo(skb)->frag_list = list->next;
2176 /* And insert new clone at head. */
2179 skb_shinfo(skb)->frag_list = clone;
2182 /* Success! Now we may commit changes to skb data. */
2187 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2188 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2191 skb_frag_unref(skb, i);
2194 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2196 *frag = skb_shinfo(skb)->frags[i];
2198 skb_frag_off_add(frag, eat);
2199 skb_frag_size_sub(frag, eat);
2207 skb_shinfo(skb)->nr_frags = k;
2211 skb->data_len -= delta;
2214 skb_zcopy_clear(skb, false);
2216 return skb_tail_pointer(skb);
2218 EXPORT_SYMBOL(__pskb_pull_tail);
2221 * skb_copy_bits - copy bits from skb to kernel buffer
2223 * @offset: offset in source
2224 * @to: destination buffer
2225 * @len: number of bytes to copy
2227 * Copy the specified number of bytes from the source skb to the
2228 * destination buffer.
2231 * If its prototype is ever changed,
2232 * check arch/{*}/net/{*}.S files,
2233 * since it is called from BPF assembly code.
2235 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2237 int start = skb_headlen(skb);
2238 struct sk_buff *frag_iter;
2241 if (offset > (int)skb->len - len)
2245 if ((copy = start - offset) > 0) {
2248 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2249 if ((len -= copy) == 0)
2255 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2257 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2259 WARN_ON(start > offset + len);
2261 end = start + skb_frag_size(f);
2262 if ((copy = end - offset) > 0) {
2263 u32 p_off, p_len, copied;
2270 skb_frag_foreach_page(f,
2271 skb_frag_off(f) + offset - start,
2272 copy, p, p_off, p_len, copied) {
2273 vaddr = kmap_atomic(p);
2274 memcpy(to + copied, vaddr + p_off, p_len);
2275 kunmap_atomic(vaddr);
2278 if ((len -= copy) == 0)
2286 skb_walk_frags(skb, frag_iter) {
2289 WARN_ON(start > offset + len);
2291 end = start + frag_iter->len;
2292 if ((copy = end - offset) > 0) {
2295 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2297 if ((len -= copy) == 0)
2311 EXPORT_SYMBOL(skb_copy_bits);
2314 * Callback from splice_to_pipe(), if we need to release some pages
2315 * at the end of the spd in case we error'ed out in filling the pipe.
2317 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2319 put_page(spd->pages[i]);
2322 static struct page *linear_to_page(struct page *page, unsigned int *len,
2323 unsigned int *offset,
2326 struct page_frag *pfrag = sk_page_frag(sk);
2328 if (!sk_page_frag_refill(sk, pfrag))
2331 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2333 memcpy(page_address(pfrag->page) + pfrag->offset,
2334 page_address(page) + *offset, *len);
2335 *offset = pfrag->offset;
2336 pfrag->offset += *len;
2341 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2343 unsigned int offset)
2345 return spd->nr_pages &&
2346 spd->pages[spd->nr_pages - 1] == page &&
2347 (spd->partial[spd->nr_pages - 1].offset +
2348 spd->partial[spd->nr_pages - 1].len == offset);
2352 * Fill page/offset/length into spd, if it can hold more pages.
2354 static bool spd_fill_page(struct splice_pipe_desc *spd,
2355 struct pipe_inode_info *pipe, struct page *page,
2356 unsigned int *len, unsigned int offset,
2360 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2364 page = linear_to_page(page, len, &offset, sk);
2368 if (spd_can_coalesce(spd, page, offset)) {
2369 spd->partial[spd->nr_pages - 1].len += *len;
2373 spd->pages[spd->nr_pages] = page;
2374 spd->partial[spd->nr_pages].len = *len;
2375 spd->partial[spd->nr_pages].offset = offset;
2381 static bool __splice_segment(struct page *page, unsigned int poff,
2382 unsigned int plen, unsigned int *off,
2384 struct splice_pipe_desc *spd, bool linear,
2386 struct pipe_inode_info *pipe)
2391 /* skip this segment if already processed */
2397 /* ignore any bits we already processed */
2403 unsigned int flen = min(*len, plen);
2405 if (spd_fill_page(spd, pipe, page, &flen, poff,
2411 } while (*len && plen);
2417 * Map linear and fragment data from the skb to spd. It reports true if the
2418 * pipe is full or if we already spliced the requested length.
2420 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2421 unsigned int *offset, unsigned int *len,
2422 struct splice_pipe_desc *spd, struct sock *sk)
2425 struct sk_buff *iter;
2427 /* map the linear part :
2428 * If skb->head_frag is set, this 'linear' part is backed by a
2429 * fragment, and if the head is not shared with any clones then
2430 * we can avoid a copy since we own the head portion of this page.
2432 if (__splice_segment(virt_to_page(skb->data),
2433 (unsigned long) skb->data & (PAGE_SIZE - 1),
2436 skb_head_is_locked(skb),
2441 * then map the fragments
2443 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2444 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2446 if (__splice_segment(skb_frag_page(f),
2447 skb_frag_off(f), skb_frag_size(f),
2448 offset, len, spd, false, sk, pipe))
2452 skb_walk_frags(skb, iter) {
2453 if (*offset >= iter->len) {
2454 *offset -= iter->len;
2457 /* __skb_splice_bits() only fails if the output has no room
2458 * left, so no point in going over the frag_list for the error
2461 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2469 * Map data from the skb to a pipe. Should handle both the linear part,
2470 * the fragments, and the frag list.
2472 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2473 struct pipe_inode_info *pipe, unsigned int tlen,
2476 struct partial_page partial[MAX_SKB_FRAGS];
2477 struct page *pages[MAX_SKB_FRAGS];
2478 struct splice_pipe_desc spd = {
2481 .nr_pages_max = MAX_SKB_FRAGS,
2482 .ops = &nosteal_pipe_buf_ops,
2483 .spd_release = sock_spd_release,
2487 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2490 ret = splice_to_pipe(pipe, &spd);
2494 EXPORT_SYMBOL_GPL(skb_splice_bits);
2496 /* Send skb data on a socket. Socket must be locked. */
2497 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2500 unsigned int orig_len = len;
2501 struct sk_buff *head = skb;
2502 unsigned short fragidx;
2507 /* Deal with head data */
2508 while (offset < skb_headlen(skb) && len) {
2512 slen = min_t(int, len, skb_headlen(skb) - offset);
2513 kv.iov_base = skb->data + offset;
2515 memset(&msg, 0, sizeof(msg));
2516 msg.msg_flags = MSG_DONTWAIT;
2518 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2526 /* All the data was skb head? */
2530 /* Make offset relative to start of frags */
2531 offset -= skb_headlen(skb);
2533 /* Find where we are in frag list */
2534 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2535 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2537 if (offset < skb_frag_size(frag))
2540 offset -= skb_frag_size(frag);
2543 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2544 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2546 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2549 ret = kernel_sendpage_locked(sk, skb_frag_page(frag),
2550 skb_frag_off(frag) + offset,
2551 slen, MSG_DONTWAIT);
2564 /* Process any frag lists */
2567 if (skb_has_frag_list(skb)) {
2568 skb = skb_shinfo(skb)->frag_list;
2571 } else if (skb->next) {
2578 return orig_len - len;
2581 return orig_len == len ? ret : orig_len - len;
2583 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2586 * skb_store_bits - store bits from kernel buffer to skb
2587 * @skb: destination buffer
2588 * @offset: offset in destination
2589 * @from: source buffer
2590 * @len: number of bytes to copy
2592 * Copy the specified number of bytes from the source buffer to the
2593 * destination skb. This function handles all the messy bits of
2594 * traversing fragment lists and such.
2597 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2599 int start = skb_headlen(skb);
2600 struct sk_buff *frag_iter;
2603 if (offset > (int)skb->len - len)
2606 if ((copy = start - offset) > 0) {
2609 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2610 if ((len -= copy) == 0)
2616 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2617 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2620 WARN_ON(start > offset + len);
2622 end = start + skb_frag_size(frag);
2623 if ((copy = end - offset) > 0) {
2624 u32 p_off, p_len, copied;
2631 skb_frag_foreach_page(frag,
2632 skb_frag_off(frag) + offset - start,
2633 copy, p, p_off, p_len, copied) {
2634 vaddr = kmap_atomic(p);
2635 memcpy(vaddr + p_off, from + copied, p_len);
2636 kunmap_atomic(vaddr);
2639 if ((len -= copy) == 0)
2647 skb_walk_frags(skb, frag_iter) {
2650 WARN_ON(start > offset + len);
2652 end = start + frag_iter->len;
2653 if ((copy = end - offset) > 0) {
2656 if (skb_store_bits(frag_iter, offset - start,
2659 if ((len -= copy) == 0)
2672 EXPORT_SYMBOL(skb_store_bits);
2674 /* Checksum skb data. */
2675 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2676 __wsum csum, const struct skb_checksum_ops *ops)
2678 int start = skb_headlen(skb);
2679 int i, copy = start - offset;
2680 struct sk_buff *frag_iter;
2683 /* Checksum header. */
2687 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2688 skb->data + offset, copy, csum);
2689 if ((len -= copy) == 0)
2695 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2697 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2699 WARN_ON(start > offset + len);
2701 end = start + skb_frag_size(frag);
2702 if ((copy = end - offset) > 0) {
2703 u32 p_off, p_len, copied;
2711 skb_frag_foreach_page(frag,
2712 skb_frag_off(frag) + offset - start,
2713 copy, p, p_off, p_len, copied) {
2714 vaddr = kmap_atomic(p);
2715 csum2 = INDIRECT_CALL_1(ops->update,
2717 vaddr + p_off, p_len, 0);
2718 kunmap_atomic(vaddr);
2719 csum = INDIRECT_CALL_1(ops->combine,
2720 csum_block_add_ext, csum,
2732 skb_walk_frags(skb, frag_iter) {
2735 WARN_ON(start > offset + len);
2737 end = start + frag_iter->len;
2738 if ((copy = end - offset) > 0) {
2742 csum2 = __skb_checksum(frag_iter, offset - start,
2744 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2745 csum, csum2, pos, copy);
2746 if ((len -= copy) == 0)
2757 EXPORT_SYMBOL(__skb_checksum);
2759 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2760 int len, __wsum csum)
2762 const struct skb_checksum_ops ops = {
2763 .update = csum_partial_ext,
2764 .combine = csum_block_add_ext,
2767 return __skb_checksum(skb, offset, len, csum, &ops);
2769 EXPORT_SYMBOL(skb_checksum);
2771 /* Both of above in one bottle. */
2773 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2776 int start = skb_headlen(skb);
2777 int i, copy = start - offset;
2778 struct sk_buff *frag_iter;
2786 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2788 if ((len -= copy) == 0)
2795 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2798 WARN_ON(start > offset + len);
2800 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2801 if ((copy = end - offset) > 0) {
2802 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2803 u32 p_off, p_len, copied;
2811 skb_frag_foreach_page(frag,
2812 skb_frag_off(frag) + offset - start,
2813 copy, p, p_off, p_len, copied) {
2814 vaddr = kmap_atomic(p);
2815 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2818 kunmap_atomic(vaddr);
2819 csum = csum_block_add(csum, csum2, pos);
2831 skb_walk_frags(skb, frag_iter) {
2835 WARN_ON(start > offset + len);
2837 end = start + frag_iter->len;
2838 if ((copy = end - offset) > 0) {
2841 csum2 = skb_copy_and_csum_bits(frag_iter,
2844 csum = csum_block_add(csum, csum2, pos);
2845 if ((len -= copy) == 0)
2856 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2858 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2862 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2863 /* See comments in __skb_checksum_complete(). */
2865 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2866 !skb->csum_complete_sw)
2867 netdev_rx_csum_fault(skb->dev, skb);
2869 if (!skb_shared(skb))
2870 skb->csum_valid = !sum;
2873 EXPORT_SYMBOL(__skb_checksum_complete_head);
2875 /* This function assumes skb->csum already holds pseudo header's checksum,
2876 * which has been changed from the hardware checksum, for example, by
2877 * __skb_checksum_validate_complete(). And, the original skb->csum must
2878 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2880 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2881 * zero. The new checksum is stored back into skb->csum unless the skb is
2884 __sum16 __skb_checksum_complete(struct sk_buff *skb)
2889 csum = skb_checksum(skb, 0, skb->len, 0);
2891 sum = csum_fold(csum_add(skb->csum, csum));
2892 /* This check is inverted, because we already knew the hardware
2893 * checksum is invalid before calling this function. So, if the
2894 * re-computed checksum is valid instead, then we have a mismatch
2895 * between the original skb->csum and skb_checksum(). This means either
2896 * the original hardware checksum is incorrect or we screw up skb->csum
2897 * when moving skb->data around.
2900 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2901 !skb->csum_complete_sw)
2902 netdev_rx_csum_fault(skb->dev, skb);
2905 if (!skb_shared(skb)) {
2906 /* Save full packet checksum */
2908 skb->ip_summed = CHECKSUM_COMPLETE;
2909 skb->csum_complete_sw = 1;
2910 skb->csum_valid = !sum;
2915 EXPORT_SYMBOL(__skb_checksum_complete);
2917 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2919 net_warn_ratelimited(
2920 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2925 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2926 int offset, int len)
2928 net_warn_ratelimited(
2929 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2934 static const struct skb_checksum_ops default_crc32c_ops = {
2935 .update = warn_crc32c_csum_update,
2936 .combine = warn_crc32c_csum_combine,
2939 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2940 &default_crc32c_ops;
2941 EXPORT_SYMBOL(crc32c_csum_stub);
2944 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2945 * @from: source buffer
2947 * Calculates the amount of linear headroom needed in the 'to' skb passed
2948 * into skb_zerocopy().
2951 skb_zerocopy_headlen(const struct sk_buff *from)
2953 unsigned int hlen = 0;
2955 if (!from->head_frag ||
2956 skb_headlen(from) < L1_CACHE_BYTES ||
2957 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2958 hlen = skb_headlen(from);
2960 if (skb_has_frag_list(from))
2965 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2968 * skb_zerocopy - Zero copy skb to skb
2969 * @to: destination buffer
2970 * @from: source buffer
2971 * @len: number of bytes to copy from source buffer
2972 * @hlen: size of linear headroom in destination buffer
2974 * Copies up to `len` bytes from `from` to `to` by creating references
2975 * to the frags in the source buffer.
2977 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2978 * headroom in the `to` buffer.
2981 * 0: everything is OK
2982 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2983 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2986 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2989 int plen = 0; /* length of skb->head fragment */
2992 unsigned int offset;
2994 BUG_ON(!from->head_frag && !hlen);
2996 /* dont bother with small payloads */
2997 if (len <= skb_tailroom(to))
2998 return skb_copy_bits(from, 0, skb_put(to, len), len);
3001 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
3006 plen = min_t(int, skb_headlen(from), len);
3008 page = virt_to_head_page(from->head);
3009 offset = from->data - (unsigned char *)page_address(page);
3010 __skb_fill_page_desc(to, 0, page, offset, plen);
3017 to->truesize += len + plen;
3018 to->len += len + plen;
3019 to->data_len += len + plen;
3021 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
3025 skb_zerocopy_clone(to, from, GFP_ATOMIC);
3027 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3032 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3033 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3035 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3037 skb_frag_ref(to, j);
3040 skb_shinfo(to)->nr_frags = j;
3044 EXPORT_SYMBOL_GPL(skb_zerocopy);
3046 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3051 if (skb->ip_summed == CHECKSUM_PARTIAL)
3052 csstart = skb_checksum_start_offset(skb);
3054 csstart = skb_headlen(skb);
3056 BUG_ON(csstart > skb_headlen(skb));
3058 skb_copy_from_linear_data(skb, to, csstart);
3061 if (csstart != skb->len)
3062 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3063 skb->len - csstart);
3065 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3066 long csstuff = csstart + skb->csum_offset;
3068 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3071 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3074 * skb_dequeue - remove from the head of the queue
3075 * @list: list to dequeue from
3077 * Remove the head of the list. The list lock is taken so the function
3078 * may be used safely with other locking list functions. The head item is
3079 * returned or %NULL if the list is empty.
3082 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3084 unsigned long flags;
3085 struct sk_buff *result;
3087 spin_lock_irqsave(&list->lock, flags);
3088 result = __skb_dequeue(list);
3089 spin_unlock_irqrestore(&list->lock, flags);
3092 EXPORT_SYMBOL(skb_dequeue);
3095 * skb_dequeue_tail - remove from the tail of the queue
3096 * @list: list to dequeue from
3098 * Remove the tail of the list. The list lock is taken so the function
3099 * may be used safely with other locking list functions. The tail item is
3100 * returned or %NULL if the list is empty.
3102 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3104 unsigned long flags;
3105 struct sk_buff *result;
3107 spin_lock_irqsave(&list->lock, flags);
3108 result = __skb_dequeue_tail(list);
3109 spin_unlock_irqrestore(&list->lock, flags);
3112 EXPORT_SYMBOL(skb_dequeue_tail);
3115 * skb_queue_purge - empty a list
3116 * @list: list to empty
3118 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3119 * the list and one reference dropped. This function takes the list
3120 * lock and is atomic with respect to other list locking functions.
3122 void skb_queue_purge(struct sk_buff_head *list)
3124 struct sk_buff *skb;
3125 while ((skb = skb_dequeue(list)) != NULL)
3128 EXPORT_SYMBOL(skb_queue_purge);
3131 * skb_rbtree_purge - empty a skb rbtree
3132 * @root: root of the rbtree to empty
3133 * Return value: the sum of truesizes of all purged skbs.
3135 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3136 * the list and one reference dropped. This function does not take
3137 * any lock. Synchronization should be handled by the caller (e.g., TCP
3138 * out-of-order queue is protected by the socket lock).
3140 unsigned int skb_rbtree_purge(struct rb_root *root)
3142 struct rb_node *p = rb_first(root);
3143 unsigned int sum = 0;
3146 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3149 rb_erase(&skb->rbnode, root);
3150 sum += skb->truesize;
3157 * skb_queue_head - queue a buffer at the list head
3158 * @list: list to use
3159 * @newsk: buffer to queue
3161 * Queue a buffer at the start of the list. This function takes the
3162 * list lock and can be used safely with other locking &sk_buff functions
3165 * A buffer cannot be placed on two lists at the same time.
3167 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3169 unsigned long flags;
3171 spin_lock_irqsave(&list->lock, flags);
3172 __skb_queue_head(list, newsk);
3173 spin_unlock_irqrestore(&list->lock, flags);
3175 EXPORT_SYMBOL(skb_queue_head);
3178 * skb_queue_tail - queue a buffer at the list tail
3179 * @list: list to use
3180 * @newsk: buffer to queue
3182 * Queue a buffer at the tail of the list. This function takes the
3183 * list lock and can be used safely with other locking &sk_buff functions
3186 * A buffer cannot be placed on two lists at the same time.
3188 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3190 unsigned long flags;
3192 spin_lock_irqsave(&list->lock, flags);
3193 __skb_queue_tail(list, newsk);
3194 spin_unlock_irqrestore(&list->lock, flags);
3196 EXPORT_SYMBOL(skb_queue_tail);
3199 * skb_unlink - remove a buffer from a list
3200 * @skb: buffer to remove
3201 * @list: list to use
3203 * Remove a packet from a list. The list locks are taken and this
3204 * function is atomic with respect to other list locked calls
3206 * You must know what list the SKB is on.
3208 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3210 unsigned long flags;
3212 spin_lock_irqsave(&list->lock, flags);
3213 __skb_unlink(skb, list);
3214 spin_unlock_irqrestore(&list->lock, flags);
3216 EXPORT_SYMBOL(skb_unlink);
3219 * skb_append - append a buffer
3220 * @old: buffer to insert after
3221 * @newsk: buffer to insert
3222 * @list: list to use
3224 * Place a packet after a given packet in a list. The list locks are taken
3225 * and this function is atomic with respect to other list locked calls.
3226 * A buffer cannot be placed on two lists at the same time.
3228 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3230 unsigned long flags;
3232 spin_lock_irqsave(&list->lock, flags);
3233 __skb_queue_after(list, old, newsk);
3234 spin_unlock_irqrestore(&list->lock, flags);
3236 EXPORT_SYMBOL(skb_append);
3238 static inline void skb_split_inside_header(struct sk_buff *skb,
3239 struct sk_buff* skb1,
3240 const u32 len, const int pos)
3244 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3246 /* And move data appendix as is. */
3247 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3248 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3250 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3251 skb_shinfo(skb)->nr_frags = 0;
3252 skb1->data_len = skb->data_len;
3253 skb1->len += skb1->data_len;
3256 skb_set_tail_pointer(skb, len);
3259 static inline void skb_split_no_header(struct sk_buff *skb,
3260 struct sk_buff* skb1,
3261 const u32 len, int pos)
3264 const int nfrags = skb_shinfo(skb)->nr_frags;
3266 skb_shinfo(skb)->nr_frags = 0;
3267 skb1->len = skb1->data_len = skb->len - len;
3269 skb->data_len = len - pos;
3271 for (i = 0; i < nfrags; i++) {
3272 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3274 if (pos + size > len) {
3275 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3279 * We have two variants in this case:
3280 * 1. Move all the frag to the second
3281 * part, if it is possible. F.e.
3282 * this approach is mandatory for TUX,
3283 * where splitting is expensive.
3284 * 2. Split is accurately. We make this.
3286 skb_frag_ref(skb, i);
3287 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3288 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3289 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3290 skb_shinfo(skb)->nr_frags++;
3294 skb_shinfo(skb)->nr_frags++;
3297 skb_shinfo(skb1)->nr_frags = k;
3301 * skb_split - Split fragmented skb to two parts at length len.
3302 * @skb: the buffer to split
3303 * @skb1: the buffer to receive the second part
3304 * @len: new length for skb
3306 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3308 int pos = skb_headlen(skb);
3310 skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3311 skb_zerocopy_clone(skb1, skb, 0);
3312 if (len < pos) /* Split line is inside header. */
3313 skb_split_inside_header(skb, skb1, len, pos);
3314 else /* Second chunk has no header, nothing to copy. */
3315 skb_split_no_header(skb, skb1, len, pos);
3317 EXPORT_SYMBOL(skb_split);
3319 /* Shifting from/to a cloned skb is a no-go.
3321 * Caller cannot keep skb_shinfo related pointers past calling here!
3323 static int skb_prepare_for_shift(struct sk_buff *skb)
3327 if (skb_cloned(skb)) {
3328 /* Save and restore truesize: pskb_expand_head() may reallocate
3329 * memory where ksize(kmalloc(S)) != ksize(kmalloc(S)), but we
3330 * cannot change truesize at this point.
3332 unsigned int save_truesize = skb->truesize;
3334 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3335 skb->truesize = save_truesize;
3341 * skb_shift - Shifts paged data partially from skb to another
3342 * @tgt: buffer into which tail data gets added
3343 * @skb: buffer from which the paged data comes from
3344 * @shiftlen: shift up to this many bytes
3346 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3347 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3348 * It's up to caller to free skb if everything was shifted.
3350 * If @tgt runs out of frags, the whole operation is aborted.
3352 * Skb cannot include anything else but paged data while tgt is allowed
3353 * to have non-paged data as well.
3355 * TODO: full sized shift could be optimized but that would need
3356 * specialized skb free'er to handle frags without up-to-date nr_frags.
3358 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3360 int from, to, merge, todo;
3361 skb_frag_t *fragfrom, *fragto;
3363 BUG_ON(shiftlen > skb->len);
3365 if (skb_headlen(skb))
3367 if (skb_zcopy(tgt) || skb_zcopy(skb))
3372 to = skb_shinfo(tgt)->nr_frags;
3373 fragfrom = &skb_shinfo(skb)->frags[from];
3375 /* Actual merge is delayed until the point when we know we can
3376 * commit all, so that we don't have to undo partial changes
3379 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3380 skb_frag_off(fragfrom))) {
3385 todo -= skb_frag_size(fragfrom);
3387 if (skb_prepare_for_shift(skb) ||
3388 skb_prepare_for_shift(tgt))
3391 /* All previous frag pointers might be stale! */
3392 fragfrom = &skb_shinfo(skb)->frags[from];
3393 fragto = &skb_shinfo(tgt)->frags[merge];
3395 skb_frag_size_add(fragto, shiftlen);
3396 skb_frag_size_sub(fragfrom, shiftlen);
3397 skb_frag_off_add(fragfrom, shiftlen);
3405 /* Skip full, not-fitting skb to avoid expensive operations */
3406 if ((shiftlen == skb->len) &&
3407 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3410 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3413 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3414 if (to == MAX_SKB_FRAGS)
3417 fragfrom = &skb_shinfo(skb)->frags[from];
3418 fragto = &skb_shinfo(tgt)->frags[to];
3420 if (todo >= skb_frag_size(fragfrom)) {
3421 *fragto = *fragfrom;
3422 todo -= skb_frag_size(fragfrom);
3427 __skb_frag_ref(fragfrom);
3428 skb_frag_page_copy(fragto, fragfrom);
3429 skb_frag_off_copy(fragto, fragfrom);
3430 skb_frag_size_set(fragto, todo);
3432 skb_frag_off_add(fragfrom, todo);
3433 skb_frag_size_sub(fragfrom, todo);
3441 /* Ready to "commit" this state change to tgt */
3442 skb_shinfo(tgt)->nr_frags = to;
3445 fragfrom = &skb_shinfo(skb)->frags[0];
3446 fragto = &skb_shinfo(tgt)->frags[merge];
3448 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3449 __skb_frag_unref(fragfrom);
3452 /* Reposition in the original skb */
3454 while (from < skb_shinfo(skb)->nr_frags)
3455 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3456 skb_shinfo(skb)->nr_frags = to;
3458 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3461 /* Most likely the tgt won't ever need its checksum anymore, skb on
3462 * the other hand might need it if it needs to be resent
3464 tgt->ip_summed = CHECKSUM_PARTIAL;
3465 skb->ip_summed = CHECKSUM_PARTIAL;
3467 /* Yak, is it really working this way? Some helper please? */
3468 skb->len -= shiftlen;
3469 skb->data_len -= shiftlen;
3470 skb->truesize -= shiftlen;
3471 tgt->len += shiftlen;
3472 tgt->data_len += shiftlen;
3473 tgt->truesize += shiftlen;
3479 * skb_prepare_seq_read - Prepare a sequential read of skb data
3480 * @skb: the buffer to read
3481 * @from: lower offset of data to be read
3482 * @to: upper offset of data to be read
3483 * @st: state variable
3485 * Initializes the specified state variable. Must be called before
3486 * invoking skb_seq_read() for the first time.
3488 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3489 unsigned int to, struct skb_seq_state *st)
3491 st->lower_offset = from;
3492 st->upper_offset = to;
3493 st->root_skb = st->cur_skb = skb;
3494 st->frag_idx = st->stepped_offset = 0;
3495 st->frag_data = NULL;
3498 EXPORT_SYMBOL(skb_prepare_seq_read);
3501 * skb_seq_read - Sequentially read skb data
3502 * @consumed: number of bytes consumed by the caller so far
3503 * @data: destination pointer for data to be returned
3504 * @st: state variable
3506 * Reads a block of skb data at @consumed relative to the
3507 * lower offset specified to skb_prepare_seq_read(). Assigns
3508 * the head of the data block to @data and returns the length
3509 * of the block or 0 if the end of the skb data or the upper
3510 * offset has been reached.
3512 * The caller is not required to consume all of the data
3513 * returned, i.e. @consumed is typically set to the number
3514 * of bytes already consumed and the next call to
3515 * skb_seq_read() will return the remaining part of the block.
3517 * Note 1: The size of each block of data returned can be arbitrary,
3518 * this limitation is the cost for zerocopy sequential
3519 * reads of potentially non linear data.
3521 * Note 2: Fragment lists within fragments are not implemented
3522 * at the moment, state->root_skb could be replaced with
3523 * a stack for this purpose.
3525 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3526 struct skb_seq_state *st)
3528 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3531 if (unlikely(abs_offset >= st->upper_offset)) {
3532 if (st->frag_data) {
3533 kunmap_atomic(st->frag_data);
3534 st->frag_data = NULL;
3540 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3542 if (abs_offset < block_limit && !st->frag_data) {
3543 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3544 return block_limit - abs_offset;
3547 if (st->frag_idx == 0 && !st->frag_data)
3548 st->stepped_offset += skb_headlen(st->cur_skb);
3550 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3551 unsigned int pg_idx, pg_off, pg_sz;
3553 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3556 pg_off = skb_frag_off(frag);
3557 pg_sz = skb_frag_size(frag);
3559 if (skb_frag_must_loop(skb_frag_page(frag))) {
3560 pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
3561 pg_off = offset_in_page(pg_off + st->frag_off);
3562 pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
3563 PAGE_SIZE - pg_off);
3566 block_limit = pg_sz + st->stepped_offset;
3567 if (abs_offset < block_limit) {
3569 st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
3571 *data = (u8 *)st->frag_data + pg_off +
3572 (abs_offset - st->stepped_offset);
3574 return block_limit - abs_offset;
3577 if (st->frag_data) {
3578 kunmap_atomic(st->frag_data);
3579 st->frag_data = NULL;
3582 st->stepped_offset += pg_sz;
3583 st->frag_off += pg_sz;
3584 if (st->frag_off == skb_frag_size(frag)) {
3590 if (st->frag_data) {
3591 kunmap_atomic(st->frag_data);
3592 st->frag_data = NULL;
3595 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3596 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3599 } else if (st->cur_skb->next) {
3600 st->cur_skb = st->cur_skb->next;
3607 EXPORT_SYMBOL(skb_seq_read);
3610 * skb_abort_seq_read - Abort a sequential read of skb data
3611 * @st: state variable
3613 * Must be called if skb_seq_read() was not called until it
3616 void skb_abort_seq_read(struct skb_seq_state *st)
3619 kunmap_atomic(st->frag_data);
3621 EXPORT_SYMBOL(skb_abort_seq_read);
3623 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3625 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3626 struct ts_config *conf,
3627 struct ts_state *state)
3629 return skb_seq_read(offset, text, TS_SKB_CB(state));
3632 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3634 skb_abort_seq_read(TS_SKB_CB(state));
3638 * skb_find_text - Find a text pattern in skb data
3639 * @skb: the buffer to look in
3640 * @from: search offset
3642 * @config: textsearch configuration
3644 * Finds a pattern in the skb data according to the specified
3645 * textsearch configuration. Use textsearch_next() to retrieve
3646 * subsequent occurrences of the pattern. Returns the offset
3647 * to the first occurrence or UINT_MAX if no match was found.
3649 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3650 unsigned int to, struct ts_config *config)
3652 struct ts_state state;
3655 config->get_next_block = skb_ts_get_next_block;
3656 config->finish = skb_ts_finish;
3658 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3660 ret = textsearch_find(config, &state);
3661 return (ret <= to - from ? ret : UINT_MAX);
3663 EXPORT_SYMBOL(skb_find_text);
3665 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3666 int offset, size_t size)
3668 int i = skb_shinfo(skb)->nr_frags;
3670 if (skb_can_coalesce(skb, i, page, offset)) {
3671 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3672 } else if (i < MAX_SKB_FRAGS) {
3674 skb_fill_page_desc(skb, i, page, offset, size);
3681 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3684 * skb_pull_rcsum - pull skb and update receive checksum
3685 * @skb: buffer to update
3686 * @len: length of data pulled
3688 * This function performs an skb_pull on the packet and updates
3689 * the CHECKSUM_COMPLETE checksum. It should be used on
3690 * receive path processing instead of skb_pull unless you know
3691 * that the checksum difference is zero (e.g., a valid IP header)
3692 * or you are setting ip_summed to CHECKSUM_NONE.
3694 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3696 unsigned char *data = skb->data;
3698 BUG_ON(len > skb->len);
3699 __skb_pull(skb, len);
3700 skb_postpull_rcsum(skb, data, len);
3703 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3705 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3707 skb_frag_t head_frag;
3710 page = virt_to_head_page(frag_skb->head);
3711 __skb_frag_set_page(&head_frag, page);
3712 skb_frag_off_set(&head_frag, frag_skb->data -
3713 (unsigned char *)page_address(page));
3714 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3718 struct sk_buff *skb_segment_list(struct sk_buff *skb,
3719 netdev_features_t features,
3720 unsigned int offset)
3722 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
3723 unsigned int tnl_hlen = skb_tnl_header_len(skb);
3724 unsigned int delta_truesize = 0;
3725 unsigned int delta_len = 0;
3726 struct sk_buff *tail = NULL;
3727 struct sk_buff *nskb, *tmp;
3730 skb_push(skb, -skb_network_offset(skb) + offset);
3732 skb_shinfo(skb)->frag_list = NULL;
3736 list_skb = list_skb->next;
3739 if (skb_shared(nskb)) {
3740 tmp = skb_clone(nskb, GFP_ATOMIC);
3744 err = skb_unclone(nskb, GFP_ATOMIC);
3755 if (unlikely(err)) {
3756 nskb->next = list_skb;
3762 delta_len += nskb->len;
3763 delta_truesize += nskb->truesize;
3765 skb_push(nskb, -skb_network_offset(nskb) + offset);
3767 skb_release_head_state(nskb);
3768 __copy_skb_header(nskb, skb);
3770 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
3771 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
3772 nskb->data - tnl_hlen,
3775 if (skb_needs_linearize(nskb, features) &&
3776 __skb_linearize(nskb))
3781 skb->truesize = skb->truesize - delta_truesize;
3782 skb->data_len = skb->data_len - delta_len;
3783 skb->len = skb->len - delta_len;
3789 if (skb_needs_linearize(skb, features) &&
3790 __skb_linearize(skb))
3798 kfree_skb_list(skb->next);
3800 return ERR_PTR(-ENOMEM);
3802 EXPORT_SYMBOL_GPL(skb_segment_list);
3804 int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
3806 if (unlikely(p->len + skb->len >= 65536))
3809 if (NAPI_GRO_CB(p)->last == p)
3810 skb_shinfo(p)->frag_list = skb;
3812 NAPI_GRO_CB(p)->last->next = skb;
3814 skb_pull(skb, skb_gro_offset(skb));
3816 NAPI_GRO_CB(p)->last = skb;
3817 NAPI_GRO_CB(p)->count++;
3818 p->data_len += skb->len;
3819 p->truesize += skb->truesize;
3822 NAPI_GRO_CB(skb)->same_flow = 1;
3828 * skb_segment - Perform protocol segmentation on skb.
3829 * @head_skb: buffer to segment
3830 * @features: features for the output path (see dev->features)
3832 * This function performs segmentation on the given skb. It returns
3833 * a pointer to the first in a list of new skbs for the segments.
3834 * In case of error it returns ERR_PTR(err).
3836 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3837 netdev_features_t features)
3839 struct sk_buff *segs = NULL;
3840 struct sk_buff *tail = NULL;
3841 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3842 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3843 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3844 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3845 struct sk_buff *frag_skb = head_skb;
3846 unsigned int offset = doffset;
3847 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3848 unsigned int partial_segs = 0;
3849 unsigned int headroom;
3850 unsigned int len = head_skb->len;
3853 int nfrags = skb_shinfo(head_skb)->nr_frags;
3858 if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3859 (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3860 /* gso_size is untrusted, and we have a frag_list with a linear
3861 * non head_frag head.
3863 * (we assume checking the first list_skb member suffices;
3864 * i.e if either of the list_skb members have non head_frag
3865 * head, then the first one has too).
3867 * If head_skb's headlen does not fit requested gso_size, it
3868 * means that the frag_list members do NOT terminate on exact
3869 * gso_size boundaries. Hence we cannot perform skb_frag_t page
3870 * sharing. Therefore we must fallback to copying the frag_list
3871 * skbs; we do so by disabling SG.
3873 if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3874 features &= ~NETIF_F_SG;
3877 __skb_push(head_skb, doffset);
3878 proto = skb_network_protocol(head_skb, NULL);
3879 if (unlikely(!proto))
3880 return ERR_PTR(-EINVAL);
3882 sg = !!(features & NETIF_F_SG);
3883 csum = !!can_checksum_protocol(features, proto);
3885 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3886 if (!(features & NETIF_F_GSO_PARTIAL)) {
3887 struct sk_buff *iter;
3888 unsigned int frag_len;
3891 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3894 /* If we get here then all the required
3895 * GSO features except frag_list are supported.
3896 * Try to split the SKB to multiple GSO SKBs
3897 * with no frag_list.
3898 * Currently we can do that only when the buffers don't
3899 * have a linear part and all the buffers except
3900 * the last are of the same length.
3902 frag_len = list_skb->len;
3903 skb_walk_frags(head_skb, iter) {
3904 if (frag_len != iter->len && iter->next)
3906 if (skb_headlen(iter) && !iter->head_frag)
3912 if (len != frag_len)
3916 /* GSO partial only requires that we trim off any excess that
3917 * doesn't fit into an MSS sized block, so take care of that
3920 partial_segs = len / mss;
3921 if (partial_segs > 1)
3922 mss *= partial_segs;
3928 headroom = skb_headroom(head_skb);
3929 pos = skb_headlen(head_skb);
3932 struct sk_buff *nskb;
3933 skb_frag_t *nskb_frag;
3937 if (unlikely(mss == GSO_BY_FRAGS)) {
3938 len = list_skb->len;
3940 len = head_skb->len - offset;
3945 hsize = skb_headlen(head_skb) - offset;
3947 if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) &&
3948 (skb_headlen(list_skb) == len || sg)) {
3949 BUG_ON(skb_headlen(list_skb) > len);
3952 nfrags = skb_shinfo(list_skb)->nr_frags;
3953 frag = skb_shinfo(list_skb)->frags;
3954 frag_skb = list_skb;
3955 pos += skb_headlen(list_skb);
3957 while (pos < offset + len) {
3958 BUG_ON(i >= nfrags);
3960 size = skb_frag_size(frag);
3961 if (pos + size > offset + len)
3969 nskb = skb_clone(list_skb, GFP_ATOMIC);
3970 list_skb = list_skb->next;
3972 if (unlikely(!nskb))
3975 if (unlikely(pskb_trim(nskb, len))) {
3980 hsize = skb_end_offset(nskb);
3981 if (skb_cow_head(nskb, doffset + headroom)) {
3986 nskb->truesize += skb_end_offset(nskb) - hsize;
3987 skb_release_head_state(nskb);
3988 __skb_push(nskb, doffset);
3992 if (hsize > len || !sg)
3995 nskb = __alloc_skb(hsize + doffset + headroom,
3996 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3999 if (unlikely(!nskb))
4002 skb_reserve(nskb, headroom);
4003 __skb_put(nskb, doffset);
4012 __copy_skb_header(nskb, head_skb);
4014 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
4015 skb_reset_mac_len(nskb);
4017 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
4018 nskb->data - tnl_hlen,
4019 doffset + tnl_hlen);
4021 if (nskb->len == len + doffset)
4022 goto perform_csum_check;
4026 if (!nskb->remcsum_offload)
4027 nskb->ip_summed = CHECKSUM_NONE;
4028 SKB_GSO_CB(nskb)->csum =
4029 skb_copy_and_csum_bits(head_skb, offset,
4033 SKB_GSO_CB(nskb)->csum_start =
4034 skb_headroom(nskb) + doffset;
4036 skb_copy_bits(head_skb, offset,
4043 nskb_frag = skb_shinfo(nskb)->frags;
4045 skb_copy_from_linear_data_offset(head_skb, offset,
4046 skb_put(nskb, hsize), hsize);
4048 skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags &
4051 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4052 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4055 while (pos < offset + len) {
4058 nfrags = skb_shinfo(list_skb)->nr_frags;
4059 frag = skb_shinfo(list_skb)->frags;
4060 frag_skb = list_skb;
4061 if (!skb_headlen(list_skb)) {
4064 BUG_ON(!list_skb->head_frag);
4066 /* to make room for head_frag. */
4070 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4071 skb_zerocopy_clone(nskb, frag_skb,
4075 list_skb = list_skb->next;
4078 if (unlikely(skb_shinfo(nskb)->nr_frags >=
4080 net_warn_ratelimited(
4081 "skb_segment: too many frags: %u %u\n",
4087 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4088 __skb_frag_ref(nskb_frag);
4089 size = skb_frag_size(nskb_frag);
4092 skb_frag_off_add(nskb_frag, offset - pos);
4093 skb_frag_size_sub(nskb_frag, offset - pos);
4096 skb_shinfo(nskb)->nr_frags++;
4098 if (pos + size <= offset + len) {
4103 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4111 nskb->data_len = len - hsize;
4112 nskb->len += nskb->data_len;
4113 nskb->truesize += nskb->data_len;
4117 if (skb_has_shared_frag(nskb) &&
4118 __skb_linearize(nskb))
4121 if (!nskb->remcsum_offload)
4122 nskb->ip_summed = CHECKSUM_NONE;
4123 SKB_GSO_CB(nskb)->csum =
4124 skb_checksum(nskb, doffset,
4125 nskb->len - doffset, 0);
4126 SKB_GSO_CB(nskb)->csum_start =
4127 skb_headroom(nskb) + doffset;
4129 } while ((offset += len) < head_skb->len);
4131 /* Some callers want to get the end of the list.
4132 * Put it in segs->prev to avoid walking the list.
4133 * (see validate_xmit_skb_list() for example)
4138 struct sk_buff *iter;
4139 int type = skb_shinfo(head_skb)->gso_type;
4140 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4142 /* Update type to add partial and then remove dodgy if set */
4143 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4144 type &= ~SKB_GSO_DODGY;
4146 /* Update GSO info and prepare to start updating headers on
4147 * our way back down the stack of protocols.
4149 for (iter = segs; iter; iter = iter->next) {
4150 skb_shinfo(iter)->gso_size = gso_size;
4151 skb_shinfo(iter)->gso_segs = partial_segs;
4152 skb_shinfo(iter)->gso_type = type;
4153 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4156 if (tail->len - doffset <= gso_size)
4157 skb_shinfo(tail)->gso_size = 0;
4158 else if (tail != segs)
4159 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4162 /* Following permits correct backpressure, for protocols
4163 * using skb_set_owner_w().
4164 * Idea is to tranfert ownership from head_skb to last segment.
4166 if (head_skb->destructor == sock_wfree) {
4167 swap(tail->truesize, head_skb->truesize);
4168 swap(tail->destructor, head_skb->destructor);
4169 swap(tail->sk, head_skb->sk);
4174 kfree_skb_list(segs);
4175 return ERR_PTR(err);
4177 EXPORT_SYMBOL_GPL(skb_segment);
4179 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4181 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4182 unsigned int offset = skb_gro_offset(skb);
4183 unsigned int headlen = skb_headlen(skb);
4184 unsigned int len = skb_gro_len(skb);
4185 unsigned int delta_truesize;
4188 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4191 lp = NAPI_GRO_CB(p)->last;
4192 pinfo = skb_shinfo(lp);
4194 if (headlen <= offset) {
4197 int i = skbinfo->nr_frags;
4198 int nr_frags = pinfo->nr_frags + i;
4200 if (nr_frags > MAX_SKB_FRAGS)
4204 pinfo->nr_frags = nr_frags;
4205 skbinfo->nr_frags = 0;
4207 frag = pinfo->frags + nr_frags;
4208 frag2 = skbinfo->frags + i;
4213 skb_frag_off_add(frag, offset);
4214 skb_frag_size_sub(frag, offset);
4216 /* all fragments truesize : remove (head size + sk_buff) */
4217 delta_truesize = skb->truesize -
4218 SKB_TRUESIZE(skb_end_offset(skb));
4220 skb->truesize -= skb->data_len;
4221 skb->len -= skb->data_len;
4224 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4226 } else if (skb->head_frag) {
4227 int nr_frags = pinfo->nr_frags;
4228 skb_frag_t *frag = pinfo->frags + nr_frags;
4229 struct page *page = virt_to_head_page(skb->head);
4230 unsigned int first_size = headlen - offset;
4231 unsigned int first_offset;
4233 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4236 first_offset = skb->data -
4237 (unsigned char *)page_address(page) +
4240 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4242 __skb_frag_set_page(frag, page);
4243 skb_frag_off_set(frag, first_offset);
4244 skb_frag_size_set(frag, first_size);
4246 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4247 /* We dont need to clear skbinfo->nr_frags here */
4249 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4250 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4255 delta_truesize = skb->truesize;
4256 if (offset > headlen) {
4257 unsigned int eat = offset - headlen;
4259 skb_frag_off_add(&skbinfo->frags[0], eat);
4260 skb_frag_size_sub(&skbinfo->frags[0], eat);
4261 skb->data_len -= eat;
4266 __skb_pull(skb, offset);
4268 if (NAPI_GRO_CB(p)->last == p)
4269 skb_shinfo(p)->frag_list = skb;
4271 NAPI_GRO_CB(p)->last->next = skb;
4272 NAPI_GRO_CB(p)->last = skb;
4273 __skb_header_release(skb);
4277 NAPI_GRO_CB(p)->count++;
4279 p->truesize += delta_truesize;
4282 lp->data_len += len;
4283 lp->truesize += delta_truesize;
4286 NAPI_GRO_CB(skb)->same_flow = 1;
4290 #ifdef CONFIG_SKB_EXTENSIONS
4291 #define SKB_EXT_ALIGN_VALUE 8
4292 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4294 static const u8 skb_ext_type_len[] = {
4295 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4296 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4299 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4301 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4302 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4304 #if IS_ENABLED(CONFIG_MPTCP)
4305 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4309 static __always_inline unsigned int skb_ext_total_length(void)
4311 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4312 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4313 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4316 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4318 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4319 skb_ext_type_len[TC_SKB_EXT] +
4321 #if IS_ENABLED(CONFIG_MPTCP)
4322 skb_ext_type_len[SKB_EXT_MPTCP] +
4327 static void skb_extensions_init(void)
4329 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4330 BUILD_BUG_ON(skb_ext_total_length() > 255);
4332 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4333 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4335 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4339 static void skb_extensions_init(void) {}
4342 void __init skb_init(void)
4344 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4345 sizeof(struct sk_buff),
4347 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4348 offsetof(struct sk_buff, cb),
4349 sizeof_field(struct sk_buff, cb),
4351 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4352 sizeof(struct sk_buff_fclones),
4354 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4356 skb_extensions_init();
4360 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4361 unsigned int recursion_level)
4363 int start = skb_headlen(skb);
4364 int i, copy = start - offset;
4365 struct sk_buff *frag_iter;
4368 if (unlikely(recursion_level >= 24))
4374 sg_set_buf(sg, skb->data + offset, copy);
4376 if ((len -= copy) == 0)
4381 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4384 WARN_ON(start > offset + len);
4386 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4387 if ((copy = end - offset) > 0) {
4388 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4389 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4394 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4395 skb_frag_off(frag) + offset - start);
4404 skb_walk_frags(skb, frag_iter) {
4407 WARN_ON(start > offset + len);
4409 end = start + frag_iter->len;
4410 if ((copy = end - offset) > 0) {
4411 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4416 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4417 copy, recursion_level + 1);
4418 if (unlikely(ret < 0))
4421 if ((len -= copy) == 0)
4432 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4433 * @skb: Socket buffer containing the buffers to be mapped
4434 * @sg: The scatter-gather list to map into
4435 * @offset: The offset into the buffer's contents to start mapping
4436 * @len: Length of buffer space to be mapped
4438 * Fill the specified scatter-gather list with mappings/pointers into a
4439 * region of the buffer space attached to a socket buffer. Returns either
4440 * the number of scatterlist items used, or -EMSGSIZE if the contents
4443 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4445 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4450 sg_mark_end(&sg[nsg - 1]);
4454 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4456 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4457 * sglist without mark the sg which contain last skb data as the end.
4458 * So the caller can mannipulate sg list as will when padding new data after
4459 * the first call without calling sg_unmark_end to expend sg list.
4461 * Scenario to use skb_to_sgvec_nomark:
4463 * 2. skb_to_sgvec_nomark(payload1)
4464 * 3. skb_to_sgvec_nomark(payload2)
4466 * This is equivalent to:
4468 * 2. skb_to_sgvec(payload1)
4470 * 4. skb_to_sgvec(payload2)
4472 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4473 * is more preferable.
4475 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4476 int offset, int len)
4478 return __skb_to_sgvec(skb, sg, offset, len, 0);
4480 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4485 * skb_cow_data - Check that a socket buffer's data buffers are writable
4486 * @skb: The socket buffer to check.
4487 * @tailbits: Amount of trailing space to be added
4488 * @trailer: Returned pointer to the skb where the @tailbits space begins
4490 * Make sure that the data buffers attached to a socket buffer are
4491 * writable. If they are not, private copies are made of the data buffers
4492 * and the socket buffer is set to use these instead.
4494 * If @tailbits is given, make sure that there is space to write @tailbits
4495 * bytes of data beyond current end of socket buffer. @trailer will be
4496 * set to point to the skb in which this space begins.
4498 * The number of scatterlist elements required to completely map the
4499 * COW'd and extended socket buffer will be returned.
4501 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4505 struct sk_buff *skb1, **skb_p;
4507 /* If skb is cloned or its head is paged, reallocate
4508 * head pulling out all the pages (pages are considered not writable
4509 * at the moment even if they are anonymous).
4511 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4512 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4515 /* Easy case. Most of packets will go this way. */
4516 if (!skb_has_frag_list(skb)) {
4517 /* A little of trouble, not enough of space for trailer.
4518 * This should not happen, when stack is tuned to generate
4519 * good frames. OK, on miss we reallocate and reserve even more
4520 * space, 128 bytes is fair. */
4522 if (skb_tailroom(skb) < tailbits &&
4523 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4531 /* Misery. We are in troubles, going to mincer fragments... */
4534 skb_p = &skb_shinfo(skb)->frag_list;
4537 while ((skb1 = *skb_p) != NULL) {
4540 /* The fragment is partially pulled by someone,
4541 * this can happen on input. Copy it and everything
4544 if (skb_shared(skb1))
4547 /* If the skb is the last, worry about trailer. */
4549 if (skb1->next == NULL && tailbits) {
4550 if (skb_shinfo(skb1)->nr_frags ||
4551 skb_has_frag_list(skb1) ||
4552 skb_tailroom(skb1) < tailbits)
4553 ntail = tailbits + 128;
4559 skb_shinfo(skb1)->nr_frags ||
4560 skb_has_frag_list(skb1)) {
4561 struct sk_buff *skb2;
4563 /* Fuck, we are miserable poor guys... */
4565 skb2 = skb_copy(skb1, GFP_ATOMIC);
4567 skb2 = skb_copy_expand(skb1,
4571 if (unlikely(skb2 == NULL))
4575 skb_set_owner_w(skb2, skb1->sk);
4577 /* Looking around. Are we still alive?
4578 * OK, link new skb, drop old one */
4580 skb2->next = skb1->next;
4587 skb_p = &skb1->next;
4592 EXPORT_SYMBOL_GPL(skb_cow_data);
4594 static void sock_rmem_free(struct sk_buff *skb)
4596 struct sock *sk = skb->sk;
4598 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4601 static void skb_set_err_queue(struct sk_buff *skb)
4603 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4604 * So, it is safe to (mis)use it to mark skbs on the error queue.
4606 skb->pkt_type = PACKET_OUTGOING;
4607 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4611 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4613 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4615 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4616 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4621 skb->destructor = sock_rmem_free;
4622 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4623 skb_set_err_queue(skb);
4625 /* before exiting rcu section, make sure dst is refcounted */
4628 skb_queue_tail(&sk->sk_error_queue, skb);
4629 if (!sock_flag(sk, SOCK_DEAD))
4630 sk->sk_error_report(sk);
4633 EXPORT_SYMBOL(sock_queue_err_skb);
4635 static bool is_icmp_err_skb(const struct sk_buff *skb)
4637 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4638 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4641 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4643 struct sk_buff_head *q = &sk->sk_error_queue;
4644 struct sk_buff *skb, *skb_next = NULL;
4645 bool icmp_next = false;
4646 unsigned long flags;
4648 spin_lock_irqsave(&q->lock, flags);
4649 skb = __skb_dequeue(q);
4650 if (skb && (skb_next = skb_peek(q))) {
4651 icmp_next = is_icmp_err_skb(skb_next);
4653 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
4655 spin_unlock_irqrestore(&q->lock, flags);
4657 if (is_icmp_err_skb(skb) && !icmp_next)
4661 sk->sk_error_report(sk);
4665 EXPORT_SYMBOL(sock_dequeue_err_skb);
4668 * skb_clone_sk - create clone of skb, and take reference to socket
4669 * @skb: the skb to clone
4671 * This function creates a clone of a buffer that holds a reference on
4672 * sk_refcnt. Buffers created via this function are meant to be
4673 * returned using sock_queue_err_skb, or free via kfree_skb.
4675 * When passing buffers allocated with this function to sock_queue_err_skb
4676 * it is necessary to wrap the call with sock_hold/sock_put in order to
4677 * prevent the socket from being released prior to being enqueued on
4678 * the sk_error_queue.
4680 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4682 struct sock *sk = skb->sk;
4683 struct sk_buff *clone;
4685 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4688 clone = skb_clone(skb, GFP_ATOMIC);
4695 clone->destructor = sock_efree;
4699 EXPORT_SYMBOL(skb_clone_sk);
4701 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4706 struct sock_exterr_skb *serr;
4709 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4711 serr = SKB_EXT_ERR(skb);
4712 memset(serr, 0, sizeof(*serr));
4713 serr->ee.ee_errno = ENOMSG;
4714 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4715 serr->ee.ee_info = tstype;
4716 serr->opt_stats = opt_stats;
4717 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4718 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4719 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4720 if (sk->sk_protocol == IPPROTO_TCP &&
4721 sk->sk_type == SOCK_STREAM)
4722 serr->ee.ee_data -= sk->sk_tskey;
4725 err = sock_queue_err_skb(sk, skb);
4731 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4735 if (likely(sysctl_tstamp_allow_data || tsonly))
4738 read_lock_bh(&sk->sk_callback_lock);
4739 ret = sk->sk_socket && sk->sk_socket->file &&
4740 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4741 read_unlock_bh(&sk->sk_callback_lock);
4745 void skb_complete_tx_timestamp(struct sk_buff *skb,
4746 struct skb_shared_hwtstamps *hwtstamps)
4748 struct sock *sk = skb->sk;
4750 if (!skb_may_tx_timestamp(sk, false))
4753 /* Take a reference to prevent skb_orphan() from freeing the socket,
4754 * but only if the socket refcount is not zero.
4756 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4757 *skb_hwtstamps(skb) = *hwtstamps;
4758 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4766 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4768 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4769 const struct sk_buff *ack_skb,
4770 struct skb_shared_hwtstamps *hwtstamps,
4771 struct sock *sk, int tstype)
4773 struct sk_buff *skb;
4774 bool tsonly, opt_stats = false;
4779 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4780 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4783 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4784 if (!skb_may_tx_timestamp(sk, tsonly))
4789 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4790 sk->sk_protocol == IPPROTO_TCP &&
4791 sk->sk_type == SOCK_STREAM) {
4792 skb = tcp_get_timestamping_opt_stats(sk, orig_skb,
4797 skb = alloc_skb(0, GFP_ATOMIC);
4799 skb = skb_clone(orig_skb, GFP_ATOMIC);
4805 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4807 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4811 *skb_hwtstamps(skb) = *hwtstamps;
4813 skb->tstamp = ktime_get_real();
4815 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4817 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4819 void skb_tstamp_tx(struct sk_buff *orig_skb,
4820 struct skb_shared_hwtstamps *hwtstamps)
4822 return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk,
4825 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4827 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4829 struct sock *sk = skb->sk;
4830 struct sock_exterr_skb *serr;
4833 skb->wifi_acked_valid = 1;
4834 skb->wifi_acked = acked;
4836 serr = SKB_EXT_ERR(skb);
4837 memset(serr, 0, sizeof(*serr));
4838 serr->ee.ee_errno = ENOMSG;
4839 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4841 /* Take a reference to prevent skb_orphan() from freeing the socket,
4842 * but only if the socket refcount is not zero.
4844 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4845 err = sock_queue_err_skb(sk, skb);
4851 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4854 * skb_partial_csum_set - set up and verify partial csum values for packet
4855 * @skb: the skb to set
4856 * @start: the number of bytes after skb->data to start checksumming.
4857 * @off: the offset from start to place the checksum.
4859 * For untrusted partially-checksummed packets, we need to make sure the values
4860 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4862 * This function checks and sets those values and skb->ip_summed: if this
4863 * returns false you should drop the packet.
4865 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4867 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4868 u32 csum_start = skb_headroom(skb) + (u32)start;
4870 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4871 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4872 start, off, skb_headroom(skb), skb_headlen(skb));
4875 skb->ip_summed = CHECKSUM_PARTIAL;
4876 skb->csum_start = csum_start;
4877 skb->csum_offset = off;
4878 skb_set_transport_header(skb, start);
4881 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4883 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4886 if (skb_headlen(skb) >= len)
4889 /* If we need to pullup then pullup to the max, so we
4890 * won't need to do it again.
4895 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4898 if (skb_headlen(skb) < len)
4904 #define MAX_TCP_HDR_LEN (15 * 4)
4906 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4907 typeof(IPPROTO_IP) proto,
4914 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4915 off + MAX_TCP_HDR_LEN);
4916 if (!err && !skb_partial_csum_set(skb, off,
4917 offsetof(struct tcphdr,
4920 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4923 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4924 off + sizeof(struct udphdr));
4925 if (!err && !skb_partial_csum_set(skb, off,
4926 offsetof(struct udphdr,
4929 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4932 return ERR_PTR(-EPROTO);
4935 /* This value should be large enough to cover a tagged ethernet header plus
4936 * maximally sized IP and TCP or UDP headers.
4938 #define MAX_IP_HDR_LEN 128
4940 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4949 err = skb_maybe_pull_tail(skb,
4950 sizeof(struct iphdr),
4955 if (ip_is_fragment(ip_hdr(skb)))
4958 off = ip_hdrlen(skb);
4965 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4967 return PTR_ERR(csum);
4970 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4973 ip_hdr(skb)->protocol, 0);
4980 /* This value should be large enough to cover a tagged ethernet header plus
4981 * an IPv6 header, all options, and a maximal TCP or UDP header.
4983 #define MAX_IPV6_HDR_LEN 256
4985 #define OPT_HDR(type, skb, off) \
4986 (type *)(skb_network_header(skb) + (off))
4988 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
5001 off = sizeof(struct ipv6hdr);
5003 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
5007 nexthdr = ipv6_hdr(skb)->nexthdr;
5009 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
5010 while (off <= len && !done) {
5012 case IPPROTO_DSTOPTS:
5013 case IPPROTO_HOPOPTS:
5014 case IPPROTO_ROUTING: {
5015 struct ipv6_opt_hdr *hp;
5017 err = skb_maybe_pull_tail(skb,
5019 sizeof(struct ipv6_opt_hdr),
5024 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
5025 nexthdr = hp->nexthdr;
5026 off += ipv6_optlen(hp);
5030 struct ip_auth_hdr *hp;
5032 err = skb_maybe_pull_tail(skb,
5034 sizeof(struct ip_auth_hdr),
5039 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
5040 nexthdr = hp->nexthdr;
5041 off += ipv6_authlen(hp);
5044 case IPPROTO_FRAGMENT: {
5045 struct frag_hdr *hp;
5047 err = skb_maybe_pull_tail(skb,
5049 sizeof(struct frag_hdr),
5054 hp = OPT_HDR(struct frag_hdr, skb, off);
5056 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5059 nexthdr = hp->nexthdr;
5060 off += sizeof(struct frag_hdr);
5071 if (!done || fragment)
5074 csum = skb_checksum_setup_ip(skb, nexthdr, off);
5076 return PTR_ERR(csum);
5079 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5080 &ipv6_hdr(skb)->daddr,
5081 skb->len - off, nexthdr, 0);
5089 * skb_checksum_setup - set up partial checksum offset
5090 * @skb: the skb to set up
5091 * @recalculate: if true the pseudo-header checksum will be recalculated
5093 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5097 switch (skb->protocol) {
5098 case htons(ETH_P_IP):
5099 err = skb_checksum_setup_ipv4(skb, recalculate);
5102 case htons(ETH_P_IPV6):
5103 err = skb_checksum_setup_ipv6(skb, recalculate);
5113 EXPORT_SYMBOL(skb_checksum_setup);
5116 * skb_checksum_maybe_trim - maybe trims the given skb
5117 * @skb: the skb to check
5118 * @transport_len: the data length beyond the network header
5120 * Checks whether the given skb has data beyond the given transport length.
5121 * If so, returns a cloned skb trimmed to this transport length.
5122 * Otherwise returns the provided skb. Returns NULL in error cases
5123 * (e.g. transport_len exceeds skb length or out-of-memory).
5125 * Caller needs to set the skb transport header and free any returned skb if it
5126 * differs from the provided skb.
5128 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5129 unsigned int transport_len)
5131 struct sk_buff *skb_chk;
5132 unsigned int len = skb_transport_offset(skb) + transport_len;
5137 else if (skb->len == len)
5140 skb_chk = skb_clone(skb, GFP_ATOMIC);
5144 ret = pskb_trim_rcsum(skb_chk, len);
5154 * skb_checksum_trimmed - validate checksum of an skb
5155 * @skb: the skb to check
5156 * @transport_len: the data length beyond the network header
5157 * @skb_chkf: checksum function to use
5159 * Applies the given checksum function skb_chkf to the provided skb.
5160 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5162 * If the skb has data beyond the given transport length, then a
5163 * trimmed & cloned skb is checked and returned.
5165 * Caller needs to set the skb transport header and free any returned skb if it
5166 * differs from the provided skb.
5168 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5169 unsigned int transport_len,
5170 __sum16(*skb_chkf)(struct sk_buff *skb))
5172 struct sk_buff *skb_chk;
5173 unsigned int offset = skb_transport_offset(skb);
5176 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5180 if (!pskb_may_pull(skb_chk, offset))
5183 skb_pull_rcsum(skb_chk, offset);
5184 ret = skb_chkf(skb_chk);
5185 skb_push_rcsum(skb_chk, offset);
5193 if (skb_chk && skb_chk != skb)
5199 EXPORT_SYMBOL(skb_checksum_trimmed);
5201 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5203 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5206 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5208 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5211 skb_release_head_state(skb);
5212 kmem_cache_free(skbuff_head_cache, skb);
5217 EXPORT_SYMBOL(kfree_skb_partial);
5220 * skb_try_coalesce - try to merge skb to prior one
5222 * @from: buffer to add
5223 * @fragstolen: pointer to boolean
5224 * @delta_truesize: how much more was allocated than was requested
5226 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5227 bool *fragstolen, int *delta_truesize)
5229 struct skb_shared_info *to_shinfo, *from_shinfo;
5230 int i, delta, len = from->len;
5232 *fragstolen = false;
5237 if (len <= skb_tailroom(to)) {
5239 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5240 *delta_truesize = 0;
5244 to_shinfo = skb_shinfo(to);
5245 from_shinfo = skb_shinfo(from);
5246 if (to_shinfo->frag_list || from_shinfo->frag_list)
5248 if (skb_zcopy(to) || skb_zcopy(from))
5251 if (skb_headlen(from) != 0) {
5253 unsigned int offset;
5255 if (to_shinfo->nr_frags +
5256 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5259 if (skb_head_is_locked(from))
5262 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5264 page = virt_to_head_page(from->head);
5265 offset = from->data - (unsigned char *)page_address(page);
5267 skb_fill_page_desc(to, to_shinfo->nr_frags,
5268 page, offset, skb_headlen(from));
5271 if (to_shinfo->nr_frags +
5272 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5275 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5278 WARN_ON_ONCE(delta < len);
5280 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5282 from_shinfo->nr_frags * sizeof(skb_frag_t));
5283 to_shinfo->nr_frags += from_shinfo->nr_frags;
5285 if (!skb_cloned(from))
5286 from_shinfo->nr_frags = 0;
5288 /* if the skb is not cloned this does nothing
5289 * since we set nr_frags to 0.
5291 for (i = 0; i < from_shinfo->nr_frags; i++)
5292 __skb_frag_ref(&from_shinfo->frags[i]);
5294 to->truesize += delta;
5296 to->data_len += len;
5298 *delta_truesize = delta;
5301 EXPORT_SYMBOL(skb_try_coalesce);
5304 * skb_scrub_packet - scrub an skb
5306 * @skb: buffer to clean
5307 * @xnet: packet is crossing netns
5309 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5310 * into/from a tunnel. Some information have to be cleared during these
5312 * skb_scrub_packet can also be used to clean a skb before injecting it in
5313 * another namespace (@xnet == true). We have to clear all information in the
5314 * skb that could impact namespace isolation.
5316 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5318 skb->pkt_type = PACKET_HOST;
5324 nf_reset_trace(skb);
5326 #ifdef CONFIG_NET_SWITCHDEV
5327 skb->offload_fwd_mark = 0;
5328 skb->offload_l3_fwd_mark = 0;
5338 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5341 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5345 * skb_gso_transport_seglen is used to determine the real size of the
5346 * individual segments, including Layer4 headers (TCP/UDP).
5348 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5350 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5352 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5353 unsigned int thlen = 0;
5355 if (skb->encapsulation) {
5356 thlen = skb_inner_transport_header(skb) -
5357 skb_transport_header(skb);
5359 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5360 thlen += inner_tcp_hdrlen(skb);
5361 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5362 thlen = tcp_hdrlen(skb);
5363 } else if (unlikely(skb_is_gso_sctp(skb))) {
5364 thlen = sizeof(struct sctphdr);
5365 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5366 thlen = sizeof(struct udphdr);
5368 /* UFO sets gso_size to the size of the fragmentation
5369 * payload, i.e. the size of the L4 (UDP) header is already
5372 return thlen + shinfo->gso_size;
5376 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5380 * skb_gso_network_seglen is used to determine the real size of the
5381 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5383 * The MAC/L2 header is not accounted for.
5385 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5387 unsigned int hdr_len = skb_transport_header(skb) -
5388 skb_network_header(skb);
5390 return hdr_len + skb_gso_transport_seglen(skb);
5394 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5398 * skb_gso_mac_seglen is used to determine the real size of the
5399 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5400 * headers (TCP/UDP).
5402 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5404 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5406 return hdr_len + skb_gso_transport_seglen(skb);
5410 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5412 * There are a couple of instances where we have a GSO skb, and we
5413 * want to determine what size it would be after it is segmented.
5415 * We might want to check:
5416 * - L3+L4+payload size (e.g. IP forwarding)
5417 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5419 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5423 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5424 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5426 * @max_len: The maximum permissible length.
5428 * Returns true if the segmented length <= max length.
5430 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5431 unsigned int seg_len,
5432 unsigned int max_len) {
5433 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5434 const struct sk_buff *iter;
5436 if (shinfo->gso_size != GSO_BY_FRAGS)
5437 return seg_len <= max_len;
5439 /* Undo this so we can re-use header sizes */
5440 seg_len -= GSO_BY_FRAGS;
5442 skb_walk_frags(skb, iter) {
5443 if (seg_len + skb_headlen(iter) > max_len)
5451 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5454 * @mtu: MTU to validate against
5456 * skb_gso_validate_network_len validates if a given skb will fit a
5457 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5460 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5462 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5464 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5467 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5470 * @len: length to validate against
5472 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5473 * length once split, including L2, L3 and L4 headers and the payload.
5475 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5477 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5479 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5481 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5483 int mac_len, meta_len;
5486 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5491 mac_len = skb->data - skb_mac_header(skb);
5492 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5493 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5494 mac_len - VLAN_HLEN - ETH_TLEN);
5497 meta_len = skb_metadata_len(skb);
5499 meta = skb_metadata_end(skb) - meta_len;
5500 memmove(meta + VLAN_HLEN, meta, meta_len);
5503 skb->mac_header += VLAN_HLEN;
5507 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5509 struct vlan_hdr *vhdr;
5512 if (unlikely(skb_vlan_tag_present(skb))) {
5513 /* vlan_tci is already set-up so leave this for another time */
5517 skb = skb_share_check(skb, GFP_ATOMIC);
5520 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5521 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5524 vhdr = (struct vlan_hdr *)skb->data;
5525 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5526 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5528 skb_pull_rcsum(skb, VLAN_HLEN);
5529 vlan_set_encap_proto(skb, vhdr);
5531 skb = skb_reorder_vlan_header(skb);
5535 skb_reset_network_header(skb);
5536 if (!skb_transport_header_was_set(skb))
5537 skb_reset_transport_header(skb);
5538 skb_reset_mac_len(skb);
5546 EXPORT_SYMBOL(skb_vlan_untag);
5548 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5550 if (!pskb_may_pull(skb, write_len))
5553 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5556 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5558 EXPORT_SYMBOL(skb_ensure_writable);
5560 /* remove VLAN header from packet and update csum accordingly.
5561 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5563 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5565 struct vlan_hdr *vhdr;
5566 int offset = skb->data - skb_mac_header(skb);
5569 if (WARN_ONCE(offset,
5570 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5575 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5579 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5581 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5582 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5584 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5585 __skb_pull(skb, VLAN_HLEN);
5587 vlan_set_encap_proto(skb, vhdr);
5588 skb->mac_header += VLAN_HLEN;
5590 if (skb_network_offset(skb) < ETH_HLEN)
5591 skb_set_network_header(skb, ETH_HLEN);
5593 skb_reset_mac_len(skb);
5597 EXPORT_SYMBOL(__skb_vlan_pop);
5599 /* Pop a vlan tag either from hwaccel or from payload.
5600 * Expects skb->data at mac header.
5602 int skb_vlan_pop(struct sk_buff *skb)
5608 if (likely(skb_vlan_tag_present(skb))) {
5609 __vlan_hwaccel_clear_tag(skb);
5611 if (unlikely(!eth_type_vlan(skb->protocol)))
5614 err = __skb_vlan_pop(skb, &vlan_tci);
5618 /* move next vlan tag to hw accel tag */
5619 if (likely(!eth_type_vlan(skb->protocol)))
5622 vlan_proto = skb->protocol;
5623 err = __skb_vlan_pop(skb, &vlan_tci);
5627 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5630 EXPORT_SYMBOL(skb_vlan_pop);
5632 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5633 * Expects skb->data at mac header.
5635 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5637 if (skb_vlan_tag_present(skb)) {
5638 int offset = skb->data - skb_mac_header(skb);
5641 if (WARN_ONCE(offset,
5642 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5647 err = __vlan_insert_tag(skb, skb->vlan_proto,
5648 skb_vlan_tag_get(skb));
5652 skb->protocol = skb->vlan_proto;
5653 skb->mac_len += VLAN_HLEN;
5655 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5657 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5660 EXPORT_SYMBOL(skb_vlan_push);
5663 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
5665 * @skb: Socket buffer to modify
5667 * Drop the Ethernet header of @skb.
5669 * Expects that skb->data points to the mac header and that no VLAN tags are
5672 * Returns 0 on success, -errno otherwise.
5674 int skb_eth_pop(struct sk_buff *skb)
5676 if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
5677 skb_network_offset(skb) < ETH_HLEN)
5680 skb_pull_rcsum(skb, ETH_HLEN);
5681 skb_reset_mac_header(skb);
5682 skb_reset_mac_len(skb);
5686 EXPORT_SYMBOL(skb_eth_pop);
5689 * skb_eth_push() - Add a new Ethernet header at the head of a packet
5691 * @skb: Socket buffer to modify
5692 * @dst: Destination MAC address of the new header
5693 * @src: Source MAC address of the new header
5695 * Prepend @skb with a new Ethernet header.
5697 * Expects that skb->data points to the mac header, which must be empty.
5699 * Returns 0 on success, -errno otherwise.
5701 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
5702 const unsigned char *src)
5707 if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
5710 err = skb_cow_head(skb, sizeof(*eth));
5714 skb_push(skb, sizeof(*eth));
5715 skb_reset_mac_header(skb);
5716 skb_reset_mac_len(skb);
5719 ether_addr_copy(eth->h_dest, dst);
5720 ether_addr_copy(eth->h_source, src);
5721 eth->h_proto = skb->protocol;
5723 skb_postpush_rcsum(skb, eth, sizeof(*eth));
5727 EXPORT_SYMBOL(skb_eth_push);
5729 /* Update the ethertype of hdr and the skb csum value if required. */
5730 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5733 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5734 __be16 diff[] = { ~hdr->h_proto, ethertype };
5736 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5739 hdr->h_proto = ethertype;
5743 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
5747 * @mpls_lse: MPLS label stack entry to push
5748 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5749 * @mac_len: length of the MAC header
5750 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
5753 * Expects skb->data at mac header.
5755 * Returns 0 on success, -errno otherwise.
5757 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5758 int mac_len, bool ethernet)
5760 struct mpls_shim_hdr *lse;
5763 if (unlikely(!eth_p_mpls(mpls_proto)))
5766 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5767 if (skb->encapsulation)
5770 err = skb_cow_head(skb, MPLS_HLEN);
5774 if (!skb->inner_protocol) {
5775 skb_set_inner_network_header(skb, skb_network_offset(skb));
5776 skb_set_inner_protocol(skb, skb->protocol);
5779 skb_push(skb, MPLS_HLEN);
5780 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5782 skb_reset_mac_header(skb);
5783 skb_set_network_header(skb, mac_len);
5784 skb_reset_mac_len(skb);
5786 lse = mpls_hdr(skb);
5787 lse->label_stack_entry = mpls_lse;
5788 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5790 if (ethernet && mac_len >= ETH_HLEN)
5791 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5792 skb->protocol = mpls_proto;
5796 EXPORT_SYMBOL_GPL(skb_mpls_push);
5799 * skb_mpls_pop() - pop the outermost MPLS header
5802 * @next_proto: ethertype of header after popped MPLS header
5803 * @mac_len: length of the MAC header
5804 * @ethernet: flag to indicate if the packet is ethernet
5806 * Expects skb->data at mac header.
5808 * Returns 0 on success, -errno otherwise.
5810 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5815 if (unlikely(!eth_p_mpls(skb->protocol)))
5818 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5822 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5823 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5826 __skb_pull(skb, MPLS_HLEN);
5827 skb_reset_mac_header(skb);
5828 skb_set_network_header(skb, mac_len);
5830 if (ethernet && mac_len >= ETH_HLEN) {
5833 /* use mpls_hdr() to get ethertype to account for VLANs. */
5834 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5835 skb_mod_eth_type(skb, hdr, next_proto);
5837 skb->protocol = next_proto;
5841 EXPORT_SYMBOL_GPL(skb_mpls_pop);
5844 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5847 * @mpls_lse: new MPLS label stack entry to update to
5849 * Expects skb->data at mac header.
5851 * Returns 0 on success, -errno otherwise.
5853 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5857 if (unlikely(!eth_p_mpls(skb->protocol)))
5860 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5864 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5865 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5867 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5870 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5874 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5877 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5881 * Expects skb->data at mac header.
5883 * Returns 0 on success, -errno otherwise.
5885 int skb_mpls_dec_ttl(struct sk_buff *skb)
5890 if (unlikely(!eth_p_mpls(skb->protocol)))
5893 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
5896 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5897 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5901 lse &= ~MPLS_LS_TTL_MASK;
5902 lse |= ttl << MPLS_LS_TTL_SHIFT;
5904 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5906 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5909 * alloc_skb_with_frags - allocate skb with page frags
5911 * @header_len: size of linear part
5912 * @data_len: needed length in frags
5913 * @max_page_order: max page order desired.
5914 * @errcode: pointer to error code if any
5915 * @gfp_mask: allocation mask
5917 * This can be used to allocate a paged skb, given a maximal order for frags.
5919 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5920 unsigned long data_len,
5925 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5926 unsigned long chunk;
5927 struct sk_buff *skb;
5931 *errcode = -EMSGSIZE;
5932 /* Note this test could be relaxed, if we succeed to allocate
5933 * high order pages...
5935 if (npages > MAX_SKB_FRAGS)
5938 *errcode = -ENOBUFS;
5939 skb = alloc_skb(header_len, gfp_mask);
5943 skb->truesize += npages << PAGE_SHIFT;
5945 for (i = 0; npages > 0; i++) {
5946 int order = max_page_order;
5949 if (npages >= 1 << order) {
5950 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5956 /* Do not retry other high order allocations */
5962 page = alloc_page(gfp_mask);
5966 chunk = min_t(unsigned long, data_len,
5967 PAGE_SIZE << order);
5968 skb_fill_page_desc(skb, i, page, 0, chunk);
5970 npages -= 1 << order;
5978 EXPORT_SYMBOL(alloc_skb_with_frags);
5980 /* carve out the first off bytes from skb when off < headlen */
5981 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5982 const int headlen, gfp_t gfp_mask)
5985 int size = skb_end_offset(skb);
5986 int new_hlen = headlen - off;
5989 size = SKB_DATA_ALIGN(size);
5991 if (skb_pfmemalloc(skb))
5992 gfp_mask |= __GFP_MEMALLOC;
5993 data = kmalloc_reserve(size +
5994 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5995 gfp_mask, NUMA_NO_NODE, NULL);
5999 size = SKB_WITH_OVERHEAD(ksize(data));
6001 /* Copy real data, and all frags */
6002 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
6005 memcpy((struct skb_shared_info *)(data + size),
6007 offsetof(struct skb_shared_info,
6008 frags[skb_shinfo(skb)->nr_frags]));
6009 if (skb_cloned(skb)) {
6010 /* drop the old head gracefully */
6011 if (skb_orphan_frags(skb, gfp_mask)) {
6015 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
6016 skb_frag_ref(skb, i);
6017 if (skb_has_frag_list(skb))
6018 skb_clone_fraglist(skb);
6019 skb_release_data(skb);
6021 /* we can reuse existing recount- all we did was
6030 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6033 skb->end = skb->head + size;
6035 skb_set_tail_pointer(skb, skb_headlen(skb));
6036 skb_headers_offset_update(skb, 0);
6040 atomic_set(&skb_shinfo(skb)->dataref, 1);
6045 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6047 /* carve out the first eat bytes from skb's frag_list. May recurse into
6050 static int pskb_carve_frag_list(struct sk_buff *skb,
6051 struct skb_shared_info *shinfo, int eat,
6054 struct sk_buff *list = shinfo->frag_list;
6055 struct sk_buff *clone = NULL;
6056 struct sk_buff *insp = NULL;
6060 pr_err("Not enough bytes to eat. Want %d\n", eat);
6063 if (list->len <= eat) {
6064 /* Eaten as whole. */
6069 /* Eaten partially. */
6070 if (skb_shared(list)) {
6071 clone = skb_clone(list, gfp_mask);
6077 /* This may be pulled without problems. */
6080 if (pskb_carve(list, eat, gfp_mask) < 0) {
6088 /* Free pulled out fragments. */
6089 while ((list = shinfo->frag_list) != insp) {
6090 shinfo->frag_list = list->next;
6093 /* And insert new clone at head. */
6096 shinfo->frag_list = clone;
6101 /* carve off first len bytes from skb. Split line (off) is in the
6102 * non-linear part of skb
6104 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6105 int pos, gfp_t gfp_mask)
6108 int size = skb_end_offset(skb);
6110 const int nfrags = skb_shinfo(skb)->nr_frags;
6111 struct skb_shared_info *shinfo;
6113 size = SKB_DATA_ALIGN(size);
6115 if (skb_pfmemalloc(skb))
6116 gfp_mask |= __GFP_MEMALLOC;
6117 data = kmalloc_reserve(size +
6118 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6119 gfp_mask, NUMA_NO_NODE, NULL);
6123 size = SKB_WITH_OVERHEAD(ksize(data));
6125 memcpy((struct skb_shared_info *)(data + size),
6126 skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6127 if (skb_orphan_frags(skb, gfp_mask)) {
6131 shinfo = (struct skb_shared_info *)(data + size);
6132 for (i = 0; i < nfrags; i++) {
6133 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6135 if (pos + fsize > off) {
6136 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6140 * We have two variants in this case:
6141 * 1. Move all the frag to the second
6142 * part, if it is possible. F.e.
6143 * this approach is mandatory for TUX,
6144 * where splitting is expensive.
6145 * 2. Split is accurately. We make this.
6147 skb_frag_off_add(&shinfo->frags[0], off - pos);
6148 skb_frag_size_sub(&shinfo->frags[0], off - pos);
6150 skb_frag_ref(skb, i);
6155 shinfo->nr_frags = k;
6156 if (skb_has_frag_list(skb))
6157 skb_clone_fraglist(skb);
6159 /* split line is in frag list */
6160 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6161 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6162 if (skb_has_frag_list(skb))
6163 kfree_skb_list(skb_shinfo(skb)->frag_list);
6167 skb_release_data(skb);
6172 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6175 skb->end = skb->head + size;
6177 skb_reset_tail_pointer(skb);
6178 skb_headers_offset_update(skb, 0);
6183 skb->data_len = skb->len;
6184 atomic_set(&skb_shinfo(skb)->dataref, 1);
6188 /* remove len bytes from the beginning of the skb */
6189 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6191 int headlen = skb_headlen(skb);
6194 return pskb_carve_inside_header(skb, len, headlen, gfp);
6196 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6199 /* Extract to_copy bytes starting at off from skb, and return this in
6202 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6203 int to_copy, gfp_t gfp)
6205 struct sk_buff *clone = skb_clone(skb, gfp);
6210 if (pskb_carve(clone, off, gfp) < 0 ||
6211 pskb_trim(clone, to_copy)) {
6217 EXPORT_SYMBOL(pskb_extract);
6220 * skb_condense - try to get rid of fragments/frag_list if possible
6223 * Can be used to save memory before skb is added to a busy queue.
6224 * If packet has bytes in frags and enough tail room in skb->head,
6225 * pull all of them, so that we can free the frags right now and adjust
6228 * We do not reallocate skb->head thus can not fail.
6229 * Caller must re-evaluate skb->truesize if needed.
6231 void skb_condense(struct sk_buff *skb)
6233 if (skb->data_len) {
6234 if (skb->data_len > skb->end - skb->tail ||
6238 /* Nice, we can free page frag(s) right now */
6239 __pskb_pull_tail(skb, skb->data_len);
6241 /* At this point, skb->truesize might be over estimated,
6242 * because skb had a fragment, and fragments do not tell
6244 * When we pulled its content into skb->head, fragment
6245 * was freed, but __pskb_pull_tail() could not possibly
6246 * adjust skb->truesize, not knowing the frag truesize.
6248 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6251 #ifdef CONFIG_SKB_EXTENSIONS
6252 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6254 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6258 * __skb_ext_alloc - allocate a new skb extensions storage
6260 * @flags: See kmalloc().
6262 * Returns the newly allocated pointer. The pointer can later attached to a
6263 * skb via __skb_ext_set().
6264 * Note: caller must handle the skb_ext as an opaque data.
6266 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6268 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6271 memset(new->offset, 0, sizeof(new->offset));
6272 refcount_set(&new->refcnt, 1);
6278 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6279 unsigned int old_active)
6281 struct skb_ext *new;
6283 if (refcount_read(&old->refcnt) == 1)
6286 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6290 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6291 refcount_set(&new->refcnt, 1);
6294 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6295 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6298 for (i = 0; i < sp->len; i++)
6299 xfrm_state_hold(sp->xvec[i]);
6307 * __skb_ext_set - attach the specified extension storage to this skb
6310 * @ext: extension storage previously allocated via __skb_ext_alloc()
6312 * Existing extensions, if any, are cleared.
6314 * Returns the pointer to the extension.
6316 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6317 struct skb_ext *ext)
6319 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6322 newlen = newoff + skb_ext_type_len[id];
6323 ext->chunks = newlen;
6324 ext->offset[id] = newoff;
6325 skb->extensions = ext;
6326 skb->active_extensions = 1 << id;
6327 return skb_ext_get_ptr(ext, id);
6331 * skb_ext_add - allocate space for given extension, COW if needed
6333 * @id: extension to allocate space for
6335 * Allocates enough space for the given extension.
6336 * If the extension is already present, a pointer to that extension
6339 * If the skb was cloned, COW applies and the returned memory can be
6340 * modified without changing the extension space of clones buffers.
6342 * Returns pointer to the extension or NULL on allocation failure.
6344 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6346 struct skb_ext *new, *old = NULL;
6347 unsigned int newlen, newoff;
6349 if (skb->active_extensions) {
6350 old = skb->extensions;
6352 new = skb_ext_maybe_cow(old, skb->active_extensions);
6356 if (__skb_ext_exist(new, id))
6359 newoff = new->chunks;
6361 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6363 new = __skb_ext_alloc(GFP_ATOMIC);
6368 newlen = newoff + skb_ext_type_len[id];
6369 new->chunks = newlen;
6370 new->offset[id] = newoff;
6372 skb->extensions = new;
6373 skb->active_extensions |= 1 << id;
6374 return skb_ext_get_ptr(new, id);
6376 EXPORT_SYMBOL(skb_ext_add);
6379 static void skb_ext_put_sp(struct sec_path *sp)
6383 for (i = 0; i < sp->len; i++)
6384 xfrm_state_put(sp->xvec[i]);
6388 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6390 struct skb_ext *ext = skb->extensions;
6392 skb->active_extensions &= ~(1 << id);
6393 if (skb->active_extensions == 0) {
6394 skb->extensions = NULL;
6397 } else if (id == SKB_EXT_SEC_PATH &&
6398 refcount_read(&ext->refcnt) == 1) {
6399 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6406 EXPORT_SYMBOL(__skb_ext_del);
6408 void __skb_ext_put(struct skb_ext *ext)
6410 /* If this is last clone, nothing can increment
6411 * it after check passes. Avoids one atomic op.
6413 if (refcount_read(&ext->refcnt) == 1)
6416 if (!refcount_dec_and_test(&ext->refcnt))
6420 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6421 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6424 kmem_cache_free(skbuff_ext_cache, ext);
6426 EXPORT_SYMBOL(__skb_ext_put);
6427 #endif /* CONFIG_SKB_EXTENSIONS */